xref: /linux/arch/x86/events/perf_event.h (revision 968e3000680713f712bcf02c51c4d7bb7d4d7685)
1 /*
2  * Performance events x86 architecture header
3  *
4  *  Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de>
5  *  Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar
6  *  Copyright (C) 2009 Jaswinder Singh Rajput
7  *  Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter
8  *  Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra
9  *  Copyright (C) 2009 Intel Corporation, <markus.t.metzger@intel.com>
10  *  Copyright (C) 2009 Google, Inc., Stephane Eranian
11  *
12  *  For licencing details see kernel-base/COPYING
13  */
14 
15 #include <linux/perf_event.h>
16 
17 #include <asm/fpu/xstate.h>
18 #include <asm/intel_ds.h>
19 #include <asm/cpu.h>
20 #include <asm/msr.h>
21 
22 /* To enable MSR tracing please use the generic trace points. */
23 
24 /*
25  *          |   NHM/WSM    |      SNB     |
26  * register -------------------------------
27  *          |  HT  | no HT |  HT  | no HT |
28  *-----------------------------------------
29  * offcore  | core | core  | cpu  | core  |
30  * lbr_sel  | core | core  | cpu  | core  |
31  * ld_lat   | cpu  | core  | cpu  | core  |
32  *-----------------------------------------
33  *
34  * Given that there is a small number of shared regs,
35  * we can pre-allocate their slot in the per-cpu
36  * per-core reg tables.
37  */
38 enum extra_reg_type {
39 	EXTRA_REG_NONE		= -1, /* not used */
40 
41 	EXTRA_REG_RSP_0		= 0,  /* offcore_response_0 */
42 	EXTRA_REG_RSP_1		= 1,  /* offcore_response_1 */
43 	EXTRA_REG_LBR		= 2,  /* lbr_select */
44 	EXTRA_REG_LDLAT		= 3,  /* ld_lat_threshold */
45 	EXTRA_REG_FE		= 4,  /* fe_* */
46 	EXTRA_REG_SNOOP_0	= 5,  /* snoop response 0 */
47 	EXTRA_REG_SNOOP_1	= 6,  /* snoop response 1 */
48 
49 	EXTRA_REG_MAX		      /* number of entries needed */
50 };
51 
52 struct event_constraint {
53 	union {
54 		unsigned long	idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
55 		u64		idxmsk64;
56 	};
57 	u64		code;
58 	u64		cmask;
59 	int		weight;
60 	int		overlap;
61 	int		flags;
62 	unsigned int	size;
63 };
64 
65 static inline bool constraint_match(struct event_constraint *c, u64 ecode)
66 {
67 	return ((ecode & c->cmask) - c->code) <= (u64)c->size;
68 }
69 
70 #define PERF_ARCH(name, val)	\
71 	PERF_X86_EVENT_##name = val,
72 
73 /*
74  * struct hw_perf_event.flags flags
75  */
76 enum {
77 #include "perf_event_flags.h"
78 };
79 
80 #undef PERF_ARCH
81 
82 #define PERF_ARCH(name, val)						\
83 	static_assert((PERF_X86_EVENT_##name & PERF_EVENT_FLAG_ARCH) ==	\
84 		      PERF_X86_EVENT_##name);
85 
86 #include "perf_event_flags.h"
87 
88 #undef PERF_ARCH
89 
90 static inline bool is_topdown_count(struct perf_event *event)
91 {
92 	return event->hw.flags & PERF_X86_EVENT_TOPDOWN;
93 }
94 
95 static inline bool is_metric_event(struct perf_event *event)
96 {
97 	u64 config = event->attr.config;
98 
99 	return ((config & ARCH_PERFMON_EVENTSEL_EVENT) == 0) &&
100 		((config & INTEL_ARCH_EVENT_MASK) >= INTEL_TD_METRIC_RETIRING)  &&
101 		((config & INTEL_ARCH_EVENT_MASK) <= INTEL_TD_METRIC_MAX);
102 }
103 
104 static inline bool is_slots_event(struct perf_event *event)
105 {
106 	return (event->attr.config & INTEL_ARCH_EVENT_MASK) == INTEL_TD_SLOTS;
107 }
108 
109 static inline bool is_topdown_event(struct perf_event *event)
110 {
111 	return is_metric_event(event) || is_slots_event(event);
112 }
113 
114 int is_x86_event(struct perf_event *event);
115 
116 static inline bool check_leader_group(struct perf_event *leader, int flags)
117 {
118 	return is_x86_event(leader) ? !!(leader->hw.flags & flags) : false;
119 }
120 
121 static inline bool is_branch_counters_group(struct perf_event *event)
122 {
123 	return check_leader_group(event->group_leader, PERF_X86_EVENT_BRANCH_COUNTERS);
124 }
125 
126 static inline bool is_pebs_counter_event_group(struct perf_event *event)
127 {
128 	return check_leader_group(event->group_leader, PERF_X86_EVENT_PEBS_CNTR);
129 }
130 
131 struct amd_nb {
132 	int nb_id;  /* NorthBridge id */
133 	int refcnt; /* reference count */
134 	struct perf_event *owners[X86_PMC_IDX_MAX];
135 	struct event_constraint event_constraints[X86_PMC_IDX_MAX];
136 };
137 
138 #define PEBS_COUNTER_MASK	((1ULL << MAX_PEBS_EVENTS) - 1)
139 #define PEBS_PMI_AFTER_EACH_RECORD BIT_ULL(60)
140 #define PEBS_OUTPUT_OFFSET	61
141 #define PEBS_OUTPUT_MASK	(3ull << PEBS_OUTPUT_OFFSET)
142 #define PEBS_OUTPUT_PT		(1ull << PEBS_OUTPUT_OFFSET)
143 #define PEBS_VIA_PT_MASK	(PEBS_OUTPUT_PT | PEBS_PMI_AFTER_EACH_RECORD)
144 
145 /*
146  * Flags PEBS can handle without an PMI.
147  *
148  * TID can only be handled by flushing at context switch.
149  * REGS_USER can be handled for events limited to ring 3.
150  *
151  */
152 #define LARGE_PEBS_FLAGS \
153 	(PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_ADDR | \
154 	PERF_SAMPLE_ID | PERF_SAMPLE_CPU | PERF_SAMPLE_STREAM_ID | \
155 	PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
156 	PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR | \
157 	PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER | \
158 	PERF_SAMPLE_PERIOD | PERF_SAMPLE_CODE_PAGE_SIZE | \
159 	PERF_SAMPLE_WEIGHT_TYPE)
160 
161 #define PEBS_GP_REGS			\
162 	((1ULL << PERF_REG_X86_AX)    | \
163 	 (1ULL << PERF_REG_X86_BX)    | \
164 	 (1ULL << PERF_REG_X86_CX)    | \
165 	 (1ULL << PERF_REG_X86_DX)    | \
166 	 (1ULL << PERF_REG_X86_DI)    | \
167 	 (1ULL << PERF_REG_X86_SI)    | \
168 	 (1ULL << PERF_REG_X86_SP)    | \
169 	 (1ULL << PERF_REG_X86_BP)    | \
170 	 (1ULL << PERF_REG_X86_IP)    | \
171 	 (1ULL << PERF_REG_X86_FLAGS) | \
172 	 (1ULL << PERF_REG_X86_R8)    | \
173 	 (1ULL << PERF_REG_X86_R9)    | \
174 	 (1ULL << PERF_REG_X86_R10)   | \
175 	 (1ULL << PERF_REG_X86_R11)   | \
176 	 (1ULL << PERF_REG_X86_R12)   | \
177 	 (1ULL << PERF_REG_X86_R13)   | \
178 	 (1ULL << PERF_REG_X86_R14)   | \
179 	 (1ULL << PERF_REG_X86_R15))
180 
181 /*
182  * Per register state.
183  */
184 struct er_account {
185 	raw_spinlock_t      lock;	/* per-core: protect structure */
186 	u64                 config;	/* extra MSR config */
187 	u64                 reg;	/* extra MSR number */
188 	atomic_t            ref;	/* reference count */
189 };
190 
191 /*
192  * Per core/cpu state
193  *
194  * Used to coordinate shared registers between HT threads or
195  * among events on a single PMU.
