Commit | Line | Data |
---|---|---|
f22f54f4 PZ |
1 | #ifdef CONFIG_CPU_SUP_INTEL |
2 | ||
3 | /* | |
b622d644 | 4 | * Intel PerfMon, used on Core and later. |
f22f54f4 PZ |
5 | */ |
6 | static 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 | ||
17 | static 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 | ||
28 | static 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 | ||
51 | static 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 | ||
67 | static 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 | ||
78 | static 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 | ||
86 | static u64 intel_pmu_event_map(int hw_event) | |
87 | { | |
88 | return intel_perfmon_event_map[hw_event]; | |
89 | } | |
90 | ||
91 | static __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 | ||
182 | static __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 | ||
273 | static __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 | ||
364 | static __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 | ||
455 | static 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 |
473 | static 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(); | |
f22f54f4 PZ |
483 | } |
484 | ||
485 | static void intel_pmu_enable_all(void) | |
486 | { | |
487 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); | |
488 | ||
489 | wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl); | |
490 | ||
491 | if (test_bit(X86_PMC_IDX_FIXED_BTS, cpuc->active_mask)) { | |
492 | struct perf_event *event = | |
493 | cpuc->events[X86_PMC_IDX_FIXED_BTS]; | |
494 | ||
495 | if (WARN_ON_ONCE(!event)) | |
496 | return; | |
497 | ||
498 | intel_pmu_enable_bts(event->hw.config); | |
499 | } | |
ca037701 PZ |
500 | |
501 | intel_pmu_pebs_enable_all(); | |
f22f54f4 PZ |
502 | } |
503 | ||
504 | static inline u64 intel_pmu_get_status(void) | |
505 | { | |
506 | u64 status; | |
507 | ||
508 | rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status); | |
509 | ||
510 | return status; | |
511 | } | |
512 | ||
513 | static inline void intel_pmu_ack_status(u64 ack) | |
514 | { | |
515 | wrmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack); | |
516 | } | |
517 | ||
ca037701 | 518 | static void intel_pmu_disable_fixed(struct hw_perf_event *hwc) |
f22f54f4 | 519 | { |
aff3d91a | 520 | int idx = hwc->idx - X86_PMC_IDX_FIXED; |
f22f54f4 PZ |
521 | u64 ctrl_val, mask; |
522 | ||
523 | mask = 0xfULL << (idx * 4); | |
524 | ||
525 | rdmsrl(hwc->config_base, ctrl_val); | |
526 | ctrl_val &= ~mask; | |
527 | (void)checking_wrmsrl(hwc->config_base, ctrl_val); | |
528 | } | |
529 | ||
ca037701 | 530 | static void intel_pmu_disable_event(struct perf_event *event) |
f22f54f4 | 531 | { |
aff3d91a PZ |
532 | struct hw_perf_event *hwc = &event->hw; |
533 | ||
534 | if (unlikely(hwc->idx == X86_PMC_IDX_FIXED_BTS)) { | |
f22f54f4 PZ |
535 | intel_pmu_disable_bts(); |
536 | intel_pmu_drain_bts_buffer(); | |
537 | return; | |
538 | } | |
539 | ||
540 | if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) { | |
aff3d91a | 541 | intel_pmu_disable_fixed(hwc); |
f22f54f4 PZ |
542 | return; |
543 | } | |
544 | ||
aff3d91a | 545 | x86_pmu_disable_event(event); |
ca037701 PZ |
546 | |
547 | if (unlikely(event->attr.precise)) | |
548 | intel_pmu_pebs_disable(hwc); | |
f22f54f4 PZ |
549 | } |
550 | ||
ca037701 | 551 | static void intel_pmu_enable_fixed(struct hw_perf_event *hwc) |
f22f54f4 | 552 | { |
aff3d91a | 553 | int idx = hwc->idx - X86_PMC_IDX_FIXED; |
f22f54f4 PZ |
554 | u64 ctrl_val, bits, mask; |
555 | int err; | |
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; | |
580 | err = checking_wrmsrl(hwc->config_base, ctrl_val); | |
581 | } | |
582 | ||
aff3d91a | 583 | static 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 PZ |
600 | if (unlikely(event->attr.precise)) |
601 | intel_pmu_pebs_enable(hwc); | |
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 | */ | |
610 | static 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 | ||
