Merge patch series "scsi: target: Allow userspace to config cmd submission"

Mike Christie <michael.christie@oracle.com> says:

The following patches were made over Linus's tree but apply over
Martin's branches. They allow userspace to configure how fabric
drivers submit cmds to backend drivers.

Right now loop and vhost use a worker thread, and the other drivers
submit from the contexts they receive/process the cmd from. For
multiple LUN cases where the target can queue more cmds than the
backend can handle then deferring to a worker thread is safest because
the backend driver can block when doing things like waiting for a free
request/tag. Deferring also helps when the target has to handle
transport level requests from the recv context.

For cases where the backend devices can queue everything the target
sends, then there is no need to defer to a workqueue and you can see a
perf boost of up to 26% for small IO workloads. For a nvme device and
vhost-scsi I can see with 4K IOs:

fio jobs        1       2       4       8       10
--------------------------------------------------
workqueue
submit        94K     190K    394K    770K    890K

direct
submit       128K     252K    488K    950K    -

Link: https://lore.kernel.org/r/1b1f7a5c-0988-45f9-b103-dfed2c0405b1@oracle.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

1 files changed, 59 insertions, 57 deletions

diff --git a/drivers/target/target_core_transport.c b/drivers/target/target_core_transport.c
index 0686882bcbda..c81def3c96df 100644
--- a/drivers/target/target_core_transport.c
+++ b/drivers/target/target_core_transport.c
@@ -1576,17 +1576,39 @@ target_cmd_parse_cdb(struct se_cmd *cmd)
 }
 EXPORT_SYMBOL(target_cmd_parse_cdb);
 
-/*
- * Used by fabric module frontends to queue tasks directly.
- * May only be used from process context.
- */
-int transport_handle_cdb_direct(
-	struct se_cmd *cmd)
+static int __target_submit(struct se_cmd *cmd)
 {
 	sense_reason_t ret;
 
 	might_sleep();
 
+	/*
+	 * Check if we need to delay processing because of ALUA
+	 * Active/NonOptimized primary access state..
+	 */
+	core_alua_check_nonop_delay(cmd);
+
+	if (cmd->t_data_nents != 0) {
+		/*
+		 * This is primarily a hack for udev and tcm loop which sends
+		 * INQUIRYs with a single page and expects the data to be
+		 * cleared.
+		 */
+		if (!(cmd->se_cmd_flags & SCF_SCSI_DATA_CDB) &&
+		    cmd->data_direction == DMA_FROM_DEVICE) {
+			struct scatterlist *sgl = cmd->t_data_sg;
+			unsigned char *buf = NULL;
+
+			BUG_ON(!sgl);
+
+			buf = kmap_local_page(sg_page(sgl));
+			if (buf) {
+				memset(buf + sgl->offset, 0, sgl->length);
+				kunmap_local(buf);
+			}
+		}
+	}
+
 	if (!cmd->se_lun) {
 		dump_stack();
 		pr_err("cmd->se_lun is NULL\n");
@@ -1614,7 +1636,6 @@ int transport_handle_cdb_direct(
 		transport_generic_request_failure(cmd, ret);
 	return 0;
 }
-EXPORT_SYMBOL(transport_handle_cdb_direct);
 
 sense_reason_t
 transport_generic_map_mem_to_cmd(struct se_cmd *cmd, struct scatterlist *sgl,
@@ -1782,53 +1803,6 @@ generic_fail:
 EXPORT_SYMBOL_GPL(target_submit_prep);
 
 /**
- * target_submit - perform final initialization and submit cmd to LIO core
- * @se_cmd: command descriptor to submit
- *
- * target_submit_prep must have been called on the cmd, and this must be
- * called from process context.
- */
-void target_submit(struct se_cmd *se_cmd)
-{
-	struct scatterlist *sgl = se_cmd->t_data_sg;
-	unsigned char *buf = NULL;
-
-	might_sleep();
-
-	if (se_cmd->t_data_nents != 0) {
-		BUG_ON(!sgl);
-		/*
-		 * A work-around for tcm_loop as some userspace code via
-		 * scsi-generic do not memset their associated read buffers,
-		 * so go ahead and do that here for type non-data CDBs.  Also
-		 * note that this is currently guaranteed to be a single SGL
-		 * for this case by target core in target_setup_cmd_from_cdb()
-		 * -> transport_generic_cmd_sequencer().
-		 */
-		if (!(se_cmd->se_cmd_flags & SCF_SCSI_DATA_CDB) &&
-		     se_cmd->data_direction == DMA_FROM_DEVICE) {
-			if (sgl)
-				buf = kmap(sg_page(sgl)) + sgl->offset;
-
-			if (buf) {
-				memset(buf, 0, sgl->length);
-				kunmap(sg_page(sgl));
-			}
-		}
-
-	}
-
-	/*
-	 * Check if we need to delay processing because of ALUA
-	 * Active/NonOptimized primary access state..
-	 */
-	core_alua_check_nonop_delay(se_cmd);
-
-	transport_handle_cdb_direct(se_cmd);
-}
-EXPORT_SYMBOL_GPL(target_submit);
-
-/**
  * target_submit_cmd - lookup unpacked lun and submit uninitialized se_cmd
  *
  * @se_cmd: command descriptor to submit
@@ -1923,7 +1897,7 @@ void target_queued_submit_work(struct work_struct *work)
 			se_plug = target_plug_device(se_dev);
 		}
 
-		target_submit(se_cmd);
+		__target_submit(se_cmd);
 	}
 
 	if (se_plug)
@@ -1934,7 +1908,7 @@ void target_queued_submit_work(struct work_struct *work)
  * target_queue_submission - queue the cmd to run on the LIO workqueue
  * @se_cmd: command descriptor to submit
  */
-void target_queue_submission(struct se_cmd *se_cmd)
+static void target_queue_submission(struct se_cmd *se_cmd)
 {
 	struct se_device *se_dev = se_cmd->se_dev;
 	int cpu = se_cmd->cpuid;
@@ -1944,7 +1918,35 @@ void target_queue_submission(struct se_cmd *se_cmd)
 	llist_add(&se_cmd->se_cmd_list, &sq->cmd_list);
 	queue_work_on(cpu, target_submission_wq, &sq->work);
 }
-EXPORT_SYMBOL_GPL(target_queue_submission);
+
+/**
+ * target_submit - perform final initialization and submit cmd to LIO core
+ * @cmd: command descriptor to submit
+ *
+ * target_submit_prep or something similar must have been called on the cmd,
+ * and this must be called from process context.
+ */
+int target_submit(struct se_cmd *se_cmd)
+{
+	const struct target_core_fabric_ops *tfo = se_cmd->se_sess->se_tpg->se_tpg_tfo;
+	struct se_dev_attrib *da = &se_cmd->se_dev->dev_attrib;
+	u8 submit_type;
+
+	if (da->submit_type == TARGET_FABRIC_DEFAULT_SUBMIT)
+		submit_type = tfo->default_submit_type;
+	else if (da->submit_type == TARGET_DIRECT_SUBMIT &&
+		 tfo->direct_submit_supp)
+		submit_type = TARGET_DIRECT_SUBMIT;
+	else
+		submit_type = TARGET_QUEUE_SUBMIT;
+
+	if (submit_type == TARGET_DIRECT_SUBMIT)
+		return __target_submit(se_cmd);
+
+	target_queue_submission(se_cmd);
+	return 0;
+}
+EXPORT_SYMBOL_GPL(target_submit);
 
 static void target_complete_tmr_failure(struct work_struct *work)
 {