Re: [PATCH v2 1/4] driver core: Ignore 0 in dev_err_probe()

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Mon, Sep 02, 2024 at 02:10:41PM +0300, Dan Carpenter wrote:
> On Mon, Sep 02, 2024 at 01:16:17PM +0300, Andy Shevchenko wrote:
> > On Sat, Aug 31, 2024 at 11:53:51AM +0300, Dan Carpenter wrote:
> > > On Sat, Aug 31, 2024 at 11:25:54AM +0300, Dan Carpenter wrote:
> > > > On Thu, Aug 22, 2024 at 04:05:38PM +0300, Andy Shevchenko wrote:
> > > > > In the similar way, ignore 0 error code (AKA "success") in
> > > > > dev_err_probe(). This helps to simplify a code such as
> > > > > 
> > > > >   if (ret < 0)
> > > > >     return dev_err_probe(int3472->dev, ret, err_msg);
> > > > > 
> > > > >   return ret;
> > > > > 
> > > > > to
> > > > > 
> > > > >   return dev_err_probe(int3472->dev, ret, err_msg);
> > > > > 
> > > > > Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
> > > > > Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> > > > 
> > > > This is a terrible idea because currently Smatch is able to detect about one
> > > > bug per month where someone unintentionally passes the wrong error variable
> > > > to dev_err_probe().
> > 
> > How many cases you know where smatch is false positive about this?
> 
> This check has a very low false positive rate.

Yes, that's my expectation as well.

> There is only one potential
> false positive in the current linux-next.  Let me add Baolin Wang to get his
> take on this.  I never mentioned reported this warning because the code was old
> when I wrote the check.
> 
> drivers/spi/spi-sprd-adi.c
>    550          ret = of_hwspin_lock_get_id(np, 0);
>    551          if (ret > 0 || (IS_ENABLED(CONFIG_HWSPINLOCK) && ret == 0)) {
> 
> Is it possible for the CONFIG_ to not be enabled but ret is zero?
> 
>    552                  sadi->hwlock =
>    553                          devm_hwspin_lock_request_specific(&pdev->dev, ret);
>    554                  if (!sadi->hwlock) {
>    555                          ret = -ENXIO;
>    556                          goto put_ctlr;
>    557                  }
>    558          } else {
>    559                  switch (ret) {
>    560                  case -ENOENT:
>    561                          dev_info(&pdev->dev, "no hardware spinlock supplied\n");
>    562                          break;
>    563                  default:
>    564                          dev_err_probe(&pdev->dev, ret, "failed to find hwlock id\n");
>                                                           ^^^
> 
>    565                          goto put_ctlr;
>    566                  }
>    567          }
> 
> 
> > I believe the number is only a few at most, which means that you may easily
> > detect this still with this change being applied, i.e. "anything that
> > terminates function flow with code 0, passed to dev_err_probe(), is
> > suspicious".
> 
> I think you mean the opposite of what you wrote?  That if we're passing zero to
> dev_err_probe() and it's the last line in a function it's *NOT* suspicious?

Yes, sorry, I meant "...terminates function flow _in the middle_...".

> Otherwise, I don't really understand the heuristic you're proposing.

-- 
With Best Regards,
Andy Shevchenko