From 7aa03d66b38e43563a7ee4f811c45695a4d59703 Mon Sep 17 00:00:00 2001 From: Nathan Hjelm Date: Tue, 22 Mar 2016 15:27:06 -0600 Subject: [PATCH 1/8] opal/memory: add support for patch based memory hooks This commit adds support for runtime binary patching. The support is broken down into two parts: util/opal_patcher.[ch] which contains the functionality for runtime patching of symbols, and mca/memory/patcher which patches the various symbols needed to provide support for memory hooks. This work is preliminary and is based off work donated by IBM. The patcher code is disabled if dlopen is disabled. Signed-off-by: Nathan Hjelm --- config/opal_configure_options.m4 | 3 +- opal/mca/memory/patcher/Makefile.am | 32 ++ opal/mca/memory/patcher/configure.m4 | 90 +++++ opal/mca/memory/patcher/memory_patcher.h | 27 ++ .../memory/patcher/memory_patcher_component.c | 276 ++++++++++++++ opal/util/Makefile.am | 4 + opal/util/opal_patcher.c | 347 ++++++++++++++++++ opal/util/opal_patcher.h | 57 +++ 8 files changed, 835 insertions(+), 1 deletion(-) create mode 100644 opal/mca/memory/patcher/Makefile.am create mode 100644 opal/mca/memory/patcher/configure.m4 create mode 100644 opal/mca/memory/patcher/memory_patcher.h create mode 100644 opal/mca/memory/patcher/memory_patcher_component.c create mode 100644 opal/util/opal_patcher.c create mode 100644 opal/util/opal_patcher.h diff --git a/config/opal_configure_options.m4 b/config/opal_configure_options.m4 index ce06b9edd25..06619dcee6b 100644 --- a/config/opal_configure_options.m4 +++ b/config/opal_configure_options.m4 @@ -283,7 +283,8 @@ else OPAL_ENABLE_DLOPEN_SUPPORT=1 AC_MSG_RESULT([yes]) fi - +AC_DEFINE_UNQUOTED(OPAL_ENABLE_DLOPEN_SUPPORT, $OPAL_ENABLE_DLOPEN_SUPPORT, + [Whether we want to enable dlopen support]) # # Heterogeneous support diff --git a/opal/mca/memory/patcher/Makefile.am b/opal/mca/memory/patcher/Makefile.am new file mode 100644 index 00000000000..ce4172617f3 --- /dev/null +++ b/opal/mca/memory/patcher/Makefile.am @@ -0,0 +1,32 @@ +# +# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana +# University Research and Technology +# Corporation. All rights reserved. +# Copyright (c) 2004-2005 The University of Tennessee and The University +# of Tennessee Research Foundation. All rights +# reserved. +# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, +# University of Stuttgart. All rights reserved. +# Copyright (c) 2004-2005 The Regents of the University of California. +# All rights reserved. +# Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. +# Copyright (c) 2015 Research Organization for Information Science +# and Technology (RIST). All rights reserved. +# Copyright (c) 2016 Los Alamos National Security, LLC. All rights +# reserved. +# $COPYRIGHT$ +# +# Additional copyrights may follow +# +# $HEADER$ +# + +# This component is only ever built statically (i.e., slurped into +# libopen-pal) -- it is never built as a DSO. +noinst_LTLIBRARIES = libmca_memory_patcher.la +libmca_memory_patcher_la_SOURCES = \ + memory_patcher.h \ + memory_patcher_component.c +libmca_memory_patcher_la_LDFLAGS = \ + -module -avoid-version $(memory_patcher_LDFLAGS) +libmca_memory_patcher_la_LIBADD = $(memory_patcher_LIBS) diff --git a/opal/mca/memory/patcher/configure.m4 b/opal/mca/memory/patcher/configure.m4 new file mode 100644 index 00000000000..bb19dd9985c --- /dev/null +++ b/opal/mca/memory/patcher/configure.m4 @@ -0,0 +1,90 @@ +# -*- shell-script -*- +# +# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana +# University Research and Technology +# Corporation. All rights reserved. +# Copyright (c) 2004-2005 The University of Tennessee and The University +# of Tennessee Research Foundation. All rights +# reserved. +# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, +# University of Stuttgart. All rights reserved. +# Copyright (c) 2004-2005 The Regents of the University of California. +# All rights reserved. +# Copyright (c) 2008-2010 Cisco Systems, Inc. All rights reserved. +# Copyright (c) 2015 Research Organization for Information Science +# and Technology (RIST). All rights reserved. +# Copyright (c) 2016 Los Alamos National Security, LLC. All rights +# reserved. +# $COPYRIGHT$ +# +# Additional copyrights may follow +# +# $HEADER$ +# +AC_DEFUN([MCA_opal_memory_patcher_PRIORITY], [40]) + +AC_DEFUN([MCA_opal_memory_patcher_COMPILE_MODE], [ + AC_MSG_CHECKING([for MCA component $2:$3 compile mode]) + $4="static" + AC_MSG_RESULT([$$4]) +]) + + +# MCA_memory_patcher_CONFIG(action-if-can-compile, +# [action-if-cant-compile]) +# ------------------------------------------------ +AC_DEFUN([MCA_opal_memory_patcher_CONFIG],[ + AC_CONFIG_FILES([opal/mca/memory/patcher/Makefile]) + + OPAL_VAR_SCOPE_PUSH([memory_patcher_have___curbrk memory_patcher_have___mmap memory_patcher_have___syscall memory_patcher_have___mmap_prototype memory_patcher_have___syscall_prototype]) + + memory_patcher_have___curbrk=0 + memory_patcher_have___mmap=0 + memory_patcher_have___mmap_prototype=0 + memory_patcher_have___syscall=0 + memory_patcher_have___syscall_prototype=0 + + AC_MSG_CHECKING([for __curbrk symbol]) + AC_LINK_IFELSE([AC_LANG_PROGRAM([extern char *__curbrk;],[char *tmp = __curbrk;])], + [AC_MSG_RESULT([yes]) + memory_patcher_have___curbrk=1], + [AC_MSG_RESULT([no])]) + AC_DEFINE_UNQUOTED([OPAL_MEMORY_PATCHER_HAVE___CURBRK], [$memory_patcher_have___curbrk], + [Whether the glibc __curbrk exists]) + + AC_MSG_CHECKING([whether __mmap prototype exists]) + AC_COMPILE_IFELSE([AC_LANG_PROGRAM([#include ],[char *tmp = __mmap (NULL, 0, 0, 0, 0, 0);])], + [AC_MSG_RESULT([yes]) + memory_patcher_have___mmap_prototype=1], + [AC_MSG_RESULT([no])]) + AC_DEFINE_UNQUOTED([OPAL_MEMORY_PATCHER_HAVE___MMAP_PROTO], [$memory_patcher_have___mmap_prototype], + [Whether the internal __mmap call has a prototype]) + + AC_MSG_CHECKING([whether __mmap symbol exists]) + AC_LINK_IFELSE([AC_LANG_PROGRAM([void *__mmap ();],[char *tmp = __mmap ();])], + [AC_MSG_RESULT([yes]) + memory_patcher_have___mmap=1], + [AC_MSG_RESULT([no])]) + AC_DEFINE_UNQUOTED([OPAL_MEMORY_PATCHER_HAVE___MMAP], [$memory_patcher_have___mmap], + [Whether the internal __mmap call exists]) + + AC_MSG_CHECKING([whether __syscall prototype exists]) + AC_COMPILE_IFELSE([AC_LANG_PROGRAM([#include ],[char *tmp = __syscall (SYS_mmap, NULL);])], + [AC_MSG_RESULT([yes]) + memory_patcher_have___syscall_prototype=1], + [AC_MSG_RESULT([no])]) + AC_DEFINE_UNQUOTED([OPAL_MEMORY_PATCHER_HAVE___SYSCALL_PROTO], [$memory_patcher_have___syscall_prototype], + [Whether the internal __syscall call has a prototype]) + + AC_MSG_CHECKING([whether __syscall symbol exists]) + AC_LINK_IFELSE([AC_LANG_PROGRAM([void *__syscall ();],[char *tmp = __syscall ();])], + [AC_MSG_RESULT([yes]) + memory_patcher_have___syscall=1], + [AC_MSG_RESULT([no])]) + AC_DEFINE_UNQUOTED([OPAL_MEMORY_PATCHER_HAVE___SYSCALL], [$memory_patcher_have___syscall], + [Whether the internal __syscall call exists]) + + [$1] + + OPAL_VAR_SCOPE_POP +]) diff --git a/opal/mca/memory/patcher/memory_patcher.h b/opal/mca/memory/patcher/memory_patcher.h new file mode 100644 index 00000000000..0d51a34922e --- /dev/null +++ b/opal/mca/memory/patcher/memory_patcher.h @@ -0,0 +1,27 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * ******** ADD IBM COPYRIGHT HERE ****** + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#if !defined(OPAL_MEMORY_PATCHER_H) +#define OPAL_MEMORY_PATCHER_H + +#include "opal_config.h" + +#include "opal/mca/memory/memory.h" +#include "opal/util/opal_patcher.h" + +typedef struct opal_memory_patcher_component_t { + opal_memory_base_component_2_0_0_t super; +} opal_memory_patcher_component_t; + +extern opal_memory_patcher_component_t mca_memory_patcher_component; + +#endif /* !defined(OPAL_MEMORY_PATCHER_H) */ diff --git a/opal/mca/memory/patcher/memory_patcher_component.c b/opal/mca/memory/patcher/memory_patcher_component.c new file mode 100644 index 00000000000..f8de71d60cb --- /dev/null +++ b/opal/mca/memory/patcher/memory_patcher_component.c @@ -0,0 +1,276 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana + * University Research and Technology + * Corporation. All rights reserved. + * Copyright (c) 2004-2005 The University of Tennessee and The University + * of Tennessee Research Foundation. All rights + * reserved. + * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, + * University of Stuttgart. All rights reserved. + * Copyright (c) 2004-2005 The Regents of the University of California. + * All rights reserved. + * Copyright (c) 2009-2014 Cisco Systems, Inc. All rights reserved. + * Copyright (c) 2013-2016 Los Alamos National Security, LLC. All rights + * reserved. + * Copyright (c) 2016 Research Organization for Information Science + * and Technology (RIST). All rights reserved. + * ******** Add IBM COPYRIGHT HERE *********** + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#include "memory_patcher.h" + +#include "opal/util/output.h" +#include "opal/util/show_help.h" +#include "opal/mca/memory/base/empty.h" +#include "opal/memoryhooks/memory.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "memory_patcher.h" +#undef opal_memory_changed + +static int patcher_open(void); +static int patcher_close(void); +static int patcher_register(void); + +opal_memory_patcher_component_t mca_memory_patcher_component = { + .super = { + .memoryc_version = { + OPAL_MEMORY_BASE_VERSION_2_0_0, + + /* Component name and version */ + .mca_component_name = "patcher", + MCA_BASE_MAKE_VERSION(component, OPAL_MAJOR_VERSION, OPAL_MINOR_VERSION, + OPAL_RELEASE_VERSION), + + /* Component open and close functions */ + .mca_open_component = patcher_open, + .mca_close_component = patcher_close, + .mca_register_component_params = patcher_register, + }, + .memoryc_data = { + /* The component is checkpoint ready */ + MCA_BASE_METADATA_PARAM_CHECKPOINT + }, + + /* Memory framework functions. */ + .memoryc_register = opal_memory_base_component_register_empty, + .memoryc_deregister = opal_memory_base_component_deregister_empty, + .memoryc_set_alignment = opal_memory_base_component_set_alignment_empty, + }, + + /* Component-specific data, filled in later (compiler will 0/NULL + it out) */ +}; + +#if OPAL_MEMORY_PATCHER_HAVE___MMAP && !OPAL_MEMORY_PATCHER_HAVE___MMAP_PROTO +/* prototype for Apple's internal mmap function */ +void *__mmap (void *start, size_t length, int prot, int flags, int fd, off_t offset); +#endif + +#if OPAL_MEMORY_PATCHER_HAVE___SYSCALL_PROTO && OPAL_MEMORY_PATCHER_HAVE___SYSCALL +/* calling __syscall is preferred on some systems when some arguments may be 64-bit. it also + * has the benefit of having an off_t return type */ +#define memory_patcher_syscall __syscall +#else +#define memory_patcher_syscall syscall +#endif + +static void *intercept_mmap(void *start, size_t length, int prot, int flags, int fd, off_t offset) +{ + OPAL_PATCHER_BEGIN; + void *result = 0; + + if (prot == PROT_NONE) { + opal_mem_hooks_release_hook (start, length, 0); + } + +#if OPAL_MEMORY_PATCHER_HAVE___MMAP + /* the darwin syscall returns an int not a long so call the underlying __mmap function */ + result = __mmap (start, length, prot, flags, fd, offset); +#else + result = (void*)(intptr_t) memory_patcher_syscall(SYS_mmap, start, length, prot, flags, fd, offset); +#endif + +// I thought we had some issue in the past with the above line for IA32, +// like maybe syscall() wouldn't handle that many arguments. But just now +// I used gcc -m32 and it worked on a recent system. But there's a possibility +// that older ia32 systems may need some other code to make the above syscall. + + OPAL_PATCHER_END; + return result; +} + +static int intercept_munmap(void *start, size_t length) +{ + OPAL_PATCHER_BEGIN; + int result = 0; + + opal_mem_hooks_release_hook (start, length, 0); + + result=memory_patcher_syscall(SYS_munmap, start, length); + + OPAL_PATCHER_END; + return result; +} + +#if defined (SYS_mremap) + +static void *intercept_mremap (void *start, size_t oldlen, size_t newlen, int flags, ...) +{ + OPAL_PATCHER_BEGIN; + void *result = 0; +#ifdef MREMAP_FIXED + va_list ap; + void *new_address; +#endif + + opal_mem_hooks_release_hook (start, oldlen, 0); + +#ifdef MREMAP_FIXED + if (flags & MREMAP_FIXED) { + va_start(ap, flags); + new_address = va_arg(ap, void*); + result=(void *)(intptr_t) memory_patcher_syscall( + SYS_mremap, start, oldlen, newlen, flags, new_address); + va_end(ap); + } else { + result=(void*)memory_patcher_syscall( + SYS_mremap, start, oldlen, newlen, flags); + } +#else + result=(void*)(intptr_t) memory_patcher_syscall(SYS_mremap, start, oldlen, newlen, flags); +#endif + + OPAL_PATCHER_END; + return result; +} + +#endif + +static int intercept_madvise (void *start, size_t length, int advice) +{ + OPAL_PATCHER_BEGIN; + int result = 0; + + if (advice == MADV_DONTNEED || +#ifdef MADV_REMOVE + advice == MADV_REMOVE || +#endif + advice == POSIX_MADV_DONTNEED) + { + opal_mem_hooks_release_hook (start, length, 0); + } + result = memory_patcher_syscall(SYS_madvise, start, length, advice); + + OPAL_PATCHER_END; + return result; +} + +#if defined SYS_brk + +#if OPAL_MEMORY_PATCHER_HAVE___CURBRK +void *__curbrk; /* in libc */ +#endif + +static int intercept_brk (void *addr) +{ + OPAL_PATCHER_BEGIN; + int result = 0; + void *old_addr, *new_addr; + +#if OPAL_MEMORY_PATCHER_HAVE___CURBRK + old_addr = __curbrk; +#else + old_addr = sbrk (0); +#endif + + /* get the current_addr */ + new_addr = (void *) (intptr_t) memory_patcher_syscall(SYS_brk, addr); + +#if OPAL_MEMORY_PATCHER_HAVE___CURBRK + /* + * Note: if we were using glibc brk/sbrk, their __curbrk would get + * updated, but since we're going straight to the syscall, we have + * to update __curbrk or else glibc won't see it. + */ + __curbrk = new_addr; +#endif + + if (new_addr < addr) { + errno = ENOMEM; + result = -1; + } else if (new_addr < old_addr) { + opal_mem_hooks_release_hook (new_addr, (intptr_t) old_addr - (intptr_t) new_addr, 0); + } + OPAL_PATCHER_END; + return result; +} + +#endif + +static int patcher_register (void) +{ + return OPAL_SUCCESS; +} + +static int patcher_open (void) +{ + static int was_executed_already = 0; + int rc; + + if (was_executed_already) { + return OPAL_SUCCESS; + } + + was_executed_already = 1; + + rc = opal_patch_symbol ("mmap", (uintptr_t) intercept_mmap); + if (OPAL_SUCCESS != rc) { + return rc; + } + + rc = opal_patch_symbol ("munmap", (uintptr_t)intercept_munmap); + if (OPAL_SUCCESS != rc) { + return rc; + } + +#if defined (SYS_mremap) + rc = opal_patch_symbol ("mremap",(uintptr_t)intercept_mremap); + if (OPAL_SUCCESS != rc) { + return rc; + } +#endif + + rc = opal_patch_symbol ("madvise", (uintptr_t)intercept_madvise); + if (OPAL_SUCCESS != rc) { + return rc; + } + +#if defined (SYS_brk) + rc = opal_patch_symbol ("brk", (uintptr_t)intercept_brk); +#endif + + return rc; +} + +static int patcher_close(void) +{ + /* NTH: is it possible to unpatch the symbols? */ + return OPAL_SUCCESS; +} diff --git a/opal/util/Makefile.am b/opal/util/Makefile.am index f80cbd50ec8..10c91ea5844 100644 --- a/opal/util/Makefile.am +++ b/opal/util/Makefile.am @@ -12,6 +12,8 @@ # Copyright (c) 2007-2015 Cisco Systems, Inc. All rights reserved. # Copyright (c) 2013 NVIDIA Corporation. All rights reserved. # Copyright (c) 2013 Intel, Inc. All rights reserved +# Copyright (c) 2016 Los Alamos National Security, LLC. All rights +# reserved. # $COPYRIGHT$ # # Additional copyrights may follow @@ -51,6 +53,7 @@ headers = \ numtostr.h \ opal_environ.h \ opal_getcwd.h \ + opal_patcher.h \ opal_pty.h \ os_dirpath.h \ os_path.h \ @@ -86,6 +89,7 @@ libopalutil_la_SOURCES = \ numtostr.c \ opal_environ.c \ opal_getcwd.c \ + opal_patcher.c \ opal_pty.c \ os_dirpath.c \ os_path.c \ diff --git a/opal/util/opal_patcher.c b/opal/util/opal_patcher.c new file mode 100644 index 00000000000..74e88c3a87d --- /dev/null +++ b/opal/util/opal_patcher.c @@ -0,0 +1,347 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * ******** ADD IBM COPYRIGHT HERE ****** + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#include "opal_patcher.h" + +#include "opal/constants.h" +#include "opal/util/sys_limits.h" +#include "opal/util/output.h" + +#include +#include +#include +#include +#include +#include +#include +#include + +#if OPAL_ENABLE_DLOPEN_SUPPORT && (defined(__i386__) || defined(__x86_64__) || defined(__ia64__)) + +static void flush_and_invalidate_cache (unsigned long a) +{ +#if defined(__i386__) + /* does not work with AMD processors */ + __asm__ volatile("mfence;clflush %0;mfence" : :"m" (*(char*)a)); +#elif defined(__x86_64__) + __asm__ volatile("mfence;clflush %0;mfence" : :"m" (*(char*)a)); +#elif defined(__ia64__) + __asm__ volatile ("fc %0;; sync.i;; srlz.i;;" : : "r"(a) : "memory"); +#endif +} + +#if defined(__ia64__) + +#define INSERT_BIT(d,p,v) do { \ + unsigned char c=*(d); \ + assert(((p) < 8) && ((p) >= 0)); \ + c&= ~(1<<(p)); \ + c|= ((v)<<(p)); \ + *(d) = c; \ + } while (0) + +static inline void +copy_instr_slot(unsigned char **dst, int *dst_bitpos, unsigned long instr_slot) +{ + for (int i = 40 ; i >= 0 ; --i) { + INSERT_BIT(*dst, *dst_bitpos, (instr_slot>>i)&1); + if (*dst_bitpos == 0) { + ++*dst; + *dst_bitpos = 7; + } else { + --*dst_bitpos; + } + } +} + +static void make_ia64_bundle (unsigned char *dst, + unsigned long i2, + unsigned long i1, + unsigned long i0, + unsigned template) +{ +/* + * each instr is 41 bits, template is 5 bits + * + * generate the bit concatenation of i2:i1:i0:t, all in all 128 bits + * + */ + + int dst_bitpos = 7; + + copy_instr_slot(&dst, &dst_bitpos, i2); + copy_instr_slot(&dst, &dst_bitpos, i1); + copy_instr_slot(&dst, &dst_bitpos, i0); + + assert(dst_bitpos == 4); + + for (int i = 4 ; i >= 0 ; --i) { + INSERT_BIT(dst, dst_bitpos, (template>>i)&1); + --dst_bitpos; + } +} +#endif /* defined(__ia64__) */ + +static int patch_code (uintptr_t func_old_addr, unsigned long func_new_addr) +{ + long pg_sz = opal_getpagesize (); + + if (mprotect((void*)(func_old_addr&~(pg_sz-1)), pg_sz, PROT_EXEC|PROT_READ|PROT_WRITE)) { + perror("mprotect failed\n"); + } + + { +#if defined(__i386__) + *(unsigned char*)(func_old_addr+0) = 0xe9; + *(unsigned int *)(func_old_addr+1) = (unsigned int)(func_new_addr - func_old_addr - 5); +#elif defined(__x86_64__) + *(unsigned short*)(func_old_addr+ 0) = 0xbb49; + *(unsigned long* )(func_old_addr+ 2) = (unsigned long) func_new_addr; + *(unsigned char*) (func_old_addr+10) = 0x41; + *(unsigned char*) (func_old_addr+11) = 0xff; + *(unsigned char*) (func_old_addr+12) = 0xe3; +#elif defined(__ia64__) + { +/* + * target64 = IP + ((i << 59 | imm39 << 20 | imm20) << 4) + * imm64 = i << 63 | imm41 << 22 | ic << 21 | imm5c << 16 | imm9d << 7 | imm7b + */ + unsigned char buf[16]; + unsigned long long imm64 = func_new_addr - func_old_addr - 16; + register unsigned long long glb_ptr __asm__("r1"); + unsigned long long nop = + (0x0ULL<<37) | /* O */ + (0x0ULL<<36) | /* i */ + (0x0ULL<<33) | /* x3 */ + (0x1ULL<<27) | /* x6 */ + (0x0ULL<< 6) | /* imm20 */ + (0x0ULL<< 0); /* qp */ + unsigned long long brl = + (0xcULL << 37) | + (((imm64>>63)&0x1ULL) << 36) | + (0x0ULL << 35) | + (0x0ULL << 33) | + (((imm64>>4)&0xFFFFFULL) << 13) | + (0x0ULL << 6) | + (0x0ULL << 0); + unsigned long long movl = + (0x6ULL << 37) | + (((glb_ptr>>63)&0x1ULL) << 36) | + (((glb_ptr>> 7)&0x1FFULL) << 27) | + (((glb_ptr>>16)&0x1FULL) << 22) | + (((glb_ptr>>21)&0x1ULL) << 21) | + (0ULL << 20) | + (((glb_ptr>> 0)&0x7FULL) << 13) | + (1ULL << 6) | + (0x0ULL << 0); + + make_ia64_bundle(buf, movl, (glb_ptr>>22)&0x1FFFFFFFFFFULL, nop, 5); + for (int i = 0 ; i < 16 ; ++i) { + ((unsigned char *)func_old_addr)[16-i-1] = buf[i]; + } + + make_ia64_bundle(buf, brl, ((imm64>>24)&0x7FFFFFFFFFULL)<<2, nop, 5); + for (int i = 0 ; i < 16 ; ++i) { + ((unsigned char *)func_old_addr+16)[16-i-1] = buf[i]; + } + } +#endif + } + + flush_and_invalidate_cache(func_old_addr+ 0); + flush_and_invalidate_cache(func_old_addr+16); + +#if 1 + if (mprotect((void*)(func_old_addr&~(pg_sz-1)), pg_sz, PROT_EXEC)) { + perror("mprotect failed\n"); + } +#endif + + return 0; +} + +int opal_patch_symbol (const char *func_symbol_name, uintptr_t func_new_addr) +{ + void *sym_addr; + char *error; + uintptr_t func_old_addr; + + /* NTH: might want to update opal/mca/dl to handle lookups in the default + * handle. */ + sym_addr = dlsym(RTLD_DEFAULT, func_symbol_name); + if ( (sym_addr == NULL) && ((error = dlerror()) != NULL) ) { + opal_output(0, "error locating symbol %s to patch. %s", func_symbol_name, + error); + return -1; + } + func_old_addr = (unsigned long)sym_addr; + +#if defined(__ia64__) + /* On IA64 addresses are all indirect */ + func_new_addr = *(unsigned long *)func_new_addr; + func_old_addr = *(unsigned long *)func_old_addr; +#endif + + patch_code(func_old_addr, func_new_addr); + return OPAL_SUCCESS; +} + +/* end of #if defined(__i386__) || defined(__x86_64__) || defined(__ia64__) */ +// ------------------------------------------------- PPC equivalent: +#elif OPAL_ENABLE_DLOPEN_SUPPORT && defined(__PPC__) + +static inline uintptr_t addr_text (uintptr_t addr) { +#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) && _CALL_ELF != 2 + struct odp_t { + uintptr_t text; + uintptr_t toc; + } *odp = (struct odp_t *) addr; + return (odp)?odp->text:0; +#else + return addr; +#endif +} + +// modify protection of memory range +static void +ModifyMemoryProtection(uintptr_t addr, size_t length, int prot) +{ + long page_size = opal_getpagesize (); + uintptr_t page_addr = (addr & ~(page_size-1)); + do { + if (mprotect((void *)page_addr, page_size, prot)) + perror("MemHook: mprotect failed"); + page_addr += page_size; + } while (page_addr < addr + length); +} + +// PowerPC instructions used in patching +// Reference: "PowerPC User Instruction Set Architecture" +unsigned int addis(unsigned int RT, unsigned int RS, unsigned int UI) { + return (15<<26) + (RT<<21) + (RS<<16) + (UI&0xffff); +} +unsigned int ori(unsigned int RT, unsigned int RS, unsigned int UI) { + return (24<<26) + (RS<<21) + (RT<<16) + (UI&0xffff); +} +unsigned int oris(unsigned int RT, unsigned int RS, unsigned int UI) { + return (25<<26) + (RS<<21) + (RT<<16) + (UI&0xffff); +} +unsigned int mtspr(unsigned int SPR, unsigned int RS) { + return (31<<26) + (RS<<21) + ((SPR&0x1f)<<16) + ((SPR>>5)<<11) + (467<<1); +} +unsigned int bcctr(unsigned int BO, unsigned int BI, unsigned int BH) { + return (19<<26) + (BO<<21) + (BI<<16) + (BH<<11) + (528<<1); +} +unsigned int rldicr(unsigned int RT, unsigned int RS, unsigned int SH, unsigned int MB) +{ + return (30<<26) + (RS<<21) + (RT<<16) + ((SH&0x1f)<<11) + ((SH>>5)<<1) + + ((MB&0x1f)<<6) + ((MB>>5)<<5) + (1<<2); +} + +static int +PatchLoadImm(uintptr_t addr, unsigned int reg, size_t value) +{ +#if defined(__PPC64__) + *(unsigned int *) (addr + 0) = addis ( reg, 0, (value >> 48)); + *(unsigned int *) (addr + 4) = ori ( reg, reg, (value >> 32)); + *(unsigned int *) (addr + 8) = rldicr( reg, reg, 32, 31); + *(unsigned int *) (addr +12) = oris ( reg, reg, (value >> 16)); + *(unsigned int *) (addr +16) = ori ( reg, reg, (value >> 0)); + return 20; +#else + *(unsigned int *) (addr + 0) = addis ( reg, 0, (value >> 16)); + *(unsigned int *) (addr + 4) = ori ( reg, reg, (value >> 0)); + return 8; +#endif +} + + +static void +apply_patch(char *patch, void *addr, int length) +{ + if (length == 0) { return; } + ModifyMemoryProtection((uintptr_t)addr, length, + PROT_EXEC|PROT_READ|PROT_WRITE); + memcpy((void *)addr, patch, length); + ModifyMemoryProtection((uintptr_t)addr, length, PROT_EXEC|PROT_READ); +} + +#define MAX_PATCH_SIZE 1024 +int opal_patch_symbol (const char *sys_func, uintptr_t hook_addr) +{ + void *addr_to_patch_hook; + int len_to_patch_hook = 0; + char patch_hook[MAX_PATCH_SIZE]; + + void *addr_to_patch_sys; + int len_to_patch_sys = 0; + char patch_sys[MAX_PATCH_SIZE]; + + int offset; +#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) + unsigned int *nop_addr; + const unsigned int nop = 0x60000000; +#endif + + // get system function address + uintptr_t sys_addr = add_text(dlsym(RTLD_NEXT, sys_func)); + if(sys_addr == 0) sys_addr = add_text(dlsym(RTLD_DEFAULT, sys_func)); + hook_addr = add_text(hook_addr); + +// Patch for hook function: +#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) + // locate reserved code space in hook function + nop_addr = (unsigned int *)hook_addr; + for (; ; nop_addr++) + if (nop_addr[0] == nop && nop_addr[1] == nop && nop_addr[2] == nop + && nop_addr[3] == nop && nop_addr[4] == nop) + break; + // generate code to restore TOC + register unsigned long toc asm("r2"); + addr_to_patch_hook = nop_addr; + len_to_patch_hook = PatchLoadImm((uintptr_t)patch_hook, 2, toc); + + // save the original code + assert(len_to_patch_hook <= MAX_PATCH_SIZE); + //memcpy(save, (void *)addr_to_patch_hook, len_to_patch_hook); // meh +#endif + +// Patch for system function: +#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) && _CALL_ELF == 2 + sys_addr += 8; + hook_addr += 8; +#endif /* _CALL_ELF == 2*/ + + addr_to_patch_sys = (void*) sys_addr; + // generate patch code + // r11 is a volatile register according to PowerPC EABI + const unsigned int gr = 11; + offset = PatchLoadImm((uintptr_t)patch_sys, gr, hook_addr); + *(unsigned int *) (patch_sys + offset + 0) = mtspr (9, gr); // 9 = CTR + *(unsigned int *) (patch_sys + offset + 4) = bcctr (20, 0, 0);// 20 = always + len_to_patch_sys = offset + 8; + + assert(len_to_patch_sys <= MAX_PATCH_SIZE); + //memcpy(save, (void *)addr_to_patch_sys, len_to_patch_sys); + + apply_patch(patch_hook, addr_to_patch_hook, len_to_patch_hook); + apply_patch(patch_sys, addr_to_patch_sys, len_to_patch_sys); + return OPAL_SUCCESS; +} + +#else + +int opal_patch_symbol (const char *sys_func, uintptr_t hook_addr) { + return OPAL_ERR_NOT_SUPPORTED; +} + +#endif diff --git a/opal/util/opal_patcher.h b/opal/util/opal_patcher.h new file mode 100644 index 00000000000..19c04c800fa --- /dev/null +++ b/opal/util/opal_patcher.h @@ -0,0 +1,57 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * ******** ADD IBM COPYRIGHT HERE ****** + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#if !defined(OPAL_PATCHER_H) +#define OPAL_PATCHER_H + +#include "opal_config.h" + +// Any function being patched in must use SYMBOLPATCH_BEGIN at the top, +// and SYMBOLPATCH_END before it returns (this is just for PPC). + +#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) && defined(OPAL_GCC_INLINE_ASSEMBLY) + +// special processing for ppc64 to save and restore TOC (r2) +// Reference: "64-bit PowerPC ELF Application Binary Interface Supplement 1.9" +#define OPAL_PATCHER_BEGIN \ + unsigned long toc_save; \ + asm volatile ("std 2, %0" : "=m" (toc_save)); \ + asm volatile ("nop; nop; nop; nop; nop"); +#define OPAL_PATCHER_END \ + asm volatile ("ld 2, %0" : : "m" (toc_save)); + +#else // !__PPC64__ + +#define OPAL_PATCHER_BEGIN +#define OPAL_PATCHER_END + +#endif + + +/** + * Patch all instances of calls to a function + * + * @param[in] func_symbol_name Name of symbol to patch + * @param[in] func_new_addr Pointer to new function to call + * + * @returns OPAL_SUCCESS on success + * @returns OPAL_ERR_NOT_AVAILABLE if symbol patching is not supported on this + * platform. + * + * This function patches any calls to the named symbol with a new function + * address. Any function that is passed into func_new_addr MUST begin with + * OPAL_PATCHER_BEGIN and end with OPAL_PATCHER_END. + */ +int opal_patch_symbol (const char *func_symbol_name, uintptr_t func_new_addr); + + +#endif /* !defined(OPAL_PATCHER_H) */ From 11e2d7886eaf90089e7bc85cad5bc1c89c38afcd Mon Sep 17 00:00:00 2001 From: Nathan Hjelm Date: Fri, 25 Mar 2016 14:03:19 -0600 Subject: [PATCH 2/8] opal/memory: update component structure This commit makes it possible to set relative priorities for components. Before the addition of the patched component there was only one component that would run on any system but that is no longer the case. When determining which component to open each component's query function is called and the one that returns the highest priority is opened. The default priority of the patcher component is set slightly higher than the old ptmalloc2/ummunotify component. This commit fixes a long-standing break in the abstration of the memory components. ompi_mpi_init.c was referencing the linux malloc hook initilize function to ensure the hooks are initialized for libmpi.so. The abstraction break has been fixed by adding a memory base function that calls the open memory component's malloc hook init function if it has one. The code is not yet complete but is intended to support ptmalloc in 2.0.0. In that case the base function will always call the ptmalloc hook init if exists. Signed-off-by: Nathan Hjelm --- ompi/runtime/ompi_mpi_init.c | 6 +- opal/mca/memory/base/base.h | 2 + opal/mca/memory/base/memory_base_open.c | 59 +++++++++++++------ opal/mca/memory/linux/hooks.c | 4 ++ opal/mca/memory/linux/memory_linux.h | 2 + .../mca/memory/linux/memory_linux_component.c | 23 +++++++- .../memory_malloc_solaris_component.c | 9 ++- opal/mca/memory/memory.h | 19 +++++- .../memory/patcher/memory_patcher_component.c | 19 ++++++ 9 files changed, 118 insertions(+), 25 deletions(-) diff --git a/ompi/runtime/ompi_mpi_init.c b/ompi/runtime/ompi_mpi_init.c index 59f365bd0da..667ae84d62e 100644 --- a/ompi/runtime/ompi_mpi_init.c +++ b/ompi/runtime/ompi_mpi_init.c @@ -98,15 +98,13 @@ #endif #include "ompi/runtime/ompi_cr.h" -#if defined(MEMORY_LINUX_PTMALLOC2) && MEMORY_LINUX_PTMALLOC2 -#include "opal/mca/memory/linux/memory_linux.h" +#include "opal/mca/memory/base/base.h" /* So this sucks, but with OPAL in its own library that is brought in implicity from libmpi, there are times when the malloc initialize hook in the memory component doesn't work. So we have to do it from here, since any MPI code is going to call MPI_Init... */ OPAL_DECLSPEC void (*__malloc_initialize_hook) (void) = - opal_memory_linux_malloc_init_hook; -#endif /* defined(MEMORY_LINUX_PTMALLOC2) && MEMORY_LINUX_PTMALLOC2 */ + opal_memory_base_malloc_init_hook; /* This is required for the boundaries of the hash tables used to store * the F90 types returned by the MPI_Type_create_f90_XXX functions. diff --git a/opal/mca/memory/base/base.h b/opal/mca/memory/base/base.h index 2316b6c4f13..63e5b568031 100644 --- a/opal/mca/memory/base/base.h +++ b/opal/mca/memory/base/base.h @@ -32,5 +32,7 @@ BEGIN_C_DECLS */ OPAL_DECLSPEC extern mca_base_framework_t opal_memory_base_framework; +OPAL_DECLSPEC void opal_memory_base_malloc_init_hook (void); + END_C_DECLS #endif /* OPAL_BASE_MEMORY_H */ diff --git a/opal/mca/memory/base/memory_base_open.c b/opal/mca/memory/base/memory_base_open.c index b27b1ae61d6..0560b93ae08 100644 --- a/opal/mca/memory/base/memory_base_open.c +++ b/opal/mca/memory/base/memory_base_open.c @@ -13,6 +13,8 @@ * Copyright (c) 2009 Cisco Systems, Inc. All rights reserved. * Copyright (c) 2016 Research Organization for Information Science * and Technology (RIST). All rights reserved. + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. * $COPYRIGHT$ * * Additional copyrights may follow @@ -43,20 +45,22 @@ static int empty_process(void) return OPAL_SUCCESS; } +static int empty_query (int *priority) +{ + *priority = 0; + return OPAL_SUCCESS; +} /* * Local variables */ static opal_memory_base_component_2_0_0_t empty_component = { - /* Don't care about the version info */ - { 0, }, - /* Don't care about the data */ - { 0, }, /* Empty / safe functions to call if no memory componet is selected */ - empty_process, - opal_memory_base_component_register_empty, - opal_memory_base_component_deregister_empty, - opal_memory_base_component_set_alignment_empty, + .memoryc_query = empty_query, + .memoryc_process = empty_process, + .memoryc_register = opal_memory_base_component_register_empty, + .memoryc_deregister = opal_memory_base_component_deregister_empty, + .memoryc_set_alignment = opal_memory_base_component_set_alignment_empty, }; @@ -66,6 +70,12 @@ static opal_memory_base_component_2_0_0_t empty_component = { opal_memory_base_component_2_0_0_t *opal_memory = &empty_component; +void opal_memory_base_malloc_init_hook (void) +{ + if (opal_memory->memoryc_init_hook) { + opal_memory->memoryc_init_hook (); + } +} /* * Function for finding and opening either all MCA components, or the one @@ -73,23 +83,36 @@ opal_memory_base_component_2_0_0_t *opal_memory = &empty_component; */ static int opal_memory_base_open(mca_base_open_flag_t flags) { + mca_base_component_list_item_t *item, *next; + opal_memory_base_component_2_0_0_t *tmp; + int priority, highest_priority = 0; int ret; - /* Open up all available components */ + /* can only be zero or one */ + OPAL_LIST_FOREACH(item, &opal_memory_base_framework.framework_components, mca_base_component_list_item_t) { + tmp = (opal_memory_base_component_2_0_0_t *) item->cli_component; + ret = tmp->memoryc_query (&priority); + if (OPAL_SUCCESS != ret || priority < highest_priority) { + continue; + } + + highest_priority = priority; + opal_memory = tmp; + } + + OPAL_LIST_FOREACH_SAFE(item, next, &opal_memory_base_framework.framework_components, mca_base_component_list_item_t) { + if ((void *) opal_memory != (void *) item->cli_component) { + mca_base_component_unload (item->cli_component, opal_memory_base_framework.framework_output); + opal_list_remove_item (&opal_memory_base_framework.framework_components, &item->super); + } + } + + /* open remaining component */ ret = mca_base_framework_components_open (&opal_memory_base_framework, flags); if (ret != OPAL_SUCCESS) { return ret; } - /* can only be zero or one */ - if (opal_list_get_size(&opal_memory_base_framework.framework_components) == 1) { - mca_base_component_list_item_t *item; - item = (mca_base_component_list_item_t*) - opal_list_get_first(&opal_memory_base_framework.framework_components); - opal_memory = (opal_memory_base_component_2_0_0_t*) - item->cli_component; - } - /* All done */ return OPAL_SUCCESS; } diff --git a/opal/mca/memory/linux/hooks.c b/opal/mca/memory/linux/hooks.c index 910d8e6e202..db608869b28 100644 --- a/opal/mca/memory/linux/hooks.c +++ b/opal/mca/memory/linux/hooks.c @@ -744,6 +744,10 @@ void opal_memory_linux_malloc_init_hook(void) check_result_t r1, lp, lpp; bool want_rcache = false, found_driver = false; + if (!opal_memory_linux_opened) { + return; + } + /* First, check for a FAKEROOT environment. If we're in a fakeroot, then access() (and likely others) have been replaced and are not safe to call here in this pre-main environment. So diff --git a/opal/mca/memory/linux/memory_linux.h b/opal/mca/memory/linux/memory_linux.h index 22685c14780..9334f3ae828 100644 --- a/opal/mca/memory/linux/memory_linux.h +++ b/opal/mca/memory/linux/memory_linux.h @@ -83,6 +83,8 @@ OPAL_DECLSPEC void opal_memory_linux_malloc_init_hook(void); OPAL_DECLSPEC void opal_memory_linux_malloc_set_alignment(int use_memalign, size_t memalign_threshold); #endif /* MEMORY_LINUX_MALLOC_ALIGN_ENABLED */ +extern bool opal_memory_linux_opened; + END_C_DECLS #endif diff --git a/opal/mca/memory/linux/memory_linux_component.c b/opal/mca/memory/linux/memory_linux_component.c index d883f10f2a9..323ab191344 100644 --- a/opal/mca/memory/linux/memory_linux_component.c +++ b/opal/mca/memory/linux/memory_linux_component.c @@ -11,7 +11,7 @@ * Copyright (c) 2004-2005 The Regents of the University of California. * All rights reserved. * Copyright (c) 2009-2014 Cisco Systems, Inc. All rights reserved. - * Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights + * Copyright (c) 2013-2016 Los Alamos National Security, LLC. All rights * reserved. * Copyright (c) 2016 Research Organization for Information Science * and Technology (RIST). All rights reserved. @@ -60,6 +60,7 @@ static int linux_open(void); static int linux_close(void); static int linux_register(void); +static int linux_query(int *); #if MEMORY_LINUX_UMMUNOTIFY static bool ummunotify_opened = false; @@ -69,6 +70,9 @@ static bool ptmalloc2_opened = false; #endif bool opal_memory_linux_disable = false; +static int mca_memory_linux_priority; + +bool opal_memory_linux_opened = false; opal_memory_linux_component_t mca_memory_linux_component = { /* First, the opal_memory_base_component_2_0_0_t */ @@ -96,6 +100,8 @@ opal_memory_linux_component_t mca_memory_linux_component = { /* Memory framework functions. These function pointer values are replaced by memory_linux_ummunotify.c at run time if we end up using ummunotify support. */ + .memoryc_init_hook = opal_memory_linux_malloc_init_hook, + .memoryc_query = linux_query, .memoryc_register = opal_memory_base_component_register_empty, .memoryc_deregister = opal_memory_base_component_deregister_empty, #if MEMORY_LINUX_MALLOC_ALIGN_ENABLED @@ -243,11 +249,25 @@ static int linux_register(void) if (0 > ret) { return ret; } + + mca_memory_linux_priority = 50; + ret = mca_base_component_var_register (&mca_memory_linux_component.super.memoryc_version, + "priority", "Priority of the linux memory hook component", + MCA_BASE_VAR_TYPE_INT, NULL, 0, 0, OPAL_INFO_LVL_5, + MCA_BASE_VAR_SCOPE_CONSTANT, &mca_memory_linux_priority); + if (0 > ret) { + return ret; + } #endif /* MEMORY_LINUX_MALLOC_ALIGN_ENABLED */ return (0 > ret) ? ret : OPAL_SUCCESS; } +static int linux_query (int *priority) +{ + *priority = mca_memory_linux_priority; + return OPAL_SUCCESS; +} static int linux_open(void) { @@ -318,6 +338,7 @@ static int linux_open(void) __malloc_hook = _opal_memory_linux_malloc_align_hook; } #endif /* MEMORY_LINUX_MALLOC_ALIGN_ENABLED */ + opal_memory_linux_opened = true; return OPAL_SUCCESS; } diff --git a/opal/mca/memory/malloc_solaris/memory_malloc_solaris_component.c b/opal/mca/memory/malloc_solaris/memory_malloc_solaris_component.c index 15fa63468d0..035d14ade41 100644 --- a/opal/mca/memory/malloc_solaris/memory_malloc_solaris_component.c +++ b/opal/mca/memory/malloc_solaris/memory_malloc_solaris_component.c @@ -12,7 +12,7 @@ * All rights reserved. * Copyright (c) 2007-2011 Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2009-2011 Cisco Systems, Inc. All rights reserved. - * Copyright (c) 2015 Los Alamos National Security, LLC. All rights + * Copyright (c) 2015-2016 Los Alamos National Security, LLC. All rights * reserved. * Copyright (c) 2016 Research Organization for Information Science * and Technology (RIST). All rights reserved. @@ -46,6 +46,7 @@ int __munmap(caddr_t addr, size_t len); #endif static int opal_memory_malloc_open(void); +static int opal_memory_malloc_query(int *); const opal_memory_base_component_2_0_0_t mca_memory_malloc_solaris_component = { /* First, the mca_component_t struct containing meta information @@ -68,6 +69,7 @@ const opal_memory_base_component_2_0_0_t mca_memory_malloc_solaris_component = { /* This component doesn't need these functions, but need to provide safe/empty register/deregister functions to call */ + .memoryc_query = opal_memory_malloc_query, .memoryc_register = opal_memory_base_component_register_empty, .memoryc_deregister = opal_memory_base_component_deregister_empty, .memoryc_set_alignment = opal_memory_base_component_set_alignment_empty, @@ -93,6 +95,11 @@ opal_memory_malloc_open(void) return OPAL_SUCCESS; } +static int opal_memory_malloc_query (int *priority) +{ + *priority = 79; + return OPAL_SUCCESS; +} /* * Three ways to call munmap. Prefered is to call __munmap, which diff --git a/opal/mca/memory/memory.h b/opal/mca/memory/memory.h index ab9ac49d78c..b5968fd2640 100644 --- a/opal/mca/memory/memory.h +++ b/opal/mca/memory/memory.h @@ -11,7 +11,7 @@ * Copyright (c) 2004-2005 The Regents of the University of California. * All rights reserved. * Copyright (c) 2009 Cisco Systems, Inc. All rights reserved. - * Copyright (c) 2015 Los Alamos National Security, LLC. All rights + * Copyright (c) 2015-2016 Los Alamos National Security, LLC. All rights * reserved. * Copyright (c) 2016 Research Organization for Information Science * and Technology (RIST). All rights reserved. @@ -78,6 +78,12 @@ BEGIN_C_DECLS */ typedef int (*opal_memory_base_component_process_fn_t)(void); +/** + * Prototype for a function that is invoked when the memory base is + * trying to select a component. This funtionality is required. + */ +typedef int (*opal_memory_base_component_query_fn_t)(int *priority); + /** * Prototype for a function that is invoked when Open MPI starts to * "care" about a specific memory region. That is, Open MPI declares @@ -119,6 +125,11 @@ typedef int (*opal_memory_base_component_deregister_fn_t)(void *base, typedef void (*opal_memory_base_component_set_alignment_fn_t)(int use_memalign, size_t memalign_threshold); +/** + * Function to be called when initializing malloc hooks + */ +typedef void (*opal_memory_base_component_init_hook_fn_t)(void); + /** * Structure for memory components. */ @@ -128,6 +139,12 @@ typedef struct opal_memory_base_component_2_0_0_t { /** MCA base data */ mca_base_component_data_t memoryc_data; + opal_memory_base_component_query_fn_t memoryc_query; + + /** This function will be called when the malloc hooks are + * initialized. It may be NULL if no hooks are needed. */ + opal_memory_base_component_init_hook_fn_t memoryc_init_hook; + /** Function to call when something has changed, as indicated by opal_memory_changed(). Will be ignored if the component does not provide an opal_memory_changed() macro that returns diff --git a/opal/mca/memory/patcher/memory_patcher_component.c b/opal/mca/memory/patcher/memory_patcher_component.c index f8de71d60cb..ad6ae05e03f 100644 --- a/opal/mca/memory/patcher/memory_patcher_component.c +++ b/opal/mca/memory/patcher/memory_patcher_component.c @@ -47,6 +47,9 @@ static int patcher_open(void); static int patcher_close(void); static int patcher_register(void); +static int patcher_query (int *); + +static int mca_memory_patcher_priority; opal_memory_patcher_component_t mca_memory_patcher_component = { .super = { @@ -69,6 +72,7 @@ opal_memory_patcher_component_t mca_memory_patcher_component = { }, /* Memory framework functions. */ + .memoryc_query = patcher_query, .memoryc_register = opal_memory_base_component_register_empty, .memoryc_deregister = opal_memory_base_component_deregister_empty, .memoryc_set_alignment = opal_memory_base_component_set_alignment_empty, @@ -226,6 +230,18 @@ static int intercept_brk (void *addr) static int patcher_register (void) { + mca_memory_patcher_priority = 80; + mca_base_component_var_register (&mca_memory_patcher_component.super.memoryc_version, + "priority", "Priority of the patcher memory hook component", + MCA_BASE_VAR_TYPE_INT, NULL, 0, 0, OPAL_INFO_LVL_5, + MCA_BASE_VAR_SCOPE_CONSTANT, &mca_memory_patcher_priority); + + return OPAL_SUCCESS; +} + +static int patcher_query (int *priority) +{ + *priority = mca_memory_patcher_priority; return OPAL_SUCCESS; } @@ -240,6 +256,9 @@ static int patcher_open (void) was_executed_already = 1; + /* set memory hooks support level */ + opal_mem_hooks_set_support (OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_MUNMAP_SUPPORT); + rc = opal_patch_symbol ("mmap", (uintptr_t) intercept_mmap); if (OPAL_SUCCESS != rc) { return rc; From 4cac623aeb8c2e85ee36ec1f939f52f1cbd8dddb Mon Sep 17 00:00:00 2001 From: Nathan Hjelm Date: Fri, 25 Mar 2016 14:08:37 -0600 Subject: [PATCH 3/8] opal/patch: add call to check if binary patching is supported Signed-off-by: Nathan Hjelm --- .../memory/patcher/memory_patcher_component.c | 6 +++++- opal/util/opal_patcher.c | 17 ++++++++++++++++- opal/util/opal_patcher.h | 4 ++++ 3 files changed, 25 insertions(+), 2 deletions(-) diff --git a/opal/mca/memory/patcher/memory_patcher_component.c b/opal/mca/memory/patcher/memory_patcher_component.c index ad6ae05e03f..1baf85a6de0 100644 --- a/opal/mca/memory/patcher/memory_patcher_component.c +++ b/opal/mca/memory/patcher/memory_patcher_component.c @@ -241,7 +241,11 @@ static int patcher_register (void) static int patcher_query (int *priority) { - *priority = mca_memory_patcher_priority; + if (opal_patch_supported ()) { + *priority = mca_memory_patcher_priority; + } else { + *priority = -1; + } return OPAL_SUCCESS; } diff --git a/opal/util/opal_patcher.c b/opal/util/opal_patcher.c index 74e88c3a87d..0463decbf52 100644 --- a/opal/util/opal_patcher.c +++ b/opal/util/opal_patcher.c @@ -195,6 +195,11 @@ int opal_patch_symbol (const char *func_symbol_name, uintptr_t func_new_addr) return OPAL_SUCCESS; } +bool opal_patch_supported (void) +{ + return true; +} + /* end of #if defined(__i386__) || defined(__x86_64__) || defined(__ia64__) */ // ------------------------------------------------- PPC equivalent: #elif OPAL_ENABLE_DLOPEN_SUPPORT && defined(__PPC__) @@ -338,10 +343,20 @@ int opal_patch_symbol (const char *sys_func, uintptr_t hook_addr) return OPAL_SUCCESS; } +bool opal_patch_supported (void) +{ + return true; +} #else -int opal_patch_symbol (const char *sys_func, uintptr_t hook_addr) { +int opal_patch_symbol (const char *sys_func, uintptr_t hook_addr) +{ return OPAL_ERR_NOT_SUPPORTED; } +bool opal_patch_supported (void) +{ + return false; +} + #endif diff --git a/opal/util/opal_patcher.h b/opal/util/opal_patcher.h index 19c04c800fa..571cbead7e5 100644 --- a/opal/util/opal_patcher.h +++ b/opal/util/opal_patcher.h @@ -53,5 +53,9 @@ */ int opal_patch_symbol (const char *func_symbol_name, uintptr_t func_new_addr); +/** + * Check if symbol patching is available + */ +bool opal_patch_supported (void); #endif /* !defined(OPAL_PATCHER_H) */ From b1670f844d1deff163119db6d779dab49eae735e Mon Sep 17 00:00:00 2001 From: Nathan Hjelm Date: Mon, 28 Mar 2016 12:16:03 -0600 Subject: [PATCH 4/8] contrib/platform: don't disable dlopen The --enable-static gives us what we want: statically linked components. Signed-off-by: Nathan Hjelm --- contrib/platform/lanl/cray_xe6/debug-common | 1 - contrib/platform/lanl/cray_xe6/optimized-common | 2 -- contrib/platform/lanl/toss/toss-common | 1 - 3 files changed, 4 deletions(-) diff --git a/contrib/platform/lanl/cray_xe6/debug-common b/contrib/platform/lanl/cray_xe6/debug-common index e7e6bcd99de..98f6a5f84e6 100644 --- a/contrib/platform/lanl/cray_xe6/debug-common +++ b/contrib/platform/lanl/cray_xe6/debug-common @@ -1,4 +1,3 @@ -enable_dlopen=no enable_mem_profile=no enable_binaries=yes enable_heterogeneous=no diff --git a/contrib/platform/lanl/cray_xe6/optimized-common b/contrib/platform/lanl/cray_xe6/optimized-common index 26877aa41bb..a6169ae173b 100644 --- a/contrib/platform/lanl/cray_xe6/optimized-common +++ b/contrib/platform/lanl/cray_xe6/optimized-common @@ -1,5 +1,3 @@ -enable_dlopen=no - enable_mem_profile=no enable_binaries=yes diff --git a/contrib/platform/lanl/toss/toss-common b/contrib/platform/lanl/toss/toss-common index 639672a3a54..ba000be23f0 100644 --- a/contrib/platform/lanl/toss/toss-common +++ b/contrib/platform/lanl/toss/toss-common @@ -1,7 +1,6 @@ # (c) 2013 Los Alamos National Security, LLC. All rights reserved. # Open MPI common configuration for TOSS/TOSS2 v1.7.x/1.8.x -enable_dlopen=no enable_binaries=yes enable_heterogeneous=no enable_shared=yes From 27f8a4e806adcdc9c506772ac66579e3d2b4f70d Mon Sep 17 00:00:00 2001 From: Nathan Hjelm Date: Mon, 28 Mar 2016 22:35:18 -0600 Subject: [PATCH 5/8] opal: add code patcher framework This commit adds a framework to abstract runtime code patching. Components in the new framework can provide functions for either patching a named function or a function pointer. The later functionality is not being used but may provide a way to allow memory hooks when dlopen functionality is disabled. This commit adds two different flavors of code patching. The first is provided by the overwrite component. This component overwrites the first several instructions of the target function with code to jump to the provided hook function. The hook is expected to provide the full functionality of the hooked function. The linux patcher component is based on the memory hooks in ucx. It only works on linux and operates by overwriting function pointers in the symbol table. In this case the hook is free to call the original function using the function pointer returned by dlsym. Both components restore the original functions when the patcher framework closes. Changes had to be made to support Power/PowerPC with the Linux dynamic loader patcher. Some of the changes: - Move code necessary for powerpc/power support to the patcher base. The code is needed by both the overwrite and linux components. - Move patch structure down to base and move the patch list to mca_patcher_base_module_t. The structure has been modified to include a function pointer to the function that will unapply the patch. This allows the mixing of multiple different types of patches in the patch_list. - Update linux patching code to keep track of the matching between got entry and original (unpatched) address. This allows us to completely clean up the patch on finalize. All patchers keep track of the changes they made so that they can be reversed when the patcher framework is closed. At this time there are bugs in the Linux dynamic loader patcher so its priority is lower than the overwrite patcher. Signed-off-by: Nathan Hjelm --- contrib/platform/lanl/cray_xe6/debug-common | 3 - .../platform/lanl/cray_xe6/optimized-common | 3 - opal/common_sym_whitelist.txt | 1 + opal/mca/memory/patcher/configure.m4 | 2 +- opal/mca/memory/patcher/memory_patcher.h | 2 +- .../memory/patcher/memory_patcher_component.c | 105 ++-- opal/mca/patcher/Makefile.am | 39 ++ opal/mca/patcher/base/Makefile.am | 25 + opal/mca/patcher/base/base.h | 83 ++++ opal/mca/patcher/base/patcher_base_frame.c | 81 +++ opal/mca/patcher/base/patcher_base_patch.c | 175 +++++++ opal/mca/patcher/linux/Makefile.am | 47 ++ opal/mca/patcher/linux/configure.m4 | 43 ++ opal/mca/patcher/linux/patcher_linux.h | 45 ++ .../patcher/linux/patcher_linux_component.c | 43 ++ opal/mca/patcher/linux/patcher_linux_module.c | 460 ++++++++++++++++++ opal/mca/patcher/overwrite/Makefile.am | 47 ++ opal/mca/patcher/overwrite/configure.m4 | 41 ++ .../mca/patcher/overwrite/patcher_overwrite.h | 32 ++ .../overwrite/patcher_overwrite_component.c | 45 ++ .../overwrite/patcher_overwrite_module.c} | 273 +++++------ opal/mca/patcher/patcher.h | 121 +++++ opal/runtime/opal_finalize.c | 2 + opal/runtime/opal_init.c | 9 + opal/util/Makefile.am | 2 - opal/util/opal_patcher.h | 61 --- 26 files changed, 1519 insertions(+), 271 deletions(-) create mode 100644 opal/mca/patcher/Makefile.am create mode 100644 opal/mca/patcher/base/Makefile.am create mode 100644 opal/mca/patcher/base/base.h create mode 100644 opal/mca/patcher/base/patcher_base_frame.c create mode 100644 opal/mca/patcher/base/patcher_base_patch.c create mode 100644 opal/mca/patcher/linux/Makefile.am create mode 100644 opal/mca/patcher/linux/configure.m4 create mode 100644 opal/mca/patcher/linux/patcher_linux.h create mode 100644 opal/mca/patcher/linux/patcher_linux_component.c create mode 100644 opal/mca/patcher/linux/patcher_linux_module.c create mode 100644 opal/mca/patcher/overwrite/Makefile.am create mode 100644 opal/mca/patcher/overwrite/configure.m4 create mode 100644 opal/mca/patcher/overwrite/patcher_overwrite.h create mode 100644 opal/mca/patcher/overwrite/patcher_overwrite_component.c rename opal/{util/opal_patcher.c => mca/patcher/overwrite/patcher_overwrite_module.c} (50%) create mode 100644 opal/mca/patcher/patcher.h delete mode 100644 opal/util/opal_patcher.h diff --git a/contrib/platform/lanl/cray_xe6/debug-common b/contrib/platform/lanl/cray_xe6/debug-common index 98f6a5f84e6..6bb806362d5 100644 --- a/contrib/platform/lanl/cray_xe6/debug-common +++ b/contrib/platform/lanl/cray_xe6/debug-common @@ -32,8 +32,5 @@ enable_mca_direct=pml-ob1 # enable development headers with_devel_headers=yes -# enable ptmalloc (enables lazy deregistration) -with_memory_manager=linux - # disable valgrind with_valgrind=no diff --git a/contrib/platform/lanl/cray_xe6/optimized-common b/contrib/platform/lanl/cray_xe6/optimized-common index a6169ae173b..937bf6b5982 100644 --- a/contrib/platform/lanl/cray_xe6/optimized-common +++ b/contrib/platform/lanl/cray_xe6/optimized-common @@ -38,8 +38,5 @@ enable_mca_direct=pml-ob1 # enable development headers with_devel_headers=yes -# enable ptmalloc (enables lazy deregistration) -with_memory_manager=linux - # disable valgrind with_valgrind=no diff --git a/opal/common_sym_whitelist.txt b/opal/common_sym_whitelist.txt index d3c525b54d9..4f2ffe1042e 100644 --- a/opal/common_sym_whitelist.txt +++ b/opal/common_sym_whitelist.txt @@ -4,3 +4,4 @@ opal_show_help_yyleng opal_show_help_yytext opal_util_keyval_yyleng opal_util_keyval_yytext +__curbrk diff --git a/opal/mca/memory/patcher/configure.m4 b/opal/mca/memory/patcher/configure.m4 index bb19dd9985c..4e2949fd636 100644 --- a/opal/mca/memory/patcher/configure.m4 +++ b/opal/mca/memory/patcher/configure.m4 @@ -21,7 +21,7 @@ # # $HEADER$ # -AC_DEFUN([MCA_opal_memory_patcher_PRIORITY], [40]) +AC_DEFUN([MCA_opal_memory_patcher_PRIORITY], [41]) AC_DEFUN([MCA_opal_memory_patcher_COMPILE_MODE], [ AC_MSG_CHECKING([for MCA component $2:$3 compile mode]) diff --git a/opal/mca/memory/patcher/memory_patcher.h b/opal/mca/memory/patcher/memory_patcher.h index 0d51a34922e..0650f30c159 100644 --- a/opal/mca/memory/patcher/memory_patcher.h +++ b/opal/mca/memory/patcher/memory_patcher.h @@ -16,7 +16,7 @@ #include "opal_config.h" #include "opal/mca/memory/memory.h" -#include "opal/util/opal_patcher.h" +#include "opal/mca/patcher/patcher.h" typedef struct opal_memory_patcher_component_t { opal_memory_base_component_2_0_0_t super; diff --git a/opal/mca/memory/patcher/memory_patcher_component.c b/opal/mca/memory/patcher/memory_patcher_component.c index 1baf85a6de0..cacefc3232f 100644 --- a/opal/mca/memory/patcher/memory_patcher_component.c +++ b/opal/mca/memory/patcher/memory_patcher_component.c @@ -28,6 +28,7 @@ #include "opal/util/output.h" #include "opal/util/show_help.h" #include "opal/mca/memory/base/empty.h" +#include "opal/mca/memory/base/base.h" #include "opal/memoryhooks/memory.h" #include @@ -95,39 +96,51 @@ void *__mmap (void *start, size_t length, int prot, int flags, int fd, off_t off #define memory_patcher_syscall syscall #endif +static void *(*original_mmap)(void *, size_t, int, int, int, off_t); + static void *intercept_mmap(void *start, size_t length, int prot, int flags, int fd, off_t offset) { OPAL_PATCHER_BEGIN; void *result = 0; if (prot == PROT_NONE) { - opal_mem_hooks_release_hook (start, length, 0); + opal_mem_hooks_release_hook (start, length, true); } + if (!original_mmap) { #if OPAL_MEMORY_PATCHER_HAVE___MMAP - /* the darwin syscall returns an int not a long so call the underlying __mmap function */ - result = __mmap (start, length, prot, flags, fd, offset); + /* the darwin syscall returns an int not a long so call the underlying __mmap function */ + result = __mmap (start, length, prot, flags, fd, offset); #else - result = (void*)(intptr_t) memory_patcher_syscall(SYS_mmap, start, length, prot, flags, fd, offset); + result = (void*)(intptr_t) memory_patcher_syscall(SYS_mmap, start, length, prot, flags, fd, offset); #endif -// I thought we had some issue in the past with the above line for IA32, -// like maybe syscall() wouldn't handle that many arguments. But just now -// I used gcc -m32 and it worked on a recent system. But there's a possibility -// that older ia32 systems may need some other code to make the above syscall. + // I thought we had some issue in the past with the above line for IA32, + // like maybe syscall() wouldn't handle that many arguments. But just now + // I used gcc -m32 and it worked on a recent system. But there's a possibility + // that older ia32 systems may need some other code to make the above syscall. + } else { + result = original_mmap (start, length, prot, flags, fd, offset); + } OPAL_PATCHER_END; return result; } +static int (*original_munmap) (void *, size_t); + static int intercept_munmap(void *start, size_t length) { OPAL_PATCHER_BEGIN; int result = 0; - opal_mem_hooks_release_hook (start, length, 0); + opal_mem_hooks_release_hook (start, length, false); - result=memory_patcher_syscall(SYS_munmap, start, length); + if (!original_munmap) { + result = memory_patcher_syscall(SYS_munmap, start, length); + } else { + result = original_munmap (start, length); + } OPAL_PATCHER_END; return result; @@ -135,38 +148,42 @@ static int intercept_munmap(void *start, size_t length) #if defined (SYS_mremap) +static void *(*original_mremap) (void *, size_t, size_t, int, ...); + static void *intercept_mremap (void *start, size_t oldlen, size_t newlen, int flags, ...) { OPAL_PATCHER_BEGIN; void *result = 0; #ifdef MREMAP_FIXED va_list ap; - void *new_address; #endif + void *new_address = NULL; - opal_mem_hooks_release_hook (start, oldlen, 0); + opal_mem_hooks_release_hook (start, oldlen, false); #ifdef MREMAP_FIXED if (flags & MREMAP_FIXED) { va_start(ap, flags); new_address = va_arg(ap, void*); - result=(void *)(intptr_t) memory_patcher_syscall( - SYS_mremap, start, oldlen, newlen, flags, new_address); + va_end(ap); - } else { - result=(void*)memory_patcher_syscall( - SYS_mremap, start, oldlen, newlen, flags); } -#else - result=(void*)(intptr_t) memory_patcher_syscall(SYS_mremap, start, oldlen, newlen, flags); #endif + if (!original_mremap) { + result = (void *)(intptr_t) memory_patcher_syscall (SYS_mremap, start, oldlen, newlen, flags, new_address); + } else { + result = original_mremap (start, oldlen, newlen, flags, new_address); + } + OPAL_PATCHER_END; return result; } #endif +static int (*original_madvise) (void *, size_t, int); + static int intercept_madvise (void *start, size_t length, int advice) { OPAL_PATCHER_BEGIN; @@ -178,9 +195,14 @@ static int intercept_madvise (void *start, size_t length, int advice) #endif advice == POSIX_MADV_DONTNEED) { - opal_mem_hooks_release_hook (start, length, 0); + opal_mem_hooks_release_hook (start, length, false); + } + + if (!original_madvise) { + result = memory_patcher_syscall(SYS_madvise, start, length, advice); + } else { + result = original_madvise (start, length, advice); } - result = memory_patcher_syscall(SYS_madvise, start, length, advice); OPAL_PATCHER_END; return result; @@ -192,6 +214,8 @@ static int intercept_madvise (void *start, size_t length, int advice) void *__curbrk; /* in libc */ #endif +static int (*original_brk) (void *); + static int intercept_brk (void *addr) { OPAL_PATCHER_BEGIN; @@ -204,23 +228,32 @@ static int intercept_brk (void *addr) old_addr = sbrk (0); #endif - /* get the current_addr */ - new_addr = (void *) (intptr_t) memory_patcher_syscall(SYS_brk, addr); + if (!original_brk) { + /* get the current_addr */ + new_addr = (void *) (intptr_t) memory_patcher_syscall(SYS_brk, addr); #if OPAL_MEMORY_PATCHER_HAVE___CURBRK - /* - * Note: if we were using glibc brk/sbrk, their __curbrk would get - * updated, but since we're going straight to the syscall, we have - * to update __curbrk or else glibc won't see it. - */ - __curbrk = new_addr; + /* + * Note: if we were using glibc brk/sbrk, their __curbrk would get + * updated, but since we're going straight to the syscall, we have + * to update __curbrk or else glibc won't see it. + */ + __curbrk = new_addr; #endif + } else { + result = original_brk (addr); +#if OPAL_MEMORY_PATCHER_HAVE___CURBRK + new_addr = __curbrk; +#else + new_addr = sbrk (0); +#endif + } if (new_addr < addr) { errno = ENOMEM; result = -1; } else if (new_addr < old_addr) { - opal_mem_hooks_release_hook (new_addr, (intptr_t) old_addr - (intptr_t) new_addr, 0); + opal_mem_hooks_release_hook (new_addr, (intptr_t) old_addr - (intptr_t) new_addr, true); } OPAL_PATCHER_END; return result; @@ -241,7 +274,7 @@ static int patcher_register (void) static int patcher_query (int *priority) { - if (opal_patch_supported ()) { + if (opal_patcher->patch_symbol) { *priority = mca_memory_patcher_priority; } else { *priority = -1; @@ -263,30 +296,30 @@ static int patcher_open (void) /* set memory hooks support level */ opal_mem_hooks_set_support (OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_MUNMAP_SUPPORT); - rc = opal_patch_symbol ("mmap", (uintptr_t) intercept_mmap); + rc = opal_patcher->patch_symbol ("mmap", (uintptr_t) intercept_mmap, (uintptr_t *) &original_mmap); if (OPAL_SUCCESS != rc) { return rc; } - rc = opal_patch_symbol ("munmap", (uintptr_t)intercept_munmap); + rc = opal_patcher->patch_symbol ("munmap", (uintptr_t)intercept_munmap, (uintptr_t *) &original_munmap); if (OPAL_SUCCESS != rc) { return rc; } #if defined (SYS_mremap) - rc = opal_patch_symbol ("mremap",(uintptr_t)intercept_mremap); + rc = opal_patcher->patch_symbol ("mremap",(uintptr_t)intercept_mremap, (uintptr_t *) &original_mremap); if (OPAL_SUCCESS != rc) { return rc; } #endif - rc = opal_patch_symbol ("madvise", (uintptr_t)intercept_madvise); + rc = opal_patcher->patch_symbol ("madvise", (uintptr_t)intercept_madvise, (uintptr_t *) &original_madvise); if (OPAL_SUCCESS != rc) { return rc; } #if defined (SYS_brk) - rc = opal_patch_symbol ("brk", (uintptr_t)intercept_brk); + rc = opal_patcher->patch_symbol ("brk", (uintptr_t)intercept_brk, (uintptr_t *) &original_brk); #endif return rc; diff --git a/opal/mca/patcher/Makefile.am b/opal/mca/patcher/Makefile.am new file mode 100644 index 00000000000..664683bad89 --- /dev/null +++ b/opal/mca/patcher/Makefile.am @@ -0,0 +1,39 @@ +# +# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana +# University Research and Technology +# Corporation. All rights reserved. +# Copyright (c) 2004-2005 The University of Tennessee and The University +# of Tennessee Research Foundation. All rights +# reserved. +# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, +# University of Stuttgart. All rights reserved. +# Copyright (c) 2004-2005 The Regents of the University of California. +# All rights reserved. +# Copyright (c) 2010 Cisco Systems, Inc. All rights reserved. +# Copyright (c) 2016 Los Alamos National Security, LLC. All rights +# reserved. +# $COPYRIGHT$ +# +# Additional copyrights may follow +# +# $HEADER$ +# + +# main library setup +noinst_LTLIBRARIES = libmca_patcher.la +libmca_patcher_la_SOURCES = + +# local files +headers = patcher.h +libmca_patcher_la_SOURCES += $(headers) + +# Conditionally install the header files +if WANT_INSTALL_HEADERS +opaldir = $(opalincludedir)/$(subdir) +nobase_opal_HEADERS = $(headers) +endif + +include base/Makefile.am + +distclean-local: + rm -f base/static-components.h diff --git a/opal/mca/patcher/base/Makefile.am b/opal/mca/patcher/base/Makefile.am new file mode 100644 index 00000000000..441e1c645d2 --- /dev/null +++ b/opal/mca/patcher/base/Makefile.am @@ -0,0 +1,25 @@ +# +# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana +# University Research and Technology +# Corporation. All rights reserved. +# Copyright (c) 2004-2005 The University of Tennessee and The University +# of Tennessee Research Foundation. All rights +# reserved. +# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, +# University of Stuttgart. All rights reserved. +# Copyright (c) 2004-2005 The Regents of the University of California. +# All rights reserved. +# Copyright (c) 2009 Cisco Systems, Inc. All rights reserved. +# Copyright (c) 2016 Los Alamos National Security, LLC. All rights +# reserved. +# $COPYRIGHT$ +# +# Additional copyrights may follow +# +# $HEADER$ +# + +headers += base/base.h + +libmca_patcher_la_SOURCES += base/patcher_base_frame.c \ + base/patcher_base_patch.c diff --git a/opal/mca/patcher/base/base.h b/opal/mca/patcher/base/base.h new file mode 100644 index 00000000000..65b48fc00c0 --- /dev/null +++ b/opal/mca/patcher/base/base.h @@ -0,0 +1,83 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana + * University Research and Technology + * Corporation. All rights reserved. + * Copyright (c) 2004-2006 The University of Tennessee and The University + * of Tennessee Research Foundation. All rights + * reserved. + * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, + * University of Stuttgart. All rights reserved. + * Copyright (c) 2004-2005 The Regents of the University of California. + * All rights reserved. + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + * + */ + +#ifndef OPAL_PATCHER_BASE_H +#define OPAL_PATCHER_BASE_H + +#include "opal_config.h" +#include "opal/mca/base/mca_base_framework.h" +#include "opal/mca/patcher/patcher.h" + + +BEGIN_C_DECLS + +#define MCA_BASE_PATCHER_MAX_PATCH 32 + +struct mca_patcher_base_patch_t; + +typedef void (*mca_patcher_base_restore_fn_t) (struct mca_patcher_base_patch_t *); + +struct mca_patcher_base_patch_t { + /** patches are list items */ + opal_list_item_t super; + /** name symbol to patch */ + char *patch_symbol; + /** address of function to call instead */ + uintptr_t patch_value; + /** original address of function */ + uintptr_t patch_orig; + /** patch data */ + unsigned char patch_data[MCA_BASE_PATCHER_MAX_PATCH]; + /** original data */ + unsigned char patch_orig_data[MCA_BASE_PATCHER_MAX_PATCH]; + /** size of patch data */ + unsigned patch_data_size; + /** function to undo the patch */ + mca_patcher_base_restore_fn_t patch_restore; +}; + +typedef struct mca_patcher_base_patch_t mca_patcher_base_patch_t; + +OBJ_CLASS_DECLARATION(mca_patcher_base_patch_t); + +/** + * Framework struct declaration for this framework + */ +OPAL_DECLSPEC extern mca_base_framework_t opal_patcher_base_framework; +OPAL_DECLSPEC int opal_patcher_base_select (void); +OPAL_DECLSPEC int mca_patcher_base_patch_hook (mca_patcher_base_module_t *module, uintptr_t hook); +OPAL_DECLSPEC void mca_base_patcher_patch_apply_binary (mca_patcher_base_patch_t *patch); + +static inline uintptr_t mca_patcher_base_addr_text (uintptr_t addr) { +#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) && _CALL_ELF != 2 + struct odp_t { + uintptr_t text; + uintptr_t toc; + } *odp = (struct odp_t *) addr; + return (odp)?odp->text:0; +#else + return addr; +#endif +} + +END_C_DECLS +#endif /* OPAL_BASE_PATCHER_H */ diff --git a/opal/mca/patcher/base/patcher_base_frame.c b/opal/mca/patcher/base/patcher_base_frame.c new file mode 100644 index 00000000000..8d685d3fa43 --- /dev/null +++ b/opal/mca/patcher/base/patcher_base_frame.c @@ -0,0 +1,81 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#include "opal_config.h" + +#include "opal/mca/patcher/patcher.h" +#include "opal/mca/patcher/base/base.h" +#include "opal/mca/patcher/base/static-components.h" + +/* + * Local variables + */ +static mca_patcher_base_module_t empty_module; + +/* + * Globals + */ +mca_patcher_base_module_t *opal_patcher = &empty_module; + +int opal_patcher_base_select (void) +{ + mca_patcher_base_module_t *best_module; + mca_patcher_base_component_t *best_component; + int rc, priority; + + rc = mca_base_select ("patcher", opal_patcher_base_framework.framework_output, + &opal_patcher_base_framework.framework_components, + (mca_base_module_t **) &best_module, (mca_base_component_t **) &best_component, + &priority); + if (OPAL_SUCCESS != rc) { + return rc; + } + + OBJ_CONSTRUCT(&best_module->patch_list, opal_list_t); + OBJ_CONSTRUCT(&best_module->patch_list_mutex, opal_mutex_t); + + if (best_module->patch_init) { + rc = best_module->patch_init (); + if (OPAL_SUCCESS != rc) { + return rc; + } + } + + opal_patcher = best_module; + + return OPAL_SUCCESS; +} + +static int opal_patcher_base_close (void) +{ + if (opal_patcher == &empty_module) { + return OPAL_SUCCESS; + } + + mca_patcher_base_patch_t *patch; + OPAL_LIST_FOREACH_REV(patch, &opal_patcher->patch_list, mca_patcher_base_patch_t) { + patch->patch_restore (patch); + } + + OPAL_LIST_DESTRUCT(&opal_patcher->patch_list); + OBJ_DESTRUCT(&opal_patcher->patch_list_mutex); + + if (opal_patcher->patch_fini) { + return opal_patcher->patch_fini (); + } + + return OPAL_SUCCESS; +} + +/* Use default register/open functions */ +MCA_BASE_FRAMEWORK_DECLARE(opal, patcher, "runtime code patching", NULL, NULL, + opal_patcher_base_close, mca_patcher_base_static_components, + 0); diff --git a/opal/mca/patcher/base/patcher_base_patch.c b/opal/mca/patcher/base/patcher_base_patch.c new file mode 100644 index 00000000000..94a0c12e387 --- /dev/null +++ b/opal/mca/patcher/base/patcher_base_patch.c @@ -0,0 +1,175 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#include "opal_config.h" + +#include "opal/mca/patcher/patcher.h" +#include "opal/mca/patcher/base/base.h" +#include "opal/util/sys_limits.h" +#include "opal/prefetch.h" +#include + +static void mca_patcher_base_patch_construct (mca_patcher_base_patch_t *patch) +{ + patch->patch_symbol = NULL; + patch->patch_data_size = 0; +} + +static void mca_patcher_base_patch_destruct (mca_patcher_base_patch_t *patch) +{ + free (patch->patch_symbol); +} + +OBJ_CLASS_INSTANCE(mca_patcher_base_patch_t, opal_list_item_t, + mca_patcher_base_patch_construct, + mca_patcher_base_patch_destruct); + +#if defined(__PPC__) + +// PowerPC instructions used in patching +// Reference: "PowerPC User Instruction Set Architecture" +static unsigned int addis(unsigned int RT, unsigned int RS, unsigned int UI) { + return (15<<26) + (RT<<21) + (RS<<16) + (UI&0xffff); +} +static unsigned int ori(unsigned int RT, unsigned int RS, unsigned int UI) { + return (24<<26) + (RS<<21) + (RT<<16) + (UI&0xffff); +} +static unsigned int oris(unsigned int RT, unsigned int RS, unsigned int UI) { + return (25<<26) + (RS<<21) + (RT<<16) + (UI&0xffff); +} +static unsigned int mtspr(unsigned int SPR, unsigned int RS) { + return (31<<26) + (RS<<21) + ((SPR&0x1f)<<16) + ((SPR>>5)<<11) + (467<<1); +} +static unsigned int bcctr(unsigned int BO, unsigned int BI, unsigned int BH) { + return (19<<26) + (BO<<21) + (BI<<16) + (BH<<11) + (528<<1); +} +static unsigned int rldicr(unsigned int RT, unsigned int RS, unsigned int SH, unsigned int MB) +{ + return (30<<26) + (RS<<21) + (RT<<16) + ((SH&0x1f)<<11) + ((SH>>5)<<1) + + ((MB&0x1f)<<6) + ((MB>>5)<<5) + (1<<2); +} + +static int PatchLoadImm (uintptr_t addr, unsigned int reg, size_t value) +{ +#if defined(__PPC64__) + *(unsigned int *) (addr + 0) = addis ( reg, 0, (value >> 48)); + *(unsigned int *) (addr + 4) = ori ( reg, reg, (value >> 32)); + *(unsigned int *) (addr + 8) = rldicr( reg, reg, 32, 31); + *(unsigned int *) (addr +12) = oris ( reg, reg, (value >> 16)); + *(unsigned int *) (addr +16) = ori ( reg, reg, (value >> 0)); + return 20; +#else + *(unsigned int *) (addr + 0) = addis ( reg, 0, (value >> 16)); + *(unsigned int *) (addr + 4) = ori ( reg, reg, (value >> 0)); + return 8; +#endif +} + +#endif + +#if defined(__i386__) || defined(__x86_64__) || defined(__ia64__) + +static void flush_and_invalidate_cache (unsigned long a) +{ +#if defined(__i386__) + /* does not work with AMD processors */ + __asm__ volatile("mfence;clflush %0;mfence" : :"m" (*(char*)a)); +#elif defined(__x86_64__) + __asm__ volatile("mfence;clflush %0;mfence" : :"m" (*(char*)a)); +#elif defined(__ia64__) + __asm__ volatile ("fc %0;; sync.i;; srlz.i;;" : : "r"(a) : "memory"); +#endif +} +#endif + +// modify protection of memory range +static void ModifyMemoryProtection (uintptr_t addr, size_t length, int prot) +{ + long page_size = opal_getpagesize (); + uintptr_t base = (addr & ~(page_size-1)); + uintptr_t bound = ((addr + length + page_size-1) & ~(page_size-1)); + + length = bound - base; + +#if defined(__PPC__) + /* NTH: is a loop necessary here? */ + do { + if (mprotect((void *)base, page_size, prot)) + perror("MemHook: mprotect failed"); + base += page_size; + } while (base < addr + length); +#else + if (mprotect((void *) base, length, prot)) { + perror("MemHook: mprotect failed"); + } +#endif +} + +static inline void apply_patch (unsigned char *patch_data, uintptr_t address, size_t data_size) +{ + ModifyMemoryProtection (address, data_size, PROT_EXEC|PROT_READ|PROT_WRITE); + memcpy ((void *) address, patch_data, data_size); +#if defined(__i386__) || defined(__x86_64__) || defined(__ia64__) + for (size_t i = 0 ; i < data_size ; i += 16) { + flush_and_invalidate_cache (address + i); + } +#endif + + ModifyMemoryProtection (address, data_size, PROT_EXEC|PROT_READ); +} + +static void mca_base_patcher_patch_unapply_binary (mca_patcher_base_patch_t *patch) +{ + apply_patch (patch->patch_orig_data, patch->patch_orig, patch->patch_data_size); +} + +void mca_base_patcher_patch_apply_binary (mca_patcher_base_patch_t *patch) +{ + memcpy (patch->patch_orig_data, (void *) patch->patch_orig, patch->patch_data_size); + apply_patch (patch->patch_data, patch->patch_orig, patch->patch_data_size); + patch->patch_restore = mca_base_patcher_patch_unapply_binary; +} + + +int mca_patcher_base_patch_hook (mca_patcher_base_module_t *module, uintptr_t hook_addr) +{ +#if defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__) + mca_patcher_base_patch_t *hook_patch; + const unsigned int nop = 0x60000000; + unsigned int *nop_addr; + + fprintf (stderr, "Patching hook @ 0x%lx\n", hook_addr); + + hook_patch = OBJ_NEW(mca_patcher_base_patch_t); + if (OPAL_UNLIKELY(NULL == hook_patch)) { + return OPAL_ERR_OUT_OF_RESOURCE; + } + + // locate reserved code space in hook function + for (nop_addr = (unsigned int *)hook_addr ; ; nop_addr++) { + if (nop_addr[0] == nop && nop_addr[1] == nop && nop_addr[2] == nop + && nop_addr[3] == nop && nop_addr[4] == nop) { + break; + } + } + // generate code to restore TOC + register unsigned long toc asm("r2"); + hook_patch->patch_orig = (uintptr_t) nop_addr; + hook_patch->patch_data_size = PatchLoadImm((uintptr_t)hook_patch->patch_data, 2, toc); + + /* put the hook patch on the patch list so it will be undone on finalize */ + opal_list_append (&module->patch_list, &hook_patch->super); + + mca_base_patcher_patch_apply_binary (hook_patch); +#endif + + return OPAL_SUCCESS; +} diff --git a/opal/mca/patcher/linux/Makefile.am b/opal/mca/patcher/linux/Makefile.am new file mode 100644 index 00000000000..a0facb5ce76 --- /dev/null +++ b/opal/mca/patcher/linux/Makefile.am @@ -0,0 +1,47 @@ +# +# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana +# University Research and Technology +# Corporation. All rights reserved. +# Copyright (c) 2004-2005 The University of Tennessee and The University +# of Tennessee Research Foundation. All rights +# reserved. +# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, +# University of Stuttgart. All rights reserved. +# Copyright (c) 2004-2005 The Regents of the University of California. +# All rights reserved. +# Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. +# Copyright (c) 2015 Research Organization for Information Science +# and Technology (RIST). All rights reserved. +# Copyright (c) 2016 Los Alamos National Security, LLC. All rights +# reserved. +# $COPYRIGHT$ +# +# Additional copyrights may follow +# +# $HEADER$ +# + +if MCA_BUILD_opal_patcher_linux_DSO +component_noinst = +component_install = mca_patcher_linux.la +else +component_noinst = libmca_patcher_linux.la +component_install = +endif + +linux_SOURCES = \ + patcher_linux.h \ + patcher_linux_module.c \ + patcher_linux_component.c + +mcacomponentdir = $(opallibdir) +mcacomponent_LTLIBRARIES = $(component_install) +mca_patcher_linux_la_SOURCES = $(linux_SOURCES) +nodist_mca_patcher_linux_la_SOURCES = $(linux_nodist_SOURCES) +mca_patcher_linux_la_LDFLAGS = -module -avoid-version + +noinst_LTLIBRARIES = $(component_noinst) +libmca_patcher_linux_la_SOURCES = $(linux_SOURCES) +nodist_libmca_patcher_linux_la_SOURCES = $(linux_nodist_SOURCES) +libmca_patcher_linux_la_LIBADD = $(patcher_linux_LIBS) +libmca_patcher_linux_la_LDFLAGS = -module -avoid-version diff --git a/opal/mca/patcher/linux/configure.m4 b/opal/mca/patcher/linux/configure.m4 new file mode 100644 index 00000000000..cd3c3d69f0f --- /dev/null +++ b/opal/mca/patcher/linux/configure.m4 @@ -0,0 +1,43 @@ +# -*- shell-script -*- +# +# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana +# University Research and Technology +# Corporation. All rights reserved. +# Copyright (c) 2004-2005 The University of Tennessee and The University +# of Tennessee Research Foundation. All rights +# reserved. +# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, +# University of Stuttgart. All rights reserved. +# Copyright (c) 2004-2005 The Regents of the University of California. +# All rights reserved. +# Copyright (c) 2008-2010 Cisco Systems, Inc. All rights reserved. +# Copyright (c) 2015 Research Organization for Information Science +# and Technology (RIST). All rights reserved. +# Copyright (c) 2016 Los Alamos National Security, LLC. All rights +# reserved. +# $COPYRIGHT$ +# +# Additional copyrights may follow +# +# $HEADER$ +# + +# MCA_patcher_linux_CONFIG(action-if-can-compile, +# [action-if-cant-compile]) +# ------------------------------------------------ +AC_DEFUN([MCA_opal_patcher_linux_CONFIG],[ + AC_CONFIG_FILES([opal/mca/patcher/linux/Makefile]) + + OPAL_VAR_SCOPE_PUSH([opal_patcher_linux_CPPFLAGS_save]) + + opal_patcher_linux_happy=no + if test $OPAL_ENABLE_DLOPEN_SUPPORT = 1; then + OPAL_CHECK_PACKAGE([patcher_linux], [dlfcn.h], [dl], [dl_iterate_phdr], [], [], [], + [opal_patcher_linux_happy=yes],[]) + AC_CHECK_HEADERS([elf.h],[],[opal_patcher_linux_happy=no]) + AC_CHECK_HEADERS([sys/auxv.h]) + fi + + AS_IF([test $opal_patcher_linux_happy = yes], [$1], [$2]) + OPAL_VAR_SCOPE_POP +]) diff --git a/opal/mca/patcher/linux/patcher_linux.h b/opal/mca/patcher/linux/patcher_linux.h new file mode 100644 index 00000000000..de4e85b35da --- /dev/null +++ b/opal/mca/patcher/linux/patcher_linux.h @@ -0,0 +1,45 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#if !defined(OPAL_PATCHER_LINUX_H) +#define OPAL_PATCHER_LINUX_H + +#include "opal_config.h" + +#include "opal/mca/patcher/base/base.h" +#include "opal/mca/patcher/patcher.h" + +#include "opal/class/opal_list.h" +#include "opal/threads/mutex.h" + +struct mca_patcher_linux_patch_got_t { + opal_list_item_t super; + void **got_entry; + void *got_orig; +}; + +typedef struct mca_patcher_linux_patch_got_t mca_patcher_linux_patch_got_t; + +OBJ_CLASS_DECLARATION(mca_patcher_linux_patch_got_t); + +struct mca_patcher_linux_patch_t { + mca_patcher_base_patch_t super; + opal_list_t patch_got_list; +}; + +typedef struct mca_patcher_linux_patch_t mca_patcher_linux_patch_t; + +OBJ_CLASS_DECLARATION(mca_patcher_linux_patch_t); + +extern mca_patcher_base_module_t mca_patcher_linux_module; +extern mca_patcher_base_component_t mca_patcher_linux_component; + +#endif /* !defined(OPAL_PATCHER_LINUX_H) */ diff --git a/opal/mca/patcher/linux/patcher_linux_component.c b/opal/mca/patcher/linux/patcher_linux_component.c new file mode 100644 index 00000000000..445ee23d6a2 --- /dev/null +++ b/opal/mca/patcher/linux/patcher_linux_component.c @@ -0,0 +1,43 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#include "patcher_linux.h" + +static int mca_patcher_linux_priority; + +static int mca_patcher_linux_register (void) +{ + mca_patcher_linux_priority = 13; + mca_base_component_var_register (&mca_patcher_linux_component.patcherc_version, + "priority", "Priority of the linux binary patcher component", + MCA_BASE_VAR_TYPE_INT, NULL, 0, 0, OPAL_INFO_LVL_5, + MCA_BASE_VAR_SCOPE_CONSTANT, &mca_patcher_linux_priority); + + return OPAL_SUCCESS; +} + +static int mca_patcher_linux_query (mca_base_module_t **module, int *priority) +{ + *module = &mca_patcher_linux_module.super; + *priority = mca_patcher_linux_priority; + return OPAL_SUCCESS; +} + +mca_patcher_base_component_t mca_patcher_linux_component = { + .patcherc_version = { + OPAL_PATCHER_BASE_VERSION_1_0_0, + .mca_component_name = "linux", + MCA_BASE_MAKE_VERSION(component, OPAL_MAJOR_VERSION, OPAL_MINOR_VERSION, + OPAL_RELEASE_VERSION), + .mca_query_component = mca_patcher_linux_query, + .mca_register_component_params = mca_patcher_linux_register, + }, +}; diff --git a/opal/mca/patcher/linux/patcher_linux_module.c b/opal/mca/patcher/linux/patcher_linux_module.c new file mode 100644 index 00000000000..2d74c4c9f0b --- /dev/null +++ b/opal/mca/patcher/linux/patcher_linux_module.c @@ -0,0 +1,460 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (C) Mellanox Technologies Ltd. 2001-2015. ALL RIGHTS RESERVED. + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ +/* + * Copied from OpenUCX + */ + +#include "patcher_linux.h" + +#include "opal/mca/patcher/base/base.h" + +#include "opal/constants.h" +#include "opal/util/sys_limits.h" +#include "opal/util/output.h" +#include "opal/prefetch.h" + +#if defined(HAVE_SYS_AUXV_H) +#include +#endif + +#include + +#include +#include +#include +#include +#include +#include +#include +#include + +static void *mca_patcher_linux_dlopen(const char *filename, int flag); + +typedef struct mca_patcher_linux_elf_strtab { + char *tab; + ElfW(Xword) size; +} mca_patcher_linux_elf_strtab_t; + +typedef struct mca_patcher_linux_elf_jmpreltab { + ElfW(Rela) *tab; + ElfW(Xword) size; +} mca_patcher_linux_elf_jmprel_t; + +typedef struct mca_patcher_linux_elf_symtab { + ElfW(Sym) *tab; + ElfW(Xword) entsz; +} mca_patcher_linux_elf_symtab_t; + +typedef struct mca_patcher_linux_dl_iter_context { + mca_patcher_linux_patch_t *patch; + bool remove; + int status; +} mca_patcher_linux_dl_iter_context_t; + +OBJ_CLASS_INSTANCE(mca_patcher_linux_patch_got_t, opal_list_item_t, NULL, NULL); + +static void mca_patcher_linux_patch_construct (mca_patcher_linux_patch_t *patch) +{ + OBJ_CONSTRUCT(&patch->patch_got_list, opal_list_t); +} + +static void mca_patcher_linux_patch_destruct (mca_patcher_linux_patch_t *patch) +{ + OPAL_LIST_DESTRUCT(&patch->patch_got_list); +} + +OBJ_CLASS_INSTANCE(mca_patcher_linux_patch_t, mca_patcher_base_patch_t, mca_patcher_linux_patch_construct, + mca_patcher_linux_patch_destruct); + +/* List of patches to be applied to additional libraries */ +static void *(*orig_dlopen) (const char *, int); + +static const ElfW(Phdr) * +mca_patcher_linux_get_phdr_dynamic(const ElfW(Phdr) *phdr, uint16_t phnum, int phent) +{ + for (uint16_t i = 0; i < phnum; ++i) { + if (phdr->p_type == PT_DYNAMIC) { + return phdr; + } + phdr = (ElfW(Phdr)*)((char*)phdr + phent); + } + return NULL; +} + +static const ElfW(Dyn)* +mca_patcher_linux_get_dynentry(ElfW(Addr) base, const ElfW(Phdr) *pdyn, uint32_t type) +{ + for (ElfW(Dyn) *dyn = (ElfW(Dyn)*)(base + pdyn->p_vaddr); dyn->d_tag; ++dyn) { + if (dyn->d_tag == type) { + return dyn; + } + } + return NULL; +} + +static void mca_patcher_linux_get_jmprel(ElfW(Addr) base, const ElfW(Phdr) *pdyn, + mca_patcher_linux_elf_jmprel_t *table) +{ + const ElfW(Dyn) *dyn; + + dyn = mca_patcher_linux_get_dynentry(base, pdyn, DT_JMPREL); + table->tab = (dyn == NULL) ? NULL : (ElfW(Rela)*)dyn->d_un.d_ptr; + dyn = mca_patcher_linux_get_dynentry(base, pdyn, DT_PLTRELSZ); + table->size = (dyn == NULL) ? 0 : dyn->d_un.d_val; +} + +static void mca_patcher_linux_get_symtab(ElfW(Addr) base, const ElfW(Phdr) *pdyn, + mca_patcher_linux_elf_symtab_t *table) +{ + const ElfW(Dyn) *dyn; + + dyn = mca_patcher_linux_get_dynentry(base, pdyn, DT_SYMTAB); + table->tab = (dyn == NULL) ? NULL : (ElfW(Sym)*)dyn->d_un.d_ptr; + dyn = mca_patcher_linux_get_dynentry(base, pdyn, DT_SYMENT); + table->entsz = (dyn == NULL) ? 0 : dyn->d_un.d_val; +} + +static void mca_patcher_linux_get_strtab(ElfW(Addr) base, const ElfW(Phdr) *pdyn, + mca_patcher_linux_elf_strtab_t *table) +{ + const ElfW(Dyn) *dyn; + + dyn = mca_patcher_linux_get_dynentry(base, pdyn, DT_STRTAB); + table->tab = (dyn == NULL) ? NULL : (char *)dyn->d_un.d_ptr; + dyn = mca_patcher_linux_get_dynentry(base, pdyn, DT_STRSZ); + table->size = (dyn == NULL) ? 0 : dyn->d_un.d_val; +} + +static void * mca_patcher_linux_get_got_entry (ElfW(Addr) base, const ElfW(Phdr) *phdr, int16_t phnum, + int phent, const char *symbol) +{ + mca_patcher_linux_elf_jmprel_t jmprel; + mca_patcher_linux_elf_symtab_t symtab; + mca_patcher_linux_elf_strtab_t strtab; + ElfW(Rela) *rela, *relaend; + const ElfW(Phdr) *dphdr; + const char *relsymname; + uint32_t relsymidx; + + dphdr = mca_patcher_linux_get_phdr_dynamic (phdr, phnum, phent); + + mca_patcher_linux_get_jmprel (base, dphdr, &jmprel); + mca_patcher_linux_get_symtab (base, dphdr, &symtab); + mca_patcher_linux_get_strtab (base, dphdr, &strtab); + + relaend = (ElfW(Rela) *)((char *)jmprel.tab + jmprel.size); + for (rela = jmprel.tab; rela < relaend; ++rela) { +#if SIZEOF_VOID_P == 8 + relsymidx = ELF64_R_SYM(rela->r_info); +#else + relsymidx = ELF32_R_SYM(rela->r_info); +#endif + relsymname = strtab.tab + symtab.tab[relsymidx].st_name; + if (!strcmp(symbol, relsymname)) { + return (void *)(base + rela->r_offset); + } + } + return NULL; +} + +static int mca_patcher_linux_get_aux_phent (void) +{ +#if !defined(HAVE_SYS_AUXV_H) +#define MCA_PATCHER_LINUX_AUXV_BUF_LEN 16 + static const char *proc_auxv_filename = "/proc/self/auxv"; + static int phent = 0; +#if SIZEOF_VOID_P == 8 + Elf64_auxv_t buffer[MCA_PATCHER_LINUX_AUXV_BUF_LEN]; +#else + Elf32_auxv_t buffer[MCA_PATCHER_LINUX_AUXV_BUF_LEN]; +#endif + unsigned count; + ssize_t nread; + int fd; + + /* Can avoid lock here - worst case we'll read the file more than once */ + if (phent == 0) { + fd = open(proc_auxv_filename, O_RDONLY); + if (fd < 0) { + opal_output_verbose (MCA_BASE_VERBOSE_ERROR, opal_patcher_base_framework.framework_output, + "failed to open '%s' for reading: %s", proc_auxv_filename, + strerror (errno)); + return OPAL_ERROR; + } + + /* Use small buffer on the stack, avoid using malloc() */ + do { + nread = read(fd, buffer, sizeof(buffer)); + if (nread < 0) { + opal_output_verbose (MCA_BASE_VERBOSE_ERROR, opal_patcher_base_framework.framework_output, + "failed to read %lu bytes from %s (ret=%ld): %s", sizeof (buffer), + proc_auxv_filename, nread, strerror (errno)); + break; + } + + count = nread / sizeof(buffer[0]); + for (unsigned i = 0 ; i < count && AT_NULL != buffer[i].a_type ; ++i) { + if (AT_PHENT == buffer[i].a_type) { + phent = buffer[i].a_un.a_val; + opal_output_verbose (MCA_BASE_VERBOSE_ERROR, opal_patcher_base_framework.framework_output, + "read phent from %s: %d", proc_auxv_filename, phent); + break; + } + } + } while ((count > 0) && (phent == 0)); + + close(fd); + } + + return phent; +#else + return getauxval (AT_PHENT); +#endif +} + +static int +mca_patcher_linux_modify_got (ElfW(Addr) base, const ElfW(Phdr) *phdr, const char *phname, + int16_t phnum, int phent, mca_patcher_linux_dl_iter_context_t *ctx) +{ + long page_size = opal_getpagesize (); + void **entry; + void *page; + int ret; + + entry = mca_patcher_linux_get_got_entry (base, phdr, phnum, phent, ctx->patch->super.patch_symbol); + if (entry == NULL) { + return OPAL_SUCCESS; + } + + page = (void *)((intptr_t)entry & ~(page_size - 1)); + ret = mprotect(page, page_size, PROT_READ|PROT_WRITE); + if (ret < 0) { + opal_output_verbose (MCA_BASE_VERBOSE_ERROR, opal_patcher_base_framework.framework_output, + "failed to modify GOT page %p to rw: %s", page, strerror (errno)); + return OPAL_ERR_NOT_SUPPORTED; + } + + if (!ctx->remove) { + if (*entry != (void *) ctx->patch->super.patch_value) { + mca_patcher_linux_patch_got_t *patch_got = OBJ_NEW(mca_patcher_linux_patch_got_t); + if (NULL == patch_got) { + return OPAL_ERR_OUT_OF_RESOURCE; + } + + opal_output_verbose (MCA_BASE_VERBOSE_TRACE, opal_patcher_base_framework.framework_output, + "patch %p (%s): modifying got entry %p. original value %p. new value %p\n", ctx->patch, + ctx->patch->super.patch_symbol, (void *) entry, *entry, (void *) ctx->patch->super.patch_value); + + patch_got->got_entry = entry; + patch_got->got_orig = *entry; + + opal_list_append (&ctx->patch->patch_got_list, &patch_got->super); + + *entry = (void *) ctx->patch->super.patch_value; + } + } else { + if (*entry == (void *) ctx->patch->super.patch_value) { + /* find the appropriate entry and restore the original value */ + mca_patcher_linux_patch_got_t *patch_got; + OPAL_LIST_FOREACH_REV(patch_got, &ctx->patch->patch_got_list, mca_patcher_linux_patch_got_t) { + if (patch_got->got_entry == entry) { + opal_output_verbose (MCA_BASE_VERBOSE_TRACE, opal_patcher_base_framework.framework_output, + "restoring got entry %p with original value %p\n", (void *) entry, patch_got->got_orig); + *entry = patch_got->got_orig; + opal_list_remove_item (&ctx->patch->patch_got_list, &patch_got->super); + OBJ_RELEASE(patch_got); + break; + } + } + } + } + + return OPAL_SUCCESS; +} + +static int mca_patcher_linux_phdr_iterator(struct dl_phdr_info *info, size_t size, void *data) +{ + mca_patcher_linux_dl_iter_context_t *ctx = data; + int phent; + + phent = mca_patcher_linux_get_aux_phent(); + if (phent <= 0) { + opal_output_verbose (MCA_BASE_VERBOSE_ERROR, opal_patcher_base_framework.framework_output, + "failed to read phent size"); + ctx->status = OPAL_ERR_NOT_SUPPORTED; + return -1; + } + + ctx->status = mca_patcher_linux_modify_got (info->dlpi_addr, info->dlpi_phdr, + info->dlpi_name, info->dlpi_phnum, + phent, ctx); + if (ctx->status == OPAL_SUCCESS) { + return 0; /* continue iteration and patch all objects */ + } else { + return -1; /* stop iteration if got a real error */ + } +} + +/* called with lock held */ +static int mca_patcher_linux_apply_patch (mca_patcher_linux_patch_t *patch) +{ + mca_patcher_linux_dl_iter_context_t ctx = { + .patch = patch, + .remove = false, + .status = OPAL_SUCCESS, + }; + + /* Avoid locks here because we don't modify ELF data structures. + * Worst case the same symbol will be written more than once. + */ + (void)dl_iterate_phdr(mca_patcher_linux_phdr_iterator, &ctx); + if (ctx.status == OPAL_SUCCESS) { + opal_output_verbose (MCA_BASE_VERBOSE_INFO, opal_patcher_base_framework.framework_output, + "modified '%s' to 0x%lx", ctx.patch->super.patch_symbol, ctx.patch->super.patch_value); + } + + return ctx.status; +} + +static int mca_patcher_linux_remove_patch (mca_patcher_linux_patch_t *patch) +{ + mca_patcher_linux_dl_iter_context_t ctx = { + .patch = patch, + .remove = true, + .status = OPAL_SUCCESS, + }; + + /* Avoid locks here because we don't modify ELF data structures. + * Worst case the same symbol will be written more than once. + */ + (void)dl_iterate_phdr(mca_patcher_linux_phdr_iterator, &ctx); + if (ctx.status == OPAL_SUCCESS) { + opal_output_verbose (MCA_BASE_VERBOSE_INFO, opal_patcher_base_framework.framework_output, + "modified '%s' to 0x%lx", ctx.patch->super.patch_symbol, ctx.patch->super.patch_value); + } + + return ctx.status; +} + +static void *mca_patcher_linux_dlopen(const char *filename, int flag) +{ + OPAL_PATCHER_BEGIN; + mca_patcher_linux_patch_t *patch; + void *handle; + + assert (orig_dlopen); + handle = orig_dlopen (filename, flag); + if (handle != NULL) { + /* + * Every time a new object is loaded, we must update its relocations + * with our list of patches (including dlopen itself). This code is less + * efficient and will modify all existing objects every time, but good + * enough. + */ + opal_mutex_lock (&mca_patcher_linux_module.patch_list_mutex); + OPAL_LIST_FOREACH(patch, &mca_patcher_linux_module.patch_list, mca_patcher_linux_patch_t) { + opal_output_verbose (MCA_BASE_VERBOSE_INFO, opal_patcher_base_framework.framework_output, + "in dlopen(), re-applying '%s' to %p", patch->super.patch_symbol, (void *) patch->super.patch_value); + /* ignore hook binary patches */ + if (!patch->super.patch_data_size) { + mca_patcher_linux_apply_patch (patch); + } + } + opal_mutex_unlock (&mca_patcher_linux_module.patch_list_mutex); + } + + OPAL_PATCHER_END; + return handle; +} + +static intptr_t mca_patcher_linux_get_orig (const char *symbol, void *replacement) +{ + const char *error; + void *func_ptr; + + func_ptr = dlsym(RTLD_DEFAULT, symbol); + if (func_ptr == replacement) { + (void)dlerror(); + func_ptr = dlsym(RTLD_NEXT, symbol); + if (func_ptr == NULL) { + error = dlerror(); + opal_output_verbose (MCA_BASE_VERBOSE_ERROR, opal_patcher_base_framework.framework_output, + "could not find address of original %s(): %s", symbol, error ? error : "Unknown error"); + } + } + + opal_output_verbose (MCA_BASE_VERBOSE_INFO, opal_patcher_base_framework.framework_output, + "original %s() is at %p", symbol, func_ptr); + + return (intptr_t) func_ptr; +} + +static int mca_patcher_linux_patch_symbol (const char *symbol_name, uintptr_t replacement, uintptr_t *orig) +{ + mca_patcher_linux_patch_t *patch = OBJ_NEW(mca_patcher_linux_patch_t); + int rc; + + if (OPAL_UNLIKELY(NULL == patch)) { + return OPAL_ERR_OUT_OF_RESOURCE; + } + + patch->super.patch_symbol = strdup (symbol_name); + if (NULL == patch->super.patch_symbol) { + OBJ_RELEASE(patch); + return OPAL_ERR_OUT_OF_RESOURCE; + } + + patch->super.patch_value = mca_patcher_base_addr_text (replacement); + patch->super.patch_restore = (mca_patcher_base_restore_fn_t) mca_patcher_linux_remove_patch; + + /* Take lock first to handle a possible race where dlopen() is called + * from another thread and we may end up not patching it. + */ + opal_mutex_lock (&mca_patcher_linux_module.patch_list_mutex); + do { + rc = mca_patcher_base_patch_hook (&mca_patcher_linux_module, patch->super.patch_value); + if (OPAL_SUCCESS != rc) { + OBJ_RELEASE(patch); + break; + } + + rc = mca_patcher_linux_apply_patch (patch); + if (OPAL_SUCCESS != rc) { + OBJ_RELEASE(patch); + break; + } + + *orig = mca_patcher_linux_get_orig (patch->super.patch_symbol, (void *) replacement); + + opal_list_append (&mca_patcher_linux_module.patch_list, &patch->super.super); + } while (0); + opal_mutex_unlock (&mca_patcher_linux_module.patch_list_mutex); + + return rc; +} + +/* called with lock held */ +static int mca_patcher_linux_install_dlopen (void) +{ + return mca_patcher_linux_patch_symbol ("dlopen", (uintptr_t) mca_patcher_linux_dlopen, + (uintptr_t *) &orig_dlopen); +} + +static int mca_patcher_linux_init (void) +{ + return mca_patcher_linux_install_dlopen (); +} + +mca_patcher_base_module_t mca_patcher_linux_module = { + .patch_init = mca_patcher_linux_init, + .patch_symbol = mca_patcher_linux_patch_symbol, +}; diff --git a/opal/mca/patcher/overwrite/Makefile.am b/opal/mca/patcher/overwrite/Makefile.am new file mode 100644 index 00000000000..e9e4a317181 --- /dev/null +++ b/opal/mca/patcher/overwrite/Makefile.am @@ -0,0 +1,47 @@ +# +# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana +# University Research and Technology +# Corporation. All rights reserved. +# Copyright (c) 2004-2005 The University of Tennessee and The University +# of Tennessee Research Foundation. All rights +# reserved. +# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, +# University of Stuttgart. All rights reserved. +# Copyright (c) 2004-2005 The Regents of the University of California. +# All rights reserved. +# Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. +# Copyright (c) 2015 Research Organization for Information Science +# and Technology (RIST). All rights reserved. +# Copyright (c) 2016 Los Alamos National Security, LLC. All rights +# reserved. +# $COPYRIGHT$ +# +# Additional copyrights may follow +# +# $HEADER$ +# + +if MCA_BUILD_opal_patcher_overwrite_DSO +component_noinst = +component_install = mca_patcher_overwrite.la +else +component_noinst = libmca_patcher_overwrite.la +component_install = +endif + +overwrite_SOURCES = \ + patcher_overwrite.h \ + patcher_overwrite_module.c \ + patcher_overwrite_component.c + +mcacomponentdir = $(opallibdir) +mcacomponent_LTLIBRARIES = $(component_install) +mca_patcher_overwrite_la_SOURCES = $(overwrite_SOURCES) +nodist_mca_patcher_overwrite_la_SOURCES = $(overwrite_nodist_SOURCES) +mca_patcher_overwrite_la_LDFLAGS = -module -avoid-version + +noinst_LTLIBRARIES = $(component_noinst) +libmca_patcher_overwrite_la_SOURCES = $(overwrite_SOURCES) +nodist_libmca_patcher_overwrite_la_SOURCES = $(overwrite_nodist_SOURCES) +libmca_patcher_overwrite_la_LIBADD = $(patcher_overwrite_LIBS) +libmca_patcher_overwrite_la_LDFLAGS = -module -avoid-version diff --git a/opal/mca/patcher/overwrite/configure.m4 b/opal/mca/patcher/overwrite/configure.m4 new file mode 100644 index 00000000000..02394cc9586 --- /dev/null +++ b/opal/mca/patcher/overwrite/configure.m4 @@ -0,0 +1,41 @@ +# -*- shell-script -*- +# +# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana +# University Research and Technology +# Corporation. All rights reserved. +# Copyright (c) 2004-2005 The University of Tennessee and The University +# of Tennessee Research Foundation. All rights +# reserved. +# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, +# University of Stuttgart. All rights reserved. +# Copyright (c) 2004-2005 The Regents of the University of California. +# All rights reserved. +# Copyright (c) 2008-2010 Cisco Systems, Inc. All rights reserved. +# Copyright (c) 2015 Research Organization for Information Science +# and Technology (RIST). All rights reserved. +# Copyright (c) 2016 Los Alamos National Security, LLC. All rights +# reserved. +# $COPYRIGHT$ +# +# Additional copyrights may follow +# +# $HEADER$ +# + +# MCA_patcher_overwrite_CONFIG(action-if-can-compile, +# [action-if-cant-compile]) +# ------------------------------------------------ +AC_DEFUN([MCA_opal_patcher_overwrite_CONFIG],[ + AC_CONFIG_FILES([opal/mca/patcher/overwrite/Makefile]) + + opal_patcher_overwrite_happy=no + if test $OPAL_ENABLE_DLOPEN_SUPPORT = 1; then + AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[ +#if !defined(__i386__) && !defined(__x86_64__) && !defined(__ia64__) && !defined(__PPC__) +#error "platform not supported" +#endif +]],[])],[opal_patcher_overwrite_happy=yes],[]) + fi + + AS_IF([test $opal_patcher_overwrite_happy = yes], [$1], [$2]) +]) diff --git a/opal/mca/patcher/overwrite/patcher_overwrite.h b/opal/mca/patcher/overwrite/patcher_overwrite.h new file mode 100644 index 00000000000..9c2ad58dfd5 --- /dev/null +++ b/opal/mca/patcher/overwrite/patcher_overwrite.h @@ -0,0 +1,32 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ +/** + * @file pather_overwrite.h + * + * This component works by overwritting the first couple instructions in + * the target function with a jump instruction to the hook function. The + * hook function will be expected to implement the functionality of the + * hooked function when using this module. + * + * Note: This component only supports x86, x86_64, ia64, and powerpc/power. + */ + +#if !defined(OPAL_PATCHER_OVERWRITE_H) +#define OPAL_PATCHER_OVERWRITE_H + +#include "opal_config.h" +#include "opal/mca/patcher/patcher.h" +#include "opal/class/opal_list.h" + +extern mca_patcher_base_module_t mca_patcher_overwrite_module; +extern mca_patcher_base_component_t mca_patcher_overwrite_component; + +#endif /* !defined(OPAL_PATCHER_OVERWRITE_H) */ diff --git a/opal/mca/patcher/overwrite/patcher_overwrite_component.c b/opal/mca/patcher/overwrite/patcher_overwrite_component.c new file mode 100644 index 00000000000..5211d4deaef --- /dev/null +++ b/opal/mca/patcher/overwrite/patcher_overwrite_component.c @@ -0,0 +1,45 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#include "patcher_overwrite.h" +#include "opal/mca/mca.h" +#include "opal/mca/base/base.h" + +static int mca_patcher_overwrite_priority; + +static int mca_patcher_overwrite_register (void) +{ + mca_patcher_overwrite_priority = 37; + mca_base_component_var_register (&mca_patcher_overwrite_component.patcherc_version, + "priority", "Priority of the overwrite binary patcher component", + MCA_BASE_VAR_TYPE_INT, NULL, 0, 0, OPAL_INFO_LVL_5, + MCA_BASE_VAR_SCOPE_CONSTANT, &mca_patcher_overwrite_priority); + + return OPAL_SUCCESS; +} + +static int mca_patcher_overwrite_query (mca_base_module_t **module, int *priority) +{ + *module = &mca_patcher_overwrite_module.super; + *priority = mca_patcher_overwrite_priority; + return OPAL_SUCCESS; +} + +mca_patcher_base_component_t mca_patcher_overwrite_component = { + .patcherc_version = { + OPAL_PATCHER_BASE_VERSION_1_0_0, + .mca_component_name = "overwrite", + MCA_BASE_MAKE_VERSION(component, OPAL_MAJOR_VERSION, OPAL_MINOR_VERSION, + OPAL_RELEASE_VERSION), + .mca_query_component = mca_patcher_overwrite_query, + .mca_register_component_params = mca_patcher_overwrite_register, + }, +}; diff --git a/opal/util/opal_patcher.c b/opal/mca/patcher/overwrite/patcher_overwrite_module.c similarity index 50% rename from opal/util/opal_patcher.c rename to opal/mca/patcher/overwrite/patcher_overwrite_module.c index 0463decbf52..79cb6ce5c72 100644 --- a/opal/util/opal_patcher.c +++ b/opal/mca/patcher/overwrite/patcher_overwrite_module.c @@ -10,11 +10,14 @@ * $HEADER$ */ -#include "opal_patcher.h" +#include "patcher_overwrite.h" + +#include "opal/mca/patcher/base/base.h" #include "opal/constants.h" #include "opal/util/sys_limits.h" #include "opal/util/output.h" +#include "opal/prefetch.h" #include #include @@ -25,19 +28,7 @@ #include #include -#if OPAL_ENABLE_DLOPEN_SUPPORT && (defined(__i386__) || defined(__x86_64__) || defined(__ia64__)) - -static void flush_and_invalidate_cache (unsigned long a) -{ -#if defined(__i386__) - /* does not work with AMD processors */ - __asm__ volatile("mfence;clflush %0;mfence" : :"m" (*(char*)a)); -#elif defined(__x86_64__) - __asm__ volatile("mfence;clflush %0;mfence" : :"m" (*(char*)a)); -#elif defined(__ia64__) - __asm__ volatile ("fc %0;; sync.i;; srlz.i;;" : : "r"(a) : "memory"); -#endif -} +#if defined(__i386__) || defined(__x86_64__) || defined(__ia64__) #if defined(__ia64__) @@ -91,24 +82,22 @@ static void make_ia64_bundle (unsigned char *dst, } #endif /* defined(__ia64__) */ -static int patch_code (uintptr_t func_old_addr, unsigned long func_new_addr) +static int mca_patcher_overwrite_apply_patch (mca_patcher_base_patch_t *patch) { - long pg_sz = opal_getpagesize (); - - if (mprotect((void*)(func_old_addr&~(pg_sz-1)), pg_sz, PROT_EXEC|PROT_READ|PROT_WRITE)) { - perror("mprotect failed\n"); - } + uintptr_t func_new_addr = patch->patch_value; { #if defined(__i386__) - *(unsigned char*)(func_old_addr+0) = 0xe9; - *(unsigned int *)(func_old_addr+1) = (unsigned int)(func_new_addr - func_old_addr - 5); + patch->patch_data_size = 5; + *(unsigned char *)(patch->patch_data+0) = 0xe9; + *(unsigned int *) (patch->patch_data+1) = (unsigned int)(func_new_addr - func_old_addr - 5); #elif defined(__x86_64__) - *(unsigned short*)(func_old_addr+ 0) = 0xbb49; - *(unsigned long* )(func_old_addr+ 2) = (unsigned long) func_new_addr; - *(unsigned char*) (func_old_addr+10) = 0x41; - *(unsigned char*) (func_old_addr+11) = 0xff; - *(unsigned char*) (func_old_addr+12) = 0xe3; + patch->patch_data_size = 13; + *(unsigned short*)(patch->patch_data + 0) = 0xbb49; + *(unsigned long* )(patch->patch_data + 2) = (unsigned long) func_new_addr; + *(unsigned char*) (patch->patch_data +10) = 0x41; + *(unsigned char*) (patch->patch_data +11) = 0xff; + *(unsigned char*) (patch->patch_data +12) = 0xe3; #elif defined(__ia64__) { /* @@ -144,109 +133,48 @@ static int patch_code (uintptr_t func_old_addr, unsigned long func_new_addr) (1ULL << 6) | (0x0ULL << 0); + patch->data_size = 32; + make_ia64_bundle(buf, movl, (glb_ptr>>22)&0x1FFFFFFFFFFULL, nop, 5); for (int i = 0 ; i < 16 ; ++i) { - ((unsigned char *)func_old_addr)[16-i-1] = buf[i]; + patch->patch_data[16-i-1] = buf[i]; } make_ia64_bundle(buf, brl, ((imm64>>24)&0x7FFFFFFFFFULL)<<2, nop, 5); for (int i = 0 ; i < 16 ; ++i) { - ((unsigned char *)func_old_addr+16)[16-i-1] = buf[i]; + patch->patch_data[32-i-1] = buf[i]; } } #endif } - flush_and_invalidate_cache(func_old_addr+ 0); - flush_and_invalidate_cache(func_old_addr+16); + mca_base_patcher_patch_apply_binary (patch); -#if 1 - if (mprotect((void*)(func_old_addr&~(pg_sz-1)), pg_sz, PROT_EXEC)) { - perror("mprotect failed\n"); - } -#endif - - return 0; -} - -int opal_patch_symbol (const char *func_symbol_name, uintptr_t func_new_addr) -{ - void *sym_addr; - char *error; - uintptr_t func_old_addr; - - /* NTH: might want to update opal/mca/dl to handle lookups in the default - * handle. */ - sym_addr = dlsym(RTLD_DEFAULT, func_symbol_name); - if ( (sym_addr == NULL) && ((error = dlerror()) != NULL) ) { - opal_output(0, "error locating symbol %s to patch. %s", func_symbol_name, - error); - return -1; - } - func_old_addr = (unsigned long)sym_addr; - -#if defined(__ia64__) - /* On IA64 addresses are all indirect */ - func_new_addr = *(unsigned long *)func_new_addr; - func_old_addr = *(unsigned long *)func_old_addr; -#endif - - patch_code(func_old_addr, func_new_addr); return OPAL_SUCCESS; } -bool opal_patch_supported (void) -{ - return true; -} - /* end of #if defined(__i386__) || defined(__x86_64__) || defined(__ia64__) */ // ------------------------------------------------- PPC equivalent: -#elif OPAL_ENABLE_DLOPEN_SUPPORT && defined(__PPC__) - -static inline uintptr_t addr_text (uintptr_t addr) { -#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) && _CALL_ELF != 2 - struct odp_t { - uintptr_t text; - uintptr_t toc; - } *odp = (struct odp_t *) addr; - return (odp)?odp->text:0; -#else - return addr; -#endif -} - -// modify protection of memory range -static void -ModifyMemoryProtection(uintptr_t addr, size_t length, int prot) -{ - long page_size = opal_getpagesize (); - uintptr_t page_addr = (addr & ~(page_size-1)); - do { - if (mprotect((void *)page_addr, page_size, prot)) - perror("MemHook: mprotect failed"); - page_addr += page_size; - } while (page_addr < addr + length); -} +#elif defined(__PPC__) // PowerPC instructions used in patching // Reference: "PowerPC User Instruction Set Architecture" -unsigned int addis(unsigned int RT, unsigned int RS, unsigned int UI) { +static unsigned int addis(unsigned int RT, unsigned int RS, unsigned int UI) { return (15<<26) + (RT<<21) + (RS<<16) + (UI&0xffff); } -unsigned int ori(unsigned int RT, unsigned int RS, unsigned int UI) { +static unsigned int ori(unsigned int RT, unsigned int RS, unsigned int UI) { return (24<<26) + (RS<<21) + (RT<<16) + (UI&0xffff); } -unsigned int oris(unsigned int RT, unsigned int RS, unsigned int UI) { +static unsigned int oris(unsigned int RT, unsigned int RS, unsigned int UI) { return (25<<26) + (RS<<21) + (RT<<16) + (UI&0xffff); } -unsigned int mtspr(unsigned int SPR, unsigned int RS) { +static unsigned int mtspr(unsigned int SPR, unsigned int RS) { return (31<<26) + (RS<<21) + ((SPR&0x1f)<<16) + ((SPR>>5)<<11) + (467<<1); } -unsigned int bcctr(unsigned int BO, unsigned int BI, unsigned int BH) { +static unsigned int bcctr(unsigned int BO, unsigned int BI, unsigned int BH) { return (19<<26) + (BO<<21) + (BI<<16) + (BH<<11) + (528<<1); } -unsigned int rldicr(unsigned int RT, unsigned int RS, unsigned int SH, unsigned int MB) +static unsigned int rldicr(unsigned int RT, unsigned int RS, unsigned int SH, unsigned int MB) { return (30<<26) + (RS<<21) + (RT<<16) + ((SH&0x1f)<<11) + ((SH>>5)<<1) + ((MB&0x1f)<<6) + ((MB>>5)<<5) + (1<<2); @@ -270,93 +198,110 @@ PatchLoadImm(uintptr_t addr, unsigned int reg, size_t value) } -static void -apply_patch(char *patch, void *addr, int length) -{ - if (length == 0) { return; } - ModifyMemoryProtection((uintptr_t)addr, length, - PROT_EXEC|PROT_READ|PROT_WRITE); - memcpy((void *)addr, patch, length); - ModifyMemoryProtection((uintptr_t)addr, length, PROT_EXEC|PROT_READ); -} - -#define MAX_PATCH_SIZE 1024 -int opal_patch_symbol (const char *sys_func, uintptr_t hook_addr) +static int mca_patcher_overwrite_apply_patch (mca_patcher_base_patch_t *patch) { - void *addr_to_patch_hook; - int len_to_patch_hook = 0; - char patch_hook[MAX_PATCH_SIZE]; - - void *addr_to_patch_sys; - int len_to_patch_sys = 0; - char patch_sys[MAX_PATCH_SIZE]; - - int offset; -#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) - unsigned int *nop_addr; - const unsigned int nop = 0x60000000; -#endif + uintptr_t sys_addr, hook_addr; + int offset, rc; // get system function address - uintptr_t sys_addr = add_text(dlsym(RTLD_NEXT, sys_func)); - if(sys_addr == 0) sys_addr = add_text(dlsym(RTLD_DEFAULT, sys_func)); - hook_addr = add_text(hook_addr); + sys_addr = mca_patcher_base_addr_text(patch->patch_orig); + hook_addr = mca_patcher_base_addr_text(patch->patch_value); // Patch for hook function: #if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) - // locate reserved code space in hook function - nop_addr = (unsigned int *)hook_addr; - for (; ; nop_addr++) - if (nop_addr[0] == nop && nop_addr[1] == nop && nop_addr[2] == nop - && nop_addr[3] == nop && nop_addr[4] == nop) - break; - // generate code to restore TOC - register unsigned long toc asm("r2"); - addr_to_patch_hook = nop_addr; - len_to_patch_hook = PatchLoadImm((uintptr_t)patch_hook, 2, toc); - - // save the original code - assert(len_to_patch_hook <= MAX_PATCH_SIZE); - //memcpy(save, (void *)addr_to_patch_hook, len_to_patch_hook); // meh -#endif + rc = mca_patcher_base_patch_hook (&mca_patcher_overwrite_module, hook_addr); + if (OPAL_SUCCESS != rc) { + return rc; + } -// Patch for system function: -#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) && _CALL_ELF == 2 +#if _CALL_ELF == 2 sys_addr += 8; hook_addr += 8; #endif /* _CALL_ELF == 2*/ +#endif - addr_to_patch_sys = (void*) sys_addr; + // Patch for system function: // generate patch code // r11 is a volatile register according to PowerPC EABI const unsigned int gr = 11; - offset = PatchLoadImm((uintptr_t)patch_sys, gr, hook_addr); - *(unsigned int *) (patch_sys + offset + 0) = mtspr (9, gr); // 9 = CTR - *(unsigned int *) (patch_sys + offset + 4) = bcctr (20, 0, 0);// 20 = always - len_to_patch_sys = offset + 8; + offset = PatchLoadImm ((uintptr_t) patch->patch_data, gr, hook_addr); + *(unsigned int *) (patch->patch_data + offset + 0) = mtspr (9, gr); // 9 = CTR + *(unsigned int *) (patch->patch_data + offset + 4) = bcctr (20, 0, 0);// 20 = always + patch->patch_data_size = offset + 8; + patch->patch_orig = sys_addr; - assert(len_to_patch_sys <= MAX_PATCH_SIZE); - //memcpy(save, (void *)addr_to_patch_sys, len_to_patch_sys); + mca_base_patcher_patch_apply_binary (patch); - apply_patch(patch_hook, addr_to_patch_hook, len_to_patch_hook); - apply_patch(patch_sys, addr_to_patch_sys, len_to_patch_sys); return OPAL_SUCCESS; } -bool opal_patch_supported (void) -{ - return true; -} -#else +#endif -int opal_patch_symbol (const char *sys_func, uintptr_t hook_addr) +static int mca_patcher_overwrite_patch_address (uintptr_t sys_addr, unsigned long hook_addr) { - return OPAL_ERR_NOT_SUPPORTED; + mca_patcher_base_patch_t *patch; + int rc; + + patch = OBJ_NEW(mca_patcher_base_patch_t); + if (OPAL_UNLIKELY(NULL == patch)) { + return OPAL_ERR_OUT_OF_RESOURCE; + } + + patch->patch_orig = sys_addr; + patch->patch_value = hook_addr; + + opal_mutex_lock (&mca_patcher_overwrite_module.patch_list_mutex); + do { + rc = mca_patcher_overwrite_apply_patch (patch); + if (OPAL_SUCCESS != rc) { + break; + } + + opal_list_append (&mca_patcher_overwrite_module.patch_list, &patch->super); + } while (0); + + opal_mutex_unlock (&mca_patcher_overwrite_module.patch_list_mutex); + + return OPAL_SUCCESS; } -bool opal_patch_supported (void) +static int mca_patcher_overwrite_patch_symbol (const char *func_symbol_name, uintptr_t func_new_addr, + uintptr_t *func_old_addr) { - return false; -} + void *sym_addr; + char *error; + uintptr_t old_addr; + + /* NTH: might want to update opal/mca/dl to handle lookups in the default + * handle. */ + sym_addr = dlsym (RTLD_NEXT, func_symbol_name); + if (NULL == sym_addr) { + sym_addr = dlsym(RTLD_DEFAULT, func_symbol_name); + if ( (sym_addr == NULL) && ((error = dlerror()) != NULL) ) { + opal_output(0, "error locating symbol %s to patch. %s", func_symbol_name, + error); + return OPAL_ERR_NOT_FOUND; + } + } + + old_addr = (unsigned long)sym_addr; +#if defined(__ia64__) + /* On IA64 addresses are all indirect */ + func_new_addr = *(unsigned long *)func_new_addr; + old_addr = *(unsigned long *) old_addr; #endif + + if (func_old_addr) { + /* we will be overwritting part of the original function. do not return + * its address */ + *func_old_addr = 0; + } + + return mca_patcher_overwrite_patch_address (old_addr, func_new_addr); +} + +mca_patcher_base_module_t mca_patcher_overwrite_module = { + .patch_symbol = mca_patcher_overwrite_patch_symbol, + .patch_address = mca_patcher_overwrite_patch_address, +}; diff --git a/opal/mca/patcher/patcher.h b/opal/mca/patcher/patcher.h new file mode 100644 index 00000000000..25af9b1376a --- /dev/null +++ b/opal/mca/patcher/patcher.h @@ -0,0 +1,121 @@ +/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ +/* + * Copyright (c) 2016 Los Alamos National Security, LLC. All rights + * reserved. + * $COPYRIGHT$ + * + * Additional copyrights may follow + * + * $HEADER$ + */ + +#ifndef OPAL_MCA_PATCHER_PATCHER_H +#define OPAL_MCA_PATCHER_PATCHER_H + +#include "opal_config.h" + +#include "opal/mca/mca.h" +#include "opal/mca/base/base.h" +#include "opal/class/opal_list.h" + +/* Any function being patched in as a hook must use SYMBOLPATCH_BEGIN at the top, + * and SYMBOLPATCH_END before it returns (this is just for PPC). */ + +#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) && defined(OPAL_GCC_INLINE_ASSEMBLY) + +/* special processing for ppc64 to save and restore TOC (r2) + * Reference: "64-bit PowerPC ELF Application Binary Interface Supplement 1.9" */ +#define OPAL_PATCHER_BEGIN \ + unsigned long toc_save; \ + asm volatile ("std 2, %0" : "=m" (toc_save)); \ + asm volatile ("nop; nop; nop; nop; nop"); +#define OPAL_PATCHER_END \ + asm volatile ("ld 2, %0" : : "m" (toc_save)); + +#else /* !__PPC64__ */ + +#define OPAL_PATCHER_BEGIN +#define OPAL_PATCHER_END + +#endif + +/** + * Make any calls to the named function redirect to a new function + * + * @param[in] func_symbol_name function to hook + * @param[in] func_new_addr function pointer of hook + * @param[out] func_old_addr address of func_symbol_name + * + * This function redirects all calls to the function func_symbol_name to + * the function pointer func_new_addr. If it is possible for the hook + * function to call the original function the patcher module will return + * the old function's address in func_old_addr. + */ +typedef int (*mca_patcher_base_patch_symbol_fn_t)(const char *func_symbol_name, uintptr_t func_new_addr, + uintptr_t *func_old_addr); + +/** + * Make any calls to a function redirect to a new function + * + * @param[in] func_symbol_name function to hook + * @param[in] func_new_addr function pointer of hook + * @param[out] func_old_addr address of func_symbol_name + * + * This function redirects all calls to the function at func_addr to + * the function pointer func_new_addr. + */ +typedef int (*mca_patcher_base_patch_address_fn_t)(uintptr_t func_addr, uintptr_t func_new_addr); + +/** + * Set up the patcher module + */ +typedef int (*mca_patcher_base_init_fn_t) (void); + +/** + * Finalize the patcher module + */ +typedef int (*mca_patcher_base_fini_fn_t) (void); + +/** + * Structure for patcher modules. + */ +typedef struct mca_patcher_base_module_t { + mca_base_module_t super; + /** list of patches */ + opal_list_t patch_list; + /** lock for patch list */ + opal_mutex_t patch_list_mutex; + /** function to call if the patcher module is used. can + * be NULL. */ + mca_patcher_base_init_fn_t patch_init; + /** function to call when patcher is unloaded. this function + * MUST clean up all active patches. can be NULL. */ + mca_patcher_base_fini_fn_t patch_fini; + /** hook a symbol. may be NULL */ + mca_patcher_base_patch_symbol_fn_t patch_symbol; + /** hook a function pointer. may be NULL */ + mca_patcher_base_patch_address_fn_t patch_address; +} mca_patcher_base_module_t; + + +OPAL_DECLSPEC extern mca_patcher_base_module_t *opal_patcher; + +/** + * Structure for patcher components. + */ +typedef struct mca_patcher_base_component_1_0_0_t { + /** MCA base component */ + mca_base_component_t patcherc_version; + /** MCA base data */ + mca_base_component_data_t patcherc_data; +} mca_patcher_base_component_1_0_0_t; + +typedef mca_patcher_base_component_1_0_0_t mca_patcher_base_component_t; + +/* + * Macro for use in components that are of type patcher + */ +#define OPAL_PATCHER_BASE_VERSION_1_0_0 \ + OPAL_MCA_BASE_VERSION_2_1_0("patcher", 1, 0, 0) + +#endif /* OPAL_MCA_PATCHER_PATCHER_H */ diff --git a/opal/runtime/opal_finalize.c b/opal/runtime/opal_finalize.c index 382deb3c324..7138d1680d7 100644 --- a/opal/runtime/opal_finalize.c +++ b/opal/runtime/opal_finalize.c @@ -41,6 +41,7 @@ #include "opal/mca/memchecker/base/base.h" #include "opal/mca/memcpy/base/base.h" #include "opal/mca/memory/base/base.h" +#include "opal/mca/patcher/base/base.h" #include "opal/mca/backtrace/base/base.h" #include "opal/mca/sec/base/base.h" #include "opal/mca/timer/base/base.h" @@ -160,6 +161,7 @@ opal_finalize(void) hooks to the bowels of the mem_free code can still occur any time between now and end of application (even post main()!) */ (void) mca_base_framework_close(&opal_memory_base_framework); + (void) mca_base_framework_close(&opal_patcher_base_framework); /* close the memcpy framework */ (void) mca_base_framework_close(&opal_memcpy_base_framework); diff --git a/opal/runtime/opal_init.c b/opal/runtime/opal_init.c index 2eade6204b6..0a19a70a8e1 100644 --- a/opal/runtime/opal_init.c +++ b/opal/runtime/opal_init.c @@ -45,6 +45,7 @@ #include "opal/datatype/opal_datatype.h" #include "opal/mca/installdirs/base/base.h" #include "opal/mca/memory/base/base.h" +#include "opal/mca/patcher/base/base.h" #include "opal/mca/memcpy/base/base.h" #include "opal/mca/hwloc/base/base.h" #include "opal/mca/sec/base/base.h" @@ -430,6 +431,14 @@ opal_init(int* pargc, char*** pargv) goto return_error; } + if (OPAL_SUCCESS != (ret = mca_base_framework_open(&opal_patcher_base_framework, 0))) { + error = "opal_patcher_base_open"; + goto return_error; + } + + /* select a patcher module. if a patcher module can not be found it is not an error. */ + (void) opal_patcher_base_select (); + /* open the memory manager components. Memory hooks may be triggered before this (any time after mem_free_init(), actually). This is a hook available for memory manager hooks diff --git a/opal/util/Makefile.am b/opal/util/Makefile.am index 10c91ea5844..5c4cb2945ed 100644 --- a/opal/util/Makefile.am +++ b/opal/util/Makefile.am @@ -53,7 +53,6 @@ headers = \ numtostr.h \ opal_environ.h \ opal_getcwd.h \ - opal_patcher.h \ opal_pty.h \ os_dirpath.h \ os_path.h \ @@ -89,7 +88,6 @@ libopalutil_la_SOURCES = \ numtostr.c \ opal_environ.c \ opal_getcwd.c \ - opal_patcher.c \ opal_pty.c \ os_dirpath.c \ os_path.c \ diff --git a/opal/util/opal_patcher.h b/opal/util/opal_patcher.h deleted file mode 100644 index 571cbead7e5..00000000000 --- a/opal/util/opal_patcher.h +++ /dev/null @@ -1,61 +0,0 @@ -/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ -/* - * Copyright (c) 2016 Los Alamos National Security, LLC. All rights - * reserved. - * ******** ADD IBM COPYRIGHT HERE ****** - * $COPYRIGHT$ - * - * Additional copyrights may follow - * - * $HEADER$ - */ - -#if !defined(OPAL_PATCHER_H) -#define OPAL_PATCHER_H - -#include "opal_config.h" - -// Any function being patched in must use SYMBOLPATCH_BEGIN at the top, -// and SYMBOLPATCH_END before it returns (this is just for PPC). - -#if (defined(__PPC64__) || defined(__powerpc64__) || defined(__PPC__)) && defined(OPAL_GCC_INLINE_ASSEMBLY) - -// special processing for ppc64 to save and restore TOC (r2) -// Reference: "64-bit PowerPC ELF Application Binary Interface Supplement 1.9" -#define OPAL_PATCHER_BEGIN \ - unsigned long toc_save; \ - asm volatile ("std 2, %0" : "=m" (toc_save)); \ - asm volatile ("nop; nop; nop; nop; nop"); -#define OPAL_PATCHER_END \ - asm volatile ("ld 2, %0" : : "m" (toc_save)); - -#else // !__PPC64__ - -#define OPAL_PATCHER_BEGIN -#define OPAL_PATCHER_END - -#endif - - -/** - * Patch all instances of calls to a function - * - * @param[in] func_symbol_name Name of symbol to patch - * @param[in] func_new_addr Pointer to new function to call - * - * @returns OPAL_SUCCESS on success - * @returns OPAL_ERR_NOT_AVAILABLE if symbol patching is not supported on this - * platform. - * - * This function patches any calls to the named symbol with a new function - * address. Any function that is passed into func_new_addr MUST begin with - * OPAL_PATCHER_BEGIN and end with OPAL_PATCHER_END. - */ -int opal_patch_symbol (const char *func_symbol_name, uintptr_t func_new_addr); - -/** - * Check if symbol patching is available - */ -bool opal_patch_supported (void); - -#endif /* !defined(OPAL_PATCHER_H) */ From 91bcab93cb444db66040a48a499f384ac263ffa0 Mon Sep 17 00:00:00 2001 From: Nathan Hjelm Date: Sat, 2 Apr 2016 12:06:15 -0600 Subject: [PATCH 6/8] opal/memory: remove ptmalloc2 This commit removes the ptmalloc2 memory hooks. This is necessary in order to support lazy registration of memory hooks. A feature that is not supported by the ptmalloc hooks but is supported by the new patcher hooks. Signed-off-by: Nathan Hjelm --- opal/mca/memory/linux/COPYRIGHT-ptmalloc2.txt | 19 - opal/mca/memory/linux/ChangeLog-ptmalloc2.txt | 181 - opal/mca/memory/linux/Makefile.am | 92 - opal/mca/memory/linux/README-open-mpi.txt | 161 - opal/mca/memory/linux/README-ptmalloc2.txt | 192 - opal/mca/memory/linux/arena.c | 805 --- opal/mca/memory/linux/configure.m4 | 236 - .../memory/linux/help-opal-memory-linux.txt | 29 - opal/mca/memory/linux/hooks.c | 899 --- opal/mca/memory/linux/lran2.h | 51 - opal/mca/memory/linux/malloc-stats.c | 169 - opal/mca/memory/linux/malloc.c | 5599 ----------------- opal/mca/memory/linux/malloc.h | 300 - opal/mca/memory/linux/memory_linux.h | 90 - .../mca/memory/linux/memory_linux_component.c | 407 -- opal/mca/memory/linux/memory_linux_munmap.c | 92 - .../mca/memory/linux/memory_linux_ptmalloc2.c | 136 - .../memory/linux/memory_linux_ummunotify.c | 235 - opal/mca/memory/linux/owner.txt | 7 - opal/mca/memory/linux/public.h | 24 - opal/mca/memory/linux/rename.h | 52 - .../mca/memory/linux/sysdeps/generic/atomic.h | 1 - .../linux/sysdeps/generic/malloc-machine.h | 68 - .../memory/linux/sysdeps/generic/thread-st.h | 48 - .../linux/sysdeps/pthread/malloc-machine.h | 132 - .../memory/linux/sysdeps/pthread/thread-st.h | 111 - .../linux/sysdeps/solaris/malloc-machine.h | 51 - .../memory/linux/sysdeps/solaris/thread-st.h | 72 - .../linux/sysdeps/sproc/malloc-machine.h | 51 - .../memory/linux/sysdeps/sproc/thread-st.h | 84 - opal/mca/memory/linux/t-test.h | 143 - opal/mca/memory/linux/t-test1.c | 285 - opal/mca/memory/linux/t-test2.c | 231 - opal/mca/memory/linux/tst-mallocstate.c | 82 - opal/mca/memory/linux/tst-mstats.c | 100 - 35 files changed, 11235 deletions(-) delete mode 100644 opal/mca/memory/linux/COPYRIGHT-ptmalloc2.txt delete mode 100644 opal/mca/memory/linux/ChangeLog-ptmalloc2.txt delete mode 100644 opal/mca/memory/linux/Makefile.am delete mode 100644 opal/mca/memory/linux/README-open-mpi.txt delete mode 100644 opal/mca/memory/linux/README-ptmalloc2.txt delete mode 100644 opal/mca/memory/linux/arena.c delete mode 100644 opal/mca/memory/linux/configure.m4 delete mode 100644 opal/mca/memory/linux/help-opal-memory-linux.txt delete mode 100644 opal/mca/memory/linux/hooks.c delete mode 100644 opal/mca/memory/linux/lran2.h delete mode 100644 opal/mca/memory/linux/malloc-stats.c delete mode 100644 opal/mca/memory/linux/malloc.c delete mode 100644 opal/mca/memory/linux/malloc.h delete mode 100644 opal/mca/memory/linux/memory_linux.h delete mode 100644 opal/mca/memory/linux/memory_linux_component.c delete mode 100644 opal/mca/memory/linux/memory_linux_munmap.c delete mode 100644 opal/mca/memory/linux/memory_linux_ptmalloc2.c delete mode 100644 opal/mca/memory/linux/memory_linux_ummunotify.c delete mode 100644 opal/mca/memory/linux/owner.txt delete mode 100644 opal/mca/memory/linux/public.h delete mode 100644 opal/mca/memory/linux/rename.h delete mode 100644 opal/mca/memory/linux/sysdeps/generic/atomic.h delete mode 100644 opal/mca/memory/linux/sysdeps/generic/malloc-machine.h delete mode 100644 opal/mca/memory/linux/sysdeps/generic/thread-st.h delete mode 100644 opal/mca/memory/linux/sysdeps/pthread/malloc-machine.h delete mode 100644 opal/mca/memory/linux/sysdeps/pthread/thread-st.h delete mode 100644 opal/mca/memory/linux/sysdeps/solaris/malloc-machine.h delete mode 100644 opal/mca/memory/linux/sysdeps/solaris/thread-st.h delete mode 100644 opal/mca/memory/linux/sysdeps/sproc/malloc-machine.h delete mode 100644 opal/mca/memory/linux/sysdeps/sproc/thread-st.h delete mode 100644 opal/mca/memory/linux/t-test.h delete mode 100644 opal/mca/memory/linux/t-test1.c delete mode 100644 opal/mca/memory/linux/t-test2.c delete mode 100644 opal/mca/memory/linux/tst-mallocstate.c delete mode 100644 opal/mca/memory/linux/tst-mstats.c diff --git a/opal/mca/memory/linux/COPYRIGHT-ptmalloc2.txt b/opal/mca/memory/linux/COPYRIGHT-ptmalloc2.txt deleted file mode 100644 index 4615c2637cc..00000000000 --- a/opal/mca/memory/linux/COPYRIGHT-ptmalloc2.txt +++ /dev/null @@ -1,19 +0,0 @@ -Copyright (c) 2001-2004 Wolfram Gloger - -Permission to use, copy, modify, distribute, and sell this software -and its documentation for any purpose is hereby granted without fee, -provided that (i) the above copyright notices and this permission -notice appear in all copies of the software and related documentation, -and (ii) the name of Wolfram Gloger may not be used in any advertising -or publicity relating to the software. - -THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF ANY KIND, -EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY -WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. - -IN NO EVENT SHALL WOLFRAM GLOGER BE LIABLE FOR ANY SPECIAL, -INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND, OR ANY -DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, -WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY -OF LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR -PERFORMANCE OF THIS SOFTWARE. diff --git a/opal/mca/memory/linux/ChangeLog-ptmalloc2.txt b/opal/mca/memory/linux/ChangeLog-ptmalloc2.txt deleted file mode 100644 index e985550c9f5..00000000000 --- a/opal/mca/memory/linux/ChangeLog-ptmalloc2.txt +++ /dev/null @@ -1,181 +0,0 @@ -2004-11-05 Wolfram Gloger - - * malloc/hooks.c (malloc_starter, memalign_starter): Call - ptmalloc_init_minimal(). - -2004-11-04 Wolfram Gloger - - * malloc/malloc.c (USE_STARTER): New macro. - * malloc/hooks.c: Use USE_STARTER. - * malloc/arena.c: Use USE_STARTER. - -2004-08-13 Ulrich Drepper - - * malloc/malloc.c: Use strong_alias instead of weak_alias wherever - possible. - -2002-12-06 Roland McGrath - - * malloc/arena.c (ptmalloc_init_minimal): New function, broken out - of ptmalloc_init. - -2002-08-23 Roland McGrath - - * malloc/hooks.c (__malloc_initialize_hook, __free_hook, - __malloc_hook, __realloc_hook, __memalign_hook, - __after_morecore_hook): Variable definitions moved to ... - * malloc/malloc.c: ... here, so as to be before all references. - -2004-10-19 Wolfram Gloger - - * malloc/hooks.c (mem2chunk_check, top_check): Handle - non-contiguous arena. Reported by Michael Dalton - [BZ #457]. Add further checks for top - chunk. - -2004-08-08 Wolfram Gloger - - * include/malloc.h (mstate): Move type declaration from here... - * malloc/malloc.h: ...to here. - (struct malloc_arena_info, struct malloc_global_info): New types. - (_int_get_arena, _int_get_arena_info, _int_get_global_info): New - functions. - * malloc/malloc.c (mSTATS, public_mSTATs, mALLINFo): Remove. - (_int_get_arena_info, _int_get_global_info): New functions. - * malloc/arena.c (_int_get_arena): New function. - * malloc/malloc-stats.c: New file. - * malloc/tst-mstats.c: New file. - * malloc/Makefile (tests): Add tst-mstats. - (distribute): Remove no-longer existing thread-m.h. - (dist-routines): Add malloc-stats. - * malloc/Versions: Add _int_get_arena, _int_get_arena_info, - _int_get_global_info. - -2004-07-25 Wolfram Gloger - - * sysdeps/generic/thread-st.h: New file. - * sysdeps/pthread/thread-st.h: New file. - * sysdeps/sproc/thread-st.h: New file. - * sysdeps/solaris/thread-st.h: New file. - * thread-st.h: Removed. - -2004-03-18 Ulrich Drepper - - * malloc/malloc.c (__posix_memalign): Correct alignment check. - Reported by Don Heller . - -2003-12-17 Jakub Jelinek - - * malloc/malloc.c (__posix_memalign): If __memalign_hook != NULL, - call it directly instead of memalign_internal. - -2003-09-27 Wolfram Gloger - - * malloc/malloc.c: Include earlier instead of - "thread-m.h", so that default parameters can be overridden in a - system-specific malloc-machine.h. Remove extra ; from extern "C" - closing brace. - * sysdeps/generic/malloc-machine.h: New file. - * malloc/thread-m.h: Removed. - -2003-09-08 Wolfram Gloger - - * malloc/malloc.c (sYSMALLOc): Move foreign sbrk accounting into - contiguous case. Bug report from Prem Gopalan - . - -2003-08-18 Art Haas - - * malloc/malloc.h: Remove unneeded ';' where closing the C++ - extern block. - -2003-06-18 Ulrich Drepper - - * malloc/malloc.c (public_mALLINFo): Initialize malloc if it - hasn't happened yet. - -2003-05-28 Roland McGrath - - * malloc/malloc.h [! __GNUC__] (__const): Define if undefined. - -2003-05-04 H.J. Lu - - * malloc/arena.c (arena_get2): Add atomic_write_barrier. - * malloc/thread-m.h: Include . - (atomic_full_barrier): Provide default. - (atomic_read_barrier): Likewise. - (atomic_write_barrier): Likewise. - -2003-05-01 Ulrich Drepper - - * malloc/malloc.c (mSTATs): Call ptmalloc_init if necessary. - -2003-01-27 Wolfram Gloger - - * malloc/hooks.c (mem2chunk_check): Check alignment of mem - pointer, not of the computed chunk. Bug report from Carlos - O'Donell . - -2002-12-27 Jakub Jelinek - - * malloc/arena.c (ptmalloc_init): Don't call next_env_entry if - _environ is NULL. - -2002-12-17 Ulrich Drepper - - * malloc/malloc.c (mALLOPt): Make sure malloc is initialized. - -2002-12-06 Roland McGrath - - * malloc/hooks.c [_LIBC && (USE___THREAD || (USE_TLS && !SHARED))] - (malloc_starter, memalign_starter, free_starter): Don't define these. - - * malloc/hooks.c (memalign_starter): New function. - * malloc/malloc.c: Declare it. - * malloc/arena.c (save_memalign_hook): New variable. - (ptmalloc_init): Set __memalign_hook to memalign_starter. - -2002-11-18 Wolfram Gloger - - * malloc/arena.c - (ptmalloc_lock_all, ptmalloc_unlock_all, ptmalloc_unlock_all2): Do - nothing if not initialized. Bug report from Marcus Brinkmann - . - -2002-10-07 Wolfram Gloger - - * malloc/malloc.c (sYSMALLOc): Only check for breakage due - to foreign sbrk()'s if arena is contiguous. Bug report from - Bruno Haible . - -2002-07-11 Wolfram Gloger - - * malloc/hooks.c: typo fix in NO_THREADS case, realloc_check - fix in HAVE_MREMAP case. - -2002-06-11 Wolfram Gloger - - * malloc/malloc.c: Fix error path when new_heap() returns NULL. - Reported by Michael Meissner . - -2002-03-29 Wolfram Gloger - - * malloc/malloc.c: Add short description and prototypes for - malloc_get_state, malloc_set_state and posix_memalign, for - consistency and to avoid warnings with -Wstrict-prototypes. - Reported by Andreas Jaeger . - -2002-03-13 Wolfram Gloger - - * malloc/malloc.c (sYSMALLOc): Don't change brk if mmap - failed. - -2002-01-18 Wolfram Gloger - - * malloc/malloc.c: Rewrite, adapted from Doug Lea's malloc-2.7.0.c. - * malloc/malloc.h: Likewise. - * malloc/arena.c: New file. - * malloc/hooks.c: New file. - * malloc/tst-mallocstate.c: New file. - * malloc/Makefile: Add new testcase tst-mallocstate. - Add arena.c and hooks.c to distribute. Fix commented CPPFLAGS. diff --git a/opal/mca/memory/linux/Makefile.am b/opal/mca/memory/linux/Makefile.am deleted file mode 100644 index c29fc5d87ff..00000000000 --- a/opal/mca/memory/linux/Makefile.am +++ /dev/null @@ -1,92 +0,0 @@ -# -# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana -# University Research and Technology -# Corporation. All rights reserved. -# Copyright (c) 2004-2005 The University of Tennessee and The University -# of Tennessee Research Foundation. All rights -# reserved. -# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, -# University of Stuttgart. All rights reserved. -# Copyright (c) 2004-2005 The Regents of the University of California. -# All rights reserved. -# Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. -# Copyright (c) 2015 Research Organization for Information Science -# and Technology (RIST). All rights reserved. -# $COPYRIGHT$ -# -# Additional copyrights may follow -# -# $HEADER$ -# - -AM_CPPFLAGS = -DMALLOC_DEBUG=0 - -AM_CPPFLAGS += \ - -D_GNU_SOURCE=1 \ - -DMALLOC_HOOKS=1 \ - -I$(srcdir)/sysdeps/pthread -# this must come *after* the threads -Is -AM_CPPFLAGS += -I$(srcdir)/sysdeps/generic - -# ptmalloc requires us to include the copyright notice in the -# software. So install it in the same place that we install ROMIO's -# copyright notices. - -docdir = $(opaldatadir)/doc -doc_DATA = COPYRIGHT-ptmalloc2.txt - -# Help file -dist_opaldata_DATA = help-opal-memory-linux.txt - -# This component is only ever built statically (i.e., slurped into -# libopen-pal) -- it is never built as a DSO. -noinst_LTLIBRARIES = libmca_memory_linux.la -libmca_memory_linux_la_SOURCES = \ - memory_linux.h \ - memory_linux_component.c -libmca_memory_linux_la_LDFLAGS = \ - -module -avoid-version $(memory_linux_LDFLAGS) -libmca_memory_linux_la_LIBADD = $(memory_linux_LIBS) - -# Do we have ptmalloc2 support? -if MEMORY_LINUX_PTMALLOC2 -libmca_memory_linux_la_SOURCES += \ - memory_linux_ptmalloc2.c \ - memory_linux_munmap.c \ - rename.h \ - malloc.c \ - malloc-stats.c \ - malloc.h -endif - -# Do we have ummunotify support? -if MEMORY_LINUX_UMMUNOTIFY -libmca_memory_linux_la_SOURCES += memory_linux_ummunotify.c public.h -endif - -# these are included directly and shouldn't be built solo -EXTRA_libmca_memory_linux_la_SOURCES = \ - arena.c \ - hooks.c - -EXTRA_DIST = \ - README-open-mpi.txt \ - README-ptmalloc2.txt \ - ChangeLog-ptmalloc2.txt \ - COPYRIGHT-ptmalloc2.txt \ - lran2.h \ - t-test.h \ - t-test1.c \ - t-test2.c \ - tst-mallocstate.c \ - tst-mstats.c \ - sysdeps/sproc/malloc-machine.h \ - sysdeps/sproc/thread-st.h \ - sysdeps/pthread/malloc-machine.h \ - sysdeps/pthread/thread-st.h \ - sysdeps/solaris/malloc-machine.h \ - sysdeps/solaris/thread-st.h \ - sysdeps/generic/malloc-machine.h \ - sysdeps/generic/thread-st.h \ - sysdeps/generic/atomic.h \ - $(doc_DATA) diff --git a/opal/mca/memory/linux/README-open-mpi.txt b/opal/mca/memory/linux/README-open-mpi.txt deleted file mode 100644 index 43e88eba8a0..00000000000 --- a/opal/mca/memory/linux/README-open-mpi.txt +++ /dev/null @@ -1,161 +0,0 @@ -30 March 2009 - -This file documents Open MPI's usage of ptmalloc2. This is perhaps -our 7,208,499th iteration of ptmalloc2 support, so let's document it -here so that some future developer might spend *slightly* less time -understanding what the heck is going on. - -See glibc documentation about malloc hooks before continuing. This is -pretty much required reading before reading the rest of this file / -having a hope of understanding what's going on here: - - http://www.gnu.org/software/libc/manual/html_mono/libc.html#Hooks-for-Malloc - -The overall goal is that we're using the Linux glibc hooks to wholly -replace the underlying allocator. We *used* to use horrid linker -tricks to interpose OMPI's ptmalloc2 symbols with the glibc ones -- -meaning that user apps would call our symbols and not the glibc ones. -But that scheme is fraught with problems, not the least of which is -that *all* MPI applications will be forced to use our overridden -allocator (not just the ones that need it, such as the ones running on -OpenFabrics-based networks). Instead, what we do here is, frankly, -quite similar to what is done in MX: we use the 4 glibc hooks to -assert our own malloc, realloc, free, and memalign functions. This -allows the decision as to whether to use this internal ptmalloc2 -allocate to be a run-time decision. This is quite important; using -this internal allocator has both benefits (allowing using -mpi_leave_pinned=1 behavior) and drawbacks (breaking some debuggers, -being unnecessary for non-OpenFabrics-based networks, etc.). - -Here's how it works... - -This component *must* be linked statically as part of libopen-pal; it -*cannot* be a DSO. Specifically, this library must be present during -pre-main() initialization phases so that its __malloc_initialize_hook -can be found and executed. Loading it as a DSO during MPI_INIT is far -too late. In configure.m4, we define the M4 macro -MCA_memory_ptmalloc2_COMPILE_MODE to always compile this component in -static mode. Yay flexible build system. - -This component provides an munmap() function that will intercept calls -to munmap() and do the Right Thing. That is fairly straightforward to -do. Intercepting the malloc/free/etc. allocator is much more -complicated. - -All the ptmalloc2 public symbols in this component have been name -shifted via the rename.h file. Hence, what used to be "malloc" is now -opal_memory_ptmalloc2_malloc. Since all the public symbols are -name-shifted, we can safely link this component in all MPI -applications. Specifically: just because this ptmalloc2 allocator is -present in all OMPI executables and user-level applications, it won't -necessarily be used -- it's a separate/run-time decision as to whether -it will be used. - -We set the __malloc_initialize_hook variable to point to -opal_memory_ptmalloc2_malloc_init_hook (in hooks.c). This function is -called by the underlying glibc allocator before any allocations occur -and before the memory allocation subsystem is setup. As such, this -function is *extremely* restricted in what it can do. It cannot call -any form of malloc, for example (which seems fairly obvious, but it's -worth mentioning :-) ). This function is one of the determining -steps as to whether we'll use the internal ptmalloc2 allocator or -not. Several checks are performed: - -- Was either the MCA params mpi_leave_pinned or - mpi_leave_pinned_pipeline set? -- Is a driver found to be active indicating that an OS-bypass network - is in effect (OpenFabrics, MX, Open-MX, ...etc.) -- Was an environment variable set indicating that we want to disable - this component? - -If the $OMPI_MCA_memory_ptmalloc2_disable or the $FAKEROOTKEY env -variables are set, we don't enable the memory hooks. - -We then use the following matrix to determine whether to enable the -memory hooks or not (explanation of the matrix is below): - - lp / lpp yes no runtime not found - yes yes yes yes yes - no yes no no no - runtime yes no runtime runtime - not found yes no runtime runtime - -lp = leave_pinned (the rows), lpp = leave_pinned_pipeline (the columns) -yes = found that variable to be set to "yes" (i.e., 1) -no = found that variable to be set to "no" (i.e., 0) -runtime = found that variable to be set to "determine at runtime" (i.e., -1) -not found = that variable was not set at all - -Hence, if we end up on a "yes" block in the matrix, we enable the -hooks. If we end up in a "no" block in the matrix, we disable the -hooks. If we end up in a "runtime" block in the matrix, then we -enable the hooks *if* we can find indications that an OS bypass -network is present and available for use (e.g., OpenFabrics, MX, -Open-MX, ...etc.). - -To be clear: sometime during process startup, this function will -definitely be called. It will either set the 4 hook functions to -point to our name-shifted ptmalloc2 functions, or it won't. If the 4 -hook functions are set, then the underlying glibc allocator will -always call our 4 functions in all the relevant places instead of -calling its own functions. Specifically: the process is calling the -underlying glibc allocator, but that underlying glibc allocator will -make function pointer callbacks to our name-shifted ptmalloc2 -functions to actually do the work. - -Note that because we know our ptmalloc will not be providing all 5 -hook variables (because we want to use the underlying glibc hook -variables), they are #if 0'ed out in our malloc.c. This has the -direct consequence that the *_hook_ini() in hooks.c are never used. -So to avoid compiler/linker warnings, I #if 0'ed those out as well. - -All the public functions in malloc.c that call hook functions were -modified to #if 0 the hook function invocations. After all, that's -something that we want the *underlying* glibc allocator to do -- but -we are putting these functions as the hooks, so we don't want to -invoke ourselves in an infinite loop! - -The next thing that happens in the startup sequence is that the -ptmalloc2 memory component's "open" function is called during -MPI_INIT. But we need to test to see if the glibc memory hooks have -been overridden before MPI_INIT was invoked. If so, we need to signal -that our allocator support may not be complete. - -Patrick Geoffray/MX suggests a simple test: malloc() 4MB and then free -it. Watch to see if our name-shifted ptmalloc2 free() function was -invoked. If it was, then all of our hooks are probably in place and -we can proceed. If not, then set flags indicating that this memory -allocator only supports MUNMAP (not FREE/CHUNK). - -We actually perform this test for malloc, realloc, and memalign. If -they all pass, then we say that the memory allocator supports -everything. If any of them fail, then we say that the memory -allocator does not support FREE/CHUNK. - -NOTE: we *used* to simply set the FREE/CHUNK support flags during our -ptmalloc2's internal ptmalloc_init() function. This is not a good -idea becaus even after our ptmalloc_init() function has been invoked, -someone may come in an override our memory hooks. Doing tests during -the ptmalloc2 memory component's open function seems to be the safest -way to test whether we *actually* support FREE/CHUNK (this is what MX -does, too). - -As stated above, we always intercept munmap() -- this is acceptable in -all environments. But we test that, too, just to be sure that the -munmap intercept is working. If we verify that it is working -properly, then we set that we have MUNMAP support. - -Much later in the init sequence during MPI_INIT, components indicate -whether they want to use mpi_leave_pinned[_pipeline] support or not. -For example, the openib BTL queries the opal_mem_hooks_support_level() -function to see if FREE and MUNMAP are supported. If they are, then -the openib BTL sets mpi_leave_pinned = 1. - -Finally, the mpool base does a final check. If -mpi_leave_pinned[_pipeline] is set to 1 and/or use_mem_hooks is set, -if FREE/MUNMAP are not set in the supported flags, then a warning is -printed. Otherwise, life continues (assumedly using -mpi_leave_pinned[_pipeline] support). - -Simple, right? - diff --git a/opal/mca/memory/linux/README-ptmalloc2.txt b/opal/mca/memory/linux/README-ptmalloc2.txt deleted file mode 100644 index fed69951346..00000000000 --- a/opal/mca/memory/linux/README-ptmalloc2.txt +++ /dev/null @@ -1,192 +0,0 @@ -ptmalloc2 - a multi-thread malloc implementation -================================================ - -Wolfram Gloger (wg@malloc.de) - -Nov 2004 - - -Introduction -============ - -This package is a modified version of Doug Lea's malloc-2.7.1pre -implementation (available seperately from ftp://g.oswego.edu/pub/misc) -that I adapted for multiple threads, while trying to avoid lock -contention as much as possible. Many thanks should go to Doug Lea -(dl@cs.oswego.edu) for the great original malloc implementation. - -As part of the GNU C library, the source files are available under the -GNU Library General Public License (see the comments in the files). -But as part of this stand-alone package, the code is also available -under the (probably less restrictive) conditions described in the file -'COPYRIGHT'. In any case, there is no warranty whatsoever for this -package. - -The current distribution should be available from: - -http://www.malloc.de/malloc/ptmalloc2.tar.gz - - -Compilation -=========== - -It should be possible to build ptmalloc2 on any UN*X-like system that -implements the sbrk(), mmap(), munmap() and mprotect() calls. If -mmap() is not available, it is only possible to produce a -non-threadsafe implementation. Since there are now several source -files, a library (libmalloc.a) is generated. See the Makefile for -examples of the compile-time options. - -Note that support for non-ANSI compilers is no longer a significant -goal. - -Several example targets are provided in the Makefile: - - o Posix threads (pthreads), compile with "make posix" - - o Posix threads with explicit initialization, compile with - "make posix-explicit" (known to be required on HPUX) - - o Posix threads without "tsd data hack" (see below), compile with - "make posix-with-tsd" - - o Solaris threads, compile with "make solaris" - - o SGI sproc() threads, compile with "make sproc" - - o no threads, compile with "make nothreads" - -For Linux: - - o make "linux-pthread" (almost the same as "make posix") - -Note that some compilers need special flags for multi-threaded code, -e.g. with Solaris cc with Posix threads, one should use: - -% make posix SYS_FLAGS='-mt' - -Some additional targets, ending in `-libc', are also provided in the -Makefile, to compare performance of the test programs to the case when -linking with the standard malloc implementation in libc. - -A potential problem remains: If any of the system-specific functions -for getting/setting thread-specific data or for locking a mutex call -one of the malloc-related functions internally, the implementation -cannot work at all due to infinite recursion. One example seems to be -Solaris 2.4. I would like to hear if this problem occurs on other -systems, and whether similar workarounds could be applied. - -For Posix threads, too, an optional hack like that has been integrated -(activated when defining USE_TSD_DATA_HACK) which depends on -`pthread_t' being convertible to an integral type (which is of course -not generally guaranteed). USE_TSD_DATA_HACK is now the default -because I haven't yet found a non-glibc pthreads system where this -hack is _not_ needed. - -*NEW* and _important_: In (currently) one place in the ptmalloc2 -source, a write memory barrier is needed, named -atomic_write_barrier(). This macro needs to be defined at the end of -malloc-machine.h. For gcc, a fallback in the form of a full memory -barrier is already defined, but you may need to add another definition -if you don't use gcc. - -Usage -===== - -Just link libmalloc.a into your application. - -Some wicked systems (e.g. HPUX apparently) won't let malloc call _any_ -thread-related functions before main(). On these systems, -USE_STARTER=2 must be defined during compilation (see "make -posix-explicit" above) and the global initialization function -ptmalloc_init() must be called explitly, preferably at the start of -main(). - -Otherwise, when using ptmalloc2, no special precautions are necessary. - -Link order is important -======================= - -On some systems, when overriding malloc and linking against shared -libraries, the link order becomes very important. E.g., when linking -C++ programs on Solaris, don't rely on libC being included by default, -but instead put `-lthread' behind `-lC' on the command line: - - CC ... libmalloc.a -lC -lthread - -This is because there are global constructors in libC that need -malloc/ptmalloc, which in turn needs to have the thread library to be -already initialized. - -Debugging hooks -=============== - -All calls to malloc(), realloc(), free() and memalign() are routed -through the global function pointers __malloc_hook, __realloc_hook, -__free_hook and __memalign_hook if they are not NULL (see the malloc.h -header file for declarations of these pointers). Therefore the malloc -implementation can be changed at runtime, if care is taken not to call -free() or realloc() on pointers obtained with a different -implementation than the one currently in effect. (The easiest way to -guarantee this is to set up the hooks before any malloc call, e.g. -with a function pointed to by the global variable -__malloc_initialize_hook). - -A useful application of the hooks is built-in into ptmalloc2: The -implementation is usually very unforgiving with respect to misuse, -such as free()ing a pointer twice or free()ing a pointer not obtained -with malloc() (these will typically crash the application -immediately). To debug in such situations, you can set the -environment variable `MALLOC_CHECK_' (note the trailing underscore). -Performance will suffer somewhat, but you will get more controlled -behaviour in the case of misuse. If MALLOC_CHECK_=0, wrong free()s -will be silently ignored, if MALLOC_CHECK_=1, diagnostics will be -printed on stderr, and if MALLOC_CHECK_=2, abort() will be called on -any error. - -You can now also tune other malloc parameters (normally adjused via -mallopt() calls from the application) with environment variables: - - MALLOC_TRIM_THRESHOLD_ for deciding to shrink the heap (in bytes) - - MALLOC_TOP_PAD_ how much extra memory to allocate on - each system call (in bytes) - - MALLOC_MMAP_THRESHOLD_ min. size for chunks allocated via - mmap() (in bytes) - - MALLOC_MMAP_MAX_ max. number of mmapped regions to use - -Tests -===== - -Two testing applications, t-test1 and t-test2, are included in this -source distribution. Both perform pseudo-random sequences of -allocations/frees, and can be given numeric arguments (all arguments -are optional): - -% t-test[12] - - n-total = total number of threads executed (default 10) - n-parallel = number of threads running in parallel (2) - n-allocs = number of malloc()'s / free()'s per thread (10000) - size-max = max. size requested with malloc() in bytes (10000) - bins = number of bins to maintain - -The first test `t-test1' maintains a completely seperate pool of -allocated bins for each thread, and should therefore show full -parallelism. On the other hand, `t-test2' creates only a single pool -of bins, and each thread randomly allocates/frees any bin. Some lock -contention is to be expected in this case, as the threads frequently -cross each others arena. - -Performance results from t-test1 should be quite repeatable, while the -behaviour of t-test2 depends on scheduling variations. - -Conclusion -========== - -I'm always interested in performance data and feedback, just send mail -to ptmalloc@malloc.de. - -Good luck! diff --git a/opal/mca/memory/linux/arena.c b/opal/mca/memory/linux/arena.c deleted file mode 100644 index 3f75a1c3af5..00000000000 --- a/opal/mca/memory/linux/arena.c +++ /dev/null @@ -1,805 +0,0 @@ -/* Malloc implementation for multiple threads without lock contention. - Copyright (C) 2001 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Wolfram Gloger , 2001. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -/* $Id: arena.c,v 1.9 2004/11/05 14:42:23 wg Exp $ */ - -/* Compile-time constants. */ - -#define HEAP_MIN_SIZE (32*1024) -#ifndef HEAP_MAX_SIZE -#define HEAP_MAX_SIZE (1024*1024) /* must be a power of two */ -#endif - -/* HEAP_MIN_SIZE and HEAP_MAX_SIZE limit the size of mmap()ed heaps - that are dynamically created for multi-threaded programs. The - maximum size must be a power of two, for fast determination of - which heap belongs to a chunk. It should be much larger than the - mmap threshold, so that requests with a size just below that - threshold can be fulfilled without creating too many heaps. */ - - -#ifndef THREAD_STATS -#define THREAD_STATS 0 -#endif - -/* If THREAD_STATS is non-zero, some statistics on mutex locking are - computed. */ - -/***************************************************************************/ - -#define top(ar_ptr) ((ar_ptr)->top) - -/* A heap is a single contiguous memory region holding (coalesceable) - malloc_chunks. It is allocated with mmap() and always starts at an - address aligned to HEAP_MAX_SIZE. Not used unless compiling with - USE_ARENAS. */ - -typedef struct _heap_info { - mstate ar_ptr; /* Arena for this heap. */ - struct _heap_info *prev; /* Previous heap. */ - size_t size; /* Current size in bytes. */ - size_t pad; /* Make sure the following data is properly aligned. */ -} heap_info; - -/* Thread specific data */ - -static tsd_key_t arena_key; -static mutex_t list_lock; - -#if THREAD_STATS -static int stat_n_heaps; -#define THREAD_STAT(x) x -#else -#define THREAD_STAT(x) do ; while(0) -#endif - -/* Mapped memory in non-main arenas (reliable only for NO_THREADS). */ -static unsigned long arena_mem; - -/* Already initialized? */ -int __malloc_initialized = -1; - -/**************************************************************************/ - -#if USE_ARENAS - -/* arena_get() acquires an arena and locks the corresponding mutex. - First, try the one last locked successfully by this thread. (This - is the common case and handled with a macro for speed.) Then, loop - once over the circularly linked list of arenas. If no arena is - readily available, create a new one. In this latter case, `size' - is just a hint as to how much memory will be required immediately - in the new arena. */ - -#define arena_get(ptr, size) do { \ - Void_t *vptr = NULL; \ - ptr = (mstate)tsd_getspecific(arena_key, vptr); \ - if(ptr && !mutex_trylock(&ptr->mutex)) { \ - THREAD_STAT(++(ptr->stat_lock_direct)); \ - } else \ - ptr = arena_get2(ptr, (size)); \ -} while(0) - -/* find the heap and corresponding arena for a given ptr */ - -#define heap_for_ptr(ptr) \ - ((heap_info *)((unsigned long)(ptr) & ~(HEAP_MAX_SIZE-1))) -#define arena_for_chunk(ptr) \ - (chunk_non_main_arena(ptr) ? heap_for_ptr(ptr)->ar_ptr : &main_arena) - -#else /* !USE_ARENAS */ - -/* There is only one arena, main_arena. */ - -#if THREAD_STATS -#define arena_get(ar_ptr, sz) do { \ - ar_ptr = &main_arena; \ - if(!mutex_trylock(&ar_ptr->mutex)) \ - ++(ar_ptr->stat_lock_direct); \ - else { \ - (void)mutex_lock(&ar_ptr->mutex); \ - ++(ar_ptr->stat_lock_wait); \ - } \ -} while(0) -#else -#define arena_get(ar_ptr, sz) do { \ - ar_ptr = &main_arena; \ - (void)mutex_lock(&ar_ptr->mutex); \ -} while(0) -#endif -#define arena_for_chunk(ptr) (&main_arena) - -#endif /* USE_ARENAS */ - -/**************************************************************************/ - -#ifndef NO_THREADS - -/* atfork support. */ - -static __malloc_ptr_t (*save_malloc_hook) __MALLOC_P ((size_t __size, - __const __malloc_ptr_t)); -# if !defined _LIBC || !defined USE_TLS || (defined SHARED && !USE___THREAD) -static __malloc_ptr_t (*save_memalign_hook) __MALLOC_P ((size_t align, - size_t __size, - __const __malloc_ptr_t)); -# endif -static void (*save_free_hook) __MALLOC_P ((__malloc_ptr_t __ptr, - __const __malloc_ptr_t)); -static Void_t* save_arena; - -/* Magic value for the thread-specific arena pointer when - malloc_atfork() is in use. */ - -#define ATFORK_ARENA_PTR ((Void_t*)-1) - -/* The following hooks are used while the `atfork' handling mechanism - is active. */ - -static Void_t* -malloc_atfork(size_t sz, const Void_t *caller) -{ - Void_t *vptr = NULL; - Void_t *victim; - - tsd_getspecific(arena_key, vptr); - if(vptr == ATFORK_ARENA_PTR) { - /* We are the only thread that may allocate at all. */ - if(save_malloc_hook != malloc_check) { - return _int_malloc(&main_arena, sz); - } else { - if(top_check()<0) - return 0; - victim = _int_malloc(&main_arena, sz+1); - return mem2mem_check(victim, sz); - } - } else { - /* Suspend the thread until the `atfork' handlers have completed. - By that time, the hooks will have been reset as well, so that - mALLOc() can be used again. */ - (void)mutex_lock(&list_lock); - (void)mutex_unlock(&list_lock); - return public_mALLOc(sz); - } -} - -static void -free_atfork(Void_t* mem, const Void_t *caller) -{ - Void_t *vptr = NULL; - mstate ar_ptr; - mchunkptr p; /* chunk corresponding to mem */ - - if (mem == 0) /* free(0) has no effect */ - return; - - p = mem2chunk(mem); /* do not bother to replicate free_check here */ - -#if HAVE_MMAP - if (chunk_is_mmapped(p)) /* release mmapped memory. */ - { - munmap_chunk(p); - return; - } -#endif - - ar_ptr = arena_for_chunk(p); - tsd_getspecific(arena_key, vptr); - if(vptr != ATFORK_ARENA_PTR) - (void)mutex_lock(&ar_ptr->mutex); - _int_free(ar_ptr, mem); - if(vptr != ATFORK_ARENA_PTR) - (void)mutex_unlock(&ar_ptr->mutex); -} - -/* The following two functions are registered via thread_atfork() to - make sure that the mutexes remain in a consistent state in the - fork()ed version of a thread. Also adapt the malloc and free hooks - temporarily, because the `atfork' handler mechanism may use - malloc/free internally (e.g. in LinuxThreads). */ - -static void -ptmalloc_lock_all __MALLOC_P((void)) -{ - mstate ar_ptr; - - if(__malloc_initialized < 1) - return; - (void)mutex_lock(&list_lock); - for(ar_ptr = &main_arena;;) { - (void)mutex_lock(&ar_ptr->mutex); - ar_ptr = ar_ptr->next; - if(ar_ptr == &main_arena) break; - } - save_malloc_hook = __malloc_hook; - save_free_hook = __free_hook; - __malloc_hook = malloc_atfork; - __free_hook = free_atfork; - /* Only the current thread may perform malloc/free calls now. */ - tsd_getspecific(arena_key, save_arena); - tsd_setspecific(arena_key, ATFORK_ARENA_PTR); -} - -static void -ptmalloc_unlock_all __MALLOC_P((void)) -{ - mstate ar_ptr; - - if(__malloc_initialized < 1) - return; - tsd_setspecific(arena_key, save_arena); - __malloc_hook = save_malloc_hook; - __free_hook = save_free_hook; - for(ar_ptr = &main_arena;;) { - (void)mutex_unlock(&ar_ptr->mutex); - ar_ptr = ar_ptr->next; - if(ar_ptr == &main_arena) break; - } - (void)mutex_unlock(&list_lock); -} - -#ifdef __linux__ - -/* In LinuxThreads, unlocking a mutex in the child process after a - fork() is currently unsafe, whereas re-initializing it is safe and - does not leak resources. Therefore, a special atfork handler is - installed for the child. */ - -static void -ptmalloc_unlock_all2 __MALLOC_P((void)) -{ - mstate ar_ptr; - - if(__malloc_initialized < 1) - return; -#if defined _LIBC || defined MALLOC_HOOKS - tsd_setspecific(arena_key, save_arena); - __malloc_hook = save_malloc_hook; - __free_hook = save_free_hook; -#endif - for(ar_ptr = &main_arena;;) { - (void)mutex_init(&ar_ptr->mutex); - ar_ptr = ar_ptr->next; - if(ar_ptr == &main_arena) break; - } - (void)mutex_init(&list_lock); -} - -#else - -#define ptmalloc_unlock_all2 ptmalloc_unlock_all - -#endif - -#endif /* !defined NO_THREADS */ - -/* Initialization routine. */ -#ifdef _LIBC -#include -extern char **_environ; - -static char * -internal_function -next_env_entry (char ***position) -{ - char **current = *position; - char *result = NULL; - - while (*current != NULL) - { - if (__builtin_expect ((*current)[0] == 'M', 0) - && (*current)[1] == 'A' - && (*current)[2] == 'L' - && (*current)[3] == 'L' - && (*current)[4] == 'O' - && (*current)[5] == 'C' - && (*current)[6] == '_') - { - result = &(*current)[7]; - - /* Save current position for next visit. */ - *position = ++current; - - break; - } - - ++current; - } - - return result; -} -#endif /* _LIBC */ - -/* Set up basic state so that _int_malloc et al can work. */ -static void -ptmalloc_init_minimal __MALLOC_P((void)) -{ -#if DEFAULT_TOP_PAD != 0 - mp_.top_pad = DEFAULT_TOP_PAD; -#endif - mp_.n_mmaps_max = DEFAULT_MMAP_MAX; - mp_.mmap_threshold = DEFAULT_MMAP_THRESHOLD; - mp_.trim_threshold = DEFAULT_TRIM_THRESHOLD; - mp_.pagesize = malloc_getpagesize; -} - - -#if !(USE_STARTER & 2) -static -#endif -void -ptmalloc_init __MALLOC_P((void)) -{ -#if __STD_C - const char* s; -#else - char* s; -#endif - int secure = 0; - - if(__malloc_initialized >= 0) return; - __malloc_initialized = 0; - - if (mp_.pagesize == 0) - ptmalloc_init_minimal(); - -#ifndef NO_THREADS -# if USE_STARTER & 1 - /* With some threads implementations, creating thread-specific data - or initializing a mutex may call malloc() itself. Provide a - simple starter version (realloc() won't work). */ - save_malloc_hook = __malloc_hook; - save_memalign_hook = __memalign_hook; - save_free_hook = __free_hook; - __malloc_hook = malloc_starter; - __memalign_hook = memalign_starter; - __free_hook = free_starter; -# ifdef _LIBC - /* Initialize the pthreads interface. */ - if (__pthread_initialize != NULL) - __pthread_initialize(); -# endif /* !defined _LIBC */ -# endif /* USE_STARTER & 1 */ -#endif /* !defined NO_THREADS */ - mutex_init(&main_arena.mutex); - main_arena.next = &main_arena; - - mutex_init(&list_lock); - tsd_key_create(&arena_key, NULL); - tsd_setspecific(arena_key, (Void_t *)&main_arena); - thread_atfork(ptmalloc_lock_all, ptmalloc_unlock_all, ptmalloc_unlock_all2); -#ifndef NO_THREADS -# if USE_STARTER & 1 - __malloc_hook = save_malloc_hook; - __memalign_hook = save_memalign_hook; - __free_hook = save_free_hook; -# endif -# if USE_STARTER & 2 - __malloc_hook = 0; - __memalign_hook = 0; - __free_hook = 0; -# endif -#endif -#ifdef _LIBC - secure = __libc_enable_secure; - s = NULL; - if (__builtin_expect (_environ != NULL, 1)) - { - char **runp = _environ; - char *envline; - - while (__builtin_expect ((envline = next_env_entry (&runp)) != NULL, - 0)) - { - size_t len = strcspn (envline, "="); - - if (envline[len] != '=') - /* This is a "MALLOC_" variable at the end of the string - without a '=' character. Ignore it since otherwise we - will access invalid memory below. */ - continue; - - switch (len) - { - case 6: - if (memcmp (envline, "CHECK_", 6) == 0) - s = &envline[7]; - break; - case 8: - if (! secure && memcmp (envline, "TOP_PAD_", 8) == 0) - mALLOPt(M_TOP_PAD, atoi(&envline[9])); - break; - case 9: - if (! secure && memcmp (envline, "MMAP_MAX_", 9) == 0) - mALLOPt(M_MMAP_MAX, atoi(&envline[10])); - break; - case 15: - if (! secure) - { - if (memcmp (envline, "TRIM_THRESHOLD_", 15) == 0) - mALLOPt(M_TRIM_THRESHOLD, atoi(&envline[16])); - else if (memcmp (envline, "MMAP_THRESHOLD_", 15) == 0) - mALLOPt(M_MMAP_THRESHOLD, atoi(&envline[16])); - } - break; - default: - break; - } - } - } -#else - if (! secure) - { - if((s = getenv("MALLOC_TRIM_THRESHOLD_"))) - mALLOPt(M_TRIM_THRESHOLD, atoi(s)); - if((s = getenv("MALLOC_TOP_PAD_"))) - mALLOPt(M_TOP_PAD, atoi(s)); - if((s = getenv("MALLOC_MMAP_THRESHOLD_"))) - mALLOPt(M_MMAP_THRESHOLD, atoi(s)); - if((s = getenv("MALLOC_MMAP_MAX_"))) - mALLOPt(M_MMAP_MAX, atoi(s)); - } - s = getenv("MALLOC_CHECK_"); -#endif - if(s) { - if(s[0]) mALLOPt(M_CHECK_ACTION, (int)(s[0] - '0')); - __malloc_check_init(); - } -#if 0 - /* OMPI Change: Don't call the initialize hook; it was us. */ - if(__malloc_initialize_hook != NULL) - (*__malloc_initialize_hook)(); -#endif - - __malloc_initialized = 1; -} - -/* There are platforms (e.g. Hurd) with a link-time hook mechanism. */ -#ifdef thread_atfork_static -thread_atfork_static(ptmalloc_lock_all, ptmalloc_unlock_all, \ - ptmalloc_unlock_all2) -#endif - - - -/* Managing heaps and arenas (for concurrent threads) */ - -#if USE_ARENAS - -#if MALLOC_DEBUG > 1 - -/* Print the complete contents of a single heap to stderr. */ - -static void -#if __STD_C -dump_heap(heap_info *heap) -#else -dump_heap(heap) heap_info *heap; -#endif -{ - char *ptr; - mchunkptr p; - - fprintf(stderr, "Heap %p, size %10lx:\n", heap, (long)heap->size); - ptr = (heap->ar_ptr != (mstate)(heap+1)) ? - (char*)(heap + 1) : (char*)(heap + 1) + sizeof(struct malloc_state); - p = (mchunkptr)(((unsigned long)ptr + MALLOC_ALIGN_MASK) & - ~MALLOC_ALIGN_MASK); - for(;;) { - fprintf(stderr, "chunk %p size %10lx", p, (long)p->size); - if(p == top(heap->ar_ptr)) { - fprintf(stderr, " (top)\n"); - break; - } else if(p->size == (0|PREV_INUSE)) { - fprintf(stderr, " (fence)\n"); - break; - } - fprintf(stderr, "\n"); - p = next_chunk(p); - } -} - -#endif /* MALLOC_DEBUG > 1 */ - -/* Create a new heap. size is automatically rounded up to a multiple - of the page size. */ - -static heap_info * -internal_function -#if __STD_C -new_heap(size_t size, size_t top_pad) -#else -new_heap(size, top_pad) size_t size, top_pad; -#endif -{ - size_t page_mask = malloc_getpagesize - 1; - char *p1, *p2; - unsigned long ul; - heap_info *h; - - if(size+top_pad < HEAP_MIN_SIZE) - size = HEAP_MIN_SIZE; - else if(size+top_pad <= HEAP_MAX_SIZE) - size += top_pad; - else if(size > HEAP_MAX_SIZE) - return 0; - else - size = HEAP_MAX_SIZE; - size = (size + page_mask) & ~page_mask; - - /* A memory region aligned to a multiple of HEAP_MAX_SIZE is needed. - No swap space needs to be reserved for the following large - mapping (on Linux, this is the case for all non-writable mappings - anyway). */ - p1 = (char *)MMAP(0, HEAP_MAX_SIZE<<1, PROT_NONE, MAP_PRIVATE|MAP_NORESERVE); - if(p1 != MAP_FAILED) { - p2 = (char *)(((unsigned long)p1 + (HEAP_MAX_SIZE-1)) & ~(HEAP_MAX_SIZE-1)); - ul = p2 - p1; - munmap(p1, ul); - munmap(p2 + HEAP_MAX_SIZE, HEAP_MAX_SIZE - ul); - } else { - /* Try to take the chance that an allocation of only HEAP_MAX_SIZE - is already aligned. */ - p2 = (char *)MMAP(0, HEAP_MAX_SIZE, PROT_NONE, MAP_PRIVATE|MAP_NORESERVE); - if(p2 == MAP_FAILED) - return 0; - if((unsigned long)p2 & (HEAP_MAX_SIZE-1)) { - munmap(p2, HEAP_MAX_SIZE); - return 0; - } - } - if(mprotect(p2, size, PROT_READ|PROT_WRITE) != 0) { - munmap(p2, HEAP_MAX_SIZE); - return 0; - } - h = (heap_info *)p2; - h->size = size; - THREAD_STAT(stat_n_heaps++); - return h; -} - -/* Grow or shrink a heap. size is automatically rounded up to a - multiple of the page size if it is positive. */ - -static int -#if __STD_C -grow_heap(heap_info *h, long diff) -#else -grow_heap(h, diff) heap_info *h; long diff; -#endif -{ - size_t page_mask = malloc_getpagesize - 1; - long new_size; - - if(diff >= 0) { - diff = (diff + page_mask) & ~page_mask; - new_size = (long)h->size + diff; - if(new_size > HEAP_MAX_SIZE) - return -1; - if(mprotect((char *)h + h->size, diff, PROT_READ|PROT_WRITE) != 0) - return -2; - } else { - new_size = (long)h->size + diff; - if(new_size < (long)sizeof(*h)) - return -1; - - if(mprotect((char *)h + new_size, -diff, PROT_NONE) != 0) - return -2; - /*fprintf(stderr, "shrink %p %08lx\n", h, new_size);*/ - } - h->size = new_size; - return 0; -} - -/* Delete a heap. */ - -#define delete_heap(heap) munmap((char*)(heap), HEAP_MAX_SIZE) - -static int -internal_function -#if __STD_C -heap_trim(heap_info *heap, size_t pad) -#else -heap_trim(heap, pad) heap_info *heap; size_t pad; -#endif -{ - mstate ar_ptr = heap->ar_ptr; - unsigned long pagesz = mp_.pagesize; - mchunkptr top_chunk = top(ar_ptr), p, bck, fwd; - heap_info *prev_heap; - long new_size, top_size, extra; - - /* Can this heap go away completely? */ - while(top_chunk == chunk_at_offset(heap, sizeof(*heap))) { - prev_heap = heap->prev; - p = chunk_at_offset(prev_heap, prev_heap->size - (MINSIZE-2*SIZE_SZ)); - assert(p->size == (0|PREV_INUSE)); /* must be fencepost */ - p = prev_chunk(p); - new_size = chunksize(p) + (MINSIZE-2*SIZE_SZ); - assert(new_size>0 && new_size<(long)(2*MINSIZE)); - if(!prev_inuse(p)) - new_size += p->prev_size; - assert(new_size>0 && new_sizesize) < pad + MINSIZE + pagesz) - break; - ar_ptr->system_mem -= heap->size; - arena_mem -= heap->size; - delete_heap(heap); - heap = prev_heap; - if(!prev_inuse(p)) { /* consolidate backward */ - p = prev_chunk(p); - unlink(p, bck, fwd); - } - assert(((unsigned long)((char*)p + new_size) & (pagesz-1)) == 0); - assert( ((char*)p + new_size) == ((char*)heap + heap->size) ); - top(ar_ptr) = top_chunk = p; - set_head(top_chunk, new_size | PREV_INUSE); - /*check_chunk(ar_ptr, top_chunk);*/ - } - top_size = chunksize(top_chunk); - extra = ((top_size - pad - MINSIZE + (pagesz-1))/pagesz - 1) * pagesz; - if(extra < (long)pagesz) - return 0; - /* Try to shrink. */ - if(grow_heap(heap, -extra) != 0) - return 0; - ar_ptr->system_mem -= extra; - arena_mem -= extra; - - /* Success. Adjust top accordingly. */ - set_head(top_chunk, (top_size - extra) | PREV_INUSE); - /*check_chunk(ar_ptr, top_chunk);*/ - return 1; -} - -static mstate -internal_function -#if __STD_C -arena_get2(mstate a_tsd, size_t size) -#else -arena_get2(a_tsd, size) mstate a_tsd; size_t size; -#endif -{ - mstate a; - int err; - - if(!a_tsd) - a = a_tsd = &main_arena; - else { - a = a_tsd->next; - if(!a) { - /* This can only happen while initializing the new arena. */ - (void)mutex_lock(&main_arena.mutex); - THREAD_STAT(++(main_arena.stat_lock_wait)); - return &main_arena; - } - } - - /* Check the global, circularly linked list for available arenas. */ - repeat: - do { - if(!mutex_trylock(&a->mutex)) { - THREAD_STAT(++(a->stat_lock_loop)); - tsd_setspecific(arena_key, (Void_t *)a); - return a; - } - a = a->next; - } while(a != a_tsd); - - /* If not even the list_lock can be obtained, try again. This can - happen during `atfork', or for example on systems where thread - creation makes it temporarily impossible to obtain _any_ - locks. */ - if(mutex_trylock(&list_lock)) { - a = a_tsd; - goto repeat; - } - (void)mutex_unlock(&list_lock); - - /* Nothing immediately available, so generate a new arena. */ - a = _int_new_arena(size); - if(!a) - return 0; - - tsd_setspecific(arena_key, (Void_t *)a); - mutex_init(&a->mutex); - err = mutex_lock(&a->mutex); /* remember result */ - - /* Add the new arena to the global list. */ - (void)mutex_lock(&list_lock); - a->next = main_arena.next; -/* OMPI: use our barriers - atomic_write_barrier (); -*/ - opal_atomic_wmb(); - main_arena.next = a; - (void)mutex_unlock(&list_lock); - - if(err) /* locking failed; keep arena for further attempts later */ - return 0; - - THREAD_STAT(++(a->stat_lock_loop)); - return a; -} - -/* Create a new arena with initial size "size". */ - -mstate -_int_new_arena(size_t size) -{ - mstate a; - heap_info *h; - char *ptr; - unsigned long misalign; - - h = new_heap(size + (sizeof(*h) + sizeof(*a) + MALLOC_ALIGNMENT), - mp_.top_pad); - if(!h) { - /* Maybe size is too large to fit in a single heap. So, just try - to create a minimally-sized arena and let _int_malloc() attempt - to deal with the large request via mmap_chunk(). */ - h = new_heap(sizeof(*h) + sizeof(*a) + MALLOC_ALIGNMENT, mp_.top_pad); - if(!h) - return 0; - } - a = h->ar_ptr = (mstate)(h+1); - malloc_init_state(a); - /*a->next = NULL;*/ - a->system_mem = a->max_system_mem = h->size; - arena_mem += h->size; -#ifdef NO_THREADS - if((unsigned long)(mp_.mmapped_mem + arena_mem + main_arena.system_mem) > - mp_.max_total_mem) - mp_.max_total_mem = mp_.mmapped_mem + arena_mem + main_arena.system_mem; -#endif - - /* Set up the top chunk, with proper alignment. */ - ptr = (char *)(a + 1); - misalign = (unsigned long)chunk2mem(ptr) & MALLOC_ALIGN_MASK; - if (misalign > 0) - ptr += MALLOC_ALIGNMENT - misalign; - top(a) = (mchunkptr)ptr; - set_head(top(a), (((char*)h + h->size) - ptr) | PREV_INUSE); - - return a; -} - -/* Obtain the arena number n. Needed in malloc_stats. */ - -mstate -_int_get_arena (int n) -{ - mstate a = &main_arena; - - while (n-- != 0) { - a = a->next; - if (a == &main_arena) - return 0; - } - return a; -} - -#endif /* USE_ARENAS */ - -/* - * Local variables: - * c-basic-offset: 2 - * End: - */ diff --git a/opal/mca/memory/linux/configure.m4 b/opal/mca/memory/linux/configure.m4 deleted file mode 100644 index e44786e1eb3..00000000000 --- a/opal/mca/memory/linux/configure.m4 +++ /dev/null @@ -1,236 +0,0 @@ -# -*- shell-script -*- -# -# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana -# University Research and Technology -# Corporation. All rights reserved. -# Copyright (c) 2004-2005 The University of Tennessee and The University -# of Tennessee Research Foundation. All rights -# reserved. -# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, -# University of Stuttgart. All rights reserved. -# Copyright (c) 2004-2005 The Regents of the University of California. -# All rights reserved. -# Copyright (c) 2008-2010 Cisco Systems, Inc. All rights reserved. -# Copyright (c) 2015 Research Organization for Information Science -# and Technology (RIST). All rights reserved. -# $COPYRIGHT$ -# -# Additional copyrights may follow -# -# $HEADER$ -# -AC_DEFUN([MCA_opal_memory_linux_PRIORITY], [40]) - -AC_DEFUN([MCA_opal_memory_linux_COMPILE_MODE], [ - AC_MSG_CHECKING([for MCA component $2:$3 compile mode]) - $4="static" - AC_MSG_RESULT([$$4]) -]) - - -# MCA_memory_linux_CONFIG(action-if-can-compile, -# [action-if-cant-compile]) -# ------------------------------------------------ -AC_DEFUN([MCA_opal_memory_linux_CONFIG],[ - AC_CONFIG_FILES([opal/mca/memory/linux/Makefile]) - - OPAL_VAR_SCOPE_PUSH([memory_linux_ptmalloc2_happy memory_linux_ummu_happy memory_linux_requested icc_major_ver icc_minor_ver memory_linux_mmap memory_linux_munmap memory_linux_LIBS_SAVE]) - - # Only allow this component to build on Linux-based systems - - AC_MSG_CHECKING([operating system]) - case $host in - *linux*) - AC_MSG_RESULT([$host -- supported]) - memory_linux_ptmalloc2_happy=yes - memory_linux_ummu_happy=yes - ;; - *) - AC_MSG_RESULT([$host -- unsupported]) - memory_linux_ptmalloc2_happy=no - memory_linux_ummu_happy=no - ;; - esac - - AS_IF([test "$with_memory_manager" = "linux"], - [memory_linux_ptmalloc2_happy=yes - memory_linux_ummu_happy=yes - memory_linux_requested=1], - [memory_linux_requested=0 - AS_IF([test -z "$with_memory_manager" || test "$with_memory_manager" = "yes"], - [memory_linux_ptmalloc2_happy=yes - memory_linux_ummu_happy=yes], - [memory_linux_ptmalloc2_happy=no - memory_linux_ummu_happy=no])]) - - - ###################################################################### - # if memory hook available - ###################################################################### - memory_hook_found=1 - AS_IF([test "$memory_hook_found" -eq 1], - [memory_hook_found=0 AC_CHECK_HEADER([malloc.h], - [AC_CHECK_FUNC([__malloc_initialize_hook], - [AC_CHECK_FUNC([__malloc_hook], - [AC_CHECK_FUNC([__realloc_hook], - [AC_CHECK_FUNC([__free_hook], - [memory_hook_found=1])])])])])]) - AC_MSG_CHECKING([whether the system can use malloc hooks]) - AS_IF([test "$memory_hook_found" = "0"], - [AC_MSG_RESULT([no])], - [AC_MSG_RESULT([yes])]) - AC_DEFINE_UNQUOTED([MEMORY_LINUX_HAVE_MALLOC_HOOK_SUPPORT], [$memory_hook_found], - [Whether the system has Memory Allocation Hooks]) - - AC_ARG_ENABLE(memory-linux-malloc-alignment, - AC_HELP_STRING([--enable-memory-linux-malloc-alignment], [Enable support for allocated memory alignment. Default: enabled if supported, disabled otherwise.])) - - malloc_align_enabled=0 - AS_IF([test "$enable_memory_linux_malloc_alignment" != "no"], - [malloc_align_enabled=$memory_hook_found]) - - AS_IF([test "$enable_memory_linux_malloc_alignment" = "yes" && test "$malloc_align_enabled" = "0"], - [AC_MSG_ERROR([memory linux malloc alignment is requested but __malloc_hook is not available])]) - AC_MSG_CHECKING([whether the memory linux will use malloc alignment]) - AS_IF([test "$malloc_align_enabled" = "0"], - [AC_MSG_RESULT([no])], - [AC_MSG_RESULT([yes])]) - - AC_DEFINE_UNQUOTED(MEMORY_LINUX_MALLOC_ALIGN_ENABLED, [$malloc_align_enabled], - [Whether the memory linux malloc alignment is enabled]) - - ###################################################################### - # ptmalloc2 - ###################################################################### - - # Per ticket #227, Intel 9.0 v20051201 on ia64 with optimization - # of -O2 or higher will bork linux in strange in mysterious ways. - # Doh! So if the compiler vendor is intel and we're on an ia64 - # box, run "icc --version" and snarf the version string. If it's - # 9.0 and the version is <= 20051201, then disable ptmalloc2. - # Executive decision: ignore optimization levels (even though -O1 - # and -O0 seem to work). The upgrade to 9.1 is free, so that's a - # better path than trying to make a much more complicated test - # here. - - AS_IF([test "$memory_linux_ptmalloc2_happy" = yes], - [case $host in - ia64-*) - AS_IF([test "$opal_c_vendor" = "intel"], - [# check for v9.0 <= 20051201 - icc_major_ver="`$CC --version | head -n 1 | awk '{ print [$]3 }'`" - icc_minor_ver="`$CC --version | head -n 1 | awk '{ print [$]4 }'`" - AS_IF([test "$icc_major_ver" = "9.0" && test "`expr $icc_minor_ver \<= 20051201`" = "1"], - [memory_linux_ptmalloc2_happy=no - AC_MSG_WARN([*** Detected Intel C compiler v9.0 <= 20051201 on ia64]) - AC_MSG_WARN([*** This compiler/platform combination has known problems with ptmalloc2]) - AC_MSG_WARN([*** Disabling ptmalloc2])])]) - ;; - esac]) - - AS_IF([test "$memory_linux_ptmalloc2_happy" = yes], - [# check for malloc.h - AC_CHECK_HEADER([malloc.h], - [memory_linux_ptmalloc2_happy=yes], - [memory_linux_ptmalloc2_happy=no])]) - - AS_IF([test "$memory_linux_ptmalloc2_happy" = yes], - [# check for init hook symbol - AC_CHECK_DECL([__malloc_initialize_hook], - [memory_linux_ptmalloc2_happy=yes], - [memory_linux_ptmalloc2_happy=no], - [AC_INCLUDES_DEFAULT - #include ])]) - - # - # See if we have sbrk prototyped - # - AS_IF([test "$memory_linux_ptmalloc2_happy" = yes], - [AC_CHECK_DECLS([sbrk])]) - - # - # Figure out how we're going to call mmap/munmap for real - # - AS_IF([test "$memory_linux_ptmalloc2_happy" = yes], - [memory_linux_mmap=0 - memory_linux_munmap=1 - - # it's nearly impossible to call mmap from syscall(), so - # only go this route if we can't get at munmap any other - # way. - AC_CHECK_HEADER([syscall.h], - [AC_CHECK_FUNCS([syscall], [], [memory_linux_munmap=0])]) - - # Always look for __munmap and __mmap - AC_CHECK_FUNCS([__munmap], [memory_linux_mmap=1]) - AC_CHECK_FUNCS([__mmap]) - - # only allow dlsym (and therefore add -ldl) if we - # really need to - AS_IF([test "$memory_linux_mmap" = "0"], - [memory_linux_LIBS_SAVE="$LIBS" - AC_CHECK_LIB([dl], - [dlsym], - [LIBS="$LIBS -ldl" - memory_linux_LIBS="-ldl" - memory_linux_mmap=1]) - AC_CHECK_FUNCS([dlsym]) - LIBS="$memory_linux_LIBS_SAVE"]) - - AS_IF([test "$memory_linux_mmap" = "0" && test "$memory_linux_munmap" = "0"], - [memory_linux_ptmalloc2_happy=no])]) - - # If all is good, save the extra libs for the wrapper - AS_IF([test "$memory_linux_ptmalloc2_happy" = yes], - [value=1], - [value=0]) - AC_DEFINE_UNQUOTED([MEMORY_LINUX_PTMALLOC2], [$value], - [Whether ptmalloc2 is supported on this system or not]) - AM_CONDITIONAL([MEMORY_LINUX_PTMALLOC2], - [test "$memory_linux_ptmalloc2_happy" = yes]) - - ###################################################################### - # ummunotify - ###################################################################### - - # Check for the relevant header - AS_IF([test "$memory_linux_ummu_happy" = yes], - [# check for linux/ummunotify.h - AC_CHECK_HEADER([linux/ummunotify.h], - [memory_linux_ummu_happy=yes], - [memory_linux_ummu_happy=no])]) - - # has the Linux declaration for ioctl - AC_CHECK_HEADERS([stropts.h]) - - # If all is good, set the header file that we want the rest of the - # code base to use - AS_IF([test "$memory_linux_ummu_happy" = yes], - [memory_base_include="linux/public.h" - value=1], - [value=0]) - AC_DEFINE_UNQUOTED([MEMORY_LINUX_UMMUNOTIFY], [$value], - [Whether ummunotify is supported on this system or not]) - AM_CONDITIONAL([MEMORY_LINUX_UMMUNOTIFY], - [test "$memory_linux_ummu_happy" = yes]) - - ###################################################################### - # post processing - ###################################################################### - - AS_IF([test "$memory_malloc_hooks_requested" = 1 && \ - test "$memory_linux_ptmalloc2_happy" = no && \ - test "$memory_linux_ummu_happy" = no], - [AC_MSG_ERROR([linux memory management requested but neither ptmalloc2 nor ummunotify are available. Aborting.])]) - AC_SUBST([memory_linux_LIBS]) - - AS_IF([test "$memory_linux_ptmalloc2_happy" = yes || \ - test "$memory_linux_ummu_happy" = yes], - [memory_base_found=1 - $1], - [memory_base_found=0 - memory_base_include= - $2]) - - OPAL_VAR_SCOPE_POP -]) diff --git a/opal/mca/memory/linux/help-opal-memory-linux.txt b/opal/mca/memory/linux/help-opal-memory-linux.txt deleted file mode 100644 index b3327c7356a..00000000000 --- a/opal/mca/memory/linux/help-opal-memory-linux.txt +++ /dev/null @@ -1,29 +0,0 @@ -# -*- text -*- -# -# Copyright (c) 2009 Sun Microsystems, Inc. All rights reserved. -# Copyright (c) 2010-2014 Cisco Systems, Inc. All rights reserved. -# $COPYRIGHT$ -# -# Additional copyrights may follow -# -# $HEADER$ -# -# This is the US/English help file for Open MPI's memory/linux component. -# -[ummunotify eaccess] -Open MPI was unable to open the UMMU notification device. This is -likely a permissions problem on the device itself. UMMU notification -support is therefore disabled in this process; an alternate memory -hook manager *may* be used instead (if available). - - Local host: %s - UMMU device: %s -# -[ummunotify open error] -Open MPI was unable to open the UMMU notification device. UMMU -notification support is therefore disabled in this process; an -alternate memory hook manager *may* be used instead (if available). - - Local host: %s - UMMU device: %s - Error: %s (%d) diff --git a/opal/mca/memory/linux/hooks.c b/opal/mca/memory/linux/hooks.c deleted file mode 100644 index db608869b28..00000000000 --- a/opal/mca/memory/linux/hooks.c +++ /dev/null @@ -1,899 +0,0 @@ -/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ -/* - * Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. - * Copyright (c) 2012-2013 Los Alamos National Security, LLC. All rights - * reserved. - * - * Additional copyrights may follow. - */ -/* Malloc implementation for multiple threads without lock contention. - Copyright (C) 2001, 2002, 2003 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Wolfram Gloger , 2001. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Library General Public License as - published by the Free Software Foundation; either version 2 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Library General Public License for more details. - - You should have received a copy of the GNU Library General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -/* $Id: hooks.c,v 1.12 2004/11/05 14:42:32 wg Exp $ */ - -#include "opal_config.h" - -#include "opal/mca/mca.h" -#include "opal/mca/memory/memory.h" -#include "opal/constants.h" -#include "opal/memoryhooks/memory.h" - -#include "opal/mca/memory/linux/memory_linux.h" - -#ifndef DEFAULT_CHECK_ACTION -#define DEFAULT_CHECK_ACTION 1 -#endif - -#ifdef HAVE_SYS_STAT_H -#include /* for stat */ -#endif /* HAVE_SYS_STAT_H */ - -/* Defined in memory_linux_component.c */ -extern bool opal_memory_linux_disable; - -/* What to do if the standard debugging hooks are in place and a - corrupt pointer is detected: do nothing (0), print an error message - (1), or call abort() (2). */ - -/* Hooks for debugging versions. The initial hooks just call the - initialization routine, then do the normal work. */ - -#if !(USE_STARTER & 2) - -/* OMPI change: these aren't used (avoid a compiler warning by if - 0'ing them out */ -#if 0 -static Void_t* -#if __STD_C -malloc_hook_ini(size_t sz, const __malloc_ptr_t caller) -#else -malloc_hook_ini(sz, caller) - size_t sz; const __malloc_ptr_t caller; -#endif -{ - __malloc_hook = NULL; - ptmalloc_init(); - return public_mALLOc(sz); -} -#endif - -/* OMPI change: these aren't used (avoid a compiler warning by if - 0'ing them out */ -#if 0 -static Void_t* -#if __STD_C -realloc_hook_ini(Void_t* ptr, size_t sz, const __malloc_ptr_t caller) -#else -realloc_hook_ini(ptr, sz, caller) - Void_t* ptr; size_t sz; const __malloc_ptr_t caller; -#endif -{ - __malloc_hook = NULL; - __realloc_hook = NULL; - ptmalloc_init(); - return public_rEALLOc(ptr, sz); -} -#endif - -/* OMPI change: these aren't used (avoid a compiler warning by if - 0'ing them out */ -#if 0 -static Void_t* -#if __STD_C -memalign_hook_ini(size_t alignment, size_t sz, const __malloc_ptr_t caller) -#else -memalign_hook_ini(alignment, sz, caller) - size_t alignment; size_t sz; const __malloc_ptr_t caller; -#endif -{ - __memalign_hook = NULL; - ptmalloc_init(); - return public_mEMALIGn(alignment, sz); -} -#endif - -#endif /* !(USE_STARTER & 2) */ - -static int check_action = DEFAULT_CHECK_ACTION; - -/* Whether we are using malloc checking. */ -static int using_malloc_checking; - -/* A flag that is set by malloc_set_state, to signal that malloc checking - must not be enabled on the request from the user (via the MALLOC_CHECK_ - environment variable). It is reset by __malloc_check_init to tell - malloc_set_state that the user has requested malloc checking. - - The purpose of this flag is to make sure that malloc checking is not - enabled when the heap to be restored was constructed without malloc - checking, and thus does not contain the required magic bytes. - Otherwise the heap would be corrupted by calls to free and realloc. If - it turns out that the heap was created with malloc checking and the - user has requested it malloc_set_state just calls __malloc_check_init - again to enable it. On the other hand, reusing such a heap without - further malloc checking is safe. */ -static int disallow_malloc_check; - -/* Activate a standard set of debugging hooks. */ -void -__malloc_check_init() -{ - if (disallow_malloc_check) { - disallow_malloc_check = 0; - return; - } - using_malloc_checking = 1; - __malloc_hook = malloc_check; - __free_hook = free_check; - __realloc_hook = realloc_check; - __memalign_hook = memalign_check; - if(check_action & 1) - fprintf(stderr, "malloc: using debugging hooks\n"); -} - -/* A simple, standard set of debugging hooks. Overhead is `only' one - byte per chunk; still this will catch most cases of double frees or - overruns. The goal here is to avoid obscure crashes due to invalid - usage, unlike in the MALLOC_DEBUG code. */ - -#define MAGICBYTE(p) ( ( ((size_t)p >> 3) ^ ((size_t)p >> 11)) & 0xFF ) - -/* Instrument a chunk with overrun detector byte(s) and convert it - into a user pointer with requested size sz. */ - -static Void_t* -internal_function -#if __STD_C -mem2mem_check(Void_t *ptr, size_t sz) -#else -mem2mem_check(ptr, sz) Void_t *ptr; size_t sz; -#endif -{ - mchunkptr p; - unsigned char* m_ptr = (unsigned char*)BOUNDED_N(ptr, sz); - size_t i; - - if (!ptr) - return ptr; - p = mem2chunk(ptr); - for(i = chunksize(p) - (chunk_is_mmapped(p) ? 2*SIZE_SZ+1 : SIZE_SZ+1); - i > sz; - i -= 0xFF) { - if(i-sz < 0x100) { - m_ptr[i] = (unsigned char)(i-sz); - break; - } - m_ptr[i] = 0xFF; - } - m_ptr[sz] = MAGICBYTE(p); - return (Void_t*)m_ptr; -} - -/* Convert a pointer to be free()d or realloc()ed to a valid chunk - pointer. If the provided pointer is not valid, return NULL. */ - -static mchunkptr -internal_function -#if __STD_C -mem2chunk_check(Void_t* mem) -#else -mem2chunk_check(mem) Void_t* mem; -#endif -{ - mchunkptr p; - INTERNAL_SIZE_T sz, c; - unsigned char magic; - - if(!aligned_OK(mem)) return NULL; - p = mem2chunk(mem); - if (!chunk_is_mmapped(p)) { - /* Must be a chunk in conventional heap memory. */ - int contig = contiguous(&main_arena); - sz = chunksize(p); - if((contig && - ((char*)p=(mp_.sbrk_base+main_arena.system_mem) )) || - szprev_size&MALLOC_ALIGN_MASK || - (contig && (char*)prev_chunk(p)size & PREV_INUSE) || - ( (((unsigned long)p - p->prev_size) & page_mask) != 0 ) || - ( (sz = chunksize(p)), ((p->prev_size + sz) & page_mask) != 0 ) ) - return NULL; - magic = MAGICBYTE(p); - for(sz -= 1; (c = ((unsigned char*)p)[sz]) != magic; sz -= c) { - if(c<=0 || sz<(c+2*SIZE_SZ)) return NULL; - } - ((unsigned char*)p)[sz] ^= 0xFF; - } - return p; -} - -/* Check for corruption of the top chunk, and try to recover if - necessary. */ - -static int -internal_function -#if __STD_C -top_check(void) -#else -top_check() -#endif -{ - mchunkptr t = top(&main_arena); - char* brk, * new_brk; - INTERNAL_SIZE_T front_misalign, sbrk_size; - unsigned long pagesz = malloc_getpagesize; - - if (t == initial_top(&main_arena) || - (!chunk_is_mmapped(t) && - chunksize(t)>=MINSIZE && - prev_inuse(t) && - (!contiguous(&main_arena) || - (char*)t + chunksize(t) == mp_.sbrk_base + main_arena.system_mem))) - return 0; - - if(check_action & 1) - fprintf(stderr, "malloc: top chunk is corrupt\n"); - if(check_action & 2) - abort(); - - /* Try to set up a new top chunk. */ - brk = MORECORE(0); - front_misalign = (unsigned long)chunk2mem(brk) & MALLOC_ALIGN_MASK; - if (front_misalign > 0) - front_misalign = MALLOC_ALIGNMENT - front_misalign; - sbrk_size = front_misalign + mp_.top_pad + MINSIZE; - sbrk_size += pagesz - ((unsigned long)(brk + sbrk_size) & (pagesz - 1)); - new_brk = (char*)(MORECORE (sbrk_size)); - if (new_brk == (char*)(MORECORE_FAILURE)) return -1; - /* Call the `morecore' hook if necessary. */ - if (__after_morecore_hook) - (*__after_morecore_hook) (); - main_arena.system_mem = (new_brk - mp_.sbrk_base) + sbrk_size; - - top(&main_arena) = (mchunkptr)(brk + front_misalign); - set_head(top(&main_arena), (sbrk_size - front_misalign) | PREV_INUSE); - - return 0; -} - -static Void_t* -#if __STD_C -malloc_check(size_t sz, const Void_t *caller) -#else -malloc_check(sz, caller) size_t sz; const Void_t *caller; -#endif -{ - Void_t *victim; - - (void)mutex_lock(&main_arena.mutex); - victim = (top_check() >= 0) ? _int_malloc(&main_arena, sz+1) : NULL; - (void)mutex_unlock(&main_arena.mutex); - return mem2mem_check(victim, sz); -} - -static void -#if __STD_C -free_check(Void_t* mem, const Void_t *caller) -#else -free_check(mem, caller) Void_t* mem; const Void_t *caller; -#endif -{ - mchunkptr p; - - if(!mem) return; - (void)mutex_lock(&main_arena.mutex); - p = mem2chunk_check(mem); - if(!p) { - (void)mutex_unlock(&main_arena.mutex); - if(check_action & 1) - fprintf(stderr, "free(): invalid pointer %p!\n", mem); - if(check_action & 2) - abort(); - return; - } -#if HAVE_MMAP - if (chunk_is_mmapped(p)) { - (void)mutex_unlock(&main_arena.mutex); - munmap_chunk(p); - return; - } -#endif -#if 0 /* Erase freed memory. */ - memset(mem, 0, chunksize(p) - (SIZE_SZ+1)); -#endif - _int_free(&main_arena, mem); - (void)mutex_unlock(&main_arena.mutex); -} - -static Void_t* -#if __STD_C -realloc_check(Void_t* oldmem, size_t bytes, const Void_t *caller) -#else -realloc_check(oldmem, bytes, caller) - Void_t* oldmem; size_t bytes; const Void_t *caller; -#endif -{ - mchunkptr oldp; - INTERNAL_SIZE_T nb, oldsize; - Void_t* newmem = 0; - - if (oldmem == 0) return malloc_check(bytes, NULL); - (void)mutex_lock(&main_arena.mutex); - oldp = mem2chunk_check(oldmem); - (void)mutex_unlock(&main_arena.mutex); - if(!oldp) { - if(check_action & 1) - fprintf(stderr, "realloc(): invalid pointer %p!\n", oldmem); - if(check_action & 2) - abort(); - return malloc_check(bytes, NULL); - } - oldsize = chunksize(oldp); - - checked_request2size(bytes+1, nb); - (void)mutex_lock(&main_arena.mutex); - -#if HAVE_MMAP - if (chunk_is_mmapped(oldp)) { -#if HAVE_MREMAP - mchunkptr newp = mremap_chunk(oldp, nb); - if(newp) - newmem = chunk2mem(newp); - else -#endif - { - /* Note the extra SIZE_SZ overhead. */ - if(oldsize - SIZE_SZ >= nb) - newmem = oldmem; /* do nothing */ - else { - /* Must alloc, copy, free. */ - if (top_check() >= 0) - newmem = _int_malloc(&main_arena, bytes+1); - if (newmem) { - MALLOC_COPY(BOUNDED_N(newmem, bytes+1), oldmem, oldsize - 2*SIZE_SZ); - munmap_chunk(oldp); - } - } - } - } else { -#endif /* HAVE_MMAP */ - if (top_check() >= 0) - newmem = _int_realloc(&main_arena, oldmem, bytes+1); -#if 0 /* Erase freed memory. */ - if(newmem) - newp = mem2chunk(newmem); - nb = chunksize(newp); - if(oldp=chunk_at_offset(newp, nb)) { - memset((char*)oldmem + 2*sizeof(mbinptr), 0, - oldsize - (2*sizeof(mbinptr)+2*SIZE_SZ+1)); - } else if(nb > oldsize+SIZE_SZ) { - memset((char*)BOUNDED_N(chunk2mem(newp), bytes) + oldsize, - 0, nb - (oldsize+SIZE_SZ)); - } -#endif -#if HAVE_MMAP - } -#endif - (void)mutex_unlock(&main_arena.mutex); - - return mem2mem_check(newmem, bytes); -} - -static Void_t* -#if __STD_C -memalign_check(size_t alignment, size_t bytes, const Void_t *caller) -#else -memalign_check(alignment, bytes, caller) - size_t alignment; size_t bytes; const Void_t *caller; -#endif -{ - INTERNAL_SIZE_T nb; - Void_t* mem; - - if (alignment <= MALLOC_ALIGNMENT) return malloc_check(bytes, NULL); - if (alignment < MINSIZE) alignment = MINSIZE; - - checked_request2size(bytes+1, nb); - (void)mutex_lock(&main_arena.mutex); - mem = (top_check() >= 0) ? _int_memalign(&main_arena, alignment, bytes+1) : - NULL; - (void)mutex_unlock(&main_arena.mutex); - return mem2mem_check(mem, bytes); -} - -#if !defined NO_THREADS && USE_STARTER - -/* The following hooks are used when the global initialization in - ptmalloc_init() hasn't completed yet. */ - -static Void_t* -#if __STD_C -malloc_starter(size_t sz, const Void_t *caller) -#else -malloc_starter(sz, caller) size_t sz; const Void_t *caller; -#endif -{ - Void_t* victim; - - ptmalloc_init_minimal(); - victim = _int_malloc(&main_arena, sz); - - return victim ? BOUNDED_N(victim, sz) : 0; -} - -static Void_t* -#if __STD_C -memalign_starter(size_t align, size_t sz, const Void_t *caller) -#else -memalign_starter(align, sz, caller) size_t align, sz; const Void_t *caller; -#endif -{ - Void_t* victim; - - ptmalloc_init_minimal(); - victim = _int_memalign(&main_arena, align, sz); - - return victim ? BOUNDED_N(victim, sz) : 0; -} - -static void -#if __STD_C -free_starter(Void_t* mem, const Void_t *caller) -#else -free_starter(mem, caller) Void_t* mem; const Void_t *caller; -#endif -{ - mchunkptr p; - - if(!mem) return; - p = mem2chunk(mem); -#if HAVE_MMAP - if (chunk_is_mmapped(p)) { - munmap_chunk(p); - return; - } -#endif - _int_free(&main_arena, mem); -} - -#endif /* !defined NO_THREADS && USE_STARTER */ - - -/* Get/set state: malloc_get_state() records the current state of all - malloc variables (_except_ for the actual heap contents and `hook' - function pointers) in a system dependent, opaque data structure. - This data structure is dynamically allocated and can be free()d - after use. malloc_set_state() restores the state of all malloc - variables to the previously obtained state. This is especially - useful when using this malloc as part of a shared library, and when - the heap contents are saved/restored via some other method. The - primary example for this is GNU Emacs with its `dumping' procedure. - `Hook' function pointers are never saved or restored by these - functions, with two exceptions: If malloc checking was in use when - malloc_get_state() was called, then malloc_set_state() calls - __malloc_check_init() if possible; if malloc checking was not in - use in the recorded state but the user requested malloc checking, - then the hooks are reset to 0. */ - -#define MALLOC_STATE_MAGIC 0x444c4541l -#define MALLOC_STATE_VERSION (0*0x100l + 2l) /* major*0x100 + minor */ - -struct malloc_save_state { - long magic; - long version; - mbinptr av[NBINS * 2 + 2]; - char* sbrk_base; - int sbrked_mem_bytes; - unsigned long trim_threshold; - unsigned long top_pad; - unsigned int n_mmaps_max; - unsigned long mmap_threshold; - int check_action; - unsigned long max_sbrked_mem; - unsigned long max_total_mem; - unsigned int n_mmaps; - unsigned int max_n_mmaps; - unsigned long mmapped_mem; - unsigned long max_mmapped_mem; - int using_malloc_checking; -}; - -Void_t* -public_gET_STATe(void) -{ - struct malloc_save_state* ms; - int i; - mbinptr b; - - ms = (struct malloc_save_state*)public_mALLOc(sizeof(*ms)); - if (!ms) - return 0; - (void)mutex_lock(&main_arena.mutex); - malloc_consolidate(&main_arena); - ms->magic = MALLOC_STATE_MAGIC; - ms->version = MALLOC_STATE_VERSION; - ms->av[0] = 0; - ms->av[1] = 0; /* used to be binblocks, now no longer used */ - ms->av[2] = top(&main_arena); - ms->av[3] = 0; /* used to be undefined */ - for(i=1; iav[2*i+2] = ms->av[2*i+3] = 0; /* empty bin */ - else { - ms->av[2*i+2] = first(b); - ms->av[2*i+3] = last(b); - } - } - ms->sbrk_base = mp_.sbrk_base; - ms->sbrked_mem_bytes = main_arena.system_mem; - ms->trim_threshold = mp_.trim_threshold; - ms->top_pad = mp_.top_pad; - ms->n_mmaps_max = mp_.n_mmaps_max; - ms->mmap_threshold = mp_.mmap_threshold; - ms->check_action = check_action; - ms->max_sbrked_mem = main_arena.max_system_mem; -#ifdef NO_THREADS - ms->max_total_mem = mp_.max_total_mem; -#else - ms->max_total_mem = 0; -#endif - ms->n_mmaps = mp_.n_mmaps; - ms->max_n_mmaps = mp_.max_n_mmaps; - ms->mmapped_mem = mp_.mmapped_mem; - ms->max_mmapped_mem = mp_.max_mmapped_mem; - ms->using_malloc_checking = using_malloc_checking; - (void)mutex_unlock(&main_arena.mutex); - return (Void_t*)ms; -} - -int -public_sET_STATe(Void_t* msptr) -{ - struct malloc_save_state* ms = (struct malloc_save_state*)msptr; - int i; - mbinptr b; - - disallow_malloc_check = 1; - ptmalloc_init(); - if(ms->magic != MALLOC_STATE_MAGIC) return -1; - /* Must fail if the major version is too high. */ - if((ms->version & ~0xffl) > (MALLOC_STATE_VERSION & ~0xffl)) return -2; - (void)mutex_lock(&main_arena.mutex); - /* There are no fastchunks. */ - clear_fastchunks(&main_arena); - set_max_fast(&main_arena, DEFAULT_MXFAST); - for (i=0; i<(int)NFASTBINS; ++i) - main_arena.fastbins[i] = 0; - for (i=0; i<(int)BINMAPSIZE; ++i) - main_arena.binmap[i] = 0; - top(&main_arena) = ms->av[2]; - main_arena.last_remainder = 0; - for(i=1; iav[2*i+2] == 0) { - assert(ms->av[2*i+3] == 0); - first(b) = last(b) = b; - } else { - if(i<(int)NSMALLBINS || ((int)largebin_index(chunksize(ms->av[2*i+2]))==i && - (int)largebin_index(chunksize(ms->av[2*i+3]))==i)) { - first(b) = ms->av[2*i+2]; - last(b) = ms->av[2*i+3]; - /* Make sure the links to the bins within the heap are correct. */ - first(b)->bk = b; - last(b)->fd = b; - /* Set bit in binblocks. */ - mark_bin(&main_arena, i); - } else { - /* Oops, index computation from chunksize must have changed. - Link the whole list into unsorted_chunks. */ - first(b) = last(b) = b; - b = unsorted_chunks(&main_arena); - ms->av[2*i+2]->bk = b; - ms->av[2*i+3]->fd = b->fd; - b->fd->bk = ms->av[2*i+3]; - b->fd = ms->av[2*i+2]; - } - } - } - mp_.sbrk_base = ms->sbrk_base; - main_arena.system_mem = ms->sbrked_mem_bytes; - mp_.trim_threshold = ms->trim_threshold; - mp_.top_pad = ms->top_pad; - mp_.n_mmaps_max = ms->n_mmaps_max; - mp_.mmap_threshold = ms->mmap_threshold; - check_action = ms->check_action; - main_arena.max_system_mem = ms->max_sbrked_mem; -#ifdef NO_THREADS - mp_.max_total_mem = ms->max_total_mem; -#endif - mp_.n_mmaps = ms->n_mmaps; - mp_.max_n_mmaps = ms->max_n_mmaps; - mp_.mmapped_mem = ms->mmapped_mem; - mp_.max_mmapped_mem = ms->max_mmapped_mem; - /* add version-dependent code here */ - if (ms->version >= 1) { - /* Check whether it is safe to enable malloc checking, or whether - it is necessary to disable it. */ - if (ms->using_malloc_checking && !using_malloc_checking && - !disallow_malloc_check) - __malloc_check_init (); - else if (!ms->using_malloc_checking && using_malloc_checking) { - __malloc_hook = 0; - __free_hook = 0; - __realloc_hook = 0; - __memalign_hook = 0; - using_malloc_checking = 0; - } - } - check_malloc_state(&main_arena); - - (void)mutex_unlock(&main_arena.mutex); - return 0; -} - - -/*------------------------------------------------------------------------- - OMPI change: Per - http://www.gnu.org/software/libc/manual/html_mono/libc.html#Hooks-for-Malloc, - we can define the __malloc_initialize_hook variable to be a - function that is invoked before the first allocation is ever - performed. We use this hook to wholly replace the underlying - allocator to our own allocator if a few conditions are met. - - Remember that this hook is called probably at the very very very - beginning of the process. MCA parameters haven't been setup yet -- - darn near nothing has been setup yet. Indeed, we're effectively in - signal context because we can't call anything that calls malloc. - So we can basically have some hard-coded tests for things to see if - we want to setup to use our internal ptmalloc2 or not. */ - -static void *opal_memory_linux_malloc_hook(size_t sz, - const __malloc_ptr_t caller) -{ - return public_mALLOc(sz); -} - -static void *opal_memory_linux_realloc_hook(Void_t* ptr, size_t sz, - const __malloc_ptr_t caller) -{ - return public_rEALLOc(ptr, sz); -} - -static void *opal_memory_linux_memalign_hook(size_t alignment, size_t sz, - const __malloc_ptr_t caller) -{ - return public_mEMALIGn(alignment, sz); -} - -static void opal_memory_linux_free_hook(__malloc_ptr_t __ptr, - const __malloc_ptr_t caller) -{ - public_fREe(__ptr); -} - -typedef enum { - RESULT_NO, - RESULT_YES, - RESULT_RUNTIME, - RESULT_NOT_FOUND -} check_result_t; - -static check_result_t check(const char *name) -{ - char *s = getenv(name); - if (NULL == s) { - return RESULT_NOT_FOUND; - } - - if ('0' == s[0] && '\0' == s[1]) { - /* A value of 0 means "don't use!" */ - return RESULT_NO; - } else if ('-' == s[0] && '1' == s[1] && '\0' == s[2]) { - /* A value of -1 means "use it if it would be advantageous */ - return RESULT_RUNTIME; - } else { - /* Any other value means "use the hooks, Luke!" */ - return RESULT_YES; - } -} - - -/* This function is called on loading libmpi in case system has Memory Allocation Hooks - * (see ompi/runtime/ompi_mpi_init.c for details) - */ -void opal_memory_linux_malloc_init_hook(void) -{ - check_result_t r1, lp, lpp; - bool want_rcache = false, found_driver = false; - - if (!opal_memory_linux_opened) { - return; - } - - /* First, check for a FAKEROOT environment. If we're in a - fakeroot, then access() (and likely others) have been replaced - and are not safe to call here in this pre-main environment. So - check for the environment markers that we're in a FAKEROOT. - And if so, return immediately. - - Note that this check was inspired by a problem with Debian's - "fakeroot" build environment that allocates memory during - stat() (see http://bugs.debian.org/531522). It may not be - necessary any more since we're using access(), not stat(). But - we'll leave the check, anyway. - - This is also an issue when using Gentoo's version of - 'fakeroot', sandbox v2.5. Sandbox environments can also be - detected fairly easily by looking for SANDBOX_ON. */ - if (getenv("FAKEROOTKEY") != NULL || - getenv("FAKED_MODE") != NULL || - getenv("SANDBOX_ON") != NULL ) { - return; - } - -#if MEMORY_LINUX_UMMUNOTIFY - /* Next, check if ummunotify is present on the system. If it is, - and if we were compile with ummunotify support, then we don't - need to do the following ptmalloc2 hacks. open/mmap on the - device may fail during init, but if /dev/ummunotify exists, we - assume that the user/administrator *wants* to use - ummunotify. */ - if (access("/dev/ummunotify", F_OK) == 0) { - return; - } -#endif - - /* Yes, checking for an MPI MCA parameter here is an abstraction - violation. Cope. Yes, even checking for *any* MCA parameter - here (without going through the MCA param API) is an - abstraction violation. Fricken' cope, will ya? - (unfortunately, there's really no good way to do this other - than this abstraction violation :-( ) */ - lp = check("OPAL_MCA_leave_pinned"); - if( RESULT_NOT_FOUND == lp ) lp = check(OPAL_MCA_PREFIX"mpi_leave_pinned"); - lpp = check("OPAL_MCA_leave_pinned_pipeline"); - if( RESULT_NOT_FOUND == lpp ) lpp = check(OPAL_MCA_PREFIX"mpi_leave_pinned_pipeline"); - - /* See if we want to disable this component. */ - r1 = check(OPAL_MCA_PREFIX"memory_linux_disable"); - if (RESULT_NOT_FOUND != r1 && RESULT_NO != r1) { - return; - } - - /* Look for sentinel files (directories) to see if various network - drivers are loaded (yes, I know, further abstraction - violations...). - - * All OpenFabrics devices will have files in - /sys/class/infiniband (even iWARP) - * Open-MX doesn't currently use a reg cache, but it might - someday. So be conservative and check for /dev/open-mx. - * MX will have one or more of /dev/myri[0-9]. Yuck. - */ - if (0 == access("/sys/class/infiniband", F_OK) || - 0 == access("/dev/open-mx", F_OK) || - 0 == access("/dev/myri0", F_OK) || - 0 == access("/dev/myri1", F_OK) || - 0 == access("/dev/myri2", F_OK) || - 0 == access("/dev/myri3", F_OK) || - 0 == access("/dev/myri4", F_OK) || - 0 == access("/dev/myri5", F_OK) || - 0 == access("/dev/myri6", F_OK) || - 0 == access("/dev/myri7", F_OK) || - 0 == access("/dev/myri8", F_OK) || - 0 == access("/dev/myri9", F_OK) || - 0 == access("/dev/ipath", F_OK) || - 0 == access("/dev/kgni0", F_OK) || - 0 == access("/dev/mic/scif", F_OK) || - 0 == access("/dev/scif", F_OK)) { - found_driver = true; - } - - /* Simple combination of the results of these two environment - variables (if both "yes" and "no" are specified, then be - conservative and assume "yes"): - - lp / lpp yes no runtime not found - yes yes yes yes yes - no yes no no no - runtime yes no runtime runtime - not found yes no runtime runtime - */ - if (RESULT_YES == lp || RESULT_YES == lpp) { - want_rcache = true; - } else if (RESULT_NO == lp || RESULT_NO == lpp) { - want_rcache = false; - } else { - want_rcache = found_driver; - } - - if (want_rcache) { - /* Initialize ptmalloc */ - ptmalloc_init(); - - /* Now set the hooks to point to our functions */ - __free_hook = opal_memory_linux_free_hook; - __malloc_hook = opal_memory_linux_malloc_hook; - __memalign_hook = opal_memory_linux_memalign_hook; - __realloc_hook = opal_memory_linux_realloc_hook; - } -} - - -/* OMPI change: prototype the function below, otherwise we'll get - warnings about it not being declared (at least in developer/debug - builds). This function is not DECLSPEC'ed because we don't want it - visible outside of this component (i.e., libopen-pal, since this - component is never built as a DSO; it's always slurped into - libopen-pal). This declaration is not in malloc.h because this - function only exists as a horrid workaround to force linkers to - pull in this .o file (see explanation below). */ -void opal_memory_linux_hook_pull(bool *want_hooks); - -/* OMPI change: add a dummy function here that will be called by the - linux component open() function. This dummy function is - necessary for when OMPI is built as --disable-shared - --enable-static --disable-dlopen, because we won't use - -Wl,--export-dynamic when building OMPI. So we need to ensure that - not only that all the symbols in this file end up in libopen-pal.a, - but they also end up in the final exectuable (so that - __malloc_initialize_hook is there, overrides the weak symbol in - glibc, ....etc.). */ -void opal_memory_linux_hook_pull(bool *want_hooks) -{ - /* Make this slightly less than a dummy function -- register the - MCA parameter here (that way we keep the name of this MCA - parameter here within this one, single file). Register solely - so that it shows up in ompi_info -- by the time we register it, - the _malloc_init_hook() has almost certainly already fired, so - whatever value was set via normal MCA mechanisms likely won't - be see if it wasn't already see by the getenv() in the - _malloc_init_hook(). */ - *want_hooks = !opal_memory_linux_disable; -} - - - -/* - * Local variables: - * c-basic-offset: 4 - * End: - */ diff --git a/opal/mca/memory/linux/lran2.h b/opal/mca/memory/linux/lran2.h deleted file mode 100644 index cea9920282a..00000000000 --- a/opal/mca/memory/linux/lran2.h +++ /dev/null @@ -1,51 +0,0 @@ -/* lran2.h - * by Wolfram Gloger 1996. - * - * A small, portable pseudo-random number generator. - */ - -#ifndef _LRAN2_H -#define _LRAN2_H - -#define LRAN2_MAX 714025l /* constants for portable */ -#define IA 1366l /* random number generator */ -#define IC 150889l /* (see e.g. `Numerical Recipes') */ - -struct lran2_st { - long x, y, v[97]; -}; - -static void -lran2_init(struct lran2_st* d, long seed) -{ - long x; - int j; - - x = (IC - seed) % LRAN2_MAX; - if(x < 0) x = -x; - for(j=0; j<97; j++) { - x = (IA*x + IC) % LRAN2_MAX; - d->v[j] = x; - } - d->x = (IA*x + IC) % LRAN2_MAX; - d->y = d->x; -} - -#ifdef __GNUC__ -__inline__ -#endif -static long -lran2(struct lran2_st* d) -{ - int j = (d->y % 97); - - d->y = d->v[j]; - d->x = (IA*d->x + IC) % LRAN2_MAX; - d->v[j] = d->x; - return d->y; -} - -#undef IA -#undef IC - -#endif diff --git a/opal/mca/memory/linux/malloc-stats.c b/opal/mca/memory/linux/malloc-stats.c deleted file mode 100644 index cfb9fb23df4..00000000000 --- a/opal/mca/memory/linux/malloc-stats.c +++ /dev/null @@ -1,169 +0,0 @@ -/* - * Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. - * - * Additional copyrights may follow. - */ -/* Malloc implementation for multiple threads; statistics printing. - Copyright (C) 2004 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Wolfram Gloger , 2004. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library; if not, write to the Free - Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA - 02111-1307 USA. */ - -/* $Id: $ */ - -/* OMPI change: Name-shift all the internal symbols */ -#include "opal/mca/memory/linux/rename.h" - -#include /* needed for malloc_stats */ - -#include - -#include "malloc.h" - -/* - Define HAVE_MMAP as true to optionally make malloc() use mmap() to - allocate very large blocks. These will be returned to the - operating system immediately after a free(). Also, if mmap - is available, it is used as a backup strategy in cases where - MORECORE fails to provide space from system. - - This malloc is best tuned to work with mmap for large requests. - If you do not have mmap, operations involving very large chunks (1MB - or so) may be slower than you'd like. -*/ - -#ifndef HAVE_MMAP -#define HAVE_MMAP 1 -#endif - -#ifdef USE_DL_PREFIX - -#define public_mSTATs dlmalloc_stats - -#else /* USE_DL_PREFIX */ -#ifdef _LIBC - -#define public_mSTATs __malloc_stats - -#else /* !_LIBC */ - -#define public_mSTATs malloc_stats - -#endif /* _LIBC */ -#endif /* USE_DL_PREFIX */ - -/* - malloc_stats(); - Prints on stderr the amount of space obtained from the system (both - via sbrk and mmap), the maximum amount (which may be more than - current if malloc_trim and/or munmap got called), and the current - number of bytes allocated via malloc (or realloc, etc) but not yet - freed. Note that this is the number of bytes allocated, not the - number requested. It will be larger than the number requested - because of alignment and bookkeeping overhead. Because it includes - alignment wastage as being in use, this figure may be greater than - zero even when no user-level chunks are allocated. - - The reported current and maximum system memory can be inaccurate if - a program makes other calls to system memory allocation functions - (normally sbrk) outside of malloc. - - malloc_stats prints only the most commonly interesting statistics. - More information can be obtained by calling mallinfo. - -*/ -void public_mSTATs __MALLOC_P((void)); - -/* - ------------------------------ malloc_stats ------------------------------ -*/ - -void public_mSTATs() -{ - int i; - mstate ar_ptr; - struct malloc_global_info mgi; - struct malloc_arena_info mai; - unsigned long in_use_b, system_b, avail_b; -#if defined(THREAD_STATS) && THREAD_STATS - long stat_lock_direct = 0, stat_lock_loop = 0, stat_lock_wait = 0; -#endif - -#if 0 - if(__malloc_initialized < 0) - ptmalloc_init (); -#endif - _int_get_global_info(&mgi); - system_b = in_use_b = mgi.mmapped_mem; -#ifdef _LIBC - _IO_flockfile (stderr); - int old_flags2 = ((_IO_FILE *) stderr)->_flags2; - ((_IO_FILE *) stderr)->_flags2 |= _IO_FLAGS2_NOTCANCEL; -#endif - for (i=0; (ar_ptr = _int_get_arena(i)); i++) { - _int_get_arena_info(ar_ptr, &mai); - avail_b = mai.fastavail + mai.binavail + mai.top_size; - fprintf(stderr, "Arena %d:\n", i); - fprintf(stderr, "system bytes = %10lu\n", - (unsigned long)mai.system_mem); - fprintf(stderr, "in use bytes = %10lu\n", - (unsigned long)(mai.system_mem - avail_b)); -#if MALLOC_DEBUG > 1 - if (i > 0) - dump_heap(heap_for_ptr(top(ar_ptr))); -#endif - system_b += mai.system_mem; - in_use_b += mai.system_mem - avail_b; -#if defined(THREAD_STATS) && THREAD_STATS - stat_lock_direct += mai.stat_lock_direct; - stat_lock_loop += mai.stat_lock_loop; - stat_lock_wait += mai.stat_lock_wait; -#endif - } -#if HAVE_MMAP - fprintf(stderr, "Total (incl. mmap):\n"); -#else - fprintf(stderr, "Total:\n"); -#endif - fprintf(stderr, "system bytes = %10lu\n", system_b); - fprintf(stderr, "in use bytes = %10lu\n", in_use_b); -#ifdef NO_THREADS - fprintf(stderr, "max system bytes = %10lu\n", - (unsigned long)mgi.max_total_mem); -#endif -#if HAVE_MMAP - fprintf(stderr, "max mmap regions = %10u\n", (unsigned int)mgi.max_n_mmaps); - fprintf(stderr, "max mmap bytes = %10lu\n", - (unsigned long)mgi.max_mmapped_mem); -#endif -#if defined(THREAD_STATS) && THREAD_STATS - fprintf(stderr, "heaps created = %10d\n", mgi.stat_n_heaps); - fprintf(stderr, "locked directly = %10ld\n", stat_lock_direct); - fprintf(stderr, "locked in loop = %10ld\n", stat_lock_loop); - fprintf(stderr, "locked waiting = %10ld\n", stat_lock_wait); - fprintf(stderr, "locked total = %10ld\n", - stat_lock_direct + stat_lock_loop + stat_lock_wait); -#endif -#ifdef _LIBC - ((_IO_FILE *) stderr)->_flags2 |= old_flags2; - _IO_funlockfile (stderr); -#endif -} - -#ifdef _LIBC -weak_alias (__malloc_stats, malloc_stats) -#endif diff --git a/opal/mca/memory/linux/malloc.c b/opal/mca/memory/linux/malloc.c deleted file mode 100644 index 55ee1f09960..00000000000 --- a/opal/mca/memory/linux/malloc.c +++ /dev/null @@ -1,5599 +0,0 @@ -/********************** BEGIN OMPI CHANGES *****************************/ -/* - * Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. - * Copyright (c) 2015 Research Organization for Information Science - * and Technology (RIST). All rights reserved. - * - * Additional copyrights may follow. - */ - -#define OPAL_DISABLE_ENABLE_MEM_DEBUG 1 -#include "opal_config.h" - -#include -#include - -#include "opal/sys/atomic.h" -#include "opal/memoryhooks/memory_internal.h" -#include "opal/mca/memory/linux/memory_linux.h" -/* Name-shift all the internal symbols */ -#include "opal/mca/memory/linux/rename.h" - -/* - * Not all systems have sbrk() declared, since it's technically not a - * POSIX function. - */ -#if !HAVE_DECL_SBRK -void *sbrk(); -#endif - - -static void *opal_memory_linux_free_ptmalloc2_sbrk(int inc) -{ - if (inc < 0) { - long oldp = (long) sbrk(0); - opal_mem_hooks_release_hook((void*) (oldp + inc), -inc, 1); - } - - return sbrk(inc); -} - -/* if we are trying to catch only allocations from and releases to the - operating system, intercept sbrk, mmap, and munmap. If we want to - intercept every call to malloc/realloc/free/etc., don't do this, as - we need to add something into each of those calls anyway. */ -#define MORECORE opal_memory_linux_free_ptmalloc2_sbrk -#define munmap(a,b) opal_memory_linux_free_ptmalloc2_munmap(a,b,1) - -/* make some non-GCC compilers happy */ -#ifndef __GNUC__ -#define __const const -#endif - -/********************* END OMPI CHANGES ******************************/ - - - - -/* Malloc implementation for multiple threads without lock contention. - Copyright (C) 1996-2002, 2003, 2004 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Wolfram Gloger - and Doug Lea , 2001. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public License as - published by the Free Software Foundation; either version 2.1 of the - License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library; see the file COPYING.LIB. If not, - write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, - Boston, MA 02111-1307, USA. */ - -/* - This is a version (aka ptmalloc2) of malloc/free/realloc written by - Doug Lea and adapted to multiple threads/arenas by Wolfram Gloger. - -* Version ptmalloc2-20011215 - $Id: malloc.c,v 1.20 2004/11/04 17:31:04 wg Exp $ - based on: - VERSION 2.7.1pre1 Sat May 12 07:41:21 2001 Doug Lea (dl at gee) - - Note: There may be an updated version of this malloc obtainable at - http://www.malloc.de/malloc/ptmalloc2.tar.gz - Check before installing! - -* Quickstart - - In order to compile this implementation, a Makefile is provided with - the ptmalloc2 distribution, which has pre-defined targets for some - popular systems (e.g. "make posix" for Posix threads). All that is - typically required with regard to compiler flags is the selection of - an appropriate malloc-machine.h include file via -I directives. - Many/most systems will additionally require USE_TSD_DATA_HACK to be - defined, so this is the default for "make posix". - -* Why use this malloc? - - This is not the fastest, most space-conserving, most portable, or - most tunable malloc ever written. However it is among the fastest - while also being among the most space-conserving, portable and tunable. - Consistent balance across these factors results in a good general-purpose - allocator for malloc-intensive programs. - - The main properties of the algorithms are: - * For large (>= 512 bytes) requests, it is a pure best-fit allocator, - with ties normally decided via FIFO (i.e. least recently used). - * For small (<= 64 bytes by default) requests, it is a caching - allocator, that maintains pools of quickly recycled chunks. - * In between, and for combinations of large and small requests, it does - the best it can trying to meet both goals at once. - * For very large requests (>= 128KB by default), it relies on system - memory mapping facilities, if supported. - - For a longer but slightly out of date high-level description, see - http://gee.cs.oswego.edu/dl/html/malloc.html - - You may already by default be using a C library containing a malloc - that is based on some version of this malloc (for example in - linux). You might still want to use the one in this file in order to - customize settings or to avoid overheads associated with library - versions. - -* Contents, described in more detail in "description of public routines" below. - - Standard (ANSI/SVID/...) functions: - malloc(size_t n); - calloc(size_t n_elements, size_t element_size); - free(Void_t* p); - realloc(Void_t* p, size_t n); - memalign(size_t alignment, size_t n); - valloc(size_t n); - mallinfo() - mallopt(int parameter_number, int parameter_value) - - Additional functions: - independent_calloc(size_t n_elements, size_t size, Void_t* chunks[]); - independent_comalloc(size_t n_elements, size_t sizes[], Void_t* chunks[]); - pvalloc(size_t n); - cfree(Void_t* p); - malloc_trim(size_t pad); - malloc_usable_size(Void_t* p); - malloc_stats(); - -* Vital statistics: - - Supported pointer representation: 4 or 8 bytes - Supported size_t representation: 4 or 8 bytes - Note that size_t is allowed to be 4 bytes even if pointers are 8. - You can adjust this by defining INTERNAL_SIZE_T - - Alignment: 2 * sizeof(size_t) (default) - (i.e., 8 byte alignment with 4byte size_t). This suffices for - nearly all current machines and C compilers. However, you can - define MALLOC_ALIGNMENT to be wider than this if necessary. - - Minimum overhead per allocated chunk: 4 or 8 bytes - Each malloced chunk has a hidden word of overhead holding size - and status information. - - Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead) - 8-byte ptrs: 24/32 bytes (including, 4/8 overhead) - - When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte - ptrs but 4 byte size) or 24 (for 8/8) additional bytes are - needed; 4 (8) for a trailing size field and 8 (16) bytes for - free list pointers. Thus, the minimum allocatable size is - 16/24/32 bytes. - - Even a request for zero bytes (i.e., malloc(0)) returns a - pointer to something of the minimum allocatable size. - - The maximum overhead wastage (i.e., number of extra bytes - allocated than were requested in malloc) is less than or equal - to the minimum size, except for requests >= mmap_threshold that - are serviced via mmap(), where the worst case wastage is 2 * - sizeof(size_t) bytes plus the remainder from a system page (the - minimal mmap unit); typically 4096 or 8192 bytes. - - Maximum allocated size: 4-byte size_t: 2^32 minus about two pages - 8-byte size_t: 2^64 minus about two pages - - It is assumed that (possibly signed) size_t values suffice to - represent chunk sizes. `Possibly signed' is due to the fact - that `size_t' may be defined on a system as either a signed or - an unsigned type. The ISO C standard says that it must be - unsigned, but a few systems are known not to adhere to this. - Additionally, even when size_t is unsigned, sbrk (which is by - default used to obtain memory from system) accepts signed - arguments, and may not be able to handle size_t-wide arguments - with negative sign bit. Generally, values that would - appear as negative after accounting for overhead and alignment - are supported only via mmap(), which does not have this - limitation. - - Requests for sizes outside the allowed range will perform an optional - failure action and then return null. (Requests may also - also fail because a system is out of memory.) - - Thread-safety: thread-safe unless NO_THREADS is defined - - Compliance: I believe it is compliant with the 1997 Single Unix Specification - (See http://www.opennc.org). Also SVID/XPG, ANSI C, and probably - others as well. - -* Synopsis of compile-time options: - - People have reported using previous versions of this malloc on all - versions of Unix, sometimes by tweaking some of the defines - below. It has been tested most extensively on Solaris and - Linux. It is also reported to work on WIN32 platforms. - People also report using it in stand-alone embedded systems. - - The implementation is in straight, hand-tuned ANSI C. It is not - at all modular. (Sorry!) It uses a lot of macros. To be at all - usable, this code should be compiled using an optimizing compiler - (for example gcc -O3) that can simplify expressions and control - paths. (FAQ: some macros import variables as arguments rather than - declare locals because people reported that some debuggers - otherwise get confused.) - - OPTION DEFAULT VALUE - - Compilation Environment options: - - __STD_C derived from C compiler defines - WIN32 NOT defined - HAVE_MEMCPY defined - USE_MEMCPY 1 if HAVE_MEMCPY is defined - HAVE_MMAP defined as 1 - MMAP_CLEARS 1 - HAVE_MREMAP 0 unless linux defined - USE_ARENAS the same as HAVE_MMAP - USE_STARTER 1 - malloc_getpagesize derived from system #includes, or 4096 if not - HAVE_USR_INCLUDE_MALLOC_H NOT defined - LACKS_UNISTD_H NOT defined unless WIN32 - LACKS_SYS_PARAM_H NOT defined unless WIN32 - LACKS_SYS_MMAN_H NOT defined unless WIN32 - - Changing default word sizes: - - INTERNAL_SIZE_T size_t - MALLOC_ALIGNMENT 2 * sizeof(INTERNAL_SIZE_T) - - Configuration and functionality options: - - USE_DL_PREFIX NOT defined - USE_PUBLIC_MALLOC_WRAPPERS NOT defined - USE_MALLOC_LOCK NOT defined - MALLOC_DEBUG NOT defined - REALLOC_ZERO_BYTES_FREES 1 - MALLOC_FAILURE_ACTION errno = ENOMEM, if __STD_C defined, else no-op - TRIM_FASTBINS 0 - FIRST_SORTED_BIN_SIZE 512 - - Options for customizing MORECORE: - - MORECORE sbrk - MORECORE_FAILURE -1 - MORECORE_CONTIGUOUS 1 - MORECORE_CANNOT_TRIM NOT defined - MORECORE_CLEARS 1 - MMAP_AS_MORECORE_SIZE (1024 * 1024) - - Tuning options that are also dynamically changeable via mallopt: - - DEFAULT_MXFAST 64 - DEFAULT_TRIM_THRESHOLD 128 * 1024 - DEFAULT_TOP_PAD 0 - DEFAULT_MMAP_THRESHOLD 128 * 1024 - DEFAULT_MMAP_MAX 65536 - - There are several other #defined constants and macros that you - probably don't want to touch unless you are extending or adapting malloc. */ - -/* - __STD_C should be nonzero if using ANSI-standard C compiler, a C++ - compiler, or a C compiler sufficiently close to ANSI to get away - with it. -*/ - -#ifndef __STD_C -#if defined(__STDC__) || defined(__cplusplus) -#define __STD_C 1 -#else -#define __STD_C 0 -#endif -#endif /*__STD_C*/ - - -/* - Void_t* is the pointer type that malloc should say it returns -*/ - -#ifndef Void_t -#if (__STD_C || defined(WIN32)) -#define Void_t void -#else -#define Void_t char -#endif -#endif /*Void_t*/ - -#if __STD_C -#include /* for size_t */ -#include /* for getenv(), abort() */ -#else -#include -#endif - -#include - -#ifdef __cplusplus -extern "C" { -#endif - -/* define LACKS_UNISTD_H if your system does not have a . */ - -/* #define LACKS_UNISTD_H */ - -#ifndef LACKS_UNISTD_H -#include -#endif - -/* define LACKS_SYS_PARAM_H if your system does not have a . */ - -/* #define LACKS_SYS_PARAM_H */ - - -#include /* needed for malloc_stats */ -#include /* needed for optional MALLOC_FAILURE_ACTION */ - - -/* - Debugging: - - Because freed chunks may be overwritten with bookkeeping fields, this - malloc will often die when freed memory is overwritten by user - programs. This can be very effective (albeit in an annoying way) - in helping track down dangling pointers. - - If you compile with -DMALLOC_DEBUG, a number of assertion checks are - enabled that will catch more memory errors. You probably won't be - able to make much sense of the actual assertion errors, but they - should help you locate incorrectly overwritten memory. The checking - is fairly extensive, and will slow down execution - noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set - will attempt to check every non-mmapped allocated and free chunk in - the course of computing the summmaries. (By nature, mmapped regions - cannot be checked very much automatically.) - - Setting MALLOC_DEBUG may also be helpful if you are trying to modify - this code. The assertions in the check routines spell out in more - detail the assumptions and invariants underlying the algorithms. - - Setting MALLOC_DEBUG does NOT provide an automated mechanism for - checking that all accesses to malloced memory stay within their - bounds. However, there are several add-ons and adaptations of this - or other mallocs available that do this. -*/ - -#if MALLOC_DEBUG -#include -#else -#undef assert -#define assert(x) ((void)0) -#endif - - -/* - INTERNAL_SIZE_T is the word-size used for internal bookkeeping - of chunk sizes. - - The default version is the same as size_t. - - While not strictly necessary, it is best to define this as an - unsigned type, even if size_t is a signed type. This may avoid some - artificial size limitations on some systems. - - On a 64-bit machine, you may be able to reduce malloc overhead by - defining INTERNAL_SIZE_T to be a 32 bit `unsigned int' at the - expense of not being able to handle more than 2^32 of malloced - space. If this limitation is acceptable, you are encouraged to set - this unless you are on a platform requiring 16byte alignments. In - this case the alignment requirements turn out to negate any - potential advantages of decreasing size_t word size. - - Implementors: Beware of the possible combinations of: - - INTERNAL_SIZE_T might be signed or unsigned, might be 32 or 64 bits, - and might be the same width as int or as long - - size_t might have different width and signedness as INTERNAL_SIZE_T - - int and long might be 32 or 64 bits, and might be the same width - To deal with this, most comparisons and difference computations - among INTERNAL_SIZE_Ts should cast them to unsigned long, being - aware of the fact that casting an unsigned int to a wider long does - not sign-extend. (This also makes checking for negative numbers - awkward.) Some of these casts result in harmless compiler warnings - on some systems. -*/ - -#ifndef INTERNAL_SIZE_T -#define INTERNAL_SIZE_T size_t -#endif - -/* The corresponding word size */ -#define SIZE_SZ (sizeof(INTERNAL_SIZE_T)) - - -/* - MALLOC_ALIGNMENT is the minimum alignment for malloc'ed chunks. - It must be a power of two at least 2 * SIZE_SZ, even on machines - for which smaller alignments would suffice. It may be defined as - larger than this though. Note however that code and data structures - are optimized for the case of 8-byte alignment. -*/ - - -#ifndef MALLOC_ALIGNMENT -#define MALLOC_ALIGNMENT (2 * SIZE_SZ) -#endif - -/* The corresponding bit mask value */ -#define MALLOC_ALIGN_MASK (MALLOC_ALIGNMENT - 1) - - - -/* - REALLOC_ZERO_BYTES_FREES should be set if a call to - realloc with zero bytes should be the same as a call to free. - This is required by the C standard. Otherwise, since this malloc - returns a unique pointer for malloc(0), so does realloc(p, 0). -*/ - -#ifndef REALLOC_ZERO_BYTES_FREES -#define REALLOC_ZERO_BYTES_FREES 1 -#endif - -/* - TRIM_FASTBINS controls whether free() of a very small chunk can - immediately lead to trimming. Setting to true (1) can reduce memory - footprint, but will almost always slow down programs that use a lot - of small chunks. - - Define this only if you are willing to give up some speed to more - aggressively reduce system-level memory footprint when releasing - memory in programs that use many small chunks. You can get - essentially the same effect by setting MXFAST to 0, but this can - lead to even greater slowdowns in programs using many small chunks. - TRIM_FASTBINS is an in-between compile-time option, that disables - only those chunks bordering topmost memory from being placed in - fastbins. -*/ - -#ifndef TRIM_FASTBINS -#define TRIM_FASTBINS 0 -#endif - - -/* - USE_DL_PREFIX will prefix all public routines with the string 'dl'. - This is necessary when you only want to use this malloc in one part - of a program, using your regular system malloc elsewhere. -*/ - -/* #define USE_DL_PREFIX */ - - -/* - Two-phase name translation. - All of the actual routines are given mangled names. - When wrappers are used, they become the public callable versions. - When DL_PREFIX is used, the callable names are prefixed. -*/ - -#ifdef USE_DL_PREFIX -#define public_cALLOc dlcalloc -#define public_fREe dlfree -#define public_cFREe dlcfree -#define public_mALLOc dlmalloc -#define public_mEMALIGn dlmemalign -#define public_rEALLOc dlrealloc -#define public_vALLOc dlvalloc -#define public_pVALLOc dlpvalloc -#define public_mALLINFo dlmallinfo -#define public_mALLOPt dlmallopt -#define public_mTRIm dlmalloc_trim -#define public_mSTATs dlmalloc_stats -#define public_mUSABLe dlmalloc_usable_size -#define public_iCALLOc dlindependent_calloc -#define public_iCOMALLOc dlindependent_comalloc -#define public_gET_STATe dlget_state -#define public_sET_STATe dlset_state -#else /* USE_DL_PREFIX */ -#ifdef _LIBC - -/* Special defines for the GNU C library. */ -#define public_cALLOc __libc_calloc -#define public_fREe __libc_free -#define public_cFREe __libc_cfree -#define public_mALLOc __libc_malloc -#define public_mEMALIGn __libc_memalign -#define public_rEALLOc __libc_realloc -#define public_vALLOc __libc_valloc -#define public_pVALLOc __libc_pvalloc -#define public_mALLINFo __libc_mallinfo -#define public_mALLOPt __libc_mallopt -#define public_mTRIm __malloc_trim -#define public_mSTATs __malloc_stats -#define public_mUSABLe __malloc_usable_size -#define public_iCALLOc __libc_independent_calloc -#define public_iCOMALLOc __libc_independent_comalloc -#define public_gET_STATe __malloc_get_state -#define public_sET_STATe __malloc_set_state -#define malloc_getpagesize __getpagesize() -#define open __open -#define mmap __mmap -#define munmap __munmap -#define mremap __mremap -#define mprotect __mprotect -#define MORECORE (*__morecore) -#define MORECORE_FAILURE 0 - -Void_t * __default_morecore (ptrdiff_t); -Void_t *(*__morecore)(ptrdiff_t) = __default_morecore; - -#else /* !_LIBC */ - /* OMPI change: put "opal_memory_ptmalloc2_" in front of all of these */ -#define public_cALLOc opal_memory_ptmalloc2_calloc -#define public_fREe opal_memory_ptmalloc2_free -#define public_cFREe opal_memory_ptmalloc2_cfree -#define public_mALLOc opal_memory_ptmalloc2_malloc -#define public_mEMALIGn opal_memory_ptmalloc2_memalign -#define public_rEALLOc opal_memory_ptmalloc2_realloc -#define public_vALLOc opal_memory_ptmalloc2_valloc -#define public_pVALLOc opal_memory_ptmalloc2_pvalloc -#define public_mALLINFo opal_memory_ptmalloc2_mallinfo -#define public_mALLOPt opal_memory_ptmalloc2_mallopt -#define public_mTRIm opal_memory_ptmalloc2_malloc_trim -#define public_mSTATs opal_memory_ptmalloc2_malloc_stats -#define public_mUSABLe opal_memory_ptmalloc2_malloc_usable_size -#define public_iCALLOc opal_memory_ptmalloc2_independent_calloc -#define public_iCOMALLOc opal_memory_ptmalloc2_independent_comalloc -#define public_gET_STATe opal_memory_ptmalloc2_malloc_get_state -#define public_sET_STATe opal_memory_ptmalloc2_malloc_set_state -#endif /* _LIBC */ -#endif /* USE_DL_PREFIX */ - - -#if !defined _LIBC && (!defined __GNUC__ || __GNUC__<3) -#ifndef __builtin_expect -#define __builtin_expect(expr, val) (expr) -#endif -#endif - -/* - HAVE_MEMCPY should be defined if you are not otherwise using - ANSI STD C, but still have memcpy and memset in your C library - and want to use them in calloc and realloc. Otherwise simple - macro versions are defined below. - - USE_MEMCPY should be defined as 1 if you actually want to - have memset and memcpy called. People report that the macro - versions are faster than libc versions on some systems. - - Even if USE_MEMCPY is set to 1, loops to copy/clear small chunks - (of <= 36 bytes) are manually unrolled in realloc and calloc. -*/ - -#define HAVE_MEMCPY - -#ifndef USE_MEMCPY -#ifdef HAVE_MEMCPY -#define USE_MEMCPY 1 -#else -#define USE_MEMCPY 0 -#endif -#endif - - -#if (__STD_C || defined(HAVE_MEMCPY)) - -#ifdef _LIBC -# include -#else -#ifdef WIN32 -/* On Win32 memset and memcpy are already declared in windows.h */ -#else -#if __STD_C -void* memset(void*, int, size_t); -void* memcpy(void*, const void*, size_t); -#else -Void_t* memset(); -Void_t* memcpy(); -#endif -#endif -#endif -#endif - -/* - MALLOC_FAILURE_ACTION is the action to take before "return 0" when - malloc fails to be able to return memory, either because memory is - exhausted or because of illegal arguments. - - By default, sets errno if running on STD_C platform, else does nothing. -*/ - -#ifndef MALLOC_FAILURE_ACTION -#if __STD_C -#define MALLOC_FAILURE_ACTION \ - errno = ENOMEM; - -#else -#define MALLOC_FAILURE_ACTION -#endif -#endif - -/* - MORECORE-related declarations. By default, rely on sbrk -*/ - - -#ifdef LACKS_UNISTD_H -#if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__NetBSD__) -#if __STD_C -extern Void_t* sbrk(ptrdiff_t); -#else -extern Void_t* sbrk(); -#endif -#endif -#endif - -/* - MORECORE is the name of the routine to call to obtain more memory - from the system. See below for general guidance on writing - alternative MORECORE functions, as well as a version for WIN32 and a - sample version for pre-OSX macos. -*/ - -#ifndef MORECORE -#define MORECORE sbrk -#endif - -/* - MORECORE_FAILURE is the value returned upon failure of MORECORE - as well as mmap. Since it cannot be an otherwise valid memory address, - and must reflect values of standard sys calls, you probably ought not - try to redefine it. -*/ - -#ifndef MORECORE_FAILURE -#define MORECORE_FAILURE (-1) -#endif - -/* - If MORECORE_CONTIGUOUS is true, take advantage of fact that - consecutive calls to MORECORE with positive arguments always return - contiguous increasing addresses. This is true of unix sbrk. Even - if not defined, when regions happen to be contiguous, malloc will - permit allocations spanning regions obtained from different - calls. But defining this when applicable enables some stronger - consistency checks and space efficiencies. -*/ - -#ifndef MORECORE_CONTIGUOUS -#define MORECORE_CONTIGUOUS 1 -#endif - -/* - Define MORECORE_CANNOT_TRIM if your version of MORECORE - cannot release space back to the system when given negative - arguments. This is generally necessary only if you are using - a hand-crafted MORECORE function that cannot handle negative arguments. -*/ - -/* #define MORECORE_CANNOT_TRIM */ - -/* MORECORE_CLEARS (default 1) - The degree to which the routine mapped to MORECORE zeroes out - memory: never (0), only for newly allocated space (1) or always - (2). The distinction between (1) and (2) is necessary because on - some systems, if the application first decrements and then - increments the break value, the contents of the reallocated space - are unspecified. -*/ - -#ifndef MORECORE_CLEARS -#define MORECORE_CLEARS 1 -#endif - - -/* - Define HAVE_MMAP as true to optionally make malloc() use mmap() to - allocate very large blocks. These will be returned to the - operating system immediately after a free(). Also, if mmap - is available, it is used as a backup strategy in cases where - MORECORE fails to provide space from system. - - This malloc is best tuned to work with mmap for large requests. - If you do not have mmap, operations involving very large chunks (1MB - or so) may be slower than you'd like. -*/ - -#ifndef HAVE_MMAP -#define HAVE_MMAP 1 - -/* - Standard unix mmap using /dev/zero clears memory so calloc doesn't - need to. -*/ - -#ifndef MMAP_CLEARS -#define MMAP_CLEARS 1 -#endif - -#else /* no mmap */ -#ifndef MMAP_CLEARS -#define MMAP_CLEARS 0 -#endif -#endif - - -/* - MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if - sbrk fails, and mmap is used as a backup (which is done only if - HAVE_MMAP). The value must be a multiple of page size. This - backup strategy generally applies only when systems have "holes" in - address space, so sbrk cannot perform contiguous expansion, but - there is still space available on system. On systems for which - this is known to be useful (i.e. most linux kernels), this occurs - only when programs allocate huge amounts of memory. Between this, - and the fact that mmap regions tend to be limited, the size should - be large, to avoid too many mmap calls and thus avoid running out - of kernel resources. -*/ - -#ifndef MMAP_AS_MORECORE_SIZE -#define MMAP_AS_MORECORE_SIZE (1024 * 1024) -#endif - -/* - Define HAVE_MREMAP to make realloc() use mremap() to re-allocate - large blocks. This is currently only possible on Linux with - kernel versions newer than 1.3.77. -*/ - -#ifndef HAVE_MREMAP -#ifdef linux -#define HAVE_MREMAP 1 -#else -#define HAVE_MREMAP 0 -#endif - -#endif /* HAVE_MMAP */ - -/* Define USE_ARENAS to enable support for multiple `arenas'. These - are allocated using mmap(), are necessary for threads and - occasionally useful to overcome address space limitations affecting - sbrk(). */ - -#ifndef USE_ARENAS -#define USE_ARENAS HAVE_MMAP -#endif - -/* USE_STARTER determines if and when the special "starter" hook - functions are used: not at all (0), during ptmalloc_init (first bit - set), or from the beginning until an explicit call to ptmalloc_init - (second bit set). This is necessary if thread-related - initialization functions (e.g. pthread_key_create) require - malloc() calls (set USE_STARTER=1), or if those functions initially - cannot be used at all (set USE_STARTER=2 and perform an explicit - ptmalloc_init() when the thread library is ready, typically at the - start of main()). */ - -#ifndef USE_STARTER -# ifndef _LIBC -# define USE_STARTER 1 -# else -# if USE___THREAD || (defined USE_TLS && !defined SHARED) - /* These routines are never needed in this configuration. */ -# define USE_STARTER 0 -# else -# define USE_STARTER (USE_TLS ? 