Copious TSAN warnings on Linux #14710
Labels
Comments
Example output: |
|
V1 review: fixes in master for this are bug fixes and would be acceptable, however this is not a V1 blocker. |
|
This might be related: https://gitlab.gnome.org/GNOME/glib/-/issues/1672 |
|
I've done some tests of this issue and it seems that on Ubuntu 22.04 TSAN has been fixed - it no longer reports such false positives. One can use provided simple test code to check that with different TSAN versions. Compile the test code as follows: test.cpp#include <iostream>
#include <thread>
#include <glib.h>
struct data {
int value;
};
static struct data *sharedData1;
static struct data *sharedData2;
void glibMainLoopThread(GMainLoop *loop) {
g_main_loop_run(loop);
}
gboolean glibCallback1(void * data) {
auto *sharedData = reinterpret_cast<struct data *>(data);
std::cout << "T: sharedData1 = " << sharedData->value << std::endl;
sharedData->value++;
return FALSE;
}
gboolean glibCallback2(void * data) {
auto *sharedData = reinterpret_cast<struct data *>(data);
std::cout << "T: sharedData2 = " << sharedData->value << std::endl;
sharedData->value++;
return FALSE;
}
int main() {
sharedData1 = new struct data({10});
GMainLoop *loop = g_main_loop_new(NULL, FALSE);
std::thread t(glibMainLoopThread, loop);
sharedData2 = new struct data({20});
// That's not a data race
// NOTE: TSAN on Ubuntu 20.04 reports it (false positive)
std::cout << "sharedData1 = " << sharedData1->value << std::endl;
std::cout << "sharedData2 = " << sharedData2->value << std::endl;
g_idle_add(glibCallback1, sharedData1);
g_idle_add(glibCallback2, sharedData2);
// Here we've got a data race (callback is already added to main loop thread)
// std::cout << "sharedData2 = " << sharedData2->value << std::endl;
std::this_thread::sleep_for(std::chrono::seconds(1));
g_main_loop_quit(loop);
t.join();
std::cout << "sharedData1 = " << sharedData1->value << std::endl;
std::cout << "sharedData2 = " << sharedData2->value << std::endl;
delete sharedData1;
delete sharedData2;
return 0;
}On Ubuntu 22.04 I've got no errors (unless commented Output from Ubuntu 22.04: With our CI the problem occurs because as for now the |
|
Issue Scrub: @woody-apple to pin us to ubuntu-22.04 in CI. @andy31415 will help check to make sure warnings aren't being suppressed afterwards. |
woody-apple
added
v1.1
build issue
and removed
V1.0
platform
linux
p1
|
After more research I've found the TRUE cause of false positives. The proper answer is here: https://stackoverflow.com/a/74328089 |
|
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Problem
We have lots and lots of TSAN data race warnings on Linux that should be addressed before it results in other potentially bigger blow-ups that are more difficult to debug later.
Example: https://github.com/project-chip/connectedhomeip/runs/5039936950?check_suite_focus=true
The text was updated successfully, but these errors were encountered: