Bug 77201 - [855GM] Firefox textdisplay problems (partly or full blank lines) after update to xf86-video-intel-2.99.911
Summary: [855GM] Firefox textdisplay problems (partly or full blank lines) after updat...
Status: RESOLVED FIXED
Alias: None
Product: xorg
Classification: Unclassified
Component: Driver/intel (show other bugs)
Version: unspecified
Hardware: x86 (IA32) Linux (All)
: medium normal
Assignee: Chris Wilson
QA Contact: Intel GFX Bugs mailing list
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2014-04-08 20:00 UTC by CarlEitsger
Modified: 2014-04-12 06:26 UTC (History)
0 users

See Also:
i915 platform:
i915 features:

Attachments
excerpt of dmesglog (1.53 KB, text/plain)
2014-04-08 20:00 UTC, CarlEitsger 		no flags Details
screenshot excerpt showing the lines in text while everything else is okay (83.63 KB, image/png)
2014-04-08 20:03 UTC, CarlEitsger 		no flags Details
Thunderbird 24.4.0 in RSS message view (even in the RSS message list there are blank lines - not shown here) (73.91 KB, image/png)
2014-04-09 16:02 UTC, CarlEitsger 		no flags Details
screenshot of a search on ixquick.com (as before - on Firefox 28.0) (72.73 KB, image/png)
2014-04-09 16:02 UTC, CarlEitsger 		no flags Details
Xorg.0.log (private information and input devices removed) (9.46 KB, text/plain)
2014-04-09 16:04 UTC, CarlEitsger 		no flags Details
the bisect log when the compile error happened (1.09 KB, text/plain)
2014-04-10 19:16 UTC, CarlEitsger 		no flags Details
the bisect log when bisect was finished (1.41 KB, text/plain)
2014-04-10 19:17 UTC, CarlEitsger 		no flags Details
Fix CA/a8 blend factors (2.77 KB, patch)
2014-04-10 19:47 UTC, Chris Wilson 		no flags Details | Splinter Review
View All

Description CarlEitsger 2014-04-08 20:00:46 UTC 
Created attachment 97093 [details]
excerpt of dmesglog

Hardware: [url=http://www.intel.com/content/www/us/en/chipsets/mainstream-chipsets/chipset-855gm.html]Intel 855GM chipset[/url] with integrated graphics. Driver: [url=https://www.archlinux.org/packages/extra/i686/xf86-video-intel/]xf86-video-intel[/url]

Software: Arch Linux

Driver versions

* xf86-video-intel-2.99.[b]911[/b]-1-i686
* xf86-video-intel-2.99.[b]911[/b]-2-i686 (current)

produce display errors on http://www.faz.net/ (and https://www.ixquick.com ) in Firefox (using a fresh profile results in same error) under KDE4, TWM, openbox and LXDE. See article text screenshot [url]http://i.imgur.com/xV4sxni.png[/url]. Missing text lines, in the headlines on hover, in the main text after load but more severe after scrolling a bit. Everything else displays fine (images, firefox' GUI).

The error does also occur under another user account (with nearly standard settings) in KDE4 (same computer). All packages up to date.

No errors with the older version xf86-video-intel-2.99.[b]910[/b]-1-i686 (to which I have downgraded now). With this and the older versions I had no such problems since months.

No errors in other programs like Chromium or Konqueror (but did not test that much, Firefox is what I use often).

It seems to be similar to bug #36326 but the effect is much more severe and appeared only in the last weeks while it worked before.
Comment 1 CarlEitsger 2014-04-08 20:01:29 UTC 
Previously reported/discussed at https://bbs.archlinux.org/viewtopic.php?pid=1401270 .
Comment 2 CarlEitsger 2014-04-08 20:03:50 UTC 
Created attachment 97094 [details]
screenshot excerpt showing the lines in text while everything else is okay
Comment 3 Chris Wilson 2014-04-09 06:03:43 UTC 
Can you perform a bisection?
Comment 4 Chris Wilson 2014-04-09 09:34:37 UTC 
Looking at firefox-24 on debian with an 845g under awesome, everything looks ok, even http;//wwww.faz.net/

Please also attach your Xorg.0.log.
Comment 5 Chris Wilson 2014-04-09 11:47:16 UTC 
Also it would be useful to confirm the bug against xf86-video-intel.git.
Comment 6 CarlEitsger 2014-04-09 16:00:59 UTC 
Thank you, Chris!

(In reply to comment #3)
> Can you perform a bisection?

Sorry, I have not really a clue what you mean. I am no programmer. Do you mean to find out the code commit date which causes this problem? I do not think so if I would have to compile packes myself I will need to stay at the old version and hope that somebody else solves this problem in the driver. I mentioned between which release versions the bad change seems to happen. It is from .910 to .911.

(In reply to comment #5)
> Also it would be useful to confirm the bug against xf86-video-intel.git.

See above, sorry, I think this is too difficult for me.


