From: torvalds@transmeta.com (Linus Torvalds)
To: linux-kernel@vger.kernel.org
Subject: Re: Intel P6 vs P7 system call performance
Date: Mon, 30 Dec 2002 21:22:14 +0000 (UTC) [thread overview]
Message-ID: <auqde6$1pv$1@penguin.transmeta.com> (raw)
In-Reply-To: 20021230182209.GA3981@core.home
In article <20021230182209.GA3981@core.home>,
Christian Leber <christian@leber.de> wrote:
>
>But now the right and interesting lines:
>
>2.5.53:
>igor3:~# ./a.out
>166.283549 cycles
>278.461609 cycles
>
>2.5.53-bk5:
>igor3:~# ./a.out
>150.895348 cycles
>279.441955 cycles
>
>The question is: are the numbers correct?
Roughly. The program I posted has some overflow errors (which you will
hit if testing expensive system calls that take >4000 cycles). They also
do an average, which is "mostly correct", but not stable if there is
some load in the machine. The right way to do timings like this is
probably to do minimums for individual calls, and then subtract out the
TSC reading overhead. See attached silly program.
>And why have int 80 also gotten faster?
Random luck. Sometimes you get cacheline alignment magic etc. Or just
because the timings aren't stable for other reasons (background
processes etc).
>Is this a valid testprogramm to find out how long a system call takes?
Not really. The results won't be stable, since you might have cache
misses, page faults, other processes, whatever.
So you'll get _somehat_ correct numbers, but they may be randomly off.
Linus
---
#include <sys/types.h>
#include <time.h>
#include <sys/time.h>
#include <sys/fcntl.h>
#include <asm/unistd.h>
#include <sys/stat.h>
#include <stdio.h>
#define rdtsc() ({ unsigned long a, d; asm volatile("rdtsc":"=a" (a), "=d" (d)); a; })
// for testing _just_ system call overhead.
//#define __NR_syscall __NR_stat64
#define __NR_syscall __NR_getpid
#define NR (100000)
int main()
{
int i, ret;
unsigned long fast = ~0UL, slow = ~0UL, overhead = ~0UL;
struct timeval x,y;
char *filename = "test";
struct stat st;
int j;
for (i = 0; i < NR; i++) {
unsigned long cycles = rdtsc();
asm volatile("");
cycles = rdtsc() - cycles;
if (cycles < overhead)
overhead = cycles;
}
printf("overhead: %6d\n", overhead);
for (j = 0; j < 10; j++)
for (i = 0; i < NR; i++) {
unsigned long cycles = rdtsc();
asm volatile("call 0xffffe000"
:"=a" (ret)
:"0" (__NR_syscall),
"b" (filename),
"c" (&st));
cycles = rdtsc() - cycles;
if (cycles < fast)
fast = cycles;
}
fast -= overhead;
printf("sysenter: %6d cycles\n", fast);
for (i = 0; i < NR; i++) {
unsigned long cycles = rdtsc();
asm volatile("int $0x80"
:"=a" (ret)
:"0" (__NR_syscall),
"b" (filename),
"c" (&st));
cycles = rdtsc() - cycles;
if (cycles < slow)
slow = cycles;
}
slow -= overhead;
