diff mbox series

[3/5] Use blk-stat infrastructure to collect per queue device stats

Message ID 20190426002050.15499-4-Jes.Sorensen@gmail.com (mailing list archive)
State New, archived
Headers show
Series RFC blk-mq device stats | expand

Commit Message

Jes Sorensen April 26, 2019, 12:20 a.m. UTC 
From: Jes Sorensen <jsorensen@fb.com>

Put request bytes into 8 buckets of requests, for read, write, and
discard.

Enable stats by writing 1 to /sys/block/<dev>/queue/histstat, disable
by writing 0.

Signed-off-by: Jes Sorensen <jsorensen@fb.com>
---
 block/blk-mq.c         | 52 ++++++++++++++++++++++++++++++++++++++++++
 block/blk-stat.h       |  1 +
 block/blk-sysfs.c      | 40 ++++++++++++++++++++++++++++++++
 include/linux/blkdev.h |  7 ++++++
 4 files changed, 100 insertions(+)
diff mbox series

Patch

diff --git a/block/blk-mq.c b/block/blk-mq.c
index e8cd2028745a..3b36e8e26feb 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -58,6 +58,52 @@  static int blk_mq_poll_stats_bkt(const struct request *rq)
 	return bucket;
 }
 
+/*
+ * 8 buckets for each of read, write, and discard
+ */
+static int blk_dev_stats_bkt(const struct request *rq)
+{
+	int grp, bucket;
+
+	grp = op_stat_group(req_op(rq));
+
+	bucket = grp + 3*(ilog2(rq->io_bytes) - 9);
+
+	if (bucket < 0)
+		return -1;
+	else if (bucket >= BLK_DEV_STATS_BKTS)
+		return grp + BLK_DEV_STATS_BKTS - 3;
+
+	return bucket;
+}
+
+/*
+ * Copy out the stats to their official location
+ */
+static void blk_dev_stats_cb(struct blk_stat_callback *cb)
+{
+	struct request_queue *q = cb->data;
+	int bucket;
+
+	for (bucket = 0; bucket < BLK_DEV_STATS_BKTS; bucket++) {
+		if (cb->stat[bucket].nr_samples) {
+			q->dev_stat[bucket].size +=
+				cb->stat[bucket].size;
+			q->dev_stat[bucket].nr_samples +=
+				cb->stat[bucket].nr_samples;
+		}
+	}
+
+	if (!blk_stat_is_active(cb))
+		blk_stat_activate_msecs(cb, 100);
+}
+
+void blk_dev_stats_free(struct request_queue *q)
+{
+	blk_stat_remove_callback(q, q->dev_cb);
+	blk_stat_free_callback(q->dev_cb);
+}
+
 /*
  * Check if any of the ctx, dispatch list or elevator
  * have pending work in this hardware queue.
@@ -2851,6 +2897,12 @@  struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 	if (!q->nr_hw_queues)
 		goto err_hctxs;
 
+	q->dev_cb = blk_stat_alloc_callback(blk_dev_stats_cb,
+					    blk_dev_stats_bkt,
+					    BLK_DEV_STATS_BKTS, q);
+	if (!q->dev_cb)
+		goto err_hctxs;
+
 	INIT_WORK(&q->timeout_work, blk_mq_timeout_work);
 	blk_queue_rq_timeout(q, set->timeout ? set->timeout : 30 * HZ);
 
diff --git a/block/blk-stat.h b/block/blk-stat.h
index ea893c4a9af1..7f0c8b737a9d 100644
--- a/block/blk-stat.h
+++ b/block/blk-stat.h
@@ -168,4 +168,5 @@  void blk_rq_stat_add(struct blk_rq_stat *, u64, u64);
 void blk_rq_stat_sum(struct blk_rq_stat *, struct blk_rq_stat *);
 void blk_rq_stat_init(struct blk_rq_stat *);
 
+void blk_dev_stats_free(struct request_queue *q);
 #endif
diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 422327089e0f..394760754bfc 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -530,6 +530,37 @@  static ssize_t queue_dax_show(struct request_queue *q, char *page)
 	return queue_var_show(blk_queue_dax(q), page);
 }
 
