Re: [PATCH mm] kfence: fix printk format for ptrdiff_t

[Christophe Leroy <christophe.leroy@csgroup.eu>]

Le 18/03/2021 à 10:14, David Laight a écrit :
> From: Christophe Leroy
>> Sent: 17 March 2021 17:35
>>
>> Le 17/03/2021 à 13:51, David Laight a écrit :
>>> From: Christophe Leroy
>>>> Sent: 16 March 2021 15:41
>>> ...
>>>>>> include/linux/types.h:typedef __kernel_ptrdiff_t	ptrdiff_t;
>>>>>>
>>>>>> And get:
>>>>>>
>>>>>>      CC      mm/kfence/report.o
>>>>>> In file included from ./include/linux/printk.h:7,
>>>>>>                     from ./include/linux/kernel.h:16,
>>>>>>                     from mm/kfence/report.c:10:
>>>>>> mm/kfence/report.c: In function 'kfence_report_error':
>>>>>> ./include/linux/kern_levels.h:5:18: warning: format '%td' expects argument
>>>>>> of type 'ptrdiff_t', but argument 6 has type 'long int' [-Wformat=]
>>>>>
>>>>> This is declared as
>>>>>            const ptrdiff_t object_index = meta ? meta - kfence_metadata : -1;
>>>>> so maybe something with that goes wrong?  What happens if you delete the
>>>>> (useless) "const" here?
>>>
>>> The obvious thing to try is changing it to 'int'.
>>> That will break 64bit builds, but if it fixes the 32bit one
>>> it will tell you what type gcc is expecting.
>>>
>>
>> Yes, if defining 'object_index' as int, gcc is happy.
>> If removing the powerpc re-definition of ptrdiff_t typedef in
>> https://elixir.bootlin.com/linux/v5.12-rc3/source/arch/powerpc/include/uapi/asm/posix_types.h , it
>> works great as well.
>>
>> So seems like gcc doesn't take into account the typedef behind ptrdiff_t, it just expects it to be
>> int on 32 bits ?
> 
> gcc never cares how ptrdiff_t (or any of the related types) is defined
> it requires int or long for the format depending on the architecture.
> The error message will say ptrdiff_t or size_t (etc) - but that is just
> in the error message.
> 
> So the ppc32 uapi definition of __kernel_ptrdiff_t is wrong.
> However it is probably set in stone.
> 

Yes it seems to be wrong. It was changed by commit d27dfd3887 ("Import pre2.0.8"), so that's long 
time ago. Before that it was an 'int' for ppc32.

gcc provides ptrdiff_t in stddef.h via __PTRDIFF_TYPE__
gcc defined __PTRDIFF_TYPE__ as 'int' at build time.

Should we fix it in arch/powerpc/include/uapi/asm/posix_types.h ? Anyway 'long' and 'int' makes no 
functionnal difference on 32 bits so there should be no impact for users if any.

Christophe