Skip to contentSkip to conversation

drm/i915: Don't write DSPSURF for old chips

PATCH In review Takashi Iwai 2 participants 6 messages Feb 27, 2012 – Mar 1, 2012

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/i915: Don't write DSPSURF for old chips

It seems that writing DSPSURF in intel_flush_display_plane() causes the blank screen on some old laptops like Dell D630 with 965GM. Since this operation is needed only for ILK+, make it conditional.

---
 drivers/gpu/drm/i915/intel_display.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_display.c +2 -1
@@ -1379,7 +1379,8 @@ static void intel_flush_display_plane(st
 				      enum plane plane)
 {
 	I915_WRITE(DSPADDR(plane), I915_READ(DSPADDR(plane)));
-	I915_WRITE(DSPSURF(plane), I915_READ(DSPSURF(plane)));
+	if (dev_priv->info->gen >= 5)
+		I915_WRITE(DSPSURF(plane), I915_READ(DSPSURF(plane)));
 }
 
 /**
#1 lore raw
Quoted 3 lines · Takashi Iwai

It seems that writing DSPSURF in intel_flush_display_plane() causes the blank screen on some old laptops like Dell D630 with 965GM. Since this operation is needed only for ILK+, make it conditional.

The specs say that DSPASURF is the latch register for updates of the DSPA registers on gen4 (including 965gm) as well. Presumably the bug is that we only partially update the DSPA registers prior to the first call to intel_flush_display_plane() which this papers over by disabling the update until a valid address is written to DSPASURF. And there is such a spurious call to intel_enable_plane() prior to us setting a valid scanout surface:

(quoted diff) +0 -0

b/drivers/gpu/drm/i915/intel_d
index 66b19d3..ea8e4a1 100644

drivers/gpu/drm/i915/intel_display.c +0 -0
@@ -5453,7 +5453,6 @@ static int i9xx_crtc_mode_set(struct drm_crtc

*crtc,

        I915_WRITE(DSPCNTR(plane), dspcntr);
        POSTING_READ(DSPCNTR(plane));

- intel_enable_plane(dev_priv, plane, pipe);

        ret = intel_pipe_set_base(crtc, x, y, old_fb);

-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#2 lore raw

At Mon, 27 Feb 2012 12:17:57 +0000,

Quoted 6 lines · Chris Wilson
Quoted 3 lines · Takashi Iwai

It seems that writing DSPSURF in intel_flush_display_plane() causes the blank screen on some old laptops like Dell D630 with 965GM. Since this operation is needed only for ILK+, make it conditional.

Quoted 7 lines

The specs say that DSPASURF is the latch register for updates of the DSPA registers on gen4 (including 965gm) as well. Presumably the bug is that we only partially update the DSPA registers prior to the first call to intel_flush_display_plane() which this papers over by disabling the update until a valid address is written to DSPASURF. And there is such a spurious call to intel_enable_plane() prior to us setting a valid scanout surface:

Sounds reasonable. FWIW, the change was first introduced in commit [b24e7179: drm/i915: add pipe/plane enable/disable functions], then in commit [efc2924e: drm/i915: Call intel_enable_plane from i9xx_crtc_mode_set (again)], it's placed into i9xx_crtc_mode_set().

This explains the fact that it was discovered only on old machines as i9xx_crtc_mode_set() is the only crtc_mode_set op calling intel_enable_plane().

BTW, the bisection leaded to a merge commit, so the bug is really depending on the activation path or timing.

I'll ask a tester to try your patch.

thanks,

Takashi

Quoted 19 lines
drivers/gpu/drm/i915/intel_display.c +0 -2
index 66b19d3..ea8e4a1 100644
@@ -5453,7 +5453,6 @@ static int i9xx_crtc_mode_set(struct drm_crtc
 *crtc,
 
        I915_WRITE(DSPCNTR(plane), dspcntr);
        POSTING_READ(DSPCNTR(plane));
-       intel_enable_plane(dev_priv, plane, pipe);
 
        ret = intel_pipe_set_base(crtc, x, y, old_fb);
 
-Chris
Signature · 2 lines
-- 
Chris Wilson, Intel Open Source Technology Centre
#3 lore raw

At Mon, 27 Feb 2012 14:08:28 +0100,

Quoted 22 lines · Takashi Iwai

At Mon, 27 Feb 2012 12:17:57 +0000,

Quoted 6 lines · Chris Wilson
Quoted 3 lines · Takashi Iwai

It seems that writing DSPSURF in intel_flush_display_plane() causes the blank screen on some old laptops like Dell D630 with 965GM. Since this operation is needed only for ILK+, make it conditional.

Quoted 7 lines

The specs say that DSPASURF is the latch register for updates of the DSPA registers on gen4 (including 965gm) as well. Presumably the bug is that we only partially update the DSPA registers prior to the first call to intel_flush_display_plane() which this papers over by disabling the update until a valid address is written to DSPASURF. And there is such a spurious call to intel_enable_plane() prior to us setting a valid scanout surface:

Sounds reasonable. FWIW, the change was first introduced in commit [b24e7179: drm/i915: add pipe/plane enable/disable functions], then in commit [efc2924e: drm/i915: Call intel_enable_plane from i9xx_crtc_mode_set (again)], it's placed into i9xx_crtc_mode_set().

Quoted 8 lines

This explains the fact that it was discovered only on old machines as i9xx_crtc_mode_set() is the only crtc_mode_set op calling intel_enable_plane().

BTW, the bisection leaded to a merge commit, so the bug is really depending on the activation path or timing.

I'll ask a tester to try your patch.

He reported back that it reduces the failure rate but doesn't fix completely. Still get a blank screen in 20% rate.

