• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
21 /* To enable MSR tracing please use the generic trace points. */
22 
23 /*
24  *          |   NHM/WSM    |      SNB     |
25  * register -------------------------------
26  *          |  HT  | no HT |  HT  | no HT |
27  *-----------------------------------------
28  * offcore  | core | core  | cpu  | core  |
29  * lbr_sel  | core | core  | cpu  | core  |
30  * ld_lat   | cpu  | core  | cpu  | core  |
31  *-----------------------------------------
32  *
33  * Given that there is a small number of shared regs,
34  * we can pre-allocate their slot in the per-cpu
35  * per-core reg tables.
36  */
37 enum extra_reg_type {
38 	EXTRA_REG_NONE		= -1, /* not used */
39 
40 	EXTRA_REG_RSP_0		= 0,  /* offcore_response_0 */
41 	EXTRA_REG_RSP_1		= 1,  /* offcore_response_1 */
42 	EXTRA_REG_LBR		= 2,  /* lbr_select */
43 	EXTRA_REG_LDLAT		= 3,  /* ld_lat_threshold */
44 	EXTRA_REG_FE		= 4,  /* fe_* */
45 	EXTRA_REG_SNOOP_0	= 5,  /* snoop response 0 */
46 	EXTRA_REG_SNOOP_1	= 6,  /* snoop response 1 */
47 
48 	EXTRA_REG_MAX		      /* number of entries needed */
49 };
50 
51 struct event_constraint {
52 	union {
53 		unsigned long	idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
54 		u64		idxmsk64;
55 	};
56 	u64		code;
57 	u64		cmask;
58 	int		weight;
59 	int		overlap;
60 	int		flags;
61 	unsigned int	size;
62 };
63 
constraint_match(struct event_constraint * c,u64 ecode)64 static inline bool constraint_match(struct event_constraint *c, u64 ecode)
65 {
66 	return ((ecode & c->cmask) - c->code) <= (u64)c->size;
67 }
68 
69 #define PERF_ARCH(name, val)	\
70 	PERF_X86_EVENT_##name = val,
71 
72 /*
73  * struct hw_perf_event.flags flags
74  */
75 enum {
76 #include "perf_event_flags.h"
77 };
78 
79 #undef PERF_ARCH
80 
81 #define PERF_ARCH(name, val)						\
82 	static_assert((PERF_X86_EVENT_##name & PERF_EVENT_FLAG_ARCH) ==	\
83 		      PERF_X86_EVENT_##name);
84 
85 #include "perf_event_flags.h"
86 
87 #undef PERF_ARCH
88 
is_topdown_count(struct perf_event * event)89 static inline bool is_topdown_count(struct perf_event *event)
90 {
91 	return event->hw.flags & PERF_X86_EVENT_TOPDOWN;
92 }
93 
is_metric_event(struct perf_event * event)94 static inline bool is_metric_event(struct perf_event *event)
95 {
96 	u64 config = event->attr.config;
97 
98 	return ((config & ARCH_PERFMON_EVENTSEL_EVENT) == 0) &&
99 		((config & INTEL_ARCH_EVENT_MASK) >= INTEL_TD_METRIC_RETIRING)  &&
100 		((config & INTEL_ARCH_EVENT_MASK) <= INTEL_TD_METRIC_MAX);
101 }
102 
is_slots_event(struct perf_event * event)103 static inline bool is_slots_event(struct perf_event *event)
104 {
105 	return (event->attr.config & INTEL_ARCH_EVENT_MASK) == INTEL_TD_SLOTS;
106 }
107 
is_topdown_event(struct perf_event * event)108 static inline bool is_topdown_event(struct perf_event *event)
109 {
110 	return is_metric_event(event) || is_slots_event(event);
111 }
112 
113 int is_x86_event(struct perf_event *event);
114 
check_leader_group(struct perf_event * leader,int flags)115 static inline bool check_leader_group(struct perf_event *leader, int flags)
116 {
117 	return is_x86_event(leader) ? !!(leader->hw.flags & flags) : false;
118 }
119 
is_branch_counters_group(struct perf_event * event)120 static inline bool is_branch_counters_group(struct perf_event *event)
121 {
122 	return check_leader_group(event->group_leader, PERF_X86_EVENT_BRANCH_COUNTERS);
123 }
124 
125 struct amd_nb {
126 	int nb_id;  /* NorthBridge id */
127 	int refcnt; /* reference count */
128 	struct perf_event *owners[X86_PMC_IDX_MAX];
129 	struct event_constraint event_constraints[X86_PMC_IDX_MAX];
130 };
131 
132 #define PEBS_COUNTER_MASK	((1ULL << MAX_PEBS_EVENTS) - 1)
133 #define PEBS_PMI_AFTER_EACH_RECORD BIT_ULL(60)
134 #define PEBS_OUTPUT_OFFSET	61
135 #define PEBS_OUTPUT_MASK	(3ull << PEBS_OUTPUT_OFFSET)
136 #define PEBS_OUTPUT_PT		(1ull << PEBS_OUTPUT_OFFSET)
137 #define PEBS_VIA_PT_MASK	(PEBS_OUTPUT_PT | PEBS_PMI_AFTER_EACH_RECORD)
138 
139 /*
140  * Flags PEBS can handle without an PMI.
141  *
142  * TID can only be handled by flushing at context switch.
143  * REGS_USER can be handled for events limited to ring 3.
144  *
145  */
146 #define LARGE_PEBS_FLAGS \
147 	(PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_ADDR | \
148 	PERF_SAMPLE_ID | PERF_SAMPLE_CPU | PERF_SAMPLE_STREAM_ID | \
149 	PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
150 	PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR | \
151 	PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER | \
152 	PERF_SAMPLE_PERIOD | PERF_SAMPLE_CODE_PAGE_SIZE | \
153 	PERF_SAMPLE_WEIGHT_TYPE)
154 
155 #define PEBS_GP_REGS			\
156 	((1ULL << PERF_REG_X86_AX)    | \
157 	 (1ULL << PERF_REG_X86_BX)    | \
158 	 (1ULL << PERF_REG_X86_CX)    | \
159 	 (1ULL << PERF_REG_X86_DX)    | \
160 	 (1ULL << PERF_REG_X86_DI)    | \
161 	 (1ULL << PERF_REG_X86_SI)    | \
162 	 (1ULL << PERF_REG_X86_SP)    | \
163 	 (1ULL << PERF_REG_X86_BP)    | \
164 	 (1ULL << PERF_REG_X86_IP)    | \
165 	 (1ULL << PERF_REG_X86_FLAGS) | \
166 	 (1ULL << PERF_REG_X86_R8)    | \
167 	 (1ULL << PERF_REG_X86_R9)    | \
168 	 (1ULL << PERF_REG_X86_R10)   | \
169 	 (1ULL << PERF_REG_X86_R11)   | \
170 	 (1ULL << PERF_REG_X86_R12)   | \
171 	 (1ULL << PERF_REG_X86_R13)   | \
172 	 (1ULL << PERF_REG_X86_R14)   | \
173 	 (1ULL << PERF_REG_X86_R15))
174 
175 /*
176  * Per register state.
177  */
178 struct er_account {
179 	raw_spinlock_t      lock;	/* per-core: protect structure */
180 	u64                 config;	/* extra MSR config */
181 	u64                 reg;	/* extra MSR number */
182 	atomic_t            ref;	/* reference count */
183 };
184 
185 /*
186  * Per core/cpu state
187  *
188  * Used to coordinate shared registers between HT threads or
189  * among events on a single PMU.
190  */
191 struct intel_shared_regs {
192 	struct er_account       regs[EXTRA_REG_MAX];
193 	int                     refcnt;		/* per-core: #HT threads */
194 	unsigned                core_id;	/* per-core: core id */
195 };
196 
197 enum intel_excl_state_type {
198 	INTEL_EXCL_UNUSED    = 0, /* counter is unused */
199 	INTEL_EXCL_SHARED    = 1, /* counter can be used by both threads */
200 	INTEL_EXCL_EXCLUSIVE = 2, /* counter can be used by one thread only */
201 };
202 
203 struct intel_excl_states {
204 	enum intel_excl_state_type state[X86_PMC_IDX_MAX];
205 	bool sched_started; /* true if scheduling has started */
206 };
207 
208 struct intel_excl_cntrs {
209 	raw_spinlock_t	lock;
210 
211 	struct intel_excl_states states[2];
212 
213 	union {
214 		u16	has_exclusive[2];
215 		u32	exclusive_present;
216 	};
217 
218 	int		refcnt;		/* per-core: #HT threads */
219 	unsigned	core_id;	/* per-core: core id */
220 };
221 
222 struct x86_perf_task_context;
223 #define MAX_LBR_ENTRIES		32
224 
225 enum {
226 	LBR_FORMAT_32		= 0x00,
227 	LBR_FORMAT_LIP		= 0x01,
228 	LBR_FORMAT_EIP		= 0x02,
229 	LBR_FORMAT_EIP_FLAGS	= 0x03,
230 	LBR_FORMAT_EIP_FLAGS2	= 0x04,
231 	LBR_FORMAT_INFO		= 0x05,
232 	LBR_FORMAT_TIME		= 0x06,
233 	LBR_FORMAT_INFO2	= 0x07,
234 	LBR_FORMAT_MAX_KNOWN    = LBR_FORMAT_INFO2,
235 };
236 
237 enum {
238 	X86_PERF_KFREE_SHARED = 0,
239 	X86_PERF_KFREE_EXCL   = 1,
240 	X86_PERF_KFREE_MAX
241 };
242 
243 struct cpu_hw_events {
244 	/*
245 	 * Generic x86 PMC bits
246 	 */
247 	struct perf_event	*events[X86_PMC_IDX_MAX]; /* in counter order */
248 	unsigned long		active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
249 	unsigned long		dirty[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
250 	int			enabled;
251 
252 	int			n_events; /* the # of events in the below arrays */
253 	int			n_added;  /* the # last events in the below arrays;
254 					     they've never been enabled yet */
255 	int			n_txn;    /* the # last events in the below arrays;
256 					     added in the current transaction */
257 	int			n_txn_pair;
258 	int			n_txn_metric;
259 	int			assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
260 	u64			tags[X86_PMC_IDX_MAX];
261 
262 	struct perf_event	*event_list[X86_PMC_IDX_MAX]; /* in enabled order */
263 	struct event_constraint	*event_constraint[X86_PMC_IDX_MAX];
264 
265 	int			n_excl; /* the number of exclusive events */
266 
267 	unsigned int		txn_flags;
268 	int			is_fake;
269 
270 	/*
271 	 * Intel DebugStore bits
272 	 */
273 	struct debug_store	*ds;
274 	void			*ds_pebs_vaddr;
275 	void			*ds_bts_vaddr;
276 	u64			pebs_enabled;
277 	int			n_pebs;
278 	int			n_large_pebs;
279 	int			n_pebs_via_pt;
280 	int			pebs_output;
281 
282 	/* Current super set of events hardware configuration */
283 	u64			pebs_data_cfg;
284 	u64			active_pebs_data_cfg;
285 	int			pebs_record_size;
286 
287 	/* Intel Fixed counter configuration */
288 	u64			fixed_ctrl_val;
289 	u64			active_fixed_ctrl_val;
290 
291 	/*
292 	 * Intel LBR bits
293 	 */
294 	int				lbr_users;
295 	int				lbr_pebs_users;
296 	struct perf_branch_stack	lbr_stack;
297 	struct perf_branch_entry	lbr_entries[MAX_LBR_ENTRIES];
298 	u64				lbr_counters[MAX_LBR_ENTRIES]; /* branch stack extra */
299 	union {
300 		struct er_account		*lbr_sel;
301 		struct er_account		*lbr_ctl;
302 	};
303 	u64				br_sel;
304 	void				*last_task_ctx;
305 	int				last_log_id;
306 	int				lbr_select;
307 	void				*lbr_xsave;
308 
309 	/*
310 	 * Intel host/guest exclude bits
311 	 */
312 	u64				intel_ctrl_guest_mask;
313 	u64				intel_ctrl_host_mask;
314 	struct perf_guest_switch_msr	guest_switch_msrs[X86_PMC_IDX_MAX];
315 
316 	/*
317 	 * Intel checkpoint mask
318 	 */
319 	u64				intel_cp_status;
320 
321 	/*
322 	 * manage shared (per-core, per-cpu) registers
323 	 * used on Intel NHM/WSM/SNB
324 	 */
325 	struct intel_shared_regs	*shared_regs;
326 	/*
327 	 * manage exclusive counter access between hyperthread
328 	 */
329 	struct event_constraint *constraint_list; /* in enable order */
330 	struct intel_excl_cntrs		*excl_cntrs;
331 	int excl_thread_id; /* 0 or 1 */
332 
333 	/*
334 	 * SKL TSX_FORCE_ABORT shadow
335 	 */
336 	u64				tfa_shadow;
337 
338 	/*
339 	 * Perf Metrics
340 	 */
341 	/* number of accepted metrics events */
342 	int				n_metric;
343 
344 	/*
345 	 * AMD specific bits
346 	 */
347 	struct amd_nb			*amd_nb;
348 	int				brs_active; /* BRS is enabled */
349 
350 	/* Inverted mask of bits to clear in the perf_ctr ctrl registers */
351 	u64				perf_ctr_virt_mask;
352 	int				n_pair; /* Large increment events */
353 
354 	void				*kfree_on_online[X86_PERF_KFREE_MAX];
355 
356 	struct pmu			*pmu;
357 };
358 
359 #define __EVENT_CONSTRAINT_RANGE(c, e, n, m, w, o, f) {	\
360 	{ .idxmsk64 = (n) },		\
361 	.code = (c),			\
362 	.size = (e) - (c),		\
363 	.cmask = (m),			\
364 	.weight = (w),			\
365 	.overlap = (o),			\
366 	.flags = f,			\
367 }
368 
369 #define __EVENT_CONSTRAINT(c, n, m, w, o, f) \
370 	__EVENT_CONSTRAINT_RANGE(c, c, n, m, w, o, f)
371 
372 #define EVENT_CONSTRAINT(c, n, m)	\
373 	__EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0)
374 
375 /*
376  * The constraint_match() function only works for 'simple' event codes
377  * and not for extended (AMD64_EVENTSEL_EVENT) events codes.
378  */
379 #define EVENT_CONSTRAINT_RANGE(c, e, n, m) \
380 	__EVENT_CONSTRAINT_RANGE(c, e, n, m, HWEIGHT(n), 0, 0)
381 
382 #define INTEL_EXCLEVT_CONSTRAINT(c, n)	\
383 	__EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT, HWEIGHT(n),\
384 			   0, PERF_X86_EVENT_EXCL)
385 
386 /*
387  * The overlap flag marks event constraints with overlapping counter
388  * masks. This is the case if the counter mask of such an event is not
389  * a subset of any other counter mask of a constraint with an equal or
390  * higher weight, e.g.:
391  *
392  *  c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
393  *  c_another1 = EVENT_CONSTRAINT(0, 0x07, 0);
394  *  c_another2 = EVENT_CONSTRAINT(0, 0x38, 0);
395  *
396  * The event scheduler may not select the correct counter in the first
397  * cycle because it needs to know which subsequent events will be
398  * scheduled. It may fail to schedule the events then. So we set the
399  * overlap flag for such constraints to give the scheduler a hint which
400  * events to select for counter rescheduling.
401  *
402  * Care must be taken as the rescheduling algorithm is O(n!) which
403  * will increase scheduling cycles for an over-committed system
404  * dramatically.  The number of such EVENT_CONSTRAINT_OVERLAP() macros
405  * and its counter masks must be kept at a minimum.
406  */
407 #define EVENT_CONSTRAINT_OVERLAP(c, n, m)	\
408 	__EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0)
409 
410 /*
411  * Constraint on the Event code.
412  */
413 #define INTEL_EVENT_CONSTRAINT(c, n)	\
414 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
415 
416 /*
417  * Constraint on a range of Event codes
418  */
419 #define INTEL_EVENT_CONSTRAINT_RANGE(c, e, n)			\
420 	EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT)
421 
422 /*
423  * Constraint on the Event code + UMask + fixed-mask
424  *
425  * filter mask to validate fixed counter events.
426  * the following filters disqualify for fixed counters:
427  *  - inv
428  *  - edge
429  *  - cnt-mask
430  *  - in_tx
431  *  - in_tx_checkpointed
432  *  The other filters are supported by fixed counters.
433  *  The any-thread option is supported starting with v3.
434  */
435 #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED)
436 #define FIXED_EVENT_CONSTRAINT(c, n)	\
437 	EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
438 
439 /*
440  * The special metric counters do not actually exist. They are calculated from
441  * the combination of the FxCtr3 + MSR_PERF_METRICS.
442  *
443  * The special metric counters are mapped to a dummy offset for the scheduler.
444  * The sharing between multiple users of the same metric without multiplexing
445  * is not allowed, even though the hardware supports that in principle.
