Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 27 Oct 2020 20:17:11 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Sean Christopherson <sean.j.christopherson@intel.com>
Cc:     Ben Gardon <bgardon@google.com>, linux-kernel@vger.kernel.org,
        kvm@vger.kernel.org, Paolo Bonzini <pbonzini@redhat.com>,
        Peter Shier <pshier@google.com>,
        Ingo Molnar <mingo@redhat.com>, Will Deacon <will@kernel.org>,
        Waiman Long <longman@redhat.com>,
        Davidlohr Bueso <dave@stgolabs.net>
Subject: Re: [PATCH 3/3] sched: Add cond_resched_rwlock
Message-ID: <20201027191711.GP2628@hirez.programming.kicks-ass.net>
References: <20201027164950.1057601-1-bgardon@google.com>
 <20201027164950.1057601-3-bgardon@google.com>
 <20201027175634.GI1021@linux.intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20201027175634.GI1021@linux.intel.com>
Precedence: bulk

On Tue, Oct 27, 2020 at 10:56:36AM -0700, Sean Christopherson wrote:
> On Tue, Oct 27, 2020 at 09:49:50AM -0700, Ben Gardon wrote:
> > Rescheduling while holding a spin lock is essential for keeping long
> > running kernel operations running smoothly. Add the facility to
> > cond_resched rwlocks.
> 
> This adds two new exports and two new macros without any in-tree users, which
> is generally frowned upon.  You and I know these will be used by KVM's new
> TDP MMU, but the non-KVM folks, and more importantly the maintainers of this
> code, are undoubtedly going to ask "why".  I.e. these patches probably belong
> in the KVM series to switch to a rwlock for the TDP MMU.

I was informed about this ;-)

> Regarding the code, it's all copy-pasted from the spinlock code and darn near
> identical.  It might be worth adding builder macros for these.

I considered mentioning them; I'm typically a fan of them, but I'm not
quite sure it's worth the effort here.

> > +int __cond_resched_rwlock_read(rwlock_t *lock)
> > +{
> > +	int resched = should_resched(PREEMPT_LOCK_OFFSET);
> > +	int ret = 0;
> > +
> > +	lockdep_assert_held(lock);
> > +
> > +	if (rwlock_needbreak(lock) || resched) {
> > +		read_unlock(lock);
> > +		if (resched)
> > +			preempt_schedule_common();
> > +		else
> > +			cpu_relax();
> > +		ret = 1;
> 
> AFAICT, this rather odd code flow from __cond_resched_lock() is an artifact of
> code changes over the years and not intentionally weird.  IMO, it would be
> cleaner and easier to read as:
> 
> 	int resched = should_resched(PREEMPT_LOCK_OFFSET);
> 
> 	lockdep_assert_held(lock);

lockdep_assert_held_read() :-)

> 
> 	if (!rwlock_needbreak(lock) && !resched)
> 		return 0;
> 
> 	read_unlock(lock);
> 	if (resched)
> 		preempt_schedule_common();
> 	else
> 		cpu_relax();
> 	read_lock(lock)
> 	return 1;
> 

I suppose that works, but then also change the existing one.