Re: [RFC v2 08/20] sysdeps/wait: Use waitid if avaliable

[Arnd Bergmann <arnd at arndb dot de>]

On Tue, Jun 25, 2019 at 3:47 PM Florian Weimer <fweimer@redhat.com> wrote:
>
> * Arnd Bergmann:
>
> > On Tue, Jun 25, 2019 at 3:29 PM Arnd Bergmann <arnd@arndb.de> wrote:
> >> On Tue, Jun 25, 2019 at 2:10 PM Florian Weimer <fweimer@redhat.com> wrote:
> >>
> >> > Does this means that RV32 will use a 32-bit struct timeval in those
> >> > system calls?  Even if everything else 64-bit?
> >>
> >> Correct. Only those four (all deprecated but still used) system calls,
> >> as we could not agree on a new interface before 5.1, and there
> >> is no urgency for deployment when they can be emulated.
> >
> > Correction: getrusage() is still a recommended interface in POSIX.1-2017
> > with no nanosecond based replacement, while wait4(), getitimer() and
> > getrusage() are all obsolete but cannot be implemented on top of other
> > POSIX system calls.
>
> This makes me rather unhappy.  I also don't see the benefit of renaming
> all time-related system calls for new architectures.

What got renamed? I was trying very hard to make it as consistent
and easy as possible. There is a strict mapping of __NR_* macros
to argument types for each system call [1], so e.g. __kernel_timespec
will always be used together with system calls named *time64(),
while the old system calls always refer to the traditional
"struct timespec {long tv_sec; long tv_nsec;}" type. This is the
same way that loff_t gets handled, so I assumed that all C libraries
would know how to deal with this well.

> Oh well.  I hope
> someone will figure out how to integrate this smoothly into glibc.

Integrating this should actually be easier than the other system calls
since you only need to implement the fallback path and not also the
call into the replacement. One major factor for not requiring replacements
for getrusage()/getitimer()/setitimer() first was actually that both glibc
and musl have plans to support old kernels and need to implement
the fallback path regardless of whether we implemented the new
version or not.

       Arnd

[1] ignoring the universally hated x32 ABI