Re: [PATCH i-g-t] tests/intel/xe_exec_threads: whitelist expected multi-queue hang dmesg warns
Kamil Konieczny <[email protected]>
| Newsgroups | org.freedesktop.lists.igt-dev |
|---|---|
| Message-ID | <[email protected]> |
Hi Himanshu, On 2026-07-21 at 18:34:05 +0530, [email protected] wrote: > From: Himanshu Girotra <[email protected]> please make subject little shorter, imho better: [PATCH i-g-t] tests/intel/xe_exec_threads: ignore multi-queue hang dmesg warns > > Multi-queue hang subtests hang one queue and reset the whole group. > Sibling jobs that never started get logged by the KMD as > "Check job timeout: ... not started" at warn level, which shows up as a > dmesg-warn in CI even though the test passes. > > These warnings are expected for multi-queue hangs (single-queue hangs > don't hit this path), so whitelist them when the subtest has both > MULTI_QUEUE and HANG set. > > Signed-off-by: Himanshu Girotra <[email protected]> > --- > tests/intel/xe_exec_threads.c | 16 ++++++++++++++++ > 1 file changed, 16 insertions(+) > > diff --git a/tests/intel/xe_exec_threads.c b/tests/intel/xe_exec_threads.c > index 0480df779..21529e863 100644 > --- a/tests/intel/xe_exec_threads.c > +++ b/tests/intel/xe_exec_threads.c > @@ -1468,6 +1468,19 @@ static void threads(int fd, int flags) > pthread_barrier_destroy(&barrier); > } > > +/* > + * Multi-queue hang subtests hang one queue of a group. The group's queues > + * are scheduled together, so sibling jobs stay queued behind the hung one and > + * never start; when the TDR resets the group, the KMD reports each unstarted > + * sibling as "Check job timeout: ... not started". These warns are expected > + * (single-queue hangs don't hit this path); whitelist them to avoid spurious > + * dmesg-warn in CI. > + */ > +static void ignore_multi_queue_hang_warnings_in_dmesg(void) > +{ > + igt_emit_ignore_dmesg_regex("Check job timeout: .*not started"); > +} > + > int igt_main() > { > const struct section { > @@ -1735,6 +1748,9 @@ int igt_main() > igt_assert(!(s->flags & BALANCER)); > } > > + if ((s->flags & MULTI_QUEUE) && (s->flags & HANG)) or just: if ((s->flags & (MULTI_QUEUE | HANG)) See also note from Priyanka. Regards, Kamil > + ignore_multi_queue_hang_warnings_in_dmesg(); > + > threads(fd, s->flags); > } > } > -- > 2.50.1 >