Re: [PATCH] iio: adc: qcom-pm8xxx-xoadc: Remove useless condition in pm8xxx_xoadc_parse_channel()

[Linus Walleij]

On Tue, Mar 14, 2023 at 10:12 PM Linus Walleij <linus.walleij@xxxxxxxxxx> wrote:
> On Tue, Mar 14, 2023 at 8:37 PM Kasumov Ruslan <xhxgldhlpfy@xxxxxxxxx> wrote:
>
> > The left side of the loop condition never becomes false.
> > hwchan cannot be NULL, because it points to elements of the
> > hw_channels array that takes one of 4 predefined values:
> > pm8018_xoadc_channels, pm8038_xoadc_channels,
> > pm8058_xoadc_channels, pm8921_xoadc_channels.
> >
> > Found by Linux Verification Center (linuxtesting.org) with SVACE.
>
> I am not impressed with that tool. See below:
>
> > Fixes: 63c3ecd946d4 ("iio: adc: add a driver for Qualcomm PM8xxx HK/XOADC")
> > Signed-off-by: Kasumov Ruslan <s02210418@xxxxxxxxxxxxx>
>
> (...)
> >         hwchan = &hw_channels[0];
> > -       while (hwchan && hwchan->datasheet_name) {
> > +       while (hwchan->datasheet_name) {
> >                 if (hwchan->pre_scale_mux == pre_scale_mux &&
> >                     hwchan->amux_channel == amux_channel)
> >                         break;
>
> NAK have you tested this on a real system?
>
> Here is the complete loop:
>
>         hwchan = &hw_channels[0];
>         while (hwchan && hwchan->datasheet_name) {
>                 if (hwchan->pre_scale_mux == pre_scale_mux &&
>                     hwchan->amux_channel == amux_channel)
>                         break;
>                 hwchan++;
>                 chid++;
>         }
>
> Notice how hwchan is used as iterator in hwchan++.
>
> What you are doing will cause a zero-pointer dereference.

Nah the AI is smarter than me this time, I'm wrong, I think :(

hwchan is indeed never NULL here, and the code immediately
after unconditionally dereferences hwchan->datasheet_name.

Who wrote this convoluted code again:
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  759)
 chid = 0;
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  760)
 hwchan = &hw_channels[0];
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  761)
 while (hwchan && hwchan->datasheet_name) {
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  762)
         if (hwchan->pre_scale_mux == pre_scale_mux &&
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  763)
             hwchan->amux_channel == amux_channel)
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  764)
                 break;
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  765)
         hwchan++;
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  766)
         chid++;
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  767)         }
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  768)
 /* The sentinel does not have a name assigned */
63c3ecd946d4a (Linus Walleij    2017-04-04 14:08:19 +0200  769)
 if (!hwchan->datasheet_name) {

Oh that guy ...

I wonder if we can make it look better and less unintuitive.

Yours,
Linus Walleij