testwsdeque failure on POWER8

On a POWER8 testwsdeque fails occasionally under load:

Wrong exit code for testwsdeque(threaded1)(expected 0 , actual 134 )
Stderr ( testwsdeque ):
internal error: FAIL: 277808608 3 13
    (GHC version 8.3.20170430 for powerpc64le_unknown_linux)

Trac metadata

Trac field	Value
Version	8.1
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Runtime System
Test case
Differential revisions
BlockedBy
Related
Blocking
CC	erikd, hvr, simonmar
Operating system
Architecture

changed weight to 5

added RTS Tbug Trac import labels

This is bad and suggests that either we're using the wrong barriers in the implementation, or the implementation of the barriers is wrong for POWER8.

Replying to [ticket:13633#comment:136091 simonmar]:

This is bad and suggests that either we're using the wrong barriers in the implementation, or the implementation of the barriers is wrong for POWER8.

I looked at the implementation of the barriers and cas() in SMP.hagain and they are correct for POWER and PowerPC.

Test testwsdeque always fails on POWER9 under load.

I'm experiencing nondeterministic CI failures on validate-x86_64-linux-deb9-unreg due to testwsdeque unexpectedly //passing//, such as in this run:

*** unexpected pass for testwsdeque(threaded1)

<elided>

Unexpected results from:
TEST="testwsdeque"

SUMMARY for test run started at Sun Feb 24 09:28:06 2019 UTC
 0:42:51 spent to go through
    6844 total tests, which gave rise to
   21127 test cases, of which
   14004 were skipped

      38 had missing libraries
    6984 expected passes
     100 expected failures

       0 caused framework failures
       0 caused framework warnings
       1 unexpected passes
       0 unexpected failures
       0 unexpected stat failures

Unexpected passes:
   rts/testwsdeque.run  testwsdeque [unexpected] (threaded1)

added CI breakage label

assigned to @trommler

I have a patch on gitlab at https://gitlab.haskell.org/ghc/ghc/commits/wip/T13633. It seems to fix POWER9. I ran the testsuite twice and did not see it fail again.

I ran the test about a dozen times on a POWER9 and did not see any more failures.

Replying to [ticket:13633#comment:168989 RyanGlScott]:

I'm experiencing nondeterministic CI failures on validate-x86_64-linux-deb9-unreg due to testwsdeque unexpectedly //passing//, such as in this run:

I'm rather puzzled by this. This test isn't marked as expect_broken as far as I can tell. Why is the testsuite driver expecting it to fail?

Fixed in !445 (merged)

closed

mentioned in commit 5c084e04

added concurrency label

added runtime crash label

mentioned in commit 2df05448

mentioned in merge request !1237 (merged)

Backported to %8.8.1 in !1237 (merged).

added Pnormal label

mentioned in merge request !2646 (closed)