[Intel-gfx] [PATCH] drm/i915: Drop vblank wait from intel_dp_link_down

[Ville Syrjälä]

On Mon, Feb 09, 2015 at 03:15:56PM +0200, Jani Nikula wrote:
> On Wed, 26 Nov 2014, Paulo Zanoni <przanoni at gmail.com> wrote:
> > 2014-11-24 13:54 GMT-02:00 Daniel Vetter <daniel.vetter at ffwll.ch>:
> >> Nothing in Bspec seems to indicate that we actually needs this, and it
> >> looks like can't work since by this point the pipe is off and so
> >> vblanks won't really happen any more.
> >>
> >> Note that Bspec mentions that it takes a vblank for this bit to
> >> change, but _only_ when enabling.
> >>
> >> Dropping this code quenches an annoying backtrace introduced by the
> >> more anal checking since
> >>
> >> commit 51e31d49c89055299e34b8f44d13f70e19aaaad1
> >> Author: Daniel Vetter <daniel.vetter at ffwll.ch>
> >> Date:   Mon Sep 15 12:36:02 2014 +0200
> >>
> >>     drm/i915: Use generic vblank wait
> >>
> >> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86095
> >> Signed-off-by: Daniel Vetter <daniel.vetter at intel.com>
> >> ---
> >>  drivers/gpu/drm/i915/intel_dp.c | 17 +----------------
> >>  1 file changed, 1 insertion(+), 16 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/i915/intel_dp.c b/drivers/gpu/drm/i915/intel_dp.c
> >> index 46731da407c0..63fcdbf90652 100644
> >> --- a/drivers/gpu/drm/i915/intel_dp.c
> >> +++ b/drivers/gpu/drm/i915/intel_dp.c
> >> @@ -3514,8 +3514,6 @@ intel_dp_link_down(struct intel_dp *intel_dp)
> >>         enum port port = intel_dig_port->port;
> >>         struct drm_device *dev = intel_dig_port->base.base.dev;
> >>         struct drm_i915_private *dev_priv = dev->dev_private;
> >> -       struct intel_crtc *intel_crtc =
> >> -               to_intel_crtc(intel_dig_port->base.base.crtc);
> >>         uint32_t DP = intel_dp->DP;
> >>
> >>         if (WARN_ON(HAS_DDI(dev)))
> >> @@ -3540,8 +3538,6 @@ intel_dp_link_down(struct intel_dp *intel_dp)
> >>
> >>         if (HAS_PCH_IBX(dev) &&
> >>             I915_READ(intel_dp->output_reg) & DP_PIPEB_SELECT) {
> >> -               struct drm_crtc *crtc = intel_dig_port->base.base.crtc;
> >> -
> >>                 /* Hardware workaround: leaving our transcoder select
> >>                  * set to transcoder B while it's off will prevent the
> >>                  * corresponding HDMI output on transcoder A.
> >> @@ -3552,18 +3548,7 @@ intel_dp_link_down(struct intel_dp *intel_dp)
> >>                  */
> >>                 DP &= ~DP_PIPEB_SELECT;
> >>                 I915_WRITE(intel_dp->output_reg, DP);
> >> -
> >> -               /* Changes to enable or select take place the vblank
> >> -                * after being written.
> >> -                */
> >> -               if (WARN_ON(crtc == NULL)) {
> >> -                       /* We should never try to disable a port without a crtc
> >> -                        * attached. For paranoia keep the code around for a
> >> -                        * bit. */
> >> -                       POSTING_READ(intel_dp->output_reg);
> >> -                       msleep(50);
> >> -               } else
> >> -                       intel_wait_for_vblank(dev, intel_crtc->pipe);
> >
> > What I can guess is that we have the vblank wait here because the
> > DP_PORT_EN bit is still enabled at this point. It would make some
> > sense to have it if the pipe were not off... So removing the waits
> > looks sane: Reviewed-by: Paulo Zanoni <paulo.r.zanoni at intel.com>
> >
> > But when I read the spec, it makes me think that maybe doing the
> > I915_WRITE above is also wrong, since the port is still enabled. Maybe
> > we should only clear bit 30 in the same write as the one that clears
> > bit 31?
> 
> Ugh. So the spec says, "When disabling the port, software must
> temporarily enable the port with transcoder select (bit #30) cleared to
> ‘0’ after disabling the port."
> 
> IIUC we should disable like we normally do, and do the w/a by enabling
> and disabling the port with DP_PIPEB_SELECT cleared *after* that. I
> think the current code is wrong, the patch is wrong, what Paulo suggests
> is wrong, and also intel_disable_hdmi() is wrong.

This code has been bugging me for a long time as well. IIRC I even had
cooked up some patches to do the re-enable as you suggest since I
read the spec the same way. But I never had enough time to test it. And
in order to really test it I would first like to actually reproduce the
problem that the workaround is supposed to fix. How else would you know
if the workaround is correct after all.

-- 
Ville Syrjälä
Intel OTC