[PATCH v2 12/21] drm/i915/dp: Use check link state work in the detect handler

[Ville Syrjälä]

On Thu, May 23, 2024 at 06:29:21PM +0300, Imre Deak wrote:
> On Thu, May 23, 2024 at 06:08:40PM +0300, Ville Syrjälä wrote:
> > On Mon, May 20, 2024 at 09:58:10PM +0300, Imre Deak wrote:
> > > Simplify things by retraining a DP link if a bad link is detected in the
> > > connector detect handler from the encoder's check link state work,
> > > similarly to how this is done after a modeset link training failure.
> > > 
> > > Signed-off-by: Imre Deak <imre.deak at intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/display/intel_dp.c | 7 ++-----
> > >  1 file changed, 2 insertions(+), 5 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c
> > > index ff4ed6bb520d8..70b00e5ae7ad7 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_dp.c
> > > +++ b/drivers/gpu/drm/i915/display/intel_dp.c
> > > @@ -5863,11 +5863,8 @@ intel_dp_detect(struct drm_connector *connector,
> > >  	 * Some external monitors do not signal loss of link synchronization
> > >  	 * with an IRQ_HPD, so force a link status check.
> > >  	 */
> > > -	if (!intel_dp_is_edp(intel_dp)) {
> > > -		ret = intel_dp_retrain_link(encoder, ctx);
> > > -		if (ret)
> > > -			return ret;
> > > -	}
> > > +	if (!intel_dp_is_edp(intel_dp))
> > > +		intel_dp_check_link_state(intel_dp);
> > 
> > I would like to see this hack nuked entirely. But that
> > could be a followup.
> 
> Okay. This tries to keep the current behavior, but can add a note that
> the above workaround can be removed after the link state is checked
> after modesets.
> 
> I also wondered about the link state check in the hotplug handler. If
> that's only a way to defer doing this from the HPD IRQ handler - which
> is now changed by patch 13 - that could be also removed eventually?

Not sure which one you want to removed exactly. I presume there
are still at least these cases we need to handle:
- long HDP dropped and came back without any userspace
  initiated modeset in between
  -> kick off a retrain from the long HPD handler
- short HPD indicated some link badness
  -> kick off a retrain from the short HDP handler
- link dropped on its own soon after modeset without
  any HPD for some reason
  -> kick off a retrain from the post modeset link check

And the one we should remove:
- something weird happened to the link and no one noticed,
  and for some random reason userspace just happens to do
  a getconnector() which ends up randomly fixing things

Did I miss anything else?

-- 
Ville Syrjälä
Intel