-
Notifications
You must be signed in to change notification settings - Fork 4.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[RISC-V] Set the HAVE_FUNCTIONAL_PTHREAD_ROBUST_MUTEXES_EXITCOD value to 0 #92324
Conversation
@@ -170,8 +170,6 @@ else() | |||
message(FATAL_ERROR "Unsupported platform. OS: ${CMAKE_SYSTEM_NAME}, arch: ${TARGET_ARCH_NAME}") | |||
endif() | |||
|
|||
if(TARGET_ARCH_NAME MATCHES "^(x86|x64|s390x|armv6|loongarch64|ppc64le)$") | |||
if(TARGET_ARCH_NAME MATCHES "^(x86|x64|s390x|armv6|loongarch64|riscv64|ppc64le)$") | |||
set_cache_value(HAVE_FUNCTIONAL_PTHREAD_ROBUST_MUTEXES_EXITCODE 0) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@kouvel Do we still need HAVE_FUNCTIONAL_PTHREAD_ROBUST_MUTEXES
? It does not look right that it is architecture specific.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It was introduced due to a couple of issues, one in Alpine Linux and one in glibc on ARM/ARM64, but I believe that both have since been fixed. I'm not sure why the cache value is being set based on architecture, it should probably be checked based on the target OS, which is currently done in mutex.hpp
in the PAL.
It was noted in #82380 (comment) that a PAL test for named mutexes fails when pthread robust mutexes are used on RISC-V, so I wonder if it actually works. @JongHeonChoi, have you checked if the corresponding test case in
runtime/src/coreclr/pal/src/configure.cmake
Lines 1012 to 1199 in e235aef
if(NOT CLR_CMAKE_HOST_ARCH_ARM AND NOT CLR_CMAKE_HOST_ARCH_ARM64) | |
set(CMAKE_REQUIRED_LIBRARIES pthread) | |
check_cxx_source_runs(" | |
// This test case verifies the pthread process-shared robust mutex's cross-process abandon detection. The parent process starts | |
// a child process that locks the mutex, the process process then waits to acquire the lock, and the child process abandons the | |
// mutex by exiting the process while holding the lock. The parent process should then be released from its wait, be assigned | |
// ownership of the lock, and be notified that the mutex was abandoned. | |
#include <sys/mman.h> | |
#include <sys/time.h> | |
#include <errno.h> | |
#include <pthread.h> | |
#include <stdio.h> | |
#include <unistd.h> | |
#include <new> | |
using namespace std; | |
struct Shm | |
{ | |
pthread_mutex_t syncMutex; | |
pthread_cond_t syncCondition; | |
pthread_mutex_t robustMutex; | |
int conditionValue; | |
Shm() : conditionValue(0) | |
{ | |
} | |
} *shm; | |
int GetFailTimeoutTime(struct timespec *timeoutTimeRef) | |
{ | |
int getTimeResult = clock_gettime(CLOCK_REALTIME, timeoutTimeRef); | |
if (getTimeResult != 0) | |
{ | |
struct timeval tv; | |
getTimeResult = gettimeofday(&tv, NULL); | |
if (getTimeResult != 0) | |
return 1; | |
timeoutTimeRef->tv_sec = tv.tv_sec; | |
timeoutTimeRef->tv_nsec = tv.tv_usec * 1000; | |
} | |
timeoutTimeRef->tv_sec += 30; | |
return 0; | |
} | |
int WaitForConditionValue(int desiredConditionValue) | |
{ | |
struct timespec timeoutTime; | |
if (GetFailTimeoutTime(&timeoutTime) != 0) | |
return 1; | |
if (pthread_mutex_timedlock(&shm->syncMutex, &timeoutTime) != 0) | |
return 1; | |
if (shm->conditionValue != desiredConditionValue) | |
{ | |
if (GetFailTimeoutTime(&timeoutTime) != 0) | |
return 1; | |
if (pthread_cond_timedwait(&shm->syncCondition, &shm->syncMutex, &timeoutTime) != 0) | |
return 1; | |
if (shm->conditionValue != desiredConditionValue) | |
return 1; | |
} | |
if (pthread_mutex_unlock(&shm->syncMutex) != 0) | |
return 1; | |
return 0; | |
} | |
int SetConditionValue(int newConditionValue) | |
{ | |
struct timespec timeoutTime; | |
if (GetFailTimeoutTime(&timeoutTime) != 0) | |
return 1; | |
if (pthread_mutex_timedlock(&shm->syncMutex, &timeoutTime) != 0) | |
return 1; | |
shm->conditionValue = newConditionValue; | |
if (pthread_cond_signal(&shm->syncCondition) != 0) | |
return 1; | |
if (pthread_mutex_unlock(&shm->syncMutex) != 0) | |