+static ssize_t queue_histstat_show(struct request_queue *q, char *page)
+{
+	return queue_var_show(test_bit(QUEUE_FLAG_HISTSTATS,
+				       &q->queue_flags), page);
+}
+
+static ssize_t queue_histstat_store(struct request_queue *q, const char *page,
+				    size_t size)
+{
+	unsigned long histstat_on;
+	ssize_t ret;
+
+	ret = queue_var_store(&histstat_on, page, size);
+	if (ret < 0)
+		return ret;
+
+	if (histstat_on) {
+		if (!blk_queue_flag_test_and_set(QUEUE_FLAG_HISTSTATS, q))
+			blk_stat_add_callback(q, q->dev_cb);
+		if (!blk_stat_is_active(q->dev_cb))
+			blk_stat_activate_msecs(q->dev_cb, 100);
+	} else {
+		if (test_bit(QUEUE_FLAG_HISTSTATS, &q->queue_flags)) {
+			blk_stat_remove_callback(q, q->dev_cb);
+			blk_queue_flag_clear(QUEUE_FLAG_HISTSTATS, q);
+		}
+	}
+
+	return ret;
+}
+
 static struct queue_sysfs_entry queue_requests_entry = {
 	.attr = {.name = "nr_requests", .mode = 0644 },
 	.show = queue_requests_show,
@@ -728,6 +759,12 @@  static struct queue_sysfs_entry throtl_sample_time_entry = {
 };
 #endif
 
+static struct queue_sysfs_entry queue_histstat_entry = {
+	.attr = {.name = "histstat", .mode = 0644 },
+	.show = queue_histstat_show,
+	.store = queue_histstat_store,
+};
+
 static struct attribute *default_attrs[] = {
 	&queue_requests_entry.attr,
 	&queue_ra_entry.attr,
@@ -767,6 +804,7 @@  static struct attribute *default_attrs[] = {
 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 	&throtl_sample_time_entry.attr,
 #endif
+	&queue_histstat_entry.attr,
 	NULL,
 };
 
@@ -837,6 +875,8 @@  static void __blk_release_queue(struct work_struct *work)
 {
 	struct request_queue *q = container_of(work, typeof(*q), release_work);
 
+	blk_dev_stats_free(q);
+
 	if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
 		blk_stat_remove_callback(q, q->poll_cb);
 	blk_stat_free_callback(q->poll_cb);
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index a2291b1d1cb0..54f7718cd6a2 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -53,6 +53,9 @@  struct blk_stat_callback;
 /* Doing classic polling */
 #define BLK_MQ_POLL_CLASSIC -1
 
+/* Must be consistent with blk_part_stats_bkt() */
+#define BLK_DEV_STATS_BKTS (3 * 8)
+
 /*
  * Maximum number of blkcg policies allowed to be registered concurrently.
  * Defined here to simplify include dependency.
@@ -478,6 +481,9 @@  struct request_queue {
 	struct blk_stat_callback	*poll_cb;
 	struct blk_rq_stat	poll_stat[BLK_MQ_POLL_STATS_BKTS];
 
+	struct blk_stat_callback	*dev_cb;
+	struct blk_rq_stat	dev_stat[BLK_DEV_STATS_BKTS];
+
 	struct timer_list	timeout;
 	struct work_struct	timeout_work;
 
@@ -593,6 +599,7 @@  struct request_queue {
 #define QUEUE_FLAG_SCSI_PASSTHROUGH 23	/* queue supports SCSI commands */
 #define QUEUE_FLAG_QUIESCED	24	/* queue has been quiesced */
 #define QUEUE_FLAG_PCI_P2PDMA	25	/* device supports PCI p2p requests */
+#define QUEUE_FLAG_HISTSTATS	26	/* Hist stats enabled if set */
 
 #define QUEUE_FLAG_MQ_DEFAULT	((1 << QUEUE_FLAG_IO_STAT) |		\
 				 (1 << QUEUE_FLAG_SAME_COMP))