> On Tue, 31 Mar 1998, Sander van Malssen wrote:
>
> > BTW, any other stats that someone would like to see? (Perhaps we
> > could finally crack that ancient
> > interrupt-handling-reported-as-idle-time bug somehow.)
>
> this one should be doable, the speed of the timer interrupts isnt very
> critical. You'll probably have to look at the interrupted EIP, and
> check wether it's _not_ the 'hlt' instruction (only if we are in the
> idle task). Then (and only then) add this tick to system time.
But it's not just a i386 problem, is it? (And even then there's the
no-hlt option, too.)
> > call SYMBOL_NAME(sys_call_table)(,%eax,4)
> > + incl kstat # kstat.syscalls++
> > movl %eax,EAX(%esp) # save the return value
>
> do we really want to have 2 cycles off from this _very_ heavily used
> code path? [== 2% slower getpid()]
Do we really need that? I don't know. I'm wondering about the same thing
for most of the kstat stuff. :-)
I suppose that if many more of there things were to be added, we'd put a
CONFIG_something around them. The highest I've managed to get is ~4700
syscals per second with a find(1) on a directory tree that was already
in the kernel cache, BTW.
Cheers,
Sander
--
Sander van Malssen -- svm@kozmix.ow.nl -- http://svm.www.cistron.nl/
* The 1-2-5 Page: http://svm.www.cistron.nl/music/ *
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu