x86, ptrace: Fix block-step
[deliverable/linux.git] / arch / x86 / kernel / cpu / perf_event_intel.c
CommitLineData
f22f54f4
PZ
1#ifdef CONFIG_CPU_SUP_INTEL
2
3/*
b622d644 4 * Intel PerfMon, used on Core and later.
f22f54f4
PZ
5 */
6static const u64 intel_perfmon_event_map[] =
7{
8 [PERF_COUNT_HW_CPU_CYCLES] = 0x003c,
9 [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0,
10 [PERF_COUNT_HW_CACHE_REFERENCES] = 0x4f2e,
11 [PERF_COUNT_HW_CACHE_MISSES] = 0x412e,
12 [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4,
13 [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5,
14 [PERF_COUNT_HW_BUS_CYCLES] = 0x013c,
15};
16
17static struct event_constraint intel_core_event_constraints[] =
18{
19 INTEL_EVENT_CONSTRAINT(0x11, 0x2), /* FP_ASSIST */
20 INTEL_EVENT_CONSTRAINT(0x12, 0x2), /* MUL */
21 INTEL_EVENT_CONSTRAINT(0x13, 0x2), /* DIV */
22 INTEL_EVENT_CONSTRAINT(0x14, 0x1), /* CYCLES_DIV_BUSY */
23 INTEL_EVENT_CONSTRAINT(0x19, 0x2), /* DELAYED_BYPASS */
24 INTEL_EVENT_CONSTRAINT(0xc1, 0x1), /* FP_COMP_INSTR_RET */
25 EVENT_CONSTRAINT_END
26};
27
28static struct event_constraint intel_core2_event_constraints[] =
29{
b622d644
PZ
30 FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
31 FIXED_EVENT_CONSTRAINT(0x003c, 1), /* CPU_CLK_UNHALTED.CORE */
32 /*
33 * Core2 has Fixed Counter 2 listed as CPU_CLK_UNHALTED.REF and event
34 * 0x013c as CPU_CLK_UNHALTED.BUS and specifies there is a fixed
35 * ratio between these counters.
36 */
37 /* FIXED_EVENT_CONSTRAINT(0x013c, 2), CPU_CLK_UNHALTED.REF */
f22f54f4
PZ
38 INTEL_EVENT_CONSTRAINT(0x10, 0x1), /* FP_COMP_OPS_EXE */
39 INTEL_EVENT_CONSTRAINT(0x11, 0x2), /* FP_ASSIST */
40 INTEL_EVENT_CONSTRAINT(0x12, 0x2), /* MUL */
41 INTEL_EVENT_CONSTRAINT(0x13, 0x2), /* DIV */
42 INTEL_EVENT_CONSTRAINT(0x14, 0x1), /* CYCLES_DIV_BUSY */
43 INTEL_EVENT_CONSTRAINT(0x18, 0x1), /* IDLE_DURING_DIV */
44 INTEL_EVENT_CONSTRAINT(0x19, 0x2), /* DELAYED_BYPASS */
45 INTEL_EVENT_CONSTRAINT(0xa1, 0x1), /* RS_UOPS_DISPATCH_CYCLES */
b622d644 46 INTEL_EVENT_CONSTRAINT(0xc9, 0x1), /* ITLB_MISS_RETIRED (T30-9) */
f22f54f4
PZ
47 INTEL_EVENT_CONSTRAINT(0xcb, 0x1), /* MEM_LOAD_RETIRED */
48 EVENT_CONSTRAINT_END
49};
50
51static struct event_constraint intel_nehalem_event_constraints[] =
52{
b622d644
PZ
53 FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
54 FIXED_EVENT_CONSTRAINT(0x003c, 1), /* CPU_CLK_UNHALTED.CORE */
55 /* FIXED_EVENT_CONSTRAINT(0x013c, 2), CPU_CLK_UNHALTED.REF */
f22f54f4
PZ
56 INTEL_EVENT_CONSTRAINT(0x40, 0x3), /* L1D_CACHE_LD */
57 INTEL_EVENT_CONSTRAINT(0x41, 0x3), /* L1D_CACHE_ST */
58 INTEL_EVENT_CONSTRAINT(0x42, 0x3), /* L1D_CACHE_LOCK */
59 INTEL_EVENT_CONSTRAINT(0x43, 0x3), /* L1D_ALL_REF */
60 INTEL_EVENT_CONSTRAINT(0x48, 0x3), /* L1D_PEND_MISS */
61 INTEL_EVENT_CONSTRAINT(0x4e, 0x3), /* L1D_PREFETCH */
62 INTEL_EVENT_CONSTRAINT(0x51, 0x3), /* L1D */
63 INTEL_EVENT_CONSTRAINT(0x63, 0x3), /* CACHE_LOCK_CYCLES */
64 EVENT_CONSTRAINT_END
65};
66
67static struct event_constraint intel_westmere_event_constraints[] =
68{
b622d644
PZ
69 FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
70 FIXED_EVENT_CONSTRAINT(0x003c, 1), /* CPU_CLK_UNHALTED.CORE */
71 /* FIXED_EVENT_CONSTRAINT(0x013c, 2), CPU_CLK_UNHALTED.REF */
f22f54f4
PZ
72 INTEL_EVENT_CONSTRAINT(0x51, 0x3), /* L1D */
73 INTEL_EVENT_CONSTRAINT(0x60, 0x1), /* OFFCORE_REQUESTS_OUTSTANDING */
74 INTEL_EVENT_CONSTRAINT(0x63, 0x3), /* CACHE_LOCK_CYCLES */
75 EVENT_CONSTRAINT_END
76};
77
78static struct event_constraint intel_gen_event_constraints[] =
79{
b622d644
PZ
80 FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
