[PATCH v2 02/18] RISC-V: Define __NR_* as __NR_*_time64/64 for 32-bit

[Alistair Francis]

On Wed, Jul 8, 2020 at 10:08 AM Adhemerval Zanella via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
>
>
> On 07/07/2020 21:09, Maciej W. Rozycki via Libc-alpha wrote:
> > Alistair,
> >
> >  I think the change heading is too cryptic and does not express the intent
> > of the change well enough.  How about:
> >
> > RISC-V: Use 64-bit-time syscall numbers with the 32-bit port
> >
> > and then maybe explain in a little more details in the change description.
> >
> > On Wed, 3 Jun 2020, Alistair Francis via Libc-alpha wrote:
> >
> >> diff --git a/sysdeps/unix/sysv/linux/riscv/sysdep.h b/sysdeps/unix/sysv/linux/riscv/sysdep.h
> >> index 83e4adf6a2..aa61e8b04d 100644
> >> --- a/sysdeps/unix/sysv/linux/riscv/sysdep.h
> >> +++ b/sysdeps/unix/sysv/linux/riscv/sysdep.h
> >> @@ -116,6 +116,67 @@
> >>
> >>  #include <sysdeps/unix/sysdep.h>
> >
> >  This file is weird, as it includes <sysdeps/unix/sysdep.h> twice, first
> > time indirectly via <sysdeps/unix/sysv/linux/generic/sysdep.h> at the top,
> > and then second time here.  So I think this second inclusion can be
> > removed (along with the preceding inclusion of <errno.h>, as it does not
> > appear to change anything), and the following conditional moved to the
> > top, just after the inclusion of <tls.h>.  Oddly <sysdeps/unix/sysdep.h>
> > has not been protected against multiple inclusion, but its contents do not
> > trigger compilation warnings if processed more than once.
>
> This multiple inclusion is done internally in some places so the file
> that wants to override a definition would first include the generic
> definitions and then #undef/#define the new one (kernel-features.h
> still does it).

I'm not sure if this is required. I split out the clean up into a
seperate patch and I don't see any regressions.

>
> I think this is kind of confusing, but this is how sysdep.h is currently
> organized.  We have been changing bit per bit, but there is a lot of
> place where inclusing is done without guards.
>
> >
> >  The two #ifdef/#ifndef __ASSEMBLER__ conditionals will then become
> > adjacent and can be merged into a single #ifdef/#else one.
> >
> >  This clean-up would probably better be made as a separate preceding
> > change.
> >
> >> +#if __riscv_xlen == 32
> >
> >  I think using __WORDSIZE here would be more consistent with the rest of
> > our code (we do use `__riscv_xlen' in a couple of places, but I think they
> > ought to be cleaned up).
> >
> >> +/* Define the __NR_futex as __NR_futex64 as RV32 doesn't have a
> >> + * __NR_futex syscall.
> >> + */
> >> +# ifndef __NR_futex
> >> +#  define __NR_futex __NR_futex_time64
> >> +# endif
> >
> >  The comment does not match the code.
> >
> >  I think it makes no sense to comment on individual entries as they all
> > repeat the same pattern and the same purpose, so an introductory comment
> > covering them all at the beginning of the conditional would be better
> > instead.  I suppose you can then reuse it for the change description too.
> >
> >> +# ifndef __NR_clock_adjtime
> >> +#  define __NR_clock_adjtime __NR_clock_adjtime64
> >> +# endif
> >> +#endif /* __riscv_xlen == 32 */
> >
> >  Since you have multiple inner conditionals separated by an empty line
> > each I think it will make sense to have an empty line as well between the
> > final one and the closing of the outer conditional.  Likewise at the
> > beginning.
>
> I think the correct way to how ARC port is doing which is:
>
>  1. Add any required syscall suppression for 32-bit off_t/time_t on
>     fixup-asm-unistd.h so arch-syscall.h will have only the required
>     definitions.  This might be the case for riscv32 since its kernel
>     ABI only supports 64-bit off_t/time_t.
>
>  2. Regenerate arch-syscall.h if it were the case.
>
>  3. On sysdep.h redefine only the syscall where the generic implementation
>     still does not have actual 64-bit time_t support:
>
>    /* "workarounds" for generic code needing to handle 64-bit time_t.  */
>
>    /* Fix sysdeps/unix/sysv/linux/clock_getcpuclockid.c.  */
>    #define __NR_clock_getres    __NR_clock_getres_time64
>    /* Fix sysdeps/nptl/lowlevellock-futex.h.  */
>    #define __NR_futex           __NR_futex_time64
>    [...]
>
>    (with proper guards it should require the #ifndef/#define)
>
>  4. Add a comment it is a workaround to handle 64-bit time_t
>     and on each #define comment for which implementation it intends to.

Thanks! I have updated this to be similar to the ARC port. I also
paraphrased the points above in the commit message (I hope that is
ok).

Alistair