xref: /openbmc/linux/arch/x86/events/perf_event.h (revision 5085e036)
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 
46 	EXTRA_REG_MAX		/* number of entries needed */
47 };
48 
49 struct event_constraint {
50 	union {
51 		unsigned long	idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
52 		u64		idxmsk64;
53 	};
54 	u64		code;
55 	u64		cmask;
56 	int		weight;
57 	int		overlap;
58 	int		flags;
59 	unsigned int	size;
60 };
61 
62 static inline bool constraint_match(struct event_constraint *c, u64 ecode)
63 {
64 	return ((ecode & c->cmask) - c->code) <= (u64)c->size;
65 }
66 
67 /*
68  * struct hw_perf_event.flags flags
69  */
70 #define PERF_X86_EVENT_PEBS_LDLAT	0x00001 /* ld+ldlat data address sampling */
71 #define PERF_X86_EVENT_PEBS_ST		0x00002 /* st data address sampling */
72 #define PERF_X86_EVENT_PEBS_ST_HSW	0x00004 /* haswell style datala, store */
73 #define PERF_X86_EVENT_PEBS_LD_HSW	0x00008 /* haswell style datala, load */
74 #define PERF_X86_EVENT_PEBS_NA_HSW	0x00010 /* haswell style datala, unknown */
75 #define PERF_X86_EVENT_EXCL		0x00020 /* HT exclusivity on counter */
76 #define PERF_X86_EVENT_DYNAMIC		0x00040 /* dynamic alloc'd constraint */
77 
78 #define PERF_X86_EVENT_EXCL_ACCT	0x00100 /* accounted EXCL event */
79 #define PERF_X86_EVENT_AUTO_RELOAD	0x00200 /* use PEBS auto-reload */
80 #define PERF_X86_EVENT_LARGE_PEBS	0x00400 /* use large PEBS */
81 #define PERF_X86_EVENT_PEBS_VIA_PT	0x00800 /* use PT buffer for PEBS */
82 #define PERF_X86_EVENT_PAIR		0x01000 /* Large Increment per Cycle */
83 #define PERF_X86_EVENT_LBR_SELECT	0x02000 /* Save/Restore MSR_LBR_SELECT */
84 #define PERF_X86_EVENT_TOPDOWN		0x04000 /* Count Topdown slots/metrics events */
85 #define PERF_X86_EVENT_PEBS_STLAT	0x08000 /* st+stlat data address sampling */
86 #define PERF_X86_EVENT_AMD_BRS		0x10000 /* AMD Branch Sampling */
87 
88 static inline bool is_topdown_count(struct perf_event *event)
89 {
90 	return event->hw.flags & PERF_X86_EVENT_TOPDOWN;
91 }
92 
93 static inline bool is_metric_event(struct perf_event *event)
94 {
95 	u64 config = event->attr.config;
96 
97 	return ((config & ARCH_PERFMON_EVENTSEL_EVENT) == 0) &&
98 		((config & INTEL_ARCH_EVENT_MASK) >= INTEL_TD_METRIC_RETIRING)  &&
99 		((config & INTEL_ARCH_EVENT_MASK) <= INTEL_TD_METRIC_MAX);
100 }
101 
102 static inline bool is_slots_event(struct perf_event *event)
103 {
104 	return (event->attr.config & INTEL_ARCH_EVENT_MASK) == INTEL_TD_SLOTS;
105 }
106 
107 static inline bool is_topdown_event(struct perf_event *event)
108 {
109 	return is_metric_event(event) || is_slots_event(event);
110 }
111 
112 struct amd_nb {
113 	int nb_id;  /* NorthBridge id */
114 	int refcnt; /* reference count */
115 	struct perf_event *owners[X86_PMC_IDX_MAX];
116 	struct event_constraint event_constraints[X86_PMC_IDX_MAX];
117 };
118 
119 #define PEBS_COUNTER_MASK	((1ULL << MAX_PEBS_EVENTS) - 1)
120 #define PEBS_PMI_AFTER_EACH_RECORD BIT_ULL(60)
121 #define PEBS_OUTPUT_OFFSET	61
122 #define PEBS_OUTPUT_MASK	(3ull << PEBS_OUTPUT_OFFSET)
123 #define PEBS_OUTPUT_PT		(1ull << PEBS_OUTPUT_OFFSET)
124 #define PEBS_VIA_PT_MASK	(PEBS_OUTPUT_PT | PEBS_PMI_AFTER_EACH_RECORD)
125 
126 /*
127  * Flags PEBS can handle without an PMI.
128  *
129  * TID can only be handled by flushing at context switch.
130  * REGS_USER can be handled for events limited to ring 3.
131  *
132  */
133 #define LARGE_PEBS_FLAGS \
134 	(PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_ADDR | \
135 	PERF_SAMPLE_ID | PERF_SAMPLE_CPU | PERF_SAMPLE_STREAM_ID | \
136 	PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
137 	PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR | \
138 	PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER | \
139 	PERF_SAMPLE_PERIOD | PERF_SAMPLE_CODE_PAGE_SIZE)
140 
141 #define PEBS_GP_REGS			\
142 	((1ULL << PERF_REG_X86_AX)    | \
143 	 (1ULL << PERF_REG_X86_BX)    | \
144 	 (1ULL << PERF_REG_X86_CX)    | \
145 	 (1ULL << PERF_REG_X86_DX)    | \
146 	 (1ULL << PERF_REG_X86_DI)    | \
147 	 (1ULL << PERF_REG_X86_SI)    | \
148 	 (1ULL << PERF_REG_X86_SP)    | \
149 	 (1ULL << PERF_REG_X86_BP)    | \
150 	 (1ULL << PERF_REG_X86_IP)    | \
151 	 (1ULL << PERF_REG_X86_FLAGS) | \
152 	 (1ULL << PERF_REG_X86_R8)    | \
153 	 (1ULL << PERF_REG_X86_R9)    | \
154 	 (1ULL << PERF_REG_X86_R10)   | \
155 	 (1ULL << PERF_REG_X86_R11)   | \
156 	 (1ULL << PERF_REG_X86_R12)   | \
157 	 (1ULL << PERF_REG_X86_R13)   | \
158 	 (1ULL << PERF_REG_X86_R14)   | \
159 	 (1ULL << PERF_REG_X86_R15))
160 
161 /*
162  * Per register state.
163  */
164 struct er_account {
165 	raw_spinlock_t      lock;	/* per-core: protect structure */
166 	u64                 config;	/* extra MSR config */
167 	u64                 reg;	/* extra MSR number */
168 	atomic_t            ref;	/* reference count */
169 };
170 
171 /*
172  * Per core/cpu state
173  *
174  * Used to coordinate shared registers between HT threads or
175  * among events on a single PMU.
