Re: [PATCH] reworked IDE/general tagged command queueing

[Jens Axboe <axboe@suse.de>]

On Wed, May 01 2002, Linus Torvalds wrote:
> 
> 
> On Wed, 1 May 2002, Jens Axboe wrote:
> >
> > I've rewritten parts of the IDE TCQ stuff to be, well, a lot better in
> > my oppinion. I had to accept that the ata_request and rq->special usage
> > sucked, it was just one big mess.
> 
> Looks good.
> 
> I would have one more comment - it would be wonderful if somebody else who
> was using tagged commands were to try to use this interface too, just to
> verify that it doesn't have any major problems.
> 
> I'll be happy to merge the generic tag stuff even without that (on the
> assumption that whatever deficiencies can be fixed later anyway), but it
> would be wonderful to have some kind of validation of the genericity.
> 
> In particular, on an IDE TCQ device we expect the queue depth to be fixed,
> but the SCSI subsystem considers queue depth to be only approximate
> (because the drive internally might split a command or something, I don't
> know). So I wonder how well this interface would interact with the
> QUEUE_FULL handling that the SCSI layer does.
> 
> Maybe it doesn't matter, and maybe some blk_queue_invalidate_tags() magic
> could do the right thing, but I'd like to get some kind of idea of whether
> others can use it well..
> 
> (If not the SCSI layer, maybe somebody could look at DAC960 or cciss and
> see whether they would seem to like the tag interface).

Yeah, that's why I said the interface was simple on purpose right now.
I'm sure it's extendable for some sort of adaptive queueing as well, I
will try to provide some examples for that. cciss will eat a lot more
commands than you think, it typically has a bigger internal queue depth
than the block layer :-). If not dac960, then SCSI is a possibility.

-- 
Jens Axboe