linux-kernel - Re: [PATCH v2] percpu-refcount: Use normal instead of RCU-sched"

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: <20191117041113.GA39597@dennisz-mbp>
Date:   Sat, 16 Nov 2019 23:11:13 -0500
From:   Dennis Zhou <dennis@...nel.org>
To:     Sebastian Andrzej Siewior <bigeasy@...utronix.de>
Cc:     Dennis Zhou <dennis@...nel.org>, linux-kernel@...r.kernel.org,
        Tejun Heo <tj@...nel.org>, Christoph Lameter <cl@...ux.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Peter Zijlstra <peterz@...radead.org>,
        "Paul E. McKenney" <paulmck@...nel.org>
Subject: Re: [PATCH v2] percpu-refcount: Use normal instead of RCU-sched"

On Fri, Nov 08, 2019 at 06:35:53PM +0100, Sebastian Andrzej Siewior wrote:
> This is a revert of commit
>    a4244454df129 ("percpu-refcount: use RCU-sched insted of normal RCU")
> 
> which claims the only reason for using RCU-sched is
>    "rcu_read_[un]lock() … are slightly more expensive than preempt_disable/enable()"
> 
> and
>     "As the RCU critical sections are extremely short, using sched-RCU
>     shouldn't have any latency implications."
> 
> The problem with using RCU-sched here is that it disables preemption and
> the release callback (called from percpu_ref_put_many()) must not
> acquire any sleeping locks like spinlock_t. This breaks PREEMPT_RT
> because some of the users acquire spinlock_t locks in their callbacks.
> 
> Using rcu_read_lock() on PREEMPTION=n kernels is not any different
> compared to rcu_read_lock_sched(). On PREEMPTION=y kernels there are
> already performance issues due to additional preemption points.
> Looking at the code, the rcu_read_lock() is just an increment and unlock
> is almost just a decrement unless there is something special to do. Both
> are functions while disabling preemption is inlined.
> Doing a small benchmark, the minimal amount of time required was mostly
> the same. The average time required was higher due to the higher MAX
> value (which could be preemption). With DEBUG_PREEMPT=y it is
> rcu_read_lock_sched() that takes a little longer due to the additional
> debug code.
> 
> Convert back to normal RCU.
> 
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
> ---
> On 2019-11-07 12:36:53 [-0500], Dennis Zhou wrote:
> > > some RCU section here invoke callbacks which acquire spinlock_t locks.
> > > This does not work on RT with disabled preemption.
> > > 
> > 
> > Yeah, so adding a bit in the commit message about why it's an issue for
> > RT kernels with disabled preemption as I don't believe this is an issue
> > for non-RT kernels.
> 
> I realized that I had partly in the commit message so I rewrote the
> second chapter hopefully covering it all now more explicit.
> 
> v1…v2: Slightly rewriting the second paragraph regarding RT
> implications.
> 
>  include/linux/percpu-refcount.h | 16 ++++++++--------
>  1 file changed, 8 insertions(+), 8 deletions(-)
> 
> diff --git a/include/linux/percpu-refcount.h b/include/linux/percpu-refcount.h
> index 7aef0abc194a2..390031e816dcd 100644
> --- a/include/linux/percpu-refcount.h
> +++ b/include/linux/percpu-refcount.h
> @@ -186,14 +186,14 @@ static inline void percpu_ref_get_many(struct percpu_ref *ref, unsigned long nr)
>  {
>  	unsigned long __percpu *percpu_count;
>  
> -	rcu_read_lock_sched();
> +	rcu_read_lock();
>  
>  	if (__ref_is_percpu(ref, &percpu_count))
>  		this_cpu_add(*percpu_count, nr);
>  	else
>  		atomic_long_add(nr, &ref->count);
>  
> -	rcu_read_unlock_sched();
> +	rcu_read_unlock();
>  }
>  
>  /**
> @@ -223,7 +223,7 @@ static inline bool percpu_ref_tryget(struct percpu_ref *ref)
>  	unsigned long __percpu *percpu_count;
>  	bool ret;
>  
> -	rcu_read_lock_sched();
> +	rcu_read_lock();
>  
>  	if (__ref_is_percpu(ref, &percpu_count)) {
>  		this_cpu_inc(*percpu_count);
> @@ -232,7 +232,7 @@ static inline bool percpu_ref_tryget(struct percpu_ref *ref)
>  		ret = atomic_long_inc_not_zero(&ref->count);
>  	}
>  
> -	rcu_read_unlock_sched();
> +	rcu_read_unlock();
>  
>  	return ret;
>  }
> @@ -257,7 +257,7 @@ static inline bool percpu_ref_tryget_live(struct percpu_ref *ref)
>  	unsigned long __percpu *percpu_count;
>  	bool ret = false;
>  
> -	rcu_read_lock_sched();
> +	rcu_read_lock();
>  
>  	if (__ref_is_percpu(ref, &percpu_count)) {
>  		this_cpu_inc(*percpu_count);
> @@ -266,7 +266,7 @@ static inline bool percpu_ref_tryget_live(struct percpu_ref *ref)
>  		ret = atomic_long_inc_not_zero(&ref->count);
>  	}
>  
> -	rcu_read_unlock_sched();
> +	rcu_read_unlock();
>  
>  	return ret;
>  }
> @@ -285,14 +285,14 @@ static inline void percpu_ref_put_many(struct percpu_ref *ref, unsigned long nr)
>  {
>  	unsigned long __percpu *percpu_count;
>  
> -	rcu_read_lock_sched();
> +	rcu_read_lock();
>  
>  	if (__ref_is_percpu(ref, &percpu_count))
>  		this_cpu_sub(*percpu_count, nr);
>  	else if (unlikely(atomic_long_sub_and_test(nr, &ref->count)))
>  		ref->release(ref);
>  
> -	rcu_read_unlock_sched();
> +	rcu_read_unlock();
>  }
>  
>  /**
> -- 
> 2.24.0
> 
> 

Sorry for sitting on this for so long. I've applied it to for-5.5.

Thanks,
Dennis