Subject: [linux-audio-dev] [PATCH] latency-profiling for 2.4.0-test3-pre2
From: Roger Larsson (roger.larsson_AT_norran.net)
Date: Mon Jul 03 2000 - 20:17:00 EEST
Hi all,
Since there has been much talk about latencies lately I updated my
latency profiling patch to the most recent kernel.
Note: I have simplified usage.
Output format (changed):
Latency <dt>ms PID <pid> % <name>
Trace; [<IP>] [<IP>]
dt
latency time from rescheduling request (need_resched = 1) until
actually done.
pid
process id
IPs
instruction pointers, decode as any (sampled!) kernel stack
trace.
Note: there might be none since they are sampled with HZ
name
the running process name. But the problem really is inside the
kernel. A process
running in user space would have been (un-)scheduled
immediately!!!
Example (output from dmesg):
Latency 708ms PID 267 % modprobe
Trace: [<c0205d7e>] [<c0205d7e>] [<c0205d7e>] [<c0205d7e>]
[<c0205d7e>]
[<c0205d7e>] [<c0205d7e>] [<c0205d80>] [<c0205d7e>] [<c0205d7e>]
With the modified format you may get symbols by:
dmesg | ksymoops
Trace: c0205d7e <__rdtsc_delay+e/18>
Trace: c0205d7e <__rdtsc_delay+e/18>
Trace: c0205d7e <__rdtsc_delay+e/18>
Trace: c0205d7e <__rdtsc_delay+e/18>
Trace: c0205d7e <__rdtsc_delay+e/18>
Trace: c0205d7e <__rdtsc_delay+e/18>
Trace: c0205d7e <__rdtsc_delay+e/18>
Trace: c0205d80 <__rdtsc_delay+10/18>
Trace: c0205d7e <__rdtsc_delay+e/18>
Trace: c0205d7e <__rdtsc_delay+e/18>
But again, note that this is NOT a stack trace. The trace is in sample
order with HZ (10 ms) run time in between.
/RogerL
-- Home page: http://www.norran.net/nra02596/
This archive was generated by hypermail 2b28 : Mon Jul 03 2000 - 20:55:27 EEST