4 : 1) -# endif -# endif -#endif - - -/* - The system page size. To the extent possible, this malloc manages - memory from the system in page-size units. Note that this value is - cached during initialization into a field of malloc_state. So even - if malloc_getpagesize is a function, it is only called once. - - The following mechanics for getpagesize were adapted from bsd/gnu - getpagesize.h. If none of the system-probes here apply, a value of - 4096 is used, which should be OK: If they don't apply, then using - the actual value probably doesn't impact performance. -*/ - - -#ifndef malloc_getpagesize - -#ifndef LACKS_UNISTD_H -# include -#endif - -# ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */ -# ifndef _SC_PAGE_SIZE -# define _SC_PAGE_SIZE _SC_PAGESIZE -# endif -# endif - -# ifdef _SC_PAGE_SIZE -# define malloc_getpagesize sysconf(_SC_PAGE_SIZE) -# else -# if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE) - extern size_t getpagesize(); -# define malloc_getpagesize getpagesize() -# else -# ifdef WIN32 /* use supplied emulation of getpagesize */ -# define malloc_getpagesize getpagesize() -# else -# ifndef LACKS_SYS_PARAM_H -# include -# endif -# ifdef EXEC_PAGESIZE -# define malloc_getpagesize EXEC_PAGESIZE -# else -# ifdef NBPG -# ifndef CLSIZE -# define malloc_getpagesize NBPG -# else -# define malloc_getpagesize (NBPG * CLSIZE) -# endif -# else -# ifdef NBPC -# define malloc_getpagesize NBPC -# else -# ifdef PAGESIZE -# define malloc_getpagesize PAGESIZE -# else /* just guess */ -# define malloc_getpagesize (4096) -# endif -# endif -# endif -# endif -# endif -# endif -# endif -#endif - -/* - This version of malloc supports the standard SVID/XPG mallinfo - routine that returns a struct containing usage properties and - statistics. It should work on any SVID/XPG compliant system that has - a /usr/include/malloc.h defining struct mallinfo. (If you'd like to - install such a thing yourself, cut out the preliminary declarations - as described above and below and save them in a malloc.h file. But - there's no compelling reason to bother to do this.) - - The main declaration needed is the mallinfo struct that is returned - (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a - bunch of fields that are not even meaningful in this version of - malloc. These fields are are instead filled by mallinfo() with - other numbers that might be of interest. - - HAVE_USR_INCLUDE_MALLOC_H should be set if you have a - /usr/include/malloc.h file that includes a declaration of struct - mallinfo. If so, it is included; else an SVID2/XPG2 compliant - version is declared below. These must be precisely the same for - mallinfo() to work. The original SVID version of this struct, - defined on most systems with mallinfo, declares all fields as - ints. But some others define as unsigned long. If your system - defines the fields using a type of different width than listed here, - you must #include your system version and #define - HAVE_USR_INCLUDE_MALLOC_H. -*/ - -/* #define HAVE_USR_INCLUDE_MALLOC_H */ - -#ifdef HAVE_USR_INCLUDE_MALLOC_H -#include "/usr/include/malloc.h" -#endif - - -/* ---------- description of public routines ------------ */ - -/* - malloc(size_t n) - Returns a pointer to a newly allocated chunk of at least n bytes, or null - if no space is available. Additionally, on failure, errno is - set to ENOMEM on ANSI C systems. - - If n is zero, malloc returns a minumum-sized chunk. (The minimum - size is 16 bytes on most 32bit systems, and 24 or 32 bytes on 64bit - systems.) On most systems, size_t is an unsigned type, so calls - with negative arguments are interpreted as requests for huge amounts - of space, which will often fail. The maximum supported value of n - differs across systems, but is in all cases less than the maximum - representable value of a size_t. -*/ -#if __STD_C -Void_t* public_mALLOc(size_t); -#else -Void_t* public_mALLOc(); -#endif -#ifdef libc_hidden_proto -libc_hidden_proto (public_mALLOc) -#endif - -/* - free(Void_t* p) - Releases the chunk of memory pointed to by p, that had been previously - allocated using malloc or a related routine such as realloc. - It has no effect if p is null. It can have arbitrary (i.e., bad!) - effects if p has already been freed. - - Unless disabled (using mallopt), freeing very large spaces will - when possible, automatically trigger operations that give - back unused memory to the system, thus reducing program footprint. -*/ -#if __STD_C -void public_fREe(Void_t*); -#else -void public_fREe(); -#endif -#ifdef libc_hidden_proto -libc_hidden_proto (public_fREe) -#endif - -/* - calloc(size_t n_elements, size_t element_size); - Returns a pointer to n_elements * element_size bytes, with all locations - set to zero. -*/ -#if __STD_C -Void_t* public_cALLOc(size_t, size_t); -#else -Void_t* public_cALLOc(); -#endif - -/* - realloc(Void_t* p, size_t n) - Returns a pointer to a chunk of size n that contains the same data - as does chunk p up to the minimum of (n, p's size) bytes, or null - if no space is available. - - The returned pointer may or may not be the same as p. The algorithm - prefers extending p when possible, otherwise it employs the - equivalent of a malloc-copy-free sequence. - - If p is null, realloc is equivalent to malloc. - - If space is not available, realloc returns null, errno is set (if on - ANSI) and p is NOT freed. - - if n is for fewer bytes than already held by p, the newly unused - space is lopped off and freed if possible. Unless the #define - REALLOC_ZERO_BYTES_FREES is set, realloc with a size argument of - zero (re)allocates a minimum-sized chunk. - - Large chunks that were internally obtained via mmap will always - be reallocated using malloc-copy-free sequences unless - the system supports MREMAP (currently only linux). - - The old unix realloc convention of allowing the last-free'd chunk - to be used as an argument to realloc is not supported. -*/ -#if __STD_C -Void_t* public_rEALLOc(Void_t*, size_t); -#else -Void_t* public_rEALLOc(); -#endif -#ifdef libc_hidden_proto -libc_hidden_proto (public_rEALLOc) -#endif - -/* - memalign(size_t alignment, size_t n); - Returns a pointer to a newly allocated chunk of n bytes, aligned - in accord with the alignment argument. - - The alignment argument should be a power of two. If the argument is - not a power of two, the nearest greater power is used. - 8-byte alignment is guaranteed by normal malloc calls, so don't - bother calling memalign with an argument of 8 or less. - - Overreliance on memalign is a sure way to fragment space. -*/ -#if __STD_C -Void_t* public_mEMALIGn(size_t, size_t); -#else -Void_t* public_mEMALIGn(); -#endif -#ifdef libc_hidden_proto -libc_hidden_proto (public_mEMALIGn) -#endif - -/* - valloc(size_t n); - Equivalent to memalign(pagesize, n), where pagesize is the page - size of the system. If the pagesize is unknown, 4096 is used. -*/ -#if __STD_C -Void_t* public_vALLOc(size_t); -#else -Void_t* public_vALLOc(); -#endif - - - -/* - mallopt(int parameter_number, int parameter_value) - Sets tunable parameters The format is to provide a - (parameter-number, parameter-value) pair. mallopt then sets the - corresponding parameter to the argument value if it can (i.e., so - long as the value is meaningful), and returns 1 if successful else - 0. SVID/XPG/ANSI defines four standard param numbers for mallopt, - normally defined in malloc.h. Only one of these (M_MXFAST) is used - in this malloc. The others (M_NLBLKS, M_GRAIN, M_KEEP) don't apply, - so setting them has no effect. But this malloc also supports four - other options in mallopt. See below for details. Briefly, supported - parameters are as follows (listed defaults are for "typical" - configurations). - - Symbol param # default allowed param values - M_MXFAST 1 64 0-80 (0 disables fastbins) - M_TRIM_THRESHOLD -1 128*1024 any (-1U disables trimming) - M_TOP_PAD -2 0 any - M_MMAP_THRESHOLD -3 128*1024 any (or 0 if no MMAP support) - M_MMAP_MAX -4 65536 any (0 disables use of mmap) -*/ -#if __STD_C -int public_mALLOPt(int, int); -#else -int public_mALLOPt(); -#endif - - -/* - mallinfo() - Returns (by copy) a struct containing various summary statistics: - - arena: current total non-mmapped bytes allocated from system - ordblks: the number of free chunks - smblks: the number of fastbin blocks (i.e., small chunks that - have been freed but not use resused or consolidated) - hblks: current number of mmapped regions - hblkhd: total bytes held in mmapped regions - usmblks: the maximum total allocated space. This will be greater - than current total if trimming has occurred. - fsmblks: total bytes held in fastbin blocks - uordblks: current total allocated space (normal or mmapped) - fordblks: total free space - keepcost: the maximum number of bytes that could ideally be released - back to system via malloc_trim. ("ideally" means that - it ignores page restrictions etc.) - - Because these fields are ints, but internal bookkeeping may - be kept as longs, the reported values may wrap around zero and - thus be inaccurate. -*/ -#if __STD_C -struct mallinfo public_mALLINFo(void); -#else -struct mallinfo public_mALLINFo(); -#endif - -/* - independent_calloc(size_t n_elements, size_t element_size, Void_t* chunks[]); - - independent_calloc is similar to calloc, but instead of returning a - single cleared space, it returns an array of pointers to n_elements - independent elements that can hold contents of size elem_size, each - of which starts out cleared, and can be independently freed, - realloc'ed etc. The elements are guaranteed to be adjacently - allocated (this is not guaranteed to occur with multiple callocs or - mallocs), which may also improve cache locality in some - applications. - - The "chunks" argument is optional (i.e., may be null, which is - probably the most typical usage). If it is null, the returned array - is itself dynamically allocated and should also be freed when it is - no longer needed. Otherwise, the chunks array must be of at least - n_elements in length. It is filled in with the pointers to the - chunks. - - In either case, independent_calloc returns this pointer array, or - null if the allocation failed. If n_elements is zero and "chunks" - is null, it returns a chunk representing an array with zero elements - (which should be freed if not wanted). - - Each element must be individually freed when it is no longer - needed. If you'd like to instead be able to free all at once, you - should instead use regular calloc and assign pointers into this - space to represent elements. (In this case though, you cannot - independently free elements.) - - independent_calloc simplifies and speeds up implementations of many - kinds of pools. It may also be useful when constructing large data - structures that initially have a fixed number of fixed-sized nodes, - but the number is not known at compile time, and some of the nodes - may later need to be freed. For example: - - struct Node { int item; struct Node* next; }; - - struct Node* build_list() { - struct Node** pool; - int n = read_number_of_nodes_needed(); - if (n <= 0) return 0; - pool = (struct Node**)(independent_calloc(n, sizeof(struct Node), 0); - if (pool == 0) die(); - // organize into a linked list... - struct Node* first = pool[0]; - for (i = 0; i < n-1; ++i) - pool[i]->next = pool[i+1]; - free(pool); // Can now free the array (or not, if it is needed later) - return first; - } -*/ -#if __STD_C -Void_t** public_iCALLOc(size_t, size_t, Void_t**); -#else -Void_t** public_iCALLOc(); -#endif - -/* - independent_comalloc(size_t n_elements, size_t sizes[], Void_t* chunks[]); - - independent_comalloc allocates, all at once, a set of n_elements - chunks with sizes indicated in the "sizes" array. It returns - an array of pointers to these elements, each of which can be - independently freed, realloc'ed etc. The elements are guaranteed to - be adjacently allocated (this is not guaranteed to occur with - multiple callocs or mallocs), which may also improve cache locality - in some applications. - - The "chunks" argument is optional (i.e., may be null). If it is null - the returned array is itself dynamically allocated and should also - be freed when it is no longer needed. Otherwise, the chunks array - must be of at least n_elements in length. It is filled in with the - pointers to the chunks. - - In either case, independent_comalloc returns this pointer array, or - null if the allocation failed. If n_elements is zero and chunks is - null, it returns a chunk representing an array with zero elements - (which should be freed if not wanted). - - Each element must be individually freed when it is no longer - needed. If you'd like to instead be able to free all at once, you - should instead use a single regular malloc, and assign pointers at - particular offsets in the aggregate space. (In this case though, you - cannot independently free elements.) - - independent_comallac differs from independent_calloc in that each - element may have a different size, and also that it does not - automatically clear elements. - - independent_comalloc can be used to speed up allocation in cases - where several structs or objects must always be allocated at the - same time. For example: - - struct Head { ... } - struct Foot { ... } - - void send_message(char* msg) { - int msglen = strlen(msg); - size_t sizes[3] = { sizeof(struct Head), msglen, sizeof(struct Foot) }; - void* chunks[3]; - if (independent_comalloc(3, sizes, chunks) == 0) - die(); - struct Head* head = (struct Head*)(chunks[0]); - char* body = (char*)(chunks[1]); - struct Foot* foot = (struct Foot*)(chunks[2]); - // ... - } - - In general though, independent_comalloc is worth using only for - larger values of n_elements. For small values, you probably won't - detect enough difference from series of malloc calls to bother. - - Overuse of independent_comalloc can increase overall memory usage, - since it cannot reuse existing noncontiguous small chunks that - might be available for some of the elements. -*/ -#if __STD_C -Void_t** public_iCOMALLOc(size_t, size_t*, Void_t**); -#else -Void_t** public_iCOMALLOc(); -#endif - - -/* - pvalloc(size_t n); - Equivalent to valloc(minimum-page-that-holds(n)), that is, - round up n to nearest pagesize. - */ -#if __STD_C -Void_t* public_pVALLOc(size_t); -#else -Void_t* public_pVALLOc(); -#endif - -/* - cfree(Void_t* p); - Equivalent to free(p). - - cfree is needed/defined on some systems that pair it with calloc, - for odd historical reasons (such as: cfree is used in example - code in the first edition of K&R). -*/ -#if __STD_C -void public_cFREe(Void_t*); -#else -void public_cFREe(); -#endif - -/* - malloc_trim(size_t pad); - - If possible, gives memory back to the system (via negative - arguments to sbrk) if there is unused memory at the `high' end of - the malloc pool. You can call this after freeing large blocks of - memory to potentially reduce the system-level memory requirements - of a program. However, it cannot guarantee to reduce memory. Under - some allocation patterns, some large free blocks of memory will be - locked between two used chunks, so they cannot be given back to - the system. - - The `pad' argument to malloc_trim represents the amount of free - trailing space to leave untrimmed. If this argument is zero, - only the minimum amount of memory to maintain internal data - structures will be left (one page or less). Non-zero arguments - can be supplied to maintain enough trailing space to service - future expected allocations without having to re-obtain memory - from the system. - - Malloc_trim returns 1 if it actually released any memory, else 0. - On systems that do not support "negative sbrks", it will always - rreturn 0. -*/ -#if __STD_C -int public_mTRIm(size_t); -#else -int public_mTRIm(); -#endif - -/* - malloc_usable_size(Void_t* p); - - Returns the number of bytes you can actually use in - an allocated chunk, which may be more than you requested (although - often not) due to alignment and minimum size constraints. - You can use this many bytes without worrying about - overwriting other allocated objects. This is not a particularly great - programming practice. malloc_usable_size can be more useful in - debugging and assertions, for example: - - p = malloc(n); - assert(malloc_usable_size(p) >= 256); - -*/ -#if __STD_C -size_t public_mUSABLe(Void_t*); -#else -size_t public_mUSABLe(); -#endif - -/* - malloc_stats(); - Prints on stderr the amount of space obtained from the system (both - via sbrk and mmap), the maximum amount (which may be more than - current if malloc_trim and/or munmap got called), and the current - number of bytes allocated via malloc (or realloc, etc) but not yet - freed. Note that this is the number of bytes allocated, not the - number requested. It will be larger than the number requested - because of alignment and bookkeeping overhead. Because it includes - alignment wastage as being in use, this figure may be greater than - zero even when no user-level chunks are allocated. - - The reported current and maximum system memory can be inaccurate if - a program makes other calls to system memory allocation functions - (normally sbrk) outside of malloc. - - malloc_stats prints only the most commonly interesting statistics. - More information can be obtained by calling mallinfo. - -*/ -#if __STD_C -void public_mSTATs(void); -#else -void public_mSTATs(); -#endif - -/* - malloc_get_state(void); - - Returns the state of all malloc variables in an opaque data - structure. -*/ -#if __STD_C -Void_t* public_gET_STATe(void); -#else -Void_t* public_gET_STATe(); -#endif - -/* - malloc_set_state(Void_t* state); - - Restore the state of all malloc variables from data obtained with - malloc_get_state(). -*/ -#if __STD_C -int public_sET_STATe(Void_t*); -#else -int public_sET_STATe(); -#endif - -#ifdef _LIBC -/* - posix_memalign(void **memptr, size_t alignment, size_t size); - - POSIX wrapper like memalign(), checking for validity of size. -*/ -int __posix_memalign(void **, size_t, size_t); -#endif - -/* mallopt tuning options */ - -/* - M_MXFAST is the maximum request size used for "fastbins", special bins - that hold returned chunks without consolidating their spaces. This - enables future requests for chunks of the same size to be handled - very quickly, but can increase fragmentation, and thus increase the - overall memory footprint of a program. - - This malloc manages fastbins very conservatively yet still - efficiently, so fragmentation is rarely a problem for values less - than or equal to the default. The maximum supported value of MXFAST - is 80. You wouldn't want it any higher than this anyway. Fastbins - are designed especially for use with many small structs, objects or - strings -- the default handles structs/objects/arrays with sizes up - to 8 4byte fields, or small strings representing words, tokens, - etc. Using fastbins for larger objects normally worsens - fragmentation without improving speed. - - M_MXFAST is set in REQUEST size units. It is internally used in - chunksize units, which adds padding and alignment. You can reduce - M_MXFAST to 0 to disable all use of fastbins. This causes the malloc - algorithm to be a closer approximation of fifo-best-fit in all cases, - not just for larger requests, but will generally cause it to be - slower. -*/ - - -/* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */ -#ifndef M_MXFAST -#define M_MXFAST 1 -#endif - -#ifndef DEFAULT_MXFAST -#define DEFAULT_MXFAST 64 -#endif - - -/* - M_TRIM_THRESHOLD is the maximum amount of unused top-most memory - to keep before releasing via malloc_trim in free(). - - Automatic trimming is mainly useful in long-lived programs. - Because trimming via sbrk can be slow on some systems, and can - sometimes be wasteful (in cases where programs immediately - afterward allocate more large chunks) the value should be high - enough so that your overall system performance would improve by - releasing this much memory. - - The trim threshold and the mmap control parameters (see below) - can be traded off with one another. Trimming and mmapping are - two different ways of releasing unused memory back to the - system. Between these two, it is often possible to keep - system-level demands of a long-lived program down to a bare - minimum. For example, in one test suite of sessions measuring - the XF86 X server on Linux, using a trim threshold of 128K and a - mmap threshold of 192K led to near-minimal long term resource - consumption. - - If you are using this malloc in a long-lived program, it should - pay to experiment with these values. As a rough guide, you - might set to a value close to the average size of a process - (program) running on your system. Releasing this much memory - would allow such a process to run in memory. Generally, it's - worth it to tune for trimming rather tham memory mapping when a - program undergoes phases where several large chunks are - allocated and released in ways that can reuse each other's - storage, perhaps mixed with phases where there are no such - chunks at all. And in well-behaved long-lived programs, - controlling release of large blocks via trimming versus mapping - is usually faster. - - However, in most programs, these parameters serve mainly as - protection against the system-level effects of carrying around - massive amounts of unneeded memory. Since frequent calls to - sbrk, mmap, and munmap otherwise degrade performance, the default - parameters are set to relatively high values that serve only as - safeguards. - - The trim value It must be greater than page size to have any useful - effect. To disable trimming completely, you can set to - (unsigned long)(-1) - - Trim settings interact with fastbin (MXFAST) settings: Unless - TRIM_FASTBINS is defined, automatic trimming never takes place upon - freeing a chunk with size less than or equal to MXFAST. Trimming is - instead delayed until subsequent freeing of larger chunks. However, - you can still force an attempted trim by calling malloc_trim. - - Also, trimming is not generally possible in cases where - the main arena is obtained via mmap. - - Note that the trick some people use of mallocing a huge space and - then freeing it at program startup, in an attempt to reserve system - memory, doesn't have the intended effect under automatic trimming, - since that memory will immediately be returned to the system. -*/ - -#define M_TRIM_THRESHOLD -1 - -#ifndef DEFAULT_TRIM_THRESHOLD -#define DEFAULT_TRIM_THRESHOLD (128 * 1024) -#endif - -/* - M_TOP_PAD is the amount of extra `padding' space to allocate or - retain whenever sbrk is called. It is used in two ways internally: - - * When sbrk is called to extend the top of the arena to satisfy - a new malloc request, this much padding is added to the sbrk - request. - - * When malloc_trim is called automatically from free(), - it is used as the `pad' argument. - - In both cases, the actual amount of padding is rounded - so that the end of the arena is always a system page boundary. - - The main reason for using padding is to avoid calling sbrk so - often. Having even a small pad greatly reduces the likelihood - that nearly every malloc request during program start-up (or - after trimming) will invoke sbrk, which needlessly wastes - time. - - Automatic rounding-up to page-size units is normally sufficient - to avoid measurable overhead, so the default is 0. However, in - systems where sbrk is relatively slow, it can pay to increase - this value, at the expense of carrying around more memory than - the program needs. -*/ - -#define M_TOP_PAD -2 - -#ifndef DEFAULT_TOP_PAD -#define DEFAULT_TOP_PAD (0) -#endif - -/* - M_MMAP_THRESHOLD is the request size threshold for using mmap() - to service a request. Requests of at least this size that cannot - be allocated using already-existing space will be serviced via mmap. - (If enough normal freed space already exists it is used instead.) - - Using mmap segregates relatively large chunks of memory so that - they can be individually obtained and released from the host - system. A request serviced through mmap is never reused by any - other request (at least not directly; the system may just so - happen to remap successive requests to the same locations). - - Segregating space in this way has the benefits that: - - 1. Mmapped space can ALWAYS be individually released back - to the system, which helps keep the system level memory - demands of a long-lived program low. - 2. Mapped memory can never become `locked' between - other chunks, as can happen with normally allocated chunks, which - means that even trimming via malloc_trim would not release them. - 3. On some systems with "holes" in address spaces, mmap can obtain - memory that sbrk cannot. - - However, it has the disadvantages that: - - 1. The space cannot be reclaimed, consolidated, and then - used to service later requests, as happens with normal chunks. - 2. It can lead to more wastage because of mmap page alignment - requirements - 3. It causes malloc performance to be more dependent on host - system memory management support routines which may vary in - implementation quality and may impose arbitrary - limitations. Generally, servicing a request via normal - malloc steps is faster than going through a system's mmap. - - The advantages of mmap nearly always outweigh disadvantages for - "large" chunks, but the value of "large" varies across systems. The - default is an empirically derived value that works well in most - systems. -*/ - -#define M_MMAP_THRESHOLD -3 - -#ifndef DEFAULT_MMAP_THRESHOLD -#define DEFAULT_MMAP_THRESHOLD (128 * 1024) -#endif - -/* - M_MMAP_MAX is the maximum number of requests to simultaneously - service using mmap. This parameter exists because - some systems have a limited number of internal tables for - use by mmap, and using more than a few of them may degrade - performance. - - The default is set to a value that serves only as a safeguard. - Setting to 0 disables use of mmap for servicing large requests. If - HAVE_MMAP is not set, the default value is 0, and attempts to set it - to non-zero values in mallopt will fail. -*/ - -#define M_MMAP_MAX -4 - -#ifndef DEFAULT_MMAP_MAX -#if HAVE_MMAP -#define DEFAULT_MMAP_MAX (65536) -#else -#define DEFAULT_MMAP_MAX (0) -#endif -#endif - -#ifdef __cplusplus -} /* end of extern "C" */ -#endif - -#include "malloc.h" - -#ifndef BOUNDED_N -#define BOUNDED_N(ptr, sz) (ptr) -#endif -#ifndef RETURN_ADDRESS -#define RETURN_ADDRESS(X_) (NULL) -#endif - -/* On some platforms we can compile internal, not exported functions better. - Let the environment provide a macro and define it to be empty if it - is not available. */ -#ifndef internal_function -# define internal_function -#endif - -/* Forward declarations. */ -struct malloc_chunk; -typedef struct malloc_chunk* mchunkptr; - -/* Internal routines. */ - -#if __STD_C - -Void_t* _int_malloc(mstate, size_t); -void _int_free(mstate, Void_t*); -Void_t* _int_realloc(mstate, Void_t*, size_t); -Void_t* _int_memalign(mstate, size_t, size_t); -Void_t* _int_valloc(mstate, size_t); -static Void_t* _int_pvalloc(mstate, size_t); -/*static Void_t* cALLOc(size_t, size_t);*/ -static Void_t** _int_icalloc(mstate, size_t, size_t, Void_t**); -static Void_t** _int_icomalloc(mstate, size_t, size_t*, Void_t**); -static int mTRIm(size_t); -static size_t mUSABLe(Void_t*); -static int mALLOPt(int, int); - -static Void_t* internal_function mem2mem_check(Void_t *p, size_t sz); -static int internal_function top_check(void); -static void internal_function munmap_chunk(mchunkptr p); -#if HAVE_MREMAP -static mchunkptr internal_function mremap_chunk(mchunkptr p, size_t new_size); -#endif - -static Void_t* malloc_check(size_t sz, const Void_t *caller); -static void free_check(Void_t* mem, const Void_t *caller); -static Void_t* realloc_check(Void_t* oldmem, size_t bytes, - const Void_t *caller); -static Void_t* memalign_check(size_t alignment, size_t bytes, - const Void_t *caller); -#ifndef NO_THREADS -# if USE_STARTER -static Void_t* malloc_starter(size_t sz, const Void_t *caller); -static Void_t* memalign_starter(size_t aln, size_t sz, const Void_t *caller); -static void free_starter(Void_t* mem, const Void_t *caller); -# endif -static Void_t* malloc_atfork(size_t sz, const Void_t *caller); -static void free_atfork(Void_t* mem, const Void_t *caller); -#endif - -#else - -Void_t* _int_malloc(); -void _int_free(); -Void_t* _int_realloc(); -Void_t* _int_memalign(); -Void_t* _int_valloc(); -Void_t* _int_pvalloc(); -/*static Void_t* cALLOc();*/ -static Void_t** _int_icalloc(); -static Void_t** _int_icomalloc(); -static int mTRIm(); -static size_t mUSABLe(); -static int mALLOPt(); - -#endif - - - - -/* ------------- Optional versions of memcopy ---------------- */ - - -#if USE_MEMCPY - -/* - Note: memcpy is ONLY invoked with non-overlapping regions, - so the (usually slower) memmove is not needed. -*/ - -#define MALLOC_COPY(dest, src, nbytes) memcpy(dest, src, nbytes) -#define MALLOC_ZERO(dest, nbytes) memset(dest, 0, nbytes) - -#else /* !USE_MEMCPY */ - -/* Use Duff's device for good zeroing/copying performance. */ - -#define MALLOC_ZERO(charp, nbytes) \ -do { \ - INTERNAL_SIZE_T* mzp = (INTERNAL_SIZE_T*)(charp); \ - unsigned long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T); \ - long mcn; \ - if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \ - switch (mctmp) { \ - case 0: for(;;) { *mzp++ = 0; \ - case 7: *mzp++ = 0; \ - case 6: *mzp++ = 0; \ - case 5: *mzp++ = 0; \ - case 4: *mzp++ = 0; \ - case 3: *mzp++ = 0; \ - case 2: *mzp++ = 0; \ - case 1: *mzp++ = 0; if(mcn <= 0) break; mcn--; } \ - } \ -} while(0) - -#define MALLOC_COPY(dest,src,nbytes) \ -do { \ - INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) src; \ - INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) dest; \ - unsigned long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T); \ - long mcn; \ - if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \ - switch (mctmp) { \ - case 0: for(;;) { *mcdst++ = *mcsrc++; \ - case 7: *mcdst++ = *mcsrc++; \ - case 6: *mcdst++ = *mcsrc++; \ - case 5: *mcdst++ = *mcsrc++; \ - case 4: *mcdst++ = *mcsrc++; \ - case 3: *mcdst++ = *mcsrc++; \ - case 2: *mcdst++ = *mcsrc++; \ - case 1: *mcdst++ = *mcsrc++; if(mcn <= 0) break; mcn--; } \ - } \ -} while(0) - -#endif - -/* ------------------ MMAP support ------------------ */ - - -#if HAVE_MMAP - -#include -#ifndef LACKS_SYS_MMAN_H -#include -#endif - -#if !defined(MAP_ANONYMOUS) && defined(MAP_ANON) -# define MAP_ANONYMOUS MAP_ANON -#endif -#if !defined(MAP_FAILED) -# define MAP_FAILED ((char*)-1) -#endif - -#ifndef MAP_NORESERVE -# ifdef MAP_AUTORESRV -# define MAP_NORESERVE MAP_AUTORESRV -# else -# define MAP_NORESERVE 0 -# endif -#endif - -/* - Nearly all versions of mmap support MAP_ANONYMOUS, - so the following is unlikely to be needed, but is - supplied just in case. -*/ - -#ifndef MAP_ANONYMOUS - -static int dev_zero_fd = -1; /* Cached file descriptor for /dev/zero. */ - -#define MMAP(addr, size, prot, flags) ((dev_zero_fd < 0) ? \ - (dev_zero_fd = open("/dev/zero", O_RDWR), \ - mmap((addr), (size), (prot), (flags), dev_zero_fd, 0)) : \ - mmap((addr), (size), (prot), (flags), dev_zero_fd, 0)) - -#else - -#define MMAP(addr, size, prot, flags) \ - (mmap((addr), (size), (prot), (flags)|MAP_ANONYMOUS, -1, 0)) - -#endif - - -#endif /* HAVE_MMAP */ - - -/* - ----------------------- Chunk representations ----------------------- -*/ - - -/* - This struct declaration is misleading (but accurate and necessary). - It declares a "view" into memory allowing access to necessary - fields at known offsets from a given base. See explanation below. -*/ - -struct malloc_chunk { - - INTERNAL_SIZE_T prev_size; /* Size of previous chunk (if free). */ - INTERNAL_SIZE_T size; /* Size in bytes, including overhead. */ - - struct malloc_chunk* fd; /* double links -- used only if free. */ - struct malloc_chunk* bk; -}; - - -/* - malloc_chunk details: - - (The following includes lightly edited explanations by Colin Plumb.) - - Chunks of memory are maintained using a `boundary tag' method as - described in e.g., Knuth or Standish. (See the paper by Paul - Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a - survey of such techniques.) Sizes of free chunks are stored both - in the front of each chunk and at the end. This makes - consolidating fragmented chunks into bigger chunks very fast. The - size fields also hold bits representing whether chunks are free or - in use. - - An allocated chunk looks like this: - - - chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - | Size of previous chunk, if allocated | | - +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - | Size of chunk, in bytes |P| - mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - | User data starts here... . - . . - . (malloc_usable_space() bytes) . - . | -nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - | Size of chunk | - +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - - - Where "chunk" is the front of the chunk for the purpose of most of - the malloc code, but "mem" is the pointer that is returned to the - user. "Nextchunk" is the beginning of the next contiguous chunk. - - Chunks always begin on even word boundries, so the mem portion - (which is returned to the user) is also on an even word boundary, and - thus at least double-word aligned. - - Free chunks are stored in circular doubly-linked lists, and look like this: - - chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - | Size of previous chunk | - +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - `head:' | Size of chunk, in bytes |P| - mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - | Forward pointer to next chunk in list | - +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - | Back pointer to previous chunk in list | - +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - | Unused space (may be 0 bytes long) . - . . - . | -nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - `foot:' | Size of chunk, in bytes | - +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - - The P (PREV_INUSE) bit, stored in the unused low-order bit of the - chunk size (which is always a multiple of two words), is an in-use - bit for the *previous* chunk. If that bit is *clear*, then the - word before the current chunk size contains the previous chunk - size, and can be used to find the front of the previous chunk. - The very first chunk allocated always has this bit set, - preventing access to non-existent (or non-owned) memory. If - prev_inuse is set for any given chunk, then you CANNOT determine - the size of the previous chunk, and might even get a memory - addressing fault when trying to do so. - - Note that the `foot' of the current chunk is actually represented - as the prev_size of the NEXT chunk. This makes it easier to - deal with alignments etc but can be very confusing when trying - to extend or adapt this code. - - The two exceptions to all this are - - 1. The special chunk `top' doesn't bother using the - trailing size field since there is no next contiguous chunk - that would have to index off it. After initialization, `top' - is forced to always exist. If it would become less than - MINSIZE bytes long, it is replenished. - - 2. Chunks allocated via mmap, which have the second-lowest-order - bit (IS_MMAPPED) set in their size fields. Because they are - allocated one-by-one, each must contain its own trailing size field. - -*/ - -/* - ---------- Size and alignment checks and conversions ---------- -*/ - -/* conversion from malloc headers to user pointers, and back */ - -#define chunk2mem(p) ((Void_t*)((char*)(p) + 2*SIZE_SZ)) -#define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ)) - -/* The smallest possible chunk */ -#define MIN_CHUNK_SIZE (sizeof(struct malloc_chunk)) - -/* The smallest size we can malloc is an aligned minimal chunk */ - -#define MINSIZE \ - (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)) - -/* Check if m has acceptable alignment */ - -#define aligned_OK(m) (((unsigned long)((m)) & (MALLOC_ALIGN_MASK)) == 0) - - -/* - Check if a request is so large that it would wrap around zero when - padded and aligned. To simplify some other code, the bound is made - low enough so that adding MINSIZE will also not wrap around zero. -*/ - -#define REQUEST_OUT_OF_RANGE(req) \ - ((unsigned long)(req) >= \ - (unsigned long)(INTERNAL_SIZE_T)(-2 * MINSIZE)) - -/* pad request bytes into a usable size -- internal version */ - -#define request2size(req) \ - (((req) + SIZE_SZ + MALLOC_ALIGN_MASK < MINSIZE) ? \ - MINSIZE : \ - ((req) + SIZE_SZ + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK) - -/* Same, except also perform argument check */ - -#define checked_request2size(req, sz) \ - if (REQUEST_OUT_OF_RANGE(req)) { \ - MALLOC_FAILURE_ACTION; \ - return 0; \ - } \ - (sz) = request2size(req); - -/* - --------------- Physical chunk operations --------------- -*/ - - -/* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */ -#define PREV_INUSE 0x1 - -/* extract inuse bit of previous chunk */ -#define prev_inuse(p) ((p)->size & PREV_INUSE) - - -/* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */ -#define IS_MMAPPED 0x2 - -/* check for mmap()'ed chunk */ -#define chunk_is_mmapped(p) ((p)->size & IS_MMAPPED) - - -/* size field is or'ed with NON_MAIN_ARENA if the chunk was obtained - from a non-main arena. This is only set immediately before handing - the chunk to the user, if necessary. */ -#define NON_MAIN_ARENA 0x4 - -/* check for chunk from non-main arena */ -#define chunk_non_main_arena(p) ((p)->size & NON_MAIN_ARENA) - - -/* - Bits to mask off when extracting size - - Note: IS_MMAPPED is intentionally not masked off from size field in - macros for which mmapped chunks should never be seen. This should - cause helpful core dumps to occur if it is tried by accident by - people extending or adapting this malloc. -*/ -#define SIZE_BITS (PREV_INUSE|IS_MMAPPED|NON_MAIN_ARENA) - -/* Get size, ignoring use bits */ -#define chunksize(p) ((p)->size & ~(SIZE_BITS)) - - -/* Ptr to next physical malloc_chunk. */ -#define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->size & ~SIZE_BITS) )) - -/* Ptr to previous physical malloc_chunk */ -#define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_size) )) - -/* Treat space at ptr + offset as a chunk */ -#define chunk_at_offset(p, s) ((mchunkptr)(((char*)(p)) + (s))) - -/* extract p's inuse bit */ -#define inuse(p)\ -((((mchunkptr)(((char*)(p))+((p)->size & ~SIZE_BITS)))->size) & PREV_INUSE) - -/* set/clear chunk as being inuse without otherwise disturbing */ -#define set_inuse(p)\ -((mchunkptr)(((char*)(p)) + ((p)->size & ~SIZE_BITS)))->size |= PREV_INUSE - -#define clear_inuse(p)\ -((mchunkptr)(((char*)(p)) + ((p)->size & ~SIZE_BITS)))->size &= ~(PREV_INUSE) - - -/* check/set/clear inuse bits in known places */ -#define inuse_bit_at_offset(p, s)\ - (((mchunkptr)(((char*)(p)) + (s)))->size & PREV_INUSE) - -#define set_inuse_bit_at_offset(p, s)\ - (((mchunkptr)(((char*)(p)) + (s)))->size |= PREV_INUSE) - -#define clear_inuse_bit_at_offset(p, s)\ - (((mchunkptr)(((char*)(p)) + (s)))->size &= ~(PREV_INUSE)) - - -/* Set size at head, without disturbing its use bit */ -#define set_head_size(p, s) ((p)->size = (((p)->size & SIZE_BITS) | (s))) - -/* Set size/use field */ -#define set_head(p, s) ((p)->size = (s)) - -/* Set size at footer (only when chunk is not in use) */ -#define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_size = (s)) - - -/* - -------------------- Internal data structures -------------------- - - All internal state is held in an instance of malloc_state defined - below. There are no other static variables, except in two optional - cases: - * If USE_MALLOC_LOCK is defined, the mALLOC_MUTEx declared above. - * If HAVE_MMAP is true, but mmap doesn't support - MAP_ANONYMOUS, a dummy file descriptor for mmap. - - Beware of lots of tricks that minimize the total bookkeeping space - requirements. The result is a little over 1K bytes (for 4byte - pointers and size_t.) -*/ - -/* - Bins - - An array of bin headers for free chunks. Each bin is doubly - linked. The bins are approximately proportionally (log) spaced. - There are a lot of these bins (128). This may look excessive, but - works very well in practice. Most bins hold sizes that are - unusual as malloc request sizes, but are more usual for fragments - and consolidated sets of chunks, which is what these bins hold, so - they can be found quickly. All procedures maintain the invariant - that no consolidated chunk physically borders another one, so each - chunk in a list is known to be preceeded and followed by either - inuse chunks or the ends of memory. - - Chunks in bins are kept in size order, with ties going to the - approximately least recently used chunk. Ordering isn't needed - for the small bins, which all contain the same-sized chunks, but - facilitates best-fit allocation for larger chunks. These lists - are just sequential. Keeping them in order almost never requires - enough traversal to warrant using fancier ordered data - structures. - - Chunks of the same size are linked with the most - recently freed at the front, and allocations are taken from the - back. This results in LRU (FIFO) allocation order, which tends - to give each chunk an equal opportunity to be consolidated with - adjacent freed chunks, resulting in larger free chunks and less - fragmentation. - - To simplify use in double-linked lists, each bin header acts - as a malloc_chunk. This avoids special-casing for headers. - But to conserve space and improve locality, we allocate - only the fd/bk pointers of bins, and then use repositioning tricks - to treat these as the fields of a malloc_chunk*. -*/ - -typedef struct malloc_chunk* mbinptr; - -/* addressing -- note that bin_at(0) does not exist */ -#define bin_at(m, i) ((mbinptr)((char*)&((m)->bins[(i)<<1]) - (SIZE_SZ<<1))) - -/* analog of ++bin */ -#define next_bin(b) ((mbinptr)((char*)(b) + (sizeof(mchunkptr)<<1))) - -/* Reminders about list directionality within bins */ -#define first(b) ((b)->fd) -#define last(b) ((b)->bk) - -/* Take a chunk off a bin list */ -#define unlink(P, BK, FD) { \ - FD = P->fd; \ - BK = P->bk; \ - FD->bk = BK; \ - BK->fd = FD; \ -} - -/* - Indexing - - Bins for sizes < 512 bytes contain chunks of all the same size, spaced - 8 bytes apart. Larger bins are approximately logarithmically spaced: - - 64 bins of size 8 - 32 bins of size 64 - 16 bins of size 512 - 8 bins of size 4096 - 4 bins of size 32768 - 2 bins of size 262144 - 1 bin of size what's left - - There is actually a little bit of slop in the numbers in bin_index - for the sake of speed. This makes no difference elsewhere. - - The bins top out around 1MB because we expect to service large - requests via mmap. -*/ - -#define NBINS 128 -#define NSMALLBINS 64 -#define SMALLBIN_WIDTH 8 -#define MIN_LARGE_SIZE 512 - -#define in_smallbin_range(sz) \ - ((unsigned long)(sz) < (unsigned long)MIN_LARGE_SIZE) - -#define smallbin_index(sz) (((unsigned)(sz)) >> 3) - -#define largebin_index(sz) \ -(((((unsigned long)(sz)) >> 6) <= 32)? 56 + (((unsigned long)(sz)) >> 6): \ - ((((unsigned long)(sz)) >> 9) <= 20)? 91 + (((unsigned long)(sz)) >> 9): \ - ((((unsigned long)(sz)) >> 12) <= 10)? 110 + (((unsigned long)(sz)) >> 12): \ - ((((unsigned long)(sz)) >> 15) <= 4)? 119 + (((unsigned long)(sz)) >> 15): \ - ((((unsigned long)(sz)) >> 18) <= 2)? 124 + (((unsigned long)(sz)) >> 18): \ - 126) - -#define bin_index(sz) \ - ((in_smallbin_range(sz)) ? smallbin_index(sz) : largebin_index(sz)) - -/* - FIRST_SORTED_BIN_SIZE is the chunk size corresponding to the - first bin that is maintained in sorted order. This must - be the smallest size corresponding to a given bin. - - Normally, this should be MIN_LARGE_SIZE. But you can weaken - best fit guarantees to sometimes speed up malloc by increasing value. - Doing this means that malloc may choose a chunk that is - non-best-fitting by up to the width of the bin. - - Some useful cutoff values: - 512 - all bins sorted - 2560 - leaves bins <= 64 bytes wide unsorted - 12288 - leaves bins <= 512 bytes wide unsorted - 65536 - leaves bins <= 4096 bytes wide unsorted - 262144 - leaves bins <= 32768 bytes wide unsorted - -1 - no bins sorted (not recommended!) -*/ - -#define FIRST_SORTED_BIN_SIZE MIN_LARGE_SIZE -/* #define FIRST_SORTED_BIN_SIZE 65536 */ - -/* - Unsorted chunks - - All remainders from chunk splits, as well as all returned chunks, - are first placed in the "unsorted" bin. They are then placed - in regular bins after malloc gives them ONE chance to be used before - binning. So, basically, the unsorted_chunks list acts as a queue, - with chunks being placed on it in free (and malloc_consolidate), - and taken off (to be either used or placed in bins) in malloc. - - The NON_MAIN_ARENA flag is never set for unsorted chunks, so it - does not have to be taken into account in size comparisons. -*/ - -/* The otherwise unindexable 1-bin is used to hold unsorted chunks. */ -#define unsorted_chunks(M) (bin_at(M, 1)) - -/* - Top - - The top-most available chunk (i.e., the one bordering the end of - available memory) is treated specially. It is never included in - any bin, is used only if no other chunk is available, and is - released back to the system if it is very large (see - M_TRIM_THRESHOLD). Because top initially - points to its own bin with initial zero size, thus forcing - extension on the first malloc request, we avoid having any special - code in malloc to check whether it even exists yet. But we still - need to do so when getting memory from system, so we make - initial_top treat the bin as a legal but unusable chunk during the - interval between initialization and the first call to - sYSMALLOc. (This is somewhat delicate, since it relies on - the 2 preceding words to be zero during this interval as well.) -*/ - -/* Conveniently, the unsorted bin can be used as dummy top on first call */ -#define initial_top(M) (unsorted_chunks(M)) - -/* - Binmap - - To help compensate for the large number of bins, a one-level index - structure is used for bin-by-bin searching. `binmap' is a - bitvector recording whether bins are definitely empty so they can - be skipped over during during traversals. The bits are NOT always - cleared as soon as bins are empty, but instead only - when they are noticed to be empty during traversal in malloc. -*/ - -/* Conservatively use 32 bits per map word, even if on 64bit system */ -#define BINMAPSHIFT 5 -#define BITSPERMAP (1U << BINMAPSHIFT) -#define BINMAPSIZE (NBINS / BITSPERMAP) - -#define idx2block(i) ((i) >> BINMAPSHIFT) -#define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT)-1)))) - -#define mark_bin(m,i) ((m)->binmap[idx2block(i)] |= idx2bit(i)) -#define unmark_bin(m,i) ((m)->binmap[idx2block(i)] &= ~(idx2bit(i))) -#define get_binmap(m,i) ((m)->binmap[idx2block(i)] & idx2bit(i)) - -/* - Fastbins - - An array of lists holding recently freed small chunks. Fastbins - are not doubly linked. It is faster to single-link them, and - since chunks are never removed from the middles of these lists, - double linking is not necessary. Also, unlike regular bins, they - are not even processed in FIFO order (they use faster LIFO) since - ordering doesn't much matter in the transient contexts in which - fastbins are normally used. - - Chunks in fastbins keep their inuse bit set, so they cannot - be consolidated with other free chunks. malloc_consolidate - releases all chunks in fastbins and consolidates them with - other free chunks. -*/ - -typedef struct malloc_chunk* mfastbinptr; - -/* offset 2 to use otherwise unindexable first 2 bins */ -#define fastbin_index(sz) ((((unsigned int)(sz)) >> 3) - 2) - -/* The maximum fastbin request size we support */ -#define MAX_FAST_SIZE 80 - -#define NFASTBINS (fastbin_index(request2size(MAX_FAST_SIZE))+1) - -/* - FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free() - that triggers automatic consolidation of possibly-surrounding - fastbin chunks. This is a heuristic, so the exact value should not - matter too much. It is defined at half the default trim threshold as a - compromise heuristic to only attempt consolidation if it is likely - to lead to trimming. However, it is not dynamically tunable, since - consolidation reduces fragmentation surrounding large chunks even - if trimming is not used. -*/ - -#define FASTBIN_CONSOLIDATION_THRESHOLD (65536UL) - -/* - Since the lowest 2 bits in max_fast don't matter in size comparisons, - they are used as flags. -*/ - -/* - FASTCHUNKS_BIT held in max_fast indicates that there are probably - some fastbin chunks. It is set true on entering a chunk into any - fastbin, and cleared only in malloc_consolidate. - - The truth value is inverted so that have_fastchunks will be true - upon startup (since statics are zero-filled), simplifying - initialization checks. -*/ - -#define FASTCHUNKS_BIT (1U) - -#define have_fastchunks(M) (((M)->max_fast & FASTCHUNKS_BIT) == 0) -#define clear_fastchunks(M) ((M)->max_fast |= FASTCHUNKS_BIT) -#define set_fastchunks(M) ((M)->max_fast &= ~FASTCHUNKS_BIT) - -/* - NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous - regions. Otherwise, contiguity is exploited in merging together, - when possible, results from consecutive MORECORE calls. - - The initial value comes from MORECORE_CONTIGUOUS, but is - changed dynamically if mmap is ever used as an sbrk substitute. -*/ - -#define NONCONTIGUOUS_BIT (2U) - -#define contiguous(M) (((M)->max_fast & NONCONTIGUOUS_BIT) == 0) -#define noncontiguous(M) (((M)->max_fast & NONCONTIGUOUS_BIT) != 0) -#define set_noncontiguous(M) ((M)->max_fast |= NONCONTIGUOUS_BIT) -#define set_contiguous(M) ((M)->max_fast &= ~NONCONTIGUOUS_BIT) - -/* - Set value of max_fast. - Use impossibly small value if 0. - Precondition: there are no existing fastbin chunks. - Setting the value clears fastchunk bit but preserves noncontiguous bit. -*/ - -#define set_max_fast(M, s) \ - (M)->max_fast = (((s) == 0)? SMALLBIN_WIDTH: request2size(s)) | \ - FASTCHUNKS_BIT | \ - ((M)->max_fast & NONCONTIGUOUS_BIT) - - -/* - ----------- Internal state representation and initialization ----------- -*/ - -struct malloc_state { - /* Serialize access. */ - mutex_t mutex; - - /* Statistics for locking. Only used if THREAD_STATS is defined. */ - long stat_lock_direct, stat_lock_loop, stat_lock_wait; - long pad0_[1]; /* try to give the mutex its own cacheline */ - - /* The maximum chunk size to be eligible for fastbin */ - INTERNAL_SIZE_T max_fast; /* low 2 bits used as flags */ - - /* Fastbins */ - mfastbinptr fastbins[NFASTBINS]; - - /* Base of the topmost chunk -- not otherwise kept in a bin */ - mchunkptr top; - - /* The remainder from the most recent split of a small request */ - mchunkptr last_remainder; - - /* Normal bins packed as described above */ - mchunkptr bins[NBINS * 2]; - - /* Bitmap of bins */ - unsigned int binmap[BINMAPSIZE]; - - /* Linked list */ - struct malloc_state *next; - - /* Memory allocated from the system in this arena. */ - INTERNAL_SIZE_T system_mem; - INTERNAL_SIZE_T max_system_mem; -}; - -struct malloc_par { - /* Tunable parameters */ - unsigned long trim_threshold; - INTERNAL_SIZE_T top_pad; - INTERNAL_SIZE_T mmap_threshold; - - /* Memory map support */ - int n_mmaps; - int n_mmaps_max; - int max_n_mmaps; - - /* Cache malloc_getpagesize */ - unsigned int pagesize; - - /* Statistics */ - INTERNAL_SIZE_T mmapped_mem; - /*INTERNAL_SIZE_T sbrked_mem;*/ - /*INTERNAL_SIZE_T max_sbrked_mem;*/ - INTERNAL_SIZE_T max_mmapped_mem; - INTERNAL_SIZE_T max_total_mem; /* only kept for NO_THREADS */ - - /* First address handed out by MORECORE/sbrk. */ - char* sbrk_base; -}; - -/* There are several instances of this struct ("arenas") in this - malloc. If you are adapting this malloc in a way that does NOT use - a static or mmapped malloc_state, you MUST explicitly zero-fill it - before using. This malloc relies on the property that malloc_state - is initialized to all zeroes (as is true of C statics). */ - -static struct malloc_state main_arena; - -/* There is only one instance of the malloc parameters. */ - -static struct malloc_par mp_; - -/* - Initialize a malloc_state struct. - - This is called only from within malloc_consolidate, which needs - be called in the same contexts anyway. It is never called directly - outside of malloc_consolidate because some optimizing compilers try - to inline it at all call points, which turns out not to be an - optimization at all. (Inlining it in malloc_consolidate is fine though.) -*/ - -#if __STD_C -static void malloc_init_state(mstate av) -#else -static void malloc_init_state(av) mstate av; -#endif -{ - int i; - mbinptr bin; - - /* Establish circular links for normal bins */ - for (i = 1; i < NBINS; ++i) { - bin = bin_at(av,i); - bin->fd = bin->bk = bin; - } - -#if MORECORE_CONTIGUOUS - if (av != &main_arena) -#endif - set_noncontiguous(av); - - set_max_fast(av, DEFAULT_MXFAST); - - av->top = initial_top(av); -} - -/* - Other internal utilities operating on mstates -*/ - -#if __STD_C -static Void_t* sYSMALLOc(INTERNAL_SIZE_T, mstate); -#ifndef MORECORE_CANNOT_TRIM -static int sYSTRIm(size_t, mstate); -#endif -static void malloc_consolidate(mstate); -static Void_t** iALLOc(mstate, size_t, size_t*, int, Void_t**); -#else -static Void_t* sYSMALLOc(); -#ifndef MORECORE_CANNOT_TRIM -static int sYSTRIm(); -#endif -static void malloc_consolidate(); -static Void_t** iALLOc(); -#endif - - -/* -------------- Early definitions for debugging hooks ---------------- */ - -/* Define and initialize the hook variables. These weak definitions must - appear before any use of the variables in a function (arena.c uses one). */ -#ifndef weak_variable -#ifndef _LIBC -#define weak_variable /**/ -#else -/* In GNU libc we want the hook variables to be weak definitions to - avoid a problem with Emacs. */ -#define weak_variable weak_function -#endif -#endif - -/* OMPI change: these aren't used */ -#if 0 -#if !