I've been experimenting with xfs on gentoo-ppc64 (hell SUSE does it, it can't be *THAT* difficult ;-) and I've noted that there are numerous I/O errors that occur while using XFS. I have created a bug with the xfs team at SGI (http://oss.sgi.com/bugzilla/show_bug.cgi?id=406) and I will continue to monitor from this side, reporting updates to this defect as neccassary. Reproducible: Always Steps to Reproduce:
I'm assisting with the gentoo-ppc64 (testing on IBM POWER5) port. In testing XFS I've found that XFS dies under moderate load. I noticed a number of I/O errors during an rsync operatation. This prompted me to unmount and run xfs_check which reported a journal needed to be commited. I umounted/remounted and got a kernel panic. Other fs's (ext2/3,reiserfs, jfs) seem to work without issue. I'm a little stumped and would like some advice as to how to continue. ------- Additional Comment #1 From Nathan Scott 2005-04-10 16:18 PDT ------- Whats your kernel version there? And what does "dies" mean in this context - kernel panic? Or just the I/O errors? Can you give more details of these I/O errors too - are they being reported by the kernel or by rsync? | This prompted me to unmount and run xfs_check which reported a journal | needed to be commited That should never happen after a clean unmount... do you see the same issues with a non-patched kernel.org kernel? You using a non-patched gcc there? thanks. ------- Additional Comment #2 From Omkhar Arasaratnam 2005-04-11 20:17 PDT ------- |Whats your kernel version there? I've tried on 2.6.9+Gentoo patches. I will try 2.6.11.6 in a couple of days (more on this below) |And what does "dies" mean in this context - kernel panic? Or just the I/O |errors? Can you give more details of these I/O errors too - are they being |reported by the kernel or by rsync? both the kernel (dmesg) and rsync (stderr) reported I/O errors. No IMMEDIATE kernel panic, but the kernel did panic after I attempted to remount the fs. Each subsequent time the server would reboot it would hang when attempting to mount the xfs filesystem - but no panic || This prompted me to unmount and run xfs_check which reported a journal || needed to be commited |That should never happen after a clean unmount... do you see the same issues |with a non-patched kernel.org kernel? You using a non-patched gcc there? The gcc is patched (I'm not sure that I know of any distro that used vanilla gcc... That said here is my idea: I'm in the middle of spinning up a couple of release discs over the next couple of days - once this is complete I will try reverting the filespace to the xfs system. Are there any special flags I should use with mkfs.xfs ? ------- Additional Comment #3 From Eric Sandeen 2005-04-11 21:08 PDT ------- can you paste in the "I/O" errors you're seeing? Some actual description of how xfs "died" would be a good place to start. dmesg messages would be most useful. Thanks, -Eric ------- Additional Comment #4 From Omkhar Arasaratnam 2005-04-12 12:44 PDT ------- I began testing again today, I started by zeroing the partition and using 2.6.11.6 (instead of the 2.6.9-gentoo kernel). Things seem to be working well thus far - will keep the bug open while I test with a couple of more kernels - thanks. ------- Additional Comment #5 From Omkhar Arasaratnam 2005-04-12 18:46 PDT ------- Latest Gentoo versioned kernel works perfectly after some intense testing - closing out (that was an easy one eh ?) ;-)
So - can we mask this for everything before 2.6.11 on ppc64?
no we can't, because 2.6.9 is the only kernel my Apple Display works correctly. benh is working on that (I hope so...)
Created attachment 56873 [details, diff] ibm_use-genkernel-3.1.6.patch
ignore patch silly bugzilla advaced to the wrong boog again
2.6.11 has been stable for a while now. I mark this bug as FIXED, as everyone should run latest stable kernels anyways. ;-)