* [PATCH 0/4] block: callback-based statistics
@ 2017-03-14 21:03 Omar Sandoval
2017-03-14 21:03 ` [PATCH 1/4] block: tear out blk-stat debugfs/sysfs entries Omar Sandoval
` (4 more replies)
0 siblings, 5 replies; 6+ messages in thread
From: Omar Sandoval @ 2017-03-14 21:03 UTC (permalink / raw)
To: linux-block; +Cc: kernel-team
From: Omar Sandoval <osandov@fb.com>
This patchset generalizes the blk-stats infrastructure to allow users to
register a callback to be called at a given time with the statistics of
requests completed during that window. Writeback throttling and hybrid
polling are converted to the new infrastructure. The ultimate goal is to
use this infrastructure for the mq I/O scheduler.
The details are in patch 4, which is the actual conversion. Patches 1-3
are preparation cleanups.
A couple of minor open issues:
- Since there is no single source of truth for stats anymore (it's
per-callback now), I got rid of the sysfs/debugfs files. If we still
want these, we could potentially expose the statistics that poll is
using.
- Polling used to use per-hctx stats, but now the stats are gathered to
a per-request_queue cache. This might make the stats noisier, but we
can improve this in the future by splitting out stats by request size
or something like that.
Omar Sandoval (4):
block: tear out blk-stat debugfs/sysfs entries
block: remove extra calls to wbt_exit()
block: change stats from array type to struct type
blk-stats: convert to callback-based statistics reporting
block/blk-core.c | 7 +-
block/blk-mq-debugfs.c | 55 ---------
block/blk-mq.c | 79 ++++++++----
block/blk-mq.h | 1 -
block/blk-stat.c | 300 ++++++++++++++++++++-------------------------
block/blk-stat.h | 98 ++++++++++++---
block/blk-sysfs.c | 31 +----
block/blk-wbt.c | 66 +++++-----
block/blk-wbt.h | 2 +-
include/linux/blk_types.h | 8 +-
include/linux/blkdev.h | 10 +-
include/trace/events/wbt.h | 20 +--
12 files changed, 331 insertions(+), 346 deletions(-)
--
2.12.0
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH 1/4] block: tear out blk-stat debugfs/sysfs entries
2017-03-14 21:03 [PATCH 0/4] block: callback-based statistics Omar Sandoval
@ 2017-03-14 21:03 ` Omar Sandoval
2017-03-14 21:03 ` [PATCH 2/4] block: remove extra calls to wbt_exit() Omar Sandoval
` (3 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Omar Sandoval @ 2017-03-14 21:03 UTC (permalink / raw)
To: linux-block; +Cc: kernel-team
From: Omar Sandoval <osandov@fb.com>
These made sense when there was a single stats buffer for the request
queue, but that's going away.
Signed-off-by: Omar Sandoval <osandov@fb.com>
---
block/blk-mq-debugfs.c | 55 --------------------------------------------------
block/blk-sysfs.c | 26 ------------------------
2 files changed, 81 deletions(-)
diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c
index f6d917977b33..18672225c417 100644
--- a/block/blk-mq-debugfs.c
+++ b/block/blk-mq-debugfs.c
@@ -322,60 +322,6 @@ static const struct file_operations hctx_io_poll_fops = {
.release = single_release,
};
-static void print_stat(struct seq_file *m, struct blk_rq_stat *stat)
-{
- seq_printf(m, "samples=%d, mean=%lld, min=%llu, max=%llu",
- stat->nr_samples, stat->mean, stat->min, stat->max);
-}
-
-static int hctx_stats_show(struct seq_file *m, void *v)
-{
- struct blk_mq_hw_ctx *hctx = m->private;
- struct blk_rq_stat stat[2];
-
- blk_stat_init(&stat[BLK_STAT_READ]);
- blk_stat_init(&stat[BLK_STAT_WRITE]);
-
- blk_hctx_stat_get(hctx, stat);
-
- seq_puts(m, "read: ");
- print_stat(m, &stat[BLK_STAT_READ]);
- seq_puts(m, "\n");
-
- seq_puts(m, "write: ");
- print_stat(m, &stat[BLK_STAT_WRITE]);
- seq_puts(m, "\n");
- return 0;
-}
-
-static int hctx_stats_open(struct inode *inode, struct file *file)
-{
- return single_open(file, hctx_stats_show, inode->i_private);
-}
-
-static ssize_t hctx_stats_write(struct file *file, const char __user *buf,
- size_t count, loff_t *ppos)
-{
- struct seq_file *m = file->private_data;
- struct blk_mq_hw_ctx *hctx = m->private;
- struct blk_mq_ctx *ctx;
- int i;
-
- hctx_for_each_ctx(hctx, ctx, i) {
- blk_stat_init(&ctx->stat[BLK_STAT_READ]);
- blk_stat_init(&ctx->stat[BLK_STAT_WRITE]);
- }
- return count;
-}
-
-static const struct file_operations hctx_stats_fops = {
- .open = hctx_stats_open,
- .read = seq_read,
- .write = hctx_stats_write,
- .llseek = seq_lseek,
- .release = single_release,
-};
-
static int hctx_dispatched_show(struct seq_file *m, void *v)
{
struct blk_mq_hw_ctx *hctx = m->private;
@@ -646,7 +592,6 @@ static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
{"sched_tags", 0400, &hctx_sched_tags_fops},
{"sched_tags_bitmap", 0400, &hctx_sched_tags_bitmap_fops},
{"io_poll", 0600, &hctx_io_poll_fops},
- {"stats", 0600, &hctx_stats_fops},
{"dispatched", 0600, &hctx_dispatched_fops},
{"queued", 0600, &hctx_queued_fops},
{"run", 0600, &hctx_run_fops},
diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index c44b321335f3..6e1cf622af21 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -503,26 +503,6 @@ static ssize_t queue_dax_show(struct request_queue *q, char *page)
return queue_var_show(blk_queue_dax(q), page);
}
-static ssize_t print_stat(char *page, struct blk_rq_stat *stat, const char *pre)
-{
- return sprintf(page, "%s samples=%llu, mean=%lld, min=%lld, max=%lld\n",
- pre, (long long) stat->nr_samples,
- (long long) stat->mean, (long long) stat->min,
- (long long) stat->max);
-}
-
-static ssize_t queue_stats_show(struct request_queue *q, char *page)
-{
- struct blk_rq_stat stat[2];
- ssize_t ret;
-
- blk_queue_stat_get(q, stat);
-
- ret = print_stat(page, &stat[BLK_STAT_READ], "read :");
- ret += print_stat(page + ret, &stat[BLK_STAT_WRITE], "write:");
- return ret;
-}
-
static struct queue_sysfs_entry queue_requests_entry = {
.attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
.show = queue_requests_show,
@@ -691,11 +671,6 @@ static struct queue_sysfs_entry queue_dax_entry = {
.show = queue_dax_show,
};
-static struct queue_sysfs_entry queue_stats_entry = {
- .attr = {.name = "stats", .mode = S_IRUGO },
- .show = queue_stats_show,
-};
-
static struct queue_sysfs_entry queue_wb_lat_entry = {
.attr = {.name = "wbt_lat_usec", .mode = S_IRUGO | S_IWUSR },
.show = queue_wb_lat_show,
@@ -733,7 +708,6 @@ static struct attribute *default_attrs[] = {
&queue_poll_entry.attr,
&queue_wc_entry.attr,
&queue_dax_entry.attr,
- &queue_stats_entry.attr,
&queue_wb_lat_entry.attr,
&queue_poll_delay_entry.attr,
NULL,
--
2.12.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH 2/4] block: remove extra calls to wbt_exit()
2017-03-14 21:03 [PATCH 0/4] block: callback-based statistics Omar Sandoval
2017-03-14 21:03 ` [PATCH 1/4] block: tear out blk-stat debugfs/sysfs entries Omar Sandoval
@ 2017-03-14 21:03 ` Omar Sandoval
2017-03-14 21:03 ` [PATCH 3/4] block: change stats from array type to struct type Omar Sandoval
` (2 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Omar Sandoval @ 2017-03-14 21:03 UTC (permalink / raw)
To: linux-block; +Cc: kernel-team
From: Omar Sandoval <osandov@fb.com>
We always call wbt_exit() from blk_release_queue(), so these are
unnecessary.
