i#1979 Mac64: Add 64-bit Mojave + XCode 10.1 build support (#3514)

A series of small changes to build with XCode 10.1 64-bit: + Updates -max_supported_os_version on Mac to 18. + Updates 64-bit Mac frame field. + Removes , from runtime options in template defines. + Removes 32-bit-only tests from the OSX label list for 64-bit. + Increases the core's preferred base to 0x1'71000000 to stay above the default 0x1' _PAGEZERO. + Adds missing DECL_EXTERNs in mangle_suspend and mangle_asynch tests. + Adds missing GLOBAL_REFs in common.decode test. + Re-defines sigcontext_t and uc_mcontext in tools.h to match this toolchain. + Adds 64-bit libelftc.a and libdwarf.a Mach-O libraries along with instructions on how they were built. + Adds -mmacosx-version-min=10.9 to drcachesim to match core/. + Fixes a format string warning in drcachesim. Includes a number of similar small changes from Shawn Denbow's first 2 commits on the project-mac64 branch in PR #2269 and PR #2273, updated for clang-format: + Move -vm_base above 4GB since __PAGEZERO takes up first 4GB by default + Increase MEMQUERY_INTERNAL_DATA_LEN + Update Mach-O parsing to check for LC_SEGMENT_64 + Change error notifications to warning to allow build to continue + Define SC_FIELD for 64-bit registers r8-r15 and update references + Update dynamorio_mach_dep_syscall to use syscall for 64-bit + Update dynamorio_mach_syscall to use syscall for 64-bit + Change code referencing eflags to xflags for cross platform compat + Misc code fix-ups Issue: #1979
DynamoRIO · Apr 15, 2019 · 229ff65 · 229ff65
1 parent 09c690b
commit 229ff65
Show file tree

Hide file tree

Showing 23 changed files with 145 additions and 70 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -242,7 +242,7 @@ endif (CMAKE_C_SIZEOF_DATA_PTR EQUAL 8)
 
 if (APPLE AND X64)
   # XXX #1979: 64-bit OSX is not supported.
-  message(FATAL_ERROR "64-bit Mac OSX is not supported")
+  message(WARNING "64-bit Mac OSX is not supported")
 endif ()
 
 option(VMKERNEL "target VMkernel (not officially supported yet)")
@@ -334,9 +334,12 @@ option(SET_PREFERRED_BASE "set a preferred library base address")
 if (WIN32 AND DEBUG)
   # apparently no numeric type so we use STRING
   set(preferred_base "0x15000000" CACHE STRING "Preferred library base address")
-else (WIN32 AND DEBUG)
+elseif (APPLE AND X64)
+  # Set to higher than _PAGEZERO which is [0..0x1'00000000).
+  set(preferred_base "0x171000000" CACHE STRING "Preferred library base address")
+else ()
   set(preferred_base "0x71000000" CACHE STRING "Preferred library base address")
-endif (WIN32 AND DEBUG)
+endif ()
 
 # for x64: PR 253624: we need our library to be next to our heap
 # for win32: not PIC so need a base
@@ -730,9 +733,10 @@ if (UNIX)
   endif ()
   if (APPLE AND CMAKE_COMPILER_IS_CLANG)
     # Ensure our binaries can run on older OSX
-    # We're building our release package on 10.11, and that XCode does not
+    # We're building our release package on 10.14, and that XCode does not
     # supply C++ libraries for below 10.9, causing drcachesim to fail to build.
     # Thus we have abandoned support for below 10.9.
+    # If this is updated, update drcachesim's as well.
     set(BASE_CFLAGS "${BASE_CFLAGS} -mmacosx-version-min=10.9")
   endif ()
   if (APPLE)

diff --git a/clients/drcachesim/CMakeLists.txt b/clients/drcachesim/CMakeLists.txt
@@ -1,5 +1,5 @@
 # **********************************************************
-# Copyright (c) 2015-2018 Google, Inc.    All rights reserved.
+# Copyright (c) 2015-2019 Google, Inc.    All rights reserved.
 # **********************************************************
 
