2.5.75 does not boot - TCQ oops
Okay, I figured out some more things.
I split my root fs in half and made two identical copies - one on reiser and one on xfs. I compiled a bunch of kernels, and tested some more.
A TCQ enabled kernel with AS, queue depth of 32 works fine on both reiser and xfs. A TCQ enabled kernel with deadline, queue depth of 32 works fine on both reiser and xfs. Note the depth of 32. I had TCQ enabled on previous <74 kernels and it worked fine, everywhere with depth 32.
However, the kernel that crashed was using the default
(The default is 8, even though the comment says 32).
I tested that kernel again with reiserfs, TCQ, the two elevators, and queue depth 8 - on-boot fsck detects corruption every time, marks the system unclean, and requires --rebuild-tree or --fix-fixable on any further mounts. I now get "Wrong amount of used blocks." message.
Have not tested depth 8 TCQ kernel with xfs, since that's my surviving root fs, and I'd like to avoid corruption there.
Have not tested queue depths other than 8 and 32.
I could test some more on reiser now that I have a backup root fs.
On Sat, 12 Jul 2003 06:46:01 -0400
Ivan Gyurdiev <[email protected]> wrote:
[...]
Quoted 2 lines
However, the kernel that crashed was using the default (The default is 8, even though the comment says 32).
Some nice testing you've done there. Yeah, the 'map vs reality' is confusing when configuring. I previously changed the 8 to 32, following the help, but for 2.5.75 changed it to 8 based on an offhand comment on lkml saying: '8 has been tested as optimal'.
Hope LinuxTag isn't drawing all the developers attention away from your problem!
Mvh
Mats Johannesson