xref: /freebsd/sys/sys/pmc.h (revision 2f513db7)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2003-2008, Joseph Koshy
5  * Copyright (c) 2007 The FreeBSD Foundation
6  * All rights reserved.
7  *
8  * Portions of this software were developed by A. Joseph Koshy under
9  * sponsorship from the FreeBSD Foundation and Google, Inc.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * $FreeBSD$
33  */
34 
35 #ifndef _SYS_PMC_H_
36 #define	_SYS_PMC_H_
37 
38 #include <dev/hwpmc/pmc_events.h>
39 #include <sys/proc.h>
40 #include <sys/counter.h>
41 #include <machine/pmc_mdep.h>
42 #include <machine/profile.h>
43 #ifdef _KERNEL
44 #include <sys/epoch.h>
45 #include <ck_queue.h>
46 #endif
47 
48 #define	PMC_MODULE_NAME		"hwpmc"
49 #define	PMC_NAME_MAX		64 /* HW counter name size */
50 #define	PMC_CLASS_MAX		8  /* max #classes of PMCs per-system */
51 
52 /*
53  * Kernel<->userland API version number [MMmmpppp]
54  *
55  * Major numbers are to be incremented when an incompatible change to
56  * the ABI occurs that older clients will not be able to handle.
57  *
58  * Minor numbers are incremented when a backwards compatible change
59  * occurs that allows older correct programs to run unchanged.  For
60  * example, when support for a new PMC type is added.
61  *
62  * The patch version is incremented for every bug fix.
63  */
64 #define	PMC_VERSION_MAJOR	0x09
65 #define	PMC_VERSION_MINOR	0x03
66 #define	PMC_VERSION_PATCH	0x0000
67 
68 #define	PMC_VERSION		(PMC_VERSION_MAJOR << 24 |		\
69 	PMC_VERSION_MINOR << 16 | PMC_VERSION_PATCH)
70 
71 #define PMC_CPUID_LEN 64
72 /* cpu model name for pmu lookup */
73 extern char pmc_cpuid[PMC_CPUID_LEN];
74 
75 /*
76  * Kinds of CPUs known.
77  *
78  * We keep track of CPU variants that need to be distinguished in
79  * some way for PMC operations.  CPU names are grouped by manufacturer
80  * and numbered sparsely in order to minimize changes to the ABI involved
81  * when new CPUs are added.
82  */
83 
84 #define	__PMC_CPUS()						\
85 	__PMC_CPU(AMD_K7,	0x00,	"AMD K7")		\
86 	__PMC_CPU(AMD_K8,	0x01,	"AMD K8")		\
87 	__PMC_CPU(INTEL_P5,	0x80,	"Intel Pentium")	\
88 	__PMC_CPU(INTEL_P6,	0x81,	"Intel Pentium Pro")	\
89 	__PMC_CPU(INTEL_CL,	0x82,	"Intel Celeron")	\
90 	__PMC_CPU(INTEL_PII,	0x83,	"Intel Pentium II")	\
91 	__PMC_CPU(INTEL_PIII,	0x84,	"Intel Pentium III")	\
92 	__PMC_CPU(INTEL_PM,	0x85,	"Intel Pentium M")	\
93 	__PMC_CPU(INTEL_PIV,	0x86,	"Intel Pentium IV")	\
94 	__PMC_CPU(INTEL_CORE,	0x87,	"Intel Core Solo/Duo")	\
95 	__PMC_CPU(INTEL_CORE2,	0x88,	"Intel Core2")		\
96 	__PMC_CPU(INTEL_CORE2EXTREME,	0x89,	"Intel Core2 Extreme")	\
97 	__PMC_CPU(INTEL_ATOM,	0x8A,	"Intel Atom")		\
98 	__PMC_CPU(INTEL_COREI7, 0x8B,   "Intel Core i7")	\
99 	__PMC_CPU(INTEL_WESTMERE, 0x8C,   "Intel Westmere")	\
100 	__PMC_CPU(INTEL_SANDYBRIDGE, 0x8D,   "Intel Sandy Bridge")	\
101 	__PMC_CPU(INTEL_IVYBRIDGE, 0x8E,   "Intel Ivy Bridge")	\
102 	__PMC_CPU(INTEL_SANDYBRIDGE_XEON, 0x8F,   "Intel Sandy Bridge Xeon")	\
103 	__PMC_CPU(INTEL_IVYBRIDGE_XEON, 0x90,   "Intel Ivy Bridge Xeon")	\
104 	__PMC_CPU(INTEL_HASWELL, 0x91,   "Intel Haswell")	\
105 	__PMC_CPU(INTEL_ATOM_SILVERMONT, 0x92,	"Intel Atom Silvermont")    \
106 	__PMC_CPU(INTEL_NEHALEM_EX, 0x93,   "Intel Nehalem Xeon 7500")	\
107 	__PMC_CPU(INTEL_WESTMERE_EX, 0x94,   "Intel Westmere Xeon E7")	\
108 	__PMC_CPU(INTEL_HASWELL_XEON, 0x95,   "Intel Haswell Xeon E5 v3") \
109 	__PMC_CPU(INTEL_BROADWELL, 0x96,   "Intel Broadwell") \
110 	__PMC_CPU(INTEL_BROADWELL_XEON, 0x97,   "Intel Broadwell Xeon") \
111 	__PMC_CPU(INTEL_SKYLAKE, 0x98,   "Intel Skylake")		\
112 	__PMC_CPU(INTEL_SKYLAKE_XEON, 0x99,   "Intel Skylake Xeon")	\
113 	__PMC_CPU(INTEL_XSCALE,	0x100,	"Intel XScale")		\
114 	__PMC_CPU(MIPS_24K,     0x200,  "MIPS 24K")		\
115 	__PMC_CPU(MIPS_OCTEON,  0x201,  "Cavium Octeon")	\
116 	__PMC_CPU(MIPS_74K,     0x202,  "MIPS 74K")		\
117 	__PMC_CPU(MIPS_BERI,	0x203,  "BERI")			\
118 	__PMC_CPU(PPC_7450,     0x300,  "PowerPC MPC7450")	\
119 	__PMC_CPU(PPC_E500,     0x340,  "PowerPC e500 Core")	\
120 	__PMC_CPU(PPC_970,      0x380,  "IBM PowerPC 970")	\
121 	__PMC_CPU(GENERIC, 	0x400,  "Generic")		\
122 	__PMC_CPU(ARMV7_CORTEX_A5,	0x500,	"ARMv7 Cortex A5")	\
123 	__PMC_CPU(ARMV7_CORTEX_A7,	0x501,	"ARMv7 Cortex A7")	\
124 	__PMC_CPU(ARMV7_CORTEX_A8,	0x502,	"ARMv7 Cortex A8")	\
125 	__PMC_CPU(ARMV7_CORTEX_A9,	0x503,	"ARMv7 Cortex A9")	\
126 	__PMC_CPU(ARMV7_CORTEX_A15,	0x504,	"ARMv7 Cortex A15")	\
127 	__PMC_CPU(ARMV7_CORTEX_A17,	0x505,	"ARMv7 Cortex A17")	\
128 	__PMC_CPU(ARMV8_CORTEX_A53,	0x600,	"ARMv8 Cortex A53")	\
129 	__PMC_CPU(ARMV8_CORTEX_A57,	0x601,	"ARMv8 Cortex A57")
130 
131 enum pmc_cputype {
132 #undef	__PMC_CPU
133 #define	__PMC_CPU(S,V,D)	PMC_CPU_##S = V,
134 	__PMC_CPUS()
135 };
136 
137 #define	PMC_CPU_FIRST	PMC_CPU_AMD_K7
138 #define	PMC_CPU_LAST	PMC_CPU_GENERIC
139 
140 /*
141  * Classes of PMCs
142  */
143 
144 #define	__PMC_CLASSES()							\
145 	__PMC_CLASS(TSC,	0x00,	"CPU Timestamp counter")	\
146 	__PMC_CLASS(K7,		0x01,	"AMD K7 performance counters")	\
147 	__PMC_CLASS(K8,		0x02,	"AMD K8 performance counters")	\
148 	__PMC_CLASS(P5,		0x03,	"Intel Pentium counters")	\
149 	__PMC_CLASS(P6,		0x04,	"Intel Pentium Pro counters")	\
150 	__PMC_CLASS(P4,		0x05,	"Intel Pentium-IV counters")	\
151 	__PMC_CLASS(IAF,	0x06,	"Intel Core2/Atom, fixed function") \
152 	__PMC_CLASS(IAP,	0x07,	"Intel Core...Atom, programmable") \
153 	__PMC_CLASS(UCF,	0x08,	"Intel Uncore fixed function")	\
154 	__PMC_CLASS(UCP,	0x09,	"Intel Uncore programmable")	\
