| Message ID | 1490494692-2416-3-git-send-email-sbates@raithlin.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 26/03/17 05:18, sbates@raithlin.com wrote: > From: Stephen Bates <sbates@raithlin.com> > > In order to bucket IO for the polling algorithm we use a sysfs entry > to set the filter value. It is signed and we will use that as follows: > > 0 : No filtering. All IO are considered in stat generation > > 0 : Filtering based on IO of exactly this size only. > < 0 : Filtering based on IO less than or equal to -1 time this value. I'd say that this is a fairly non-trivial semantic meanning to this... Is there any use for the size exact match filter? If not then I suggest we always make it (<=) in its semantics... > Use this member to implement a new bucket function for the io polling > stats. > > Signed-off-by: Stephen Bates <sbates@raithlin.com> > --- > block/blk-mq.c | 17 +++++++++++++++-- > block/blk-sysfs.c | 30 ++++++++++++++++++++++++++++++ > include/linux/blkdev.h | 1 + > 3 files changed, 46 insertions(+), 2 deletions(-) > > diff --git a/block/blk-mq.c b/block/blk-mq.c > index 5ff66f2..5ea9481 100644 > --- a/block/blk-mq.c > +++ b/block/blk-mq.c > @@ -2327,6 +2327,18 @@ static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set, > blk_mq_sysfs_register(q); > } > > +int blk_mq_poll_stats_bucket(const struct request *rq) > +{ > + int val = rq->q->poll_size; > + > + if ((val == 0) || > + (val > 0 && blk_rq_bytes(rq) == val) || > + (val < 0 && blk_rq_bytes(rq) <= -val)) > + return (int)rq_data_dir(rq); > + > + return -1; > +} > + > struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, > struct request_queue *q) > { > @@ -2338,7 +2350,7 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, > goto err_exit; > > q->poll_cb = blk_stat_alloc_callback(blk_mq_poll_stats_fn, > - blk_stat_rq_ddir, 2, q); > + blk_mq_poll_stats_bucket, 2, q); > if (!q->poll_cb) > goto err_exit; > > @@ -2387,9 +2399,10 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, > q->nr_requests = set->queue_depth; > > /* > - * Default to classic polling > + * Default to classic polling. Default to considering all IO. > */ > q->poll_nsec = -1; > + q->poll_size = 0; > > if (set->ops->complete) > blk_queue_softirq_done(q, set->ops->complete); > diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c > index fa831cb..000d5db 100644 > --- a/block/blk-sysfs.c > +++ b/block/blk-sysfs.c > @@ -394,6 +394,29 @@ static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page, > return count; > } > > +static ssize_t queue_poll_size_show(struct request_queue *q, char *page) > +{ > + return sprintf(page, "%d\n", q->poll_size); > +} > + > +static ssize_t queue_poll_size_store(struct request_queue *q, const char *page, > + size_t count) > +{ > + int err, val; > + > + if (!q->mq_ops || !q->mq_ops->poll) > + return -EINVAL; > + > + err = kstrtoint(page, 10, &val); > + if (err < 0) > + return err; > + > + q->poll_size = val; > + > + return count; > +} > + > + > static ssize_t queue_poll_show(struct request_queue *q, char *page) > { > return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page); > @@ -654,6 +677,12 @@ static struct queue_sysfs_entry queue_poll_entry = { > .store = queue_poll_store, > }; > > +static struct queue_sysfs_entry queue_poll_size_entry = { > + .attr = {.name = "io_poll_size", .mode = S_IRUGO | S_IWUSR }, > + .show = queue_poll_size_show, > + .store = queue_poll_size_store, > +}; > + > static struct queue_sysfs_entry queue_poll_delay_entry = { > .attr = {.name = "io_poll_delay", .mode = S_IRUGO | S_IWUSR }, > .show = queue_poll_delay_show, > @@ -710,6 +739,7 @@ static struct attribute *default_attrs[] = { > &queue_dax_entry.attr, > &queue_wb_lat_entry.attr, > &queue_poll_delay_entry.attr, > + &queue_poll_size_entry.attr, > NULL, > }; > > diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h > index 1a7dc42..7ff5388 100644 > --- a/include/linux/blkdev.h > +++ b/include/linux/blkdev.h > @@ -517,6 +517,7 @@ struct request_queue { > > unsigned int rq_timeout; > int poll_nsec; > + int poll_size; > > struct blk_stat_callback *poll_cb; > struct blk_rq_stat poll_stat[2]; >
