xref: /freebsd/share/man/man9/LOCK_PROFILING.9 (revision 90aad5d3e44d137ce80d2ec257de07c7c1367bd3)
1.\"-
2.\" Copyright (c) 2004 Dag-Erling Coïdan Smørgrav
3.\" Copyright (c) 2005 Robert N. M. Watson
4.\" Copyright (c) 2006 Kip Macy
5.\" All rights reserved.
6.\"
7.\" Redistribution and use in source and binary forms, with or without
8.\" modification, are permitted provided that the following conditions
9.\" are met:
10.\" 1. Redistributions of source code must retain the above copyright
11.\"    notice, this list of conditions and the following disclaimer.
12.\" 2. Redistributions in binary form must reproduce the above copyright
13.\"    notice, this list of conditions and the following disclaimer in the
14.\"    documentation and/or other materials provided with the distribution.
15.\" 3. The name of the author may not be used to endorse or promote products
16.\"    derived from this software without specific prior written permission.
17.\"
18.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28.\" SUCH DAMAGE.
29.\"
30.Dd March 7, 2012
31.Dt LOCK_PROFILING 9
32.Os
33.Sh NAME
34.Nm LOCK_PROFILING
35.Nd kernel lock profiling support
36.Sh SYNOPSIS
37.Cd "options LOCK_PROFILING"
38.Sh DESCRIPTION
39The
40.Dv LOCK_PROFILING
41kernel option adds support for measuring and reporting lock use and
42contention statistics.
43These statistics are collated by
44.Dq acquisition point .
45Acquisition points are
46distinct places in the kernel source code (identified by source file
47name and line number) where a lock is acquired.
48.Pp
49For each acquisition point, the following statistics are accumulated:
50.Bl -bullet
51.It
52The longest time the lock was ever continuously held after being
53acquired at this point.
54.It
55The total time the lock was held after being acquired at this point.
56.It
57The total time that threads have spent waiting to acquire the lock.
58.It
59The total number of non-recursive acquisitions.
60.It
61The total number of times the lock was already held by another thread
62when this point was reached, requiring a spin or a sleep.
63.It
64The total number of times another thread tried to acquire the lock
65while it was held after having been acquired at this point.
66.El
67.Pp
68In addition, the average hold time and average wait time are derived
69from the total hold time
70and total wait time respectively and the number of acquisitions.
71.Pp
72The
73.Dv LOCK_PROFILING
74kernel option also adds the following
75.Xr sysctl 8
76variables to control and monitor the profiling code:
77.Bl -tag -width indent
78.It Va debug.lock.prof.enable
79Enable or disable the lock profiling code.
80This defaults to 0 (off).
81.It Va debug.lock.prof.reset
82Reset the current lock profiling buffers.
83.It Va debug.lock.prof.stats
84The actual profiling statistics in plain text.
85The columns are as follows, from left to right:
86.Bl -tag -width ".Va cnt_hold"
87.It Va max
88The longest continuous hold time in microseconds.
89.It Va wait_max
90The longest continuous wait time in microseconds.
91.It Va total
92The total (accumulated) hold time in microseconds.
93.It Va wait_total
94The total (accumulated) wait time in microseconds.
95.It Va count
96The total number of acquisitions.
97.It Va avg
98The average hold time in microseconds, derived from the total hold time
99and the number of acquisitions.
100.It Va wait_avg
101The average wait time in microseconds, derived from the total wait time
102and the number of acquisitions.
103.It Va cnt_hold
104The number of times the lock was held and another thread attempted to
105acquire the lock.
106.It Va cnt_lock
107The number of times the lock was already held when this point was
108reached.
109.It Va name
110The name of the acquisition point, derived from the source file name
111and line number, followed by the name of the lock in parentheses.
112.El
113.It Va debug.lock.prof.rejected
114The number of acquisition points that were ignored after the table
115filled up.
116.It Va debug.lock.prof.skipspin
117Disable or enable the lock profiling code for the spin locks.
118This defaults to 0 (do profiling for the spin locks).
119.It Va debug.lock.prof.skipcount
120Do sampling approximately every N lock acquisitions.
121.El
122.Sh SEE ALSO
123.Xr sysctl 8 ,
124.Xr mutex 9
125.Sh HISTORY
126Mutex profiling support appeared in
127.Fx 5.0 .
128Generalized lock profiling support appeared in
129.Fx 7.0 .
130.Sh AUTHORS
131.An -nosplit
132The
133.Nm MUTEX_PROFILING
134code was written by
135.An Eivind Eklund Aq Mt eivind@FreeBSD.org ,
136.An Dag-Erling Sm\(/orgrav Aq Mt des@FreeBSD.org
137and
138.An Robert Watson Aq Mt rwatson@FreeBSD.org .
139The
140.Nm
141code was written by
142.An Kip Macy Aq Mt kmacy@FreeBSD.org .
143This manual page was written by
144.An Dag-Erling Sm\(/orgrav Aq Mt des@FreeBSD.org .
145.Sh NOTES
146The
147.Dv LOCK_PROFILING
148option increases the size of
149.Vt "struct lock_object" ,
150so a kernel built with that option will not work with modules built
151without it.
152.Pp
153The
154.Dv LOCK_PROFILING
155option also prevents inlining of the mutex code, which can result in a
156fairly severe performance penalty.
157This is, however, not always the case.
158.Dv LOCK_PROFILING
159can introduce a substantial performance overhead that is easily
160monitorable using other profiling tools, so combining profiling tools
161with
162.Dv LOCK_PROFILING
163is not recommended.
164.Pp
165Measurements are made and stored in nanoseconds using
166.Xr nanotime 9 ,
167(on architectures without a synchronized TSC) but are presented in microseconds.
168This should still be sufficient for the locks one would be most
169interested in profiling (those that are held long and/or acquired
170often).
171.Pp
172.Dv LOCK_PROFILING
173should generally not be used in combination with other debugging options, as
174the results may be strongly affected by interactions between the features.
175In particular,
176.Dv LOCK_PROFILING
177will report higher than normal
178.Xr uma 9
179lock contention when run with
180.Dv INVARIANTS
181due to extra locking that occurs when
182.Dv INVARIANTS
183is present; likewise, using it in combination with
184.Dv WITNESS
185will lead to much higher lock hold times and contention in profiling output.
186