616 | static 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 | */ | |
646 | static 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; | |
666 | again: | |
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 PZ |
676 | |
677 | /* | |
678 | * PEBS overflow sets bit 62 in the global status register | |
679 | */ | |
680 | if (__test_and_clear_bit(62, (unsigned long *)&status)) | |
681 | x86_pmu.drain_pebs(regs); | |
682 | ||
984b3f57 | 683 | for_each_set_bit(bit, (unsigned long *)&status, X86_PMC_IDX_MAX) { |
f22f54f4 PZ |
684 | struct perf_event *event = cpuc->events[bit]; |
685 | ||
f22f54f4 PZ |
686 | if (!test_bit(bit, cpuc->active_mask)) |
687 | continue; | |
688 | ||
689 | if (!intel_pmu_save_and_restart(event)) | |
690 | continue; | |
691 | ||
692 | data.period = event->hw.last_period; | |
693 | ||
694 | if (perf_event_overflow(event, 1, &data, regs)) | |
71e2d282 | 695 | x86_pmu_stop(event); |
f22f54f4 PZ |
696 | } |
697 | ||
698 | intel_pmu_ack_status(ack); | |
699 | ||
700 | /* | |
701 | * Repeat if there is more work to be done: | |
702 | */ | |
703 | status = intel_pmu_get_status(); | |
704 | if (status) | |
705 | goto again; | |
706 | ||
3fb2b8dd PZ |
707 | done: |
708 | intel_pmu_enable_all(); | |
f22f54f4 PZ |
709 | return 1; |
710 | } | |
711 | ||
f22f54f4 | 712 | static struct event_constraint * |
ca037701 | 713 | intel_bts_constraints(struct perf_event *event) |
f22f54f4 | 714 | { |
ca037701 PZ |
715 | struct hw_perf_event *hwc = &event->hw; |
716 | unsigned int hw_event, bts_event; | |
f22f54f4 | 717 | |
ca037701 PZ |
718 | hw_event = hwc->config & INTEL_ARCH_EVENT_MASK; |
719 | bts_event = x86_pmu.event_map(PERF_COUNT_HW_BRANCH_INSTRUCTIONS); | |
f22f54f4 | 720 | |
ca037701 | 721 | if (unlikely(hw_event == bts_event && hwc->sample_period == 1)) |
f22f54f4 | 722 | return &bts_constraint; |
ca037701 | 723 | |
f22f54f4 PZ |
724 | return NULL; |
725 | } | |
726 | ||
727 | static struct event_constraint * | |
728 | intel_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event) | |
729 | { | |
730 | struct event_constraint *c; | |
731 | ||
ca037701 PZ |
732 | c = intel_bts_constraints(event); |
733 | if (c) | |
734 | return c; | |
735 | ||
736 | c = intel_pebs_constraints(event); | |
f22f54f4 PZ |
737 | if (c) |
738 | return c; | |
739 | ||
740 | return x86_get_event_constraints(cpuc, event); | |
741 | } | |
742 | ||
743 | static __initconst struct x86_pmu core_pmu = { | |
744 | .name = "core", | |
745 | .handle_irq = x86_pmu_handle_irq, | |
746 | .disable_all = x86_pmu_disable_all, | |
747 | .enable_all = x86_pmu_enable_all, | |
748 | .enable = x86_pmu_enable_event, | |
749 | .disable = x86_pmu_disable_event, | |
750 | .eventsel = MSR_ARCH_PERFMON_EVENTSEL0, | |
751 | .perfctr = MSR_ARCH_PERFMON_PERFCTR0, | |
752 | .event_map = intel_pmu_event_map, | |
753 | .raw_event = intel_pmu_raw_event, | |
754 | .max_events = ARRAY_SIZE(intel_perfmon_event_map), | |
755 | .apic = 1, | |
756 | /* | |
757 | * Intel PMCs cannot be accessed sanely above 32 bit width, | |
758 | * so we install an artificial 1<<31 period regardless of | |
759 | * the generic event period: | |
760 | */ | |
761 | .max_period = (1ULL << 31) - 1, | |
762 | .get_event_constraints = intel_get_event_constraints, | |
763 | .event_constraints = intel_core_event_constraints, | |
764 | }; | |
765 | ||
766 | static __initconst struct x86_pmu intel_pmu = { | |
767 | .name = "Intel", | |
768 | .handle_irq = intel_pmu_handle_irq, | |
769 | .disable_all = intel_pmu_disable_all, | |
770 | .enable_all = intel_pmu_enable_all, | |
771 | .enable = intel_pmu_enable_event, | |
772 | .disable = intel_pmu_disable_event, | |
773 | .eventsel = MSR_ARCH_PERFMON_EVENTSEL0, | |
774 | .perfctr = MSR_ARCH_PERFMON_PERFCTR0, | |
775 | .event_map = intel_pmu_event_map, | |
776 | .raw_event = intel_pmu_raw_event, | |
777 | .max_events = ARRAY_SIZE(intel_perfmon_event_map), | |
778 | .apic = 1, | |
779 | /* | |
780 | * Intel PMCs cannot be accessed sanely above 32 bit width, | |
781 | * so we install an artificial 1<<31 period regardless of | |
