Skip to contentSkip to conversation

i915 blank issue on kernel 3.1.0

DISCUSSION Woody Suwalski 2 participants 5 messages Sep 21, 2011 – Sep 25, 2011

lore
Share & tools
lore
#0 lore raw

Chris, I do not know if it is i915 driver or X'drm or what.. Problem exists on 3.1.0, no problem on 3.0.3

Intel 945GJME 8086:27ac, intel 2.15 x driver, libdrm-intel 2.4.26, Debian wheezy.

Recently I have noticed that on Asus EeePCs once X screensaver kicks in - system is hosed. Screen goes black and I can not wake it - even restarting X does not work, only reboot.

Over ssh I have captured some action from drm.debug=0x0e:

<7>[  787.953399] [drm:intel_panel_get_backlight], get backlight PWM = 312
<7>[  787.953421] [drm:intel_panel_set_backlight], set backlight PWM = 0
<7>[  788.232114] [drm:intel_update_fbc],
<7>[  788.232125] [drm:i9xx_get_fifo_size], FIFO size - (0x00001d9c) A: 28
<7>[  788.232134] [drm:intel_calculate_wm], FIFO entries required for mode: 15
<7>[  788.232142] [drm:intel_calculate_wm], FIFO watermark level: 11
<7>[  788.232149] [drm:i9xx_get_fifo_size], FIFO size - (0x00001d9c) B: 31
<7>[  788.232157] [drm:i9xx_update_wm], FIFO watermarks - A: 11, B: 29
<7>[  788.232165] [drm:i9xx_update_wm], self-refresh entries: 64
<7>[  788.232171] [drm:i9xx_update_wm], Setting FIFO watermarks - A: 11, B: 29, C: 2, SR 63
<7>[  788.232180] [drm:i9xx_update_wm], memory self refresh enabled
<7>[  788.232320] [drm:intel_panel_get_backlight], get backlight PWM = 0
<7>[  788.232393] [drm:intel_panel_set_backlight], set backlight PWM = 0
<7>[  788.232623] [drm:intel_panel_set_backlight], set backlight PWM = 0
<7>[ 1132.017589] [drm:i9xx_get_fifo_size], FIFO size - (0x00001d9c) A: 28
<7>[ 1132.017614] [drm:intel_calculate_wm], FIFO entries required for mode: 15
<7>[ 1132.017629] [drm:intel_calculate_wm], FIFO watermark level: 11
<7>[ 1132.017643] [drm:i9xx_get_fifo_size], FIFO size - (0x00001d9c) B: 31
<7>[ 1132.017658] [drm:i9xx_update_wm], FIFO watermarks - A: 11, B: 29
<7>[ 1132.017674] [drm:i9xx_update_wm], self-refresh entries: 64
<7>[ 1132.017687] [drm:i9xx_update_wm], Setting FIFO watermarks - A: 11, B: 29, C: 2, SR 63
<7>[ 1132.017704] [drm:i9xx_update_wm], memory self refresh enabled
<7>[ 1132.036216] [drm:intel_update_fbc],
<7>[ 1132.036248] [drm:intel_lvds_enable], applying panel-fitter: 8, 0
<7>[ 1132.038081] [drm:intel_panel_set_backlight], set backlight PWM = 312
<7>[ 1132.038299] [drm:intel_panel_set_backlight], set backlight PWM = 0

It looks that the problem is the last line - sets back to 0???

I have tried to write 312 back to /sys/class/backlight/intel-backlight/ bl-power or brightness, and touch uevent - no help. On 3.0.3 system I see that there is no /sys/../intel-backlight, just eeepc.
And on touching a key it produces
drm:intel_lvds_enable, followed by set backlight PWM=312 (but skips resetting it back to 0 right after enable).

So:
is there a way to fix "intel_panel" to call lvds_enable logic? Is there a way to force usage of eeepc backlight instead (as a workaround)?

Thanks, Woody Suwalski

#1 lore raw
Quoted 2 lines · Woody Suwalski

Chris, I do not know if it is i915 driver or X'drm or what.. Problem exists on 3.1.0, no problem on 3.0.3

Any way you could narrow it down a bit more? A bisect should sort it out pretty quickly, given that you've got working code in 3.0 and not-working code in 3.1-rc. You can start by manually bisecting across the 3.1 release candidates, and then narrow it down from there by bisecting only across the drivers/gpu/drm/i915 directory.

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

Re: Regression: i915 blank issue on kernel 3.1.0-rc3

Quoted 9 lines · Keith Packard
Quoted 2 lines · Woody Suwalski

Chris, I do not know if it is i915 driver or X'drm or what.. Problem exists on 3.1.0, no problem on 3.0.3

Any way you could narrow it down a bit more? A bisect should sort it out pretty quickly, given that you've got working code in 3.0 and not-working code in 3.1-rc. You can start by manually bisecting across the 3.1 release candidates, and then narrow it down from there by bisecting only across the drivers/gpu/drm/i915 directory.

Yes, I have narrowed it down. I did not run bisect, as it takes me 4hrs for kernel build on Eeepc...

The problem has appeared in 3.1-rc3, it is a regression.
I have tested -rc1 and -rc2, they are OK ( not that there was much activity regarding i915 in -rc2). The problem is still present in -rc7.

I see these entries in the changelog, these are Keith's additions to -rc3 which maight impact the blanker:

Keith Packard (6):
drm/i915: Wait for LVDS panel power sequence drm/i915: Leave LVDS registers unlocked drm/i915: Fix PCH port pipe select in CPT disable paths drm/i915: Remove unused 'reg' argument to dp_pipe_enabled drm/i915: Can't do accurate vblank timestamps with UMS drm/i915: Cannot set clock gating under UMS

I can do a fast incremental rebuild of my current -rc3 tree, if Keith could send me some patches...

Thanks, Woody

#3 lore raw

Re: Regression: i915 blank issue on kernel 3.1.0-rc3

In reply to Keith Packard
Quoted 10 lines · Keith Packard
Quoted 2 lines · Woody Suwalski

Chris, I do not know if it is i915 driver or X'drm or what.. Problem exists on 3.1.0, no problem on 3.0.3

Any way you could narrow it down a bit more? A bisect should sort it out pretty quickly, given that you've got working code in 3.0 and not-working code in 3.1-rc. You can start by manually bisecting across the 3.1 release candidates, and then narrow it down from there by bisecting only across the drivers/gpu/drm/i915 directory.

I see that now more people are reporting a blanker problem. I guess kernel.org downtime is really impacting testing...

Anyway, for Eeepc reverting the

aaa6fd2a004147bf32fce05720938236de3361d9

seems to be fixing the issue. Which is not of course the way we want to go ;-(

Woody

#4 lore raw

Re: Regression: i915 blank issue on kernel 3.1.0-rc3

Quoted 6 lines · Woody Suwalski

Anyway, for Eeepc reverting the

aaa6fd2a004147bf32fce05720938236de3361d9

seems to be fixing the issue. Which is not of course the way we want to go ;-(

That's better than something broken in the middle of mode setting.

Matthew - this is your most excellent patch which adds a native backlight mechanism. What other info do you need to help figure out how to make it play well on other machines? We've got a handful of regressions about backligh in the 3.1 RC series, and I'd hate to have to revert this...

Signature · 1 line
[email protected]