Re: [PATCH] srcu: Remove srcu_queue_delayed_work_on()

[Joel Fernandes]

On Wed, Dec 12, 2018 at 07:55:46AM -0800, Paul E. McKenney wrote:
> On Wed, Dec 12, 2018 at 10:05:19PM +0800, Boqun Feng wrote:
> > Hi Joel,
> > 
> > On Tue, Dec 11, 2018 at 05:40:16PM -0800, Joel Fernandes wrote:
> > > On Tue, Dec 11, 2018 at 12:12:38PM +0100, Sebastian Andrzej Siewior wrote:
> > > > srcu_queue_delayed_work_on() disables preemption (and therefore CPU
> > > > hotplug in RCU's case) and then checks based on its own accounting if a
> > > > CPU is online. If the CPU is online it uses queue_delayed_work_on()
> > > > otherwise it fallbacks to queue_delayed_work().
> > > > The problem here is that queue_work() on -RT does not work with disabled
> > > > preemption.
> > > > 
> > > > queue_work_on() works also on an offlined CPU. queue_delayed_work_on()
> > > > has the problem that it is possible to program a timer on an offlined
> > > > CPU. This timer will fire once the CPU is online again. But until then,
> > > > the timer remains programmed and nothing will happen.
> > > > Add a local timer which will fire (as requested per delay) on the local
> > > > CPU and then enqueue the work on the specific CPU.
> > > > 
> > > > RCUtorture testing with SRCU-P for 24h showed no problems.
> > > > 
> > > > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
> > > > ---
> > > >  include/linux/srcutree.h |  3 ++-
> > > >  kernel/rcu/srcutree.c    | 57 ++++++++++++++++++----------------------
> > > >  kernel/rcu/tree.c        |  4 ---
> > > >  kernel/rcu/tree.h        |  8 ------
> > > >  4 files changed, 27 insertions(+), 45 deletions(-)
> > > > 
> > > > diff --git a/include/linux/srcutree.h b/include/linux/srcutree.h
> > > > index 6f292bd3e7db7..0faa978c98807 100644
> > > > --- a/include/linux/srcutree.h
> > > > +++ b/include/linux/srcutree.h
> > > > @@ -45,7 +45,8 @@ struct srcu_data {
> > > >  	unsigned long srcu_gp_seq_needed;	/* Furthest future GP needed. */
> > > >  	unsigned long srcu_gp_seq_needed_exp;	/* Furthest future exp GP. */
> > > >  	bool srcu_cblist_invoking;		/* Invoking these CBs? */
> > > > -	struct delayed_work work;		/* Context for CB invoking. */
> > > > +	struct timer_list delay_work;		/* Delay for CB invoking */
> > > > +	struct work_struct work;		/* Context for CB invoking. */
> > > >  	struct rcu_head srcu_barrier_head;	/* For srcu_barrier() use. */
> > > >  	struct srcu_node *mynode;		/* Leaf srcu_node. */
> > > >  	unsigned long grpmask;			/* Mask for leaf srcu_node */
> > > > diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c
> > > > index 3600d88d8956b..7f041f2435df9 100644
> > > > --- a/kernel/rcu/srcutree.c
> > > > +++ b/kernel/rcu/srcutree.c
> > > > @@ -58,6 +58,7 @@ static bool __read_mostly srcu_init_done;
> > > >  static void srcu_invoke_callbacks(struct work_struct *work);
> > > >  static void srcu_reschedule(struct srcu_struct *ssp, unsigned long delay);
> > > >  static void process_srcu(struct work_struct *work);
> > > > +static void srcu_delay_timer(struct timer_list *t);
> > > >  
> > > >  /* Wrappers for lock acquisition and release, see raw_spin_lock_rcu_node(). */
> > > >  #define spin_lock_rcu_node(p)					\
> > > > @@ -156,7 +157,8 @@ static void init_srcu_struct_nodes(struct srcu_struct *ssp, bool is_static)
> > > >  			snp->grphi = cpu;
> > > >  		}
> > > >  		sdp->cpu = cpu;
> > > > -		INIT_DELAYED_WORK(&sdp->work, srcu_invoke_callbacks);
> > > > +		INIT_WORK(&sdp->work, srcu_invoke_callbacks);
> > > > +		timer_setup(&sdp->delay_work, srcu_delay_timer, 0);
> > > >  		sdp->ssp = ssp;
> > > >  		sdp->grpmask = 1 << (cpu - sdp->mynode->grplo);
> > > >  		if (is_static)
> > > > @@ -386,13 +388,19 @@ void _cleanup_srcu_struct(struct srcu_struct *ssp, bool quiesced)
> > > >  	} else {
> > > >  		flush_delayed_work(&ssp->work);
> > > >  	}
> > > > -	for_each_possible_cpu(cpu)
> > > > +	for_each_possible_cpu(cpu) {
> > > > +		struct srcu_data *sdp = per_cpu_ptr(ssp->sda, cpu);
> > > > +
> > > >  		if (quiesced) {
> > > > -			if (WARN_ON(delayed_work_pending(&per_cpu_ptr(ssp->sda, cpu)->work)))
> > > > +			if (WARN_ON(timer_pending(&sdp->delay_work)))
> > > > +				return; /* Just leak it! */
> > > > +			if (WARN_ON(work_pending(&sdp->work)))
> > > >  				return; /* Just leak it! */
> > > >  		} else {
