]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - arch/x86/events/intel/cstate.c
Documentation/l1tf: Fix small spelling typo
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / events / intel / cstate.c
1 /*
2 * Support cstate residency counters
3 *
4 * Copyright (C) 2015, Intel Corp.
5 * Author: Kan Liang (kan.liang@intel.com)
6 *
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Library General Public
9 * License as published by the Free Software Foundation; either
10 * version 2 of the License, or (at your option) any later version.
11 *
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Library General Public License for more details.
16 *
17 */
18
19 /*
20 * This file export cstate related free running (read-only) counters
21 * for perf. These counters may be use simultaneously by other tools,
22 * such as turbostat. However, it still make sense to implement them
23 * in perf. Because we can conveniently collect them together with
24 * other events, and allow to use them from tools without special MSR
25 * access code.
26 *
27 * The events only support system-wide mode counting. There is no
28 * sampling support because it is not supported by the hardware.
29 *
30 * According to counters' scope and category, two PMUs are registered
31 * with the perf_event core subsystem.
32 * - 'cstate_core': The counter is available for each physical core.
33 * The counters include CORE_C*_RESIDENCY.
34 * - 'cstate_pkg': The counter is available for each physical package.
35 * The counters include PKG_C*_RESIDENCY.
36 *
37 * All of these counters are specified in the IntelĀ® 64 and IA-32
38 * Architectures Software Developer.s Manual Vol3b.
39 *
40 * Model specific counters:
41 * MSR_CORE_C1_RES: CORE C1 Residency Counter
42 * perf code: 0x00
43 * Available model: SLM,AMT,GLM
44 * Scope: Core (each processor core has a MSR)
45 * MSR_CORE_C3_RESIDENCY: CORE C3 Residency Counter
46 * perf code: 0x01
47 * Available model: NHM,WSM,SNB,IVB,HSW,BDW,SKL,GLM
48 * Scope: Core
49 * MSR_CORE_C6_RESIDENCY: CORE C6 Residency Counter
50 * perf code: 0x02
51 * Available model: SLM,AMT,NHM,WSM,SNB,IVB,HSW,BDW
52 * SKL,KNL,GLM
53 * Scope: Core
54 * MSR_CORE_C7_RESIDENCY: CORE C7 Residency Counter
55 * perf code: 0x03
56 * Available model: SNB,IVB,HSW,BDW,SKL
57 * Scope: Core
58 * MSR_PKG_C2_RESIDENCY: Package C2 Residency Counter.
59 * perf code: 0x00
60 * Available model: SNB,IVB,HSW,BDW,SKL,KNL,GLM
61 * Scope: Package (physical package)
62 * MSR_PKG_C3_RESIDENCY: Package C3 Residency Counter.
63 * perf code: 0x01
64 * Available model: NHM,WSM,SNB,IVB,HSW,BDW,SKL,KNL
65 * GLM
66 * Scope: Package (physical package)
67 * MSR_PKG_C6_RESIDENCY: Package C6 Residency Counter.
68 * perf code: 0x02
69 * Available model: SLM,AMT,NHM,WSM,SNB,IVB,HSW,BDW
70 * SKL,KNL,GLM
71 * Scope: Package (physical package)
72 * MSR_PKG_C7_RESIDENCY: Package C7 Residency Counter.
73 * perf code: 0x03
74 * Available model: NHM,WSM,SNB,IVB,HSW,BDW,SKL
75 * Scope: Package (physical package)
76 * MSR_PKG_C8_RESIDENCY: Package C8 Residency Counter.
77 * perf code: 0x04
78 * Available model: HSW ULT only
79 * Scope: Package (physical package)
80 * MSR_PKG_C9_RESIDENCY: Package C9 Residency Counter.
81 * perf code: 0x05
82 * Available model: HSW ULT only
83 * Scope: Package (physical package)
84 * MSR_PKG_C10_RESIDENCY: Package C10 Residency Counter.
