xref: /linux/arch/x86/events/perf_event.h (revision 0c8a32eed1625a65798286fb73fea8710a908545)
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/intel_ds.h>
18 
19 /* To enable MSR tracing please use the generic trace points. */
20 
21 /*
22  *          |   NHM/WSM    |      SNB     |
23  * register -------------------------------
24  *          |  HT  | no HT |  HT  | no HT |
25  *-----------------------------------------
26  * offcore  | core | core  | cpu  | core  |
27  * lbr_sel  | core | core  | cpu  | core  |
28  * ld_lat   | cpu  | core  | cpu  | core  |
29  *-----------------------------------------
30  *
31  * Given that there is a small number of shared regs,
32  * we can pre-allocate their slot in the per-cpu
33  * per-core reg tables.
34  */
35 enum extra_reg_type {
36 	EXTRA_REG_NONE  = -1,	/* not used */
37 
38 	EXTRA_REG_RSP_0 = 0,	/* offcore_response_0 */
39 	EXTRA_REG_RSP_1 = 1,	/* offcore_response_1 */
40 	EXTRA_REG_LBR   = 2,	/* lbr_select */
41 	EXTRA_REG_LDLAT = 3,	/* ld_lat_threshold */
42 	EXTRA_REG_FE    = 4,    /* fe_* */
43 
44 	EXTRA_REG_MAX		/* number of entries needed */
45 };
46 
47 struct event_constraint {
48 	union {
49 		unsigned long	idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
50 		u64		idxmsk64;
51 	};
52 	u64		code;
53 	u64		cmask;
54 	int		weight;
55 	int		overlap;
56 	int		flags;
57 	unsigned int	size;
58 };
59 
60 static inline bool constraint_match(struct event_constraint *c, u64 ecode)
61 {
62 	return ((ecode & c->cmask) - c->code) <= (u64)c->size;
63 }
64 
65 /*
66  * struct hw_perf_event.flags flags
67  */
68 #define PERF_X86_EVENT_PEBS_LDLAT	0x0001 /* ld+ldlat data address sampling */
69 #define PERF_X86_EVENT_PEBS_ST		0x0002 /* st data address sampling */
70 #define PERF_X86_EVENT_PEBS_ST_HSW	0x0004 /* haswell style datala, store */
71 #define PERF_X86_EVENT_PEBS_LD_HSW	0x0008 /* haswell style datala, load */
72 #define PERF_X86_EVENT_PEBS_NA_HSW	0x0010 /* haswell style datala, unknown */
73 #define PERF_X86_EVENT_EXCL		0x0020 /* HT exclusivity on counter */
74 #define PERF_X86_EVENT_DYNAMIC		0x0040 /* dynamic alloc'd constraint */
75 #define PERF_X86_EVENT_RDPMC_ALLOWED	0x0080 /* grant rdpmc permission */
76 #define PERF_X86_EVENT_EXCL_ACCT	0x0100 /* accounted EXCL event */
77 #define PERF_X86_EVENT_AUTO_RELOAD	0x0200 /* use PEBS auto-reload */
78 #define PERF_X86_EVENT_LARGE_PEBS	0x0400 /* use large PEBS */
79 #define PERF_X86_EVENT_PEBS_VIA_PT	0x0800 /* use PT buffer for PEBS */
80 #define PERF_X86_EVENT_PAIR		0x1000 /* Large Increment per Cycle */
81 #define PERF_X86_EVENT_LBR_SELECT	0x2000 /* Save/Restore MSR_LBR_SELECT */
82 #define PERF_X86_EVENT_TOPDOWN		0x4000 /* Count Topdown slots/metrics events */
83 
84 static inline bool is_topdown_count(struct perf_event *event)
85 {
86 	return event->hw.flags & PERF_X86_EVENT_TOPDOWN;
87 }
88 
89 static inline bool is_metric_event(struct perf_event *event)
90 {
91 	u64 config = event->attr.config;
92 
93 	return ((config & ARCH_PERFMON_EVENTSEL_EVENT) == 0) &&
94 		((config & INTEL_ARCH_EVENT_MASK) >= INTEL_TD_METRIC_RETIRING)  &&
95 		((config & INTEL_ARCH_EVENT_MASK) <= INTEL_TD_METRIC_MAX);
96 }
97 
98 static inline bool is_slots_event(struct perf_event *event)
99 {
100 	return (event->attr.config & INTEL_ARCH_EVENT_MASK) == INTEL_TD_SLOTS;
101 }
102 
103 static inline bool is_topdown_event(struct perf_event *event)
104 {
105 	return is_metric_event(event) || is_slots_event(event);
106 }
107 
108 struct amd_nb {
109 	int nb_id;  /* NorthBridge id */
110 	int refcnt; /* reference count */
111 	struct perf_event *owners[X86_PMC_IDX_MAX];
112 	struct event_constraint event_constraints[X86_PMC_IDX_MAX];
113 };
114 
115 #define PEBS_COUNTER_MASK	((1ULL << MAX_PEBS_EVENTS) - 1)
116 #define PEBS_PMI_AFTER_EACH_RECORD BIT_ULL(60)
117 #define PEBS_OUTPUT_OFFSET	61
118 #define PEBS_OUTPUT_MASK	(3ull << PEBS_OUTPUT_OFFSET)
119 #define PEBS_OUTPUT_PT		(1ull << PEBS_OUTPUT_OFFSET)
120 #define PEBS_VIA_PT_MASK	(PEBS_OUTPUT_PT | PEBS_PMI_AFTER_EACH_RECORD)
121 
122 /*
123  * Flags PEBS can handle without an PMI.
124  *
125  * TID can only be handled by flushing at context switch.
126  * REGS_USER can be handled for events limited to ring 3.
127  *
128  */
129 #define LARGE_PEBS_FLAGS \
130 	(PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_ADDR | \
131 	PERF_SAMPLE_ID | PERF_SAMPLE_CPU | PERF_SAMPLE_STREAM_ID | \
132 	PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
133 	PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR | \
134 	PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER | \
135 	PERF_SAMPLE_PERIOD | PERF_SAMPLE_CODE_PAGE_SIZE)
136 
137 #define PEBS_GP_REGS			\
138 	((1ULL << PERF_REG_X86_AX)    | \
139 	 (1ULL << PERF_REG_X86_BX)    | \
140 	 (1ULL << PERF_REG_X86_CX)    | \
141 	 (1ULL << PERF_REG_X86_DX)    | \
142 	 (1ULL << PERF_REG_X86_DI)    | \
143 	 (1ULL << PERF_REG_X86_SI)    | \
144 	 (1ULL << PERF_REG_X86_SP)    | \
145 	 (1ULL << PERF_REG_X86_BP)    | \
146 	 (1ULL << PERF_REG_X86_IP)    | \
147 	 (1ULL << PERF_REG_X86_FLAGS) | \
148 	 (1ULL << PERF_REG_X86_R8)    | \
149 	 (1ULL << PERF_REG_X86_R9)    | \
150 	 (1ULL << PERF_REG_X86_R10)   | \
151 	 (1ULL << PERF_REG_X86_R11)   | \
152 	 (1ULL << PERF_REG_X86_R12)   | \
153 	 (1ULL << PERF_REG_X86_R13)   | \
154 	 (1ULL << PERF_REG_X86_R14)   | \
155 	 (1ULL << PERF_REG_X86_R15))
156 
157 /*
158  * Per register state.
159  */
160 struct er_account {
161 	raw_spinlock_t      lock;	/* per-core: protect structure */
162 	u64                 config;	/* extra MSR config */
163 	u64                 reg;	/* extra MSR number */
164 	atomic_t            ref;	/* reference count */
165 };
166 
167 /*
168  * Per core/cpu state
169  *
170  * Used to coordinate shared registers between HT threads or
171  * among events on a single PMU.
172  */
173 struct intel_shared_regs {
174 	struct er_account       regs[EXTRA_REG_MAX];
175 	int                     refcnt;		/* per-core: #HT threads */
176 	unsigned                core_id;	/* per-core: core id */
177 };
178 
179 enum intel_excl_state_type {
180 	INTEL_EXCL_UNUSED    = 0, /* counter is unused */
181 	INTEL_EXCL_SHARED    = 1, /* counter can be used by both threads */
182 	INTEL_EXCL_EXCLUSIVE = 2, /* counter can be used by one thread only */
183 };
184 
185 struct intel_excl_states {
186 	enum intel_excl_state_type state[X86_PMC_IDX_MAX];
187 	bool sched_started; /* true if scheduling has started */
188 };
189 
190 struct intel_excl_cntrs {
191 	raw_spinlock_t	lock;
192 
193 	struct intel_excl_states states[2];
194 
195 	union {
196 		u16	has_exclusive[2];
197 		u32	exclusive_present;
198 	};
199 
200 	int		refcnt;		/* per-core: #HT threads */
201 	unsigned	core_id;	/* per-core: core id */
202 };
203 
204 struct x86_perf_task_context;
205 #define MAX_LBR_ENTRIES		32
206 
207 enum {
208 	LBR_FORMAT_32		= 0x00,
209 	LBR_FORMAT_LIP		= 0x01,
210 	LBR_FORMAT_EIP		= 0x02,
211 	LBR_FORMAT_EIP_FLAGS	= 0x03,
212 	LBR_FORMAT_EIP_FLAGS2	= 0x04,
213 	LBR_FORMAT_INFO		= 0x05,
214 	LBR_FORMAT_TIME		= 0x06,
215 	LBR_FORMAT_MAX_KNOWN    = LBR_FORMAT_TIME,
216 };
217 
218 enum {
219 	X86_PERF_KFREE_SHARED = 0,
220 	X86_PERF_KFREE_EXCL   = 1,
221 	X86_PERF_KFREE_MAX
222 };
223 
224 struct cpu_hw_events {
225 	/*
226 	 * Generic x86 PMC bits
227 	 */
228 	struct perf_event	*events[X86_PMC_IDX_MAX]; /* in counter order */
229 	unsigned long		active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
230 	unsigned long		running[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
231 	int			enabled;
232 
233 	int			n_events; /* the # of events in the below arrays */
234 	int			n_added;  /* the # last events in the below arrays;
235 					     they've never been enabled yet */
236 	int			n_txn;    /* the # last events in the below arrays;
237 					     added in the current transaction */
238 	int			n_txn_pair;
239 	int			n_txn_metric;
240 	int			assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
241 	u64			tags[X86_PMC_IDX_MAX];
242 
243 	struct perf_event	*event_list[X86_PMC_IDX_MAX]; /* in enabled order */
244 	struct event_constraint	*event_constraint[X86_PMC_IDX_MAX];
245 
246 	int			n_excl; /* the number of exclusive events */
247 
248 	unsigned int		txn_flags;
249 	int			is_fake;
250 
251 	/*
252 	 * Intel DebugStore bits
253 	 */
254 	struct debug_store	*ds;
255 	void			*ds_pebs_vaddr;
256 	void			*ds_bts_vaddr;
257 	u64			pebs_enabled;
258 	int			n_pebs;
259 	int			n_large_pebs;
260 	int			n_pebs_via_pt;
261 	int			pebs_output;
262 
263 	/* Current super set of events hardware configuration */
264 	u64			pebs_data_cfg;
265 	u64			active_pebs_data_cfg;
266 	int			pebs_record_size;
267 
268 	/*
269 	 * Intel LBR bits
270 	 */
271 	int				lbr_users;
272 	int				lbr_pebs_users;
273 	struct perf_branch_stack	lbr_stack;
274 	struct perf_branch_entry	lbr_entries[MAX_LBR_ENTRIES];
275 	union {
276 		struct er_account		*lbr_sel;
277 		struct er_account		*lbr_ctl;
278 	};
279 	u64				br_sel;
280 	void				*last_task_ctx;
281 	int				last_log_id;
282 	int				lbr_select;
283 	void				*lbr_xsave;
284 
285 	/*
286 	 * Intel host/guest exclude bits
287 	 */
288 	u64				intel_ctrl_guest_mask;
289 	u64				intel_ctrl_host_mask;
290 	struct perf_guest_switch_msr	guest_switch_msrs[X86_PMC_IDX_MAX];
291 
292 	/*
293 	 * Intel checkpoint mask
294 	 */
295 	u64				intel_cp_status;
296 
297 	/*
298 	 * manage shared (per-core, per-cpu) registers
299 	 * used on Intel NHM/WSM/SNB
300 	 */
301 	struct intel_shared_regs	*shared_regs;
302 	/*
303 	 * manage exclusive counter access between hyperthread
304 	 */
305 	struct event_constraint *constraint_list; /* in enable order */
306 	struct intel_excl_cntrs		*excl_cntrs;
307 	int excl_thread_id; /* 0 or 1 */
308 
309 	/*
310 	 * SKL TSX_FORCE_ABORT shadow
311 	 */
312 	u64				tfa_shadow;
313 
314 	/*
315 	 * Perf Metrics
316 	 */
317 	/* number of accepted metrics events */
318 	int				n_metric;
319 
320 	/*
321 	 * AMD specific bits
322 	 */
323 	struct amd_nb			*amd_nb;
324 	/* Inverted mask of bits to clear in the perf_ctr ctrl registers */
325 	u64				perf_ctr_virt_mask;
326 	int				n_pair; /* Large increment events */
327 
328 	void				*kfree_on_online[X86_PERF_KFREE_MAX];
329 };
330 
331 #define __EVENT_CONSTRAINT_RANGE(c, e, n, m, w, o, f) {	\
332 	{ .idxmsk64 = (n) },		\
333 	.code = (c),			\
334 	.size = (e) - (c),		\
335 	.cmask = (m),			\
336 	.weight = (w),			\
337 	.overlap = (o),			\
338 	.flags = f,			\
339 }
340 
341 #define __EVENT_CONSTRAINT(c, n, m, w, o, f) \
342 	__EVENT_CONSTRAINT_RANGE(c, c, n, m, w, o, f)
343 
344 #define EVENT_CONSTRAINT(c, n, m)	\
345 	__EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0)
346 
347 /*
348  * The constraint_match() function only works for 'simple' event codes
349  * and not for extended (AMD64_EVENTSEL_EVENT) events codes.
350  */
351 #define EVENT_CONSTRAINT_RANGE(c, e, n, m) \
352 	__EVENT_CONSTRAINT_RANGE(c, e, n, m, HWEIGHT(n), 0, 0)
353 
354 #define INTEL_EXCLEVT_CONSTRAINT(c, n)	\
355 	__EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT, HWEIGHT(n),\
356 			   0, PERF_X86_EVENT_EXCL)
357 
358 /*
359  * The overlap flag marks event constraints with overlapping counter
360  * masks. This is the case if the counter mask of such an event is not
361  * a subset of any other counter mask of a constraint with an equal or
362  * higher weight, e.g.:
363  *
364  *  c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
365  *  c_another1 = EVENT_CONSTRAINT(0, 0x07, 0);
366  *  c_another2 = EVENT_CONSTRAINT(0, 0x38, 0);
367  *
368  * The event scheduler may not select the correct counter in the first
369  * cycle because it needs to know which subsequent events will be
370  * scheduled. It may fail to schedule the events then. So we set the
371  * overlap flag for such constraints to give the scheduler a hint which
372  * events to select for counter rescheduling.
373  *
374  * Care must be taken as the rescheduling algorithm is O(n!) which
375  * will increase scheduling cycles for an over-committed system
376  * dramatically.  The number of such EVENT_CONSTRAINT_OVERLAP() macros
377  * and its counter masks must be kept at a minimum.
378  */
379 #define EVENT_CONSTRAINT_OVERLAP(c, n, m)	\
380 	__EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0)
381 
382 /*
383  * Constraint on the Event code.
384  */
385 #define INTEL_EVENT_CONSTRAINT(c, n)	\
386 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
387 
388 /*
389  * Constraint on a range of Event codes
390  */
391 #define INTEL_EVENT_CONSTRAINT_RANGE(c, e, n)			\
392 	EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT)
393 
394 /*
395  * Constraint on the Event code + UMask + fixed-mask
396  *
397  * filter mask to validate fixed counter events.
398  * the following filters disqualify for fixed counters:
399  *  - inv
400  *  - edge
401  *  - cnt-mask
402  *  - in_tx
403  *  - in_tx_checkpointed
404  *  The other filters are supported by fixed counters.
405  *  The any-thread option is supported starting with v3.
406  */
407 #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED)
408 #define FIXED_EVENT_CONSTRAINT(c, n)	\
409 	EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
410 
411 /*
412  * The special metric counters do not actually exist. They are calculated from
413  * the combination of the FxCtr3 + MSR_PERF_METRICS.
414  *
415  * The special metric counters are mapped to a dummy offset for the scheduler.
416  * The sharing between multiple users of the same metric without multiplexing
417  * is not allowed, even though the hardware supports that in principle.
