Open cjjdespres opened 1 year ago
Attn @mpirvu. Perhaps they're related to #15601, given the timing?
That's a fair number of failures. Do these runs have the latest changes related to localSyncCompiles? If we suspect that localSyncCompiles are at fault we can grind with "-XX:-JITServerLocalSyncCompiles" and see if the failures are still present.
No, the latest changes haven't been picked up in the nightly builds yet. I'll run some grinders to see if the problem is sensitive to LocalSyncCompiles
.
I haven't seen any failures in the nightly tests since the latest LocalSyncCompiles changes were picked up, though it's only been a day. I've also run tests of DaaLoadTest_all_special_5m_27
on ppc64le_linux with JDK17 and found that the memory corruption failures happen only when JITServerLocalSyncCompiles
is enabled, both before and after the latest changes (with a 2/50 failure rate), and not when it is disabled.
Could you please add verbose logs at the client and try to reproduce on pLinux. -Xjit:verbose={compil*},vlog=vlog.txt
It doesn't seem to be reproducible with verbose logs on, like the segfaults.
From
/job/Test_openjdk17_j9_special.system_ppc64le_linux_jit_Personal/341
/job/Test_openjdk8_j9_special.system_ppc64le_linux_jit_Personal/421
/job/Test_openjdk17_j9_special.system_x86-64_linux_jit_Personal/356
/job/Test_openjdk8_j9_special.system_ppc64le_linux_jit_Personal/420
/job/Test_openjdk17_j9_special.system_ppc64le_linux_jit_Personal/340
/job/Test_openjdk11_j9_special.system_ppc64le_linux_jit_Personal/421
failures in
DaaLoadTest_all_special_5m_27
DaaLoadTest_daa1_special_5m_26
LambdaLoadTest_special_J9_5m_25
MathLoadTest_all_special_5m_25
MathLoadTest_all_special_5m_27
MathLoadTest_autosimd_special_5m_23
MathLoadTest_bigdecimal_special_5m_25
MathLoadTest_bigdecimal_special_5m_26
MauveMultiThrdLoad_special_5m_13
MauveMultiThrdLoad_special_5m_24
MauveMultiThrdLoad_special_5m_25
All of them are memory corruption/malloc errors. A few console logs: