Bug 87171 - [855GM] KMS regression in 3.18
Summary: [855GM] KMS regression in 3.18
Status: CLOSED FIXED
Alias: None
Product: DRI
Classification: Unclassified
Component: DRM/Intel (show other bugs)
Version: unspecified
Hardware: x86 (IA32) Linux (All)
: high critical
Assignee: Ville Syrjala
QA Contact: Intel GFX Bugs mailing list
URL:
Whiteboard:
Keywords: bisected, regression
Depends on:
Blocks:
 
Reported: 2014-12-09 21:23 UTC by Bruno
Modified: 2016-12-13 08:45 UTC (History)
2 users (show)

See Also:
i915 platform:
i915 features:


Attachments
Displayed result with 3.18 (1.92 MB, image/jpeg)
2014-12-09 21:23 UTC, Bruno
no flags Details
Expected result, with 3.17 (1.97 MB, image/jpeg)
2014-12-09 21:24 UTC, Bruno
no flags Details
intel_reg_dump, with 3.17 (10.64 KB, text/plain)
2014-12-09 21:25 UTC, Bruno
no flags Details
intel_reg_dump, with 3.18 (10.64 KB, text/plain)
2014-12-09 21:41 UTC, Bruno
no flags Details

Description Bruno 2014-12-09 21:23:49 UTC
Created attachment 110648 [details]
Displayed result with 3.18

On Acer TravelMate 66x with i855GM switching from 3.17 to 3.18 kernel scanout to LVDS becomes mis-configured.

It looks like pixels are being displayed "interlaced" both horizontally and vertically so that first pixel is at its expected location, pixel right of it is two pixes further to the right, pixel at half screen (or is it 1/4 of screen) is located right below first pixel.

See screenshots for visual representation. Further debug details will follow with some delay, as well as bisection.
Comment 1 Bruno 2014-12-09 21:24:29 UTC
Created attachment 110649 [details]
Expected result, with 3.17
Comment 2 Bruno 2014-12-09 21:25:48 UTC
Created attachment 110650 [details]
intel_reg_dump, with 3.17
Comment 3 Bruno 2014-12-09 21:32:04 UTC
Kernel log extract, with 3.18:
[    0.893782] Linux agpgart interface v0.103
[    0.893893] agpgart-intel 0000:00:00.0: Intel 855GM Chipset
[    0.893960] agpgart-intel 0000:00:00.0: detected gtt size: 131072K total, 131072K mappable
[    0.894634] agpgart-intel 0000:00:00.0: detected 32768K stolen memory
[    0.894992] agpgart-intel 0000:00:00.0: AGP aperture is 128M @ 0xe8000000
[    0.895201] [drm] Initialized drm 1.1.0 20060810
[    0.897913] [drm] Memory usable by graphics device = 128M
[    0.897973] [drm] Replacing VGA console driver
[    0.898927] Console: switching to colour dummy device 80x25
[    0.901093] [drm] Supports vblank timestamp caching Rev 2 (21.10.2013).
[    0.901109] [drm] Driver supports precise vblank timestamp query.
[    0.901357] vgaarb: device changed decodes: PCI:0000:00:02.0,olddecodes=io+mem,decodes=io+mem:owns=io+mem
[    0.990220] [drm] GMBUS [i915 gmbus panel] timed out, falling back to bit banging on pin 3
[    1.020916] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[    1.047516] [drm] initialized overlay support
[    1.047632] [drm] Initialized i915 1.6.0 20140905 for 0000:00:02.0 on minor 0
[    1.100281] [drm] GMBUS [i915 gmbus vga] timed out, falling back to bit banging on pin 2
[    1.121264] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[    1.200224] [drm] GMBUS [i915 gmbus dpb] timed out, falling back to bit banging on pin 5
[    1.215077] fbcon: inteldrmfb (fb0) is primary device
[    1.810166] Console: switching to colour frame buffer device 175x65
[    1.820984] i915 0000:00:02.0: fb0: inteldrmfb frame buffer device
[    1.821053] i915 0000:00:02.0: registered panic notifier
[  122.817020] [drm:i9xx_check_fifo_underruns] *ERROR* pipe B underrun
[  122.831035] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  155.129274] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  432.772681] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  498.863614] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  524.688752] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  526.163628] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  543.921330] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  545.472861] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  562.710908] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  565.496726] [drm:i9xx_check_fifo_underruns] *ERROR* pipe B underrun
[  565.510848] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[  624.522935] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe B underrun
Comment 4 Bruno 2014-12-09 21:41:15 UTC
Created attachment 110651 [details]
intel_reg_dump, with 3.18
Comment 5 Bruno 2014-12-09 21:48:32 UTC
Better description of the output with 3.18:

