Skip to contentSkip to conversation

i915 wakes from suspend to RAM with blank screen after commit cd9dde44f47501394b9f0715b6a36a92aa74c0d0

DISCUSSION Lubos Kolouch 3 participants 13 messages Jan 21, 2012 – Feb 7, 2012

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

Hello,

After upgrading kernel from 3.1.5 to 3.2.0 the notebook cannot wake from suspend to RAM properly. The problem stays even with 3.3-rc1.

It wakes, but the screen is blank and computer needs to be rebooted.

When woken, computer is working otherwise (can ping it, ssh to it etc.)

I git bisected it to commit cd9dde44f47501394b9f0715b6a36a92aa74c0d0 is the first bad commit commit cd9dde44f47501394b9f0715b6a36a92aa74c0d0

Author: Adam Jackson <[email protected]>
Date:   Fri Oct 14 12:43:49 2011 -0400

drm/i915/dp: Fix the math in intel_dp_link_required

The previous code was confused about units, which is pretty reasonable given that the units themselves are confusing.

:040000 040000 3c0a0f38fc6482a401340e03fa3fbd687d11eda4
294673b782b3ecc93540f46f7721d385bd30cfac M drivers

Reproducible: Always

Steps to Reproduce:
1. suspend to RAM
2. start the computer
Actual Results:
Screen is blank

Expected Results:
Screen should be normal as before suspend

#1 lore raw
Quoted 5 lines · Lubos Kolouch

I git bisected it to commit cd9dde44f47501394b9f0715b6a36a92aa74c0d0 is the first bad commit commit cd9dde44f47501394b9f0715b6a36a92aa74c0d0

Author: Adam Jackson <[email protected]>
Date:   Fri Oct 14 12:43:49 2011 -0400

Did you try reverting this single commit on top of 3.3-rc1?

And, can you collect the dmesg output with the kernel parameter drm.debug=0xe added to the boot line, both messages at boot time *and* messages during resume would be needed. If it works at boot time, I'm quite surprised it somehow doesn't work at resume...

Signature · 1 line
[email protected]
#2 lore raw

Keith Packard, Sat, 21 Jan 2012 10:15:24 -0800:

Quoted 8 lines
Quoted 4 lines · Lubos Kolouch

I git bisected it to commit cd9dde44f47501394b9f0715b6a36a92aa74c0d0 is the first bad commit commit cd9dde44f47501394b9f0715b6a36a92aa74c0d0

Author: Adam Jackson <[email protected]>
Date:   Fri Oct 14 12:43:49 2011 -0400
Quoted 6 lines

Did you try reverting this single commit on top of 3.3-rc1?

And, can you collect the dmesg output with the kernel parameter drm.debug=0xe added to the boot line, both messages at boot time *and* messages during resume would be needed. If it works at boot time, I'm quite surprised it somehow doesn't work at resume...

I tried now and when I revert this line :

- return (pixel_clock * bpp + 7) / 8;
+ return (pixel_clock * bpp + 9) / 10;

resume works OK for me with 3.3-rc1.

Before change (failed)
----------------------
dmesg after boot - http://paste.pocoo.org/show/538343/
dmesg after failed resume - http://paste.pocoo.org/show/538344/

After change (OK)
-----------------
dmesg after boot - http://paste.pocoo.org/show/538345/
dmesg after OK resume - http://paste.pocoo.org/show/538346/

Luboš Kolouch

#3 lore raw
In reply to Keith Packard
Quoted 8 lines · Keith Packard
Quoted 5 lines · Lubos Kolouch

I git bisected it to commit cd9dde44f47501394b9f0715b6a36a92aa74c0d0 is the first bad commit commit cd9dde44f47501394b9f0715b6a36a92aa74c0d0

Author: Adam Jackson <[email protected]>
Date:   Fri Oct 14 12:43:49 2011 -0400
Quoted 6 lines

Did you try reverting this single commit on top of 3.3-rc1?

And, can you collect the dmesg output with the kernel parameter drm.debug=0xe added to the boot line, both messages at boot time *and* messages during resume would be needed. If it works at boot time, I'm quite surprised it somehow doesn't work at resume...

I tried now and when I revert this line :

- return (pixel_clock * bpp + 7) / 8;
+ return (pixel_clock * bpp + 9) / 10;

resume works OK for me with 3.3-rc1.

