]>
Commit | Line | Data |
---|---|---|
de0428a7 KW |
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 | ||
1c2ac3fd PZ |
17 | #if 0 |
18 | #undef wrmsrl | |
19 | #define wrmsrl(msr, val) \ | |
20 | do { \ | |
21 | unsigned int _msr = (msr); \ | |
22 | u64 _val = (val); \ | |
23 | trace_printk("wrmsrl(%x, %Lx)\n", (unsigned int)(_msr), \ | |
24 | (unsigned long long)(_val)); \ | |
25 | native_write_msr((_msr), (u32)(_val), (u32)(_val >> 32)); \ | |
26 | } while (0) | |
27 | #endif | |
28 | ||
de0428a7 KW |
29 | /* |
30 | * | NHM/WSM | SNB | | |
31 | * register ------------------------------- | |
32 | * | HT | no HT | HT | no HT | | |
33 | *----------------------------------------- | |
34 | * offcore | core | core | cpu | core | | |
35 | * lbr_sel | core | core | cpu | core | | |
36 | * ld_lat | cpu | core | cpu | core | | |
37 | *----------------------------------------- | |
38 | * | |
39 | * Given that there is a small number of shared regs, | |
40 | * we can pre-allocate their slot in the per-cpu | |
41 | * per-core reg tables. | |
42 | */ | |
43 | enum extra_reg_type { | |
44 | EXTRA_REG_NONE = -1, /* not used */ | |
45 | ||
46 | EXTRA_REG_RSP_0 = 0, /* offcore_response_0 */ | |
47 | EXTRA_REG_RSP_1 = 1, /* offcore_response_1 */ | |
b36817e8 | 48 | EXTRA_REG_LBR = 2, /* lbr_select */ |
f20093ee | 49 | EXTRA_REG_LDLAT = 3, /* ld_lat_threshold */ |
de0428a7 KW |
50 | |
51 | EXTRA_REG_MAX /* number of entries needed */ | |
52 | }; | |
53 | ||
54 | struct event_constraint { | |
55 | union { | |
56 | unsigned long idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; | |
57 | u64 idxmsk64; | |
58 | }; | |
59 | u64 code; | |
60 | u64 cmask; | |
61 | int weight; | |
bc1738f6 | 62 | int overlap; |
9fac2cf3 | 63 | int flags; |
de0428a7 | 64 | }; |
f20093ee | 65 | /* |
2f7f73a5 | 66 | * struct hw_perf_event.flags flags |
f20093ee SE |
67 | */ |
68 | #define PERF_X86_EVENT_PEBS_LDLAT 0x1 /* ld+ldlat data address sampling */ | |
9ad64c0f | 69 | #define PERF_X86_EVENT_PEBS_ST 0x2 /* st data address sampling */ |
86a04461 | 70 | #define PERF_X86_EVENT_PEBS_ST_HSW 0x4 /* haswell style datala, store */ |
2f7f73a5 | 71 | #define PERF_X86_EVENT_COMMITTED 0x8 /* event passed commit_txn */ |
86a04461 AK |
72 | #define PERF_X86_EVENT_PEBS_LD_HSW 0x10 /* haswell style datala, load */ |
73 | #define PERF_X86_EVENT_PEBS_NA_HSW 0x20 /* haswell style datala, unknown */ | |
7911d3f7 AL |
74 | #define PERF_X86_EVENT_RDPMC_ALLOWED 0x40 /* grant rdpmc permission */ |
75 | ||
de0428a7 KW |
76 | |
77 | struct amd_nb { | |
78 | int nb_id; /* NorthBridge id */ | |
79 | int refcnt; /* reference count */ | |
80 | struct perf_event *owners[X86_PMC_IDX_MAX]; | |
81 | struct event_constraint event_constraints[X86_PMC_IDX_MAX]; | |
82 | }; | |
83 | ||
84 | /* The maximal number of PEBS events: */ | |
70ab7003 | 85 | #define MAX_PEBS_EVENTS 8 |
de0428a7 KW |
86 | |
87 | /* | |
88 | * A debug store configuration. | |
89 | * | |
90 | * We only support architectures that use 64bit fields. | |
91 | */ | |
92 | struct debug_store { | |
93 | u64 bts_buffer_base; | |
94 | u64 bts_index; | |
95 | u64 bts_absolute_maximum; | |
96 | u64 bts_interrupt_threshold; | |
97 | u64 pebs_buffer_base; | |
98 | u64 pebs_index; | |
99 | u64 pebs_absolute_maximum; | |
100 | u64 pebs_interrupt_threshold; | |
101 | u64 pebs_event_reset[MAX_PEBS_EVENTS]; | |
102 | }; | |
103 | ||
104 | /* | |
105 | * Per register state. | |
106 | */ | |
107 | struct er_account { | |
108 | raw_spinlock_t lock; /* per-core: protect structure */ | |
109 | u64 config; /* extra MSR config */ | |
110 | u64 reg; /* extra MSR number */ | |
111 | atomic_t ref; /* reference count */ | |
112 | }; | |
113 | ||
114 | /* | |
115 | * Per core/cpu state | |
116 | * | |
117 | * Used to coordinate shared registers between HT threads or | |
