| Message ID | 20230209153204.901518530@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | fold per-CPU vmstats remotely | expand |
On Thu, Feb 09, 2023 at 12:02:00PM -0300, Marcelo Tosatti wrote: > +#ifdef CONFIG_HAVE_CMPXCHG_LOCAL > +/* Flush counters remotely if CPU uses cmpxchg to update its per-CPU counters */ > +static void vmstat_shepherd(struct work_struct *w) > +{ > + int cpu; > + > + cpus_read_lock(); > + for_each_online_cpu(cpu) { > + cpu_vm_stats_fold(cpu); Nitpick: IIUC this line is the only change with CONFIG_HAVE_CMPXCHG_LOCAL to replace the queuing. Would it be cleaner to move the ifdef into vmstat_shepherd, then, and keep the common logic? > + cond_resched(); > + } > + cpus_read_unlock(); > + > + schedule_delayed_work(&shepherd, > + round_jiffies_relative(sysctl_stat_interval)); > +} > +#else > static void vmstat_shepherd(struct work_struct *w) > { > int cpu; > @@ -2026,6 +2043,7 @@ static void vmstat_shepherd(struct work_ > schedule_delayed_work(&shepherd, > round_jiffies_relative(sysctl_stat_interval)); > } > +#endif > > static void __init start_shepherd_timer(void) > { > > >
On Thu, Mar 02, 2023 at 04:01:07PM -0500, Peter Xu wrote: > On Thu, Feb 09, 2023 at 12:02:00PM -0300, Marcelo Tosatti wrote: > > +#ifdef CONFIG_HAVE_CMPXCHG_LOCAL > > +/* Flush counters remotely if CPU uses cmpxchg to update its per-CPU counters */ > > +static void vmstat_shepherd(struct work_struct *w) > > +{ > > + int cpu; > > + > > + cpus_read_lock(); > > + for_each_online_cpu(cpu) { > > + cpu_vm_stats_fold(cpu); > > Nitpick: IIUC this line is the only change with CONFIG_HAVE_CMPXCHG_LOCAL > to replace the queuing. Would it be cleaner to move the ifdef into > vmstat_shepherd, then, and keep the common logic? https://lore.kernel.org/lkml/20221223144150.GA79369@lothringen/ Could have #ifdef CONFIG_HAVE_CMPXCHG_LOCAL int cpu_flush_vm_stats(int cpu) { return cpu_vm_stats_fold(cpu); } #else int cpu_flush_vm_stats(int cpu) { struct delayed_work *dw = &per_cpu(vmstat_work, cpu); if (!delayed_work_pending(dw) && need_update(cpu)) queue_delayed_work_on(cpu, mm_percpu_wq, dw, 0); } #endif static void vmstat_shepherd(struct work_struct *w) { int cpu; cpus_read_lock(); for_each_online_cpu(cpu) { cpu_flush_vm_stats(cpu); cond_resched(); } cpus_read_unlock(); schedule_delayed_work(&shepherd, round_jiffies_relative(sysctl_stat_interval)); } This looks really awkward to me. But then, we don't want schedule_delayed_work if !CONFIG_HAVE_CMPXCHG_LOCAL. The common part would be the cpus_read_lock and for_each_online_cpu loop. So it seems the current separation is quite readable (unless you have a suggestion). > > + cond_resched(); > > + } > > + cpus_read_unlock(); > > + > > + schedule_delayed_work(&shepherd, > > + round_jiffies_relative(sysctl_stat_interval)); > > +} > > +#else > > static void vmstat_shepherd(struct work_struct *w) > > { > > int cpu; > > @@ -2026,6 +2043,7 @@ static void vmstat_shepherd(struct work_ > > schedule_delayed_work(&shepherd, > > round_jiffies_relative(sysctl_stat_interval)); > > } > > +#endif > > > > static void __init start_shepherd_timer(void) > > { > > > > > > > > -- > Peter Xu > >
On Thu, Mar 02, 2023 at 06:16:42PM -0300, Marcelo Tosatti wrote: > On Thu, Mar 02, 2023 at 04:01:07PM -0500, Peter Xu wrote: > > On Thu, Feb 09, 2023 at 12:02:00PM -0300, Marcelo Tosatti wrote: > > > +#ifdef CONFIG_HAVE_CMPXCHG_LOCAL > > > +/* Flush counters remotely if CPU uses cmpxchg to update its per-CPU counters */ > > > +static void vmstat_shepherd(struct work_struct *w) > > > +{ > > > + int cpu; > > > + > > > + cpus_read_lock(); > > > + for_each_online_cpu(cpu) { > > > + cpu_vm_stats_fold(cpu); > > > > Nitpick: IIUC this line is the only change with CONFIG_HAVE_CMPXCHG_LOCAL > > to replace the queuing. Would it be cleaner to move the ifdef into > > vmstat_shepherd, then, and keep the common logic? > > https://lore.kernel.org/lkml/20221223144150.GA79369@lothringen/ :-) [...] > So it seems the current separation is quite readable > (unless you have a suggestion). No, feel free to ignore any of my nitpicks when you don't think proper. :) Keeping it is fine to me.
Index: linux-2.6/mm/vmstat.c =================================================================== --- linux-2.6.orig/mm/vmstat.c +++ linux-2.6/mm/vmstat.c @@ -2007,6 +2007,23 @@ static void vmstat_shepherd(struct work_ static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd); +#ifdef CONFIG_HAVE_CMPXCHG_LOCAL +/* Flush counters remotely if CPU uses cmpxchg to update its per-CPU counters */ +static void vmstat_shepherd(struct work_struct *w) +{ + int cpu; + + cpus_read_lock(); + for_each_online_cpu(cpu) { + cpu_vm_stats_fold(cpu); + cond_resched(); + } + cpus_read_unlock(); + + schedule_delayed_work(&shepherd, + round_jiffies_relative(sysctl_stat_interval)); +} +#else static void vmstat_shepherd(struct work_struct *w) { int cpu; @@ -2026,6 +2043,7 @@ static void vmstat_shepherd(struct work_ schedule_delayed_work(&shepherd, round_jiffies_relative(sysctl_stat_interval)); } +#endif static void __init start_shepherd_timer(void) {
Now that the counters are modified via cmpxchg both CPU locally (via the account functions), and remotely (via cpu_vm_stats_fold), its possible to switch vmstat_shepherd to perform the per-CPU vmstats folding remotely. This fixes the following two problems: 1. A customer provided some evidence which indicates that the idle tick was stopped; albeit, CPU-specific vmstat counters still remained populated. Thus one can only assume quiet_vmstat() was not invoked on return to the idle loop. If I understand correctly, I suspect this divergence might erroneously prevent a reclaim attempt by kswapd. If the number of zone specific free pages are below their per-cpu drift value then zone_page_state_snapshot() is used to compute a more accurate view of the aforementioned statistic. Thus any task blocked on the NUMA node specific pfmemalloc_wait queue will be unable to make significant progress via direct reclaim unless it is killed after being woken up by kswapd (see throttle_direct_reclaim()) 2. With a SCHED_FIFO task that busy loops on a given CPU, and kworker for that CPU at SCHED_OTHER priority, queuing work to sync per-vmstats will either cause that work to never execute, or stalld (i.e. stall daemon) boosts kworker priority which causes a latency violation Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>