Optimize syslog_write performance #6618
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
xiaoxiang781216
force-pushed
the
syslog
branch
4 times, most recently
from
311327d
to
4ac0df7
Compare
xiaoxiang781216
force-pushed
the
syslog
branch
2 times, most recently
from
042ae9a
to
cb000fc
Compare
Since up_nputs can be replaced by chip vendor, they can provide the new implementation of up_nputs to gain the better performance. But, yes, it isn't fast by default because the default implementation simply forward the string to up_putc one by one.
remove the lock just to make syslog become the signal safety. |
and change up_puts as a simple macro Signed-off-by: Xiang Xiao <[email protected]>
Signed-off-by: Xiang Xiao <[email protected]>
Signed-off-by: Xiang Xiao <[email protected]>
to avoid the problem when is called from interrupt/signal handler Signed-off-by: Xiang Xiao <[email protected]>
pkarashchenko
approved these changes
Jul 16, 2022
tmedicci
added a commit
to tmedicci/incubator-nuttx
that referenced
this pull request
May 26, 2023
Considering SMP, multiples CPU would be able to call up_nputs simultaneously. This function is most used by syslog and, in spite of the syslog supporting buffered output, the implementation of the up_nputs allows multiple threads to run concurrently, messing with the log messages. Please check apache#6618 for more information about the locking mechanism.
tmedicci
added a commit
to tmedicci/incubator-nuttx
that referenced
this pull request
May 26, 2023
Considering SMP, multiples CPU would be able to call up_nputs simultaneously. This function is most used by syslog and, in spite of the syslog supporting buffered output, the implementation of the up_nputs allows multiple threads to run concurrently, messing with the log messages. Please check apache#6618 for more information about the locking mechanism.
pussuw
added a commit
to tiiuae/nuttx
that referenced
this pull request
Oct 22, 2024
Blocking while running a signal handler is not advisable, instead write the log string character by character. There is also a potential for a deadlock, as discussed in apache#6618 Note: querying for rtcb->sigdeliver is not 100% ideal, as it only tells _if_ a signal handler has been queued, not if it is running. However, it makes syslog safe / usable which is a debug feature anyhow.
pussuw
added a commit
to tiiuae/nuttx
that referenced
this pull request
Oct 22, 2024
Blocking while running a signal handler is not advisable, instead write the log string character by character. There is also a potential for a deadlock, as discussed in apache#6618 Note: querying for rtcb->sigdeliver is not 100% ideal, as it only tells _if_ a signal handler has been queued, not if it is running. However, it makes syslog safe / usable which is a debug feature anyhow.
pussuw
added a commit
to tiiuae/nuttx
that referenced
this pull request
Oct 23, 2024
Blocking while running a signal handler is not advisable, instead write the log string character by character. There is also a potential for a deadlock, as discussed in apache#6618 Note: querying for rtcb->sigdeliver is not 100% ideal, as it only tells _if_ a signal handler has been queued, not if it is running. However, it makes syslog safe / usable which is a debug feature anyhow.
xiaoxiang781216
pushed a commit
that referenced
this pull request
Oct 23, 2024
Blocking while running a signal handler is not advisable, instead write the log string character by character. There is also a potential for a deadlock, as discussed in #6618 Note: querying for rtcb->sigdeliver is not 100% ideal, as it only tells _if_ a signal handler has been queued, not if it is running. However, it makes syslog safe / usable which is a debug feature anyhow.
stbenn
pushed a commit
to stbenn/nuttx
that referenced
this pull request
Oct 25, 2024
Initial STM32H5 Commit Initial commit of what I deemed essential files for bringing up the STM32H5. src/stm32h5/hardware files were edited by me, but need review. files in src/stm32h5 all need review and edits. include/stm32h5 files need review, some were edited by me. Add Nucleo-H563ZI Folder Add the board folder for the nucleo-h563zi. Right now this is largely a copy of the stm32l562e-dk configuration. Some files may be deleted in the future. Also made minor modifications to arch/arm/src/stm32h5/Kconfig file. hardware/stm32h562xx_rcc.h update Finished register and bit mapping for STM32H5 RCC Rename hardware/stm32h5_rcc.h Renamed stm32h562xx_rcc.h to stm32h5_rcc.h. The RCC register is the same for all versions of the STM32H5. Defined rcc_enableperipherals functions Defined all the functions wihtin rcc_enableperipherals. Getting started on stm32h5_stdclockconfig. Incremental STM32H5 RCC Updates Incremental Updates apache#2 Added stm32h5_lse.c and stm32h5_lsi.c files. Incremental updates to board.h, stm32h5xx_rcc.c, and hardware/stm32h5_rcc.h Incremental Updates apache#3 Added stm32h5_hsi48.c and stm32h5_hsi48.h files. Incremental updates to board.h, stm32h5xx_rcc.c, and hardware/stm32h5_rcc.h. Renamed hardware crs file. Fixed lse.c and lsi.c for STM32H5. Incremental Updates apache#4 Updated setting of VOS for STM32H5. Added HSIDIV definition to hardware/stm32h5_rcc.h for potential of changing HSIDIV from default. Changed board.h to use HSI of 32 MHz, which is the default. We still set SYSCLK to the max of 250MHz. First STM32H5 PWR Commit Rewrote hardware/stm32h5_pwr.h. Added stm32h5_pwr.c and stm32h5_pwr.h. Made minor changes to RCC files based on PWR peripheral. PWR Peripheral Changes Removed enablesmps function. LDO or SMPS is decided by hardware. Removed enablepwrclk. There is no PWREN for the STM32H5. Rewrote adustvcore. vcore must be adjusted incrementally. Incremental Updates apache#5 Changed stm32 to stm32h5 in pwr.c. Added additional logic for selecting PLL sources. Added additional logic for enabling LSE or LSI. Set VCORE properly with stm32h5_pwr.c function. Fixes to adjustvcore function. STM32H5 Power and RCC cleanup Fixed some errors with private functions and incorrect preprocessor variables. Changed adjustvcore to not select intermediate VOS levels. Figure 49 in RM shows changing directly from VOS3 to VOS1. Added function adjustvos_ext for externally supplied VCORE. However I'm not sure if VOS should be incremented, then voltage incremented, then frequency incremented, or if VOS should be incremented one by one to final setting, then adjust voltage, then frequency. adjustvos does the former. Won't be used in stdclockconfig. STM32H5 serial update This commit primarily adds functionality taken from the stm32g4 lpuart implementation. The template I used, from the stm32l5, already had the LPUART in there but did not calculate the baud correctly. Added more USARTS and UARTS supported by STM32H5. Minor changes to chip.h, stm32h5_start.c, and Kconfig. STM32H5 Serial Update apache#2 Added support for additional USARTS and UARTS on STM32H5. Other minor serial updates. Build Fixes Various fixes to get the stm32h5 arch to build. Many changes to follow. But for now, Nuttx builds. Remove unnecessary hardware files from STM32H5 directory More build changes Even more build fixes Minor fixes in stm32h5_rcc.c and stm32h5_pwr.c. Changed nucleo-h563zi defconfig to use std clock config. This resulted in errors that were fixed here. Also added stm32h5_lse.c and stm32h5_lsi.c to Make.defs. Removed legacy pinmap. It is deprecated and should not be used on new designs. Confirmed hardware crs and i2c files are correct. Will keep them for now. IRQ info for STM32H52, STM32H53, STM32H56, STM32H57 libcxx: fix compile error from ServiceManager.cpp:17: /home/ligd/platform/dev/apps/external/android/frameworks/native/libs/binder/ndk/include_cpp/android/binder_to_string.h:71:24: error: expected nested-name-specifier before numeric constant 71 | template <typename _U> | ^~ /home/ligd/platform/dev/apps/external/android/frameworks/native/libs/binder/ndk/include_cpp/android/binder_to_string.h:71:24: error: expected ‘>’ before numeric constant In file included from /home/ligd/platform/dev/apps/external/android/frameworks/native/libs/binder/aidl/android/os/ConnectionInfo.h:3, from /home/ligd/platform/dev/apps/external/android/frameworks/native/libs/binder/aidl/android/os/IServiceManager.h:3, from /home/ligd/platform/dev/apps/external/android/frameworks/native/libs/binder/aidl/android/os/BnServiceManager.h:4: /home/ligd/platform/dev/apps/external/android/frameworks/native/libs/binder/ndk/include_cpp/android/binder_to_string.h:72:56: error: no matching function for call to ‘declval<1>()’ 72 | static auto _test(int) -> decltype(std::declval<_U>().toString(), std::true_type()); | ~~~~~~~~~~~~~~~~^~ In file included from /home/ligd/platform/dev/nuttx/include/libcxx/__type_traits/is_convertible.h:18, Signed-off-by: ligd <[email protected]> libc string:Separate code. Separate the code that follows the BSD license into independent files. Signed-off-by: yangguangcai <[email protected]> arch/sim/cmake: remove the host specific -U when HOSTSRCS fix macos compile hostfs.c compile issue. /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX15.0.sdk/usr/include/_string.h:131:62: error: expected function body after function declarator 131 | char *stpncpy(char *__dst, const char *__src, size_t __n) __OSX_AVAILABLE_STARTING(__MAC_10_7, __IPHONE_4_3); | ^ Signed-off-by: buxiasen <[email protected]> Revert "libc/lib_bzero:Add bzero prototype." This reverts commit 908814a. In macos, memset will be automatic optmize to bzero, caused dead loop, as we not using bzero, macro re-define should ablt to cover the requirements. Signed-off-by: buxiasen <[email protected]> arhc/arm64: vector table may be far away form arm64_fatal_handle use 33-bit (+/-4GB) pc-relative addressing to load the address of arm64_fatal_handle Signed-off-by: lipengfei28 <[email protected]> sim: fix asan address space conflict Modify the starting position of the elf segment to 0x5000000 ==2561587==Shadow memory range interleaves with an existing memory mapping. ASan cannot proceed correctly. ABORTING. ==2561587==ASan shadow was supposed to be located in the [0x1ffff000-0x3fffffff] range. ==2561587==Process memory map follows: Signed-off-by: yinshengkai <[email protected]> arm64/toolchains:Add the following kasan compilation options Signed-off-by: wangmingrong1 <[email protected]> remove unused variable 'cpu_freq' Signed-off-by: lipengfei28 <[email protected]> drivers/timers/arch_alarm.c: Remove ndelay_accurate Using ONESHOT_CURRENT retrieves the tick number multiplied by tick time; thus it doesn't give the accurate monotonic time - it is quantized by the tick time. This cannot be used as a ndelay timer, it would always loop at least to the end of the ongoing tick. Revert the up_udelay to use the original "coarse" looping. The "accurate" udelay, if such is needed, should either be done under arch specific code, or there should be a function for getting the accurate time that is available for all the platforms. Signed-off-by: Jukka Laitinen <[email protected]> boards/imx93-evk: Define CONFIG_BOARD_LOOPSPERMSEC Use value measured with 1.8GHz CPU speed Signed-off-by: Jukka Laitinen <[email protected]> arch/x86_64:Fix variable used before assignment Signed-off-by: liwenxiang1 <[email protected]> arch/arm64: vector table 2K align Signed-off-by: lipengfei28 <[email protected]> arm/build: suppress LOAD RWX linker warning Add --no-warn-rwx-segments in case of RAM boot mode to linker to suppress the below warning: "nuttx has a LOAD segment with RWX permissions" Signed-off-by: Jinliang Li <[email protected]> arch/arm64/src/imx9/imx9_lpspi.c: Fix 9-16 bit transfers Signed-off-by: Jukka Laitinen <[email protected]> arch/arm64/src/imx9/imx9_lpspi.c: Small cache operation optimization There is no need to invalidate