176  */
177 struct intel_shared_regs {
178 	struct er_account       regs[EXTRA_REG_MAX];
179 	int                     refcnt;		/* per-core: #HT threads */
180 	unsigned                core_id;	/* per-core: core id */
181 };
182 
183 enum intel_excl_state_type {
184 	INTEL_EXCL_UNUSED    = 0, /* counter is unused */
185 	INTEL_EXCL_SHARED    = 1, /* counter can be used by both threads */
186 	INTEL_EXCL_EXCLUSIVE = 2, /* counter can be used by one thread only */
187 };
188 
189 struct intel_excl_states {
190 	enum intel_excl_state_type state[X86_PMC_IDX_MAX];
191 	bool sched_started; /* true if scheduling has started */
192 };
193 
194 struct intel_excl_cntrs {
195 	raw_spinlock_t	lock;
196 
197 	struct intel_excl_states states[2];
198 
199 	union {
200 		u16	has_exclusive[2];
201 		u32	exclusive_present;
202 	};
203 
204 	int		refcnt;		/* per-core: #HT threads */
205 	unsigned	core_id;	/* per-core: core id */
206 };
207 
208 struct x86_perf_task_context;
209 #define MAX_LBR_ENTRIES		32
210 
211 enum {
212 	LBR_FORMAT_32		= 0x00,
213 	LBR_FORMAT_LIP		= 0x01,
214 	LBR_FORMAT_EIP		= 0x02,
215 	LBR_FORMAT_EIP_FLAGS	= 0x03,
216 	LBR_FORMAT_EIP_FLAGS2	= 0x04,
217 	LBR_FORMAT_INFO		= 0x05,
218 	LBR_FORMAT_TIME		= 0x06,
219 	LBR_FORMAT_INFO2	= 0x07,
220 	LBR_FORMAT_MAX_KNOWN    = LBR_FORMAT_INFO2,
221 };
222 
223 enum {
224 	X86_PERF_KFREE_SHARED = 0,
225 	X86_PERF_KFREE_EXCL   = 1,
226 	X86_PERF_KFREE_MAX
227 };
228 
229 struct cpu_hw_events {
230 	/*
231 	 * Generic x86 PMC bits
232 	 */
233 	struct perf_event	*events[X86_PMC_IDX_MAX]; /* in counter order */
234 	unsigned long		active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
235 	unsigned long		dirty[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
236 	int			enabled;
237 
238 	int			n_events; /* the # of events in the below arrays */
239 	int			n_added;  /* the # last events in the below arrays;
240 					     they've never been enabled yet */
241 	int			n_txn;    /* the # last events in the below arrays;
242 					     added in the current transaction */
243 	int			n_txn_pair;
244 	int			n_txn_metric;
245 	int			assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
246 	u64			tags[X86_PMC_IDX_MAX];
247 
248 	struct perf_event	*event_list[X86_PMC_IDX_MAX]; /* in enabled order */
249 	struct event_constraint	*event_constraint[X86_PMC_IDX_MAX];
250 
251 	int			n_excl; /* the number of exclusive events */
252 
253 	unsigned int		txn_flags;
254 	int			is_fake;
255 
256 	/*
257 	 * Intel DebugStore bits
258 	 */
259 	struct debug_store	*ds;
260 	void			*ds_pebs_vaddr;
261 	void			*ds_bts_vaddr;
262 	u64			pebs_enabled;
263 	int			n_pebs;
264 	int			n_large_pebs;
265 	int			n_pebs_via_pt;
266 	int			pebs_output;
267 
268 	/* Current super set of events hardware configuration */
269 	u64			pebs_data_cfg;
270 	u64			active_pebs_data_cfg;
271 	int			pebs_record_size;
272 
273 	/*
274 	 * Intel LBR bits
275 	 */
276 	int				lbr_users;
277 	int				lbr_pebs_users;
278 	struct perf_branch_stack	lbr_stack;
279 	struct perf_branch_entry	lbr_entries[MAX_LBR_ENTRIES];
280 	union {
281 		struct er_account		*lbr_sel;
282 		struct er_account		*lbr_ctl;
283 	};
284 	u64				br_sel;
285 	void				*last_task_ctx;
286 	int				last_log_id;
287 	int				lbr_select;
288 	void				*lbr_xsave;
289 
290 	/*
291 	 * Intel host/guest exclude bits
292 	 */
293 	u64				intel_ctrl_guest_mask;
294 	u64				intel_ctrl_host_mask;
295 	struct perf_guest_switch_msr	guest_switch_msrs[X86_PMC_IDX_MAX];
296 
297 	/*
298 	 * Intel checkpoint mask
299 	 */
300 	u64				intel_cp_status;
301 
302 	/*
303 	 * manage shared (per-core, per-cpu) registers
304 	 * used on Intel NHM/WSM/SNB
305 	 */
306 	struct intel_shared_regs	*shared_regs;
307 	/*
308 	 * manage exclusive counter access between hyperthread
309 	 */
310 	struct event_constraint *constraint_list; /* in enable order */
311 	struct intel_excl_cntrs		*excl_cntrs;
312 	int excl_thread_id; /* 0 or 1 */
313 
314 	/*
315 	 * SKL TSX_FORCE_ABORT shadow
316 	 */
317 	u64				tfa_shadow;
318 
319 	/*
320 	 * Perf Metrics
321 	 */
322 	/* number of accepted metrics events */
323 	int				n_metric;
324 
325 	/*
326 	 * AMD specific bits
327 	 */
328 	struct amd_nb			*amd_nb;
329 	int				brs_active; /* BRS is enabled */
330 
331 	/* Inverted mask of bits to clear in the perf_ctr ctrl registers */
332 	u64				perf_ctr_virt_mask;
333 	int				n_pair; /* Large increment events */
334 
335 	void				*kfree_on_online[X86_PERF_KFREE_MAX];
336 
337 	struct pmu			*pmu;
338 };
339 
340 #define __EVENT_CONSTRAINT_RANGE(c, e, n, m, w, o, f) {	\
341 	{ .idxmsk64 = (n) },		\
342 	.code = (c),			\
343 	.size = (e) - (c),		\
344 	.cmask = (m),			\
345 	.weight = (w),			\
346 	.overlap = (o),			\
347 	.flags = f,			\
348 }
349 
350 #define __EVENT_CONSTRAINT(c, n, m, w, o, f) \
351 	__EVENT_CONSTRAINT_RANGE(c, c, n, m, w, o, f)
352 
353 #define EVENT_CONSTRAINT(c, n, m)	\
354 	__EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0)
355 
356 /*
357  * The constraint_match() function only works for 'simple' event codes
358  * and not for extended (AMD64_EVENTSEL_EVENT) events codes.
359  */
360 #define EVENT_CONSTRAINT_RANGE(c, e, n, m) \
361 	__EVENT_CONSTRAINT_RANGE(c, e, n, m, HWEIGHT(n), 0, 0)
362 
363 #define INTEL_EXCLEVT_CONSTRAINT(c, n)	\
364 	__EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT, HWEIGHT(n),\
365 			   0, PERF_X86_EVENT_EXCL)
366 
367 /*
368  * The overlap flag marks event constraints with overlapping counter
369  * masks. This is the case if the counter mask of such an event is not
370  * a subset of any other counter mask of a constraint with an equal or
371  * higher weight, e.g.:
372  *
373  *  c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
374  *  c_another1 = EVENT_CONSTRAINT(0, 0x07, 0);
375  *  c_another2 = EVENT_CONSTRAINT(0, 0x38, 0);
376  *
377  * The event scheduler may not select the correct counter in the first
378  * cycle because it needs to know which subsequent events will be
379  * scheduled. It may fail to schedule the events then. So we set the
380  * overlap flag for such constraints to give the scheduler a hint which
381  * events to select for counter rescheduling.