 # Redistribution and use in source and binary forms, with or without
@@ -357,6 +357,10 @@ macro(add_win32_flags target)
     if (NOT cur MATCHES "-std=")
       append_property_string(TARGET ${target} COMPILE_FLAGS "-std=c++11")
     endif ()
+    if (APPLE)
+      # Match the core/ flags.
+      append_property_string(TARGET ${target} COMPILE_FLAGS "-mmacosx-version-min=10.9")
+    endif ()
   endif ()
 endmacro ()
 

diff --git a/clients/drcachesim/reader/file_reader.h b/clients/drcachesim/reader/file_reader.h
@@ -1,5 +1,5 @@
 /* **********************************************************
- * Copyright (c) 2016-2018 Google, Inc.  All rights reserved.
+ * Copyright (c) 2016-2019 Google, Inc.  All rights reserved.
  * **********************************************************/
 
 /*
@@ -54,7 +54,7 @@
 #    ifdef WINDOWS
 #        define ZHEX64_FORMAT_STRING "%016I64x"
 #    else
-#        if defined(__i386__) || defined(__arm__)
+#        if defined(__i386__) || defined(__arm__) || defined(__APPLE__)
 #            define ZHEX64_FORMAT_STRING "%016llx"
 #        else
 #            define ZHEX64_FORMAT_STRING "%016lx"

diff --git a/core/arch/aarch64/codec.txt b/core/arch/aarch64/codec.txt
@@ -1193,4 +1193,3 @@ x101101011000000000101xxxxxxxxxx  cls     wx0 : wx5
 01101110xx1xxxxx101000xxxxxxxxxx     umlsl2    q0 : q5 q16 bhs_sz
 00101110xx1xxxxx110000xxxxxxxxxx     umull     q0 : d5 d16 bhs_sz
 01101110xx1xxxxx110000xxxxxxxxxx     umull2    q0 : q5 q16 bhs_sz
-
diff --git a/core/arch/x86/x86.asm b/core/arch/x86/x86.asm
@@ -1345,7 +1345,7 @@ GLOBAL_LABEL(dynamorio_condvar_wake_and_jmp:)
         * so transparency should be ok so long as the app's stack is valid.
         */
         mov      REG_XDI, REG_XCX /* save across syscall */
-        mov      REG_XAX, DWORD [REG_XAX] /* load mach_synch_t->sem */
+        mov      REG_XAX, PTRSZ [REG_XAX] /* load mach_synch_t->sem */
 #  ifdef X64
         mov      ARG1, REG_XAX
         mov      eax, MACH_semaphore_signal_all_trap

diff --git a/core/drlibc/drlibc_x86.asm b/core/drlibc/drlibc_x86.asm
@@ -42,6 +42,9 @@
 
 #include "../asm_defines.asm"
 #include "x86_asm_defines.asm" /* PUSHGPR, POPGPR, etc. */
+#ifdef MACOS
+# include "include/syscall_mach.h" /* SYSCALL_NUM_MARKER_MACH */
+#endif
 START_FILE
 
