Techwizz/kernel-fxtec-pro1x
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

blk-throttle: track read and write request individually

Browse source 
In mixed read/write workload on SSD, write latency is much lower than
read. But now we only track and record read latency and then use it as
threshold base for both read and write io latency accounting. As a
result, write io latency will always be considered as good and
bad_bio_cnt is much smaller than 20% of bio_cnt. That is to mean, the
tg to be checked will be treated as idle most of the time and still let
others dispatch more ios, even it is truly running under low limit and
wants its low limit to be guaranteed, which is not we expected in fact.
So track read and write request individually, which can bring more
precise latency control for low limit idle detection.

Signed-off-by: Joseph Qi <qijiang.qj@alibaba-inc.com>
Reviewed-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This commit is contained in:
Joseph Qi 2017-11-21 09:38:30 +08:00 committed by Jens Axboe
parent a13553c777
commit b889bf66d0
1 changed files with 78 additions and 54 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

132
block/blk-throttle.c
View file

@ -216,9 +216,9 @@ struct throtl_data
unsigned int scale;
struct latency_bucket tmp_buckets[LATENCY_BUCKET_SIZE];
struct avg_latency_bucket avg_buckets[LATENCY_BUCKET_SIZE];
struct latency_bucket __percpu *latency_buckets;
struct latency_bucket tmp_buckets[2][LATENCY_BUCKET_SIZE];
struct avg_latency_bucket avg_buckets[2][LATENCY_BUCKET_SIZE];
struct latency_bucket __percpu *latency_buckets[2];
unsigned long last_calculate_time;
unsigned long filtered_latency;
@ -2050,10 +2050,10 @@ static void blk_throtl_update_idletime(struct throtl_grp *tg)
#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
static void throtl_update_latency_buckets(struct throtl_data *td)
{
struct avg_latency_bucket avg_latency[LATENCY_BUCKET_SIZE];
int i, cpu;
unsigned long last_latency = 0;
unsigned long latency;
struct avg_latency_bucket avg_latency[2][LATENCY_BUCKET_SIZE];
int i, cpu, rw;
unsigned long last_latency[2] = { 0 };
unsigned long latency[2];
if (!blk_queue_nonrot(td->queue))
return;
@ -2062,56 +2062,67 @@ static void throtl_update_latency_buckets(struct throtl_data *td)
td->last_calculate_time = jiffies;
memset(avg_latency, 0, sizeof(avg_latency));
for (i = 0; i < LATENCY_BUCKET_SIZE; i++) {
struct latency_bucket *tmp = &td->tmp_buckets[i];
for (rw = READ; rw <= WRITE; rw++) {
for (i = 0; i < LATENCY_BUCKET_SIZE; i++) {
struct latency_bucket *tmp = &td->tmp_buckets[rw][i];
for_each_possible_cpu(cpu) {
struct latency_bucket *bucket;
for_each_possible_cpu(cpu) {
struct latency_bucket *bucket;
/* this isn't race free, but ok in practice */
bucket = per_cpu_ptr(td->latency_buckets, cpu);
tmp->total_latency += bucket[i].total_latency;
tmp->samples += bucket[i].samples;
bucket[i].total_latency = 0;
bucket[i].samples = 0;
}
/* this isn't race free, but ok in practice */
bucket = per_cpu_ptr(td->latency_buckets[rw],
cpu);
tmp->total_latency += bucket[i].total_latency;
tmp->samples += bucket[i].samples;
bucket[i].total_latency = 0;
bucket[i].samples = 0;
}
if (tmp->samples >= 32) {
int samples = tmp->samples;
if (tmp->samples >= 32) {
int samples = tmp->samples;
latency = tmp->total_latency;
latency[rw] = tmp->total_latency;
tmp->total_latency = 0;
tmp->samples = 0;
latency /= samples;
if (latency == 0)
continue;
avg_latency[i].latency = latency;
tmp->total_latency = 0;
tmp->samples = 0;
latency[rw] /= samples;
if (latency[rw] == 0)
continue;
avg_latency[rw][i].latency = latency[rw];
}
}
}
for (i = 0; i < LATENCY_BUCKET_SIZE; i++) {
if (!avg_latency[i].latency) {
if (td->avg_buckets[i].latency < last_latency)
td->avg_buckets[i].latency = last_latency;
continue;
for (rw = READ; rw <= WRITE; rw++) {
for (i = 0; i < LATENCY_BUCKET_SIZE; i++) {
if (!avg_latency[rw][i].latency) {
if (td->avg_buckets[rw][i].latency < last_latency[rw])
td->avg_buckets[rw][i].latency =
last_latency[rw];
continue;
}
if (!td->avg_buckets[rw][i].valid)
latency[rw] = avg_latency[rw][i].latency;
else
latency[rw] = (td->avg_buckets[rw][i].latency * 7 +
avg_latency[rw][i].latency) >> 3;
td->avg_buckets[rw][i].latency = max(latency[rw],
last_latency[rw]);
td->avg_buckets[rw][i].valid = true;
last_latency[rw] = td->avg_buckets[rw][i].latency;
}
if (!td->avg_buckets[i].valid)
latency = avg_latency[i].latency;
else
latency = (td->avg_buckets[i].latency * 7 +
avg_latency[i].latency) >> 3;
td->avg_buckets[i].latency = max(latency, last_latency);
td->avg_buckets[i].valid = true;
last_latency = td->avg_buckets[i].latency;
}
for (i = 0; i < LATENCY_BUCKET_SIZE; i++)
throtl_log(&td->service_queue,
"Latency bucket %d: latency=%ld, valid=%d", i,
td->avg_buckets[i].latency, td->avg_buckets[i].valid);
"Latency bucket %d: read latency=%ld, read valid=%d, "
"write latency=%ld, write valid=%d", i,
td->avg_buckets[READ][i].latency,
td->avg_buckets[READ][i].valid,
td->avg_buckets[WRITE][i].latency,
td->avg_buckets[WRITE][i].valid);
}
#else
static inline void throtl_update_latency_buckets(struct throtl_data *td)
@ -2258,16 +2269,17 @@ static void throtl_track_latency(struct throtl_data *td, sector_t size,
struct latency_bucket *latency;
int index;
if (!td || td->limit_index != LIMIT_LOW || op != REQ_OP_READ ||
if (!td || td->limit_index != LIMIT_LOW ||
!(op == REQ_OP_READ || op == REQ_OP_WRITE) ||
!blk_queue_nonrot(td->queue))
return;
index = request_bucket_index(size);
latency = get_cpu_ptr(td->latency_buckets);
latency = get_cpu_ptr(td->latency_buckets[op]);
latency[index].total_latency += time;
latency[index].samples++;
put_cpu_ptr(td->latency_buckets);
put_cpu_ptr(td->latency_buckets[op]);
}
void blk_throtl_stat_add(struct request *rq, u64 time_ns)
@ -2286,6 +2298,7 @@ void blk_throtl_bio_endio(struct bio *bio)
unsigned long finish_time;
unsigned long start_time;
unsigned long lat;
int rw = bio_data_dir(bio);
tg = bio->bi_cg_private;
if (!tg)
@ -2314,7 +2327,7 @@ void blk_throtl_bio_endio(struct bio *bio)
bucket = request_bucket_index(
blk_stat_size(&bio->bi_issue_stat));
threshold = tg->td->avg_buckets[bucket].latency +
threshold = tg->td->avg_buckets[rw][bucket].latency +
tg->latency_target;
if (lat > threshold)
tg->bad_bio_cnt++;
@ -2407,9 +2420,16 @@ int blk_throtl_init(struct request_queue *q)
td = kzalloc_node(sizeof(*td), GFP_KERNEL, q->node);
if (!td)
return -ENOMEM;
td->latency_buckets = __alloc_percpu(sizeof(struct latency_bucket) *
td->latency_buckets[READ] = __alloc_percpu(sizeof(struct latency_bucket) *
LATENCY_BUCKET_SIZE, __alignof__(u64));
if (!td->latency_buckets) {
if (!td->latency_buckets[READ]) {
kfree(td);
return -ENOMEM;
}
td->latency_buckets[WRITE] = __alloc_percpu(sizeof(struct latency_bucket) *
LATENCY_BUCKET_SIZE, __alignof__(u64));
if (!td->latency_buckets[WRITE]) {
free_percpu(td->latency_buckets[READ]);
kfree(td);
return -ENOMEM;
}
@ -2428,7 +2448,8 @@ int blk_throtl_init(struct request_queue *q)
/* activate policy */
ret = blkcg_activate_policy(q, &blkcg_policy_throtl);
if (ret) {
free_percpu(td->latency_buckets);
free_percpu(td->latency_buckets[READ]);
free_percpu(td->latency_buckets[WRITE]);
kfree(td);
}
return ret;
@ -2439,7 +2460,8 @@ void blk_throtl_exit(struct request_queue *q)
BUG_ON(!q->td);
throtl_shutdown_wq(q);
blkcg_deactivate_policy(q, &blkcg_policy_throtl);
free_percpu(q->td->latency_buckets);
free_percpu(q->td->latency_buckets[READ]);
free_percpu(q->td->latency_buckets[WRITE]);
kfree(q->td);
}
@ -2457,8 +2479,10 @@ void blk_throtl_register_queue(struct request_queue *q)
} else {
td->throtl_slice = DFL_THROTL_SLICE_HD;
td->filtered_latency = LATENCY_FILTERED_HD;
for (i = 0; i < LATENCY_BUCKET_SIZE; i++)
td->avg_buckets[i].latency = DFL_HD_BASELINE_LATENCY;
for (i = 0; i < LATENCY_BUCKET_SIZE; i++) {
td->avg_buckets[READ][i].latency = DFL_HD_BASELINE_LATENCY;
td->avg_buckets[WRITE][i].latency = DFL_HD_BASELINE_LATENCY;
}
}
#ifndef CONFIG_BLK_DEV_THROTTLING_LOW
/* if no low limit, use previous default */