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

[Florian Weimer <fweimer at redhat dot com>]

* Arnd Bergmann:

[Exposing futex as __NR_futex for 64-bit-only-time_t]

> Is there any long-term advantage for your approach that I'm
> missing?

Userspace that calls futex and does not care about timeouts would just
work on RV32 without any porting.

For example, consider this code in glib:

| static void
| g_futex_wait (const volatile gint *address,
|               gint                 value)
| {
|   syscall (__NR_futex, address, (gsize) FUTEX_WAIT_PRIVATE, (gsize) value, NULL);
| }

This will not work on RV32 for no good reason at all.  You would
actually have to add an #ifdef for RV32 to fix it because you clearly
don't want to do this:

| static void
| g_futex_wait (const volatile gint *address,
|               gint                 value)
| {
| #ifdef __NR_futex_time64
|   syscall (__NR_futex_time64, address, (gsize) FUTEX_WAIT_PRIVATE, (gsize) value, NULL);
| #else
|   syscall (__NR_futex, address, (gsize) FUTEX_WAIT_PRIVATE, (gsize) value, NULL);
| #endif
| }

because that would break if the library was used on an older kernel
(older than the kernel headers installed at build time).

Maybe you could use this:

| static void
| g_futex_wait (const volatile gint *address,
|               gint                 value)
| {
| #ifdef __NR_futex
|   syscall (__NR_futex_time64, address, (gsize) FUTEX_WAIT_PRIVATE, (gsize) value, NULL);
| #else
|   syscall (__NR_futex_time64, address, (gsize) FUTEX_WAIT_PRIVATE, (gsize) value, NULL);
| #endif
| }

But this would still break if people actually go ahead with the removal
of the 32-bit system calls (something I think is quite impossible to do,
but some people seem to disagree).

Fallback on ENOSYS requires introducing global variable to avoid
pointless future system calls.

Thanks,
Florian