(USE_STARTER & 2) -# define free_hook_ini NULL -/* Forward declarations. */ -static Void_t* malloc_hook_ini __MALLOC_P ((size_t sz, - const __malloc_ptr_t caller)); -static Void_t* realloc_hook_ini __MALLOC_P ((Void_t* ptr, size_t sz, - const __malloc_ptr_t caller)); -static Void_t* memalign_hook_ini __MALLOC_P ((size_t alignment, size_t sz, - const __malloc_ptr_t caller)); -#else -# define free_hook_ini free_starter -# define malloc_hook_ini malloc_starter -# define realloc_hook_ini NULL -# define memalign_hook_ini memalign_starter -#endif -#endif - -/* OMPI change: we don't want any of these -- we want to use the - underlying allocator's symbols */ -#if 0 -void weak_variable (*__malloc_initialize_hook) __MALLOC_PMT ((void)) = NULL; -void weak_variable (*__free_hook) __MALLOC_PMT ((__malloc_ptr_t __ptr, - const __malloc_ptr_t)) - = free_hook_ini; -__malloc_ptr_t weak_variable (*__malloc_hook) - __MALLOC_PMT ((size_t __size, const __malloc_ptr_t)) = malloc_hook_ini; -__malloc_ptr_t weak_variable (*__realloc_hook) - __MALLOC_PMT ((__malloc_ptr_t __ptr, size_t __size, const __malloc_ptr_t)) - = realloc_hook_ini; -__malloc_ptr_t weak_variable (*__memalign_hook) - __MALLOC_PMT ((size_t __alignment, size_t __size, const __malloc_ptr_t)) - = memalign_hook_ini; -void weak_variable (*__after_morecore_hook) __MALLOC_P ((void)) = NULL; -#endif - - -/* ------------------- Support for multiple arenas -------------------- */ -#include "arena.c" - -/* - Debugging support - - These routines make a number of assertions about the states - of data structures that should be true at all times. If any - are not true, it's very likely that a user program has somehow - trashed memory. (It's also possible that there is a coding error - in malloc. In which case, please report it!) -*/ - -#if ! MALLOC_DEBUG - -#define check_chunk(A,P) -#define check_free_chunk(A,P) -#define check_inuse_chunk(A,P) -#define check_remalloced_chunk(A,P,N) -#define check_malloced_chunk(A,P,N) -#define check_malloc_state(A) - -#else - -#define check_chunk(A,P) do_check_chunk(A,P) -#define check_free_chunk(A,P) do_check_free_chunk(A,P) -#define check_inuse_chunk(A,P) do_check_inuse_chunk(A,P) -#define check_remalloced_chunk(A,P,N) do_check_remalloced_chunk(A,P,N) -#define check_malloced_chunk(A,P,N) do_check_malloced_chunk(A,P,N) -#define check_malloc_state(A) do_check_malloc_state(A) - -/* - Properties of all chunks -*/ - -#if __STD_C -static void do_check_chunk(mstate av, mchunkptr p) -#else -static void do_check_chunk(av, p) mstate av; mchunkptr p; -#endif -{ - unsigned long sz = chunksize(p); - /* min and max possible addresses assuming contiguous allocation */ - char* max_address = (char*)(av->top) + chunksize(av->top); - char* min_address = max_address - av->system_mem; - - if (!chunk_is_mmapped(p)) { - - /* Has legal address ... */ - if (p != av->top) { - if (contiguous(av)) { - assert(((char*)p) >= min_address); - assert(((char*)p + sz) <= ((char*)(av->top))); - } - } - else { - /* top size is always at least MINSIZE */ - assert((unsigned long)(sz) >= MINSIZE); - /* top predecessor always marked inuse */ - assert(prev_inuse(p)); - } - - } - else { -#if HAVE_MMAP - /* address is outside main heap */ - if (contiguous(av) && av->top != initial_top(av)) { - assert(((char*)p) < min_address || ((char*)p) > max_address); - } - /* chunk is page-aligned */ - assert(((p->prev_size + sz) & (mp_.pagesize-1)) == 0); - /* mem is aligned */ - assert(aligned_OK(chunk2mem(p))); -#else - /* force an appropriate assert violation if debug set */ - assert(!chunk_is_mmapped(p)); -#endif - } -} - -/* - Properties of free chunks -*/ - -#if __STD_C -static void do_check_free_chunk(mstate av, mchunkptr p) -#else -static void do_check_free_chunk(av, p) mstate av; mchunkptr p; -#endif -{ - INTERNAL_SIZE_T sz = p->size & ~(PREV_INUSE|NON_MAIN_ARENA); - mchunkptr next = chunk_at_offset(p, sz); - - do_check_chunk(av, p); - - /* Chunk must claim to be free ... */ - assert(!inuse(p)); - assert (!chunk_is_mmapped(p)); - - /* Unless a special marker, must have OK fields */ - if ((unsigned long)(sz) >= MINSIZE) - { - assert((sz & MALLOC_ALIGN_MASK) == 0); - assert(aligned_OK(chunk2mem(p))); - /* ... matching footer field */ - assert(next->prev_size == sz); - /* ... and is fully consolidated */ - assert(prev_inuse(p)); - assert (next == av->top || inuse(next)); - - /* ... and has minimally sane links */ - assert(p->fd->bk == p); - assert(p->bk->fd == p); - } - else /* markers are always of size SIZE_SZ */ - assert(sz == SIZE_SZ); -} - -/* - Properties of inuse chunks -*/ - -#if __STD_C -static void do_check_inuse_chunk(mstate av, mchunkptr p) -#else -static void do_check_inuse_chunk(av, p) mstate av; mchunkptr p; -#endif -{ - mchunkptr next; - - do_check_chunk(av, p); - - if (chunk_is_mmapped(p)) - return; /* mmapped chunks have no next/prev */ - - /* Check whether it claims to be in use ... */ - assert(inuse(p)); - - next = next_chunk(p); - - /* ... and is surrounded by OK chunks. - Since more things can be checked with free chunks than inuse ones, - if an inuse chunk borders them and debug is on, it's worth doing them. - */ - if (!prev_inuse(p)) { - /* Note that we cannot even look at prev unless it is not inuse */ - mchunkptr prv = prev_chunk(p); - assert(next_chunk(prv) == p); - do_check_free_chunk(av, prv); - } - - if (next == av->top) { - assert(prev_inuse(next)); - assert(chunksize(next) >= MINSIZE); - } - else if (!inuse(next)) - do_check_free_chunk(av, next); -} - -/* - Properties of chunks recycled from fastbins -*/ - -#if __STD_C -static void do_check_remalloced_chunk(mstate av, mchunkptr p, INTERNAL_SIZE_T s) -#else -static void do_check_remalloced_chunk(av, p, s) -mstate av; mchunkptr p; INTERNAL_SIZE_T s; -#endif -{ - INTERNAL_SIZE_T sz = p->size & ~(PREV_INUSE|NON_MAIN_ARENA); - - if (!chunk_is_mmapped(p)) { - assert(av == arena_for_chunk(p)); - if (chunk_non_main_arena(p)) - assert(av != &main_arena); - else - assert(av == &main_arena); - } - - do_check_inuse_chunk(av, p); - - /* Legal size ... */ - assert((sz & MALLOC_ALIGN_MASK) == 0); - assert((unsigned long)(sz) >= MINSIZE); - /* ... and alignment */ - assert(aligned_OK(chunk2mem(p))); - /* chunk is less than MINSIZE more than request */ - assert((long)(sz) - (long)(s) >= 0); - assert((long)(sz) - (long)(s + MINSIZE) < 0); -} - -/* - Properties of nonrecycled chunks at the point they are malloced -*/ - -#if __STD_C -static void do_check_malloced_chunk(mstate av, mchunkptr p, INTERNAL_SIZE_T s) -#else -static void do_check_malloced_chunk(av, p, s) -mstate av; mchunkptr p; INTERNAL_SIZE_T s; -#endif -{ - /* same as recycled case ... */ - do_check_remalloced_chunk(av, p, s); - - /* - ... plus, must obey implementation invariant that prev_inuse is - always true of any allocated chunk; i.e., that each allocated - chunk borders either a previously allocated and still in-use - chunk, or the base of its memory arena. This is ensured - by making all allocations from the the `lowest' part of any found - chunk. This does not necessarily hold however for chunks - recycled via fastbins. - */ - - assert(prev_inuse(p)); -} - - -/* - Properties of malloc_state. - - This may be useful for debugging malloc, as well as detecting user - programmer errors that somehow write into malloc_state. - - If you are extending or experimenting with this malloc, you can - probably figure out how to hack this routine to print out or - display chunk addresses, sizes, bins, and other instrumentation. -*/ - -static void do_check_malloc_state(mstate av) -{ - int i; - mchunkptr p; - mchunkptr q; - mbinptr b; - unsigned int binbit; - int empty; - unsigned int idx; - INTERNAL_SIZE_T size; - unsigned long total = 0; - int max_fast_bin; - - /* internal size_t must be no wider than pointer type */ - assert(sizeof(INTERNAL_SIZE_T) <= sizeof(char*)); - - /* alignment is a power of 2 */ - assert((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT-1)) == 0); - - /* cannot run remaining checks until fully initialized */ - if (av->top == 0 || av->top == initial_top(av)) - return; - - /* pagesize is a power of 2 */ - assert((mp_.pagesize & (mp_.pagesize-1)) == 0); - - /* A contiguous main_arena is consistent with sbrk_base. */ - if (av == &main_arena && contiguous(av)) - assert((char*)mp_.sbrk_base + av->system_mem == - (char*)av->top + chunksize(av->top)); - - /* properties of fastbins */ - - /* max_fast is in allowed range */ - assert((av->max_fast & ~1) <= request2size(MAX_FAST_SIZE)); - - max_fast_bin = fastbin_index(av->max_fast); - - for (i = 0; i < NFASTBINS; ++i) { - p = av->fastbins[i]; - - /* all bins past max_fast are empty */ - if (i > max_fast_bin) - assert(p == 0); - - while (p != 0) { - /* each chunk claims to be inuse */ - do_check_inuse_chunk(av, p); - total += chunksize(p); - /* chunk belongs in this bin */ - assert(fastbin_index(chunksize(p)) == i); - p = p->fd; - } - } - - if (total != 0) - assert(have_fastchunks(av)); - else if (!have_fastchunks(av)) - assert(total == 0); - - /* check normal bins */ - for (i = 1; i < NBINS; ++i) { - b = bin_at(av,i); - - /* binmap is accurate (except for bin 1 == unsorted_chunks) */ - if (i >= 2) { - binbit = get_binmap(av,i); - empty = last(b) == b; - if (!binbit) - assert(empty); - else if (!empty) - assert(binbit); - } - - for (p = last(b); p != b; p = p->bk) { - /* each chunk claims to be free */ - do_check_free_chunk(av, p); - size = chunksize(p); - total += size; - if (i >= 2) { - /* chunk belongs in bin */ - idx = bin_index(size); - assert(idx == i); - /* lists are sorted */ - if ((unsigned long) size >= (unsigned long)(FIRST_SORTED_BIN_SIZE)) { - assert(p->bk == b || - (unsigned long)chunksize(p->bk) >= - (unsigned long)chunksize(p)); - } - } - /* chunk is followed by a legal chain of inuse chunks */ - for (q = next_chunk(p); - (q != av->top && inuse(q) && - (unsigned long)(chunksize(q)) >= MINSIZE); - q = next_chunk(q)) - do_check_inuse_chunk(av, q); - } - } - - /* top chunk is OK */ - check_chunk(av, av->top); - - /* sanity checks for statistics */ - -#ifdef NO_THREADS - assert(total <= (unsigned long)(mp_.max_total_mem)); - assert(mp_.n_mmaps >= 0); -#endif - assert(mp_.n_mmaps <= mp_.n_mmaps_max); - assert(mp_.n_mmaps <= mp_.max_n_mmaps); - - assert((unsigned long)(av->system_mem) <= - (unsigned long)(av->max_system_mem)); - - assert((unsigned long)(mp_.mmapped_mem) <= - (unsigned long)(mp_.max_mmapped_mem)); - -#ifdef NO_THREADS - assert((unsigned long)(mp_.max_total_mem) >= - (unsigned long)(mp_.mmapped_mem) + (unsigned long)(av->system_mem)); -#endif -} -#endif - - -/* ----------------- Support for debugging hooks -------------------- */ -#include "hooks.c" - -/* ----------- Routines dealing with system allocation -------------- */ - -/* - sysmalloc handles malloc cases requiring more memory from the system. - On entry, it is assumed that av->top does not have enough - space to service request for nb bytes, thus requiring that av->top - be extended or replaced. -*/ - -#if __STD_C -static Void_t* sYSMALLOc(INTERNAL_SIZE_T nb, mstate av) -#else -static Void_t* sYSMALLOc(nb, av) INTERNAL_SIZE_T nb; mstate av; -#endif -{ - mchunkptr old_top; /* incoming value of av->top */ - INTERNAL_SIZE_T old_size; /* its size */ - char* old_end; /* its end address */ - - long size; /* arg to first MORECORE or mmap call */ - char* brk; /* return value from MORECORE */ - - long correction; /* arg to 2nd MORECORE call */ - char* snd_brk; /* 2nd return val */ - - INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of new space */ - INTERNAL_SIZE_T end_misalign; /* partial page left at end of new space */ - char* aligned_brk; /* aligned offset into brk */ - - mchunkptr p; /* the allocated/returned chunk */ - mchunkptr remainder; /* remainder from allocation */ - unsigned long remainder_size; /* its size */ - - unsigned long sum; /* for updating stats */ - - size_t pagemask = mp_.pagesize - 1; - - -#if HAVE_MMAP - - /* - If have mmap, and the request size meets the mmap threshold, and - the system supports mmap, and there are few enough currently - allocated mmapped regions, try to directly map this request - rather than expanding top. - */ - - if ((unsigned long)(nb) >= (unsigned long)(mp_.mmap_threshold) && - (mp_.n_mmaps < mp_.n_mmaps_max)) { - - char* mm; /* return value from mmap call*/ - - /* - Round up size to nearest page. For mmapped chunks, the overhead - is one SIZE_SZ unit larger than for normal chunks, because there - is no following chunk whose prev_size field could be used. - */ - size = (nb + SIZE_SZ + MALLOC_ALIGN_MASK + pagemask) & ~pagemask; - - /* Don't try if size wraps around 0 */ - if ((unsigned long)(size) > (unsigned long)(nb)) { - - mm = (char*)(MMAP(0, size, PROT_READ|PROT_WRITE, MAP_PRIVATE)); - - if (mm != MAP_FAILED) { - - /* - The offset to the start of the mmapped region is stored - in the prev_size field of the chunk. This allows us to adjust - returned start address to meet alignment requirements here - and in memalign(), and still be able to compute proper - address argument for later munmap in free() and realloc(). - */ - - front_misalign = (INTERNAL_SIZE_T)chunk2mem(mm) & MALLOC_ALIGN_MASK; - if (front_misalign > 0) { - correction = MALLOC_ALIGNMENT - front_misalign; - p = (mchunkptr)(mm + correction); - p->prev_size = correction; - set_head(p, (size - correction) |IS_MMAPPED); - } - else { - p = (mchunkptr)mm; - set_head(p, size|IS_MMAPPED); - } - - /* update statistics */ - - if (++mp_.n_mmaps > mp_.max_n_mmaps) - mp_.max_n_mmaps = mp_.n_mmaps; - - sum = mp_.mmapped_mem += size; - if (sum > (unsigned long)(mp_.max_mmapped_mem)) - mp_.max_mmapped_mem = sum; -#ifdef NO_THREADS - sum += av->system_mem; - if (sum > (unsigned long)(mp_.max_total_mem)) - mp_.max_total_mem = sum; -#endif - - check_chunk(av, p); - - return chunk2mem(p); - } - } - } -#endif - - /* Record incoming configuration of top */ - - old_top = av->top; - old_size = chunksize(old_top); - old_end = (char*)(chunk_at_offset(old_top, old_size)); - - brk = snd_brk = (char*)(MORECORE_FAILURE); - - /* - If not the first time through, we require old_size to be - at least MINSIZE and to have prev_inuse set. - */ - - assert((old_top == initial_top(av) && old_size == 0) || - ((unsigned long) (old_size) >= MINSIZE && - prev_inuse(old_top) && - ((unsigned long)old_end & pagemask) == 0)); - - /* Precondition: not enough current space to satisfy nb request */ - assert((unsigned long)(old_size) < (unsigned long)(nb + MINSIZE)); - - /* Precondition: all fastbins are consolidated */ - assert(!have_fastchunks(av)); - - - if (av != &main_arena) { - - heap_info *old_heap, *heap; - size_t old_heap_size; - - /* First try to extend the current heap. */ - old_heap = heap_for_ptr(old_top); - old_heap_size = old_heap->size; - if (grow_heap(old_heap, MINSIZE + nb - old_size) == 0) { - av->system_mem += old_heap->size - old_heap_size; - arena_mem += old_heap->size - old_heap_size; -#if 0 - if(mmapped_mem + arena_mem + sbrked_mem > max_total_mem) - max_total_mem = mmapped_mem + arena_mem + sbrked_mem; -#endif - set_head(old_top, (((char *)old_heap + old_heap->size) - (char *)old_top) - | PREV_INUSE); - } - else if ((heap = new_heap(nb + (MINSIZE + sizeof(*heap)), mp_.top_pad))) { - /* Use a newly allocated heap. */ - heap->ar_ptr = av; - heap->prev = old_heap; - av->system_mem += heap->size; - arena_mem += heap->size; -#if 0 - if((unsigned long)(mmapped_mem + arena_mem + sbrked_mem) > max_total_mem) - max_total_mem = mmapped_mem + arena_mem + sbrked_mem; -#endif - /* Set up the new top. */ - top(av) = chunk_at_offset(heap, sizeof(*heap)); - set_head(top(av), (heap->size - sizeof(*heap)) | PREV_INUSE); - - /* Setup fencepost and free the old top chunk. */ - /* The fencepost takes at least MINSIZE bytes, because it might - become the top chunk again later. Note that a footer is set - up, too, although the chunk is marked in use. */ - old_size -= MINSIZE; - set_head(chunk_at_offset(old_top, old_size + 2*SIZE_SZ), 0|PREV_INUSE); - if (old_size >= MINSIZE) { - set_head(chunk_at_offset(old_top, old_size), (2*SIZE_SZ)|PREV_INUSE); - set_foot(chunk_at_offset(old_top, old_size), (2*SIZE_SZ)); - set_head(old_top, old_size|PREV_INUSE|NON_MAIN_ARENA); - _int_free(av, chunk2mem(old_top)); - } else { - set_head(old_top, (old_size + 2*SIZE_SZ)|PREV_INUSE); - set_foot(old_top, (old_size + 2*SIZE_SZ)); - } - } - - } else { /* av == main_arena */ - - - /* Request enough space for nb + pad + overhead */ - - size = nb + mp_.top_pad + MINSIZE; - - /* - If contiguous, we can subtract out existing space that we hope to - combine with new space. We add it back later only if - we don't actually get contiguous space. - */ - - if (contiguous(av)) - size -= old_size; - - /* - Round to a multiple of page size. - If MORECORE is not contiguous, this ensures that we only call it - with whole-page arguments. And if MORECORE is contiguous and - this is not first time through, this preserves page-alignment of - previous calls. Otherwise, we correct to page-align below. - */ - - size = (size + pagemask) & ~pagemask; - - /* - Don't try to call MORECORE if argument is so big as to appear - negative. Note that since mmap takes size_t arg, it may succeed - below even if we cannot call MORECORE. - */ - - if (size > 0) - brk = (char*)(MORECORE(size)); - - if (brk != (char*)(MORECORE_FAILURE)) { - /* Call the `morecore' hook if necessary. */ - if (__after_morecore_hook) - (*__after_morecore_hook) (); - } else { - /* - If have mmap, try using it as a backup when MORECORE fails or - cannot be used. This is worth doing on systems that have "holes" in - address space, so sbrk cannot extend to give contiguous space, but - space is available elsewhere. Note that we ignore mmap max count - and threshold limits, since the space will not be used as a - segregated mmap region. - */ - -#if HAVE_MMAP - /* Cannot merge with old top, so add its size back in */ - if (contiguous(av)) - size = (size + old_size + pagemask) & ~pagemask; - - /* If we are relying on mmap as backup, then use larger units */ - if ((unsigned long)(size) < (unsigned long)(MMAP_AS_MORECORE_SIZE)) - size = MMAP_AS_MORECORE_SIZE; - - /* Don't try if size wraps around 0 */ - if ((unsigned long)(size) > (unsigned long)(nb)) { - - char *mbrk = (char*)(MMAP(0, size, PROT_READ|PROT_WRITE, MAP_PRIVATE)); - - if (mbrk != MAP_FAILED) { - - /* We do not need, and cannot use, another sbrk call to find end */ - brk = mbrk; - snd_brk = brk + size; - - /* - Record that we no longer have a contiguous sbrk region. - After the first time mmap is used as backup, we do not - ever rely on contiguous space since this could incorrectly - bridge regions. - */ - set_noncontiguous(av); - } - } -#endif - } - - if (brk != (char*)(MORECORE_FAILURE)) { - if (mp_.sbrk_base == 0) - mp_.sbrk_base = brk; - av->system_mem += size; - - /* - If MORECORE extends previous space, we can likewise extend top size. - */ - - if (brk == old_end && snd_brk == (char*)(MORECORE_FAILURE)) - set_head(old_top, (size + old_size) | PREV_INUSE); - - else if (contiguous(av) && old_size && brk < old_end) { - /* Oops! Someone else killed our space.. Can't touch anything. */ - assert(0); - } - - /* - Otherwise, make adjustments: - - * If the first time through or noncontiguous, we need to call sbrk - just to find out where the end of memory lies. - - * We need to ensure that all returned chunks from malloc will meet - MALLOC_ALIGNMENT - - * If there was an intervening foreign sbrk, we need to adjust sbrk - request size to account for fact that we will not be able to - combine new space with existing space in old_top. - - * Almost all systems internally allocate whole pages at a time, in - which case we might as well use the whole last page of request. - So we allocate enough more memory to hit a page boundary now, - which in turn causes future contiguous calls to page-align. - */ - - else { - front_misalign = 0; - end_misalign = 0; - correction = 0; - aligned_brk = brk; - - /* handle contiguous cases */ - if (contiguous(av)) { - - /* Count foreign sbrk as system_mem. */ - if (old_size) - av->system_mem += brk - old_end; - - /* Guarantee alignment of first new chunk made from this space */ - - front_misalign = (INTERNAL_SIZE_T)chunk2mem(brk) & MALLOC_ALIGN_MASK; - if (front_misalign > 0) { - - /* - Skip over some bytes to arrive at an aligned position. - We don't need to specially mark these wasted front bytes. - They will never be accessed anyway because - prev_inuse of av->top (and any chunk created from its start) - is always true after initialization. - */ - - correction = MALLOC_ALIGNMENT - front_misalign; - aligned_brk += correction; - } - - /* - If this isn't adjacent to existing space, then we will not - be able to merge with old_top space, so must add to 2nd request. - */ - - correction += old_size; - - /* Extend the end address to hit a page boundary */ - end_misalign = (INTERNAL_SIZE_T)(brk + size + correction); - correction += ((end_misalign + pagemask) & ~pagemask) - end_misalign; - - assert(correction >= 0); - snd_brk = (char*)(MORECORE(correction)); - - /* - If can't allocate correction, try to at least find out current - brk. It might be enough to proceed without failing. - - Note that if second sbrk did NOT fail, we assume that space - is contiguous with first sbrk. This is a safe assumption unless - program is multithreaded but doesn't use locks and a foreign sbrk - occurred between our first and second calls. - */ - - if (snd_brk == (char*)(MORECORE_FAILURE)) { - correction = 0; - snd_brk = (char*)(MORECORE(0)); - } else - /* Call the `morecore' hook if necessary. */ - if (__after_morecore_hook) - (*__after_morecore_hook) (); - } - - /* handle non-contiguous cases */ - else { - /* MORECORE/mmap must correctly align */ - assert(((unsigned long)chunk2mem(brk) & MALLOC_ALIGN_MASK) == 0); - - /* Find out current end of memory */ - if (snd_brk == (char*)(MORECORE_FAILURE)) { - snd_brk = (char*)(MORECORE(0)); - } - } - - /* Adjust top based on results of second sbrk */ - if (snd_brk != (char*)(MORECORE_FAILURE)) { - av->top = (mchunkptr)aligned_brk; - set_head(av->top, (snd_brk - aligned_brk + correction) | PREV_INUSE); - av->system_mem += correction; - - /* - If not the first time through, we either have a - gap due to foreign sbrk or a non-contiguous region. Insert a - double fencepost at old_top to prevent consolidation with space - we don't own. These fenceposts are artificial chunks that are - marked as inuse and are in any case too small to use. We need - two to make sizes and alignments work out. - */ - - if (old_size != 0) { - /* - Shrink old_top to insert fenceposts, keeping size a - multiple of MALLOC_ALIGNMENT. We know there is at least - enough space in old_top to do this. - */ - old_size = (old_size - 4*SIZE_SZ) & ~MALLOC_ALIGN_MASK; - set_head(old_top, old_size | PREV_INUSE); - - /* - Note that the following assignments completely overwrite - old_top when old_size was previously MINSIZE. This is - intentional. We need the fencepost, even if old_top otherwise gets - lost. - */ - chunk_at_offset(old_top, old_size )->size = - (2*SIZE_SZ)|PREV_INUSE; - - chunk_at_offset(old_top, old_size + 2*SIZE_SZ)->size = - (2*SIZE_SZ)|PREV_INUSE; - - /* If possible, release the rest. */ - if (old_size >= MINSIZE) { - _int_free(av, chunk2mem(old_top)); - } - - } - } - } - - /* Update statistics */ -#ifdef NO_THREADS - sum = av->system_mem + mp_.mmapped_mem; - if (sum > (unsigned long)(mp_.max_total_mem)) - mp_.max_total_mem = sum; -#endif - - } - - } /* if (av != &main_arena) */ - - if ((unsigned long)av->system_mem > (unsigned long)(av->max_system_mem)) - av->max_system_mem = av->system_mem; - check_malloc_state(av); - - /* finally, do the allocation */ - p = av->top; - size = chunksize(p); - - /* check that one of the above allocation paths succeeded */ - if ((unsigned long)(size) >= (unsigned long)(nb + MINSIZE)) { - remainder_size = size - nb; - remainder = chunk_at_offset(p, nb); - av->top = remainder; - set_head(p, nb | PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0)); - set_head(remainder, remainder_size | PREV_INUSE); - check_malloced_chunk(av, p, nb); - return chunk2mem(p); - } - - /* catch all failure paths */ - MALLOC_FAILURE_ACTION; - return 0; -} - - -#ifndef MORECORE_CANNOT_TRIM -/* - sYSTRIm is an inverse of sorts to sYSMALLOc. It gives memory back - to the system (via negative arguments to sbrk) if there is unused - memory at the `high' end of the malloc pool. It is called - automatically by free() when top space exceeds the trim - threshold. It is also called by the public malloc_trim routine. It - returns 1 if it actually released any memory, else 0. -*/ - -#if __STD_C -static int sYSTRIm(size_t pad, mstate av) -#else -static int sYSTRIm(pad, av) size_t pad; mstate av; -#endif -{ - long top_size; /* Amount of top-most memory */ - long extra; /* Amount to release */ - long released; /* Amount actually released */ - char* current_brk; /* address returned by pre-check sbrk call */ - char* new_brk; /* address returned by post-check sbrk call */ - size_t pagesz; - - pagesz = mp_.pagesize; - top_size = chunksize(av->top); - - /* Release in pagesize units, keeping at least one page */ - extra = ((top_size - pad - MINSIZE + (pagesz-1)) / pagesz - 1) * pagesz; - - if (extra > 0) { - - /* - Only proceed if end of memory is where we last set it. - This avoids problems if there were foreign sbrk calls. - */ - current_brk = (char*)(MORECORE(0)); - if (current_brk == (char*)(av->top) + top_size) { - - /* - Attempt to release memory. We ignore MORECORE return value, - and instead call again to find out where new end of memory is. - This avoids problems if first call releases less than we asked, - of if failure somehow altered brk value. (We could still - encounter problems if it altered brk in some very bad way, - but the only thing we can do is adjust anyway, which will cause - some downstream failure.) - */ - - MORECORE(-extra); - /* Call the `morecore' hook if necessary. */ - if (__after_morecore_hook) - (*__after_morecore_hook) (); - new_brk = (char*)(MORECORE(0)); - - if (new_brk != (char*)MORECORE_FAILURE) { - released = (long)(current_brk - new_brk); - - if (released != 0) { - /* Success. Adjust top. */ - av->system_mem -= released; - set_head(av->top, (top_size - released) | PREV_INUSE); - check_malloc_state(av); - return 1; - } - } - } - } - return 0; -} -#endif - -#ifdef HAVE_MMAP - -static void -internal_function -#if __STD_C -munmap_chunk(mchunkptr p) -#else -munmap_chunk(p) mchunkptr p; -#endif -{ - INTERNAL_SIZE_T size = chunksize(p); - int ret; - - assert (chunk_is_mmapped(p)); -#if 0 - assert(! ((char*)p >= mp_.sbrk_base && (char*)p < mp_.sbrk_base + mp_.sbrked_mem)); - assert((mp_.n_mmaps > 0)); -#endif - assert(((p->prev_size + size) & (mp_.pagesize-1)) == 0); - - mp_.n_mmaps--; - mp_.mmapped_mem -= (size + p->prev_size); - - ret = munmap((char *)p - p->prev_size, size + p->prev_size); - - /* munmap returns non-zero on failure */ - assert(ret == 0); -} - -#if HAVE_MREMAP - -static mchunkptr -internal_function -#if __STD_C -mremap_chunk(mchunkptr p, size_t new_size) -#else -mremap_chunk(p, new_size) mchunkptr p; size_t new_size; -#endif -{ - size_t page_mask = mp_.pagesize - 1; - INTERNAL_SIZE_T offset = p->prev_size; - INTERNAL_SIZE_T size = chunksize(p); - char *cp; - - assert (chunk_is_mmapped(p)); -#if 0 - assert(! ((char*)p >= mp_.sbrk_base && (char*)p < mp_.sbrk_base + mp_.sbrked_mem)); - assert((mp_.n_mmaps > 0)); -#endif - assert(((size + offset) & (mp_.pagesize-1)) == 0); - - /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */ - new_size = (new_size + offset + SIZE_SZ + page_mask) & ~page_mask; - - /* OMPI change - take pessimistic approach and assume going to move */ - opal_mem_hooks_release_hook((char *)p - offset, size, 1); - cp = (char *)mremap((char *)p - offset, size + offset, new_size, - MREMAP_MAYMOVE); - - if (cp == MAP_FAILED) return 0; - - p = (mchunkptr)(cp + offset); - - assert(aligned_OK(chunk2mem(p))); - - assert((p->prev_size == offset)); - set_head(p, (new_size - offset)|IS_MMAPPED); - - mp_.mmapped_mem -= size + offset; - mp_.mmapped_mem += new_size; - if ((unsigned long)mp_.mmapped_mem > (unsigned long)mp_.max_mmapped_mem) - mp_.max_mmapped_mem = mp_.mmapped_mem; -#ifdef NO_THREADS - if ((unsigned long)(mp_.mmapped_mem + arena_mem + main_arena.system_mem) > - mp_.max_total_mem) - mp_.max_total_mem = mp_.mmapped_mem + arena_mem + main_arena.system_mem; -#endif - return p; -} - -#endif /* HAVE_MREMAP */ - -#endif /* HAVE_MMAP */ - -/*------------------------ Public wrappers. --------------------------------*/ - -Void_t* -public_mALLOc(size_t bytes) -{ - mstate ar_ptr; - Void_t *victim; - - /* OMPI change: the hook is us -- don't call the hook */ -#if 0 - __malloc_ptr_t (*hook) __MALLOC_P ((size_t, __const __malloc_ptr_t)) = - __malloc_hook; - if (hook != NULL) - return (*hook)(bytes, RETURN_ADDRESS (0)); -#endif - - /* OMPI change: put in a flag so that we can know that this function - was invoked. This flag is checked in the memory/linux component - init to ensure that this ptmalloc is actually being used. Used a - simple "extern" here to get the flag symbol rather than putting - it in a new .h file that would only contain a small number of - symbols. */ - mca_memory_linux_component.malloc_invoked = true; - - arena_get(ar_ptr, bytes); - if(!ar_ptr) - return 0; - victim = _int_malloc(ar_ptr, bytes); - if(!victim) { - /* Maybe the failure is due to running out of mmapped areas. */ - if(ar_ptr != &main_arena) { - (void)mutex_unlock(&ar_ptr->mutex); - (void)mutex_lock(&main_arena.mutex); - victim = _int_malloc(&main_arena, bytes); - (void)mutex_unlock(&main_arena.mutex); - } else { -#if USE_ARENAS - /* ... or sbrk() has failed and there is still a chance to mmap() */ - ar_ptr = arena_get2(ar_ptr->next ? ar_ptr : 0, bytes); - (void)mutex_unlock(&main_arena.mutex); - if(ar_ptr) { - victim = _int_malloc(ar_ptr, bytes); - (void)mutex_unlock(&ar_ptr->mutex); - } -#endif - } - } else - (void)mutex_unlock(&ar_ptr->mutex); - assert(!victim || chunk_is_mmapped(mem2chunk(victim)) || - ar_ptr == arena_for_chunk(mem2chunk(victim))); - - return victim; -} -#ifdef libc_hidden_def -libc_hidden_def(public_mALLOc) -#endif - -void -public_fREe(Void_t* mem) -{ - mstate ar_ptr; - mchunkptr p; /* chunk corresponding to mem */ - - /* OMPI change: the hook is us -- don't call the hook */ -#if 0 - void (*hook) __MALLOC_P ((__malloc_ptr_t, __const __malloc_ptr_t)) = - __free_hook; - if (hook != NULL) { - (*hook)(mem, RETURN_ADDRESS (0)); - return; - } -#endif - - /* OMPI change: put in a flag so that we can know that this function - was invoked. This flag is checked in the memory/linux component - init to ensure that this ptmalloc is actually being used. Used a - simple "extern" here to get the flag symbol rather than putting - it in a new .h file that would only contain a small number of - symbols. */ - mca_memory_linux_component.free_invoked = true; - - if (mem == 0) /* free(0) has no effect */ - return; - - p = mem2chunk(mem); - -#if HAVE_MMAP - if (chunk_is_mmapped(p)) /* release mmapped memory. */ - { - munmap_chunk(p); - return; - } -#endif - - ar_ptr = arena_for_chunk(p); -#if THREAD_STATS - if(!mutex_trylock(&ar_ptr->mutex)) - ++(ar_ptr->stat_lock_direct); - else { - (void)mutex_lock(&ar_ptr->mutex); - ++(ar_ptr->stat_lock_wait); - } -#else - (void)mutex_lock(&ar_ptr->mutex); -#endif - _int_free(ar_ptr, mem); - (void)mutex_unlock(&ar_ptr->mutex); -} -#ifdef libc_hidden_def -libc_hidden_def (public_fREe) -#endif - -Void_t* -public_rEALLOc(Void_t* oldmem, size_t bytes) -{ - mstate ar_ptr; - INTERNAL_SIZE_T nb; /* padded request size */ - - mchunkptr oldp; /* chunk corresponding to oldmem */ - INTERNAL_SIZE_T oldsize; /* its size */ - - Void_t* newp; /* chunk to return */ - - /* OMPI change: the hook is us -- don't call the hook */ -#if 0 - __malloc_ptr_t (*hook) __MALLOC_P ((__malloc_ptr_t, size_t, - __const __malloc_ptr_t)) = - __realloc_hook; - if (hook != NULL) - return (*hook)(oldmem, bytes, RETURN_ADDRESS (0)); -#endif - - /* OMPI change: put in a flag so that we can know that this function - was invoked. This flag is checked in the memory/linux component - init to ensure that this ptmalloc is actually being used. Used a - simple "extern" here to get the flag symbol rather than putting - it in a new .h file that would only contain a small number of - symbols. */ - mca_memory_linux_component.realloc_invoked = true; - -#if REALLOC_ZERO_BYTES_FREES - if (bytes == 0 && oldmem != NULL) { public_fREe(oldmem); return 0; } -#endif - - /* realloc of null is supposed to be same as malloc */ - if (oldmem == 0) return public_mALLOc(bytes); - - oldp = mem2chunk(oldmem); - oldsize = chunksize(oldp); - - checked_request2size(bytes, nb); - -#if HAVE_MMAP - if (chunk_is_mmapped(oldp)) - { - Void_t* newmem; - -#if HAVE_MREMAP - newp = mremap_chunk(oldp, nb); - if(newp) { - return chunk2mem(newp); - } -#endif - /* Note the extra SIZE_SZ overhead. */ - if(oldsize - SIZE_SZ >= nb) { - return oldmem; /* do nothing */ - } - /* Must alloc, copy, free. */ - newmem = public_mALLOc(bytes); - if (newmem == 0) return 0; /* propagate failure */ - MALLOC_COPY(newmem, oldmem, oldsize - 2*SIZE_SZ); - munmap_chunk(oldp); - return newmem; - } -#endif - - ar_ptr = arena_for_chunk(oldp); -#if THREAD_STATS - if(!mutex_trylock(&ar_ptr->mutex)) - ++(ar_ptr->stat_lock_direct); - else { - (void)mutex_lock(&ar_ptr->mutex); - ++(ar_ptr->stat_lock_wait); - } -#else - (void)mutex_lock(&ar_ptr->mutex); -#endif - -#ifndef NO_THREADS - /* As in malloc(), remember this arena for the next allocation. */ - tsd_setspecific(arena_key, (Void_t *)ar_ptr); -#endif - - newp = _int_realloc(ar_ptr, oldmem, bytes); - - (void)mutex_unlock(&ar_ptr->mutex); - assert(!newp || chunk_is_mmapped(mem2chunk(newp)) || - ar_ptr == arena_for_chunk(mem2chunk(newp))); - return newp; -} -#ifdef libc_hidden_def -libc_hidden_def (public_rEALLOc) -#endif - -Void_t* -public_mEMALIGn(size_t alignment, size_t bytes) -{ - mstate ar_ptr; - Void_t *p; - - /* OMPI change: the hook is us -- don't call the hook */ -#if 0 - __malloc_ptr_t (*hook) __MALLOC_PMT ((size_t, size_t, - __const __malloc_ptr_t)) = - __memalign_hook; - if (hook != NULL) - return (*hook)(alignment, bytes, RETURN_ADDRESS (0)); -#endif - - /* OMPI change: put in a flag so that we can know that this function - was invoked. This flag is checked in the memory/linux component - init to ensure that this ptmalloc is actually being used. Used a - simple "extern" here to get the flag symbol rather than putting - it in a new .h file that would only contain a small number of - symbols. */ - mca_memory_linux_component.memalign_invoked = true; - - /* If need less alignment than we give anyway, just relay to malloc */ - if (alignment <= MALLOC_ALIGNMENT) return public_mALLOc(bytes); - - /* Otherwise, ensure that it is at least a minimum chunk size */ - if (alignment < MINSIZE) alignment = MINSIZE; - - arena_get(ar_ptr, bytes + alignment + MINSIZE); - if(!ar_ptr) - return 0; - p = _int_memalign(ar_ptr, alignment, bytes); - (void)mutex_unlock(&ar_ptr->mutex); - if(!p) { - /* Maybe the failure is due to running out of mmapped areas. */ - if(ar_ptr != &main_arena) { - (void)mutex_lock(&main_arena.mutex); - p = _int_memalign(&main_arena, alignment, bytes); - (void)mutex_unlock(&main_arena.mutex); - } else { -#if USE_ARENAS - /* ... or sbrk() has failed and there is still a chance to mmap() */ - ar_ptr = arena_get2(ar_ptr->next ? ar_ptr : 0, bytes); - if(ar_ptr) { - p = _int_memalign(ar_ptr, alignment, bytes); - (void)mutex_unlock(&ar_ptr->mutex); - } -#endif - } - } - assert(!p || chunk_is_mmapped(mem2chunk(p)) || - ar_ptr == arena_for_chunk(mem2chunk(p))); - return p; -} -#ifdef libc_hidden_def -libc_hidden_def (public_mEMALIGn) -#endif - -Void_t* -public_vALLOc(size_t bytes) -{ - mstate ar_ptr; - Void_t *p; - - if(__malloc_initialized < 0) - ptmalloc_init (); - arena_get(ar_ptr, bytes + mp_.pagesize + MINSIZE); - if(!ar_ptr) - return 0; - p = _int_valloc(ar_ptr, bytes); - (void)mutex_unlock(&ar_ptr->mutex); - return p; -} - -Void_t* -public_pVALLOc(size_t bytes) -{ - mstate ar_ptr; - Void_t *p; - - if(__malloc_initialized < 0) - ptmalloc_init (); - arena_get(ar_ptr, bytes + 2*mp_.pagesize + MINSIZE); - p = _int_pvalloc(ar_ptr, bytes); - (void)mutex_unlock(&ar_ptr->mutex); - return p; -} - -Void_t* -public_cALLOc(size_t n, size_t elem_size) -{ - mstate av; - mchunkptr oldtop, p; - INTERNAL_SIZE_T bytes, sz, csz, oldtopsize; - Void_t* mem; - unsigned long clearsize; - unsigned long nclears; - INTERNAL_SIZE_T* d; - /* OMPI change: the hook is us -- don't call the hook */ -#if 0 - __malloc_ptr_t (*hook) __MALLOC_PMT ((size_t, __const __malloc_ptr_t)) = - __malloc_hook; -#endif - - /* size_t is unsigned so the behavior on overflow is defined. */ - bytes = n * elem_size; -#define HALF_INTERNAL_SIZE_T \ - (((INTERNAL_SIZE_T) 1) << (8 * sizeof (INTERNAL_SIZE_T) / 2)) - if (__builtin_expect ((n | elem_size) >= HALF_INTERNAL_SIZE_T, 0)) { - if (elem_size != 0 && bytes / elem_size != n) { - MALLOC_FAILURE_ACTION; - return 0; - } - } - - /* OMPI change: the hook is us -- don't call the hook */ -#if 0 - if (hook != NULL) { - sz = bytes; - mem = (*hook)(sz, RETURN_ADDRESS (0)); - if(mem == 0) - return 0; -#ifdef HAVE_MEMCPY - return memset(mem, 0, sz); -#else - while(sz > 0) ((char*)mem)[--sz] = 0; /* rather inefficient */ - return mem; -#endif - } -#endif - - sz = bytes; - - arena_get(av, sz); - if(!av) - return 0; - - /* Check if we hand out the top chunk, in which case there may be no - need to clear. */ -#if MORECORE_CLEARS - oldtop = top(av); - oldtopsize = chunksize(top(av)); -#if MORECORE_CLEARS < 2 - /* Only newly allocated memory is guaranteed to be cleared. */ - if (av == &main_arena && - oldtopsize < (INTERNAL_SIZE_T)(mp_.sbrk_base + av->max_system_mem - (char *)oldtop)) - oldtopsize = (mp_.sbrk_base + av->max_system_mem - (char *)oldtop); -#endif -#endif - mem = _int_malloc(av, sz); - - /* Only clearing follows, so we can unlock early. */ - (void)mutex_unlock(&av->mutex); - - assert(!mem || chunk_is_mmapped(mem2chunk(mem)) || - av == arena_for_chunk(mem2chunk(mem))); - - if (mem == 0) { - /* Maybe the failure is due to running out of mmapped areas. */ - if(av != &main_arena) { - (void)mutex_lock(&main_arena.mutex); - mem = _int_malloc(&main_arena, sz); - (void)mutex_unlock(&main_arena.mutex); - } else { -#if USE_ARENAS - /* ... or sbrk() has failed and there is still a chance to mmap() */ - (void)mutex_lock(&main_arena.mutex); - av = arena_get2(av->next ? av : 0, sz); - (void)mutex_unlock(&main_arena.mutex); - if(av) { - mem = _int_malloc(av, sz); - (void)mutex_unlock(&av->mutex); - } -#endif - } - if (mem == 0) return 0; - } - p = mem2chunk(mem); - - /* Two optional cases in which clearing not necessary */ -#if HAVE_MMAP - if (chunk_is_mmapped(p)) { - return mem; - } -#endif - - csz = chunksize(p); - -#if MORECORE_CLEARS - if (p == oldtop && csz > oldtopsize) { - /* clear only the bytes from non-freshly-sbrked memory */ - csz = oldtopsize; - } -#endif - - /* Unroll clear of <= 36 bytes (72 if 8byte sizes). We know that - contents have an odd number of INTERNAL_SIZE_T-sized words; - minimally 3. */ - d = (INTERNAL_SIZE_T*)mem; - clearsize = csz - SIZE_SZ; - nclears = clearsize / sizeof(INTERNAL_SIZE_T); - assert(nclears >= 3); - - if (nclears > 9) - MALLOC_ZERO(d, clearsize); - - else { - *(d+0) = 0; - *(d+1) = 0; - *(d+2) = 0; - if (nclears > 4) { - *(d+3) = 0; - *(d+4) = 0; - if (nclears > 6) { - *(d+5) = 0; - *(d+6) = 0; - if (nclears > 8) { - *(d+7) = 0; - *(d+8) = 0; - } - } - } - } - - return mem; -} - -Void_t** -public_iCALLOc(size_t n, size_t elem_size, Void_t** chunks) -{ - mstate ar_ptr; - Void_t** m; - - arena_get(ar_ptr, n*elem_size); - if(!ar_ptr) - return 0; - - m = _int_icalloc(ar_ptr, n, elem_size, chunks); - (void)mutex_unlock(&ar_ptr->mutex); - return m; -} - -Void_t** -public_iCOMALLOc(size_t n, size_t sizes[], Void_t** chunks) -{ - mstate ar_ptr; - Void_t** m; - - arena_get(ar_ptr, 0); - if(!ar_ptr) - return 0; - - m = _int_icomalloc(ar_ptr, n, sizes, chunks); - (void)mutex_unlock(&ar_ptr->mutex); - return m; -} - -#ifndef _LIBC - -void -public_cFREe(Void_t* m) -{ - public_fREe(m); -} - -#endif /* _LIBC */ - -int -public_mTRIm(size_t s) -{ - int result; - - (void)mutex_lock(&main_arena.mutex); - result = mTRIm(s); - (void)mutex_unlock(&main_arena.mutex); - return result; -} - -size_t -public_mUSABLe(Void_t* m) -{ - size_t result; - - result = mUSABLe(m); - return result; -} - -/* This exists mainly for backward compatibility. Calling - _int_get_arena_info() directly is more useful. */ -struct mallinfo -public_mALLINFo() -{ - struct malloc_arena_info mai; - struct mallinfo m; - size_t avail; - - if(__malloc_initialized < 0) - ptmalloc_init (); - _int_get_arena_info(&main_arena, &mai); - /* Account for top */ - avail = mai.fastavail + mai.binavail + mai.top_size; - m.smblks = mai.nfastblocks; - m.ordblks = mai.nbinblocks + 1; - m.fordblks = avail; - m.uordblks = mai.system_mem - avail; - m.arena = mai.system_mem; - m.hblks = mp_.n_mmaps; - m.hblkhd = mp_.mmapped_mem; - m.fsmblks = mai.fastavail; - m.keepcost = mai.top_size; - m.usmblks = mp_.max_total_mem; - return m; -} - -int -public_mALLOPt(int p, int v) -{ - int result; - result = mALLOPt(p, v); - return result; -} - -/* - ------------------------------ malloc ------------------------------ -*/ - -/* With Intel Composer XE V12.1.0, release 2011.6.233, any launch */ -/* fails, even before main(), due to a bug in the vectorizer (see */ -/* https://svn.open-mpi.org/trac/ompi/changeset/25290). The fix is */ -/* to disable vectorization by reducing the optimization level to */ -/* -O1 for _int_malloc(). The only reliable method to identify */ -/* release 2011.6.233 is the predefined __INTEL_COMPILER_BUILD_DATE */ -/* macro, which will have the value 20110811 (Linux, Windows, and */ -/* Mac OS X). (The predefined __INTEL_COMPILER macro is nonsense, */ -/* 9999, and both the 2011.6.233 and 2011.7.256 releases identify */ -/* themselves as V12.1.0 from the -v command line option.) */ - -#ifdef __INTEL_COMPILER_BUILD_DATE -# if __INTEL_COMPILER_BUILD_DATE == 20110811 -# pragma GCC optimization_level 1 -# endif -#endif - -Void_t* -_int_malloc(mstate av, size_t bytes) -{ - INTERNAL_SIZE_T nb; /* normalized request size */ - unsigned int idx; /* associated bin index */ - mbinptr bin; /* associated bin */ - mfastbinptr* fb; /* associated fastbin */ - - mchunkptr victim; /* inspected/selected chunk */ - INTERNAL_SIZE_T size; /* its size */ - int victim_index; /* its bin index */ - - mchunkptr remainder; /* remainder from a split */ - unsigned long remainder_size; /* its size */ - - unsigned int block; /* bit map traverser */ - unsigned int bit; /* bit map traverser */ - unsigned int map; /* current word of binmap */ - - mchunkptr fwd; /* misc temp for linking */ - mchunkptr bck; /* misc temp for linking */ - - /* - Convert request size to internal form by adding SIZE_SZ bytes - overhead plus possibly more to obtain necessary alignment and/or - to obtain a size of at least MINSIZE, the smallest allocatable - size. Also, checked_request2size traps (returning 0) request sizes - that are so large that they wrap around zero when padded and - aligned. - */ - - checked_request2size(bytes, nb); - - /* - If the size qualifies as a fastbin, first check corresponding bin. - This code is safe to execute even if av is not yet initialized, so we - can try it without checking, which saves some time on this fast path. - */ - - if ((unsigned long)(nb) <= (unsigned long)(av->max_fast)) { - fb = &(av->fastbins[(fastbin_index(nb))]); - if ( (victim = *fb) != 0) { - *fb = victim->fd; - check_remalloced_chunk(av, victim, nb); - return chunk2mem(victim); - } - } - - /* - If a small request, check regular bin. Since these "smallbins" - hold one size each, no searching within bins is necessary. - (For a large request, we need to wait until unsorted chunks are - processed to find best fit. But for small ones, fits are exact - anyway, so we can check now, which is faster.) - */ - - if (in_smallbin_range(nb)) { - idx = smallbin_index(nb); - bin = bin_at(av,idx); - - if ( (victim = last(bin)) != bin) { - if (victim == 0) /* initialization check */ - malloc_consolidate(av); - else { - bck = victim->bk; - set_inuse_bit_at_offset(victim, nb); - bin->bk = bck; - bck->fd = bin; - - if (av != &main_arena) - victim->size |= NON_MAIN_ARENA; - check_malloced_chunk(av, victim, nb); - return chunk2mem(victim); - } - } - } - - /* - If this is a large request, consolidate fastbins before continuing. - While it might look excessive to kill all fastbins before - even seeing if there is space available, this avoids - fragmentation problems normally associated with fastbins. - Also, in practice, programs tend to have runs of either small or - large requests, but less often mixtures, so consolidation is not - invoked all that often in most programs. And the programs that - it is called frequently in otherwise tend to fragment. - */ - - else { - idx = largebin_index(nb); - if (have_fastchunks(av)) - malloc_consolidate(av); - } - - /* - Process recently freed or remaindered chunks, taking one only if - it is exact fit, or, if this a small request, the chunk is remainder from - the most recent non-exact fit. Place other traversed chunks in - bins. Note that this step is the only place in any routine where - chunks are placed in bins. - - The outer loop here is needed because we might not realize until - near the end of malloc that we should have consolidated, so must - do so and retry. This happens at most once, and only when we would - otherwise need to expand memory to service a "small" request. - */ - - for(;;) { - - while ( (victim = unsorted_chunks(av)->bk) != unsorted_chunks(av)) { - bck = victim->bk; - size = chunksize(victim); - - /* - If a small request, try to use last remainder if it is the - only chunk in unsorted bin. This helps promote locality for - runs of consecutive small requests. This is the only - exception to best-fit, and applies only when there is - no exact fit for a small chunk. - */ - - if (in_smallbin_range(nb) && - bck == unsorted_chunks(av) && - victim == av->last_remainder && - (unsigned long)(size) > (unsigned long)(nb + MINSIZE)) { - - /* split and reattach remainder */ - remainder_size = size - nb; - remainder = chunk_at_offset(victim, nb); - unsorted_chunks(av)->bk = unsorted_chunks(av)->fd = remainder; - av->last_remainder = remainder; - remainder->bk = remainder->fd = unsorted_chunks(av); - - set_head(victim, nb | PREV_INUSE | - (av != &main_arena ? NON_MAIN_ARENA : 0)); - set_head(remainder, remainder_size | PREV_INUSE); - set_foot(remainder, remainder_size); - - check_malloced_chunk(av, victim, nb); - return chunk2mem(victim); - } - - /* remove from unsorted list */ - unsorted_chunks(av)->bk = bck; - bck->fd = unsorted_chunks(av); - - /* Take now instead of binning if exact fit */ - - if (size == nb) { - set_inuse_bit_at_offset(victim, size); - if (av != &main_arena) - victim->size |= NON_MAIN_ARENA; - check_malloced_chunk(av, victim, nb); - return chunk2mem(victim); - } - - /* place chunk in bin */ - - if (in_smallbin_range(size)) { - victim_index = smallbin_index(size); - bck = bin_at(av, victim_index); - fwd = bck->fd; - } - else { - victim_index = largebin_index(size); - bck = bin_at(av, victim_index); - fwd = bck->fd; - - if (fwd != bck) { - /* if smaller than smallest, place first */ - assert((bck->bk->size & NON_MAIN_ARENA) == 0); - if ((unsigned long)(size) < (unsigned long)(bck->bk->size)) { - fwd = bck; - bck = bck->bk; - } - else if ((unsigned long)(size) >= - (unsigned long)(FIRST_SORTED_BIN_SIZE)) { - - /* maintain large bins in sorted order */ - size |= PREV_INUSE; /* Or with inuse bit to speed comparisons */ - assert((fwd->size & NON_MAIN_ARENA) == 0); - while ((unsigned long)(size) < (unsigned long)(fwd->size)) { - fwd = fwd->fd; - assert((fwd->size & NON_MAIN_ARENA) == 0); - } - bck = fwd->bk; - } - } - } - - mark_bin(av, victim_index); - victim->bk = bck; - victim->fd = fwd; - fwd->bk = victim; - bck->fd = victim; - } - - /* - If a large request, scan through the chunks of current bin in - sorted order to find smallest that fits. This is the only step - where an unbounded number of chunks might be scanned without doing - anything useful with them. However the lists tend to be short. - */ - - if (!in_smallbin_range(nb)) { - bin = bin_at(av, idx); - - for (victim = last(bin); victim != bin; victim = victim->bk) { - size = chunksize(victim); - - if ((unsigned long)(size) >= (unsigned long)(nb)) { - remainder_size = size - nb; - unlink(victim, bck, fwd); - - /* Exhaust */ - if (remainder_size < MINSIZE) { - set_inuse_bit_at_offset(victim, size); - if (av != &main_arena) - victim->size |= NON_MAIN_ARENA; - check_malloced_chunk(av, victim, nb); - return chunk2mem(victim); - } - /* Split */ - else { - remainder = chunk_at_offset(victim, nb); - unsorted_chunks(av)->bk = unsorted_chunks(av)->fd = remainder; - remainder->bk = remainder->fd = unsorted_chunks(av); - set_head(victim, nb | PREV_INUSE | - (av != &main_arena ? NON_MAIN_ARENA : 0)); - set_head(remainder, remainder_size | PREV_INUSE); - set_foot(remainder, remainder_size); - check_malloced_chunk(av, victim, nb); - return chunk2mem(victim); - } - } - } - } - - /* - Search for a chunk by scanning bins, starting with next largest - bin. This search is strictly by best-fit; i.e., the smallest - (with ties going to approximately the least recently used) chunk - that fits is selected. - - The bitmap avoids needing to check that most blocks are nonempty. - The particular case of skipping all bins during warm-up phases - when no chunks have been returned yet is faster than it might look. - */ - - ++idx; - bin = bin_at(av,idx); - block = idx2block(idx); - map = av->binmap[block]; - bit = idx2bit(idx); - - for (;;) { - - /* Skip rest of block if there are no more set bits in this block. */ - if (bit > map || bit == 0) { - do { - if (++block >= BINMAPSIZE) /* out of bins */ - goto use_top; - } while ( (map = av->binmap[block]) == 0); - - bin = bin_at(av, (block << BINMAPSHIFT)); - bit = 1; - } - - /* Advance to bin with set bit. There must be one. */ - while ((bit & map) == 0) { - bin = next_bin(bin); - bit <<= 1; - assert(bit != 0); - } - - /* Inspect the bin. It is likely to be non-empty */ - victim = last(bin); - - /* If a false alarm (empty bin), clear the bit. */ - if (victim == bin) { - av->binmap[block] = map &= ~bit; /* Write through */ - bin = next_bin(bin); - bit <<= 1; - } - - else { - size = chunksize(victim); - - /* We know the first chunk in this bin is big enough to use. */ - assert((unsigned long)(size) >= (unsigned long)(nb)); - - remainder_size = size - nb; - - /* unlink */ - bck = victim->bk; - bin->bk = bck; - bck->fd = bin; - - /* Exhaust */ - if (remainder_size < MINSIZE) { - set_inuse_bit_at_offset(victim, size); - if (av != &main_arena) - victim->size |= NON_MAIN_ARENA; - check_malloced_chunk(av, victim, nb); - return chunk2mem(victim); - } - - /* Split */ - else { - remainder = chunk_at_offset(victim, nb); - - unsorted_chunks(av)->bk = unsorted_chunks(av)->fd = remainder; - remainder->bk = remainder->fd = unsorted_chunks(av); - /* advertise as last remainder */ - if (in_smallbin_range(nb)) - av->last_remainder = remainder; - - set_head(victim, nb | PREV_INUSE | - (av != &main_arena ? NON_MAIN_ARENA : 0)); - set_head(remainder, remainder_size | PREV_INUSE); - set_foot(remainder, remainder_size); - check_malloced_chunk(av, victim, nb); - return chunk2mem(victim); - } - } - } - - use_top: - /* - If large enough, split off the chunk bordering the end of memory - (held in av->top). Note that this is in accord with the best-fit - search rule. In effect, av->top is treated as larger (and thus - less well fitting) than any other available chunk since it can - be extended to be as large as necessary (up to system - limitations). - - We require that av->top always exists (i.e., has size >= - MINSIZE) after initialization, so if it would otherwise be - exhuasted by current request, it is replenished. (The main - reason for ensuring it exists is that we may need MINSIZE space - to put in fenceposts in sysmalloc.) - */ - - victim = av->top; - size = chunksize(victim); - - if ((unsigned long)(size) >= (unsigned long)(nb + MINSIZE)) { - remainder_size = size - nb; - remainder = chunk_at_offset(victim, nb); - av->top = remainder; - set_head(victim, nb | PREV_INUSE | - (av != &main_arena ? NON_MAIN_ARENA : 0)); - set_head(remainder, remainder_size | PREV_INUSE); - - check_malloced_chunk(av, victim, nb); - return chunk2mem(victim); - } - - /* - If there is space available in fastbins, consolidate and retry, - to possibly avoid expanding memory. This can occur only if nb is - in smallbin range so we didn't consolidate upon entry. - */ - - else if (have_fastchunks(av)) { - assert(in_smallbin_range(nb)); - malloc_consolidate(av); - idx = smallbin_index(nb); /* restore original bin index */ - } - - /* - Otherwise, relay to handle system-dependent cases - */ - else - return sYSMALLOc(nb, av); - } -} - -/* - ------------------------------ free ------------------------------ -*/ - -void -_int_free(mstate av, Void_t* mem) -{ - mchunkptr p; /* chunk corresponding to mem */ - INTERNAL_SIZE_T size; /* its size */ - mfastbinptr* fb; /* associated fastbin */ - mchunkptr nextchunk; /* next contiguous chunk */ - INTERNAL_SIZE_T nextsize; /* its size */ - int nextinuse; /* true if nextchunk is used */ - INTERNAL_SIZE_T prevsize; /* size of previous contiguous chunk */ - mchunkptr bck; /* misc temp for linking */ - mchunkptr fwd; /* misc temp for linking */ - - - /* free(0) has no effect */ - if (mem != 0) { - p = mem2chunk(mem); - size = chunksize(p); - - check_inuse_chunk(av, p); - - /* - If eligible, place chunk on a fastbin so it can be found - and used quickly in malloc. - */ - - if ((unsigned long)(size) <= (unsigned long)(av->max_fast) - -#if TRIM_FASTBINS - /* - If TRIM_FASTBINS set, don't place chunks - bordering top into fastbins - */ - && (chunk_at_offset(p, size) != av->top) -#endif - ) { - - set_fastchunks(av); - fb = &(av->fastbins[fastbin_index(size)]); - p->fd = *fb; - *fb = p; - } - - /* - Consolidate other non-mmapped chunks as they arrive. - */ - - else if (!chunk_is_mmapped(p)) { - nextchunk = chunk_at_offset(p, size); - nextsize = chunksize(nextchunk); - assert(nextsize > 0); - - /* consolidate backward */ - if (!prev_inuse(p)) { - prevsize = p->prev_size; - size += prevsize; - p = chunk_at_offset(p, -((long) prevsize)); - unlink(p, bck, fwd); - } - - if (nextchunk != av->top) { - /* get and clear inuse bit */ - nextinuse = inuse_bit_at_offset(nextchunk, nextsize); - - /* consolidate forward */ - if (!nextinuse) { - unlink(nextchunk, bck, fwd); - size += nextsize; - } else - clear_inuse_bit_at_offset(nextchunk, 0); - - /* - Place the chunk in unsorted chunk list. Chunks are - not placed into regular bins until after they have - been given one chance to be used in malloc. - */ - - bck = unsorted_chunks(av); - fwd = bck->fd; - p->bk = bck; - p->fd = fwd; - bck->fd = p; - fwd->bk = p; - - set_head(p, size | PREV_INUSE); - set_foot(p, size); - - check_free_chunk(av, p); - } - - /* - If the chunk borders the current high end of memory, - consolidate into top - */ - - else { - size += nextsize; - set_head(p, size | PREV_INUSE); - av->top = p; - check_chunk(av, p); - } - - /* - If freeing a large space, consolidate possibly-surrounding - chunks. Then, if the total unused topmost memory exceeds trim - threshold, ask malloc_trim to reduce top. - - Unless max_fast is 0, we don't know if there are fastbins - bordering top, so we cannot tell for sure whether threshold - has been reached unless fastbins are consolidated. But we - don't want to consolidate on each free. As a compromise, - consolidation is performed if FASTBIN_CONSOLIDATION_THRESHOLD - is reached. - */ - - if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) { - if (have_fastchunks(av)) - malloc_consolidate(av); - - if (av == &main_arena) { -#ifndef MORECORE_CANNOT_TRIM - if ((unsigned long)(chunksize(av->top)) >= - (unsigned long)(mp_.trim_threshold)) - sYSTRIm(mp_.top_pad, av); -#endif - } else { - /* Always try heap_trim(), even if the top chunk is not - large, because the corresponding heap might go away. */ - heap_info *heap = heap_for_ptr(top(av)); - - assert(heap->ar_ptr == av); - heap_trim(heap, mp_.top_pad); - } - } - - } - /* - If the chunk was allocated via mmap, release via munmap(). Note - that if HAVE_MMAP is false but chunk_is_mmapped is true, then - user must have overwritten memory. There's nothing we can do to - catch this error unless MALLOC_DEBUG is set, in which case - check_inuse_chunk (above) will have triggered error. - */ - - else { -#if HAVE_MMAP - int ret; - INTERNAL_SIZE_T offset = p->prev_size; - mp_.n_mmaps--; - mp_.mmapped_mem -= (size + offset); - ret = munmap((char*)p - offset, size + offset); - /* munmap returns non-zero on failure */ - assert(ret == 0); -#endif - } - } -} - -/* - ------------------------- malloc_consolidate ------------------------- - - malloc_consolidate is a specialized version of free() that tears - down chunks held in fastbins. Free itself cannot be used for this - purpose since, among other things, it might place chunks back onto - fastbins. So, instead, we need to use a minor variant of the same - code. - - Also, because this routine needs to be called the first time through - malloc anyway, it turns out to be the perfect place to trigger - initialization code. -*/ - -#if __STD_C -static void malloc_consolidate(mstate av) -#else -static void malloc_consolidate(av) mstate av; -#endif -{ - mfastbinptr* fb; /* current fastbin being consolidated */ - mfastbinptr* maxfb; /* last fastbin (for loop control) */ - mchunkptr p; /* current chunk being consolidated */ - mchunkptr nextp; /* next chunk to consolidate */ - mchunkptr unsorted_bin; /* bin header */ - mchunkptr first_unsorted; /* chunk to link to */ - - /* These have same use as in free() */ - mchunkptr nextchunk; - INTERNAL_SIZE_T size; - INTERNAL_SIZE_T nextsize; - INTERNAL_SIZE_T prevsize; - int nextinuse; - mchunkptr bck; - mchunkptr fwd; - - /* - If max_fast is 0, we know that av hasn't - yet been initialized, in which case do so below - */ - - if (av->max_fast != 0) { - clear_fastchunks(av); - - unsorted_bin = unsorted_chunks(av); - - /* - Remove each chunk from fast bin and consolidate it, placing it - then in unsorted bin. Among other reasons for doing this, - placing in unsorted bin avoids needing to calculate actual bins - until malloc is sure that chunks aren't immediately going to be - reused anyway. - */ - - maxfb = &(av->fastbins[fastbin_index(av->max_fast)]); - fb = &(av->fastbins[0]); - do { - if ( (p = *fb) != 0) { - *fb = 0; - - do { - check_inuse_chunk(av, p); - nextp = p->fd; - - /* Slightly streamlined version of consolidation code in free() */ - size = p->size & ~(PREV_INUSE|NON_MAIN_ARENA); - nextchunk = chunk_at_offset(p, size); - nextsize = chunksize(nextchunk); - - if (!prev_inuse(p)) { - prevsize = p->prev_size; - size += prevsize; - p = chunk_at_offset(p, -((long) prevsize)); - unlink(p, bck, fwd); - } - - if (nextchunk != av->top) { - nextinuse = inuse_bit_at_offset(nextchunk, nextsize); - - if (!nextinuse) { - size += nextsize; - unlink(nextchunk, bck, fwd); - } else - clear_inuse_bit_at_offset(nextchunk, 0); - - first_unsorted = unsorted_bin->fd; - unsorted_bin->fd = p; - first_unsorted->bk = p; - - set_head(p, size | PREV_INUSE); - p->bk = unsorted_bin; - p->fd = first_unsorted; - set_foot(p, size); - } - - else { - size += nextsize; - set_head(p, size | PREV_INUSE); - av->top = p; - } - - } while ( (p = nextp) != 0); - - } - } while (fb++ != maxfb); - } - else { - malloc_init_state(av); - check_malloc_state(av); - } -} - -/* - ------------------------------ realloc ------------------------------ -*/ - -Void_t* -_int_realloc(mstate av, Void_t* oldmem, size_t bytes) -{ - INTERNAL_SIZE_T nb; /* padded request size */ - - mchunkptr oldp; /* chunk corresponding to oldmem */ - INTERNAL_SIZE_T