[PATCH] D129089: [BOLT] Fix concurrent hash table modification in the instrumentation runtime

[Vladislav Khmelevsky via Phabricator via llvm-commits]

yota9 added a comment.

@michoecho 
Oh, you are totally right about the locks, I've briefly checked that moment and made a wrong assumption, thank you!
As for tryLocks I'm totally agree with you that basically it should have the same behaviour as lock. But I don't see how __bolt_instr_data_dump could be called from signal handler. Basically we've got 2 situation when this function is called - when DT_FINI is called or (my situation) when we've got forked process waiting for the parent to be dead. The DT_FINI (probably) never get called from the signal handler and our fork doesn't have signal handlers. 
Plus when the deadlock occurs the parent process is already normally exited at that moment, so no deadlock should be at that moment.. So I'm still confused why it happens sometimes.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D129089/new/

https://reviews.llvm.org/D129089