Signed-off-by: Omar Sandoval <osandov@fb.com>
---
block/blk-core.c | 1 -
block/blk-mq.c | 2 --
2 files changed, 3 deletions(-)
diff --git a/block/blk-core.c b/block/blk-core.c
index 0eeb99ef654f..82425017c9b8 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -889,7 +889,6 @@ int blk_init_allocated_queue(struct request_queue *q)
q->exit_rq_fn(q, q->fq->flush_rq);
out_free_flush_queue:
blk_free_flush_queue(q->fq);
- wbt_exit(q);
return -ENOMEM;
}
EXPORT_SYMBOL(blk_init_allocated_queue);
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 159187a28d66..eb84daf550f4 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2430,8 +2430,6 @@ void blk_mq_free_queue(struct request_queue *q)
list_del_init(&q->all_q_node);
mutex_unlock(&all_q_mutex);
- wbt_exit(q);
-
blk_mq_del_queue_tag_set(q);
blk_mq_exit_hw_queues(q, set, set->nr_hw_queues);
--
2.12.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH 3/4] block: change stats from array type to struct type
2017-03-14 21:03 [PATCH 0/4] block: callback-based statistics Omar Sandoval
2017-03-14 21:03 ` [PATCH 1/4] block: tear out blk-stat debugfs/sysfs entries Omar Sandoval
2017-03-14 21:03 ` [PATCH 2/4] block: remove extra calls to wbt_exit() Omar Sandoval
@ 2017-03-14 21:03 ` Omar Sandoval
2017-03-14 21:03 ` [PATCH 4/4] blk-stats: convert to callback-based statistics reporting Omar Sandoval
2017-03-14 21:07 ` [PATCH 0/4] block: callback-based statistics Omar Sandoval
4 siblings, 0 replies; 6+ messages in thread
From: Omar Sandoval @ 2017-03-14 21:03 UTC (permalink / raw)
To: linux-block; +Cc: kernel-team
From: Omar Sandoval <osandov@fb.com>
This is cleanup in preparation for callback-based stats with a couple of
(minor) advantages:
1. Allocating a percpu array type is awkward.
2. It makes the callback type signature more clear. With `struct
blk_rq_stat *`, the only way to know that you're getting an array of
read and write stats is to look at the implementation or other users.
With the struct type, it's immediately obvious.
Note that I left q->rq_stats and ctx->stat alone since those are going
away anyways.
Signed-off-by: Omar Sandoval <osandov@fb.com>
---
block/blk-mq.c | 14 +++++++-------
block/blk-stat.c | 34 +++++++++++++++++-----------------
block/blk-stat.h | 4 ++--
block/blk-wbt.c | 28 ++++++++++++++--------------
include/linux/blk_types.h | 5 +++++
include/trace/events/wbt.h | 20 ++++++++++----------
6 files changed, 55 insertions(+), 50 deletions(-)
diff --git a/block/blk-mq.c b/block/blk-mq.c
index eb84daf550f4..e16f8d420683 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2743,7 +2743,7 @@ static unsigned long blk_mq_poll_nsecs(struct request_queue *q,
struct blk_mq_hw_ctx *hctx,
struct request *rq)
{
- struct blk_rq_stat stat[2];
+ struct blk_stats stats;
unsigned long ret = 0;
/*
@@ -2757,8 +2757,8 @@ static unsigned long blk_mq_poll_nsecs(struct request_queue *q,
* We don't have to do this once per IO, should optimize this
* to just use the current window of stats until it changes
*/
- memset(&stat, 0, sizeof(stat));
- blk_hctx_stat_get(hctx, stat);
+ memset(&stats, 0, sizeof(stats));
+ blk_hctx_stat_get(hctx, &stats);
/*
* As an optimistic guess, use half of the mean service time
@@ -2768,10 +2768,10 @@ static unsigned long blk_mq_poll_nsecs(struct request_queue *q,
* important on devices where the completion latencies are longer
* than ~10 usec.
*/
- if (req_op(rq) == REQ_OP_READ && stat[BLK_STAT_READ].nr_samples)
- ret = (stat[BLK_STAT_READ].mean + 1) / 2;
- else if (req_op(rq) == REQ_OP_WRITE && stat[BLK_STAT_WRITE].nr_samples)
- ret = (stat[BLK_STAT_WRITE].mean + 1) / 2;
+ if (req_op(rq) == REQ_OP_READ && stats.read.nr_samples)
+ ret = (stats.read.mean + 1) / 2;
+ else if (req_op(rq) == REQ_OP_WRITE && stats.write.nr_samples)
+ ret = (stats.write.mean + 1) / 2;
return ret;
}
diff --git a/block/blk-stat.c b/block/blk-stat.c
index 9b43efb8933f..40f6c90e432a 100644
--- a/block/blk-stat.c
+++ b/block/blk-stat.c
@@ -48,15 +48,15 @@ static void blk_stat_sum(struct blk_rq_stat *dst, struct blk_rq_stat *src)
dst->nr_samples += src->nr_samples;
}
-static void blk_mq_stat_get(struct request_queue *q, struct blk_rq_stat *dst)
+static void blk_mq_stat_get(struct request_queue *q, struct blk_stats *stats)
{
struct blk_mq_hw_ctx *hctx;
struct blk_mq_ctx *ctx;
uint64_t latest = 0;
int i, j, nr;
- blk_stat_init(&dst[BLK_STAT_READ]);
- blk_stat_init(&dst[BLK_STAT_WRITE]);
+ blk_stat_init(&stats->read);
+ blk_stat_init(&stats->write);
nr = 0;
do {
@@ -89,12 +89,12 @@ static void blk_mq_stat_get(struct request_queue *q, struct blk_rq_stat *dst)
queue_for_each_hw_ctx(q, hctx, i) {
hctx_for_each_ctx(hctx, ctx, j) {
if (ctx->stat[BLK_STAT_READ].time == newest) {
- blk_stat_sum(&dst[BLK_STAT_READ],
+ blk_stat_sum(&stats->read,
&ctx->stat[BLK_STAT_READ]);
nr++;
}
if (ctx->stat[BLK_STAT_WRITE].time == newest) {
- blk_stat_sum(&dst[BLK_STAT_WRITE],
+ blk_stat_sum(&stats->write,
&ctx->stat[BLK_STAT_WRITE]);
nr++;
}
@@ -106,24 +106,24 @@ static void blk_mq_stat_get(struct request_queue *q, struct blk_rq_stat *dst)
*/
} while (!nr);
- dst[BLK_STAT_READ].time = dst[BLK_STAT_WRITE].time = latest;
+ stats->read.time = stats->write.time = latest;
}
-void blk_queue_stat_get(struct request_queue *q, struct blk_rq_stat *dst)
+void blk_queue_stat_get(struct request_queue *q, struct blk_stats *stats)
{
if (q->mq_ops)
- blk_mq_stat_get(q, dst);
+ blk_mq_stat_get(q, stats);
else {
blk_stat_flush_batch(&q->rq_stats[BLK_STAT_READ]);
blk_stat_flush_batch(&q->rq_stats[BLK_STAT_WRITE]);
- memcpy(&dst[BLK_STAT_READ], &q->rq_stats[BLK_STAT_READ],
- sizeof(struct blk_rq_stat));
- memcpy(&dst[BLK_STAT_WRITE], &q->rq_stats[BLK_STAT_WRITE],
- sizeof(struct blk_rq_stat));
+ memcpy(&stats->read, &q->rq_stats[BLK_STAT_READ],
+ sizeof(struct blk_rq_stat));
+ memcpy(&stats->write, &q->rq_stats[BLK_STAT_WRITE],
+ sizeof(struct blk_rq_stat));
}
}
-void blk_hctx_stat_get(struct blk_mq_hw_ctx *hctx, struct blk_rq_stat *dst)
+void blk_hctx_stat_get(struct blk_mq_hw_ctx *hctx, struct blk_stats *stats)
{
struct blk_mq_ctx *ctx;
unsigned int i, nr;
@@ -151,13 +151,13 @@ void blk_hctx_stat_get(struct blk_mq_hw_ctx *hctx, struct blk_rq_stat *dst)
hctx_for_each_ctx(hctx, ctx, i) {
if (ctx->stat[BLK_STAT_READ].time == newest) {
- blk_stat_sum(&dst[BLK_STAT_READ],
- &ctx->stat[BLK_STAT_READ]);
+ blk_stat_sum(&stats->read,
+ &ctx->stat[BLK_STAT_READ]);
nr++;
}
if (ctx->stat[BLK_STAT_WRITE].time == newest) {
- blk_stat_sum(&dst[BLK_STAT_WRITE],
- &ctx->stat[BLK_STAT_WRITE]);
+ blk_stat_sum(&stats->write,
+ &ctx->stat[BLK_STAT_WRITE]);
nr++;
}
}
diff --git a/block/blk-stat.h b/block/blk-stat.h
index a2050a0a5314..a24439aab710 100644
--- a/block/blk-stat.h
+++ b/block/blk-stat.h
@@ -21,8 +21,8 @@ enum {
};
void blk_stat_add(struct blk_rq_stat *, struct request *);
-void blk_hctx_stat_get(struct blk_mq_hw_ctx *, struct blk_rq_stat *);
-void blk_queue_stat_get(struct request_queue *, struct blk_rq_stat *);
+void blk_hctx_stat_get(struct blk_mq_hw_ctx *, struct blk_stats *);
+void blk_queue_stat_get(struct request_queue *, struct blk_stats *);
void blk_stat_clear(struct request_queue *);
void blk_stat_init(struct blk_rq_stat *);
bool blk_stat_is_current(struct blk_rq_stat *);
diff --git a/block/blk-wbt.c b/block/blk-wbt.c
index 1aedb1f7ee0c..5e81068f7c52 100644
--- a/block/blk-wbt.c
+++ b/block/blk-wbt.c
@@ -247,7 +247,7 @@ static bool calc_wb_limits(struct rq_wb *rwb)
return ret;
}
-static inline bool stat_sample_valid(struct blk_rq_stat *stat)
+static inline bool stat_sample_valid(struct blk_stats *stats)
{
/*
* We need at least one read sample, and a minimum of
@@ -255,8 +255,8 @@ static inline bool stat_sample_valid(struct blk_rq_stat *stat)
* that it's writes impacting us, and not just some sole read on
* a device that is in a lower power state.
*/
- return stat[BLK_STAT_READ].nr_samples >= 1 &&
- stat[BLK_STAT_WRITE].nr_samples >= RWB_MIN_WRITE_SAMPLES;
+ return (stats->read.nr_samples >= 1 &&
+ stats->write.nr_samples >= RWB_MIN_WRITE_SAMPLES);
}
static u64 rwb_sync_issue_lat(struct rq_wb *rwb)
@@ -277,7 +277,7 @@ enum {
LAT_EXCEEDED,
};
-static int __latency_exceeded(struct rq_wb *rwb, struct blk_rq_stat *stat)
+static int __latency_exceeded(struct rq_wb *rwb, struct blk_stats *stats)
{
struct backing_dev_info *bdi = rwb->queue->backing_dev_info;
u64 thislat;
@@ -293,7 +293,7 @@ static int __latency_exceeded(struct rq_wb *rwb, struct blk_rq_stat *stat)
*/
thislat = rwb_sync_issue_lat(rwb);
if (thislat > rwb->cur_win_nsec ||
- (thislat > rwb->min_lat_nsec && !stat[BLK_STAT_READ].nr_samples)) {
+ (thislat > rwb->min_lat_nsec && !stats->read.nr_samples)) {
trace_wbt_lat(bdi, thislat);
return LAT_EXCEEDED;
}
@@ -301,14 +301,14 @@ static int __latency_exceeded(struct rq_wb *rwb, struct blk_rq_stat *stat)
/*
* No read/write mix, if stat isn't valid
*/
- if (!stat_sample_valid(stat)) {
+ if (!stat_sample_valid(stats)) {
/*
* If we had writes in this stat window and the window is
* current, we're only doing writes. If a task recently
* waited or still has writes in flights, consider us doing
* just writes as well.
*/
- if ((stat[BLK_STAT_WRITE].nr_samples && blk_stat_is_current(stat)) ||
+ if ((stats->write.nr_samples && blk_stat_is_current(&stats->read)) ||
wb_recent_wait(rwb) || wbt_inflight(rwb))
return LAT_UNKNOWN_WRITES;
return LAT_UNKNOWN;
@@ -317,24 +317,24 @@ static int __latency_exceeded(struct rq_wb *rwb, struct blk_rq_stat *stat)
/*
* If the 'min' latency exceeds our target, step down.
*/
- if (stat[BLK_STAT_READ].min > rwb->min_lat_nsec) {
- trace_wbt_lat(bdi, stat[BLK_STAT_READ].min);
- trace_wbt_stat(bdi, stat);
+ if (stats->read.min > rwb->min_lat_nsec) {
+ trace_wbt_lat(bdi, stats->read.min);
+ trace_wbt_stat(bdi, stats);
return LAT_EXCEEDED;
}
if (rwb->scale_step)
- trace_wbt_stat(bdi, stat);
+ trace_wbt_stat(bdi, stats);
return LAT_OK;
}
static int latency_exceeded(struct rq_wb *rwb)
{
- struct blk_rq_stat stat[2];
+ struct blk_stats stats;
- blk_queue_stat_get(rwb->queue, stat);
- return __latency_exceeded(rwb, stat);
+ blk_queue_stat_get(rwb->queue, &stats);
+ return __latency_exceeded(rwb, &stats);
}
static void rwb_trace_step(struct rq_wb *rwb, const char *msg)
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index d703acb55d0f..5e4c5380edf0 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -299,4 +299,9 @@ struct blk_rq_stat {
s64 time;
};
+struct blk_stats {
+ struct blk_rq_stat read;
+ struct blk_rq_stat write;
+};
+
#endif /* __LINUX_BLK_TYPES_H */
diff --git a/include/trace/events/wbt.h b/include/trace/events/wbt.h
index 3c518e455680..5b7f8ef73532 100644
--- a/include/trace/events/wbt.h
+++ b/include/trace/events/wbt.h
@@ -13,9 +13,9 @@
*/
TRACE_EVENT(wbt_stat,
- TP_PROTO(struct backing_dev_info *bdi, struct blk_rq_stat *stat),
+ TP_PROTO(struct backing_dev_info *bdi, struct blk_stats *stats),
- TP_ARGS(bdi, stat),
+ TP_ARGS(bdi, stats),
TP_STRUCT__entry(
__array(char, name, 32)
@@ -33,14 +33,14 @@ TRACE_EVENT(wbt_stat,
TP_fast_assign(
strncpy(__entry->name, dev_name(bdi->dev), 32);
- __entry->rmean = stat[0].mean;
- __entry->rmin = stat[0].min;
- __entry->rmax = stat[0].max;
- __entry->rnr_samples = stat[0].nr_samples;
- __entry->wmean = stat[1].mean;
- __entry->wmin = stat[1].min;
- __entry->wmax = stat[1].max;
- __entry->wnr_samples = stat[1].nr_samples;
+ __entry->rmean = stats->read.mean;
+ __entry->rmin = stats->read.min;
+ __entry->rmax = stats->read.max;
+ __entry->rnr_samples = stats->read.nr_samples;
+ __entry->wmean = stats->write.mean;
+ __entry->wmin = stats->write.min;
+ __entry->wmax = stats->write.max;
+ __entry->wnr_samples = stats->write.nr_samples;
),
TP_printk("%s: rmean=%llu, rmin=%llu, rmax=%llu, rsamples=%llu, "
--
2.12.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH 4/4] blk-stats: convert to callback-based statistics reporting
2017-03-14 21:03 [PATCH 0/4] block: callback-based statistics Omar Sandoval
` (2 preceding siblings ...)