118 | * among events on a single PMU. | |
119 | */ | |
120 | struct intel_shared_regs { | |
121 | struct er_account regs[EXTRA_REG_MAX]; | |
122 | int refcnt; /* per-core: #HT threads */ | |
123 | unsigned core_id; /* per-core: core id */ | |
124 | }; | |
125 | ||
126 | #define MAX_LBR_ENTRIES 16 | |
127 | ||
128 | struct cpu_hw_events { | |
129 | /* | |
130 | * Generic x86 PMC bits | |
131 | */ | |
132 | struct perf_event *events[X86_PMC_IDX_MAX]; /* in counter order */ | |
133 | unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; | |
134 | unsigned long running[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; | |
135 | int enabled; | |
136 | ||
c347a2f1 PZ |
137 | int n_events; /* the # of events in the below arrays */ |
138 | int n_added; /* the # last events in the below arrays; | |
139 | they've never been enabled yet */ | |
140 | int n_txn; /* the # last events in the below arrays; | |
141 | added in the current transaction */ | |
de0428a7 KW |
142 | int assign[X86_PMC_IDX_MAX]; /* event to counter assignment */ |
143 | u64 tags[X86_PMC_IDX_MAX]; | |
144 | struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */ | |
145 | ||
146 | unsigned int group_flag; | |
5a425294 | 147 | int is_fake; |
de0428a7 KW |
148 | |
149 | /* | |
150 | * Intel DebugStore bits | |
151 | */ | |
152 | struct debug_store *ds; | |
153 | u64 pebs_enabled; | |
154 | ||
155 | /* | |
156 | * Intel LBR bits | |
157 | */ | |
158 | int lbr_users; | |
159 | void *lbr_context; | |
160 | struct perf_branch_stack lbr_stack; | |
161 | struct perf_branch_entry lbr_entries[MAX_LBR_ENTRIES]; | |
b36817e8 | 162 | struct er_account *lbr_sel; |
3e702ff6 | 163 | u64 br_sel; |
de0428a7 | 164 | |
144d31e6 GN |
165 | /* |
166 | * Intel host/guest exclude bits | |
167 | */ | |
168 | u64 intel_ctrl_guest_mask; | |
169 | u64 intel_ctrl_host_mask; | |
170 | struct perf_guest_switch_msr guest_switch_msrs[X86_PMC_IDX_MAX]; | |
171 | ||
2b9e344d PZ |
172 | /* |
173 | * Intel checkpoint mask | |
174 | */ | |
175 | u64 intel_cp_status; | |
176 | ||
de0428a7 KW |
177 | /* |
178 | * manage shared (per-core, per-cpu) registers | |
179 | * used on Intel NHM/WSM/SNB | |
180 | */ | |
181 | struct intel_shared_regs *shared_regs; | |
182 | ||
183 | /* | |
184 | * AMD specific bits | |
185 | */ | |
1018faa6 JR |
186 | struct amd_nb *amd_nb; |
187 | /* Inverted mask of bits to clear in the perf_ctr ctrl registers */ | |
188 | u64 perf_ctr_virt_mask; | |
de0428a7 KW |
189 | |
190 | void *kfree_on_online; | |
191 | }; | |
192 | ||
9fac2cf3 | 193 | #define __EVENT_CONSTRAINT(c, n, m, w, o, f) {\ |
de0428a7 KW |
194 | { .idxmsk64 = (n) }, \ |
195 | .code = (c), \ | |
196 | .cmask = (m), \ | |
197 | .weight = (w), \ | |
bc1738f6 | 198 | .overlap = (o), \ |
9fac2cf3 | 199 | .flags = f, \ |
de0428a7 KW |
200 | } |
201 | ||
202 | #define EVENT_CONSTRAINT(c, n, m) \ | |
9fac2cf3 | 203 | __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 0, 0) |
bc1738f6 RR |
204 | |
205 | /* | |
206 | * The overlap flag marks event constraints with overlapping counter | |
207 | * masks. This is the case if the counter mask of such an event is not | |
208 | * a subset of any other counter mask of a constraint with an equal or | |
209 | * higher weight, e.g.: | |
210 | * | |
211 | * c_overlaps = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0); | |
212 | * c_another1 = EVENT_CONSTRAINT(0, 0x07, 0); | |
213 | * c_another2 = EVENT_CONSTRAINT(0, 0x38, 0); | |
214 | * | |
215 | * The event scheduler may not select the correct counter in the first | |
216 | * cycle because it needs to know which subsequent events will be | |
217 | * scheduled. It may fail to schedule the events then. So we set the | |
218 | * overlap flag for such constraints to give the scheduler a hint which | |
219 | * events to select for counter rescheduling. | |