 DECL_EXTERN(unexpected_return)
@@ -214,6 +217,8 @@ GLOBAL_LABEL(dynamorio_mach_dep_syscall:)
         cmp      REG_XBX, 3
         je       mach_dep_syscall_ready
         mov      ARG4, ARG6
+mach_dep_syscall_ready:
+        syscall
 #  else
         push     REG_XBP
         push     REG_XSI
@@ -246,10 +251,8 @@ mach_dep_syscall_0args:
         push     0 /* extra slot */
         /* clear the top half so we can always consider the result 64-bit */
         mov      edx, 0
-#  endif
         /* mach dep syscalls use interrupt 0x82 */
         int      HEX(82)
-#  ifndef X64
         lea      esp, [7*ARG_SZ + esp] /* must not change flags */
         pop      REG_XDI
         pop      REG_XSI
@@ -279,6 +282,7 @@ GLOBAL_LABEL(dynamorio_mach_syscall:)
         /* reverse order so we don't clobber earlier args */
         mov      REG_XBX, ARG2 /* put num_args where we can reference it longer */
         mov      rax, ARG1 /* sysnum: only need eax, but need rax to use ARG1 (or movzx) */
+        or       eax, SYSCALL_NUM_MARKER_MACH
         cmp      REG_XBX, 0
         je       dynamorio_mach_syscall_ready
         mov      ARG1, ARG3
@@ -291,6 +295,8 @@ GLOBAL_LABEL(dynamorio_mach_syscall:)
         cmp      REG_XBX, 3
         je       dynamorio_mach_syscall_ready
         mov      ARG4, ARG6
+dynamorio_mach_syscall_ready:
+        syscall
 #  else
         push     REG_XBP
         push     REG_XSI
@@ -314,12 +320,8 @@ dynamorio_mach_syscall_1args:
         mov      ebx, [16+12 + esp] /* arg1 */
 dynamorio_mach_syscall_0args:
         mov      eax, [16+ 4 + esp] /* sysnum */
-#  ifdef X64
-        or       eax, SYSCALL_NUM_MARKER_MACH
-#  else
         /* The sysnum is passed as a negative number */
         neg      eax
-#  endif
         /* args are on stack, w/ an extra slot (retaddr of syscall wrapper) */
         lea      REG_XSP, [-2*ARG_SZ + REG_XSP] /* maintain align-16: retaddr-5th below */
         /* args are on stack, w/ an extra slot (retaddr of syscall wrapper) */
@@ -328,7 +330,6 @@ dynamorio_mach_syscall_0args:
         push     ecx
         push     ebx
         push     0 /* extra slot */
-#  endif
         /* If we use ADDRTAKEN_LABEL and GLOBAL_REF we get text relocation
          * complaints so we instead do this hack:
          */
@@ -341,7 +342,6 @@ dynamorio_mach_syscall_next:
          * This implies that we can't return 64-bit in 32-bit mode.
          */
         sysenter
-#  ifndef X64
         lea      esp, [7*ARG_SZ + esp] /* must not change flags */
         pop      REG_XDI
         pop      REG_XSI

diff --git a/core/optionsx.h b/core/optionsx.h
@@ -687,7 +687,7 @@ DYNAMIC_OPTION(bool, pause_via_loop,
 
     /* For MacOS, set to 0 to disable the check */
     OPTION_DEFAULT(uint, max_supported_os_version,
-        IF_WINDOWS_ELSE(105, IF_MACOS_ELSE(17, 0)),
+        IF_WINDOWS_ELSE(105, IF_MACOS_ELSE(18, 0)),
         /* case 447, defaults to supporting NT, 2000, XP, 2003, and Vista.
          * Windows 7 added with i#218
          * Windows 8 added with i#565
@@ -1430,11 +1430,13 @@ DYNAMIC_OPTION(bool, pause_via_loop,
      * new default -vm_size of 0x20000000 we want to stay below our various
      * preferred addresses of 0x7xxx0000 so we keep the base plus offset plus
      * size below that.
+     * For a 64-bit process on MacOS __PAGEZERO takes up the first 4GB by default.
      * We ignore this for x64 if -vm_base_near_app and the app is far away.
      */
     OPTION_DEFAULT(uint_addr, vm_base,
                    IF_VMX86_ELSE(IF_X64_ELSE(0x40000000,0x10800000),
-                                 IF_WINDOWS_ELSE(0x16000000, 0x3f000000)),
+                                 IF_WINDOWS_ELSE(0x16000000, IF_MACOS_ELSE(
+                                 IF_X64_ELSE(0x120000000,0x3f000000),0x3f000000))),
                    "preferred base address hint")
      /* FIXME: we need to find a good location with no conflict with DLLs or apps allocations */
     OPTION_DEFAULT(uint_addr, vm_max_offset,

diff --git a/core/unix/memquery.h b/core/unix/memquery.h
@@ -64,7 +64,7 @@ typedef struct _memquery_iter_t {
      * without using static data and limiting to one iterator (and having
      * to unprotect and reprotect if in .data).
      */
-#define MEMQUERY_INTERNAL_DATA_LEN 96 /* 84 bytes needed for Mac 10.9.1 */
+#define MEMQUERY_INTERNAL_DATA_LEN 116 /* 104 bytes needed for MacOS 64-bit */
     char internal[MEMQUERY_INTERNAL_DATA_LEN];
 } memquery_iter_t;
 