2017-03-14 21:03 ` [PATCH 3/4] block: change stats from array type to struct type Omar Sandoval
@ 2017-03-14 21:03 ` Omar Sandoval
2017-03-14 21:07 ` [PATCH 0/4] block: callback-based statistics Omar Sandoval
4 siblings, 0 replies; 6+ messages in thread
From: Omar Sandoval @ 2017-03-14 21:03 UTC (permalink / raw)
To: linux-block; +Cc: kernel-team
From: Omar Sandoval <osandov@fb.com>
Currently, statistics are gathered in ~0.13s windows, and users grab the
statistics whenever they need them. This is not ideal for both in-tree
users:
1. Writeback throttling wants its own dynamically sized window of
statistics. Since the blk-stats statistics are reset after every
window and the wbt windows don't line up with the blk-stats windows,
wbt doesn't see every I/O.
2. Polling currently grabs the statistics on every I/O. Again, depending
on how the window lines up, we may miss some I/Os. It's also
unnecessary overhead to get the statistics on every I/O; the hybrid
polling heuristic would be just as happy with the statistics from the
previous full window.
This reworks the blk-stats infrastructure to be callback-based: users
register a callback that they want called at a given time with all of
the statistics from the window during which the callback was active. The
callbacks are kept on an RCU list, and each callback has percpu stats
buffers. There will only be a few users, so the overhead on the I/O
completion side is low. The stats flushing is also simplified
considerably: since the timer function is responsible for clearing the
statistics, we don't have to worry about stale statistics.
wbt is a trivial conversion. After the conversion, the windowing problem
mentioned above is fixed.
For polling, we register an extra callback that caches the previous
window's statistics in the struct request_queue for the hybrid polling
heuristic to use.
Signed-off-by: Omar Sandoval <osandov@fb.com>
---
block/blk-core.c | 6 +-
block/blk-mq.c | 77 ++++++++----
block/blk-mq.h | 1 -
block/blk-stat.c | 300 +++++++++++++++++++++-------------------------
block/blk-stat.h | 98 ++++++++++++---
block/blk-sysfs.c | 5 +
block/blk-wbt.c | 48 +++-----
block/blk-wbt.h | 2 +-
include/linux/blk_types.h | 3 -
include/linux/blkdev.h | 10 +-
10 files changed, 307 insertions(+), 243 deletions(-)
diff --git a/block/blk-core.c b/block/blk-core.c
index 82425017c9b8..0f4ae118e220 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -852,6 +852,10 @@ static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
int blk_init_allocated_queue(struct request_queue *q)
{
+ q->stats = blk_alloc_queue_stats();
+ if (!q->stats)
+ return -ENOMEM;
+
q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
if (!q->fq)
return -ENOMEM;
@@ -2692,7 +2696,7 @@ void blk_finish_request(struct request *req, int error)
struct request_queue *q = req->q;
if (req->rq_flags & RQF_STATS)
- blk_stat_add(&q->rq_stats[rq_data_dir(req)], req);
+ blk_stat_add(req);
if (req->rq_flags & RQF_QUEUED)
blk_queue_end_tag(q, req);
diff --git a/block/blk-mq.c b/block/blk-mq.c
index e16f8d420683..559f9b0f24a1 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -39,6 +39,10 @@
static DEFINE_MUTEX(all_q_mutex);
static LIST_HEAD(all_q_list);
+static void blk_mq_poll_stats_start(struct request_queue *q);
+static void blk_mq_poll_stats_fn(struct blk_stat_callback *cb,
+ struct blk_stats *stats);
+
/*
* Check if any of the ctx's have pending work in this hardware queue
*/
@@ -432,15 +436,8 @@ static void blk_mq_ipi_complete_request(struct request *rq)
static void blk_mq_stat_add(struct request *rq)
{
if (rq->rq_flags & RQF_STATS) {
- /*
- * We could rq->mq_ctx here, but there's less of a risk
- * of races if we have the completion event add the stats
- * to the local software queue.
- */
- struct blk_mq_ctx *ctx;
-
- ctx = __blk_mq_get_ctx(rq->q, raw_smp_processor_id());
- blk_stat_add(&ctx->stat[rq_data_dir(rq)], rq);
+ blk_mq_poll_stats_start(rq->q);
+ blk_stat_add(rq);
}
}
@@ -2039,8 +2036,6 @@ static void blk_mq_init_cpu_queues(struct request_queue *q,
spin_lock_init(&__ctx->lock);
INIT_LIST_HEAD(&__ctx->rq_list);
__ctx->queue = q;
- blk_stat_init(&__ctx->stat[BLK_STAT_READ]);
- blk_stat_init(&__ctx->stat[BLK_STAT_WRITE]);
/* If the cpu isn't online, the cpu is mapped to first hctx */
if (!cpu_online(i))
@@ -2338,6 +2333,14 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
/* mark the queue as mq asap */
q->mq_ops = set->ops;
+ q->stats = blk_alloc_queue_stats();
+ if (!q->stats)
+ goto err_exit;
+
+ q->poll_cb = blk_stat_alloc_callback(blk_mq_poll_stats_fn, q);
+ if (!q->poll_cb)
+ goto err_exit;
+
q->queue_ctx = alloc_percpu(struct blk_mq_ctx);
if (!q->queue_ctx)
goto err_exit;
@@ -2739,28 +2742,54 @@ void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues)
}
EXPORT_SYMBOL_GPL(blk_mq_update_nr_hw_queues);
+/* Enable polling stats and return whether they were already enabled. */
+static bool blk_poll_stats_enable(struct request_queue *q)
+{
+ if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags) ||
+ test_and_set_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
+ return true;
+ blk_stat_add_callback(q, q->poll_cb);
+ return false;
+}
+
+static void blk_mq_poll_stats_start(struct request_queue *q)
+{
+ /*
+ * We don't arm the callback if polling stats are not enabled or the
+ * callback has already been armed.
+ */
+ if (!test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags) ||
+ timer_pending(&q->poll_cb->timer))
+ return;
+
+ blk_stat_arm_callback(q->poll_cb, jiffies + msecs_to_jiffies(100));
+}
+
+static void blk_mq_poll_stats_fn(struct blk_stat_callback *cb,
+ struct blk_stats *stats)
+{
+ struct request_queue *q = cb->data;
+
+ if (stats->read.nr_samples)
+ q->poll_stats.read = stats->read;
+ if (stats->write.nr_samples)
+ q->poll_stats.write = stats->write;
+}
+
static unsigned long blk_mq_poll_nsecs(struct request_queue *q,
struct blk_mq_hw_ctx *hctx,
struct request *rq)
{
- struct blk_stats stats;
unsigned long ret = 0;
/*
* If stats collection isn't on, don't sleep but turn it on for
* future users
*/
- if (!blk_stat_enable(q))
+ if (!blk_poll_stats_enable(q))
return 0;
/*
- * We don't have to do this once per IO, should optimize this
- * to just use the current window of stats until it changes
- */
- memset(&stats, 0, sizeof(stats));
- blk_hctx_stat_get(hctx, &stats);
-
- /*
* As an optimistic guess, use half of the mean service time
* for this type of request. We can (and should) make this smarter.
