Bug 19129

Summary: Visiting Engadget with Firefox 3.5 causes X to hang with radeon driver
Product: xorg Reporter: eruditehermit <eruditehermit>
Component: Driver/RadeonAssignee: xf86-video-ati maintainers <xorg-driver-ati>
Status: RESOLVED INVALID QA Contact: Xorg Project Team <xorg-team>
Severity: normal    
Priority: medium CC: christopher.m.penalver, manuel.lauss
Version: 7.4 (2008.09)   
Hardware: x86 (IA32)   
OS: Linux (All)   
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
xorg logfile
none
Debugging patch: Make RENDER acceleration synchronous
none
Alternative synchronous debugging patch none

Description eruditehermit 2008-12-16 23:24:46 UTC
Created attachment 21229 [details]
xorg logfile

When I visit Engadget (www.engadget.com) with Firefox 3.1, X freezes and I am unable to move the mouse pointer. The system is totally unresponsive and I am not able to ctrl+alt+backspace to restart X and I am not able to use magic alt+sysreq key sequences to reboot either. This problem is always reproducable. My hardware is a radeon 9600 mobility rv350 AGP. I am using xserver 1.5.2, Xorg 7.4 in Ubuntu/intrepid. This effects both the radeon driver version in intrepid (6.9.0+git20081003.f9826a56-0ubuntu2) and current git master as of 20081216 (December 16, 2008). I have attached my Xorg.0.log with this message.
Comment 1 Michel Dänzer 2008-12-17 08:35:25 UTC
Does it also happen without Option "GARTSize", "AccelDFS" and "DynamicClocks"? If yes, does Option "RenderAccel" "off" or "AGPMode" "1" (or "2", or "8" if it's AGPv3) or "BusType" "PCI" avoid the problem?
Comment 2 eruditehermit 2008-12-22 01:53:53 UTC
I debugged the problem further with your suggestions and the results are as follows: the problem persists when I disable GARTSize, AccelDFS and DynamicClocks. It also persists when I try AGP Modes 1x, 2x and 4x (I don't think my system is not AGP 8x capable). However, disabling RenderAccel seems to stop the crash from occurring but I can't be 100% sure. I tested reloading the Engadget page about 30-40 times in Firefox 3.1 with RenderAccel disabled and It didn't crash. Then I re-enabled RenderAccel and I managed to get a crash on the first load of Engadget. When I have RenderAccel disabled, X is very slow at drawing to the screen and it is not suitable for normal use as a workaround. I hope this helps.

Thank you for taking the time to help me with this. Please let me know if there is anything else I can do.
Comment 3 eruditehermit 2009-04-08 14:52:40 UTC
The bug still persists with Ubuntu Jaunty 9.04 with radeon 6.12.1 and xserver 1.6. In addition to the complete freeze described before, sometimes when the system is not responsive, I will be able to move the mouse cursor but the movement is slow/jerky and clicks and keystrokes or any input is not processed. The magic reset keys and restarting the xserver don't work and the only way to restore the system is to power cycle.
Comment 4 Michel Dänzer 2009-04-16 03:18:41 UTC
Created attachment 24846 [details] [review]
Debugging patch: Make RENDER acceleration synchronous

In order to try and narrow this down, please apply this xf86-video-ati patch (note that it'll probably decrease performance in general, so don't leave it applied otherwise), then reproduce the problem. If the system doesn't freeze completely, log in from another system via ssh, attach gdb to the X server process and enter the command

bt full

Then attach the output here. This will probably only be useful if all X server binaries have debugging symbols though.
Comment 5 Alex Deucher 2009-06-01 14:11:13 UTC
Does this still happen with xf86-video-ati from git master or the 6.12-branch?
Comment 6 eruditehermit 2009-06-05 08:39:30 UTC
The crash no longer seems to happen with the latest master code. I don't know what commit changed the behaviour however.
Comment 7 Alex Deucher 2009-06-05 09:09:20 UTC
I would guess:
22e39392297fa11003df90c175db3c449d8f9853
Comment 8 eruditehermit 2009-06-12 08:23:34 UTC
I was mistaken when closing this bug. This is still a problem with git Master of radeon as of 2009-06-12 (June 12th 2009). I am still experiencing crashes, most easily reproducable by visiting engadget.com with firefox 3.5b4 or newer. The details of the system are Ubuntu Jaunty 9.04, latest radeon git master, kernel 2.6.28, rv350 ATI mobility radeon 9600 pci ID: 1002:4e50. I will be trying the patch suggested and provide more information soon.
Comment 9 eruditehermit 2009-07-10 00:04:04 UTC
In response to comment #4 which stated that I should apply the asynchronous patch and try to SSH into the box when crashed to dump the log file, I am unable to login via SSH from another machine because the system is completely hung. The machine is unreachable. Please let me know if there is anything else that I can try to figure out what is going wrong.
Comment 10 Michel Dänzer 2009-07-10 00:10:57 UTC
Did you ever try if

