powerpc/tm:tm-resched-dscr in ubuntu_kernel_selftests flaky on P8 node gulpin

[Po-Hsu Lin]

Issue found with Focal hwe-5.15.0-66.73~20.04.1 on Power8 node "gulpin".

Test failed with:
  test: tm_resched_dscr
  tags: git_version:bfd31f0-dirty
  Binding to cpu 8
  main test running as pid 2542
  Check DSCR TM context switch:
  !! killing tm_resched_dscr
  !! child died by signal 15
  failure: tm_resched_dscr

Nothing interesting in dmesg / syslog:
$ dmesg | tail
[ 20.556968] IPv6: ADDRCONF(NETDEV_CHANGE): enP34p1s0f0: link becomes ready
[ 20.572182] tg3 0022:01:00.2 enP34p1s0f2: Link is up at 1000 Mbps, full duplex
[ 20.572187] tg3 0022:01:00.2 enP34p1s0f2: Flow control is off for TX and off for RX
[ 20.572191] tg3 0022:01:00.2 enP34p1s0f2: EEE is disabled
[ 20.572201] IPv6: ADDRCONF(NETDEV_CHANGE): enP34p1s0f2: link becomes ready
[ 24.598633] kauditd_printk_skb: 19 callbacks suppressed
[ 24.598640] audit: type=1400 audit(1676957767.768:31): apparmor="DENIED" operation="open" profile="/usr/sbin/ntpd" name="/snap/bin/" pid=2171 comm="ntpd" requested_mask="r" denied_mask="r" fsuid=0 ouid=0
[ 27.682671] loop3: detected capacity change from 0 to 8
[ 28.618195] fbcon: Taking over console
[ 28.747706] Console: switching to colour frame buffer device 128x48

This issue does not exist with 5.15.0-60.66~20.04.1
  test: tm_resched_dscr
  tags: git_version:bfd31f0-dirty
  Binding to cpu 8
  main test running as pid 28806
  Check DSCR TM context switch: OK
  success: tm_resched_dscr

We didn't catch this issue in Jammy 5.15 as Power8 was not supported there, and this test was skipped on the Power9 node that got tested with Jammy (looks like because of the lack of PPC_FEATURE2_HTM)

[Ubuntu Kernel Bot (ubuntu-kernel-bot)]

This bug is missing log files that will aid in diagnosing the problem. While running an Ubuntu kernel (not a mainline or third-party kernel) please enter the following command in a terminal window:

apport-collect 2007908

and then change the status of the bug to 'Confirmed'.

If, due to the nature of the issue you have encountered, you are unable to run this command, please add a comment stating that fact and change the bug status to 'Confirmed'.

This change has been made by an automated script, maintained by the Ubuntu Kernel Team.

[Po-Hsu Lin (cypressyew)]

stress tested with the following command on node gulpin with -66 shows this test is a bit flaky:

failcnt=0
for i in `seq 1 100`; do
   time sudo ./tm-resched-dscr
   if [ $? -ne 0 ]; then
        let "failcnt=$failcnt+1"
   fi
done
echo $failcnt

There are 27 failures in 100 attempts.

Next is to stress test with -60

[Po-Hsu Lin (cypressyew)]

Here is the test output with -66.

[Po-Hsu Lin (cypressyew)]

stress tested with 5.15.0-60-generic, there are 20 failures reported in 100 attempts.

So it looks like a flaky test instead of a regression. I was too lucky to get multiple failures in a row.