Re: [PATCH 04/19] sched: Prepare for Core-wide rq->lock

From: Peter Zijlstra
Date: Tue May 04 2021 - 03:39:10 EST


On Thu, Apr 29, 2021 at 01:39:54PM -0700, Josh Don wrote:

> > > +void double_rq_lock(struct rq *rq1, struct rq *rq2)
> > > +{
> > > + lockdep_assert_irqs_disabled();
> > > +
> > > + if (rq1->cpu > rq2->cpu)
> >
> > It's still a bit hard for me to digest this function, I guess using (rq->cpu)
> > can't guarantee the sequence of locking when coresched is enabled.
> >
> > - cpu1 and cpu7 shares lockA
> > - cpu2 and cpu8 shares lockB
> >
> > double_rq_lock(1,8) leads to lock(A) and lock(B)
> > double_rq_lock(7,2) leads to lock(B) and lock(A)

Good one!

> > change to below to avoid ABBA?
> > + if (__rq_lockp(rq1) > __rq_lockp(rq2))

This, however, is broken badly, not only does it suffer the problem Josh
pointed out, it also breaks the rq->__lock ordering vs
__sched_core_flip(), which was the whole reason the ordering needed
changing in the first place.

> I'd propose an alternative but
> similar idea: order by core, then break ties by ordering on cpu.
>
> +#ifdef CONFIG_SCHED_CORE
> + if (rq1->core->cpu > rq2->core->cpu)
> + swap(rq1, rq2);
> + else if (rq1->core->cpu == rq2->core->cpu && rq1->cpu > rq2->cpu)
> + swap(rq1, rq2);
> +#else
> if (rq1->cpu > rq2->cpu)
> swap(rq1, rq2);
> +#endif

I've written it like so:

static inline bool rq_order_less(struct rq *rq1, struct rq *rq2)
{
#ifdef CONFIG_SCHED_CORE
if (rq1->core->cpu < rq2->core->cpu)
return true;
if (rq1->core->cpu > rq2->core->cpu)
return false;
#endif
return rq1->cpu < rq2->cpu;
}

/*
* double_rq_lock - safely lock two runqueues
*/
void double_rq_lock(struct rq *rq1, struct rq *rq2)
{
lockdep_assert_irqs_disabled();

if (rq_order_less(rq2, rq1))
swap(rq1, rq2);

raw_spin_rq_lock(rq1);
if (rq_lockp(rq1) == rq_lockp(rq2))
return;

raw_spin_rq_lock_nested(rq2, SINGLE_DEPTH_NESTING);
}