Re: [PATCH v2 0/7] Remove deep-pkgc subtest

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Mon, May 25, 2026 at 08:03:39PM +0530, Thasleem, Mohammed wrote:
> 
> On 25-05-2026 07:48 pm, Ville Syrjälä wrote:
> > On Mon, May 25, 2026 at 07:41:29PM +0530, Thasleem, Mohammed wrote:
> >> On 17-04-2026 06:04 pm, Ville Syrjälä wrote:
> >>> On Fri, Apr 17, 2026 at 02:04:59PM +0530, Mohammed Thasleem wrote:
> >>>> Remove the deep-pkgc subtest as it depends on other display IPs which makes
> >>>> testing unreliable. Current validation uses Package C-state counters that
> >>>> are affected by unrelated system components and requires platform-specific
> >>>> checks that vary across different platforms.
> >>>> A redesigned test with proper display IP isolation will be implemented
> >>>> in a future patch.
> >>> What is the actual plan for that? How would it even be implemented
> >>> and when should we expect to get it?
> >> It's in discussion with hardware design team to come up with hooks for
> >> better validation and tracking.
> > And how are you going to catch regression on current systems?
> The removed deep-pkgc subtest was providing unreliable results due to 
> its dependency on unrelated system components, making it more likely to 
> produce false positives rather than catch actual regressions.

Are you saying the CI setups are unstable enough that we sometimes
reach deep pkgc and sometimes we don't, on the same system? Someone
should fix that so that there isn't random ping pong. Either the
systems have been configured correctly and can reach deep pkgc, or
they have not and never reach those states.

I know it's a real pain to figure out what is preventing the pkgc
states due to lack of good feedback from the SoC level stuff, and
the pmc debugfs stuff is an undocumented mess. But it should still
be possible to get there in most cases.

> And we will test pkg c tests as part of manual validation.

Aren't you removing those tests here? And on which platforms
does this manual testing happen? I also have no idea where the
results of this manual testing are tracked.

I would also like to know which systems in CI can actually
reach deep pkgc states and which can't because that really does
affect how the display hardware works, and might be important
if we eg. get underruns.

> >
> >>>> Mohammed Thasleem (7):
> >>>>     Revert "tests/intel/kms_pm_dc: Ensure eDP detection before skipping
> >>>>       test"
> >>>>     Revert "tests/intel/kms_pm_dc: Add polling for deep-pkgc"
> >>>>     Revert "tests/intel/kms_pm_dc: Update test duration to 4 seconds"
> >>>>     Revert "tests/intel/kms_pm_dc: Update VRR handling and eDP output
> >>>>       check"
> >>>>     Revert "tests/intel/kms_pm_dc: Add time unit macros and update delay
> >>>>       calculation"
> >>>>     Revert "tests/intel/kms_pm_dc: Add a new test to validate the deep
> >>>>       sleep state during extended vblank"
> >>>>     tests/intel/kms_pm_dc: Remove unused pkgc functions
> >>>>
> >>>>    tests/intel/kms_pm_dc.c | 98 -----------------------------------------
> >>>>    1 file changed, 98 deletions(-)
> >>>>
> >>>> -- 
> >>>> 2.43.0

-- 
Ville Syrjälä
Intel