1 .\" Copyright (c) 2003-2005 Joseph Koshy
2 .\" All rights reserved.
4 .\" Redistribution and use in source and binary forms, with or without
5 .\" modification, are permitted provided that the following conditions
7 .\" 1. Redistributions of source code must retain the above copyright
8 .\" notice, this list of conditions and the following disclaimer.
9 .\" 2. Redistributions in binary form must reproduce the above copyright
10 .\" notice, this list of conditions and the following disclaimer in the
11 .\" documentation and/or other materials provided with the distribution.
13 .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 .Dd September 28, 2005
32 .Nd "Hardware Performance Monitoring Counter support"
34 .Cd "options HWPMC_HOOKS"
37 Additionally, for i386 systems:
42 driver virtualizes the hardware performance monitoring facilities in
43 modern CPUs and provides support for using these facilities from
46 The driver supports multi-processor systems.
48 PMCs are allocated using the
49 .Dv PMC_OP_PMCALLOCATE
52 .Dv PMC_OP_PMCALLOCATE
53 request will return an integer handle to the requesting process.
54 Subsequent operations on the allocated PMC use this handle to denote
56 A process that has successfully allocated a PMC is termed an
59 PMCs may be allocated to operate in process-private or in system-wide
61 .Bl -tag -width ".Em Process-private"
62 .It Em Process-private
63 In process-private mode, a PMC is active only when a thread belonging
64 to a process it is attached to is scheduled on a CPU.
66 In system-wide mode, a PMC operates independently of processes and
67 measures hardware events for the system as a whole.
72 driver supports the use of hardware PMCs for counting or for
74 .Bl -tag -width ".Em Counting"
76 In counting modes, the PMCs count hardware events.
77 These counts are retrievable using the
79 system call on all architectures.
80 Some architectures offer faster methods of reading these counts.
82 In sampling modes, the PMCs are configured to sample the CPU
83 instruction pointer after a configurable number of hardware events
85 These instruction pointer samples are usually directed to a log file
86 for subsequent analysis.
89 These modes of operation are orthogonal; a PMC may be configured to
90 operate in one of four modes:
91 .Bl -tag -width indent
92 .It Process-private, counting
93 These PMCs count hardware events whenever a thread in their attached process is
95 These PMCs normally count from zero, but the initial count may be
99 Applications can read the value of the PMC anytime using the
102 .It Process-private, sampling
103 These PMCs sample the target processes instruction pointer after they
104 have seen the configured number of hardware events.
105 The PMCs only count events when a thread belonging to their attached
107 The desired frequency of sampling is set using the
109 operation prior to starting the PMC.
110 Log files are configured using the
111 .Dv PMC_OP_CONFIGURELOG
113 .It System-wide, counting
114 These PMCs count hardware events seen by them independent of the
115 processes that are executing.
116 The current count on these PMCs can be read using the
119 These PMCs normally count from zero, but the initial count may be
123 .It System-wide, sampling
124 These PMCs will periodically sample the instruction pointer of the CPU
125 they are allocated on, and will write the sample to a log for further
127 The desired frequency of sampling is set using the
129 operation prior to starting the PMC.
130 Log files are configured using the
131 .Dv PMC_OP_CONFIGURELOG
134 System-wide statistical sampling can only be enabled by a process with
135 super-user privileges.
138 Processes are allowed to allocate as many PMCs as the hardware and
139 current operating conditions permit.
140 Processes may mix allocations of system-wide and process-private
142 Multiple processes may be using PMCs simultaneously.
144 Allocated PMCs are started using the
146 operation, and stopped using the
149 Stopping and starting a PMC is permitted at any time the owner process
150 has a valid handle to the PMC.
152 Process-private PMCs need to be attached to a target process before
154 Attaching a process to a PMC is done using the
157 An already attached PMC may be detached from its target process
163 operation on an as yet unattached PMC will cause it to be attached
164 to its owner process.
165 The following rules determine whether a given process may attach
166 a PMC to another target process:
169 A non-jailed process with super-user privileges is allowed to attach
170 to any other process in the system.
172 Other processes are only allowed to attach to targets that they would
173 be able to attach to for debugging (as determined by
177 PMCs are released using
178 .Dv PMC_OP_PMCRELEASE .
180 .Dv PMC_OP_PMCRELEASE
181 operation the handle to the PMC will become invalid.
184 .Dv PMC_OP_PMCALLOCATE
185 operation supports the following flags that modify the behavior
187 .Bl -tag -width indent
188 .It Dv PMC_F_DESCENDANTS
189 This modifier is valid only for a PMC being allocated in process-private
191 It signifies that the PMC will track hardware events for its
192 target process and the target's current and future descendants.
194 This modifier is valid only for a PMC being allocated in system-wide
196 It signifies that the PMC's sampling interrupt is to be used to drive
199 .It Dv PMC_F_LOG_PROCCSW
200 This modifier is valid only for a PMC being allocated in process-private
202 When this modifier is present, at every context switch,
204 will log a record containing the number of hardware events
205 seen by the target process when it was scheduled on the CPU.
206 .It Dv PMC_F_LOG_PROCEXIT
207 This modifier is valid only for a PMC being allocated in process-private
209 With this modifier present,
211 will maintain per-process counts for each target process attached to
213 At process exit time, a record containing the target process' PID and
214 the accumulated per-process count for that process will be written to the
219 .Dv PMC_F_LOG_PROCEXIT
221 .Dv PMC_F_LOG_PROCCSW
222 may be used in combination with modifier
223 .Dv PMC_F_DESCENDANTS
224 to track the behavior of complex pipelines of processes.
226 .Dv PMC_F_LOG_PROCEXIT
228 .Dv PMC_F_LOG_PROCCSW
229 cannot be started until their owner process has configured a log file.
233 driver may deliver signals to processes that have allocated PMCs:
234 .Bl -tag -width ".Dv SIGBUS"
238 operation was attempted on a process-private PMC that does not have
239 attached target processes.
243 driver is being unloaded from the kernel.
246 The recommended way for application programs to use the facilities of
249 driver is using the API provided by the
255 driver operates using a system call number that is dynamically
256 allotted to it when it is loaded into the kernel.
260 driver supports the following operations:
261 .Bl -tag -width indent
262 .It Dv PMC_OP_CONFIGURELOG
263 Configure a log file for sampling mode PMCs.
264 .It Dv PMC_OP_FLUSHLOG
265 Transfer buffered log data inside
267 to a configured output file.
268 This operation returns to the caller after the write operation
270 .It Dv PMC_OP_GETCPUINFO
271 Retrieve information about the number of CPUs on the system and
272 the number of hardware performance monitoring counters available per-CPU.
273 .It Dv PMC_OP_GETDRIVERSTATS
274 Retrieve module statistics (for analyzing the behavior of
277 .It Dv PMC_OP_GETMODULEVERSION
278 Retrieve the version number of API.
279 .It Dv PMC_OP_GETPMCINFO
280 Retrieve information about the current state of the PMCs on a
282 .It Dv PMC_OP_PMCADMIN
283 Set the administrative state (i.e., whether enabled or disabled) for
284 the hardware PMCs managed by the
287 .It Dv PMC_OP_PMCALLOCATE
288 Allocate and configure a PMC.
289 On successful allocation, a handle to the PMC (a small integer)
291 .It Dv PMC_OP_PMCATTACH
292 Attach a process mode PMC to a target process.
293 The PMC will be active whenever a thread in the target process is
297 .Dv PMC_F_DESCENDANTS
298 flag had been specified at PMC allocation time, then the PMC is
299 attached to all current and future descendants of the target process.
300 .It Dv PMC_OP_PMCDETACH
301 Detach a PMC from its target process.
302 .It Dv PMC_OP_PMCRELEASE
305 Read and write a PMC.
306 This operation is valid only for PMCs configured in counting modes.
307 .It Dv PMC_OP_SETCOUNT
308 Set the initial count (for counting mode PMCs) or the desired sampling
309 rate (for sampling mode PMCs).
310 .It Dv PMC_OP_PMCSTART
312 .It Dv PMC_OP_PMCSTOP
314 .It Dv PMC_OP_WRITELOG
315 Insert a timestamped user record into the log file.
317 .Ss i386 Specific API
318 Some i386 family CPUs support the RDPMC instruction which allows a
319 user process to read a PMC value without needing to invoke a
322 On such CPUs, the machine address associated with an allocated PMC is
323 retrievable using the
324 .Dv PMC_OP_PMCX86GETMSR
326 .Bl -tag -width indent
327 .It Dv PMC_OP_PMCX86GETMSR
328 Retrieve the MSR (machine specific register) number associated with
329 the given PMC handle.
331 The PMC needs to be in process-private mode and allocated without the
332 .Dv PMC_F_DESCENDANTS
333 modifier flag, and should be attached only to its owner process at the
336 .Ss amd64 Specific API
337 AMD64 CPUs support the RDPMC instruction which allows a
338 user process to read a PMC value without needing to invoke a
341 The machine address associated with an allocated PMC is
342 retrievable using the
343 .Dv PMC_OP_PMCX86GETMSR
345 .Bl -tag -width indent
346 .It Dv PMC_OP_PMCX86GETMSR
347 Retrieve the MSR (machine specific register) number associated with
348 the given PMC handle.