85 * perf code: 0x06
86 * Available model: HSW ULT, GLM
87 * Scope: Package (physical package)
88 *
89 */
90
91 #include <linux/module.h>
92 #include <linux/slab.h>
93 #include <linux/perf_event.h>
94 #include <linux/nospec.h>
95 #include <asm/cpu_device_id.h>
96 #include <asm/intel-family.h>
97 #include "../perf_event.h"
98
99 MODULE_LICENSE("GPL");
100
101 #define DEFINE_CSTATE_FORMAT_ATTR(_var, _name, _format) \
102 static ssize_t __cstate_##_var##_show(struct kobject *kobj, \
103 struct kobj_attribute *attr, \
104 char *page) \
105 { \
106 BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \
107 return sprintf(page, _format "\n"); \
108 } \
109 static struct kobj_attribute format_attr_##_var = \
110 __ATTR(_name, 0444, __cstate_##_var##_show, NULL)
111
112 static ssize_t cstate_get_attr_cpumask(struct device *dev,
113 struct device_attribute *attr,
114 char *buf);
115
116 /* Model -> events mapping */
117 struct cstate_model {
118 unsigned long core_events;
119 unsigned long pkg_events;
120 unsigned long quirks;
121 };
122
123 /* Quirk flags */
124 #define SLM_PKG_C6_USE_C7_MSR (1UL << 0)
125 #define KNL_CORE_C6_MSR (1UL << 1)
126
127 struct perf_cstate_msr {
128 u64 msr;
129 struct perf_pmu_events_attr *attr;
130 };
131
132
133 /* cstate_core PMU */
134 static struct pmu cstate_core_pmu;
135 static bool has_cstate_core;
136
137 enum perf_cstate_core_events {
138 PERF_CSTATE_CORE_C1_RES = 0,
139 PERF_CSTATE_CORE_C3_RES,
140 PERF_CSTATE_CORE_C6_RES,
141 PERF_CSTATE_CORE_C7_RES,
142
143 PERF_CSTATE_CORE_EVENT_MAX,
144 };
145
146 PMU_EVENT_ATTR_STRING(c1-residency, evattr_cstate_core_c1, "event=0x00");
147 PMU_EVENT_ATTR_STRING(c3-residency, evattr_cstate_core_c3, "event=0x01");
148 PMU_EVENT_ATTR_STRING(c6-residency, evattr_cstate_core_c6, "event=0x02");
149 PMU_EVENT_ATTR_STRING(c7-residency, evattr_cstate_core_c7, "event=0x03");
150
151 static struct perf_cstate_msr core_msr[] = {
152 [PERF_CSTATE_CORE_C1_RES] = { MSR_CORE_C1_RES, &evattr_cstate_core_c1 },
153 [PERF_CSTATE_CORE_C3_RES] = { MSR_CORE_C3_RESIDENCY, &evattr_cstate_core_c3 },
154 [PERF_CSTATE_CORE_C6_RES] = { MSR_CORE_C6_RESIDENCY, &evattr_cstate_core_c6 },
155 [PERF_CSTATE_CORE_C7_RES] = { MSR_CORE_C7_RESIDENCY, &evattr_cstate_core_c7 },
156 };
157
158 static struct attribute *core_events_attrs[PERF_CSTATE_CORE_EVENT_MAX + 1] = {
159 NULL,
160 };
161
162 static struct attribute_group core_events_attr_group = {
163 .name = "events",
164 .attrs = core_events_attrs,
165 };
166
167 DEFINE_CSTATE_FORMAT_ATTR(core_event, event, "config:0-63");
168 static struct attribute *core_format_attrs[] = {
169 &format_attr_core_event.attr,
170 NULL,
171 };
172
173 static struct attribute_group core_format_attr_group = {
174 .name = "format",
175 .attrs = core_format_attrs,
176 };
177
178 static cpumask_t cstate_core_cpu_mask;
179 static DEVICE_ATTR(cpumask, S_IRUGO, cstate_get_attr_cpumask, NULL);
180
181 static struct attribute *cstate_cpumask_attrs[] = {
182 &dev_attr_cpumask.attr,
183 NULL,
184 };
185
186 static struct attribute_group cpumask_attr_group = {
187 .attrs = cstate_cpumask_attrs,
188 };
189
190 static const struct attribute_group *core_attr_groups[] = {
191 &core_events_attr_group,
192 &core_format_attr_group,
193 &cpumask_attr_group,
194 NULL,
195 };
196
197 /* cstate_pkg PMU */
198 static struct pmu cstate_pkg_pmu;
199 static bool has_cstate_pkg;
200
201 enum perf_cstate_pkg_events {
202 PERF_CSTATE_PKG_C2_RES = 0,
203 PERF_CSTATE_PKG_C3_RES,
204 PERF_CSTATE_PKG_C6_RES,
205 PERF_CSTATE_PKG_C7_RES,
206 PERF_CSTATE_PKG_C8_RES,
207 PERF_CSTATE_PKG_C9_RES,
208 PERF_CSTATE_PKG_C10_RES,
209
210 PERF_CSTATE_PKG_EVENT_MAX,
211 };
212
213 PMU_EVENT_ATTR_STRING(c2-residency, evattr_cstate_pkg_c2, "event=0x00");
214 PMU_EVENT_ATTR_STRING(c3-residency, evattr_cstate_pkg_c3, "event=0x01");
215 PMU_EVENT_ATTR_STRING(c6-residency, evattr_cstate_pkg_c6, "event=0x02");
216 PMU_EVENT_ATTR_STRING(c7-residency, evattr_cstate_pkg_c7, "event=0x03");
217 PMU_EVENT_ATTR_STRING(c8-residency, evattr_cstate_pkg_c8, "event=0x04");
218 PMU_EVENT_ATTR_STRING(c9-residency, evattr_cstate_pkg_c9, "event=0x05");
219 PMU_EVENT_ATTR_STRING(c10-residency, evattr_cstate_pkg_c10, "event=0x06");
220
221 static struct perf_cstate_msr pkg_msr[] = {
222 [PERF_CSTATE_PKG_C2_RES] = { MSR_PKG_C2_RESIDENCY, &evattr_cstate_pkg_c2 },
223 [PERF_CSTATE_PKG_C3_RES] = { MSR_PKG_C3_RESIDENCY, &evattr_cstate_pkg_c3 },
224 [PERF_CSTATE_PKG_C6_RES] = { MSR_PKG_C6_RESIDENCY, &evattr_cstate_pkg_c6 },
225 [PERF_CSTATE_PKG_C7_RES] = { MSR_PKG_C7_RESIDENCY, &evattr_cstate_pkg_c7 },
226 [PERF_CSTATE_PKG_C8_RES] = { MSR_PKG_C8_RESIDENCY, &evattr_cstate_pkg_c8 },
227 [PERF_CSTATE_PKG_C9_RES] = { MSR_PKG_C9_RESIDENCY, &evattr_cstate_pkg_c9 },
228 [PERF_CSTATE_PKG_C10_RES] = { MSR_PKG_C10_RESIDENCY, &evattr_cstate_pkg_c10 },
229 };
230
231 static struct attribute *pkg_events_attrs[PERF_CSTATE_PKG_EVENT_MAX + 1] = {
232 NULL,
233 };
234
235 static struct attribute_group pkg_events_attr_group = {
236 .name = "events",
237 .attrs = pkg_events_attrs,
238 };
239
240 DEFINE_CSTATE_FORMAT_ATTR(pkg_event, event, "config:0-63");
241 static struct attribute *pkg_format_attrs[] = {
242 &format_attr_pkg_event.attr,
243 NULL,
244 };
245 static struct attribute_group pkg_format_attr_group = {
246 .name = "format",
247 .attrs = pkg_format_attrs,
248 };
249
250 static cpumask_t cstate_pkg_cpu_mask;
251
252 static const struct attribute_group *pkg_attr_groups[] = {
253 &pkg_events_attr_group,
254 &pkg_format_attr_group,
255 &cpumask_attr_group,
256 NULL,
257 };
258
259 static ssize_t cstate_get_attr_cpumask(struct device *dev,
260 struct device_attribute *attr,
261 char *buf)
262 {
263 struct pmu *pmu = dev_get_drvdata(dev);
264
265 if (pmu == &cstate_core_pmu)
266 return cpumap_print_to_pagebuf(true, buf, &cstate_core_cpu_mask);
267 else if (pmu == &cstate_pkg_pmu)
268 return cpumap_print_to_pagebuf(true, buf, &cstate_pkg_cpu_mask);
269 else
270 return 0;
271 }
272
273 static int cstate_pmu_event_init(struct perf_event *event)
274 {
275 u64 cfg = event->attr.config;
276 int cpu;
277
278 if (event->attr.type != event->pmu->type)
279 return -ENOENT;
280
281 /* unsupported modes and filters */
282 if (event->attr.exclude_user ||
283 event->attr.exclude_kernel ||
284 event->attr.exclude_hv ||
285 event->attr.exclude_idle ||
286 event->attr.exclude_host ||
287 event->attr.exclude_guest ||
288 event->attr.sample_period) /* no sampling */
289 return -EINVAL;
290
291 if (event->cpu < 0)
292 return -EINVAL;
293