* For instance, if the completion latencies are tight, we can
@@ -2768,10 +2797,10 @@ static unsigned long blk_mq_poll_nsecs(struct request_queue *q,
* important on devices where the completion latencies are longer
* than ~10 usec.
*/
- if (req_op(rq) == REQ_OP_READ && stats.read.nr_samples)
- ret = (stats.read.mean + 1) / 2;
- else if (req_op(rq) == REQ_OP_WRITE && stats.write.nr_samples)
- ret = (stats.write.mean + 1) / 2;
+ if (req_op(rq) == REQ_OP_READ && q->poll_stats.read.nr_samples)
+ ret = (q->poll_stats.read.mean + 1) / 2;
+ else if (req_op(rq) == REQ_OP_WRITE && q->poll_stats.write.nr_samples)
+ ret = (q->poll_stats.write.mean + 1) / 2;
return ret;
}
diff --git a/block/blk-mq.h b/block/blk-mq.h
index b79f9a7d8cf6..8d49c06fc520 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -20,7 +20,6 @@ struct blk_mq_ctx {
/* incremented at completion time */
unsigned long ____cacheline_aligned_in_smp rq_completed[2];
- struct blk_rq_stat stat[2];
struct request_queue *queue;
struct kobject kobj;
diff --git a/block/blk-stat.c b/block/blk-stat.c
index 40f6c90e432a..a71cfb35712d 100644
--- a/block/blk-stat.c
+++ b/block/blk-stat.c
@@ -4,11 +4,24 @@
* Copyright (C) 2016 Jens Axboe
*/
#include <linux/kernel.h>
+#include <linux/rculist.h>
#include <linux/blk-mq.h>
#include "blk-stat.h"
#include "blk-mq.h"
+struct blk_queue_stats {
+ struct list_head callbacks;
+ spinlock_t lock;
+};
+
+static void blk_stat_init(struct blk_rq_stat *stat)
+{
+ stat->min = -1ULL;
+ stat->max = stat->nr_samples = stat->mean = 0;
+ stat->batch = stat->nr_batch = 0;
+}
+
static void blk_stat_flush_batch(struct blk_rq_stat *stat)
{
const s32 nr_batch = READ_ONCE(stat->nr_batch);
@@ -48,209 +61,164 @@ static void blk_stat_sum(struct blk_rq_stat *dst, struct blk_rq_stat *src)
dst->nr_samples += src->nr_samples;
}
-static void blk_mq_stat_get(struct request_queue *q, struct blk_stats *stats)
+static void __blk_stat_add(struct blk_rq_stat *stat, u64 value)
{
- struct blk_mq_hw_ctx *hctx;
- struct blk_mq_ctx *ctx;
- uint64_t latest = 0;
- int i, j, nr;
-
- blk_stat_init(&stats->read);
- blk_stat_init(&stats->write);
-
- nr = 0;
- do {
- uint64_t newest = 0;
-
- queue_for_each_hw_ctx(q, hctx, i) {
- hctx_for_each_ctx(hctx, ctx, j) {
- blk_stat_flush_batch(&ctx->stat[BLK_STAT_READ]);
- blk_stat_flush_batch(&ctx->stat[BLK_STAT_WRITE]);
-
- if (!ctx->stat[BLK_STAT_READ].nr_samples &&
- !ctx->stat[BLK_STAT_WRITE].nr_samples)
- continue;
- if (ctx->stat[BLK_STAT_READ].time > newest)
- newest = ctx->stat[BLK_STAT_READ].time;
- if (ctx->stat[BLK_STAT_WRITE].time > newest)
- newest = ctx->stat[BLK_STAT_WRITE].time;
- }
- }
+ stat->min = min(stat->min, value);
+ stat->max = max(stat->max, value);
- /*
- * No samples
- */
- if (!newest)
- break;
-
- if (newest > latest)
- latest = newest;
-
- queue_for_each_hw_ctx(q, hctx, i) {
- hctx_for_each_ctx(hctx, ctx, j) {
- if (ctx->stat[BLK_STAT_READ].time == newest) {
- blk_stat_sum(&stats->read,
- &ctx->stat[BLK_STAT_READ]);
- nr++;
- }
- if (ctx->stat[BLK_STAT_WRITE].time == newest) {
- blk_stat_sum(&stats->write,
- &ctx->stat[BLK_STAT_WRITE]);
- nr++;
- }
- }
- }
- /*
- * If we race on finding an entry, just loop back again.
- * Should be very rare.
- */
- } while (!nr);
+ if (stat->batch + value < stat->batch ||
+ stat->nr_batch + 1 == BLK_RQ_STAT_BATCH)
+ blk_stat_flush_batch(stat);
- stats->read.time = stats->write.time = latest;
+ stat->batch += value;
+ stat->nr_batch++;
}
-void blk_queue_stat_get(struct request_queue *q, struct blk_stats *stats)
+static struct blk_rq_stat *blk_stat_ddir(struct blk_stats __percpu *stats,
+ int ddir)
{
- if (q->mq_ops)
- blk_mq_stat_get(q, stats);
- else {
- blk_stat_flush_batch(&q->rq_stats[BLK_STAT_READ]);
- blk_stat_flush_batch(&q->rq_stats[BLK_STAT_WRITE]);
- memcpy(&stats->read, &q->rq_stats[BLK_STAT_READ],
- sizeof(struct blk_rq_stat));
- memcpy(&stats->write, &q->rq_stats[BLK_STAT_WRITE],
- sizeof(struct blk_rq_stat));
- }
+ if (ddir == READ)
+ return &this_cpu_ptr(stats)->read;
+ else
+ return &this_cpu_ptr(stats)->write;
}
-void blk_hctx_stat_get(struct blk_mq_hw_ctx *hctx, struct blk_stats *stats)
+void blk_stat_add(struct request *rq)
{
- struct blk_mq_ctx *ctx;
- unsigned int i, nr;
-
- nr = 0;
- do {
- uint64_t newest = 0;
-
- hctx_for_each_ctx(hctx, ctx, i) {
- blk_stat_flush_batch(&ctx->stat[BLK_STAT_READ]);
- blk_stat_flush_batch(&ctx->stat[BLK_STAT_WRITE]);
+ struct request_queue *q = rq->q;
+ struct blk_stat_callback *cb;
+ struct blk_rq_stat *stat;
+ int ddir = rq_data_dir(rq);
+ s64 now, value;
- if (!ctx->stat[BLK_STAT_READ].nr_samples &&
- !ctx->stat[BLK_STAT_WRITE].nr_samples)
- continue;
+ now = __blk_stat_time(ktime_to_ns(ktime_get()));
+ if (now < blk_stat_time(&rq->issue_stat))
+ return;
- if (ctx->stat[BLK_STAT_READ].time > newest)
- newest = ctx->stat[BLK_STAT_READ].time;
- if (ctx->stat[BLK_STAT_WRITE].time > newest)
- newest = ctx->stat[BLK_STAT_WRITE].time;
- }
+ value = now - blk_stat_time(&rq->issue_stat);
- if (!newest)
- break;
-
- hctx_for_each_ctx(hctx, ctx, i) {
- if (ctx->stat[BLK_STAT_READ].time == newest) {
- blk_stat_sum(&stats->read,
- &ctx->stat[BLK_STAT_READ]);
- nr++;
- }
- if (ctx->stat[BLK_STAT_WRITE].time == newest) {
- blk_stat_sum(&stats->write,
- &ctx->stat[BLK_STAT_WRITE]);
- nr++;
- }
+ rcu_read_lock();
+ list_for_each_entry_rcu(cb, &q->stats->callbacks, list) {
+ if (timer_pending(&cb->timer)) {
+ stat = blk_stat_ddir(cb->stats, ddir);
+ __blk_stat_add(stat, value);
}
- /*
- * If we race on finding an entry, just loop back again.
