igt@kms_plane@pixel-format-pipe-a-planes failed on EHL 4K@60hz

(kms_plane:4136) CRITICAL: Test assertion failure function capture_crc, file ../tests/kms_plane.c:526:
(kms_plane:4136) CRITICAL: Failed assertion: !(crc->has_valid_frame && crc->frame != vblank)
(kms_plane:4136) CRITICAL: Got CRC for the wrong frame (got 640, expected 639). CRC buffer overflow?
(kms_plane:4136) igt_core-INFO: Stack trace:
(kms_plane:4136) igt_core-INFO:   #0 ../lib/igt_core.c:1712 __igt_fail_assert()
(kms_plane:4136) igt_core-INFO:   #1 [capture_crc.isra.0+0x4f]
(kms_plane:4136) igt_core-INFO:   #2 ../tests/kms_plane.c:568 capture_format_crcs.constprop.0()
(kms_plane:4136) igt_core-INFO:   #3 ../tests/kms_plane.c:653 test_format_plane_colors.constprop.0()
(kms_plane:4136) igt_core-INFO:   #4 ../tests/kms_plane.c:680 test_pixel_formats.constprop.0()
(kms_plane:4136) igt_core-INFO:   #5 ../tests/kms_plane.c:947 run_tests_for_pipe_plane.constprop.0()
(kms_plane:4136) igt_core-INFO:   #6 ../tests/kms_plane.c:1020 __real_main1007()
(kms_plane:4136) igt_core-INFO:   #7 ../tests/kms_plane.c:1007 main()
(kms_plane:4136) igt_core-INFO:   #8 ../csu/libc-start.c:342 __libc_start_main()
(kms_plane:4136) igt_core-INFO:   #9 ../sysdeps/x86_64/start.S:122 _start()

it seems to be a regression of commit 8eddbad0(@vsyrjala), the problem's gone once rolling back. My understanding is that i915 skips two vblank/flip events for gen8+, and this w/a is triggered when enabling/disabling pipe crc, which seems to interfere with the pending event querying and caused the problem.

changed the description

This should just indicate the test is too slow and the crc buffer overflowed (did you get a message about that in dmesg?) while the test was rendering the next fb. But on such modern hw the test should be using a 64x64 fb so the rendering really should be quit fast, even if it has to do some expensive format conversions.

In general this bug report is lacking logs so impossible to say what's really going on.

I did not see the "user space read too slow and overflow" message in dmesg.

what log in particular do you need? i will re-run the test and capture it

dmesg_kms_plane_pixel.log

kms_plane_pixel.log

Is this still seen to happen? I made for this issues this fix some time ago fd6d58ed

closed