Re: [PATCH v2 6/8] drm/i915/reset: move gt related stuff out of display reset

[Jani Nikula <jani.nikula@intel.com>]

On Wed, 26 Feb 2025, Matt Roper <matthew.d.roper@intel.com> wrote:
> On Wed, Feb 26, 2025 at 12:38:40PM +0200, Jani Nikula wrote:
>> On Tue, 25 Feb 2025, Matt Roper <matthew.d.roper@intel.com> wrote:
>> > On Tue, Feb 25, 2025 at 01:14:20PM +0200, Jani Nikula wrote:
>> >> Move the checks for whether display reset is needed as well as
>> >> I915_RESET_MODESET flag handling to gt side of things.
>> >> 
>> >> Cc: Matt Roper <matthew.d.roper@intel.com>
>> >> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
>> >> ---
>> >>  .../drm/i915/display/intel_display_reset.c    | 15 --------------
>> >>  drivers/gpu/drm/i915/gt/intel_reset.c         | 20 +++++++++++++++++++
>> >>  2 files changed, 20 insertions(+), 15 deletions(-)
>> >> 
>> >> diff --git a/drivers/gpu/drm/i915/display/intel_display_reset.c b/drivers/gpu/drm/i915/display/intel_display_reset.c
>> >> index b7962f90c21c..362436cd280f 100644
>> >> --- a/drivers/gpu/drm/i915/display/intel_display_reset.c
>> >> +++ b/drivers/gpu/drm/i915/display/intel_display_reset.c
>> >> @@ -37,15 +37,6 @@ void intel_display_reset_prepare(struct intel_display *display)
>> >>  	if (!HAS_DISPLAY(display))
>> >>  		return;
>> >>  
>> >> -	/* reset doesn't touch the display */
>> >> -	if (!intel_display_reset_test(display) &&
>> >> -	    !gpu_reset_clobbers_display(display))
>> >> -		return;
>> >> -
>> >> -	/* We have a modeset vs reset deadlock, defensively unbreak it. */
>> >
>> > Doesn't this comment more accurately apply to the 'if' condition below
>> > rather than to the flag updates and wakeup we do before that?  Assuming
>> > I'm understanding correctly, it seems like the comment should stay here
>> > and not move to the other file --- saying "We have a ... deadlock" is
>> > only true if we still have a pending pin after we've done that other
>> > stuff.  The unbreaking part (by wedging) is still located here too.
>> 
>> I'm... not sure.
>> 
>> Commit d59cf7bb73f3 ("drm/i915/display: Use dma_fence interfaces instead
>> of i915_sw_fence") seems relevant. We no longer have anyone waiting on
>> I915_RESET_MODESET, and I think we could probably remove the bit from
>> reset flags altogether, and handle this locally in
>> intel_gt_reset_global(). Right?
>
> Yeah, I believe you're right.

So I modified the approach again [1]. Sorry, and thanks for bearing with
me!

BR,
Jani.


[1] https://lore.kernel.org/r/cover.1741001054.git.jani.nikula@intel.com


-- 
Jani Nikula, Intel