xwayland + radeon = consistent filesystem corruption Re: I'm the only one getting hard drive errors, right?

Tue Sep 4 19:37:47 PDT 2012

I reproduced the problem with chromium under X, not running wayland.  Woo.
Just by loading up chromium and playing back a youtube video for a while.

(After doing a long smartctl test and fsck of the partition.)

[ 4043.097706] EXT4-fs error (device sda1): ext4_ext_search_left:1275: inode #21374007: comm flush-8:0: ix (10742) != EXT_FIRST_INDEX (0) (depth 1)!
[ 4043.097711] Aborting journal on device sda1-8.
[ 4043.097873] EXT4-fs (sda1): Remounting filesystem read-only
[ 4043.097883] EXT4-fs error (device sda1) in ext4_da_writepages:3033: IO failure
[ 4043.098011] EXT4-fs (sda1): ext4_da_writepages: jbd2_start: 962 pages, ino 21374007; err -30

Same inode again.  Which I guess I didn't mention on this list turned out
to be a file in chromium's cache:

$ sudo find . -inum 21374007 -print
./home/darxus/.cache/chromium/Default/Cache/data_3

Still curious... wtf... but at least it's not wayland.  I'll continue this
on the ext4 list.

On 09/03, darxus at chaosreigns.com wrote:
> [732715.730069] EXT4-fs error (device sda1): ext4_ext_search_left:1275: inode #21374007: comm flush-8:0: ix (10742) != EXT_FIRST_INDEX (0) (depth 1)!
> [732715.730084] Aborting journal on device sda1-8.
> [732715.730269] EXT4-fs (sda1): Remounting filesystem read-only
> [732715.730278] EXT4-fs error (device sda1) in ext4_da_writepages:3033: IO failure
> [732715.730440] EXT4-fs (sda1): ext4_da_writepages: jbd2_start: 589 pages, ino 21374007; err -30
> 
> This hasn't happened in three months.  The last time I saw it was the
> last time I ran xwayland.  While correlation does not imply causation,
> and it *could* be a coincidence, I'm really not willing to entertain
> that possibility as realistic.
> 
> This time I used RAOF's X DDX (and updated the xwayland instructions
> for Radeon / ATI to use it).  Last time I was using timon37's DDX.  I don't
> know if they share code.  I don't know if they're at fault.
> 
> I was using the DRM backend.  I ran "make install" as a non-root user, and
> then set weston-launch as owned by root and +s, and ran weston-launch.  I
> did not have xserver set suid root.
> 
> I was playing a video on youtube in chromium vix xwayland when X crashed
> (taking firefox, the only other X client I was running, out with it.)
> And was chatting with folks in IRC about X's failure to respawn when I
> realized my filesystem had been remounted readonly.  Then dug the above
> output out of dmesg.
> 
> I was working on updating my "state of wayland" page to say that wayland
> was looking pretty usable now :/
> 
> fsck said lots of scary things after rebooting, I had to manually confirm
> it wanted to do many of them.  I have photos if anyone is interested
> in details.  Lots of "Free blocks count wrong for group... Fix(y)?" and
> "Free inodes count wrong for group... Fix(y)?"  A "Block bitmap
> differences..."
> 
> I don't know for sure if I lost anything, but have not yet
> seen evidence that I did.  I have pretty good backups.
> 
> 
> 12:15 < pq> either xwayland triggers some fs bug, or triggers a gfx driver
> bug, which then scribbles over kernel memory - or faulty hw. Can't know.
> 12:16 < soreau> either way, it's a fairly serious problem
> 
> I agree with this assessment.
> 
> 
> So far, I think it has only affected the filesystem I was using at the time
> (I basically only use one partition per linux install).  So I may be
> willing to do more testing on a dedicated testing partition.
> 
> This graphics card needed to go on ubuntu's grub gfxpayload blacklist,
> because for some reason retaining the graphics mode from grub to X breaks
> on some graphics cards, including this one.  Seems unlikely to be directly
> related, just trying to provide all possibly relevant info I have.  The bug
> for this was:  https://bugs.launchpad.net/ubuntu/+source/linux/+bug/971204
> 
> I was running an up to date ubuntu oneric install.  
> 
> lspci output:
> 
> 05:00.0 VGA compatible controller: Advanced Micro Devices [AMD] nee ATI Barts XT [Radeon HD 6800 Series] (prog-if 00 [VGA controller])
>         Subsystem: Hightech Information System Ltd. Device 2010
>         Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR+ FastB2B- DisINTx+
>         Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
>         Latency: 0, Cache Line Size: 64 bytes
>         Interrupt: pin A routed to IRQ 46
>         Region 0: Memory at d0000000 (64-bit, prefetchable) [size=256M]
>         Region 2: Memory at fbfc0000 (64-bit, non-prefetchable) [size=128K]
>         Region 4: I/O ports at e000 [size=256]
>         Expansion ROM at fbfa0000 [disabled] [size=128K]
>         Capabilities: <access denied>
>         Kernel driver in use: radeon
>         Kernel modules: radeon
> 
> 
> I had the latest get masters of everything as of 2012-09-03 08:36 -0400.  
> weston commit 8538b22ff4ad8879b4e3288be053508167562859
> wayland commit 2be6e0ed142bac669398a9ad26d336666fa53216
> raof's xf86-video-ati xwayland branch commit 8dc07e63eaf8909f7046bf746a119ec749352441
> 
> On 05/30, darxus at chaosreigns.com wrote:
> > I was just playing with weston master under X, and timon37's radeon ddx,
> > weston crashed, when I tried to delete the wayland lock I got an error
> > that the filesystem was readonly, dmesg said:
> > 
> > [  496.347230] EXT4-fs error (device sda1): ext4_ext_search_left:1275: inode #21374007: comm flush-8:0:ix (10742) != EXT_FIRST_INDEX (0) (depth 1)!
> > [  496.347236] Aborting journal on device sda1-8.
> > [  496.347383] EXT4-fs (sda1): Remounting filesystem read-only
> > 
> > Which is pretty scary.  
> > 
> > I feel like this *might* have happened before when I was playing with
> > weston, but I definitely don't have enough information to suggest there is
> > any real correlation.  Or knowledge of what exactly is going on here to
> > know if a correlation is even possible.
> > 
> > I had xserver set suid root, out of habit from before I fixed the bug
> > complaining about not getting master, which seems like it might have made
> > something like this possible.  
> > 
> > I do have good backups.  
> 
> -- 
> "If you would be a real seeker after truth, it is necessary that at
> least once in your life you doubt, as far as possible, all things."
> - Rene Descartes
> http://www.ChaosReigns.com
> _______________________________________________
> wayland-devel mailing list
> wayland-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/wayland-devel
> 

-- 
"When we remember we are all mad, the mysteries of life disappear and
life stands explained." - Mark Twain
http://www.ChaosReigns.com