382  *
383  * Care must be taken as the rescheduling algorithm is O(n!) which
384  * will increase scheduling cycles for an over-committed system
385  * dramatically.  The number of such EVENT_CONSTRAINT_OVERLAP() macros
386  * and its counter masks must be kept at a minimum.
387  */
388 #define EVENT_CONSTRAINT_OVERLAP(c, n, m)	\
389 	__EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0)
390 
391 /*
392  * Constraint on the Event code.
393  */
394 #define INTEL_EVENT_CONSTRAINT(c, n)	\
395 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
396 
397 /*
398  * Constraint on a range of Event codes
399  */
400 #define INTEL_EVENT_CONSTRAINT_RANGE(c, e, n)			\
401 	EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT)
402 
403 /*
404  * Constraint on the Event code + UMask + fixed-mask
405  *
406  * filter mask to validate fixed counter events.
407  * the following filters disqualify for fixed counters:
408  *  - inv
409  *  - edge
410  *  - cnt-mask
411  *  - in_tx
412  *  - in_tx_checkpointed
413  *  The other filters are supported by fixed counters.
414  *  The any-thread option is supported starting with v3.
415  */
416 #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED)
417 #define FIXED_EVENT_CONSTRAINT(c, n)	\
418 	EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
419 
420 /*
421  * The special metric counters do not actually exist. They are calculated from
422  * the combination of the FxCtr3 + MSR_PERF_METRICS.
423  *
424  * The special metric counters are mapped to a dummy offset for the scheduler.
425  * The sharing between multiple users of the same metric without multiplexing
426  * is not allowed, even though the hardware supports that in principle.
427  */
428 
429 #define METRIC_EVENT_CONSTRAINT(c, n)					\
430 	EVENT_CONSTRAINT(c, (1ULL << (INTEL_PMC_IDX_METRIC_BASE + n)),	\
431 			 INTEL_ARCH_EVENT_MASK)
432 
433 /*
434  * Constraint on the Event code + UMask
435  */
436 #define INTEL_UEVENT_CONSTRAINT(c, n)	\
437 	EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
438 
439 /* Constraint on specific umask bit only + event */
440 #define INTEL_UBIT_EVENT_CONSTRAINT(c, n)	\
441 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|(c))
442 
443 /* Like UEVENT_CONSTRAINT, but match flags too */
444 #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n)	\
445 	EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS)
446 
447 #define INTEL_EXCLUEVT_CONSTRAINT(c, n)	\
448 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK, \
449 			   HWEIGHT(n), 0, PERF_X86_EVENT_EXCL)
450 
451 #define INTEL_PLD_CONSTRAINT(c, n)	\
452 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
453 			   HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT)
454 
455 #define INTEL_PSD_CONSTRAINT(c, n)	\
456 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
457 			   HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_STLAT)
458 
459 #define INTEL_PST_CONSTRAINT(c, n)	\
460 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
461 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST)
462 
463 /* Event constraint, but match on all event flags too. */
464 #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \
465 	EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
466 
467 #define INTEL_FLAGS_EVENT_CONSTRAINT_RANGE(c, e, n)			\
468 	EVENT_CONSTRAINT_RANGE(c, e, n, ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS)
469 
470 /* Check only flags, but allow all event/umask */
471 #define INTEL_ALL_EVENT_CONSTRAINT(code, n)	\
472 	EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS)
473 
474 /* Check flags and event code, and set the HSW store flag */
475 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \
476 	__EVENT_CONSTRAINT(code, n, 			\
477 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
478 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
479 
480 /* Check flags and event code, and set the HSW load flag */
481 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \
482 	__EVENT_CONSTRAINT(code, n,			\
483 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
484 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
485 
486 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD_RANGE(code, end, n) \
487 	__EVENT_CONSTRAINT_RANGE(code, end, n,				\
488 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
489 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
490 
491 #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_XLD(code, n) \
492 	__EVENT_CONSTRAINT(code, n,			\
493 			  ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \
494 			  HWEIGHT(n), 0, \
495 			  PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
496 
497 /* Check flags and event code/umask, and set the HSW store flag */
498 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \
499 	__EVENT_CONSTRAINT(code, n, 			\
500 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
501 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
502 
503 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XST(code, n) \
504 	__EVENT_CONSTRAINT(code, n,			\
505 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
506 			  HWEIGHT(n), 0, \
507 			  PERF_X86_EVENT_PEBS_ST_HSW|PERF_X86_EVENT_EXCL)
508 
509 /* Check flags and event code/umask, and set the HSW load flag */
510 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \
511 	__EVENT_CONSTRAINT(code, n, 			\
512 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
513 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW)
514 
515 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_XLD(code, n) \
516 	__EVENT_CONSTRAINT(code, n,			\
517 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
518 			  HWEIGHT(n), 0, \
519 			  PERF_X86_EVENT_PEBS_LD_HSW|PERF_X86_EVENT_EXCL)
520 
521 /* Check flags and event code/umask, and set the HSW N/A flag */
522 #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \
523 	__EVENT_CONSTRAINT(code, n, 			\
524 			  INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \
525 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW)
526 
527 
528 /*
529  * We define the end marker as having a weight of -1
530  * to enable blacklisting of events using a counter bitmask
531  * of zero and thus a weight of zero.
532  * The end marker has a weight that cannot possibly be
533  * obtained from counting the bits in the bitmask.
534  */
535 #define EVENT_CONSTRAINT_END { .weight = -1 }
536 
537 /*
538  * Check for end marker with weight == -1
539  */
540 #define for_each_event_constraint(e, c)	\
541 	for ((e) = (c); (e)->weight != -1; (e)++)
542 
543 /*
544  * Extra registers for specific events.
545  *
546  * Some events need large masks and require external MSRs.
547  * Those extra MSRs end up being shared for all events on
548  * a PMU and sometimes between PMU of sibling HT threads.
549  * In either case, the kernel needs to handle conflicting
550  * accesses to those extra, shared, regs. The data structure
551  * to manage those registers is stored in cpu_hw_event.
552  */
553 struct extra_reg {
554 	unsigned int		event;
555 	unsigned int		msr;
556 	u64			config_mask;
557 	u64			valid_mask;
558 	int			idx;  /* per_xxx->regs[] reg index */
559 	bool			extra_msr_access;
560 };
561 
562 #define EVENT_EXTRA_REG(e, ms, m, vm, i) {	\
563 	.event = (e),			\
564 	.msr = (ms),			\
565 	.config_mask = (m),		\
566 	.valid_mask = (vm),		\
567 	.idx = EXTRA_REG_##i,		\
568 	.extra_msr_access = true,	\
569 	}
570 
571 #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx)	\
572 	EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
573 
574 #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \
575 	EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \
576 			ARCH_PERFMON_EVENTSEL_UMASK, vm, idx)
577 
578 #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \
579 	INTEL_UEVENT_EXTRA_REG(c, \
580 			       MSR_PEBS_LD_LAT_THRESHOLD, \
581 			       0xffff, \
582 			       LDLAT)
583 
584 #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
585 
586 union perf_capabilities {
587 	struct {
588 		u64	lbr_format:6;
589 		u64	pebs_trap:1;
590 		u64	pebs_arch_reg:1;
591 		u64	pebs_format:4;
592 		u64	smm_freeze:1;
593 		/*
594 		 * PMU supports separate counter range for writing
595 		 * values > 32bit.
