Bug 85864

Please attach your xorg log and dmesg output.

(In reply to Jonathan Schuster from comment #0)
> This only happened as of the upgrade to 7.5 (specifically, using the most
> recent version of this package on Arch: 
> https://www.archlinux.org/packages/extra/x86_64/xf86-video-ati/).

What was the latest version where it didn't happen? Can you bisect the Git change between the two versions causing the problem?

For some reason it's not happening at all now, after upgrading back to 7.5. If it does happen again, though, I'll be sure to attach dmesg output and the xorg log.

Bisecting could be difficult even if it starts happening again, since I have no way to reliably reproduce it.

Created attachment 109109 [details]
Xorg log

The symptoms just reappeared again. Here's the Xorg log; dmesg output in next attachment.

Created attachment 109110 [details]
dmesg output

dmesg output after seeing the issue again.

I see a similar thing but rarely and *only* in Firefox and Thunderbird.
There seem to be two different effects: One where every instance of one character is corrupted and one where only one is corrupted; that's the one I see. See attached screenshot. Scrolling will sometimes correct the character; having Firefox redraw the page always fixes it.
This is with KDE on a 7850 with pretty much up-to-date Debian jessie but kernel 3.17-1.

Created attachment 109120 [details]
Font corruption example

Nothing interesting in my dmesg but there's many of these from hours ago:

[351665.023972] AMD-Vi: Event logged [IO_PAGE_FAULT device=04:00.0 domain=0x0019 address=0x00000000044c3f40 flags=0x0020]
[351665.023975] AMD-Vi: Event logged [IO_PAGE_FAULT device=04:00.0 domain=0x0019 address=0x00000000044c3fc0 flags=0x0020]
[351665.023978] AMD-Vi: Event logged [IO_PAGE_FAULT device=04:00.0 domain=0x0019 address=0x00000000044c3f80 flags=0x0020]

Created attachment 109145 [details]
3.18-git dmesg

[ I'm one of the posters from the linked forum posts in comment#1 ]

In the interest of making my system actually usable (font corruption + very consistent gpu lock ups), I went ahead and upgraded to kernel 3.18 from git.

After using the system for half a day or so, it appears stable, but not quite there. While a few apps (Chromium, XBMC, emacs, etc) open I just had another lock up. Although this time, it did not totally corrupt the running system and forcing a full power off. I was able to restart my login manager to get back into X to post this comment.

Attached is my demsg output from seconds after the crash.

(In reply to Adam Flott from comment #9)
> After using the system for half a day or so, it appears stable, but not
> quite there.

This bug report is about font corruption, not about stability issues. Please file your own report about your stability issues.

(In reply to Jonathan Schuster from comment #3)
> For some reason it's not happening at all now, after upgrading back to 7.5.

Are you sure that it doesn't happen with older versions of the driver? I think this is actually more likely a bug in the glamor glyph cache code.

(In reply to Michel Dänzer from comment #11)
> (In reply to Jonathan Schuster from comment #3)
> > For some reason it's not happening at all now, after upgrading back to 7.5.
> 
> Are you sure that it doesn't happen with older versions of the driver? I
> think this is actually more likely a bug in the glamor glyph cache code.

I don't recall seeing this issue until I did an upgrade a month or so ago, and it happens often enough these days that it's hard to imagine I wouldn't notice it. Still, I've just downgraded to 7.4 again (the 1:7.4.0-3 Arch package, from mid July), and I'll post logs if I see the issue happening again.

It's also possible this is a bug in a completely different piece of software, but if that's the case, I wouldn't know where to start looking. If anyone has suggestions for other logs to look at, etc., I'd be happy to take a look.

(In reply to Michel Dänzer from comment #11)
> (In reply to Jonathan Schuster from comment #3)
> > For some reason it's not happening at all now, after upgrading back to 7.5.
> 
> Are you sure that it doesn't happen with older versions of the driver? I
> think this is actually more likely a bug in the glamor glyph cache code.

You were right. It just happened again with 7.4. I'll add a new Xorg log and dmesg output.

Created attachment 109284 [details]
7.4 dmesg

Created attachment 109285 [details]
7.4 Xorg log

Created attachment 109678 [details]
Thunderbird screenshot

Interesting case: Two "/" corrupted, another is fine.

The same happens in Opnesuse 13.2 The graphic card is a Radeon R7 240. Affected are primarily Firefox and Thunderbird. It is always one particular letter, that becomes distorted throughout the text. Increasing or decrasing the font-size lets everything go back to normal - but not for long.

This happened to me after I upgraded Xorg in the hope I will get performance improvement. Last upgrade seems to be xorg 1.16.1 -  1.16.1.901. However, I did whole system upgrade so might have installed a new library of something or other that uses a different X path and triggers the bug. I also enabled glamor. I have occasoinal corrupted characters but while the issue is noticeable it is not reliably reproducible.


01:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc. [AMD/ATI] Redwood XT [Radeon HD 5670/5690/5730] [1002:68d8]

Is this still an issue with xserver 1.17/git?

Alex: I can probably 1.17/git either tomorrow or early next week. 

Has anyone found a reliable way to reproduce the bug, though? Right now the only way I can report if it's gone or not is to run my computer for a few days and see if it shows up at all or not, which is not a particularly fast turnaround time.

Jonathan, did you get a chance to test a newer xserver? Some people in another related bug report claimed it's fixed for them in xserver Git master.

I don't remember when I last updated, but I've apparently been running this version all this week:

X.Org X Server 1.16.99.901 (1.17.0 RC 1)

I haven't seen the issue pop up at all this week, so it appears to be fixed.

(In reply to Jonathan Schuster from comment #22)
> I haven't seen the issue pop up at all this week, so it appears to be fixed.

Great, thanks, resolving.