196  */
197 struct intel_shared_regs {
198 	struct er_account       regs[EXTRA_REG_MAX];
199 	int                     refcnt;		/* per-core: #HT threads */
200 	unsigned                core_id;	/* per-core: core id */
201 };
202 
203 enum intel_excl_state_type {
204 	INTEL_EXCL_UNUSED    = 0, /* counter is unused */
205 	INTEL_EXCL_SHARED    = 1, /* counter can be used by both threads */
206 	INTEL_EXCL_EXCLUSIVE = 2, /* counter can be used by one thread only */
207 };
208 
209 struct intel_excl_states {
210 	enum intel_excl_state_type state[X86_PMC_IDX_MAX];
211 	bool sched_started; /* true if scheduling has started */
212 };
213 
214 struct intel_excl_cntrs {
215 	raw_spinlock_t	lock;
216 
217 	struct intel_excl_states states[2];
218 
219 	union {
220 		u16	has_exclusive[2];
221 		u32	exclusive_present;
222 	};
223 
224 	int		refcnt;		/* per-core: #HT threads */
225 	unsigned	core_id;	/* per-core: core id */
226 };
227 
228 struct x86_perf_task_context;
229 #define MAX_LBR_ENTRIES		32
230 
231 enum {
232 	LBR_FORMAT_32		= 0x00,
233 	LBR_FORMAT_LIP		= 0x01,
234 	LBR_FORMAT_EIP		= 0x02,
235 	LBR_FORMAT_EIP_FLAGS	= 0x03,
236 	LBR_FORMAT_EIP_FLAGS2	= 0x04,
237 	LBR_FORMAT_INFO		= 0x05,
238 	LBR_FORMAT_TIME		= 0x06,
239 	LBR_FORMAT_INFO2	= 0x07,
240 	LBR_FORMAT_MAX_KNOWN    = LBR_FORMAT_INFO2,
241 };
242 
243 enum {
244 	X86_PERF_KFREE_SHARED = 0,
245 	X86_PERF_KFREE_EXCL   = 1,
246 	X86_PERF_KFREE_MAX
247 };
248 
249 struct cpu_hw_events {
250 	/*
251 	 * Generic x86 PMC bits
252 	 */
253 	struct perf_event	*events[X86_PMC_IDX_MAX]; /* in counter order */
254 	unsigned long		active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
255 	unsigned long		dirty[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
256 	int			enabled;
257 
258 	int			n_events; /* the # of events in the below arrays */
259 	int			n_added;  /* the # last events in the below arrays;
260 					     they've never been enabled yet */
261 	int			n_txn;    /* the # last events in the below arrays;
262 					     added in the current transaction */
263 	int			n_txn_pair;
264 	int			n_txn_metric;
265 	int			assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
266 	u64			tags[X86_PMC_IDX_MAX];
267 
268 	struct perf_event	*event_list[X86_PMC_IDX_MAX]; /* in enabled order */
269 	struct event_constraint	*event_constraint[X86_PMC_IDX_MAX];
270 
271 	int			n_excl; /* the number of exclusive events */
272 
273 	unsigned int		txn_flags;
274 	int			is_fake;
275 
276 	/*
277 	 * Intel DebugStore bits
278 	 */
279 	struct debug_store	*ds;
280 	void			*ds_pebs_vaddr;
281 	void			*ds_bts_vaddr;
282 	u64			pebs_enabled;
283 	int			n_pebs;
284 	int			n_large_pebs;
285 	int			n_pebs_via_pt;
286 	int			pebs_output;
287 
288 	/* Current super set of events hardware configuration */
289 	u64			pebs_data_cfg;
290 	u64			active_pebs_data_cfg;
291 	int			pebs_record_size;
292 
293 	/* Intel Fixed counter configuration */
294 	u64			fixed_ctrl_val;
295 	u64			active_fixed_ctrl_val;
296 
297 	/*
298 	 * Intel LBR bits
299 	 */
300 	int				lbr_users;
301 	int				lbr_pebs_users;
302 	struct perf_branch_stack	lbr_stack;
303 	struct perf_branch_entry	lbr_entries[MAX_LBR_ENTRIES];
304 	u64				lbr_counters[MAX_LBR_ENTRIES]; /* branch stack extra */
305 	union {
306 		struct er_account		*lbr_sel;
307 		struct er_account		*lbr_ctl;
308 	};
309 	u64				br_sel;
310 	void				*last_task_ctx;
311 	int				last_log_id;
312 	int				lbr_select;
313 	void				*lbr_xsave;
314 
315 	/*
316 	 * Intel host/guest exclude bits
317 	 */
318 	u64				intel_ctrl_guest_mask;
319 	u64				intel_ctrl_host_mask;
320 	struct perf_guest_switch_msr	guest_switch_msrs[X86_PMC_IDX_MAX];
321 
322 	/*
323 	 * Intel checkpoint mask
324 	 */
325 	u64				intel_cp_status;
326 
327 	/*
328 	 * manage shared (per-core, per-cpu) registers
329 	 * used on Intel NHM/WSM/SNB
330 	 */
331 	struct intel_shared_regs	*shared_regs;
332 	/*
333 	 * manage exclusive counter access between hyperthread
334 	 */
335 	struct event_constraint *constraint_list; /* in enable order */
336 	struct intel_excl_cntrs		*excl_cntrs;
337 	int excl_thread_id; /* 0 or 1 */
338 
339 	/*
340 	 * SKL TSX_FORCE_ABORT shadow
341 	 */
342 	u64				tfa_shadow;
343 
344 	/*
345 	 * Perf Metrics
346 	 */
347 	/* number of accepted metrics events */
348 	int				n_metric;
349 
350 	/*
351 	 * AMD specific bits
352 	 */
353 	struct amd_nb			*amd_nb;
354 	int				brs_active; /* BRS is enabled */
355 
356 	/* Inverted mask of bits to clear in the perf_ctr ctrl registers */
357 	u64				perf_ctr_virt_mask;
358 	int				n_pair; /* Large increment events */
359 
360 	void				*kfree_on_online[X86_PERF_KFREE_MAX];
361 
362 	struct pmu			*pmu;
363 };
364 
365 #define __EVENT_CONSTRAINT_RANGE(c, e, n, m, w, o, f) {	\
366 	{ .idxmsk64 = (n) },		\
367 	.code = (c),			\
368 	.size = (e) - (c),		\
369 	.cmask = (m),			\
370 	.weight = (w),			\
371 	.overlap = (o),			\
372 	.flags = f,			\
373 }
374 
375 #define __EVENT_CONSTRAINT(c, n, m, w, o, f) \
376 	__EVENT_CONSTRAINT_RANGE(c, c, n, m, w, o, f)
377 
378 #define EVENT_CONSTRAINT(c, n, m)	\
379 	__EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0)
380 
381 /*
382  * The constraint_match() function only works for 'simple' event codes
383  * and not for extended (AMD64_EVENTSEL_EVENT) events codes.
384  */
385 #define EVENT_CONSTRAINT_RANGE(c, e, n, m) \
386 	__EVENT_CONSTRAINT_RANGE(c, e, n, m, HWEIGHT(n), 0, 0)
387 
388 #define INTEL_EXCLEVT_CONSTRAINT(c, n)	\
389 	__EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT, HWEIGHT(n),\
390 			   0, PERF_X86_EVENT_EXCL)
391 
392 /*
393  * The overlap flag marks event constraints with overlapping counter
394  * masks. This is the case if the counter mask of such an event is not
395  * a subset of any other counter mask of a constraint with an equal or
396  * higher weight, e.g.:
397  *
398  *  c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
399  *  c_another1 = EVENT_CONSTRAINT(0, 0x07, 0);
400  *  c_another2 = EVENT_CONSTRAINT(0, 0x38, 0);
401  *
402  * The event scheduler may not select the correct counter in the first
403  * cycle because it needs to know which subsequent events will be
404  * scheduled. It may fail to schedule the events then. So we set the
405  * overlap flag for such constraints to give the scheduler a hint which
406  * events to select for counter rescheduling.
407  *
408  * Care must be taken as the rescheduling algorithm is O(n!) which
409  * will increase scheduling cycles for an over-committed system
410  * dramatically.  The number of such EVENT_CONSTRAINT_OVERLAP() macros
411  * and its counter masks must be kept at a minimum.
412  */
413 #define EVENT_CONSTRAINT_OVERLAP(c, n, m)	\
414 	__EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0)
415 
416 /*
417  * Constraint on the Event code.
418  */
419 #define INTEL_EVENT_CONSTRAINT(c, n)	\
420 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
421 
422 /*
423  * Constraint on a range of Event codes
424  */
425 #define INTEL_EVENT_CONSTRAINT_RANGE(c, e, n)			\
426 	EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT)
427 
428 /*
429  * Constraint on the Event code + UMask + fixed-mask
430  *
431  * filter mask to validate fixed counter events.
432  * the following filters disqualify for fixed counters:
433  *  - inv
434  *  - edge
435  *  - cnt-mask
436  *  - in_tx
437  *  - in_tx_checkpointed
438  *  The other filters are supported by fixed counters.
439  *  The any-thread option is supported starting with v3.