81 FIXED_EVENT_CONSTRAINT(0x003c, 1), /* CPU_CLK_UNHALTED.CORE */
82 /* FIXED_EVENT_CONSTRAINT(0x013c, 2), CPU_CLK_UNHALTED.REF */
f22f54f4
PZ
83 EVENT_CONSTRAINT_END
84};
85
86static u64 intel_pmu_event_map(int hw_event)
87{
88 return intel_perfmon_event_map[hw_event];
89}
90
91static __initconst u64 westmere_hw_cache_event_ids
92 [PERF_COUNT_HW_CACHE_MAX]
93 [PERF_COUNT_HW_CACHE_OP_MAX]
94 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
95{
96 [ C(L1D) ] = {
97 [ C(OP_READ) ] = {
98 [ C(RESULT_ACCESS) ] = 0x010b, /* MEM_INST_RETIRED.LOADS */
99 [ C(RESULT_MISS) ] = 0x0151, /* L1D.REPL */
100 },
101 [ C(OP_WRITE) ] = {
102 [ C(RESULT_ACCESS) ] = 0x020b, /* MEM_INST_RETURED.STORES */
103 [ C(RESULT_MISS) ] = 0x0251, /* L1D.M_REPL */
104 },
105 [ C(OP_PREFETCH) ] = {
106 [ C(RESULT_ACCESS) ] = 0x014e, /* L1D_PREFETCH.REQUESTS */
107 [ C(RESULT_MISS) ] = 0x024e, /* L1D_PREFETCH.MISS */
108 },
109 },
110 [ C(L1I ) ] = {
111 [ C(OP_READ) ] = {
112 [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */
113 [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */
114 },
115 [ C(OP_WRITE) ] = {
116 [ C(RESULT_ACCESS) ] = -1,
117 [ C(RESULT_MISS) ] = -1,
118 },
119 [ C(OP_PREFETCH) ] = {
120 [ C(RESULT_ACCESS) ] = 0x0,
121 [ C(RESULT_MISS) ] = 0x0,
122 },
123 },
124 [ C(LL ) ] = {
125 [ C(OP_READ) ] = {
126 [ C(RESULT_ACCESS) ] = 0x0324, /* L2_RQSTS.LOADS */
127 [ C(RESULT_MISS) ] = 0x0224, /* L2_RQSTS.LD_MISS */
128 },
129 [ C(OP_WRITE) ] = {
130 [ C(RESULT_ACCESS) ] = 0x0c24, /* L2_RQSTS.RFOS */
131 [ C(RESULT_MISS) ] = 0x0824, /* L2_RQSTS.RFO_MISS */
132 },
133 [ C(OP_PREFETCH) ] = {
134 [ C(RESULT_ACCESS) ] = 0x4f2e, /* LLC Reference */
135 [ C(RESULT_MISS) ] = 0x412e, /* LLC Misses */
136 },
137 },
138 [ C(DTLB) ] = {
139 [ C(OP_READ) ] = {
140 [ C(RESULT_ACCESS) ] = 0x010b, /* MEM_INST_RETIRED.LOADS */
141 [ C(RESULT_MISS) ] = 0x0108, /* DTLB_LOAD_MISSES.ANY */
142 },
143 [ C(OP_WRITE) ] = {
144 [ C(RESULT_ACCESS) ] = 0x020b, /* MEM_INST_RETURED.STORES */
145 [ C(RESULT_MISS) ] = 0x010c, /* MEM_STORE_RETIRED.DTLB_MISS */
146 },
147 [ C(OP_PREFETCH) ] = {
148 [ C(RESULT_ACCESS) ] = 0x0,
149 [ C(RESULT_MISS) ] = 0x0,
150 },
151 },
152 [ C(ITLB) ] = {
153 [ C(OP_READ) ] = {
154 [ C(RESULT_ACCESS) ] = 0x01c0, /* INST_RETIRED.ANY_P */
155 [ C(RESULT_MISS) ] = 0x0185, /* ITLB_MISSES.ANY */
156 },
157 [ C(OP_WRITE) ] = {
158 [ C(RESULT_ACCESS) ] = -1,
159 [ C(RESULT_MISS) ] = -1,
160 },
161 [ C(OP_PREFETCH) ] = {
162 [ C(RESULT_ACCESS) ] = -1,
163 [ C(RESULT_MISS) ] = -1,
164 },
165 },
166 [ C(BPU ) ] = {
167 [ C(OP_READ) ] = {
168 [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ALL_BRANCHES */
169 [ C(RESULT_MISS) ] = 0x03e8, /* BPU_CLEARS.ANY */
170 },
171 [ C(OP_WRITE) ] = {
172 [ C(RESULT_ACCESS) ] = -1,
173 [ C(RESULT_MISS) ] = -1,
174 },
175 [ C(OP_PREFETCH) ] = {
176 [ C(RESULT_ACCESS) ] = -1,
177 [ C(RESULT_MISS) ] = -1,
178 },
179 },
180};
181
182static __initconst u64 nehalem_hw_cache_event_ids
183 [PERF_COUNT_HW_CACHE_MAX]
184 [PERF_COUNT_HW_CACHE_OP_MAX]
185 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
186{
187 [ C(L1D) ] = {
188 [ C(OP_READ) ] = {
189 [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */
190 [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */
191 },
192 [ C(OP_WRITE) ] = {
193 [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */
194 [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */
195 },
196 [ C(OP_PREFETCH) ] = {
197 [ C(RESULT_ACCESS) ] = 0x014e, /* L1D_PREFETCH.REQUESTS */
198 [ C(RESULT_MISS) ] = 0x024e, /* L1D_PREFETCH.MISS */
