Re: [PATCH 06/15 v3] regulator: max8973: Let core handle GPIO descriptor
From: Linus Walleij
Date: Thu Dec 06 2018 - 06:48:03 EST
On Thu, Dec 6, 2018 at 11:34 AM Charles Keepax
<ckeepax@xxxxxxxxxxxxxxxxxxxxx> wrote:
> On Thu, Dec 06, 2018 at 09:58:30AM +0100, Linus Walleij wrote:
> > On Wed, Dec 5, 2018 at 4:33 PM Charles Keepax
> > <ckeepax@xxxxxxxxxxxxxxxxxxxxx> wrote:
> > > On Wed, Dec 05, 2018 at 03:42:06PM +0100, Linus Walleij wrote:
> > > > On Wed, Dec 5, 2018 at 2:43 PM Charles Keepax
> > > > <ckeepax@xxxxxxxxxxxxxxxxxxxxx> wrote:
> > > > > On Wed, Dec 05, 2018 at 01:47:12PM +0100, Linus Walleij wrote:
> > > > > > @@ -775,10 +779,13 @@ static int max8973_probe(struct i2c_client *client,
> > > > > > /*
> > > > > > * We do not let the core switch this regulator on/off,
> > > > > > * we just leave it on.
> > > > > > + *
> > > > > > + * Do not use devm* here: the regulator core takes over the
> > > > > > + * lifecycle management of the GPIO descriptor.
> > > > > > */
> > > > > > - gpiod = devm_gpiod_get_optional(&client->dev,
> > > > > > - "maxim,enable",
> > > > > > - GPIOD_OUT_HIGH);
> > > > > > + gpiod = gpiod_get_optional(&client->dev,
> > > > > > + "maxim,enable",
> > > > > > + GPIOD_OUT_HIGH);
> > > > >
> > > > > My comment on v2 still stands here, the GPIO is not passed to
> > > > > the regulator core on this path.
> > > >
> > > > Patch 01 should take care of that, did you check it?
> > >
> > > Yeah, patch 1 makes the regulator core consistently handle GPIOs
> > > that are passed into it.
> >
> > Sorry. I confused this patch for the max77686 patch. That
> > one was fixed with patch 01...
> >
> > > However, on the MAX77621 path in this
> > > switch statement the GPIO is never passed into the regulator
> > > core, so the core never takes ownership of it, so the driver still
> > > needs to manage the lifetime.
> > >
> > > It should be a pretty easy fix though as commented on v2, again
> > > apologies for the slow review.
> >
> > OK I switch it to devm_ as you suggested, as we implemented
> > gpiod_unhinge it's a piece of cake nowadays.
> >
>
> You shouldn't really need to use unhinge, you can just use devm
> on the path that doesn't pass it to the core and not on the
> one that does. You just need to update the error case below it to
> use config->ena_gpiod rather than gpiod.
Indeed I just think it will be confusing when people read the code.
It's better consistency if its just devm_* and the we unhinge the
one we pass to the regulator core IMO.
Yours,
Linus Walleij