[PATCH 0/4] dm-mpath: bug fixes around pg-init handling

[Kiyoshi Ueda <k-ueda@ct.jp.nec.com>]

Hi Alasdair and multipath developers,

I've started to work on removing the multipath internal I/O queue,
since request-based dm-core now has a generic queue and multipath
can use it instead of having its own queue.

As a first step, I'm separating pg-init handling from queued I/O
handling.  During code inspection for that work, I found some bugs
in the current pg-init handling codes.
This patch-set fixes those bugs (and implies preparations for
the work of removing the multipath internal queue).

I don't have any active-passive storage, so I've just done compile
and boot tests.  Please review, and test on your storages.

This patch-set was made on top of 2.6.33-rc6 + Alasdair's patches below:
  dm-mpath-pass-struct-pgpath-to-pg-init-done.patch
  dm-mpath-skip-activate_path-for-failed-paths.patch
  dm-mpath-Remove-suspended-flag-from-struct-multipath.patch

Summary of the patch-set:
  1/4: dm-mpath: don't clear m->queue_io until all activations complete
  2/4: dm-mpath: must wait for pg-init completion in postsuspend
  3/4: dm-mpath: separate pg-init handling from process_queued_ios()
  4/4: dm-mpath: move initial pg-init kick into __switch_pg()

 drivers/md/dm-mpath.c |  131 +++++++++++++++++++++++++++++++++++++-------------
 1 file changed, 97 insertions(+), 34 deletions(-)

Thanks,
Kiyoshi Ueda