From: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx>
Historically, pSCSI devices have been the ones that required target-core
to enforce a per se_device->depth_left. This patch changes target-core
to no longer (by default) enforce a per se_device->depth_left or sleep in
transport_tcq_window_closed() when we out of queue slots.
It adds se_device->enforce_dev_tcq bitfield + se_subsystem_dev->su_dev_flags
SDF_ENFORCE_DEVICE_TCQ to signal the legacy usage in target-core with pSCSI.
So now aside from this special case, IBLOCK, FILEIO and RAMDISK with
->enforce_dev_tcq=0 by default.
It also moves all depth_left access under se_device->execute_task_lock
for the legacy case, and converts ->depth_left into a non atomic_t.
Cc: Christoph Hellwig <hch@xxxxxx>
Cc: Roland Dreier <roland@xxxxxxxxxxxxxxx>
Cc: Joern Engel <joern@xxxxxxxxx>
Signed-off-by: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx>
---
drivers/target/target_core_device.c | 6 ++-
drivers/target/target_core_pscsi.c | 4 ++
drivers/target/target_core_transport.c | 57 +++++++++++++++++++------------
include/target/target_core_base.h | 3 +-
include/target/target_core_transport.h | 1 +
5 files changed, 46 insertions(+), 25 deletions(-)
diff --git a/drivers/target/target_core_device.c b/drivers/target/target_core_device.c
index 5e10beb..ed06be4 100644
--- a/drivers/target/target_core_device.c
+++ b/drivers/target/target_core_device.c
@@ -1164,10 +1164,12 @@ int se_dev_set_queue_depth(struct se_device *dev, u32 queue_depth)
}
dev->se_sub_dev->se_dev_attrib.queue_depth = dev->queue_depth = queue_depth;
+ spin_lock_irq(&dev->execute_task_lock);
if (queue_depth > orig_queue_depth)
- atomic_add(queue_depth - orig_queue_depth, &dev->depth_left);
+ dev->depth_left += (queue_depth - orig_queue_depth);
else if (queue_depth < orig_queue_depth)
- atomic_sub(orig_queue_depth - queue_depth, &dev->depth_left);
+ dev->depth_left -= (orig_queue_depth - queue_depth);
+ spin_unlock_irq(&dev->execute_task_lock);
pr_debug("dev[%p]: SE Device TCQ Depth changed to: %u\n",
dev, queue_depth);
diff --git a/drivers/target/target_core_pscsi.c b/drivers/target/target_core_pscsi.c
index 53060ed..be5b6dd 100644
--- a/drivers/target/target_core_pscsi.c
+++ b/drivers/target/target_core_pscsi.c
@@ -350,6 +350,10 @@ static struct se_device *pscsi_add_device_to_list(
* scsi_device_put() and the pdv->pdv_sd cleared.
*/
pdv->pdv_sd = sd;
+ /*
+ * Make pSCSI enforce per se_device TCQ per scsi_device->queue_depth
+ */
+ se_dev->su_dev_flags |= SDF_ENFORCE_DEVICE_TCQ;
dev = transport_add_device_to_core_hba(hba, &pscsi_template,
se_dev, dev_flags, pdv,
diff --git a/drivers/target/target_core_transport.c b/drivers/target/target_core_transport.c
index 8bff7e71..bfe625b 100644
--- a/drivers/target/target_core_transport.c
+++ b/drivers/target/target_core_transport.c
@@ -708,12 +708,12 @@ void transport_complete_task(struct se_task *task, int success)
struct se_cmd *cmd = task->task_se_cmd;
struct se_device *dev = cmd->se_dev;
unsigned long flags;
-#if 0
- pr_debug("task: %p CDB: 0x%02x obj_ptr: %p\n", task,
- cmd->t_task_cdb[0], dev);
-#endif
- if (dev)
- atomic_inc(&dev->depth_left);
+
+ if (dev->enforce_dev_tcq) {
+ spin_lock_irq(&dev->execute_task_lock);
+ dev->depth_left++;
+ spin_unlock_irq(&dev->execute_task_lock);
+ }
spin_lock_irqsave(&cmd->t_state_lock, flags);
task->task_flags &= ~TF_ACTIVE;
@@ -996,7 +996,7 @@ void transport_dump_dev_state(
}
*bl += sprintf(b + *bl, " Execute/Left/Max Queue Depth: %d/%d/%d",
- atomic_read(&dev->execute_tasks), atomic_read(&dev->depth_left),
+ atomic_read(&dev->execute_tasks), dev->depth_left,
dev->queue_depth);
*bl += sprintf(b + *bl, " SectorSize: %u MaxSectors: %u\n",
dev->se_sub_dev->se_dev_attrib.block_size, dev->se_sub_dev->se_dev_attrib.max_sectors);
@@ -1358,9 +1358,16 @@ struct se_device *transport_add_device_to_core_hba(
spin_lock_init(&dev->se_port_lock);
spin_lock_init(&dev->se_tmr_lock);
spin_lock_init(&dev->qf_cmd_lock);
-
- dev->queue_depth = dev_limits->queue_depth;
- atomic_set(&dev->depth_left, dev->queue_depth);
+ /*
+ * Only setup and enforce the se_device queue depth for
+ * pSCSI backends using SDF_ENFORCE_DEVICE_TCQ.
