Skip to contentSkip to conversation

blk-mq: Implement runtime power management

PATCH Superseded Bart Van Assche 2 participants 12 messages Sep 18, 2018 – Sep 20, 2018

Subsystems (2)

v8 of 11 Superseded Latest is v11 →

lore
Share & tools
lore
#0 lore raw

v8 cover blk-mq: Implement runtime power management

Hello Jens,

One of the pieces that is missing before blk-mq can be made the default is implementing runtime power management support for blk-mq. This patch series not only implements runtime power management for blk-mq but also fixes a starvation issue in the power management code for the legacy block layer. Please consider this patch series for the upstream kernel.

Thanks,

Bart.

Changes compared to v7:
- Addressed Jianchao's feedback about patch "Make blk_get_request() block for non-PM requests while suspended".
- Added two new patches - one that documents the functions that iterate over requests and one that introduces a new function that iterates over all requests associated with a queue.

Changes compared to v6:
- Left out the patches that split RQF_PREEMPT in three flags.
- Left out the patch that introduces the SCSI device state SDEV_SUSPENDED.
- Left out the patch that introduces blk_pm_runtime_exit().
- Restored the patch that changes the PREEMPT_ONLY flag into a counter.

Changes compared to v5:
- Introduced a new flag RQF_DV that replaces RQF_PREEMPT for SCSI domain validation.
- Introduced a new request queue state QUEUE_FLAG_DV_ONLY for SCSI domain validation.
- Instead of using SDEV_QUIESCE for both runtime suspend and SCSI domain validation, use that state for domain validation only and introduce a new state for runtime suspend, namely SDEV_QUIESCE.
- Reallow system suspend during SCSI domain validation.
- Moved the runtime resume call from the request allocation code into blk_queue_enter().
- Instead of relying on q_usage_counter, iterate over the tag set to determine whether or not any requests are in flight.

Changes compared to v4:
- Dropped the patches "Give RQF_PREEMPT back its original meaning" and "Serialize queue freezing and blk_pre_runtime_suspend()".
- Replaced "percpu_ref_read()" with "percpu_is_in_use()".
- Inserted pm_request_resume() calls in the block layer request allocation code such that the context that submits a request no longer has to call pm_runtime_get().

Changes compared to v3:
- Avoid adverse interactions between system-wide suspend/resume and runtime power management by changing the PREEMPT_ONLY flag into a counter.
- Give RQF_PREEMPT back its original meaning, namely that it is only set for ide_preempt requests.
- Remove the flag BLK_MQ_REQ_PREEMPT.
- Removed the pm_request_resume() call.

Changes compared to v2:
- Fixed the build for CONFIG_BLOCK=n.
- Added a patch that introduces percpu_ref_read() in the percpu-counter code.
- Added a patch that makes it easier to detect missing pm_runtime_get*() calls.
- Addressed Jianchao's feedback including the comment about runtime overhead of switching a per-cpu counter to atomic mode.

Changes compared to v1:
- Moved the runtime power management code into a separate file.
- Addressed Ming's feedback.

Bart Van Assche (6):
block: Move power management code into a new source file block, scsi: Change the preempt-only flag into a counter
block: Split blk_pm_add_request() and blk_pm_put_request()
block: Schedule runtime resume earlier
block: Make blk_get_request() block for non-PM requests while
suspended
blk-mq: Enable support for runtime power management

Bart Van Assche (8):
blk-mq: Document the functions that iterate over requests
blk-mq: Introduce blk_mq_queue_rq_iter()
block: Move power management code into a new source file block, scsi: Change the preempt-only flag into a counter
block: Split blk_pm_add_request() and blk_pm_put_request()
block: Schedule runtime resume earlier
block: Make blk_get_request() block for non-PM requests while
suspended
blk-mq: Enable support for runtime power management

 block/Kconfig           |   3 +
 block/Makefile          |   1 +
 block/blk-core.c        | 271 +++++-----------------------------------
 block/blk-mq-debugfs.c  |  10 +-
 block/blk-mq-tag.c      |  72 ++++++++++-
 block/blk-mq-tag.h      |   2 +
 block/blk-mq.c          |   2 +
 block/blk-pm.c          | 242 +++++++++++++++++++++++++++++++++++
 block/blk-pm.h          |  69 ++++++++++
 block/elevator.c        |  22 +---
 drivers/scsi/scsi_lib.c |  11 +-
 drivers/scsi/scsi_pm.c  |   1 +
 drivers/scsi/sd.c       |   1 +
 drivers/scsi/sr.c       |   1 +
 include/linux/blk-pm.h  |  24 ++++
 include/linux/blkdev.h  |  37 ++----
 16 files changed, 472 insertions(+), 297 deletions(-)
 create mode 100644 block/blk-pm.c
 create mode 100644 block/blk-pm.h
 create mode 100644 include/linux/blk-pm.h
Signature · 1 line
2.18.0
#1 lore raw

PATCH v8 1/8 blk-mq: Document the functions that iterate over requests

Make it easier to understand the purpose of the functions that iterate over requests by documenting their purpose. Fix two minor spelling mistakes in comments in these functions.