199 },
200 },
201 [ C(L1I ) ] = {
202 [ C(OP_READ) ] = {
203 [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */
204 [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */
205 },
206 [ C(OP_WRITE) ] = {
207 [ C(RESULT_ACCESS) ] = -1,
208 [ C(RESULT_MISS) ] = -1,
209 },
210 [ C(OP_PREFETCH) ] = {
211 [ C(RESULT_ACCESS) ] = 0x0,
212 [ C(RESULT_MISS) ] = 0x0,
213 },
214 },
215 [ C(LL ) ] = {
216 [ C(OP_READ) ] = {
217 [ C(RESULT_ACCESS) ] = 0x0324, /* L2_RQSTS.LOADS */
218 [ C(RESULT_MISS) ] = 0x0224, /* L2_RQSTS.LD_MISS */
219 },
220 [ C(OP_WRITE) ] = {
221 [ C(RESULT_ACCESS) ] = 0x0c24, /* L2_RQSTS.RFOS */
222 [ C(RESULT_MISS) ] = 0x0824, /* L2_RQSTS.RFO_MISS */
223 },
224 [ C(OP_PREFETCH) ] = {
225 [ C(RESULT_ACCESS) ] = 0x4f2e, /* LLC Reference */
226 [ C(RESULT_MISS) ] = 0x412e, /* LLC Misses */
227 },
228 },
229 [ C(DTLB) ] = {
230 [ C(OP_READ) ] = {
231 [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */
232 [ C(RESULT_MISS) ] = 0x0108, /* DTLB_LOAD_MISSES.ANY */
233 },
234 [ C(OP_WRITE) ] = {
235 [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */
236 [ C(RESULT_MISS) ] = 0x010c, /* MEM_STORE_RETIRED.DTLB_MISS */
237 },
238 [ C(OP_PREFETCH) ] = {
239 [ C(RESULT_ACCESS) ] = 0x0,
240 [ C(RESULT_MISS) ] = 0x0,
241 },
242 },
243 [ C(ITLB) ] = {
244 [ C(OP_READ) ] = {
245 [ C(RESULT_ACCESS) ] = 0x01c0, /* INST_RETIRED.ANY_P */
246 [ C(RESULT_MISS) ] = 0x20c8, /* ITLB_MISS_RETIRED */
247 },
248 [ C(OP_WRITE) ] = {
249 [ C(RESULT_ACCESS) ] = -1,
250 [ C(RESULT_MISS) ] = -1,
251 },
252 [ C(OP_PREFETCH) ] = {
253 [ C(RESULT_ACCESS) ] = -1,
254 [ C(RESULT_MISS) ] = -1,
255 },
256 },
257 [ C(BPU ) ] = {
258 [ C(OP_READ) ] = {
259 [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ALL_BRANCHES */
260 [ C(RESULT_MISS) ] = 0x03e8, /* BPU_CLEARS.ANY */
261 },
262 [ C(OP_WRITE) ] = {
263 [ C(RESULT_ACCESS) ] = -1,
264 [ C(RESULT_MISS) ] = -1,
265 },
266 [ C(OP_PREFETCH) ] = {
267 [ C(RESULT_ACCESS) ] = -1,
268 [ C(RESULT_MISS) ] = -1,
269 },
270 },
271};
272
273static __initconst u64 core2_hw_cache_event_ids
274 [PERF_COUNT_HW_CACHE_MAX]
275 [PERF_COUNT_HW_CACHE_OP_MAX]
276 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
277{
278 [ C(L1D) ] = {
279 [ C(OP_READ) ] = {
280 [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */
281 [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */
282 },
283 [ C(OP_WRITE) ] = {
284 [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */
285 [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */
286 },
287 [ C(OP_PREFETCH) ] = {
288 [ C(RESULT_ACCESS) ] = 0x104e, /* L1D_PREFETCH.REQUESTS */
289 [ C(RESULT_MISS) ] = 0,
290 },
291 },
292 [ C(L1I ) ] = {
293 [ C(OP_READ) ] = {
294 [ C(RESULT_ACCESS) ] = 0x0080, /* L1I.READS */
295 [ C(RESULT_MISS) ] = 0x0081, /* L1I.MISSES */
296 },
297 [ C(OP_WRITE) ] = {
298 [ C(RESULT_ACCESS) ] = -1,
299 [ C(RESULT_MISS) ] = -1,
300 },
301 [ C(OP_PREFETCH) ] = {
302 [ C(RESULT_ACCESS) ] = 0,
303 [ C(RESULT_MISS) ] = 0,
304 },
305 },
306 [ C(LL ) ] = {
307 [ C(OP_READ) ] = {
308 [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */
309 [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */
310 },
311 [ C(OP_WRITE) ] = {
312 [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */
313 [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */
314 },
315 [ C(OP_PREFETCH) ] = {
316 [ C(RESULT_ACCESS) ] = 0,
317 [ C(RESULT_MISS) ] = 0,
318 },
319 },
320 [ C(DTLB) ] = {
321 [ C(OP_READ) ] = {
322 [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */
323 [ C(RESULT_MISS) ] = 0x0208, /* DTLB_MISSES.MISS_LD */
324 },
