NetBSD Documentation: Kernel Profiling HOWTO
This is a description of how kernel profiling works, and how to
use it. This documentation was written by Matthias Drochner
How does it work (top)
Two different sets of data about the behaviour of the profiled code are recorded independently: the frequency of every function call (call graph profiling) and the time spent in each function, estimated by the probability of the program counter being within the function when sampled at a random time which is in turn estimated by the fraction of profiling timer interrupts occurring while the function in question is executing. The gprof(1) utility interprets the data. There are some limitations caused by the missing correlation between the two data sets which are noted in the BUGS section of its man page.
Kernel profiling and user program profiling are mostly similar; there are only small differences in the way the profiling data are accessed and how the profiling is controlled.
The data related to kernel profiling are located
within a global structure
_gmonparam which is
during system initialization.
The user level control program kgmon(8) uses
sysctl(3) calls for
control and data access, and partly kvm(3) accesses (also in the
standard case where a live kernel is profiled!).
Call graph recording (top)
The profiling flag (
-pg) causes the
compiler to issue a call to
mcount() on every
This is dispatched by machine specific glue to
which is implemented in
the address the function was called from, and
the address of the called function itself.
For every (
encountered during the profiling, a
struct tostruct is
allocated from the array pointed to by
The entries are simply allocated from the beginning
to the end in the order of first use. Some magic
the size of the array from the kernel's text size -- it
seems that this is kind of an "educated guess".
entries contain the address of the called function
together with a histogramming counter. Entries belonging to the
same calling address form a linked list. The list
heads (ie the index of the first entry within the
array belonging to a particular calling address) are located
in a second data array
which is indexed by the calling address
divided by some value (which should not be larger than the
minimal distance of two calls within the code - see also the
Note that for standard function calls there is only
that the typical list consists of one member only.
Statistical profiling (top)
If profiling is started, a profiling timer interrupt
is set up which calls
This should be a timer independent of the normal system
clock to avoid interferences with functions running
synchronously to system clock ticks.
statclock() is used for
both user program and kernel profiling.
The program counter at the time of the interruption,
divided by some value again, is used as index into
_gmonparam.kcount and the
corresponding cell is incremented.
How to use it (top)
- Build and boot a profiling kernel. For this, use
-pflag to config(1). Otherwise, there is nothing special. The profiling kernel will print a message like:
Profiling kernel, textsize=1611256 [f0100000..f02895f8]after autoconfiguration, before the root filesystem is mounted.
- Make sure
/netbsdis the currently running kernel, or use the
-Noption in the kgmon(8) invocations below.
- Start profiling with
- Run some applications which stress the part of the system you want to investigate.
- Stop profiling with
kgmon -pto write the collected data into a file
- Before measuring different things, reset the
in-kernel profiling data buffers with
- To get a human readable interpretation from
% gprof /netbsd gmon.out > gprof.txt
- Try to make sense of the data.
Back to NetBSD Documentation: Kernel