Hi folks,
Just a heads up that the scsi-mq alpha branch has been updated to Jen's
latest linux-block/new-queue containing hch's recent blk-mq
improvements, along with Alexander's patch for the is_flush_fua +
queue_depth=1 bug.
The branch @ v3.12-rc3 is available here:
https://git.kernel.org/cgit/linux/kernel/git/nab/target-pending.git/log/?h=scsi-mq
The main changes include dropping all of the blk_mq_alloc_request() +
blk_mq_free_request() changes that are now unnecessary with hch's
improvements in place, along with other miscellaneous cleanups.
--nab
On Wed, Oct 09, 2013 at 12:12:51PM -0700, Nicholas A. Bellinger wrote:
> Just a heads up that the scsi-mq alpha branch has been updated to Jen's
> latest linux-block/new-queue containing hch's recent blk-mq
> improvements, along with Alexander's patch for the is_flush_fua +
> queue_depth=1 bug.
Oh, I have not expected it could be picked up. In fact I agree it needs
to be updated with what you spotted:
"Also, these extra increments should probably happen only when the passed
queue_depth == 1 && reserved_tags == 0."
What is the criteria for patches to include in your tree? I would suggest
to consider this one https://lkml.org/lkml/2013/8/9/90 if it fits.
--
Regards,
Alexander Gordeev
[email protected]
On Wed, 2013-10-09 at 21:46 +0200, Alexander Gordeev wrote:
> On Wed, Oct 09, 2013 at 12:12:51PM -0700, Nicholas A. Bellinger wrote:
> > Just a heads up that the scsi-mq alpha branch has been updated to Jen's
> > latest linux-block/new-queue containing hch's recent blk-mq
> > improvements, along with Alexander's patch for the is_flush_fua +
> > queue_depth=1 bug.
>
>
> Oh, I have not expected it could be picked up. In fact I agree it needs
> to be updated with what you spotted:
>
> "Also, these extra increments should probably happen only when the passed
> queue_depth == 1 && reserved_tags == 0."
Yes, of course. Fixing that up now in blk_mq_init_queue().
>
> What is the criteria for patches to include in your tree? I would suggest
> to consider this one https://lkml.org/lkml/2013/8/9/90 if it fits.
>
Since it's libata related, I'd like Tejun (CC'ed) to be comfortable with
the approach first..
As I'm still using IDE for my rootfs with scsi-mq, I'd be fine with
applying it for AHCI testing purposes if there are no objections.
Care to send out an updated version..?
--nab
> What is the criteria for patches to include in your tree? I would suggest
> to consider this one https://lkml.org/lkml/2013/8/9/90 if it fits.
For this one you probably want to send a patch to Jens to move blk-mq-tag.h
to include/linux first instead of doing the relative include hack.
Also the blk_mq_*tag* routines you use aren't exported, so a modular build of
the driver with that patch applied will fail.
On Thu, Oct 10, 2013 at 09:17:14PM +0200, Christoph Hellwig wrote:
> > What is the criteria for patches to include in your tree? I would suggest
> > to consider this one https://lkml.org/lkml/2013/8/9/90 if it fits.
>
> For this one you probably want to send a patch to Jens to move blk-mq-tag.h
> to include/linux first instead of doing the relative include hack.
I wonder, if blk-mq- prefix should remain?.. This s code seems pretty much
generic to me.
> Also the blk_mq_*tag* routines you use aren't exported, so a modular build of
> the driver with that patch applied will fail.
Yep, makes sense. Thanks for pointing out.
--
Regards,
Alexander Gordeev
[email protected]
On Thu, Oct 10, 2013 at 09:32:41PM +0200, Alexander Gordeev wrote:
> I wonder, if blk-mq- prefix should remain?.. This s code seems pretty much
> generic to me.
Seems like nowdays you should just use the percpu-ida allocator directly,
Shaohua Li hast just sent patches to switch blk-mq over to it as well.
On 10/11/2013 06:19 AM, Christoph Hellwig wrote:
> On Thu, Oct 10, 2013 at 09:32:41PM +0200, Alexander Gordeev wrote:
>> I wonder, if blk-mq- prefix should remain?.. This s code seems pretty much
>> generic to me.
>
> Seems like nowdays you should just use the percpu-ida allocator directly,
> Shaohua Li hast just sent patches to switch blk-mq over to it as well.
Yeah, that was on the list of things to do and then test.
--
Jens Axboe