440  */
441 #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED)
442 #define FIXED_EVENT_CONSTRAINT(c, n)	\
443 	EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
444 
445 /*
446  * The special metric counters do not actually exist. They are calculated from
447  * the combination of the FxCtr3 + MSR_PERF_METRICS.
448  *
449  * The special metric counters are mapped to a dummy offset for the scheduler.
450  * The sharing between multiple users of the same metric without multiplexing
451  * is not allowed, even though the hardware supports that in principle.
452  */
453 
454 #define METRIC_EVENT_CONSTRAINT(c, n)					\
455 	EVENT_CONSTRAINT(c, (1ULL << (INTEL_PMC_IDX_METRIC_BASE + n)),	\
456 			 INTEL_ARCH_EVENT_MASK)
457 
458 /*
459  * Constraint on the Event code + UMask
460  */
461 #define INTEL_UEVENT_CONSTRAINT(c, n)	\
462 	EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
463 
464 /* Constraint on specific umask bit only + event */
465 #define INTEL_UBIT_EVENT_CONSTRAINT(c, n)	\
466 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|(c))
467 
468 /* Like UEVENT_CONSTRAINT, but match flags too */
469 #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n)	\
470 	EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
471 
472 #define INTEL_EXCLUEVT_CONSTRAINT(c, n)	\
473 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK, \
474 			   HWEIGHT(n), 0, PERF_X86_EVENT_EXCL)
475 
476 #define INTEL_PLD_CONSTRAINT(c, n)	\
477 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
478 			   HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
479 
480 #define INTEL_PSD_CONSTRAINT(c, n)	\
481 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
482 			   HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_STLAT)
483 
484 #define INTEL_PST_CONSTRAINT(c, n)	\
485 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
486 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
487 
488 #define INTEL_HYBRID_LAT_CONSTRAINT(c, n)	\
489 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
490 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LAT_HYBRID)
491 
492 #define INTEL_HYBRID_LDLAT_CONSTRAINT(c, n)	\
493 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
494 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LAT_HYBRID|PERF_X86_EVENT_PEBS_LD_HSW)
495 
496 #define INTEL_HYBRID_STLAT_CONSTRAINT(c, n)	\
497 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
498 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LAT_HYBRID|PERF_X86_EVENT_PEBS_ST_HSW)
499 
500 /* Event constraint, but match on all event flags too. */
501 #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
502 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
503 
504 #define INTEL_FLAGS_EVENT_CONSTRAINT_RANGE(c, e, n)			\
505 	EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
506 
507 /* Check only flags, but allow all event/umask */
508 #define INTEL_ALL_EVENT_CONSTRAINT(code, n)	\
509 	EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
510 
511 /* Check flags and event code, and set the HSW store flag */
512 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
513 	__EVENT_CONSTRAINT(code, n, 			\
514 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
515 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
516 
517 /* Check flags and event code, and set the HSW load flag */
518 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
519 	__EVENT_CONSTRAINT(code, n,			\
520 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
521 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
522 
523 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD_RANGE(code, end, n) \
524 	__EVENT_CONSTRAINT_RANGE(code, end, n,				\
525 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
526 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
527 
528 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_XLD(code, n) \
529 	__EVENT_CONSTRAINT(code, n,			\
530 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
531 			  HWEIGHT(n), 0, \
532 			  PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
533 
534 /* Check flags and event code/umask, and set the HSW store flag */
535 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
536 	__EVENT_CONSTRAINT(code, n, 			\
537 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
538 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
539 
540 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XST(code, n) \
541 	__EVENT_CONSTRAINT(code, n,			\
542 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
543 			  HWEIGHT(n), 0, \
544 			  PERF_X86_EVENT_PEBS_ST_HSW|PERF_X86_EVENT_EXCL)
545 
546 /* Check flags and event code/umask, and set the HSW load flag */
547 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
548 	__EVENT_CONSTRAINT(code, n, 			\
549 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
550 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
551 
552 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XLD(code, n) \
553 	__EVENT_CONSTRAINT(code, n,			\
554 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
555 			  HWEIGHT(n), 0, \
556 			  PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
557 
558 /* Check flags and event code/umask, and set the HSW N/A flag */
559 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
560 	__EVENT_CONSTRAINT(code, n, 			\
561 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
562 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
563 
564 
565 /*
566  * We define the end marker as having a weight of -1
567  * to enable blacklisting of events using a counter bitmask
568  * of zero and thus a weight of zero.
569  * The end marker has a weight that cannot possibly be
570  * obtained from counting the bits in the bitmask.
571  */
572 #define EVENT_CONSTRAINT_END { .weight = -1 }
573 
574 /*
575  * Check for end marker with weight == -1
576  */
577 #define for_each_event_constraint(e, c)	\
578 	for ((e) = (c); (e)->weight != -1; (e)++)
579 
580 /*
581  * Extra registers for specific events.
582  *
583  * Some events need large masks and require external MSRs.
584  * Those extra MSRs end up being shared for all events on
585  * a PMU and sometimes between PMU of sibling HT threads.
586  * In either case, the kernel needs to handle conflicting
587  * accesses to those extra, shared, regs. The data structure
588  * to manage those registers is stored in cpu_hw_event.
589  */
590 struct extra_reg {
591 	unsigned int		event;
592 	unsigned int		msr;
593 	u64			config_mask;
594 	u64			valid_mask;
595 	int			idx;  /* per_xxx->regs[] reg index */
596 	bool			extra_msr_access;
597 };
598 
599 #define EVENT_EXTRA_REG(e, ms, m, vm, i) {	\
600 	.event = (e),			\
601 	.msr = (ms),			\
602 	.config_mask = (m),		\
603 	.valid_mask = (vm),		\
604 	.idx = EXTRA_REG_##i,		\
605 	.extra_msr_access = true,	\
606 	}
607 
608 #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx)	\
609 	EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
610 
611 #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
612 	EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
613 			ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
614 
615 #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
616 	INTEL_UEVENT_EXTRA_REG(c, \
617 			       MSR_PEBS_LD_LAT_THRESHOLD, \
618 			       0xffff, \
619 			       LDLAT)
620 
621 #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
622 
623 union perf_capabilities {
624 	struct {
625 		u64	lbr_format:6;
626 		u64	pebs_trap:1;
627 		u64	pebs_arch_reg:1;
628 		u64	pebs_format:4;
629 		u64	smm_freeze:1;
630 		/*
631 		 * PMU supports separate counter range for writing
632 		 * values > 32bit.
