Re: WARNING: suspicious RCU usage in idtentry_exit

From: "Paul E. McKenney" <paulmck@kernel.org>
To: Dmitry Vyukov <dvyukov@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	syzbot <syzbot+3ae5eaae0809ee311e75@syzkaller.appspotmail.com>,
	Paolo Bonzini <pbonzini@redhat.com>,
	Borislav Petkov <bp@alien8.de>, "H. Peter Anvin" <hpa@zytor.com>,
	LKML <linux-kernel@vger.kernel.org>,
	Andy Lutomirski <luto@kernel.org>, Ingo Molnar <mingo@kernel.org>,
	syzkaller-bugs <syzkaller-bugs@googlegroups.com>,
	the arch/x86 maintainers <x86@kernel.org>
Subject: Re: WARNING: suspicious RCU usage in idtentry_exit
Date: Fri, 29 May 2020 07:05:21 -0700	[thread overview]
Message-ID: <20200529140521.GA2869@paulmck-ThinkPad-P72> (raw)
In-Reply-To: <CACT4Y+bVPSZVkWJquu5gk11ymhirsFvVFYHkexjqgBkgga379w@mail.gmail.com>

On Fri, May 29, 2020 at 08:20:12AM +0200, Dmitry Vyukov wrote:
> On Thu, May 28, 2020 at 10:48 PM Paul E. McKenney <paulmck@kernel.org> wrote:
> >
> > On Thu, May 28, 2020 at 10:19:02PM +0200, Thomas Gleixner wrote:
> > > Paul,
> > >
> > > "Paul E. McKenney" <paulmck@kernel.org> writes:
> > > > On Thu, May 28, 2020 at 03:33:44PM +0200, Thomas Gleixner wrote:
> > > >> syzbot <syzbot+3ae5eaae0809ee311e75@syzkaller.appspotmail.com> writes:
> > > >> Weird. I have no idea how that thing is an EQS here.
> > > >
> > > > No argument on the "Weird" part!  ;-)
> > > >
> > > > Is this a NO_HZ_FULL=y kernel?
> > >
> > > No, it has only NO_HZ_IDLE.
> > >
> > >   https://syzkaller.appspot.com/x/.config?x=47b0740d89299c10
> >
> > OK, from the .config, another suggestion is to build the kernel
> > with CONFIG_RCU_EQS_DEBUG=y.  This still requires that this issue be
> > reproduced, but it might catch the problem earlier.
> 
> How much does it slow down execution? If we enable it on syzbot, it
> will affect all fuzzing done by syzbot always.
> It can tolerate significant slowdown and it's far from a production
> kernel (it enables KASAN, KCOV, LOCKDEP and more). But I am still
> asking because some debugging features are built without performance
> in mind at all (like let's just drop a global lock in every
> kmalloc/free, which may be too much even for a standard debug build).

It is an extra WARN_ON_ONCE() with a simple comparison, but on almost
every kernel entry/exit path.

So not something you want in production, but much lighter weight than
any of the tools you listed above.

Full disclosure:  It usually fires for new architectures or for new
timer hardware/drivers.  Which might allow you to enable it selectively.

							Thanx, Paul

> > > > If so, one possibility is that the call
> > > > to rcu_user_exit() went missing somehow.  If not, then RCU should have
> > > > been watching userspace execution.
> > > >
> > > > Again, the only thing I can think of (should this prove to be
> > > > reproducible) is the rcu_dyntick trace event.
> > >
> > > :)
> > >
> > > Thanks,
> > >
> > >         tglx
> >
> >                                                         Thanx, Paul
> >
> > --
> > You received this message because you are subscribed to the Google Groups "syzkaller-bugs" group.
> > To unsubscribe from this group and stop receiving emails from it, send an email to syzkaller-bugs+unsubscribe@googlegroups.com.
> > To view this discussion on the web visit https://groups.google.com/d/msgid/syzkaller-bugs/20200528204839.GR2869%40paulmck-ThinkPad-P72.