Deadlock in forkProcess
As part of `forkProcess` we discard all tasks except the one task that remains. As part of this we call `freeTask`, and `freeTask` in turn calls `closeCondition` and `closeMutex`, which are very thin wrappers around `pthread_cond_destroy` and `phread_mutex_destroy`. However, the behaviour of these functions is undefined when there are currently threads blocked on these condition variables/mutexes. In reality, this undefined behaviour often (though not always) results in a deadlock. Unfortunately, I don't have a minimal test case to demonstrate this, but in the large system on which I am testing this I am seeing these deadlocks rather frequently.
For the global mutex we don't attempt to call `pthread_cond_destroy` or `pthread_mutex_destroy`, but instead re-initialize them. The patch
https://phabricator.haskell.org/D59
does precisely that for the condition variables and mutexes associated with tasks. This is *somewhat* or a long way around, because we will then subsequently still call `pthread_..._destroy`, but it means that we don't have to mess with `freeTask`.
<details><summary>Trac metadata</summary>
| Trac field | Value |
| ---------------------- | ------------ |
| Version | 7.8.2 |
| Type | Bug |
| TypeOfFailure | OtherFailure |
| Priority | normal |
| Resolution | Unresolved |
| Component | Compiler |
| Test case | |
| Differential revisions | |
| BlockedBy | |
| Related | |
| Blocking | |
| CC | |
| Operating system | |
| Architecture | |
</details>
<!-- {"blocked_by":[],"summary":"Deadlock in forkProcess","status":"New","operating_system":"","component":"Compiler","related":[],"milestone":"","resolution":"Unresolved","owner":{"tag":"Unowned"},"version":"7.8.2","keywords":[],"differentials":[],"test_case":"","architecture":"","cc":[""],"type":"Bug","description":"As part of `forkProcess` we discard all tasks except the one task that remains. As part of this we call `freeTask`, and `freeTask` in turn calls `closeCondition` and `closeMutex`, which are very thin wrappers around `pthread_cond_destroy` and `phread_mutex_destroy`. However, the behaviour of these functions is undefined when there are currently threads blocked on these condition variables/mutexes. In reality, this undefined behaviour often (though not always) results in a deadlock. Unfortunately, I don't have a minimal test case to demonstrate this, but in the large system on which I am testing this I am seeing these deadlocks rather frequently.\r\n\r\nFor the global mutex we don't attempt to call `pthread_cond_destroy` or `pthread_mutex_destroy`, but instead re-initialize them. The patch \r\n\r\nhttps://phabricator.haskell.org/D59\r\n\r\ndoes precisely that for the condition variables and mutexes associated with tasks. This is ''somewhat'' or a long way around, because we will then subsequently still call `pthread_..._destroy`, but it means that we don't have to mess with `freeTask`. ","type_of_failure":"OtherFailure","blocking":[]} -->
issue