[GHC] #14310: Assertion triggered by STM invariant.

[GHC]

#14310: Assertion triggered by STM invariant.
----------------------------------+--------------------------------------
        Reporter:  mbw            |                Owner:  (none)
            Type:  bug            |               Status:  new
        Priority:  high           |            Milestone:  8.4.1
       Component:  Compiler       |              Version:  8.2.1
      Resolution:                 |             Keywords:
Operating System:  Linux          |         Architecture:  x86_64 (amd64)
 Type of failure:  Runtime crash  |            Test Case:
      Blocked By:                 |             Blocking:
 Related Tickets:                 |  Differential Rev(s):
       Wiki Page:                 |
----------------------------------+--------------------------------------

Comment (by bgamari):

 So I think I see one issue: in the event that
 `validate_and_acquire_ownership(cap, trec, (!use_read_phase), true);` in
 `stmCommitTransaction` returns `false` we will never unlock the invariants
 that we previously locked.

 I suspect we'll want something like this,
 {{{#!patch
 diff --git a/rts/STM.c b/rts/STM.c
 index 5c8fd4ff40..02ac22519b 100644
 --- a/rts/STM.c
 +++ b/rts/STM.c
 @@ -1441,6 +1441,17 @@ StgBool stmCommitTransaction(Capability *cap,
 StgTRecHeader *trec) {
      } else {
          revert_ownership(cap, trec, false);
      }
 +  } else {
 +    // we encountered inconsistency; unlock the invariants that we locked
 as we
 +    // are giving up on committing.
 +    if (touched_invariants) {
 +        StgInvariantCheckQueue *q = trec -> invariants_to_check;
 +        while (q != END_INVARIANT_CHECK_QUEUE) {
 +            StgAtomicInvariant *inv = q -> invariant;
 +            unlock_inv(inv);
 +            q = q -> next_queue_entry;
 +        }
 +    }
    }

    unlock_stm(trec);
 }}}

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14310#comment:10>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler