summaryrefslogtreecommitdiff
path: root/block/blk-cgroup.c
diff options
context:
space:
mode:
authorTejun Heo <tj@kernel.org>2020-09-01 21:52:52 +0300
committerJens Axboe <axboe@kernel.dk>2020-09-02 04:38:32 +0300
commit5160a5a53c0c4ae3708959d9465ea43ad5d90542 (patch)
tree15bbeba67e50df5dbbb8edab6de77a5c6677f99c /block/blk-cgroup.c
parentc421a3eb2e27402c14131a14390551ae6cdb15fa (diff)
downloadlinux-5160a5a53c0c4ae3708959d9465ea43ad5d90542.tar.xz
blk-iocost: implement delay adjustment hysteresis
Curently, iocost syncs the delay duration to the outstanding debt amount, which seemed enough to protect the system from anon memory hogs. However, that was mostly because the delay calcuation was using hweight_inuse which quickly converges towards zero under debt for delay duration calculation, often pusnishing debtors overly harshly for longer than deserved. The previous patch fixed the delay calcuation and now the protection against anonymous memory hogs isn't enough because the effect of delay is indirect and non-linear and a huge amount of future debt can accumulate abruptly while unthrottled. This patch implements delay hysteresis so that delay is decayed exponentially over time instead of getting cleared immediately as debt is paid off. While the overall behavior is similar to the blk-cgroup implementation used by blk-iolatency, a lot of the details are different and due to the empirical nature of the mechanism, it's challenging to adapt the mechanism for one controller without negatively impacting the other. As the delay is gradually decayed now, there's no point in running it from its own hrtimer. Periodic updates are now performed from ioc_timer_fn() and the dedicated hrtimer is removed. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Diffstat (limited to 'block/blk-cgroup.c')
-rw-r--r--block/blk-cgroup.c23
1 files changed, 17 insertions, 6 deletions
diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index c195365c9817..d33dd6be1d9c 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -1613,16 +1613,24 @@ static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now)
static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay)
{
unsigned long pflags;
+ bool clamp;
u64 now = ktime_to_ns(ktime_get());
u64 exp;
u64 delay_nsec = 0;
int tok;
while (blkg->parent) {
- if (atomic_read(&blkg->use_delay)) {
+ int use_delay = atomic_read(&blkg->use_delay);
+
+ if (use_delay) {
+ u64 this_delay;
+
blkcg_scale_delay(blkg, now);
- delay_nsec = max_t(u64, delay_nsec,
- atomic64_read(&blkg->delay_nsec));
+ this_delay = atomic64_read(&blkg->delay_nsec);
+ if (this_delay > delay_nsec) {
+ delay_nsec = this_delay;
+ clamp = use_delay > 0;
+ }
}
blkg = blkg->parent;
}
@@ -1634,10 +1642,13 @@ static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay)
* Let's not sleep for all eternity if we've amassed a huge delay.
* Swapping or metadata IO can accumulate 10's of seconds worth of
* delay, and we want userspace to be able to do _something_ so cap the
- * delays at 1 second. If there's 10's of seconds worth of delay then
- * the tasks will be delayed for 1 second for every syscall.
+ * delays at 0.25s. If there's 10's of seconds worth of delay then the
+ * tasks will be delayed for 0.25 second for every syscall. If
+ * blkcg_set_delay() was used as indicated by negative use_delay, the
+ * caller is responsible for regulating the range.
*/
- delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC);
+ if (clamp)
+ delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC);
if (use_memdelay)
psi_memstall_enter(&pflags);