    Option "BusType" "PCI"

works around the problem?
Comment 11 Michel Dänzer 2009-07-10 02:16:41 UTC
Created attachment 27555 [details] [review]
Alternative synchronous debugging patch

This is an alternative debugging patch which may or may not be better wrt turning the GPU lockup into a hard hang.
Comment 12 eruditehermit 2009-07-10 14:32:23 UTC
Running with the alternative synchronous debugging patch seems to stop the system from hanging so I can't get a log of the hang. When I revert back to the unpatched version, I still get the hang. It also appears that running with the Option "BusType" "PCI" stops the hang now, however there is a performance hit so it is not a long term solution.
Comment 13 eruditehermit 2009-08-04 06:24:50 UTC
The crash seems to be occurring because I have full page zoom enabled. When the page in firefox is scaled by a non integer value, it causes the crash but when the page is zoomed at 100%, or 200% it loads without crashing. Perhaps there is a rounding problem with some part of the driver that is causing the issue.
Comment 14 Manuel Lauss 2010-04-03 07:59:06 UTC
Any progress on this?

I have an identical setup (rv350, intel AGP), which freezes the system
hard when browsing image-intensive sites with firefox (i.e. open a few
images on railpictures.net and switch tabs a few times).

Disabling AGP completely gets rid of it, but makes Xv uselessly slow.

(ati-git as of 04/02/2010, w. 2.6.34-rc3 KMS)
Comment 15 Da Fox 2010-06-17 00:04:02 UTC
(In reply to comment #14)
> Any progress on this?
> 
> I have an identical setup (rv350, intel AGP), which freezes the system
> hard when browsing image-intensive sites with firefox (i.e. open a few
> images on railpictures.net and switch tabs a few times).
> 
> Disabling AGP completely gets rid of it, but makes Xv uselessly slow.
> 
> (ati-git as of 04/02/2010, w. 2.6.34-rc3 KMS)

You may be experiencing bug #28402 instead. Does this still happen
when switching to an older kernel (2.6.33 works for me, for example)?
Are there interesting messages in your logs?
Comment 16 Manuel Lauss 2010-06-17 00:21:54 UTC
(In reply to comment #15)
> (In reply to comment #14)
> > Any progress on this?
> > 
> > I have an identical setup (rv350, intel AGP), which freezes the system
> > hard when browsing image-intensive sites with firefox (i.e. open a few
> > images on railpictures.net and switch tabs a few times).
> > 
> > Disabling AGP completely gets rid of it, but makes Xv uselessly slow.
> > 
> > (ati-git as of 04/02/2010, w. 2.6.34-rc3 KMS)
> 
> You may be experiencing bug #28402 instead. Does this still happen

Looks like it, yes.


> when switching to an older kernel (2.6.33 works for me, for example)?
> Are there interesting messages in your logs?

Nothing in the logs, nothing over netconsole, just a hard hang.  Sometimes
very fast, sometimes it takes a whole 2 days.

It happens on all kernels (went back to 2.6.28), with or without KMS.  Initially the hangs started to appear with a git snapshot of ati-drivers from
mid-april 2009 (UMS).
Comment 17 Christopher M. Penalver 2016-02-25 21:29:52 UTC
eruditehermit, Ubuntu Jaunty reached EOL on October 23, 2010. For more on this, please see https://wiki.ubuntu.com/Releases .

If this is reproducible in a supported release, it will help immensely if you filed a new report with Ubuntu by ensuring you have the package xdiagnose installed, and that you click the Yes button for attaching additional debugging information running the following from a terminal:
ubuntu-bug xorg

Also, please feel free to subscribe me to it.

For more on why this is helpful, please see https://wiki.ubuntu.com/ReportingBugs.

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.