633 		 */
634 		u64	full_width_write:1;
635 		u64     pebs_baseline:1;
636 		u64	perf_metrics:1;
637 		u64	pebs_output_pt_available:1;
638 		u64	pebs_timing_info:1;
639 		u64	anythread_deprecated:1;
640 		u64	rdpmc_metrics_clear:1;
641 	};
642 	u64	capabilities;
643 };
644 
645 struct x86_pmu_quirk {
646 	struct x86_pmu_quirk *next;
647 	void (*func)(void);
648 };
649 
650 union x86_pmu_config {
651 	struct {
652 		u64 event:8,
653 		    umask:8,
654 		    usr:1,
655 		    os:1,
656 		    edge:1,
657 		    pc:1,
658 		    interrupt:1,
659 		    __reserved1:1,
660 		    en:1,
661 		    inv:1,
662 		    cmask:8,
663 		    event2:4,
664 		    __reserved2:4,
665 		    go:1,
666 		    ho:1;
667 	} bits;
668 	u64 value;
669 };
670 
671 #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
672 
673 enum {
674 	x86_lbr_exclusive_lbr,
675 	x86_lbr_exclusive_bts,
676 	x86_lbr_exclusive_pt,
677 	x86_lbr_exclusive_max,
678 };
679 
680 #define PERF_PEBS_DATA_SOURCE_MAX	0x100
681 #define PERF_PEBS_DATA_SOURCE_MASK	(PERF_PEBS_DATA_SOURCE_MAX - 1)
682 #define PERF_PEBS_DATA_SOURCE_GRT_MAX	0x10
683 #define PERF_PEBS_DATA_SOURCE_GRT_MASK	(PERF_PEBS_DATA_SOURCE_GRT_MAX - 1)
684 
685 #define X86_HYBRID_PMU_ATOM_IDX		0
686 #define X86_HYBRID_PMU_CORE_IDX		1
687 #define X86_HYBRID_PMU_TINY_IDX		2
688 
689 enum hybrid_pmu_type {
690 	not_hybrid,
691 	hybrid_small		= BIT(X86_HYBRID_PMU_ATOM_IDX),
692 	hybrid_big		= BIT(X86_HYBRID_PMU_CORE_IDX),
693 	hybrid_tiny		= BIT(X86_HYBRID_PMU_TINY_IDX),
694 
695 	/* The belows are only used for matching */
696 	hybrid_big_small	= hybrid_big   | hybrid_small,
697 	hybrid_small_tiny	= hybrid_small | hybrid_tiny,
698 	hybrid_big_small_tiny	= hybrid_big   | hybrid_small_tiny,
699 };
700 
701 struct x86_hybrid_pmu {
702 	struct pmu			pmu;
703 	const char			*name;
704 	enum hybrid_pmu_type		pmu_type;
705 	cpumask_t			supported_cpus;
706 	union perf_capabilities		intel_cap;
707 	u64				intel_ctrl;
708 	u64				pebs_events_mask;
709 	u64				config_mask;
710 	union {
711 			u64		cntr_mask64;
712 			unsigned long	cntr_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
713 	};
714 	union {
715 			u64		fixed_cntr_mask64;
716 			unsigned long	fixed_cntr_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
717 	};
718 	struct event_constraint		unconstrained;
719 
720 	u64				hw_cache_event_ids
721 					[PERF_COUNT_HW_CACHE_MAX]
722 					[PERF_COUNT_HW_CACHE_OP_MAX]
723 					[PERF_COUNT_HW_CACHE_RESULT_MAX];
724 	u64				hw_cache_extra_regs
725 					[PERF_COUNT_HW_CACHE_MAX]
726 					[PERF_COUNT_HW_CACHE_OP_MAX]
727 					[PERF_COUNT_HW_CACHE_RESULT_MAX];
728 	struct event_constraint		*event_constraints;
729 	struct event_constraint		*pebs_constraints;
730 	struct extra_reg		*extra_regs;
731 
732 	unsigned int			late_ack	:1,
733 					mid_ack		:1,
734 					enabled_ack	:1;
735 
736 	u64				pebs_data_source[PERF_PEBS_DATA_SOURCE_MAX];
737 };
738 
739 static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu)
740 {
741 	return container_of(pmu, struct x86_hybrid_pmu, pmu);
742 }
743 
744 extern struct static_key_false perf_is_hybrid;
745 #define is_hybrid()		static_branch_unlikely(&perf_is_hybrid)
746 
747 #define hybrid(_pmu, _field)				\
748 (*({							\
749 	typeof(&x86_pmu._field) __Fp = &x86_pmu._field;	\
750 							\
751 	if (is_hybrid() && (_pmu))			\
752 		__Fp = &hybrid_pmu(_pmu)->_field;	\
753 							\
754 	__Fp;						\
755 }))
756 
757 #define hybrid_var(_pmu, _var)				\
758 (*({							\
759 	typeof(&_var) __Fp = &_var;			\
760 							\
761 	if (is_hybrid() && (_pmu))			\
762 		__Fp = &hybrid_pmu(_pmu)->_var;		\
763 							\
764 	__Fp;						\
765 }))
766 
767 #define hybrid_bit(_pmu, _field)			\
768 ({							\
769 	bool __Fp = x86_pmu._field;			\
770 							\
771 	if (is_hybrid() && (_pmu))			\
772 		__Fp = hybrid_pmu(_pmu)->_field;	\
773 							\
774 	__Fp;						\
775 })
776 
777 /*
778  * struct x86_pmu - generic x86 pmu
779  */
780 struct x86_pmu {
781 	/*
782 	 * Generic x86 PMC bits
783 	 */
784 	const char	*name;
785 	int		version;
786 	int		(*handle_irq)(struct pt_regs *);
787 	void		(*disable_all)(void);
788 	void		(*enable_all)(int added);
789 	void		(*enable)(struct perf_event *);
790 	void		(*disable)(struct perf_event *);
791 	void		(*assign)(struct perf_event *event, int idx);
792 	void		(*add)(struct perf_event *);
793 	void		(*del)(struct perf_event *);
794 	void		(*read)(struct perf_event *event);
795 	int		(*set_period)(struct perf_event *event);
796 	u64		(*update)(struct perf_event *event);
797 	int		(*hw_config)(struct perf_event *event);
798 	int		(*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
799 	void		(*late_setup)(void);
800 	unsigned	eventsel;
801 	unsigned	perfctr;
802 	unsigned	fixedctr;
803 	int		(*addr_offset)(int index, bool eventsel);
804 	int		(*rdpmc_index)(int index);
805 	u64		(*event_map)(int);
806 	int		max_events;
807 	u64		config_mask;
808 	union {
809 			u64		cntr_mask64;
810 			unsigned long	cntr_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
811 	};
812 	union {
813 			u64		fixed_cntr_mask64;
814 			unsigned long	fixed_cntr_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
815 	};
816 	int		cntval_bits;
817 	u64		cntval_mask;
818 	union {
819 			unsigned long events_maskl;
820 			unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
821 	};
822 	int		events_mask_len;
823 	int		apic;
824 	u64		max_period;
825 	struct event_constraint *
826 			(*get_event_constraints)(struct cpu_hw_events *cpuc,
827 						 int idx,
828 						 struct perf_event *event);
829 
830 	void		(*put_event_constraints)(struct cpu_hw_events *cpuc,
831 						 struct perf_event *event);
832 
833 	void		(*start_scheduling)(struct cpu_hw_events *cpuc);
834 
835 	void		(*commit_scheduling)(struct cpu_hw_events *cpuc, int idx, int cntr);
836 
837 	void		(*stop_scheduling)(struct cpu_hw_events *cpuc);
838 
839 	struct event_constraint *event_constraints;
840 	struct x86_pmu_quirk *quirks;
841 	void		(*limit_period)(struct perf_event *event, s64 *l);
842 
843 	/* PMI handler bits */
844 	unsigned int	late_ack		:1,
845 			mid_ack			:1,
846 			enabled_ack		:1;
847 	/*
848 	 * sysfs attrs
849 	 */
850 	int		attr_rdpmc_broken;
851 	int		attr_rdpmc;
852 	struct attribute **format_attrs;
853 
854 	ssize_t		(*events_sysfs_show)(char *page, u64 config);
855 	const struct attribute_group **attr_update;
856 
857 	unsigned long	attr_freeze_on_smi;
858 
859 	/*
860 	 * CPU Hotplug hooks
861 	 */
862 	int		(*cpu_prepare)(int cpu);
863 	void		(*cpu_starting)(int cpu);
864 	void		(*cpu_dying)(int cpu);
865 	void		(*cpu_dead)(int cpu);
866 
867 	void		(*check_microcode)(void);
868 	void		(*sched_task)(struct perf_event_pmu_context *pmu_ctx,
869 				      struct task_struct *task, bool sched_in);
870 
871 	/*
872 	 * Intel Arch Perfmon v2+
873 	 */
874 	u64			intel_ctrl;
875 	union perf_capabilities intel_cap;
876 
877 	/*
878 	 * Intel DebugStore bits
879 	 */
880 	unsigned int	bts			:1,
881 			bts_active		:1,
882 			pebs			:1,
883 			pebs_active		:1,
884 			pebs_broken		:1,
885 			pebs_prec_dist		:1,
886 			pebs_no_tlb		:1,
887 			pebs_no_isolation	:1,
888 			pebs_block		:1,
889 			pebs_ept		:1;
890 	int		pebs_record_size;
891 	int		pebs_buffer_size;
892 	u64		pebs_events_mask;
893 	void		(*drain_pebs)(struct pt_regs *regs, struct perf_sample_data *data);
894 	struct event_constraint *pebs_constraints;
895 	void		(*pebs_aliases)(struct perf_event *event);
896 	u64		(*pebs_latency_data)(struct perf_event *event, u64 status);
897 	unsigned long	large_pebs_flags;
898 	u64		rtm_abort_event;
899 	u64		pebs_capable;
900 
901 	/*
902 	 * Intel LBR
903 	 */
904 	unsigned int	lbr_tos, lbr_from, lbr_to,
905 			lbr_info, lbr_nr;	   /* LBR base regs and size */
906 	union {
907 		u64	lbr_sel_mask;		   /* LBR_SELECT valid bits */
908 		u64	lbr_ctl_mask;		   /* LBR_CTL valid bits */
909 	};
910 	union {
911 		const int	*lbr_sel_map;	   /* lbr_select mappings */
912 		int		*lbr_ctl_map;	   /* LBR_CTL mappings */
913 	};
914 	u64		lbr_callstack_users;	   /* lbr callstack system wide users */
915 	bool		lbr_double_abort;	   /* duplicated lbr aborts */
916 	bool		lbr_pt_coexist;		   /* (LBR|BTS) may coexist with PT */
917 
918 	unsigned int	lbr_has_info:1;
919 	unsigned int	lbr_has_tsx:1;
920 	unsigned int	lbr_from_flags:1;
921 	unsigned int	lbr_to_cycles:1;
922 
923 	/*
924 	 * Intel Architectural LBR CPUID Enumeration
925 	 */
926 	unsigned int	lbr_depth_mask:8;
927 	unsigned int	lbr_deep_c_reset:1;
928 	unsigned int	lbr_lip:1;
929 	unsigned int	lbr_cpl:1;
930 	unsigned int	lbr_filter:1;
931 	unsigned int	lbr_call_stack:1;
932 	unsigned int	lbr_mispred:1;
933 	unsigned int	lbr_timed_lbr:1;
934 	unsigned int	lbr_br_type:1;
935 	unsigned int	lbr_counters:4;
936 
937 	void		(*lbr_reset)(void);
938 	void		(*lbr_read)(struct cpu_hw_events *cpuc);
939 	void		(*lbr_save)(void *ctx);
940 	void		(*lbr_restore)(void *ctx);
941 
942 	/*
943 	 * Intel PT/LBR/BTS are exclusive
944 	 */
945 	atomic_t	lbr_exclusive[x86_lbr_exclusive_max];
946 
947 	/*
948 	 * Intel perf metrics
949 	 */
950 	int		num_topdown_events;
951 
952 	/*
953 	 * AMD bits
954 	 */
955 	unsigned int	amd_nb_constraints : 1;
956 	u64		perf_ctr_pair_en;
957 
958 	/*
959 	 * Extra registers for events
960 	 */
961 	struct extra_reg *extra_regs;
962 	unsigned int flags;
963 
964 	/*
965 	 * Intel host/guest support (KVM)
966 	 */
967 	struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr, void *data);
968 
969 	/*
970 	 * Check period value for PERF_EVENT_IOC_PERIOD ioctl.