+ */
+ if (se_dev->su_dev_flags & SDF_ENFORCE_DEVICE_TCQ) {
+ dev->queue_depth = dev->depth_left = dev_limits->queue_depth;
+ dev->enforce_dev_tcq = 1;
+ pr_debug("Using dev->enforce_dev_tcq: 1 for %s backend"
+ " device: %p\n", transport->name, dev);
+ }
atomic_set(&dev->dev_ordered_id, 0);
se_dev_set_default_attribs(dev, dev_limits);
@@ -2163,18 +2170,21 @@ static int __transport_execute_tasks(struct se_device *dev)
struct se_cmd *cmd = NULL;
struct se_task *task = NULL;
unsigned long flags;
-
/*
- * Check if there is enough room in the device and HBA queue to send
- * struct se_tasks to the selected transport.
+ * By default, enforce_dev_tcq is only enabled for pSCSI devices
+ * All other device backends (including IBLOCK and FILEIO)
+ * set enforce_dev_tcq=0, and dispatch new tasks as soon as
+ * possible.
*/
check_depth:
- if (!atomic_read(&dev->depth_left))
- return transport_tcq_window_closed(dev);
-
- dev->dev_tcq_window_closed = 0;
-
spin_lock_irq(&dev->execute_task_lock);
+ if (dev->enforce_dev_tcq) {
+ if (!dev->depth_left) {
+ spin_unlock_irq(&dev->execute_task_lock);
+ return transport_tcq_window_closed(dev);
+ }
+ dev->dev_tcq_window_closed = 0;
+ }
if (list_empty(&dev->execute_task_list)) {
spin_unlock_irq(&dev->execute_task_lock);
return 0;
@@ -2182,12 +2192,11 @@ check_depth:
task = list_first_entry(&dev->execute_task_list,
struct se_task, t_execute_list);
__transport_remove_task_from_execute_queue(task, dev);
+ if (dev->enforce_dev_tcq)
+ dev->depth_left--;
spin_unlock_irq(&dev->execute_task_lock);
- atomic_dec(&dev->depth_left);
-
cmd = task->task_se_cmd;
-
spin_lock_irqsave(&cmd->t_state_lock, flags);
task->task_flags |= (TF_ACTIVE | TF_SENT);
atomic_inc(&cmd->t_task_cdbs_sent);
@@ -2208,7 +2217,11 @@ check_depth:
spin_unlock_irqrestore(&cmd->t_state_lock, flags);
atomic_set(&cmd->t_transport_sent, 0);
transport_stop_tasks_for_cmd(cmd);
- atomic_inc(&dev->depth_left);
+ if (dev->enforce_dev_tcq) {
+ spin_lock_irq(&dev->execute_task_lock);
+ dev->depth_left++;
+ spin_unlock_irq(&dev->execute_task_lock);
+ }
transport_generic_request_failure(cmd);
}
diff --git a/include/target/target_core_base.h b/include/target/target_core_base.h
index 81bd34c..0fe8183 100644
--- a/include/target/target_core_base.h
+++ b/include/target/target_core_base.h
@@ -710,11 +710,13 @@ struct se_device {
u32 dev_tcq_window_closed;
/* Physical device queue depth */
u32 queue_depth;
+ u32 depth_left;
/* Used for SPC-2 reservations enforce of ISIDs */
u64 dev_res_bin_isid;
t10_task_attr_index_t dev_task_attr_type;
/* Pointer to transport specific device structure */
void *dev_ptr;
+ unsigned int enforce_dev_tcq:1;
u32 dev_index;
u64 creation_time;
u32 num_resets;
@@ -725,7 +727,6 @@ struct se_device {
/* Active commands on this virtual SE device */
atomic_t active_cmds;
atomic_t simple_cmds;
- atomic_t depth_left;
atomic_t dev_ordered_id;
atomic_t dev_tur_active;
atomic_t execute_tasks;
diff --git a/include/target/target_core_transport.h b/include/target/target_core_transport.h
index b3e80e7..6d2f329 100644
--- a/include/target/target_core_transport.h
+++ b/include/target/target_core_transport.h
@@ -21,6 +21,7 @@
#define SDF_EMULATED_VPD_UNIT_SERIAL 0x00000002
#define SDF_USING_UDEV_PATH 0x00000004
#define SDF_USING_ALIAS 0x00000008
+#define SDF_ENFORCE_DEVICE_TCQ 0x00000010
/*
* struct se_device->dev_flags
--
1.7.2.5
--
To unsubscribe from this list: send the line "unsubscribe target-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
