Re: ✗ CI.xeFULL: failure for Intel Xe OA IGT's (rev7)

["Dixit, Ashutosh" <ashutosh.dixit@intel.com>]

On Mon, 01 Jul 2024 12:56:28 -0700, Umesh Nerlige Ramappa wrote:
>
> >  IGT changes
> >
> >    Possible regressions
> >
> >     * igt@kms_pm_rpm@i2c:
> >
> >          * shard-dg2-set2: [3]PASS -> [4]FAIL
> >
> >     * {igt@xe_oa@non-zero-reason@rcs-0} (NEW):
> >
> >          * shard-dg2-set2: NOTRUN -> [5]FAIL +1 other test fail
>
> This is something that is an open bug even in i915 and will fail with Xe as
> well. Should we even merge this test as it will create a bunch of JIRAs
> every now and then.

I am also thinking what to do about these failures. There are more in
"CI.xeBAT" (since the HAX patch added the tests to
xe-fast-feedback.testlist, though this will not happen in real CI):

	Possible regressions

	* {igt@xe_oa@mi-rpc} (NEW):

	 * bat-atsm-2: NOTRUN -> INCOMPLETE +1 other test incomplete

	* {igt@xe_oa@mi-rpc@ccs-0} (NEW):

	 * bat-pvc-2: NOTRUN -> INCOMPLETE +1 other test incomplete
	 * bat-dg2-oem2: NOTRUN -> INCOMPLETE +1 other test incomplete

	* {igt@xe_oa@mmio-triggered-reports} (NEW):

	 * bat-adlp-7: NOTRUN -> SKIP

	* {igt@xe_oa@non-zero-reason} (NEW):

	 * bat-pvc-2: NOTRUN -> FAIL +6 other tests fail
	 * bat-dg2-oem2: NOTRUN -> FAIL +1 other test fail

	* {igt@xe_oa@oa-formats} (NEW):

	 * bat-pvc-2: NOTRUN -> TIMEOUT +1 other test timeout

	* {igt@xe_oa@oa-tlb-invalidate} (NEW):

	 * {bat-lnl-1}: NOTRUN -> SKIP +1 other test skip
	 * bat-pvc-2: NOTRUN -> SKIP +1 other test skip

	* {igt@xe_oa@polling-small-buf} (NEW):

	 * bat-atsm-2: NOTRUN -> FAIL

	* {igt@xe_oa@rc6-disable} (NEW):

	 * bat-atsm-2: NOTRUN -> SKIP

Mostly similar to those we see in i915 for Xe1.

One idea is to run these xe_oa IGT's only for Xe2 where things seem to be
more stable, that is only for xe KMD supported platforms. Things seem to be
more stable for Xe2, maybe because we don't know how to check Xe2 PEC
counters, like we do for Xe1 in sanity_check_reports. Or simplify the
checks for Xe1, e.g. get rid of sanity_check_reports? And simplify some of
the other asserts.

Or merge these first and as part of fixing the bugs see what to do for the
tests, so that we'll have a record in git of what we did and why?

Thanks.
--
Ashutosh