Re: Processes receive SIGSEGV if TCQ is enabled

[Thomas Schlichter <schlicht@uni-mannheim.de>]

[-- Attachment #1: Type: text/plain, Size: 1577 bytes --]

Hi,

On Friday 31 October 2003 00:40, Nick Piggin wrote:
> Hi,
> If you're testing IDE TCQ, please try the following patch and use the
> default io scheduler. It won't fix anything, but it poisons requests
> so we can sometimes tell if they are being used in the wrong places.
> I have seen warnings that lead me to believe this might be happening.
> Its against 2.6.0-test9-mm1. Report any stack traces you see. Thanks.

OK, I tested 2.6.0-test9-mm1 + your patch, but it seems not to print any 
messages or stack traces, even if many processes are killed after setting TCQ 
depth to 1.

Today, however, I got reiserfs corruption messages in the logs, but 
fsck.reiserfs could not find any corruption on the next boot, so I think this 
is not a real corruption but just reading the wrong data...

The messages are something like:

Oct 31 07:57:53 bigboss kernel: is_tree_node: node level 2120 does not match 
to the expected one 1
Oct 31 07:57:53 bigboss kernel: vs-5150: search_by_key: invalid format found 
in block 642544. Fsck?
Oct 31 07:57:53 bigboss kernel: vs-13070: reiserfs_read_locked_inode: i/o 
failure occurred trying to find stat data of [4767268 4767269 0x0 SD]
Oct 31 07:57:53 bigboss kernel: is_tree_node: node level 2120 does not match 
to the expected one 1
Oct 31 07:57:53 bigboss kernel: vs-5150: search_by_key: invalid format found 
in block 642544. Fsck?
Oct 31 07:57:53 bigboss kernel: vs-13070: reiserfs_read_locked_inode: i/o 
failure occurred trying to find stat data of [4767268 4767269 0x0 SD]

> Nick

  Thomas

[-- Attachment #2: signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]