Re: [PATCH 04/10] workqueue: Move nr_active handling into helpers

[Tejun Heo]

On Tue, Dec 26, 2023 at 05:12:55PM +0800, Lai Jiangshan wrote:
> On Wed, Dec 20, 2023 at 3:25 PM Tejun Heo <tj@xxxxxxxxxx> wrote:
> 
> > +static bool pwq_tryinc_nr_active(struct pool_workqueue *pwq)
> >  {
> > -       struct work_struct *work = list_first_entry(&pwq->inactive_works,
> > -                                                   struct work_struct, entry);
> > +       struct workqueue_struct *wq = pwq->wq;
> > +       struct worker_pool *pool = pwq->pool;
> > +       bool obtained;
> >
> > -       pwq_activate_work(pwq, work);
> > +       lockdep_assert_held(&pool->lock);
> > +
> > +       obtained = pwq->nr_active < wq->max_active;
> 
> It is better to use READ_ONCE(wq->max_active) here in case
> the compiler issues code to calculate "obtained" multi-time.
> 
> Theoretically, READ_ONCE(wq->max_active) is recommended
> from the patch1 in pwq_dec_nr_in_flight() and __queue_work()
> since there is no wq->mutex for them.

Yeah, good point. Lemme add WRITE/READ_ONCE() around max_active accesses.

Thanks.

-- 
tejun