155 	__PMC_CLASS(XSCALE,	0x0A,	"Intel XScale counters")	\
156 	__PMC_CLASS(MIPS24K,	0x0B,	"MIPS 24K")			\
157 	__PMC_CLASS(OCTEON,	0x0C,	"Cavium Octeon")		\
158 	__PMC_CLASS(PPC7450,	0x0D,	"Motorola MPC7450 class")	\
159 	__PMC_CLASS(PPC970,	0x0E,	"IBM PowerPC 970 class")	\
160 	__PMC_CLASS(SOFT,	0x0F,	"Software events")		\
161 	__PMC_CLASS(ARMV7,	0x10,	"ARMv7")			\
162 	__PMC_CLASS(ARMV8,	0x11,	"ARMv8")			\
163 	__PMC_CLASS(MIPS74K,	0x12,	"MIPS 74K")			\
164 	__PMC_CLASS(E500,	0x13,	"Freescale e500 class")		\
165 	__PMC_CLASS(BERI,	0x14,	"MIPS BERI")
166 
167 enum pmc_class {
168 #undef  __PMC_CLASS
169 #define	__PMC_CLASS(S,V,D)	PMC_CLASS_##S = V,
170 	__PMC_CLASSES()
171 };
172 
173 #define	PMC_CLASS_FIRST	PMC_CLASS_TSC
174 #define	PMC_CLASS_LAST	PMC_CLASS_E500
175 
176 /*
177  * A PMC can be in the following states:
178  *
179  * Hardware states:
180  *   DISABLED   -- administratively prohibited from being used.
181  *   FREE       -- HW available for use
182  * Software states:
183  *   ALLOCATED  -- allocated
184  *   STOPPED    -- allocated, but not counting events
185  *   RUNNING    -- allocated, and in operation; 'pm_runcount'
186  *                 holds the number of CPUs using this PMC at
187  *                 a given instant
188  *   DELETED    -- being destroyed
189  */
190 
191 #define	__PMC_HWSTATES()			\
192 	__PMC_STATE(DISABLED)			\
193 	__PMC_STATE(FREE)
194 
195 #define	__PMC_SWSTATES()			\
196 	__PMC_STATE(ALLOCATED)			\
197 	__PMC_STATE(STOPPED)			\
198 	__PMC_STATE(RUNNING)			\
199 	__PMC_STATE(DELETED)
200 
201 #define	__PMC_STATES()				\
202 	__PMC_HWSTATES()			\
203 	__PMC_SWSTATES()
204 
205 enum pmc_state {
206 #undef	__PMC_STATE
207 #define	__PMC_STATE(S)	PMC_STATE_##S,
208 	__PMC_STATES()
209 	__PMC_STATE(MAX)
210 };
211 
212 #define	PMC_STATE_FIRST	PMC_STATE_DISABLED
213 #define	PMC_STATE_LAST	PMC_STATE_DELETED
214 
215 /*
216  * An allocated PMC may used as a 'global' counter or as a
217  * 'thread-private' one.  Each such mode of use can be in either
218  * statistical sampling mode or in counting mode.  Thus a PMC in use
219  *
220  * SS i.e., SYSTEM STATISTICAL  -- system-wide statistical profiling
221  * SC i.e., SYSTEM COUNTER      -- system-wide counting mode
222  * TS i.e., THREAD STATISTICAL  -- thread virtual, statistical profiling
223  * TC i.e., THREAD COUNTER      -- thread virtual, counting mode
224  *
225  * Statistical profiling modes rely on the PMC periodically delivering
226  * a interrupt to the CPU (when the configured number of events have
227  * been measured), so the PMC must have the ability to generate
228  * interrupts.
229  *
230  * In counting modes, the PMC counts its configured events, with the
231  * value of the PMC being read whenever needed by its owner process.
232  *
233  * The thread specific modes "virtualize" the PMCs -- the PMCs appear
234  * to be thread private and count events only when the profiled thread
235  * actually executes on the CPU.
236  *
237  * The system-wide "global" modes keep the PMCs running all the time
238  * and are used to measure the behaviour of the whole system.
239  */
240 
241 #define	__PMC_MODES()				\
242 	__PMC_MODE(SS,	0)			\
243 	__PMC_MODE(SC,	1)			\
244 	__PMC_MODE(TS,	2)			\
245 	__PMC_MODE(TC,	3)
246 
247 enum pmc_mode {
248 #undef	__PMC_MODE
249 #define	__PMC_MODE(M,N)	PMC_MODE_##M = N,
250 	__PMC_MODES()
251 };
252 
253 #define	PMC_MODE_FIRST	PMC_MODE_SS
254 #define	PMC_MODE_LAST	PMC_MODE_TC
255 
256 #define	PMC_IS_COUNTING_MODE(mode)				\
257 	((mode) == PMC_MODE_SC || (mode) == PMC_MODE_TC)
258 #define	PMC_IS_SYSTEM_MODE(mode)				\
259 	((mode) == PMC_MODE_SS || (mode) == PMC_MODE_SC)
260 #define	PMC_IS_SAMPLING_MODE(mode)				\
261 	((mode) == PMC_MODE_SS || (mode) == PMC_MODE_TS)
262 #define	PMC_IS_VIRTUAL_MODE(mode)				\
263 	((mode) == PMC_MODE_TS || (mode) == PMC_MODE_TC)
264 
265 /*
266  * PMC row disposition
267  */
268 
269 #define	__PMC_DISPOSITIONS(N)					\
270 	__PMC_DISP(STANDALONE)	/* global/disabled counters */	\
271 	__PMC_DISP(FREE)	/* free/available */		\
272 	__PMC_DISP(THREAD)	/* thread-virtual PMCs */	\
273 	__PMC_DISP(UNKNOWN)	/* sentinel */
274 
275 enum pmc_disp {
276 #undef	__PMC_DISP
277 #define	__PMC_DISP(D)	PMC_DISP_##D ,
278 	__PMC_DISPOSITIONS()
279 };
280 
281 #define	PMC_DISP_FIRST	PMC_DISP_STANDALONE
282 #define	PMC_DISP_LAST	PMC_DISP_THREAD
283 
284 /*
285  * Counter capabilities
286  *
287  * __PMC_CAPS(NAME, VALUE, DESCRIPTION)
288  */
289 
290 #define	__PMC_CAPS()							\
291 	__PMC_CAP(INTERRUPT,	0, "generate interrupts")		\
292 	__PMC_CAP(USER,		1, "count user-mode events")		\
293 	__PMC_CAP(SYSTEM,	2, "count system-mode events")		\
294 	__PMC_CAP(EDGE,		3, "do edge detection of events")	\
295 	__PMC_CAP(THRESHOLD,	4, "ignore events below a threshold")	\
296 	__PMC_CAP(READ,		5, "read PMC counter")			\
297 	__PMC_CAP(WRITE,	6, "reprogram PMC counter")		\
298 	__PMC_CAP(INVERT,	7, "invert comparison sense")		\
299 	__PMC_CAP(QUALIFIER,	8, "further qualify monitored events")	\
300 	__PMC_CAP(PRECISE,	9, "perform precise sampling")		\
301 	__PMC_CAP(TAGGING,	10, "tag upstream events")		\
302 	__PMC_CAP(CASCADE,	11, "cascade counters")
303 
304 enum pmc_caps
305 {
306 #undef	__PMC_CAP
307 #define	__PMC_CAP(NAME, VALUE, DESCR)	PMC_CAP_##NAME = (1 << VALUE) ,
308 	__PMC_CAPS()
309 };
310 
311 #define	PMC_CAP_FIRST		PMC_CAP_INTERRUPT
312 #define	PMC_CAP_LAST		PMC_CAP_CASCADE
313 
314 /*
315  * PMC Event Numbers
316  *
317  * These are generated from the definitions in "dev/hwpmc/pmc_events.h".