printf("int0x80: %6d cycles\n", slow);
printf(" %6d cycles difference\n", slow-fast);
printf("factor %f\n", (double) slow / fast);
}
next prev parent reply other threads:[~2002-12-30 21:14 UTC|newest]
Thread overview: 304+ messages / expand[flat|nested] mbox.gz Atom feed top
2002-12-09 8:30 Intel P6 vs P7 system call performance Mike Hayward
2002-12-09 15:40 ` erich
2002-12-09 17:48 ` Linus Torvalds
2002-12-09 19:36 ` Dave Jones
2002-12-09 19:46 ` H. Peter Anvin
2002-12-28 20:37 ` Ville Herva
2002-12-29 2:05 ` Christian Leber
2002-12-30 18:22 ` Christian Leber
2002-12-30 21:22 ` Linus Torvalds [this message]
2002-12-30 11:29 ` Dave Jones
2002-12-17 0:47 ` Linus Torvalds
2002-12-17 1:03 ` Dave Jones
2002-12-17 2:36 ` Linus Torvalds
2002-12-17 5:55 ` Linus Torvalds
2002-12-17 6:09 ` Linus Torvalds
2002-12-17 6:18 ` Linus Torvalds
2002-12-19 14:03 ` Shuji YAMAMURA
2002-12-17 6:19 ` GrandMasterLee
2002-12-17 6:43 ` dean gaudet
2002-12-17 16:50 ` Linus Torvalds
2002-12-17 19:11 ` H. Peter Anvin
2002-12-17 21:39 ` Benjamin LaHaise
2002-12-17 21:41 ` H. Peter Anvin
2002-12-17 21:53 ` Benjamin LaHaise
2002-12-18 23:53 ` Pavel Machek
2002-12-19 22:18 ` H. Peter Anvin
2002-12-19 22:21 ` Pavel Machek
2002-12-19 22:23 ` H. Peter Anvin
2002-12-19 22:26 ` Pavel Machek
2002-12-19 22:30 ` H. Peter Anvin
2002-12-19 22:34 ` Pavel Machek
2002-12-19 22:36 ` H. Peter Anvin
2002-12-17 19:12 ` H. Peter Anvin
2002-12-17 19:26 ` Martin J. Bligh
2002-12-17 20:51 ` Alan Cox
2002-12-17 20:16 ` H. Peter Anvin
2002-12-17 20:49 ` Alan Cox
2002-12-17 20:12 ` H. Peter Anvin
2002-12-17 9:45 ` Andre Hedrick
2002-12-17 12:40 ` Dave Jones
2002-12-17 23:18 ` Andre Hedrick
2002-12-17 15:12 ` Alan Cox
2002-12-18 23:55 ` Pavel Machek
2002-12-19 22:17 ` H. Peter Anvin
2002-12-17 10:53 ` Ulrich Drepper
2002-12-17 11:17 ` dada1
2002-12-17 17:33 ` Ulrich Drepper
2002-12-17 17:06 ` Linus Torvalds
2002-12-17 17:55 ` Ulrich Drepper
2002-12-17 18:01 ` Linus Torvalds
2002-12-17 19:23 ` Alan Cox
2002-12-17 18:48 ` Ulrich Drepper
2002-12-17 19:19 ` H. Peter Anvin
2002-12-17 19:44 ` Alan Cox
2002-12-17 19:52 ` Richard B. Johnson
2002-12-17 19:54 ` H. Peter Anvin
2002-12-17 19:58 ` Linus Torvalds
2002-12-18 7:20 ` Kai Henningsen
2002-12-17 18:49 ` Linus Torvalds
2002-12-17 19:09 ` Ross Biro
2002-12-17 21:34 ` Benjamin LaHaise
2002-12-17 21:36 ` H. Peter Anvin
2002-12-17 21:50 ` Benjamin LaHaise
2002-12-18 23:59 ` Pavel Machek
2002-12-17 16:12 ` Hugh Dickins
2002-12-17 16:33 ` Richard B. Johnson
2002-12-17 17:47 ` Linus Torvalds
2002-12-17 16:54 ` Hugh Dickins
2002-12-17 17:07 ` Linus Torvalds
2002-12-17 17:19 ` Matti Aarnio
2002-12-17 17:55 ` Linus Torvalds
2002-12-17 18:24 ` Linus Torvalds