418  */
419 
420 #define METRIC_EVENT_CONSTRAINT(c, n)					\
421 	EVENT_CONSTRAINT(c, (1ULL << (INTEL_PMC_IDX_METRIC_BASE + n)),	\
422 			 INTEL_ARCH_EVENT_MASK)
423 
424 /*
425  * Constraint on the Event code + UMask
426  */
427 #define INTEL_UEVENT_CONSTRAINT(c, n)	\
428 	EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
429 
430 /* Constraint on specific umask bit only + event */
431 #define INTEL_UBIT_EVENT_CONSTRAINT(c, n)	\
432 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|(c))
433 
434 /* Like UEVENT_CONSTRAINT, but match flags too */
435 #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n)	\
436 	EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
437 
438 #define INTEL_EXCLUEVT_CONSTRAINT(c, n)	\
439 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK, \
440 			   HWEIGHT(n), 0, PERF_X86_EVENT_EXCL)
441 
442 #define INTEL_PLD_CONSTRAINT(c, n)	\
443 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
444 			   HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
445 
446 #define INTEL_PST_CONSTRAINT(c, n)	\
447 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
448 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
449 
450 /* Event constraint, but match on all event flags too. */
451 #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
452 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
453 
454 #define INTEL_FLAGS_EVENT_CONSTRAINT_RANGE(c, e, n)			\
455 	EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
456 
457 /* Check only flags, but allow all event/umask */
458 #define INTEL_ALL_EVENT_CONSTRAINT(code, n)	\
459 	EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
460 
461 /* Check flags and event code, and set the HSW store flag */
462 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
463 	__EVENT_CONSTRAINT(code, n, 			\
464 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
465 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
466 
467 /* Check flags and event code, and set the HSW load flag */
468 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
469 	__EVENT_CONSTRAINT(code, n,			\
470 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
471 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
472 
473 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD_RANGE(code, end, n) \
474 	__EVENT_CONSTRAINT_RANGE(code, end, n,				\
475 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
476 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
477 
478 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_XLD(code, n) \
479 	__EVENT_CONSTRAINT(code, n,			\
480 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
481 			  HWEIGHT(n), 0, \
482 			  PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
483 
484 /* Check flags and event code/umask, and set the HSW store flag */
485 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
486 	__EVENT_CONSTRAINT(code, n, 			\
487 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
488 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
489 
490 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XST(code, n) \
491 	__EVENT_CONSTRAINT(code, n,			\
492 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
493 			  HWEIGHT(n), 0, \
494 			  PERF_X86_EVENT_PEBS_ST_HSW|PERF_X86_EVENT_EXCL)
495 
496 /* Check flags and event code/umask, and set the HSW load flag */
497 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
498 	__EVENT_CONSTRAINT(code, n, 			\
499 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
500 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
501 
502 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XLD(code, n) \
503 	__EVENT_CONSTRAINT(code, n,			\
504 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
505 			  HWEIGHT(n), 0, \
506 			  PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
507 
508 /* Check flags and event code/umask, and set the HSW N/A flag */
509 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
510 	__EVENT_CONSTRAINT(code, n, 			\
511 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
512 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
513 
514 
515 /*
516  * We define the end marker as having a weight of -1
517  * to enable blacklisting of events using a counter bitmask
518  * of zero and thus a weight of zero.
519  * The end marker has a weight that cannot possibly be
520  * obtained from counting the bits in the bitmask.
521  */
522 #define EVENT_CONSTRAINT_END { .weight = -1 }
523 
524 /*
525  * Check for end marker with weight == -1
526  */
527 #define for_each_event_constraint(e, c)	\
528 	for ((e) = (c); (e)->weight != -1; (e)++)
529 
530 /*
531  * Extra registers for specific events.
532  *
533  * Some events need large masks and require external MSRs.
534  * Those extra MSRs end up being shared for all events on
535  * a PMU and sometimes between PMU of sibling HT threads.
536  * In either case, the kernel needs to handle conflicting
537  * accesses to those extra, shared, regs. The data structure
538  * to manage those registers is stored in cpu_hw_event.
539  */
540 struct extra_reg {
541 	unsigned int		event;
542 	unsigned int		msr;
543 	u64			config_mask;
544 	u64			valid_mask;
545 	int			idx;  /* per_xxx->regs[] reg index */
546 	bool			extra_msr_access;
547 };
548 
549 #define EVENT_EXTRA_REG(e, ms, m, vm, i) {	\
550 	.event = (e),			\
551 	.msr = (ms),			\
552 	.config_mask = (m),		\
553 	.valid_mask = (vm),		\
554 	.idx = EXTRA_REG_##i,		\
555 	.extra_msr_access = true,	\
556 	}
557 
558 #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx)	\
559 	EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
560 
561 #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
562 	EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
563 			ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
564 
565 #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
566 	INTEL_UEVENT_EXTRA_REG(c, \
567 			       MSR_PEBS_LD_LAT_THRESHOLD, \
568 			       0xffff, \
569 			       LDLAT)
570 
571 #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
572 
573 union perf_capabilities {
574 	struct {
575 		u64	lbr_format:6;
576 		u64	pebs_trap:1;
577 		u64	pebs_arch_reg:1;
578 		u64	pebs_format:4;
579 		u64	smm_freeze:1;
580 		/*
581 		 * PMU supports separate counter range for writing
582 		 * values > 32bit.