318  */
319 
320 enum pmc_event {
321 #undef	__PMC_EV
322 #undef	__PMC_EV_BLOCK
323 #define	__PMC_EV_BLOCK(C,V)	PMC_EV_ ## C ## __BLOCK_START = (V) - 1 ,
324 #define	__PMC_EV(C,N)		PMC_EV_ ## C ## _ ## N ,
325 	__PMC_EVENTS()
326 };
327 
328 /*
329  * PMC SYSCALL INTERFACE
330  */
331 
332 /*
333  * "PMC_OPS" -- these are the commands recognized by the kernel
334  * module, and are used when performing a system call from userland.
335  */
336 #define	__PMC_OPS()							\
337 	__PMC_OP(CONFIGURELOG, "Set log file")				\
338 	__PMC_OP(FLUSHLOG, "Flush log file")				\
339 	__PMC_OP(GETCPUINFO, "Get system CPU information")		\
340 	__PMC_OP(GETDRIVERSTATS, "Get driver statistics")		\
341 	__PMC_OP(GETMODULEVERSION, "Get module version")		\
342 	__PMC_OP(GETPMCINFO, "Get per-cpu PMC information")		\
343 	__PMC_OP(PMCADMIN, "Set PMC state")				\
344 	__PMC_OP(PMCALLOCATE, "Allocate and configure a PMC")		\
345 	__PMC_OP(PMCATTACH, "Attach a PMC to a process")		\
346 	__PMC_OP(PMCDETACH, "Detach a PMC from a process")		\
347 	__PMC_OP(PMCGETMSR, "Get a PMC's hardware address")		\
348 	__PMC_OP(PMCRELEASE, "Release a PMC")				\
349 	__PMC_OP(PMCRW, "Read/Set a PMC")				\
350 	__PMC_OP(PMCSETCOUNT, "Set initial count/sampling rate")	\
351 	__PMC_OP(PMCSTART, "Start a PMC")				\
352 	__PMC_OP(PMCSTOP, "Stop a PMC")					\
353 	__PMC_OP(WRITELOG, "Write a cookie to the log file")		\
354 	__PMC_OP(CLOSELOG, "Close log file")				\
355 	__PMC_OP(GETDYNEVENTINFO, "Get dynamic events list")
356 
357 
358 enum pmc_ops {
359 #undef	__PMC_OP
360 #define	__PMC_OP(N, D)	PMC_OP_##N,
361 	__PMC_OPS()
362 };
363 
364 
365 /*
366  * Flags used in operations on PMCs.
367  */
368 
369 #define	PMC_F_UNUSED1		0x00000001 /* unused */
370 #define	PMC_F_DESCENDANTS	0x00000002 /*OP ALLOCATE track descendants */
371 #define	PMC_F_LOG_PROCCSW	0x00000004 /*OP ALLOCATE track ctx switches */
372 #define	PMC_F_LOG_PROCEXIT	0x00000008 /*OP ALLOCATE log proc exits */
373 #define	PMC_F_NEWVALUE		0x00000010 /*OP RW write new value */
374 #define	PMC_F_OLDVALUE		0x00000020 /*OP RW get old value */
375 
376 /* V2 API */
377 #define	PMC_F_CALLCHAIN		0x00000080 /*OP ALLOCATE capture callchains */
378 #define	PMC_F_USERCALLCHAIN	0x00000100 /*OP ALLOCATE use userspace stack */
379 
380 /* internal flags */
381 #define	PMC_F_ATTACHED_TO_OWNER	0x00010000 /*attached to owner*/
382 #define	PMC_F_NEEDS_LOGFILE	0x00020000 /*needs log file */
383 #define	PMC_F_ATTACH_DONE	0x00040000 /*attached at least once */
384 
385 #define	PMC_CALLCHAIN_DEPTH_MAX	512
386 
387 #define	PMC_CC_F_USERSPACE	0x01	   /*userspace callchain*/
388 
389 /*
390  * Cookies used to denote allocated PMCs, and the values of PMCs.
391  */
392 
393 typedef uint32_t	pmc_id_t;
394 typedef uint64_t	pmc_value_t;
395 
396 #define	PMC_ID_INVALID		(~ (pmc_id_t) 0)
397 
398 /*
399  * PMC IDs have the following format:
400  *
401  * +-----------------------+-------+-----------+
402  * |   CPU      | PMC MODE | CLASS | ROW INDEX |
403  * +-----------------------+-------+-----------+
404  *
405  * where CPU is 12 bits, MODE 8, CLASS 4, and ROW INDEX 8  Field 'CPU'
406  * is set to the requested CPU for system-wide PMCs or PMC_CPU_ANY for
407  * process-mode PMCs.  Field 'PMC MODE' is the allocated PMC mode.
408  * Field 'PMC CLASS' is the class of the PMC.  Field 'ROW INDEX' is the
409  * row index for the PMC.
410  *
411  * The 'ROW INDEX' ranges over 0..NWPMCS where NHWPMCS is the total
412  * number of hardware PMCs on this cpu.
413  */
414 
415 
416 #define	PMC_ID_TO_ROWINDEX(ID)	((ID) & 0xFF)
417 #define	PMC_ID_TO_CLASS(ID)	(((ID) & 0xF00) >> 8)
418 #define	PMC_ID_TO_MODE(ID)	(((ID) & 0xFF000) >> 12)
419 #define	PMC_ID_TO_CPU(ID)	(((ID) & 0xFFF00000) >> 20)
420 #define	PMC_ID_MAKE_ID(CPU,MODE,CLASS,ROWINDEX)			\
421 	((((CPU) & 0xFFF) << 20) | (((MODE) & 0xFF) << 12) |	\
422 	(((CLASS) & 0xF) << 8) | ((ROWINDEX) & 0xFF))
423 
424 /*
425  * Data structures for system calls supported by the pmc driver.
