[PATCH: 1/1] sh4: avoid spurious gcc warning

[PATCH: 1/1] sh4: avoid spurious gcc warning

[Michael.Karcher]

Prevent sizeof-pointer-div warning in SH4 intc macros

Gcc warns about the pattern sizeof(void*)/sizeof(void), as it looks like
the abuse of a pattern to calculate the array size. This pattern appears
in the unevaluated part of the ternary operator in _INTC_ARRAY if the
parameter is NULL.

The replacement uses an alternate approach to return 0 in case of NULL
which does not generate the pattern sizeof(void*)/sizeof(void), but still
emits the warning if _INTC_ARRAY is called with a nonarray parameter.

This patch is required for successful compilation with -Werror enabled.

The idea to use _Generic for type distinction is taken from Comment #7
in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108483 by Jakub Jelinek

Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
---

diff --git a/include/linux/sh_intc.h b/include/linux/sh_intc.h
index c255273b0281..d7a7ffb60a34 100644
--- a/include/linux/sh_intc.h
+++ b/include/linux/sh_intc.h
@@ -97,7 +97,7 @@ struct intc_hw_desc {
      unsigned int nr_subgroups;
  };

-#define _INTC_ARRAY(a) a, __same_type(a, NULL) ? 0 : sizeof(a)/sizeof(*a)
+#define _INTC_ARRAY(a) a, sizeof(a)/(_Generic((a), typeof(NULL): 
0xFFFFFFFFU, default: sizeof(*a)))

  #define INTC_HW_DESC(vectors, groups, mask_regs,    \
               prio_regs,    sense_regs, ack_regs)    \

Re: [PATCH: 1/1] sh4: avoid spurious gcc warning

[Randy Dunlap]

Hi--

It's just sh: AFAICT. The patch fixes the build error for me on sh2,
using gcc 12.1.0 from the kernel.org crosstool builds.

On 1/21/23 16:15, Michael.Karcher wrote:
> Prevent sizeof-pointer-div warning in SH4 intc macros
> 
> Gcc warns about the pattern sizeof(void*)/sizeof(void), as it looks like
> the abuse of a pattern to calculate the array size. This pattern appears
> in the unevaluated part of the ternary operator in _INTC_ARRAY if the
> parameter is NULL.
> 
> The replacement uses an alternate approach to return 0 in case of NULL
> which does not generate the pattern sizeof(void*)/sizeof(void), but still
> emits the warning if _INTC_ARRAY is called with a nonarray parameter.
> 
> This patch is required for successful compilation with -Werror enabled.
> 
> The idea to use _Generic for type distinction is taken from Comment #7
> in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108483 by Jakub Jelinek
> 
> Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
> ---
> 
> diff --git a/include/linux/sh_intc.h b/include/linux/sh_intc.h
> index c255273b0281..d7a7ffb60a34 100644
> --- a/include/linux/sh_intc.h
> +++ b/include/linux/sh_intc.h
> @@ -97,7 +97,7 @@ struct intc_hw_desc {
>      unsigned int nr_subgroups;
>  };
> 
> -#define _INTC_ARRAY(a) a, __same_type(a, NULL) ? 0 : sizeof(a)/sizeof(*a)
> +#define _INTC_ARRAY(a) a, sizeof(a)/(_Generic((a), typeof(NULL): 0xFFFFFFFFU, default: sizeof(*a)))

s/: / : / in 2 places.

Tested-by: Randy Dunlap <rdunlap@infradead.org> # build-tested

How far back in gcc versions does this work?

Thanks.

> 
>  #define INTC_HW_DESC(vectors, groups, mask_regs,    \
>               prio_regs,    sense_regs, ack_regs)    \
> 

-- 
~Randy

Re: [PATCH: 1/1] sh4: avoid spurious gcc warning

[Michael Karcher]

Am 22.01.2023 um 08:00 schrieb Randy Dunlap:
>> -#define _INTC_ARRAY(a) a, __same_type(a, NULL) ? 0 : sizeof(a)/sizeof(*a)
>> +#define _INTC_ARRAY(a) a, sizeof(a)/(_Generic((a), typeof(NULL): 0xFFFFFFFFU, default: sizeof(*a)))
> s/: / : / in 2 places.
>
> Tested-by: Randy Dunlap <rdunlap@infradead.org> # build-tested

Thanks for your confirmation! Are you sure about the space before the 
colon? The colon in this case terminates a case descriptor for the 
type-level switch construction using "_Generic". It says: "In case 'a' 
has the 'type of NULL', divide by 0xFFFFFFFFU, in all other cases, 
divide by the size of a single array element". It's not a colon of the 
ternary ?: operator, in which case I would agree with the space before it.