325 [ C(OP_WRITE) ] = {
326 [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */
327 [ C(RESULT_MISS) ] = 0x0808, /* DTLB_MISSES.MISS_ST */
328 },
329 [ C(OP_PREFETCH) ] = {
330 [ C(RESULT_ACCESS) ] = 0,
331 [ C(RESULT_MISS) ] = 0,
332 },
333 },
334 [ C(ITLB) ] = {
335 [ C(OP_READ) ] = {
336 [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */
337 [ C(RESULT_MISS) ] = 0x1282, /* ITLBMISSES */
338 },
339 [ C(OP_WRITE) ] = {
340 [ C(RESULT_ACCESS) ] = -1,
341 [ C(RESULT_MISS) ] = -1,
342 },
343 [ C(OP_PREFETCH) ] = {
344 [ C(RESULT_ACCESS) ] = -1,
345 [ C(RESULT_MISS) ] = -1,
346 },
347 },
348 [ C(BPU ) ] = {
349 [ C(OP_READ) ] = {
350 [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */
351 [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */
352 },
353 [ C(OP_WRITE) ] = {
354 [ C(RESULT_ACCESS) ] = -1,
355 [ C(RESULT_MISS) ] = -1,
356 },
357 [ C(OP_PREFETCH) ] = {
358 [ C(RESULT_ACCESS) ] = -1,
359 [ C(RESULT_MISS) ] = -1,
360 },
361 },
362};
363
364static __initconst u64 atom_hw_cache_event_ids
365 [PERF_COUNT_HW_CACHE_MAX]
366 [PERF_COUNT_HW_CACHE_OP_MAX]
367 [PERF_COUNT_HW_CACHE_RESULT_MAX] =
368{
369 [ C(L1D) ] = {
370 [ C(OP_READ) ] = {
371 [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE.LD */
372 [ C(RESULT_MISS) ] = 0,
373 },
374 [ C(OP_WRITE) ] = {
375 [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE.ST */
376 [ C(RESULT_MISS) ] = 0,
377 },
378 [ C(OP_PREFETCH) ] = {
379 [ C(RESULT_ACCESS) ] = 0x0,
380 [ C(RESULT_MISS) ] = 0,
381 },
382 },
383 [ C(L1I ) ] = {
384 [ C(OP_READ) ] = {
385 [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */
386 [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */
387 },
388 [ C(OP_WRITE) ] = {
389 [ C(RESULT_ACCESS) ] = -1,
390 [ C(RESULT_MISS) ] = -1,
391 },
392 [ C(OP_PREFETCH) ] = {
393 [ C(RESULT_ACCESS) ] = 0,
394 [ C(RESULT_MISS) ] = 0,
395 },
396 },
397 [ C(LL ) ] = {
398 [ C(OP_READ) ] = {
399 [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */
400 [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */
401 },
402 [ C(OP_WRITE) ] = {
403 [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */
404 [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */
405 },
406 [ C(OP_PREFETCH) ] = {
407 [ C(RESULT_ACCESS) ] = 0,
408 [ C(RESULT_MISS) ] = 0,
409 },
410 },
411 [ C(DTLB) ] = {
412 [ C(OP_READ) ] = {
413 [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE_LD.MESI (alias) */
414 [ C(RESULT_MISS) ] = 0x0508, /* DTLB_MISSES.MISS_LD */
415 },
416 [ C(OP_WRITE) ] = {
417 [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE_ST.MESI (alias) */
418 [ C(RESULT_MISS) ] = 0x0608, /* DTLB_MISSES.MISS_ST */
419 },
420 [ C(OP_PREFETCH) ] = {
421 [ C(RESULT_ACCESS) ] = 0,
422 [ C(RESULT_MISS) ] = 0,
423 },
424 },
425 [ C(ITLB) ] = {
426 [ C(OP_READ) ] = {
427 [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */
428 [ C(RESULT_MISS) ] = 0x0282, /* ITLB.MISSES */
429 },
430 [ C(OP_WRITE) ] = {
431 [ C(RESULT_ACCESS) ] = -1,
432 [ C(RESULT_MISS) ] = -1,
433 },
434 [ C(OP_PREFETCH) ] = {
435 [ C(RESULT_ACCESS) ] = -1,
436 [ C(RESULT_MISS) ] = -1,
437 },
438 },
439 [ C(BPU ) ] = {
440 [ C(OP_READ) ] = {
441 [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */
442 [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */
443 },
444 [ C(OP_WRITE) ] = {
445 [ C(RESULT_ACCESS) ] = -1,
446 [ C(RESULT_MISS) ] = -1,
447 },
448 [ C(OP_PREFETCH) ] = {
449 [ C(RESULT_ACCESS) ] = -1,
450 [ C(RESULT_MISS) ] = -1,
451 },
452 },
453};
454
455static u64 intel_pmu_raw_event(u64 hw_event)
456{
457#define CORE_EVNTSEL_EVENT_MASK 0x000000FFULL
