1 .\" Copyright (c) 2003-2008 Joseph Koshy. All rights reserved.
3 .\" Redistribution and use in source and binary forms, with or without
4 .\" modification, are permitted provided that the following conditions
6 .\" 1. Redistributions of source code must retain the above copyright
7 .\" notice, this list of conditions and the following disclaimer.
8 .\" 2. Redistributions in binary form must reproduce the above copyright
9 .\" notice, this list of conditions and the following disclaimer in the
10 .\" documentation and/or other materials provided with the distribution.
12 .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
13 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
14 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
15 .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
16 .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
17 .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
18 .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
19 .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
20 .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
21 .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 .Nd library for accessing hardware performance monitoring counters
39 provides a programming interface that allows applications to use
40 hardware performance counters to gather performance data about
41 specific processes or for the system as a whole.
42 The library is implemented using the lower-level facilities offered by
47 Performance monitoring counters (PMCs) are represented by the library
48 using a software abstraction.
51 PMCs can have two scopes:
55 These PMCs measure events in a whole-system manner, i.e., independent
56 of the currently executing thread.
57 System scope PMCs are allocated on specific CPUs and do not
59 Non-privileged process are allowed to allocate system scope PMCs if the
62 .Va security.bsd.unprivileged_syspmcs
66 These PMCs only measure hardware events when the processes they are
67 attached to are executing on a CPU.
68 In an SMP system, process scope PMCs migrate between CPUs along with
69 their target processes.
72 Orthogonal to PMC scope, PMCs may be allocated in one of two
76 Counting PMCs measure events according to their scope
78 The application needs to explicitly read these counters
79 to retrieve their value.
81 Sampling PMCs cause the CPU to be periodically interrupted
82 and information about its state of execution to be collected.
83 Sampling PMCs are used to profile specific processes and kernel
84 threads or to profile the system as a whole.
87 The scope and operational mode for a software PMC are specified at
89 An application is allowed to allocate multiple PMCs subject
90 to availability of hardware resources.
92 The library uses human-readable strings to name the event being
94 The syntax used for specifying a hardware event along with additional
95 event specific qualifiers (if any) is described in detail in section
96 .Sx "EVENT SPECIFIERS"
99 PMCs are associated with the process that allocated them and
100 will be automatically reclaimed by the system when the process exits.
101 Additionally, process-scope PMCs have to be attached to one or more
102 target processes before they can perform measurements.
103 A process-scope PMC may be attached to those target processes
104 that its owner process would otherwise be permitted to debug.
105 An owner process may attach PMCs to itself allowing
106 it to measure its own behavior.
107 Additionally, on some machine architectures, such self-attached PMCs
108 may be read cheaply using specialized instructions supported by the
111 Certain kinds of PMCs require that a log file be configured before
116 System scope sampling PMCs.
118 Process scope sampling PMCs.
120 Process scope counting PMCs that have been configured to report PMC
121 readings on process context switches or process exits.
124 Up to one log file may be configured per owner process.
125 Events logged to a log file may be subsequently analyzed using the
129 The CPUs known to the PMC library are named by the
130 .Vt "enum pmc_cputype"
132 Supported CPUs include:
134 .Bl -tag -width "Li PMC_CPU_INTEL_CORE2" -compact
135 .It Li PMC_CPU_AMD_K7
138 .It Li PMC_CPU_AMD_K8
141 .It Li PMC_CPU_INTEL_ATOM
144 CPUs and other CPUs conforming to version 3 of the
146 performance measurement architecture.
147 .It Li PMC_CPU_INTEL_CORE
152 CPUs, and other CPUs conforming to version 1 of the
154 performance measurement architecture.
155 .It Li PMC_CPU_INTEL_CORE2
161 CPUs, and other CPUs conforming to version 2 of the
163 performance measurement architecture.
164 .It Li PMC_CPU_INTEL_P5
168 .It Li PMC_CPU_INTEL_P6
172 .It Li PMC_CPU_INTEL_PII
173 .Tn "Intel Pentium II"
175 .It Li PMC_CPU_INTEL_PIII
176 .Tn "Intel Pentium III"
178 .It Li PMC_CPU_INTEL_PIV
179 .Tn "Intel Pentium 4"
181 .It Li PMC_CPU_INTEL_PM
182 .Tn "Intel Pentium M"
186 PMC supported by this library are named by the
189 Supported PMC kinds include:
191 .Bl -tag -width "Li PMC_CLASS_IAF" -compact
193 Fixed function hardware counters presents in CPUs conforming to the
195 performance measurement architecture version 2 and later.