426  */
427 
428 /*
429  * OP PMCALLOCATE
430  *
431  * Allocate a PMC on the named CPU.
432  */
433 
434 #define	PMC_CPU_ANY	~0
435 
436 struct pmc_op_pmcallocate {
437 	uint32_t	pm_caps;	/* PMC_CAP_* */
438 	uint32_t	pm_cpu;		/* CPU number or PMC_CPU_ANY */
439 	enum pmc_class	pm_class;	/* class of PMC desired */
440 	enum pmc_event	pm_ev;		/* [enum pmc_event] desired */
441 	uint32_t	pm_flags;	/* additional modifiers PMC_F_* */
442 	enum pmc_mode	pm_mode;	/* desired mode */
443 	pmc_id_t	pm_pmcid;	/* [return] process pmc id */
444 	pmc_value_t	pm_count;	/* initial/sample count */
445 
446 	union pmc_md_op_pmcallocate pm_md; /* MD layer extensions */
447 };
448 
449 /*
450  * OP PMCADMIN
451  *
452  * Set the administrative state (i.e., whether enabled or disabled) of
453  * a PMC 'pm_pmc' on CPU 'pm_cpu'.  Note that 'pm_pmc' specifies an
454  * absolute PMC number and need not have been first allocated by the
455  * calling process.
456  */
457 
458 struct pmc_op_pmcadmin {
459 	int		pm_cpu;		/* CPU# */
460 	uint32_t	pm_flags;	/* flags */
461 	int		pm_pmc;         /* PMC# */
462 	enum pmc_state  pm_state;	/* desired state */
463 };
464 
465 /*
466  * OP PMCATTACH / OP PMCDETACH
467  *
468  * Attach/detach a PMC and a process.
469  */
470 
471 struct pmc_op_pmcattach {
472 	pmc_id_t	pm_pmc;		/* PMC to attach to */
473 	pid_t		pm_pid;		/* target process */
474 };
475 
476 /*
477  * OP PMCSETCOUNT
478  *
479  * Set the sampling rate (i.e., the reload count) for statistical counters.
480  * 'pm_pmcid' need to have been previously allocated using PMCALLOCATE.
481  */
482 
483 struct pmc_op_pmcsetcount {
484 	pmc_value_t	pm_count;	/* initial/sample count */
485 	pmc_id_t	pm_pmcid;	/* PMC id to set */
486 };
487 
488 
489 /*
490  * OP PMCRW
491  *
492  * Read the value of a PMC named by 'pm_pmcid'.  'pm_pmcid' needs
493  * to have been previously allocated using PMCALLOCATE.
494  */
495 
496 
497 struct pmc_op_pmcrw {
498 	uint32_t	pm_flags;	/* PMC_F_{OLD,NEW}VALUE*/
499 	pmc_id_t	pm_pmcid;	/* pmc id */
500 	pmc_value_t	pm_value;	/* new&returned value */
501 };
502 
503 
504 /*
505  * OP GETPMCINFO
506  *
507  * retrieve PMC state for a named CPU.  The caller is expected to
508  * allocate 'npmc' * 'struct pmc_info' bytes of space for the return
509  * values.
510  */
511 
512 struct pmc_info {
513 	char		pm_name[PMC_NAME_MAX]; /* pmc name */
514 	enum pmc_class	pm_class;	/* enum pmc_class */
515 	int		pm_enabled;	/* whether enabled */
516 	enum pmc_disp	pm_rowdisp;	/* FREE, THREAD or STANDLONE */
517 	pid_t		pm_ownerpid;	/* owner, or -1 */
518 	enum pmc_mode	pm_mode;	/* current mode [enum pmc_mode] */
519 	enum pmc_event	pm_event;	/* current event */
520 	uint32_t	pm_flags;	/* current flags */
521 	pmc_value_t	pm_reloadcount;	/* sampling counters only */
522 };
523 
524 struct pmc_op_getpmcinfo {
525 	int32_t		pm_cpu;		/* 0 <= cpu < mp_maxid */
526 	struct pmc_info	pm_pmcs[];	/* space for 'npmc' structures */
527 };
528 
529 
530 /*
531  * OP GETCPUINFO
532  *
533  * Retrieve system CPU information.
534  */
535 
536 
537 struct pmc_classinfo {
538 	enum pmc_class	pm_class;	/* class id */
539 	uint32_t	pm_caps;	/* counter capabilities */
540 	uint32_t	pm_width;	/* width of the PMC */
541 	uint32_t	pm_num;		/* number of PMCs in class */
542 };
543 
544 struct pmc_op_getcpuinfo {
545 	enum pmc_cputype pm_cputype; /* what kind of CPU */
546 	uint32_t	pm_ncpu;    /* max CPU number */
547 	uint32_t	pm_npmc;    /* #PMCs per CPU */
548 	uint32_t	pm_nclass;  /* #classes of PMCs */
549 	struct pmc_classinfo  pm_classes[PMC_CLASS_MAX];
550 };
551 
552 /*
553  * OP CONFIGURELOG
554  *
555  * Configure a log file for writing system-wide statistics to.
556  */
557 
558 struct pmc_op_configurelog {
559 	int		pm_flags;
560 	int		pm_logfd;   /* logfile fd (or -1) */
561 };
562 
563 /*
564  * OP GETDRIVERSTATS
565  *
566  * Retrieve pmc(4) driver-wide statistics.
567  */
568 #ifdef _KERNEL
569 struct pmc_driverstats {
570 	counter_u64_t	pm_intr_ignored;	/* #interrupts ignored */
571 	counter_u64_t	pm_intr_processed;	/* #interrupts processed */
572 	counter_u64_t	pm_intr_bufferfull;	/* #interrupts with ENOSPC */
573 	counter_u64_t	pm_syscalls;		/* #syscalls */
574 	counter_u64_t	pm_syscall_errors;	/* #syscalls with errors */
575 	counter_u64_t	pm_buffer_requests;	/* #buffer requests */
576 	counter_u64_t	pm_buffer_requests_failed; /* #failed buffer requests */
577 	counter_u64_t	pm_log_sweeps;		/* #sample buffer processing
578 						   passes */
579 	counter_u64_t	pm_merges;		/* merged k+u */
580 	counter_u64_t	pm_overwrites;		/* UR overwrites */
581 };
582 #endif
583 
584 struct pmc_op_getdriverstats {
585 	unsigned int	pm_intr_ignored;	/* #interrupts ignored */
586 	unsigned int	pm_intr_processed;	/* #interrupts processed */
587 	unsigned int	pm_intr_bufferfull;	/* #interrupts with ENOSPC */
588 	unsigned int	pm_syscalls;		/* #syscalls */
589 	unsigned int	pm_syscall_errors;	/* #syscalls with errors */
590 	unsigned int	pm_buffer_requests;	/* #buffer requests */
591 	unsigned int	pm_buffer_requests_failed; /* #failed buffer requests */
592 	unsigned int	pm_log_sweeps;		/* #sample buffer processing
593 						   passes */
594 };
595 
596 /*
597  * OP RELEASE / OP START / OP STOP
598  *
599  * Simple operations on a PMC id.
600  */
601 
602 struct pmc_op_simple {
603 	pmc_id_t	pm_pmcid;
604 };
605 
606 /*
607  * OP WRITELOG
608  *
609  * Flush the current log buffer and write 4 bytes of user data to it.