596 		 */
597 		u64	full_width_write:1;
598 		u64     pebs_baseline:1;
599 		u64	perf_metrics:1;
600 		u64	pebs_output_pt_available:1;
601 		u64	anythread_deprecated:1;
602 	};
603 	u64	capabilities;
604 };
605 
606 struct x86_pmu_quirk {
607 	struct x86_pmu_quirk *next;
608 	void (*func)(void);
609 };
610 
611 union x86_pmu_config {
612 	struct {
613 		u64 event:8,
614 		    umask:8,
615 		    usr:1,
616 		    os:1,
617 		    edge:1,
618 		    pc:1,
619 		    interrupt:1,
620 		    __reserved1:1,
621 		    en:1,
622 		    inv:1,
623 		    cmask:8,
624 		    event2:4,
625 		    __reserved2:4,
626 		    go:1,
627 		    ho:1;
628 	} bits;
629 	u64 value;
630 };
631 
632 #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value
633 
634 enum {
635 	x86_lbr_exclusive_lbr,
636 	x86_lbr_exclusive_bts,
637 	x86_lbr_exclusive_pt,
638 	x86_lbr_exclusive_max,
639 };
640 
641 struct x86_hybrid_pmu {
642 	struct pmu			pmu;
643 	const char			*name;
644 	u8				cpu_type;
645 	cpumask_t			supported_cpus;
646 	union perf_capabilities		intel_cap;
647 	u64				intel_ctrl;
648 	int				max_pebs_events;
649 	int				num_counters;
650 	int				num_counters_fixed;
651 	struct event_constraint		unconstrained;
652 
653 	u64				hw_cache_event_ids
654 					[PERF_COUNT_HW_CACHE_MAX]
655 					[PERF_COUNT_HW_CACHE_OP_MAX]
656 					[PERF_COUNT_HW_CACHE_RESULT_MAX];
657 	u64				hw_cache_extra_regs
658 					[PERF_COUNT_HW_CACHE_MAX]
659 					[PERF_COUNT_HW_CACHE_OP_MAX]
660 					[PERF_COUNT_HW_CACHE_RESULT_MAX];
661 	struct event_constraint		*event_constraints;
662 	struct event_constraint		*pebs_constraints;
663 	struct extra_reg		*extra_regs;
664 
665 	unsigned int			late_ack	:1,
666 					mid_ack		:1,
667 					enabled_ack	:1;
668 };
669 
670 static __always_inline struct x86_hybrid_pmu *hybrid_pmu(struct pmu *pmu)
671 {
672 	return container_of(pmu, struct x86_hybrid_pmu, pmu);
673 }
674 
675 extern struct static_key_false perf_is_hybrid;
676 #define is_hybrid()		static_branch_unlikely(&perf_is_hybrid)
677 
678 #define hybrid(_pmu, _field)				\
679 (*({							\
680 	typeof(&x86_pmu._field) __Fp = &x86_pmu._field;	\
681 							\
682 	if (is_hybrid() && (_pmu))			\
683 		__Fp = &hybrid_pmu(_pmu)->_field;	\
684 							\
685 	__Fp;						\
686 }))
687 
688 #define hybrid_var(_pmu, _var)				\
689 (*({							\
690 	typeof(&_var) __Fp = &_var;			\
691 							\
692 	if (is_hybrid() && (_pmu))			\
693 		__Fp = &hybrid_pmu(_pmu)->_var;		\
694 							\
695 	__Fp;						\
696 }))
697 
698 #define hybrid_bit(_pmu, _field)			\
699 ({							\
700 	bool __Fp = x86_pmu._field;			\
701 							\
702 	if (is_hybrid() && (_pmu))			\
703 		__Fp = hybrid_pmu(_pmu)->_field;	\
704 							\
705 	__Fp;						\
706 })
707 
708 enum hybrid_pmu_type {
709 	hybrid_big		= 0x40,
710 	hybrid_small		= 0x20,
711 
712 	hybrid_big_small	= hybrid_big | hybrid_small,
713 };
714 
715 #define X86_HYBRID_PMU_ATOM_IDX		0
716 #define X86_HYBRID_PMU_CORE_IDX		1
717 
718 #define X86_HYBRID_NUM_PMUS		2
719 
720 /*
721  * struct x86_pmu - generic x86 pmu
722  */
723 struct x86_pmu {
724 	/*
725 	 * Generic x86 PMC bits
726 	 */
727 	const char	*name;
728 	int		version;
729 	int		(*handle_irq)(struct pt_regs *);
730 	void		(*disable_all)(void);
731 	void		(*enable_all)(int added);
732 	void		(*enable)(struct perf_event *);
733 	void		(*disable)(struct perf_event *);
734 	void		(*assign)(struct perf_event *event, int idx);
735 	void		(*add)(struct perf_event *);
736 	void		(*del)(struct perf_event *);
737 	void		(*read)(struct perf_event *event);
738 	int		(*hw_config)(struct perf_event *event);
739 	int		(*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
740 	unsigned	eventsel;
741 	unsigned	perfctr;
742 	int		(*addr_offset)(int index, bool eventsel);
743 	int		(*rdpmc_index)(int index);
744 	u64		(*event_map)(int);
745 	int		max_events;
746 	int		num_counters;
747 	int		num_counters_fixed;
748 	int		cntval_bits;
749 	u64		cntval_mask;
750 	union {
751 			unsigned long events_maskl;
752 			unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)];
753 	};
754 	int		events_mask_len;
755 	int		apic;
756 	u64		max_period;
757 	struct event_constraint *
758 			(*get_event_constraints)(struct cpu_hw_events *cpuc,
759 						 int idx,
760 						 struct perf_event *event);
761 
762 	void		(*put_event_constraints)(struct cpu_hw_events *cpuc,
763 						 struct perf_event *event);
764 
765 	void		(*start_scheduling)(struct cpu_hw_events *cpuc);
766 
767 	void		(*commit_scheduling)(struct cpu_hw_events *cpuc, int idx, int cntr);
768 
769 	void		(*stop_scheduling)(struct cpu_hw_events *cpuc);
770 
771 	struct event_constraint *event_constraints;
772 	struct x86_pmu_quirk *quirks;
773 	int		perfctr_second_write;
774 	u64		(*limit_period)(struct perf_event *event, u64 l);
775 
776 	/* PMI handler bits */
777 	unsigned int	late_ack		:1,
778 			mid_ack			:1,
779 			enabled_ack		:1;
780 	/*