350 The PMC needs to be in process-private mode and allocated without the
351 .Dv PMC_F_DESCENDANTS
352 modifier flag, and should be attached only to its owner process at the
355 .Sh SYSCTL VARIABLES AND LOADER TUNABLES
358 is influenced by the following
363 .Bl -tag -width indent
364 .It Va kern.hwpmc.debugflags Pq string, read-write
365 (Only available if the
367 driver was compiled with
369 Control the verbosity of debug messages from the
372 .It Va kern.hwpmc.hashsize Pq integer, read-only
373 The number of rows in the hash tables used to keep track of owner and
376 .It Va kern.hwpmc.logbuffersize Pq integer, read-only
377 The size in kilobytes of each log buffer used by
380 The default buffer size is 4KB.
381 .It Va kern.hwpmc.mtxpoolsize Pq integer, read-only
382 The size of the spin mutex pool used by the PMC driver.
384 .It Va kern.hwpmc.nbuffers Pq integer, read-only
385 The number of log buffers used by
389 .It Va kern.hwpmc.nsamples Pq integer, read-only
390 The number of entries in the per-CPU ring buffer used during sampling.
392 .It Va security.bsd.unprivileged_syspmcs Pq boolean, read-write
393 If set to non-zero, allow unprivileged processes to allocate system-wide
395 The default value is 0.
396 .It Va security.bsd.unprivileged_proc_debug Pq boolean, read-write
399 driver will only allow privileged processes to attach PMCs to other
403 These variables may be set in the kernel environment using
408 .Sh SECURITY CONSIDERATIONS
409 PMCs may be used to monitor the actual behavior of the system on hardware.
410 In situations where this constitutes an undesirable information leak,
411 the following options are available:
417 .Va security.bsd.unprivileged_syspmcs
419 This ensures that unprivileged processes cannot allocate system-wide
420 PMCs and thus cannot observe the hardware behavior of the system
422 This tunable may also be set at boot time using
428 driver into the kernel.
433 .Va security.bsd.unprivileged_proc_debug
435 This will ensure that an unprivileged process cannot attach a PMC
436 to any process other than itself and thus cannot observe the hardware
437 behavior of other processes with the same credentials.
440 System administrators should note that on IA-32 platforms
442 makes the content of the IA-32 TSC counter available to all processes
443 via the RDTSC instruction.
444 .Sh IMPLEMENTATION NOTES
446 The kernel driver requires all physical CPUs in an SMP system to have
447 identical performance monitoring counter hardware.
449 Historically, on the x86 architecture,
451 has permitted user processes running at a processor CPL of 3 to
452 read the TSC using the RDTSC instruction.
455 driver preserves this behavior.
456 .Ss Intel P4/HTT Handling
457 On CPUs with HTT support, Intel P4 PMCs are capable of qualifying
458 only a subset of hardware events on a per-logical CPU basis.
459 Consequently, if HTT is enabled on a system with Intel Pentium P4
462 driver will reject allocation requests for process-private PMCs that
463 request counting of hardware events that cannot be counted separately
464 for each logical CPU.
465 .Ss Intel Pentium-Pro Handling
466 Writing a value to the PMC MSRs found in Intel Pentium-Pro style PMCs
468 .Tn "Intel Pentium Pro" ,
474 processors) will replicate bit 31 of the
475 value being written into the upper 8 bits of the MSR,
476 bringing down the usable width of these PMCs to 31 bits.
477 For process-virtual PMCs, the
479 driver implements a workaround in software and makes the corrected 64
480 bit count available via the
483 Processes that intend to use RDPMC instructions directly or
484 that intend to write values larger than 2^31 into these PMCs with
486 need to be aware of this hardware limitation.
489 .It "hwpmc: [class/npmc/capabilities]..."
490 Announce the presence of
494 with capabilities described by bit string
496 .It "hwpmc: kernel version (0x%x) does not match module version (0x%x)."
497 The module loading process failed because a version mismatch was detected
498 between the currently executing kernel and the module being loaded.
499 .It "hwpmc: this kernel has not been compiled with 'options HWPMC_HOOKS'."
500 The module loading process failed because the currently executing kernel
501 was not configured with the required configuration option
503 .It "hwpmc: tunable hashsize=%d must be greater than zero."
504 A negative value was supplied for tunable
505 .Va kern.hwpmc.hashsize .