2002-12-17 18:33 ` Ulrich Drepper
2002-12-17 18:30 ` Ulrich Drepper
2002-12-17 19:04 ` Linus Torvalds
2002-12-17 19:19 ` Ulrich Drepper
2002-12-17 19:28 ` Linus Torvalds
2002-12-17 19:32 ` H. Peter Anvin
2002-12-17 19:44 ` Linus Torvalds
2002-12-17 19:53 ` Ulrich Drepper
2002-12-17 20:01 ` Linus Torvalds
2002-12-17 20:17 ` Ulrich Drepper
2002-12-18 4:15 ` Linus Torvalds
2002-12-18 4:15 ` Linus Torvalds
2002-12-18 4:39 ` H. Peter Anvin
2002-12-18 4:49 ` Linus Torvalds
2002-12-18 6:38 ` Linus Torvalds
2002-12-18 13:17 ` Richard B. Johnson
2002-12-18 13:40 ` Horst von Brand
2002-12-18 13:47 ` Sean Neakums
2002-12-18 14:10 ` Horst von Brand
2002-12-18 14:51 ` dada1
2002-12-18 19:12 ` Mark Mielke
2002-12-18 15:52 ` Alan Cox
2002-12-18 16:41 ` Dave Jones
2002-12-18 16:49 ` Freezing.. (was Re: Intel P6 vs P7 system call performance) Linus Torvalds
2002-12-18 16:56 ` Larry McVoy
2002-12-18 16:58 ` Dave Jones
2002-12-18 17:41 ` Linus Torvalds
2002-12-18 18:03 ` Jeff Garzik
2002-12-18 18:09 ` Mike Dresser
2002-12-23 12:34 ` Kai Henningsen
2002-12-18 19:08 ` Alan Cox
2002-12-18 19:23 ` Larry McVoy
2002-12-18 19:30 ` Alan Cox
2002-12-18 19:33 ` Larry McVoy
2002-12-18 19:42 ` Alan Cox
2002-12-18 19:45 ` Larry McVoy
2002-12-18 20:39 ` John Bradford
2002-12-18 22:08 ` Larry McVoy
2002-12-18 22:37 ` John Bradford
2002-12-19 1:09 ` Alan Cox
2002-12-19 0:37 ` Russell King
2002-12-19 0:58 ` Jeff Garzik
2002-12-19 1:43 ` Martin J. Bligh
2002-12-19 10:50 ` Dave Jones
2002-12-19 0:59 ` John Bradford
2002-12-19 10:27 ` Dave Jones
2002-12-19 1:17 ` Linus Torvalds
2002-12-19 0:08 ` Alan Cox
2002-12-19 0:53 ` John Bradford
2002-12-19 13:17 ` Stephen Satchell
2002-12-19 5:34 ` Timothy D. Witham
2002-12-19 6:43 ` Andrew Morton
2002-12-19 5:45 ` Timothy D. Witham
2002-12-19 7:05 ` Martin J. Bligh
2002-12-19 6:08 ` Timothy D. Witham
2002-12-18 19:50 ` Oliver Xymoron
2002-12-18 17:06 ` Eli Carter
2002-12-18 17:08 ` Andrew Morton
2002-12-18 18:25 ` John Alvord
2002-12-18 18:41 ` Intel P6 vs P7 system call performance Horst von Brand
2002-12-17 19:26 ` Alan Cox
2002-12-17 18:57 ` Ulrich Drepper
2002-12-17 19:10 ` Linus Torvalds
2002-12-17 19:21 ` H. Peter Anvin
2002-12-17 19:37 ` Linus Torvalds
2002-12-17 19:43 ` H. Peter Anvin
2002-12-17 20:07 ` Matti Aarnio
2002-12-17 20:10 ` H. Peter Anvin
2002-12-17 19:59 ` Matti Aarnio
2002-12-17 20:06 ` Ulrich Drepper
2002-12-17 20:35 ` Daniel Jacobowitz
2002-12-18 0:20 ` Linus Torvalds
2002-12-18 0:38 ` Ulrich Drepper
2002-12-18 7:41 ` Kai Henningsen
2002-12-18 13:00 ` Rogier Wolff
2002-12-17 19:47 ` Dave Jones
2002-12-18 12:57 ` Rogier Wolff
2002-12-19 0:14 ` Pavel Machek
2002-12-17 21:38 ` Benjamin LaHaise
2002-12-17 21:41 ` H. Peter Anvin