610  */
611 
612 struct pmc_op_writelog {
613 	uint32_t	pm_userdata;
614 };
615 
616 /*
617  * OP GETMSR
618  *
619  * Retrieve the machine specific address associated with the allocated
620  * PMC.  This number can be used subsequently with a read-performance-counter
621  * instruction.
622  */
623 
624 struct pmc_op_getmsr {
625 	uint32_t	pm_msr;		/* machine specific address */
626 	pmc_id_t	pm_pmcid;	/* allocated pmc id */
627 };
628 
629 /*
630  * OP GETDYNEVENTINFO
631  *
632  * Retrieve a PMC dynamic class events list.
633  */
634 
635 struct pmc_dyn_event_descr {
636 	char		pm_ev_name[PMC_NAME_MAX];
637 	enum pmc_event	pm_ev_code;
638 };
639 
640 struct pmc_op_getdyneventinfo {
641 	enum pmc_class			pm_class;
642 	unsigned int			pm_nevent;
643 	struct pmc_dyn_event_descr	pm_events[PMC_EV_DYN_COUNT];
644 };
645 
646 #ifdef _KERNEL
647 
648 #include <sys/malloc.h>
649 #include <sys/sysctl.h>
650 #include <sys/_cpuset.h>
651 
652 #include <machine/frame.h>
653 
654 #define	PMC_HASH_SIZE				1024
655 #define	PMC_MTXPOOL_SIZE			2048
656 #define	PMC_LOG_BUFFER_SIZE			256
657 #define	PMC_NLOGBUFFERS_PCPU			32
658 #define	PMC_NSAMPLES				256
659 #define	PMC_CALLCHAIN_DEPTH			128
660 #define	PMC_THREADLIST_MAX			128
661 
662 #define PMC_SYSCTL_NAME_PREFIX "kern." PMC_MODULE_NAME "."
663 
664 /*
665  * Locking keys
666  *
667  * (b) - pmc_bufferlist_mtx (spin lock)
668  * (k) - pmc_kthread_mtx (sleep lock)
669  * (o) - po->po_mtx (spin lock)
670  * (g) - global_epoch_preempt (epoch)
671  * (p) - pmc_sx (sx)
672  */
673 
674 /*
675  * PMC commands
676  */
677 
678 struct pmc_syscall_args {
679 	register_t	pmop_code;	/* one of PMC_OP_* */
680 	void		*pmop_data;	/* syscall parameter */
681 };
682 
683 /*
684  * Interface to processor specific s1tuff
685  */
686 
687 /*
688  * struct pmc_descr
689  *
690  * Machine independent (i.e., the common parts) of a human readable
691  * PMC description.
692  */
693 
694 struct pmc_descr {
695 	char		pd_name[PMC_NAME_MAX]; /* name */
696 	uint32_t	pd_caps;	/* capabilities */
697 	enum pmc_class	pd_class;	/* class of the PMC */
698 	uint32_t	pd_width;	/* width in bits */
699 };
700 
701 /*
702  * struct pmc_target
703  *
704  * This structure records all the target processes associated with a
705  * PMC.
706  */
707 
708 struct pmc_target {
709 	LIST_ENTRY(pmc_target)	pt_next;
710 	struct pmc_process	*pt_process; /* target descriptor */
711 };
712 
713 /*
714  * struct pmc
715  *
716  * Describes each allocated PMC.
717  *
718  * Each PMC has precisely one owner, namely the process that allocated
719  * the PMC.
720  *
721  * A PMC may be attached to multiple target processes.  The
722  * 'pm_targets' field links all the target processes being monitored
723  * by this PMC.
724  *
725  * The 'pm_savedvalue' field is protected by a mutex.
726  *
727  * On a multi-cpu machine, multiple target threads associated with a
728  * process-virtual PMC could be concurrently executing on different
729  * CPUs.  The 'pm_runcount' field is atomically incremented every time
730  * the PMC gets scheduled on a CPU and atomically decremented when it
731  * get descheduled.  Deletion of a PMC is only permitted when this
732  * field is '0'.
733  *
734  */
735 struct pmc_pcpu_state {
736 	uint8_t pps_stalled;
737 	uint8_t pps_cpustate;
738 } __aligned(CACHE_LINE_SIZE);
739 struct pmc {
740 	LIST_HEAD(,pmc_target)	pm_targets;	/* list of target processes */
741 	LIST_ENTRY(pmc)		pm_next;	/* owner's list */
742 
743 	/*
744 	 * System-wide PMCs are allocated on a CPU and are not moved
745 	 * around.  For system-wide PMCs we record the CPU the PMC was
746 	 * allocated on in the 'CPU' field of the pmc ID.
747 	 *
748 	 * Virtual PMCs run on whichever CPU is currently executing
749 	 * their targets' threads.  For these PMCs we need to save
750 	 * their current PMC counter values when they are taken off
751 	 * CPU.
752 	 */
753 
754 	union {
755 		pmc_value_t	pm_savedvalue;	/* Virtual PMCS */
756 	} pm_gv;
757 
758 	/*
759 	 * For sampling mode PMCs, we keep track of the PMC's "reload
760 	 * count", which is the counter value to be loaded in when
761 	 * arming the PMC for the next counting session.  For counting
762 	 * modes on PMCs that are read-only (e.g., the x86 TSC), we
763 	 * keep track of the initial value at the start of
764 	 * counting-mode operation.
765 	 */
766 
767 	union {
768 		pmc_value_t	pm_reloadcount;	/* sampling PMC modes */
769 		pmc_value_t	pm_initial;	/* counting PMC modes */
770 	} pm_sc;
771 
772 	struct pmc_pcpu_state *pm_pcpu_state;
773 	volatile cpuset_t pm_cpustate;	/* CPUs where PMC should be active */
774 	uint32_t	pm_caps;	/* PMC capabilities */
775 	enum pmc_event	pm_event;	/* event being measured */
776 	uint32_t	pm_flags;	/* additional flags PMC_F_... */
777 	struct pmc_owner *pm_owner;	/* owner thread state */
778 	counter_u64_t		pm_runcount;	/* #cpus currently on */
779 	enum pmc_state	pm_state;	/* current PMC state */
780 	uint32_t	pm_overflowcnt;	/* count overflow interrupts */
781 
782 	/*
783 	 * The PMC ID field encodes the row-index for the PMC, its
784 	 * mode, class and the CPU# associated with the PMC.
785 	 */
786 
787 	pmc_id_t	pm_id;		/* allocated PMC id */
788 	enum pmc_class pm_class;
789 
790 	/* md extensions */
791 	union pmc_md_pmc	pm_md;
792 };
793 
794 /*
795  * Accessor macros for 'struct pmc'
796  */
797 
798 #define	PMC_TO_MODE(P)		PMC_ID_TO_MODE((P)->pm_id)
799 #define	PMC_TO_CLASS(P)		PMC_ID_TO_CLASS((P)->pm_id)
800 #define	PMC_TO_ROWINDEX(P)	PMC_ID_TO_ROWINDEX((P)->pm_id)
801 #define	PMC_TO_CPU(P)		PMC_ID_TO_CPU((P)->pm_id)
802 
803 /*
804  * struct pmc_threadpmcstate
805  *
806  * Record per-PMC, per-thread state.
807  */
808 struct pmc_threadpmcstate {
809 	pmc_value_t	pt_pmcval;	/* per-thread reload count */
810 };
811 
812 /*
813  * struct pmc_thread
814  *
815  * Record a 'target' thread being profiled.