Cc Christoph Hellwig <[email protected]>
Cc Ming Lei <[email protected]>
Cc Jianchao Wang <[email protected]>
Cc Hannes Reinecke <[email protected]>
Cc Johannes Thumshirn <[email protected]>
---
 block/blk-mq-tag.c | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)
block/blk-mq-tag.c +26 -2
index 94e1ed667b6e..ef3acb4a80e0 100644
@@ -232,13 +232,19 @@ static bool bt_iter(struct sbitmap *bitmap, unsigned int bitnr, void *data)
 
 	/*
 	 * We can hit rq == NULL here, because the tagging functions
-	 * test and set the bit before assining ->rqs[].
+	 * test and set the bit before assigning ->rqs[].
 	 */
 	if (rq && rq->q == hctx->queue)
 		iter_data->fn(hctx, rq, iter_data->data, reserved);
 	return true;
 }
 
+/*
+ * Call function @fn(@hctx, rq, @data, @reserved) for each request queued on
+ * @hctx that has been assigned a driver tag. @reserved indicates whether @bt
+ * is the breserved_tags member or the bitmap_tags member of struct
+ * blk_mq_tags.
+ */
 static void bt_for_each(struct blk_mq_hw_ctx *hctx, struct sbitmap_queue *bt,
 			busy_iter_fn *fn, void *data, bool reserved)
 {
@@ -280,6 +286,11 @@ static bool bt_tags_iter(struct sbitmap *bitmap, unsigned int bitnr, void *data)
 	return true;
 }
 
+/*
+ * Call function @fn(rq, @data, @reserved) for each request in @tags that has
+ * been started. @reserved indicates whether @bt is the breserved_tags member
+ * or the bitmap_tags member of struct blk_mq_tags.
+ */
 static void bt_tags_for_each(struct blk_mq_tags *tags, struct sbitmap_queue *bt,
 			     busy_tag_iter_fn *fn, void *data, bool reserved)
 {
@@ -294,6 +305,10 @@ static void bt_tags_for_each(struct blk_mq_tags *tags, struct sbitmap_queue *bt,
 		sbitmap_for_each_set(&bt->sb, bt_tags_iter, &iter_data);
 }
 
+/*
+ * Call @fn(rq, @priv, reserved) for each started request in @tags. 'reserved'
+ * indicates whether or not 'rq' is a reserved request.
+ */
 static void blk_mq_all_tag_busy_iter(struct blk_mq_tags *tags,
 		busy_tag_iter_fn *fn, void *priv)
 {
@@ -302,6 +317,10 @@ static void blk_mq_all_tag_busy_iter(struct blk_mq_tags *tags,
 	bt_tags_for_each(tags, &tags->bitmap_tags, fn, priv, false);
 }
 
+/*
+ * Call @fn(rq, @priv, reserved) for each request in @tagset. 'reserved'
+ * indicates whether or not 'rq' is a reserved request.
+ */
 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
 		busy_tag_iter_fn *fn, void *priv)
 {
@@ -314,6 +333,11 @@ void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
 }
 EXPORT_SYMBOL(blk_mq_tagset_busy_iter);
 
+/*
+ * Call @fn(rq, @priv, reserved) for each request associated with request
+ * queue @q or any queue it shares tags with and that has been assigned a
+ * driver tag. 'reserved' indicates whether or not 'rq' is a reserved request.
+ */
 void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn,
 		void *priv)
 {
@@ -337,7 +361,7 @@ void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn,
 		struct blk_mq_tags *tags = hctx->tags;
 
 		/*
-		 * If not software queues are currently mapped to this
+		 * If no software queues are currently mapped to this
 		 * hardware queue, there's nothing to check
 		 */
 		if (!blk_mq_hw_queue_mapped(hctx))
Signature · 1 line
2.18.0
#2 lore raw

PATCH v8 2/8 blk-mq: Introduce blk_mq_queue_rq_iter()

In reply to Bart Van Assche

This function will be used in the patch "Make blk_get_request() block for non-PM requests while suspended".

Cc Christoph Hellwig <[email protected]>
Cc Ming Lei <[email protected]>
Cc Jianchao Wang <[email protected]>
Cc Hannes Reinecke <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Cc Alan Stern <[email protected]>
---
 block/blk-mq-tag.c | 44 ++++++++++++++++++++++++++++++++++++++++++++
 block/blk-mq-tag.h |  2 ++
 2 files changed, 46 insertions(+)
block/blk-mq-tag.c +44 -0
index ef3acb4a80e0..cf8537017f78 100644
@@ -374,6 +374,50 @@ void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn,
 	rcu_read_unlock();
 }
 
+/*
+ * Call @fn(rq, @priv, reserved) for each request associated with request
+ * queue @q or any queue that it shares tags with and that has been assigned a
+ * tag. 'reserved' indicates whether or not 'rq' is a reserved request. In
+ * contrast to blk_mq_queue_tag_busy_iter(), if an I/O scheduler has been
+ * associated with @q, this function also iterates over requests that have
+ * been assigned a scheduler tag but that have not yet been assigned a driver
+ * tag.
+ */
+void blk_mq_queue_rq_iter(struct request_queue *q, busy_iter_fn *fn, void *priv)
+{
+	struct blk_mq_hw_ctx *hctx;
+	int i;
+
+	/*
+	 * __blk_mq_update_nr_hw_queues will update the nr_hw_queues and
+	 * queue_hw_ctx after freeze the queue. So we could use q_usage_counter
+	 * to avoid race with it. __blk_mq_update_nr_hw_queues will users
+	 * synchronize_rcu to ensure all of the users go out of the critical
+	 * section below and see zeroed q_usage_counter.
+	 */
+	rcu_read_lock();
+	if (percpu_ref_is_zero(&q->q_usage_counter)) {
+		rcu_read_unlock();
+		return;
+	}
+
+	queue_for_each_hw_ctx(q, hctx, i) {
+		struct blk_mq_tags *tags = hctx->sched_tags ? : hctx->tags;
+
+		/*
+		 * If no software queues are currently mapped to this
+		 * hardware queue, there's nothing to check
+		 */
+		if (!blk_mq_hw_queue_mapped(hctx))
+			continue;
+
+		if (tags->nr_reserved_tags)
+			bt_for_each(hctx, &tags->breserved_tags, fn, priv, true);
+		bt_for_each(hctx, &tags->bitmap_tags, fn, priv, false);
+	}
+	rcu_read_unlock();
+}
+
 static int bt_alloc(struct sbitmap_queue *bt, unsigned int depth,
 		    bool round_robin, int node)
 {
block/blk-mq-tag.h +2 -0
index 61deab0b5a5a..25e62997ed6c 100644
@@ -35,6 +35,8 @@ extern int blk_mq_tag_update_depth(struct blk_mq_hw_ctx *hctx,
 extern void blk_mq_tag_wakeup_all(struct blk_mq_tags *tags, bool);
 void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn,
 		void *priv);
+void blk_mq_queue_rq_iter(struct request_queue *q, busy_iter_fn *fn,
+			  void *priv);
 
 static inline struct sbq_wait_state *bt_wait_ptr(struct sbitmap_queue *bt,
 						 struct blk_mq_hw_ctx *hctx)
Signature · 1 line
2.18.0
#3 lore raw

PATCH v8 3/8 block: Move power management code into a new source file

In reply to Bart Van Assche

Move the code for runtime power management from blk-core.c into the new source file blk-pm.c. Move the corresponding declarations from <linux/blkdev.h> into <linux/blk-pm.h>. For CONFIG_PM=n, leave out the declarations of the functions that are not used in that mode. This patch not only reduces the number of #ifdefs in the block layer core code but also reduces the size of header file <linux/blkdev.h> and hence should help to reduce the build time of the Linux kernel if CONFIG_PM is not defined.

Cc Christoph Hellwig <[email protected]>
Cc Ming Lei <[email protected]>
Cc Jianchao Wang <[email protected]>
Cc Hannes Reinecke <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Cc Alan Stern <[email protected]>
---
 block/Kconfig          |   3 +
 block/Makefile         |   1 +
 block/blk-core.c       | 196 +----------------------------------------
 block/blk-pm.c         | 188 +++++++++++++++++++++++++++++++++++++++
 block/blk-pm.h         |  43 +++++++++
 block/elevator.c       |  22 +----
 drivers/scsi/scsi_pm.c |   1 +
 drivers/scsi/sd.c      |   1 +
 drivers/scsi/sr.c      |   1 +
 include/linux/blk-pm.h |  24 +++++
 include/linux/blkdev.h |  23 -----
 11 files changed, 264 insertions(+), 239 deletions(-)
 create mode 100644 block/blk-pm.c
 create mode 100644 block/blk-pm.h
 create mode 100644 include/linux/blk-pm.h
block/Kconfig +3 -0
index 1f2469a0123c..85263e7bded6 100644
@@ -228,4 +228,7 @@ config BLK_MQ_RDMA
 	depends on BLOCK && INFINIBAND
 	default y
 
+config BLK_PM
+	def_bool BLOCK && PM
+
 source block/Kconfig.iosched
block/Makefile +1 -0
index 572b33f32c07..27eac600474f 100644
@@ -37,3 +37,4 @@ obj-$(CONFIG_BLK_WBT)		+= blk-wbt.o
 obj-$(CONFIG_BLK_DEBUG_FS)	+= blk-mq-debugfs.o
 obj-$(CONFIG_BLK_DEBUG_FS_ZONED)+= blk-mq-debugfs-zoned.o
 obj-$(CONFIG_BLK_SED_OPAL)	+= sed-opal.o
+obj-$(CONFIG_BLK_PM)		+= blk-pm.o
block/blk-core.c +1 -195

214 lines not on this page · raw message

block/blk-pm.c new file +188 -0

188 lines not on this page · raw message

block/blk-pm.h new file +43 -0

43 lines not on this page · raw message

block/elevator.c +1 -21
index 6a06b5d040e5..e18ac68626e3 100644
@@ -41,6 +41,7 @@
 
 #include "blk.h"
 #include "blk-mq-sched.h"
+#include "blk-pm.h"
 #include "blk-wbt.h"
 
 static DEFINE_SPINLOCK(elv_list_lock);
@@ -557,27 +558,6 @@ void elv_bio_merged(struct request_queue *q, struct request *rq,
 		e->type->ops.sq.elevator_bio_merged_fn(q, rq, bio);
 }
 
-#ifdef CONFIG_PM
-static void blk_pm_requeue_request(struct request *rq)
-{
-	if (rq->q->dev && !(rq->rq_flags & RQF_PM))
-		rq->q->nr_pending--;
-}
-
-static void blk_pm_add_request(struct request_queue *q, struct request *rq)
-{
-	if (q->dev && !(rq->rq_flags & RQF_PM) && q->nr_pending++ == 0 &&
-	    (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
-		pm_request_resume(q->dev);
-}
-#else
-static inline void blk_pm_requeue_request(struct request *rq) {}
-static inline void blk_pm_add_request(struct request_queue *q,
-				      struct request *rq)
-{
-}
-#endif
-
 void elv_requeue_request(struct request_queue *q, struct request *rq)
 {
 	/*
drivers/scsi/scsi_pm.c +1 -0
index b44c1bb687a2..a2b4179bfdf7 100644
@@ -8,6 +8,7 @@
 #include <linux/pm_runtime.h>
 #include <linux/export.h>
 #include <linux/async.h>
+#include <linux/blk-pm.h>
 
 #include <scsi/scsi.h>
 #include <scsi/scsi_device.h>
drivers/scsi/sd.c +1 -0
index b79b366a94f7..64514e8359e4 100644
@@ -45,6 +45,7 @@
 #include <linux/init.h>
 #include <linux/blkdev.h>
 #include <linux/blkpg.h>
+#include <linux/blk-pm.h>
 #include <linux/delay.h>
 #include <linux/mutex.h>
 #include <linux/string_helpers.h>
drivers/scsi/sr.c +1 -0
index d0389b20574d..4f07b3410595 100644
@@ -43,6 +43,7 @@
 #include <linux/interrupt.h>
 #include <linux/init.h>
 #include <linux/blkdev.h>
+#include <linux/blk-pm.h>
 #include <linux/mutex.h>
 #include <linux/slab.h>
 #include <linux/pm_runtime.h>
include/linux/blk-pm.h new file +24 -0
new file mode 100644
index 000000000000..b80c65aba249
@@ -0,0 +1,24 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+#ifndef _BLK_PM_H_
+#define _BLK_PM_H_
+
+struct device;
+struct request_queue;
+
+/*
+ * block layer runtime pm functions
+ */
+#ifdef CONFIG_PM
+extern void blk_pm_runtime_init(struct request_queue *q, struct device *dev);
+extern int blk_pre_runtime_suspend(struct request_queue *q);
+extern void blk_post_runtime_suspend(struct request_queue *q, int err);
+extern void blk_pre_runtime_resume(struct request_queue *q);
+extern void blk_post_runtime_resume(struct request_queue *q, int err);
+extern void blk_set_runtime_active(struct request_queue *q);
+#else
+static inline void blk_pm_runtime_init(struct request_queue *q,
+				       struct device *dev) {}
+#endif
+
+#endif /* _BLK_PM_H_ */
include/linux/blkdev.h +0 -23
index 6980014357d4..e25ce42c982d 100644
@@ -1280,29 +1280,6 @@ struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id,
 extern void blk_put_queue(struct request_queue *);
 extern void blk_set_queue_dying(struct request_queue *);
 
-/*
- * block layer runtime pm functions
- */
-#ifdef CONFIG_PM
-extern void blk_pm_runtime_init(struct request_queue *q, struct device *dev);
-extern int blk_pre_runtime_suspend(struct request_queue *q);
-extern void blk_post_runtime_suspend(struct request_queue *q, int err);
-extern void blk_pre_runtime_resume(struct request_queue *q);
-extern void blk_post_runtime_resume(struct request_queue *q, int err);
-extern void blk_set_runtime_active(struct request_queue *q);
-#else
-static inline void blk_pm_runtime_init(struct request_queue *q,
-	struct device *dev) {}
-static inline int blk_pre_runtime_suspend(struct request_queue *q)
-{
-	return -ENOSYS;
-}
-static inline void blk_post_runtime_suspend(struct request_queue *q, int err) {}
-static inline void blk_pre_runtime_resume(struct request_queue *q) {}
-static inline void blk_post_runtime_resume(struct request_queue *q, int err) {}
-static inline void blk_set_runtime_active(struct request_queue *q) {}
-#endif
-
 /*
  * blk_plug permits building a queue of related requests by holding the I/O
  * fragments for a short period. This allows merging of sequential requests
Signature · 1 line
2.18.0
#4 lore raw

PATCH v8 4/8 block, scsi: Change the preempt-only flag into a counter

In reply to Bart Van Assche

The RQF_PREEMPT flag is used for three purposes:
- In the SCSI core, for making sure that power management requests are executed even if a device is in the "quiesced" state.
- For domain validation by SCSI drivers that use the parallel port.
- In the IDE driver, for IDE preempt requests.
Rename "preempt-only" into "pm-only" because the primary purpose of this mode is power management. Since the power management core may but does not have to resume a runtime suspended device before performing system-wide suspend and since a later patch will set "pm-only" mode as long as a block device is runtime suspended, make it possible to set "pm-only" mode from more than one context. Since with this change scsi_device_quiesce() is no longer idempotent, make that function return early if it is called for a quiesced queue.

Cc Martin K. Petersen <[email protected]>
Cc Ming Lei <[email protected]>
Cc Jianchao Wang <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Cc Alan Stern <[email protected]>
---
 block/blk-core.c        | 35 ++++++++++++++++++-----------------
 block/blk-mq-debugfs.c  | 10 +++++++++-
 drivers/scsi/scsi_lib.c | 11 +++++++----
 include/linux/blkdev.h  | 14 +++++++++-----
 4 files changed, 43 insertions(+), 27 deletions(-)
block/blk-core.c +18 -17
index 6d4dd176bd9d..1a691f5269bb 100644
@@ -422,24 +422,25 @@ void blk_sync_queue(struct request_queue *q)
 EXPORT_SYMBOL(blk_sync_queue);
 
 /**
- * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
+ * blk_set_pm_only - increment pm_only counter
  * @q: request queue pointer
- *
- * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
- * set and 1 if the flag was already set.
  */
-int blk_set_preempt_only(struct request_queue *q)
+void blk_set_pm_only(struct request_queue *q)
 {
-	return blk_queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY, q);
+	atomic_inc(&q->pm_only);
 }
-EXPORT_SYMBOL_GPL(blk_set_preempt_only);
+EXPORT_SYMBOL_GPL(blk_set_pm_only);
 
-void blk_clear_preempt_only(struct request_queue *q)
+void blk_clear_pm_only(struct request_queue *q)
 {
-	blk_queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY, q);
-	wake_up_all(&q->mq_freeze_wq);
+	int pm_only;
+
+	pm_only = atomic_dec_return(&q->pm_only);
+	WARN_ON_ONCE(pm_only < 0);
+	if (pm_only == 0)
+		wake_up_all(&q->mq_freeze_wq);
 }
-EXPORT_SYMBOL_GPL(blk_clear_preempt_only);
+EXPORT_SYMBOL_GPL(blk_clear_pm_only);
 
