[PATCH v2] linux: posix_fadvise: restore implementation for xtensa

[Vineet Gupta]

On Thursday 09 January 2014 11:55 AM, Baruch Siach wrote:
> Hi Vineet,
> 
> On Thu, Jan 09, 2014 at 11:28:19AM +0530, Vineet Gupta wrote:
>> On Sunday 05 January 2014 03:14 PM, Baruch Siach wrote:
>>> Commit ee84b8b400 (linux: posix_fadvise: use new SYSCALL_ALIGN_64BIT)
>>> removed posix_fadvise implementation for xtensa, since xtensa does not
>>> define __NR_fadvise64. Reuse the ARM support code to restore xtensa support.
>>>
>>> This commit is based Mike Frysinger's suggested patch.
>>>
>>> Cc: Mike Frysinger <vapier at gentoo.org>
>>> Signed-off-by: Baruch Siach <baruch at tkos.co.il>
>>> ---
>>> v2: Use Mike's approach, v1 was broken
>>> ---
>>>  libc/sysdeps/linux/common/posix_fadvise.c | 15 +++++++++++----
>>>  1 file changed, 11 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/libc/sysdeps/linux/common/posix_fadvise.c b/libc/sysdeps/linux/common/posix_fadvise.c
>>> index e102ce7afe31..25c294178e5e 100644
>>> --- a/libc/sysdeps/linux/common/posix_fadvise.c
>>> +++ b/libc/sysdeps/linux/common/posix_fadvise.c
>>> @@ -10,18 +10,25 @@
>>>  
>>>  #include <sys/syscall.h>
>>>  
>>> -#if defined(__NR_fadvise64) || defined(__NR_arm_fadvise64_64)
>>> +#ifdef __NR_arm_fadvise64_64
>>> +/* We handle the 64bit alignment issue which is why the arm guys renamed their
>>> + * syscall in the first place.  So rename it back.
>>> + */
>>> +# define __NR_fadvise64_64 __NR_arm_fadvise64_64
>>> +#endif
>>> +
>>> +#if defined(__NR_fadvise64) || defined(__NR_fadvise64_64)
>>>  # include <fcntl.h>
>>>  # include <endian.h>
>>>  # include <bits/wordsize.h>
>>>  
>>> -# ifdef __NR_arm_fadvise64_64
>>> +# ifdef __NR_fadvise64_64
>>>  int posix_fadvise64(int fd, off64_t offset, off64_t len, int advice);
>>>  # endif
>>>  
>>>  int posix_fadvise(int fd, off_t offset, off_t len, int advice)
>>>  {
>>> -# ifdef __NR_arm_fadvise64_64
>>> +# ifdef __NR_fadvise64_64
>>>  	return posix_fadvise64(fd, offset, len, advice);
>>>  # else
>>>  	int ret;
>>> @@ -41,7 +48,7 @@ int posix_fadvise(int fd, off_t offset, off_t len, int advice)
>>>  	return 0;
>>>  #  endif
>>>  }
>>> -# if defined __UCLIBC_HAS_LFS__ && ((!defined __NR_fadvise64_64 && !defined __NR_arm_fadvise64_64) || __WORDSIZE == 64)
>>> +# if defined __UCLIBC_HAS_LFS__ && (!defined __NR_fadvise64_64 || __WORDSIZE == 64)
>>>  strong_alias(posix_fadvise,posix_fadvise64)
>>>  # endif
>>>  #endif
>>>
>>
>> Thx for this fix, it unborks e2fsprogs build for ARC. However this doesn't build
>> for !LFS. It seems !LFS breakage was introduced with 347567313 "add
>> posix_fadvise() for arm" and confined only to ARM. Now that everyone uses same
>> code, the damage is noticeable elsewhere too.
>>
>> The twist now is that, arches such as ARC/METAG are "no-legacy-syscalls-kernel ABI
>> - so we only have __NR_fadvise64_64 (even for !LFS)
>>
>> I'll try to come up with a patch later today but just wanted to bring this to
>> attention anyways.
> 
> Thanks for testing. Looking forward to your patch. It's needed for xtensa as 
> well, I guess. I really wonder whether this !LFS is worth the trouble.  How 
> much do you actually save when disabling LFS?

Personally I don't care much, however customers sometimes run into it and if the
feature is there we might as well keep it working.

-Vineet