220 | * | |
221 | * Care must be taken as the rescheduling algorithm is O(n!) which | |
222 | * will increase scheduling cycles for an over-commited system | |
223 | * dramatically. The number of such EVENT_CONSTRAINT_OVERLAP() macros | |
224 | * and its counter masks must be kept at a minimum. | |
225 | */ | |
226 | #define EVENT_CONSTRAINT_OVERLAP(c, n, m) \ | |
9fac2cf3 | 227 | __EVENT_CONSTRAINT(c, n, m, HWEIGHT(n), 1, 0) |
de0428a7 KW |
228 | |
229 | /* | |
230 | * Constraint on the Event code. | |
231 | */ | |
232 | #define INTEL_EVENT_CONSTRAINT(c, n) \ | |
233 | EVENT_CONSTRAINT(c, n, ARCH_PERFMON_EVENTSEL_EVENT) | |
234 | ||
235 | /* | |
236 | * Constraint on the Event code + UMask + fixed-mask | |
237 | * | |
238 | * filter mask to validate fixed counter events. | |
239 | * the following filters disqualify for fixed counters: | |
240 | * - inv | |
241 | * - edge | |
242 | * - cnt-mask | |
3a632cb2 AK |
243 | * - in_tx |
244 | * - in_tx_checkpointed | |
de0428a7 KW |
245 | * The other filters are supported by fixed counters. |
246 | * The any-thread option is supported starting with v3. | |
247 | */ | |
3a632cb2 | 248 | #define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_IN_TX|HSW_IN_TX_CHECKPOINTED) |
de0428a7 | 249 | #define FIXED_EVENT_CONSTRAINT(c, n) \ |
3a632cb2 | 250 | EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS) |
de0428a7 KW |
251 | |
252 | /* | |
253 | * Constraint on the Event code + UMask | |
254 | */ | |
255 | #define INTEL_UEVENT_CONSTRAINT(c, n) \ | |
256 | EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK) | |
257 | ||
7550ddff AK |
258 | /* Like UEVENT_CONSTRAINT, but match flags too */ |
259 | #define INTEL_FLAGS_UEVENT_CONSTRAINT(c, n) \ | |
260 | EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS) | |
261 | ||
f20093ee | 262 | #define INTEL_PLD_CONSTRAINT(c, n) \ |
86a04461 | 263 | __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \ |
f20093ee SE |
264 | HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LDLAT) |
265 | ||
9ad64c0f | 266 | #define INTEL_PST_CONSTRAINT(c, n) \ |
86a04461 | 267 | __EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \ |
9ad64c0f SE |
268 | HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST) |
269 | ||
86a04461 AK |
270 | /* Event constraint, but match on all event flags too. */ |
271 | #define INTEL_FLAGS_EVENT_CONSTRAINT(c, n) \ | |
272 | EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS) | |
273 | ||
274 | /* Check only flags, but allow all event/umask */ | |
275 | #define INTEL_ALL_EVENT_CONSTRAINT(code, n) \ | |
276 | EVENT_CONSTRAINT(code, n, X86_ALL_EVENT_FLAGS) | |
277 | ||
278 | /* Check flags and event code, and set the HSW store flag */ | |
279 | #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_ST(code, n) \ | |
280 | __EVENT_CONSTRAINT(code, n, \ | |
281 | ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \ | |
282 | HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW) | |
283 | ||
284 | /* Check flags and event code, and set the HSW load flag */ | |
285 | #define INTEL_FLAGS_EVENT_CONSTRAINT_DATALA_LD(code, n) \ | |
286 | __EVENT_CONSTRAINT(code, n, \ | |
287 | ARCH_PERFMON_EVENTSEL_EVENT|X86_ALL_EVENT_FLAGS, \ | |
288 | HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW) | |
289 | ||
290 | /* Check flags and event code/umask, and set the HSW store flag */ | |
291 | #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(code, n) \ | |
292 | __EVENT_CONSTRAINT(code, n, \ | |
293 | INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \ | |
f9134f36 AK |
294 | HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW) |
295 | ||
86a04461 AK |
296 | /* Check flags and event code/umask, and set the HSW load flag */ |
297 | #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(code, n) \ | |