816  */
817 struct pmc_thread {
818 	LIST_ENTRY(pmc_thread) pt_next;		/* linked list */
819 	struct thread	*pt_td;			/* target thread */
820 	struct pmc_threadpmcstate pt_pmcs[];	/* per-PMC state */
821 };
822 
823 /*
824  * struct pmc_process
825  *
826  * Record a 'target' process being profiled.
827  *
828  * The target process being profiled could be different from the owner
829  * process which allocated the PMCs.  Each target process descriptor
830  * is associated with NHWPMC 'struct pmc *' pointers.  Each PMC at a
831  * given hardware row-index 'n' will use slot 'n' of the 'pp_pmcs[]'
832  * array.  The size of this structure is thus PMC architecture
833  * dependent.
834  *
835  */
836 
837 struct pmc_targetstate {
838 	struct pmc	*pp_pmc;   /* target PMC */
839 	pmc_value_t	pp_pmcval; /* per-process value */
840 };
841 
842 struct pmc_process {
843 	LIST_ENTRY(pmc_process) pp_next;	/* hash chain */
844 	LIST_HEAD(,pmc_thread) pp_tds;		/* list of threads */
845 	struct mtx	*pp_tdslock;		/* lock on pp_tds thread list */
846 	int		pp_refcnt;		/* reference count */
847 	uint32_t	pp_flags;		/* flags PMC_PP_* */
848 	struct proc	*pp_proc;		/* target process */
849 	struct pmc_targetstate pp_pmcs[];       /* NHWPMCs */
850 };
851 
852 #define	PMC_PP_ENABLE_MSR_ACCESS	0x00000001
853 
854 /*
855  * struct pmc_owner
856  *
857  * We associate a PMC with an 'owner' process.
858  *
859  * A process can be associated with 0..NCPUS*NHWPMC PMCs during its
860  * lifetime, where NCPUS is the numbers of CPUS in the system and
861  * NHWPMC is the number of hardware PMCs per CPU.  These are
862  * maintained in the list headed by the 'po_pmcs' to save on space.
863  *
864  */
865 
866 struct pmc_owner  {
867 	LIST_ENTRY(pmc_owner)	po_next;	/* hash chain */
868 	CK_LIST_ENTRY(pmc_owner)	po_ssnext;	/* (g/p) list of SS PMC owners */
869 	LIST_HEAD(, pmc)	po_pmcs;	/* owned PMC list */
870 	TAILQ_HEAD(, pmclog_buffer) po_logbuffers; /* (o) logbuffer list */
871 	struct mtx		po_mtx;		/* spin lock for (o) */
872 	struct proc		*po_owner;	/* owner proc */
873 	uint32_t		po_flags;	/* (k) flags PMC_PO_* */
874 	struct proc		*po_kthread;	/* (k) helper kthread */
875 	struct file		*po_file;	/* file reference */
876 	int			po_error;	/* recorded error */
877 	short			po_sscount;	/* # SS PMCs owned */
878 	short			po_logprocmaps;	/* global mappings done */
879 	struct pmclog_buffer	*po_curbuf[MAXCPU];	/* current log buffer */
880 };
881 
882 #define	PMC_PO_OWNS_LOGFILE		0x00000001 /* has a log file */
883 #define	PMC_PO_SHUTDOWN			0x00000010 /* in the process of shutdown */
884 #define	PMC_PO_INITIAL_MAPPINGS_DONE	0x00000020
885 
886 /*
887  * struct pmc_hw -- describe the state of the PMC hardware
888  *
889  * When in use, a HW PMC is associated with one allocated 'struct pmc'
890  * pointed to by field 'phw_pmc'.  When inactive, this field is NULL.
891  *
892  * On an SMP box, one or more HW PMC's in process virtual mode with
893  * the same 'phw_pmc' could be executing on different CPUs.  In order
894  * to handle this case correctly, we need to ensure that only
895  * incremental counts get added to the saved value in the associated
896  * 'struct pmc'.  The 'phw_save' field is used to keep the saved PMC
897  * value at the time the hardware is started during this context
898  * switch (i.e., the difference between the new (hardware) count and
899  * the saved count is atomically added to the count field in 'struct
900  * pmc' at context switch time).
901  *
902  */
903 
904 struct pmc_hw {
905 	uint32_t	phw_state;	/* see PHW_* macros below */
906 	struct pmc	*phw_pmc;	/* current thread PMC */
907 };
908 
909 #define	PMC_PHW_RI_MASK		0x000000FF
910 #define	PMC_PHW_CPU_SHIFT	8
911 #define	PMC_PHW_CPU_MASK	0x0000FF00
912 #define	PMC_PHW_FLAGS_SHIFT	16
913 #define	PMC_PHW_FLAGS_MASK	0xFFFF0000
914 
915 #define	PMC_PHW_INDEX_TO_STATE(ri)	((ri) & PMC_PHW_RI_MASK)
916 #define	PMC_PHW_STATE_TO_INDEX(state)	((state) & PMC_PHW_RI_MASK)
917 #define	PMC_PHW_CPU_TO_STATE(cpu)	(((cpu) << PMC_PHW_CPU_SHIFT) & \
918 	PMC_PHW_CPU_MASK)
919 #define	PMC_PHW_STATE_TO_CPU(state)	(((state) & PMC_PHW_CPU_MASK) >> \
920 	PMC_PHW_CPU_SHIFT)
921 #define	PMC_PHW_FLAGS_TO_STATE(flags)	(((flags) << PMC_PHW_FLAGS_SHIFT) & \
922 	PMC_PHW_FLAGS_MASK)
923 #define	PMC_PHW_STATE_TO_FLAGS(state)	(((state) & PMC_PHW_FLAGS_MASK) >> \
924 	PMC_PHW_FLAGS_SHIFT)
925 #define	PMC_PHW_FLAG_IS_ENABLED		(PMC_PHW_FLAGS_TO_STATE(0x01))
926 #define	PMC_PHW_FLAG_IS_SHAREABLE	(PMC_PHW_FLAGS_TO_STATE(0x02))
927 
928 /*
929  * struct pmc_sample
930  *
931  * Space for N (tunable) PC samples and associated control data.
932  */
933 
934 struct pmc_sample {
935 	uint16_t		ps_nsamples;	/* callchain depth */
936 	uint16_t		ps_nsamples_actual;
937 	uint16_t		ps_cpu;		/* cpu number */
938 	uint16_t		ps_flags;	/* other flags */
939 	lwpid_t			ps_tid;		/* thread id */
940 	pid_t			ps_pid;		/* process PID or -1 */
941 	int		ps_ticks; /* ticks at sample time */
942 	/* pad */
943 	struct thread		*ps_td;		/* which thread */
944 	struct pmc		*ps_pmc;	/* interrupting PMC */
945 	uintptr_t		*ps_pc;		/* (const) callchain start */
946 	uint64_t		ps_tsc;		/* tsc value */
947 };
948 
949 #define 	PMC_SAMPLE_FREE		((uint16_t) 0)
950 #define 	PMC_USER_CALLCHAIN_PENDING	((uint16_t) 0xFFFF)
951 
952 struct pmc_samplebuffer {
953 	volatile uint64_t		ps_prodidx; /* producer index */
954 	volatile uint64_t		ps_considx; /* consumer index */
955 	uintptr_t		*ps_callchains;	/* all saved call chains */
956 	struct pmc_sample	ps_samples[];	/* array of sample entries */
957 };
958 
959 #define PMC_CONS_SAMPLE(psb)					\
960 	(&(psb)->ps_samples[(psb)->ps_considx & pmc_sample_mask])
961 
962 #define PMC_CONS_SAMPLE_OFF(psb, off)							\
963 	(&(psb)->ps_samples[(off) & pmc_sample_mask])
964 
965 #define PMC_PROD_SAMPLE(psb)					\
966 	(&(psb)->ps_samples[(psb)->ps_prodidx & pmc_sample_mask])
967 
968 /*
969  * struct pmc_cpustate
970  *
971  * A CPU is modelled as a collection of HW PMCs with space for additional
972  * flags.
