Hi,
On Mon, Mar 28, 2022 at 12:11:49PM -0700, Linus Torvalds wrote:
> On Mon, Mar 28, 2022 at 6:08 AM Linus Walleij <linus.walleij@xxxxxxxxxx> wrote:
> >
> > git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl.git
> > tags/pinctrl-v5.18-1
>
> Hmm.
>
> This clashes badly with the fact that we in the meantime have enabled
> -Warray-bounds, and I got
>
> drivers/pinctrl/nuvoton/pinctrl-npcm7xx.c: In function ‘npcmgpio_irq_handler’:
> ./include/linux/find.h:40:23: error: array subscript ‘long unsigned
> int[0]’ is partly outside array bounds of ‘u32[1]’ {aka ‘unsigned
> int[1]’} [-Werror=array-bounds]
> 40 | val = *addr & GENMASK(size - 1, offset);
> | ^~~~~
> drivers/pinctrl/nuvoton/pinctrl-npcm7xx.c:219:13: note: while
> referencing ‘sts’
> 219 | u32 sts, en, bit;
> | ^~~
>
> as a result.
>
> Was this not in linux-next?
It was — but when it was noticed, the fix went through the IRQ tree, on
top of a refactoring that happened there (the switch to generic_handle_domain_irq):
https://lore.kernel.org/lkml/164751044707.389.16417510835118111853.tip-bot2@tip-bot2/
So… the issue should resolve itself when the IRQ tree is pulled.
> Or was the array bounds checking not there?
>
> Anyway, that cast to "const void *"
>
> for_each_set_bit(bit, (const void *)&sts,
>
> in that driver is completely wrong.
>
> The bit operations are defined in arrays of 'unsigned long', and you
> can't just cast the issue away, because the end result is not the same
> on a big-endian machine.
>
> I fixed it up in the merge, but what really confuses me (apart from
> the apparent lack of testing in linux-next) is that I don't actually
> see what made this happen now, and not before. Maybe that's why it
> didn't show up in linux-next: it's some odd gcc heisenbug.
>
> Because there seems to be no actual changes to that driver that would
> explain why I get the warning now, but not before the pull.
When I added the pinctrl-wpcm450 driver, I changed the pinctrl/nuvoton/
directory to obj-y and exposed the pinctrl-npcm7xx to CI bot testing.
The bug existed, untouched, since the driver was added a few years ago.
Jonathan
Attachment:
signature.asc
Description: PGP signature