About top half of framebuffer is visible though pixel lines are not wrapped as one would expect, they are wrapped somewhere around the middle (but not exactly) so each pixel line from framebuffer fills two lines on monitor.

From looking at difference between reg_dumper outputs under 3.17 and 3.18, the major difference is:
3.17:
  DSPASTRIDE: 0x00002000 (8192 bytes)
3.18:
  DSPASTRIDE: 0x00001600 (5632 bytes)

Full difference:
3.17:
  DSPASTRIDE: 0x00002000 (8192 bytes)
  DSPABASE: 0x01000000
  DSPBCNTR: 0x58000000 (disabled, pipe A)
  DSPBBASE: 0x00020000
  VGACNTRL: 0x80000000 (disabled)
3.18
  DSPASTRIDE: 0x00001600 (5632 bytes)
  DSPABASE: 0x00020000
  DSPBCNTR: 0x00000000 (disabled, pipe A)
  DSPBBASE: 0x00000000
  VGACNTRL: 0xe104008e (disabled)
Comment 6 Ville Syrjala 2014-12-10 09:19:35 UTC
Can you try this 'intel_reg_write 0x20cc 0x40000' ?
Comment 7 Bruno 2014-12-10 17:31:59 UTC
(In reply to Ville Syrjala from comment #6)
> Can you try this 'intel_reg_write 0x20cc 0x40000' ?

This made no visual change, no any difference in intel_reg_dumper output.

On execution it reported old value of 0x4 and new value of 0x0.

Going to perform bisection unless you have some better/quicker suggestion.
Comment 8 Ville Syrjala 2014-12-10 17:54:59 UTC
(In reply to Bruno from comment #7)
> (In reply to Ville Syrjala from comment #6)
> > Can you try this 'intel_reg_write 0x20cc 0x40000' ?
> 
> This made no visual change, no any difference in intel_reg_dumper output.
> 
> On execution it reported old value of 0x4 and new value of 0x0.

So not trickle feed related.

> 
> Going to perform bisection unless you have some better/quicker suggestion.

I guess it wouldn't hurt to check the watermarks too. Any difference between the kernels with this?
intel_reg_read -c 2 0x20d8

If that's not it either, then a bisect would be appreciated.
Comment 9 Bruno 2014-12-10 19:24:24 UTC
> I guess it wouldn't hurt to check the watermarks too. Any difference between
> the kernels with this?
> intel_reg_read -c 2 0x20d8

There is:
- 3.17
  0x20D8 : 0x1080101
               ^
  0x20DC : 0x102
- 3.18
  0x20D8 : 0x11F0108
               ^
  0x20DC : 0x102

> If that's not it either, then a bisect would be appreciated.

Operation in progress :) 2/13 iterations done
Comment 10 Bruno 2014-12-12 22:25:05 UTC
Done bisecting.

Offending commit is:
commit 69769f9a422bfc62e17399da3590c5e31ac37f24
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date:   Fri Aug 15 01:22:08 2014 +0300

    drm/i915: Preserve VGACNTR bits from the BIOS
    
    My Fujistsu-Siemens Lifebook S6010 doesn't like to resume from
    S3 unless VGACNTR has been restore to the original value. The BIOS
    value in this case was 0x0124008E. Setting the "VGA disable" bit
    doesn't interfere with the S3 resume fortunately.
    
    Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
    Tested-by: Thomas Richter <richter@rus.uni-stuttgart.de>
    Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Reverting just this commit from 3.18 makes the display issue go away.