- * Should be very rare, as the window is only updated
- * occasionally
- */
- } while (!nr);
+ }
+ rcu_read_unlock();
}
-static void __blk_stat_init(struct blk_rq_stat *stat, s64 time_now)
+static void blk_stat_timer_fn(unsigned long data)
{
- stat->min = -1ULL;
- stat->max = stat->nr_samples = stat->mean = 0;
- stat->batch = stat->nr_batch = 0;
- stat->time = time_now & BLK_STAT_NSEC_MASK;
-}
+ struct blk_stat_callback *cb = (void *)data;
+ struct blk_stats stats;
+ int i;
-void blk_stat_init(struct blk_rq_stat *stat)
-{
- __blk_stat_init(stat, ktime_to_ns(ktime_get()));
-}
+ blk_stat_init(&stats.read);
+ blk_stat_init(&stats.write);
-static bool __blk_stat_is_current(struct blk_rq_stat *stat, s64 now)
-{
- return (now & BLK_STAT_NSEC_MASK) == (stat->time & BLK_STAT_NSEC_MASK);
+ for_each_online_cpu(i) {
+ struct blk_stats *cpu_stats;
+
+ cpu_stats = per_cpu_ptr(cb->stats, i);
+ blk_stat_sum(&stats.read, &cpu_stats->read);
+ blk_stat_sum(&stats.write, &cpu_stats->write);
+ blk_stat_init(&cpu_stats->read);
+ blk_stat_init(&cpu_stats->write);
+ }
+
+ cb->fn(cb, &stats);
}
-bool blk_stat_is_current(struct blk_rq_stat *stat)
+struct blk_stat_callback *
+blk_stat_alloc_callback(void (*fn)(struct blk_stat_callback *, struct blk_stats *),
+ void *data)
{
- return __blk_stat_is_current(stat, ktime_to_ns(ktime_get()));
+ struct blk_stat_callback *cb;
+
+ cb = kmalloc(sizeof(*cb), GFP_KERNEL);
+ if (!cb)
+ return NULL;
+
+ cb->stats = alloc_percpu(struct blk_stats);
+ if (!cb->stats) {
+ kfree(cb);
+ return NULL;
+ }
+
+ cb->fn = fn;
+ cb->data = data;
+ setup_timer(&cb->timer, blk_stat_timer_fn, (unsigned long)cb);
+
+ return cb;
}
+EXPORT_SYMBOL_GPL(blk_stat_alloc_callback);
-void blk_stat_add(struct blk_rq_stat *stat, struct request *rq)
+void blk_stat_add_callback(struct request_queue *q,
+ struct blk_stat_callback *cb)
{
- s64 now, value;
+ int i;
- now = __blk_stat_time(ktime_to_ns(ktime_get()));
- if (now < blk_stat_time(&rq->issue_stat))
- return;
+ for_each_possible_cpu(i) {
+ struct blk_stats *stats = per_cpu_ptr(cb->stats, i);
- if (!__blk_stat_is_current(stat, now))
- __blk_stat_init(stat, now);
+ blk_stat_init(&stats->read);
+ blk_stat_init(&stats->write);
+ }
- value = now - blk_stat_time(&rq->issue_stat);
- if (value > stat->max)
- stat->max = value;
- if (value < stat->min)
- stat->min = value;
+ spin_lock(&q->stats->lock);
+ list_add_tail_rcu(&cb->list, &q->stats->callbacks);
+ set_bit(QUEUE_FLAG_STATS, &q->queue_flags);
+ spin_unlock(&q->stats->lock);
+}
+EXPORT_SYMBOL_GPL(blk_stat_add_callback);
- if (stat->batch + value < stat->batch ||
- stat->nr_batch + 1 == BLK_RQ_STAT_BATCH)
- blk_stat_flush_batch(stat);
+void blk_stat_remove_callback(struct request_queue *q,
+ struct blk_stat_callback *cb)
+{
+ spin_lock(&q->stats->lock);
+ list_del_rcu(&cb->list);
+ if (list_empty(&q->stats->callbacks))
+ clear_bit(QUEUE_FLAG_STATS, &q->queue_flags);
+ spin_unlock(&q->stats->lock);
- stat->batch += value;
- stat->nr_batch++;
+ del_timer_sync(&cb->timer);
}
+EXPORT_SYMBOL_GPL(blk_stat_remove_callback);
-void blk_stat_clear(struct request_queue *q)
+static void blk_stat_free_callback_rcu(struct rcu_head *head)
{
- if (q->mq_ops) {
- struct blk_mq_hw_ctx *hctx;
- struct blk_mq_ctx *ctx;
- int i, j;
-
- queue_for_each_hw_ctx(q, hctx, i) {
- hctx_for_each_ctx(hctx, ctx, j) {
- blk_stat_init(&ctx->stat[BLK_STAT_READ]);
- blk_stat_init(&ctx->stat[BLK_STAT_WRITE]);
- }
- }
- } else {
- blk_stat_init(&q->rq_stats[BLK_STAT_READ]);
- blk_stat_init(&q->rq_stats[BLK_STAT_WRITE]);
- }
+ struct blk_stat_callback *cb;
+
+ cb = container_of(head, struct blk_stat_callback, rcu);
+ free_percpu(cb->stats);
}
-void blk_stat_set_issue_time(struct blk_issue_stat *stat)
+void blk_stat_free_callback(struct blk_stat_callback *cb)
{
- stat->time = (stat->time & BLK_STAT_MASK) |
- (ktime_to_ns(ktime_get()) & BLK_STAT_TIME_MASK);
+ call_rcu(&cb->rcu, blk_stat_free_callback_rcu);
}
+EXPORT_SYMBOL_GPL(blk_stat_free_callback);
-/*
- * Enable stat tracking, return whether it was enabled
- */
-bool blk_stat_enable(struct request_queue *q)
+struct blk_queue_stats *blk_alloc_queue_stats(void)
{
- if (!test_bit(QUEUE_FLAG_STATS, &q->queue_flags)) {
- set_bit(QUEUE_FLAG_STATS, &q->queue_flags);
- return false;
- }
+ struct blk_queue_stats *stats;
+
+ stats = kmalloc(sizeof(*stats), GFP_KERNEL);
+ if (!stats)
+ return NULL;
+
+ INIT_LIST_HEAD(&stats->callbacks);
+ spin_lock_init(&stats->lock);
+
+ return stats;
+}
+
+void blk_free_queue_stats(struct blk_queue_stats *stats)
+{
+ if (!stats)
+ return;
+
+ WARN_ON(!list_empty(&stats->callbacks));
- return true;
+ kfree(stats);
}
diff --git a/block/blk-stat.h b/block/blk-stat.h
index a24439aab710..fd6ba9183b17 100644
--- a/block/blk-stat.h
+++ b/block/blk-stat.h
@@ -1,11 +1,11 @@
#ifndef BLK_STAT_H
#define BLK_STAT_H
-/*
- * ~0.13s window as a power-of-2 (2^27 nsecs)
- */
-#define BLK_STAT_NSEC 134217728ULL
-#define BLK_STAT_NSEC_MASK ~(BLK_STAT_NSEC - 1)
+#include <linux/kernel.h>
+#include <linux/blkdev.h>
+#include <linux/ktime.h>
+#include <linux/rcupdate.h>
+#include <linux/timer.h>
/*
* Upper 3 bits can be used elsewhere
@@ -15,19 +15,27 @@
#define BLK_STAT_TIME_MASK ((1ULL << BLK_STAT_SHIFT) - 1)
#define BLK_STAT_MASK ~BLK_STAT_TIME_MASK
-enum {
- BLK_STAT_READ = 0,
- BLK_STAT_WRITE,
+#define BLK_RQ_STAT_BATCH 64
+
+struct blk_stat_callback {
+ struct list_head list;
+ struct timer_list timer;
+ struct blk_stats __percpu *stats;
+ void (*fn)(struct blk_stat_callback *, struct blk_stats *);
+ void *data;
+ struct rcu_head rcu;
};
-void blk_stat_add(struct blk_rq_stat *, struct request *);
-void blk_hctx_stat_get(struct blk_mq_hw_ctx *, struct blk_stats *);
-void blk_queue_stat_get(struct request_queue *, struct blk_stats *);
-void blk_stat_clear(struct request_queue *);
-void blk_stat_init(struct blk_rq_stat *);
-bool blk_stat_is_current(struct blk_rq_stat *);
-void blk_stat_set_issue_time(struct blk_issue_stat *);
