On Thu, Sep 10, 2020 at 9:59 AM Krzysztof Kozlowski <krzk@xxxxxxxxxx> wrote:
> On Thu, 10 Sep 2020 at 08:52, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote:
> > On Thursday, September 10, 2020, Peter Rosin <peda@xxxxxxxxxx> wrote:
> >> On 2020-09-09 21:57, Krzysztof Kozlowski wrote:
> >> > On Wed, 9 Sep 2020 at 20:36, Jonathan Cameron <jic23@xxxxxxxxxx> wrote:
> >> >> On Sat, 29 Aug 2020 08:47:16 +0200
> >> >> Krzysztof Kozlowski <krzk@xxxxxxxxxx> wrote:
...
> >> >>> @@ -596,12 +594,9 @@ static int stm32_adc_core_switches_probe(struct device *dev,
> >> >>> priv->booster = devm_regulator_get_optional(dev, "booster");
> >> >>> if (IS_ERR(priv->booster)) {
> >> >>> ret = PTR_ERR(priv->booster);
> >> >>> - if (ret != -ENODEV) {
> >> >>> - if (ret != -EPROBE_DEFER)
> >> >>> - dev_err(dev, "can't get booster %d\n",
> >> >>> - ret);
> >> >>> - return ret;
> >> >>> - }
> >> >>> + if (ret != -ENODEV)
> >> >>> + dev_err_probe(dev, ret, "can't get booster\n");
> >> >>
> >> >> This tripped a warning and got the patch dropped because we no longer
> >> >> return on error.
> >>
> >> As Jonathan already said, we no longer return in this hunk. I.e., you have
> >> clobbered the error path.
> >
> >
> > Exactly my point why I proposed _must_check in the first place.
>
> That was not exactly that point as you did not mention possible errors
> but only "miss the opportunity to optimize". Optimization is different
> things than a mistake.
Yes, and that's what happened here. You missed optimization which led
to an error.
And this is a good showcase to see how dev_err_probe() may be misused
because of overlooking subtle details.
Perhaps we can do
static inline __must_check dev_err_probe_ret(...)
{
return dev_err_probe(...);
}
(or other way around, introduce dev_err_probe_noret(), yes, name sucks)
--
With Best Regards,
Andy Shevchenko