Before change (failed)
----------------------
dmesg after boot - http://paste.pocoo.org/show/538343/
dmesg after failed resume - http://paste.pocoo.org/show/538344/

After change (OK)
-----------------
dmesg after boot - http://paste.pocoo.org/show/538345/
dmesg after OK resume - http://paste.pocoo.org/show/538346/

Luboš Kolouch

#4 lore raw
Quoted 2 lines · Lubos Kolouch

dmesg after boot - http://paste.pocoo.org/show/538343/
dmesg after failed resume - http://paste.pocoo.org/show/538344/

Ok, that's weird -- we're choosing a different display port link speed on resume than at boot time. I have a theory that intel_dp_mode_fixup is accidentally using the current bpp value when computing the values needed for the new mode, and that this value may differ between boot and resume.

Here's a patch to dump some debug info out to test this theory.

Quoted 1 line

From 2b16db9fccc6927f3e09020144d5148acc025f58 Mon Sep 17 00:00:00 2001

From: Keith Packard <[email protected]>
Date: Sat, 21 Jan 2012 12:48:38 -0800
Subject: [PATCH] drm/i915: Output a bit of debug info for display port
bandwidth

Something weird is going on with link bandwidth testing; this may help sort out what's wrong.

---
 drivers/gpu/drm/i915/intel_dp.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)
drivers/gpu/drm/i915/intel_dp.c +3 -0
index db3b461..796d534 100644
@@ -697,10 +697,13 @@ intel_dp_mode_fixup(struct drm_encoder *encoder, struct drm_display_mode *mode,
 		mode->clock = intel_dp->panel_fixed_mode->clock;
 	}
 
+	DRM_DEBUG_KMS("using bpp %d to find link_bw and lane_count\n", bpp);
 	for (lane_count = 1; lane_count <= max_lane_count; lane_count <<= 1) {
 		for (clock = 0; clock <= max_clock; clock++) {
 			int link_avail = intel_dp_max_data_rate(intel_dp_link_clock(bws[clock]), lane_count);
 
+			DRM_DEBUG_KMS("lane_count %d clock %d link_avail %d link_required %d\n",
+				      lane_count, clock, link_avail, intel_dp_link_required(intel_dp, mode->clock, bpp));
 			if (intel_dp_link_required(intel_dp, mode->clock, bpp)
 					<= link_avail) {
 				intel_dp->link_bw = bws[clock];
Signature · 1 line
1.7.8.3

Please try this on the broken kernel, sending along the dmesg output at boot and resume time again.

Signature · 1 line
[email protected]
#5 lore raw
Quoted 5 lines · Keith Packard
Quoted 2 lines · Lubos Kolouch

dmesg after boot - http://paste.pocoo.org/show/538343/
dmesg after failed resume - http://paste.pocoo.org/show/538344/

Quoted 8 lines

Ok, that's weird -- we're choosing a different display port link speed on resume than at boot time. I have a theory that intel_dp_mode_fixup is accidentally using the current bpp value when computing the values needed for the new mode, and that this value may differ between boot and resume.

Here's a patch to dump some debug info out to test this theory.

Thanks, here you go:
dmesg after boot - http://paste.pocoo.org/show/538625/
dmesg after failed resume - http://paste.pocoo.org/show/538626/

Lubos

#6 lore raw
Quoted 3 lines · Lubos Kolouch

Thanks, here you go:
dmesg after boot - http://paste.pocoo.org/show/538625/
dmesg after failed resume - http://paste.pocoo.org/show/538626/

Ok, looks like my theory is correct. We're fitting a 24bpp mode to an 18bpp data rate, which doesn't work.

I'm in the middle of bisecting a bluetooth failure in 3.3-rc1, so I don't have a working kernel at present, but here's a section of intel_dp.c in the intel_dp_mode_fixup function:

int max_clock = intel_dp_max_link_bw(intel_dp) == DP_LINK_BW_2_7 ? 1 : 0;
- int bpp = mode->private_flags & INTEL_MODE_DP_FORCE_6BPC ? 18 : 0;
+ int bpp = mode->private_flags & INTEL_MODE_DP_FORCE_6BPC ? 18 : 24;
static int bws[2] = { DP_LINK_BW_1_62, DP_LINK_BW_2_7 };

(this is completely hand-generated, so 'patch' will almost certainly not work)

This will force it to use the correct bpp when computing the bandwidth used by the new mode. This should fix your problem. If so, I'll have a larger patch which stops the underlying function from (almost always incorrectly) using the current bpp when computing bandwidth.