458#define CORE_EVNTSEL_UNIT_MASK 0x0000FF00ULL
459#define CORE_EVNTSEL_EDGE_MASK 0x00040000ULL
460#define CORE_EVNTSEL_INV_MASK 0x00800000ULL
461#define CORE_EVNTSEL_REG_MASK 0xFF000000ULL
462
463#define CORE_EVNTSEL_MASK \
464 (INTEL_ARCH_EVTSEL_MASK | \
465 INTEL_ARCH_UNIT_MASK | \
466 INTEL_ARCH_EDGE_MASK | \
467 INTEL_ARCH_INV_MASK | \
468 INTEL_ARCH_CNT_MASK)
469
470 return hw_event & CORE_EVNTSEL_MASK;
471}
472
f22f54f4
PZ
473static void intel_pmu_disable_all(void)
474{
475 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
476
477 wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0);
478
479 if (test_bit(X86_PMC_IDX_FIXED_BTS, cpuc->active_mask))
480 intel_pmu_disable_bts();
ca037701
PZ
481
482 intel_pmu_pebs_disable_all();
caff2bef 483 intel_pmu_lbr_disable_all();
f22f54f4
PZ
484}
485
486static void intel_pmu_enable_all(void)
487{
488 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
489
d329527e
PZ
490 intel_pmu_pebs_enable_all();
491 intel_pmu_lbr_enable_all();
f22f54f4
PZ
492 wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl);
493
494 if (test_bit(X86_PMC_IDX_FIXED_BTS, cpuc->active_mask)) {
495 struct perf_event *event =
496 cpuc->events[X86_PMC_IDX_FIXED_BTS];
497
498 if (WARN_ON_ONCE(!event))
499 return;
500
501 intel_pmu_enable_bts(event->hw.config);
502 }
503}
504
505static inline u64 intel_pmu_get_status(void)
506{
507 u64 status;
508
509 rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
510
511 return status;
512}
513
514static inline void intel_pmu_ack_status(u64 ack)
515{
516 wrmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack);
517}
518
ca037701 519static void intel_pmu_disable_fixed(struct hw_perf_event *hwc)
f22f54f4 520{
aff3d91a 521 int idx = hwc->idx - X86_PMC_IDX_FIXED;
f22f54f4
PZ
522 u64 ctrl_val, mask;
523
524 mask = 0xfULL << (idx * 4);
525
526 rdmsrl(hwc->config_base, ctrl_val);
527 ctrl_val &= ~mask;
7645a24c 528 wrmsrl(hwc->config_base, ctrl_val);
f22f54f4
PZ
529}
530
ca037701 531static void intel_pmu_disable_event(struct perf_event *event)
f22f54f4 532{
aff3d91a
PZ
533 struct hw_perf_event *hwc = &event->hw;
534
535 if (unlikely(hwc->idx == X86_PMC_IDX_FIXED_BTS)) {
f22f54f4
PZ
536 intel_pmu_disable_bts();
537 intel_pmu_drain_bts_buffer();
538 return;
539 }
540
541 if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) {
aff3d91a 542 intel_pmu_disable_fixed(hwc);
f22f54f4
PZ
543 return;
544 }
545
aff3d91a 546 x86_pmu_disable_event(event);
ca037701
PZ
547
548 if (unlikely(event->attr.precise))
ef21f683 549 intel_pmu_pebs_disable(event);
f22f54f4
PZ
550}
551
ca037701 552static void intel_pmu_enable_fixed(struct hw_perf_event *hwc)
f22f54f4 553{
aff3d91a 554 int idx = hwc->idx - X86_PMC_IDX_FIXED;
f22f54f4 555 u64 ctrl_val, bits, mask;
f22f54f4
PZ
556
557 /*
558 * Enable IRQ generation (0x8),
559 * and enable ring-3 counting (0x2) and ring-0 counting (0x1)
560 * if requested:
561 */
562 bits = 0x8ULL;
563 if (hwc->config & ARCH_PERFMON_EVENTSEL_USR)
564 bits |= 0x2;
565 if (hwc->config & ARCH_PERFMON_EVENTSEL_OS)
566 bits |= 0x1;
567
568 /*
569 * ANY bit is supported in v3 and up
570 */
571 if (x86_pmu.version > 2 && hwc->config & ARCH_PERFMON_EVENTSEL_ANY)
572 bits |= 0x4;
573
574 bits <<= (idx * 4);
575 mask = 0xfULL << (idx * 4);
576
577 rdmsrl(hwc->config_base, ctrl_val);
578 ctrl_val &= ~mask;
579 ctrl_val |= bits;
7645a24c 580 wrmsrl(hwc->config_base, ctrl_val);
f22f54f4
PZ
581}
582
aff3d91a 583static void intel_pmu_enable_event(struct perf_event *event)
f22f54f4 584{
aff3d91a
PZ
585 struct hw_perf_event *hwc = &event->hw;
586