If you confirm that you want a space before the colon in this case as 
well, I'm going to add it, though.

> How far back in gcc versions does this work?

I tested the support of _Generic on Compiler Explorer at godbolt.org. 
This construction is rejected by gcc 4.8, but accepted by gcc 4.9.

Kind regards,
   Michael Karcher

Re: [PATCH: 1/1] sh4: avoid spurious gcc warning

[John Paul Adrian Glaubitz]

Hi Michael!

On 1/22/23 01:15, Michael.Karcher wrote:
> Prevent sizeof-pointer-div warning in SH4 intc macros
> 
> Gcc warns about the pattern sizeof(void*)/sizeof(void), as it looks like
> the abuse of a pattern to calculate the array size. This pattern appears
> in the unevaluated part of the ternary operator in _INTC_ARRAY if the
> parameter is NULL.
> 
> The replacement uses an alternate approach to return 0 in case of NULL
> which does not generate the pattern sizeof(void*)/sizeof(void), but still
> emits the warning if _INTC_ARRAY is called with a nonarray parameter.
> 
> This patch is required for successful compilation with -Werror enabled.
> 
> The idea to use _Generic for type distinction is taken from Comment #7
> in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108483 by Jakub Jelinek
> 
> Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
> ---
> 
> diff --git a/include/linux/sh_intc.h b/include/linux/sh_intc.h
> index c255273b0281..d7a7ffb60a34 100644
> --- a/include/linux/sh_intc.h
> +++ b/include/linux/sh_intc.h
> @@ -97,7 +97,7 @@ struct intc_hw_desc {
>       unsigned int nr_subgroups;
>   };
> 
> -#define _INTC_ARRAY(a) a, __same_type(a, NULL) ? 0 : sizeof(a)/sizeof(*a)
> +#define _INTC_ARRAY(a) a, sizeof(a)/(_Generic((a), typeof(NULL): 0xFFFFFFFFU, default: sizeof(*a)))
> 
>   #define INTC_HW_DESC(vectors, groups, mask_regs,    \
>                prio_regs,    sense_regs, ack_regs)    \

The title should probably be "arch/sh: avoid spurious gcc warning" since it's not
a problem special to sh4 but affects the whole arch/sh sub-folder which covers
all SuperH and J-Core targets.

Can you rephrase the title accordingly?

Adrian

-- 
  .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
   `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

Re: [PATCH: 1/1] sh4: avoid spurious gcc warning

[Jakub Jelinek]

On Sun, Jan 22, 2023 at 12:33:41PM +0100, Michael Karcher wrote:
> Am 22.01.2023 um 08:00 schrieb Randy Dunlap:
> > > -#define _INTC_ARRAY(a) a, __same_type(a, NULL) ? 0 : sizeof(a)/sizeof(*a)
> > > +#define _INTC_ARRAY(a) a, sizeof(a)/(_Generic((a), typeof(NULL): 0xFFFFFFFFU, default: sizeof(*a)))
> > s/: / : / in 2 places.
> > 
> > Tested-by: Randy Dunlap <rdunlap@infradead.org> # build-tested
> 
> Thanks for your confirmation! Are you sure about the space before the colon?

No, it should be without those, see various other _Generic uses in
include/linux/
All those are formatted on one line for each case, so for the above macro it
would be
#define _INTC_ARRAY(a) (a), sizeof(a)/(_Generic((a),			\
				       typeof(NULL):	-1,		\
				       default:		sizeof(*(a)))
or so.
Anyway, two comments:
1) I'd use -1 as that would be after promotion to size_t the largest size_t
   unlike 0xFFFFFFFFU; of course, as for the void * case a can't be an array,
   any value > sizeof(void*) will do
2) if *a and a is fine (i.e. argument of the macro has to be really simple or
   wrapped in ()s, then perhaps (a) as first operand to _Generic isn't needed
   either, or use (a) in the two spots (sizeof(a) is of course fine) and
   *(a)

> The colon in this case terminates a case descriptor for the type-level
> switch construction using "_Generic". It says: "In case 'a' has the 'type of
> NULL', divide by 0xFFFFFFFFU, in all other cases, divide by the size of a
> single array element". It's not a colon of the ternary ?: operator, in which
> case I would agree with the space before it.
> 
> If you confirm that you want a space before the colon in this case as well,
> I'm going to add it, though.
> 
> > How far back in gcc versions does this work?
> 
> I tested the support of _Generic on Compiler Explorer at godbolt.org. This
> construction is rejected by gcc 4.8, but accepted by gcc 4.9.

Yeah, introduced in gcc 4.9, as I think kernel minimum version is 5.1, that is fine.
And various headers already use _Generic.

	Jakub

Re: [PATCH: 1/1] sh4: avoid spurious gcc warning

[Randy Dunlap]

On 1/22/23 04:42, Jakub Jelinek wrote:
> On Sun, Jan 22, 2023 at 12:33:41PM +0100, Michael Karcher wrote:
>> Am 22.01.2023 um 08:00 schrieb Randy Dunlap:
>>>> -#define _INTC_ARRAY(a) a, __same_type(a, NULL) ? 0 : sizeof(a)/sizeof(*a)
>>>> +#define _INTC_ARRAY(a) a, sizeof(a)/(_Generic((a), typeof(NULL): 0xFFFFFFFFU, default: sizeof(*a)))
>>> s/: / : / in 2 places.
>>>
>>> Tested-by: Randy Dunlap <rdunlap@infradead.org> # build-tested
>>
>> Thanks for your confirmation! Are you sure about the space before the colon?

Nope, my bad. Thanks, Jakub.

> No, it should be without those, see various other _Generic uses in
> include/linux/
> All those are formatted on one line for each case, so for the above macro it
> would be
> #define _INTC_ARRAY(a) (a), sizeof(a)/(_Generic((a),			\
> 				       typeof(NULL):	-1,		\
> 				       default:		sizeof(*(a)))
> or so.
> Anyway, two comments:
> 1) I'd use -1 as that would be after promotion to size_t the largest size_t
>    unlike 0xFFFFFFFFU; of course, as for the void * case a can't be an array,
>    any value > sizeof(void*) will do
> 2) if *a and a is fine (i.e. argument of the macro has to be really simple or
>    wrapped in ()s, then perhaps (a) as first operand to _Generic isn't needed
>    either, or use (a) in the two spots (sizeof(a) is of course fine) and
>    *(a)
> 
>> The colon in this case terminates a case descriptor for the type-level
>> switch construction using "_Generic". It says: "In case 'a' has the 'type of
>> NULL', divide by 0xFFFFFFFFU, in all other cases, divide by the size of a
>> single array element". It's not a colon of the ternary ?: operator, in which
>> case I would agree with the space before it.
>>
>> If you confirm that you want a space before the colon in this case as well,
>> I'm going to add it, though.
>>
>>> How far back in gcc versions does this work?
>>
>> I tested the support of _Generic on Compiler Explorer at godbolt.org. This
>> construction is rejected by gcc 4.8, but accepted by gcc 4.9.
> 
> Yeah, introduced in gcc 4.9, as I think kernel minimum version is 5.1, that is fine.
> And various headers already use _Generic.

and thanks for that info also.

-- 
~Randy

Re: [PATCH: 1/1] sh4: avoid spurious gcc warning

[Geert Uytterhoeven]

