This patch ** is against Christoph's core-for-3.17 branch
which includes scsi-mq V2.

The purpose of releasing it now is to facilitate testing of
scsi-mq . The fio user space program seems to be the tool
of choice for such tests. With fio and four (pseudo) devices
I have observed 1.2 M IOPS on my equipment. Rob Elliott has
better results than that.


ChangeLog:
   - add host_lock option whose default value is 0 which
     removes the host_lock around all queued commands
   - accept delay=-1 (_hi_) or -2 which use a tasklet to invoke
     the scsi_done callback into the mid-layer. The default
     is still delay=1 which uses a timer to delay 1 jiffy
   - wire .change_queue_depth and .change_queue_type
     functions to better simulate queueing in a modern LLD
   - add SCSI_DEBUG_OPT_Q_NOISE (0x200) mask to only produce
     debug output associated with queue full, plus from
     .change_queue_depth and .change_queue_type functions
   - add SCSI_DEBUG_OPT_ALL_TSF (0x400) mask which reports
     all queued_arr fulls at TASK_SET_FULL, otherwise
     SCSI_MLQUEUE_HOST_BUSY is returned
   - add SCSI_DEBUG_OPT_RARE_TSF (0x800) mask which works
     together with the every_nth option (> 0) to count
     occurrences of num_in_q==queue_depth. When every_nth
     is reached the victim (a command) yields TASK SET FULL
   - clean up many debug messages.


IMO the scsi-mq queue_depth handling doesn't break but does
not converge promptly (and in some cases seems to be steady
state unstable) in the face of TASK_SET_FULL statuses.
This can be seen with this sequence:
   # cd /sys/bus/pseudo/drivers/scsi_debug
   # echo 0x600 > opts
< then in another window 'tail -f /var/log/syslog'>
< then in yet another window run a fio test against scsi_debug>
   # echo 200 > max_queue
< if the tail shows nothing, try writing a lower value into max_queue>

max_queue starts at 576. Depending on the fio test, the new
value written into max_queue needed to prompt the TSF trip
point will vary. You will now it when you see it :-)


The PI and provisioning parts of scsi_debug have been written
by Martin Petersen and some of its bitmaps may be racy without
the host_lock. I'm hoping that when Martin has time, he will
re-visit that code.

** scsi_debug in the core-for-3.17 branch already has Hannes'
    64 bit LUN changes and Akinobu Mita's "allow huge transfer
    length for read/write commands". So expect some noise if
    this patch is applied to other trees (such as vanilla lk
    3.15.2). There is no real conflict but the patch command
    gets upset.

Signed-off-by: Douglas Gilbert <dgilbert@interlog.com>
Tested-by: Robert Elliott <elliott@hp.com>