On 25/06/2020 15:01, Hannes Reinecke wrote:
Quite some 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@xxxxxxx>
---
drivers/scsi/scsi_lib.c | 10 +++++++++-
include/scsi/scsi_host.h | 11 +++++++++++
2 files changed, 20 insertions(+), 1 deletion(-)
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index bd378e1bd3fc..a752806af70b 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -1887,7 +1887,9 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost)
else
tag_set->ops = &scsi_mq_ops_no_commit;
tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
- tag_set->queue_depth = shost->can_queue;
+ tag_set->queue_depth =
+ shost->can_queue + shost->nr_reserved_cmds;
+ tag_set->reserved_tags = shost->nr_reserved_cmds;
tag_set->cmd_size = cmd_size;
tag_set->numa_node = NUMA_NO_NODE;
tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
@@ -1910,6 +1912,9 @@ void scsi_mq_destroy_tags(struct Scsi_Host *shost)
* @op_flags: request allocation flags
*
* Allocates a SCSI command for internal LLDD use.
+ * If 'nr_reserved_commands' is spectified by the host the
/s/spectified/specified/
I suppose making this usable by drivers which don't set
nr_reserved_commands is helpful (if not slightly strange). In that we
can move drivers which don't use rq->tag currently - like pm8001 - to
use it, and not have to worry about guessing some nr_reserved_commands
to specify.
+ * command will be allocated from the reserved tag pool;
+ * otherwise the normal tag pool will be used.
*/
struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
int data_direction, int op_flags)
@@ -1919,6 +1924,9 @@ struct scsi_cmnd *scsi_get_internal_cmd(struct scsi_device *sdev,
blk_mq_req_flags_t flags = 0;
unsigned int op = REQ_INTERNAL | op_flags;
+ if (sdev->host->nr_reserved_cmds)
+ flags = BLK_MQ_REQ_RESERVED; > +
op |= (data_direction == DMA_TO_DEVICE) ?
REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN;
rq = blk_mq_alloc_request(sdev->request_queue, op, flags);
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
index 46ef8cccc982..b94938e87e3a 100644
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -590,6 +590,11 @@ struct Scsi_Host {
unsigned short max_cmd_len;
int this_id;
+
+ /*
+ * Number of commands this host can handle at the same time.
+ * This excludes reserved commands as specified by nr_reserved_cmds.
+ */
int can_queue;
short cmd_per_lun;
short unsigned int sg_tablesize;
@@ -606,6 +611,12 @@ struct Scsi_Host {
* is nr_hw_queues * can_queue.
If we are going to go this way, then could you update the comment on
Scsi_Host.nr_hw_queues here? Currently we have "it is assumed that each
hardware queue has a depth of can_queue..."
Thanks,
John
*/
unsigned nr_hw_queues;
+
+ /*
+ * Number of reserved commands to allocate, if any.
+ */
+ unsigned nr_reserved_cmds;
+
