[Intel-gfx,REGRESSION] Re: i915 driver crashes on T540p if docking station attached

Commit Message

Daniel Vetter July 30, 2015, 3:54 p.m. UTC

On Thu, Jul 30, 2015 at 5:32 PM, Theodore Ts'o <[e-mail]> wrote:
> On Thu, Jul 30, 2015 at 04:40:02PM +0200, Daniel Vetter wrote:
>> On Wed, Jul 29, 2015 at 10:18:16PM -0700, Linus Torvalds wrote:
>> >  drivers/gpu/drm/drm_atomic_helper.c | 8 +++++---
>> >  1 file changed, 5 insertions(+), 3 deletions(-)
>> >
>> > diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c
>> > index 5b59d5ad7d1c..aac212297b49 100644
>> > --- a/drivers/gpu/drm/drm_atomic_helper.c
>> > +++ b/drivers/gpu/drm/drm_atomic_helper.c
>> > @@ -230,10 +230,12 @@ update_connector_routing(struct drm_atomic_state *state, int conn_idx)
>> >     }
>> >
>> >     connector_state->best_encoder = new_encoder;
>> > -   idx = drm_crtc_index(connector_state->crtc);
>> > +   if (connector_state->crtc) {
>> > +           idx = drm_crtc_index(connector_state->crtc);
>> >
>> > -   crtc_state = state->crtc_states[idx];
>> > -   crtc_state->mode_changed = true;
>> > +           crtc_state = state->crtc_states[idx];
>> > +           crtc_state->mode_changed = true;
>> > +   }
>>
>> This shouldn't happen since if it does we ended up stealing the encoder
>> from the connector itself (we do check for connector_state->crtc earlier)
>> and that would be a bug. I haven't figured out a precise theory but my
>> guess is on the best_encoder selection, and indeed dp mst encoder
>> selection seems to have gone belly up in 4.2 with the bisected commit.
>
> Well, I just tested Linus's patch and it works.

That's sersiously surprising if you mean display and everything actually
works. Is dpms on/off and suspend and all that also still working? Can you
please changed the check into a

if (!connector_state->crtc)
        return 0;

so that we don't blow up on the debug line below and then grab dmesg with
drm.debug=0x1e when this happens? Note there will be lots of noise you
might need to dig out full dmesg from logs.

> BTW, is there any chance that I can suspend my laptop, and then move
> it from my docking station at home (where I have a Dell 30" display)
> to my docking station at work (where I have a Dell 24" display), and
> actually have the new monitor be detected?  For at least the past
> year, I have to reboot in order to be able to use the external
> monitor?  This used to work, but it's been a very long-standing
> regression.  I undrstand that Multi-stream DP is a evil horrible hack,
> and supporting it is painful, but this used to work, and it hasn't in
> a long time.  :-(

Hm we seem to not reprobe mst state on resume. The quick hack below should
help (but totally untested since still no dp mst hub here).
-Daniel

Patch

diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
index 884b4f9b81c4..c0677c83a0e9 100644
--- a/drivers/gpu/drm/i915/i915_drv.c
+++ b/drivers/gpu/drm/i915/i915_drv.c
@@ -775,6 +775,9 @@  static int i915_drm_resume(struct drm_device *dev)
 	/* Config may have changed between suspend and resume */
 	drm_helper_hpd_irq_event(dev);
 
+	dev_priv->short_hpd_port_mask = ~0;
+	queue_work(dev_priv->dp_wq, &dev_priv->dig_port_work);
+
 	intel_opregion_init(dev);
 
 	intel_fbdev_set_suspend(dev, FBINFO_STATE_RUNNING, false);