294 if (event->pmu == &cstate_core_pmu) {
295 if (cfg >= PERF_CSTATE_CORE_EVENT_MAX)
296 return -EINVAL;
297 if (!core_msr[cfg].attr)
298 return -EINVAL;
299 event->hw.event_base = core_msr[cfg].msr;
300 cpu = cpumask_any_and(&cstate_core_cpu_mask,
301 topology_sibling_cpumask(event->cpu));
302 } else if (event->pmu == &cstate_pkg_pmu) {
303 if (cfg >= PERF_CSTATE_PKG_EVENT_MAX)
304 return -EINVAL;
305 cfg = array_index_nospec((unsigned long)cfg, PERF_CSTATE_PKG_EVENT_MAX);
306 if (!pkg_msr[cfg].attr)
307 return -EINVAL;
308 event->hw.event_base = pkg_msr[cfg].msr;
309 cpu = cpumask_any_and(&cstate_pkg_cpu_mask,
310 topology_core_cpumask(event->cpu));
311 } else {
312 return -ENOENT;
313 }
314
315 if (cpu >= nr_cpu_ids)
316 return -ENODEV;
317
318 event->cpu = cpu;
319 event->hw.config = cfg;
320 event->hw.idx = -1;
321 return 0;
322 }
323
324 static inline u64 cstate_pmu_read_counter(struct perf_event *event)
325 {
326 u64 val;
327
328 rdmsrl(event->hw.event_base, val);
329 return val;
330 }
331
332 static void cstate_pmu_event_update(struct perf_event *event)
333 {
334 struct hw_perf_event *hwc = &event->hw;
335 u64 prev_raw_count, new_raw_count;
336
337 again:
338 prev_raw_count = local64_read(&hwc->prev_count);
339 new_raw_count = cstate_pmu_read_counter(event);
340
341 if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
342 new_raw_count) != prev_raw_count)
343 goto again;
344
345 local64_add(new_raw_count - prev_raw_count, &event->count);
346 }
347
348 static void cstate_pmu_event_start(struct perf_event *event, int mode)
349 {
350 local64_set(&event->hw.prev_count, cstate_pmu_read_counter(event));
351 }
352
353 static void cstate_pmu_event_stop(struct perf_event *event, int mode)
354 {
355 cstate_pmu_event_update(event);
356 }
357
358 static void cstate_pmu_event_del(struct perf_event *event, int mode)
359 {
360 cstate_pmu_event_stop(event, PERF_EF_UPDATE);
361 }
362
363 static int cstate_pmu_event_add(struct perf_event *event, int mode)
364 {
365 if (mode & PERF_EF_START)
366 cstate_pmu_event_start(event, mode);
367
368 return 0;
369 }
370
371 /*
372 * Check if exiting cpu is the designated reader. If so migrate the
373 * events when there is a valid target available
374 */
375 static int cstate_cpu_exit(unsigned int cpu)
376 {
377 unsigned int target;
378
379 if (has_cstate_core &&
380 cpumask_test_and_clear_cpu(cpu, &cstate_core_cpu_mask)) {
381
382 target = cpumask_any_but(topology_sibling_cpumask(cpu), cpu);
383 /* Migrate events if there is a valid target */
384 if (target < nr_cpu_ids) {
385 cpumask_set_cpu(target, &cstate_core_cpu_mask);
386 perf_pmu_migrate_context(&cstate_core_pmu, cpu, target);
387 }
388 }
389
390 if (has_cstate_pkg &&
391 cpumask_test_and_clear_cpu(cpu, &cstate_pkg_cpu_mask)) {
392
393 target = cpumask_any_but(topology_core_cpumask(cpu), cpu);
394 /* Migrate events if there is a valid target */
395 if (target < nr_cpu_ids) {
396 cpumask_set_cpu(target, &cstate_pkg_cpu_mask);
397 perf_pmu_migrate_context(&cstate_pkg_pmu, cpu, target);
398 }
399 }
400 return 0;
401 }
402
403 static int cstate_cpu_init(unsigned int cpu)
404 {
405 unsigned int target;
406
407 /*
408 * If this is the first online thread of that core, set it in
409 * the core cpu mask as the designated reader.