446  */
447 
448 #define METRIC_EVENT_CONSTRAINT(c, n)					\
449 	EVENT_CONSTRAINT(c, (1ULL << (INTEL_PMC_IDX_METRIC_BASE + n)),	\
450 			 INTEL_ARCH_EVENT_MASK)
451 
452 /*
453  * Constraint on the Event code + UMask
454  */
455 #define INTEL_UEVENT_CONSTRAINT(c, n)	\
456 	EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
457 
458 /* Constraint on specific umask bit only + event */
459 #define INTEL_UBIT_EVENT_CONSTRAINT(c, n)	\
460 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|(c))
461 
462 /* Like UEVENT_CONSTRAINT, but match flags too */
463 #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n)	\
464 	EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
465 
466 #define INTEL_EXCLUEVT_CONSTRAINT(c, n)	\
467 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK, \
468 			   HWEIGHT(n), 0, PERF_X86_EVENT_EXCL)
469 
470 #define INTEL_PLD_CONSTRAINT(c, n)	\
471 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
472 			   HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
473 
474 #define INTEL_PSD_CONSTRAINT(c, n)	\
475 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
476 			   HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_STLAT)
477 
478 #define INTEL_PST_CONSTRAINT(c, n)	\
479 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
480 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
481 
482 #define INTEL_HYBRID_LAT_CONSTRAINT(c, n)	\
483 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
484 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LAT_HYBRID)
485 
486 #define INTEL_HYBRID_LDLAT_CONSTRAINT(c, n)	\
487 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
488 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LAT_HYBRID|PERF_X86_EVENT_PEBS_LD_HSW)
489 
490 #define INTEL_HYBRID_STLAT_CONSTRAINT(c, n)	\
491 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
492 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LAT_HYBRID|PERF_X86_EVENT_PEBS_ST_HSW)
493 
494 /* Event constraint, but match on all event flags too. */
495 #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
496 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
497 
498 #define INTEL_FLAGS_EVENT_CONSTRAINT_RANGE(c, e, n)			\
499 	EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
500 
501 /* Check only flags, but allow all event/umask */
502 #define INTEL_ALL_EVENT_CONSTRAINT(code, n)	\
503 	EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
504 
505 /* Check flags and event code, and set the HSW store flag */
506 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
507 	__EVENT_CONSTRAINT(code, n, 			\
508 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
509 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
510 
511 /* Check flags and event code, and set the HSW load flag */
512 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
513 	__EVENT_CONSTRAINT(code, n,			\
514 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
515 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
516 
517 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD_RANGE(code, end, n) \
518 	__EVENT_CONSTRAINT_RANGE(code, end, n,				\
519 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
520 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
521 
522 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_XLD(code, n) \
523 	__EVENT_CONSTRAINT(code, n,			\
524 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
525 			  HWEIGHT(n), 0, \
526 			  PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
527 
528 /* Check flags and event code/umask, and set the HSW store flag */
529 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
530 	__EVENT_CONSTRAINT(code, n, 			\
531 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
532 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
533 
534 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XST(code, n) \
535 	__EVENT_CONSTRAINT(code, n,			\
536 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
537 			  HWEIGHT(n), 0, \
538 			  PERF_X86_EVENT_PEBS_ST_HSW|PERF_X86_EVENT_EXCL)
539 
540 /* Check flags and event code/umask, and set the HSW load flag */
541 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
542 	__EVENT_CONSTRAINT(code, n, 			\
543 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
544 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
545 
546 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XLD(code, n) \
547 	__EVENT_CONSTRAINT(code, n,			\
548 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
549 			  HWEIGHT(n), 0, \
550 			  PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
551 
552 /* Check flags and event code/umask, and set the HSW N/A flag */
553 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
554 	__EVENT_CONSTRAINT(code, n, 			\
555 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
556 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
557 
558 
559 /*
560  * We define the end marker as having a weight of -1
561  * to enable blacklisting of events using a counter bitmask
562  * of zero and thus a weight of zero.
563  * The end marker has a weight that cannot possibly be
564  * obtained from counting the bits in the bitmask.
565  */
566 #define EVENT_CONSTRAINT_END { .weight = -1 }
567 
568 /*
569  * Check for end marker with weight == -1
570  */
571 #define for_each_event_constraint(e, c)	\
572 	for ((e) = (c); (e)->weight != -1; (e)++)
573 
574 /*
575  * Extra registers for specific events.
576  *
577  * Some events need large masks and require external MSRs.
578  * Those extra MSRs end up being shared for all events on
579  * a PMU and sometimes between PMU of sibling HT threads.
580  * In either case, the kernel needs to handle conflicting
581  * accesses to those extra, shared, regs. The data structure
582  * to manage those registers is stored in cpu_hw_event.
583  */
584 struct extra_reg {
585 	unsigned int		event;
586 	unsigned int		msr;
587 	u64			config_mask;
588 	u64			valid_mask;
589 	int			idx;  /* per_xxx->regs[] reg index */
590 	bool			extra_msr_access;
591 };
592 
593 #define EVENT_EXTRA_REG(e, ms, m, vm, i) {	\
594 	.event = (e),			\
595 	.msr = (ms),			\
596 	.config_mask = (m),		\
597 	.valid_mask = (vm),		\
598 	.idx = EXTRA_REG_##i,		\
599 	.extra_msr_access = true,	\
600 	}
601 
602 #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx)	\
603 	EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
604 
605 #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
606 	EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
607 			ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
608 
609 #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
610 	INTEL_UEVENT_EXTRA_REG(c, \
611 			       MSR_PEBS_LD_LAT_THRESHOLD, \
612 			       0xffff, \
613 			       LDLAT)
614 
615 #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
616 
617 union perf_capabilities {
618 	struct {
619 		u64	lbr_format:6;
620 		u64	pebs_trap:1;
621 		u64	pebs_arch_reg:1;
622 		u64	pebs_format:4;
623 		u64	smm_freeze:1;
624 		/*
625 		 * PMU supports separate counter range for writing
626 		 * values > 32bit.
627 		 */
628 		u64	full_width_write:1;
629 		u64     pebs_baseline:1;
630 		u64	perf_metrics:1;
631 		u64	pebs_output_pt_available:1;
632 		u64	pebs_timing_info:1;
633 		u64	anythread_deprecated:1;
634 	};
635 	u64	capabilities;
636 };
637 
638 struct x86_pmu_quirk {
639 	struct x86_pmu_quirk *next;
640 	void (*func)(void);
641 };
642 
643 union x86_pmu_config {
644 	struct {
645 		u64 event:8,
646 		    umask:8,
647 		    usr:1,
648 		    os:1,
649 		    edge:1,
650 		    pc:1,
651 		    interrupt:1,
652 		    __reserved1:1,
653 		    en:1,
654 		    inv:1,
655 		    cmask:8,
656 		    event2:4,
657 		    __reserved2:4,
658 		    go:1,
659 		    ho:1;
660 	} bits;
661 	u64 value;
662 };
663 
664 #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
665 
666 enum {
667 	x86_lbr_exclusive_lbr,
668 	x86_lbr_exclusive_bts,
669 	x86_lbr_exclusive_pt,
670 	x86_lbr_exclusive_max,
671 };
672 
673 #define PERF_PEBS_DATA_SOURCE_MAX	0x100
674 #define PERF_PEBS_DATA_SOURCE_MASK	(PERF_PEBS_DATA_SOURCE_MAX - 1)
675 #define PERF_PEBS_DATA_SOURCE_GRT_MAX	0x10
676 #define PERF_PEBS_DATA_SOURCE_GRT_MASK	(PERF_PEBS_DATA_SOURCE_GRT_MAX - 1)
677 
678 enum hybrid_cpu_type {
679 	HYBRID_INTEL_NONE,
680 	HYBRID_INTEL_ATOM	= 0x20,
681 	HYBRID_INTEL_CORE	= 0x40,
682 };
683 
684 enum hybrid_pmu_type {
685 	not_hybrid,
686 	hybrid_small		= BIT(0),
687 	hybrid_big		= BIT(1),
688 
689 	hybrid_big_small	= hybrid_big | hybrid_small, /* only used for matching */
690 };
691 
692 #define X86_HYBRID_PMU_ATOM_IDX		0
693 #define X86_HYBRID_PMU_CORE_IDX		1
694 
695 #define X86_HYBRID_NUM_PMUS		2
696 
697 struct x86_hybrid_pmu {
698 	struct pmu			pmu;
699 	const char			*name;
700 	enum hybrid_pmu_type		pmu_type;
701 	cpumask_t			supported_cpus;
702 	union perf_capabilities		intel_cap;
703 	u64				intel_ctrl;
704 	u64				pebs_events_mask;
705 	u64				config_mask;
706 	union {
707 			u64		cntr_mask64;
708 			unsigned long	cntr_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
709 	};
710 	union {
711 			u64		fixed_cntr_mask64;
712 			unsigned long	fixed_cntr_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
713 	};
714 	struct event_constraint		unconstrained;
715 
716 	u64				hw_cache_event_ids
717 					[PERF_COUNT_HW_CACHE_MAX]
718 					[PERF_COUNT_HW_CACHE_OP_MAX]
719 					[PERF_COUNT_HW_CACHE_RESULT_MAX];
720 	u64				hw_cache_extra_regs
721 					[PERF_COUNT_HW_CACHE_MAX]
722 					[PERF_COUNT_HW_CACHE_OP_MAX]
723 					[PERF_COUNT_HW_CACHE_RESULT_MAX];
724 	struct event_constraint		*event_constraints;
725 	struct event_constraint		*pebs_constraints;
726 	struct extra_reg		*extra_regs;
727 
728 	unsigned int			late_ack	:1,
729 					mid_ack		:1,
730 					enabled_ack	:1;
731 
732 	u64				pebs_data_source[PERF_PEBS_DATA_SOURCE_MAX];
733 };
734 
hybrid_pmu(struct pmu * pmu)735 static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu)
736 {
737 	return container_of(pmu, struct x86_hybrid_pmu, pmu);
738 }
739 
740 extern struct static_key_false perf_is_hybrid;
741 #define is_hybrid()		static_branch_unlikely(&perf_is_hybrid)
742 
743 #define hybrid(_pmu, _field)				\
744 (*({							\
745 	typeof(&x86_pmu._field) __Fp = &x86_pmu._field;	\
746 							\
747 	if (is_hybrid() && (_pmu))			\
748 		__Fp = &hybrid_pmu(_pmu)->_field;	\
749 							\
750 	__Fp;						\
751 }))
752 
753 #define hybrid_var(_pmu, _var)				\
754 (*({							\
755 	typeof(&_var) __Fp = &_var;			\
756 							\
757 	if (is_hybrid() && (_pmu))			\
758 		__Fp = &hybrid_pmu(_pmu)->_var;		\
759 							\
760 	__Fp;						\
761 }))
762 
763 #define hybrid_bit(_pmu, _field)			\
764 ({							\
765 	bool __Fp = x86_pmu._field;			\
766 							\
767 	if (is_hybrid() && (_pmu))			\
768 		__Fp = hybrid_pmu(_pmu)->_field;	\
769 							\
770 	__Fp;						\
771 })
772 
773 /*
774  * struct x86_pmu - generic x86 pmu
775  */
776 struct x86_pmu {
777 	/*
778 	 * Generic x86 PMC bits
779 	 */
780 	const char	*name;
781 	int		version;
782 	int		(*handle_irq)(struct pt_regs *);
783 	void		(*disable_all)(void);
784 	void		(*enable_all)(int added);
785 	void		(*enable)(struct perf_event *);
786 	void		(*disable)(struct perf_event *);
787 	void		(*assign)(struct perf_event *event, int idx);
788 	void		(*add)(struct perf_event *);
789 	void		(*del)(struct perf_event *);
790 	void		(*read)(struct perf_event *event);
791 	int		(*set_period)(struct perf_event *event);
792 	u64		(*update)(struct perf_event *event);
793 	int		(*hw_config)(struct perf_event *event);
794 	int		(*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
795 	unsigned	eventsel;
796 	unsigned	perfctr;
797 	unsigned	fixedctr;
798 	int		(*addr_offset)(int index, bool eventsel);
799 	int		(*rdpmc_index)(int index);
800 	u64		(*event_map)(int);
801 	int		max_events;
802 	u64		config_mask;
803 	union {
804 			u64		cntr_mask64;
805 			unsigned long	cntr_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
806 	};
807 	union {
808 			u64		fixed_cntr_mask64;
809 			unsigned long	fixed_cntr_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
810 	};
811 	int		cntval_bits;
812 	u64		cntval_mask;
813 	union {
814 			unsigned long events_maskl;
815 			unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
816 	};
817 	int		events_mask_len;
818 	int		apic;
819 	u64		max_period;
820 	struct event_constraint *
821 			(*get_event_constraints)(struct cpu_hw_events *cpuc,
822 						 int idx,
823 						 struct perf_event *event);
824 
825 	void		(*put_event_constraints)(struct cpu_hw_events *cpuc,
826 						 struct perf_event *event);
827 
828 	void		(*start_scheduling)(struct cpu_hw_events *cpuc);
829 
830 	void		(*commit_scheduling)(struct cpu_hw_events *cpuc, int idx, int cntr);
831 
832 	void		(*stop_scheduling)(struct cpu_hw_events *cpuc);
833 
834 	struct event_constraint *event_constraints;
835 	struct x86_pmu_quirk *quirks;
836 	void		(*limit_period)(struct perf_event *event, s64 *l);
837 
838 	/* PMI handler bits */
839 	unsigned int	late_ack		:1,
840 			mid_ack			:1,
841 			enabled_ack		:1;
842 	/*
843 	 * sysfs attrs
844 	 */
845 	int		attr_rdpmc_broken;
846 	int		attr_rdpmc;
847 	struct attribute **format_attrs;
848 
849 	ssize_t		(*events_sysfs_show)(char *page, u64 config);
850 	const struct attribute_group **attr_update;
851 
852 	unsigned long	attr_freeze_on_smi;
853 
854 	/*
855 	 * CPU Hotplug hooks
856 	 */
857 	int		(*cpu_prepare)(int cpu);
858 	void		(*cpu_starting)(int cpu);
859 	void		(*cpu_dying)(int cpu);
860 	void		(*cpu_dead)(int cpu);
861 
862 	void		(*check_microcode)(void);
863 	void		(*sched_task)(struct perf_event_pmu_context *pmu_ctx,
864 				      bool sched_in);
865 
866 	/*
867 	 * Intel Arch Perfmon v2+
868 	 */
869 	u64			intel_ctrl;
870 	union perf_capabilities intel_cap;
871 
872 	/*
873 	 * Intel DebugStore bits
874 	 */
875 	unsigned int	bts			:1,
876 			bts_active		:1,
877 			pebs			:1,
878 			pebs_active		:1,
879 			pebs_broken		:1,
880 			pebs_prec_dist		:1,
881 			pebs_no_tlb		:1,
882 			pebs_no_isolation	:1,
883 			pebs_block		:1,
884 			pebs_ept		:1;
885 	int		pebs_record_size;
886 	int		pebs_buffer_size;
887 	u64		pebs_events_mask;
888 	void		(*drain_pebs)(struct pt_regs *regs, struct perf_sample_data *data);
889 	struct event_constraint *pebs_constraints;
890 	void		(*pebs_aliases)(struct perf_event *event);
891 	u64		(*pebs_latency_data)(struct perf_event *event, u64 status);
892 	unsigned long	large_pebs_flags;
893 	u64		rtm_abort_event;
894 	u64		pebs_capable;
895 
896 	/*
897 	 * Intel LBR
898 	 */
899 	unsigned int	lbr_tos, lbr_from, lbr_to,
900 			lbr_info, lbr_nr;	   /* LBR base regs and size */
901 	union {
902 		u64	lbr_sel_mask;		   /* LBR_SELECT valid bits */
903 		u64	lbr_ctl_mask;		   /* LBR_CTL valid bits */
904 	};
905 	union {
906 		const int	*lbr_sel_map;	   /* lbr_select mappings */
907 		int		*lbr_ctl_map;	   /* LBR_CTL mappings */
908 	};
909 	bool		lbr_double_abort;	   /* duplicated lbr aborts */
910 	bool		lbr_pt_coexist;		   /* (LBR|BTS) may coexist with PT */
911 
912 	unsigned int	lbr_has_info:1;
913 	unsigned int	lbr_has_tsx:1;
914 	unsigned int	lbr_from_flags:1;
915 	unsigned int	lbr_to_cycles:1;
916 
917 	/*
918 	 * Intel Architectural LBR CPUID Enumeration
919 	 */
920 	unsigned int	lbr_depth_mask:8;
921 	unsigned int	lbr_deep_c_reset:1;
922 	unsigned int	lbr_lip:1;
923 	unsigned int	lbr_cpl:1;
924 	unsigned int	lbr_filter:1;
925 	unsigned int	lbr_call_stack:1;
926 	unsigned int	lbr_mispred:1;
927 	unsigned int	lbr_timed_lbr:1;
928 	unsigned int	lbr_br_type:1;
929 	unsigned int	lbr_counters:4;
930 
931 	void		(*lbr_reset)(void);
932 	void		(*lbr_read)(struct cpu_hw_events *cpuc);
933 	void		(*lbr_save)(void *ctx);
934 	void		(*lbr_restore)(void *ctx);
935 
936 	/*
937 	 * Intel PT/LBR/BTS are exclusive
938 	 */
939 	atomic_t	lbr_exclusive[x86_lbr_exclusive_max];
940 
941 	/*
942 	 * Intel perf metrics
943 	 */
944 	int		num_topdown_events;
945 
946 	/*
947 	 * perf task context (i.e. struct perf_event_pmu_context::task_ctx_data)
948 	 * switch helper to bridge calls from perf/core to perf/x86.