Hi Jakub,

On Sun, Jan 22, 2023 at 1:47 PM Jakub Jelinek <jakub@redhat.com> wrote:
> On Sun, Jan 22, 2023 at 12:33:41PM +0100, Michael Karcher wrote:
> > Am 22.01.2023 um 08:00 schrieb Randy Dunlap:
> > > > -#define _INTC_ARRAY(a) a, __same_type(a, NULL) ? 0 : sizeof(a)/sizeof(*a)
> > > > +#define _INTC_ARRAY(a) a, sizeof(a)/(_Generic((a), typeof(NULL): 0xFFFFFFFFU, default: sizeof(*a)))
> > > s/: / : / in 2 places.
> > >
> > > Tested-by: Randy Dunlap <rdunlap@infradead.org> # build-tested
> >
> > Thanks for your confirmation! Are you sure about the space before the colon?
>
> No, it should be without those, see various other _Generic uses in
> include/linux/
> All those are formatted on one line for each case, so for the above macro it
> would be
> #define _INTC_ARRAY(a) (a), sizeof(a)/(_Generic((a),                    \
>                                        typeof(NULL):    -1,             \
>                                        default:         sizeof(*(a)))
> or so.
> Anyway, two comments:
> 1) I'd use -1 as that would be after promotion to size_t the largest size_t
>    unlike 0xFFFFFFFFU; of course, as for the void * case a can't be an array,
>    any value > sizeof(void*) will do

Or SIZE_MAX.

include/linux/limits.h:#define SIZE_MAX (~(size_t)0)

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH: 1/1] sh4: avoid spurious gcc warning

[Jakub Jelinek]

On Mon, Jan 23, 2023 at 04:06:27PM +0000, David Laight wrote:
> From: Michael.Karcher
> > Sent: 22 January 2023 00:15
> > 
> > Prevent sizeof-pointer-div warning in SH4 intc macros
> > 
> > Gcc warns about the pattern sizeof(void*)/sizeof(void), as it looks like
> > the abuse of a pattern to calculate the array size. This pattern appears
> > in the unevaluated part of the ternary operator in _INTC_ARRAY if the
> > parameter is NULL.
> > 
> > The replacement uses an alternate approach to return 0 in case of NULL
> > which does not generate the pattern sizeof(void*)/sizeof(void), but still
> > emits the warning if _INTC_ARRAY is called with a nonarray parameter.
> > 
> > This patch is required for successful compilation with -Werror enabled.
> > 
> > The idea to use _Generic for type distinction is taken from Comment #7
> > in https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108483 by Jakub Jelinek
> > 
> > Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
> > ---
> > 
> > diff --git a/include/linux/sh_intc.h b/include/linux/sh_intc.h
> > index c255273b0281..d7a7ffb60a34 100644
> > --- a/include/linux/sh_intc.h
> > +++ b/include/linux/sh_intc.h
> > @@ -97,7 +97,7 @@ struct intc_hw_desc {
> >       unsigned int nr_subgroups;
> >   };
> > 
> > -#define _INTC_ARRAY(a) a, __same_type(a, NULL) ? 0 : sizeof(a)/sizeof(*a)
> 
> FWIW it is (currently) enough to add 0 to the top or bottom
> of the division.

If you don't want the warning at all, sure.  But if you want the compiler
to warn if you use the macro on a (non-void *) pointer rather than array,
what has been posted is needed.

	Jakub

Re: [PATCH: 1/1] sh4: avoid spurious gcc warning

[Michael Karcher]

Am 23.01.2023 um 17:11 schrieb Jakub Jelinek:

> On Mon, Jan 23, 2023 at 04:06:27PM +0000, David Laight wrote:
>> From: Michael.Karcher
>>> -#define _INTC_ARRAY(a) a, __same_type(a, NULL) ? 0 : sizeof(a)/sizeof(*a)
>> FWIW it is (currently) enough to add 0 to the top or bottom
>> of the division.
> If you don't want the warning at all, sure.  But if you want the compiler
> to warn if you use the macro on a (non-void *) pointer rather than array,
> what has been posted is needed.

Exactly. I actually had sizeof(a)/(sizeof(*a) + 0) at first, but a test showed
that it would silently generate invalid code on

   struct intc_mask_reg singleton = {...};
   _INTC_ARRAY(&singleton)

If it would expand to "&singleton, 1", it would be fine, but it will
expand to "&singleton, 0", as sizeof(intc_mask_reg*) is smaller than
sizeof(intc_mask_reg). The version I posted generates the intended warning
(upgraded to an error with -Werror) in that case. The old version also
generated the intended warning in this case, and not generating a warning
here is a regression I didn't want to be responsible for.

Kind regards,
   Michael Karcher