Re: [PATCH v4 08/13] LoongArch: Linux Syscall Interface

[Adhemerval Zanella <adhemerval.zanella@linaro.org>]

On 15/05/2022 04:53, caiyinyu wrote:
> 
> 在 2022/5/11 下午3:02, Arnd Bergmann 写道:
>> On Mon, May 9, 2022 at 4:26 AM caiyinyu <caiyinyu@loongson.cn> wrote:
>>
>>> +#define __NR_clone 220
>>> +#define __NR_clone3 435
>>> +#define __NR_fstat 80
>>> +#define __NR_newfstatat 79
>> stat() and newfstatat() are not in the current ABI for the kernel, this needs
>> to be updated, and the libc wrapper updated to call statx() instead.
>>
>> clone() is still an open question, this may have to be implemented on top
>> of clone3()
>>
>>         Arnd
> 
> 
> Currently, glibc is not enable FSTATAT_USE_STATX in 64 bits system.

In fact I had the wrong assumption that __NR_newfstatat would continue t be
used on 64-bit architectures (since it requires slight less memory copy
to implement stat/lstat/fstat/fstatat.

> Shall we modify sysdeps/unix/sysv/linux/fstatat64.c to support statx like:
> 
> ===
> diff --git a/sysdeps/unix/sysv/linux/fstatat64.c b/sysdeps/unix/sysv/linux/fstatat64.c
> index 2ab914380d..bd3b20644e 100644
> --- a/sysdeps/unix/sysv/linux/fstatat64.c
> +++ b/sysdeps/unix/sysv/linux/fstatat64.c
> @@ -42,7 +42,7 @@ _Static_assert (sizeof (__blkcnt_t) == sizeof (__blkcnt64_t),
> 
>  #if (__WORDSIZE == 32 \
>       && (!defined __SYSCALL_WORDSIZE || __SYSCALL_WORDSIZE == 32)) \
> -     || defined STAT_HAS_TIME32
> +     || defined STAT_HAS_TIME32 || !defined __NR_newfstatat

It makes sense.

>  # define FSTATAT_USE_STATX 1
> 
>  static inline int
> ===
> 
> or add fstatat64.c in loongarch port??

No, if kernel interface for newers 64-bit ports to only provide statx we should
adapt the generic code to so.

> By now no other architectures have their own fstatat64.c.
>