xfs
[Top] [All Lists]

Re: [PATCH 4/5] xfs: use GFP_NOFS for page cache allocation

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 4/5] xfs: use GFP_NOFS for page cache allocation
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Fri, 16 Jul 2010 01:21:49 -0400
Cc: xfs@xxxxxxxxxxx
In-reply-to: <1279154300-2018-5-git-send-email-david@xxxxxxxxxxxxx>
References: <1279154300-2018-1-git-send-email-david@xxxxxxxxxxxxx> <1279154300-2018-5-git-send-email-david@xxxxxxxxxxxxx>
User-agent: Mutt/1.5.20 (2009-08-17)
On Thu, Jul 15, 2010 at 10:38:19AM +1000, Dave Chinner wrote:
> From: Dave Chinner <dchinner@xxxxxxxxxx>
> 
> Avoid a lockdep warning by preventing page cache allocation from
> recursing back into the filesystem during memory reclaim.

Yeah, I already have this in my queue too.  Noticed that we don't
need it because of i_mutex as suggested in the stack overflow thread,
but for the XFS ilock.

> @@ -1501,8 +1501,9 @@ xfs_vm_write_begin(
>       void                    **fsdata)
>  {
>       *pagep = NULL;
> -     return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
> -                                                             xfs_get_blocks);
> +     return block_write_begin(file, mapping, pos, len,
> +                              (flags | AOP_FLAG_NOFS),
> +                              pagep, fsdata, xfs_get_blocks);


No need for the bracing:

        return block_write_begin(file, mapping, pos, len, flags | AOP_FLAG_NOFS,
                                 pagep, fsdata, xfs_get_blocks);


And with my truncate rework in the vfs tree we'll get rejection galore in
linux-next, but Stephen has been pretty good at handling these..

Reviewed-by: Christoph Hellwig <hch@xxxxxx>

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