linux-kernel - Re: [RFC] Make need_resched() return true when rcu_urgent

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]

Date:   Thu, 19 Jul 2018 02:26:40 +0200
From:   Frederic Weisbecker <frederic@...nel.org>
To:     "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>
Cc:     David Woodhouse <dwmw2@...radead.org>,
        Christian Borntraeger <borntraeger@...ibm.com>,
        Peter Zijlstra <peterz@...radead.org>, mhillenb@...zon.de,
        linux-kernel <linux-kernel@...r.kernel.org>,
        kvm <kvm@...r.kernel.org>
Subject: Re: [RFC] Make need_resched() return true when rcu_urgent_qs
 requested

On Wed, Jul 18, 2018 at 01:17:00PM -0700, Paul E. McKenney wrote:
> On Wed, Jul 18, 2018 at 09:41:05PM +0200, David Woodhouse wrote:
> > 
> > 
> > On Wed, 2018-07-18 at 09:37 -0700, Paul E. McKenney wrote:
> > > On Wed, Jul 18, 2018 at 06:01:51PM +0200, David Woodhouse wrote:
> > > > 
> > > > On Wed, 2018-07-18 at 08:36 -0700, Paul E. McKenney wrote:
> > > > > 
> > > > > And I finally did get some near misses from an earlier commit, so we
> > > > > should consider your patch to be officially off the hook.
> > > >
> > > > Yay, I like it when it's not my fault. I'll redo it with the ifdef
> > > > CONFIG_NO_HZ_FULL.
> > >
> > > Hey, I didn't say it wasn't your fault, only that it -officially- wasn't
> > > your fault.  ;-)
> > 
> > I can live with being innocent until proven guilty.
> > 
> > > > 
> > > > What should it do for the !CONFIG_NO_HZ_FULL case? The existing call in
> > > > guest_enter_irqoff() clearly wasn't actually doing the right thing
> > > > anyway, hence the need for the need_resched() patch in $SUBJECT... so
> > > > should I just leave it doing nothing in guest_enter_irqoff()? 
> > >
> > > One starting point would be the combination of your patch and my
> > > patch, with -rcu commit IDs and diff below.  But yes, it needs to be
> > > !CONFIG_NO_HZ_FULL.  And no, I am not at all confident that I actually
> > > found all the places needing change in the core code, so this needs some
> > > serious review both by the KVM guys and the NO_HZ_FULL guys.
> > 
> > Right, that looks fairly much like the version I'd ended up with. So my
> > question was...
> > 
> > > --- a/include/linux/context_tracking.h
> > > +++ b/include/linux/context_tracking.h
> > > @@ -118,12 +118,12 @@ static inline void guest_enter_irqoff(void)
> > >  	 * one time slice). Lets treat guest mode as quiescent state, just like
> > >  	 * we do with user-mode execution.
> > >  	 */
> > 
> > ...if we change this to something like...
> > 
> > #ifdef CONFIG_NO_HZ_FULL
> > > +	rcu_kvm_enter();
> > #else
> > > 	if (!context_tracking_cpu_is_enabled())
> > > 		rcu_virt_note_context_switch(smp_processor_id());
> > #endif
> > 
> > ... do you actually want me to keep the #else case there? It blatantly
> > wasn't working anyway for us, perhaps because the condition was false?
> > That's why I started fixing need_resched() in the first place, and that
> > fix ought to cover whatever this call to rcu_virt_note_context_switch()
> > was supposed to be doing?
> 
> My thought would be something like this:
> 
> 	if (context_tracking_cpu_is_enabled())
> 		rcu_kvm_enter();
> 	else
> 		rcu_virt_note_context_switch(smp_processor_id());
> 
> The reason I believe that this is the right approach is that even when
> you have CONFIG_NO_HZ_FULL=y, some CPUs will still be nohz_full=n CPUs,
> so you don't want to take the extra overhead on those CPUs.
> 
> But I could easily be confused here, so I am adding Frederic for his
> thoughts.

Hmm, actually rcu_user_enter()/rcu_user_exit() are already called upon guest
entry/exit :-)

Now I must confess the code leading there in __context_tracking_enter/exit is
not very obvious but that part applies to both CONTEXT_USER and CONTEXT_GUEST cases.
I should probably add a few comments to clarify.