Re: [PATCH] gpiolib: annotate struct acpi_gpio_mapping with __counted_by_ptr

["Ard Biesheuvel" <ardb@kernel.org>]

On Thu, 1 Oct 2026, at 23:24, Linus Walleij wrote:
> On Thu, Oct 1, 2026 at 10:05 PM Bill Wendling <morbo@google.com> wrote:
>> On Thu, Oct 1, 2026 at 12:56 PM Linus Walleij <linusw@kernel.org> wrote:
>> > On Mon, Sep 28, 2026 at 9:22 AM Bill Wendling <morbo@google.com> wrote:
>> > > On Sun, Sep 27, 2026 at 11:38 PM Bill Wendling <morbo@google.com> wrote:
>> > > >
>> > > > The 'data' pointer field in 'struct acpi_gpio_mapping' is associated
>> > > > with the 'size' field, which represents the number of elements of
>> > > > type 'struct acpi_gpio_params' allocated for 'data'.
>> > > >
>> > > > To improve bounds checking via CONFIG_UBSAN_BOUNDS and
>> > > > CONFIG_FORTIFY_SOURCE, annotate 'data' with the __counted_by_ptr
>> > > > attribute.
>> > > >
>> > > > Analysis of allocation, assignment, and access points shows that the
>> > > > pointer is never accessed before the count is set, which guarantees that
>> > > > this annotation is safe and will not cause runtime panics or
>> > > > false-positive bounds checks.
>> > > >
>> > > > Cc: codemender-patching+linux@google.com
>> > > > Assisted-by: LLM
>> > > > Signed-off-by: Bill Wendling <morbo@google.com>
>> > > > ---
>> > > >  include/linux/gpio/consumer.h | 2 +-
>> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
>> > > >
>> > > > diff --git a/include/linux/gpio/consumer.h b/include/linux/gpio/consumer.h
>> > > > index fceeefd5f893..2b80cf7aa7e7 100644
>> > > > --- a/include/linux/gpio/consumer.h
>> > > > +++ b/include/linux/gpio/consumer.h
>> > > > @@ -667,7 +667,7 @@ struct acpi_gpio_params {
>> > > >
>> > > >  struct acpi_gpio_mapping {
>> > > >         const char *name;
>> > > > -       const struct acpi_gpio_params *data;
>> > > > +       const struct acpi_gpio_params *data __counted_by_ptr(size);
>> > > >         unsigned int size;
>> > > >
>> > > >  /* Ignore IoRestriction field */
>> > >
>> > > There's a problem with 'drivers/firmware/efi/libstub/Makefile'. Clang
>> > > needs a compiler flag to support the "__counted_by_ptr" attribute
>> > > referencing a field *after* the pointer, like in this patch. However,
>> > > the Makefile blasts the flag away for x86 platforms. Below is a hack
>> > > that copies the part of the top-level Makefile that adds the flag. I
>> > > don't think that's a good solution. The comment in the driver's
>> > > Makefile says that the stub code executes before the kernel does,
>> > > which I assume is why a lot of the flags are blown away... In any
>> > > event, I'm not sure how best to address this.
>> >
>> > But is this a problem with the current patch?
>> >
>> > Does libefistub use <linux/gpio/consumer.h> in any way, shape
>> > or form?
>> >
>> It's being #included transitively:
>>
>> In file included from drivers/firmware/efi/libstub/efi-stub-helper.c:12:
>> In file included from ./include/linux/efi.h:20:
>> In file included from ./include/linux/rtc.h:18:
>> In file included from ./include/linux/nvmem-provider.h:16:
>> ./include/linux/gpio/consumer.h:670:55: error: use of undeclared
>> identifier 'size'; did you
>>       mean 'ksize'?
>>   670 |         const struct acpi_gpio_params *data __counted_by_ptr(size);
>>       |                                                              ^~~~
>>       |                                                              ksize
>> ././include/linux/compiler_types.h:392:64: note: expanded from macro
>> '__counted_by_ptr'
>>   392 | #define __counted_by_ptr(member)
>> __attribute__((__counted_by__(member)))
>>       |
>>        ^~~~~~
>> ./include/linux/slab.h:602:8: note: 'ksize' declared here
>>   602 | size_t ksize(const void *objp);
>>       |        ^
>
> Hm I see.
>
> Certainly Jeremy or Ard will have an idea about how to solve this,
> so paging them in.
>

libstub code never executes in the context of the kernel, but only in
the context of the boot firmware. Generally, we disable instrumentation
there that has a significant runtime component, basically because we
cannot crash or panic the kernel before we have even booted it.

Can we just #define __counted_by_ptr(...) to nothing when building
from that Makefile?