IDE DMA timeouts and reiserfs stability
Hi,
I am using 2.4.5-pre1. Over the course of the last two weeks I have had DMA timeouts occur twice. Both times corrupted my fs. While this is not ideal, its not unexpected as things stand now. I have seen at least three other reports on lkml about errors of this type - suspect that 2.4's ide is a little fragile in some corner cases...
Contrary to normal practice, after an IO error and fsck is a very wise thing to do. Can we automate this process. ie can reiserfs detect that it has experienced IO error(s) and set fsck required bits (two bits) in the SB? It would also be nice to be able to manually set these bits. This way a script could be triggered at boot (from initrd for those of us with reiserfs boot disks) to do something like this
reiserfsck -a this should check each FS an does a --check when the bits are set
to 01. It changes the SB bits as follows (logging to <dev>01.log) 00 - fs is ok
10 - fix-fixable run required (logging to <dev>10.log) 11 - rebuild-tree required (logging to <dev>11.log)
writing those logs could be a bit of a catch 22...
then the script would call reiserfsck -a again to do the work (if required) and ask if its ok to do the fix-fixable or rebuild-tree
Think reiserfsck is getting good enough for this, and it would probably avoid many of the problem currently popping up on the list.
Thoughts?
Ed Tomlinson <[email protected]>
PS. Chris, with the fix you supplied for LVM, snapshots work 100% of the time when I put them on hda or hde and fail 100% on hdg...
Quoted 4 lines
DMA timeouts occur twice. Both times corrupted my fs. While this is not ideal, its not unexpected as things stand now. I have seen at least three other reports on lkml about errors of this type - suspect that 2.4's ide is a little fragile in some corner cases...
The DMA handling retry code seems to be rather broken. We also have unexplained traces of IDE recursing which would corrupt the sg tables and make rather a mess of the IO
So yes, there is some debugging work needed on the current IDE layer
Quoted 7 lines · Ed Tomlinson
Hi,
I am using 2.4.5-pre1. Over the course of the last two weeks I have had DMA timeouts occur twice. Both times corrupted my fs. While this is not ideal, its not unexpected as things stand now. I have seen at least three other reports on lkml about errors of this type - suspect that 2.4's ide is a little fragile in some corner cases...
Out of curiousity, can you reproduce the problem in a 2.4 kernel with Andre Hedrick's IDE patch for 2.4 applied?
It works for me, but that's just me. (ftp://ftp.kernel.org/pub/linux/kernel/people/hedrick/)
Signature · 3 lines
Chris Dukes The more I administer Solaris the more I believe it deserves the reputation HPUX has and vice versa. -- me.