Re: [PATCH] powerpc: Fix compiler warning on some syscalls

[Adhemerval Zanella <azanella at linux dot vnet dot ibm dot com>]

On 06-01-2015 11:18, Marek Polacek wrote:
> On Tue, Jan 06, 2015 at 11:10:38AM -0200, Adhemerval Zanella wrote:
>> GCC 5.0 emits an warning when using sizeof on array function parameters
>> and powerpc internal syscall macros add an check for such cases.  More
>> specifically, on powerpc64 and powerpc32 sysdep.h:
>>
>>   if (__builtin_classify_type (__arg3) != 5 && sizeof (__arg3) > 8) \
>>           __illegally_sized_syscall_arg3 (); \
>>
>> And for sysdeps/unix/sysv/linux/utimensat.c build GCC emits:
>>
>> error: âsizeofâ on array function parameter âtspâ will return size of
>> âconst struct timespec *â
>>
>> This patch adds explicit casts to struct pointers to avoid the warnings.
>>
>> Checked on powerpc64 and powerpc32. Ok to push?
>>
>> PS: it turned out my earlier checks were incorrect and I didn't see the
>> issue when I updated my GCC because I used --disable-werror.
>>
>> --
>>
>> 	* sysdeps/unix/sysv/linux/futimens.c (futimens): Cast timespec struct
>> 	argument to pointer.
>> 	* sysdeps/unix/sysv/linux/utimensat.c (utimensat): Likewise.
>> 	* sysdeps/unix/sysv/linux/futimesat.c (futimesat): Cast timeval struct
>> 	argument to pointer.
>> 	* sysdeps/unix/sysv/linux/utimes.c (__utimeS): Likewise.
>>
>> --
>>
>> diff --git a/sysdeps/unix/sysv/linux/futimens.c b/sysdeps/unix/sysv/linux/futimens.c
>> index c7d03a8..c0e7219 100644
>> --- a/sysdeps/unix/sysv/linux/futimens.c
>> +++ b/sysdeps/unix/sysv/linux/futimens.c
>> @@ -37,7 +37,11 @@ futimens (int fd, const struct timespec tsp[2])
>>        __set_errno (EBADF);
>>        return -1;
>>      }
>> -  return INLINE_SYSCALL (utimensat, 4, fd, NULL, tsp, 0);
>> +  /* Some archs (powerpc) add arguments type and size check using sizeof
>> +     and without a cast the compiler might emit an warning about using
>> +     sizeof on a struct (where the builtin returns the pointer size).  */
>> +  return INLINE_SYSCALL (utimensat, 4, fd, NULL, (const struct timespec*)tsp,
>> +			 0);
>>  #else
>>    __set_errno (ENOSYS);
>>    return -1;
>> diff --git a/sysdeps/unix/sysv/linux/futimesat.c b/sysdeps/unix/sysv/linux/futimesat.c
>> index ac96e2a..f7d5645 100644
>> --- a/sysdeps/unix/sysv/linux/futimesat.c
>> +++ b/sysdeps/unix/sysv/linux/futimesat.c
>> @@ -28,13 +28,13 @@
>>  /* Change the access time of FILE relative to FD to TVP[0] and
>>     the modification time of FILE to TVP[1].  */
>>  int
>> -futimesat (fd, file, tvp)
>> -     int fd;
>> -     const char *file;
>> -     const struct timeval tvp[2];
>> +futimesat (int fd, const char *file, const struct timeval tvp[2])
>>  {
>>    if (file == NULL)
>>      return __futimes (fd, tvp);
>>  
>> -  return INLINE_SYSCALL (futimesat, 3, fd, file, tvp);
>> +  /* Some archs (powerpc) add arguments type and size check using sizeof
>> +     and without a cast the compiler might emit an warning about using
>> +     sizeof on a struct (where the builtin returns the pointer size).  */
>> +  return INLINE_SYSCALL (futimesat, 3, fd, file, (const struct timeval*)tvp);
> I think it'd be better to just change the function parameter to const
> struct timeval *tvp; the warning is meant to detect a case when a sizeof
> is applied to a function parameter declared as an array.
>
> 	Marek
>
I personally don't have a preference, this suggestion was given by Joseph in a
previous threads requesting for comments.  I only see that pushing your suggestion
will require more changes (the headers and manual).