[CI] igt@pm_rps@* - fail - Test assertion failure function waitboost - Failed assertion: *_freqs[CUR] [==|<] *_freqs[*]

Submitted by Martin Peres @mupuf

Assigned to Intel GFX Bugs mailing list

Link to original bug (#102250)

Description

The test igt@pm_rps@reset hits the following assertion on our haswell system:

(pm_rps:2652) CRITICAL: Test assertion failure function waitboost, file pm_rps.c:614:
(pm_rps:2652) CRITICAL: Failed assertion: boost_freqs[CUR] == boost_freqs[MAX]
(pm_rps:2652) CRITICAL: error: 1300 != 1250

Full logs: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_2968/shard-hsw2/igt@pm_rps@reset.html

added CI feature: power/Other platform: BSW/CHT platform: BXT platform: BYT platform: CFL platform: CNL platform: GLK platform: HSW platform: KBL platform: SKL priority::medium severity::normal + 1 deleted label

Chris Wilson @ickle said:

The test assumes that the hw cannot overrule our requests, and that we sample during the boost (which itself is deferred to a task). It's not even a sensible test, it says that we must always boost a wait -- but we may wish to change that policy.

Chris Wilson @ickle closed a related bug:

*** Bug 102248 has been marked as a duplicate of this bug. ***

Martin Peres @mupuf said:

Also seen on Sandybridge.

So far, the failure rate on both HSW and SNB is close to 100% for all tests (SNB results will appear in an hour): https://intel-gfx-ci.01.org/cibuglog/index.html%3Faction_failures_history=204.html

Jani Saarinen @jani.saarinen said:

Fixed an dmerged in IGT.
https://patchwork.freedesktop.org/series/28966/

https://cgit.freedesktop.org/xorg/app/intel-gpu-tools/commit/?id=5ce65a9a51f17e0183e3e4f8943981ee7b96cadd

Jani Saarinen @jani.saarinen said:

Latest 2 shards show green:
https://intel-gfx-ci.01.org/tree/drm-tip/shards-all.html

Martin Peres @mupuf said:

Been good for > 20 runs. Closing.

Marta Löfstedt @marta said:

Now on APL- and KBL-shards:

(pm_rps:1459) CRITICAL: Test assertion failure function waitboost, file pm_rps.c:644:
(pm_rps:1459) CRITICAL: Failed assertion: boost_freqs[CUR] == boost_freqs[BOOST]
(pm_rps:1459) CRITICAL: error: 250 != 750
Subtest waitboost failed.

(pm_rps:1458) CRITICAL: Test assertion failure function waitboost, file pm_rps.c:645:
(pm_rps:1458) CRITICAL: Failed assertion: post_freqs[CUR] < post_freqs[MAX]
(pm_rps:1458) CRITICAL: error: 950 >= 950
Subtest waitboost failed.

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3245/shard-apl1/igt@pm_rps@waitboost.html

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3243/shard-kbl5/igt@pm_rps@waitboost.html

krisman@collabora.co.uk said:

I can reproduce a funny one:

root@cubi2:/work/igt-gpu-tools/build# tests/gem_sync --r basic-store-all
IGT-Version: 1.19-gf4de2cdf7fd5 (x86_64) (Linux: 4.14.0-rc5.intel-boxes+ x86_64)
Using Execlists submission
Completed 11264 cycles: 487.101 us
Subtest basic-store-all: SUCCESS (5.534s)
pm_rps@resetroot@cubi2:/work/igt-gpu-tools/build# tests/pm_rps --r reset
IGT-Version: 1.19-gf4de2cdf7fd5 (x86_64) (Linux: 4.14.0-rc5.intel-boxes+ x86_64)
(pm_rps:1415) CRITICAL: Test assertion failure function waitboost, file ../../tests/pm_rps.c:615:
(pm_rps:1415) CRITICAL: Failed assertion: post_freqs[CUR] < post_freqs[MAX]
(pm_rps:1415) CRITICAL: error: 1000 >= 1000

On APL.

krisman@collabora.co.uk said:

(In reply to krisman from comment 8)

I can reproduce a funny one:

root@cubi2:/work/igt-gpu-tools/build# tests/gem_sync --r basic-store-all
IGT-Version: 1.19-gf4de2cdf7fd5 (x86_64) (Linux: 4.14.0-rc5.intel-boxes+
x86_64)
Using Execlists submission
Completed 11264 cycles: 487.101 us
Subtest basic-store-all: SUCCESS (5.534s)
pm_rps@resetroot@cubi2:/work/igt-gpu-tools/build# tests/pm_rps --r reset
IGT-Version: 1.19-gf4de2cdf7fd5 (x86_64) (Linux: 4.14.0-rc5.intel-boxes+
x86_64)
(pm_rps:1415) CRITICAL: Test assertion failure function waitboost, file
../../tests/pm_rps.c:615:
(pm_rps:1415) CRITICAL: Failed assertion: post_freqs[CUR] < post_freqs[MAX]
(pm_rps:1415) CRITICAL: error: 1000 >= 1000

On APL.