Signature · 1 line
[email protected]
#7 lore raw
Quoted 18 lines · Keith Packard
Quoted 3 lines · Lubos Kolouch

Thanks, here you go:
dmesg after boot - http://paste.pocoo.org/show/538625/
dmesg after failed resume - http://paste.pocoo.org/show/538626/

Ok, looks like my theory is correct. We're fitting a 24bpp mode to an 18bpp data rate, which doesn't work.

I'm in the middle of bisecting a bluetooth failure in 3.3-rc1, so I don't have a working kernel at present, but here's a section of intel_dp.c in the intel_dp_mode_fixup function:

  	int max_clock = intel_dp_max_link_bw(intel_dp) == DP_LINK_BW_2_7 ? 1 : 0;
(quoted diff) +1 -1
-	int bpp = mode->private_flags & INTEL_MODE_DP_FORCE_6BPC ? 18 : 0;
+	int bpp = mode->private_flags & INTEL_MODE_DP_FORCE_6BPC ? 18 : 24;
 	static int bws[2] = { DP_LINK_BW_1_62, DP_LINK_BW_2_7 };
Quoted 9 lines

(this is completely hand-generated, so 'patch' will almost certainly not work)

This will force it to use the correct bpp when computing the bandwidth used by the new mode. This should fix your problem. If so, I'll have a larger patch which stops the underlying function from (almost always incorrectly) using the current bpp when computing bandwidth.

Signature · 1 line
[email protected]

Thanks, I applied it and it seems to solve the issue!

dmesg after boot - http://paste.pocoo.org/show/539115/
dmesg after OK resume - http://paste.pocoo.org/show/539117/

Lubos

#9 lore raw
Quoted 8 lines · Keith Packard
Quoted 2 lines · Lubos Kolouch

dmesg after boot - http://paste.pocoo.org/show/539115/
dmesg after OK resume - http://paste.pocoo.org/show/539117/

I've posted a couple of more comprehensive patches that you should have received; I'd like to know if you can test them to see if they resolve the issue as well.

1) applied the first patch, resume works OK dmesg after boot : http://paste.pocoo.org/show/541205/
dmesg after resume : http://paste.pocoo.org/show/541206/

2) applied second patch on top, resume works OK dmesg after boot: http://paste.pocoo.org/show/541477/
dmesg after resume : http://paste.pocoo.org/show/541479/

Lubos

#10 lore raw
Quoted 10 lines · Lubos Kolouch
Quoted 8 lines · Keith Packard
Quoted 2 lines · Lubos Kolouch

dmesg after boot - http://paste.pocoo.org/show/539115/
dmesg after OK resume - http://paste.pocoo.org/show/539117/

I've posted a couple of more comprehensive patches that you should have received; I'd like to know if you can test them to see if they resolve the issue as well.

Quoted 7 lines

1) applied the first patch, resume works OK dmesg after boot : http://paste.pocoo.org/show/541205/
dmesg after resume : http://paste.pocoo.org/show/541206/

2) applied second patch on top, resume works OK dmesg after boot: http://paste.pocoo.org/show/541477/
dmesg after resume : http://paste.pocoo.org/show/541479/

Thanks!

I'll append a 'Tested-by: Lubos Kolouch <[email protected]>'

to these patches when they go into the kernel.

Signature · 1 line
[email protected]
#11 lore raw
Quoted 19 lines · Keith Packard
Quoted 10 lines · Lubos Kolouch
Quoted 8 lines · Keith Packard
Quoted 2 lines · Lubos Kolouch

dmesg after boot - http://paste.pocoo.org/show/539115/
dmesg after OK resume - http://paste.pocoo.org/show/539117/

I've posted a couple of more comprehensive patches that you should have received; I'd like to know if you can test them to see if they resolve the issue as well.

Quoted 7 lines

1) applied the first patch, resume works OK dmesg after boot : http://paste.pocoo.org/show/541205/
dmesg after resume : http://paste.pocoo.org/show/541206/

2) applied second patch on top, resume works OK dmesg after boot: http://paste.pocoo.org/show/541477/
dmesg after resume : http://paste.pocoo.org/show/541479/

Quoted 6 lines

Thanks!

I'll append a 'Tested-by: Lubos Kolouch<[email protected]>'

to these patches when they go into the kernel.

Hi Keith,

Do you plan on submitting a request to have these patches go in to the 3.2 stable kernel as well?

Thanks,

Joe