971 	 */
972 	int (*check_period) (struct perf_event *event, u64 period);
973 
974 	int (*aux_output_match) (struct perf_event *event);
975 
976 	void (*filter)(struct pmu *pmu, int cpu, bool *ret);
977 	/*
978 	 * Hybrid support
979 	 *
980 	 * Most PMU capabilities are the same among different hybrid PMUs.
981 	 * The global x86_pmu saves the architecture capabilities, which
982 	 * are available for all PMUs. The hybrid_pmu only includes the
983 	 * unique capabilities.
984 	 */
985 	int				num_hybrid_pmus;
986 	struct x86_hybrid_pmu		*hybrid_pmu;
987 	enum intel_cpu_type (*get_hybrid_cpu_type)	(void);
988 };
989 
990 struct x86_perf_task_context_opt {
991 	int lbr_callstack_users;
992 	int lbr_stack_state;
993 	int log_id;
994 };
995 
996 struct x86_perf_task_context {
997 	u64 lbr_sel;
998 	int tos;
999 	int valid_lbrs;
1000 	struct x86_perf_task_context_opt opt;
1001 	struct lbr_entry lbr[MAX_LBR_ENTRIES];
1002 };
1003 
1004 struct x86_perf_task_context_arch_lbr {
1005 	struct x86_perf_task_context_opt opt;
1006 	struct lbr_entry entries[];
1007 };
1008 
1009 /*
1010  * Add padding to guarantee the 64-byte alignment of the state buffer.
1011  *
1012  * The structure is dynamically allocated. The size of the LBR state may vary
1013  * based on the number of LBR registers.
1014  *
1015  * Do not put anything after the LBR state.
1016  */
1017 struct x86_perf_task_context_arch_lbr_xsave {
1018 	struct x86_perf_task_context_opt		opt;
1019 
1020 	union {
1021 		struct xregs_state			xsave;
1022 		struct {
1023 			struct fxregs_state		i387;
1024 			struct xstate_header		header;
1025 			struct arch_lbr_state		lbr;
1026 		} __attribute__ ((packed, aligned (XSAVE_ALIGNMENT)));
1027 	};
1028 };
1029 
1030 #define x86_add_quirk(func_)						\
1031 do {									\
1032 	static struct x86_pmu_quirk __quirk __initdata = {		\
1033 		.func = func_,						\
1034 	};								\
1035 	__quirk.next = x86_pmu.quirks;					\
1036 	x86_pmu.quirks = &__quirk;					\
1037 } while (0)
1038 
1039 /*
1040  * x86_pmu flags
1041  */
1042 #define PMU_FL_NO_HT_SHARING	0x1 /* no hyper-threading resource sharing */
1043 #define PMU_FL_HAS_RSP_1	0x2 /* has 2 equivalent offcore_rsp regs   */
1044 #define PMU_FL_EXCL_CNTRS	0x4 /* has exclusive counter requirements  */
1045 #define PMU_FL_EXCL_ENABLED	0x8 /* exclusive counter active */
1046 #define PMU_FL_PEBS_ALL		0x10 /* all events are valid PEBS events */
1047 #define PMU_FL_TFA		0x20 /* deal with TSX force abort */
1048 #define PMU_FL_PAIR		0x40 /* merge counters for large incr. events */
1049 #define PMU_FL_INSTR_LATENCY	0x80 /* Support Instruction Latency in PEBS Memory Info Record */
1050 #define PMU_FL_MEM_LOADS_AUX	0x100 /* Require an auxiliary event for the complete memory info */
1051 #define PMU_FL_RETIRE_LATENCY	0x200 /* Support Retire Latency in PEBS */
1052 #define PMU_FL_BR_CNTR		0x400 /* Support branch counter logging */
1053 
1054 #define EVENT_VAR(_id)  event_attr_##_id
1055 #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
1056 
1057 #define EVENT_ATTR(_name, _id)						\
1058 static struct perf_pmu_events_attr EVENT_VAR(_id) = {			\
1059 	.attr		= __ATTR(_name, 0444, events_sysfs_show, NULL),	\
1060 	.id		= PERF_COUNT_HW_##_id,				\
1061 	.event_str	= NULL,						\
1062 };
1063 
1064 #define EVENT_ATTR_STR(_name, v, str)					\
1065 static struct perf_pmu_events_attr event_attr_##v = {			\
1066 	.attr		= __ATTR(_name, 0444, events_sysfs_show, NULL),	\
1067 	.id		= 0,						\
1068 	.event_str	= str,						\
1069 };
1070 
1071 #define EVENT_ATTR_STR_HT(_name, v, noht, ht)				\
1072 static struct perf_pmu_events_ht_attr event_attr_##v = {		\
1073 	.attr		= __ATTR(_name, 0444, events_ht_sysfs_show, NULL),\
1074 	.id		= 0,						\
1075 	.event_str_noht	= noht,						\
1076 	.event_str_ht	= ht,						\
1077 }
1078 
1079 #define EVENT_ATTR_STR_HYBRID(_name, v, str, _pmu)			\
1080 static struct perf_pmu_events_hybrid_attr event_attr_##v = {		\
1081 	.attr		= __ATTR(_name, 0444, events_hybrid_sysfs_show, NULL),\
1082 	.id		= 0,						\
1083 	.event_str	= str,						\
1084 	.pmu_type	= _pmu,						\
1085 }
1086 
1087 #define FORMAT_HYBRID_PTR(_id) (&format_attr_hybrid_##_id.attr.attr)
1088 
1089 #define FORMAT_ATTR_HYBRID(_name, _pmu)					\
1090 static struct perf_pmu_format_hybrid_attr format_attr_hybrid_##_name = {\
1091 	.attr		= __ATTR_RO(_name),				\
1092 	.pmu_type	= _pmu,						\
1093 }
1094 
1095 struct pmu *x86_get_pmu(unsigned int cpu);
1096 extern struct x86_pmu x86_pmu __read_mostly;
1097 
1098 DECLARE_STATIC_CALL(x86_pmu_set_period, *x86_pmu.set_period);
1099 DECLARE_STATIC_CALL(x86_pmu_update,     *x86_pmu.update);
1100 DECLARE_STATIC_CALL(x86_pmu_drain_pebs,	*x86_pmu.drain_pebs);
1101 DECLARE_STATIC_CALL(x86_pmu_late_setup,	*x86_pmu.late_setup);
1102 
1103 static __always_inline struct x86_perf_task_context_opt *task_context_opt(void *ctx)
1104 {
1105 	if (static_cpu_has(X86_FEATURE_ARCH_LBR))
1106 		return &((struct x86_perf_task_context_arch_lbr *)ctx)->opt;
1107 
1108 	return &((struct x86_perf_task_context *)ctx)->opt;
1109 }
1110 
1111 static inline bool x86_pmu_has_lbr_callstack(void)
1112 {
1113 	return  x86_pmu.lbr_sel_map &&
1114 		x86_pmu.lbr_sel_map[PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT] > 0;
1115 }
1116 
1117 DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
1118 DECLARE_PER_CPU(u64 [X86_PMC_IDX_MAX], pmc_prev_left);
1119 
1120 int x86_perf_event_set_period(struct perf_event *event);
1121 
1122 /*
1123  * Generalized hw caching related hw_event table, filled
1124  * in on a per model basis. A value of 0 means
1125  * 'not supported', -1 means 'hw_event makes no sense on
1126  * this CPU', any other value means the raw hw_event
1127  * ID.