It seems i915_vgacntrl_reg() [VGACNTRL on i855GM] still has some effect on LVDS configuration when VGA_DISP_DISABLE bit is set.

In i915_reg.h the following bits are defined:
# define VGA_DISP_DISABLE                       (1 << 31)
# define VGA_2X_MODE                            (1 << 30)
# define VGA_PIPE_B_SELECT                      (1 << 29)

Of those bits VGA_2X_MODE might very well be related to the end result I'm seeing here. (my VGA mode is set to 2X mode if I'm not mistaken...)
I'm going to check VGACNTRL values as handled by this commit and check if zeroing VGA_2X_MODE or VGA_PIPE_B_SELECT is sufficient to get proper output.
Comment 11 Bruno 2014-12-12 22:27:33 UTC
Note: the pipe underrun complaints are still there though:
[    0.894671] [drm] Initialized drm 1.1.0 20060810
[    0.897376] [drm] Memory usable by graphics device = 128M
[    0.897437] [drm] Replacing VGA console driver
[    0.900581] [drm] Supports vblank timestamp caching Rev 2 (21.10.2013).
[    0.900596] [drm] Driver supports precise vblank timestamp query.
[    0.990221] [drm] GMBUS [i915 gmbus panel] timed out, falling back to bit banging on pin 3
[    1.020888] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[    1.047522] [drm] initialized overlay support
[    1.047637] [drm] Initialized i915 1.6.0 20140905 for 0000:00:02.0 on minor 0
[    1.100281] [drm] GMBUS [i915 gmbus vga] timed out, falling back to bit banging on pin 2
[    1.121256] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[    1.200225] [drm] GMBUS [i915 gmbus dpb] timed out, falling back to bit banging on pin 5
[    1.215081] fbcon: inteldrmfb (fb0) is primary device
[    1.740977] i915 0000:00:02.0: fb0: inteldrmfb frame buffer device
[    1.741047] i915 0000:00:02.0: registered panic notifier
[   26.876525] [drm:i9xx_check_fifo_underruns] *ERROR* pipe B underrun
[   26.904260] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[   56.110048] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
[   61.611917] [drm:i9xx_set_fifo_underrun_reporting] *ERROR* pipe A underrun
Comment 12 Bruno 2014-12-13 10:49:40 UTC
The VGACNTRL value as saved early on:
  0x6104008e

That has both VGA_2X_MODE and VGA_PIPE_B_SELECT bits set (plus a whole bunch of bits on the less significant bytes).

With patch 69769f9a422bfc62e17399da3590c5e31ac37f24 included VGACNTRL would be set to 0xe104008e.
If I mask VGA_2X_MODE I get proper output on my 1440x1050 LVDS display.

Thus the following patch on top of 3.18 fixes the issue for me (as an alternative to reverting commit 69769f9a42):
diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index 9cb5c95..fc4e16d 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -12937,7 +12937,7 @@ static void i915_disable_vga(struct drm_device *dev)
 	 * Fujitsu-Siemens Lifebook S6010 (830) has problems resuming
 	 * from S3 without preserving (some of?) the other bits.
 	 */
-	I915_WRITE(vga_reg, dev_priv->bios_vgacntr | VGA_DISP_DISABLE);
+	I915_WRITE(vga_reg, (dev_priv->bios_vgacntr | VGA_DISP_DISABLE) &
+	                    ~VGA_2X_MODE);
 	POSTING_READ(vga_reg);
 }
Comment 13 Bruno 2015-01-12 20:29:18 UTC
Fixed in v3.19-rc4 with commit 01f5a6261cea395f72877aeb7c2fe2d42e1b1e00
Comment 14 Jani Nikula 2015-01-13 09:17:32 UTC
(In reply to Bruno from comment #13)
> Fixed in v3.19-rc4 with commit 01f5a6261cea395f72877aeb7c2fe2d42e1b1e00

Oh sorry, I failed to close the bug when I committed. Thanks for the report and the follow-up.
Comment 15 Jari Tahvanainen 2016-12-13 08:45:02 UTC
Closing resolved+fixed. Verified in v3.19-rc4 by reporter.


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.