 /**
  * __blk_run_queue_uncond - run a queue whether or not it has been stopped
@@ -918,7 +919,7 @@ EXPORT_SYMBOL(blk_alloc_queue);
  */
 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
 {
-	const bool preempt = flags & BLK_MQ_REQ_PREEMPT;
+	const bool pm = flags & BLK_MQ_REQ_PREEMPT;
 
 	while (true) {
 		bool success = false;
@@ -926,11 +927,11 @@ int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
 		rcu_read_lock();
 		if (percpu_ref_tryget_live(&q->q_usage_counter)) {
 			/*
-			 * The code that sets the PREEMPT_ONLY flag is
-			 * responsible for ensuring that that flag is globally
-			 * visible before the queue is unfrozen.
+			 * The code that increments the pm_only counter is
+			 * responsible for ensuring that that counter is
+			 * globally visible before the queue is unfrozen.
 			 */
-			if (preempt || !blk_queue_preempt_only(q)) {
+			if (pm || !blk_queue_pm_only(q)) {
 				success = true;
 			} else {
 				percpu_ref_put(&q->q_usage_counter);
@@ -955,7 +956,7 @@ int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
 
 		wait_event(q->mq_freeze_wq,
 			   (atomic_read(&q->mq_freeze_depth) == 0 &&
-			    (preempt || !blk_queue_preempt_only(q))) ||
+			    (pm || !blk_queue_pm_only(q))) ||
 			   blk_queue_dying(q));
 		if (blk_queue_dying(q))
 			return -ENODEV;
block/blk-mq-debugfs.c +9 -1
index cb1e6cf7ac48..a5ea86835fcb 100644
@@ -102,6 +102,14 @@ static int blk_flags_show(struct seq_file *m, const unsigned long flags,
 	return 0;
 }
 
+static int queue_pm_only_show(void *data, struct seq_file *m)
+{
+	struct request_queue *q = data;
+
+	seq_printf(m, "%d\n", atomic_read(&q->pm_only));
+	return 0;
+}
+
 #define QUEUE_FLAG_NAME(name) [QUEUE_FLAG_##name] = #name
 static const char *const blk_queue_flag_name[] = {
 	QUEUE_FLAG_NAME(QUEUED),
@@ -132,7 +140,6 @@ static const char *const blk_queue_flag_name[] = {
 	QUEUE_FLAG_NAME(REGISTERED),
 	QUEUE_FLAG_NAME(SCSI_PASSTHROUGH),
 	QUEUE_FLAG_NAME(QUIESCED),
-	QUEUE_FLAG_NAME(PREEMPT_ONLY),
 };
 #undef QUEUE_FLAG_NAME
 
@@ -209,6 +216,7 @@ static ssize_t queue_write_hint_store(void *data, const char __user *buf,
 static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
 	{ "poll_stat", 0400, queue_poll_stat_show },
 	{ "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
+	{ "pm_only", 0600, queue_pm_only_show, NULL },
 	{ "state", 0600, queue_state_show, queue_state_write },
 	{ "write_hints", 0600, queue_write_hint_show, queue_write_hint_store },
 	{ "zone_wlock", 0400, queue_zone_wlock_show, NULL },
drivers/scsi/scsi_lib.c +7 -4
index eb97d2dd3651..62348412ed1b 100644
@@ -3046,11 +3046,14 @@ scsi_device_quiesce(struct scsi_device *sdev)
 	 */
 	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
 
-	blk_set_preempt_only(q);
+	if (sdev->quiesced_by == current)
+		return 0;
+
+	blk_set_pm_only(q);
 
 	blk_mq_freeze_queue(q);
 	/*
-	 * Ensure that the effect of blk_set_preempt_only() will be visible
+	 * Ensure that the effect of blk_set_pm_only() will be visible
 	 * for percpu_ref_tryget() callers that occur after the queue
 	 * unfreeze even if the queue was already frozen before this function
 	 * was called. See also https://lwn.net/Articles/573497/.
@@ -3063,7 +3066,7 @@ scsi_device_quiesce(struct scsi_device *sdev)
 	if (err == 0)
 		sdev->quiesced_by = current;
 	else
-		blk_clear_preempt_only(q);
+		blk_clear_pm_only(q);
 	mutex_unlock(&sdev->state_mutex);
 
 	return err;
@@ -3088,7 +3091,7 @@ void scsi_device_resume(struct scsi_device *sdev)
 	mutex_lock(&sdev->state_mutex);
 	WARN_ON_ONCE(!sdev->quiesced_by);
 	sdev->quiesced_by = NULL;
-	blk_clear_preempt_only(sdev->request_queue);
+	blk_clear_pm_only(sdev->request_queue);
 	if (sdev->sdev_state == SDEV_QUIESCE)
 		scsi_device_set_state(sdev, SDEV_RUNNING);
 	mutex_unlock(&sdev->state_mutex);
include/linux/blkdev.h +9 -5
index e25ce42c982d..5f6c36b5058c 100644
@@ -504,6 +504,12 @@ struct request_queue {
 	 * various queue flags, see QUEUE_* below
 	 */
 	unsigned long		queue_flags;
+	/*
+	 * Number of contexts that have called blk_set_pm_only(). If this
+	 * counter is above zero then only RQF_PM and RQF_PREEMPT requests are
+	 * processed.
+	 */
+	atomic_t		pm_only;
 
