usr.sbin/pmcstat/pmcstat.8

.\" Copyright (c) 2003-2008 Joseph Koshy
.\" Copyright (c) 2007 The FreeBSD Foundation
.\" All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\"    notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\"    notice, this list of conditions and the following disclaimer in the
.\"    documentation and/or other materials provided with the distribution.
.\"
.\" This software is provided by Joseph Koshy ``as is'' and
.\" any express or implied warranties, including, but not limited to, the
.\" implied warranties of merchantability and fitness for a particular purpose
.\" are disclaimed.  in no event shall Joseph Koshy be liable
.\" for any direct, indirect, incidental, special, exemplary, or consequential
.\" damages (including, but not limited to, procurement of substitute goods
.\" or services; loss of use, data, or profits; or business interruption)
.\" however caused and on any theory of liability, whether in contract, strict
.\" liability, or tort (including negligence or otherwise) arising in any way
.\" out of the use of this software, even if advised of the possibility of
.\" such damage.
.\"
.Dd May 31, 2023
.Dt PMCSTAT 8
.Os
.Sh NAME
.Nm pmcstat
.Nd "performance measurement with performance monitoring hardware"
.Sh SYNOPSIS
.Nm
.Op Fl A
.Op Fl C
.Op Fl D Ar pathname
.Op Fl E
.Op Fl F Ar pathname
.Op Fl G Ar pathname
.Op Fl I
.Op Fl L
.Op Fl M Ar mapfilename
.Op Fl N
.Op Fl O Ar logfilename
.Op Fl P Ar event-spec
.Op Fl R Ar logfilename
.Op Fl S Ar event-spec
.Op Fl T
.Op Fl U
.Op Fl W
.Op Fl a Ar pathname
.Op Fl c Ar cpu-spec
.Op Fl d
.Op Fl e
.Op Fl f Ar pluginopt
.Op Fl g
.Op Fl i Ar lwp
.Op Fl l Ar secs
.Op Fl m Ar pathname
.Op Fl n Ar rate
.Op Fl o Ar outputfile
.Op Fl p Ar event-spec
.Op Fl q
.Op Fl r Ar fsroot
.Op Fl s Ar event-spec
.Op Fl t Ar process-spec
.Op Fl u Ar event-spec
.Op Fl v
.Op Fl w Ar secs
.Op Fl z Ar graphdepth
.Op Ar command Op Ar args
.Sh DESCRIPTION
The
.Nm
utility measures system performance using the facilities provided by
.Xr hwpmc 4 .
.Pp
The
.Nm
utility can measure both hardware events seen by the system as a
whole, and those seen when a specified set of processes are executing
on the system's CPUs.
If a specific set of processes is being targeted (for example,
if the
.Fl t Ar process-spec
option is specified, or if a command line is specified using
.Ar command ) ,
then measurement occurs till
.Ar command
exits, or till all target processes specified by the
.Fl t Ar process-spec
options exit, or till the
.Nm
utility is interrupted by the user.
If a specific set of processes is not targeted for measurement, then
.Nm
will perform system-wide measurements till interrupted by the
user.
.Pp
A given invocation of
.Nm
can mix allocations of system-mode and process-mode PMCs, of both
counting and sampling flavors.
The values of all counting PMCs are printed in human readable form
at regular intervals by
.Nm .
The format of
.Nm Ns 's
human-readable textual output is not stable, and could change
in the future.
The output of sampling PMCs may be configured to go to a log file for
subsequent offline analysis, or, at the expense of greater
overhead, may be configured to be printed in text form on the fly.
.Pp
Hardware events to measure are specified to
.Nm
using event specifier strings
.Ar event-spec .
The syntax of these event specifiers is machine dependent and is
documented in
.Xr pmc 3 .
.Pp
A process-mode PMC may be configured to be inheritable by the target
process' current and future children.
.Sh OPTIONS
The following options are available:
.Bl -tag -width indent
.It Fl A
Skip symbol lookup and display address instead.
.It Fl C
Toggle between showing cumulative or incremental counts for
subsequent counting mode PMCs specified on the command line.
The default is to show incremental counts.
.It Fl D Ar pathname
Create files with per-program samples in the directory named
by
.Ar pathname .
The default is to create these files in the current directory.
.It Fl E
Toggle showing per-process counts at the time a tracked process
exits for subsequent process-mode PMCs specified on the command line.
This option is useful for mapping the performance characteristics of a
complex pipeline of processes when used in conjunction with the
.Fl d
option.
The default is to not to enable per-process tracking.
.It Fl F Ar pathname
Print calltree (Kcachegrind) information to file
.Ar pathname .
If argument
.Ar pathname
is a
.Dq Li -
this information is sent to the output file specified by the
.Fl o
option.
.It Fl G Ar pathname
Print callchain information to file
.Ar pathname .
If argument
.Ar pathname
is a
.Dq Li -
this information is sent to the output file specified by the
.Fl o
option.
.It Fl I
Show the offset of the instruction pointer into the symbol.
.It Fl L
List all event names.
.It Fl M Ar mapfilename
Write the mapping between executable objects encountered in the event
log and the abbreviated pathnames used for
.Xr gprof 1
profiles to file
.Ar mapfilename .