583 		 */
584 		u64	full_width_write:1;
585 		u64     pebs_baseline:1;
586 		u64	perf_metrics:1;
587 		u64	pebs_output_pt_available:1;
588 		u64	anythread_deprecated:1;
589 	};
590 	u64	capabilities;
591 };
592 
593 struct x86_pmu_quirk {
594 	struct x86_pmu_quirk *next;
595 	void (*func)(void);
596 };
597 
598 union x86_pmu_config {
599 	struct {
600 		u64 event:8,
601 		    umask:8,
602 		    usr:1,
603 		    os:1,
604 		    edge:1,
605 		    pc:1,
606 		    interrupt:1,
607 		    __reserved1:1,
608 		    en:1,
609 		    inv:1,
610 		    cmask:8,
611 		    event2:4,
612 		    __reserved2:4,
613 		    go:1,
614 		    ho:1;
615 	} bits;
616 	u64 value;
617 };
618 
619 #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
620 
621 enum {
622 	x86_lbr_exclusive_lbr,
623 	x86_lbr_exclusive_bts,
624 	x86_lbr_exclusive_pt,
625 	x86_lbr_exclusive_max,
626 };
627 
628 /*
629  * struct x86_pmu - generic x86 pmu
630  */
631 struct x86_pmu {
632 	/*
633 	 * Generic x86 PMC bits
634 	 */
635 	const char	*name;
636 	int		version;
637 	int		(*handle_irq)(struct pt_regs *);
638 	void		(*disable_all)(void);
639 	void		(*enable_all)(int added);
640 	void		(*enable)(struct perf_event *);
641 	void		(*disable)(struct perf_event *);
642 	void		(*add)(struct perf_event *);
643 	void		(*del)(struct perf_event *);
644 	void		(*read)(struct perf_event *event);
645 	int		(*hw_config)(struct perf_event *event);
646 	int		(*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
647 	unsigned	eventsel;
648 	unsigned	perfctr;
649 	int		(*addr_offset)(int index, bool eventsel);
650 	int		(*rdpmc_index)(int index);
651 	u64		(*event_map)(int);
652 	int		max_events;
653 	int		num_counters;
654 	int		num_counters_fixed;
655 	int		cntval_bits;
656 	u64		cntval_mask;
657 	union {
658 			unsigned long events_maskl;
659 			unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
660 	};
661 	int		events_mask_len;
662 	int		apic;
663 	u64		max_period;
664 	struct event_constraint *
665 			(*get_event_constraints)(struct cpu_hw_events *cpuc,
666 						 int idx,
667 						 struct perf_event *event);
668 
669 	void		(*put_event_constraints)(struct cpu_hw_events *cpuc,
670 						 struct perf_event *event);
671 
672 	void		(*start_scheduling)(struct cpu_hw_events *cpuc);
673 
674 	void		(*commit_scheduling)(struct cpu_hw_events *cpuc, int idx, int cntr);
675 
676 	void		(*stop_scheduling)(struct cpu_hw_events *cpuc);
677 
678 	struct event_constraint *event_constraints;
679 	struct x86_pmu_quirk *quirks;
680 	int		perfctr_second_write;
681 	u64		(*limit_period)(struct perf_event *event, u64 l);
682 
683 	/* PMI handler bits */
684 	unsigned int	late_ack		:1,
685 			enabled_ack		:1,
686 			counter_freezing	:1;
687 	/*
688 	 * sysfs attrs
689 	 */
690 	int		attr_rdpmc_broken;
691 	int		attr_rdpmc;
692 	struct attribute **format_attrs;
693 
694 	ssize_t		(*events_sysfs_show)(char *page, u64 config);
695 	const struct attribute_group **attr_update;
696 
697 	unsigned long	attr_freeze_on_smi;
698 
699 	/*
700 	 * CPU Hotplug hooks
701 	 */
702 	int		(*cpu_prepare)(int cpu);
703 	void		(*cpu_starting)(int cpu);
704 	void		(*cpu_dying)(int cpu);
705 	void		(*cpu_dead)(int cpu);
706 
707 	void		(*check_microcode)(void);
708 	void		(*sched_task)(struct perf_event_context *ctx,
709 				      bool sched_in);
710 
711 	/*
712 	 * Intel Arch Perfmon v2+
713 	 */
714 	u64			intel_ctrl;
715 	union perf_capabilities intel_cap;
716 
717 	/*
718 	 * Intel DebugStore bits
719 	 */
720 	unsigned int	bts			:1,
721 			bts_active		:1,
722 			pebs			:1,
723 			pebs_active		:1,
724 			pebs_broken		:1,
725 			pebs_prec_dist		:1,
726 			pebs_no_tlb		:1,
727 			pebs_no_isolation	:1;
728 	int		pebs_record_size;
729 	int		pebs_buffer_size;
730 	int		max_pebs_events;
731 	void		(*drain_pebs)(struct pt_regs *regs, struct perf_sample_data *data);
732 	struct event_constraint *pebs_constraints;
733 	void		(*pebs_aliases)(struct perf_event *event);
734 	unsigned long	large_pebs_flags;
735 	u64		rtm_abort_event;
736 
737 	/*
738 	 * Intel LBR
739 	 */
740 	unsigned int	lbr_tos, lbr_from, lbr_to,
741 			lbr_info, lbr_nr;	   /* LBR base regs and size */
742 	union {
743 		u64	lbr_sel_mask;		   /* LBR_SELECT valid bits */
744 		u64	lbr_ctl_mask;		   /* LBR_CTL valid bits */
745 	};
746 	union {
747 		const int	*lbr_sel_map;	   /* lbr_select mappings */
748 		int		*lbr_ctl_map;	   /* LBR_CTL mappings */
749 	};
750 	bool		lbr_double_abort;	   /* duplicated lbr aborts */
751 	bool		lbr_pt_coexist;		   /* (LBR|BTS) may coexist with PT */
752 
753 	/*
754 	 * Intel Architectural LBR CPUID Enumeration
755 	 */
756 	unsigned int	lbr_depth_mask:8;
757 	unsigned int	lbr_deep_c_reset:1;
758 	unsigned int	lbr_lip:1;
759 	unsigned int	lbr_cpl:1;
760 	unsigned int	lbr_filter:1;
761 	unsigned int	lbr_call_stack:1;
762 	unsigned int	lbr_mispred:1;
763 	unsigned int	lbr_timed_lbr:1;
764 	unsigned int	lbr_br_type:1;
765 
766 	void		(*lbr_reset)(void);
767 	void		(*lbr_read)(struct cpu_hw_events *cpuc);
768 	void		(*lbr_save)(void *ctx);
769 	void		(*lbr_restore)(void *ctx);
770 
771 	/*
772 	 * Intel PT/LBR/BTS are exclusive
773 	 */
774 	atomic_t	lbr_exclusive[x86_lbr_exclusive_max];
775 
776 	/*
777 	 * Intel perf metrics
778 	 */
779 	u64		(*update_topdown_event)(struct perf_event *event);
780 	int		(*set_topdown_event_period)(struct perf_event *event);
781 
782 	/*
783 	 * perf task context (i.e. struct perf_event_context::task_ctx_data)
784 	 * switch helper to bridge calls from perf/core to perf/x86.