298 | __EVENT_CONSTRAINT(code, n, \ | |
299 | INTEL_ARCH_EVENT_MASK|X86_ALL_EVENT_FLAGS, \ | |
300 | HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_LD_HSW) | |
301 | ||
302 | /* Check flags and event code/umask, and set the HSW N/A flag */ | |
303 | #define INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(code, n) \ | |
304 | __EVENT_CONSTRAINT(code, n, \ | |
305 | INTEL_ARCH_EVENT_MASK|INTEL_ARCH_EVENT_MASK, \ | |
306 | HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_NA_HSW) | |
307 | ||
308 | ||
cf30d52e MD |
309 | /* |
310 | * We define the end marker as having a weight of -1 | |
311 | * to enable blacklisting of events using a counter bitmask | |
312 | * of zero and thus a weight of zero. | |
313 | * The end marker has a weight that cannot possibly be | |
314 | * obtained from counting the bits in the bitmask. | |
315 | */ | |
316 | #define EVENT_CONSTRAINT_END { .weight = -1 } | |
de0428a7 | 317 | |
cf30d52e MD |
318 | /* |
319 | * Check for end marker with weight == -1 | |
320 | */ | |
de0428a7 | 321 | #define for_each_event_constraint(e, c) \ |
cf30d52e | 322 | for ((e) = (c); (e)->weight != -1; (e)++) |
de0428a7 KW |
323 | |
324 | /* | |
325 | * Extra registers for specific events. | |
326 | * | |
327 | * Some events need large masks and require external MSRs. | |
328 | * Those extra MSRs end up being shared for all events on | |
329 | * a PMU and sometimes between PMU of sibling HT threads. | |
330 | * In either case, the kernel needs to handle conflicting | |
331 | * accesses to those extra, shared, regs. The data structure | |
332 | * to manage those registers is stored in cpu_hw_event. | |
333 | */ | |
334 | struct extra_reg { | |
335 | unsigned int event; | |
336 | unsigned int msr; | |
337 | u64 config_mask; | |
338 | u64 valid_mask; | |
339 | int idx; /* per_xxx->regs[] reg index */ | |
338b522c | 340 | bool extra_msr_access; |
de0428a7 KW |
341 | }; |
342 | ||
343 | #define EVENT_EXTRA_REG(e, ms, m, vm, i) { \ | |
338b522c KL |
344 | .event = (e), \ |
345 | .msr = (ms), \ | |
346 | .config_mask = (m), \ | |
347 | .valid_mask = (vm), \ | |
348 | .idx = EXTRA_REG_##i, \ | |
349 | .extra_msr_access = true, \ | |
de0428a7 KW |
350 | } |
351 | ||
352 | #define INTEL_EVENT_EXTRA_REG(event, msr, vm, idx) \ | |
353 | EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT, vm, idx) | |
354 | ||
f20093ee SE |
355 | #define INTEL_UEVENT_EXTRA_REG(event, msr, vm, idx) \ |
356 | EVENT_EXTRA_REG(event, msr, ARCH_PERFMON_EVENTSEL_EVENT | \ | |
357 | ARCH_PERFMON_EVENTSEL_UMASK, vm, idx) | |
358 | ||
359 | #define INTEL_UEVENT_PEBS_LDLAT_EXTRA_REG(c) \ | |
360 | INTEL_UEVENT_EXTRA_REG(c, \ | |
361 | MSR_PEBS_LD_LAT_THRESHOLD, \ | |
362 | 0xffff, \ | |
363 | LDLAT) | |
364 | ||
de0428a7 KW |
365 | #define EVENT_EXTRA_END EVENT_EXTRA_REG(0, 0, 0, 0, RSP_0) |
366 | ||
367 | union perf_capabilities { | |
368 | struct { | |
369 | u64 lbr_format:6; | |
370 | u64 pebs_trap:1; | |
371 | u64 pebs_arch_reg:1; | |
372 | u64 pebs_format:4; | |
373 | u64 smm_freeze:1; | |
069e0c3c AK |
374 | /* |
375 | * PMU supports separate counter range for writing | |
376 | * values > 32bit. | |
377 | */ | |
378 | u64 full_width_write:1; | |
de0428a7 KW |
379 | }; |
380 | u64 capabilities; | |
381 | }; | |
382 | ||
c1d6f42f PZ |
383 | struct x86_pmu_quirk { |
384 | struct x86_pmu_quirk *next; | |
385 | void (*func)(void); | |
386 | }; | |
387 | ||
f9b4eeb8 PZ |
388 | union x86_pmu_config { |
389 | struct { | |
390 | u64 event:8, | |
391 | umask:8, | |
392 | usr:1, | |
393 | os:1, | |
394 | edge:1, | |
395 | pc:1, | |
396 | interrupt:1, | |
397 | __reserved1:1, | |
398 | en:1, | |
399 | inv:1, | |
400 | cmask:8, | |
401 | event2:4, | |
402 | __reserved2:4, | |
403 | go:1, | |
404 | ho:1; | |
405 | } bits; | |
406 | u64 value; | |
407 | }; | |
