2023-11-25 19:59:57

by Tejun Heo

Subject: Re: [PATCH 12/36] sched_ext: Implement BPF extensible scheduler class

Hello,

On Thu, Nov 23, 2023 at 09:07:35AM +0100, Andrea Righi wrote:
> On Fri, Nov 10, 2023 at 04:47:38PM -1000, Tejun Heo wrote:
...
> > +#ifdef CONFIG_SCHED_CLASS_EXT
> > + p->scx.dsq = NULL;
> > + INIT_LIST_HEAD(&p->scx.dsq_node);
> > + p->scx.flags = 0;
> > + p->scx.weight = 0;
> > + p->scx.sticky_cpu = -1;
> > + p->scx.holding_cpu = -1;
> > + p->scx.kf_mask = 0;
> > + atomic64_set(&p->scx.ops_state, 0);
>
> We probably need atomic_long_set() here or in 32-bit arches (such as
> armhf) we get this:
>
> kernel/sched/core.c:4564:22: error: passing argument 1 of ‘atomic64_set’ from incompatible pointer type [-Werror=incompatible-pointer-types]
> 4564 | atomic64_set(&p->scx.ops_state, 0);
> | ^~~~~~~~~~~~~~~~~
> | |
> | atomic_long_t * {aka atomic_t *}
>
...
> > +static void set_next_task_scx(struct rq *rq, struct task_struct *p, bool first)
> > +{
> > + if (p->scx.flags & SCX_TASK_QUEUED) {
> > + WARN_ON_ONCE(atomic64_read(&p->scx.ops_state) != SCX_OPSS_NONE);
>
> Ditto. Even if this line is replaced later by
> "[PATCH 31/36] sched_ext: Implement core-sched support"
>
> > + dispatch_dequeue(&rq->scx, p);
> > + }
> > +
> > + p->se.exec_start = rq_clock_task(rq);
> > +}

Sorry about that. I updated them and will include the changes in the next
iteration. Will test 32bit build too next time.

Thanks.

--
tejun

2023-11-26 09:06:47

by Andrea Righi

Subject: Re: [PATCH 12/36] sched_ext: Implement BPF extensible scheduler class

On Sat, Nov 25, 2023 at 09:59:43AM -1000, Tejun Heo wrote:
> Hello,
>
> On Thu, Nov 23, 2023 at 09:07:35AM +0100, Andrea Righi wrote:
> > On Fri, Nov 10, 2023 at 04:47:38PM -1000, Tejun Heo wrote:
> ...
> > > +#ifdef CONFIG_SCHED_CLASS_EXT
> > > + p->scx.dsq = NULL;
> > > + INIT_LIST_HEAD(&p->scx.dsq_node);
> > > + p->scx.flags = 0;
> > > + p->scx.weight = 0;
> > > + p->scx.sticky_cpu = -1;
> > > + p->scx.holding_cpu = -1;
> > > + p->scx.kf_mask = 0;
> > > + atomic64_set(&p->scx.ops_state, 0);
> >
> > We probably need atomic_long_set() here or in 32-bit arches (such as
> > armhf) we get this:
> >
> > kernel/sched/core.c:4564:22: error: passing argument 1 of ‘atomic64_set’ from incompatible pointer type [-Werror=incompatible-pointer-types]
> > 4564 | atomic64_set(&p->scx.ops_state, 0);
> > | ^~~~~~~~~~~~~~~~~
> > | |
> > | atomic_long_t * {aka atomic_t *}
> >
> ...
> > > +static void set_next_task_scx(struct rq *rq, struct task_struct *p, bool first)
> > > +{
> > > + if (p->scx.flags & SCX_TASK_QUEUED) {
> > > + WARN_ON_ONCE(atomic64_read(&p->scx.ops_state) != SCX_OPSS_NONE);
> >
> > Ditto. Even if this line is replaced later by
> > "[PATCH 31/36] sched_ext: Implement core-sched support"
> >
> > > + dispatch_dequeue(&rq->scx, p);
> > > + }
> > > +
> > > + p->se.exec_start = rq_clock_task(rq);
> > > +}
>
> Sorry about that. I updated them and will include the changes in the next
> iteration. Will test 32bit build too next time.
>
> Thanks.

No problem, if you don't I'll test it for you. :)

-Andrea