785 	 * See struct pmu::swap_task_ctx() usage for examples;
786 	 */
787 	void		(*swap_task_ctx)(struct perf_event_context *prev,
788 					 struct perf_event_context *next);
789 
790 	/*
791 	 * AMD bits
792 	 */
793 	unsigned int	amd_nb_constraints : 1;
794 	u64		perf_ctr_pair_en;
795 
796 	/*
797 	 * Extra registers for events
798 	 */
799 	struct extra_reg *extra_regs;
800 	unsigned int flags;
801 
802 	/*
803 	 * Intel host/guest support (KVM)
804 	 */
805 	struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr);
806 
807 	/*
808 	 * Check period value for PERF_EVENT_IOC_PERIOD ioctl.
809 	 */
810 	int (*check_period) (struct perf_event *event, u64 period);
811 
812 	int (*aux_output_match) (struct perf_event *event);
813 };
814 
815 struct x86_perf_task_context_opt {
816 	int lbr_callstack_users;
817 	int lbr_stack_state;
818 	int log_id;
819 };
820 
821 struct x86_perf_task_context {
822 	u64 lbr_sel;
823 	int tos;
824 	int valid_lbrs;
825 	struct x86_perf_task_context_opt opt;
826 	struct lbr_entry lbr[MAX_LBR_ENTRIES];
827 };
828 
829 struct x86_perf_task_context_arch_lbr {
830 	struct x86_perf_task_context_opt opt;
831 	struct lbr_entry entries[];
832 };
833 
834 /*
835  * Add padding to guarantee the 64-byte alignment of the state buffer.
836  *
837  * The structure is dynamically allocated. The size of the LBR state may vary
838  * based on the number of LBR registers.
839  *
840  * Do not put anything after the LBR state.
841  */
842 struct x86_perf_task_context_arch_lbr_xsave {
843 	struct x86_perf_task_context_opt		opt;
844 
845 	union {
846 		struct xregs_state			xsave;
847 		struct {
848 			struct fxregs_state		i387;
849 			struct xstate_header		header;
850 			struct arch_lbr_state		lbr;
851 		} __attribute__ ((packed, aligned (XSAVE_ALIGNMENT)));
852 	};
853 };
854 
855 #define x86_add_quirk(func_)						\
856 do {									\
857 	static struct x86_pmu_quirk __quirk __initdata = {		\
858 		.func = func_,						\
859 	};								\
860 	__quirk.next = x86_pmu.quirks;					\
861 	x86_pmu.quirks = &__quirk;					\
862 } while (0)
863 
864 /*
865  * x86_pmu flags
866  */
867 #define PMU_FL_NO_HT_SHARING	0x1 /* no hyper-threading resource sharing */
868 #define PMU_FL_HAS_RSP_1	0x2 /* has 2 equivalent offcore_rsp regs   */
869 #define PMU_FL_EXCL_CNTRS	0x4 /* has exclusive counter requirements  */
870 #define PMU_FL_EXCL_ENABLED	0x8 /* exclusive counter active */
871 #define PMU_FL_PEBS_ALL		0x10 /* all events are valid PEBS events */
872 #define PMU_FL_TFA		0x20 /* deal with TSX force abort */
873 #define PMU_FL_PAIR		0x40 /* merge counters for large incr. events */
874 
875 #define EVENT_VAR(_id)  event_attr_##_id
876 #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
877 
878 #define EVENT_ATTR(_name, _id)						\
879 static struct perf_pmu_events_attr EVENT_VAR(_id) = {			\
880 	.attr		= __ATTR(_name, 0444, events_sysfs_show, NULL),	\
881 	.id		= PERF_COUNT_HW_##_id,				\
882 	.event_str	= NULL,						\
883 };
884 
885 #define EVENT_ATTR_STR(_name, v, str)					\
886 static struct perf_pmu_events_attr event_attr_##v = {			\
887 	.attr		= __ATTR(_name, 0444, events_sysfs_show, NULL),	\
888 	.id		= 0,						\
889 	.event_str	= str,						\
890 };
891 
892 #define EVENT_ATTR_STR_HT(_name, v, noht, ht)				\
893 static struct perf_pmu_events_ht_attr event_attr_##v = {		\
894 	.attr		= __ATTR(_name, 0444, events_ht_sysfs_show, NULL),\
895 	.id		= 0,						\
896 	.event_str_noht	= noht,						\
897 	.event_str_ht	= ht,						\
898 }
899 
900 struct pmu *x86_get_pmu(void);
901 extern struct x86_pmu x86_pmu __read_mostly;
902 
903 static __always_inline struct x86_perf_task_context_opt *task_context_opt(void *ctx)
904 {
905 	if (static_cpu_has(X86_FEATURE_ARCH_LBR))
906 		return &((struct x86_perf_task_context_arch_lbr *)ctx)->opt;
907 
908 	return &((struct x86_perf_task_context *)ctx)->opt;
909 }
910 
911 static inline bool x86_pmu_has_lbr_callstack(void)
912 {
913 	return  x86_pmu.lbr_sel_map &&
914 		x86_pmu.lbr_sel_map[PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT] > 0;
915 }
916 
917 DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
918 
919 int x86_perf_event_set_period(struct perf_event *event);
920 
921 /*
922  * Generalized hw caching related hw_event table, filled
923  * in on a per model basis. A value of 0 means
924  * 'not supported', -1 means 'hw_event makes no sense on
925  * this CPU', any other value means the raw hw_event
926  * ID.
927  */
928 
929 #define C(x) PERF_COUNT_HW_CACHE_##x
930 
931 extern u64 __read_mostly hw_cache_event_ids
932 				[PERF_COUNT_HW_CACHE_MAX]
933 				[PERF_COUNT_HW_CACHE_OP_MAX]
934 				[PERF_COUNT_HW_CACHE_RESULT_MAX];
935 extern u64 __read_mostly hw_cache_extra_regs
936 				[PERF_COUNT_HW_CACHE_MAX]
937 				[PERF_COUNT_HW_CACHE_OP_MAX]
938 				[PERF_COUNT_HW_CACHE_RESULT_MAX];
939 
940 u64 x86_perf_event_update(struct perf_event *event);
941 
942 static inline unsigned int x86_pmu_config_addr(int index)
943 {
944 	return x86_pmu.eventsel + (x86_pmu.addr_offset ?