410 */
411 target = cpumask_any_and(&cstate_core_cpu_mask,
412 topology_sibling_cpumask(cpu));
413
414 if (has_cstate_core && target >= nr_cpu_ids)
415 cpumask_set_cpu(cpu, &cstate_core_cpu_mask);
416
417 /*
418 * If this is the first online thread of that package, set it
419 * in the package cpu mask as the designated reader.
420 */
421 target = cpumask_any_and(&cstate_pkg_cpu_mask,
422 topology_core_cpumask(cpu));
423 if (has_cstate_pkg && target >= nr_cpu_ids)
424 cpumask_set_cpu(cpu, &cstate_pkg_cpu_mask);
425
426 return 0;
427 }
428
429 static struct pmu cstate_core_pmu = {
430 .attr_groups = core_attr_groups,
431 .name = "cstate_core",
432 .task_ctx_nr = perf_invalid_context,
433 .event_init = cstate_pmu_event_init,
434 .add = cstate_pmu_event_add,
435 .del = cstate_pmu_event_del,
436 .start = cstate_pmu_event_start,
437 .stop = cstate_pmu_event_stop,
438 .read = cstate_pmu_event_update,
439 .capabilities = PERF_PMU_CAP_NO_INTERRUPT,
440 .module = THIS_MODULE,
441 };
442
443 static struct pmu cstate_pkg_pmu = {
444 .attr_groups = pkg_attr_groups,
445 .name = "cstate_pkg",
446 .task_ctx_nr = perf_invalid_context,
447 .event_init = cstate_pmu_event_init,
448 .add = cstate_pmu_event_add,
449 .del = cstate_pmu_event_del,
450 .start = cstate_pmu_event_start,
451 .stop = cstate_pmu_event_stop,
452 .read = cstate_pmu_event_update,
453 .capabilities = PERF_PMU_CAP_NO_INTERRUPT,
454 .module = THIS_MODULE,
455 };
456
457 static const struct cstate_model nhm_cstates __initconst = {
458 .core_events = BIT(PERF_CSTATE_CORE_C3_RES) |
459 BIT(PERF_CSTATE_CORE_C6_RES),
460
461 .pkg_events = BIT(PERF_CSTATE_PKG_C3_RES) |
462 BIT(PERF_CSTATE_PKG_C6_RES) |
463 BIT(PERF_CSTATE_PKG_C7_RES),
464 };
465
466 static const struct cstate_model snb_cstates __initconst = {
467 .core_events = BIT(PERF_CSTATE_CORE_C3_RES) |
468 BIT(PERF_CSTATE_CORE_C6_RES) |
469 BIT(PERF_CSTATE_CORE_C7_RES),
470
471 .pkg_events = BIT(PERF_CSTATE_PKG_C2_RES) |
472 BIT(PERF_CSTATE_PKG_C3_RES) |
473 BIT(PERF_CSTATE_PKG_C6_RES) |
474 BIT(PERF_CSTATE_PKG_C7_RES),
475 };
476
477 static const struct cstate_model hswult_cstates __initconst = {
478 .core_events = BIT(PERF_CSTATE_CORE_C3_RES) |
479 BIT(PERF_CSTATE_CORE_C6_RES) |
480 BIT(PERF_CSTATE_CORE_C7_RES),
481