> > > > -			flush_delayed_work(&per_cpu_ptr(ssp->sda, cpu)->work);
> > > > +			del_timer_sync(&sdp->delay_work);
> > > > +			flush_work(&sdp->work);
> > > >  		}
> > > > +	}
> > > >  	if (WARN_ON(rcu_seq_state(READ_ONCE(ssp->srcu_gp_seq)) != SRCU_STATE_IDLE) ||
> > > >  	    WARN_ON(srcu_readers_active(ssp))) {
> > > >  		pr_info("%s: Active srcu_struct %p state: %d\n",
> > > > @@ -463,39 +471,23 @@ static void srcu_gp_start(struct srcu_struct *ssp)
> > > >  	WARN_ON_ONCE(state != SRCU_STATE_SCAN1);
> > > >  }
> > > >  
> > > > -/*
> > > > - * Track online CPUs to guide callback workqueue placement.
> > > > - */
> > > > -DEFINE_PER_CPU(bool, srcu_online);
> > > >  
> > > > -void srcu_online_cpu(unsigned int cpu)
> > > > +static void srcu_delay_timer(struct timer_list *t)
> > > >  {
> > > > -	WRITE_ONCE(per_cpu(srcu_online, cpu), true);
> > > > +	struct srcu_data *sdp = container_of(t, struct srcu_data, delay_work);
> > > > +
> > > > +	queue_work_on(sdp->cpu, rcu_gp_wq, &sdp->work);
> > > >  }
> > > >  
> > > > -void srcu_offline_cpu(unsigned int cpu)
> > > > -{
> > > > -	WRITE_ONCE(per_cpu(srcu_online, cpu), false);
> > > > -}
> > > > -
> > > > -/*
> > > > - * Place the workqueue handler on the specified CPU if online, otherwise
> > > > - * just run it whereever.  This is useful for placing workqueue handlers
> > > > - * that are to invoke the specified CPU's callbacks.
> > > > - */
> > > > -static bool srcu_queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
> > > > -				       struct delayed_work *dwork,
> > > > +static void srcu_queue_delayed_work_on(struct srcu_data *sdp,
> > > >  				       unsigned long delay)
> > > >  {
> > > > -	bool ret;
> > > > +	if (!delay) {
> > > > +		queue_work_on(sdp->cpu, rcu_gp_wq, &sdp->work);
> > > > +		return;
> > > > +	}
> > > >  
> > > > -	preempt_disable();
> > > > -	if (READ_ONCE(per_cpu(srcu_online, cpu)))
> > > > -		ret = queue_delayed_work_on(cpu, wq, dwork, delay);
> > > > -	else
> > > > -		ret = queue_delayed_work(wq, dwork, delay);
> > > > -	preempt_enable();
> > > 
> > > The deleted code looks like 'cpu' could be offlined.
> > > 
> > > Question for my clarification: According to sync_rcu_exp_select_cpus, we have
> > > to disable preemption before calling queue_work_on to ensure the CPU is online.
> > > 
> > > Also same is said in Boqun's presentation on the topic:
> > > https://linuxplumbersconf.org/event/2/contributions/158/attachments/68/79/workqueue_and_cpu_hotplug.pdf
> > > 
> > > Calling queue_work_on on an offline CPU sounds like could be problematic. So
> > > in your patch, don't you still need to disable preemption around
> > > queue_work_on, or the very least check if said CPU is online before calling
> > > queue_work_on?
> > > 
> > 
> > I should be the one who answers this ;-)
> > So I found something after my LPC topic, that is queue_work_on() may
> > work well even if racing with a cpu offline. The reason is that in the
> > cpu offline hook for workqueue only switches the percpu pool into an
> > unbound one, so as long as the related cpu has been online once, we are
> > fine here. 
> > 
> > Please note this is only my own analysis, and I'm the one who told
> > Sebastian and Paul about this.
> > 
> > I should send an email to check with workqueue maintainers about this,
> > but I didn't find the time. Sorry about this.. But let's do this right
> > now, while we at it.
> > 
> > (Cc TJ)
> > 
> > So Jiangshan and TJ, what's your opion on this one? If we call a
> > queue_work_on() at a place where that target cpu may be offlined, I
> > think we have the guarantee that the work will be eventually executed
> > even if the cpu is never online again, right? In other words, if a cpu
> > has been online once, queue_work_on() on it will be free from racing
> > with cpu hotplug.
> > 
> > Am I right about this, or did I miss something subtle?
> 
> Well, what I was relying on was Thomas Gleixner's assertion that it should
> work and would be fixed if it didn't.  ;-)

FWIW I did a simple manual test calling queue_work_on on an offline CPU to
see what happens and it appears to be working fine. On a 4 CPU system, I
offline CPU 3 and the work ends up executing on CPU 0 instead. But I haven't
done any stress tests to know of any possible races with doing so.

thanks,

- Joel