diff --git a/core/unix/module_macho.c b/core/unix/module_macho.c
@@ -99,7 +99,7 @@ module_walk_program_headers(app_pc base, size_t view_size, bool at_map, bool dyn
     cmd = (struct load_command *)(hdr + 1);
     cmd_stop = (struct load_command *)((byte *)cmd + hdr->sizeofcmds);
     while (cmd < cmd_stop) {
-        if (cmd->cmd == LC_SEGMENT) {
+        if (cmd->cmd == LC_SEGMENT || cmd->cmd == LC_SEGMENT_64) {
             segment_command_t *seg = (segment_command_t *)cmd;
             found_seg = true;
             LOG(GLOBAL, LOG_VMAREAS, 4, "%s: segment %s addr=0x%x sz=0x%x file=0x%x\n",
@@ -171,7 +171,7 @@ module_walk_program_headers(app_pc base, size_t view_size, bool at_map, bool dyn
             /* Now that we have the load delta, we can add the abs addr segments */
             cmd = (struct load_command *)(hdr + 1);
             while (cmd < cmd_stop) {
-                if (cmd->cmd == LC_SEGMENT) {
+                if (cmd->cmd == LC_SEGMENT || cmd->cmd == LC_SEGMENT_64) {
                     segment_command_t *seg = (segment_command_t *)cmd;
                     if (strcmp(seg->segname, "__PAGEZERO") == 0 && seg->initprot == 0) {
                         /* skip */
@@ -210,10 +210,10 @@ module_walk_program_headers(app_pc base, size_t view_size, bool at_map, bool dyn
                     /* even if stripped, dynamic symbols are in this table */
                     struct symtab_command *symtab = (struct symtab_command *)cmd;
                     out_data->symtab =
-                        (app_pc)symtab->symoff + load_delta + linkedit_delta;
+                        (app_pc)(symtab->symoff + load_delta + linkedit_delta);
                     out_data->num_syms = symtab->nsyms;
                     out_data->strtab =
-                        (app_pc)symtab->stroff + load_delta + linkedit_delta;
+                        (app_pc)(symtab->stroff + load_delta + linkedit_delta);
                     out_data->strtab_sz = symtab->strsize;
                 } else if (cmd->cmd == LC_UUID) {
                     memcpy(out_data->uuid, ((struct uuid_command *)cmd)->uuid,
@@ -268,7 +268,7 @@ module_entry_point(app_pc base, ptr_int_t load_delta)
 #ifdef X64
             const x86_thread_state64_t *reg =
                 (const x86_thread_state64_t *)((char *)cmd + LC_UNIXTHREAD_REGS_OFFS);
-            return (app_pc)reg->__rip + load_delta
+            return (app_pc)reg->__rip + load_delta;
 #else
             const i386_thread_state_t *reg =
                 (const i386_thread_state_t *)((byte *)cmd + LC_UNIXTHREAD_REGS_OFFS);

diff --git a/core/unix/os.c b/core/unix/os.c
@@ -2858,10 +2858,12 @@ replace_thread_id(thread_id_t old, thread_id_t new)
     ASSERT(is_thread_tls_initialized());
     DOCHECK(1, {
         thread_id_t old_tid;
-        READ_TLS_INT_SLOT_IMM(TLS_THREAD_ID_OFFSET, old_tid);
+        IF_LINUX_ELSE(READ_TLS_INT_SLOT_IMM(TLS_THREAD_ID_OFFSET, old_tid),
+                      READ_TLS_SLOT_IMM(TLS_THREAD_ID_OFFSET, old_tid));
         ASSERT(old_tid == old);
     });
-    WRITE_TLS_INT_SLOT_IMM(TLS_THREAD_ID_OFFSET, new_tid);
+    IF_LINUX_ELSE(WRITE_TLS_INT_SLOT_IMM(TLS_THREAD_ID_OFFSET, new_tid),
+                  WRITE_TLS_SLOT_IMM(TLS_THREAD_ID_OFFSET, new_tid));
 #else
     int i;
     d_r_mutex_lock(&tls_lock);
@@ -4865,7 +4867,7 @@ syscall_successful(priv_mcontext_t *mc, int normalized_sysnum)
          */
         return ((ptr_int_t)MCXT_SYSCALL_RES(mc) >= 0);
     } else
-        return !TEST(EFLAGS_CF, mc->eflags);
+        return !TEST(EFLAGS_CF, mc->xflags);
 #else
     if (normalized_sysnum == IF_X64_ELSE(SYS_mmap, SYS_mmap2) ||
 #    if !defined(ARM) && !defined(X64)
@@ -4891,7 +4893,7 @@ set_success_return_val(dcontext_t *dcontext, reg_t val)
     /* On MacOS, success is determined by CF, except for Mach syscalls, but
      * there it doesn't hurt to set CF.
      */
-    mc->eflags &= ~(EFLAGS_CF);
+    mc->xflags &= ~(EFLAGS_CF);
 #endif
     MCXT_SYSCALL_RES(mc) = val;
 }
@@ -4903,7 +4905,7 @@ set_failure_return_val(dcontext_t *dcontext, uint errno_val)
     priv_mcontext_t *mc = get_mcontext(dcontext);
 #ifdef MACOS
     /* On MacOS, success is determined by CF, and errno is positive */
-    mc->eflags |= EFLAGS_CF;
+    mc->xflags |= EFLAGS_CF;
     MCXT_SYSCALL_RES(mc) = errno_val;
 #else
     MCXT_SYSCALL_RES(mc) = -(int)errno_val;