If this option is not specified, mapping information is not written.
Argument
.Ar mapfilename
may be a
.Dq Li -
in which case this mapping information is sent to the output
file configured by the
.Fl o
option.
.It Fl N
Toggle capturing callchain information for subsequent sampling PMCs.
The default is for sampling PMCs to capture callchain information.
.It Fl O Ar logfilename
Send logging output to file
.Ar logfilename .
If
.Ar logfilename
is of the form
.Ar hostname Ns : Ns Ar port ,
where
.Ar hostname
does not start with a
.Ql \&.
or a
.Ql / ,
then
.Nm
will open a network socket to host
.Ar hostname
on port
.Ar port .
.Pp
If the
.Fl O
option is not specified and one of the logging options is requested,
then
.Nm
will print a textual form of the logged events to the configured
output file.
.It Fl P Ar event-spec
Allocate a process mode sampling PMC measuring hardware events
specified in
.Ar event-spec .
.It Fl R Ar logfilename
Perform offline analysis using sampling data in file
.Ar logfilename .
.It Fl S Ar event-spec
Allocate a system mode sampling PMC measuring hardware events
specified in
.Ar event-spec .
.It Fl T
Use a
.Xr top 1 Ns -like
mode for sampling PMCs.
The following hotkeys can be used:
.Pp
.Bl -tag -compact -width "Ctrl+a" -offset 4n
.It Ic A
Toggle symbol resolution
.Sm off
.It Ic Ctrl + a
.Sm on
Switch to accumulative mode
.Sm off
.It Ic Ctrl + d
.Sm on
Switch to delta mode
.It Ic f
Represent the
.Dq f
cost under
threshold as a dot (calltree only)
.It Ic I
Toggle showing offsets into symbols
.It Ic m
Merge PMCs
.It Ic n
Change view
.It Ic p
Show next PMC
.It Ic q
Quit
.It Ic Space
Pause
.El
.It Fl U
Toggle capturing user-space call traces while in kernel mode.
The default is for sampling PMCs to capture user-space callchain information
while in user-space mode, and kernel callchain information while in kernel mode.
.It Fl W
Toggle logging the incremental counts seen by the threads of a
tracked process each time they are scheduled on a CPU.
This is an experimental feature intended to help analyse the
dynamic behaviour of processes in the system.
It may incur substantial overhead if enabled.
The default is for this feature to be disabled.
.It Fl a Ar pathname
Perform a symbol and file:line lookup for each address in each
callgraph and save the output to
.Ar pathname .
Unlike
.Fl m
that only resolves the first symbol in the graph, this resolves
every node in the callgraph, or prints out addresses if no
lookup information is available.
This option requires the
.Fl R
option to read in samples that were previously collected and
saved with the
.Fl O
option.
.It Fl c Ar cpu-spec
Set the cpus for subsequent system mode PMCs specified on the
command line to
.Ar cpu-spec .
Argument
.Ar cpu-spec
is a comma separated list of CPU numbers, or the literal
.Sq *
denoting all available CPUs.
The default is to allocate system mode PMCs on all available
CPUs.
.It Fl d
Toggle between process mode PMCs measuring events for the target
process' current and future children or only measuring events for
the target process.
The default is to measure events for the target process alone.
(it has to be passed in the command line prior to
.Fl p ,
.Fl s ,
.Fl P ,
or
.Fl S ) .
.It Fl e
Specify that the gprof profile files will use a wide history counter.
These files are produced in a format compatible with
.Xr gprof 1 .
However, other tools that cannot fully parse a BSD-style
gmon header might be unable to correctly parse these files.
.It Fl f Ar pluginopt
Pass option string to the active plugin.
.br
threshold=<float> do not display cost under specified value (Top).
.br
skiplink=0|1 replace node with cost under threshold by a dot (Top).
.It Fl g
Produce profiles in a format compatible with
.Xr gprof 1 .
A separate profile file is generated for each executable object
encountered.
Profile files are placed in sub-directories named by their PMC
event name.
.It Fl i Ar lwp
Filter on thread ID
.Ar lwp ,
which you can get from
.Xr ps 1
.Fl o
.Li lwp .
.It Fl l Ar secs
Set system-wide performance measurement duration for
.Ar secs
seconds.
The argument
.Ar secs
may be a fractional value.
.It Fl m Ar pathname
Print the sampled PCs with the name, the start and ending addresses
of the function within they live.
The
.Ar pathname
argument is mandatory and indicates where the information will be stored.
If argument
.Ar pathname
is a
.Dq Li -
this information is sent to the output file specified by the
.Fl o
option.
This option requires the
.Fl R
option to read in samples that were previously collected and
saved with the
.Fl O
option.
.It Fl n Ar rate
Set the default sampling rate for subsequent sampling mode
PMCs specified on the command line.
The default is to configure PMCs to sample the CPU's instruction
pointer every 65536 events.
.It Fl o Ar outputfile
Send counter readings and textual representations of logged data
to file
.Ar outputfile .
The default is to send output to
.Pa stderr
when collecting live data and to
.Pa stdout
when processing a pre-existing logfile.