197 Programmable hardware counters present in CPUs conforming to the
199 performance measurement architecture version 1 and later.
201 Programmable hardware counters present in
205 Programmable hardware counters present in
209 Programmable hardware counters present in
210 .Tn "Intel Pentium 4"
213 Programmable hardware counters present in
218 Programmable hardware counters present in
228 The timestamp counter on i386 and amd64 architecture CPUs.
229 .It Li PMC_CLASS_SOFT
233 Capabilities of performance monitoring hardware are denoted using
237 Supported capabilities include:
239 .Bl -tag -width "Li PMC_CAP_INTERRUPT" -compact
240 .It Li PMC_CAP_CASCADE
241 The ability to cascade counters.
243 The ability to count negated to asserted transitions of the hardware
244 conditions being probed for.
245 .It Li PMC_CAP_INTERRUPT
246 The ability to interrupt the CPU.
247 .It Li PMC_CAP_INVERT
248 The ability to invert the sense of the hardware conditions being
250 .It Li PMC_CAP_PRECISE
251 The ability to perform precise sampling.
252 .It Li PMC_CAP_QUALIFIER
253 The hardware allows monitored to be further qualified in some
254 system dependent way.
256 The ability to read from performance counters.
257 .It Li PMC_CAP_SYSTEM
258 The ability to restrict counting of hardware events to when the CPU is
259 running privileged code.
260 .It Li PMC_CAP_THRESHOLD
261 The ability to ignore simultaneous hardware events below a
262 programmable threshold.
264 The ability to restrict counting of hardware events to those when the
265 CPU is running unprivileged code.
267 The ability to write to performance counters.
269 .Ss CPU Naming Conventions
270 CPUs are named using small integers from zero up to, but
271 excluding, the value returned by function
273 On platforms supporting sparsely numbered CPUs not all the numbers in
274 this range will denote valid CPUs.
275 Operations on non-existent CPUs will return an error.
276 .Ss Functional Grouping of the API
277 This section contains a brief overview of the available functionality
279 Each function listed here is described further in its own manual page.
282 .Bl -tag -width 6n -compact
283 .It Fn pmc_disable , Fn pmc_enable
284 Administratively disable (enable) specific performance monitoring
286 Counters that are disabled will not be available to applications to
289 .It "Convenience Functions"
290 .Bl -tag -width 6n -compact
291 .It Fn pmc_event_names_of_class
292 Returns a list of event names supported by a given PMC type.
293 .It Fn pmc_name_of_capability
296 flag to a human-readable string.
297 .It Fn pmc_name_of_class
300 constant to a human-readable string.
301 .It Fn pmc_name_of_cputype
302 Return a human-readable name for a CPU type.
303 .It Fn pmc_name_of_disposition
304 Return a human-readable string describing a PMC's disposition.
305 .It Fn pmc_name_of_event
306 Convert a numeric event code to a human-readable string.
307 .It Fn pmc_name_of_mode
310 constant to a human-readable name.
311 .It Fn pmc_name_of_state
312 Return a human-readable string describing a PMC's current state.
314 .It "Library Initialization"
315 .Bl -tag -width 6n -compact
317 Initialize the library.
318 This function must be called before any other library function.
320 .It "Log File Handling"
321 .Bl -tag -width 6n -compact
322 .It Fn pmc_configure_logfile
323 Configure a log file for
325 to write logged events to.
326 .It Fn pmc_flush_logfile
327 Flush all pending log data in
330 .It Fn pmc_close_logfile
331 Flush all pending log data and close
335 Append arbitrary user data to the current log file.
338 .Bl -tag -width 6n -compact
339 .It Fn pmc_allocate , Fn pmc_release
340 Allocate (free) a PMC.
341 .It Fn pmc_attach , Fn pmc_detach
342 Attach (detach) a process scope PMC to a target.
343 .It Fn pmc_read , Fn pmc_write , Fn pmc_rw
344 Read (write) a value from (to) a PMC.
345 .It Fn pmc_start , Fn pmc_stop
346 Start (stop) a software PMC.
348 Set the reload value for a sampling PMC.
351 .Bl -tag -width 6n -compact
352 .It Fn pmc_capabilities
353 Retrieve the capabilities for a given PMC.
355 Retrieve information about the CPUs and PMC hardware present in the
357 .It Fn pmc_get_driver_stats
358 Retrieve statistics maintained by
361 Determine the greatest possible CPU number on the system.
363 Return the number of hardware PMCs present in a given CPU.
365 Return information about the state of a given CPU's PMCs.
367 Determine the width of a hardware counter in bits.
369 .It "x86 Architecture Specific API"
370 .Bl -tag -width 6n -compact
372 Returns the processor model specific register number
375 Applications may then use the x86
377 instruction to directly read the contents of the PMC.