diff --git a/core/unix/os_public.h b/core/unix/os_public.h
@@ -105,7 +105,6 @@ typedef kernel_sigcontext_t sigcontext_t;
 #        define SC_R13 SC_FIELD(r13)
 #        define SC_R14 SC_FIELD(r14)
 #        define SC_R15 SC_FIELD(r15)
-
 #        ifdef MACOS
 #            define SC_XFLAGS SC_FIELD(rflags)
 #        else

diff --git a/core/unix/signal.c b/core/unix/signal.c
@@ -2603,14 +2603,14 @@ thread_set_self_context(void *cxt)
         buf.xsp = sc->SC_XSP - XSP_SZ; /* extra slot for retaddr */
         buf.xip = sc->SC_XIP;
 #    ifdef X64
-        buf.r8 = sc->r8;
-        buf.r9 = sc->r9;
-        buf.r10 = sc->r10;
-        buf.r11 = sc->r11;
-        buf.r12 = sc->r12;
-        buf.r13 = sc->r13;
-        buf.r14 = sc->r14;
-        buf.r15 = sc->r15;
+        buf.r8 = sc->SC_R8;
+        buf.r9 = sc->SC_R9;
+        buf.r10 = sc->SC_R10;
+        buf.r11 = sc->SC_R11;
+        buf.r12 = sc->SC_R12;
+        buf.r13 = sc->SC_R13;
+        buf.r14 = sc->SC_R14;
+        buf.r15 = sc->SC_R15;
 #    endif
         dr_longjmp(&buf, sc->SC_XAX);
         return;
@@ -3066,15 +3066,17 @@ fixup_rtframe_pointers(dcontext_t *dcontext, int sig, sigframe_rt_t *f_old,
     /* 32-bit kernel copies to aligned buf first */
     IF_X64(ASSERT(ALIGNED(f_new->uc.uc_mcontext.fpstate, 16)));
 #elif defined(MACOS)
-#    ifndef X64
+#    ifdef X64
+    ASSERT_NOT_IMPLEMENTED(false);
+#    else
     f_new->pinfo = &(f_new->info);
     f_new->puc = &(f_new->uc);
-#    endif
     f_new->puc->uc_mcontext =
         (IF_X64_ELSE(_STRUCT_MCONTEXT64, _STRUCT_MCONTEXT32) *)&f_new->mc;
     LOG(THREAD, LOG_ASYNCH, 3, "\tf_new=" PFX ", &handler=" PFX "\n", f_new,
         &f_new->handler);
     ASSERT(!for_app || ALIGNED(&f_new->handler, 16));
+#    endif
 #endif /* X86 && LINUX */
 }
 
@@ -3262,9 +3264,13 @@ copy_frame_to_stack(dcontext_t *dcontext, int sig, sigframe_rt_t *frame, byte *s
     }
 
 #ifdef MACOS
+#    ifdef X64
+    ASSERT_NOT_IMPLEMENTED(false);
+#    else
     /* Update handler field, which is passed to the libc trampoline, to app */
     ASSERT(info->app_sigaction[sig] != NULL);
     ((sigframe_rt_t *)sp)->handler = (app_pc)info->app_sigaction[sig]->handler;
+#    endif
 #endif
 }
 
