Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 14 Nov 2022 12:41:48 +0000
From:   Will Deacon <will@kernel.org>
To:     Jan Kara <jack@suse.cz>
Cc:     Waiman Long <longman@redhat.com>,
        Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
        LKML <linux-kernel@vger.kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Steven Rostedt <rostedt@goodmis.org>,
        Mel Gorman <mgorman@suse.de>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Catalin Marinas <catalin.marinas@arm.com>
Subject: Re: Crash with PREEMPT_RT on aarch64 machine
Message-ID: <20221114124147.GA30263@willie-the-truck>
References: <20221103115444.m2rjglbkubydidts@quack3>
 <Y2U+Je+LICO2HkNY@linutronix.de>
 <20221107135636.biouna36osqc4rik@quack3>
 <Y2kf6tcX47Cl7q0W@linutronix.de>
 <359cc93a-fce0-5af2-0fd5-81999fad186b@redhat.com>
 <20221109125756.GA24388@willie-the-truck>
 <20221109154023.cx2d4y3e7zqnuo35@quack3>
 <20221111142742.rh677sdwu55aeeno@quack3>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20221111142742.rh677sdwu55aeeno@quack3>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On Fri, Nov 11, 2022 at 03:27:42PM +0100, Jan Kara wrote:
> On Wed 09-11-22 16:40:23, Jan Kara wrote:
> > On Wed 09-11-22 12:57:57, Will Deacon wrote:
> > > On Mon, Nov 07, 2022 at 11:49:01AM -0500, Waiman Long wrote:
> > > > On 11/7/22 10:10, Sebastian Andrzej Siewior wrote:
> > > > > + locking, arm64
> > > > > 
> > > > > On 2022-11-07 14:56:36 [+0100], Jan Kara wrote:
> > > > > > > spinlock_t and raw_spinlock_t differ slightly in terms of locking.
> > > > > > > rt_spin_lock() has the fast path via try_cmpxchg_acquire(). If you
> > > > > > > enable CONFIG_DEBUG_RT_MUTEXES then you would force the slow path which
> > > > > > > always acquires the rt_mutex_base::wait_lock (which is a raw_spinlock_t)
> > > > > > > while the actual lock is modified via cmpxchg.
> > > > > > So I've tried enabling CONFIG_DEBUG_RT_MUTEXES and indeed the corruption
> > > > > > stops happening as well. So do you suspect some bug in the CPU itself?
> > > > > If it is only enabling CONFIG_DEBUG_RT_MUTEXES (and not whole lockdep)
> > > > > then it looks very suspicious.
> > > > > CONFIG_DEBUG_RT_MUTEXES enables a few additional checks but the main
> > > > > part is that rt_mutex_cmpxchg_acquire() + rt_mutex_cmpxchg_release()
> > > > > always fail (and so the slowpath under a raw_spinlock_t is done).
> > > > > 
> > > > > So if it is really the fast path (rt_mutex_cmpxchg_acquire()) then it
> > > > > somehow smells like the CPU is misbehaving.
> > > > > 
> > > > > Could someone from the locking/arm64 department check if the locking in
> > > > > RT-mutex (rtlock_lock()) is correct?
> > > > > 
> > > > > rtmutex locking uses try_cmpxchg_acquire(, ptr, ptr) for the fastpath
> > > > > (and try_cmpxchg_release(, ptr, ptr) for unlock).
> > > > > Now looking at it again, I don't see much difference compared to what
> > > > > queued_spin_trylock() does except the latter always operates on 32bit
> > > > > value instead a pointer.
> > > > 
> > > > Both the fast path of queued spinlock and rt_spin_lock are using
> > > > try_cmpxchg_acquire(), the only difference I saw is the size of the data to
> > > > be cmpxchg'ed. qspinlock uses 32-bit integer whereas rt_spin_lock uses
> > > > 64-bit pointer. So I believe it is more on how the arm64 does cmpxchg. I
> > > > believe there are two different ways of doing it depending on whether LSE
> > > > atomics is available in the platform. So exactly what arm64 system is being
> > > > used here and what hardware capability does it have?
> > > 
> > > I'd be more inclined to be suspicious of the slowpath tbh, as we need to
> > > make sure that we have acquire semantics on all paths where the lock can
> > > be taken. Looking at the rtmutex code, this really isn't obvious to me --
> > > for example, try_to_take_rt_mutex() appears to be able to return via the
> > > 'takeit' label without acquire semantics and it looks like we might be
> > > relying on the caller's subsequent _unlock_ of the wait_lock for ordering,
> > > but that will give us release semantics which aren't correct.
> > > 
> > > As a quick hack, can you try chucking a barrier into rt_mutex_set_owner()?
> > 
> > Bingo! This patch fixes the crashes for me.
> 
> So I suppose this is not an official fix, is it? Sebastian, it appears to
> be a bug in rtmutex implementation in the end AFAIU ;)

Right, somebody needs to go audit all the acquisition paths on the slow-path
and make sure they all have acquire semantics. The trick is doing that
without incurring unnecessary overhead, e.g. by making use of dependency
ordering where it already exists.

Will

> 
> > > diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
> > > index 7779ee8abc2a..dd6a66c90f53 100644
> > > --- a/kernel/locking/rtmutex.c
> > > +++ b/kernel/locking/rtmutex.c
> > > @@ -98,6 +98,7 @@ rt_mutex_set_owner(struct rt_mutex_base *lock, struct task_struct *owner)
> > >                 val |= RT_MUTEX_HAS_WAITERS;
> > >  
> > >         WRITE_ONCE(lock->owner, (struct task_struct *)val);
> > > +       smp_mb();
> > >  }
> > >  
> > >  static __always_inline void clear_rt_mutex_waiters(struct rt_mutex_base *lock)
> 
> 								Honza
> -- 
> Jan Kara <jack@suse.com>
> SUSE Labs, CR