the RX buffer before every transfer. It is never gets dirty, so it is good to invalidate initially after allocation, and after each transfer. Signed-off-by: Jukka Laitinen <[email protected]> libxx: C++ low level library select LIBSUPCXX by default. Signed-off-by: cuiziwei <[email protected]> nuttx/sim: Fix m64 build error. LD: nuttx nuttx.rel: in function `ff_dct32_float_sse2': (.text+0x66f9e): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_32' defined in .bss.ff_cos_32 section in nuttx.rel (.text+0x66fa7): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_32' defined in .bss.ff_cos_32 section in nuttx.rel (.text+0x672a6): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_16' defined in .bss.ff_cos_16 section in nuttx.rel (.text+0x672ae): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_16' defined in .bss.ff_cos_16 section in nuttx.rel nuttx.rel: in function `ff_imdct_calc_sse': (.text+0x67905): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_64' defined in .bss.ff_cos_64 section in nuttx.rel (.text+0x67948): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_128' defined in .bss.ff_cos_128 section in nuttx.rel (.text+0x67988): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_256' defined in .bss.ff_cos_256 section in nuttx.rel (.text+0x679c8): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_512' defined in .bss.ff_cos_512 section in nuttx.rel (.text+0x67a08): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_1024' defined in .bss.ff_cos_1024 section in nuttx.rel (.text+0x67a48): relocation truncated to fit: R_X86_64_32S against symbol `ff_cos_2048' defined in .bss.ff_cos_2048 section in nuttx.rel (.text+0x67a88): additional relocation overflows omitted from the output Signed-off-by: cuiziwei <[email protected]> tls.h: list.h should depends on CONFIG_PTHREAD_ATFORK Signed-off-by: ligd <[email protected]> bluetooth: fix bt missing header files nuttx/wqueue.h Signed-off-by: ligd <[email protected]> lib_gdbstub: fix container of Signed-off-by: buxiasen <[email protected]> Signed-off-by: ligd <[email protected]> container_of: fix compile failed cause of list.h not support container_of Signed-off-by: ligd <[email protected]> nuttx/arch:Enabling ARCH_MATH_H is required when compiling sim with the 13.2 version of the toolchain. Signed-off-by: cuiziwei <[email protected]> Signed-off-by: ligd <[email protected]> arm/stm32f401rc-rs485: Add support to WS2812 addressable LED Signed-off-by: Rodrigo Sim <[email protected]> syslog: Don't allow blocking when in signal handler Blocking while running a signal handler is not advisable, instead write the log string character by character. There is also a potential for a deadlock, as discussed in apache#6618 Note: querying for rtcb->sigdeliver is not 100% ideal, as it only tells _if_ a signal handler has been queued, not if it is running. However, it makes syslog safe / usable which is a debug feature anyhow. boards/risc-v: Remove ref to riscv_internal.h `riscv_internal.h` is a private chip level header file, and it should not be included in the board files. Signed-off-by: Huang Qi <[email protected]> boards/esp32s3: Merge MCUboot and "simple-boot" linker scripts To make it easier to keep the linker scripts updated for both MCUboot and "simple-boot", this commit merges them into a single linker script with macros to enable/disable specific sections. task_exit.c: Add missing sched_note_stop() A regression from apache#13728 ; sched_note_stop() is never called for tasks that exit normally via exit(). nuttx: Add LIBSUPCXX_TOOLCHAIN to link the prebuilt library provide by toolchain. Signed-off-by: cuiziwei <[email protected]> serial/gdbstub:Adjust serial port gdbstub Kconfig dependencies Signed-off-by: anjiahao <[email protected]> gdbstub:fix typo Signed-off-by: anjiahao <[email protected]> coredump: coredump_add_memory_region need use flags Signed-off-by: anjiahao <[email protected]> arm64: fix fvp smp faild to boot reason: we should give a busy wait addr This commit fixes the regression