oldsize; /* its size */ - - mchunkptr newp; /* chunk to return */ - INTERNAL_SIZE_T newsize; /* its size */ - Void_t* newmem; /* corresponding user mem */ - - mchunkptr next; /* next contiguous chunk after oldp */ - - mchunkptr remainder; /* extra space at end of newp */ - unsigned long remainder_size; /* its size */ - - mchunkptr bck; /* misc temp for linking */ - mchunkptr fwd; /* misc temp for linking */ - - unsigned long copysize; /* bytes to copy */ - unsigned int ncopies; /* INTERNAL_SIZE_T words to copy */ - INTERNAL_SIZE_T* s; /* copy source */ - INTERNAL_SIZE_T* d; /* copy destination */ - - -#if REALLOC_ZERO_BYTES_FREES - if (bytes == 0) { - _int_free(av, oldmem); - return 0; - } -#endif - - /* realloc of null is supposed to be same as malloc */ - if (oldmem == 0) return _int_malloc(av, bytes); - - checked_request2size(bytes, nb); - - oldp = mem2chunk(oldmem); - oldsize = chunksize(oldp); - - check_inuse_chunk(av, oldp); - - if (!chunk_is_mmapped(oldp)) { - - if ((unsigned long)(oldsize) >= (unsigned long)(nb)) { - /* already big enough; split below */ - newp = oldp; - newsize = oldsize; - } - - else { - next = chunk_at_offset(oldp, oldsize); - - /* Try to expand forward into top */ - if (next == av->top && - (unsigned long)(newsize = oldsize + chunksize(next)) >= - (unsigned long)(nb + MINSIZE)) { - set_head_size(oldp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); - av->top = chunk_at_offset(oldp, nb); - set_head(av->top, (newsize - nb) | PREV_INUSE); - check_inuse_chunk(av, oldp); - return chunk2mem(oldp); - } - - /* Try to expand forward into next chunk; split off remainder below */ - else if (next != av->top && - !inuse(next) && - (unsigned long)(newsize = oldsize + chunksize(next)) >= - (unsigned long)(nb)) { - newp = oldp; - unlink(next, bck, fwd); - } - - /* allocate, copy, free */ - else { - newmem = _int_malloc(av, nb - MALLOC_ALIGN_MASK); - if (newmem == 0) - return 0; /* propagate failure */ - - newp = mem2chunk(newmem); - newsize = chunksize(newp); - - /* - Avoid copy if newp is next chunk after oldp. - */ - if (newp == next) { - newsize += oldsize; - newp = oldp; - } - else { - /* - Unroll copy of <= 36 bytes (72 if 8byte sizes) - We know that contents have an odd number of - INTERNAL_SIZE_T-sized words; minimally 3. - */ - - copysize = oldsize - SIZE_SZ; - s = (INTERNAL_SIZE_T*)(oldmem); - d = (INTERNAL_SIZE_T*)(newmem); - ncopies = copysize / sizeof(INTERNAL_SIZE_T); - assert(ncopies >= 3); - - if (ncopies > 9) - MALLOC_COPY(d, s, copysize); - - else { - *(d+0) = *(s+0); - *(d+1) = *(s+1); - *(d+2) = *(s+2); - if (ncopies > 4) { - *(d+3) = *(s+3); - *(d+4) = *(s+4); - if (ncopies > 6) { - *(d+5) = *(s+5); - *(d+6) = *(s+6); - if (ncopies > 8) { - *(d+7) = *(s+7); - *(d+8) = *(s+8); - } - } - } - } - - _int_free(av, oldmem); - check_inuse_chunk(av, newp); - return chunk2mem(newp); - } - } - } - - /* If possible, free extra space in old or extended chunk */ - - assert((unsigned long)(newsize) >= (unsigned long)(nb)); - - remainder_size = newsize - nb; - - if (remainder_size < MINSIZE) { /* not enough extra to split off */ - set_head_size(newp, newsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); - set_inuse_bit_at_offset(newp, newsize); - } - else { /* split remainder */ - remainder = chunk_at_offset(newp, nb); - set_head_size(newp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); - set_head(remainder, remainder_size | PREV_INUSE | - (av != &main_arena ? NON_MAIN_ARENA : 0)); - /* Mark remainder as inuse so free() won't complain */ - set_inuse_bit_at_offset(remainder, remainder_size); - _int_free(av, chunk2mem(remainder)); - } - - check_inuse_chunk(av, newp); - return chunk2mem(newp); - } - - /* - Handle mmap cases - */ - - else { -#if HAVE_MMAP - -#if HAVE_MREMAP - INTERNAL_SIZE_T offset = oldp->prev_size; - size_t pagemask = mp_.pagesize - 1; - char *cp; - unsigned long sum; - - /* Note the extra SIZE_SZ overhead */ - newsize = (nb + offset + SIZE_SZ + pagemask) & ~pagemask; - - /* don't need to remap if still within same page */ - if (oldsize == newsize - offset) - return oldmem; - - /* OMPI change - take pessimistic approach and assume going to move */ - opal_mem_hooks_release_hook((char *)oldp - offset, oldsize + offset, 1); - cp = (char*)mremap((char*)oldp - offset, oldsize + offset, newsize, 1); - - if (cp != MAP_FAILED) { - - newp = (mchunkptr)(cp + offset); - set_head(newp, (newsize - offset)|IS_MMAPPED); - - assert(aligned_OK(chunk2mem(newp))); - assert((newp->prev_size == offset)); - - /* update statistics */ - sum = mp_.mmapped_mem += newsize - oldsize; - if (sum > (unsigned long)(mp_.max_mmapped_mem)) - mp_.max_mmapped_mem = sum; -#ifdef NO_THREADS - sum += main_arena.system_mem; - if (sum > (unsigned long)(mp_.max_total_mem)) - mp_.max_total_mem = sum; -#endif - - return chunk2mem(newp); - } -#endif - - /* Note the extra SIZE_SZ overhead. */ - if ((unsigned long)(oldsize) >= (unsigned long)(nb + SIZE_SZ)) - newmem = oldmem; /* do nothing */ - else { - /* Must alloc, copy, free. */ - newmem = _int_malloc(av, nb - MALLOC_ALIGN_MASK); - if (newmem != 0) { - MALLOC_COPY(newmem, oldmem, oldsize - 2*SIZE_SZ); - _int_free(av, oldmem); - } - } - return newmem; - -#else - /* If !HAVE_MMAP, but chunk_is_mmapped, user must have overwritten mem */ - check_malloc_state(av); - MALLOC_FAILURE_ACTION; - return 0; -#endif - } -} - -/* - ------------------------------ memalign ------------------------------ -*/ - -Void_t* -_int_memalign(mstate av, size_t alignment, size_t bytes) -{ - INTERNAL_SIZE_T nb; /* padded request size */ - char* m; /* memory returned by malloc call */ - mchunkptr p; /* corresponding chunk */ - char* brk; /* alignment point within p */ - mchunkptr newp; /* chunk to return */ - INTERNAL_SIZE_T newsize; /* its size */ - INTERNAL_SIZE_T leadsize; /* leading space before alignment point */ - mchunkptr remainder; /* spare room at end to split off */ - unsigned long remainder_size; /* its size */ - INTERNAL_SIZE_T size; - - /* If need less alignment than we give anyway, just relay to malloc */ - - if (alignment <= MALLOC_ALIGNMENT) return _int_malloc(av, bytes); - - /* Otherwise, ensure that it is at least a minimum chunk size */ - - if (alignment < MINSIZE) alignment = MINSIZE; - - /* Make sure alignment is power of 2 (in case MINSIZE is not). */ - if ((alignment & (alignment - 1)) != 0) { - size_t a = MALLOC_ALIGNMENT * 2; - while ((unsigned long)a < (unsigned long)alignment) a <<= 1; - alignment = a; - } - - checked_request2size(bytes, nb); - - /* - Strategy: find a spot within that chunk that meets the alignment - request, and then possibly free the leading and trailing space. - */ - - - /* Call malloc with worst case padding to hit alignment. */ - - m = (char*)(_int_malloc(av, nb + alignment + MINSIZE)); - - if (m == 0) return 0; /* propagate failure */ - - p = mem2chunk(m); - - if ((((unsigned long)(m)) % alignment) != 0) { /* misaligned */ - - /* - Find an aligned spot inside chunk. Since we need to give back - leading space in a chunk of at least MINSIZE, if the first - calculation places us at a spot with less than MINSIZE leader, - we can move to the next aligned spot -- we've allocated enough - total room so that this is always possible. - */ - - brk = (char*)mem2chunk(((unsigned long)(m + alignment - 1)) & - -((signed long) alignment)); - if ((unsigned long)(brk - (char*)(p)) < MINSIZE) - brk += alignment; - - newp = (mchunkptr)brk; - leadsize = brk - (char*)(p); - newsize = chunksize(p) - leadsize; - - /* For mmapped chunks, just adjust offset */ - if (chunk_is_mmapped(p)) { - newp->prev_size = p->prev_size + leadsize; - set_head(newp, newsize|IS_MMAPPED); - return chunk2mem(newp); - } - - /* Otherwise, give back leader, use the rest */ - set_head(newp, newsize | PREV_INUSE | - (av != &main_arena ? NON_MAIN_ARENA : 0)); - set_inuse_bit_at_offset(newp, newsize); - set_head_size(p, leadsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); - _int_free(av, chunk2mem(p)); - p = newp; - - assert (newsize >= nb && - (((unsigned long)(chunk2mem(p))) % alignment) == 0); - } - - /* Also give back spare room at the end */ - if (!chunk_is_mmapped(p)) { - size = chunksize(p); - if ((unsigned long)(size) > (unsigned long)(nb + MINSIZE)) { - remainder_size = size - nb; - remainder = chunk_at_offset(p, nb); - set_head(remainder, remainder_size | PREV_INUSE | - (av != &main_arena ? NON_MAIN_ARENA : 0)); - set_head_size(p, nb); - _int_free(av, chunk2mem(remainder)); - } - } - - check_inuse_chunk(av, p); - return chunk2mem(p); -} - -#if 0 -/* - ------------------------------ calloc ------------------------------ -*/ - -#if __STD_C -Void_t* cALLOc(size_t n_elements, size_t elem_size) -#else -Void_t* cALLOc(n_elements, elem_size) size_t n_elements; size_t elem_size; -#endif -{ - mchunkptr p; - unsigned long clearsize; - unsigned long nclears; - INTERNAL_SIZE_T* d; - - Void_t* mem = mALLOc(n_elements * elem_size); - - if (mem != 0) { - p = mem2chunk(mem); - -#if MMAP_CLEARS - if (!chunk_is_mmapped(p)) /* don't need to clear mmapped space */ -#endif - { - /* - Unroll clear of <= 36 bytes (72 if 8byte sizes) - We know that contents have an odd number of - INTERNAL_SIZE_T-sized words; minimally 3. - */ - - d = (INTERNAL_SIZE_T*)mem; - clearsize = chunksize(p) - SIZE_SZ; - nclears = clearsize / sizeof(INTERNAL_SIZE_T); - assert(nclears >= 3); - - if (nclears > 9) - MALLOC_ZERO(d, clearsize); - - else { - *(d+0) = 0; - *(d+1) = 0; - *(d+2) = 0; - if (nclears > 4) { - *(d+3) = 0; - *(d+4) = 0; - if (nclears > 6) { - *(d+5) = 0; - *(d+6) = 0; - if (nclears > 8) { - *(d+7) = 0; - *(d+8) = 0; - } - } - } - } - } - } - return mem; -} -#endif /* 0 */ - -/* - ------------------------- independent_calloc ------------------------- -*/ - -Void_t** -#if __STD_C -_int_icalloc(mstate av, size_t n_elements, size_t elem_size, Void_t* chunks[]) -#else -_int_icalloc(av, n_elements, elem_size, chunks) -mstate av; size_t n_elements; size_t elem_size; Void_t* chunks[]; -#endif -{ - size_t sz[1] = {elem_size}; - /* opts arg of 3 means all elements are same size, and should be cleared */ - return iALLOc(av, n_elements, sz, 3, chunks); -} - -/* - ------------------------- independent_comalloc ------------------------- -*/ - -Void_t** -#if __STD_C -_int_icomalloc(mstate av, size_t n_elements, size_t sizes[], Void_t* chunks[]) -#else -_int_icomalloc(av, n_elements, sizes, chunks) -mstate av; size_t n_elements; size_t sizes[]; Void_t* chunks[]; -#endif -{ - return iALLOc(av, n_elements, sizes, 0, chunks); -} - - -/* - ------------------------------ ialloc ------------------------------ - ialloc provides common support for independent_X routines, handling all of - the combinations that can result. - - The opts arg has: - bit 0 set if all elements are same size (using sizes[0]) - bit 1 set if elements should be zeroed -*/ - - -static Void_t** -#if __STD_C -iALLOc(mstate av, size_t n_elements, size_t* sizes, int opts, Void_t* chunks[]) -#else -iALLOc(av, n_elements, sizes, opts, chunks) -mstate av; size_t n_elements; size_t* sizes; int opts; Void_t* chunks[]; -#endif -{ - INTERNAL_SIZE_T element_size; /* chunksize of each element, if all same */ - INTERNAL_SIZE_T contents_size; /* total size of elements */ - INTERNAL_SIZE_T array_size; /* request size of pointer array */ - Void_t* mem; /* malloced aggregate space */ - mchunkptr p; /* corresponding chunk */ - INTERNAL_SIZE_T remainder_size; /* remaining bytes while splitting */ - Void_t** marray; /* either "chunks" or malloced ptr array */ - mchunkptr array_chunk; /* chunk for malloced ptr array */ - int mmx; /* to disable mmap */ - INTERNAL_SIZE_T size; - INTERNAL_SIZE_T size_flags; - size_t i; - - /* Ensure initialization/consolidation */ - if (have_fastchunks(av)) malloc_consolidate(av); - - /* compute array length, if needed */ - if (chunks != 0) { - if (n_elements == 0) - return chunks; /* nothing to do */ - marray = chunks; - array_size = 0; - } - else { - /* if empty req, must still return chunk representing empty array */ - if (n_elements == 0) - return (Void_t**) _int_malloc(av, 0); - marray = 0; - array_size = request2size(n_elements * (sizeof(Void_t*))); - } - - /* compute total element size */ - if (opts & 0x1) { /* all-same-size */ - element_size = request2size(*sizes); - contents_size = n_elements * element_size; - } - else { /* add up all the sizes */ - element_size = 0; - contents_size = 0; - for (i = 0; i != n_elements; ++i) - contents_size += request2size(sizes[i]); - } - - /* subtract out alignment bytes from total to minimize overallocation */ - size = contents_size + array_size - MALLOC_ALIGN_MASK; - - /* - Allocate the aggregate chunk. - But first disable mmap so malloc won't use it, since - we would not be able to later free/realloc space internal - to a segregated mmap region. - */ - mmx = mp_.n_mmaps_max; /* disable mmap */ - mp_.n_mmaps_max = 0; - mem = _int_malloc(av, size); - mp_.n_mmaps_max = mmx; /* reset mmap */ - if (mem == 0) - return 0; - - p = mem2chunk(mem); - assert(!chunk_is_mmapped(p)); - remainder_size = chunksize(p); - - if (opts & 0x2) { /* optionally clear the elements */ - MALLOC_ZERO(mem, remainder_size - SIZE_SZ - array_size); - } - - size_flags = PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0); - - /* If not provided, allocate the pointer array as final part of chunk */ - if (marray == 0) { - array_chunk = chunk_at_offset(p, contents_size); - marray = (Void_t**) (chunk2mem(array_chunk)); - set_head(array_chunk, (remainder_size - contents_size) | size_flags); - remainder_size = contents_size; - } - - /* split out elements */ - for (i = 0; ; ++i) { - marray[i] = chunk2mem(p); - if (i != n_elements-1) { - if (element_size != 0) - size = element_size; - else - size = request2size(sizes[i]); - remainder_size -= size; - set_head(p, size | size_flags); - p = chunk_at_offset(p, size); - } - else { /* the final element absorbs any overallocation slop */ - set_head(p, remainder_size | size_flags); - break; - } - } - -#if MALLOC_DEBUG - if (marray != chunks) { - /* final element must have exactly exhausted chunk */ - if (element_size != 0) - assert(remainder_size == element_size); - else - assert(remainder_size == request2size(sizes[i])); - check_inuse_chunk(av, mem2chunk(marray)); - } - - for (i = 0; i != n_elements; ++i) - check_inuse_chunk(av, mem2chunk(marray[i])); -#endif - - return marray; -} - - -/* - ------------------------------ valloc ------------------------------ -*/ - -Void_t* -#if __STD_C -_int_valloc(mstate av, size_t bytes) -#else -_int_valloc(av, bytes) mstate av; size_t bytes; -#endif -{ - /* Ensure initialization/consolidation */ - if (have_fastchunks(av)) malloc_consolidate(av); - return _int_memalign(av, mp_.pagesize, bytes); -} - -/* - ------------------------------ pvalloc ------------------------------ -*/ - - -Void_t* -#if __STD_C -_int_pvalloc(mstate av, size_t bytes) -#else -_int_pvalloc(av, bytes) mstate av, size_t bytes; -#endif -{ - size_t pagesz; - - /* Ensure initialization/consolidation */ - if (have_fastchunks(av)) malloc_consolidate(av); - pagesz = mp_.pagesize; - return _int_memalign(av, pagesz, (bytes + pagesz - 1) & ~(pagesz - 1)); -} - - -/* - ------------------------------ malloc_trim ------------------------------ -*/ - -#if __STD_C -int mTRIm(size_t pad) -#else -int mTRIm(pad) size_t pad; -#endif -{ - mstate av = &main_arena; /* already locked */ - - /* Ensure initialization/consolidation */ - malloc_consolidate(av); - -#ifndef MORECORE_CANNOT_TRIM - return sYSTRIm(pad, av); -#else - return 0; -#endif -} - - -/* - ------------------------- malloc_usable_size ------------------------- -*/ - -#if __STD_C -size_t mUSABLe(Void_t* mem) -#else -size_t mUSABLe(mem) Void_t* mem; -#endif -{ - mchunkptr p; - if (mem != 0) { - p = mem2chunk(mem); - if (chunk_is_mmapped(p)) - return chunksize(p) - 2*SIZE_SZ; - else if (inuse(p)) - return chunksize(p) - SIZE_SZ; - } - return 0; -} - -/* - ---------------------- internal mallinfo ----------------------------- -*/ - -void _int_get_arena_info(mstate av, struct malloc_arena_info *mai) -{ - size_t i; - mbinptr b; - mchunkptr p; - size_t binavail = 0; - size_t fastavail = 0; - int nbinblocks = 0; - int nfastblocks = 0; - - (void)mutex_lock(&av->mutex); - - /* Ensure initialization */ - if (av->top == 0) malloc_consolidate(av); - - check_malloc_state(av); - - /* traverse fastbins */ - for (i = 0; i < NFASTBINS; ++i) { - for (p = av->fastbins[i]; p != 0; p = p->fd) { - ++nfastblocks; - fastavail += chunksize(p); - } - } - - /* traverse regular bins */ - for (i = 1; i < NBINS; ++i) { - b = bin_at(av, i); - for (p = last(b); p != b; p = p->bk) { - ++nbinblocks; - binavail += chunksize(p); - } - } - - mai->nfastblocks = nfastblocks; - mai->nbinblocks = nbinblocks; - mai->fastavail = fastavail; - mai->binavail = binavail; - mai->top_size = chunksize(av->top); - mai->system_mem = av->system_mem; - mai->max_system_mem = av->max_system_mem; - mai->stat_lock_direct = av->stat_lock_direct; - mai->stat_lock_loop = av->stat_lock_loop; - mai->stat_lock_wait = av->stat_lock_wait; - - (void)mutex_unlock(&av->mutex); -} - -void -_int_get_global_info (struct malloc_global_info *mgi) -{ - mgi->n_mmaps = mp_.n_mmaps; - mgi->max_n_mmaps = mp_.max_n_mmaps; - mgi->mmapped_mem = mp_.mmapped_mem; - mgi->max_mmapped_mem = mp_.max_mmapped_mem; - mgi->max_total_mem = mp_.max_total_mem; -#if THREAD_STATS - mgi->stat_n_heaps = stat_n_heaps; -#else - mgi->stat_n_heaps = 0; -#endif -} - -/* - ------------------------------ malloc_stats ------------------------------ -*/ - -/* Now in separate file, malloc-stats.c. */ - -/* - ------------------------------ mallopt ------------------------------ -*/ - -#if __STD_C -int mALLOPt(int param_number, int value) -#else -int mALLOPt(param_number, value) int param_number; int value; -#endif -{ - mstate av = &main_arena; - int res = 1; - - if(__malloc_initialized < 0) - ptmalloc_init (); - (void)mutex_lock(&av->mutex); - /* Ensure initialization/consolidation */ - malloc_consolidate(av); - - switch(param_number) { - case M_MXFAST: - if (value >= 0 && value <= MAX_FAST_SIZE) { - set_max_fast(av, value); - } - else - res = 0; - break; - - case M_TRIM_THRESHOLD: - mp_.trim_threshold = value; - break; - - case M_TOP_PAD: - mp_.top_pad = value; - break; - - case M_MMAP_THRESHOLD: -#if USE_ARENAS - /* Forbid setting the threshold too high. */ - if((unsigned long)value > HEAP_MAX_SIZE/2) - res = 0; - else -#endif - mp_.mmap_threshold = value; - break; - - case M_MMAP_MAX: -#if !HAVE_MMAP - if (value != 0) - res = 0; - else -#endif - mp_.n_mmaps_max = value; - break; - - case M_CHECK_ACTION: - check_action = value; - break; - } - (void)mutex_unlock(&av->mutex); - return res; -} - - -/* - -------------------- Alternative MORECORE functions -------------------- -*/ - - -/* - General Requirements for MORECORE. - - The MORECORE function must have the following properties: - - If MORECORE_CONTIGUOUS is false: - - * MORECORE must allocate in multiples of pagesize. It will - only be called with arguments that are multiples of pagesize. - - * MORECORE(0) must return an address that is at least - MALLOC_ALIGNMENT aligned. (Page-aligning always suffices.) - - else (i.e. If MORECORE_CONTIGUOUS is true): - - * Consecutive calls to MORECORE with positive arguments - return increasing addresses, indicating that space has been - contiguously extended. - - * MORECORE need not allocate in multiples of pagesize. - Calls to MORECORE need not have args of multiples of pagesize. - - * MORECORE need not page-align. - - In either case: - - * MORECORE may allocate more memory than requested. (Or even less, - but this will generally result in a malloc failure.) - - * MORECORE must not allocate memory when given argument zero, but - instead return one past the end address of memory from previous - nonzero call. This malloc does NOT call MORECORE(0) - until at least one call with positive arguments is made, so - the initial value returned is not important. - - * Even though consecutive calls to MORECORE need not return contiguous - addresses, it must be OK for malloc'ed chunks to span multiple - regions in those cases where they do happen to be contiguous. - - * MORECORE need not handle negative arguments -- it may instead - just return MORECORE_FAILURE when given negative arguments. - Negative arguments are always multiples of pagesize. MORECORE - must not misinterpret negative args as large positive unsigned - args. You can suppress all such calls from even occurring by defining - MORECORE_CANNOT_TRIM, - - There is some variation across systems about the type of the - argument to sbrk/MORECORE. If size_t is unsigned, then it cannot - actually be size_t, because sbrk supports negative args, so it is - normally the signed type of the same width as size_t (sometimes - declared as "intptr_t", and sometimes "ptrdiff_t"). It doesn't much - matter though. Internally, we use "long" as arguments, which should - work across all reasonable possibilities. - - Additionally, if MORECORE ever returns failure for a positive - request, and HAVE_MMAP is true, then mmap is used as a noncontiguous - system allocator. This is a useful backup strategy for systems with - holes in address spaces -- in this case sbrk cannot contiguously - expand the heap, but mmap may be able to map noncontiguous space. - - If you'd like mmap to ALWAYS be used, you can define MORECORE to be - a function that always returns MORECORE_FAILURE. - - If you are using this malloc with something other than sbrk (or its - emulation) to supply memory regions, you probably want to set - MORECORE_CONTIGUOUS as false. As an example, here is a custom - allocator kindly contributed for pre-OSX macOS. It uses virtually - but not necessarily physically contiguous non-paged memory (locked - in, present and won't get swapped out). You can use it by - uncommenting this section, adding some #includes, and setting up the - appropriate defines above: - - #define MORECORE osMoreCore - #define MORECORE_CONTIGUOUS 0 - - There is also a shutdown routine that should somehow be called for - cleanup upon program exit. - - #define MAX_POOL_ENTRIES 100 - #define MINIMUM_MORECORE_SIZE (64 * 1024) - static int next_os_pool; - void *our_os_pools[MAX_POOL_ENTRIES]; - - void *osMoreCore(int size) - { - void *ptr = 0; - static void *sbrk_top = 0; - - if (size > 0) - { - if (size < MINIMUM_MORECORE_SIZE) - size = MINIMUM_MORECORE_SIZE; - if (CurrentExecutionLevel() == kTaskLevel) - ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0); - if (ptr == 0) - { - return (void *) MORECORE_FAILURE; - } - // save ptrs so they can be freed during cleanup - our_os_pools[next_os_pool] = ptr; - next_os_pool++; - ptr = (void *) ((((unsigned long) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK); - sbrk_top = (char *) ptr + size; - return ptr; - } - else if (size < 0) - { - // we don't currently support shrink behavior - return (void *) MORECORE_FAILURE; - } - else - { - return sbrk_top; - } - } - - // cleanup any allocated memory pools - // called as last thing before shutting down driver - - void osCleanupMem(void) - { - void **ptr; - - for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++) - if (*ptr) - { - PoolDeallocate(*ptr); - *ptr = 0; - } - } - -*/ - -/* OMPI: Need to expose our own posix_memalign, or the wrong one will - be used */ -# include - -/* We need a wrapper function for one of the additions of POSIX. */ -int -posix_memalign (void **memptr, size_t alignment, size_t size) -{ - void *mem; - - /* OMPI change: the hook is us -- don't call the hook */ -#if 0 - __malloc_ptr_t (*hook) __MALLOC_PMT ((size_t, size_t, - __const __malloc_ptr_t)) = - __memalign_hook; -#endif - - /* Test whether the SIZE argument is valid. It must be a power of - two multiple of sizeof (void *). */ - if (alignment % sizeof (void *) != 0 - || !powerof2 (alignment / sizeof (void *)) != 0 - || alignment == 0) - return EINVAL; - - /* OMPI change: the hook is us -- don't call the hook */ -#if 0 - /* Call the hook here, so that caller is posix_memalign's caller - and not posix_memalign itself. */ - if (hook != NULL) - mem = (*hook)(alignment, size, RETURN_ADDRESS (0)); - else - mem = public_mEMALIGn (alignment, size); -#else - mem = public_mEMALIGn (alignment, size); -#endif - - if (mem != NULL) { - *memptr = mem; - return 0; - } - - return ENOMEM; -} - -#ifdef _LIBC -weak_alias (__posix_memalign, posix_memalign) - -strong_alias (__libc_calloc, __calloc) weak_alias (__libc_calloc, calloc) -strong_alias (__libc_free, __cfree) weak_alias (__libc_free, cfree) -strong_alias (__libc_free, __free) strong_alias (__libc_free, free) -strong_alias (__libc_malloc, __malloc) strong_alias (__libc_malloc, malloc) -strong_alias (__libc_memalign, __memalign) -weak_alias (__libc_memalign, memalign) -strong_alias (__libc_realloc, __realloc) strong_alias (__libc_realloc, realloc) -strong_alias (__libc_valloc, __valloc) weak_alias (__libc_valloc, valloc) -strong_alias (__libc_pvalloc, __pvalloc) weak_alias (__libc_pvalloc, pvalloc) -strong_alias (__libc_mallinfo, __mallinfo) -weak_alias (__libc_mallinfo, mallinfo) -strong_alias (__libc_mallopt, __mallopt) weak_alias (__libc_mallopt, mallopt) - -weak_alias (__malloc_usable_size, malloc_usable_size) -weak_alias (__malloc_trim, malloc_trim) -weak_alias (__malloc_get_state, malloc_get_state) -weak_alias (__malloc_set_state, malloc_set_state) - -#endif /* _LIBC */ - -/* ------------------------------------------------------------ -History: - -[see ftp://g.oswego.edu/pub/misc/malloc.c for the history of dlmalloc] - -*/ -/* - * Local variables: - * c-basic-offset: 2 - * End: - */ diff --git a/opal/mca/memory/linux/malloc.h b/opal/mca/memory/linux/malloc.h deleted file mode 100644 index bd0430e3b2f..00000000000 --- a/opal/mca/memory/linux/malloc.h +++ /dev/null @@ -1,300 +0,0 @@ -/* Prototypes and definition for malloc implementation. - Copyright (C) 1996,97,99,2000,2002,2003,2004 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library; if not, write to the Free - Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA - 02111-1307 USA. */ - -#ifndef _MALLOC_H -#define _MALLOC_H 1 - -/* add the opal config header file for the OPAL_DECLSPEC */ -#include "opal_config.h" - - -#ifdef _LIBC -#include -#endif - -/* - $Id: malloc.h,v 1.7 2004/08/08 12:34:57 wg Exp $ - `ptmalloc2', a malloc implementation for multiple threads without - lock contention, by Wolfram Gloger . - - VERSION 2.7.0 - - This work is mainly derived from malloc-2.7.0 by Doug Lea - , which is available from: - - ftp://gee.cs.oswego.edu/pub/misc/malloc.c - - This trimmed-down header file only provides function prototypes and - the exported data structures. For more detailed function - descriptions and compile-time options, see the source file - `malloc.c'. -*/ - -#if defined(__STDC__) || defined (__cplusplus) -# include -# define __malloc_ptr_t void * -#else -# undef size_t -# define size_t unsigned int -# undef ptrdiff_t -# define ptrdiff_t int -# define __malloc_ptr_t char * -#endif - -#ifdef _LIBC -/* Used by GNU libc internals. */ -# define __malloc_size_t size_t -# define __malloc_ptrdiff_t ptrdiff_t -#elif !defined __attribute_malloc__ -# define __attribute_malloc__ -#endif - -#ifdef __GNUC__ - -/* GCC can always grok prototypes. For C++ programs we add throw() - to help it optimize the function calls. But this works only with - gcc 2.8.x and egcs. */ -# if defined __cplusplus && (__GNUC__ >= 3 || __GNUC_MINOR__ >= 8) -# define __THROW throw () -# else -#ifdef __THROW -# undef __THROW -#endif -# define __THROW -# endif -# define __MALLOC_P(args) args __THROW -/* This macro will be used for functions which might take C++ callback - functions. */ -# define __MALLOC_PMT(args) args - -#else /* Not GCC. */ - -# define __THROW - -# if (defined __STDC__ && __STDC__) || defined __cplusplus - -# define __MALLOC_P(args) args -# define __MALLOC_PMT(args) args - -# ifndef __const -# define __const const -# endif - -# else /* Not ANSI C or C++. */ - -# define __MALLOC_P(args) () /* No prototypes. */ -# define __MALLOC_PMT(args) () - -# ifndef __const -# define __const -# endif - -# endif /* ANSI C or C++. */ - -#endif /* GCC. */ - -#ifndef NULL -# ifdef __cplusplus -# define NULL 0 -# else -# define NULL ((__malloc_ptr_t) 0) -# endif -#endif - -#ifdef __cplusplus -extern "C" { -#endif - -/* Nonzero if the malloc is already initialized. */ -#ifdef _LIBC -/* In the GNU libc we rename the global variable - `__malloc_initialized' to `__libc_malloc_initialized'. */ -# define __malloc_initialized __libc_malloc_initialized -#endif -extern int __malloc_initialized; - -/* Allocate SIZE bytes of memory. */ -OPAL_DECLSPEC extern __malloc_ptr_t malloc __MALLOC_P ((size_t __size)) __attribute_malloc__; - -/* Allocate NMEMB elements of SIZE bytes each, all initialized to 0. */ -OPAL_DECLSPEC extern __malloc_ptr_t calloc __MALLOC_P ((size_t __nmemb, size_t __size)) - __attribute_malloc__; - -/* Re-allocate the previously allocated block in __ptr, making the new - block SIZE bytes long. */ -OPAL_DECLSPEC extern __malloc_ptr_t realloc __MALLOC_P ((__malloc_ptr_t __ptr, - size_t __size)) - __attribute_malloc__; - -/* Free a block allocated by `malloc', `realloc' or `calloc'. */ -OPAL_DECLSPEC extern void free __MALLOC_P ((__malloc_ptr_t __ptr)); - -/* Free a block allocated by `calloc'. */ -OPAL_DECLSPEC extern void cfree __MALLOC_P ((__malloc_ptr_t __ptr)); - -/* Allocate SIZE bytes allocated to ALIGNMENT bytes. */ -OPAL_DECLSPEC extern __malloc_ptr_t memalign __MALLOC_P ((size_t __alignment, size_t __size)); - -/* Allocate SIZE bytes on a page boundary. */ -OPAL_DECLSPEC extern __malloc_ptr_t valloc __MALLOC_P ((size_t __size)) __attribute_malloc__; - -/* Equivalent to valloc(minimum-page-that-holds(n)), that is, round up - __size to nearest pagesize. */ -OPAL_DECLSPEC extern __malloc_ptr_t pvalloc __MALLOC_P ((size_t __size)) - __attribute_malloc__; - -/* Underlying allocation function; successive calls should return - contiguous pieces of memory. */ -OPAL_DECLSPEC extern __malloc_ptr_t (*__morecore) __MALLOC_PMT ((ptrdiff_t __size)); - -/* Default value of `__morecore'. */ -OPAL_DECLSPEC extern __malloc_ptr_t __default_morecore __MALLOC_P ((ptrdiff_t __size)) - __attribute_malloc__; - -/* SVID2/XPG mallinfo structure */ - -struct mallinfo { - int arena; /* non-mmapped space allocated from system */ - int ordblks; /* number of free chunks */ - int smblks; /* number of fastbin blocks */ - int hblks; /* number of mmapped regions */ - int hblkhd; /* space in mmapped regions */ - int usmblks; /* maximum total allocated space */ - int fsmblks; /* space available in freed fastbin blocks */ - int uordblks; /* total allocated space */ - int fordblks; /* total free space */ - int keepcost; /* top-most, releasable (via malloc_trim) space */ -}; - -/* Returns a copy of the updated current mallinfo. */ -OPAL_DECLSPEC extern struct mallinfo mallinfo __MALLOC_P ((void)); - -/* SVID2/XPG mallopt options */ -#ifndef M_MXFAST -# define M_MXFAST 1 /* maximum request size for "fastbins" */ -#endif -#ifndef M_NLBLKS -# define M_NLBLKS 2 /* UNUSED in this malloc */ -#endif -#ifndef M_GRAIN -# define M_GRAIN 3 /* UNUSED in this malloc */ -#endif -#ifndef M_KEEP -# define M_KEEP 4 /* UNUSED in this malloc */ -#endif - -/* mallopt options that actually do something */ -#define M_TRIM_THRESHOLD -1 -#define M_TOP_PAD -2 -#define M_MMAP_THRESHOLD -3 -#define M_MMAP_MAX -4 -#define M_CHECK_ACTION -5 - -/* General SVID/XPG interface to tunable parameters. */ -OPAL_DECLSPEC extern int mallopt __MALLOC_P ((int __param, int __val)); - -/* Release all but __pad bytes of freed top-most memory back to the - system. Return 1 if successful, else 0. */ -OPAL_DECLSPEC extern int malloc_trim __MALLOC_P ((size_t __pad)); - -/* Report the number of usable allocated bytes associated with allocated - chunk __ptr. */ -OPAL_DECLSPEC extern size_t malloc_usable_size __MALLOC_P ((__malloc_ptr_t __ptr)); - -/* Prints brief summary statistics on stderr. */ -OPAL_DECLSPEC extern void malloc_stats __MALLOC_P ((void)); - -/* Record the state of all malloc variables in an opaque data structure. */ -OPAL_DECLSPEC extern __malloc_ptr_t malloc_get_state __MALLOC_P ((void)); - -/* Restore the state of all malloc variables from data obtained with - malloc_get_state(). */ -OPAL_DECLSPEC extern int malloc_set_state __MALLOC_P ((__malloc_ptr_t __ptr)); - -/* Called once when malloc is initialized; redefining this variable in - the application provides the preferred way to set up the hook - pointers. */ -OPAL_DECLSPEC extern void (*__malloc_initialize_hook) __MALLOC_PMT ((void)); -/* Hooks for debugging and user-defined versions. */ -OPAL_DECLSPEC extern void (*__free_hook) __MALLOC_PMT ((__malloc_ptr_t __ptr, - __const __malloc_ptr_t)); -OPAL_DECLSPEC extern __malloc_ptr_t (*__malloc_hook) __MALLOC_PMT ((size_t __size, - __const __malloc_ptr_t)); -OPAL_DECLSPEC extern __malloc_ptr_t (*__realloc_hook) __MALLOC_PMT ((__malloc_ptr_t __ptr, - size_t __size, - __const __malloc_ptr_t)); -OPAL_DECLSPEC extern __malloc_ptr_t (*__memalign_hook) __MALLOC_PMT ((size_t __alignment, - size_t __size, - __const __malloc_ptr_t)); -OPAL_DECLSPEC extern void (*__after_morecore_hook) __MALLOC_PMT ((void)); - -/* Activate a standard set of debugging hooks. */ -OPAL_DECLSPEC extern void __malloc_check_init __MALLOC_P ((void)); - -/* Internal routines, operating on "arenas". */ -struct malloc_state; -typedef struct malloc_state *mstate; - -OPAL_DECLSPEC extern mstate _int_new_arena __MALLOC_P ((size_t __ini_size)); -OPAL_DECLSPEC extern __malloc_ptr_t _int_malloc __MALLOC_P ((mstate __m, size_t __size)); -OPAL_DECLSPEC extern void _int_free __MALLOC_P ((mstate __m, __malloc_ptr_t __ptr)); -OPAL_DECLSPEC extern __malloc_ptr_t _int_realloc __MALLOC_P ((mstate __m, - __malloc_ptr_t __ptr, - size_t __size)); -OPAL_DECLSPEC extern __malloc_ptr_t _int_memalign __MALLOC_P ((mstate __m, size_t __alignment, - size_t __size)); -/* Return arena number __n, or 0 if out of bounds. Arena 0 is the - main arena. */ -OPAL_DECLSPEC extern mstate _int_get_arena __MALLOC_P ((int __n)); - -/* Implementation-specific mallinfo. More detailed than mallinfo, and - also works for size_t wider than int. */ -struct malloc_arena_info { - int nfastblocks; /* number of freed "fastchunks" */ - int nbinblocks; /* number of available chunks in bins */ - size_t fastavail; /* total space in freed "fastchunks" */ - size_t binavail; /* total space in binned chunks */ - size_t top_size; /* size of top chunk */ - size_t system_mem; /* bytes allocated from system in this arena */ - size_t max_system_mem; /* max. bytes allocated from system */ - /* Statistics for locking. Only kept if THREAD_STATS is defined - at compile time. */ - long stat_lock_direct, stat_lock_loop, stat_lock_wait; -}; - -struct malloc_global_info { - int n_mmaps; /* number of mmap'ed chunks */ - int max_n_mmaps; /* max. number of mmap'ed chunks reached */ - size_t mmapped_mem; /* total bytes allocated in mmap'ed chunks */ - size_t max_mmapped_mem; /* max. bytes allocated in mmap'ed chunks */ - size_t max_total_mem; /* only kept for NO_THREADS */ - int stat_n_heaps; /* only kept if THREAD_STATS is defined */ -}; - -OPAL_DECLSPEC extern void _int_get_arena_info __MALLOC_P ((mstate __m, - struct malloc_arena_info *__ma)); -OPAL_DECLSPEC extern void _int_get_global_info __MALLOC_P ((struct malloc_global_info *__m)); - -OPAL_DECLSPEC extern int posix_memalign (void **memptr, size_t alignment, size_t size); - -#ifdef __cplusplus -} /* end of extern "C" */ -#endif - -#endif /* malloc.h */ diff --git a/opal/mca/memory/linux/memory_linux.h b/opal/mca/memory/linux/memory_linux.h deleted file mode 100644 index 9334f3ae828..00000000000 --- a/opal/mca/memory/linux/memory_linux.h +++ /dev/null @@ -1,90 +0,0 @@ -/* - * Copyright (c) 2010 Cisco Systems, Inc. All rights reserved. - * $COPYRIGHT$ - * - * Additional copyrights may follow - * - * $HEADER$ - */ - -#ifndef OPAL_MEMORY_LINUX_H -#define OPAL_MEMORY_LINUX_H - -#include "opal_config.h" - -#include "opal/mca/memory/memory.h" - -BEGIN_C_DECLS - -/* Component structure */ - -typedef struct opal_memory_linux_component_t { - opal_memory_base_component_2_0_0_t super; - - /* Component data */ - int verbose_level; - int enable_ummunotify; - int enable_ptmalloc2; - -#if MEMORY_LINUX_UMMUNOTIFY - /* Ummunotify-specific data */ - int ummunotify_fd; -#endif - -#if MEMORY_LINUX_MALLOC_ALIGN_ENABLED - int use_memalign; - size_t memalign_threshold; -#endif - -#if MEMORY_LINUX_PTMALLOC2 - /* Ptmalloc2-specific data. Note that these variables are all marked as volatile. - * This is needed because of what may be a buggy optimization in the GCC 4.9.2 - * compilers and later. These variables are used in different code paths which the - * compiler is not aware of. - * Extra details located at these URLs: - * Open MPI User List: http://www.open-mpi.org/community/lists/users/2015/06/27039.php - * Bug Discussion: https://github.com/open-mpi/ompi/pull/625 - * GCC Discussion: https://gcc.gnu.org/ml/gcc-bugs/2015-06/msg00757.html - */ - volatile bool free_invoked; - volatile bool malloc_invoked; - volatile bool realloc_invoked; - volatile bool memalign_invoked; - volatile bool munmap_invoked; -#endif -} opal_memory_linux_component_t; - -/* memory_linux_component.c */ - -extern opal_memory_linux_component_t mca_memory_linux_component; - - -#if MEMORY_LINUX_UMMUNOTIFY -/* memory_linux_ummunotify.c */ -int opal_memory_linux_ummunotify_open(void); -int opal_memory_linux_ummunotify_close(void); -#endif /* MEMORY_LINUX_UMMUNOTIFY */ - -#if MEMORY_LINUX_PTMALLOC2 -/* memory_linux_ptmalloc2.c */ -int opal_memory_linux_ptmalloc2_open(void); -int opal_memory_linux_ptmalloc2_close(void); - -/* memory_linux_munmap.c */ -OPAL_DECLSPEC int opal_memory_linux_free_ptmalloc2_munmap(void *start, size_t length, int from_alloc); -OPAL_DECLSPEC int munmap(void* addr, size_t len); -#endif /* !MEMORY_LINUX_PTMALLOC2 */ - -#if MEMORY_LINUX_HAVE_MALLOC_HOOK_SUPPORT -OPAL_DECLSPEC void opal_memory_linux_malloc_init_hook(void); -#endif /* MEMORY_LINUX_HAVE_MALLOC_HOOK_SUPPORT */ - -#if MEMORY_LINUX_MALLOC_ALIGN_ENABLED -OPAL_DECLSPEC void opal_memory_linux_malloc_set_alignment(int use_memalign, size_t memalign_threshold); -#endif /* MEMORY_LINUX_MALLOC_ALIGN_ENABLED */ - -extern bool opal_memory_linux_opened; - -END_C_DECLS - -#endif diff --git a/opal/mca/memory/linux/memory_linux_component.c b/opal/mca/memory/linux/memory_linux_component.c deleted file mode 100644 index 323ab191344..00000000000 --- a/opal/mca/memory/linux/memory_linux_component.c +++ /dev/null @@ -1,407 +0,0 @@ -/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ -/* - * Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana - * University Research and Technology - * Corporation. All rights reserved. - * Copyright (c) 2004-2005 The University of Tennessee and The University - * of Tennessee Research Foundation. All rights - * reserved. - * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, - * University of Stuttgart. All rights reserved. - * Copyright (c) 2004-2005 The Regents of the University of California. - * All rights reserved. - * Copyright (c) 2009-2014 Cisco Systems, Inc. All rights reserved. - * Copyright (c) 2013-2016 Los Alamos National Security, LLC. All rights - * reserved. - * Copyright (c) 2016 Research Organization for Information Science - * and Technology (RIST). All rights reserved. - * $COPYRIGHT$ - * - * Additional copyrights may follow - * - * $HEADER$ - */ - -/* This component basically fronts two different memory management - schemes: the Linux "ummunotify" kernel module and hooking in a - substitute ptmalloc2 allocator. Both of these mechanisms are - unified under a single component because the "memory" framework - both only allows one component to be selected, and that one - component must be compile-time linked into libopen-pal. Hence, if - we want to try to use either one of these mechanisms, we have to - have them both in a single component. - - When using ptmalloc2, the goal of this component is to wholly - replace the underlying allocator with our internal ptmalloc2 - allocator. See the file README-open-mpi.txt for details of how it - works. - - When using ummunotify, we can probe to find out when the MMU map - has been changed (i.e., memory has been released back to the OS). */ - -#include "opal_config.h" - -#if HAVE_MALLOC_H -#include -#endif - -#include "opal/constants.h" -#include "opal/mca/base/mca_base_var.h" -#include "opal/mca/memory/memory.h" -#include "opal/mca/memory/base/empty.h" -#include "opal/memoryhooks/memory.h" -#include "opal/util/output.h" -#include "opal/util/show_help.h" - -#include "opal/mca/memory/linux/memory_linux.h" -#undef opal_memory_changed -#include "opal/mca/memory/linux/public.h" - -static int linux_open(void); -static int linux_close(void); -static int linux_register(void); -static int linux_query(int *); - -#if MEMORY_LINUX_UMMUNOTIFY -static bool ummunotify_opened = false; -#endif -#if MEMORY_LINUX_PTMALLOC2 -static bool ptmalloc2_opened = false; -#endif - -bool opal_memory_linux_disable = false; -static int mca_memory_linux_priority; - -bool opal_memory_linux_opened = false; - -opal_memory_linux_component_t mca_memory_linux_component = { - /* First, the opal_memory_base_component_2_0_0_t */ - { - /* First, the mca_component_t struct containing meta - information about the component itself */ - .memoryc_version = { - OPAL_MEMORY_BASE_VERSION_2_0_0, - - /* Component name and version */ - .mca_component_name = "linux", - MCA_BASE_MAKE_VERSION(component, OPAL_MAJOR_VERSION, OPAL_MINOR_VERSION, - OPAL_RELEASE_VERSION), - - /* Component open and close functions */ - .mca_open_component = linux_open, - .mca_close_component = linux_close, - .mca_register_component_params = linux_register, - }, - .memoryc_data = { - /* The component is checkpoint ready */ - MCA_BASE_METADATA_PARAM_CHECKPOINT - }, - - /* Memory framework functions. These function pointer values - are replaced by memory_linux_ummunotify.c at run time if we - end up using ummunotify support. */ - .memoryc_init_hook = opal_memory_linux_malloc_init_hook, - .memoryc_query = linux_query, - .memoryc_register = opal_memory_base_component_register_empty, - .memoryc_deregister = opal_memory_base_component_deregister_empty, -#if MEMORY_LINUX_MALLOC_ALIGN_ENABLED - .memoryc_set_alignment = opal_memory_linux_malloc_set_alignment, -#else - .memoryc_set_alignment = opal_memory_base_component_set_alignment_empty, -#endif - }, - - /* Component-specific data, filled in later (compiler will 0/NULL - it out) */ -}; - -static bool ptmalloc2_available = MEMORY_LINUX_PTMALLOC2; -static bool ummunotify_available = MEMORY_LINUX_UMMUNOTIFY; - -#if MEMORY_LINUX_MALLOC_ALIGN_ENABLED - -static void *(*prev_malloc_hook)(size_t, const void *); - -/* This is a memory allocator hook. The purpose of this is to make - * every malloc aligned. - * There two basic cases here: - * - * 1. Memory manager for Open MPI is enabled. Then memalign below will - * be overridden by __memalign_hook which is set to - * opal_memory_linux_memalign_hook. Thus, _malloc_hook is going to - * use opal_memory_linux_memalign_hook. - * - * 2. No memory manager support. The memalign below is just regular glibc - * memalign which will be called through __malloc_hook instead of malloc. - */ -static void *_opal_memory_linux_malloc_align_hook(size_t sz, const void* caller); - -static mca_base_var_enum_value_t align_values[] = { - {-1, "disabled"}, - {0, "0"}, - {32, "32"}, - {64, "64"}, - {0, NULL} -}; -#endif /* MEMORY_LINUX_MALLOC_ALIGN_ENABLED */ - - -/* - * Register MCA params - */ -static int linux_register(void) -{ -#if MEMORY_LINUX_MALLOC_ALIGN_ENABLED - mca_base_var_enum_t *new_enum; -#endif - int ret; - /* Information only */ - ret = mca_base_component_var_register (&mca_memory_linux_component.super.memoryc_version, - "ptmalloc2_available", - "Whether ptmalloc2 support is included in Open MPI or not (1 = yes, 0 = no)", - MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0, - OPAL_INFO_LVL_3, - MCA_BASE_VAR_SCOPE_CONSTANT, - &ptmalloc2_available); - if (0 > ret) { - return ret; - } - - ret = mca_base_component_var_register (&mca_memory_linux_component.super.memoryc_version, - "ummunotify_available", - "Whether ummunotify support is included in Open MPI or not (1 = yes, 0 = no)", - MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0, - OPAL_INFO_LVL_3, - MCA_BASE_VAR_SCOPE_CONSTANT, - &ummunotify_available); - if (0 > ret) { - return ret; - } - - /* Allow user to manually enable/disable */ - mca_memory_linux_component.enable_ptmalloc2 = -1; - ret = mca_base_component_var_register (&mca_memory_linux_component.super.memoryc_version, - "ptmalloc2_enable", - "Whether to enable ptmalloc2 support or not (negative = try to enable, but continue even if support is not available, 0 = do not enable support, positive = try to enable and fail if support is not available)", - MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE, - OPAL_INFO_LVL_3, - MCA_BASE_VAR_SCOPE_ALL_EQ, - &mca_memory_linux_component.enable_ptmalloc2); - if (0 > ret) { - return ret; - } - - mca_memory_linux_component.enable_ummunotify = -1; - ret = mca_base_component_var_register (&mca_memory_linux_component.super.memoryc_version, - "ummunotify_enable", - "Whether to enable ummunotify support or not (negative = try to enable, but continue even if support is not available, 0 = do not enable support, positive = try to enable and fail if support is not available)", - MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE, - OPAL_INFO_LVL_3, - MCA_BASE_VAR_SCOPE_ALL_EQ, - &mca_memory_linux_component.enable_ummunotify); - if (0 > ret) { - return ret; - } - - opal_memory_linux_disable = false; - (void) mca_base_component_var_register (&mca_memory_linux_component.super.memoryc_version, - "disable", - "If this MCA parameter is set to 1 **VIA ENVIRONMENT VARIABLE ONLY*** (this MCA parameter *CANNOT* be set in a file or on the mpirun command line!), this component will be disabled and will not attempt to use either ummunotify or memory hook support", - MCA_BASE_VAR_TYPE_BOOL, NULL, 0, MCA_BASE_VAR_FLAG_ENVIRONMENT_ONLY, - OPAL_INFO_LVL_3, - MCA_BASE_VAR_SCOPE_READONLY, - &opal_memory_linux_disable); - if (0 > ret) { - return ret; - } - -#if MEMORY_LINUX_MALLOC_ALIGN_ENABLED - (void)mca_base_var_enum_create("memory_linux_memalign", align_values, &new_enum); - mca_memory_linux_component.use_memalign = -1; - ret = mca_base_component_var_register(&mca_memory_linux_component.super.memoryc_version, - "memalign", - "[64 | 32 | 0] - Enable memory alignment for all malloc calls.", - MCA_BASE_VAR_TYPE_INT, - new_enum, - 0, - 0, - OPAL_INFO_LVL_5, - MCA_BASE_VAR_SCOPE_READONLY, - &mca_memory_linux_component.use_memalign); - OBJ_RELEASE(new_enum); - if (0 > ret) { - return ret; - } - - mca_memory_linux_component.memalign_threshold = 12288; - ret = mca_base_component_var_register(&mca_memory_linux_component.super.memoryc_version, - "memalign_threshold", - "Allocating memory more than memory_linux_memalign_threshold" - "bytes will automatically be aligned to the value of memory_linux_memalign bytes." - "(default: 12288)", - MCA_BASE_VAR_TYPE_SIZE_T, - NULL, - 0, - 0, - OPAL_INFO_LVL_5, - MCA_BASE_VAR_SCOPE_READONLY, - &mca_memory_linux_component.memalign_threshold); - if (0 > ret) { - return ret; - } - - mca_memory_linux_priority = 50; - ret = mca_base_component_var_register (&mca_memory_linux_component.super.memoryc_version, - "priority", "Priority of the linux memory hook component", - MCA_BASE_VAR_TYPE_INT, NULL, 0, 0, OPAL_INFO_LVL_5, - MCA_BASE_VAR_SCOPE_CONSTANT, &mca_memory_linux_priority); - if (0 > ret) { - return ret; - } -#endif /* MEMORY_LINUX_MALLOC_ALIGN_ENABLED */ - - return (0 > ret) ? ret : OPAL_SUCCESS; -} - -static int linux_query (int *priority) -{ - *priority = mca_memory_linux_priority; - return OPAL_SUCCESS; -} - -static int linux_open(void) -{ - const int *verbose = NULL; - int i; - - i = mca_base_var_find("opal", "memory", NULL, "base_verbose"); - mca_base_var_get_value(i, &verbose, NULL, NULL); - mca_memory_linux_component.verbose_level = verbose ? verbose[0] : 0; - - /* Try initializing ummunotify first; if that fails, try - ptmalloc2. */ -#if MEMORY_LINUX_UMMUNOTIFY - if (mca_memory_linux_component.enable_ummunotify) { - if (mca_memory_linux_component.verbose_level >= 10) { - opal_output(0, "memory:linux: attempting to initialize ummunotify support"); - } - if (OPAL_SUCCESS == opal_memory_linux_ummunotify_open()) { - ummunotify_opened = true; - if (mca_memory_linux_component.verbose_level >= 10) { - opal_output(0, "memory:linux: ummunotify successfully initialized; we'll use that"); - } - goto done; - } - if (mca_memory_linux_component.verbose_level >= 10) { - opal_output(0, "memory:linux: ummunotify failed to initialize"); - } - } -#endif - -#if MEMORY_LINUX_PTMALLOC2 - if (mca_memory_linux_component.enable_ptmalloc2) { - if (mca_memory_linux_component.verbose_level >= 10) { - opal_output(0, "memory:linux: attempting to initialize ptmalloc2 support"); - } - if (OPAL_SUCCESS == opal_memory_linux_ptmalloc2_open()) { - ptmalloc2_opened = true; - if (mca_memory_linux_component.verbose_level >= 10) { - opal_output(0, "memory:linux: ptmalloc2 successfully initialized; we'll use that"); - } - goto done; - } - if (mca_memory_linux_component.verbose_level >= 10) { - opal_output(0, "memory:linux: ptmalloc2 failed to initialize"); - } - } -#endif - - /* We can return OPAL_ERR_NOT_AVAILABLE if nothing is - available; that will make the MCA base silently disregard this - component. */ - - if (mca_memory_linux_component.verbose_level >= 10) { - opal_output(0, "memory:linux: no memory hooks available in this process"); - } - return OPAL_ERR_NOT_AVAILABLE; - -done: - -#if MEMORY_LINUX_MALLOC_ALIGN_ENABLED - /* save original call */ - prev_malloc_hook = NULL; - - if (mca_memory_linux_component.use_memalign > 0 && - (opal_mem_hooks_support_level() & - (OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_CHUNK_SUPPORT)) != 0) { - prev_malloc_hook = __malloc_hook; - __malloc_hook = _opal_memory_linux_malloc_align_hook; - } -#endif /* MEMORY_LINUX_MALLOC_ALIGN_ENABLED */ - opal_memory_linux_opened = true; - - return OPAL_SUCCESS; -} - -static int linux_close(void) -{ - int v = mca_memory_linux_component.verbose_level; - -#if MEMORY_LINUX_MALLOC_ALIGN_ENABLED - /* restore original call */ - if (prev_malloc_hook) { - __malloc_hook = prev_malloc_hook; - prev_malloc_hook = NULL; - } -#endif /* MEMORY_LINUX_MALLOC_ALIGN_ENABLED */ - -#if MEMORY_LINUX_UMMUNOTIFY - if (ummunotify_opened) { - if (v >= 10) { - opal_output(0, "memory:linux: shutting down ummunotify support"); - } - opal_memory_linux_ummunotify_close(); - ummunotify_opened = false; - } -#endif -#if MEMORY_LINUX_PTMALLOC2 - if (ptmalloc2_opened) { - if (v >= 10) { - opal_output(0, "memory:linux: shutting down ptmalloc2 support"); - } - opal_memory_linux_ptmalloc2_close(); - ptmalloc2_opened = false; - } -#endif - - return OPAL_SUCCESS; -} - -#if MEMORY_LINUX_MALLOC_ALIGN_ENABLED -void opal_memory_linux_malloc_set_alignment(int use_memalign, size_t memalign_threshold) -{ - /* ignore cases when this capability is enabled explicitly using - * mca variables - */ - if ((NULL == prev_malloc_hook) && (-1 == mca_memory_linux_component.use_memalign)) { - if (use_memalign == 0 || use_memalign == 32 || use_memalign == 64) { - mca_memory_linux_component.use_memalign = use_memalign; - mca_memory_linux_component.memalign_threshold = memalign_threshold; - if ((opal_mem_hooks_support_level() & - (OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_CHUNK_SUPPORT)) != 0) { - prev_malloc_hook = __malloc_hook; - __malloc_hook = _opal_memory_linux_malloc_align_hook; - } - } - } -} - -static void *_opal_memory_linux_malloc_align_hook(size_t sz, const void* caller) -{ - if (sz < mca_memory_linux_component.memalign_threshold) { - return prev_malloc_hook(sz, caller); - } else { - return memalign(mca_memory_linux_component.use_memalign, sz); - } -} -#endif /* MEMORY_LINUX_MALLOC_ALIGN_ENABLED */ diff --git a/opal/mca/memory/linux/memory_linux_munmap.c b/opal/mca/memory/linux/memory_linux_munmap.c deleted file mode 100644 index bb1783973a6..00000000000 --- a/opal/mca/memory/linux/memory_linux_munmap.c +++ /dev/null @@ -1,92 +0,0 @@ -/* - * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana - * University Research and Technology - * Corporation. All rights reserved. - * Copyright (c) 2004-2005 The University of Tennessee and The University - * of Tennessee Research Foundation. All rights - * reserved. - * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, - * University of Stuttgart. All rights reserved. - * Copyright (c) 2004-2005 The Regents of the University of California. - * All rights reserved. - * Copyright (c) 2010 Cisco Systems, Inc. All rights reserved. - * $COPYRIGHT$ - * - * Additional copyrights may follow - * - * $HEADER$ - */ - -#include "opal_config.h" - -#include -#include -#include -#if defined(HAVE___MUNMAP) -/* here so we only include others if we absolutely have to */ -#elif defined(HAVE_SYSCALL) -#include -#include -#endif -#if defined(HAVE_DLSYM) -#ifndef __USE_GNU -#define __USE_GNU -#endif -#include -#endif - -#include "opal/memoryhooks/memory_internal.h" - -#include "memory_linux.h" - -/* - * munmap is always intercepted - */ -#if defined(HAVE___MUNMAP) -int __munmap(void* addr, size_t len); -#endif - - -/* intercept munmap, as the user can give back memory that way as well. */ -OPAL_DECLSPEC int munmap(void* addr, size_t len) -{ - return opal_memory_linux_free_ptmalloc2_munmap(addr, len, 0); -} - - -/* three ways to call munmap. Prefered is to just call syscall, so - that we can intercept both munmap and __munmap. If that isn't - possible, try calling __munmap from munmap and let __munmap go. If - that doesn't work, try dlsym */ -int opal_memory_linux_free_ptmalloc2_munmap(void *start, size_t length, - int from_alloc) -{ -#if !defined(HAVE___MUNMAP) && \ - !(defined(HAVE_SYSCALL) && defined(__NR_munmap)) && defined(HAVE_DLSYM) - static int (*realmunmap)(void*, size_t); -#endif - - mca_memory_linux_component.munmap_invoked = true; - - opal_mem_hooks_release_hook(start, length, from_alloc); - -#if defined(HAVE___MUNMAP) - return __munmap(start, length); -#elif defined(HAVE_SYSCALL) && defined(__NR_munmap) - return syscall(__NR_munmap, start, length); -#elif defined(HAVE_DLSYM) - if (NULL == realmunmap) { - union { - int (*munmap_fp)(void*, size_t); - void *munmap_p; - } tmp; - - tmp.munmap_p = dlsym(RTLD_NEXT, "munmap"); - realmunmap = tmp.munmap_fp; - } - - return realmunmap(start, length); -#else - #error "Can not determine how to call munmap" -#endif -} diff --git a/opal/mca/memory/linux/memory_linux_ptmalloc2.c b/opal/mca/memory/linux/memory_linux_ptmalloc2.c deleted file mode 100644 index 00a50dc77c7..00000000000 --- a/opal/mca/memory/linux/memory_linux_ptmalloc2.c +++ /dev/null @@ -1,136 +0,0 @@ -/* - * Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana - * University Research and Technology - * Corporation. All rights reserved. - * Copyright (c) 2004-2005 The University of Tennessee and The University - * of Tennessee Research Foundation. All rights - * reserved. - * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, - * University of Stuttgart. All rights reserved. - * Copyright (c) 2004-2005 The Regents of the University of California. - * All rights reserved. - * Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. - * $COPYRIGHT$ - * - * Additional copyrights may follow - * - * $HEADER$ - */ - -#include "opal_config.h" - -#include -#include -#include - -#include "opal/constants.h" -#include "opal/util/output.h" -#include "opal/memoryhooks/memory.h" -#include "opal/memoryhooks/memory_internal.h" - -#include "opal/mca/memory/linux/memory_linux.h" - - -/* Need to call a function in hooks.c to ensure that all those symbols - get pulled in at link time (e.g., when building libmpi.a, so that - those symbols end up in the final executable -- especially if we - use --disable-dlopen and therefore -Wl,--export-dynamic isn't used - when we build OMPI). */ -extern void opal_memory_linux_hook_pull(bool *want_hooks); - - -/* - * Try to initialize ptmalloc2 - */ -int opal_memory_linux_ptmalloc2_open(void) -{ - int val = 0; - void *p; - bool want_hooks = true; - - /* Call a [somewhat] dummy function in hooks.c. ***Do not remove - this call!