perf, intel: Use GO/HO bits in perf-ctr
[deliverable/linux.git] / arch / x86 / kernel / cpu / perf_event.h
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 <pzijlstr@redhat.com>
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 /*
18 * | NHM/WSM | SNB |
19 * register -------------------------------
20 * | HT | no HT | HT | no HT |
21 *-----------------------------------------
22 * offcore | core | core | cpu | core |
23 * lbr_sel | core | core | cpu | core |
24 * ld_lat | cpu | core | cpu | core |
25 *-----------------------------------------
26 *
27 * Given that there is a small number of shared regs,
28 * we can pre-allocate their slot in the per-cpu
29 * per-core reg tables.
30 */
31 enum extra_reg_type {
32 EXTRA_REG_NONE = -1, /* not used */
33
34 EXTRA_REG_RSP_0 = 0, /* offcore_response_0 */
35 EXTRA_REG_RSP_1 = 1, /* offcore_response_1 */
36
37 EXTRA_REG_MAX /* number of entries needed */
38 };
39
40 struct event_constraint {
41 union {
42 unsigned long idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
43 u64 idxmsk64;
44 };
45 u64 code;
46 u64 cmask;
47 int weight;
48 };
49
50 struct amd_nb {
51 int nb_id; /* NorthBridge id */
52 int refcnt; /* reference count */
53 struct perf_event *owners[X86_PMC_IDX_MAX];
54 struct event_constraint event_constraints[X86_PMC_IDX_MAX];
55 };
56
57 /* The maximal number of PEBS events: */
58 #define MAX_PEBS_EVENTS 4
59
60 /*
61 * A debug store configuration.
62 *
63 * We only support architectures that use 64bit fields.
64 */
65 struct debug_store {
66 u64 bts_buffer_base;
67 u64 bts_index;
68 u64 bts_absolute_maximum;
69 u64 bts_interrupt_threshold;
70 u64 pebs_buffer_base;
71 u64 pebs_index;
72 u64 pebs_absolute_maximum;
73 u64 pebs_interrupt_threshold;
74 u64 pebs_event_reset[MAX_PEBS_EVENTS];
75 };
76
77 /*
78 * Per register state.
79 */
80 struct er_account {
81 raw_spinlock_t lock; /* per-core: protect structure */
82 u64 config; /* extra MSR config */
83 u64 reg; /* extra MSR number */
84 atomic_t ref; /* reference count */
85 };
86
87 /*
88 * Per core/cpu state
89 *
90 * Used to coordinate shared registers between HT threads or
91 * among events on a single PMU.
92 */
93 struct intel_shared_regs {
94 struct er_account regs[EXTRA_REG_MAX];
95 int refcnt; /* per-core: #HT threads */
96 unsigned core_id; /* per-core: core id */
97 };
98
99 #define MAX_LBR_ENTRIES 16
100
101 struct cpu_hw_events {
102 /*
103 * Generic x86 PMC bits
104 */
105 struct perf_event *events[X86_PMC_IDX_MAX]; /* in counter order */
106 unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
107 unsigned long running[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
108 int enabled;
109
110 int n_events;
111 int n_added;
112 int n_txn;
113 int assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
114 u64 tags[X86_PMC_IDX_MAX];
115 struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */
116
117 unsigned int group_flag;
118
119 /*
120 * Intel DebugStore bits
121 */
122 struct debug_store *ds;
123 u64 pebs_enabled;
124
125 /*
126 * Intel LBR bits
127 */
128 int lbr_users;
129 void *lbr_context;
130 struct perf_branch_stack lbr_stack;
131 struct perf_branch_entry lbr_entries[MAX_LBR_ENTRIES];
132
133 /*
134 * Intel host/guest exclude bits
135 */
136 u64 intel_ctrl_guest_mask;
137 u64 intel_ctrl_host_mask;
138 struct perf_guest_switch_msr guest_switch_msrs[X86_PMC_IDX_MAX];
139
140 /*
141 * manage shared (per-core, per-cpu) registers
142 * used on Intel NHM/WSM/SNB
143 */
144 struct intel_shared_regs *shared_regs;
145
146 /*
147 * AMD specific bits
148 */
149 struct amd_nb *amd_nb;
150
151 void *kfree_on_online;
152 };
153
154 #define __EVENT_CONSTRAINT(c, n, m, w) {\
155 { .idxmsk64 = (n) }, \
156 .code = (c), \
157 .cmask = (m), \
158 .weight = (w), \
159 }
160
161 #define EVENT_CONSTRAINT(c, n, m) \
162 __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n))
163
164 /*
165 * Constraint on the Event code.
166 */
167 #define INTEL_EVENT_CONSTRAINT(c, n) \
168 EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT)
169
170 /*
171 * Constraint on the Event code + UMask + fixed-mask
172 *
173 * filter mask to validate fixed counter events.
174 * the following filters disqualify for fixed counters:
175 * - inv
176 * - edge
177 * - cnt-mask
178 * The other filters are supported by fixed counters.
179 * The any-thread option is supported starting with v3.
180 */
181 #define FIXED_EVENT_CONSTRAINT(c, n) \
182 EVENT_CONSTRAINT(c, (1ULL << (32+n)), X86_RAW_EVENT_MASK)
183
184 /*
185 * Constraint on the Event code + UMask
186 */
187 #define INTEL_UEVENT_CONSTRAINT(c, n) \
188 EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK)
189
190 #define EVENT_CONSTRAINT_END \
191 EVENT_CONSTRAINT(0, 0, 0)
192
193 #define for_each_event_constraint(e, c) \
194 for ((e) = (c); (e)->weight; (e)++)
195
196 /*
197 * Extra registers for specific events.
198 *
199 * Some events need large masks and require external MSRs.
200 * Those extra MSRs end up being shared for all events on
201 * a PMU and sometimes between PMU of sibling HT threads.
202 * In either case, the kernel needs to handle conflicting
203 * accesses to those extra, shared, regs. The data structure
204 * to manage those registers is stored in cpu_hw_event.
205 */
206 struct extra_reg {
207 unsigned int event;
208 unsigned int msr;
209 u64 config_mask;
210 u64 valid_mask;
211 int idx; /* per_xxx->regs[] reg index */
212 };
213
214 #define EVENT_EXTRA_REG(e, ms, m, vm, i) { \
215 .event = (e), \
216 .msr = (ms), \
217 .config_mask = (m), \
218 .valid_mask = (vm), \
219 .idx = EXTRA_REG_##i \
220 }
221
222 #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx) \
223 EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx)
224
225 #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0)
226
227 union perf_capabilities {
228 struct {
229 u64 lbr_format:6;
230 u64 pebs_trap:1;
231 u64 pebs_arch_reg:1;
232 u64 pebs_format:4;
233 u64 smm_freeze:1;
234 };
235 u64 capabilities;
236 };
237
238 /*
239 * struct x86_pmu - generic x86 pmu
240 */
241 struct x86_pmu {
242 /*
243 * Generic x86 PMC bits
244 */
245 const char *name;
246 int version;
247 int (*handle_irq)(struct pt_regs *);
248 void (*disable_all)(void);
249 void (*enable_all)(int added);
250 void (*enable)(struct perf_event *);
251 void (*disable)(struct perf_event *);
252 int (*hw_config)(struct perf_event *event);
253 int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
254 unsigned eventsel;
255 unsigned perfctr;
256 u64 (*event_map)(int);
257 int max_events;
258 int num_counters;
259 int num_counters_fixed;
260 int cntval_bits;
261 u64 cntval_mask;
262 int apic;
263 u64 max_period;
264 struct event_constraint *
265 (*get_event_constraints)(struct cpu_hw_events *cpuc,
266 struct perf_event *event);
267
268 void (*put_event_constraints)(struct cpu_hw_events *cpuc,
269 struct perf_event *event);
270 struct event_constraint *event_constraints;
271 void (*quirks)(void);
272 int perfctr_second_write;
273
274 int (*cpu_prepare)(int cpu);
275 void (*cpu_starting)(int cpu);
276 void (*cpu_dying)(int cpu);
277 void (*cpu_dead)(int cpu);
278
279 /*
280 * Intel Arch Perfmon v2+
281 */
282 u64 intel_ctrl;
283 union perf_capabilities intel_cap;
284
285 /*
286 * Intel DebugStore bits
287 */
288 int bts, pebs;
289 int bts_active, pebs_active;
290 int pebs_record_size;
291 void (*drain_pebs)(struct pt_regs *regs);
292 struct event_constraint *pebs_constraints;
293
294 /*
295 * Intel LBR
296 */
297 unsigned long lbr_tos, lbr_from, lbr_to; /* MSR base regs */
298 int lbr_nr; /* hardware stack size */
299
300 /*
301 * Extra registers for events
302 */
303 struct extra_reg *extra_regs;
304 unsigned int er_flags;
305
306 /*
307 * Intel host/guest support (KVM)
308 */
309 struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr);
310 };
311
312 #define ERF_NO_HT_SHARING 1
313 #define ERF_HAS_RSP_1 2
314
315 extern struct x86_pmu x86_pmu __read_mostly;
316
317 DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
318
319 int x86_perf_event_set_period(struct perf_event *event);
320
321 /*
322 * Generalized hw caching related hw_event table, filled
323 * in on a per model basis. A value of 0 means
324 * 'not supported', -1 means 'hw_event makes no sense on
325 * this CPU', any other value means the raw hw_event
326 * ID.
327 */
328
329 #define C(x) PERF_COUNT_HW_CACHE_##x
330
331 extern u64 __read_mostly hw_cache_event_ids
332 [PERF_COUNT_HW_CACHE_MAX]
333 [PERF_COUNT_HW_CACHE_OP_MAX]
334 [PERF_COUNT_HW_CACHE_RESULT_MAX];
335 extern u64 __read_mostly hw_cache_extra_regs
336 [PERF_COUNT_HW_CACHE_MAX]
337 [PERF_COUNT_HW_CACHE_OP_MAX]
338 [PERF_COUNT_HW_CACHE_RESULT_MAX];
339
340 u64 x86_perf_event_update(struct perf_event *event);
341
342 static inline int x86_pmu_addr_offset(int index)
343 {
344 int offset;
345
346 /* offset = X86_FEATURE_PERFCTR_CORE ? index << 1 : index */
347 alternative_io(ASM_NOP2,
348 "shll $1, %%eax",
349 X86_FEATURE_PERFCTR_CORE,
350 "=a" (offset),
351 "a" (index));
352
353 return offset;
354 }
355
356 static inline unsigned int x86_pmu_config_addr(int index)
357 {
358 return x86_pmu.eventsel + x86_pmu_addr_offset(index);
359 }
360
361 static inline unsigned int x86_pmu_event_addr(int index)
362 {
363 return x86_pmu.perfctr + x86_pmu_addr_offset(index);
364 }
365
366 int x86_setup_perfctr(struct perf_event *event);
367
368 int x86_pmu_hw_config(struct perf_event *event);
369
370 void x86_pmu_disable_all(void);
371
372 static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc,
373 u64 enable_mask)
374 {
375 if (hwc->extra_reg.reg)
376 wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config);
377 wrmsrl(hwc->config_base, hwc->config | enable_mask);
378 }
379
380 void x86_pmu_enable_all(int added);
381
382 int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign);
383
384 void x86_pmu_stop(struct perf_event *event, int flags);
385
386 static inline void x86_pmu_disable_event(struct perf_event *event)
387 {
388 struct hw_perf_event *hwc = &event->hw;
389
390 wrmsrl(hwc->config_base, hwc->config);
391 }
392
393 void x86_pmu_enable_event(struct perf_event *event);
394
395 int x86_pmu_handle_irq(struct pt_regs *regs);
396
397 extern struct event_constraint emptyconstraint;
398
399 extern struct event_constraint unconstrained;
400
401 #ifdef CONFIG_CPU_SUP_AMD
402
403 int amd_pmu_init(void);
404
405 #else /* CONFIG_CPU_SUP_AMD */
406
407 static inline int amd_pmu_init(void)
408 {
409 return 0;
410 }
411
412 #endif /* CONFIG_CPU_SUP_AMD */
413
414 #ifdef CONFIG_CPU_SUP_INTEL
415
416 int intel_pmu_save_and_restart(struct perf_event *event);
417
418 struct event_constraint *
419 x86_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event);
420
421 struct intel_shared_regs *allocate_shared_regs(int cpu);
422
423 int intel_pmu_init(void);
424
425 void init_debug_store_on_cpu(int cpu);
426
427 void fini_debug_store_on_cpu(int cpu);
428
429 void release_ds_buffers(void);
430
431 void reserve_ds_buffers(void);
432
433 extern struct event_constraint bts_constraint;
434
435 void intel_pmu_enable_bts(u64 config);
436
437 void intel_pmu_disable_bts(void);
438
439 int intel_pmu_drain_bts_buffer(void);
440
441 extern struct event_constraint intel_core2_pebs_event_constraints[];
442
443 extern struct event_constraint intel_atom_pebs_event_constraints[];
444
445 extern struct event_constraint intel_nehalem_pebs_event_constraints[];
446
447 extern struct event_constraint intel_westmere_pebs_event_constraints[];
448
449 extern struct event_constraint intel_snb_pebs_event_constraints[];
450
451 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
452
453 void intel_pmu_pebs_enable(struct perf_event *event);
454
455 void intel_pmu_pebs_disable(struct perf_event *event);
456
457 void intel_pmu_pebs_enable_all(void);
458
459 void intel_pmu_pebs_disable_all(void);
460
461 void intel_ds_init(void);
462
463 void intel_pmu_lbr_reset(void);
464
465 void intel_pmu_lbr_enable(struct perf_event *event);
466
467 void intel_pmu_lbr_disable(struct perf_event *event);
468
469 void intel_pmu_lbr_enable_all(void);
470
471 void intel_pmu_lbr_disable_all(void);
472
473 void intel_pmu_lbr_read(void);
474
475 void intel_pmu_lbr_init_core(void);
476
477 void intel_pmu_lbr_init_nhm(void);
478
479 void intel_pmu_lbr_init_atom(void);
480
481 int p4_pmu_init(void);
482
483 int p6_pmu_init(void);
484
485 #else /* CONFIG_CPU_SUP_INTEL */
486
487 static inline void reserve_ds_buffers(void)
488 {
489 }
490
491 static inline void release_ds_buffers(void)
492 {
493 }
494
495 static inline int intel_pmu_init(void)
496 {
497 return 0;
498 }
499
500 static inline struct intel_shared_regs *allocate_shared_regs(int cpu)
501 {
502 return NULL;
503 }
504
505 #endif /* CONFIG_CPU_SUP_INTEL */
This page took 0.040945 seconds and 5 git commands to generate.