xfs
[Top] [All Lists]

Re: inconsistent lock state on 2.6.30?

To: Sage Weil <sage@xxxxxxxxxxxx>
Subject: Re: inconsistent lock state on 2.6.30?
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Fri, 26 Jun 2009 13:35:58 -0400
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
In-reply-to: <Pine.LNX.4.64.0906241438100.24641@xxxxxxxxxxxxxxxxxx>
References: <Pine.LNX.4.64.0906230925420.20462@xxxxxxxxxxxxxxxxxx> <20090623170844.GA23971@xxxxxxxxxxxxx> <Pine.LNX.4.64.0906241438100.24641@xxxxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.18 (2008-05-17)
On Wed, Jun 24, 2009 at 02:40:42PM -0700, Sage Weil wrote:
> [ 7822.230090] =================================
> [ 7822.230208] [ INFO: inconsistent lock state ]
> [ 7822.230208] 2.6.30 #22
> [ 7822.230208] ---------------------------------
> [ 7822.230208] inconsistent {RECLAIM_FS-ON-W} -> {IN-RECLAIM_FS-W} usage.
> [ 7822.230208] kswapd0/290 [HC0[0]:SC0[0]:HE1:SE1] takes:
> [ 7822.230208]  (&(&ip->i_iolock)->mr_lock){++++?+}, at: [<ffffffff803af53a>] 
> xfs_ilock+0x27/0x79
> [ 7822.230208] {RECLAIM_FS-ON-W} state was registered at:
> [ 7822.230208]   [<ffffffff8025c297>] mark_held_locks+0x4d/0x6b
> [ 7822.230208]   [<ffffffff8025c35d>] lockdep_trace_alloc+0xa8/0xc3
> [ 7822.230208]   [<ffffffff80287c4e>] __alloc_pages_internal+0x6d/0x457
> [ 7822.230208]   [<ffffffff802a7eb6>] alloc_pages_current+0xbe/0xc6
> [ 7822.230208]   [<ffffffff80281c19>] grab_cache_page_write_begin+0x5e/0xa2
> [ 7822.230208]   [<ffffffff802d2aa5>] block_write_begin+0x3d/0xcf
> [ 7822.230208]   [<ffffffff803cb634>] xfs_vm_write_begin+0x25/0x27
> [ 7822.230208]   [<ffffffff802825ba>] generic_file_buffered_write+0x139/0x2ff
> [ 7822.230208]   [<ffffffff803d172a>] xfs_write+0x4de/0x717

That's actually a different but slightly related one.


But thinking about I came to the conclusion that both the previous and
this one actually are false positives

Both of them actually fit into the earlier reports of i_lock dependencies
inside the inode relcaim path causing problems for normal runtime use
of the fs.  But unlike the previous mmap path where we really have
exclusive lock chains I'm not so sure about this one.  Give me some time
to sort out the reclaim path which I'll need to do anyway for the
various nfs-related issues hitting the list.

<Prev in Thread] Current Thread [Next in Thread>