Re: [PATCH 21/27] [AARCH64] ILP32: introduce syscalls that pass off_t

[Yury Norov]

On Tue, Jun 21, 2016 at 10:35:27AM +0000, Joseph Myers wrote:
> On Tue, 21 Jun 2016, Yury Norov wrote:
> 
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/fallocate.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/fallocate64.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/ftruncate.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/ftruncate64.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/llseek.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/lseek.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/mmap.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/posix_fadvise.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/posix_fadvise64.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/pread.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/pread64.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/pwrite.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/pwrite64.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/readahead.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/truncate.c
> >  create mode 100644 sysdeps/unix/sysv/linux/aarch64/ilp32/truncate64.c
> 
> I don't like how you need so many ilp32 files.
> 
> Presumably this is a new convention to be followed for all future ilp32 
> ABIs on 64-bit architectures.  Meaning that you should have some sysdeps 
> macros to say whether this convention is in use and then make either the 
> sysdeps/unix/sysv/linux files, or the .../generic files, or a new 
> architecture-independent sysdeps directory, implement that convention.
> 
> Note also how Adhemerval recently unified pread / pwrite implementations.  
> Adding new files for those functions goes against that unification.
> 
> -- 
> Joseph S. Myers
> joseph@xxxxxxxxxxxxxxxx

Hi Joseph,

There's RISC-V convention that implements very similar approach to
aarch64/ilp32.
https://github.com/manuelafm/riscv-gnu-toolchain/commits/master

But as I understand, there is much work there to be done before upstreaming.

So for now I think it's simpler to have this ABI in sysdeps, and be in
touch with RISC-V team to have aarch64/ilp32 compatible to RISC-V.
After RISC-V merge, we can switch aarch64/ilp32 to it and drop
unneeded sysdeps.

If you mean something different, could you explain it in details, or
suggest an action plan for such rework.

Yury.