Re: [PATCH v3 3/6] parisc: add system call table generation support

[Arnd Bergmann]

On Mon, Oct 8, 2018 at 3:02 PM Eugene Syromiatnikov <esyr@xxxxxxxxxx> wrote:
> > On Mon, 8 Oct 2018 at 10:47, Firoz Khan <firoz.khan@xxxxxxxxxx> wrote:
> > > diff --git a/arch/parisc/kernel/syscalls/syscall.tbl b/arch/parisc/kernel/syscalls/syscall.tbl
>
> > > +84      common  lstat                           sys_newlstat                    compat_sys_newlstat
> > > +85      common  readlink                        sys_readlink
> > > +86      common  uselib                          sys_ni_syscall
>
> Why uselib is declared, contrary to all the skipped syscalls below,
> that were sys_ni_syscall previously? Only __NR_socketcall was explicitly
> undefined in arch/parisc/include/uapi/asm/unistd.h.

Good catch, I didn't see that in my earlier review. I agree we want the
files to be identical to the version they replace, so the macros need to
be there for now.

We may later decide to clean this up and remove all __NR_* that have
no entry point, but the conversion to the new table format needs to
otherwise be a nop.

> > > +87      common  swapon                          sys_swapon
> > > +88      common  reboot                          sys_reboot
> > > +89      common  mmap2                           sys_mmap2
> > > +90      common  mmap                            sys_mmap
> > > +91      common  munmap                          sys_munmap
> > > +92      common  truncate                        sys_truncate                    compat_sys_truncate
> > > +93      common  ftruncate                       sys_ftruncate                   compat_sys_ftruncate
> > > +94      common  fchmod                          sys_fchmod
> > > +95      common  fchown                          sys_fchown
> > > +96      common  getpriority                     sys_getpriority
> > > +97      common  setpriority                     sys_setpriority
> > > +98      common  recv                            sys_recv
> > > +99      common  statfs                          sys_statfs                      compat_sys_statfs
> > > +100     common  fstatfs                         sys_fstatfs                     compat_sys_fstatfs
> > > +101     common  stat64                          sys_stat64
>
> It is probably worth adding a comment here that syscall 102 was
> socketcall, in order to make reason for this jump in syscall numeration
> self-evident.

+1

In general, I'd argue we want to keep all the nontrivial comments that
were present in either unistd.h or syscall_table.S.

> > > +103     common  syslog                          sys_syslog
> > > +104     common  setitimer                       sys_setitimer                   compat_sys_setitimer
> > > +105     common  getitimer                       sys_getitimer                   compat_sys_getitimer
> > > +106     common  capget                          sys_capget
> > > +107     common  capset                          sys_capset
>
> > > +108     32      pread64                         parisc_pread64
> > > +108     64      pread64                         sys_pread64                     parisc_pread64
>
> It would be probably nice to have some syntax that would allow avoid
> this duplication (as compat handler on 64 bit and native on 32 bit are
> the same).

I think I would prefer to have the compat table be generated with the '32'
abi rather than the '64' abi, so we end up with

108     32      pread64                         parisc_pread64
   parisc_pread64
108     64      pread64                         sys_pread64

I think this makes more sense, in particular on the other architectures
that have different macro names in some cases. When we do this,
the entries could get compressed to

108     32      pread64                         parisc_pread64
108     64      pread64                         sys_pread64

> > > +348     common  pwritev2                        sys_pwritev2                    compat_sys_pwritev2
> > > +349     common  statx                           sys_statx
> > > +350    common  io_pgetevents                   sys_io_pgetevents               compat_sys_io_pgetevents
> > > \ No newline at end of file

As others have commented several times, Firoz still needs to fix
the missing newline.

Firoz, please fix all the newlines before you repost any further
patches.

      Arnd