n 12 March 2013 08:14, Vineet Gupta <[email protected]> wrote:
> On Monday 11 March 2013 10:26 PM, Markos Chandras wrote:
>> On 11 March 2013 13:21, Vineet Gupta <[email protected]> wrote:
>>> sys/time.h has libc_hidden_proto(utimes) which generates the hidden
>>> "__GI_utimes" symbol reference, and common/utimes.c has a
>>> libc_hidden_def(utimes) which generates the exported "utimes" alias.
>>>
>>> As part of no-legacy-syscall kernel ABI, Commit 80dc2ed05
>>> "utimes: Use utimensat if arch does not have the utimes syscall"
>>> introduced a new wrapper, but missed the corresponding libc_hidden_def,
>>> causing Busybox (1.20.0) link to fail (for ARC).
>>>
>>> ---------------->8-------------------
>>> ....
>>> touch.c:(.text.touch_main+0xdc): undefined reference to `utimes'
>>> touch.c:(.text.touch_main+0x114): undefined reference to `utimes'
>>> libbb/lib.a(copy_file.o): In function `copy_file':
>>> copy_file.c:(.text.copy_file+0x446): undefined reference to `utimes'
>>> collect2: ld returned 1 exit status
>>> ---------------->8-------------------
>>>
>>> Signed-off-by: Vineet Gupta <[email protected]>
>>> Cc: Markos Chandras <[email protected]>
>>> ---
>>> libc/sysdeps/linux/common/utimes.c | 2 +-
>>> 1 files changed, 1 insertions(+), 1 deletions(-)
>>>
>>> diff --git a/libc/sysdeps/linux/common/utimes.c
>>> b/libc/sysdeps/linux/common/utimes.c
>>> index a459d15..b5e76cc 100644
>>> --- a/libc/sysdeps/linux/common/utimes.c
>>> +++ b/libc/sysdeps/linux/common/utimes.c
>>> @@ -51,6 +51,6 @@ int utimes(const char *file, const struct timeval tvp[2])
>>> }
>>> #endif
>>>
>>> -#if defined __NR_utimes || defined __NR_utime
>>> +#if defined __NR_utimensat || defined __NR_utimes || defined __NR_utime
>>> libc_hidden_def(utimes)
>>> #endif
>>> --
>>> 1.7.4.1
>>>
>>> _______________________________________________
>>> uClibc mailing list
>>> [email protected]
>>> http://lists.busybox.net/mailman/listinfo/uclibc
>>
>> Well, yeah your fix is correct but the linking problem is not because
>> we don't have libc_hidden_def(utimes) for new arches. You most likely
>> build uClibc without __UCLIBC_HAS_STUBS__ and this is why you are
>> seeing the linking problems. We currently have the following code in
>> libc/sysdeps/linux/common/stubs.c
>>
>> #if !defined __NR_utime && !defined __NR_utimes
>> /*make_stub(utime) obsoleted */
>> make_stub(utimes)
>> #endif
>>
>> So, if __UCLIBC_HAS_STUBS__ is enabled, you will have a utimes symbol
>> in your libc but it will actually return ENOSYS. So whilst we provide
>> a utimes function for new architectures, we never actually use it and
>> we create a stub instead. This is wrong so I believe your patch needs
>> to fix the libc/sysdeps/linux/common/stubs.c entry as well.
>>
>
> Thanks for reviewing Markos. I'll fix up the patch per above. Any chance you
> can
> comment on the other couple patches so that I can re-spin the full v2 series.
>
> Also per another discussion on lkml yesterday
> (http://www.spinics.net/lists/kernel/msg1496470.html), the no-legacy-abi need
> not
> send IPC_64 to kernel. I see that you fixed it with 6c2fa3451f35f8 + fix to
> metag
> to define it's own version - but I think a better solution is to do that
> within
> libc/misc/sysvipc/ipc.h using __ARCH_HAS_DEPRECATED_SYSCALLS__
That could work just as fine I guess.
>
> Yet another change I'm thinking it to make the align64 optional - as it is
> not a
> hard requirement for all arches (e.g. ARC) and leads to bit of space wastage.
>
> -Vineet
I think the current implementation shouldn't cause any problems no
matter what your arch requires. I presume you mean that for 64-bit
structs you don't need 64-bit alignment. Then this struct
__longlong_aligned { long long x; } (from align64bit.h) should return
4 for you and no space is wasted in your case.
--
Regards,
Markos Chandras
_______________________________________________
uClibc mailing list
[email protected]
http://lists.busybox.net/mailman/listinfo/uclibc
