Re: 2.6.19-rc2-mm1

[Nick Piggin <nickpiggin@yahoo.com.au>]

Badari Pulavarty wrote:
> On Thu, 2006-10-19 at 03:32 +1000, Nick Piggin wrote:
> 
>>Badari Pulavarty wrote:
>>
>>>On Thu, 2006-10-19 at 01:48 +1000, Nick Piggin wrote:
>>>
>>>
>>>>Badari Pulavarty wrote:
>>>>
>>>>
>>>>>On Mon, 2006-10-16 at 23:06 -0700, Andrew Morton wrote:
>>>>>
>>>>>
>>>>>
>>>>>>ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.19-rc2/2.6.19-rc2-mm1/
>>>>>>
>>>>>>
>>>>>>- Added the hwmon and i2c trees to the -mm lineup.  These are quilt-style
>>>>>>trees, maintained by Jean Delvare.
>>>>>
>>>>>
>>>>>
>>>>>LTP writev tests seems to lockup the machine. reiserfs issue ?
>>>>
>>>>...
>>>>
>>>>
>>>>
>>>>>BUG: soft lockup detected on CPU#2!
>>>>>
>>>>>Call Trace:
>>>>><IRQ>  [<ffffffff8024a4ba>] softlockup_tick+0xfa/0x120
>>>>>[<ffffffff8022e10f>] __do_softirq+0x5f/0xd0
>>>>>[<ffffffff80232067>] update_process_times+0x57/0x90
>>>>>[<ffffffff80217e84>] smp_local_timer_interrupt+0x34/0x60
>>>>>[<ffffffff802185db>] smp_apic_timer_interrupt+0x4b/0x80
>>>>>[<ffffffff80326ce0>] __copy_user_nocache+0x20/0x150
>>>>>[<ffffffff8020a7e6>] apic_timer_interrupt+0x66/0x70
>>>>><EOI>  [<ffffffff802b8de0>] reiserfs_get_block+0x0/0x10c0
>>>>>[<ffffffff80295198>] __block_prepare_write+0x158/0x470
>>>>>[<ffffffff802b8de0>] reiserfs_get_block+0x0/0x10c0
>>>>>[<ffffffff802954ca>] block_prepare_write+0x1a/0x30
>>>>>[<ffffffff802b7cea>] reiserfs_prepare_write+0xca/0x140
>>>>>[<ffffffff8024e9d2>] generic_file_buffered_write+0x2b2/0x610
>>>>
>>>>This is likely to be a reiserfs interaction with the pagecache write
>>>>deadlock fixes. Chris Mason just now identified a couple of issues
>>>>and is going to work on a fix.
>>>>
>>>
>>>
>>>No. seems to be generic issue .. (happens with ext3 also) :(
>>
>>I think I may have missed a fix for ext3 ordered and journalled too
>>(I've just sent a patch to Andrew privately).
>>
>>Sorry. Can you try with ext2?
> 
> 
> 
> No luck with ext2 either ..

Hmm OK, it could be an issue with the way the iovecs are being walked. Maybe.
I wouldn't have thought that should fire the softlockup detector though.
It could walk a large number of iovec segments (4096) with preempt disabled,
but that's not 10s worth.

And it looks like it is getting stuck in commit_write (but the traces are
a bit wonky, could be some faults happening too).

Do you have CONFIG_DEBUG_VM turned on? (hmm, I notice fault_in_pages_readable
in that case should be have a 2nd argument of seglen rather than bytes, but
that shouldn't be causing your problem)

-- 
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com