945 				   x86_pmu.addr_offset(index, true) : index);
946 }
947 
948 static inline unsigned int x86_pmu_event_addr(int index)
949 {
950 	return x86_pmu.perfctr + (x86_pmu.addr_offset ?
951 				  x86_pmu.addr_offset(index, false) : index);
952 }
953 
954 static inline int x86_pmu_rdpmc_index(int index)
955 {
956 	return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
957 }
958 
959 int x86_add_exclusive(unsigned int what);
960 
961 void x86_del_exclusive(unsigned int what);
962 
963 int x86_reserve_hardware(void);
964 
965 void x86_release_hardware(void);
966 
967 int x86_pmu_max_precise(void);
968 
969 void hw_perf_lbr_event_destroy(struct perf_event *event);
970 
971 int x86_setup_perfctr(struct perf_event *event);
972 
973 int x86_pmu_hw_config(struct perf_event *event);
974 
975 void x86_pmu_disable_all(void);
976 
977 static inline bool is_counter_pair(struct hw_perf_event *hwc)
978 {
979 	return hwc->flags & PERF_X86_EVENT_PAIR;
980 }
981 
982 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
983 					  u64 enable_mask)
984 {
985 	u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
986 
987 	if (hwc->extra_reg.reg)
988 		wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
989 
990 	/*
991 	 * Add enabled Merge event on next counter
992 	 * if large increment event being enabled on this counter
993 	 */
994 	if (is_counter_pair(hwc))
995 		wrmsrl(x86_pmu_config_addr(hwc->idx + 1), x86_pmu.perf_ctr_pair_en);
996 
997 	wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
998 }
999 
1000 void x86_pmu_enable_all(int added);
1001 
1002 int perf_assign_events(struct event_constraint **constraints, int n,
1003 			int wmin, int wmax, int gpmax, int *assign);
1004 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
1005 
1006 void x86_pmu_stop(struct perf_event *event, int flags);
1007 
1008 static inline void x86_pmu_disable_event(struct perf_event *event)
1009 {
1010 	struct hw_perf_event *hwc = &event->hw;
1011 
1012 	wrmsrl(hwc->config_base, hwc->config);
1013 
1014 	if (is_counter_pair(hwc))
1015 		wrmsrl(x86_pmu_config_addr(hwc->idx + 1), 0);
1016 }
1017 
1018 void x86_pmu_enable_event(struct perf_event *event);
1019 
1020 int x86_pmu_handle_irq(struct pt_regs *regs);
1021 
1022 extern struct event_constraint emptyconstraint;
1023 
1024 extern struct event_constraint unconstrained;
1025 
1026 static inline bool kernel_ip(unsigned long ip)
1027 {
1028 #ifdef CONFIG_X86_32
1029 	return ip > PAGE_OFFSET;
1030 #else
1031 	return (long)ip < 0;
1032 #endif
1033 }
1034 
1035 /*
1036  * Not all PMUs provide the right context information to place the reported IP
1037  * into full context. Specifically segment registers are typically not
1038  * supplied.
1039  *
1040  * Assuming the address is a linear address (it is for IBS), we fake the CS and
1041  * vm86 mode using the known zero-based code segment and 'fix up' the registers
1042  * to reflect this.
1043  *
1044  * Intel PEBS/LBR appear to typically provide the effective address, nothing
1045  * much we can do about that but pray and treat it like a linear address.
1046  */
1047 static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
1048 {
1049 	regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
1050 	if (regs->flags & X86_VM_MASK)
1051 		regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
1052 	regs->ip = ip;
1053 }
1054 
1055 ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
1056 ssize_t intel_event_sysfs_show(char *page, u64 config);
1057 
1058 ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
1059 			  char *page);
1060 ssize_t events_ht_sysfs_show(struct device *dev, struct device_attribute *attr,
1061 			  char *page);
1062 
1063 #ifdef CONFIG_CPU_SUP_AMD
1064 
1065 int amd_pmu_init(void);
1066 
1067 #else /* CONFIG_CPU_SUP_AMD */
1068 
1069 static inline int amd_pmu_init(void)
1070 {
1071 	return 0;
1072 }
1073 
1074 #endif /* CONFIG_CPU_SUP_AMD */
1075 
1076 static inline int is_pebs_pt(struct perf_event *event)
1077 {
1078 	return !!(event->hw.flags & PERF_X86_EVENT_PEBS_VIA_PT);
1079 }
1080 
1081 #ifdef CONFIG_CPU_SUP_INTEL
1082 
1083 static inline bool intel_pmu_has_bts_period(struct perf_event *event, u64 period)
1084 {
1085 	struct hw_perf_event *hwc = &event->hw;
1086 	unsigned int hw_event, bts_event;
1087 
1088 	if (event->attr.freq)
1089 		return false;
1090 
1091 	hw_event = hwc->config & INTEL_ARCH_EVENT_MASK;
1092 	bts_event = x86_pmu.event_map(PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
1093 
1094 	return hw_event == bts_event && period == 1;
1095 }
1096 
1097 static inline bool intel_pmu_has_bts(struct perf_event *event)
1098 {
1099 	struct hw_perf_event *hwc = &event->hw;
1100 
1101 	return intel_pmu_has_bts_period(event, hwc->sample_period);
1102 }
1103 
1104 int intel_pmu_save_and_restart(struct perf_event *event);
1105 
1106 struct event_constraint *
1107 x86_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
1108 			  struct perf_event *event);
1109 
1110 extern int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu);
1111 extern void intel_cpuc_finish(struct cpu_hw_events *cpuc);