973  */
974 
975 struct pmc_cpu {
976 	uint32_t	pc_state;	/* physical cpu number + flags */
977 	struct pmc_samplebuffer *pc_sb[3]; /* space for samples */
978 	struct pmc_hw	*pc_hwpmcs[];	/* 'npmc' pointers */
979 };
980 
981 #define	PMC_PCPU_CPU_MASK		0x000000FF
982 #define	PMC_PCPU_FLAGS_MASK		0xFFFFFF00
983 #define	PMC_PCPU_FLAGS_SHIFT		8
984 #define	PMC_PCPU_STATE_TO_CPU(S)	((S) & PMC_PCPU_CPU_MASK)
985 #define	PMC_PCPU_STATE_TO_FLAGS(S)	(((S) & PMC_PCPU_FLAGS_MASK) >> PMC_PCPU_FLAGS_SHIFT)
986 #define	PMC_PCPU_FLAGS_TO_STATE(F)	(((F) << PMC_PCPU_FLAGS_SHIFT) & PMC_PCPU_FLAGS_MASK)
987 #define	PMC_PCPU_CPU_TO_STATE(C)	((C) & PMC_PCPU_CPU_MASK)
988 #define	PMC_PCPU_FLAG_HTT		(PMC_PCPU_FLAGS_TO_STATE(0x1))
989 
990 /*
991  * struct pmc_binding
992  *
993  * CPU binding information.
994  */
995 
996 struct pmc_binding {
997 	int	pb_bound;	/* is bound? */
998 	int	pb_cpu;		/* if so, to which CPU */
999 };
1000 
1001 
1002 struct pmc_mdep;
1003 
1004 /*
1005  * struct pmc_classdep
1006  *
1007  * PMC class-dependent operations.
1008  */
1009 struct pmc_classdep {
1010 	uint32_t	pcd_caps;	/* class capabilities */
1011 	enum pmc_class	pcd_class;	/* class id */
1012 	int		pcd_num;	/* number of PMCs */
1013 	int		pcd_ri;		/* row index of the first PMC in class */
1014 	int		pcd_width;	/* width of the PMC */
1015 
1016 	/* configuring/reading/writing the hardware PMCs */
1017 	int (*pcd_config_pmc)(int _cpu, int _ri, struct pmc *_pm);
1018 	int (*pcd_get_config)(int _cpu, int _ri, struct pmc **_ppm);
1019 	int (*pcd_read_pmc)(int _cpu, int _ri, pmc_value_t *_value);
1020 	int (*pcd_write_pmc)(int _cpu, int _ri, pmc_value_t _value);
1021 
1022 	/* pmc allocation/release */
1023 	int (*pcd_allocate_pmc)(int _cpu, int _ri, struct pmc *_t,
1024 		const struct pmc_op_pmcallocate *_a);
1025 	int (*pcd_release_pmc)(int _cpu, int _ri, struct pmc *_pm);
1026 
1027 	/* starting and stopping PMCs */
1028 	int (*pcd_start_pmc)(int _cpu, int _ri);
1029 	int (*pcd_stop_pmc)(int _cpu, int _ri);
1030 
1031 	/* description */
1032 	int (*pcd_describe)(int _cpu, int _ri, struct pmc_info *_pi,
1033 		struct pmc **_ppmc);
1034 
1035 	/* class-dependent initialization & finalization */
1036 	int (*pcd_pcpu_init)(struct pmc_mdep *_md, int _cpu);
1037 	int (*pcd_pcpu_fini)(struct pmc_mdep *_md, int _cpu);
1038 
1039 	/* machine-specific interface */
1040 	int (*pcd_get_msr)(int _ri, uint32_t *_msr);
1041 };
1042 
1043 /*
1044  * struct pmc_mdep
1045  *
1046  * Machine dependent bits needed per CPU type.
1047  */
1048 
1049 struct pmc_mdep  {
1050 	uint32_t	pmd_cputype;    /* from enum pmc_cputype */
1051 	uint32_t	pmd_npmc;	/* number of PMCs per CPU */
1052 	uint32_t	pmd_nclass;	/* number of PMC classes present */
1053 
1054 	/*
1055 	 * Machine dependent methods.
1056 	 */
1057 
1058 	/* per-cpu initialization and finalization */
1059 	int (*pmd_pcpu_init)(struct pmc_mdep *_md, int _cpu);
1060 	int (*pmd_pcpu_fini)(struct pmc_mdep *_md, int _cpu);
1061 
1062 	/* thread context switch in/out */
1063 	int (*pmd_switch_in)(struct pmc_cpu *_p, struct pmc_process *_pp);
1064 	int (*pmd_switch_out)(struct pmc_cpu *_p, struct pmc_process *_pp);
1065 
1066 	/* handle a PMC interrupt */
1067 	int (*pmd_intr)(struct trapframe *_tf);
1068 
1069 	/*
1070 	 * PMC class dependent information.
1071 	 */
1072 	struct pmc_classdep pmd_classdep[];
1073 };
1074 
1075 /*
1076  * Per-CPU state.  This is an array of 'mp_ncpu' pointers
1077  * to struct pmc_cpu descriptors.