2002-12-17 18:39 ` Jeff Dike
2002-12-17 19:05 ` Linus Torvalds
2002-12-18 5:34 ` Jeremy Fitzhardinge
2002-12-18 5:38 ` H. Peter Anvin
2002-12-18 15:50 ` Alan Cox
2002-12-18 23:51 ` Pavel Machek
2002-12-13 15:45 ` William Lee Irwin III
2002-12-13 16:49 ` Mike Hayward
2002-12-14 0:55 ` GrandMasterLee
2002-12-14 4:41 ` Mike Dresser
2002-12-14 4:53 ` Mike Dresser
2002-12-14 10:01 ` Dave Jones
2002-12-14 17:48 ` Mike Dresser
2002-12-14 18:36 ` GrandMasterLee
2002-12-15 2:03 ` J.A. Magallon
2002-12-15 21:59 ` Pavel Machek
2002-12-15 22:37 ` William Lee Irwin III
2002-12-15 22:43 ` Pavel Machek
-- strict thread matches above, loose matches on Subject: below --
2003-01-10 18:08 Gabriel Paubert
2002-12-30 13:06 Manfred Spraul
2002-12-30 14:54 ` Andi Kleen
2002-12-22 15:45 Nakajima, Jun
2002-12-22 12:33 Mikael Pettersson
2002-12-22 16:00 ` Jamie Lokier
2002-12-19 18:46 billyrose
2002-12-19 16:10 billyrose
2002-12-19 15:20 bart
2002-12-19 14:57 bart
2002-12-19 14:40 billyrose
2002-12-19 15:11 ` Richard B. Johnson
2002-12-19 13:55 bart
2002-12-19 19:37 ` Linus Torvalds
2002-12-19 22:10 ` Jamie Lokier
2002-12-19 22:16 ` H. Peter Anvin
2002-12-19 22:22 ` Linus Torvalds
2002-12-19 22:26 ` H. Peter Anvin
2002-12-19 22:49 ` Linus Torvalds
2002-12-19 23:30 ` Linus Torvalds
2002-12-22 11:08 ` James H. Cloos Jr.
2002-12-22 18:49 ` Linus Torvalds
2002-12-22 19:07 ` Ulrich Drepper
2002-12-22 19:34 ` Linus Torvalds
2002-12-22 19:51 ` Ulrich Drepper
2002-12-22 20:50 ` James H. Cloos Jr.
2002-12-22 20:56 ` Ulrich Drepper
2002-12-22 19:17 ` Ulrich Drepper
2002-12-20 10:08 ` Ulrich Drepper
2002-12-20 12:06 ` Jamie Lokier
2002-12-20 16:47 ` Linus Torvalds
2002-12-20 23:38 ` Jamie Lokier
2002-12-20 23:50 ` H. Peter Anvin
2002-12-21 0:09 ` Linus Torvalds
2002-12-21 17:18 ` Jamie Lokier
2002-12-21 19:39 ` Linus Torvalds
2002-12-22 2:18 ` Jamie Lokier
2002-12-22 3:11 ` Linus Torvalds
2002-12-22 10:13 ` Ingo Molnar
2002-12-22 15:32 ` Jamie Lokier
2002-12-22 18:53 ` Linus Torvalds
2002-12-23 5:03 ` Linus Torvalds
2002-12-23 7:14 ` Ulrich Drepper
2002-12-23 23:27 ` Petr Vandrovec
2002-12-24 0:22 ` Stephen Rothwell
2002-12-24 4:10 ` Linus Torvalds
2002-12-24 8:05 ` Rogier Wolff
2002-12-24 18:51 ` Linus Torvalds
2002-12-24 21:10 ` Rogier Wolff
2002-12-27 16:14 ` Kai Henningsen
2002-12-24 19:36 ` Linus Torvalds
2002-12-24 20:20 ` Ingo Molnar
2002-12-24 20:27 ` Linus Torvalds
2002-12-24 20:31 ` Ingo Molnar
2002-12-24 20:39 ` Linus Torvalds
2002-12-28 2:05 ` H. Peter Anvin
2002-12-28 2:04 ` H. Peter Anvin
2002-12-26 7:47 ` Pavel Machek
2003-01-10 11:30 ` Gabriel Paubert
2003-01-10 17:11 ` Linus Torvalds
2002-12-22 10:23 ` Ingo Molnar
2002-12-19 13:22 bart
2002-12-19 13:38 ` Dave Jones
