Messages in this thread |  | | Date | Sun, 12 Jan 1997 03:40:29 -0500 (EST) | From | Jon Lewis <> | Subject | need profiling help |
| |
I've been having some strange problems that appear to be IP filtering causing a system to slow down to the point that interrupts aren't serviced in a timely manor. This prompted me to start playing with kernel profiling on my home system in preparation for doing it on the troubled system.
I started by using the modified profiler program Linus posted about 2 months ago, but the output wasn't too illuminating:
port-28:~# ./kprofile do_no_page Floating point exception
I booted with profile=2 and have /proc/profile: -rw-r--r-- 1 root root 812812 Jan 12 03:24 /proc/profile
I have a symlink from /System.map that points to my current kernel's System.map... /System.map -> /kernels/System.map-2.0.27
What step have I missed?
I then tried the readprofile-2.0 package, and it appears to work:
port-28:/home/jlewis# readprofile -m /System.map | sort -rn | head -5 1767 total 0.0022 166 do_wp_page 0.2530 96 read_profile 0.3750 82 do_no_page 0.0949 61 con_write 0.0109
One thing that's odd though, is Linus suggested doing: echo >/proc/profile to clear the profile info between benchmarks. If I do that, then /proc/profile stays 0 length and readprofile won't work until I reboot.
Also, what about functions in modules? Can the insmod load maps be merged with the System.map readprofile uses...or does the kernel have to be as unmodular as possible for profiling? One of the systems I'm interested in profiling has a driver that has to be used as a module. ------------------------------------------------------------------ Jon Lewis <jlewis@fdt.net> | Unsolicited commercial e-mail will Network Administrator | be proof-read for $199/hr. ________Finger jlewis@inorganic5.fdt.net for PGP public key_______
|  |