408 | ||
409 | #define X86_CONFIG(args...) ((union x86_pmu_config){.bits = {args}}).value | |
410 | ||
48070342 AS |
411 | enum { |
412 | x86_lbr_exclusive_lbr, | |
413 | x86_lbr_exclusive_pt, | |
414 | x86_lbr_exclusive_max, | |
415 | }; | |
416 | ||
de0428a7 KW |
417 | /* |
418 | * struct x86_pmu - generic x86 pmu | |
419 | */ | |
420 | struct x86_pmu { | |
421 | /* | |
422 | * Generic x86 PMC bits | |
423 | */ | |
424 | const char *name; | |
425 | int version; | |
426 | int (*handle_irq)(struct pt_regs *); | |
427 | void (*disable_all)(void); | |
428 | void (*enable_all)(int added); | |
429 | void (*enable)(struct perf_event *); | |
430 | void (*disable)(struct perf_event *); | |
431 | int (*hw_config)(struct perf_event *event); | |
432 | int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign); | |
433 | unsigned eventsel; | |
434 | unsigned perfctr; | |
4c1fd17a | 435 | int (*addr_offset)(int index, bool eventsel); |
0fbdad07 | 436 | int (*rdpmc_index)(int index); |
de0428a7 KW |
437 | u64 (*event_map)(int); |
438 | int max_events; | |
439 | int num_counters; | |
440 | int num_counters_fixed; | |
441 | int cntval_bits; | |
442 | u64 cntval_mask; | |
ffb871bc GN |
443 | union { |
444 | unsigned long events_maskl; | |
445 | unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)]; | |
446 | }; | |
447 | int events_mask_len; | |
de0428a7 KW |
448 | int apic; |
449 | u64 max_period; | |
450 | struct event_constraint * | |
451 | (*get_event_constraints)(struct cpu_hw_events *cpuc, | |
452 | struct perf_event *event); | |
453 | ||
454 | void (*put_event_constraints)(struct cpu_hw_events *cpuc, | |
455 | struct perf_event *event); | |
456 | struct event_constraint *event_constraints; | |
c1d6f42f | 457 | struct x86_pmu_quirk *quirks; |
de0428a7 | 458 | int perfctr_second_write; |
72db5596 | 459 | bool late_ack; |
294fe0f5 | 460 | unsigned (*limit_period)(struct perf_event *event, unsigned l); |
de0428a7 | 461 | |
0c9d42ed PZ |
462 | /* |
463 | * sysfs attrs | |
464 | */ | |
e97df763 | 465 | int attr_rdpmc_broken; |
0c9d42ed | 466 | int attr_rdpmc; |
641cc938 | 467 | struct attribute **format_attrs; |
f20093ee | 468 | struct attribute **event_attrs; |
0c9d42ed | 469 | |
a4747393 | 470 | ssize_t (*events_sysfs_show)(char *page, u64 config); |
1a6461b1 | 471 | struct attribute **cpu_events; |
a4747393 | 472 | |
0c9d42ed PZ |
473 | /* |
474 | * CPU Hotplug hooks | |
475 | */ | |
de0428a7 KW |
476 | int (*cpu_prepare)(int cpu); |
477 | void (*cpu_starting)(int cpu); | |
478 | void (*cpu_dying)(int cpu); | |
479 | void (*cpu_dead)(int cpu); | |
c93dc84c PZ |
480 | |
481 | void (*check_microcode)(void); | |
ba532500 YZ |
482 | void (*sched_task)(struct perf_event_context *ctx, |
483 | bool sched_in); | |
de0428a7 KW |
484 | |
485 | /* | |
486 | * Intel Arch Perfmon v2+ | |
487 | */ | |
488 | u64 intel_ctrl; | |
489 | union perf_capabilities intel_cap; | |
490 | ||
491 | /* | |
492 | * Intel DebugStore bits | |
493 | */ | |
597ed953 | 494 | unsigned int bts :1, |
3e0091e2 PZ |
495 | bts_active :1, |
496 | pebs :1, | |
497 | pebs_active :1, | |
498 | pebs_broken :1; | |
de0428a7 KW |
499 | int pebs_record_size; |
500 | void (*drain_pebs)(struct pt_regs *regs); | |
501 | struct event_constraint *pebs_constraints; | |
0780c927 | 502 | void (*pebs_aliases)(struct perf_event *event); |
70ab7003 | 503 | int max_pebs_events; |
de0428a7 KW |
504 | |
505 | /* | |
506 | * Intel LBR | |
507 | */ | |
508 | unsigned long lbr_tos, lbr_from, lbr_to; /* MSR base regs */ | |
509 | int lbr_nr; /* hardware stack size */ | |
b36817e8 SE |
510 | u64 lbr_sel_mask; /* LBR_SELECT valid bits */ |
511 | const int *lbr_sel_map; /* lbr_select mappings */ | |
b7af41a1 | 512 | bool lbr_double_abort; /* duplicated lbr aborts */ |
de0428a7 | 513 | |
48070342 AS |
514 | /* |
