Re: [PATCH v2 2/5] media: i2c: ov5645: Use runtime PM

[Sakari Ailus]

Hi Prabhakar,

On Thu, Oct 27, 2022 at 06:37:58PM +0000, Sakari Ailus wrote:
> Hi Prabhakar,
> 
> On Thu, Oct 27, 2022 at 05:32:07PM +0100, Lad, Prabhakar wrote:
> > Hi Sakari,
> > 
> > On Thu, Oct 27, 2022 at 1:47 PM Sakari Ailus
> > <sakari.ailus@xxxxxxxxxxxxxxx> wrote:
> > >
> > > Hi Prabhakar,
> > >
> > > On Thu, Oct 27, 2022 at 01:01:52PM +0100, Lad, Prabhakar wrote:
> > > > Hi Sakari,
> > > >
> > > > On Thu, Oct 27, 2022 at 12:20 PM Sakari Ailus
> > > > <sakari.ailus@xxxxxxxxxxxxxxx> wrote:
> > > > >
> > > > > Hi Prabhakar,
> > > > >
> > > > > One more comment.
> > > > >
> > > > > On Fri, Oct 14, 2022 at 07:34:56PM +0100, Prabhakar wrote:
> > > > > > @@ -1209,12 +1190,16 @@ static int ov5645_probe(struct i2c_client *client)
> > > > > >
> > > > > >       dev_info(dev, "OV5645 detected at address 0x%02x\n", client->addr);
> > > > > >
> > > > > > +     pm_runtime_set_active(dev);
> > > > > > +     pm_runtime_get_noresume(dev);
> > > > > > +     pm_runtime_enable(dev);
> > > > >
> > > > > You won't gain anything by eanbling runtime PM here. Just move it to the
> > > > > end of the function before the rest of the calls. Error handling becomes
> > > > > more simple.
> > > > >
> > > > If I move the above calls below I get the below warning:
> > > >
> > > > [    2.633386] ov5645 0-003c: Runtime PM usage count underflow!
> > > >
> > > > This is because of the last patch which moves ov5645_entity_init_cfg()
> > > > before registering the subdev. ov5645_entity_init_cfg() calls s_ctrl
> > > > due to which we are seeing the above message. Please let me know how
> > > > to proceed on this.
> > >
> > > Ah. Yes, this is a problem with the usage pattern of
> > > pm_runtime_get_if_in_use(). But please don't change that.
> > >
> > > You can still move enabling runtime PM later in the function.
> > >
> > Agreed, the final version looks like below:
> > 
> >     pm_runtime_set_active(dev);
> >     pm_runtime_get_noresume(dev);
> > 
> 
> You'll have to enable runtime PM here, before pm_runtime_get_if_in_use()
> gets called.
> 
> I'll see if it could be made to work in a sensible way when runtime PM
> isn't enabled yet.

There are various ways how runtime PM interface functions generally work,
and generally return an error when runtime PM is disabled. Incrementing the
usage_count when runtime PM is disabled would make
pm_runtime_get_if_in_use() very special and not match what the rest would
do. Therefore I think it's best to keep this in the driver. After all, mo
other driver needs this in the media tree, which is the major user of the
function.

-- 
Kind regards,

Sakari Ailus