-bool blk_stat_enable(struct request_queue *);
+struct blk_queue_stats *blk_alloc_queue_stats(void);
+void blk_free_queue_stats(struct blk_queue_stats *);
+
+void blk_stat_add(struct request *);
+
+static inline void blk_stat_set_issue_time(struct blk_issue_stat *stat)
+{
+ stat->time = ((stat->time & BLK_STAT_MASK) |
+ (ktime_to_ns(ktime_get()) & BLK_STAT_TIME_MASK));
+}
static inline u64 __blk_stat_time(u64 time)
{
@@ -39,4 +47,62 @@ static inline u64 blk_stat_time(struct blk_issue_stat *stat)
return __blk_stat_time(stat->time);
}
+/**
+ * blk_stat_alloc_callback() - Allocate a block statistics callback.
+ * @fn: Callback function.
+ * @data: Value for the @data field of the &struct blk_stat_callback.
+ *
+ * Return: &struct blk_stat_callback on success or NULL on ENOMEM.
+ */
+struct blk_stat_callback *
+blk_stat_alloc_callback(void (*fn)(struct blk_stat_callback *, struct blk_stats *),
+ void *data);
+
+/**
+ * blk_stat_add_callback() - Add a block statistics callback to be run on a
+ * request queue.
+ * @q: The request queue.
+ * @cb: The callback.
+ *
+ * Note that a single &struct blk_stat_callback can only be added to a single
+ * &struct request_queue.
+ */
+void blk_stat_add_callback(struct request_queue *q,
+ struct blk_stat_callback *cb);
+
+/**
+ * blk_stat_remove_callback() - Remove a block statistics callback from a
+ * request queue.
+ * @q: The request queue.
+ * @cb: The callback.
+ *
+ * When this returns, the callback is not running on any CPUs and will not be
+ * called again unless readded.
+ */
+void blk_stat_remove_callback(struct request_queue *q,
+ struct blk_stat_callback *cb);
+
+/**
+ * blk_stat_free_callback() - Free a block statistics callback.
+ * @cb: The callback.
+ *
+ * @cb may be NULL, in which case this does nothing. If it is not NULL, @cb must
+ * not be associated with a request queue. I.e., if it was previously added with
+ * blk_stat_add_callback(), it must also have been removed since then with
+ * blk_stat_remove_callback().
+ */
+void blk_stat_free_callback(struct blk_stat_callback *cb);
+
+/**
+ * blk_stat_arm_callback() - Set a block statistics callback to fire at a given
+ * time.
+ * @cb: The callback.
+ * @expires: The time (in jiffies) at which to run.
+ */
+static inline void blk_stat_arm_callback(struct blk_stat_callback *cb,
+ unsigned long expires)
+{
+ mod_timer(&cb->timer, expires);
+}
+
#endif
diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 6e1cf622af21..fa831cb2fc30 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -785,6 +785,9 @@ static void blk_release_queue(struct kobject *kobj)
container_of(kobj, struct request_queue, kobj);
wbt_exit(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);
bdi_put(q->backing_dev_info);
blkcg_exit_queue(q);
@@ -793,6 +796,8 @@ static void blk_release_queue(struct kobject *kobj)
elevator_exit(q->elevator);
}
+ blk_free_queue_stats(q->stats);
+
blk_exit_rl(&q->root_rl);
if (q->queue_tags)
diff --git a/block/blk-wbt.c b/block/blk-wbt.c
index 5e81068f7c52..695bb92be82b 100644
--- a/block/blk-wbt.c
+++ b/block/blk-wbt.c
@@ -277,7 +277,7 @@ enum {
LAT_EXCEEDED,
};
-static int __latency_exceeded(struct rq_wb *rwb, struct blk_stats *stats)
+static int latency_exceeded(struct rq_wb *rwb, struct blk_stats *stats)
{
struct backing_dev_info *bdi = rwb->queue->backing_dev_info;
u64 thislat;
@@ -308,8 +308,8 @@ static int __latency_exceeded(struct rq_wb *rwb, struct blk_stats *stats)
* waited or still has writes in flights, consider us doing
* just writes as well.
*/
- if ((stats->write.nr_samples && blk_stat_is_current(&stats->read)) ||
- wb_recent_wait(rwb) || wbt_inflight(rwb))
+ if (stats->write.nr_samples || wb_recent_wait(rwb) ||
+ wbt_inflight(rwb))
return LAT_UNKNOWN_WRITES;
return LAT_UNKNOWN;
}
@@ -329,14 +329,6 @@ static int __latency_exceeded(struct rq_wb *rwb, struct blk_stats *stats)
return LAT_OK;
}
-static int latency_exceeded(struct rq_wb *rwb)
-{
- struct blk_stats stats;
-
- blk_queue_stat_get(rwb->queue, &stats);
- return __latency_exceeded(rwb, &stats);
-}
-
static void rwb_trace_step(struct rq_wb *rwb, const char *msg)
{
struct backing_dev_info *bdi = rwb->queue->backing_dev_info;
@@ -355,7 +347,6 @@ static void scale_up(struct rq_wb *rwb)
rwb->scale_step--;
rwb->unknown_cnt = 0;
- blk_stat_clear(rwb->queue);
rwb->scaled_max = calc_wb_limits(rwb);
@@ -385,7 +376,6 @@ static void scale_down(struct rq_wb *rwb, bool hard_throttle)
rwb->scaled_max = false;
rwb->unknown_cnt = 0;
- blk_stat_clear(rwb->queue);
calc_wb_limits(rwb);
rwb_trace_step(rwb, "step down");
}
@@ -412,16 +402,16 @@ static void rwb_arm_timer(struct rq_wb *rwb)
}
expires = jiffies + nsecs_to_jiffies(rwb->cur_win_nsec);
- mod_timer(&rwb->window_timer, expires);
+ blk_stat_arm_callback(rwb->cb, expires);
}
-static void wb_timer_fn(unsigned long data)
+static void wb_timer_fn(struct blk_stat_callback *cb, struct blk_stats *stats)
{
- struct rq_wb *rwb = (struct rq_wb *) data;
+ struct rq_wb *rwb = cb->data;
unsigned int inflight = wbt_inflight(rwb);
int status;
- status = latency_exceeded(rwb);
+ status = latency_exceeded(rwb, stats);
trace_wbt_timer(rwb->queue->backing_dev_info, status, rwb->scale_step,
inflight);
@@ -614,7 +604,7 @@ enum wbt_flags wbt_wait(struct rq_wb *rwb, struct bio *bio, spinlock_t *lock)
__wbt_wait(rwb, bio->bi_opf, lock);
- if (!timer_pending(&rwb->window_timer))
+ if (!timer_pending(&rwb->cb->timer))
rwb_arm_timer(rwb);
if (current_is_kswapd())
@@ -675,7 +665,7 @@ void wbt_disable_default(struct request_queue *q)
struct rq_wb *rwb = q->rq_wb;
if (rwb && rwb->enable_state == WBT_STATE_ON_DEFAULT) {
- del_timer_sync(&rwb->window_timer);
+ blk_stat_remove_callback(q, rwb->cb);
rwb->win_nsec = rwb->min_lat_nsec = 0;
wbt_update_limits(rwb);
}
@@ -699,24 +689,23 @@ int wbt_init(struct request_queue *q)
struct rq_wb *rwb;
int i;
- /*
- * For now, we depend on the stats window being larger than
- * our monitoring window. Ensure that this isn't inadvertently
- * violated.