515 | * Intel PT/LBR/BTS are exclusive | |
516 | */ | |
517 | atomic_t lbr_exclusive[x86_lbr_exclusive_max]; | |
518 | ||
de0428a7 KW |
519 | /* |
520 | * Extra registers for events | |
521 | */ | |
522 | struct extra_reg *extra_regs; | |
523 | unsigned int er_flags; | |
144d31e6 GN |
524 | |
525 | /* | |
526 | * Intel host/guest support (KVM) | |
527 | */ | |
528 | struct perf_guest_switch_msr *(*guest_get_msrs)(int *nr); | |
de0428a7 KW |
529 | }; |
530 | ||
e18bf526 YZ |
531 | struct x86_perf_task_context { |
532 | u64 lbr_from[MAX_LBR_ENTRIES]; | |
533 | u64 lbr_to[MAX_LBR_ENTRIES]; | |
534 | int lbr_callstack_users; | |
535 | int lbr_stack_state; | |
536 | }; | |
537 | ||
c1d6f42f PZ |
538 | #define x86_add_quirk(func_) \ |
539 | do { \ | |
540 | static struct x86_pmu_quirk __quirk __initdata = { \ | |
541 | .func = func_, \ | |
542 | }; \ | |
543 | __quirk.next = x86_pmu.quirks; \ | |
544 | x86_pmu.quirks = &__quirk; \ | |
545 | } while (0) | |
546 | ||
de0428a7 KW |
547 | #define ERF_NO_HT_SHARING 1 |
548 | #define ERF_HAS_RSP_1 2 | |
549 | ||
3a54aaa0 SE |
550 | #define EVENT_VAR(_id) event_attr_##_id |
551 | #define EVENT_PTR(_id) &event_attr_##_id.attr.attr | |
552 | ||
553 | #define EVENT_ATTR(_name, _id) \ | |
554 | static struct perf_pmu_events_attr EVENT_VAR(_id) = { \ | |
555 | .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \ | |
556 | .id = PERF_COUNT_HW_##_id, \ | |
557 | .event_str = NULL, \ | |
558 | }; | |
559 | ||
560 | #define EVENT_ATTR_STR(_name, v, str) \ | |
561 | static struct perf_pmu_events_attr event_attr_##v = { \ | |
562 | .attr = __ATTR(_name, 0444, events_sysfs_show, NULL), \ | |
563 | .id = 0, \ | |
564 | .event_str = str, \ | |
565 | }; | |
566 | ||
de0428a7 KW |
567 | extern struct x86_pmu x86_pmu __read_mostly; |
568 | ||
e9d7f7cd YZ |
569 | static inline bool x86_pmu_has_lbr_callstack(void) |
570 | { | |
571 | return x86_pmu.lbr_sel_map && | |
572 | x86_pmu.lbr_sel_map[PERF_SAMPLE_BRANCH_CALL_STACK_SHIFT] > 0; | |
573 | } | |
574 | ||
de0428a7 KW |
575 | DECLARE_PER_CPU(struct cpu_hw_events, cpu_hw_events); |
576 | ||
577 | int x86_perf_event_set_period(struct perf_event *event); | |
578 | ||
579 | /* | |
580 | * Generalized hw caching related hw_event table, filled | |
581 | * in on a per model basis. A value of 0 means | |
582 | * 'not supported', -1 means 'hw_event makes no sense on | |
583 | * this CPU', any other value means the raw hw_event | |
584 | * ID. | |
585 | */ | |
586 | ||
587 | #define C(x) PERF_COUNT_HW_CACHE_##x | |
588 | ||
589 | extern u64 __read_mostly hw_cache_event_ids | |
590 | [PERF_COUNT_HW_CACHE_MAX] | |
591 | [PERF_COUNT_HW_CACHE_OP_MAX] | |
592 | [PERF_COUNT_HW_CACHE_RESULT_MAX]; | |
593 | extern u64 __read_mostly hw_cache_extra_regs | |
594 | [PERF_COUNT_HW_CACHE_MAX] | |
595 | [PERF_COUNT_HW_CACHE_OP_MAX] | |
596 | [PERF_COUNT_HW_CACHE_RESULT_MAX]; | |
597 | ||
598 | u64 x86_perf_event_update(struct perf_event *event); | |
599 | ||
de0428a7 KW |
600 | static inline unsigned int x86_pmu_config_addr(int index) |
601 | { | |
4c1fd17a JS |
602 | return x86_pmu.eventsel + (x86_pmu.addr_offset ? |
603 | x86_pmu.addr_offset(index, true) : index); | |
de0428a7 KW |
604 | } |
605 | ||
606 | static inline unsigned int x86_pmu_event_addr(int index) | |
607 | { | |
4c1fd17a JS |
608 | return x86_pmu.perfctr + (x86_pmu.addr_offset ? |
609 | x86_pmu.addr_offset(index, false) : index); | |
de0428a7 KW |
610 | } |
611 | ||
0fbdad07 JS |
612 | static inline int x86_pmu_rdpmc_index(int index) |
613 | { | |
614 | return x86_pmu.rdpmc_index ? x86_pmu.rdpmc_index(index) : index; | |
615 | } | |
616 | ||
48070342 AS |
617 | int x86_add_exclusive(unsigned int what); |
618 | ||
619 | void x86_del_exclusive(unsigned int what); | |
620 | ||
621 | void hw_perf_lbr_event_destroy(struct perf_event *event); | |
622 | ||
de0428a7 KW |
623 | int x86_setup_perfctr(struct perf_event *event); |
624 | ||
625 | int x86_pmu_hw_config(struct perf_event *event); | |
626 | ||
627 | void x86_pmu_disable_all(void); | |
628 | ||
629 | static inline void __x86_pmu_enable_event(struct hw_perf_event *hwc, | |
630 | u64 enable_mask) | |
631 | { | |
1018faa6 JR |
632 | u64 disable_mask = __this_cpu_read(cpu_hw_events.perf_ctr_virt_mask); |
633 | ||
de0428a7 KW |
634 | if (hwc->extra_reg.reg) |
635 | wrmsrl(hwc->extra_reg.reg, hwc->extra_reg.config); | |
1018faa6 | 636 | wrmsrl(hwc->config_base, (hwc->config | enable_mask) & ~disable_mask); |
de0428a7 KW |
637 | } |
638 | ||
639 | void x86_pmu_enable_all(int added); | |
640 | ||
43b45780 | 641 | int perf_assign_events(struct perf_event **events, int n, |
4b4969b1 | 642 | int wmin, int wmax, int *assign); |
de0428a7 KW |
643 | int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign); |
644 | ||
645 | void x86_pmu_stop(struct perf_event *event, int flags); | |
646 | ||
647 | static inline void x86_pmu_disable_event(struct perf_event *event) | |
648 | { | |
649 | struct hw_perf_event *hwc = &event->hw; | |
650 | ||
651 | wrmsrl(hwc->config_base, hwc->config); | |
652 | } | |
653 | ||
654 | void x86_pmu_enable_event(struct perf_event *event); | |
655 | ||
656 | int x86_pmu_handle_irq(struct pt_regs *regs); | |
657 | ||
658 | extern struct event_constraint emptyconstraint; | |
659 | ||
660 | extern struct event_constraint unconstrained; | |
661 | ||
3e702ff6 SE |
662 | static inline bool kernel_ip(unsigned long ip) |
663 | { | |
664 | #ifdef CONFIG_X86_32 | |
665 | return ip > PAGE_OFFSET; | |
666 | #else | |
667 | return (long)ip < 0; | |
668 | #endif | |
669 | } | |
670 | ||
d07bdfd3 PZ |
671 | /* |
672 | * Not all PMUs provide the right context information to place the reported IP | |
673 | * into full context. Specifically segment registers are typically not | |
674 | * supplied. | |
675 | * | |
676 | * Assuming the address is a linear address (it is for IBS), we fake the CS and | |
677 | * vm86 mode using the known zero-based code segment and 'fix up' the registers | |
678 | * to reflect this. | |
679 | * | |
680 | * Intel PEBS/LBR appear to typically provide the effective address, nothing | |
681 | * much we can do about that but pray and treat it like a linear address. | |
682 | */ | |
683 | static inline void set_linear_ip(struct pt_regs *regs, unsigned long ip) | |
684 | { | |
685 | regs->cs = kernel_ip(ip) ? __KERNEL_CS : __USER_CS; | |
686 | if (regs->flags & X86_VM_MASK) | |
687 | regs->flags ^= (PERF_EFLAGS_VM | X86_VM_MASK); | |
688 | regs->ip = ip; | |
689 | } | |
690 | ||
0bf79d44 | 691 | ssize_t x86_event_sysfs_show(char *page, u64 config, u64 event); |
20550a43 | 692 | ssize_t intel_event_sysfs_show(char *page, u64 config); |
43c032fe | 693 | |
de0428a7 KW |
694 | #ifdef CONFIG_CPU_SUP_AMD |
695 | ||
696 | int amd_pmu_init(void); | |
697 | ||
698 | #else /* CONFIG_CPU_SUP_AMD */ | |
699 | ||
700 | static inline int amd_pmu_init(void) | |
701 | { | |
702 | return 0; | |
703 | } | |
704 | ||
705 | #endif /* CONFIG_CPU_SUP_AMD */ | |
706 | ||
707 | #ifdef CONFIG_CPU_SUP_INTEL | |
708 | ||
48070342 AS |
709 | static inline bool intel_pmu_needs_lbr_smpl(struct perf_event *event) |
710 | { | |
711 | /* user explicitly requested branch sampling */ | |
712 | if (has_branch_stack(event)) | |
713 | return true; | |
714 | ||
715 | /* implicit branch sampling to correct PEBS skid */ | |
716 | if (x86_pmu.intel_cap.pebs_trap && event->attr.precise_ip > 1 && | |
717 | x86_pmu.intel_cap.pebs_format < 2) | |
718 | return true; | |
719 | ||
720 | return false; | |
721 | } | |
722 | ||
723 | static inline bool intel_pmu_has_bts(struct perf_event *event) | |
724 | { | |
725 | if (event->attr.config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS && | |
726 | !event->attr.freq && event->hw.sample_period == 1) | |
727 | return true; | |
728 | ||
729 | return false; | |
730 | } | |
731 | ||
de0428a7 KW |
732 | int intel_pmu_save_and_restart(struct perf_event *event); |
733 | ||
734 | struct event_constraint * | |
735 | x86_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event); | |
736 | ||
737 | struct intel_shared_regs *allocate_shared_regs(int cpu); | |
738 | ||
739 | int intel_pmu_init(void); | |
740 | ||
741 | void init_debug_store_on_cpu(int cpu); | |
742 | ||
743 | void fini_debug_store_on_cpu(int cpu); | |
744 | ||
745 | void release_ds_buffers(void); | |
746 | ||
747 | void reserve_ds_buffers(void); | |
748 | ||
749 | extern struct event_constraint bts_constraint; | |
750 | ||
751 | void intel_pmu_enable_bts(u64 config); | |
752 | ||
753 | void intel_pmu_disable_bts(void); | |
754 | ||
755 | int intel_pmu_drain_bts_buffer(void); | |
756 | ||
757 | extern struct event_constraint intel_core2_pebs_event_constraints[]; | |
758 | ||
759 | extern struct event_constraint intel_atom_pebs_event_constraints[]; | |
760 | ||
1fa64180 YZ |
761 | extern struct event_constraint intel_slm_pebs_event_constraints[]; |
762 | ||
de0428a7 KW |
763 | extern struct event_constraint intel_nehalem_pebs_event_constraints[]; |
764 | ||
765 | extern struct event_constraint intel_westmere_pebs_event_constraints[]; | |
766 | ||
767 | extern struct event_constraint intel_snb_pebs_event_constraints[]; | |
768 | ||
20a36e39 SE |
769 | extern struct event_constraint intel_ivb_pebs_event_constraints[]; |
770 | ||
3044318f AK |
771 | extern struct event_constraint intel_hsw_pebs_event_constraints[]; |
772 | ||
de0428a7 KW |
773 | struct event_constraint *intel_pebs_constraints(struct perf_event *event); |
774 | ||
775 | void intel_pmu_pebs_enable(struct perf_event *event); | |
776 | ||
777 | void intel_pmu_pebs_disable(struct perf_event *event); | |
778 | ||
779 | void intel_pmu_pebs_enable_all(void); | |
780 | ||
781 | void intel_pmu_pebs_disable_all(void); | |
782 | ||
783 | void intel_ds_init(void); | |
784 | ||
2a0ad3b3 YZ |
785 | void intel_pmu_lbr_sched_task(struct perf_event_context *ctx, bool sched_in); |
786 | ||
de0428a7 KW |
787 | void intel_pmu_lbr_reset(void); |
788 | ||
789 | void intel_pmu_lbr_enable(struct perf_event *event); | |
790 | ||
791 | void intel_pmu_lbr_disable(struct perf_event *event); | |
792 | ||
793 | void intel_pmu_lbr_enable_all(void); | |
794 | ||
795 | void intel_pmu_lbr_disable_all(void); | |
796 | ||
797 | void intel_pmu_lbr_read(void); | |
798 | ||
799 | void intel_pmu_lbr_init_core(void); | |
800 | ||
801 | void intel_pmu_lbr_init_nhm(void); | |
802 | ||
803 | void intel_pmu_lbr_init_atom(void); | |
804 | ||
c5cc2cd9 SE |
805 | void intel_pmu_lbr_init_snb(void); |
806 | ||
e9d7f7cd YZ |
807 | void intel_pmu_lbr_init_hsw(void); |
808 | ||
60ce0fbd SE |
809 | int intel_pmu_setup_lbr_filter(struct perf_event *event); |
810 | ||
52ca9ced AS |
811 | void intel_pt_interrupt(void); |
812 | ||
de0428a7 KW |
813 | int p4_pmu_init(void); |
814 | ||
815 | int p6_pmu_init(void); | |
816 | ||
e717bf4e VW |
817 | int knc_pmu_init(void); |
818 | ||
f20093ee SE |
819 | ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr, |
820 | char *page); | |
821 | ||
de0428a7 KW |
822 | #else /* CONFIG_CPU_SUP_INTEL */ |
823 | ||
824 | static inline void reserve_ds_buffers(void) | |
825 | { | |
826 | } | |
827 | ||
828 | static inline void release_ds_buffers(void) | |
829 | { | |
830 | } | |
831 | ||
832 | static inline int intel_pmu_init(void) | |
833 | { | |
834 | return 0; | |
835 | } | |
836 | ||
837 | static inline struct intel_shared_regs *allocate_shared_regs(int cpu) | |
838 | { | |
839 | return NULL; | |
840 | } | |
841 | ||
842 | #endif /* CONFIG_CPU_SUP_INTEL */ |