Messages in this thread | | | From | KOSAKI Motohiro <> | Subject | Re: [PATCH 1/3] Reintroduce zone_reclaim_interval for when zone_reclaim() scans and fails to avoid CPU spinning at 100% on NUMA | Date | Tue, 9 Jun 2009 17:45:02 +0900 (JST) |
| |
Hi
> > > @@ -1192,6 +1192,15 @@ static struct ctl_table vm_table[] = { > > > .extra1 = &zero, > > > }, > > > { > > > + .ctl_name = CTL_UNNUMBERED, > > > + .procname = "zone_reclaim_interval", > > > + .data = &zone_reclaim_interval, > > > + .maxlen = sizeof(zone_reclaim_interval), > > > + .mode = 0644, > > > + .proc_handler = &proc_dointvec_jiffies, > > > + .strategy = &sysctl_jiffies, > > > + }, > > > > hmmm, I think nobody can know proper interval settings on his own systems. > > I agree with Wu. It can be hidden. > > > > For the few users that case, I expect the majority of those will choose > either 0 or the default value of 30. They might want to alter this while > setting zone_reclaim_mode if they don't understand the different values > it can have for example. > > My preference would be that this not exist at all but the > scan-avoidance-heuristic has to be perfect to allow that.
Ah, I didn't concern interval==0. thanks. I can ack this now, but please add documentation about interval==0 meaning?
> > > @@ -2414,6 +2426,16 @@ int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) > > > ret = __zone_reclaim(zone, gfp_mask, order); > > > zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); > > > > > > + if (!ret) { > > > + /* > > > + * We were unable to reclaim enough pages to stay on node and > > > + * unable to detect in advance that the scan would fail. Allow > > > + * off node accesses for zone_reclaim_inteval jiffies before > > > + * trying zone_reclaim() again > > > + */ > > > + zone->zone_reclaim_failure = jiffies; > > > > Oops, this simple assignment don't care jiffies round-trip. > > > > Here it is just recording the jiffies value. The real smarts with the counter > use time_before() which I assumed could handle jiffie wrap-arounds. Even > if it doesn't, the consequence is that one scan will occur that could have > been avoided around the time of the jiffie wraparound. The value will then > be reset and it will be fine.
time_before() assume two argument are enough nearly time. if we use 32bit cpu and HZ=1000, about jiffies wraparound about one month.
Then,
1. zone reclaim failure occur 2. system works fine for one month 3. jiffies wrap and time_before() makes mis-calculation.
I think.
| |