781 	 * sysfs attrs
782 	 */
783 	int		attr_rdpmc_broken;
784 	int		attr_rdpmc;
785 	struct attribute **format_attrs;
786 
787 	ssize_t		(*events_sysfs_show)(char *page, u64 config);
788 	const struct attribute_group **attr_update;
789 
790 	unsigned long	attr_freeze_on_smi;
791 
792 	/*
793 	 * CPU Hotplug hooks
794 	 */
795 	int		(*cpu_prepare)(int cpu);
796 	void		(*cpu_starting)(int cpu);
797 	void		(*cpu_dying)(int cpu);
798 	void		(*cpu_dead)(int cpu);
799 
800 	void		(*check_microcode)(void);
801 	void		(*sched_task)(struct perf_event_context *ctx,
802 				      bool sched_in);
803 
804 	/*
805 	 * Intel Arch Perfmon v2+
806 	 */
807 	u64			intel_ctrl;
808 	union perf_capabilities intel_cap;
809 
810 	/*
811 	 * Intel DebugStore bits
812 	 */
813 	unsigned int	bts			:1,
814 			bts_active		:1,
815 			pebs			:1,
816 			pebs_active		:1,
817 			pebs_broken		:1,
818 			pebs_prec_dist		:1,
819 			pebs_no_tlb		:1,
820 			pebs_no_isolation	:1,
821 			pebs_block		:1;
822 	int		pebs_record_size;
823 	int		pebs_buffer_size;
824 	int		max_pebs_events;
825 	void		(*drain_pebs)(struct pt_regs *regs, struct perf_sample_data *data);
826 	struct event_constraint *pebs_constraints;
827 	void		(*pebs_aliases)(struct perf_event *event);
828 	unsigned long	large_pebs_flags;
829 	u64		rtm_abort_event;
830 
831 	/*
832 	 * Intel LBR
833 	 */
834 	unsigned int	lbr_tos, lbr_from, lbr_to,
835 			lbr_info, lbr_nr;	   /* LBR base regs and size */
836 	union {
837 		u64	lbr_sel_mask;		   /* LBR_SELECT valid bits */
838 		u64	lbr_ctl_mask;		   /* LBR_CTL valid bits */
839 	};
840 	union {
841 		const int	*lbr_sel_map;	   /* lbr_select mappings */
842 		int		*lbr_ctl_map;	   /* LBR_CTL mappings */
843 	};
844 	bool		lbr_double_abort;	   /* duplicated lbr aborts */
845 	bool		lbr_pt_coexist;		   /* (LBR|BTS) may coexist with PT */
846 
847 	unsigned int	lbr_has_info:1;
848 	unsigned int	lbr_has_tsx:1;
849 	unsigned int	lbr_from_flags:1;
850 	unsigned int	lbr_to_cycles:1;
851 
852 	/*
853 	 * Intel Architectural LBR CPUID Enumeration
854 	 */
855 	unsigned int	lbr_depth_mask:8;
856 	unsigned int	lbr_deep_c_reset:1;
857 	unsigned int	lbr_lip:1;
858 	unsigned int	lbr_cpl:1;
859 	unsigned int	lbr_filter:1;
860 	unsigned int	lbr_call_stack:1;
861 	unsigned int	lbr_mispred:1;
862 	unsigned int	lbr_timed_lbr:1;
863 	unsigned int	lbr_br_type:1;
864 
865 	void		(*lbr_reset)(void);
866 	void		(*lbr_read)(struct cpu_hw_events *cpuc);
867 	void		(*lbr_save)(void *ctx);
868 	void		(*lbr_restore)(void *ctx);
869 
870 	/*
871 	 * Intel PT/LBR/BTS are exclusive
872 	 */
873 	atomic_t	lbr_exclusive[x86_lbr_exclusive_max];
874 
875 	/*
876 	 * Intel perf metrics
877 	 */
878 	int		num_topdown_events;
879 	u64		(*update_topdown_event)(struct perf_event *event);
880 	int		(*set_topdown_event_period)(struct perf_event *event);
881 
882 	/*
883 	 * perf task context (i.e. struct perf_event_context::task_ctx_data)
884 	 * switch helper to bridge calls from perf/core to perf/x86.
885 	 * See struct pmu::swap_task_ctx() usage for examples;
886 	 */
887 	void		(*swap_task_ctx)(struct perf_event_context *prev,
888 					 struct perf_event_context *next);
889 
890 	/*
891 	 * AMD bits
892 	 */
893 	unsigned int	amd_nb_constraints : 1;
894 	u64		perf_ctr_pair_en;
895 
896 	/*
897 	 * Extra registers for events
898 	 */
899 	struct extra_reg *extra_regs;
900 	unsigned int flags;
901 
902 	/*
903 	 * Intel host/guest support (KVM)
904 	 */
905 	struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr);
906 
907 	/*
908 	 * Check period value for PERF_EVENT_IOC_PERIOD ioctl.
909 	 */
910 	int (*check_period) (struct perf_event *event, u64 period);
911 
912 	int (*aux_output_match) (struct perf_event *event);
913 
914 	int (*filter_match)(struct perf_event *event);
915 	/*
916 	 * Hybrid support
917 	 *
918 	 * Most PMU capabilities are the same among different hybrid PMUs.
919 	 * The global x86_pmu saves the architecture capabilities, which
920 	 * are available for all PMUs. The hybrid_pmu only includes the
921 	 * unique capabilities.
922 	 */
923 	int				num_hybrid_pmus;
924 	struct x86_hybrid_pmu		*hybrid_pmu;
925 	u8 (*get_hybrid_cpu_type)	(void);
926 };
927 
928 struct x86_perf_task_context_opt {
929 	int lbr_callstack_users;
930 	int lbr_stack_state;
931 	int log_id;
932 };
933 
934 struct x86_perf_task_context {
935 	u64 lbr_sel;
936 	int tos;
937 	int valid_lbrs;
938 	struct x86_perf_task_context_opt opt;
939 	struct lbr_entry lbr[MAX_LBR_ENTRIES];
940 };
941 
942 struct x86_perf_task_context_arch_lbr {
943 	struct x86_perf_task_context_opt opt;
944 	struct lbr_entry entries[];
945 };
946 
947 /*
948  * Add padding to guarantee the 64-byte alignment of the state buffer.
949  *
950  * The structure is dynamically allocated. The size of the LBR state may vary
951  * based on the number of LBR registers.
952  *
953  * Do not put anything after the LBR state.
954  */
955 struct x86_perf_task_context_arch_lbr_xsave {
956 	struct x86_perf_task_context_opt		opt;
957 
958 	union {
959 		struct xregs_state			xsave;
960 		struct {
961 			struct fxregs_state		i387;
962 			struct xstate_header		header;
963 			struct arch_lbr_state		lbr;
964 		} __attribute__ ((packed, aligned (XSAVE_ALIGNMENT)));
965 	};
966 };
967 
968 #define x86_add_quirk(func_)						\
969 do {									\
970 	static struct x86_pmu_quirk __quirk __initdata = {		\
971 		.func = func_,						\
972 	};								\
973 	__quirk.next = x86_pmu.quirks;					\
974 	x86_pmu.quirks = &__quirk;					\
975 } while (0)
976 
977 /*
978  * x86_pmu flags
979  */
980 #define PMU_FL_NO_HT_SHARING	0x1 /* no hyper-threading resource sharing */
981 #define PMU_FL_HAS_RSP_1	0x2 /* has 2 equivalent offcore_rsp regs   */
982 #define PMU_FL_EXCL_CNTRS	0x4 /* has exclusive counter requirements  */
983 #define PMU_FL_EXCL_ENABLED	0x8 /* exclusive counter active */
984 #define PMU_FL_PEBS_ALL		0x10 /* all events are valid PEBS events */
985 #define PMU_FL_TFA		0x20 /* deal with TSX force abort */
986 #define PMU_FL_PAIR		0x40 /* merge counters for large incr. events */
987 #define PMU_FL_INSTR_LATENCY	0x80 /* Support Instruction Latency in PEBS Memory Info Record */
988 #define PMU_FL_MEM_LOADS_AUX	0x100 /* Require an auxiliary event for the complete memory info */
989 
990 #define EVENT_VAR(_id)  event_attr_##_id
991 #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
992 
993 #define EVENT_ATTR(_name, _id)						\
994 static struct perf_pmu_events_attr EVENT_VAR(_id) = {			\
995 	.attr		= __ATTR(_name, 0444, events_sysfs_show, NULL),	\
996 	.id		= PERF_COUNT_HW_##_id,				\
997 	.event_str	= NULL,						\
998 };
999 
1000 #define EVENT_ATTR_STR(_name, v, str)					\
1001 static struct perf_pmu_events_attr event_attr_##v = {			\
1002 	.attr		= __ATTR(_name, 0444, events_sysfs_show, NULL),	\
1003 	.id		= 0,						\
1004 	.event_str	= str,						\
1005 };
1006 
1007 #define EVENT_ATTR_STR_HT(_name, v, noht, ht)				\
1008 static struct perf_pmu_events_ht_attr event_attr_##v = {		\
1009 	.attr		= __ATTR(_name, 0444, events_ht_sysfs_show, NULL),\
1010 	.id		= 0,						\
1011 	.event_str_noht	= noht,						\
1012 	.event_str_ht	= ht,						\
1013 }
1014 
1015 #define EVENT_ATTR_STR_HYBRID(_name, v, str, _pmu)			\
1016 static struct perf_pmu_events_hybrid_attr event_attr_##v = {		\
1017 	.attr		= __ATTR(_name, 0444, events_hybrid_sysfs_show, NULL),\
1018 	.id		= 0,						\
1019 	.event_str	= str,						\
1020 	.pmu_type	= _pmu,						\
1021 }
1022 
1023 #define FORMAT_HYBRID_PTR(_id) (&format_attr_hybrid_##_id.attr.attr)
1024 
1025 #define FORMAT_ATTR_HYBRID(_name, _pmu)					\
1026 static struct perf_pmu_format_hybrid_attr format_attr_hybrid_##_name = {\
1027 	.attr		= __ATTR_RO(_name),				\
1028 	.pmu_type	= _pmu,						\
1029 }
1030 
1031 struct pmu *x86_get_pmu(unsigned int cpu);
1032 extern struct x86_pmu x86_pmu __read_mostly;
1033 
1034 static __always_inline struct x86_perf_task_context_opt *task_context_opt(void *ctx)
1035 {
1036 	if (static_cpu_has(X86_FEATURE_ARCH_LBR))
1037 		return &((struct x86_perf_task_context_arch_lbr *)ctx)->opt;
1038 
1039 	return &((struct x86_perf_task_context *)ctx)->opt;
1040 }
1041 
1042 static inline bool x86_pmu_has_lbr_callstack(void)
1043 {
1044 	return  x86_pmu.lbr_sel_map &&
1045 		x86_pmu.lbr_sel_map[PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT] > 0;
1046 }
1047 
1048 DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
1049 
1050 int x86_perf_event_set_period(struct perf_event *event);
1051 
1052 /*
1053  * Generalized hw caching related hw_event table, filled
1054  * in on a per model basis. A value of 0 means
1055  * 'not supported', -1 means 'hw_event makes no sense on
1056  * this CPU', any other value means the raw hw_event
1057  * ID.
1058  */
1059 
1060 #define C(x) PERF_COUNT_HW_CACHE_##x
1061 
1062 extern u64 __read_mostly hw_cache_event_ids
1063 				[PERF_COUNT_HW_CACHE_MAX]
1064 				[PERF_COUNT_HW_CACHE_OP_MAX]
1065 				[PERF_COUNT_HW_CACHE_RESULT_MAX];
1066 extern u64 __read_mostly hw_cache_extra_regs
1067 				[PERF_COUNT_HW_CACHE_MAX]
1068 				[PERF_COUNT_HW_CACHE_OP_MAX]
1069 				[PERF_COUNT_HW_CACHE_RESULT_MAX];
1070 
1071 u64 x86_perf_event_update(struct perf_event *event);
1072 
1073 static inline unsigned int x86_pmu_config_addr(int index)
1074 {
1075 	return x86_pmu.eventsel + (x86_pmu.addr_offset ?
1076 				   x86_pmu.addr_offset(index, true) : index);
1077 }
1078 
1079 static inline unsigned int x86_pmu_event_addr(int index)
1080 {
1081 	return x86_pmu.perfctr + (x86_pmu.addr_offset ?
1082 				  x86_pmu.addr_offset(index, false) : index);
1083 }
1084 
1085 static inline int x86_pmu_rdpmc_index(int index)
1086 {
1087 	return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index;
1088 }
1089 
1090 bool check_hw_exists(struct pmu *pmu, int num_counters,
1091 		     int num_counters_fixed);
1092 
1093 int x86_add_exclusive(unsigned int what);
1094 
1095 void x86_del_exclusive(unsigned int what);
1096 
1097 int x86_reserve_hardware(void);
1098 
1099 void x86_release_hardware(void);
1100 
1101 int x86_pmu_max_precise(void);
1102 
1103 void hw_perf_lbr_event_destroy(struct perf_event *event);
1104 
1105 int x86_setup_perfctr(struct perf_event *event);
1106 
1107 int x86_pmu_hw_config(struct perf_event *event);
1108 
1109 void x86_pmu_disable_all(void);
1110 
1111 static inline bool has_amd_brs(struct hw_perf_event *hwc)
1112 {
1113 	return hwc->flags & PERF_X86_EVENT_AMD_BRS;
1114 }
1115 
1116 static inline bool is_counter_pair(struct hw_perf_event *hwc)
1117 {
1118 	return hwc->flags & PERF_X86_EVENT_PAIR;
1119 }
1120 
1121 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
1122 					  u64 enable_mask)
1123 {
1124 	u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
1125 
1126 	if (hwc->extra_reg.reg)
1127 		wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
1128 
1129 	/*
1130 	 * Add enabled Merge event on next counter
1131 	 * if large increment event being enabled on this counter
1132 	 */
1133 	if (is_counter_pair(hwc))
1134 		wrmsrl(x86_pmu_config_addr(hwc->idx + 1), x86_pmu.perf_ctr_pair_en);
1135 
1136 	wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask);
1137 }
1138 
1139 void x86_pmu_enable_all(int added);
1140 
1141 int perf_assign_events(struct event_constraint **constraints, int n,
1142 			int wmin, int wmax, int gpmax, int *assign);
1143 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
1144 
1145 void x86_pmu_stop(struct perf_event *event, int flags);
1146 
1147 static inline void x86_pmu_disable_event(struct perf_event *event)
1148 {
1149 	u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask);
1150 	struct hw_perf_event *hwc = &event->hw;
1151 
1152 	wrmsrl(hwc->config_base, hwc->config & ~disable_mask);
1153 
1154 	if (is_counter_pair(hwc))
1155 		wrmsrl(x86_pmu_config_addr(hwc->idx + 1), 0);
1156 }
1157 
1158 void x86_pmu_enable_event(struct perf_event *event);
1159 
1160 int x86_pmu_handle_irq(struct pt_regs *regs);
1161 
1162 void x86_pmu_show_pmu_cap(int num_counters, int num_counters_fixed,
1163 			  u64 intel_ctrl);
1164 
1165 void x86_pmu_update_cpu_context(struct pmu *pmu, int cpu);
1166 
1167 extern struct event_constraint emptyconstraint;
1168 
1169 extern struct event_constraint unconstrained;
1170 
1171 static inline bool kernel_ip(unsigned long ip)
1172 {
1173 #ifdef CONFIG_X86_32
1174 	return ip > PAGE_OFFSET;
1175 #else
1176 	return (long)ip < 0;
1177 #endif
1178 }
1179 
1180 /*
1181  * Not all PMUs provide the right context information to place the reported IP
1182  * into full context. Specifically segment registers are typically not
1183  * supplied.
1184  *
1185  * Assuming the address is a linear address (it is for IBS), we fake the CS and
1186  * vm86 mode using the known zero-based code segment and 'fix up' the registers
1187  * to reflect this.
1188  *
1189  * Intel PEBS/LBR appear to typically provide the effective address, nothing
1190  * much we can do about that but pray and treat it like a linear address.
1191  */
1192 static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip)
1193 {
1194 	regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS;
1195 	if (regs->flags & X86_VM_MASK)
1196 		regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK);
1197 	regs->ip = ip;
1198 }
1199 
1200 ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event);
1201 ssize_t intel_event_sysfs_show(char *page, u64 config);
1202 
1203 ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
1204 			  char *page);
1205 ssize_t events_ht_sysfs_show(struct device *dev, struct device_attribute *attr,
1206 			  char *page);
1207 ssize_t events_hybrid_sysfs_show(struct device *dev,
1208 				 struct device_attribute *attr,
1209 				 char *page);
1210 
1211 static inline bool fixed_counter_disabled(int i, struct pmu *pmu)
1212 {
1213 	u64 intel_ctrl = hybrid(pmu, intel_ctrl);
1214 
1215 	return !(intel_ctrl >> (i + INTEL_PMC_IDX_FIXED));
1216 }
1217 
1218 #ifdef CONFIG_CPU_SUP_AMD
1219 
1220 int amd_pmu_init(void);
1221 
1222 #ifdef CONFIG_PERF_EVENTS_AMD_BRS
1223 int amd_brs_init(void);
1224 void amd_brs_disable(void);
1225 void amd_brs_enable(void);
1226 void amd_brs_enable_all(void);
1227 void amd_brs_disable_all(void);
1228 void amd_brs_drain(void);
1229 void amd_brs_lopwr_init(void);
1230 void amd_brs_disable_all(void);
1231 int amd_brs_setup_filter(struct perf_event *event);
1232 void amd_brs_reset(void);
1233 
1234 static inline void amd_pmu_brs_add(struct perf_event *event)
1235 {
1236 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
1237 
1238 	perf_sched_cb_inc(event->ctx->pmu);
1239 	cpuc->lbr_users++;
1240 	/*
1241 	 * No need to reset BRS because it is reset
1242 	 * on brs_enable() and it is saturating
1243 	 */
1244 }
1245 
1246 static inline void amd_pmu_brs_del(struct perf_event *event)
1247 {
1248 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
1249 
1250 	cpuc->lbr_users--;
1251 	WARN_ON_ONCE(cpuc->lbr_users < 0);
1252 
1253 	perf_sched_cb_dec(event->ctx->pmu);
1254 }
1255 
1256 void amd_pmu_brs_sched_task(struct perf_event_context *ctx, bool sched_in);
1257 #else
1258 static inline int amd_brs_init(void)
1259 {
1260 	return 0;
1261 }
1262 static inline void amd_brs_disable(void) {}
1263 static inline void amd_brs_enable(void) {}
1264 static inline void amd_brs_drain(void) {}
1265 static inline void amd_brs_lopwr_init(void) {}
1266 static inline void amd_brs_disable_all(void) {}
1267 static inline int amd_brs_setup_filter(struct perf_event *event)
1268 {
1269 	return 0;
1270 }
1271 static inline void amd_brs_reset(void) {}
1272 
1273 static inline void amd_pmu_brs_add(struct perf_event *event)
1274 {
1275 }
1276 
1277 static inline void amd_pmu_brs_del(struct perf_event *event)
1278 {
1279 }
1280 
1281 static inline void amd_pmu_brs_sched_task(struct perf_event_context *ctx, bool sched_in)
1282 {
1283 }
1284 
1285 static inline void amd_brs_enable_all(void)
1286 {
1287 }
1288 
1289 #endif
1290 
1291 #else /* CONFIG_CPU_SUP_AMD */
1292 
1293 static inline int amd_pmu_init(void)
1294 {
1295 	return 0;
1296 }
1297 
1298 static inline int amd_brs_init(void)
1299 {
1300 	return -EOPNOTSUPP;
1301 }
1302 
1303 static inline void amd_brs_drain(void)
1304 {
1305 }
1306 
1307 static inline void amd_brs_enable_all(void)
1308 {
1309 }
1310 
1311 static inline void amd_brs_disable_all(void)
1312 {
1313 }
1314 #endif /* CONFIG_CPU_SUP_AMD */
1315 
1316 static inline int is_pebs_pt(struct perf_event *event)
1317 {
1318 	return !!(event->hw.flags & PERF_X86_EVENT_PEBS_VIA_PT);
1319 }
1320 
1321 #ifdef CONFIG_CPU_SUP_INTEL
1322 
1323 static inline bool intel_pmu_has_bts_period(struct perf_event *event, u64 period)
1324 {
1325 	struct hw_perf_event *hwc = &event->hw;
1326 	unsigned int hw_event, bts_event;
1327 
1328 	if (event->attr.freq)
1329 		return false;
1330 
1331 	hw_event = hwc->config & INTEL_ARCH_EVENT_MASK;
1332 	bts_event = x86_pmu.event_map(PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
1333 
1334 	return hw_event == bts_event && period == 1;
1335 }
1336 
1337 static inline bool intel_pmu_has_bts(struct perf_event *event)
1338 {
1339 	struct hw_perf_event *hwc = &event->hw;
1340 
1341 	return intel_pmu_has_bts_period(event, hwc->sample_period);
1342 }
1343 
1344 static __always_inline void __intel_pmu_pebs_disable_all(void)
1345 {
1346 	wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1347 }
1348 
1349 static __always_inline void __intel_pmu_arch_lbr_disable(void)
1350 {
1351 	wrmsrl(MSR_ARCH_LBR_CTL, 0);
1352 }
1353 
1354 static __always_inline void __intel_pmu_lbr_disable(void)
1355 {
1356 	u64 debugctl;
1357 
1358 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
1359 	debugctl &= ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_FREEZE_LBRS_ON_PMI);
1360 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
1361 }
1362 
1363 int intel_pmu_save_and_restart(struct perf_event *event);
1364 
1365 struct event_constraint *
1366 x86_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
1367 			  struct perf_event *event);
1368 
1369 extern int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu);
1370 extern void intel_cpuc_finish(struct cpu_hw_events *cpuc);
1371 
1372 int intel_pmu_init(void);
1373 
1374 void init_debug_store_on_cpu(int cpu);
1375 
1376 void fini_debug_store_on_cpu(int cpu);
1377 
1378 void release_ds_buffers(void);
1379 
1380 void reserve_ds_buffers(void);
1381 
1382 void release_lbr_buffers(void);
1383 
1384 void reserve_lbr_buffers(void);
1385 
1386 extern struct event_constraint bts_constraint;
1387 extern struct event_constraint vlbr_constraint;
1388 
1389 void intel_pmu_enable_bts(u64 config);
1390 
1391 void intel_pmu_disable_bts(void);
1392 
1393 int intel_pmu_drain_bts_buffer(void);
1394 
1395 extern struct event_constraint intel_core2_pebs_event_constraints[];
1396 
1397 extern struct event_constraint intel_atom_pebs_event_constraints[];
1398 
1399 extern struct event_constraint intel_slm_pebs_event_constraints[];
1400 
1401 extern struct event_constraint intel_glm_pebs_event_constraints[];
1402 
1403 extern struct event_constraint intel_glp_pebs_event_constraints[];
1404 
1405 extern struct event_constraint intel_grt_pebs_event_constraints[];
1406 
1407 extern struct event_constraint intel_nehalem_pebs_event_constraints[];
1408 
1409 extern struct event_constraint intel_westmere_pebs_event_constraints[];
1410 
1411 extern struct event_constraint intel_snb_pebs_event_constraints[];
1412 
1413 extern struct event_constraint intel_ivb_pebs_event_constraints[];
1414 
1415 extern struct event_constraint intel_hsw_pebs_event_constraints[];
1416 
1417 extern struct event_constraint intel_bdw_pebs_event_constraints[];
1418 
1419 extern struct event_constraint intel_skl_pebs_event_constraints[];
1420 
1421 extern struct event_constraint intel_icl_pebs_event_constraints[];
1422 
1423 extern struct event_constraint intel_spr_pebs_event_constraints[];
1424 
1425 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
1426 
1427 void intel_pmu_pebs_add(struct perf_event *event);
1428 
1429 void intel_pmu_pebs_del(struct perf_event *event);
1430 
1431 void intel_pmu_pebs_enable(struct perf_event *event);
1432 
1433 void intel_pmu_pebs_disable(struct perf_event *event);
1434 
1435 void intel_pmu_pebs_enable_all(void);
1436 
1437 void intel_pmu_pebs_disable_all(void);
1438 
1439 void intel_pmu_pebs_sched_task(struct perf_event_context *ctx, bool sched_in);
1440 
1441 void intel_pmu_auto_reload_read(struct perf_event *event);
1442 
1443 void intel_pmu_store_pebs_lbrs(struct lbr_entry *lbr);
1444 
1445 void intel_ds_init(void);
1446 
1447 void intel_pmu_lbr_swap_task_ctx(struct perf_event_context *prev,
1448 				 struct perf_event_context *next);
1449 
1450 void intel_pmu_lbr_sched_task(struct perf_event_context *ctx, bool sched_in);
1451 
1452 u64 lbr_from_signext_quirk_wr(u64 val);
1453 
1454 void intel_pmu_lbr_reset(void);
1455 
1456 void intel_pmu_lbr_reset_32(void);
1457 
1458 void intel_pmu_lbr_reset_64(void);
1459 
1460 void intel_pmu_lbr_add(struct perf_event *event);
1461 
1462 void intel_pmu_lbr_del(struct perf_event *event);
1463 
1464 void intel_pmu_lbr_enable_all(bool pmi);
1465 
1466 void intel_pmu_lbr_disable_all(void);
1467 
1468 void intel_pmu_lbr_read(void);
1469 
1470 void intel_pmu_lbr_read_32(struct cpu_hw_events *cpuc);
1471 
1472 void intel_pmu_lbr_read_64(struct cpu_hw_events *cpuc);
1473 
1474 void intel_pmu_lbr_save(void *ctx);
1475 
1476 void intel_pmu_lbr_restore(void *ctx);
1477 
1478 void intel_pmu_lbr_init_core(void);
1479 
1480 void intel_pmu_lbr_init_nhm(void);
1481 
1482 void intel_pmu_lbr_init_atom(void);
1483 
1484 void intel_pmu_lbr_init_slm(void);
1485 
1486 void intel_pmu_lbr_init_snb(void);
1487 
1488 void intel_pmu_lbr_init_hsw(void);
1489 
1490 void intel_pmu_lbr_init_skl(void);
1491 
1492 void intel_pmu_lbr_init_knl(void);
1493 
1494 void intel_pmu_lbr_init(void);
1495 
1496 void intel_pmu_arch_lbr_init(void);
1497 
1498 void intel_pmu_pebs_data_source_nhm(void);
1499 
1500 void intel_pmu_pebs_data_source_skl(bool pmem);
1501 
1502 int intel_pmu_setup_lbr_filter(struct perf_event *event);
1503 
1504 void intel_pt_interrupt(void);
1505 
1506 int intel_bts_interrupt(void);
1507 
1508 void intel_bts_enable_local(void);
1509 
1510 void intel_bts_disable_local(void);
1511 
1512 int p4_pmu_init(void);
1513 
1514 int p6_pmu_init(void);
1515 
1516 int knc_pmu_init(void);
1517 
1518 static inline int is_ht_workaround_enabled(void)
1519 {
1520 	return !!(x86_pmu.flags & PMU_FL_EXCL_ENABLED);
1521 }
1522 
1523 #else /* CONFIG_CPU_SUP_INTEL */
1524 
1525 static inline void reserve_ds_buffers(void)
1526 {
1527 }
1528 
1529 static inline void release_ds_buffers(void)
1530 {
1531 }
1532 
1533 static inline void release_lbr_buffers(void)
1534 {
1535 }
1536 
1537 static inline void reserve_lbr_buffers(void)
1538 {
1539 }
1540 
1541 static inline int intel_pmu_init(void)
1542 {
1543 	return 0;
1544 }
1545 
1546 static inline int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu)
1547 {
1548 	return 0;
1549 }
1550 
1551 static inline void intel_cpuc_finish(struct cpu_hw_events *cpuc)
1552 {
1553 }
1554 
1555 static inline int is_ht_workaround_enabled(void)
1556 {
1557 	return 0;
1558 }
1559 #endif /* CONFIG_CPU_SUP_INTEL */
1560 
1561 #if ((defined CONFIG_CPU_SUP_CENTAUR) || (defined CONFIG_CPU_SUP_ZHAOXIN))
1562 int zhaoxin_pmu_init(void);
1563 #else
1564 static inline int zhaoxin_pmu_init(void)
1565 {
1566 	return 0;
1567 }
1568 #endif /*CONFIG_CPU_SUP_CENTAUR or CONFIG_CPU_SUP_ZHAOXIN*/
1569