linux-kernel - Re: [PATCH v3 2/3] sched/cpuacct: optimize away RCU read lock

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] [day] [month] [year] [list]

Message-ID: <20220312174457.GN4285@paulmck-ThinkPad-P17-Gen-1>
Date:   Sat, 12 Mar 2022 09:44:57 -0800
From:   "Paul E. McKenney" <paulmck@...nel.org>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     Marek Szyprowski <m.szyprowski@...sung.com>,
        Chengming Zhou <zhouchengming@...edance.com>, mingo@...hat.com,
        vincent.guittot@...aro.org, bristot@...hat.com,
        zhaolei@...fujitsu.com, tj@...nel.org, lizefan.x@...edance.com,
        hannes@...xchg.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3 2/3] sched/cpuacct: optimize away RCU read lock

On Sat, Mar 12, 2022 at 01:15:33PM +0100, Peter Zijlstra wrote:
> On Thu, Mar 10, 2022 at 07:01:52AM -0800, Paul E. McKenney wrote:
> 
> > > > > > ./include/linux/cgroup.h:481 suspicious rcu_dereference_check() usage!
> > > > > 
> > > > > Arguably, with the flavours folded again, rcu_dereference_check() ought
> > > > > to default include rcu_read_lock_sched_held() or its equivalent I
> > > > > suppose.
> > > > > 
> > > > > Paul?
> > > > 
> > > > That would reduce the number of warnings, but it also would hide bugs.
> > > > 
> > > > So, are you sure you really want this?
> > > 
> > > I don't understand... Since the flavours got merged regular RCU has it's
> > > quescent state held off by preempt_disable. So how can relying on that
> > > cause bugs?
> > 
> > Somene forgets an rcu_read_lock() and there happens to be something
> > like a preempt_disable() that by coincidence covers that particular
> > rcu_dereference().  The kernel therefore doesn't complain.  That someone
> > goes on to other things, maybe even posthumously.  Then some time later
> > the preempt_disable() goes away, for good and sufficient reasons.
> > 
> > Good luck figuring out where to put the needed rcu_read_lock() and
> > rcu_read_unlock().
> 
> Well, that's software engineering for you.

My point exactly!!!

>                                            Also in that case the warning
> will work as expected. Then figuring out how to fix it is not the
> problem of the warning -- that worked as advertised.
> 
> (also, I don't think it'll be too hard, you just gotta figure out which
> object is rcu protected -- the warning gives you this, where the lookup
> happens -- again the warning helps, and how long it's used for, all
> relatively well definted things)

Without in any way agreeing with that assessment of difficulty, especially
in the general case...  It is -way- easier just to tell RCU what your
design rules are for the code in question.

> I don't see a problem. No bugs hidden.

C'mon, Peter!

There really was a bug hidden.  That someone intended to add some
calls to rcu_read_lock() and rcu_read_unlock() in the proper places.
Their failure to add them really was a bug.

That bug was hidden by: (1) There being a preempt_disable() or
whatever that by coincidence happened to be covering the part of the
code containing the rcu_dereference() and (2) Your proposed change that
would make rcu_dereference() unable to detect that bug.

And that bug can be quite bad.  Given your proposed change, RCU
cannot detect this bug:


	/* Preemption is enabled. */
	/* There should be an rcu_read_lock() here. */
	preempt_disable();
	p = rcu_dereference(gp);
	do_something_with(p);
	preempt_enable();
	/* Without the rcu_read_lock(), *p is history. */
	do_something_else_with(p);
	/* There should be an rcu_read_unlock() here. */

> > > And if we can rely on that, then surely rcu_dereferenced_check() ought
> > > to play by the same rules, otherwise we get silly warnings like these at
> > > hand.
> > > 
> > > Specifically, we removed the rcu_read_lock() here because this has
> > > rq->lock held, which is a raw_spinlock_t which very much implies preempt
> > > disable, on top of that, it's also an IRQ-safe lock and thus IRQs will
> > > be disabled.
> > > 
> > > There is no possible way for RCU to make progress.
> > 
> > Then let's have that particular rcu_dereference_check() explicitly state
> > what it needs, which seems to be either rcu_read_lock() on the one hand.
> > Right now, that could be just this:
> > 
> > 	p = rcu_dereference_check(gp, rcu_read_lock_sched_held());
> > 
> > Or am I missing something here?
> 
> That will work; I just don't agree with it. Per the rules of RCU it is
> entirely correct to mix rcu_read_lock() and preempt_disable() (or
> anything that implies the same). So I strongly feel that
> rcu_dereference() should not warn about obviously correct code. Why
> would we need to special case this ?

This use case might well be entirely correct, but it is most certainly
not the common case.

Therefore, my answer to this requested chance in rcu_dereference()
semantics is "no".

							Thanx, Paul