Re: [PATCH 01/14] gpio: pca953x: move suspend/resume to suspend_noirq/resume_noirq

[Bartosz Golaszewski <brgl@xxxxxxxx>]

On Thu, Feb 8, 2024 at 5:19 PM Thomas Richard
<thomas.richard@xxxxxxxxxxx> wrote:
>
> On 1/28/24 01:12, Linus Walleij wrote:
> > On Fri, Jan 19, 2024 at 6:01 PM Thomas Richard
> > <thomas.richard@xxxxxxxxxxx> wrote:
> >> On 1/16/24 08:43, Tony Lindgren wrote:
> >>> * Thomas Richard <thomas.richard@xxxxxxxxxxx> [240115 16:16]:
> >>>> Some IOs can be needed during suspend_noirq/resume_noirq.
> >>>> So move suspend/resume callbacks to noirq.
> >>>
> >>> So have you checked that the pca953x_save_context() and restore works
> >>> this way? There's i2c traffic and regulators may sleep.. I wonder if
> >>> you instead just need to leave gpio-pca953x enabled in some cases
> >>> instead?
> >>>
> >>
> >> Yes I tested it, and it works (with my setup).
> >> But this patch may have an impact for other people.
> >> How could I leave it enabled in some cases ?
> >
> > I guess you could define both pca953x_suspend() and
> > pca953x_suspend_noirq() and selectively bail out on one
> > path on some systems?
>
> Yes.
>
> What do you think if I use a property like for example "ti,pm-noirq" to
> select the right path ?
> Is a property relevant for this use case ?
>

I prefer a new property than calling of_machine_is_compatible().
Please do run it by the DT maintainers, I think it should be fine.
Maybe even don't limit it to TI but make it a generic property.

Bart

> Regards,
>
> >
> > Worst case using if (of_machine_is_compatible("my,machine"))...
> >
> > Yours,
> > Linus Walleij
> --
> Thomas Richard, Bootlin
> Embedded Linux and Kernel engineering
> https://bootlin.com
>