https://bugs.llvm.org/show_bug.cgi?id=43813
Bug ID: 43813
Summary: Insufficient locking in
std::atomic_compare_exchange_strong for
std::shared_ptr
Product: libc++
Version: 9.0
Hardware: All
OS: All
Status: NEW
Severity: enhancement
Priority: P
Component: All Bugs
Assignee: unassignedclangb...@nondot.org
Reporter: lahav.sch.l...@gmail.com
CC: llvm-bugs@lists.llvm.org, mclow.li...@gmail.com
Hi,
The implementation of 'std::atomic_compare_exchange_strong(shared_ptr* p,
shared_ptr* v, shared_ptr w)' only locks the mutex related to 'p', but not the
one for 'v' as well.
This means the following code (Which should have no race conditions):
#include <memory>
#include <atomic>
#include <unistd.h>
#include <pthread.h>
std::shared_ptr<int> g_p;
static void* thread_run(void*) {
auto w = std::make_shared<int>();
while (true) {
auto p = std::make_shared<int>();
std::atomic_compare_exchange_strong(&p, &g_p, w);
}
return nullptr;
}
int main() {
pthread_t tid;
pthread_create(&tid, nullptr, thread_run, nullptr);
auto w = std::make_shared<int>();
while(true) {
std::atomic_store(&g_p, w);
}
}
may attempt to access the same 'g_p' instance from multiple threads - The main
thread (Via std::atomic_store()) will properly lock the matching mutex, but the
2nd thread will not, causing a race condition.
I believe the mutex for 'v' should be locked as well, with special care as to
not introduce a deadlock now that 2 mutexes are being locked.
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
