Subject: [linux-audio-dev] Re: lowish-lat for 2.4.0-test10-pre3
From: Andrew Morton (andrewm_AT_uow.edu.au)
Date: Sun Oct 22 2000 - 03:49:47 EEST
Andi Kleen wrote:
>
> On Sun, Oct 22, 2000 at 12:58:00AM +1100, Andrew Morton wrote:
> > reschedule:
> > + sti
> > call SYMBOL_NAME(schedule) # test
>
> The sti can be removed because one of the first things schedule() does is a
> sti [forgot to mention that earlier]
Yup. But it's the rare path and I was being paranoid :)
schedule() actually does *two* sti's. One is
open-coded in the tq scheduler code and the other
is subtly buried inside release_kernel_lock(). It should
be hoisted right up to the start of the function.
BTW: regarding this comment:
> On the Mendocino the `cli' adds ~15 cycles to system calls. The
> cunning removal of a conditional jump from the syscall and interrupt
> fastpath reduces this to 13 cycles.
>
> So `getpid()' now takes 1.005x as long as it used to. So shoot me.
The arithmetic was wrong. 22nSec on 530nSec is 1.04x.
This archive was generated by hypermail 2b28 : Sun Oct 22 2000 - 04:25:13 EEST