Re: [RFC][PATCH] Runtime switching of the idle function [take 2]

[Lee Revell]

On Tue, 2005-11-29 at 21:51 +0100, Andi Kleen wrote:
> On Tue, Nov 29, 2005 at 03:35:39PM -0500, Lee Revell wrote:
> > On Tue, 2005-11-29 at 20:53 +0100, Andi Kleen wrote:
> > > We're mostly addressing it - there are problems left, but
> > > overall it's looking good. The remaining problem is 
> > > an education issue of users to not use RDTSC directly, 
> > > but use gettimeofday/clock_gettime 
> > 
> > No the issue is to make gettimeofday fast enough that the people who
> > currently have to use the TSC can use it.  Right now it's 1500-3000 nsec
> > or so, Vojtech mentioned that he has a patch that could reduce that to
> 
> It's only that slow if the hardware can't do better.
> 
> And the kernel makes it only slow when using RDTSC directly
> is unsafe - so if you use it directly thinking the kernel cheats
> you for your cycles you're just shoting yourself in the own foot.
> 
> > 150-300 nsec.
> 
> If you have capable hardware it can already do much better.
> 

But on my system gettimeofday uses the TSC and it's still ~35x slower
than RDTSC:

rlrevell@mindpipe:~$ ./timetest 
rdtsc: 10000 calls in 1079 usecs
gettimeofday: 10000 calls in 36628 usecs

#include <stdio.h>
#include <stdlib.h>
#include <sys/time.h>

typedef unsigned long long cycles_t;

#define rdtscll(val) \
     __asm__ __volatile__("rdtsc" : "=A" (val))

static inline cycles_t get_cycles_tsc (void)
{
	unsigned long long ret;

	rdtscll(ret);
	return ret;
}

static inline cycles_t get_cycles_gtod (void)
{
       struct timeval tv;
       gettimeofday (&tv, NULL);

       return tv.tv_usec;
}

int main (void) {
	int i;
	cycles_t start_time;
	start_time= get_cycles_gtod();
	for (i = 0; i < 10000; i++) {
		get_cycles_tsc();
	}
	printf("rdtsc: %i calls in %llu usecs\n", i, get_cycles_gtod() - start_time);
	start_time = get_cycles_gtod();
	for (i = 0; i < 10000; i++) {
		get_cycles_gtod();
	}
	printf("gettimeofday: %i calls in %llu usecs\n", i, get_cycles_gtod() - start_time);
	return 0;
}


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/