Re: [PATCH 2/5] xfs: use per-filesystem I/O completion workqueues

[Dave Chinner <david@xxxxxxxxxxxxx>]

On Tue, Nov 08, 2011 at 03:56:16AM -0500, Christoph Hellwig wrote:
> The new concurrency managed workqueues are cheap enough that we can create
> per-filesystem instead of global workqueues.  This allows us to remove the
> trylock or defer scheme on the ilock, which has the potential of delaying
> size updates, and is not helpful once we start to log the inode size.

So why does the per-mount workqueues allow removal of the trylock?

IOWs, it might be worthwhile pointing to the commit (77d7a0c "xfs:
Non-blocking inode locking in IO completion") to indicate that the
functionality being removed was introduced to avoid IO completion
deadlocks between dependent filesystems (e.g. XFS on loop device on
XFS) and that moving to per-mount completion queues removes that
dependency....

Cheers,

Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx