[PATCH 0/3] drain_workqueue vs scsi_flush_work

[Dan Williams <dan.j.williams@intel.com>]

Hi Tejun, James,

While trying to use drain_workqueue() in libsas I hit the
WARN_ON_ONCE(!is_chained_work(wq)) in __queue_work().  However, as I
mention in patch-2 libsas expects that new unchained work can be
submitted during a drain.  So, patch-2 is a hack to accept new work when
drain_workqueue() is used publicly as a "stronger flush", while keeping
the same warn+drop behavior for internal usages of drain_workqueue() in
the destroy_workqueue() case.

Smoke tested on sas topoloogy of:
host-->ata
host-->expander-->ata
host-->expander-->expander-->ata

--
Dan


---

Dan Williams (3):
      workqueue: promote workqueue_lock to hard-irq safe
      workqueue: defer work to a draining queue
      scsi: use drain_workqueue

 drivers/scsi/hosts.c      |    8 ++-
 drivers/scsi/isci/host.c  |    3 -
 include/linux/workqueue.h |    3 +
 include/scsi/scsi_host.h  |    2 -
 kernel/workqueue.c        |  114 ++++++++++++++++++++++++++++++++-------------
 5 files changed, 89 insertions(+), 41 deletions(-)