1128  */
1129 
1130 #define C(x) PERF_COUNT_HW_CACHE_##x
1131 
1132 extern u64 __read_mostly hw_cache_event_ids
1133 				[PERF_COUNT_HW_CACHE_MAX]
1134 				[PERF_COUNT_HW_CACHE_OP_MAX]
1135 				[PERF_COUNT_HW_CACHE_RESULT_MAX];
1136 extern u64 __read_mostly hw_cache_extra_regs
1137 				[PERF_COUNT_HW_CACHE_MAX]
1138 				[PERF_COUNT_HW_CACHE_OP_MAX]
1139 				[PERF_COUNT_HW_CACHE_RESULT_MAX];
1140 
1141 u64 x86_perf_event_update(struct perf_event *event);
1142 
1143 static inline u64 intel_pmu_topdown_event_update(struct perf_event *event, u64 *val)
1144 {
1145 	return x86_perf_event_update(event);
1146 }
1147 DECLARE_STATIC_CALL(intel_pmu_update_topdown_event, intel_pmu_topdown_event_update);
1148 
1149 static inline unsigned int x86_pmu_config_addr(int index)
1150 {
1151 	return x86_pmu.eventsel + (x86_pmu.addr_offset ?
1152 				   x86_pmu.addr_offset(index, true) : index);
1153 }
1154 
1155 static inline unsigned int x86_pmu_event_addr(int index)
1156 {
1157 	return x86_pmu.perfctr + (x86_pmu.addr_offset ?
1158 				  x86_pmu.addr_offset(index, false) : index);
1159 }
1160 
1161 static inline unsigned int x86_pmu_fixed_ctr_addr(int index)
1162 {
1163 	return x86_pmu.fixedctr + (x86_pmu.addr_offset ?
1164 				   x86_pmu.addr_offset(index, false) : index);
1165 }
1166 
1167 static inline int x86_pmu_rdpmc_index(int index)
1168 {
1169 	return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
1170 }
1171 
1172 bool check_hw_exists(struct pmu *pmu, unsigned long *cntr_mask,
1173 		     unsigned long *fixed_cntr_mask);
1174 
1175 int x86_add_exclusive(unsigned int what);
1176 
1177 void x86_del_exclusive(unsigned int what);
1178 
1179 int x86_reserve_hardware(void);
1180 
1181 void x86_release_hardware(void);
1182 
1183 int x86_pmu_max_precise(void);
1184 
1185 void hw_perf_lbr_event_destroy(struct perf_event *event);
1186 
1187 int x86_setup_perfctr(struct perf_event *event);
1188 
1189 int x86_pmu_hw_config(struct perf_event *event);
1190 
1191 void x86_pmu_disable_all(void);
1192 
1193 static inline bool has_amd_brs(struct hw_perf_event *hwc)
1194 {
1195 	return hwc->flags & PERF_X86_EVENT_AMD_BRS;
1196 }
1197 
1198 static inline bool is_counter_pair(struct hw_perf_event *hwc)
1199 {
1200 	return hwc->flags & PERF_X86_EVENT_PAIR;
1201 }
1202 
1203 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
1204 					  u64 enable_mask)
1205 {
1206 	u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
1207 
1208 	if (hwc->extra_reg.reg)
1209 		wrmsrq(hwc->extra_reg.reg, hwc->extra_reg.config);
1210 
1211 	/*
1212 	 * Add enabled Merge event on next counter
1213 	 * if large increment event being enabled on this counter
1214 	 */
1215 	if (is_counter_pair(hwc))
1216 		wrmsrq(x86_pmu_config_addr(hwc->idx + 1), x86_pmu.perf_ctr_pair_en);
1217 
1218 	wrmsrq(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
1219 }
1220 
1221 void x86_pmu_enable_all(int added);
1222 
1223 int perf_assign_events(struct event_constraint **constraints, int n,
1224 			int wmin, int wmax, int gpmax, int *assign);
1225 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
1226 
1227 void x86_pmu_stop(struct perf_event *event, int flags);
1228 
1229 static inline void x86_pmu_disable_event(struct perf_event *event)
1230 {
1231 	u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
1232 	struct hw_perf_event *hwc = &event->hw;
1233 
1234 	wrmsrq(hwc->config_base, hwc->config & ~disable_mask);
1235 
1236 	if (is_counter_pair(hwc))
1237 		wrmsrq(x86_pmu_config_addr(hwc->idx + 1), 0);
1238 }
1239 
1240 void x86_pmu_enable_event(struct perf_event *event);
1241 
1242 int x86_pmu_handle_irq(struct pt_regs *regs);
1243 
1244 void x86_pmu_show_pmu_cap(struct pmu *pmu);
1245 
1246 static inline int x86_pmu_num_counters(struct pmu *pmu)
1247 {
1248 	return hweight64(hybrid(pmu, cntr_mask64));
1249 }
1250 
1251 static inline int x86_pmu_max_num_counters(struct pmu *pmu)
1252 {
1253 	return fls64(hybrid(pmu, cntr_mask64));
1254 }
1255 
1256 static inline int x86_pmu_num_counters_fixed(struct pmu *pmu)
1257 {
1258 	return hweight64(hybrid(pmu, fixed_cntr_mask64));
1259 }
1260 
1261 static inline int x86_pmu_max_num_counters_fixed(struct pmu *pmu)
1262 {
1263 	return fls64(hybrid(pmu, fixed_cntr_mask64));
1264 }
1265 
1266 static inline u64 x86_pmu_get_event_config(struct perf_event *event)
1267 {
1268 	return event->attr.config & hybrid(event->pmu, config_mask);
1269 }
1270 
1271 extern struct event_constraint emptyconstraint;
1272 
1273 extern struct event_constraint unconstrained;
1274 
1275 static inline bool kernel_ip(unsigned long ip)
1276 {
1277 #ifdef CONFIG_X86_32
1278 	return ip > PAGE_OFFSET;
1279 #else
1280 	return (long)ip < 0;
1281 #endif
1282 }
1283 
1284 /*
1285  * Not all PMUs provide the right context information to place the reported IP
1286  * into full context. Specifically segment registers are typically not
1287  * supplied.
1288  *
1289  * Assuming the address is a linear address (it is for IBS), we fake the CS and
1290  * vm86 mode using the known zero-based code segment and 'fix up' the registers
1291  * to reflect this.
1292  *
1293  * Intel PEBS/LBR appear to typically provide the effective address, nothing
1294  * much we can do about that but pray and treat it like a linear address.