782 | * the generic event period: | |
783 | */ | |
784 | .max_period = (1ULL << 31) - 1, | |
3f6da390 PZ |
785 | .get_event_constraints = intel_get_event_constraints, |
786 | ||
787 | .cpu_starting = init_debug_store_on_cpu, | |
788 | .cpu_dying = fini_debug_store_on_cpu, | |
f22f54f4 PZ |
789 | }; |
790 | ||
791 | static __init int intel_pmu_init(void) | |
792 | { | |
793 | union cpuid10_edx edx; | |
794 | union cpuid10_eax eax; | |
795 | unsigned int unused; | |
796 | unsigned int ebx; | |
797 | int version; | |
798 | ||
799 | if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) { | |
800 | /* check for P6 processor family */ | |
801 | if (boot_cpu_data.x86 == 6) { | |
802 | return p6_pmu_init(); | |
803 | } else { | |
804 | return -ENODEV; | |
805 | } | |
806 | } | |
807 | ||
808 | /* | |
809 | * Check whether the Architectural PerfMon supports | |
810 | * Branch Misses Retired hw_event or not. | |
811 | */ | |
812 | cpuid(10, &eax.full, &ebx, &unused, &edx.full); | |
813 | if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED) | |
814 | return -ENODEV; | |
815 | ||
816 | version = eax.split.version_id; | |
817 | if (version < 2) | |
818 | x86_pmu = core_pmu; | |
819 | else | |
820 | x86_pmu = intel_pmu; | |
821 | ||
822 | x86_pmu.version = version; | |
823 | x86_pmu.num_events = eax.split.num_events; | |
824 | x86_pmu.event_bits = eax.split.bit_width; | |
825 | x86_pmu.event_mask = (1ULL << eax.split.bit_width) - 1; | |
826 | ||
827 | /* | |
828 | * Quirk: v2 perfmon does not report fixed-purpose events, so | |
829 | * assume at least 3 events: | |
830 | */ | |
831 | if (version > 1) | |
832 | x86_pmu.num_events_fixed = max((int)edx.split.num_events_fixed, 3); | |
833 | ||
ca037701 PZ |
834 | intel_ds_init(); |
835 | ||
f22f54f4 PZ |
836 | /* |
837 | * Install the hw-cache-events table: | |
838 | */ | |
839 | switch (boot_cpu_data.x86_model) { | |
840 | case 14: /* 65 nm core solo/duo, "Yonah" */ | |
841 | pr_cont("Core events, "); | |
842 | break; | |
843 | ||
844 | case 15: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */ | |
845 | case 22: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */ | |
846 | case 23: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */ | |
847 | case 29: /* six-core 45 nm xeon "Dunnington" */ | |
848 | memcpy(hw_cache_event_ids, core2_hw_cache_event_ids, | |
849 | sizeof(hw_cache_event_ids)); | |
850 | ||
851 | x86_pmu.event_constraints = intel_core2_event_constraints; | |
852 | pr_cont("Core2 events, "); | |
853 | break; | |
854 | ||
855 | case 26: /* 45 nm nehalem, "Bloomfield" */ | |
856 | case 30: /* 45 nm nehalem, "Lynnfield" */ | |
857 | memcpy(hw_cache_event_ids, nehalem_hw_cache_event_ids, | |
858 | sizeof(hw_cache_event_ids)); | |
859 | ||
860 | x86_pmu.event_constraints = intel_nehalem_event_constraints; | |
861 | pr_cont("Nehalem/Corei7 events, "); | |
862 | break; | |
b622d644 | 863 | case 28: /* Atom */ |
f22f54f4 PZ |
864 | memcpy(hw_cache_event_ids, atom_hw_cache_event_ids, |
865 | sizeof(hw_cache_event_ids)); | |
866 | ||
867 | x86_pmu.event_constraints = intel_gen_event_constraints; | |
868 | pr_cont("Atom events, "); | |
869 | break; | |
870 | ||
871 | case 37: /* 32 nm nehalem, "Clarkdale" */ | |
872 | case 44: /* 32 nm nehalem, "Gulftown" */ | |
873 | memcpy(hw_cache_event_ids, westmere_hw_cache_event_ids, | |
874 | sizeof(hw_cache_event_ids)); | |
875 | ||
876 | x86_pmu.event_constraints = intel_westmere_event_constraints; | |
877 | pr_cont("Westmere events, "); | |
878 | break; | |
b622d644 | 879 | |
f22f54f4 PZ |
880 | default: |
881 | /* | |
882 | * default constraints for v2 and up | |
883 | */ | |
884 | x86_pmu.event_constraints = intel_gen_event_constraints; | |
885 | pr_cont("generic architected perfmon, "); | |
886 | } | |
887 | return 0; | |
888 | } | |
889 | ||
890 | #else /* CONFIG_CPU_SUP_INTEL */ | |
891 | ||
892 | static int intel_pmu_init(void) | |
893 | { | |
894 | return 0; | |
895 | } | |
896 | ||
897 | #endif /* CONFIG_CPU_SUP_INTEL */ |