1078  */
1079 
1080 extern struct pmc_cpu **pmc_pcpu;
1081 
1082 /* driver statistics */
1083 extern struct pmc_driverstats pmc_stats;
1084 
1085 #if	defined(HWPMC_DEBUG)
1086 #include <sys/ktr.h>
1087 
1088 /* debug flags, major flag groups */
1089 struct pmc_debugflags {
1090 	int	pdb_CPU;
1091 	int	pdb_CSW;
1092 	int	pdb_LOG;
1093 	int	pdb_MDP;
1094 	int	pdb_MOD;
1095 	int	pdb_OWN;
1096 	int	pdb_PMC;
1097 	int	pdb_PRC;
1098 	int	pdb_SAM;
1099 };
1100 
1101 extern struct pmc_debugflags pmc_debugflags;
1102 
1103 #define	KTR_PMC			KTR_SUBSYS
1104 
1105 #define	PMC_DEBUG_STRSIZE		128
1106 #define	PMC_DEBUG_DEFAULT_FLAGS		{ 0, 0, 0, 0, 0, 0, 0, 0, 0 }
1107 
1108 #define	PMCDBG0(M, N, L, F) do {					\
1109 	if (pmc_debugflags.pdb_ ## M & (1 << PMC_DEBUG_MIN_ ## N))	\
1110 		CTR0(KTR_PMC, #M ":" #N ":" #L  ": " F);		\
1111 } while (0)
1112 #define	PMCDBG1(M, N, L, F, p1) do {					\
1113 	if (pmc_debugflags.pdb_ ## M & (1 << PMC_DEBUG_MIN_ ## N))	\
1114 		CTR1(KTR_PMC, #M ":" #N ":" #L  ": " F, p1);		\
1115 } while (0)
1116 #define	PMCDBG2(M, N, L, F, p1, p2) do {				\
1117 	if (pmc_debugflags.pdb_ ## M & (1 << PMC_DEBUG_MIN_ ## N))	\
1118 		CTR2(KTR_PMC, #M ":" #N ":" #L  ": " F, p1, p2);	\
1119 } while (0)
1120 #define	PMCDBG3(M, N, L, F, p1, p2, p3) do {				\
1121 	if (pmc_debugflags.pdb_ ## M & (1 << PMC_DEBUG_MIN_ ## N))	\
1122 		CTR3(KTR_PMC, #M ":" #N ":" #L  ": " F, p1, p2, p3);	\
1123 } while (0)
1124 #define	PMCDBG4(M, N, L, F, p1, p2, p3, p4) do {			\
1125 	if (pmc_debugflags.pdb_ ## M & (1 << PMC_DEBUG_MIN_ ## N))	\
1126 		CTR4(KTR_PMC, #M ":" #N ":" #L  ": " F, p1, p2, p3, p4);\
1127 } while (0)
1128 #define	PMCDBG5(M, N, L, F, p1, p2, p3, p4, p5) do {			\
1129 	if (pmc_debugflags.pdb_ ## M & (1 << PMC_DEBUG_MIN_ ## N))	\
1130 		CTR5(KTR_PMC, #M ":" #N ":" #L  ": " F, p1, p2, p3, p4,	\
1131 		    p5);						\
1132 } while (0)
1133 #define	PMCDBG6(M, N, L, F, p1, p2, p3, p4, p5, p6) do {		\
1134 	if (pmc_debugflags.pdb_ ## M & (1 << PMC_DEBUG_MIN_ ## N))	\
1135 		CTR6(KTR_PMC, #M ":" #N ":" #L  ": " F, p1, p2, p3, p4,	\
1136 		    p5, p6);						\
1137 } while (0)
1138 
1139 /* Major numbers */
1140 #define	PMC_DEBUG_MAJ_CPU		0 /* cpu switches */
1141 #define	PMC_DEBUG_MAJ_CSW		1 /* context switches */
1142 #define	PMC_DEBUG_MAJ_LOG		2 /* logging */
1143 #define	PMC_DEBUG_MAJ_MDP		3 /* machine dependent */
1144 #define	PMC_DEBUG_MAJ_MOD		4 /* misc module infrastructure */
1145 #define	PMC_DEBUG_MAJ_OWN		5 /* owner */
1146 #define	PMC_DEBUG_MAJ_PMC		6 /* pmc management */
1147 #define	PMC_DEBUG_MAJ_PRC		7 /* processes */
1148 #define	PMC_DEBUG_MAJ_SAM		8 /* sampling */
1149 
1150 /* Minor numbers */
1151 
1152 /* Common (8 bits) */
1153 #define	PMC_DEBUG_MIN_ALL		0 /* allocation */
1154 #define	PMC_DEBUG_MIN_REL		1 /* release */
1155 #define	PMC_DEBUG_MIN_OPS		2 /* ops: start, stop, ... */
1156 #define	PMC_DEBUG_MIN_INI		3 /* init */
1157 #define	PMC_DEBUG_MIN_FND		4 /* find */
1158 
1159 /* MODULE */
1160 #define	PMC_DEBUG_MIN_PMH	       14 /* pmc_hook */
1161 #define	PMC_DEBUG_MIN_PMS	       15 /* pmc_syscall */
1162 
1163 /* OWN */
1164 #define	PMC_DEBUG_MIN_ORM		8 /* owner remove */
1165 #define	PMC_DEBUG_MIN_OMR		9 /* owner maybe remove */
1166 
1167 /* PROCESSES */
1168 #define	PMC_DEBUG_MIN_TLK		8 /* link target */
1169 #define	PMC_DEBUG_MIN_TUL		9 /* unlink target */
1170 #define	PMC_DEBUG_MIN_EXT	       10 /* process exit */
1171 #define	PMC_DEBUG_MIN_EXC	       11 /* process exec */
1172 #define	PMC_DEBUG_MIN_FRK	       12 /* process fork */
1173 #define	PMC_DEBUG_MIN_ATT	       13 /* attach/detach */
1174 #define	PMC_DEBUG_MIN_SIG	       14 /* signalling */
1175 
1176 /* CONTEXT SWITCHES */
1177 #define	PMC_DEBUG_MIN_SWI		8 /* switch in */
1178 #define	PMC_DEBUG_MIN_SWO		9 /* switch out */
1179 
1180 /* PMC */
1181 #define	PMC_DEBUG_MIN_REG		8 /* pmc register */
1182 #define	PMC_DEBUG_MIN_ALR		9 /* allocate row */
1183 
1184 /* MACHINE DEPENDENT LAYER */
1185 #define	PMC_DEBUG_MIN_REA		8 /* read */
1186 #define	PMC_DEBUG_MIN_WRI		9 /* write */
1187 #define	PMC_DEBUG_MIN_CFG	       10 /* config */
1188 #define	PMC_DEBUG_MIN_STA	       11 /* start */
1189 #define	PMC_DEBUG_MIN_STO	       12 /* stop */
1190 #define	PMC_DEBUG_MIN_INT	       13 /* interrupts */
1191 
1192 /* CPU */
1193 #define	PMC_DEBUG_MIN_BND		8 /* bind */
1194 #define	PMC_DEBUG_MIN_SEL		9 /* select */
1195 
1196 /* LOG */
1197 #define	PMC_DEBUG_MIN_GTB		8 /* get buf */
1198 #define	PMC_DEBUG_MIN_SIO		9 /* schedule i/o */
1199 #define	PMC_DEBUG_MIN_FLS	       10 /* flush */
1200 #define	PMC_DEBUG_MIN_SAM	       11 /* sample */
1201 #define	PMC_DEBUG_MIN_CLO	       12 /* close */
1202 
1203 #else
1204 #define	PMCDBG0(M, N, L, F)		/* nothing */
1205 #define	PMCDBG1(M, N, L, F, p1)
1206 #define	PMCDBG2(M, N, L, F, p1, p2)
1207 #define	PMCDBG3(M, N, L, F, p1, p2, p3)
1208 #define	PMCDBG4(M, N, L, F, p1, p2, p3, p4)
1209 #define	PMCDBG5(M, N, L, F, p1, p2, p3, p4, p5)
1210 #define	PMCDBG6(M, N, L, F, p1, p2, p3, p4, p5, p6)
1211 #endif
1212 
1213 /* declare a dedicated memory pool */
1214 MALLOC_DECLARE(M_PMC);
1215 
1216 /*
1217  * Functions
1218  */
1219 
1220 struct pmc_mdep *pmc_md_initialize(void);	/* MD init function */
1221 void	pmc_md_finalize(struct pmc_mdep *_md);	/* MD fini function */
1222 int	pmc_getrowdisp(int _ri);
1223 int	pmc_process_interrupt(int _ring, struct pmc *_pm, struct trapframe *_tf);
1224 int	pmc_save_kernel_callchain(uintptr_t *_cc, int _maxsamples,
1225     struct trapframe *_tf);
1226 int	pmc_save_user_callchain(uintptr_t *_cc, int _maxsamples,
1227     struct trapframe *_tf);
1228 struct pmc_mdep *pmc_mdep_alloc(int nclasses);
1229 void pmc_mdep_free(struct pmc_mdep *md);
1230 uint64_t pmc_rdtsc(void);
1231 #endif /* _KERNEL */
1232 #endif /* _SYS_PMC_H_ */
1233