https://sourceware.org/bugzilla/show_bug.cgi?id=34544
--- Comment #3 from Jaipaul Cheernam <jaipaul.cheernam at est dot tech> ---
The testjob_launch fix in 0.197 doesn't fully resolve this.
run-stackprof-system.sh still fails intermittently on the Yocto autobuilder
(qemu on virtualized CI workers), reproduced on both qemux86-64 and qemuarm64:
Failed ptests: {'elfutils': ['run-stackprof-system.sh']}
The background loop is running and perf works, but the plain -v run still
summarizes 0 samples in its window, so the grep -E '^[0-9]+ ' assertion fails:
Starting stack profile collection systemwide
find_dwfl newly seen pid 13028 (sh): ...
+ grep -E '^[0-9]+ ' test.out <- no match, no per-PID summary
FAIL: run-stackprof-system.sh
Meanwhile, in the same run, the other variants collected samples and passed
(e.g. gprof: received 9976 samples; arm64 user: received 1983 samples). So it's
not a capability issue — the plain systemwide run with
its shorter timeout 10/--kill-after=5 just intermittently loses the race,
likely worsened by exclude_kernel=1 plus CPU contention starving the single
busy-loop.
Suggestion: treat "0 samples in the window" as a SKIP (exit 77), consistent
with the existing prerequisite skips, rather than a hard FAIL — and/or give the
plain run the same longer timeout as the gprof run.
Logs:
- qemux86-64:
https://valkyrie.yocto.io/pub/non-release/20261027-47/testresults/qemux86-64-ptest/elfutils.log
(https://valkyrie.yocto.io/pub/non-release/20261027-47/testresults/qemux86-64-ptest/elfutils.log)
- qemuarm64:
https://valkyrie.yocto.io/pub/non-release/20261029-77/testresults/qemuarm64-ptest/elfutils.log
(https://valkyrie.yocto.io/pub/non-release/20261029-77/testresults/qemuarm64-ptest/elfutils.log)
--
You are receiving this mail because:
You are on the CC list for the bug.