1295  */
1296 static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
1297 {
1298 	regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
1299 	if (regs->flags & X86_VM_MASK)
1300 		regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
1301 	regs->ip = ip;
1302 }
1303 
1304 /*
1305  * x86control flow change classification
1306  * x86control flow changes include branches, interrupts, traps, faults
1307  */
1308 enum {
1309 	X86_BR_NONE		= 0,      /* unknown */
1310 
1311 	X86_BR_USER		= 1 << 0, /* branch target is user */
1312 	X86_BR_KERNEL		= 1 << 1, /* branch target is kernel */
1313 
1314 	X86_BR_CALL		= 1 << 2, /* call */
1315 	X86_BR_RET		= 1 << 3, /* return */
1316 	X86_BR_SYSCALL		= 1 << 4, /* syscall */
1317 	X86_BR_SYSRET		= 1 << 5, /* syscall return */
1318 	X86_BR_INT		= 1 << 6, /* sw interrupt */
1319 	X86_BR_IRET		= 1 << 7, /* return from interrupt */
1320 	X86_BR_JCC		= 1 << 8, /* conditional */
1321 	X86_BR_JMP		= 1 << 9, /* jump */
1322 	X86_BR_IRQ		= 1 << 10,/* hw interrupt or trap or fault */
1323 	X86_BR_IND_CALL		= 1 << 11,/* indirect calls */
1324 	X86_BR_ABORT		= 1 << 12,/* transaction abort */
1325 	X86_BR_IN_TX		= 1 << 13,/* in transaction */
1326 	X86_BR_NO_TX		= 1 << 14,/* not in transaction */
1327 	X86_BR_ZERO_CALL	= 1 << 15,/* zero length call */
1328 	X86_BR_CALL_STACK	= 1 << 16,/* call stack */
1329 	X86_BR_IND_JMP		= 1 << 17,/* indirect jump */
1330 
1331 	X86_BR_TYPE_SAVE	= 1 << 18,/* indicate to save branch type */
1332 
1333 };
1334 
1335 #define X86_BR_PLM (X86_BR_USER | X86_BR_KERNEL)
1336 #define X86_BR_ANYTX (X86_BR_NO_TX | X86_BR_IN_TX)
1337 
1338 #define X86_BR_ANY       \
1339 	(X86_BR_CALL    |\
1340 	 X86_BR_RET     |\
1341 	 X86_BR_SYSCALL |\
1342 	 X86_BR_SYSRET  |\
1343 	 X86_BR_INT     |\
1344 	 X86_BR_IRET    |\
1345 	 X86_BR_JCC     |\
1346 	 X86_BR_JMP	 |\
1347 	 X86_BR_IRQ	 |\
1348 	 X86_BR_ABORT	 |\
1349 	 X86_BR_IND_CALL |\
1350 	 X86_BR_IND_JMP  |\
1351 	 X86_BR_ZERO_CALL)
1352 
1353 #define X86_BR_ALL (X86_BR_PLM | X86_BR_ANY)
1354 
1355 #define X86_BR_ANY_CALL		 \
1356 	(X86_BR_CALL		|\
1357 	 X86_BR_IND_CALL	|\
1358 	 X86_BR_ZERO_CALL	|\
1359 	 X86_BR_SYSCALL		|\
1360 	 X86_BR_IRQ		|\
1361 	 X86_BR_INT)
1362 
1363 int common_branch_type(int type);
1364 int branch_type(unsigned long from, unsigned long to, int abort);
1365 int branch_type_fused(unsigned long from, unsigned long to, int abort,
1366 		      int *offset);
1367 
1368 ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
1369 ssize_t intel_event_sysfs_show(char *page, u64 config);
1370 
1371 ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
1372 			  char *page);
1373 ssize_t events_ht_sysfs_show(struct device *dev, struct device_attribute *attr,
1374 			  char *page);
1375 ssize_t events_hybrid_sysfs_show(struct device *dev,
1376 				 struct device_attribute *attr,
1377 				 char *page);
1378 
1379 static inline bool fixed_counter_disabled(int i, struct pmu *pmu)
1380 {
1381 	u64 intel_ctrl = hybrid(pmu, intel_ctrl);
1382 
1383 	return !(intel_ctrl >> (i + INTEL_PMC_IDX_FIXED));
1384 }
1385 
1386 #ifdef CONFIG_CPU_SUP_AMD
1387 
1388 int amd_pmu_init(void);
1389 
1390 int amd_pmu_lbr_init(void);
1391 void amd_pmu_lbr_reset(void);
1392 void amd_pmu_lbr_read(void);
1393 void amd_pmu_lbr_add(struct perf_event *event);
1394 void amd_pmu_lbr_del(struct perf_event *event);
1395 void amd_pmu_lbr_sched_task(struct perf_event_pmu_context *pmu_ctx,
1396 			    struct task_struct *task, bool sched_in);
1397 void amd_pmu_lbr_enable_all(void);
1398 void amd_pmu_lbr_disable_all(void);
1399 int amd_pmu_lbr_hw_config(struct perf_event *event);
1400 
1401 static __always_inline void __amd_pmu_lbr_disable(void)
1402 {
1403 	u64 dbg_ctl, dbg_extn_cfg;
1404 
1405 	rdmsrq(MSR_AMD_DBG_EXTN_CFG, dbg_extn_cfg);
1406 	wrmsrq(MSR_AMD_DBG_EXTN_CFG, dbg_extn_cfg & ~DBG_EXTN_CFG_LBRV2EN);
1407 
1408 	if (cpu_feature_enabled(X86_FEATURE_AMD_LBR_PMC_FREEZE)) {
1409 		rdmsrq(MSR_IA32_DEBUGCTLMSR, dbg_ctl);
1410 		wrmsrq(MSR_IA32_DEBUGCTLMSR, dbg_ctl & ~DEBUGCTLMSR_FREEZE_LBRS_ON_PMI);
1411 	}
1412 }
1413 
1414 #ifdef CONFIG_PERF_EVENTS_AMD_BRS
1415 
1416 #define AMD_FAM19H_BRS_EVENT 0xc4 /* RETIRED_TAKEN_BRANCH_INSTRUCTIONS */
1417 
1418 int amd_brs_init(void);
1419 void amd_brs_disable(void);
1420 void amd_brs_enable(void);
1421 void amd_brs_enable_all(void);
1422 void amd_brs_disable_all(void);
1423 void amd_brs_drain(void);
1424 void amd_brs_lopwr_init(void);
1425 int amd_brs_hw_config(struct perf_event *event);
1426 void amd_brs_reset(void);
1427 
1428 static inline void amd_pmu_brs_add(struct perf_event *event)
1429 {
1430 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
1431 
1432 	perf_sched_cb_inc(event->pmu);
1433 	cpuc->lbr_users++;
1434 	/*
1435 	 * No need to reset BRS because it is reset
1436 	 * on brs_enable() and it is saturating
1437 	 */
1438 }
1439 
1440 static inline void amd_pmu_brs_del(struct perf_event *event)
1441 {
1442 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
1443 
1444 	cpuc->lbr_users--;
1445 	WARN_ON_ONCE(cpuc->lbr_users < 0);
1446 
1447 	perf_sched_cb_dec(event->pmu);
1448 }
1449 
1450 void amd_pmu_brs_sched_task(struct perf_event_pmu_context *pmu_ctx,
1451 			    struct task_struct *task, bool sched_in);
1452 #else
1453 static inline int amd_brs_init(void)
1454 {
1455 	return 0;
1456 }
1457 static inline void amd_brs_disable(void) {}
1458 static inline void amd_brs_enable(void) {}
1459 static inline void amd_brs_drain(void) {}
1460 static inline void amd_brs_lopwr_init(void) {}
1461 static inline void amd_brs_disable_all(void) {}
1462 static inline int amd_brs_hw_config(struct perf_event *event)
1463 {
1464 	return 0;
1465 }
1466 static inline void amd_brs_reset(void) {}
1467 
1468 static inline void amd_pmu_brs_add(struct perf_event *event)
1469 {
1470 }
1471 
1472 static inline void amd_pmu_brs_del(struct perf_event *event)
1473 {
1474 }
1475 
1476 static inline void amd_pmu_brs_sched_task(struct perf_event_pmu_context *pmu_ctx,
1477 					  struct task_struct *task, bool sched_in)
1478 {
1479 }
1480 
1481 static inline void amd_brs_enable_all(void)
1482 {
1483 }
1484 
1485 #endif
1486 
1487 #else /* CONFIG_CPU_SUP_AMD */
1488 
1489 static inline int amd_pmu_init(void)
1490 {
1491 	return 0;
1492 }
1493 
1494 static inline int amd_brs_init(void)
1495 {
1496 	return -EOPNOTSUPP;
1497 }
1498 
1499 static inline void amd_brs_drain(void)
1500 {
1501 }
1502 
1503 static inline void amd_brs_enable_all(void)
1504 {
1505 }
1506 
1507 static inline void amd_brs_disable_all(void)
1508 {
1509 }
1510 #endif /* CONFIG_CPU_SUP_AMD */
1511 
1512 static inline int is_pebs_pt(struct perf_event *event)
1513 {
1514 	return !!(event->hw.flags & PERF_X86_EVENT_PEBS_VIA_PT);
1515 }
1516 
1517 #ifdef CONFIG_CPU_SUP_INTEL
1518 
1519 static inline bool intel_pmu_has_bts_period(struct perf_event *event, u64 period)
1520 {
1521 	struct hw_perf_event *hwc = &event->hw;
1522 	unsigned int hw_event, bts_event;
1523 
1524 	if (event->attr.freq)
1525 		return false;
1526 
1527 	hw_event = hwc->config & INTEL_ARCH_EVENT_MASK;
1528 	bts_event = x86_pmu.event_map(PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
1529 
1530 	return hw_event == bts_event && period == 1;
1531 }
1532 
1533 static inline bool intel_pmu_has_bts(struct perf_event *event)
1534 {
1535 	struct hw_perf_event *hwc = &event->hw;
1536 
1537 	return intel_pmu_has_bts_period(event, hwc->sample_period);
1538 }
1539 
1540 static __always_inline void __intel_pmu_pebs_disable_all(void)
1541 {
1542 	wrmsrq(MSR_IA32_PEBS_ENABLE, 0);
1543 }
1544 
1545 static __always_inline void __intel_pmu_arch_lbr_disable(void)
1546 {
1547 	wrmsrq(MSR_ARCH_LBR_CTL, 0);
1548 }
1549 
1550 static __always_inline void __intel_pmu_lbr_disable(void)
1551 {
1552 	u64 debugctl;
1553 
1554 	rdmsrq(MSR_IA32_DEBUGCTLMSR, debugctl);
1555 	debugctl &= ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_FREEZE_LBRS_ON_PMI);
1556 	wrmsrq(MSR_IA32_DEBUGCTLMSR, debugctl);
1557 }
1558 
1559 int intel_pmu_save_and_restart(struct perf_event *event);
1560 
1561 struct event_constraint *
1562 x86_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
1563 			  struct perf_event *event);
1564 
1565 extern int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu);
1566 extern void intel_cpuc_finish(struct cpu_hw_events *cpuc);
1567 
1568 int intel_pmu_init(void);
1569 
1570 void init_debug_store_on_cpu(int cpu);
1571 
1572 void fini_debug_store_on_cpu(int cpu);
1573 
1574 void release_ds_buffers(void);
1575 
1576 void reserve_ds_buffers(void);
1577 
1578 void release_lbr_buffers(void);
1579 
1580 void reserve_lbr_buffers(void);
1581 
1582 extern struct event_constraint bts_constraint;
1583 extern struct event_constraint vlbr_constraint;
1584 
1585 void intel_pmu_enable_bts(u64 config);
1586 
1587 void intel_pmu_disable_bts(void);
1588 
1589 int intel_pmu_drain_bts_buffer(void);
1590 
1591 u64 grt_latency_data(struct perf_event *event, u64 status);
1592 
1593 u64 cmt_latency_data(struct perf_event *event, u64 status);
1594 
1595 u64 lnl_latency_data(struct perf_event *event, u64 status);
1596 
1597 u64 arl_h_latency_data(struct perf_event *event, u64 status);
1598 
1599 extern struct event_constraint intel_core2_pebs_event_constraints[];
1600 
1601 extern struct event_constraint intel_atom_pebs_event_constraints[];
1602 
1603 extern struct event_constraint intel_slm_pebs_event_constraints[];
1604 
1605 extern struct event_constraint intel_glm_pebs_event_constraints[];
1606 
1607 extern struct event_constraint intel_glp_pebs_event_constraints[];
1608 
1609 extern struct event_constraint intel_grt_pebs_event_constraints[];
1610 
1611 extern struct event_constraint intel_nehalem_pebs_event_constraints[];
1612 
1613 extern struct event_constraint intel_westmere_pebs_event_constraints[];
1614 
1615 extern struct event_constraint intel_snb_pebs_event_constraints[];
1616 
1617 extern struct event_constraint intel_ivb_pebs_event_constraints[];
1618 
1619 extern struct event_constraint intel_hsw_pebs_event_constraints[];
1620 
1621 extern struct event_constraint intel_bdw_pebs_event_constraints[];
1622 
1623 extern struct event_constraint intel_skl_pebs_event_constraints[];
1624 
1625 extern struct event_constraint intel_icl_pebs_event_constraints[];
1626 
1627 extern struct event_constraint intel_glc_pebs_event_constraints[];
1628 
1629 extern struct event_constraint intel_lnc_pebs_event_constraints[];
1630 
1631 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
1632 
1633 void intel_pmu_pebs_add(struct perf_event *event);
1634 
1635 void intel_pmu_pebs_del(struct perf_event *event);
1636 
1637 void intel_pmu_pebs_enable(struct perf_event *event);
1638 
1639 void intel_pmu_pebs_disable(struct perf_event *event);
1640 
1641 void intel_pmu_pebs_enable_all(void);
1642 
1643 void intel_pmu_pebs_disable_all(void);
1644 
1645 void intel_pmu_pebs_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in);
1646 
1647 void intel_pmu_drain_pebs_buffer(void);
1648 
1649 void intel_pmu_store_pebs_lbrs(struct lbr_entry *lbr);
1650 
1651 void intel_ds_init(void);
1652 
1653 void intel_pmu_lbr_save_brstack(struct perf_sample_data *data,
1654 				struct cpu_hw_events *cpuc,
1655 				struct perf_event *event);
1656 
1657 void intel_pmu_lbr_sched_task(struct perf_event_pmu_context *pmu_ctx,
1658 			      struct task_struct *task, bool sched_in);
1659 
1660 u64 lbr_from_signext_quirk_wr(u64 val);
1661 
1662 void intel_pmu_lbr_reset(void);
1663 
1664 void intel_pmu_lbr_reset_32(void);
1665 
1666 void intel_pmu_lbr_reset_64(void);
1667 
1668 void intel_pmu_lbr_add(struct perf_event *event);
1669 
1670 void intel_pmu_lbr_del(struct perf_event *event);
1671 
1672 void intel_pmu_lbr_enable_all(bool pmi);
1673 
1674 void intel_pmu_lbr_disable_all(void);
1675 
1676 void intel_pmu_lbr_read(void);
1677 
1678 void intel_pmu_lbr_read_32(struct cpu_hw_events *cpuc);
1679 
1680 void intel_pmu_lbr_read_64(struct cpu_hw_events *cpuc);
1681 
1682 void intel_pmu_lbr_save(void *ctx);
1683 
1684 void intel_pmu_lbr_restore(void *ctx);
1685 
1686 void intel_pmu_lbr_init_core(void);
1687 
1688 void intel_pmu_lbr_init_nhm(void);
1689 
1690 void intel_pmu_lbr_init_atom(void);
1691 
1692 void intel_pmu_lbr_init_slm(void);
1693 
1694 void intel_pmu_lbr_init_snb(void);
1695 
1696 void intel_pmu_lbr_init_hsw(void);
1697 
1698 void intel_pmu_lbr_init_skl(void);
1699 
1700 void intel_pmu_lbr_init_knl(void);
1701 
1702 void intel_pmu_lbr_init(void);
1703 
1704 void intel_pmu_arch_lbr_init(void);
1705 
1706 void intel_pmu_pebs_data_source_nhm(void);
1707 
1708 void intel_pmu_pebs_data_source_skl(bool pmem);
1709 
1710 void intel_pmu_pebs_data_source_adl(void);
1711 
1712 void intel_pmu_pebs_data_source_grt(void);
1713 
1714 void intel_pmu_pebs_data_source_mtl(void);
1715 
1716 void intel_pmu_pebs_data_source_arl_h(void);
1717 
1718 void intel_pmu_pebs_data_source_cmt(void);
1719 
1720 void intel_pmu_pebs_data_source_lnl(void);
1721 
1722 int intel_pmu_setup_lbr_filter(struct perf_event *event);
1723 
1724 void intel_pt_interrupt(void);
1725 
1726 int intel_bts_interrupt(void);
1727 
1728 void intel_bts_enable_local(void);
1729 
1730 void intel_bts_disable_local(void);
1731 
1732 int p4_pmu_init(void);
1733 
1734 int p6_pmu_init(void);
1735 
1736 int knc_pmu_init(void);
1737 
1738 static inline int is_ht_workaround_enabled(void)
1739 {
1740 	return !!(x86_pmu.flags & PMU_FL_EXCL_ENABLED);
1741 }
1742 
1743 static inline u64 intel_pmu_pebs_mask(u64 cntr_mask)
1744 {
1745 	return MAX_PEBS_EVENTS_MASK & cntr_mask;
1746 }
1747 
1748 static inline int intel_pmu_max_num_pebs(struct pmu *pmu)
1749 {
1750 	static_assert(MAX_PEBS_EVENTS == 32);
1751 	return fls((u32)hybrid(pmu, pebs_events_mask));
1752 }
1753 
1754 #else /* CONFIG_CPU_SUP_INTEL */
1755 
1756 static inline void reserve_ds_buffers(void)
1757 {
1758 }
1759 
1760 static inline void release_ds_buffers(void)
1761 {
1762 }
1763 
1764 static inline void release_lbr_buffers(void)
1765 {
1766 }
1767 
1768 static inline void reserve_lbr_buffers(void)
1769 {
1770 }
1771 
1772 static inline int intel_pmu_init(void)
1773 {
1774 	return 0;
1775 }
1776 
1777 static inline int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu)
1778 {
1779 	return 0;
1780 }
1781 
1782 static inline void intel_cpuc_finish(struct cpu_hw_events *cpuc)
1783 {
1784 }
1785 
1786 static inline int is_ht_workaround_enabled(void)
1787 {
1788 	return 0;
1789 }
1790 #endif /* CONFIG_CPU_SUP_INTEL */
1791 
1792 #if ((defined CONFIG_CPU_SUP_CENTAUR) || (defined CONFIG_CPU_SUP_ZHAOXIN))
1793 int zhaoxin_pmu_init(void);
1794 #else
1795 static inline int zhaoxin_pmu_init(void)
1796 {
1797 	return 0;
1798 }
1799 #endif /*CONFIG_CPU_SUP_CENTAUR or CONFIG_CPU_SUP_ZHAOXIN*/
1800