587 if (unlikely(hwc->idx == X86_PMC_IDX_FIXED_BTS)) {
f22f54f4
PZ
588 if (!__get_cpu_var(cpu_hw_events).enabled)
589 return;
590
591 intel_pmu_enable_bts(hwc->config);
592 return;
593 }
594
595 if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) {
aff3d91a 596 intel_pmu_enable_fixed(hwc);
f22f54f4
PZ
597 return;
598 }
599
ca037701 600 if (unlikely(event->attr.precise))
ef21f683 601 intel_pmu_pebs_enable(event);
ca037701 602
aff3d91a 603 __x86_pmu_enable_event(hwc);
f22f54f4
PZ
604}
605
606/*
607 * Save and restart an expired event. Called by NMI contexts,
608 * so it has to be careful about preempting normal event ops:
609 */
610static int intel_pmu_save_and_restart(struct perf_event *event)
611{
cc2ad4ba
PZ
612 x86_perf_event_update(event);
613 return x86_perf_event_set_period(event);
f22f54f4
PZ
614}
615
616static void intel_pmu_reset(void)
617{
618 struct debug_store *ds = __get_cpu_var(cpu_hw_events).ds;
619 unsigned long flags;
620 int idx;
621
622 if (!x86_pmu.num_events)
623 return;
624
625 local_irq_save(flags);
626
627 printk("clearing PMU state on CPU#%d\n", smp_processor_id());
628
629 for (idx = 0; idx < x86_pmu.num_events; idx++) {
630 checking_wrmsrl(x86_pmu.eventsel + idx, 0ull);
631 checking_wrmsrl(x86_pmu.perfctr + idx, 0ull);
632 }
633 for (idx = 0; idx < x86_pmu.num_events_fixed; idx++) {
634 checking_wrmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, 0ull);
635 }
636 if (ds)
637 ds->bts_index = ds->bts_buffer_base;
638
639 local_irq_restore(flags);
640}
641
642/*
643 * This handler is triggered by the local APIC, so the APIC IRQ handling
644 * rules apply:
645 */
646static int intel_pmu_handle_irq(struct pt_regs *regs)
647{
648 struct perf_sample_data data;
649 struct cpu_hw_events *cpuc;
650 int bit, loops;
651 u64 ack, status;
652
dc1d628a 653 perf_sample_data_init(&data, 0);
f22f54f4
PZ
654
655 cpuc = &__get_cpu_var(cpu_hw_events);
656
3fb2b8dd 657 intel_pmu_disable_all();
f22f54f4
PZ
658 intel_pmu_drain_bts_buffer();
659 status = intel_pmu_get_status();
660 if (!status) {
3fb2b8dd 661 intel_pmu_enable_all();
f22f54f4
PZ
662 return 0;
663 }
664
665 loops = 0;
666again:
667 if (++loops > 100) {
668 WARN_ONCE(1, "perfevents: irq loop stuck!\n");
669 perf_event_print_debug();
670 intel_pmu_reset();
3fb2b8dd 671 goto done;
f22f54f4
PZ
672 }
673
674 inc_irq_stat(apic_perf_irqs);
675 ack = status;
ca037701 676
caff2bef
PZ
677 intel_pmu_lbr_read();
678
ca037701
PZ
679 /*
680 * PEBS overflow sets bit 62 in the global status register
681 */
682 if (__test_and_clear_bit(62, (unsigned long *)&status))
683 x86_pmu.drain_pebs(regs);
684
984b3f57 685 for_each_set_bit(bit, (unsigned long *)&status, X86_PMC_IDX_MAX) {
f22f54f4
PZ
686 struct perf_event *event = cpuc->events[bit];
687
f22f54f4
PZ
688 if (!test_bit(bit, cpuc->active_mask))
689 continue;
690
691 if (!intel_pmu_save_and_restart(event))
692 continue;
693
694 data.period = event->hw.last_period;
695
696 if (perf_event_overflow(event, 1, &data, regs))
71e2d282 697 x86_pmu_stop(event);
f22f54f4
PZ
698 }
699
700 intel_pmu_ack_status(ack);
701
702 /*
703 * Repeat if there is more work to be done:
704 */
705 status = intel_pmu_get_status();
706 if (status)
707 goto again;
708
3fb2b8dd
PZ
709done:
710 intel_pmu_enable_all();
f22f54f4
PZ
711 return 1;
712}
713
f22f54f4 714static struct event_constraint *
ca037701 715intel_bts_constraints(struct perf_event *event)
f22f54f4 716{
ca037701
PZ
717 struct hw_perf_event *hwc = &event->hw;
718 unsigned int hw_event, bts_event;
f22f54f4 719
ca037701
PZ
720 hw_event = hwc->config & INTEL_ARCH_EVENT_MASK;