949 	 * See struct pmu::swap_task_ctx() usage for examples;
950 	 */
951 	void		(*swap_task_ctx)(struct perf_event_pmu_context *prev_epc,
952 					 struct perf_event_pmu_context *next_epc);
953 
954 	/*
955 	 * AMD bits
956 	 */
957 	unsigned int	amd_nb_constraints : 1;
958 	u64		perf_ctr_pair_en;
959 
960 	/*
961 	 * Extra registers for events
962 	 */
963 	struct extra_reg *extra_regs;
964 	unsigned int flags;
965 
966 	/*
967 	 * Intel host/guest support (KVM)
968 	 */
969 	struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr, void *data);
970 
971 	/*
972 	 * Check period value for PERF_EVENT_IOC_PERIOD ioctl.
973 	 */
974 	int (*check_period) (struct perf_event *event, u64 period);
975 
976 	int (*aux_output_match) (struct perf_event *event);
977 
978 	void (*filter)(struct pmu *pmu, int cpu, bool *ret);
979 	/*
980 	 * Hybrid support
981 	 *
982 	 * Most PMU capabilities are the same among different hybrid PMUs.
983 	 * The global x86_pmu saves the architecture capabilities, which
984 	 * are available for all PMUs. The hybrid_pmu only includes the
985 	 * unique capabilities.
986 	 */
987 	int				num_hybrid_pmus;
988 	struct x86_hybrid_pmu		*hybrid_pmu;
989 	enum hybrid_cpu_type (*get_hybrid_cpu_type)	(void);
990 };
991 
992 struct x86_perf_task_context_opt {
993 	int lbr_callstack_users;
994 	int lbr_stack_state;
995 	int log_id;
996 };
997 
998 struct x86_perf_task_context {
999 	u64 lbr_sel;
1000 	int tos;
1001 	int valid_lbrs;
1002 	struct x86_perf_task_context_opt opt;
1003 	struct lbr_entry lbr[MAX_LBR_ENTRIES];
1004 };
1005 
1006 struct x86_perf_task_context_arch_lbr {
1007 	struct x86_perf_task_context_opt opt;
1008 	struct lbr_entry entries[];
1009 };
1010 
1011 /*
1012  * Add padding to guarantee the 64-byte alignment of the state buffer.
1013  *
1014  * The structure is dynamically allocated. The size of the LBR state may vary
1015  * based on the number of LBR registers.
1016  *
1017  * Do not put anything after the LBR state.
1018  */
1019 struct x86_perf_task_context_arch_lbr_xsave {
1020 	struct x86_perf_task_context_opt		opt;
1021 
1022 	union {
1023 		struct xregs_state			xsave;
1024 		struct {
1025 			struct fxregs_state		i387;
1026 			struct xstate_header		header;
1027 			struct arch_lbr_state		lbr;
1028 		} __attribute__ ((packed, aligned (XSAVE_ALIGNMENT)));
1029 	};
1030 };
1031 
1032 #define x86_add_quirk(func_)						\
1033 do {									\
1034 	static struct x86_pmu_quirk __quirk __initdata = {		\
1035 		.func = func_,						\
1036 	};								\
1037 	__quirk.next = x86_pmu.quirks;					\
1038 	x86_pmu.quirks = &__quirk;					\
1039 } while (0)
1040 
1041 /*
1042  * x86_pmu flags
1043  */
1044 #define PMU_FL_NO_HT_SHARING	0x1 /* no hyper-threading resource sharing */
1045 #define PMU_FL_HAS_RSP_1	0x2 /* has 2 equivalent offcore_rsp regs   */
1046 #define PMU_FL_EXCL_CNTRS	0x4 /* has exclusive counter requirements  */
1047 #define PMU_FL_EXCL_ENABLED	0x8 /* exclusive counter active */
1048 #define PMU_FL_PEBS_ALL		0x10 /* all events are valid PEBS events */
1049 #define PMU_FL_TFA		0x20 /* deal with TSX force abort */
1050 #define PMU_FL_PAIR		0x40 /* merge counters for large incr. events */
1051 #define PMU_FL_INSTR_LATENCY	0x80 /* Support Instruction Latency in PEBS Memory Info Record */
1052 #define PMU_FL_MEM_LOADS_AUX	0x100 /* Require an auxiliary event for the complete memory info */
1053 #define PMU_FL_RETIRE_LATENCY	0x200 /* Support Retire Latency in PEBS */
1054 #define PMU_FL_BR_CNTR		0x400 /* Support branch counter logging */
1055 
1056 #define EVENT_VAR(_id)  event_attr_##_id
1057 #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
1058 
1059 #define EVENT_ATTR(_name, _id)						\
1060 static struct perf_pmu_events_attr EVENT_VAR(_id) = {			\
1061 	.attr		= __ATTR(_name, 0444, events_sysfs_show, NULL),	\
1062 	.id		= PERF_COUNT_HW_##_id,				\
1063 	.event_str	= NULL,						\
1064 };
1065 
1066 #define EVENT_ATTR_STR(_name, v, str)					\
1067 static struct perf_pmu_events_attr event_attr_##v = {			\
1068 	.attr		= __ATTR(_name, 0444, events_sysfs_show, NULL),	\
1069 	.id		= 0,						\
1070 	.event_str	= str,						\
1071 };
1072 
1073 #define EVENT_ATTR_STR_HT(_name, v, noht, ht)				\
1074 static struct perf_pmu_events_ht_attr event_attr_##v = {		\
1075 	.attr		= __ATTR(_name, 0444, events_ht_sysfs_show, NULL),\
1076 	.id		= 0,						\
1077 	.event_str_noht	= noht,						\
1078 	.event_str_ht	= ht,						\
1079 }
1080 
1081 #define EVENT_ATTR_STR_HYBRID(_name, v, str, _pmu)			\
1082 static struct perf_pmu_events_hybrid_attr event_attr_##v = {		\
1083 	.attr		= __ATTR(_name, 0444, events_hybrid_sysfs_show, NULL),\
1084 	.id		= 0,						\
1085 	.event_str	= str,						\
1086 	.pmu_type	= _pmu,						\
1087 }
1088 
1089 #define FORMAT_HYBRID_PTR(_id) (&format_attr_hybrid_##_id.attr.attr)
1090 
1091 #define FORMAT_ATTR_HYBRID(_name, _pmu)					\
1092 static struct perf_pmu_format_hybrid_attr format_attr_hybrid_##_name = {\
1093 	.attr		= __ATTR_RO(_name),				\
1094 	.pmu_type	= _pmu,						\
1095 }
1096 
1097 struct pmu *x86_get_pmu(unsigned int cpu);
1098 extern struct x86_pmu x86_pmu __read_mostly;
1099 
1100 DECLARE_STATIC_CALL(x86_pmu_set_period, *x86_pmu.set_period);
1101 DECLARE_STATIC_CALL(x86_pmu_update,     *x86_pmu.update);
1102 DECLARE_STATIC_CALL(x86_pmu_drain_pebs,	*x86_pmu.drain_pebs);
1103 
task_context_opt(void * ctx)1104 static __always_inline struct x86_perf_task_context_opt *task_context_opt(void *ctx)
1105 {
1106 	if (static_cpu_has(X86_FEATURE_ARCH_LBR))
1107 		return &((struct x86_perf_task_context_arch_lbr *)ctx)->opt;
1108 
1109 	return &((struct x86_perf_task_context *)ctx)->opt;
1110 }
1111 
x86_pmu_has_lbr_callstack(void)1112 static inline bool x86_pmu_has_lbr_callstack(void)
1113 {
1114 	return  x86_pmu.lbr_sel_map &&
1115 		x86_pmu.lbr_sel_map[PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT] > 0;
1116 }
1117 
1118 DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
1119 DECLARE_PER_CPU(u64 [X86_PMC_IDX_MAX], pmc_prev_left);
1120 
1121 int x86_perf_event_set_period(struct perf_event *event);
1122 
1123 /*
1124  * Generalized hw caching related hw_event table, filled
1125  * in on a per model basis. A value of 0 means
1126  * 'not supported', -1 means 'hw_event makes no sense on
1127  * this CPU', any other value means the raw hw_event
1128  * ID.
1129  */
1130 
1131 #define C(x) PERF_COUNT_HW_CACHE_##x
1132 
1133 extern u64 __read_mostly hw_cache_event_ids
1134 				[PERF_COUNT_HW_CACHE_MAX]
1135 				[PERF_COUNT_HW_CACHE_OP_MAX]
1136 				[PERF_COUNT_HW_CACHE_RESULT_MAX];
1137 extern u64 __read_mostly hw_cache_extra_regs
1138 				[PERF_COUNT_HW_CACHE_MAX]
1139 				[PERF_COUNT_HW_CACHE_OP_MAX]
1140 				[PERF_COUNT_HW_CACHE_RESULT_MAX];
1141 
1142 u64 x86_perf_event_update(struct perf_event *event);
1143 
x86_pmu_config_addr(int index)1144 static inline unsigned int x86_pmu_config_addr(int index)
1145 {
1146 	return x86_pmu.eventsel + (x86_pmu.addr_offset ?