Which is no longer reproducible after the patch.

This has also not seen on shards in the past ~90 runs. Closing.

Marta Löfstedt @marta said:

https://intel-gfx-ci.01.org/tree/drm-tip/IGT_4029/shard-apl1/igt@pm_rps@min-max-config-idle.html

(pm_rps:23290) CRITICAL: Test assertion failure function idle_check, file pm_rps.c:528:
(pm_rps:23290) CRITICAL: Failed assertion: freqs[CUR] == freqs[RPn]
(pm_rps:23290) CRITICAL: Last errno: 22, Invalid argument
(pm_rps:23290) CRITICAL: error: 433 != 100
Subtest min-max-config-idle failed.

Since it is the same assert I reopen this bug although it is a new subtest.

Chris Wilson @ickle said:

(In reply to Marta Löfstedt from comment 10)

https://intel-gfx-ci.01.org/tree/drm-tip/IGT_4029/shard-apl1/igt@pm_rps@min-
max-config-idle.html

(pm_rps:23290) CRITICAL: Test assertion failure function idle_check, file
pm_rps.c:528:
(pm_rps:23290) CRITICAL: Failed assertion: freqs[CUR] == freqs[RPn]
(pm_rps:23290) CRITICAL: Last errno: 22, Invalid argument
(pm_rps:23290) CRITICAL: error: 433 != 100
Subtest min-max-config-idle failed.

Since it is the same assert I reopen this bug although it is a new subtest.

This test is unlike the reset scenario, it's a very basic failure. Can we move this to a new report so I don't get confused?

Marta Löfstedt @marta said:

The test filed on this bug no longer fail as described in this bug. I will archive and close this.

Marta Löfstedt @marta said:

Restored, but so far only the on the igt@pm_rps@waitboost subtest.

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3634/shard-apl5/igt@pm_rps@waitboost.html

(pm_rps:4799) CRITICAL: Test assertion failure function waitboost, file pm_rps.c:548:
(pm_rps:4799) CRITICAL: Failed assertion: boost_freqs[CUR] == boost_freqs[BOOST]
(pm_rps:4799) CRITICAL: error: 250 != 750
Subtest waitboost failed.

Marta Löfstedt @marta said:

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3642/shard-glkb3/igt@pm_rps@waitboost.html

(pm_rps:8957) CRITICAL: Test assertion failure function waitboost, file pm_rps.c:549:
(pm_rps:8957) CRITICAL: Failed assertion: post_freqs[CUR] < post_freqs[MAX]
(pm_rps:8957) CRITICAL: error: 750 >= 750
Subtest waitboost failed.

Marta Löfstedt @marta closed a related bug:

*** Bug 105186 has been marked as a duplicate of this bug. ***

Marta Löfstedt @marta said:

IGT_4234: 2018-02-10 / 338 runs ago

Marta Löfstedt @marta said:

Reopend due to:
https://intel-gfx-ci.01.org/tree/drm-tip/IGT_4385/shard-hsw2/igt@pm_rps@reset.html

(pm_rps:21511) CRITICAL: Test assertion failure function waitboost, file ../tests/pm_rps.c:568:
(pm_rps:21511) CRITICAL: Failed assertion: boost_freqs[CUR] == boost_freqs[BOOST]
(pm_rps:21511) CRITICAL: error: 750 != 1300
Subtest reset failed.

Marta Löfstedt @marta said:

https://intel-gfx-ci.01.org/tree/drm-tip/drmtip_12/fi-skl-6600u/igt@pm_rps@waitboost.html

(pm_rps:1247) CRITICAL: Test assertion failure function waitboost, file ../tests/pm_rps.c:568:
(pm_rps:1247) CRITICAL: Failed assertion: boost_freqs[CUR] == boost_freqs[BOOST]
(pm_rps:1247) CRITICAL: error: 300 != 1050
Subtest waitboost failed.

Marta Löfstedt @marta said:

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_4050/shard-apl5/igt@pm_rps@reset.html

(pm_rps:1294) CRITICAL: Test assertion failure function waitboost, file ../tests/pm_rps.c:568:
(pm_rps:1294) CRITICAL: Failed assertion: boost_freqs[CUR] == boost_freqs[BOOST]
(pm_rps:1294) CRITICAL: error: 250 != 750
Subtest reset failed.