- */
- BUILD_BUG_ON(RWB_WINDOW_NSEC > BLK_STAT_NSEC);
BUILD_BUG_ON(WBT_NR_BITS > BLK_STAT_RES_BITS);
rwb = kzalloc(sizeof(*rwb), GFP_KERNEL);
if (!rwb)
return -ENOMEM;
+ rwb->cb = blk_stat_alloc_callback(wb_timer_fn, rwb);
+ if (!rwb->cb) {
+ kfree(rwb);
+ return -ENOMEM;
+ }
+
for (i = 0; i < WBT_NUM_RWQ; i++) {
atomic_set(&rwb->rq_wait[i].inflight, 0);
init_waitqueue_head(&rwb->rq_wait[i].wait);
}
- setup_timer(&rwb->window_timer, wb_timer_fn, (unsigned long) rwb);
rwb->wc = 1;
rwb->queue_depth = RWB_DEF_DEPTH;
rwb->last_comp = rwb->last_issue = jiffies;
@@ -726,10 +715,10 @@ int wbt_init(struct request_queue *q)
wbt_update_limits(rwb);
/*
- * Assign rwb, and turn on stats tracking for this queue
+ * Assign rwb and add the stats callback.
*/
q->rq_wb = rwb;
- blk_stat_enable(q);
+ blk_stat_add_callback(q, rwb->cb);
rwb->min_lat_nsec = wbt_default_latency_nsec(q);
@@ -744,7 +733,8 @@ void wbt_exit(struct request_queue *q)
struct rq_wb *rwb = q->rq_wb;
if (rwb) {
- del_timer_sync(&rwb->window_timer);
+ blk_stat_remove_callback(q, rwb->cb);
+ blk_stat_free_callback(rwb->cb);
q->rq_wb = NULL;
kfree(rwb);
}
diff --git a/block/blk-wbt.h b/block/blk-wbt.h
index 65f1de519f67..591ff2f4b2ee 100644
--- a/block/blk-wbt.h
+++ b/block/blk-wbt.h
@@ -81,7 +81,7 @@ struct rq_wb {
u64 win_nsec; /* default window size */
u64 cur_win_nsec; /* current window size */
- struct timer_list window_timer;
+ struct blk_stat_callback *cb;
s64 sync_issue;
void *sync_cookie;
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 5e4c5380edf0..14789bdb3338 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -287,8 +287,6 @@ struct blk_issue_stat {
u64 time;
};
-#define BLK_RQ_STAT_BATCH 64
-
struct blk_rq_stat {
s64 mean;
u64 min;
@@ -296,7 +294,6 @@ struct blk_rq_stat {
s32 nr_samples;
s32 nr_batch;
u64 batch;
- s64 time;
};
struct blk_stats {
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 5a7da607ca04..a093449adbb8 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -40,6 +40,8 @@ struct blkcg_gq;
struct blk_flush_queue;
struct pr_ops;
struct rq_wb;
+struct blk_queue_stats;
+struct blk_stat_callback;
#define BLKDEV_MIN_RQ 4
#define BLKDEV_MAX_RQ 128 /* Default maximum */
@@ -388,6 +390,7 @@ struct request_queue {
int nr_rqs[2]; /* # allocated [a]sync rqs */
int nr_rqs_elvpriv; /* # allocated rqs w/ elvpriv */
+ struct blk_queue_stats *stats;
struct rq_wb *rq_wb;
/*
@@ -505,8 +508,6 @@ struct request_queue {
unsigned int nr_sorted;
unsigned int in_flight[2];
- struct blk_rq_stat rq_stats[2];
-
/*
* Number of active block driver functions for which blk_drain_queue()
* must wait. Must be incremented around functions that unlock the
@@ -516,6 +517,10 @@ struct request_queue {
unsigned int rq_timeout;
int poll_nsec;
+
+ struct blk_stat_callback *poll_cb;
+ struct blk_stats poll_stats;
+
struct timer_list timeout;
struct work_struct timeout_work;
struct list_head timeout_list;
@@ -611,6 +616,7 @@ struct request_queue {
#define QUEUE_FLAG_DAX 26 /* device supports DAX */
#define QUEUE_FLAG_STATS 27 /* track rq completion times */
#define QUEUE_FLAG_RESTART 28 /* queue needs restart at completion */
+#define QUEUE_FLAG_POLL_STATS 29 /* collecting stats for hybrid polling */
#define QUEUE_FLAG_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \
(1 << QUEUE_FLAG_STACKABLE) | \
--
2.12.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH 0/4] block: callback-based statistics
2017-03-14 21:03 [PATCH 0/4] block: callback-based statistics Omar Sandoval
` (3 preceding siblings ...)
2017-03-14 21:03 ` [PATCH 4/4] blk-stats: convert to callback-based statistics reporting Omar Sandoval
@ 2017-03-14 21:07 ` Omar Sandoval
4 siblings, 0 replies; 6+ messages in thread
From: Omar Sandoval @ 2017-03-14 21:07 UTC (permalink / raw)
To: linux-block; +Cc: kernel-team
On Tue, Mar 14, 2017 at 02:03:27PM -0700, Omar Sandoval wrote:
> From: Omar Sandoval <osandov@fb.com>
>
> This patchset generalizes the blk-stats infrastructure to allow users to
> register a callback to be called at a given time with the statistics of
> requests completed during that window. Writeback throttling and hybrid
> polling are converted to the new infrastructure. The ultimate goal is to
> use this infrastructure for the mq I/O scheduler.
>
> The details are in patch 4, which is the actual conversion. Patches 1-3
> are preparation cleanups.
FYI, the diff of block/blk-stat.[ch] is a pain to read. It's easier to
just apply the patches and look at the whole files.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2017-03-14 21:07 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-03-14 21:03 [PATCH 0/4] block: callback-based statistics Omar Sandoval
2017-03-14 21:03 ` [PATCH 1/4] block: tear out blk-stat debugfs/sysfs entries Omar Sandoval
2017-03-14 21:03 ` [PATCH 2/4] block: remove extra calls to wbt_exit() Omar Sandoval
2017-03-14 21:03 ` [PATCH 3/4] block: change stats from array type to struct type Omar Sandoval
2017-03-14 21:03 ` [PATCH 4/4] blk-stats: convert to callback-based statistics reporting Omar Sandoval
2017-03-14 21:07 ` [PATCH 0/4] block: callback-based statistics Omar Sandoval
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).