from apache#13640 Signed-off-by: hujun5 <[email protected]> CI: Enable sim-02 build when we create or update a Complex PR CI Build Job sim-02 was disabled to reduce our usage of GitHub Runners, to comply with ASF Policy: apache#14376 (comment) However this causes the Scheduled Merge Job to fail, due to reduced CI Checks: https://github.com/NuttX/nuttx/actions/runs/11490041505/job/31980056690#step:7:465 This PR re-enables sim-02 when we create or update a Complex PR. arch/Kconfig: remove ARCH_MATH_H if LIBCXX Because some libraries do require a full libm implementation. Signed-off-by: zhanghongyu <[email protected]> Documentation: migrate README.txt from boards and fixes for mps boards migrate some README.txt form boards/ and fixes for mps boards rst samv7: fix QSPI build Commit 313d6df caused the following build error: CC: fixedmath/lib_b16atan2.c chip/sam_qspi.c: In function 'qspi_memory': chip/sam_qspi.c:1552:7: warning: implicit declaration of function 'IS_ALIGNED' [-Wimplicit-function-declaration] 1552 | IS_ALIGNED((uintptr_t)meminfo->buffer, 4) && | ^~~~~~~~~~ In file included from chip/sam_qspi.c:41: chip/sam_qspi.c: In function 'qspi_alloc': chip/sam_qspi.c:1591:21: warning: implicit declaration of function 'ALIGN_UP' [-Wimplicit-function-declaration] 1591 | return kmm_malloc(ALIGN_UP(buflen, 4)); This was caused by missing include of nuttx.h header defining ALIGN_UP and IS_ALIGNED. Signed-off-by: Michal Lenc <[email protected]> mmcsd: SDIO_CAPS_4BIT_ONLY set buswidth MMCSD_SCR_BUSWIDTH_4BIT uint8_t buswidth:4; /* Bus widths supported (SD only) */ Signed-off-by: zhangshoukui <[email protected]> armv8m/clang.cmake: add armv8m clang config Its makefile is implemented in arch/arm/src/armv8-m/Toolchain.defs as follows: ifeq ($(CONFIG_ARM_TOOLCHAIN_CLANG),y) ifeq ($(CONFIG_ARCH_CORTEXM23),y) TOOLCHAIN_CLANG_CONFIG = armv8m.main_soft_nofp else ifeq ($(CONFIG_ARCH_CORTEXM33),y) ifeq ($(CONFIG_ARCH_FPU),y) TOOLCHAIN_CLANG_CONFIG = armv8m.main_hard_fp else TOOLCHAIN_CLANG_CONFIG = armv8m.main_soft_nofp endif else ifeq ($(CONFIG_ARCH_CORTEXM35P),y) ifeq ($(CONFIG_ARCH_FPU),y) TOOLCHAIN_CLANG_CONFIG = armv8m.main_hard_fp else TOOLCHAIN_CLANG_CONFIG = armv8m.main_soft_nofp endif else ifeq ($(CONFIG_ARCH_CORTEXM55),y) ifeq ($(CONFIG_ARCH_FPU),y) TOOLCHAIN_CLANG_CONFIG = armv8.1m.main_hard_fp else TOOLCHAIN_CLANG_CONFIG = armv8.1m.main_soft_nofp_nomve endif else ifeq ($(CONFIG_ARCH_CORTEXM85),y) ifeq ($(CONFIG_ARCH_FPU),y) TOOLCHAIN_CLANG_CONFIG = armv8.1m.main_hard_fp else TOOLCHAIN_CLANG_CONFIG = armv8.1m.main_soft_nofp_nomve endif endif Signed-off-by: wangmingrong1 <[email protected]> Writing documentation related to SPI slave. Fix build issues Fix xtensa build error with choice LIBSUPCXX by default. Signed-off-by: cuiziwei <[email protected]> sim/cmake: compatible when nuttx COMPILE_OPTIONS is not set yet Signed-off-by: buxiasen <[email protected]> Fix cdcncm printf formatter compiler warning esp32s3: Increase the init task stask size when using NSH After recent changes on nuttx-apps (not limited to, but related to nuttx-apps#2738, for instance), the stack usage for the NSH task increased, causing stack overflows under specific situations (when running `ps` command, for instance). This commit increases the init task stack size to avoid it. Please note that, even before these changes, the stack usage of the NSH task was around 90% and, then, increasing the stack size of it was recommended. kconfig: Add link parameters that can print remaining memory information LD: nuttx Memory region Used Size Region Size %age Used flash: 284272 B 512 KB 54.22% sram1: 13296 B 2 MB 0.63% sram2: 0 GB 2 MB 0.00% CP: nuttx.hex CP: nuttx.bin Signed-off-by: wangmingrong1 <[email protected]> Fixed selection of irq file. Added flash.ld script to nucleo-h563zi/scripts folder. Changed Make.defs to use it. Minor change to Kconfig regarding flash configurations. Various changes Fix include guards.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Impact
syslog_write
Testing
Pass CI