1112 
1113 int intel_pmu_init(void);
1114 
1115 void init_debug_store_on_cpu(int cpu);
1116 
1117 void fini_debug_store_on_cpu(int cpu);
1118 
1119 void release_ds_buffers(void);
1120 
1121 void reserve_ds_buffers(void);
1122 
1123 void release_lbr_buffers(void);
1124 
1125 extern struct event_constraint bts_constraint;
1126 extern struct event_constraint vlbr_constraint;
1127 
1128 void intel_pmu_enable_bts(u64 config);
1129 
1130 void intel_pmu_disable_bts(void);
1131 
1132 int intel_pmu_drain_bts_buffer(void);
1133 
1134 extern struct event_constraint intel_core2_pebs_event_constraints[];
1135 
1136 extern struct event_constraint intel_atom_pebs_event_constraints[];
1137 
1138 extern struct event_constraint intel_slm_pebs_event_constraints[];
1139 
1140 extern struct event_constraint intel_glm_pebs_event_constraints[];
1141 
1142 extern struct event_constraint intel_glp_pebs_event_constraints[];
1143 
1144 extern struct event_constraint intel_nehalem_pebs_event_constraints[];
1145 
1146 extern struct event_constraint intel_westmere_pebs_event_constraints[];
1147 
1148 extern struct event_constraint intel_snb_pebs_event_constraints[];
1149 
1150 extern struct event_constraint intel_ivb_pebs_event_constraints[];
1151 
1152 extern struct event_constraint intel_hsw_pebs_event_constraints[];
1153 
1154 extern struct event_constraint intel_bdw_pebs_event_constraints[];
1155 
1156 extern struct event_constraint intel_skl_pebs_event_constraints[];
1157 
1158 extern struct event_constraint intel_icl_pebs_event_constraints[];
1159 
1160 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
1161 
1162 void intel_pmu_pebs_add(struct perf_event *event);
1163 
1164 void intel_pmu_pebs_del(struct perf_event *event);
1165 
1166 void intel_pmu_pebs_enable(struct perf_event *event);
1167 
1168 void intel_pmu_pebs_disable(struct perf_event *event);
1169 
1170 void intel_pmu_pebs_enable_all(void);
1171 
1172 void intel_pmu_pebs_disable_all(void);
1173 
1174 void intel_pmu_pebs_sched_task(struct perf_event_context *ctx, bool sched_in);
1175 
1176 void intel_pmu_auto_reload_read(struct perf_event *event);
1177 
1178 void intel_pmu_store_pebs_lbrs(struct lbr_entry *lbr);
1179 
1180 void intel_ds_init(void);
1181 
1182 void intel_pmu_lbr_swap_task_ctx(struct perf_event_context *prev,
1183 				 struct perf_event_context *next);
1184 
1185 void intel_pmu_lbr_sched_task(struct perf_event_context *ctx, bool sched_in);
1186 
1187 u64 lbr_from_signext_quirk_wr(u64 val);
1188 
1189 void intel_pmu_lbr_reset(void);
1190 
1191 void intel_pmu_lbr_reset_32(void);
1192 
1193 void intel_pmu_lbr_reset_64(void);
1194 
1195 void intel_pmu_lbr_add(struct perf_event *event);
1196 
1197 void intel_pmu_lbr_del(struct perf_event *event);
1198 
1199 void intel_pmu_lbr_enable_all(bool pmi);
1200 
1201 void intel_pmu_lbr_disable_all(void);
1202 
1203 void intel_pmu_lbr_read(void);
1204 
1205 void intel_pmu_lbr_read_32(struct cpu_hw_events *cpuc);
1206 
1207 void intel_pmu_lbr_read_64(struct cpu_hw_events *cpuc);
1208 
1209 void intel_pmu_lbr_save(void *ctx);
1210 
1211 void intel_pmu_lbr_restore(void *ctx);
1212 
1213 void intel_pmu_lbr_init_core(void);
1214 
1215 void intel_pmu_lbr_init_nhm(void);
1216 
1217 void intel_pmu_lbr_init_atom(void);
1218 
1219 void intel_pmu_lbr_init_slm(void);
1220 
1221 void intel_pmu_lbr_init_snb(void);
1222 
1223 void intel_pmu_lbr_init_hsw(void);
1224 
1225 void intel_pmu_lbr_init_skl(void);
1226 
1227 void intel_pmu_lbr_init_knl(void);
1228 
1229 void intel_pmu_arch_lbr_init(void);
1230 
1231 void intel_pmu_pebs_data_source_nhm(void);
1232 
1233 void intel_pmu_pebs_data_source_skl(bool pmem);
1234 
1235 int intel_pmu_setup_lbr_filter(struct perf_event *event);
1236 
1237 void intel_pt_interrupt(void);
1238 
1239 int intel_bts_interrupt(void);
1240 
1241 void intel_bts_enable_local(void);
1242 
1243 void intel_bts_disable_local(void);
1244 
1245 int p4_pmu_init(void);
1246 
1247 int p6_pmu_init(void);
1248 
1249 int knc_pmu_init(void);
1250 
1251 static inline int is_ht_workaround_enabled(void)
1252 {
1253 	return !!(x86_pmu.flags & PMU_FL_EXCL_ENABLED);
1254 }
1255 
1256 #else /* CONFIG_CPU_SUP_INTEL */
1257 
1258 static inline void reserve_ds_buffers(void)
1259 {
1260 }
1261 
1262 static inline void release_ds_buffers(void)
1263 {
1264 }
1265 
1266 static inline void release_lbr_buffers(void)
1267 {
1268 }
1269 
1270 static inline int intel_pmu_init(void)
1271 {
1272 	return 0;
1273 }
1274 
1275 static inline int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu)
1276 {
1277 	return 0;
1278 }
1279 
1280 static inline void intel_cpuc_finish(struct cpu_hw_events *cpuc)
1281 {
1282 }
1283 
1284 static inline int is_ht_workaround_enabled(void)
1285 {
1286 	return 0;
1287 }
1288 #endif /* CONFIG_CPU_SUP_INTEL */
1289 
1290 #if ((defined CONFIG_CPU_SUP_CENTAUR) || (defined CONFIG_CPU_SUP_ZHAOXIN))
1291 int zhaoxin_pmu_init(void);
1292 #else
1293 static inline int zhaoxin_pmu_init(void)
1294 {
1295 	return 0;
1296 }
1297 #endif /*CONFIG_CPU_SUP_CENTAUR or CONFIG_CPU_SUP_ZHAOXIN*/
1298