linux-kernel - Re: [PATCH] rcu: shrink each possible cpu krcp

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200818232554.GA2850477@google.com>
Date:   Tue, 18 Aug 2020 19:25:54 -0400
From:   Joel Fernandes <joel@...lfernandes.org>
To:     Uladzislau Rezki <urezki@...il.com>
Cc:     "Paul E. McKenney" <paulmck@...nel.org>, qiang.zhang@...driver.com,
        Josh Triplett <josh@...htriplett.org>,
        Steven Rostedt <rostedt@...dmis.org>,
        Mathieu Desnoyers <mathieu.desnoyers@...icios.com>,
        Lai Jiangshan <jiangshanlai@...il.com>,
        rcu <rcu@...r.kernel.org>, LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] rcu: shrink each possible cpu krcp

On Tue, Aug 18, 2020 at 11:55:11PM +0200, Uladzislau Rezki wrote:
> > On Tue, Aug 18, 2020 at 03:00:35PM -0400, Joel Fernandes wrote:
> > > On Tue, Aug 18, 2020 at 1:18 PM Paul E. McKenney <paulmck@...nel.org> wrote:
> > > >
> > > > On Mon, Aug 17, 2020 at 06:03:54PM -0400, Joel Fernandes wrote:
> > > > > On Fri, Aug 14, 2020 at 2:51 PM Uladzislau Rezki <urezki@...il.com> wrote:
> > > > > >
> > > > > > > From: Zqiang <qiang.zhang@...driver.com>
> > > > > > >
> > > > > > > Due to cpu hotplug. some cpu may be offline after call "kfree_call_rcu"
> > > > > > > func, if the shrinker is triggered at this time, we should drain each
> > > > > > > possible cpu "krcp".
> > > > > > >
> > > > > > > Signed-off-by: Zqiang <qiang.zhang@...driver.com>
> > > > > > > ---
> > > > > > >  kernel/rcu/tree.c | 6 +++---
> > > > > > >  1 file changed, 3 insertions(+), 3 deletions(-)
> > > > > > >
> > > > > > > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > > > > > > index 8ce77d9ac716..619ccbb3fe4b 100644
> > > > > > > --- a/kernel/rcu/tree.c
> > > > > > > +++ b/kernel/rcu/tree.c
> > > > > > > @@ -3443,7 +3443,7 @@ kfree_rcu_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
> > > > > > >       unsigned long count = 0;
> > > > > > >
> > > > > > >       /* Snapshot count of all CPUs */
> > > > > > > -     for_each_online_cpu(cpu) {
> > > > > > > +     for_each_possible_cpu(cpu) {
> > > > > > >               struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
> > > > > > >
> > > > > > >               count += READ_ONCE(krcp->count);
> > > > > > > @@ -3458,7 +3458,7 @@ kfree_rcu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
> > > > > > >       int cpu, freed = 0;
> > > > > > >       unsigned long flags;
> > > > > > >
> > > > > > > -     for_each_online_cpu(cpu) {
> > > > > > > +     for_each_possible_cpu(cpu) {
> > > > > > >               int count;
> > > > > > >               struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
> > > > > > >
> > > > > > > @@ -3491,7 +3491,7 @@ void __init kfree_rcu_scheduler_running(void)
> > > > > > >       int cpu;
> > > > > > >       unsigned long flags;
> > > > > > >
> > > > > > > -     for_each_online_cpu(cpu) {
> > > > > > > +     for_each_possible_cpu(cpu) {
> > > > > > >               struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
> > > > > > >
> > > > > > >               raw_spin_lock_irqsave(&krcp->lock, flags);
> > > > > > >
> > > > > > I agree that it can happen.
> > > > > >
> > > > > > Joel, what is your view?
> > > > >
> > > > > Yes I also think it is possible. The patch LGTM. Another fix could be
> > > > > to drain the caches in the CPU offline path and save the memory. But
> > > > > then it will take hit during __get_free_page(). If CPU
> > > > > offlining/online is not frequent, then it will save the lost memory.
> > > > >
> > > > > I wonder how other per-cpu caches in the kernel work in such scenarios.
> > > > >
> > > > > Thoughts?
> > > >
> > > > Do I count this as an ack or a review?  If not, what precisely would