 	/*
 	 * ida allocated id for this queue.  Used to index queues from
@@ -698,7 +704,6 @@ struct request_queue {
 #define QUEUE_FLAG_REGISTERED  26	/* queue has been registered to a disk */
 #define QUEUE_FLAG_SCSI_PASSTHROUGH 27	/* queue supports SCSI commands */
 #define QUEUE_FLAG_QUIESCED    28	/* queue has been quiesced */
-#define QUEUE_FLAG_PREEMPT_ONLY	29	/* only process REQ_PREEMPT requests */
 
 #define QUEUE_FLAG_DEFAULT	((1 << QUEUE_FLAG_IO_STAT) |		\
 				 (1 << QUEUE_FLAG_SAME_COMP)	|	\
@@ -736,12 +741,11 @@ bool blk_queue_flag_test_and_clear(unsigned int flag, struct request_queue *q);
 	((rq)->cmd_flags & (REQ_FAILFAST_DEV|REQ_FAILFAST_TRANSPORT| \
 			     REQ_FAILFAST_DRIVER))
 #define blk_queue_quiesced(q)	test_bit(QUEUE_FLAG_QUIESCED, &(q)->queue_flags)
-#define blk_queue_preempt_only(q)				\
-	test_bit(QUEUE_FLAG_PREEMPT_ONLY, &(q)->queue_flags)
+#define blk_queue_pm_only(q)	atomic_read(&(q)->pm_only)
 #define blk_queue_fua(q)	test_bit(QUEUE_FLAG_FUA, &(q)->queue_flags)
 
-extern int blk_set_preempt_only(struct request_queue *q);
-extern void blk_clear_preempt_only(struct request_queue *q);
+extern void blk_set_pm_only(struct request_queue *q);
+extern void blk_clear_pm_only(struct request_queue *q);
 
 static inline int queue_in_flight(struct request_queue *q)
 {
Signature · 1 line
2.18.0
#5 lore raw

PATCH v8 5/8 block: Split blk_pm_add_request() and blk_pm_put_request()

In reply to Bart Van Assche

Move the pm_request_resume() and pm_runtime_mark_last_busy() calls into two new functions and thereby separate legacy block layer code from code that works for both the legacy block layer and blk-mq. A later patch will add calls to the new functions in the blk-mq code.

Cc Martin K. Petersen <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Ming Lei <[email protected]>
Cc Jianchao Wang <[email protected]>
Cc Hannes Reinecke <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Cc Alan Stern <[email protected]>
---
 block/blk-core.c |  1 +
 block/blk-pm.h   | 36 +++++++++++++++++++++++++++++++-----
 block/elevator.c |  1 +
 3 files changed, 33 insertions(+), 5 deletions(-)
block/blk-core.c +1 -0
index 1a691f5269bb..fd91e9bf2893 100644
@@ -1744,6 +1744,7 @@ void __blk_put_request(struct request_queue *q, struct request *req)
 
 	blk_req_zone_write_unlock(req);
 	blk_pm_put_request(req);
+	blk_pm_mark_last_busy(req);
 
 	elv_completed_request(q, req);
 
block/blk-pm.h +31 -5
index 1ffc8ef203ec..a8564ea72a41 100644
@@ -6,8 +6,23 @@
 #include <linux/pm_runtime.h>
 
 #ifdef CONFIG_PM
+static inline void blk_pm_request_resume(struct request_queue *q)
+{
+	if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
+		       q->rpm_status == RPM_SUSPENDING))
+		pm_request_resume(q->dev);
+}
+
+static inline void blk_pm_mark_last_busy(struct request *rq)
+{
+	if (rq->q->dev && !(rq->rq_flags & RQF_PM))
+		pm_runtime_mark_last_busy(rq->q->dev);
+}
+
 static inline void blk_pm_requeue_request(struct request *rq)
 {
+	lockdep_assert_held(rq->q->queue_lock);
+
 	if (rq->q->dev && !(rq->rq_flags & RQF_PM))
 		rq->q->nr_pending--;
 }
@@ -15,17 +30,28 @@ static inline void blk_pm_requeue_request(struct request *rq)
 static inline void blk_pm_add_request(struct request_queue *q,
 				      struct request *rq)
 {
-	if (q->dev && !(rq->rq_flags & RQF_PM) && q->nr_pending++ == 0 &&
-	    (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
-		pm_request_resume(q->dev);
+	lockdep_assert_held(q->queue_lock);
+
+	if (q->dev && !(rq->rq_flags & RQF_PM))
+		q->nr_pending++;
 }
 
 static inline void blk_pm_put_request(struct request *rq)
 {
-	if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
-		pm_runtime_mark_last_busy(rq->q->dev);
+	lockdep_assert_held(rq->q->queue_lock);
+
+	if (rq->q->dev && !(rq->rq_flags & RQF_PM))
+		--rq->q->nr_pending;
 }
 #else
+static inline void blk_pm_request_resume(struct request_queue *q)
+{
+}
+
+static inline void blk_pm_mark_last_busy(struct request *rq)
+{
+}
+
 static inline void blk_pm_requeue_request(struct request *rq)
 {
 }
block/elevator.c +1 -0
index e18ac68626e3..1c992bf6cfb1 100644
@@ -601,6 +601,7 @@ void __elv_add_request(struct request_queue *q, struct request *rq, int where)
 	trace_block_rq_insert(q, rq);
 
 	blk_pm_add_request(q, rq);
+	blk_pm_request_resume(q);
 
 	rq->q = q;
 
Signature · 1 line
2.18.0
#6 lore raw

PATCH v8 6/8 block: Schedule runtime resume earlier

In reply to Bart Van Assche

Instead of scheduling runtime resume of a request queue after a request has been queued, schedule asynchronous resume during request allocation. The new pm_request_resume() calls occur after blk_queue_enter() has increased the q_usage_counter request queue member. This change is needed for a later patch that will make request allocation block while the queue status is not RPM_ACTIVE.