2002-12-19 14:22 ` Jamie Lokier
2002-12-19 16:56 ` Dave Jones
2002-12-19 19:29 ` H. Peter Anvin
2002-12-18 23:51 billyrose
2002-12-19 13:10 ` Richard B. Johnson
2002-12-18 12:55 Terje Eggestad
2002-12-18 20:14 ` H. Peter Anvin
2002-12-18 20:25 ` Richard B. Johnson
2002-12-18 20:26 ` H. Peter Anvin
2002-12-18 22:28 ` Jamie Lokier
2002-12-18 22:37 ` Linus Torvalds
2002-12-18 22:57 ` Linus Torvalds
2002-12-20 0:53 ` Daniel Jacobowitz
2002-12-20 1:47 ` Linus Torvalds
2002-12-20 2:37 ` Daniel Jacobowitz
2002-12-18 22:39 ` H. Peter Anvin
2002-12-18 1:30 Nakajima, Jun
2002-12-18 1:54 ` Ulrich Drepper
2002-12-18 3:36 ` H. Peter Anvin
2002-12-18 4:05 ` Linus Torvalds
2002-12-18 4:36 ` H. Peter Anvin
2002-12-18 4:07 ` Linus Torvalds
2002-12-18 4:40 ` Stephen Rothwell
2002-12-18 4:52 ` Linus Torvalds
2002-12-18 4:53 ` Andrew Morton
2002-12-18 19:12 ` Andrew Morton
2002-12-18 23:45 ` Pavel Machek
2002-12-20 3:05 ` Alan Cox
2002-12-20 4:03 ` Stephen Rothwell
2002-12-18 6:00 ` Brian Gerst
2002-12-17 16:32 Manfred Spraul
2002-12-17 17:13 ` Richard B. Johnson
2002-12-17 17:19 ` Richard B. Johnson
2002-12-17 17:37 ` Mikael Pettersson
2002-12-17 16:14 John Reiser
2002-12-17 16:01 John Reiser
[not found] <20021209193649.GC10316@suse.de.suse.lists.linux.kernel>
[not found] ` <Pine.LNX.4.44.0212161639310.1623-100000@penguin.transmeta.com.suse.lists.linux.kernel>
2002-12-17 8:56 ` Andi Kleen
2002-12-17 16:57 ` Linus Torvalds
2002-12-18 5:25 ` Brian Gerst
2002-12-18 6:06 ` Linus Torvalds
2002-12-21 11:24 ` Ingo Molnar
2002-12-21 17:28 ` Jamie Lokier
2002-12-21 16:07 ` Christian Leber
2002-12-15 8:43 scott thomason
2002-12-15 4:06 Albert D. Cahalan
2002-12-15 22:01 ` Pavel Machek
2002-12-16 7:33 ` Albert D. Cahalan
2002-12-16 11:17 ` Pavel Machek
2002-12-16 17:54 ` Mark Mielke
2002-12-16 16:07 ` Jonah Sherman
2002-12-17 4:10 ` David Schwartz
2002-12-17 8:02 ` Helge Hafting
2002-12-16 19:55 ` H. Peter Anvin
2002-12-13 21:52 Margit Schubert-While
2002-12-13 19:32 Dieter Nützel
2002-12-13 17:51 Margit Schubert-While
2002-12-11 12:48 Terje Eggestad
2002-12-11 18:50 ` H. Peter Anvin
2002-12-12 9:42 ` Terje Eggestad
2002-12-12 10:06 ` Arjan van de Ven
2002-12-12 10:31 ` Terje Eggestad
2002-12-12 19:03 ` H. Peter Anvin
2002-12-12 20:36 ` Mark Mielke
2002-12-12 20:56 ` J.A. Magallon
2002-12-12 20:12 ` Zac Hansen
2002-12-13 9:21 ` Terje Eggestad
2002-12-13 15:58 ` Ville Herva
2002-12-13 21:57 ` Terje Eggestad
2002-12-13 22:53 ` H. Peter Anvin
2002-12-12 20:56 ` Vojtech Pavlik
2002-12-09 7:01 Samium Gromoff
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='auqde6$1pv$1@penguin.transmeta.com' \
--to=torvalds@transmeta.com \
--cc=linux-kernel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.