> > > > you like the submitter to do differently?
> > > 
> > > Hi Paul,
> > > The patch is correct and is definitely an improvement. I was thinking
> > > about whether we should always do what the patch is doing when
> > > offlining CPUs to save memory but now I feel that may not be that much
> > > of a win to justify more complexity.
> > > 
> > > You can take it with my ack:
> > > 
> > > Acked-by: Joel Fernandes <joel@...lfernandes.org>
> > 
> > Thank you all!  I wordsmithed a bit as shown below, so please let
> > me know if I messed anything up.
> > 
> > 							Thanx, Paul
> > 
> > ------------------------------------------------------------------------
> > 
> > commit fe5d89cc025b3efe682cac122bc4d39f4722821e
> > Author: Zqiang <qiang.zhang@...driver.com>
> > Date:   Fri Aug 14 14:45:57 2020 +0800
> > 
> >     rcu: Shrink each possible cpu krcp
> >     
> >     CPUs can go offline shortly after kfree_call_rcu() has been invoked,
> >     which can leave memory stranded until those CPUs come back online.
> >     This commit therefore drains the kcrp of each CPU, not just the
> >     ones that happen to be online.
> >     
> >     Acked-by: Joel Fernandes <joel@...lfernandes.org>
> >     Signed-off-by: Zqiang <qiang.zhang@...driver.com>
> >     Signed-off-by: Paul E. McKenney <paulmck@...nel.org>
> > 
> > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > index 02ca8e5..d9f90f6 100644
> > --- a/kernel/rcu/tree.c
> > +++ b/kernel/rcu/tree.c
> > @@ -3500,7 +3500,7 @@ kfree_rcu_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
> >  	unsigned long count = 0;
> >  
> >  	/* Snapshot count of all CPUs */
> > -	for_each_online_cpu(cpu) {
> > +	for_each_possible_cpu(cpu) {
> >  		struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
> >  
> >  		count += READ_ONCE(krcp->count);
> > @@ -3515,7 +3515,7 @@ kfree_rcu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
> >  	int cpu, freed = 0;
> >  	unsigned long flags;
> >  
> > -	for_each_online_cpu(cpu) {
> > +	for_each_possible_cpu(cpu) {
> >  		int count;
> >  		struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
> >  
> > @@ -3548,7 +3548,7 @@ void __init kfree_rcu_scheduler_running(void)
> >  	int cpu;
> >  	unsigned long flags;
> >  
> > -	for_each_online_cpu(cpu) {
> > +	for_each_possible_cpu(cpu) {
> >  		struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
> >  
> >  		raw_spin_lock_irqsave(&krcp->lock, flags);
> >
> 
> Should we just clean a krc of a CPU when it goes offline?
> 
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index b8ccd7b5af82..6decb9ad2421 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -2336,10 +2336,15 @@ int rcutree_dead_cpu(unsigned int cpu)
>  {
>         struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
>         struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
> +       struct kfree_rcu_cpu *krcp;
>  
>         if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
>                 return 0;
>  
> +       /* Drain the kcrp of this CPU. IRQs should be disabled? */
> +       krcp = this_cpu_ptr(&krc)
> +       schedule_delayed_work(&krcp->monitor_work, 0);
> +
> 
> A cpu can be offlined and its krp will be stuck until a shrinker is involved.
> Maybe be never.
> 

Yes that is a bug as we discussed on IRC, thanks for following up as well.

We need to acquire the krcp->lock too if no monitor is scheduled then nothing
to do so it does not race with the kfree_rcu_work. So same as what shrinker
does:

		raw_spin_lock_irqsave(&krcp->lock, flags);
		if (krcp->monitor_todo)
			kfree_rcu_drain_unlock(krcp, flags);
		else
			raw_spin_unlock_irqrestore(&krcp->lock, flags);

thanks!

 - Joel