[igt-dev] ✗ Fi.CI.BAT: failure for igt/i915/i915_pm_lpsp enable pm_lpsp for platforms till Gen11 v2

[Imre Deak]

On Tue, Apr 09, 2019 at 06:11:27PM +0530, Gupta, Anshuman wrote:
> Hi Imre ,
> Could you please review the i915_pm_lpsp series.
> "https://patchwork.freedesktop.org/series/59048/"
> 
> There is non-edp subtest igt failure for broxton.
> "https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_2797/fi-bxt-j4205/igt@i915_pm_lpsp@non-edp.html"
> 
> This broxton machine has a DP1 connector and connected on Pipe A and to
> DDI A. (https://intel-gfx-ci.01.org/hardware/fi-bxt-j4205/fi-bxt-j4205/i915_display_info.txt)
> 
> 
> Current implementation of lpsp_is_enabled() is checking for state of power
> well 2 for gen9 and power well 3 for gen 11, these power well should be
> turned off in order to lpsp enabled.
> 
> But as above broxton machine has connected to port DDI A, that  could be the
> reason power well 2 is not active and non-edp igt subtest get failed on this
> machine.
> 
> What should be the correct way to check lpsp_is_enabled(), could you
> please provide your suggestion.

Power well#2 is not enabled when the test fails, which is expected for
any outputs on DDI-A regardless of the output being eDP or not.

So the test is broken with its assumption that only eDP outputs will
leave PW#2 disabled.

To fix that You need to determine if the enabled output (regardless
whether it's DP/eDP or HDMI) is using pipe A/port A and expect PW#2 to
be disabled if so. You can use i915_display_info to get at the port the
output is using.

Note that audio could also use PW#2 but the driver doesn't enable it on
port A ever (even if the display would support it).

There is also a PW#2 enable/disable sequence during/after the test exits
it would be good to know what causes that (by instrumenting the PW#2
enabling in the driver).

--Imre