380 .Ss Signal Handling Requirements
381 Applications using PMCs are required to handle the following signals:
382 .Bl -tag -width ".Dv SIGBUS"
386 module is unloaded using
388 processes that have PMCs allocated to them will be sent a
394 driver will send a PMC owning process a
399 If any process-mode PMC allocated by it loses all its
402 If the driver encounters an error when writing log data to a
404 This error may be retrieved by a subsequent call to
405 .Fn pmc_flush_logfile .
408 .Ss Typical Program Flow
411 An application would first invoke function
413 to allow the library to initialize itself.
415 Signal handling would then be set up.
417 Next the application would allocate the PMCs it desires using function
420 Initial values for PMCs may be set using function
423 If a log file is necessary for the PMCs to work, it would
424 be configured using function
425 .Fn pmc_configure_logfile .
427 Process scope PMCs would then be attached to their target processes
431 The PMCs would then be started using function
434 Once started, the values of counting PMCs may be read using function
436 For PMCs that write events to the log file, this logged data would be
437 read and parsed using the
441 PMCs are stopped using function
443 and process scope PMCs are detached from their targets using
447 Before the process exits, its may release its PMCs using function
449 Any configured log file may be closed using function
450 .Fn pmc_configure_logfile .
453 Event specifiers are strings comprising of an event name, followed by
454 optional parameters modifying the semantics of the hardware event
456 Event names are PMC architecture dependent, but the PMC library defines
457 machine independent aliases for commonly used events.
459 Event specifiers spellings are case-insensitive and space characters,
460 periods, underscores and hyphens are considered equivalent to each other.
461 Thus the event specifiers
462 .Qq "Example Event" ,
463 .Qq "example-event" ,
467 .Ss PMC Architecture Dependent Events
468 PMC architecture dependent event specifiers are described in the
469 following manual pages:
470 .Bl -column " PMC_CLASS_TSC " "MANUAL PAGE "
471 .It Em "PMC Class" Ta Em "Manual Page"
472 .It Li PMC_CLASS_IAF Ta Xr pmc.iaf 3
473 .It Li PMC_CLASS_IAP Ta Xr pmc.atom 3 , Xr pmc.core 3 , Xr pmc.core2 3
474 .It Li PMC_CLASS_K7 Ta Xr pmc.k7 3
475 .It Li PMC_CLASS_K8 Ta Xr pmc.k8 3
476 .It Li PMC_CLASS_P4 Ta Xr pmc.p4 3
477 .It Li PMC_CLASS_P5 Ta Xr pmc.p5 3
478 .It Li PMC_CLASS_P6 Ta Xr pmc.p6 3
479 .It Li PMC_CLASS_TSC Ta Xr pmc.tsc 3
481 .Ss Event Name Aliases
482 Event name aliases are PMC-independent names for commonly used events.
483 The following aliases are known to this version of the
486 .Bl -tag -width indent
488 Measure the number of branches retired.
489 .It Li branch-mispredicts
490 Measure the number of retired branches that were mispredicted.
492 Measure processor cycles.
493 This event is implemented using the processor's Time Stamp Counter
496 Measure the number of data cache misses.
498 Measure the number of instruction cache misses.
500 Measure the number of instructions retired.
502 Measure the number of interrupts seen.
503 .It Li unhalted-cycles
504 Measure the number of cycles the processor is not in a halted
508 The interface between the
512 driver is intended to be private to the implementation and may
514 In order to ease forward compatibility with future versions of the
516 driver, applications are urged to dynamically link with the
529 .Xr pmc.haswelluc 3 ,
530 .Xr pmc.haswellxeon 3 ,
532 .Xr pmc.ivybridge 3 ,
533 .Xr pmc.ivybridgexeon 3 ,
541 .Xr pmc.sandybridge 3 ,
542 .Xr pmc.sandybridgeuc 3 ,
543 .Xr pmc.sandybridgexeon 3 ,
547 .Xr pmc.westmereuc 3 ,
551 .Xr pmc_capabilities 3 ,
552 .Xr pmc_configure_logfile 3 ,
554 .Xr pmc_event_names_of_class 3 ,
555 .Xr pmc_get_driver_stats 3 ,
558 .Xr pmc_name_of_capability 3 ,
569 library first appeared in
574 library was written by
575 .An Joseph Koshy Aq Mt jkoshy@FreeBSD.org .