Messages in this thread Patch in this message | | | From | Chengming Zhou <> | Subject | [PATCH] blk-iocost: fix false positive lagging | Date | Thu, 26 May 2022 21:35:54 +0800 |
| |
I found many false positive lagging during iocost test.
Since iocg->vtime will be advanced to (vnow - margins.target) in hweight_after_donation(), which called throw away excess, the iocg->done_vtime will also be advanced that much.
period_at_vtime <--period_vtime--> vnow | | ---------------------------------------------------> |<--->| margins.target |-> vtime, done_vtime
If that iocg has some inflight io when vnow, but its done_vtime is before period_at_vtime, ioc_timer_fn() will think it has lagging io, even these io maybe issued just before now.
This patch change the condition to check if vdone is before (period_at_vtime - margins.target) instead of period_at_vtime.
But there is another problem that this patch doesn't fix. Since vtime will be advanced, we can't check if vtime is after (vnow - MAX_LAGGING_PERIODS * period_vtime) to tell whether this iocg pin lagging for too long.
Maybe we can add lagging_periods in iocg to record how many periods this iocg pin lagging, but I don't know when to clean it.
Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com> --- block/blk-iocost.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/block/blk-iocost.c b/block/blk-iocost.c index 33a11ba971ea..42e301b7527b 100644 --- a/block/blk-iocost.c +++ b/block/blk-iocost.c @@ -2259,7 +2259,7 @@ static void ioc_timer_fn(struct timer_list *timer) time_after64(vtime, vdone) && time_after64(vtime, now.vnow - MAX_LAGGING_PERIODS * period_vtime) && - time_before64(vdone, now.vnow - period_vtime)) + time_before64(vdone, ioc->period_at_vtime - ioc->margins.target)) nr_lagging++; /* -- 2.36.1
| |