Bring aac out from under Giant:

- the mutex aac_io_lock protects the main codepaths which handle queues and
  hardware registers.  Only one acquire/release is done in the top-half and
  the taskqueue.  This mutex also applies to the userland command path and
  CAM data path.
- Move the taskqueue to the new Giant-free version.
- Register the disk device with DISKFLAG_NOGIANT so the top-half processing
  runs without Giant.
- Move the dynamic command allocator to the worker thread to avoid locking
  issues with bus_dmamem_alloc().

This gives about 20% improvement in most of my benchmarks.

1 files changed, 4 insertions, 0 deletions

diff --git a/sys/dev/aac/aac_cam.c b/sys/dev/aac/aac_cam.c
index e1874be..b7e22b1 100644
--- a/sys/dev/aac/aac_cam.c
+++ b/sys/dev/aac/aac_cam.c
@@ -286,7 +286,9 @@ aac_cam_action(struct cam_sim *sim, union ccb *ccb)
 
 	/* Async ops that require communcation with the controller */
 
+	AAC_LOCK_ACQUIRE(&sc->aac_io_lock);
 	if (aac_alloc_command(sc, &cm)) {
+		AAC_LOCK_RELEASE(&sc->aac_io_lock);
 		xpt_freeze_simq(sim, 1);
 		ccb->ccb_h.status = CAM_REQUEUE_REQ;
 		xpt_done(ccb);
@@ -402,6 +404,8 @@ aac_cam_action(struct cam_sim *sim, union ccb *ccb)
 	aac_enqueue_ready(cm);
 	aac_startio(cm->cm_sc);
 
+	AAC_LOCK_RELEASE(&sc->aac_io_lock);
+
 	return;
 }