Re: 回复: [PATCH v3] kvfree_rcu: Release page cache under memory pressure

[Uladzislau Rezki]

> 发件人: Uladzislau Rezki <urezki@xxxxxxxxx>
> 发送时间: 2021年2月2日 3:57
> 收件人: Zhang, Qiang
> 抄送: urezki@xxxxxxxxx; paulmck@xxxxxxxxxx; joel@xxxxxxxxxxxxxxxxx; rcu@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> 主题: Re: [PATCH v3] kvfree_rcu: Release page cache under memory pressure
> 
> [Please note: This e-mail is from an EXTERNAL e-mail address]
> 
> Hello, Zqiang.
> 
> > From: Zqiang <qiang.zhang@xxxxxxxxxxxxx>
> >
> > Add free per-cpu existing krcp's page cache operation, when
> > the system is under memory pressure.
> >
> > Signed-off-by: Zqiang <qiang.zhang@xxxxxxxxxxxxx>
> > ---
> >  kernel/rcu/tree.c | 26 ++++++++++++++++++++++++++
> >  1 file changed, 26 insertions(+)
> >
> > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > index c1ae1e52f638..644b0f3c7b9f 100644
> > --- a/kernel/rcu/tree.c
> > +++ b/kernel/rcu/tree.c
> > @@ -3571,17 +3571,41 @@ void kvfree_call_rcu(struct rcu_head *head, rcu_callback_t func)
> >  }
> >  EXPORT_SYMBOL_GPL(kvfree_call_rcu);
> >
> > +static int free_krc_page_cache(struct kfree_rcu_cpu *krcp)
> > +{
> > +     unsigned long flags;
> > +     struct llist_node *page_list, *pos, *n;
> > +     int freed = 0;
> > +
> > +     raw_spin_lock_irqsave(&krcp->lock, flags);
> > +     page_list = llist_del_all(&krcp->bkvcache);
> > +     krcp->nr_bkv_objs = 0;
> > +     raw_spin_unlock_irqrestore(&krcp->lock, flags);
> > +
> > +     llist_for_each_safe(pos, n, page_list) {
> > +             free_page((unsigned long)pos);
> > +             freed++;
> > +     }
> > +
> > +     return freed;
> > +}
> > +
> >  static unsigned long
> >  kfree_rcu_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
> >  {
> >       int cpu;
> >       unsigned long count = 0;
> > +     unsigned long flags;
> >
> >       /* Snapshot count of all CPUs */
> >       for_each_possible_cpu(cpu) {
> >               struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
> >
> >               count += READ_ONCE(krcp->count);
> > +
> > +             raw_spin_lock_irqsave(&krcp->lock, flags);
> > +             count += krcp->nr_bkv_objs;
> > +             raw_spin_unlock_irqrestore(&krcp->lock, flags);
> >       }
> >
> >       return count;
> > @@ -3598,6 +3622,8 @@ kfree_rcu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
> >               struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
> >
> >               count = krcp->count;
> > +             count += free_krc_page_cache(krcp);
> > +
> >               raw_spin_lock_irqsave(&krcp->lock, flags);
> >               if (krcp->monitor_todo)
> >                       kfree_rcu_drain_unlock(krcp, flags);
> > --
> > 2.17.1
> >>
> >Thank you for your patch!
> >
> >I spent some time to see how the patch behaves under low memory condition.
> >To simulate it, i used "rcuscale" tool with below parameters:
> >
> >../rcutorture/bin/kvm.sh --torture rcuscale --allcpus --duration 10 --kconfig >CONFIG_NR_CPUS=64 \
> >--bootargs "rcuscale.kfree_rcu_test=1 rcuscale.kfree_nthreads=16 >rcuscale.holdoff=20 rcuscale.kfree_loops=10000 \
> >torture.disable_onoff_at_boot" --trust-make
> >
> >64 CPUs + 512 MB of memory. In general, my test system was running on edge
> >hitting an out of memory sometimes, but could be considered as stable in
> >regards to a test completion and taken time, so both were pretty solid.
> >
> >You can find a comparison on a plot, that can be downloaded following
> >a link: wget >ftp://vps418301.ovh.net/incoming/release_page_cache_under_low_memory.png
> >
> >In short, i see that a patched version can lead to longer test completion,
> >whereas the default variant is stable on almost all runs. After some analysis
> >and further digging i came to conclusion that a shrinker free_krc_page_cache()
> >concurs with run_page_cache_worker(krcp) running from kvfree_rcu() context.
> >
> >i.e. During the test a page shrinker is pretty active, because of low memory
> >condition. Our callback drains it whereas kvfree_rcu() part refill it right
> >away making kind of vicious circle.
> >
> >So, a run_page_cache_worker() should be backoff for some time when a system
> >runs into a low memory condition or high pressure:
> >
> >diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> >index 7077d73fcb53..446723b9646b 100644
> >--- a/kernel/rcu/tree.c
> >+++ b/kernel/rcu/tree.c
> >@@ -3163,7 +3163,7 @@ struct kfree_rcu_cpu {
> >        bool initialized;
> >        int count;
> >
> >-       struct work_struct page_cache_work;
> >+       struct delayed_work page_cache_work;
> >        atomic_t work_in_progress;
> >        struct hrtimer hrtimer;
> >
> >@@ -3419,7 +3419,7 @@ schedule_page_work_fn(struct hrtimer *t)
> >        struct kfree_rcu_cpu *krcp =
> >                container_of(t, struct kfree_rcu_cpu, hrtimer);
> >
> >-       queue_work(system_highpri_wq, &krcp->page_cache_work);
> >+       queue_delayed_work(system_highpri_wq, &krcp->page_cache_work, 0);
> >        return HRTIMER_NORESTART;
> > }
> >
> >@@ -3428,7 +3428,7 @@ static void fill_page_cache_func(struct work_struct *work)
> >        struct kvfree_rcu_bulk_data *bnode;
> >        struct kfree_rcu_cpu *krcp =
> >                container_of(work, struct kfree_rcu_cpu,
> >-                       page_cache_work);
> >+                       page_cache_work.work);
> >        unsigned long flags;
> >        bool pushed;
> >        int i;
> >@@ -3452,15 +3452,22 @@ static void fill_page_cache_func(struct work_struct >*work)
> >        atomic_set(&krcp->work_in_progress, 0);
> > }
> >
> >+static bool backoff_page_cache_fill;
> >+
> > static void
> > run_page_cache_worker(struct kfree_rcu_cpu *krcp)
> > {
> >        if (rcu_scheduler_active == RCU_SCHEDULER_RUNNING &&
> >                        !atomic_xchg(&krcp->work_in_progress, 1)) {
> >-               hrtimer_init(&krcp->hrtimer, CLOCK_MONOTONIC,
> >-                       HRTIMER_MODE_REL);
> >-               krcp->hrtimer.function = schedule_page_work_fn;
> >-               hrtimer_start(&krcp->hrtimer, 0, HRTIMER_MODE_REL);
> >+               if (READ_ONCE(backoff_page_cache_fill)) {
> >+                       queue_delayed_work(system_highpri_wq, &krcp->page_cache_work, >HZ);
> >+                       WRITE_ONCE(backoff_page_cache_fill, false);
> >+               } else {
> >+                       hrtimer_init(&krcp->hrtimer, CLOCK_MONOTONIC,
> >+                               HRTIMER_MODE_REL);
> >+                       krcp->hrtimer.function = schedule_page_work_fn;
> >+                       hrtimer_start(&krcp->hrtimer, 0, HRTIMER_MODE_REL);
> >+               }
> >        }
> > }
> >
> >@@ -3644,6 +3651,8 @@ kfree_rcu_shrink_count(struct shrinker *shrink, struct >shrink_control *sc)
> >                raw_spin_unlock_irqrestore(&krcp->lock, flags);
> >        }
> >
> >+       // Low memory condition, limit a page cache worker activity.
> >+       WRITE_ONCE(backoff_page_cache_fill, true);
> >        return count;
> > }
> 
> >@@ -4634,7 +4643,7 @@ static void __init kfree_rcu_batch_init(void)
> >                }
> >
> >                INIT_DELAYED_WORK(&krcp->monitor_work, kfree_rcu_monitor);
> >-               INIT_WORK(&krcp->page_cache_work, fill_page_cache_func);
> >+               INIT_DELAYED_WORK(&krcp->page_cache_work, fill_page_cache_func);
> >                krcp->initialized = true;
> >        }
> >        if (register_shrinker(&kfree_rcu_shrinker))
> >
> >below patch fixes it. We should backoff the page fill worker keeping it empty
> >until the situation with memory consumption is normalized.
> 
>   Maybe can cancel the timer and cancel work in the rcu_shrink_count function, after
>   the rcu_shrink_scan function is executed and recover timer.
>   what do you think?
I think it is just easier to schedule a periodic delayed work with HZ
interval. If a system is in a tight memory situation, that periodic
work will not be harmful from CPU cycles point of view. Once it is
settled, the hrtimer takes over to speedup the process of cache filling.

>    
>    The other question is, why not call queue_work function directly to fill page cache, 
>    but through a hrtimer, what is the purpose of this design?
> 
It is done that way because the kvfree_rcu() is supposed that it could
be invoked from the scheduler part. It means that if a CPU holds rq->lock 
queuing the work or placing it into rq will lead to deadlock.

--
Vlad Rezki