Re: [PATCH v4 09/13] LoongArch: Linux ABI

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

On 11/05/2022 04:08, Arnd Bergmann wrote:
> On Mon, May 9, 2022 at 4:26 AM caiyinyu <caiyinyu@loongson.cn> wrote:
>> +
>> +#include <bits/wordsize.h>
>> +
>> +/* In 64-bit ISA files are always with 64bit off_t and F_*LK64 are the same as
>> +   non-64-bit versions.  It will need to be revised for 128-bit.  */
>> +#if __WORDSIZE == 64
>> +#define __O_LARGEFILE 0
>> +
>> +#define F_GETLK64 5  /* Get record locking info.  */
>> +#define F_SETLK64 6  /* Set record locking info (non-blocking).  */
>> +#define F_SETLKW64 7 /* Set record locking info (blocking).  */
>> +#endif
>> +
>> +struct flock
>> +{
>> +  short int l_type;   /* Type of lock: F_RDLCK, F_WRLCK, or F_UNLCK.  */
>> +  short int l_whence; /* Where `l_start' is relative to (like `lseek').  */
>> +#ifndef __USE_FILE_OFFSET64
>> +  __off_t l_start; /* Offset where the lock begins.  */
>> +  __off_t l_len;   /* Size of the locked area; zero means until EOF.  */
>> +#else
>> +  __off64_t l_start; /* Offset where the lock begins.  */
>> +  __off64_t l_len;   /* Size of the locked area; zero means until EOF.  */
>> +#endif
>> +  __pid_t l_pid; /* Process holding the lock.  */
>> +};
>> +
>> +#ifdef __USE_LARGEFILE64
>> +struct flock64
>> +{
>> +  short int l_type;   /* Type of lock: F_RDLCK, F_WRLCK, or F_UNLCK.  */
>> +  short int l_whence; /* Where `l_start' is relative to (like `lseek').  */
>> +  __off64_t l_start;  /* Offset where the lock begins.  */
>> +  __off64_t l_len;    /* Size of the locked area; zero means until EOF.  */
>> +  __pid_t l_pid;      /* Process holding the lock.  */
>> +};
>> +#endif
> 
> Not sure why these are provided here, the kernel has no loongarch64 specific
> interface for flock in the current version, so either this refers to
> an older version
> of the kernel port, or it is something that could be generalized into an
> architecture-independent file in glibc.

Indeed this is a missing spot in Linux support, the generic
sysdeps/unix/sysv/linux/bits/fcntl.h is just a #error stating each
architecture should define its own.

glibc does have a generic interface on generic subfolder,
sysdeps/unix/sysv/linux/generic/bits/fcntl.h; however loongarch is
not using it because it does not add the folder on 
sysdeps/unix/sysv/linux/loongarch/Implies. NIOS2, for instance,
which as the original architecture that pushed for the generic
folder (as a easier way to provide the new UAPI kernel definitions)
has:

$ cat ./sysdeps/unix/sysv/linux/nios2/Implies
nios2/nptl
unix/sysv/linux/generic/wordsize-32
unix/sysv/linux/generic

Making loongarch use it will change some implementation selections,
so it would require to check it it is ok for the architecture.

My long term plan is to just remove the generic and subfolder to avoid
this exact confusion.