506 .It "hwpmc: tunable logbuffersize=%d must be greater than zero."
507 A negative value was supplied for tunable
508 .Va kern.hwpmc.logbuffersize .
509 .It "hwpmc: tunable nlogbuffers=%d must be greater than zero."
510 A negative value was supplied for tunable
511 .Va kern.hwpmc.nlogbuffers .
512 .It "hwpmc: tunable nsamples=%d out of range."
513 The value for tunable
514 .Va kern.hwpmc.nsamples
515 was negative or greater than 65535.
522 The API and ABI documented in this manual page may change in
524 The recommended method of accessing this driver is using the
528 A command issued to the
530 driver may fail with the following errors:
534 .Dv PMC_OP_CONFIGURELOG
535 operation was requested while an existing log was active.
537 A DISABLE operation was requested using the
539 request for a set of hardware resources currently in use for
540 process-private PMCs.
544 operation was requested on an active system mode PMC.
548 operation was requested for a target process that already had another
549 PMC using the same hardware resources attached to it.
553 request writing a new value was issued on a PMC that was active.
556 .Dv PMC_OP_PMCSETCOUNT
557 request was issued on a PMC that was active.
561 operation was requested without a log file being configured for a
563 .Dv PMC_F_LOG_PROCCSW
565 .Dv PMC_F_LOG_PROCEXIT
570 request was reissued for a target process that already is the target
573 A bad address was passed in to the driver.
575 A process specified an invalid PMC handle.
577 An invalid CPU number was passed in for a
578 .Dv PMC_OP_GETPMCINFO
581 An invalid CPU number was passed in for a
585 An invalid operation request was passed in for a
589 An invalid PMC ID was passed in for a
593 A suitable PMC matching the parameters passed in to a
594 .Dv PMC_OP_PMCALLOCATE
595 request could not be allocated.
597 An invalid PMC mode was requested during a
598 .Dv PMC_OP_PMCALLOCATE
601 An invalid CPU number was specified during a
602 .Dv PMC_OP_PMCALLOCATE
609 request for a process-private PMC.
615 request for a system-wide PMC.
620 .Dv PMC_OP_PMCALLOCATE
621 request contained unknown flags.
623 A PMC allocated for system-wide operation was specified with a
631 request specified an illegal process ID.
635 request was issued for a PMC not attached to the target process.
641 request contained illegal flags.
644 .Dv PMC_OP_PMCX86GETMSR
645 operation was requested for a PMC not in process-virtual mode, or
646 for a PMC that is not solely attached to its owner process, or for
647 a PMC that was allocated with flag
648 .Dv PMC_F_DESCENDANTS .
650 (On Intel Pentium 4 CPUs with HTT support)
651 An allocation request for
652 a process-private PMC was issued for an event that does not support
653 counting on a per-logical CPU basis.
655 The system was not able to allocate kernel memory.
659 .Dv PMC_OP_PMCX86GETMSR
660 operation was requested for hardware that does not support reading
661 PMCs directly with the RDPMC instruction.
664 .Dv PMC_OP_GETPMCINFO
665 operation was requested for a disabled CPU.
667 A system-wide PMC on a disabled CPU was requested to be allocated with
668 .Dv PMC_OP_PMCALLOCATE .
674 request was issued for a system-wide PMC that was allocated on a
675 currently disabled CPU.
678 .Dv PMC_OP_PMCALLOCATE
679 request was issued for PMC capabilities not supported
680 by the specified PMC class.
683 A sampling mode PMC was requested on a CPU lacking an APIC.
687 request was issued by a process without super-user
688 privilege or by a jailed super-user process.
692 operation was issued for a target process that the current process
693 does not have permission to attach to.
695 (i386 and amd64 architectures)
698 operation was issued on a PMC whose MSR has been retrieved using
699 .Dv PMC_OP_PMCX86GETMSR .
701 A process issued a PMC operation request without having allocated any
704 A process issued a PMC operation request after the PMC was detached
705 from all of its target processes.
709 request specified a non-existent process ID.
711 The target process for a
713 operation is not being monitored by the
730 driver first appeared in
733 The driver samples the state of the kernel's logical processor support
734 at the time of initialization (i.e., at module load time).
735 On CPUs supporting logical processors, the driver could misbehave if
736 logical processors are subsequently enabled or disabled while the
739 On the i386 architecture, the driver requires that the local APIC on the
740 CPU be enabled for sampling mode to be supported.
741 Many single-processor motherboards keep the APIC disabled in BIOS; on
744 will not support sampling PMCs.