721 bts_event = x86_pmu.event_map(PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
f22f54f4 722
ca037701 723 if (unlikely(hw_event == bts_event && hwc->sample_period == 1))
f22f54f4 724 return &bts_constraint;
ca037701 725
f22f54f4
PZ
726 return NULL;
727}
728
729static struct event_constraint *
730intel_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
731{
732 struct event_constraint *c;
733
ca037701
PZ
734 c = intel_bts_constraints(event);
735 if (c)
736 return c;
737
738 c = intel_pebs_constraints(event);
f22f54f4
PZ
739 if (c)
740 return c;
741
742 return x86_get_event_constraints(cpuc, event);
743}
744
745static __initconst struct x86_pmu core_pmu = {
746 .name = "core",
747 .handle_irq = x86_pmu_handle_irq,
748 .disable_all = x86_pmu_disable_all,
749 .enable_all = x86_pmu_enable_all,
750 .enable = x86_pmu_enable_event,
751 .disable = x86_pmu_disable_event,
a072738e
CG
752 .hw_config = x86_hw_config,
753 .schedule_events = x86_schedule_events,
f22f54f4
PZ
754 .eventsel = MSR_ARCH_PERFMON_EVENTSEL0,
755 .perfctr = MSR_ARCH_PERFMON_PERFCTR0,
756 .event_map = intel_pmu_event_map,
757 .raw_event = intel_pmu_raw_event,
758 .max_events = ARRAY_SIZE(intel_perfmon_event_map),
759 .apic = 1,
760 /*
761 * Intel PMCs cannot be accessed sanely above 32 bit width,
762 * so we install an artificial 1<<31 period regardless of
763 * the generic event period:
764 */
765 .max_period = (1ULL << 31) - 1,
766 .get_event_constraints = intel_get_event_constraints,
767 .event_constraints = intel_core_event_constraints,
768};
769
74846d35
PZ
770static void intel_pmu_cpu_starting(int cpu)
771{
772 init_debug_store_on_cpu(cpu);
773 /*
774 * Deal with CPUs that don't clear their LBRs on power-up.
775 */
776 intel_pmu_lbr_reset();
777}
778
779static void intel_pmu_cpu_dying(int cpu)
780{
781 fini_debug_store_on_cpu(cpu);
782}
783
f22f54f4
PZ
784static __initconst struct x86_pmu intel_pmu = {
785 .name = "Intel",
786 .handle_irq = intel_pmu_handle_irq,
787 .disable_all = intel_pmu_disable_all,
788 .enable_all = intel_pmu_enable_all,
789 .enable = intel_pmu_enable_event,
790 .disable = intel_pmu_disable_event,
a072738e
CG
791 .hw_config = x86_hw_config,
792 .schedule_events = x86_schedule_events,
f22f54f4
PZ
793 .eventsel = MSR_ARCH_PERFMON_EVENTSEL0,
794 .perfctr = MSR_ARCH_PERFMON_PERFCTR0,
795 .event_map = intel_pmu_event_map,
796 .raw_event = intel_pmu_raw_event,
797 .max_events = ARRAY_SIZE(intel_perfmon_event_map),
798 .apic = 1,
799 /*
800 * Intel PMCs cannot be accessed sanely above 32 bit width,
801 * so we install an artificial 1<<31 period regardless of
802 * the generic event period:
803 */
804 .max_period = (1ULL << 31) - 1,
3f6da390
PZ
805 .get_event_constraints = intel_get_event_constraints,
806
74846d35
PZ
807 .cpu_starting = intel_pmu_cpu_starting,
808 .cpu_dying = intel_pmu_cpu_dying,
f22f54f4
PZ
809};
810
3c44780b
PZ
811static void intel_clovertown_quirks(void)
812{
813 /*
814 * PEBS is unreliable due to:
815 *
816 * AJ67 - PEBS may experience CPL leaks
817 * AJ68 - PEBS PMI may be delayed by one event
818 * AJ69 - GLOBAL_STATUS[62] will only be set when DEBUGCTL[12]
819 * AJ106 - FREEZE_LBRS_ON_PMI doesn't work in combination with PEBS
820 *
821 * AJ67 could be worked around by restricting the OS/USR flags.
822 * AJ69 could be worked around by setting PMU_FREEZE_ON_PMI.
823 *
824 * AJ106 could possibly be worked around by not allowing LBR
825 * usage from PEBS, including the fixup.
826 * AJ68 could possibly be worked around by always programming
827 * a pebs_event_reset[0] value and coping with the lost events.
