Message ID | 20200430131904.5847-2-hare@suse.de (mailing list archive) |
---|---|
State | Changes Requested |
Headers | show |
Series | scsi: enable reserved commands for LLDDs | expand |
On 30/04/2020 14:18, Hannes Reinecke wrote: > From: Hannes Reinecke <hare@suse.com> > > Quite a lot of drivers are using management commands internally, which > typically use the same hardware tag pool (ie they are being allocated > from the same hardware resources) as the 'normal' I/O commands. > These commands are set aside before allocating the block-mq tag bitmap, > so they'll never show up as busy in the tag map. > The block-layer, OTOH, already has 'reserved_tags' to handle precisely > this situation. > So this patch adds a new field 'nr_reserved_cmds' to the SCSI host > template to instruct the block layer to set aside a tag space for these > management commands by using reserved_tags. > > Signed-off-by: Hannes Reinecke <hare@suse.com> It may be worth adding this field to scsi_host_template. And we should also prob mention this in Documentation/scsi/scsi_mid_low_api.txt Apart from that, thanks: Reviewed-by: John Garry <john.garry@huawei.com> > --- > drivers/scsi/scsi_lib.c | 1 + > include/scsi/scsi_host.h | 6 ++++++ > 2 files changed, 7 insertions(+) > > diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c > index 47835c4b4ee0..5358f553f526 100644 > --- a/drivers/scsi/scsi_lib.c > +++ b/drivers/scsi/scsi_lib.c > @@ -1885,6 +1885,7 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost) > shost->tag_set.ops = &scsi_mq_ops_no_commit; > shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1; > shost->tag_set.queue_depth = shost->can_queue; > + shost->tag_set.reserved_tags = shost->nr_reserved_cmds; > shost->tag_set.cmd_size = cmd_size; > shost->tag_set.numa_node = NUMA_NO_NODE; > shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; > diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h > index 822e8cda8d9b..37bb7d74e4c4 100644 > --- a/include/scsi/scsi_host.h > +++ b/include/scsi/scsi_host.h > @@ -599,6 +599,12 @@ struct Scsi_Host { > * is nr_hw_queues * can_queue. > */ > unsigned nr_hw_queues; > + > + /* > + * Number of commands to reserve, if any. > + */ > + unsigned nr_reserved_cmds; > + > unsigned active_mode:2; > unsigned unchecked_isa_dma:1; > >
On 4/30/20 4:15 PM, John Garry wrote: > On 30/04/2020 14:18, Hannes Reinecke wrote: >> From: Hannes Reinecke <hare@suse.com> >> >> Quite a lot of drivers are using management commands internally, which >> typically use the same hardware tag pool (ie they are being allocated >> from the same hardware resources) as the 'normal' I/O commands. >> These commands are set aside before allocating the block-mq tag bitmap, >> so they'll never show up as busy in the tag map. >> The block-layer, OTOH, already has 'reserved_tags' to handle precisely >> this situation. >> So this patch adds a new field 'nr_reserved_cmds' to the SCSI host >> template to instruct the block layer to set aside a tag space for these >> management commands by using reserved_tags. >> >> Signed-off-by: Hannes Reinecke <hare@suse.com> > > It may be worth adding this field to scsi_host_template. And we should > also prob mention this in Documentation/scsi/scsi_mid_low_api.txt > Right, indeed, will be doing so. Haven't done it as long as this is still an RFC; guess when we'll get the SAS bits sorted (hint, hint :-) and no further objections are coming wrt the overall design I'll be sending out a 'real' patchset with the documentation bits sorted, too. > Apart from that, thanks: > > Reviewed-by: John Garry <john.garry@huawei.com> > Thanks for the review! Cheers, Hannes
On 2020-04-30 06:18, Hannes Reinecke wrote: > From: Hannes Reinecke <hare@suse.com> > > Quite a lot of drivers are using management commands internally, which > typically use the same hardware tag pool (ie they are being allocated > from the same hardware resources) as the 'normal' I/O commands. > These commands are set aside before allocating the block-mq tag bitmap, > so they'll never show up as busy in the tag map. > The block-layer, OTOH, already has 'reserved_tags' to handle precisely > this situation. > So this patch adds a new field 'nr_reserved_cmds' to the SCSI host > template to instruct the block layer to set aside a tag space for these > management commands by using reserved_tags. Reviewed-by: Bart Van Assche <bvanassche@acm.org>
On Thu, Apr 30, 2020 at 03:18:24PM +0200, Hannes Reinecke wrote: > diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c > index 47835c4b4ee0..5358f553f526 100644 > --- a/drivers/scsi/scsi_lib.c > +++ b/drivers/scsi/scsi_lib.c > @@ -1885,6 +1885,7 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost) > shost->tag_set.ops = &scsi_mq_ops_no_commit; > shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1; > shost->tag_set.queue_depth = shost->can_queue; > + shost->tag_set.reserved_tags = shost->nr_reserved_cmds; Insteda of just passing through the value I think we should remove them from can_queue here - as seen in the few patches the current behavior of summing both up seems to cause a fair amount of confusion. Also I'd merge this with the patch to actually allocate reserved command, as that is one actual unit of useful functionality.
On 5/1/20 7:48 PM, Christoph Hellwig wrote: > On Thu, Apr 30, 2020 at 03:18:24PM +0200, Hannes Reinecke wrote: >> diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c >> index 47835c4b4ee0..5358f553f526 100644 >> --- a/drivers/scsi/scsi_lib.c >> +++ b/drivers/scsi/scsi_lib.c >> @@ -1885,6 +1885,7 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost) >> shost->tag_set.ops = &scsi_mq_ops_no_commit; >> shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1; >> shost->tag_set.queue_depth = shost->can_queue; >> + shost->tag_set.reserved_tags = shost->nr_reserved_cmds; > > Insteda of just passing through the value I think we should remove > them from can_queue here - as seen in the few patches the current > behavior of summing both up seems to cause a fair amount of > confusion. > Yes, I would very much prefer that. > Also I'd merge this with the patch to actually allocate reserved > command, as that is one actual unit of useful functionality. > Right, will do. Cheers, Hannes
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index 47835c4b4ee0..5358f553f526 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -1885,6 +1885,7 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost) shost->tag_set.ops = &scsi_mq_ops_no_commit; shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1; shost->tag_set.queue_depth = shost->can_queue; + shost->tag_set.reserved_tags = shost->nr_reserved_cmds; shost->tag_set.cmd_size = cmd_size; shost->tag_set.numa_node = NUMA_NO_NODE; shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h index 822e8cda8d9b..37bb7d74e4c4 100644 --- a/include/scsi/scsi_host.h +++ b/include/scsi/scsi_host.h @@ -599,6 +599,12 @@ struct Scsi_Host { * is nr_hw_queues * can_queue. */ unsigned nr_hw_queues; + + /* + * Number of commands to reserve, if any. + */ + unsigned nr_reserved_cmds; + unsigned active_mode:2; unsigned unchecked_isa_dma:1;