*** See comment at the beginning of this file - explaining why it is here. It will also check to see if an - environment variable has been set to disable this component - (note that OPAL_ERR_NOT_AVAILABLE is a special return value - that will silently fail the open component call; all others - will issue an error). */ - opal_memory_linux_hook_pull(&want_hooks); - if (!want_hooks) { - return OPAL_ERR_NOT_AVAILABLE; - } - - /* We will also provide malloc/free support if we've been - activated. We don't exclusively rely on the - __malloc_initialize_hook() previously being called because it's - possible that our hook was called, but then someone else reset - the hooks to point to something else (i.e., before MPI_INIT). - So explicitly test here if our hooks are still in place. If - they are, then enable FREE|CHUNK_SUPPORT. If not, then don't - enable that support -- just leave it at MUNMAP_SUPPORT. - - (Look in hooks.c for the __malloc_initialize_hook setup) */ - - /* Do a simple set of tests to see if our hooks are still the ones - installed. Explicitly reset the flags indicating that our - functions were invoked */ - p = malloc(1024 * 1024 * 4); - if (NULL == p) { - return OPAL_ERR_OUT_OF_RESOURCE; - } - p = realloc(p, 1024 * 1024 * 4 + 32); - if (NULL == p) { - return OPAL_ERR_OUT_OF_RESOURCE; - } - free(p); - p = memalign(4, 1024 * 1024); - if (NULL == p) { - return OPAL_ERR_OUT_OF_RESOURCE; - } - free(p); - -#if HAVE_POSIX_MEMALIGN - /* Double check for posix_memalign, too */ - if (mca_memory_linux_component.memalign_invoked) { - mca_memory_linux_component.memalign_invoked = false; - if (0 != posix_memalign(&p, sizeof(void*), 1024 * 1024)) { - return OPAL_ERR_IN_ERRNO; - } - free(p); - } -#endif - - if (mca_memory_linux_component.malloc_invoked && - mca_memory_linux_component.realloc_invoked && - mca_memory_linux_component.memalign_invoked && - mca_memory_linux_component.free_invoked) { - /* Happiness; our functions were invoked */ - val |= OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_CHUNK_SUPPORT; - } - - /* Check if our mmap layering is working */ - p = mmap(NULL, 4096, PROT_READ, (MAP_ANONYMOUS | MAP_PRIVATE), -1, 0); - if (MAP_FAILED == p) { - return OPAL_ERR_OUT_OF_RESOURCE; - } - munmap(p, 4096); - if (mca_memory_linux_component.munmap_invoked) { - val |= OPAL_MEMORY_MUNMAP_SUPPORT; - } - - /* All done */ - if (val > 0) { - opal_mem_hooks_set_support(val); - return OPAL_SUCCESS; - } - return OPAL_ERR_NOT_AVAILABLE; -} - - -int opal_memory_linux_ptmalloc2_close(void) -{ - /* Nothing to do, really. This function exists just for - symmetry. */ - - return OPAL_SUCCESS; -} diff --git a/opal/mca/memory/linux/memory_linux_ummunotify.c b/opal/mca/memory/linux/memory_linux_ummunotify.c deleted file mode 100644 index 4f85416d0c8..00000000000 --- a/opal/mca/memory/linux/memory_linux_ummunotify.c +++ /dev/null @@ -1,235 +0,0 @@ -/* - * Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana - * University Research and Technology - * Corporation. All rights reserved. - * Copyright (c) 2004-2005 The University of Tennessee and The University - * of Tennessee Research Foundation. All rights - * reserved. - * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, - * University of Stuttgart. All rights reserved. - * Copyright (c) 2004-2005 The Regents of the University of California. - * All rights reserved. - * Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. - * $COPYRIGHT$ - * - * Additional copyrights may follow - * - * $HEADER$ - */ - -#include "opal_config.h" - -#include -#include -#include -#include -#include -#ifdef HAVE_STROPTS_H -#include -#endif -#include -#include -#include - -#include - -#include "opal_stdint.h" -#include "opal/constants.h" -#include "opal/util/output.h" -#include "opal/util/show_help.h" -#include "opal/mca/memory/memory.h" -#include "opal/memoryhooks/memory.h" -#include "opal/memoryhooks/memory_internal.h" - -#include "opal/mca/memory/linux/memory_linux.h" -#include "opal/mca/memory/linux/public.h" - -#define DEV_UMMUNOTIFY "/dev/ummunotify" - - -/* - * Local functions - */ -static int ummunotify_process(void); -static int ummunotify_register(void *start, size_t len, uint64_t cookie); -static int ummunotify_deregister(void *start, size_t len, uint64_t cookie); - - -/* - * Local variables - */ -static bool initialized = false; - - -/* - * Global variables (these need to be global variables rather than in - * the component struct because they are accessed in the - * opal_memory_changed() macro defined in public.h, and we don't want - * to have to include the component structure definition in public.h). - */ -uint64_t opal_memory_linux_ummunotify_counter_last_value = 0; -volatile uint64_t *opal_memory_linux_ummunotify_counter = - &opal_memory_linux_ummunotify_counter_last_value; - - -int opal_memory_linux_ummunotify_open(void) -{ - uint64_t *p; - - /* Just to be safe... */ - opal_memory_linux_ummunotify_counter_last_value = 0; - opal_memory_linux_ummunotify_counter = - &opal_memory_linux_ummunotify_counter_last_value; - - /* Open the device. Try to give a meaningful error message if - we're unable to open it. */ - mca_memory_linux_component.ummunotify_fd = - open(DEV_UMMUNOTIFY, O_RDONLY | O_NONBLOCK); - if (mca_memory_linux_component.ummunotify_fd < 0) { - char hostname[HOST_NAME_MAX]; - gethostname(hostname, sizeof(hostname)); - - if (EACCES == errno) { - /* This will get a proper show_help when merged into the - linux component */ - opal_show_help("help-opal-memory-linux.txt", - "ummunotify eaccess", true, - hostname, DEV_UMMUNOTIFY); - } else if (ENOENT != errno) { - /* Don't print an error if DEV_UMMUNOTIFY simply doesn't exist */ - opal_show_help("help-opal-memory-linux.txt", - "ummunotify open error", true, - hostname, DEV_UMMUNOTIFY, - strerror(errno), errno); - } - return OPAL_ERR_NOT_SUPPORTED; - } - - p = mmap(NULL, sizeof(*opal_memory_linux_ummunotify_counter), - PROT_READ, MAP_SHARED, - mca_memory_linux_component.ummunotify_fd, 0); - if (MAP_FAILED == opal_memory_linux_ummunotify_counter) { - close(mca_memory_linux_component.ummunotify_fd); - mca_memory_linux_component.ummunotify_fd = -1; - return OPAL_ERR_NOT_SUPPORTED; - } - opal_memory_linux_ummunotify_counter = p; - - /* If everything went well, tell OMPI that we have full support - for the memory hooks and fill in the component function - pointers */ - opal_mem_hooks_set_support(OPAL_MEMORY_FREE_SUPPORT | - OPAL_MEMORY_CHUNK_SUPPORT | - OPAL_MEMORY_MUNMAP_SUPPORT); - mca_memory_linux_component.super.memoryc_process = ummunotify_process; - mca_memory_linux_component.super.memoryc_register = ummunotify_register; - mca_memory_linux_component.super.memoryc_deregister = ummunotify_deregister; - initialized = true; - - return OPAL_SUCCESS; -} - - -/* - * Called during opal_finalize (usually during MPI_FINALIZE) to tear - * down anything that can/should be torn down to disable this - * component. The application may continue for a while after - * MPI_FINALIZE, so we should do as much as possible to disable - * anything we enabled during ummunotify_open(). - */ -int opal_memory_linux_ummunotify_close(void) -{ - if (initialized && mca_memory_linux_component.ummunotify_fd >= 0) { - munmap((void*) opal_memory_linux_ummunotify_counter, - sizeof(*opal_memory_linux_ummunotify_counter)); - close(mca_memory_linux_component.ummunotify_fd); - mca_memory_linux_component.ummunotify_fd = -1; - opal_memory_linux_ummunotify_counter = - &opal_memory_linux_ummunotify_counter_last_value; - initialized = false; - } - - return OPAL_SUCCESS; -} - -/* - * Called when opal_memory_changed() returns 1 - */ -static int ummunotify_process(void) -{ - int n; - unsigned int i; - struct ummunotify_event events[128]; - - /* Loop reading from the ummunot fd until there's nothing left to - read. If we get a LAST event, re-record the counter. */ - while (initialized) { - n = read(mca_memory_linux_component.ummunotify_fd, - &events, sizeof(events)); - if (n <= 0) { - return (EAGAIN == errno) ? OPAL_SUCCESS : OPAL_ERR_IN_ERRNO; - } - - for (i = 0; i < n / sizeof(events[0]); ++i) { - switch (events[i].type) { - case UMMUNOTIFY_EVENT_TYPE_INVAL: - /* 0 => this callback did not come from malloc */ - OPAL_OUTPUT((-1, "ummunot: invalidate start %p, end %p", - (void*) events[i].hint_start, - (void*) events[i].hint_end)); - opal_mem_hooks_release_hook((void *) (uintptr_t) events[i].hint_start, - events[i].hint_end - events[i].hint_start, - 0); - break; - - case UMMUNOTIFY_EVENT_TYPE_LAST: - opal_memory_linux_ummunotify_counter_last_value = - events[i].user_cookie_counter; - /* Are there more events to read? */ - if (opal_memory_linux_ummunotify_counter_last_value == - *opal_memory_linux_ummunotify_counter) { - OPAL_OUTPUT((-1, "ummunot: LAST; done")); - return OPAL_SUCCESS; - } - OPAL_OUTPUT((-1, "ummunot: LAST; but looping around")); - break; - } - } - } - - /* Will only get here if this component has not been - initialized */ - return OPAL_SUCCESS; -} - -static int ummunotify_register(void *start, size_t len, uint64_t cookie) -{ - struct ummunotify_register_ioctl r; - r.reserved = 0; - r.start = (unsigned long) start; - r.end = (unsigned long) start + len; - r.user_cookie = cookie; - - OPAL_OUTPUT((-1, "ummunot: register %p - %p", - start, ((char*) start) + len)); - if (initialized && ioctl(mca_memory_linux_component.ummunotify_fd, - UMMUNOTIFY_REGISTER_REGION, &r)) { - OPAL_OUTPUT((-1, "Error in ioctl register!")); - return OPAL_ERR_IN_ERRNO; - } - - return OPAL_SUCCESS; -} - -static int ummunotify_deregister(void *start, size_t len, uint64_t cookie) -{ - OPAL_OUTPUT((-1, "ummunot: deregister %p - %p", - start, ((char*) start) + len)); - if (initialized && ioctl(mca_memory_linux_component.ummunotify_fd, - UMMUNOTIFY_UNREGISTER_REGION, &cookie)) { - OPAL_OUTPUT((-1, "Error in ioctl unregister!")); - return OPAL_ERR_IN_ERRNO; - } - - return OPAL_SUCCESS; -} diff --git a/opal/mca/memory/linux/owner.txt b/opal/mca/memory/linux/owner.txt deleted file mode 100644 index 2662d53d1d7..00000000000 --- a/opal/mca/memory/linux/owner.txt +++ /dev/null @@ -1,7 +0,0 @@ -# -# owner/status file -# owner: institution that is responsible for this package -# status: e.g. active, maintenance, unmaintained -# -owner: MELLANOX,CISCO -status: maintenance diff --git a/opal/mca/memory/linux/public.h b/opal/mca/memory/linux/public.h deleted file mode 100644 index cc30e8c42a3..00000000000 --- a/opal/mca/memory/linux/public.h +++ /dev/null @@ -1,24 +0,0 @@ -/* - * Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved. - * $COPYRIGHT$ - * - * Additional copyrights may follow - * - * $HEADER$ - */ - -#ifndef OPAL_MEMORY_LINUX_PUBLIC_H -#define OPAL_MEMORY_LINUX_PUBLIC_H - -#include "opal_config.h" - -#include - -OPAL_DECLSPEC extern volatile uint64_t *opal_memory_linux_ummunotify_counter; -OPAL_DECLSPEC extern uint64_t opal_memory_linux_ummunotify_counter_last_value; - -#define opal_memory_changed() \ - (opal_memory_linux_ummunotify_counter_last_value != \ - *opal_memory_linux_ummunotify_counter) - -#endif diff --git a/opal/mca/memory/linux/rename.h b/opal/mca/memory/linux/rename.h deleted file mode 100644 index 03bd5ce3224..00000000000 --- a/opal/mca/memory/linux/rename.h +++ /dev/null @@ -1,52 +0,0 @@ -/* - * Copyright (c) 2009 Cisco Systems, Inc. All rights reserved. - * $COPYRIGHT$ - * - * Additional copyrights may follow - * - * $HEADER$ - */ - -/* Name-shift all the internal ptmalloc22 symbols to guarantee to not - conflict / confuse / override the internal glibc symbols. */ - -#define __default_morecore opal_memory_ptmalloc2_default_morecore - -#define _int_malloc opal_memory_ptmalloc2_int_malloc -#define _int_free opal_memory_ptmalloc2_int_free -#define _int_realloc opal_memory_ptmalloc2_int_realloc -#define _int_memalign opal_memory_ptmalloc2_int_memalign -#define _int_valloc opal_memory_ptmalloc2_int_valloc -#define _int_pvalloc opal_memory_ptmalloc2_int_pvalloc -#define _int_icalloc opal_memory_ptmalloc2_int_icalloc -#define _int_icomalloc opal_memory_ptmalloc2_int_icomalloc - -#define mTRIm opal_memory_ptmalloc2_mTRIm -#define mUSABLe opal_memory_ptmalloc2_mUSABLe -#define mALLOPt opal_memory_ptmalloc2_mALLOPt - -#define mem2mem_check opal_memory_ptmalloc2_mem2mem_check -#define top_check opal_memory_ptmalloc2_top_check -#define munmap_chunk opal_memory_ptmalloc2_munmap_chunk -#define mremap_chunk opal_memory_ptmalloc2_mremap_chunk - -#define malloc_check opal_memory_ptmalloc2_malloc_check -#define free_check opal_memory_ptmalloc2_free_check -#define realloc_check opal_memory_ptmalloc2_realloc_check -#define memalign_check opal_memory_ptmalloc2_memalign_check - -#define malloc_starter opal_memory_ptmalloc2_malloc_starter -#define memalign_starter opal_memory_ptmalloc2_memalign_starter -#define free_starter opal_memory_ptmalloc2_free_starter - -#define malloc_atfork opal_memory_ptmalloc2_malloc_atfork -#define free_atfork opal_memory_ptmalloc2_free_atfork - -#define _int_get_arena opal_memory_ptmalloc2_int_get_arena -#define _int_get_arena_info opal_memory_ptmalloc2_int_get_arena_info -#define _int_get_global_info opal_memory_ptmalloc2_int_get_global_info -#define _int_new_arena opal_memory_ptmalloc2_int_new_arena -#define __malloc_check_init opal_memory_ptmalloc2_malloc_check_init -#define malloc_stats opal_memory_ptmalloc2_malloc_stats - -#define posix_memalign opal_memory_ptmalloc2_posix_memalign diff --git a/opal/mca/memory/linux/sysdeps/generic/atomic.h b/opal/mca/memory/linux/sysdeps/generic/atomic.h deleted file mode 100644 index a3aeed13e89..00000000000 --- a/opal/mca/memory/linux/sysdeps/generic/atomic.h +++ /dev/null @@ -1 +0,0 @@ -/* Empty placeholder */ diff --git a/opal/mca/memory/linux/sysdeps/generic/malloc-machine.h b/opal/mca/memory/linux/sysdeps/generic/malloc-machine.h deleted file mode 100644 index 345137060d0..00000000000 --- a/opal/mca/memory/linux/sysdeps/generic/malloc-machine.h +++ /dev/null @@ -1,68 +0,0 @@ -/* Basic platform-independent macro definitions for mutexes, - thread-specific data and parameters for malloc. - Copyright (C) 2003 Free Software Foundation, Inc. - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library; if not, write to the Free - Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA - 02111-1307 USA. */ - -#ifndef _GENERIC_MALLOC_MACHINE_H -#define _GENERIC_MALLOC_MACHINE_H - -#include - -#ifndef mutex_init /* No threads, provide dummy macros */ - -# define NO_THREADS - -/* The mutex functions used to do absolutely nothing, i.e. lock, - trylock and unlock would always just return 0. However, even - without any concurrently active threads, a mutex can be used - legitimately as an `in use' flag. To make the code that is - protected by a mutex async-signal safe, these macros would have to - be based on atomic test-and-set operations, for example. */ -typedef int mutex_t; - -# define mutex_init(m) (*(m) = 0) -# define mutex_lock(m) ((*(m) = 1), 0) -# define mutex_trylock(m) (*(m) ? 1 : ((*(m) = 1), 0)) -# define mutex_unlock(m) (*(m) = 0) - -typedef void *tsd_key_t; -# define tsd_key_create(key, destr) do {} while(0) -# define tsd_setspecific(key, data) ((key) = (data)) -# define tsd_getspecific(key, vptr) (vptr = (key)) - -# define thread_atfork(prepare, parent, child) do {} while(0) - -#endif /* !defined mutex_init */ - -#ifndef atomic_full_barrier -# define atomic_full_barrier() __asm ("" ::: "memory") -#endif - -#ifndef atomic_read_barrier -# define atomic_read_barrier() atomic_full_barrier () -#endif - -#ifndef atomic_write_barrier -# define atomic_write_barrier() atomic_full_barrier () -#endif - -#ifndef DEFAULT_TOP_PAD -# define DEFAULT_TOP_PAD 131072 -#endif - -#endif /* !defined(_GENERIC_MALLOC_MACHINE_H) */ diff --git a/opal/mca/memory/linux/sysdeps/generic/thread-st.h b/opal/mca/memory/linux/sysdeps/generic/thread-st.h deleted file mode 100644 index 0243774b7cf..00000000000 --- a/opal/mca/memory/linux/sysdeps/generic/thread-st.h +++ /dev/null @@ -1,48 +0,0 @@ -/* - * $Id:$ - * Generic version: no threads. - * by Wolfram Gloger 2004 - */ - -#include - -struct thread_st { - char *sp; /* stack pointer, can be 0 */ - void (*func)(struct thread_st* st); /* must be set by user */ - int id; - int flags; - struct user_data u; -}; - -static void -thread_init(void) -{ - printf("No threads.\n"); -} - -/* Create a thread. */ -static int -thread_create(struct thread_st *st) -{ - st->flags = 0; - st->id = 1; - st->func(st); - return 0; -} - -/* Wait for one of several subthreads to finish. */ -static void -wait_for_thread(struct thread_st st[], int n_thr, - int (*end_thr)(struct thread_st*)) -{ - int i; - for(i=0; i. - -Permission to use, copy, modify, distribute, and sell this software -and its documentation for any purpose is hereby granted without fee, -provided that (i) the above copyright notices and this permission -notice appear in all copies of the software and related documentation, -and (ii) the name of Wolfram Gloger may not be used in any advertising -or publicity relating to the software. - -THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF ANY KIND, -EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY -WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. - -IN NO EVENT SHALL WOLFRAM GLOGER BE LIABLE FOR ANY SPECIAL, -INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND, OR ANY -DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, -WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY -OF LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR -PERFORMANCE OF THIS SOFTWARE. -*/ - -#ifndef _PTHREAD_MALLOC_MACHINE_H -#define _PTHREAD_MALLOC_MACHINE_H - -#include - -#undef thread_atfork_static - -/* Use fast inline spinlocks with gcc. */ -#if (defined __i386__ || defined __x86_64__) && defined __GNUC__ && \ - !defined USE_NO_SPINLOCKS - -#include -#include - -typedef struct { - volatile unsigned int lock; - int pad0_; -} mutex_t; - -#define MUTEX_INITIALIZER { 0 } -#define mutex_init(m) ((m)->lock = 0) -static inline int mutex_lock(mutex_t *m) { - int cnt = 0, r; - struct timespec tm; - - for(;;) { - __asm__ __volatile__ - ("xchgl %0, %1" - : "=r"(r), "=m"(m->lock) - : "0"(1), "m"(m->lock) - : "memory"); - if(!r) - return 0; - if(cnt < 50) { - sched_yield(); - cnt++; - } else { - tm.tv_sec = 0; - tm.tv_nsec = 2000001; - nanosleep(&tm, NULL); - cnt = 0; - } - } -} -static inline int mutex_trylock(mutex_t *m) { - int r; - - __asm__ __volatile__ - ("xchgl %0, %1" - : "=r"(r), "=m"(m->lock) - : "0"(1), "m"(m->lock) - : "memory"); - return r; -} -static inline int mutex_unlock(mutex_t *m) { - m->lock = 0; - __asm __volatile__ ("" : : : "memory"); - return 0; -} - -#else - -/* Normal pthread mutex. */ -typedef pthread_mutex_t mutex_t; - -#define MUTEX_INITIALIZER PTHREAD_MUTEX_INITIALIZER -#define mutex_init(m) pthread_mutex_init(m, NULL) -#define mutex_lock(m) pthread_mutex_lock(m) -#define mutex_trylock(m) pthread_mutex_trylock(m) -#define mutex_unlock(m) pthread_mutex_unlock(m) - -#endif /* (__i386__ || __x86_64__) && __GNUC__ && !USE_NO_SPINLOCKS */ - -/* thread specific data */ -#if defined(__sgi) || defined(USE_TSD_DATA_HACK) - -/* Hack for thread-specific data, e.g. on Irix 6.x. We can't use - pthread_setspecific because that function calls malloc() itself. - The hack only works when pthread_t can be converted to an integral - type. */ - -typedef void *tsd_key_t[256]; -#define tsd_key_create(key, destr) do { \ - int i; \ - for(i=0; i<256; i++) (*key)[i] = 0; \ -} while(0) -#define tsd_setspecific(key, data) \ - (key[(unsigned)pthread_self() % 256] = (data)) -#define tsd_getspecific(key, vptr) \ - (vptr = key[(unsigned)pthread_self() % 256]) - -#else - -typedef pthread_key_t tsd_key_t; - -#define tsd_key_create(key, destr) pthread_key_create(key, destr) -#define tsd_setspecific(key, data) pthread_setspecific(key, data) -#define tsd_getspecific(key, vptr) (vptr = pthread_getspecific(key)) - -#endif - -/* at fork */ -#define thread_atfork(prepare, parent, child) \ - pthread_atfork(prepare, parent, child) - -#include - -#endif /* !defined(_MALLOC_MACHINE_H) */ diff --git a/opal/mca/memory/linux/sysdeps/pthread/thread-st.h b/opal/mca/memory/linux/sysdeps/pthread/thread-st.h deleted file mode 100644 index f97a0a35527..00000000000 --- a/opal/mca/memory/linux/sysdeps/pthread/thread-st.h +++ /dev/null @@ -1,111 +0,0 @@ -/* - * $Id: thread-st.h$ - * pthread version - * by Wolfram Gloger 2004 - */ - -#include -#include - -pthread_cond_t finish_cond = PTHREAD_COND_INITIALIZER; -pthread_mutex_t finish_mutex = PTHREAD_MUTEX_INITIALIZER; - -#ifndef USE_PTHREADS_STACKS -#define USE_PTHREADS_STACKS 0 -#endif - -#ifndef STACKSIZE -#define STACKSIZE 32768 -#endif - -struct thread_st { - char *sp; /* stack pointer, can be 0 */ - void (*func)(struct thread_st* st); /* must be set by user */ - pthread_t id; - int flags; - struct user_data u; -}; - -static void -thread_init(void) -{ - printf("Using posix threads.\n"); - pthread_cond_init(&finish_cond, NULL); - pthread_mutex_init(&finish_mutex, NULL); -} - -static void * -thread_wrapper(void *ptr) -{ - struct thread_st *st = (struct thread_st*)ptr; - - /*printf("begin %p\n", st->sp);*/ - st->func(st); - pthread_mutex_lock(&finish_mutex); - st->flags = 1; - pthread_mutex_unlock(&finish_mutex); - pthread_cond_signal(&finish_cond); - /*printf("end %p\n", st->sp);*/ - return NULL; -} - -/* Create a thread. */ -static int -thread_create(struct thread_st *st) -{ - st->flags = 0; - { - pthread_attr_t* attr_p = 0; -#if USE_PTHREADS_STACKS - pthread_attr_t attr; - - pthread_attr_init (&attr); - if(!st->sp) - st->sp = malloc(STACKSIZE+16); - if(!st->sp) - return -1; - if(pthread_attr_setstacksize(&attr, STACKSIZE)) - fprintf(stderr, "error setting stacksize"); - else - pthread_attr_setstackaddr(&attr, st->sp + STACKSIZE); - /*printf("create %p\n", st->sp);*/ - attr_p = &attr; -#endif - return pthread_create(&st->id, attr_p, thread_wrapper, st); - } - return 0; -} - -/* Wait for one of several subthreads to finish. */ -static void -wait_for_thread(struct thread_st st[], int n_thr, - int (*end_thr)(struct thread_st*)) -{ - int i; - - pthread_mutex_lock(&finish_mutex); - for(;;) { - int term = 0; - for(i=0; i 0) - break; - pthread_cond_wait(&finish_cond, &finish_mutex); - } - pthread_mutex_unlock(&finish_mutex); -} - -/* - * Local variables: - * tab-width: 4 - * End: - */ diff --git a/opal/mca/memory/linux/sysdeps/solaris/malloc-machine.h b/opal/mca/memory/linux/sysdeps/solaris/malloc-machine.h deleted file mode 100644 index 00e33b08014..00000000000 --- a/opal/mca/memory/linux/sysdeps/solaris/malloc-machine.h +++ /dev/null @@ -1,51 +0,0 @@ -/* Basic platform-independent macro definitions for mutexes, - thread-specific data and parameters for malloc. - Solaris threads version. - Copyright (C) 2004 Wolfram Gloger . - -Permission to use, copy, modify, distribute, and sell this software -and its documentation for any purpose is hereby granted without fee, -provided that (i) the above copyright notices and this permission -notice appear in all copies of the software and related documentation, -and (ii) the name of Wolfram Gloger may not be used in any advertising -or publicity relating to the software. - -THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF ANY KIND, -EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY -WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. - -IN NO EVENT SHALL WOLFRAM GLOGER BE LIABLE FOR ANY SPECIAL, -INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND, OR ANY -DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, -WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY -OF LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR -PERFORMANCE OF THIS SOFTWARE. -*/ - -#ifndef _SOLARIS_MALLOC_MACHINE_H -#define _SOLARIS_MALLOC_MACHINE_H - -#include - -typedef thread_t thread_id; - -#define MUTEX_INITIALIZER { 0 } -#define mutex_init(m) mutex_init(m, USYNC_THREAD, NULL) - -/* - * Hack for thread-specific data on Solaris. We can't use thr_setspecific - * because that function calls malloc() itself. - */ -typedef void *tsd_key_t[256]; -#define tsd_key_create(key, destr) do { \ - int i; \ - for(i=0; i<256; i++) (*key)[i] = 0; \ -} while(0) -#define tsd_setspecific(key, data) (key[(unsigned)thr_self() % 256] = (data)) -#define tsd_getspecific(key, vptr) (vptr = key[(unsigned)thr_self() % 256]) - -#define thread_atfork(prepare, parent, child) do {} while(0) - -#include - -#endif /* !defined(_SOLARIS_MALLOC_MACHINE_H) */ diff --git a/opal/mca/memory/linux/sysdeps/solaris/thread-st.h b/opal/mca/memory/linux/sysdeps/solaris/thread-st.h deleted file mode 100644 index dbb4b097e8b..00000000000 --- a/opal/mca/memory/linux/sysdeps/solaris/thread-st.h +++ /dev/null @@ -1,72 +0,0 @@ -/* - * $Id:$ - * Solaris version - * by Wolfram Gloger 2004 - */ - -#include -#include - -#ifndef STACKSIZE -#define STACKSIZE 32768 -#endif - -struct thread_st { - char *sp; /* stack pointer, can be 0 */ - void (*func)(struct thread_st* st); /* must be set by user */ - thread_id id; - int flags; - struct user_data u; -}; - -static void -thread_init(void) -{ - printf("Using Solaris threads.\n"); -} - -static void * -thread_wrapper(void *ptr) -{ - struct thread_st *st = (struct thread_st*)ptr; - - /*printf("begin %p\n", st->sp);*/ - st->func(st); - /*printf("end %p\n", st->sp);*/ - return NULL; -} - -/* Create a thread. */ -static int -thread_create(struct thread_st *st) -{ - st->flags = 0; - if(!st->sp) - st->sp = malloc(STACKSIZE); - if(!st->sp) return -1; - thr_create(st->sp, STACKSIZE, thread_wrapper, st, THR_NEW_LWP, &st->id); - return 0; -} - -/* Wait for one of several subthreads to finish. */ -static void -wait_for_thread(struct thread_st st[], int n_thr, - int (*end_thr)(struct thread_st*)) -{ - int i; - thread_t id; - - thr_join(0, &id, NULL); - for(i=0; i. - -Permission to use, copy, modify, distribute, and sell this software -and its documentation for any purpose is hereby granted without fee, -provided that (i) the above copyright notices and this permission -notice appear in all copies of the software and related documentation, -and (ii) the name of Wolfram Gloger may not be used in any advertising -or publicity relating to the software. - -THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF ANY KIND, -EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY -WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. - -IN NO EVENT SHALL WOLFRAM GLOGER BE LIABLE FOR ANY SPECIAL, -INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND, OR ANY -DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, -WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY -OF LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR -PERFORMANCE OF THIS SOFTWARE. -*/ - -#ifndef _SPROC_MALLOC_MACHINE_H -#define _SPROC_MALLOC_MACHINE_H - -#include -#include -#include -#include - -typedef abilock_t mutex_t; - -#define MUTEX_INITIALIZER { 0 } -#define mutex_init(m) init_lock(m) -#define mutex_lock(m) (spin_lock(m), 0) -#define mutex_trylock(m) acquire_lock(m) -#define mutex_unlock(m) release_lock(m) - -typedef int tsd_key_t; -int tsd_key_next; -#define tsd_key_create(key, destr) ((*key) = tsd_key_next++) -#define tsd_setspecific(key, data) (((void **)(&PRDA->usr_prda))[key] = data) -#define tsd_getspecific(key, vptr) (vptr = ((void **)(&PRDA->usr_prda))[key]) - -#define thread_atfork(prepare, parent, child) do {} while(0) - -#include - -#endif /* !defined(_SPROC_MALLOC_MACHINE_H) */ diff --git a/opal/mca/memory/linux/sysdeps/sproc/thread-st.h b/opal/mca/memory/linux/sysdeps/sproc/thread-st.h deleted file mode 100644 index a512f921306..00000000000 --- a/opal/mca/memory/linux/sysdeps/sproc/thread-st.h +++ /dev/null @@ -1,84 +0,0 @@ -/* - * $Id:$ - * sproc version - * by Wolfram Gloger 2001, 2004 - */ - -#include -#include -#include - -#ifndef STACKSIZE -#define STACKSIZE 32768 -#endif - -struct thread_st { - char *sp; /* stack pointer, can be 0 */ - void (*func)(struct thread_st* st); /* must be set by user */ - thread_id id; - int flags; - struct user_data u; -}; - -static void -thread_init(void) -{ - printf("Using sproc() threads.\n"); -} - -static void -thread_wrapper(void *ptr, size_t stack_len) -{ - struct thread_st *st = (struct thread_st*)ptr; - - /*printf("begin %p\n", st->sp);*/ - st->func(st); - /*printf("end %p\n", st->sp);*/ -} - -/* Create a thread. */ -static int -thread_create(struct thread_st *st) -{ - st->flags = 0; - if(!st->sp) - st->sp = malloc(STACKSIZE); - if(!st->sp) return -1; - st->id = sprocsp(thread_wrapper, PR_SALL, st, st->sp+STACKSIZE, STACKSIZE); - if(st->id < 0) { - return -1; - } - return 0; -} - -/* Wait for one of several subthreads to finish. */ -static void -wait_for_thread(struct thread_st st[], int n_thr, - int (*end_thr)(struct thread_st*)) -{ - int i; - int id; - - int status = 0; - id = wait(&status); - if(status != 0) { - if(WIFSIGNALED(status)) - printf("thread %id terminated by signal %d\n", - id, WTERMSIG(status)); - else - printf("thread %id exited with status %d\n", - id, WEXITSTATUS(status)); - } - for(i=0; i 0 - -static void -mem_init(unsigned char *ptr, unsigned long size) -{ - unsigned long i, j; - - if(size == 0) return; - for(i=0; i>8)) & 0xFF); - } - j = (unsigned long)ptr ^ (size-1); - ptr[size-1] = ((j ^ (j>>8)) & 0xFF); -} - -static int -mem_check(unsigned char *ptr, unsigned long size) -{ - unsigned long i, j; - - if(size == 0) return 0; - for(i=0; i>8)) & 0xFF)) return 1; - } - j = (unsigned long)ptr ^ (size-1); - if(ptr[size-1] != ((j ^ (j>>8)) & 0xFF)) return 2; - return 0; -} - -static int -zero_check(unsigned* ptr, unsigned long size) -{ - unsigned char* ptr2; - - while(size >= sizeof(*ptr)) { - if(*ptr++ != 0) - return -1; - size -= sizeof(*ptr); - } - ptr2 = (unsigned char*)ptr; - while(size > 0) { - if(*ptr2++ != 0) - return -1; - --size; - } - return 0; -} - -#endif /* TEST > 0 */ - -/* Allocate a bin with malloc(), realloc() or memalign(). r must be a - random number >= 1024. */ - -static void -bin_alloc(struct bin *m, unsigned long size, int r) -{ -#if TEST > 0 - if(mem_check(m->ptr, m->size)) { - printf("memory corrupt!\n"); - exit(1); - } -#endif - r %= 1024; - /*printf("%d ", r);*/ - if(r < 4) { /* memalign */ - if(m->size > 0) free(m->ptr); - m->ptr = (unsigned char *)memalign(sizeof(int) << r, size); - } else if(r < 20) { /* calloc */ - if(m->size > 0) free(m->ptr); - m->ptr = (unsigned char *)calloc(size, 1); -#if TEST > 0 - if(zero_check((unsigned*)m->ptr, size)) { - long i; - for(i=0; iptr[i] != 0) - break; - printf("calloc'ed memory non-zero (ptr=%p, i=%ld)!\n", m->ptr, i); - exit(1); - } -#endif - } else if(r < 100 && m->size < REALLOC_MAX) { /* realloc */ - if(m->size == 0) m->ptr = NULL; - m->ptr = realloc(m->ptr, size); - } else { /* plain malloc */ - if(m->size > 0) free(m->ptr); - m->ptr = (unsigned char *)malloc(size); - } - if(!m->ptr) { - printf("out of memory (r=%d, size=%ld)!\n", r, (long)size); - exit(1); - } - m->size = size; -#if TEST > 0 - mem_init(m->ptr, m->size); -#endif -} - -/* Free a bin. */ - -static void -bin_free(struct bin *m) -{ - if(m->size == 0) return; -#if TEST > 0 - if(mem_check(m->ptr, m->size)) { - printf("memory corrupt!\n"); - exit(1); - } -#endif - free(m->ptr); - m->size = 0; -} - -/* - * Local variables: - * tab-width: 4 - * End: - */ diff --git a/opal/mca/memory/linux/t-test1.c b/opal/mca/memory/linux/t-test1.c deleted file mode 100644 index 15dc7c6aaea..00000000000 --- a/opal/mca/memory/linux/t-test1.c +++ /dev/null @@ -1,285 +0,0 @@ -/* - * $Id: t-test1.c,v 1.2 2004/11/04 14:58:45 wg Exp $ - * by Wolfram Gloger 1996-1999, 2001, 2004 - * A multi-thread test for malloc performance, maintaining one pool of - * allocated bins per thread. - */ - -#if (defined __STDC__ && __STDC__) || defined __cplusplus -# include -#endif -#include -#include -#include -#include -#include -#include -#include - -#if !USE_MALLOC -#include -#else -#include "malloc.h" -#endif - -#include "lran2.h" -#include "t-test.h" - -struct user_data { - int bins, max; - unsigned long size; - long seed; -}; -#include "thread-st.h" - -#define N_TOTAL 10 -#ifndef N_THREADS -#define N_THREADS 2 -#endif -#ifndef N_TOTAL_PRINT -#define N_TOTAL_PRINT 50 -#endif -#ifndef MEMORY -#define MEMORY 8000000l -#endif -#define SIZE 10000 -#define I_MAX 10000 -#define ACTIONS_MAX 30 -#ifndef TEST_FORK -#define TEST_FORK 0 -#endif - -#define RANDOM(d,s) (lran2(d) % (s)) - -struct bin_info { - struct bin *m; - unsigned long size, bins; -}; - -#if TEST > 0 - -void -bin_test(struct bin_info *p) -{ - int b; - - for(b=0; bbins; b++) { - if(mem_check(p->m[b].ptr, p->m[b].size)) { - printf("memory corrupt!\n"); - abort(); - } - } -} - -#endif - -void -malloc_test(struct thread_st *st) -{ - int b, i, j, actions, pid = 1; - struct bin_info p; - struct lran2_st ld; /* data for random number generator */ - - lran2_init(&ld, st->u.seed); -#if TEST_FORK>0 - if(RANDOM(&ld, TEST_FORK) == 0) { - int status; - -#if !USE_THR - pid = fork(); -#else - pid = fork1(); -#endif - if(pid > 0) { - /*printf("forked, waiting for %d...\n", pid);*/ - waitpid(pid, &status, 0); - printf("done with %d...\n", pid); - if(!WIFEXITED(status)) { - printf("child term with signal %d\n", WTERMSIG(status)); - exit(1); - } - return; - } - exit(0); - } -#endif - p.m = (struct bin *)malloc(st->u.bins*sizeof(*p.m)); - p.bins = st->u.bins; - p.size = st->u.size; - for(b=0; bu.max;) { -#if TEST > 1 - bin_test(&p); -#endif - actions = RANDOM(&ld, ACTIONS_MAX); -#if USE_MALLOC && MALLOC_DEBUG - if(actions < 2) { mallinfo(); } -#endif - for(j=0; j 2 - bin_test(&p); -#endif - } -#if 0 /* Test illegal free()s while setting MALLOC_CHECK_ */ - for(j=0; j<8; j++) { - b = RANDOM(&ld, p.bins); - if(p.m[b].ptr) { - int offset = (RANDOM(&ld, 11) - 5)*8; - char *rogue = (char*)(p.m[b].ptr) + offset; - /*printf("p=%p rogue=%p\n", p.m[b].ptr, rogue);*/ - free(rogue); - } - } -#endif - i += actions; - } - for(b=0; bid); -#endif - if(n_total >= n_total_max) { - n_running--; - } else if(st->u.seed++, thread_create(st)) { - printf("Creating thread #%d failed.\n", n_total); - } else { - n_total++; - if(n_total%N_TOTAL_PRINT == 0) - printf("n_total = %d\n", n_total); - } - return 0; -} - -#if 0 -/* Protect address space for allocation of n threads by LinuxThreads. */ -static void -protect_stack(int n) -{ - char buf[2048*1024]; - char* guard; - size_t guard_size = 2*2048*1024UL*(n+2); - - buf[0] = '\0'; - guard = (char*)(((unsigned long)buf - 4096)& ~4095UL) - guard_size; - printf("Setting up stack guard at %p\n", guard); - if(mmap(guard, guard_size, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, - -1, 0) - != guard) - printf("failed!\n"); -} -#endif - -int -main(int argc, char *argv[]) -{ - int i, bins; - int n_thr=N_THREADS; - int i_max=I_MAX; - unsigned long size=SIZE; - struct thread_st *st; - -#if USE_MALLOC && USE_STARTER==2 - ptmalloc_init(); - printf("ptmalloc_init\n"); -#endif - - if(argc > 1) n_total_max = atoi(argv[1]); - if(n_total_max < 1) n_thr = 1; - if(argc > 2) n_thr = atoi(argv[2]); - if(n_thr < 1) n_thr = 1; - if(n_thr > 100) n_thr = 100; - if(argc > 3) i_max = atoi(argv[3]); - - if(argc > 4) size = atol(argv[4]); - if(size < 2) size = 2; - - bins = MEMORY/(size*n_thr); - if(argc > 5) bins = atoi(argv[5]); - if(bins < 4) bins = 4; - - /*protect_stack(n_thr);*/ - - thread_init(); - printf("total=%d threads=%d i_max=%d size=%ld bins=%d\n", - n_total_max, n_thr, i_max, size, bins); - - st = (struct thread_st *)malloc(n_thr*sizeof(*st)); - if(!st) exit(-1); - -#if !defined NO_THREADS && (defined __sun__ || defined sun) - /* I know of no other way to achieve proper concurrency with Solaris. */ - thr_setconcurrency(n_thr); -#endif - - /* Start all n_thr threads. */ - for(i=0; i0;) { - wait_for_thread(st, n_thr, my_end_thread); - } - for(i=0; i -#endif -#include -#include -#include -#include -#include -#include - -#if !USE_MALLOC -#include -#else -#include "malloc.h" -#endif - -#include "lran2.h" -#include "t-test.h" - -struct user_data { - int max; - unsigned long size; - long seed; -}; -#include "thread-st.h" -#include "malloc-machine.h" /* for mutex */ - -#define N_TOTAL 10 -#ifndef N_THREADS -#define N_THREADS 2 -#endif -#ifndef N_TOTAL_PRINT -#define N_TOTAL_PRINT 50 -#endif -#define STACKSIZE 32768 -#ifndef MEMORY -#define MEMORY 8000000l -#endif -#define SIZE 10000 -#define I_MAX 10000 -#define BINS_PER_BLOCK 20 - -#define RANDOM(d,s) (lran2(d) % (s)) - -struct block { - struct bin b[BINS_PER_BLOCK]; - mutex_t mutex; -} *blocks; - -int n_blocks; - -#if TEST > 0 - -void -bin_test(void) -{ - int b, i; - - for(b=0; bu.seed); - for(i=0; i<=st->u.max;) { -#if TEST > 1 - bin_test(); -#endif - bl = &blocks[RANDOM(&ld, n_blocks)]; - r = RANDOM(&ld, 1024); - if(r < 200) { /* free only */ - mutex_lock(&bl->mutex); - for(b=0; bb[b]); - mutex_unlock(&bl->mutex); - i += BINS_PER_BLOCK; - } else { /* alloc/realloc */ - /* Generate random numbers in advance. */ - for(b=0; bu.size) + 1; - rnum[b] = lran2(&ld); - } - mutex_lock(&bl->mutex); - for(b=0; bb[b], rsize[b], rnum[b]); - mutex_unlock(&bl->mutex); - i += BINS_PER_BLOCK; - } -#if TEST > 2 - bin_test(); -#endif - } -} - -int n_total=0, n_total_max=N_TOTAL, n_running; - -int -my_end_thread(struct thread_st *st) -{ - /* Thread st has finished. Start a new one. */ -#if 0 - printf("Thread %lx terminated.\n", (long)st->id); -#endif - if(n_total >= n_total_max) { - n_running--; - } else if(st->u.seed++, thread_create(st)) { - printf("Creating thread #%d failed.\n", n_total); - } else { - n_total++; - if(n_total%N_TOTAL_PRINT == 0) - printf("n_total = %d\n", n_total); - } - return 0; -} - -int -main(int argc, char *argv[]) -{ - int i, j, bins; - int n_thr=N_THREADS; - int i_max=I_MAX; - unsigned long size=SIZE; - struct thread_st *st; - -#if USE_MALLOC && USE_STARTER==2 - ptmalloc_init(); - printf("ptmalloc_init\n"); -#endif - - if(argc > 1) n_total_max = atoi(argv[1]); - if(n_total_max < 1) n_thr = 1; - if(argc > 2) n_thr = atoi(argv[2]); - if(n_thr < 1) n_thr = 1; - if(n_thr > 100) n_thr = 100; - if(argc > 3) i_max = atoi(argv[3]); - - if(argc > 4) size = atol(argv[4]); - if(size < 2) size = 2; - - bins = MEMORY/size; - if(argc > 5) bins = atoi(argv[5]); - if(bins < BINS_PER_BLOCK) bins = BINS_PER_BLOCK; - - n_blocks = bins/BINS_PER_BLOCK; - blocks = (struct block *)malloc(n_blocks*sizeof(*blocks)); - if(!blocks) - exit(1); - - thread_init(); - printf("total=%d threads=%d i_max=%d size=%ld bins=%d\n", - n_total_max, n_thr, i_max, size, n_blocks*BINS_PER_BLOCK); - - for(i=0; i0;) { - wait_for_thread(st, n_thr, my_end_thread); - } - - for(i=0; i, 2001. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library; if not, write to the Free - Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA - 02111-1307 USA. */ - -#include -#include -#include "malloc.h" - -static int errors = 0; - -static void -merror (const char *msg) -{ - ++errors; - printf ("Error: %s\n", msg); -} - -int -main (void) -{ - void *p1, *p2; - void *save_state; - long i; - - errno = 0; - - p1 = malloc (10); - if (p1 == NULL) - merror ("malloc (10) failed."); - - p2 = malloc (20); - if (p2 == NULL) - merror ("malloc (20) failed."); - - free (malloc (10)); - - for (i=0; i<100; ++i) - { - save_state = malloc_get_state (); - if (save_state == NULL) - { - merror ("malloc_get_state () failed."); - break; - } - /*free (malloc (10)); This could change the top chunk! */ - malloc_set_state (save_state); - p1 = realloc (p1, i*4 + 4); - if (p1 == NULL) - merror ("realloc (i*4) failed."); - free (save_state); - } - - p1 = realloc (p1, 40); - free (p2); - p2 = malloc (10); - if (p2 == NULL) - merror ("malloc (10) failed."); - free (p1); - - return errors != 0; -} - -/* - * Local variables: - * c-basic-offset: 2 - * End: - */ diff --git a/opal/mca/memory/linux/tst-mstats.c b/opal/mca/memory/linux/tst-mstats.c deleted file mode 100644 index 5d843128413..00000000000 --- a/opal/mca/memory/linux/tst-mstats.c +++ /dev/null @@ -1,100 +0,0 @@ -/* Copyright (C) 2004 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Wolfram Gloger , 2004. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library; if not, write to the Free - Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA - 02111-1307 USA. */ - -#include -#include -#include "malloc.h" - -static int errors = 0; - -static void -merror (const char *msg) -{ - ++errors; - printf ("Error: %s\n", msg); -} - -int -main (void) -{ - void *p1, *p2; - long i; - mstate a; - struct malloc_arena_info mai; - int nfree; - unsigned long navail; - - errno = 0; - - malloc_stats(); /* check that it works even without initialization */ - a = _int_get_arena(0); - if (!a) { - merror ("Can't get main arena."); - return 1; - } - free (malloc (10)); - _int_get_arena_info(a, &mai); - printf("nfree = %d\navail = %lu\nfastavail = %lu\ntop_size = %lu\n", - mai.nbinblocks + mai.nfastblocks, - (unsigned long)mai.binavail, - (unsigned long)mai.fastavail, - (unsigned long)mai.top_size); - if (mai.nfastblocks+mai.nbinblocks < 1) - merror ("initial _int_get_arena_info() failed."); - nfree = mai.nbinblocks + mai.nfastblocks; - navail = mai.binavail + mai.fastavail; - - p1 = malloc (10); - if (p1 == NULL) - merror ("malloc (10) failed."); - p2 = malloc (30); - if (p2 == NULL) - merror ("malloc (30) failed."); - - free (malloc (10)); - - for (i=0; i<100; ++i) - { - p1 = realloc (p1, i*7 + 3); - if (p1 == NULL) - merror ("realloc (i*7 + 3) failed."); - } - free (p2); - - _int_get_arena_info(a, &mai); - printf("nfree = %d\navail = %lu\nfastavail = %lu\ntop_size = %lu\n", - mai.nbinblocks + mai.nfastblocks, - (unsigned long)mai.binavail, - (unsigned long)mai.fastavail, - (unsigned long)mai.top_size); - /* Assume that no memory is returned to the system from these small - chunks. */ - if (mai.nbinblocks+mai.nfastblocks < nfree || - mai.binavail+mai.fastavail < navail) - merror ("final _int_get_arena_info() failed."); - malloc_stats(); - - return errors != 0; -} - -/* - * Local variables: - * c-basic-offset: 2 - * End: - */ From 80ec79cfc84b76999d98f9772bc122e007fbecaf Mon Sep 17 00:00:00 2001 From: Nathan Hjelm Date: Fri, 8 Apr 2016 21:07:28 -0600 Subject: [PATCH 7/8] memory/patcher: updates to memory hooks This commit fixes bugs that can cause crashes and memory corruption when the mremap hook is called. The problem occurs because of the ellipses (...) in the mremap intercept function. The ellipses cover the optional new_addr argument on Linux. This commit removes the ellipses and adds an explicit 5th argument. This commit also adds a hook for shmdt. The code only works on Linux at the moment as it needs to read /proc/self/maps to determine the size of the shared memory segment. Additionally, this commit removes the mmap hook. There is no apparent benefit for detecting mmap(..., PROT_NONE, ...) and it seems to cause problems when threads are in use. Signed-off-by: Nathan Hjelm --- .../memory/patcher/memory_patcher_component.c | 131 +++++++++++++++--- 1 file changed, 111 insertions(+), 20 deletions(-) diff --git a/opal/mca/memory/patcher/memory_patcher_component.c b/opal/mca/memory/patcher/memory_patcher_component.c index cacefc3232f..0b029ad15de 100644 --- a/opal/mca/memory/patcher/memory_patcher_component.c +++ b/opal/mca/memory/patcher/memory_patcher_component.c @@ -83,11 +83,6 @@ opal_memory_patcher_component_t mca_memory_patcher_component = { it out) */ }; -#if OPAL_MEMORY_PATCHER_HAVE___MMAP && !OPAL_MEMORY_PATCHER_HAVE___MMAP_PROTO -/* prototype for Apple's internal mmap function */ -void *__mmap (void *start, size_t length, int prot, int flags, int fd, off_t offset); -#endif - #if OPAL_MEMORY_PATCHER_HAVE___SYSCALL_PROTO && OPAL_MEMORY_PATCHER_HAVE___SYSCALL /* calling __syscall is preferred on some systems when some arguments may be 64-bit. it also * has the benefit of having an off_t return type */ @@ -96,6 +91,13 @@ void *__mmap (void *start, size_t length, int prot, int flags, int fd, off_t off #define memory_patcher_syscall syscall #endif +#if 0 + +#if OPAL_MEMORY_PATCHER_HAVE___MMAP && !OPAL_MEMORY_PATCHER_HAVE___MMAP_PROTO +/* prototype for Apple's internal mmap function */ +void *__mmap (void *start, size_t length, int prot, int flags, int fd, off_t offset); +#endif + static void *(*original_mmap)(void *, size_t, int, int, int, off_t); static void *intercept_mmap(void *start, size_t length, int prot, int flags, int fd, off_t offset) @@ -127,6 +129,8 @@ static void *intercept_mmap(void *start, size_t length, int prot, int flags, int return result; } +#endif + static int (*original_munmap) (void *, size_t); static int intercept_munmap(void *start, size_t length) @@ -148,27 +152,22 @@ static int intercept_munmap(void *start, size_t length) #if defined (SYS_mremap) -static void *(*original_mremap) (void *, size_t, size_t, int, ...); +/* on linux this function has an optional extra argument but ... can not be used here because it + * causes issues when intercepting a 4-argument mremap call */ +static void *(*original_mremap) (void *, size_t, size_t, int, void *); -static void *intercept_mremap (void *start, size_t oldlen, size_t newlen, int flags, ...) +static void *intercept_mremap (void *start, size_t oldlen, size_t newlen, int flags, void *new_address) { OPAL_PATCHER_BEGIN; - void *result = 0; -#ifdef MREMAP_FIXED - va_list ap; -#endif - void *new_address = NULL; - - opal_mem_hooks_release_hook (start, oldlen, false); + void *result = MAP_FAILED; -#ifdef MREMAP_FIXED - if (flags & MREMAP_FIXED) { - va_start(ap, flags); - new_address = va_arg(ap, void*); + if (MAP_FAILED != start && oldlen > 0) { + opal_mem_hooks_release_hook (start, oldlen, true); + } - va_end(ap); + if (!(flags & MREMAP_FIXED)) { + new_address = NULL; } -#endif if (!original_mremap) { result = (void *)(intptr_t) memory_patcher_syscall (SYS_mremap, start, oldlen, newlen, flags, new_address); @@ -261,6 +260,87 @@ static int intercept_brk (void *addr) #endif +#if defined(SYS_shmdt) && defined(__linux__) + +#include +#include +#include + +static size_t memory_patcher_get_shm_seg_size (const void *shmaddr) +{ + unsigned long start_addr, end_addr; + char *ptr, *newline; + char buffer[1024]; + size_t seg_size = 0; + int fd; + + seg_size = 0; + + fd = open ("/proc/self/maps", O_RDONLY); + assert (fd >= 0); + + for (size_t read_offset = 0 ; ; ) { + ssize_t nread = read(fd, buffer + read_offset, sizeof(buffer) - 1 - read_offset); + if (nread <= 0) { + if (errno == EINTR) { + continue; + } + + break; + } else { + buffer[nread + read_offset] = '\0'; + } + + ptr = buffer; + while ( (newline = strchr(ptr, '\n')) != NULL ) { + /* 00400000-0040b000 r-xp ... \n */ + int ret = sscanf(ptr, "%lx-%lx ", &start_addr, &end_addr); + if (ret != 2) { + continue; + } + + if (start_addr == (uintptr_t)shmaddr) { + seg_size = end_addr - start_addr; + goto out_close; + } + + newline = strchr(ptr, '\n'); + if (newline == NULL) { + break; + } + + ptr = newline + 1; + } + + read_offset = strlen(ptr); + memmove(buffer, ptr, read_offset); + } + + out_close: + close(fd); + return seg_size; +} + +static int (*original_shmdt) (const void *); + +static int intercept_shmdt (const void *shmaddr) +{ + OPAL_PATCHER_BEGIN; + int result; + + opal_mem_hooks_release_hook (shmaddr, memory_patcher_get_shm_seg_size (shmaddr), false); + + if (original_shmdt) { + result = original_shmdt (shmaddr); + } else { + result = memory_patcher_syscall (SYS_shmdt, shmaddr); + } + + OPAL_PATCHER_END; + return result; +} +#endif + static int patcher_register (void) { mca_memory_patcher_priority = 80; @@ -296,10 +376,14 @@ static int patcher_open (void) /* set memory hooks support level */ opal_mem_hooks_set_support (OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_MUNMAP_SUPPORT); +#if 0 + /* NTH: the only reason to hook mmap would be to detect memory protection. this does not invalidate + * any cache entries in the region. */ rc = opal_patcher->patch_symbol ("mmap", (uintptr_t) intercept_mmap, (uintptr_t *) &original_mmap); if (OPAL_SUCCESS != rc) { return rc; } +#endif rc = opal_patcher->patch_symbol ("munmap", (uintptr_t)intercept_munmap, (uintptr_t *) &original_munmap); if (OPAL_SUCCESS != rc) { @@ -318,6 +402,13 @@ static int patcher_open (void) return rc; } +#if defined(SYS_shmdt) && defined(__linux__) + rc = opal_patcher->patch_symbol ("shmdt", (uintptr_t) intercept_shmdt, (uintptr_t *) &original_shmdt); + if (OPAL_SUCCESS != rc) { + return rc; + } +#endif + #if defined (SYS_brk) rc = opal_patcher->patch_symbol ("brk", (uintptr_t)intercept_brk, (uintptr_t *) &original_brk); #endif From c2b6fbb1243f7347360ec626f261b4c290f24cb6 Mon Sep 17 00:00:00 2001 From: Nathan Hjelm Date: Fri, 8 Apr 2016 14:38:09 -0600 Subject: [PATCH 8/8] opal/memory: move initialization to first rcache creation Because of the removal of the linux memory component it is no longer necessary to initialize the memory component in opal_init(). This commit moves the initialization to the creation of the first rcache component. Signed-off-by: Nathan Hjelm --- opal/mca/btl/openib/btl_openib_component.c | 49 +++++++-------------- opal/mca/btl/ugni/btl_ugni_component.c | 22 ---------- opal/mca/rcache/base/base.h | 1 + opal/mca/rcache/base/rcache_base_create.c | 50 +++++++++++++--------- opal/mca/rcache/base/rcache_base_frame.c | 9 +++- opal/memoryhooks/memory.c | 9 +--- opal/runtime/opal_finalize.c | 8 ---- opal/runtime/opal_init.c | 9 ---- 8 files changed, 57 insertions(+), 100 deletions(-) diff --git a/opal/mca/btl/openib/btl_openib_component.c b/opal/mca/btl/openib/btl_openib_component.c index c7dc6c5b40e..d2477f608f6 100644 --- a/opal/mca/btl/openib/btl_openib_component.c +++ b/opal/mca/btl/openib/btl_openib_component.c @@ -2557,39 +2557,6 @@ btl_openib_component_init(int *num_btl_modules, goto no_btls; } - /* If we are using ptmalloc2 and there are no posix threads - available, this will cause memory corruption. Refuse to run. - Right now, ptmalloc2 is the only memory manager that we have on - OS's that support OpenFabrics that provide both FREE and MUNMAP - support, so the following test is [currently] good enough... */ - value = opal_mem_hooks_support_level(); - - /* If we have a memory manager available, and - opal_leave_pinned==-1, then unless the user explicitly set - opal_leave_pinned_pipeline==0, then set opal_leave_pinned to 1. - - We have a memory manager if we have both FREE and MUNMAP - support */ - if ((OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_MUNMAP_SUPPORT) == - ((OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_MUNMAP_SUPPORT) & value)) { - if (0 == opal_leave_pinned_pipeline && - -1 == opal_leave_pinned) { - opal_leave_pinned = 1; - } - } else { - opal_leave_pinned = 0; - opal_leave_pinned_pipeline = 0; - } - -#if OPAL_CUDA_SUPPORT - if (mca_btl_openib_component.cuda_want_gdr && (0 == opal_leave_pinned)) { - opal_show_help("help-mpi-btl-openib.txt", - "CUDA_gdr_and_nopinned", true, - opal_process_info.nodename); - goto no_btls; - } -#endif /* OPAL_CUDA_SUPPORT */ - index = mca_base_var_find("ompi", "btl", "openib", "max_inline_data"); if (index >= 0) { if (OPAL_SUCCESS == mca_base_var_get_value(index, NULL, &source, NULL)) { @@ -2926,6 +2893,22 @@ btl_openib_component_init(int *num_btl_modules, mca_btl_openib_component.if_exclude_list = NULL; } + /* If we are using ptmalloc2 and there are no posix threads + available, this will cause memory corruption. Refuse to run. + Right now, ptmalloc2 is the only memory manager that we have on + OS's that support OpenFabrics that provide both FREE and MUNMAP + support, so the following test is [currently] good enough... */ + value = opal_mem_hooks_support_level(); + +#if OPAL_CUDA_SUPPORT + if (mca_btl_openib_component.cuda_want_gdr && (0 == opal_leave_pinned)) { + opal_show_help("help-mpi-btl-openib.txt", + "CUDA_gdr_and_nopinned", true, + opal_process_info.nodename); + goto no_btls; + } +#endif /* OPAL_CUDA_SUPPORT */ + mca_btl_openib_component.memory_registration_verbose = opal_output_open(NULL); opal_output_set_verbosity (mca_btl_openib_component.memory_registration_verbose, mca_btl_openib_component.memory_registration_verbose_level); diff --git a/opal/mca/btl/ugni/btl_ugni_component.c b/opal/mca/btl/ugni/btl_ugni_component.c index acf0b073ec7..04076a16d6b 100644 --- a/opal/mca/btl/ugni/btl_ugni_component.c +++ b/opal/mca/btl/ugni/btl_ugni_component.c @@ -331,26 +331,6 @@ btl_ugni_component_close(void) return OPAL_SUCCESS; } -static void mca_btl_ugni_autoset_leave_pinned (void) { - if (MCA_BTL_UGNI_RCACHE_UDREG != mca_btl_ugni_component.rcache_type) { - int value = opal_mem_hooks_support_level(); - if ((OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_MUNMAP_SUPPORT) == - ((OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_MUNMAP_SUPPORT) & value)) { - /* Set leave pinned to 1 if leave pinned pipeline is not set */ - if (-1 == opal_leave_pinned) { - opal_leave_pinned = !opal_leave_pinned_pipeline; - } - } else { - opal_leave_pinned = 0; - opal_leave_pinned_pipeline = 0; - } - } else if (-1 == opal_leave_pinned) { - /* if udreg is in use we can set leave pinned without checking for the - * memory hooks. */ - opal_leave_pinned = !opal_leave_pinned_pipeline; - } -} - static mca_btl_base_module_t ** mca_btl_ugni_component_init (int *num_btl_modules, bool enable_progress_threads, @@ -409,8 +389,6 @@ mca_btl_ugni_component_init (int *num_btl_modules, } } - mca_btl_ugni_autoset_leave_pinned (); - mca_btl_ugni_module.super.btl_rdma_pipeline_send_length = mca_btl_ugni_module.super.btl_eager_limit; for (i = 0 ; i < mca_btl_ugni_component.ugni_num_btls ; ++i) { diff --git a/opal/mca/rcache/base/base.h b/opal/mca/rcache/base/base.h index d4bfeb27ee3..a36b7bd0799 100644 --- a/opal/mca/rcache/base/base.h +++ b/opal/mca/rcache/base/base.h @@ -29,6 +29,7 @@ #include "opal/class/opal_list.h" #include "opal/mca/mca.h" #include "opal/mca/rcache/rcache.h" +#include "opal/mca/memory/base/base.h" BEGIN_C_DECLS diff --git a/opal/mca/rcache/base/rcache_base_create.c b/opal/mca/rcache/base/rcache_base_create.c index 1219677d66b..39456515309 100644 --- a/opal/mca/rcache/base/rcache_base_create.c +++ b/opal/mca/rcache/base/rcache_base_create.c @@ -43,24 +43,6 @@ mca_rcache_base_module_t* mca_rcache_base_module_create (const char* name, void mca_base_component_list_item_t *cli; mca_rcache_base_selected_module_t *sm; - OPAL_LIST_FOREACH(cli, &opal_rcache_base_framework.framework_components, mca_base_component_list_item_t) { - component = (mca_rcache_base_component_t *) cli->cli_component; - if(0 == strcmp(component->rcache_version.mca_component_name, name)) { - module = component->rcache_init (resources); - break; - } - } - - if ( NULL == module ) { - return NULL; - } - - sm = OBJ_NEW(mca_rcache_base_selected_module_t); - sm->rcache_component = component; - sm->rcache_module = module; - sm->user_data = user_data; - opal_list_append(&mca_rcache_base_modules, (opal_list_item_t*) sm); - /* on the very first creation of a module we init the memory callback */ if (!mca_rcache_base_used_mem_hooks) { @@ -71,11 +53,21 @@ mca_rcache_base_module_t* mca_rcache_base_module_create (const char* name, void * coded this more succinctly, but this is more clear. Do not * check memory hooks if the rcache does not provide an * range invalidation function.. */ - if ((opal_leave_pinned > 0 || opal_leave_pinned_pipeline) && - module->rcache_invalidate_range) { + if (opal_leave_pinned != 0 || opal_leave_pinned_pipeline) { + /* open the memory manager components. Memory hooks may be + triggered before this (any time after mem_free_init(), + actually). This is a hook available for memory manager hooks + without good initialization routine support */ + (void) mca_base_framework_open (&opal_memory_base_framework, 0); + } + + if (opal_leave_pinned != 0 || opal_leave_pinned_pipeline) { if ((OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_MUNMAP_SUPPORT) == ((OPAL_MEMORY_FREE_SUPPORT | OPAL_MEMORY_MUNMAP_SUPPORT) & opal_mem_hooks_support_level())) { + if (-1 == opal_leave_pinned) { + opal_leave_pinned = !opal_leave_pinned_pipeline; + } opal_mem_hooks_register_release(mca_rcache_base_mem_cb, NULL); } else { opal_show_help("help-rcache-base.txt", "leave pinned failed", @@ -90,6 +82,24 @@ mca_rcache_base_module_t* mca_rcache_base_module_create (const char* name, void } } + OPAL_LIST_FOREACH(cli, &opal_rcache_base_framework.framework_components, mca_base_component_list_item_t) { + component = (mca_rcache_base_component_t *) cli->cli_component; + if(0 == strcmp(component->rcache_version.mca_component_name, name)) { + module = component->rcache_init (resources); + break; + } + } + + if ( NULL == module ) { + return NULL; + } + + sm = OBJ_NEW(mca_rcache_base_selected_module_t); + sm->rcache_component = component; + sm->rcache_module = module; + sm->user_data = user_data; + opal_list_append(&mca_rcache_base_modules, (opal_list_item_t*) sm); + return module; } diff --git a/opal/mca/rcache/base/rcache_base_frame.c b/opal/mca/rcache/base/rcache_base_frame.c index 5e7e5ed0b11..339e0e77aa1 100644 --- a/opal/mca/rcache/base/rcache_base_frame.c +++ b/opal/mca/rcache/base/rcache_base_frame.c @@ -10,7 +10,7 @@ * University of Stuttgart. All rights reserved. * Copyright (c) 2004-2005 The Regents of the University of California. * All rights reserved. - * Copyright (c) 2012-2013 Los Alamos National Security, LLC. + * Copyright (c) 2012-2016 Los Alamos National Security, LLC. * All rights reserved * Copyright (c) 2015-2016 Research Organization for Information Science * and Technology (RIST). All rights reserved. @@ -92,6 +92,13 @@ static int mca_rcache_base_close(void) /* deregister memory free callback */ if (mca_rcache_base_used_mem_hooks) { opal_mem_hooks_unregister_release(mca_rcache_base_mem_cb); + + /* close the memory manager components. Registered hooks can + still be fired any time between now and the call to + opal_mem_free_finalize(), and callbacks from the memory manager + hooks to the bowels of the mem_free code can still occur any + time between now and end of application (even post main()!) */ + (void) mca_base_framework_close (&opal_memory_base_framework); } /* All done */ diff --git a/opal/memoryhooks/memory.c b/opal/memoryhooks/memory.c index e72ca291570..00e7404f9f9 100644 --- a/opal/memoryhooks/memory.c +++ b/opal/memoryhooks/memory.c @@ -106,7 +106,7 @@ opal_mem_hooks_set_support(int support) void opal_mem_hooks_release_hook(void *buf, size_t length, bool from_alloc) { - opal_list_item_t *item; + callback_list_item_t *cbitem, *next; if (!release_run_callbacks) return; @@ -121,12 +121,7 @@ opal_mem_hooks_release_hook(void *buf, size_t length, bool from_alloc) */ opal_atomic_lock(&release_lock); - item = opal_list_get_first(&release_cb_list); - while(item != opal_list_get_end(&release_cb_list)) { - opal_list_item_t* next = opal_list_get_next(item); - callback_list_item_t *cbitem = (callback_list_item_t*) item; - item = next; - + OPAL_LIST_FOREACH_SAFE(cbitem, next, &release_cb_list, callback_list_item_t) { opal_atomic_unlock(&release_lock); cbitem->cbfunc(buf, length, cbitem->cbdata, (bool) from_alloc); opal_atomic_lock(&release_lock); diff --git a/opal/runtime/opal_finalize.c b/opal/runtime/opal_finalize.c index 7138d1680d7..04a580ecac6 100644 --- a/opal/runtime/opal_finalize.c +++ b/opal/runtime/opal_finalize.c @@ -40,7 +40,6 @@ #include "opal/mca/installdirs/base/base.h" #include "opal/mca/memchecker/base/base.h" #include "opal/mca/memcpy/base/base.h" -#include "opal/mca/memory/base/base.h" #include "opal/mca/patcher/base/base.h" #include "opal/mca/backtrace/base/base.h" #include "opal/mca/sec/base/base.h" @@ -154,13 +153,6 @@ opal_finalize(void) (void) mca_base_framework_close(&opal_backtrace_base_framework); (void) mca_base_framework_close(&opal_memchecker_base_framework); - - /* close the memory manager components. Registered hooks can - still be fired any time between now and the call to - opal_mem_free_finalize(), and callbacks from the memory manager - hooks to the bowels of the mem_free code can still occur any - time between now and end of application (even post main()!) */ - (void) mca_base_framework_close(&opal_memory_base_framework); (void) mca_base_framework_close(&opal_patcher_base_framework); /* close the memcpy framework */ diff --git a/opal/runtime/opal_init.c b/opal/runtime/opal_init.c index 0a19a70a8e1..6140b776529 100644 --- a/opal/runtime/opal_init.c +++ b/opal/runtime/opal_init.c @@ -439,15 +439,6 @@ opal_init(int* pargc, char*** pargv) /* select a patcher module. if a patcher module can not be found it is not an error. */ (void) opal_patcher_base_select (); - /* open the memory manager components. Memory hooks may be - triggered before this (any time after mem_free_init(), - actually). This is a hook available for memory manager hooks - without good initialization routine support */ - if (OPAL_SUCCESS != (ret = mca_base_framework_open(&opal_memory_base_framework, 0))) { - error = "opal_memory_base_open"; - goto return_error; - } - /* initialize the memory manager / tracker */ if (OPAL_SUCCESS != (ret = opal_mem_hooks_init())) { error = "opal_mem_hooks_init";