Cc Christoph Hellwig <[email protected]>
Cc Ming Lei <[email protected]>
Cc Jianchao Wang <[email protected]>
Cc Hannes Reinecke <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Cc Alan Stern <[email protected]>
---
 block/blk-core.c | 2 ++
 block/elevator.c | 1 -
 2 files changed, 2 insertions(+), 1 deletion(-)
block/blk-core.c +2 -0
index fd91e9bf2893..18b874d5c9c9 100644
@@ -942,6 +942,8 @@ int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
 		if (success)
 			return 0;
 
+		blk_pm_request_resume(q);
+
 		if (flags & BLK_MQ_REQ_NOWAIT)
 			return -EBUSY;
 
block/elevator.c +0 -1
index 1c992bf6cfb1..e18ac68626e3 100644
@@ -601,7 +601,6 @@ void __elv_add_request(struct request_queue *q, struct request *rq, int where)
 	trace_block_rq_insert(q, rq);
 
 	blk_pm_add_request(q, rq);
-	blk_pm_request_resume(q);
 
 	rq->q = q;
 
Signature · 1 line
2.18.0
#7 lore raw

Re: [PATCH v8 6/8] block: Schedule runtime resume earlier

Quoted 24 lines · Bart Van Assche

Instead of scheduling runtime resume of a request queue after a request has been queued, schedule asynchronous resume during request allocation. The new pm_request_resume() calls occur after blk_queue_enter() has increased the q_usage_counter request queue member. This change is needed for a later patch that will make request allocation block while the queue status is not RPM_ACTIVE.

Cc Christoph Hellwig <[email protected]>
Cc Ming Lei <[email protected]>
Cc Jianchao Wang <[email protected]>
Cc Hannes Reinecke <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Cc Alan Stern <[email protected]>
---
 block/blk-core.c | 2 ++
 block/elevator.c | 1 -
 2 files changed, 2 insertions(+), 1 deletion(-)
block/blk-core.c +0 -0
index fd91e9bf2893..18b874d5c9c9 100644

block/blk-core.c int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) · line 945

Quoted 4 lines
block/blk-core.c +1 -0
@@ -942,6 +942,8 @@ int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
 		if (success)
 			return 0;
 
+		blk_pm_request_resume(q);

Looks this patch may introduce the following race between queue freeze and runtime suspend:

--------------------------------------------------------------------------
CPU0 CPU1 CPU2
--------------------------------------------------------------------------

blk_freeze_queue()

					blk_mq_alloc_request()
						blk_queue_enter()
							blk_pm_request_resume()
							wait_event()

												blk_pre_runtime_suspend()
													->blk_set_pm_only
												...
												blk_post_runtime_suspend()

...
blk_mq_unfreeze_queue()
--------------------------------------------------------------------------
- CPU0: queue is frozen

- CPU1: one new request comes, and see queue is frozen, but queue isn't runtime-suspended yet, then blk_pm_request_resume() does nothing. So this allocation is blocked in wait_event().

- CPU2: runtime suspend comes, and queue becomes runtime-suspended now

- CPU0: queue is unfreeze, but the new request allocation in CPU1 may never be done because the queue is runtime suspended, and wait_event() won't return. And the expected result is that the queue becomes active and the allocation on CPU1 is done immediately.

Thanks,
Ming

#8 lore raw

Re: [PATCH v8 6/8] block: Schedule runtime resume earlier

Quoted 42 lines · Ming Lei

Looks this patch may introduce the following race between queue freeze and
runtime suspend:

-------------------------------------------------------------------
-------
CPU0 CPU1
CPU2
-------------------------------------------------------------------
-------

blk_freeze_queue()

					blk_mq_alloc_request()
						blk_queue_enter()
							blk_pm_request_>resume()
							wait_event()

									
			blk_pre_runtime_suspend()
									
				->blk_set_pm_only
									
			...
									
			blk_post_runtime_suspend()

...
blk_mq_unfreeze_queue()
-------------------------------------------------------------------
-------
- CPU0: queue is frozen

- CPU1: one new request comes, and see queue is frozen, but queue isn't
runtime-suspended yet, then blk_pm_request_resume() does nothing. So this
allocation is blocked in wait_event().

- CPU2: runtime suspend comes, and queue becomes runtime-suspended now

Quoted 7 lines

- CPU0: queue is unfreeze, but the new request allocation in CPU1 may never
be done because the queue is runtime suspended, and wait_event() won't return.
And the expected result is that the queue becomes active and the allocation on
CPU1 is done immediately.

Hello Ming,

Just like for the scenario Jianchao reported, I will address this by only allowing the suspend to proceed if q_usage_counter ==0.

Bart.

#9 lore raw

Re: [PATCH v8 6/8] block: Schedule runtime resume earlier

Quoted 52 lines · Bart Van Assche
Quoted 43 lines · Ming Lei

Looks this patch may introduce the following race between queue freeze and
runtime suspend:

-------------------------------------------------------------------
-------
CPU0 CPU1
CPU2
-------------------------------------------------------------------
-------

blk_freeze_queue()

blk_mq_alloc_request()
blk_queue_enter()
blk_pm_request_
resume()

							wait_event()

									
			blk_pre_runtime_suspend()
									
				->blk_set_pm_only
									
			...
									
			blk_post_runtime_suspend()

...
blk_mq_unfreeze_queue()
-------------------------------------------------------------------
-------
- CPU0: queue is frozen

- CPU1: one new request comes, and see queue is frozen, but queue isn't
runtime-suspended yet, then blk_pm_request_resume() does nothing. So this
allocation is blocked in wait_event().

