On 16/10/2018 05:28, Martin K. Petersen wrote:
John,
As I mentioned in the thread that spawned from this, we actually can't
expose multiple hw queues at the moment. And, if we did, we find a
performance drop due to having to go back to manage this IPTT
internally.
So how to handle? We're goin
John,
> As I mentioned in the thread that spawned from this, we actually can't
> expose multiple hw queues at the moment. And, if we did, we find a
> performance drop due to having to go back to manage this IPTT
> internally.
>
> So how to handle? We're going to continue to work towards exposing
On Fri, Oct 12, 2018 at 10:02:57AM +0100, John Garry wrote:
> Hi Ming,
>
> > In theory, you still may generate and manage the IPTT in the LLDD by
> > simply ignoring rq->tag, meantime enabling SCSI_MQ with 16 hw queues.
> >
>
> Well at the moment we can't expose all 16 hw queues to upper layer a
On 11/10/2018 02:58, Martin K. Petersen wrote:
John,
Hi Martin,
However it does block us in future from enabling SCSI MQ in the driver.
We're going to remove the legacy I/O path so I'm not particularly keen
on merging something that's going in the opposite direction.
What I really meant
Hi Ming,
In theory, you still may generate and manage the IPTT in the LLDD by
simply ignoring rq->tag, meantime enabling SCSI_MQ with 16 hw queues.
Well at the moment we can't expose all 16 hw queues to upper layer
anyway, due to ordering restiction imposed by HW on LLDD. We have a plan
to
On Thu, Oct 11, 2018 at 03:07:33PM +0100, John Garry wrote:
> On 11/10/2018 14:32, Ming Lei wrote:
> > On Thu, Oct 11, 2018 at 02:12:11PM +0100, John Garry wrote:
> > > On 11/10/2018 11:15, Christoph Hellwig wrote:
> > > > On Thu, Oct 11, 2018 at 10:59:11AM +0100, John Garry wrote:
> > > > >
> > >
On 11/10/2018 14:32, Ming Lei wrote:
On Thu, Oct 11, 2018 at 02:12:11PM +0100, John Garry wrote:
On 11/10/2018 11:15, Christoph Hellwig wrote:
On Thu, Oct 11, 2018 at 10:59:11AM +0100, John Garry wrote:
blk-mq tags are always per-host (which has actually caused problems for
ATA, which is now
On Thu, Oct 11, 2018 at 02:12:11PM +0100, John Garry wrote:
> On 11/10/2018 11:15, Christoph Hellwig wrote:
> > On Thu, Oct 11, 2018 at 10:59:11AM +0100, John Garry wrote:
> > >
> > > > blk-mq tags are always per-host (which has actually caused problems for
> > > > ATA, which is now using its own
On 11/10/2018 11:15, Christoph Hellwig wrote:
On Thu, Oct 11, 2018 at 10:59:11AM +0100, John Garry wrote:
blk-mq tags are always per-host (which has actually caused problems for
ATA, which is now using its own per-device tags).
So, for example, if Scsi_host.can_queue = 2048 and Scsi_host.nr
On Thu, Oct 11, 2018 at 10:59:11AM +0100, John Garry wrote:
>
> > blk-mq tags are always per-host (which has actually caused problems for
> > ATA, which is now using its own per-device tags).
> >
>
> So, for example, if Scsi_host.can_queue = 2048 and Scsi_host.nr_hw_queues =
> 16, then rq tags a
On 11/10/2018 07:36, Christoph Hellwig wrote:
On Wed, Oct 10, 2018 at 09:58:25PM -0400, Martin K. Petersen wrote:
This is because the IPTT index must be a unique value per HBA. However,
if we switched to SCSI MQ, the block layer tag becomes unique per queue,
and not per HBA.
That doesn't sound
On Wed, Oct 10, 2018 at 09:58:25PM -0400, Martin K. Petersen wrote:
> > This is because the IPTT index must be a unique value per HBA. However,
> > if we switched to SCSI MQ, the block layer tag becomes unique per queue,
> > and not per HBA.
>
> That doesn't sound right.
blk-mq tags are always pe
John,
> However it does block us in future from enabling SCSI MQ in the driver.
We're going to remove the legacy I/O path so I'm not particularly keen
on merging something that's going in the opposite direction.
> This is because the IPTT index must be a unique value per HBA. However,
> if we
On 24/09/2018 16:06, John Garry wrote:
This patchset introduces mostly more minor/obscure bugfixes for the
driver.
Hi Martin,
Can you please consider merging this patchset?
Thanks,
John
Also included is an optimisation to use the block layer tag for the IPTT
indexing. This quite a nice opt
This patchset introduces mostly more minor/obscure bugfixes for the
driver.
Also included is an optimisation to use the block layer tag for the IPTT
indexing. This quite a nice optimisation as it means we don't have to
evaluate this in the driver - it was a bit of a bottle-neck.
However it does b
15 matches
Mail list logo