1147 				   x86_pmu.addr_offset(index, true) : index);
1148 }
1149 
x86_pmu_event_addr(int index)1150 static inline unsigned int x86_pmu_event_addr(int index)
1151 {
1152 	return x86_pmu.perfctr + (x86_pmu.addr_offset ?
1153 				  x86_pmu.addr_offset(index, false) : index);
1154 }
1155 
x86_pmu_fixed_ctr_addr(int index)1156 static inline unsigned int x86_pmu_fixed_ctr_addr(int index)
1157 {
1158 	return x86_pmu.fixedctr + (x86_pmu.addr_offset ?
1159 				   x86_pmu.addr_offset(index, false) : index);
1160 }
1161 
x86_pmu_rdpmc_index(int index)1162 static inline int x86_pmu_rdpmc_index(int index)
1163 {
1164 	return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
1165 }
1166 
1167 bool check_hw_exists(struct pmu *pmu, unsigned long *cntr_mask,
1168 		     unsigned long *fixed_cntr_mask);
1169 
1170 int x86_add_exclusive(unsigned int what);
1171 
1172 void x86_del_exclusive(unsigned int what);
1173 
1174 int x86_reserve_hardware(void);
1175 
1176 void x86_release_hardware(void);
1177 
1178 int x86_pmu_max_precise(void);
1179 
1180 void hw_perf_lbr_event_destroy(struct perf_event *event);
1181 
1182 int x86_setup_perfctr(struct perf_event *event);
1183 
1184 int x86_pmu_hw_config(struct perf_event *event);
1185 
1186 void x86_pmu_disable_all(void);
1187 
has_amd_brs(struct hw_perf_event * hwc)1188 static inline bool has_amd_brs(struct hw_perf_event *hwc)
1189 {
1190 	return hwc->flags & PERF_X86_EVENT_AMD_BRS;
1191 }
1192 
is_counter_pair(struct hw_perf_event * hwc)1193 static inline bool is_counter_pair(struct hw_perf_event *hwc)
1194 {
1195 	return hwc->flags & PERF_X86_EVENT_PAIR;
1196 }
1197 
__x86_pmu_enable_event(struct hw_perf_event * hwc,u64 enable_mask)1198 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
1199 					  u64 enable_mask)
1200 {
1201 	u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
1202 
1203 	if (hwc->extra_reg.reg)
1204 		wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
1205 
1206 	/*
1207 	 * Add enabled Merge event on next counter
1208 	 * if large increment event being enabled on this counter
1209 	 */
1210 	if (is_counter_pair(hwc))
1211 		wrmsrl(x86_pmu_config_addr(hwc->idx + 1), x86_pmu.perf_ctr_pair_en);
1212 
1213 	wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
1214 }
1215 
1216 void x86_pmu_enable_all(int added);
1217 
1218 int perf_assign_events(struct event_constraint **constraints, int n,
1219 			int wmin, int wmax, int gpmax, int *assign);
1220 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
1221 
1222 void x86_pmu_stop(struct perf_event *event, int flags);
1223 
x86_pmu_disable_event(struct perf_event * event)1224 static inline void x86_pmu_disable_event(struct perf_event *event)
1225 {
1226 	u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
1227 	struct hw_perf_event *hwc = &event->hw;
1228 
1229 	wrmsrl(hwc->config_base, hwc->config & ~disable_mask);
1230 
1231 	if (is_counter_pair(hwc))
1232 		wrmsrl(x86_pmu_config_addr(hwc->idx + 1), 0);
1233 }
1234 
1235 void x86_pmu_enable_event(struct perf_event *event);
1236 
1237 int x86_pmu_handle_irq(struct pt_regs *regs);
1238 
1239 void x86_pmu_show_pmu_cap(struct pmu *pmu);
1240 
x86_pmu_num_counters(struct pmu * pmu)1241 static inline int x86_pmu_num_counters(struct pmu *pmu)
1242 {
1243 	return hweight64(hybrid(pmu, cntr_mask64));
1244 }
1245 
x86_pmu_max_num_counters(struct pmu * pmu)1246 static inline int x86_pmu_max_num_counters(struct pmu *pmu)
1247 {
1248 	return fls64(hybrid(pmu, cntr_mask64));
1249 }
1250 
x86_pmu_num_counters_fixed(struct pmu * pmu)1251 static inline int x86_pmu_num_counters_fixed(struct pmu *pmu)
1252 {
1253 	return hweight64(hybrid(pmu, fixed_cntr_mask64));
1254 }
1255 
x86_pmu_max_num_counters_fixed(struct pmu * pmu)1256 static inline int x86_pmu_max_num_counters_fixed(struct pmu *pmu)
1257 {
1258 	return fls64(hybrid(pmu, fixed_cntr_mask64));
1259 }
1260 
x86_pmu_get_event_config(struct perf_event * event)1261 static inline u64 x86_pmu_get_event_config(struct perf_event *event)
1262 {
1263 	return event->attr.config & hybrid(event->pmu, config_mask);
1264 }
1265 
1266 extern struct event_constraint emptyconstraint;
1267 
1268 extern struct event_constraint unconstrained;
1269 
kernel_ip(unsigned long ip)1270 static inline bool kernel_ip(unsigned long ip)
1271 {
1272 #ifdef CONFIG_X86_32
1273 	return ip > PAGE_OFFSET;
1274 #else
1275 	return (long)ip < 0;
1276 #endif
1277 }
1278 
1279 /*
1280  * Not all PMUs provide the right context information to place the reported IP
1281  * into full context. Specifically segment registers are typically not
1282  * supplied.
1283  *
1284  * Assuming the address is a linear address (it is for IBS), we fake the CS and
1285  * vm86 mode using the known zero-based code segment and 'fix up' the registers
1286  * to reflect this.
1287  *
1288  * Intel PEBS/LBR appear to typically provide the effective address, nothing
1289  * much we can do about that but pray and treat it like a linear address.
1290  */
set_linear_ip(struct pt_regs * regs,unsigned long ip)1291 static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
1292 {
1293 	regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
1294 	if (regs->flags & X86_VM_MASK)
1295 		regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
1296 	regs->ip = ip;
1297 }
1298 
1299 /*
1300  * x86control flow change classification
1301  * x86control flow changes include branches, interrupts, traps, faults
1302  */
1303 enum {
1304 	X86_BR_NONE		= 0,      /* unknown */
1305 
1306 	X86_BR_USER		= 1 << 0, /* branch target is user */
1307 	X86_BR_KERNEL		= 1 << 1, /* branch target is kernel */
1308 
1309 	X86_BR_CALL		= 1 << 2, /* call */
1310 	X86_BR_RET		= 1 << 3, /* return */
1311 	X86_BR_SYSCALL		= 1 << 4, /* syscall */
1312 	X86_BR_SYSRET		= 1 << 5, /* syscall return */
1313 	X86_BR_INT		= 1 << 6, /* sw interrupt */
1314 	X86_BR_IRET		= 1 << 7, /* return from interrupt */
1315 	X86_BR_JCC		= 1 << 8, /* conditional */
1316 	X86_BR_JMP		= 1 << 9, /* jump */
1317 	X86_BR_IRQ		= 1 << 10,/* hw interrupt or trap or fault */
1318 	X86_BR_IND_CALL		= 1 << 11,/* indirect calls */
1319 	X86_BR_ABORT		= 1 << 12,/* transaction abort */
1320 	X86_BR_IN_TX		= 1 << 13,/* in transaction */
1321 	X86_BR_NO_TX		= 1 << 14,/* not in transaction */
1322 	X86_BR_ZERO_CALL	= 1 << 15,/* zero length call */
1323 	X86_BR_CALL_STACK	= 1 << 16,/* call stack */
1324 	X86_BR_IND_JMP		= 1 << 17,/* indirect jump */
1325 
1326 	X86_BR_TYPE_SAVE	= 1 << 18,/* indicate to save branch type */
1327 
1328 };
1329 
1330 #define X86_BR_PLM (X86_BR_USER | X86_BR_KERNEL)
1331 #define X86_BR_ANYTX (X86_BR_NO_TX | X86_BR_IN_TX)
1332 
1333 #define X86_BR_ANY       \
1334 	(X86_BR_CALL    |\
1335 	 X86_BR_RET     |\
1336 	 X86_BR_SYSCALL |\
1337 	 X86_BR_SYSRET  |\
1338 	 X86_BR_INT     |\
1339 	 X86_BR_IRET    |\
1340 	 X86_BR_JCC     |\
1341 	 X86_BR_JMP	 |\
1342 	 X86_BR_IRQ	 |\
1343 	 X86_BR_ABORT	 |\
1344 	 X86_BR_IND_CALL |\
1345 	 X86_BR_IND_JMP  |\
1346 	 X86_BR_ZERO_CALL)
1347 
1348 #define X86_BR_ALL (X86_BR_PLM | X86_BR_ANY)
1349 
1350 #define X86_BR_ANY_CALL		 \
1351 	(X86_BR_CALL		|\
1352 	 X86_BR_IND_CALL	|\
1353 	 X86_BR_ZERO_CALL	|\
1354 	 X86_BR_SYSCALL		|\
1355 	 X86_BR_IRQ		|\
1356 	 X86_BR_INT)
1357 
1358 int common_branch_type(int type);
1359 int branch_type(unsigned long from, unsigned long to, int abort);
1360 int branch_type_fused(unsigned long from, unsigned long to, int abort,
1361 		      int *offset);
1362 
1363 ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
1364 ssize_t intel_event_sysfs_show(char *page, u64 config);
1365 
1366 ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
1367 			  char *page);
1368 ssize_t events_ht_sysfs_show(struct device *dev, struct device_attribute *attr,
1369 			  char *page);
1370 ssize_t events_hybrid_sysfs_show(struct device *dev,
1371 				 struct device_attribute *attr,
1372 				 char *page);
1373 
fixed_counter_disabled(int i,struct pmu * pmu)1374 static inline bool fixed_counter_disabled(int i, struct pmu *pmu)
1375 {
1376 	u64 intel_ctrl = hybrid(pmu, intel_ctrl);
1377 
1378 	return !(intel_ctrl >> (i + INTEL_PMC_IDX_FIXED));
1379 }
1380 
1381 #ifdef CONFIG_CPU_SUP_AMD
1382 
1383 int amd_pmu_init(void);
1384 
1385 int amd_pmu_lbr_init(void);
1386 void amd_pmu_lbr_reset(void);
1387 void amd_pmu_lbr_read(void);
1388 void amd_pmu_lbr_add(struct perf_event *event);
1389 void amd_pmu_lbr_del(struct perf_event *event);
1390 void amd_pmu_lbr_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in);
1391 void amd_pmu_lbr_enable_all(void);
1392 void amd_pmu_lbr_disable_all(void);
1393 int amd_pmu_lbr_hw_config(struct perf_event *event);
1394 
__amd_pmu_lbr_disable(void)1395 static __always_inline void __amd_pmu_lbr_disable(void)
1396 {
1397 	u64 dbg_ctl, dbg_extn_cfg;
1398 
1399 	rdmsrl(MSR_AMD_DBG_EXTN_CFG, dbg_extn_cfg);
1400 	wrmsrl(MSR_AMD_DBG_EXTN_CFG, dbg_extn_cfg & ~DBG_EXTN_CFG_LBRV2EN);
1401 
1402 	if (cpu_feature_enabled(X86_FEATURE_AMD_LBR_PMC_FREEZE)) {
1403 		rdmsrl(MSR_IA32_DEBUGCTLMSR, dbg_ctl);
1404 		wrmsrl(MSR_IA32_DEBUGCTLMSR, dbg_ctl & ~DEBUGCTLMSR_FREEZE_LBRS_ON_PMI);
1405 	}
1406 }
1407 
1408 #ifdef CONFIG_PERF_EVENTS_AMD_BRS
1409 
1410 #define AMD_FAM19H_BRS_EVENT 0xc4 /* RETIRED_TAKEN_BRANCH_INSTRUCTIONS */
1411 
1412 int amd_brs_init(void);
1413 void amd_brs_disable(void);
1414 void amd_brs_enable(void);
1415 void amd_brs_enable_all(void);
1416 void amd_brs_disable_all(void);
1417 void amd_brs_drain(void);
1418 void amd_brs_lopwr_init(void);
1419 int amd_brs_hw_config(struct perf_event *event);
1420 void amd_brs_reset(void);
1421 
amd_pmu_brs_add(struct perf_event * event)1422 static inline void amd_pmu_brs_add(struct perf_event *event)
1423 {
1424 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
1425 
1426 	perf_sched_cb_inc(event->pmu);
1427 	cpuc->lbr_users++;
1428 	/*
1429 	 * No need to reset BRS because it is reset
1430 	 * on brs_enable() and it is saturating
1431 	 */
1432 }
1433 
amd_pmu_brs_del(struct perf_event * event)1434 static inline void amd_pmu_brs_del(struct perf_event *event)
1435 {
1436 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
1437 
1438 	cpuc->lbr_users--;
1439 	WARN_ON_ONCE(cpuc->lbr_users < 0);
1440 
1441 	perf_sched_cb_dec(event->pmu);
1442 }
1443 
1444 void amd_pmu_brs_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in);
1445 #else
amd_brs_init(void)1446 static inline int amd_brs_init(void)
1447 {
1448 	return 0;
1449 }
amd_brs_disable(void)1450 static inline void amd_brs_disable(void) {}
amd_brs_enable(void)1451 static inline void amd_brs_enable(void) {}
amd_brs_drain(void)1452 static inline void amd_brs_drain(void) {}
amd_brs_lopwr_init(void)1453 static inline void amd_brs_lopwr_init(void) {}
amd_brs_disable_all(void)1454 static inline void amd_brs_disable_all(void) {}
amd_brs_hw_config(struct perf_event * event)1455 static inline int amd_brs_hw_config(struct perf_event *event)
1456 {
1457 	return 0;
1458 }
amd_brs_reset(void)1459 static inline void amd_brs_reset(void) {}
1460 
amd_pmu_brs_add(struct perf_event * event)1461 static inline void amd_pmu_brs_add(struct perf_event *event)
1462 {
1463 }
1464 
amd_pmu_brs_del(struct perf_event * event)1465 static inline void amd_pmu_brs_del(struct perf_event *event)
1466 {
1467 }
1468 
amd_pmu_brs_sched_task(struct perf_event_pmu_context * pmu_ctx,bool sched_in)1469 static inline void amd_pmu_brs_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in)
1470 {
1471 }
1472 
amd_brs_enable_all(void)1473 static inline void amd_brs_enable_all(void)
1474 {
1475 }
1476 
1477 #endif
1478 
1479 #else /* CONFIG_CPU_SUP_AMD */
1480 
amd_pmu_init(void)1481 static inline int amd_pmu_init(void)
1482 {
1483 	return 0;
1484 }
1485 
amd_brs_init(void)1486 static inline int amd_brs_init(void)
1487 {
1488 	return -EOPNOTSUPP;
1489 }
1490 
amd_brs_drain(void)1491 static inline void amd_brs_drain(void)
1492 {
1493 }
1494 
amd_brs_enable_all(void)1495 static inline void amd_brs_enable_all(void)
1496 {
1497 }
1498 
amd_brs_disable_all(void)1499 static inline void amd_brs_disable_all(void)
1500 {
1501 }
1502 #endif /* CONFIG_CPU_SUP_AMD */
1503 
is_pebs_pt(struct perf_event * event)1504 static inline int is_pebs_pt(struct perf_event *event)
1505 {
1506 	return !!(event->hw.flags & PERF_X86_EVENT_PEBS_VIA_PT);
1507 }
1508 
1509 #ifdef CONFIG_CPU_SUP_INTEL
1510 
intel_pmu_has_bts_period(struct perf_event * event,u64 period)1511 static inline bool intel_pmu_has_bts_period(struct perf_event *event, u64 period)
1512 {
1513 	struct hw_perf_event *hwc = &event->hw;
1514 	unsigned int hw_event, bts_event;
1515 
1516 	if (event->attr.freq)
1517 		return false;
1518 
1519 	hw_event = hwc->config & INTEL_ARCH_EVENT_MASK;
1520 	bts_event = x86_pmu.event_map(PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
1521 
1522 	return hw_event == bts_event && period == 1;
1523 }
1524 
intel_pmu_has_bts(struct perf_event * event)1525 static inline bool intel_pmu_has_bts(struct perf_event *event)
1526 {
1527 	struct hw_perf_event *hwc = &event->hw;
1528 
1529 	return intel_pmu_has_bts_period(event, hwc->sample_period);
1530 }
1531 
__intel_pmu_pebs_disable_all(void)1532 static __always_inline void __intel_pmu_pebs_disable_all(void)
1533 {
1534 	wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1535 }
1536 
__intel_pmu_arch_lbr_disable(void)1537 static __always_inline void __intel_pmu_arch_lbr_disable(void)
1538 {
1539 	wrmsrl(MSR_ARCH_LBR_CTL, 0);
1540 }
1541 
__intel_pmu_lbr_disable(void)1542 static __always_inline void __intel_pmu_lbr_disable(void)
1543 {
1544 	u64 debugctl;
1545 
1546 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
1547 	debugctl &= ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_FREEZE_LBRS_ON_PMI);
1548 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
1549 }
1550 
1551 int intel_pmu_save_and_restart(struct perf_event *event);
1552 
1553 struct event_constraint *
1554 x86_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
1555 			  struct perf_event *event);
1556 
1557 extern int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu);
1558 extern void intel_cpuc_finish(struct cpu_hw_events *cpuc);
1559 
1560 int intel_pmu_init(void);
1561 
1562 void init_debug_store_on_cpu(int cpu);
1563 
1564 void fini_debug_store_on_cpu(int cpu);
1565 
1566 void release_ds_buffers(void);
1567 
1568 void reserve_ds_buffers(void);
1569 
1570 void release_lbr_buffers(void);
1571 
1572 void reserve_lbr_buffers(void);
1573 
1574 extern struct event_constraint bts_constraint;
1575 extern struct event_constraint vlbr_constraint;
1576 
1577 void intel_pmu_enable_bts(u64 config);
1578 
1579 void intel_pmu_disable_bts(void);
1580 
1581 int intel_pmu_drain_bts_buffer(void);
1582 
1583 u64 grt_latency_data(struct perf_event *event, u64 status);
1584 
1585 u64 cmt_latency_data(struct perf_event *event, u64 status);
1586 
1587 u64 lnl_latency_data(struct perf_event *event, u64 status);
1588 
1589 extern struct event_constraint intel_core2_pebs_event_constraints[];
1590 
1591 extern struct event_constraint intel_atom_pebs_event_constraints[];
1592 
1593 extern struct event_constraint intel_slm_pebs_event_constraints[];
1594 
1595 extern struct event_constraint intel_glm_pebs_event_constraints[];
1596 
1597 extern struct event_constraint intel_glp_pebs_event_constraints[];
1598 
1599 extern struct event_constraint intel_grt_pebs_event_constraints[];
1600 
1601 extern struct event_constraint intel_nehalem_pebs_event_constraints[];
1602 
1603 extern struct event_constraint intel_westmere_pebs_event_constraints[];
1604 
1605 extern struct event_constraint intel_snb_pebs_event_constraints[];
1606 
1607 extern struct event_constraint intel_ivb_pebs_event_constraints[];
1608 
1609 extern struct event_constraint intel_hsw_pebs_event_constraints[];
1610 
1611 extern struct event_constraint intel_bdw_pebs_event_constraints[];
1612 
1613 extern struct event_constraint intel_skl_pebs_event_constraints[];
1614 
1615 extern struct event_constraint intel_icl_pebs_event_constraints[];
1616 
1617 extern struct event_constraint intel_glc_pebs_event_constraints[];
1618 
1619 extern struct event_constraint intel_lnc_pebs_event_constraints[];
1620 
1621 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
1622 
1623 void intel_pmu_pebs_add(struct perf_event *event);
1624 
1625 void intel_pmu_pebs_del(struct perf_event *event);
1626 
1627 void intel_pmu_pebs_enable(struct perf_event *event);
1628 
1629 void intel_pmu_pebs_disable(struct perf_event *event);
1630 
1631 void intel_pmu_pebs_enable_all(void);
1632 
1633 void intel_pmu_pebs_disable_all(void);
1634 
1635 void intel_pmu_pebs_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in);
1636 
1637 void intel_pmu_drain_pebs_buffer(void);
1638 
1639 void intel_pmu_store_pebs_lbrs(struct lbr_entry *lbr);
1640 
1641 void intel_ds_init(void);
1642 
1643 void intel_pmu_lbr_save_brstack(struct perf_sample_data *data,
1644 				struct cpu_hw_events *cpuc,
1645 				struct perf_event *event);
1646 
1647 void intel_pmu_lbr_swap_task_ctx(struct perf_event_pmu_context *prev_epc,
1648 				 struct perf_event_pmu_context *next_epc);
1649 
1650 void intel_pmu_lbr_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in);
1651 
1652 u64 lbr_from_signext_quirk_wr(u64 val);
1653 
1654 void intel_pmu_lbr_reset(void);
1655 
1656 void intel_pmu_lbr_reset_32(void);
1657 
1658 void intel_pmu_lbr_reset_64(void);
1659 
1660 void intel_pmu_lbr_add(struct perf_event *event);
1661 
1662 void intel_pmu_lbr_del(struct perf_event *event);
1663 
1664 void intel_pmu_lbr_enable_all(bool pmi);
1665 
1666 void intel_pmu_lbr_disable_all(void);
1667 
1668 void intel_pmu_lbr_read(void);
1669 
1670 void intel_pmu_lbr_read_32(struct cpu_hw_events *cpuc);
1671 
1672 void intel_pmu_lbr_read_64(struct cpu_hw_events *cpuc);
1673 
1674 void intel_pmu_lbr_save(void *ctx);
1675 
1676 void intel_pmu_lbr_restore(void *ctx);
1677 
1678 void intel_pmu_lbr_init_core(void);
1679 
1680 void intel_pmu_lbr_init_nhm(void);
1681 
1682 void intel_pmu_lbr_init_atom(void);
1683 
1684 void intel_pmu_lbr_init_slm(void);
1685 
1686 void intel_pmu_lbr_init_snb(void);
1687 
1688 void intel_pmu_lbr_init_hsw(void);
1689 
1690 void intel_pmu_lbr_init_skl(void);
1691 
1692 void intel_pmu_lbr_init_knl(void);
1693 
1694 void intel_pmu_lbr_init(void);
1695 
1696 void intel_pmu_arch_lbr_init(void);
1697 
1698 void intel_pmu_pebs_data_source_nhm(void);
1699 
1700 void intel_pmu_pebs_data_source_skl(bool pmem);
1701 
1702 void intel_pmu_pebs_data_source_adl(void);
1703 
1704 void intel_pmu_pebs_data_source_grt(void);
1705 
1706 void intel_pmu_pebs_data_source_mtl(void);
1707 
1708 void intel_pmu_pebs_data_source_cmt(void);
1709 
1710 void intel_pmu_pebs_data_source_lnl(void);
1711 
1712 int intel_pmu_setup_lbr_filter(struct perf_event *event);
1713 
1714 void intel_pt_interrupt(void);
1715 
1716 int intel_bts_interrupt(void);
1717 
1718 void intel_bts_enable_local(void);
1719 
1720 void intel_bts_disable_local(void);
1721 
1722 int p4_pmu_init(void);
1723 
1724 int p6_pmu_init(void);
1725 
1726 int knc_pmu_init(void);
1727 
is_ht_workaround_enabled(void)1728 static inline int is_ht_workaround_enabled(void)
1729 {
1730 	return !!(x86_pmu.flags & PMU_FL_EXCL_ENABLED);
1731 }
1732 
intel_pmu_pebs_mask(u64 cntr_mask)1733 static inline u64 intel_pmu_pebs_mask(u64 cntr_mask)
1734 {
1735 	return MAX_PEBS_EVENTS_MASK & cntr_mask;
1736 }
1737 
intel_pmu_max_num_pebs(struct pmu * pmu)1738 static inline int intel_pmu_max_num_pebs(struct pmu *pmu)
1739 {
1740 	static_assert(MAX_PEBS_EVENTS == 32);
1741 	return fls((u32)hybrid(pmu, pebs_events_mask));
1742 }
1743 
1744 #else /* CONFIG_CPU_SUP_INTEL */
1745 
reserve_ds_buffers(void)1746 static inline void reserve_ds_buffers(void)
1747 {
1748 }
1749 
release_ds_buffers(void)1750 static inline void release_ds_buffers(void)
1751 {
1752 }
1753 
release_lbr_buffers(void)1754 static inline void release_lbr_buffers(void)
1755 {
1756 }
1757 
reserve_lbr_buffers(void)1758 static inline void reserve_lbr_buffers(void)
1759 {
1760 }
1761 
intel_pmu_init(void)1762 static inline int intel_pmu_init(void)
1763 {
1764 	return 0;
1765 }
1766 
intel_cpuc_prepare(struct cpu_hw_events * cpuc,int cpu)1767 static inline int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu)
1768 {
1769 	return 0;
1770 }
1771 
intel_cpuc_finish(struct cpu_hw_events * cpuc)1772 static inline void intel_cpuc_finish(struct cpu_hw_events *cpuc)
1773 {
1774 }
1775 
is_ht_workaround_enabled(void)1776 static inline int is_ht_workaround_enabled(void)
1777 {
1778 	return 0;
1779 }
1780 #endif /* CONFIG_CPU_SUP_INTEL */
1781 
1782 #if ((defined CONFIG_CPU_SUP_CENTAUR) || (defined CONFIG_CPU_SUP_ZHAOXIN))
1783 int zhaoxin_pmu_init(void);
1784 #else
zhaoxin_pmu_init(void)1785 static inline int zhaoxin_pmu_init(void)
1786 {
1787 	return 0;
1788 }
1789 #endif /*CONFIG_CPU_SUP_CENTAUR or CONFIG_CPU_SUP_ZHAOXIN*/
1790