Any other clue?

thanks,

Takashi

Quoted 25 lines

thanks,

Takashi

Quoted 19 lines
drivers/gpu/drm/i915/intel_display.c +0 -2
index 66b19d3..ea8e4a1 100644
@@ -5453,7 +5453,6 @@ static int i9xx_crtc_mode_set(struct drm_crtc
 *crtc,
 
        I915_WRITE(DSPCNTR(plane), dspcntr);
        POSTING_READ(DSPCNTR(plane));
-       intel_enable_plane(dev_priv, plane, pipe);
 
        ret = intel_pipe_set_base(crtc, x, y, old_fb);
 
-Chris
Signature · 2 lines
-- 
Chris Wilson, Intel Open Source Technology Centre
Reviewed-by Daniel Vetter Reviewed-by
#4 lore raw
Quoted 31 lines · Takashi Iwai

At Mon, 27 Feb 2012 14:08:28 +0100,

Quoted 29 lines · Takashi Iwai

At Mon, 27 Feb 2012 12:17:57 +0000,

Quoted 6 lines · Chris Wilson
Quoted 3 lines · Takashi Iwai

It seems that writing DSPSURF in intel_flush_display_plane() causes the blank screen on some old laptops like Dell D630 with 965GM. Since this operation is needed only for ILK+, make it conditional.

Quoted 7 lines

The specs say that DSPASURF is the latch register for updates of the DSPA registers on gen4 (including 965gm) as well. Presumably the bug is that we only partially update the DSPA registers prior to the first call to intel_flush_display_plane() which this papers over by disabling the update until a valid address is written to DSPASURF. And there is such a spurious call to intel_enable_plane() prior to us setting a valid scanout surface:

Sounds reasonable. FWIW, the change was first introduced in commit [b24e7179: drm/i915: add pipe/plane enable/disable functions], then in commit [efc2924e: drm/i915: Call intel_enable_plane from i9xx_crtc_mode_set (again)], it's placed into i9xx_crtc_mode_set().

This explains the fact that it was discovered only on old machines as i9xx_crtc_mode_set() is the only crtc_mode_set op calling intel_enable_plane().

BTW, the bisection leaded to a merge commit, so the bug is really depending on the activation path or timing.

Quoted 6 lines
Quoted 1 line

I'll ask a tester to try your patch.

He reported back that it reduces the failure rate but doesn't fix completely. Still get a blank screen in 20% rate.

Any other clue?

I haven't yet found anything else in the same vein as this, but the 965gm does ring a warning bell for this recent regression:

commit 5ca0c34ae28344b6b4ca3036bc82f89c8db16a59

Author: Dave Airlie <[email protected]>
Date:   Thu Feb 23 15:33:40 2012 +0000

drm/i915: fix mode set on load pipe. (v2)

Booted my i965 machine and it started printing the unsupported pixel format of 0 message (once I added content to it).

Oh looksie here, we pass 0. fix.

v2: compile it.

Buzilla: https://bugs.freedesktop.org/show_bug.cgi?id=45966

which is currently sitting in airlied/drm-fixes.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#5 lore raw

At Wed, 29 Feb 2012 23:54:46 +0000,

Quoted 56 lines · Chris Wilson
Quoted 31 lines · Takashi Iwai

At Mon, 27 Feb 2012 14:08:28 +0100,

Quoted 29 lines · Takashi Iwai

At Mon, 27 Feb 2012 12:17:57 +0000,

Quoted 6 lines · Chris Wilson
Quoted 3 lines · Takashi Iwai

It seems that writing DSPSURF in intel_flush_display_plane() causes the blank screen on some old laptops like Dell D630 with 965GM. Since this operation is needed only for ILK+, make it conditional.

Quoted 7 lines

The specs say that DSPASURF is the latch register for updates of the DSPA registers on gen4 (including 965gm) as well. Presumably the bug is that we only partially update the DSPA registers prior to the first call to intel_flush_display_plane() which this papers over by disabling the update until a valid address is written to DSPASURF. And there is such a spurious call to intel_enable_plane() prior to us setting a valid scanout surface:

Sounds reasonable. FWIW, the change was first introduced in commit [b24e7179: drm/i915: add pipe/plane enable/disable functions], then in commit [efc2924e: drm/i915: Call intel_enable_plane from i9xx_crtc_mode_set (again)], it's placed into i9xx_crtc_mode_set().

This explains the fact that it was discovered only on old machines as i9xx_crtc_mode_set() is the only crtc_mode_set op calling intel_enable_plane().

BTW, the bisection leaded to a merge commit, so the bug is really depending on the activation path or timing.

Quoted 6 lines
Quoted 1 line

I'll ask a tester to try your patch.

He reported back that it reduces the failure rate but doesn't fix completely. Still get a blank screen in 20% rate.

Any other clue?

I haven't yet found anything else in the same vein as this, but the 965gm does ring a warning bell for this recent regression:

commit 5ca0c34ae28344b6b4ca3036bc82f89c8db16a59

Author: Dave Airlie <[email protected]>
Date:   Thu Feb 23 15:33:40 2012 +0000

drm/i915: fix mode set on load pipe. (v2)

Booted my i965 machine and it started printing the unsupported pixel format of 0 message (once I added content to it).

Oh looksie here, we pass 0. fix.

v2: compile it.

Quoted 8 lines

Buzilla: https://bugs.freedesktop.org/show_bug.cgi?id=45966

which is currently sitting in airlied/drm-fixes.

Hm, this must be irrelevant (inapplicable) because the tests were done with 3.1, 3.2 (and SLE11-SP2 kernels which has backports from 3.3-rc2 but without these fixes). The bug seems to have been introduced between 3.0 and 3.1.

thanks,

Takashi