Re: [PATCH -next v2 1/2] pinctrl: ocelot: add missing destroy_workqueue() in error path in ocelot_pinctrl_probe()

[Yang Yingliang]

Hi,

On 2022/9/17 16:54, andy.shevchenko@xxxxxxxxx wrote:
> Sat, Sep 17, 2022 at 10:46:33AM +0800, Yang Yingliang kirjoitti:
> > Add the missing destroy_workqueue() before return from ocelot_pinctrl_probe()
> > in error path.
> >
> > Fixes: c297561bc98a ("pinctrl: ocelot: Fix interrupt controller")
> > Signed-off-by: Yang Yingliang <yangyingliang@xxxxxxxxxx>
> > ---
> > v2:
> >    move alloc_ordered_workqueue() after ocelot_pinctrl_register().
> Why? What will happen if user space start using pins before workqueue is allocated?
This is a suggestion from Horatiu Vultur.

And I think the workqueue is used in ocelot_irq_unmask_level(), but 
before ocelot_gpiochip_register() calling,
the irq is not prepared, the work won't be queued before 
ocelot_gpiochip_register().
> If you really want to have it correct, you need either drop all devm_ calls
> after allocating workqueue, or wrap destroying into devm.
I am trying to add devm APIs for allocating workqueue.

Thanks,
Yang
>