blk_flush_plug in the MD code, was Re: [xfs-masters] linux-next: manual merge of the xfs tree with Linus' tree

On Wed, Mar 30, 2011 at 01:43:55PM +0200, Jens Axboe wrote:

On 2011-03-29 22:08, Christoph Hellwig wrote:

quoted

FYI, the blk_flush_plug call in ufs_truncate also seem to be almost
certainly incorrect as it's followed by a yield.

That only leaves the RAID code as remaining modular users of it, and
I suspect it really is something that shouldn't need to be exported.

Agree, once things have settled down we can reap the remaining and keep
it internal.

In fact I think the blk_flush_plug calls in the raid1/raid10 code can
simply be removed without a replacement.  They are in wait_even_loops
which schedule before/after the call so we do get the implicit context
switch plugs there.  Even more they are generally called from the
make_request handler, which for MD is in the loop around
__generic_make_request and thus can't actually unplug in a sensible
way.