Subtests that measure time of resume after engine reset require results
from at least 9 reset-resume cycles for reasonable calculation of a median
value to be compared against presumed limits. On most Gen12+ platforms,
as well as on some older ones like JSL, CHV, ILK or ELK, the current limit
of 5 seconds for collecting those results occurs too short. Moreover,
upstream CI results from kms, reset-stress and unwedge-stress subtests
show GPU resume after reset noticeably slower when running on some Gen12+
platforms.
Skip if less than 9 results from reset-resume cycles are collected. Raise
the 5 seconds loop limit to 20 seconds, and break the loop as soon as more
than 8 results are collected. Having the test results revisited with the
above in place, relax expectations on resume-after-reset time for Gen12+
similar to Gen5-. Also, limit the scope of exercised engines to physical,
skipping virtual, to avoid exceeding per test timeout applied by
igt_runner, and limit the amount of debug info on stderr to avoid
exceeding per test disk space limit.
v2: Skip, not warn, on too few reset-resume cycles, may be handled as
expected skip on extremely slow platforms (Krzysztof),
- split out a change in handling of not enough measurements to a
separate patch (Kamil),
- reword commit message to be more distinct from other patches in
series (Kamil),
- reword commit message and description so they no longer state the
scope of the issue is limited to Gen12+, and list other (non-Gen12+)
platforms found also affected,
- disable massive engine debugging info on stderr to avoid incompletes
on disk space limit exceeded.
Janusz Krzysztofik (5):
tests/gem_eio: Skip on too few reset-resume cycles
tests/gem_eio: Try harder to measure median resume time
tests/gem_eio: Adjust for slow resume after reset on Gen12+
tests/gem_eio: Iterate over physical rings in stress subtests
tests/gem_eio: Avoid exceeding CI disk space limit
tests/intel/gem_eio.c | 47 ++++++++++++++++++++++++-------------------
1 file changed, 26 insertions(+), 21 deletions(-)
--
2.51.0