Attachments follow. I noticed that it also occurs in Thunderbird.
Comment 7 CarlEitsger 2014-04-09 16:02:03 UTC 
Created attachment 97140 [details]
Thunderbird 24.4.0 in RSS message view (even in the RSS message list there are blank lines - not shown here)
Comment 8 CarlEitsger 2014-04-09 16:02:35 UTC 
Created attachment 97141 [details]
screenshot of a search on ixquick.com (as before - on Firefox 28.0)
Comment 9 CarlEitsger 2014-04-09 16:04:38 UTC 
Created attachment 97142 [details]
Xorg.0.log (private information and input devices removed)
Comment 10 Chris Wilson 2014-04-09 17:05:20 UTC 
git bisect is a tool to identify a commit that introduces a bug.

Start by cloning the xf86-video-intel.git repository,
$ git clone git://anongit.freedesktop.org/xorg/driver/xf86-video-intel
$ cd xf86-video-intel
$ ./autogen.sh --prefix=/usr
$ make && sudo make install

Next time you update your system pkg, it will overwrite our temporary install. autogen is likely to complain about missing development packages, you will have to query arch how best to install those.

Check that you can reproduce the bug.
$ git bisect start
$ git bisect bad
$ git checkout 2.99.910
$ make && sudo make install

Check that you cannot reproduce the bug
$ git bisect good

git will then select a commit roughly halfway between, you then
$ make && sudo make install

Retest, if the result is good:
$ git bisect good

and if bad (lack of lines being drawn):
$ git bisect bad

Repeat until git announces a winner.
Comment 11 Chris Wilson 2014-04-10 17:16:11 UTC 
Any luck in compiling xf86-video-intel.git?
Comment 12 CarlEitsger 2014-04-10 19:16:08 UTC 
Created attachment 97177 [details]
the bisect log when the compile error happened
Comment 13 CarlEitsger 2014-04-10 19:16:53 UTC 
I had a compile error for 

[142f8461944b294dbc3fb0a7bf607bccf0bccc1f] sna/gen2: Tidy blend factor selection for the source 

Final error was:

make: *** [all] Error 2

See attachment 97177 [details] for the log at this state.

I did a 

git reset --hard HEAD~1

then (according to the man page).
Comment 14 CarlEitsger 2014-04-10 19:17:38 UTC 
Created attachment 97178 [details]
the bisect log when bisect was finished
Comment 15 CarlEitsger 2014-04-10 19:19:24 UTC 
After this I could successfully finish the bisecting:

Bisecting: 0 revisions left to test after this (roughly 0 steps)
[142f8461944b294dbc3fb0a7bf607bccf0bccc1f] sna/gen2: Tidy blend factor selection for the source

See attachment 97178 [details] for the log.

Do you need and other information?
Comment 16 CarlEitsger 2014-04-10 19:24:57 UTC 
just a note: git's bisecting was not that hard. Thanks for the excellent guide! I only had to install the packages git and xorg-server-devel (with arch's pacman).
Comment 17 Chris Wilson 2014-04-10 19:47:22 UTC 
Created attachment 97184 [details] [review]
Fix CA/a8 blend factors

Thanks, that was very useful.

Can you please apply this patch (git apply < /path/to/download) and give it a quick spin?
Comment 18 CarlEitsger 2014-04-10 23:23:27 UTC 
I have applied attachment 97184 [details] [review] to the newes git version. It removes the problem of missing lines.

I do not know if it is relevant:  for some reason I get wrong mouse cursors with the latest git version (they show the cursor for the last action - no update, e.g. a waiting clock).
Comment 19 Chris Wilson 2014-04-11 05:42:43 UTC 
Hmm, still updating here, but 855gm is special...

commit a79eb2e29a158910785580409c51a585f25876a7
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Thu Apr 10 20:41:57 2014 +0100

    sna/gen2: Fix CA/a8 blend factors
    
    This is a revert of
    
    commit 142f8461944b294dbc3fb0a7bf607bccf0bccc1f [2.99.911]
    Author: Chris Wilson <chris@chris-wilson.co.uk>
    Date:   Mon Mar 10 15:20:16 2014 +0000
    
        sna/gen2: Tidy blend factor selection for the source
    
    as that broke glyph rendering in firefox for instance.
    
    Reported-by: Carl Eitsger <4607vrfcr84spd21f08@weg-werf-email.de>
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=77201
    Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Comment 20 Chris Wilson 2014-04-11 06:35:33 UTC 
Just tested on a 845g (which is different at a hw to the 855gm, but uses the same code path for simplicity). I saw some lag in cursor updates, but it turned out just to be awesome waiting for a timeout before switching back from the clock to the arrow.
Comment 21 CarlEitsger 2014-04-11 23:17:37 UTC 
Confirm, no blank lines in the current HEAD - 9ae8213. Great!  Set to "fixed" - or is this too early?!

By the way: blank lines also appeared in pidgin (saw them with version e8be2a4) - not only in Mozilla products (does pidgin use a gecko for rendering?.

Regarding the cursors: have opened bug 77351
Comment 22 Chris Wilson 2014-04-12 06:26:50 UTC 
(In reply to comment #21)
> By the way: blank lines also appeared in pidgin (saw them with version
> e8be2a4) - not only in Mozilla products (does pidgin use a gecko for
> rendering?.

They both use cairo, but in this case it is the actual font and font filtering that was important.

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.