[Bug 621437] [NEW] libata stalling under heavy IO

[Ynot]

Public bug reported:

Lucid Server i386

uname -a
Linux tony-server 2.6.32-24-generic-pae #41-Ubuntu SMP Thu Aug 19 02:43:57 UTC 2010 i686 GNU/Linux

/proc/version_signature
Ubuntu 2.6.32-24.41-generic-pae 2.6.32.15+drm33.5


Under heavy IO load to a RAID5 array, I'm experiencing high iowait

Syslog reports:

Aug 20 21:44:17 tony-server kernel: [ 3902.000092] ata6.01: exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x6 frozen
Aug 20 21:44:17 tony-server kernel: [ 3902.013734] ata6.01: failed command: READ DMA
Aug 20 21:44:17 tony-server kernel: [ 3902.027159] ata6.01: cmd c8/00:20:3f:6b:6a/00:00:00:00:00/fb tag 0 dma 16384 in
Aug 20 21:44:17 tony-server kernel: [ 3902.027161]          res 40/00:ff:00:00:00/00:00:00:00:00/50 Emask 0x4 (timeout)
Aug 20 21:44:17 tony-server kernel: [ 3902.079963] ata6.01: status: { DRDY }
Aug 20 21:44:22 tony-server kernel: [ 3907.132036] ata6: link is slow to respond, please be patient (ready=0)
Aug 20 21:44:27 tony-server kernel: [ 3912.116037] ata6: device not ready (errno=-16), forcing hardreset
Aug 20 21:44:27 tony-server kernel: [ 3912.116053] ata6: soft resetting link
Aug 20 21:44:28 tony-server kernel: [ 3912.304511] ata6.00: configured for UDMA/33
Aug 20 21:44:28 tony-server kernel: [ 3912.320846] ata6.01: configured for UDMA/33
Aug 20 21:44:28 tony-server kernel: [ 3912.320859] ata6.01: device reported invalid CHS sector 0
Aug 20 21:44:28 tony-server kernel: [ 3912.320873] ata6: EH complete

These occur repeatedly every few minutes while RAID array is being written to
If I forcefully limit the client write speed to 500 KiB/s or less, problem goes away

RAID array is EXT4 exported over NFS, and all disk activity is initiated
by client

Searching around, I came across this
http://lkml.org/lkml/2010/3/9/345
(This is linux 2.6.34, but the errors look identical to me)

There's also a patch a few posts in from the above link

Quick look at the Lucid kernel git repo suggests this patch is not applied to the Lucid kernel
http://kernel.ubuntu.com/git?p=ubuntu/ubuntu-lucid.git;a=blob_plain;f=drivers/ata/libata-sff.c;hb=HEAD
(I hope I'm looking in the right tree...?)

ProblemType: Bug
DistroRelease: Ubuntu 10.04
Package: linux-image-2.6.32-24-generic-pae 2.6.32-24.41
Regression: Yes
Reproducible: No
ProcVersionSignature: Ubuntu 2.6.32-24.41-generic-pae 2.6.32.15+drm33.5
Uname: Linux 2.6.32-24-generic-pae i686
Architecture: i386
Date: Fri Aug 20 22:57:54 2010
ProcEnviron:
 LANG=en_GB.UTF-8
 SHELL=/bin/bash
SourcePackage: linux

** Affects: linux (Ubuntu)
     Importance: Undecided
         Status: New


** Tags: apport-bug i386 lucid

-- 
libata stalling under heavy IO
https://bugs.launchpad.net/bugs/621437
You received this bug notification because you are a member of Kernel
Bugs, which is subscribed to linux in ubuntu.