Bug 15284 - [radeon] corruption in firefox and cs checker warning
Summary: [radeon] corruption in firefox and cs checker warning
Status: RESOLVED CODE_FIX
Alias: None
Product: Drivers
Classification: Unclassified
Component: Video(DRI - non Intel) (show other bugs)
Hardware: All Linux
: P1 normal
Assignee: drivers_video-dri
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2010-02-12 13:23 UTC by dev
Modified: 2010-02-20 12:23 UTC (History)
4 users (show)

See Also:
Kernel Version: drm-radeon-testing
Subsystem:
Regression: No
Bisected commit-id:


Attachments
visual corruption in firefox (181.86 KB, image/png)
2010-02-12 13:23 UTC, dev
Details
modprobe log (5.93 KB, text/plain)
2010-02-12 13:25 UTC, dev
Details
other visual corruption in firefox (320.26 KB, image/png)
2010-02-12 13:35 UTC, dev
Details

Description dev 2010-02-12 13:23:54 UTC
Created attachment 25003 [details]
visual corruption in firefox

from drm-radeon-testing tree @commit 7cb72ef4d39978e6e07415a2d552b06d567c3079
lots of the following in dmesg:
radeon 0000:01:00.0: r600_packet3_check:1055 invalid cmd stream 140
[drm:radeon_cs_ioctl] *ERROR* Invalid command stream !
radeon 0000:01:00.0: z/stencil with no depth buffer
radeon 0000:01:00.0: r600_packet3_check:1055 invalid cmd stream 159
[drm:radeon_cs_ioctl] *ERROR* Invalid command stream !
radeon 0000:01:00.0: z/stencil with no depth buffer

Card is an Radeon 4850(RV770), Screenshots from corruption in firefox attached.
Comment 1 dev 2010-02-12 13:25:18 UTC
Created attachment 25004 [details]
modprobe log
Comment 2 dev 2010-02-12 13:35:34 UTC
Created attachment 25005 [details]
other visual corruption in firefox
Comment 3 Jérôme Glisse 2010-02-12 15:42:18 UTC
Please add more context, which desktop are you running (kde, gnome ? ...) Do you have desktop effect enabled ? Is there a GL application running ?
Comment 4 Johannes Hirte 2010-02-15 00:04:29 UTC
I can confirm the "Invalid command stream" errors. In the log I found mostly 

[drm:radeon_cs_ioctl] *ERROR* Invalid command stream !
radeon 0000:01:00.0: z/stencil with no depth buffer                                                                       
radeon 0000:01:00.0: r600_packet3_check:1055 invalid cmd stream 142

and one

[drm:radeon_cs_ioctl] *ERROR* Invalid command stream !                                                                    
radeon 0000:01:00.0: z/stencil with no depth buffer                                                                       
radeon 0000:01:00.0: r600_packet3_check:1055 invalid cmd stream 161

The screen corruption occured in akregator (KDE newsreader) here. 

My kernel is latest git (linux-2.6) with the drm-code from drm-radeon-testing. I'm running KDE 4.4 (with composite enabled). Libdrm, mesa and radeon driver are latest git, KMS is enabled. Hardware is a RV635 PRO AGP [Radeon HD 3650].
Comment 5 Johannes Hirte 2010-02-17 13:11:23 UTC
with latest updates from drm-radeon-testing I get now:

radeon 0000:01:00.0: z/stencil with no depth buffer                                                                       
radeon 0000:01:00.0: r600_packet3_check:1055 invalid cmd stream 142                                                       
[drm:radeon_cs_ioctl] *ERROR* Invalid command stream !                                                                    
------------[ cut here ]------------                                                                                      
WARNING: at drivers/gpu/drm/radeon/radeon_fence.c:159 radeon_fence_signaled+0x61/0x90 [radeon]()                          
Hardware name: To Be Filled By O.E.M.                                                                                     
Querying an unemited fence : ffff88001f6b9680 !                                                                           
Modules linked in: snd_seq_midi snd_emu10k1_synth snd_emux_synth snd_seq_virmidi snd_seq_midi_emul snd_seq_oss snd_seq_midi_event snd_seq snd_pcm_oss snd_mixer_oss aes_x86_64 aes_generic xts gf128mul dm_crypt radeon ttm drm_kms_helper drm i2c_algo_bit snd_emu10k1 snd_rawmidi snd_ac97_codec ac97_bus snd_pcm snd_seq_device snd_timer snd_page_alloc snd_util_mem snd_hwdep i2c_amd8111 ohci_hcd sata_sil sr_mod amd64_edac_mod k8temp snd edac_core hwmon i2c_amd756 uhci_hcd edac_mce_amd sg                                                                                                   
Pid: 2384, comm: X Not tainted 2.6.33-rc8-00070-g96737d6 #45
Call Trace:
 [<ffffffff810305ff>] ? warn_slowpath_common+0x76/0x8c
 [<ffffffff81030667>] ? warn_slowpath_fmt+0x40/0x45
 [<ffffffffa016916c>] ? radeon_fence_signaled+0x61/0x90 [radeon]
 [<ffffffffa01691df>] ? radeon_fence_wait+0x44/0x300 [radeon]
 [<ffffffff8108a2ac>] ? get_partial_node+0x15/0x81
 [<ffffffffa017b4c9>] ? radeon_cs_parser_init+0x1cb/0x293 [radeon]
 [<ffffffffa017aaae>] ? radeon_ib_get+0x124/0x1ec [radeon]
 [<ffffffffa017b78e>] ? radeon_cs_ioctl+0x0/0x19e [radeon]
 [<ffffffffa017b82e>] ? radeon_cs_ioctl+0xa0/0x19e [radeon]
 [<ffffffffa00f9970>] ? drm_ioctl+0x21a/0x300 [drm]
 [<ffffffff81009fe9>] ? save_i387_xstate+0x99/0x1ae
 [<ffffffff8109e622>] ? vfs_ioctl+0x23/0x93
 [<ffffffff8109eb69>] ? do_vfs_ioctl+0x454/0x498
 [<ffffffff81093549>] ? vfs_read+0x133/0x148
 [<ffffffff8109ebe9>] ? sys_ioctl+0x3c/0x5c
 [<ffffffff81001e6b>] ? system_call_fastpath+0x16/0x1b
---[ end trace 9376ac3cb6e5c7db ]---
radeon 0000:01:00.0: z/stencil with no depth buffer
radeon 0000:01:00.0: r600_packet3_check:1055 invalid cmd stream 142
[drm:radeon_cs_ioctl] *ERROR* Invalid command stream !
------------[ cut here ]------------
WARNING: at drivers/gpu/drm/radeon/radeon_fence.c:159 radeon_fence_signaled+0x61/0x90 [radeon]()
Hardware name: To Be Filled By O.E.M.
Querying an unemited fence : ffff88002680d780 !
Modules linked in: snd_seq_midi snd_emu10k1_synth snd_emux_synth snd_seq_virmidi snd_seq_midi_emul snd_seq_oss snd_seq_midi_event snd_seq snd_pcm_oss snd_mixer_oss aes_x86_64 aes_generic xts gf128mul dm_crypt radeon ttm drm_kms_helper drm i2c_algo_bit snd_emu10k1 snd_rawmidi snd_ac97_codec ac97_bus snd_pcm snd_seq_device snd_timer snd_page_alloc snd_util_mem snd_hwdep i2c_amd8111 ohci_hcd sata_sil sr_mod amd64_edac_mod k8temp snd edac_core hwmon i2c_amd756 uhci_hcd edac_mce_amd sg
Pid: 2384, comm: X Tainted: G        W  2.6.33-rc8-00070-g96737d6 #45
Call Trace:
 [<ffffffff810305ff>] ? warn_slowpath_common+0x76/0x8c
 [<ffffffff81030667>] ? warn_slowpath_fmt+0x40/0x45
 [<ffffffffa016916c>] ? radeon_fence_signaled+0x61/0x90 [radeon]
 [<ffffffffa01691df>] ? radeon_fence_wait+0x44/0x300 [radeon]
 [<ffffffffa017aaae>] ? radeon_ib_get+0x124/0x1ec [radeon]
 [<ffffffffa017b78e>] ? radeon_cs_ioctl+0x0/0x19e [radeon]
 [<ffffffffa017b82e>] ? radeon_cs_ioctl+0xa0/0x19e [radeon]
 [<ffffffffa00f9970>] ? drm_ioctl+0x21a/0x300 [drm]
 [<ffffffff8109e622>] ? vfs_ioctl+0x23/0x93
 [<ffffffff8109eb69>] ? do_vfs_ioctl+0x454/0x498
 [<ffffffff81093549>] ? vfs_read+0x133/0x148
 [<ffffffff8109ebe9>] ? sys_ioctl+0x3c/0x5c
 [<ffffffff81001e6b>] ? system_call_fastpath+0x16/0x1b
---[ end trace 9376ac3cb6e5c7dc ]---

So additional to the Invalid command stream errors come now the radeon fence warnings.
Comment 6 Jérôme Glisse 2010-02-17 21:40:30 UTC
Possible trigger, haven't tested yet
gtkperf -c 500 and then choose the GtkDrawingArea - Circles test
Comment 7 Johannes Hirte 2010-02-18 01:41:43 UTC
(In reply to comment #6)
> Possible trigger, haven't tested yet
> gtkperf -c 500 and then choose the GtkDrawingArea - Circles test

Can confirm this, gtkperf triggers this error/warning reliable.
Comment 8 Johannes Hirte 2010-02-20 00:35:48 UTC
seems to be fixed in the xf86-video-ati driver in commit a3b730eceb522c7ac1ef3dd6f6c7d773118d03f7
Comment 9 Matthias Treydte 2010-02-20 09:58:35 UTC
Same story here, except that I have a HD4850 card. Problems are gone as of

>
> http://cgit.freedesktop.org/xorg/driver/xf86-video-ati/commit/?id=a3b730eceb522c7ac1ef3dd6f6c7d773118d03f7
Comment 10 Jérôme Glisse 2010-02-20 12:23:22 UTC
Sorry i forgot to update this bug, yes the mentioned commit fix the userspace bug the kernel was complaining about. I am closing this bug.

Note You need to log in before you can comment on or make changes to this bug.