>> >> In order to bucket IO for the polling algorithm we use a sysfs entry >> to set the filter value. It is signed and we will use that as follows: >> >> 0 : No filtering. All IO are considered in stat generation >> > 0 : Filtering based on IO of exactly this size only. >> < 0 : Filtering based on IO less than or equal to -1 time this value.> > > I'd say that this is a fairly non-trivial semantic meanning to this... > > Is there any use for the size exact match filter? If not then > I suggest we always make it (<=) in its semantics... Thanks for the review Sagi. I’d be OK going with <=0 as the exact match would normally be for minimal IO sizes (where <= and = are the same thing). I will see what other feedback I get and aim to do a respin soon… Stephen
On 03/28/2017 01:38 PM, Stephen Bates wrote: >>> >>> In order to bucket IO for the polling algorithm we use a sysfs entry >>> to set the filter value. It is signed and we will use that as follows: >>> >>> 0 : No filtering. All IO are considered in stat generation >>> > 0 : Filtering based on IO of exactly this size only. >>> < 0 : Filtering based on IO less than or equal to -1 time this value.> >> >> I'd say that this is a fairly non-trivial semantic meanning to this... >> >> Is there any use for the size exact match filter? If not then >> I suggest we always make it (<=) in its semantics... > > Thanks for the review Sagi. I’d be OK going with <=0 as the exact > match would normally be for minimal IO sizes (where <= and = are the > same thing). I will see what other feedback I get and aim to do a > respin soon… No tunables for this, please. There's absolutely no reason why we should need it.
>> >> Thanks for the review Sagi. I’d be OK going with <=0 as the exact >> match would normally be for minimal IO sizes (where <= and = are the >> same thing). I will see what other feedback I get and aim to do a >> respin soon… > > No tunables for this, please. There's absolutely no reason why we should > need it. Jens – by this you mean you want to only bucket IO that are exactly the minimum block size supported by the underlying block device? I was envisioning we might want to relax that in certain cases (e.g. bucket 4KB and below going to a 512B device). Stephen
On 03/28/2017 01:58 PM, Stephen Bates wrote: >>> >>> Thanks for the review Sagi. I’d be OK going with <=0 as the exact >>> match would normally be for minimal IO sizes (where <= and = are the >>> same thing). I will see what other feedback I get and aim to do a >>> respin soon… >> >> No tunables for this, please. There's absolutely no reason why we should >> need it. > > Jens – by this you mean you want to only bucket IO that are exactly > the minimum block size supported by the underlying block device? I was > envisioning we might want to relax that in certain cases (e.g. bucket > 4KB and below going to a 512B device). Sorry, the above was a bit terse. I think a much better solution would be to create a number of buckets (per data direction) and do stats on all of them. The buckets would cover a reasonable range of request sizes. Then when you poll for a given request, we can base our timed number on the data direction AND size of it. You can get pretty far with a few buckets: 512b 4k 8k 16k 32k 64k 128k and you could even have your time estimation function turn these into something sane. Or just use a composite of buckets, if you wish.
[Retrying as my new setup secretly converted to html format without telling me. Apologies for the resend.] >>> >>> Thanks for the review Sagi. I’d be OK going with <=0 as the exact >>> match would normally be for minimal IO sizes (where <= and = are the >>> same thing). I will see what other feedback I get and aim to do a >>> respin soon… >> >> No tunables for this, please. There's absolutely no reason why we should >> need it. > > Jens – by this you mean you want to only bucket IO that are exactly > the minimum block size supported by the underlying block device? I was > envisioning we might want to relax that in certain cases (e.g. bucket > 4KB and below going to a 512B device). > Sorry, the above was a bit terse. I think a much better solution would > be to create a number of buckets (per data direction) and do stats on > all of them. The buckets would cover a reasonable range of request > sizes. Then when you poll for a given request, we can base our timed > number on the data direction AND size of it. You can get pretty far with > a few buckets: > > 512b > 4k > 8k > 16k > 32k > 64k > 128k > > and you could even have your time estimation function turn these into > something sane. Or just use a composite of buckets, if you wish. I did go down this path initially but then moved away from it since we were focusing only on the smaller IO size. However I can definitely take a look at this again as I agree that it could be more useful in the long run. I would like to keep my first patch in this series alive since I do think having the option to not bucket an IO is a useful thing to have. I’ll take all the feedback to date and work on a v2. Thanks! Stephen
diff --git a/block/blk-mq.c b/block/blk-mq.c index 5ff66f2..5ea9481 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -2327,6 +2327,18 @@ static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set, blk_mq_sysfs_register(q); } +int blk_mq_poll_stats_bucket(const struct request *rq) +{ + int val = rq->q->poll_size; + + if ((val == 0) || + (val > 0 && blk_rq_bytes(rq) == val) || + (val < 0 && blk_rq_bytes(rq) <= -val)) + return (int)rq_data_dir(rq); + + return -1; +} + struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, struct request_queue *q) { @@ -2338,7 +2350,7 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, goto err_exit; q->poll_cb = blk_stat_alloc_callback(blk_mq_poll_stats_fn, - blk_stat_rq_ddir, 2, q); + blk_mq_poll_stats_bucket, 2, q); if (!q->poll_cb) goto err_exit; @@ -2387,9 +2399,10 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, q->nr_requests = set->queue_depth; /* - * Default to classic polling + * Default to classic polling. Default to considering all IO. */ q->poll_nsec = -1; + q->poll_size = 0; if (set->ops->complete) blk_queue_softirq_done(q, set->ops->complete); diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c index fa831cb..000d5db 100644 --- a/block/blk-sysfs.c +++ b/block/blk-sysfs.c @@ -394,6 +394,29 @@ static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page, return count; } +static ssize_t queue_poll_size_show(struct request_queue *q, char *page) +{ + return sprintf(page, "%d\n", q->poll_size); +} + +static ssize_t queue_poll_size_store(struct request_queue *q, const char *page, + size_t count) +{ + int err, val; + + if (!q->mq_ops || !q->mq_ops->poll) + return -EINVAL; + + err = kstrtoint(page, 10, &val); + if (err < 0) + return err; + + q->poll_size = val; + + return count; +} + + static ssize_t queue_poll_show(struct request_queue *q, char *page) { return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page); @@ -654,6 +677,12 @@ static struct queue_sysfs_entry queue_poll_entry = { .store = queue_poll_store, }; +static struct queue_sysfs_entry queue_poll_size_entry = { + .attr = {.name = "io_poll_size", .mode = S_IRUGO | S_IWUSR }, + .show = queue_poll_size_show, + .store = queue_poll_size_store, +}; + static struct queue_sysfs_entry queue_poll_delay_entry = { .attr = {.name = "io_poll_delay", .mode = S_IRUGO | S_IWUSR }, .show = queue_poll_delay_show, @@ -710,6 +739,7 @@ static struct attribute *default_attrs[] = { &queue_dax_entry.attr, &queue_wb_lat_entry.attr, &queue_poll_delay_entry.attr, + &queue_poll_size_entry.attr, NULL, }; diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h index 1a7dc42..7ff5388 100644 --- a/include/linux/blkdev.h +++ b/include/linux/blkdev.h @@ -517,6 +517,7 @@ struct request_queue { unsigned int rq_timeout; int poll_nsec; + int poll_size; struct blk_stat_callback *poll_cb; struct blk_rq_stat poll_stat[2];