RE: [PATCH] compatibility syscall layer (lets try again)

[Paul Mackerras <paulus@samba.org>]

Linus Torvalds writes:

> Note that I've not committed the patch to my tree at all, and as far as I
> am concerned this is in somebody elses court (ie somebody that cares about
> restarting). I don't have any strong feelings either way about how
> restarting should work - and I'd like to have somebody take it up and
> testing it as well as having architecture maintainers largely sign off on
> this approach.

There is a simpler way to solve the nanosleep problem which doesn't
involve any more restart magic than we have been using for years.
That is to define a new sys_new_nanosleep system call which takes one
argument which is a pointer to the time to sleep.  If the sleep gets
interrupted by a pending signal, the kernel sys_new_nanosleep will
write back the remaining time (overwriting the requested time) and
return -ERESTARTNOHAND.  The glibc nanosleep() then looks like this:

int nanosleep(const struct timespec *req, struct timespec *rem)
{
	*rem = *req;
	return new_nanosleep(rem);
}

Any reason why this can't work?

(BTW this is Rusty's idea. :)

Regards,
Paul.