<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/134344>134344</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            [TSAN] Threadsanitizer false negative
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            false-negative
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          jaidevshastri
      </td>
    </tr>
</table>

<pre>
    Hi, 

TSAN reports no race when enabled with `-fsanitize=thread` on the test program below. This behavior has been observed with precompiled llvm+clang binaries (v16) and compiled llvm project binaries (v16.0.6).

```
#include <stdio.h>
#include <pthread.h>

pthread_mutex_t l;
int x = 1;

void* writer1(void* _) {
    pthread_mutex_lock(&l);              
    x = 0; //Wx1
    pthread_mutex_unlock(&l); 
    return NULL;
}
void* writer2(void* _) {
    pthread_mutex_lock(&l);             
 pthread_mutex_unlock(&l);
    x = 1; //Wx2
    return NULL;
}

int main() {
    pthread_t thr[2];
    pthread_mutex_init(&l, NULL);

 pthread_create(&thr[0], NULL, writer1, NULL);
    pthread_create(&thr[1], NULL, writer2, NULL);

    pthread_join(thr[0], NULL);
 pthread_join(thr[1], NULL);

    pthread_mutex_destroy(&l);
    return 0;
}
```
When writer1 acquires the lock `l` before writer2, the epoch information seems to be transferred such that the happens-before (HB) analysis on the shared memory `x` returns true: Wx1 HB Wx2. TSAN thus fails to assert the race between Wx1 and Wx2.
When writer2 acquires the lock `l` before writer1, the happens-before analysis correctly asserts the lack of ordering between Wx1 and Wx2, thereby reporting race.

Could the above behavior be clarified? Specifically, I would like to know if this is a limitation of pure HB analysis or a limitation/bug of TSAN.

I am also attaching my understanding of the two scheduling sequences for the test program and the vector clock information for reference.

[Tsan-norace-seq.pdf](https://github.com/user-attachments/files/19598094/Tsan-norace-seq.pdf)
[Tsan-race-sequence.pdf](https://github.com/user-attachments/files/19598095/Tsan-race-sequence.pdf)

Any information on this issue would be much appreciated. 

Thanks!
Jaidev
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJysVlGP2joT_TXmZQRynGSBBx7C7ofaT1Vfulf7WE2cCXHXsVPbgeX--isnYRcK6r0PBaQIe-bMmZn4eNB7tTdEG5ZvWf40wz401m1-oKro4Bv0walZaavT5pNi4hEYLxgvnr8VX8FRZ13wYCw4lATHhgyQwVJTBUcVGmAPfF57NCqov4mlT6FxhBV74GANhIYgkA_QObt32EJJ2h4X8NwoDyU1eFDWQYPxDxmwpSd3OCN3jqRtOxVDaX1omdhKjWYPpTLoFHlgYnVIHphYA5oKroxjxB8kw6_GC76IDosxR_bApx8vmEiVkbqvCFj66EOl7KJh6f9utroxxfdNXkwr39s-0Nv3AJqlW8YLZQK8AUufIBkXGC8OVlVMFHB0KpBLIqlp5XvMgy2jHQDANaa28pWJFRMPmok1S7dw9Zl8xmA8bjOxY2L38pbchevNLeBk6Cj0zsDXv758mUgvn254iz_DO7r8K7Gr3JLL3MTvKU8daFGZAfAeywChcSzfCpY_fcS65qSMCu-MHscwZ2aXGUhHGGi0HFF5RH13efxo-i8olzFvUZK7KOIulwugH3bI-x6Tc9i7psk907uFqcgHZ0_3ujW1hF_14-KwvUQdmcoBKH_2ypEf5CL2P4qKjhJSUm0dXWYcTaizsgFlautaDMoa8ESth2ChJAgOja_JOarA97KB0GAY_BrsOjJ-PqEysfq0HcUD9ckrf1Ys32B0bqm17hS5vEUuY0oeguuJpQW8vCXwaQsvb2IBg1aGpvdQo9IDE_Se3Bh3UM6SwjFqXHSLahX9rgsh_mshknMhfknoPQ9pnSMZ9GliMQGifAVbg3UVORV19JbShOyoPE3aHw1jApNiPtpeVwMclvZAHyJeEkiNTtWKKpbu4FtHUtVKotaniPoZjoOrVq8U6_Nq7BFUDSFeBcoDglatCmM_bQ1d7yjW96M57sqGiV3Z76NlLP7E7jNgC6i9BQwBZRPJtyfoTUXOBzRVXLD1eC8dLXjZUNXruOrpZ09GkofautuLK9YnLh5IButADt25fAWjl6OaXAQ5Xy_59tmjmRsbKzj39HPRVfVwvlZNCJ1naTGK2V6Fpi8X0rZM7HpPbj4m0JIJnoldrTTFZ7LO1yu-zpjY3UMW64uo550hrT8UOD8HvgUfY_OiMKerwgyHauix72l6CUqCNh5N7OIlrzBQtTiPHQ2aV89EvLj-P0wos2qTVut0jTPaJMssXWYJX2azZiOSjBNWeZo9ZJxzKnOJlIjliotSLNc4UxvBRc6z-E2TJFvIpOIJXyervM7XWCHLOLWo9CKODAvr9rOB5SZJszTLZhpL0n4Ym4SoUXuaG9pjUAdiIl4ZM7eJnvOy33uWca188B9YQQU9DF3xDWX5EzwP6nkelhwMkHCGnPVOb37TnXEGGh7zabphYjcQHjo0cj5sxD8BAAD__4y1JeE">