828 *
829 * But taken together it might just make sense to not enable PEBS on
830 * these chips.
831 */
832 printk(KERN_WARNING "PEBS disabled due to CPU errata.\n");
833 x86_pmu.pebs = 0;
834 x86_pmu.pebs_constraints = NULL;
835}
836
f22f54f4
PZ
837static __init int intel_pmu_init(void)
838{
839 union cpuid10_edx edx;
840 union cpuid10_eax eax;
841 unsigned int unused;
842 unsigned int ebx;
843 int version;
844
845 if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) {
a072738e
CG
846 switch (boot_cpu_data.x86) {
847 case 0x6:
848 return p6_pmu_init();
849 case 0xf:
850 return p4_pmu_init();
851 }
f22f54f4 852 return -ENODEV;
f22f54f4
PZ
853 }
854
855 /*
856 * Check whether the Architectural PerfMon supports
857 * Branch Misses Retired hw_event or not.
858 */
859 cpuid(10, &eax.full, &ebx, &unused, &edx.full);
860 if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED)
861 return -ENODEV;
862
863 version = eax.split.version_id;
864 if (version < 2)
865 x86_pmu = core_pmu;
866 else
867 x86_pmu = intel_pmu;
868
869 x86_pmu.version = version;
870 x86_pmu.num_events = eax.split.num_events;
871 x86_pmu.event_bits = eax.split.bit_width;
872 x86_pmu.event_mask = (1ULL << eax.split.bit_width) - 1;
873
874 /*
875 * Quirk: v2 perfmon does not report fixed-purpose events, so
876 * assume at least 3 events:
877 */
878 if (version > 1)
879 x86_pmu.num_events_fixed = max((int)edx.split.num_events_fixed, 3);
880
8db909a7
PZ
881 /*
882 * v2 and above have a perf capabilities MSR
883 */
884 if (version > 1) {
885 u64 capabilities;
886
887 rdmsrl(MSR_IA32_PERF_CAPABILITIES, capabilities);
888 x86_pmu.intel_cap.capabilities = capabilities;
889 }
890
ca037701
PZ
891 intel_ds_init();
892
f22f54f4
PZ
893 /*
894 * Install the hw-cache-events table:
895 */
896 switch (boot_cpu_data.x86_model) {
897 case 14: /* 65 nm core solo/duo, "Yonah" */
898 pr_cont("Core events, ");
899 break;
900
901 case 15: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */
3c44780b 902 x86_pmu.quirks = intel_clovertown_quirks;
f22f54f4
PZ
903 case 22: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */
904 case 23: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */
905 case 29: /* six-core 45 nm xeon "Dunnington" */
906 memcpy(hw_cache_event_ids, core2_hw_cache_event_ids,
907 sizeof(hw_cache_event_ids));
908
caff2bef
PZ
909 intel_pmu_lbr_init_core();
910
f22f54f4
PZ
911 x86_pmu.event_constraints = intel_core2_event_constraints;
912 pr_cont("Core2 events, ");
913 break;
914
915 case 26: /* 45 nm nehalem, "Bloomfield" */
916 case 30: /* 45 nm nehalem, "Lynnfield" */
917 memcpy(hw_cache_event_ids, nehalem_hw_cache_event_ids,
918 sizeof(hw_cache_event_ids));
919
caff2bef
PZ
920 intel_pmu_lbr_init_nhm();
921
f22f54f4
PZ
922 x86_pmu.event_constraints = intel_nehalem_event_constraints;
923 pr_cont("Nehalem/Corei7 events, ");
924 break;
caff2bef 925
b622d644 926 case 28: /* Atom */
f22f54f4
PZ
927 memcpy(hw_cache_event_ids, atom_hw_cache_event_ids,
928 sizeof(hw_cache_event_ids));
929
caff2bef
PZ
930 intel_pmu_lbr_init_atom();
931
f22f54f4
PZ
932 x86_pmu.event_constraints = intel_gen_event_constraints;
933 pr_cont("Atom events, ");
934 break;
935
936 case 37: /* 32 nm nehalem, "Clarkdale" */
937 case 44: /* 32 nm nehalem, "Gulftown" */
938 memcpy(hw_cache_event_ids, westmere_hw_cache_event_ids,
939 sizeof(hw_cache_event_ids));
940
caff2bef
PZ
941 intel_pmu_lbr_init_nhm();
942
f22f54f4
PZ
943 x86_pmu.event_constraints = intel_westmere_event_constraints;
944 pr_cont("Westmere events, ");
945 break;
b622d644 946
f22f54f4
PZ
947 default:
948 /*
949 * default constraints for v2 and up
950 */
951 x86_pmu.event_constraints = intel_gen_event_constraints;
952 pr_cont("generic architected perfmon, ");
953 }
954 return 0;
955}
956
957#else /* CONFIG_CPU_SUP_INTEL */
958
959static int intel_pmu_init(void)
960{
961 return 0;
962}
963
964#endif /* CONFIG_CPU_SUP_INTEL */
This page took 0.073104 seconds and 5 git commands to generate.