- CPU2: runtime suspend comes, and queue becomes runtime-suspended now

Quoted 7 lines

- CPU0: queue is unfreeze, but the new request allocation in CPU1 may never
be done because the queue is runtime suspended, and wait_event() won't return.
And the expected result is that the queue becomes active and the allocation on
CPU1 is done immediately.

Quoted 4 lines

Hello Ming,

Just like for the scenario Jianchao reported, I will address this by only allowing the suspend to proceed if q_usage_counter == 0.

Hi Bart,

But .q_usage_counter has been zero already in the case I described, no one grabs a queue ref before starting runtime suspend.

Also, if there is in-progress PM request, the .q_usage_counter can be non-zero, but it should be fine to start the suspend.

Thanks,
Ming

#10 lore raw

PATCH v8 7/8 block: Make blk_get_request() block for non-PM requests while suspended

In reply to Bart Van Assche

Instead of allowing requests that are not power management requests to enter the queue in runtime suspended status (RPM_SUSPENDED), make the blk_get_request() caller block. This change fixes a starvation
issue: it is now guaranteed that power management requests will be executed no matter how many blk_get_request() callers are waiting. For blk-mq, instead of maintaining the q->nr_pending counter, rely on q->q_usage_counter. Call pm_runtime_mark_last_busy() every time a request finishes instead of only if the queue depth drops to zero.

Cc Christoph Hellwig <[email protected]>
Cc Ming Lei <[email protected]>
Cc Jianchao Wang <[email protected]>
Cc Hannes Reinecke <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Cc Alan Stern <[email protected]>
---
 block/blk-core.c | 37 ++++++-------------------
 block/blk-pm.c   | 70 ++++++++++++++++++++++++++++++++++++++++++++----
 2 files changed, 73 insertions(+), 34 deletions(-)
block/blk-core.c +8 -29
index 18b874d5c9c9..ae092ca121d5 100644
@@ -2747,30 +2747,6 @@ void blk_account_io_done(struct request *req, u64 now)
 	}
 }
 
-#ifdef CONFIG_PM
-/*
- * Don't process normal requests when queue is suspended
- * or in the process of suspending/resuming
- */
-static bool blk_pm_allow_request(struct request *rq)
-{
-	switch (rq->q->rpm_status) {
-	case RPM_RESUMING:
-	case RPM_SUSPENDING:
-		return rq->rq_flags & RQF_PM;
-	case RPM_SUSPENDED:
-		return false;
-	default:
-		return true;
-	}
-}
-#else
-static bool blk_pm_allow_request(struct request *rq)
-{
-	return true;
-}
-#endif
-
 void blk_account_io_start(struct request *rq, bool new_io)
 {
 	struct hd_struct *part;
@@ -2816,11 +2792,14 @@ static struct request *elv_next_request(struct request_queue *q)
 
 	while (1) {
 		list_for_each_entry(rq, &q->queue_head, queuelist) {
-			if (blk_pm_allow_request(rq))
-				return rq;
-
-			if (rq->rq_flags & RQF_SOFTBARRIER)
-				break;
+#ifdef CONFIG_PM
+			/*
+			 * If a request gets queued in state RPM_SUSPENDED
+			 * then that's a kernel bug.
+			 */
+			WARN_ON_ONCE(q->rpm_status == RPM_SUSPENDED);
+#endif
+			return rq;
 		}
 
 		/*
block/blk-pm.c +65 -5

112 lines not on this page · raw message

Signature · 1 line
2.18.0
#11 lore raw

PATCH v8 8/8 blk-mq: Enable support for runtime power management

In reply to Bart Van Assche

Now that the blk-mq core processes power management requests (marked with RQF_PREEMPT) in other states than RPM_ACTIVE, enable runtime power management for blk-mq.

Cc Christoph Hellwig <[email protected]>
Cc Ming Lei <[email protected]>
Cc Jianchao Wang <[email protected]>
Cc Hannes Reinecke <[email protected]>
Cc Johannes Thumshirn <[email protected]>
Cc Alan Stern <[email protected]>
---
 block/blk-mq.c | 2 ++
 block/blk-pm.c | 6 ------
 2 files changed, 2 insertions(+), 6 deletions(-)
block/blk-mq.c +2 -0
index 85a1c1a59c72..20fdd78b75c7 100644
@@ -33,6 +33,7 @@
 #include "blk-mq.h"
 #include "blk-mq-debugfs.h"
 #include "blk-mq-tag.h"
+#include "blk-pm.h"
 #include "blk-stat.h"
 #include "blk-mq-sched.h"
 #include "blk-rq-qos.h"
@@ -475,6 +476,7 @@ static void __blk_mq_free_request(struct request *rq)
 	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
 	const int sched_tag = rq->internal_tag;
 
+	blk_pm_mark_last_busy(rq);
 	if (rq->tag != -1)
 		blk_mq_put_tag(hctx, hctx->tags, ctx, rq->tag);
 	if (sched_tag != -1)
block/blk-pm.c +0 -6
index dc9bc45b0db5..2a7a9c0f5b99 100644
@@ -30,12 +30,6 @@
  */
 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
 {
-	/* Don't enable runtime PM for blk-mq until it is ready */
-	if (q->mq_ops) {
-		pm_runtime_disable(dev);
-		return;
-	}
-
 	q->dev = dev;
 	q->rpm_status = RPM_ACTIVE;
 	pm_runtime_set_autosuspend_delay(q->dev, -1);
Signature · 1 line
2.18.0