@@ -4741,7 +4747,7 @@ is_safe_read_ucxt(kernel_ucontext_t *ucxt)
 /* stub in x86.asm passes our xsp to us */
 #    ifdef MACOS
 void
-master_signal_handler_C(handler_t handler, int style, int sig, kernel_siginfo_t *info,
+master_signal_handler_C(handler_t handler, int style, int sig, kernel_siginfo_t *siginfo,
                         kernel_ucontext_t *ucxt, byte *xsp)
 #    else
 void

diff --git a/core/unix/signal_macos.c b/core/unix/signal_macos.c
@@ -241,14 +241,14 @@ dump_sigcontext(dcontext_t *dcontext, sigcontext_t *sc)
     LOG(THREAD, LOG_ASYNCH, 1, "\txcx=" PFX "\n", sc->SC_XCX);
     LOG(THREAD, LOG_ASYNCH, 1, "\txax=" PFX "\n", sc->SC_XAX);
 #ifdef X64
-    LOG(THREAD, LOG_ASYNCH, 1, "\t r8=" PFX "\n", sc->r8);
-    LOG(THREAD, LOG_ASYNCH, 1, "\t r9=" PFX "\n", sc->r8);
-    LOG(THREAD, LOG_ASYNCH, 1, "\tr10=" PFX "\n", sc->r10);
-    LOG(THREAD, LOG_ASYNCH, 1, "\tr11=" PFX "\n", sc->r11);
-    LOG(THREAD, LOG_ASYNCH, 1, "\tr12=" PFX "\n", sc->r12);
-    LOG(THREAD, LOG_ASYNCH, 1, "\tr13=" PFX "\n", sc->r13);
-    LOG(THREAD, LOG_ASYNCH, 1, "\tr14=" PFX "\n", sc->r14);
-    LOG(THREAD, LOG_ASYNCH, 1, "\tr15=" PFX "\n", sc->r15);
+    LOG(THREAD, LOG_ASYNCH, 1, "\t r8=" PFX "\n", sc->SC_R8);
+    LOG(THREAD, LOG_ASYNCH, 1, "\t r9=" PFX "\n", sc->SC_R8);
+    LOG(THREAD, LOG_ASYNCH, 1, "\tr10=" PFX "\n", sc->SC_R10);
+    LOG(THREAD, LOG_ASYNCH, 1, "\tr11=" PFX "\n", sc->SC_R11);
+    LOG(THREAD, LOG_ASYNCH, 1, "\tr12=" PFX "\n", sc->SC_R12);
+    LOG(THREAD, LOG_ASYNCH, 1, "\tr13=" PFX "\n", sc->SC_R13);
+    LOG(THREAD, LOG_ASYNCH, 1, "\tr14=" PFX "\n", sc->SC_R14);
+    LOG(THREAD, LOG_ASYNCH, 1, "\tr15=" PFX "\n", sc->SC_R15);
 #endif
 
     LOG(THREAD, LOG_ASYNCH, 1, "\txip=" PFX "\n", sc->SC_XIP);

diff --git a/core/unix/signal_private.h b/core/unix/signal_private.h
@@ -53,6 +53,9 @@
 #elif defined(MACOS)
 #    include "../globals.h" /* this defines _XOPEN_SOURCE for Mac */
 #    include <signal.h>     /* after globals.h, for _XOPEN_SOURCE from os_exports.h */
+#    ifdef X64
+#        include <sys/_types/_ucontext64.h> /* for _STRUCT_UCONTEXT64 */
+#    endif
 #endif
 
 #include "os_private.h"
@@ -187,7 +190,7 @@ typedef _STRUCT_UCONTEXT /* == __darwin_ucontext */ kernel_ucontext_t;
 #    define SIGMASK_FROM_UCXT(ucxt) ((kernel_sigset_t *)&((ucxt)->uc_sigmask))
 #endif
 
-#ifdef LINUX
+#if defined(LINUX) || defined(X64)
 #    define SIGINFO_FROM_RT_FRAME(frame) (&(frame)->info)
 #elif defined(MACOS)
 /* Make sure to access through pinfo rather than info as on Mac the info