return 1; | |
return 0; | |
} | |
void DoTest_Child(); | |
int DoTest() | |
{ | |
// Map some shared memory | |
void *shmBuffer = mmap(NULL, 4096, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_SHARED, -1, 0); | |
if (shmBuffer == MAP_FAILED) | |
return 1; | |
shm = new(shmBuffer) Shm; | |
// Create sync mutex | |
pthread_mutexattr_t syncMutexAttributes; | |
if (pthread_mutexattr_init(&syncMutexAttributes) != 0) | |
return 1; | |
if (pthread_mutexattr_setpshared(&syncMutexAttributes, PTHREAD_PROCESS_SHARED) != 0) | |
return 1; | |
if (pthread_mutex_init(&shm->syncMutex, &syncMutexAttributes) != 0) | |
return 1; | |
if (pthread_mutexattr_destroy(&syncMutexAttributes) != 0) | |
return 1; | |
// Create sync condition | |
pthread_condattr_t syncConditionAttributes; | |
if (pthread_condattr_init(&syncConditionAttributes) != 0) | |
return 1; | |
if (pthread_condattr_setpshared(&syncConditionAttributes, PTHREAD_PROCESS_SHARED) != 0) | |
return 1; | |
if (pthread_cond_init(&shm->syncCondition, &syncConditionAttributes) != 0) | |
return 1; | |
if (pthread_condattr_destroy(&syncConditionAttributes) != 0) | |
return 1; | |
// Create the robust mutex that will be tested | |
pthread_mutexattr_t robustMutexAttributes; | |
if (pthread_mutexattr_init(&robustMutexAttributes) != 0) | |
return 1; | |
if (pthread_mutexattr_setpshared(&robustMutexAttributes, PTHREAD_PROCESS_SHARED) != 0) | |
return 1; | |
if (pthread_mutexattr_setrobust(&robustMutexAttributes, PTHREAD_MUTEX_ROBUST) != 0) | |
return 1; | |
if (pthread_mutex_init(&shm->robustMutex, &robustMutexAttributes) != 0) | |
return 1; | |
if (pthread_mutexattr_destroy(&robustMutexAttributes) != 0) | |
return 1; | |
// Start child test process | |
int error = fork(); | |
if (error == -1) | |
return 1; | |
if (error == 0) | |
{ | |
DoTest_Child(); | |
return -1; | |
} | |
// Wait for child to take a lock | |
WaitForConditionValue(1); | |
// Wait to try to take a lock. Meanwhile, child abandons the robust mutex. | |
struct timespec timeoutTime; | |
if (GetFailTimeoutTime(&timeoutTime) != 0) | |
return 1; | |
error = pthread_mutex_timedlock(&shm->robustMutex, &timeoutTime); | |
if (error != EOWNERDEAD) // expect to be notified that the robust mutex was abandoned | |
return 1; | |
if (pthread_mutex_consistent(&shm->robustMutex) != 0) | |
return 1; | |
if (pthread_mutex_unlock(&shm->robustMutex) != 0) | |
return 1; | |
if (pthread_mutex_destroy(&shm->robustMutex) != 0) | |
return 1; | |
return 0; | |
} | |
void DoTest_Child() | |
{ | |
// Lock the robust mutex | |
struct timespec timeoutTime; | |
if (GetFailTimeoutTime(&timeoutTime) != 0) | |
return; | |
if (pthread_mutex_timedlock(&shm->robustMutex, &timeoutTime) != 0) | |
return; | |
// Notify parent that robust mutex is locked | |
if (SetConditionValue(1) != 0) | |
return; | |
// Wait a short period to let the parent block on waiting for a lock | |
sleep(1); | |
// Abandon the mutex by exiting the process while holding the lock. Parent's wait should be released by EOWNERDEAD. | |
} | |
int main() | |
{ | |
int result = DoTest(); | |
return result >= 0 ? result : 0; | |
}" HAVE_FUNCTIONAL_PTHREAD_ROBUST_MUTEXES) | |
set(CMAKE_REQUIRED_LIBRARIES) | |
endif() |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@kouvel Yes. I checked the return value(result >= 0 ? result : 0;
) of the test case in VisionFive2 target.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sounds good, thanks! I'll look into whether this check can be removed, as I'm not aware of any current cases where it would fail.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. I've added pal tests results with this change too in #75749 (comment).
The
HAVE_FUNCTIONAL_PTHREAD_ROBUST_MUTEXES_EXITCODE
value is also set to 0 in theVisionFive2
target ofRISCV64
.cc @wscho77 @HJLeee @JongHeonChoi @t-mustafin @alpencolt @gbalykov @clamp03 @sirntar @yurai007