Re: "Bug" in 2.0.x/2.1.x
Linus Torvalds (torvalds@cs.helsinki.fi)
Wed, 18 Dec 1996 10:39:33 +0200 (EET)
On Tue, 17 Dec 1996, Andi Kleen wrote:
>
> In kernel/sched.c in Linux 2.0.x/2.1.x I see this:
>
> asmlinkage int sys_sched_rr_get_interval(pid_t pid, struct timespec *interval)
> {
> struct timespec t;
>
> t.tv_sec = 0;
> t.tv_nsec = 0; /* <-- Linus, please fill correct value in here */
> return -ENOSYS; /* and then delete this line. Thanks! */
> return copy_to_user(interval, &t, sizeof(struct timespec)) ? -EFAULT : 0;
> }
>
> See the problem? ;) I can't supply a patch because I don't know the length
> of a time slice in Linux (it is architecture dependent, isn't it?) and I don't
> have a POSIX.4 copy at hand.
It's defined by "DEF_PRIORITY" in ticks, currently
#define DEF_PRIORITY (20*HZ/100) /* 200 ms time slice */
so the correct value would probably be
t.tv_nsec = DEF_PRIORITY*1000000/HZ;
but that's just the system "default" value, and I suspect that the
function should return the per-process thing which is more like
t.tv_sec = current->priority * 1000000/HZ;
could you test it out and see if it works for you?
I don't want to change any of the RT stuff as I don't have any
test-programs that care about it, and as such I don't have any personal
interest..
Linus