.It Fl p Ar event-spec
Allocate a process mode counting PMC measuring hardware events
specified in
.Ar event-spec .
.It Fl q
Decrease verbosity.
.It Fl r Ar fsroot
Set the top of the filesystem hierarchy under which executables
are located to argument
.Ar fsroot .
The default is
.Pa / .
.It Fl s Ar event-spec
Allocate a system mode counting PMC measuring hardware events
specified in
.Ar event-spec .
.It Fl t Ar process-spec
Attach process mode PMCs to the processes named by argument
.Ar process-spec .
Argument
.Ar process-spec
may be a non-negative integer denoting a specific process id, or a
regular expression for selecting processes based on their command names.
.It Fl u Ar event-spec
Provide short description of event.
.It Fl v
Increase verbosity.
.It Fl w Ar secs
Print the values of all counting mode PMCs or sampling mode PMCs
for top mode every
.Ar secs
seconds.
The argument
.Ar secs
may be a fractional value.
The default interval is 5 seconds.
.It Fl z Ar graphdepth
When printing system-wide callgraphs, limit callgraphs to the depth
specified by argument
.Ar graphdepth .
.El
.Pp
If
.Ar command
is specified, it is executed using
.Xr execvp 3 .
.Sh EXAMPLES
To perform system-wide statistical sampling on an AMD Athlon CPU with
samples taken every 32768 instruction retirals and data being sampled
to file
.Pa sample.stat ,
use:
.Dl "pmcstat -O sample.stat -n 32768 -S k7-retired-instructions"
.Pp
To execute
.Nm firefox
and measure the number of data cache misses suffered
by it and its children every 12 seconds on an AMD Athlon, use:
.Dl "pmcstat -d -w 12 -p k7-dc-misses firefox"
.Pp
To measure instructions retired for all processes named
.Dq emacs
use:
.Dl "pmcstat -t '^emacs$' -p instructions"
.Pp
To measure instructions retired for processes named
.Dq emacs
for a period of 10 seconds use:
.Dl "pmcstat -t '^emacs$' -p instructions sleep 10"
.Pp
To count instruction tlb-misses on CPUs 0 and 2 on a Intel
Pentium Pro/Pentium III SMP system use:
.Dl "pmcstat -c 0,2 -s p6-itlb-miss"
.Pp
To collect profiling information for a specific process with pid 1234
based on instruction cache misses seen by it use:
.Dl "pmcstat -P ic-misses -t 1234 -O /tmp/sample.out"
.Pp
To perform system-wide sampling on all configured processors
based on processor instructions retired use:
.Dl "pmcstat -S instructions -O /tmp/sample.out"
If callgraph capture is not desired use:
.Dl "pmcstat -N -S instructions -O /tmp/sample.out"
.Pp
To send the generated event log to a remote machine use:
.Dl "pmcstat -S instructions -O remotehost:port"
On the remote machine, the sample log can be collected using
.Xr nc 1 :
.Dl "nc -l remotehost port > /tmp/sample.out"
.Pp
To generate
.Xr gprof 1
compatible profiles from a sample file use:
.Dl "pmcstat -R /tmp/sample.out -g"
.Pp
To print a system-wide profile with callgraphs to file
.Pa "foo.graph"
use:
.Dl "pmcstat -R /tmp/sample.out -G foo.graph"
.Sh DIAGNOSTICS
If option
.Fl v
is specified,
.Nm
may issue the following diagnostic messages:
.Bl -diag
.It "#callchain/dubious-frames"
The number of callchain records that had an
.Dq impossible
value for a return address.
.It "#exec handling errors"
The number of
.Xr exec 2
events in the log file that named executables that could not be
analyzed.
.It "#exec/elf"
The number of
.Xr exec 2
events that named ELF executables.
.It "#exec/unknown"
The number of
.Xr exec 2
events that named executables with unrecognized formats.
.It "#samples/total"
The total number of samples in the log file.
.It "#samples/unclaimed"
The number of samples that could not be correlated to a known
executable object (i.e., to an executable, shared library, the
kernel or the runtime loader).
.It "#samples/unknown-object"
The number of samples that were associated with an executable
with an unrecognized object format.
.El
.Pp
.Ex -std
.Sh COMPATIBILITY
Due to the limitations of the
.Pa gmon.out
file format,
.Xr gprof 1
compatible profiles generated by the
.Fl g
option do not contain information about calls that cross executable
boundaries.
The generated
.Pa gmon.out
files are also only meaningful for native executables.
.Sh SEE ALSO
.Xr gprof 1 ,
.Xr nc 1 ,
.Xr execvp 3 ,
.Xr pmc 3 ,
.Xr pmclog 3 ,
.Xr hwpmc 4 ,
.Xr pmccontrol 8 ,
.Xr sysctl 8
.Sh HISTORY
The
.Nm
utility first appeared in
.Fx 6.0 .
.Sh AUTHORS
.An Joseph Koshy Aq Mt jkoshy@FreeBSD.org
.Sh BUGS
The
.Nm
utility cannot yet analyse
.Xr hwpmc 4
logs generated by non-native architectures.