Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 12 Jan 1999 21:41:32 -0600 (CST)
From:      Mohit Aron <aron@cs.rice.edu>
To:        freebsd-hackers@FreeBSD.ORG
Subject:   Performance counters on Pentium Pro's
Message-ID:  <199901130341.VAA14056@cs.rice.edu>

next in thread | raw e-mail | index | archive | help
Hi,
	it seems that the driver for reading Pentium Pro performance counters 
in FreeBSD-2.2.6 doesn't work (it works with a Pentium though). I've already
configured and compiled my kernel with support for the driver and I'm trying to
read the data cache misses. Below are some sample commands with the example 
program provided in /usr/share/examples/perfmon. 

# ./perfmon -o -u -e -l 2 -s 1 72
1: 3298535393077
2: 3298535358033
total: 6597070751110.000000
mean: 3298535375555.000000
clocks (at 299-MHz): 603885293

# ./perfmon -o -u -e -l 2 -s 1 0 
1: 3298534883328
2: 3298534883328
total: 6597069766656.000000
mean: 3298534883328.000000
clocks (at 299-MHz): 603491159

# ./perfmon -o -u -l 2 -s 1 72
1: 3298535418313
2: 3298535807953
total: 6597071226266.000000
mean: 3298535613133.000000
clocks (at 299-MHz): 603927055


Notice that using event 72 (or 0x48 as suggested in 
/usr/include/machine/perfmon.h for reading data cache misses) seems to throw
out a large number. This number seems to remain relatively unchanged when I 
use other events (the 2nd example above uses 0). In fact the output of the
program even seems to be similar when I don't use the '-e' flag which is 
supposed to count the number of events rather than their duration.


I've also contacted the person who wrote the code (Garrett Wollman) and he
agrees that he had some questionable results with Pentium Pro in the past but
never tracked them down. Does someone have a fix for this problem ?



- Mohit

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199901130341.VAA14056>