Received-SPF: pass (google.com: domain of linux-kernel+bounces-25229-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Precedence: bulk
MIME-Version: 1.0
References: <cover.1700502145.git.andreyknvl@google.com> <9f81ffcc4bb422ebb6326a65a770bf1918634cbb.1700502145.git.andreyknvl@google.com>
 <ZZUlgs69iTTlG8Lh@localhost.localdomain> <87sf34lrn3.fsf@linux.intel.com>
 <CANpmjNNdWwGsD3JRcEqpq_ywwDFoxsBjz6n=6vL5YksNsPyqHw@mail.gmail.com>
 <ZZ_gssjTCyoWjjhP@tassilo> <ZaA8oQG-stLAVTbM@elver.google.com>
 <CA+fCnZeS=OrqSK4QVUVdS6PwzGrpg8CBj8i2Uq=VMgMcNg1FYw@mail.gmail.com>
 <CANpmjNOoidtyeQ76274SWtTYR4zZPdr1DnxhLaagHGXcKwPOhA@mail.gmail.com>
 <ZaG56XTDwPfkqkJb@elver.google.com> <ZaHmQU5DouedI9kS@tassilo>
In-Reply-To: <ZaHmQU5DouedI9kS@tassilo>
From: Marco Elver <elver@google.com>
Date: Sat, 13 Jan 2024 10:12:21 +0100
Message-ID: <CANpmjNO-q4pjS4z=W8xVLHTs72FNq+TR+-=QBmkP=HOQy6UHmg@mail.gmail.com>
Subject: Re: [PATCH v4 12/22] lib/stackdepot: use read/write lock
To: Andi Kleen <ak@linux.intel.com>
Cc: Andrey Konovalov <andreyknvl@gmail.com>, Oscar Salvador <osalvador@suse.de>, andrey.konovalov@linux.dev, 
	Andrew Morton <akpm@linux-foundation.org>, Alexander Potapenko <glider@google.com>, 
	Dmitry Vyukov <dvyukov@google.com>, Vlastimil Babka <vbabka@suse.cz>, kasan-dev@googlegroups.com, 
	Evgenii Stepanov <eugenis@google.com>, linux-mm@kvack.org, linux-kernel@vger.kernel.org, 
	Andrey Konovalov <andreyknvl@google.com>
Content-Type: text/plain; charset="UTF-8"

On Sat, 13 Jan 2024 at 02:24, Andi Kleen <ak@linux.intel.com> wrote:
>
> On Fri, Jan 12, 2024 at 11:15:05PM +0100, Marco Elver wrote:
> > +             /*
> > +              * Stack traces of size 0 are never saved, and we can simply use
> > +              * the size field as an indicator if this is a new unused stack
> > +              * record in the freelist.
> > +              */
> > +             stack->size = 0;
>
> I would use WRITE_ONCE here too, at least for TSan.

This is written with the pool_lock held.

> > +             return NULL;
> > +
> > +     /*
> > +      * We maintain the invariant that the elements in front are least
> > +      * recently used, and are therefore more likely to be associated with an
> > +      * RCU grace period in the past. Consequently it is sufficient to only
> > +      * check the first entry.
> > +      */
> > +     stack = list_first_entry(&free_stacks, struct stack_record, free_list);
> > +     if (stack->size && !poll_state_synchronize_rcu(stack->rcu_state))
>
> READ_ONCE (also for TSan, and might be safer long term in case the
> compiler considers some fancy code transformation)

And this is also only read with the pool_lock held, so it's impossible
that there'd be a data race due to size. (And if there is a data race,
I'd want KCSAN to tell us because that'd be a bug then.)
depot_pop_free() can't be used w/o the lock because it's manipulating
the freelist.
To be sure, I'm adding a lockdep_assert_held() to depot_pop_free().

> > +             return NULL;
> >
> > +             stack = depot_pop_free();
> > +             if (WARN_ON(!stack))
>
> Won't you get nesting problems here if this triggers due to the print?
> I assume the nmi safe printk won't consider it like an NMI.
>
> >       counters[DEPOT_COUNTER_FREELIST_SIZE]++;
> >       counters[DEPOT_COUNTER_FREES]++;
> >       counters[DEPOT_COUNTER_INUSE]--;
> > +
> > +     printk_deferred_exit();
>
> Ah this handles the WARN_ON? Should be ok then.

Yes, the pool_lock critical sections are surrounded by printk_deferred.

Thanks,
-- Marco