On Wed, Jul 18, 2018 at 11:14 AM Simon Horman <horms@xxxxxxxxxxxx> wrote:
> On Tue, Jul 17, 2018 at 02:57:27PM +0000, Phil Edworthy wrote:
> > On 17 July 2018 15:47, Andy Shevchenko wrote:
> > > On Tue, 2018-07-17 at 14:40 +0000, Phil Edworthy wrote:
> > > > On 17 July 2018 15:19, Andy Shevchenko wrote:
> > > > > On Tue, 2018-07-17 at 12:42 +0000, Phil Edworthy wrote:
> > > > >
> > > > > > > While your point sounds valid (don't remember how clk_get() is
> > > > > > > implemented), NULL is also OK to have.
> > > > > >
> > > > > > Ok as in there is no bus clock, right?
> > > > > > So it should be:
> > > > > > if (!IS_ERR_OR_NULL (dev->busclk))
> > > > >
> > > > > Nope, NULL is no error case for optional clock.
> > > >
> > > > I must be missing something here...
> > >
> > > See how clk_prepare_enable() is implemented.
> > Ok, if busclk is NULL the code can safely call clk_prepare_enable()
> >
> > > > I agree that NULL for an optional clock is not an error. However, the
> > > > code above is now:
> > > > + if (prepare) {
> > > > + /* Optional bus clock */
> > >
> > > > + if (!IS_ERR_OR_NULL(dev->busclk)) {
> > >
> > > Check for NULL is redundant.
> > >
> > > > + ret = clk_prepare_enable(dev->busclk);
> > > > + if (ret)
> > > > + return ret;
> > > > + }
> > > > +
> > > > return clk_prepare_enable(dev->clk);
> > > > + }
> > > >
> > > > So, if you have a valid busclk, it gets enabled, otherwise it is left
> > > > alone.
> > >
> > So the code as sent in the original email is correct (aside from Geert's
> > comments about EPROBE_DEFER handling).
> >
> > Maybe I need some coffee :\
> > Thanks
> > Phil
>
> My point is that errors should be treated as errors.
>
> In i2c_dw_prepare_clk() the following appears:
>
> if (IS_ERR(dev->clk))
> return PTR_ERR(dev->clk);
>
> So dev->clk being an error value is treated as an error that
> is passed up to the caller.
>
> But in your patch (and the snippet below) dev->busclk is treated
> as the optional clock not being present. Even if the error stored
> nothing to do with the clock not being present - f.e. ENOMEM or
> as Geert mentioned elsewhere, EPROBE_DEFER.
>
> Assuming the absense of the optional clock is indicated by ENOENT,
> in my view correct code would include something like:
>
> ...
>
> if (IS_ERR(dev->clk))
> return PTR_ERR(dev->clk);
>
> if (IS_ERR(dev->buslck) && PTR_ERR(dev->busclk) != -ENOENT)
> return PTR_ERR(dev->busclk);
>
> ...
As this is a commonly-used construct, perhaps it would be good to introduce
clk_get_optional() (cfr. gpiod_get_optional()) and devm_clk_get_optional(),
which would return NULL instead of -ENOENT?
Then it becomes a simple check for IS_ERR() in the driver.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
