On 11/16/20 10:01 AM, Bart Van Assche wrote:

On 11/16/20 9:17 AM, Christoph Hellwig wrote:

quoted

On Sun, Nov 15, 2020 at 07:04:54PM -0800, Bart Van Assche wrote:

quoted

Do not modify sdev->request_queue. Remove the sdev->request_queue
assignment. That assignment is superfluous because scsi_mq_alloc_queue()
only has one caller and that caller calls scsi_mq_alloc_queue() as
follows:

    sdev->request_queue = scsi_mq_alloc_queue(sdev);

This looks ok to me.  But is there any good to keep scsi_mq_alloc_queue
around at all?  It is so trivial that it can be open coded in the
currently only caller, as well as a new one if added.

Hi Christoph,

A later patch in this series introduces a second call to
scsi_mq_alloc_queue(). Do we really want to have multiple functions that
set QUEUE_FLAG_SCSI_PASSTHROUGH? I'm concerned that if the logic for
creating a SCSI queue would ever be changed that only the copy in the
SCSI core would be updated but not the copy in the SPI code.

(replying to my own email)

Hi Christoph,

Is this something that you feel strongly about? I can make this change
but that would require reaching out again to someone who owns an SPI
setup for testing this patch series since I do not own an SPI setup
myself ...

Thanks,

Bart.