Created attachment 46657 [details]
Sysrq-t

See avocado.suse.de. 

Looks like a problem in the ext3 journaling code. 
Not sure if Jan is back already, he was in China this week. Chris, will 
you assign please? 

this happens also with reiser. on mac.suse.de, single cpu g5. I cant switch
console anymore, mouse moves still.

Actually everything seems to be waiting for kjournald to finish. kjournald trace
looks like:
kjournald     D c0000000000145a8 11264  3672      1          3993  2752 (L-TLB)
Call Trace:
[c00000007e27f670] [c00000007e27f740] 0xc00000007e27f740 (unreliable)
[c00000007e27f840] [c000000000011b10] .__switch_to+0xd0/0x160
[c00000007e27f8d0] [c0000000003d44a0] .schedule+0x590/0xe90
[c00000007e27fa10] [c0000000003d4df0] .io_schedule+0x50/0x90
[c00000007e27faa0] [c0000000000cd988] .sync_buffer+0x68/0x80
[c00000007e27fb20] [c0000000003d5788] .__wait_on_bit+0xc8/0x140
[c00000007e27fbd0] [c0000000003d5ad0] .out_of_line_wait_on_bit+0x90/0xc0
[c00000007e27fcb0] [c0000000000cc9b0] .__wait_on_buffer+0x30/0x50
[c00000007e27fd30] [c00000000018d254] .journal_commit_transaction+0x6a4/0x1580
[c00000007e27fe80] [c00000000019102c] .kjournald+0x11c/0x290
[c00000007e27ff90] [c0000000000145a8] .kernel_thread+0x4c/0x68

So kjournald seems to be waiting for IO completion and that never happens... I
guess Jens should know more about this ;) Also the fact that it happens both on
ext3 and reiser seems to confirm that it's probably a problem below a filesystem
layer.

Which vanilla kernel did you test (and did you use barriers there as well, we
put them on by default)?

In any case, try and reproduce with -o barrier=none (or barrier=0 for ext3) as
well, thanks!

I just build all -rc* patches without suse patches. I havent tried to narrow it
further, G5 is busy with other testing.

avocado currently runs 2.6.13-rc6 with the minimal set of required patches. 

When mounting with barrier=0 the problem does not occur any more. 

Andreas, can you try mounting with barriers again and do:

# echo 2 > /sys/block/sda/queue/iosched/max_depth
# echo 2 > /sys/block/sdb/queue/iosched/max_depth

and see if it still hangs?

That appears to be working as well. 

Thanks for testing, I will change the max_depth default to 2 then.

Marking as duplicate, no need to track two different bugs for the same issue.

*** This bug has been marked as a duplicate of 106103 ***