482 .pkg_events = BIT(PERF_CSTATE_PKG_C2_RES) |
483 BIT(PERF_CSTATE_PKG_C3_RES) |
484 BIT(PERF_CSTATE_PKG_C6_RES) |
485 BIT(PERF_CSTATE_PKG_C7_RES) |
486 BIT(PERF_CSTATE_PKG_C8_RES) |
487 BIT(PERF_CSTATE_PKG_C9_RES) |
488 BIT(PERF_CSTATE_PKG_C10_RES),
489 };
490
491 static const struct cstate_model slm_cstates __initconst = {
492 .core_events = BIT(PERF_CSTATE_CORE_C1_RES) |
493 BIT(PERF_CSTATE_CORE_C6_RES),
494
495 .pkg_events = BIT(PERF_CSTATE_PKG_C6_RES),
496 .quirks = SLM_PKG_C6_USE_C7_MSR,
497 };
498
499
500 static const struct cstate_model knl_cstates __initconst = {
501 .core_events = BIT(PERF_CSTATE_CORE_C6_RES),
502
503 .pkg_events = BIT(PERF_CSTATE_PKG_C2_RES) |
504 BIT(PERF_CSTATE_PKG_C3_RES) |
505 BIT(PERF_CSTATE_PKG_C6_RES),
506 .quirks = KNL_CORE_C6_MSR,
507 };
508
509
510 static const struct cstate_model glm_cstates __initconst = {
511 .core_events = BIT(PERF_CSTATE_CORE_C1_RES) |
512 BIT(PERF_CSTATE_CORE_C3_RES) |
513 BIT(PERF_CSTATE_CORE_C6_RES),
514
515 .pkg_events = BIT(PERF_CSTATE_PKG_C2_RES) |
516 BIT(PERF_CSTATE_PKG_C3_RES) |
517 BIT(PERF_CSTATE_PKG_C6_RES) |
518 BIT(PERF_CSTATE_PKG_C10_RES),
519 };
520
521
522 #define X86_CSTATES_MODEL(model, states) \
523 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long) &(states) }
524
525 static const struct x86_cpu_id intel_cstates_match[] __initconst = {
526 X86_CSTATES_MODEL(INTEL_FAM6_NEHALEM, nhm_cstates),
527 X86_CSTATES_MODEL(INTEL_FAM6_NEHALEM_EP, nhm_cstates),
528 X86_CSTATES_MODEL(INTEL_FAM6_NEHALEM_EX, nhm_cstates),
529
530 X86_CSTATES_MODEL(INTEL_FAM6_WESTMERE, nhm_cstates),
531 X86_CSTATES_MODEL(INTEL_FAM6_WESTMERE_EP, nhm_cstates),
532 X86_CSTATES_MODEL(INTEL_FAM6_WESTMERE_EX, nhm_cstates),
533
534 X86_CSTATES_MODEL(INTEL_FAM6_SANDYBRIDGE, snb_cstates),
535 X86_CSTATES_MODEL(INTEL_FAM6_SANDYBRIDGE_X, snb_cstates),
536
537 X86_CSTATES_MODEL(INTEL_FAM6_IVYBRIDGE, snb_cstates),
538 X86_CSTATES_MODEL(INTEL_FAM6_IVYBRIDGE_X, snb_cstates),
539
540 X86_CSTATES_MODEL(INTEL_FAM6_HASWELL_CORE, snb_cstates),
541 X86_CSTATES_MODEL(INTEL_FAM6_HASWELL_X, snb_cstates),
542 X86_CSTATES_MODEL(INTEL_FAM6_HASWELL_GT3E, snb_cstates),
543
544 X86_CSTATES_MODEL(INTEL_FAM6_HASWELL_ULT, hswult_cstates),
545
546 X86_CSTATES_MODEL(INTEL_FAM6_ATOM_SILVERMONT1, slm_cstates),
547 X86_CSTATES_MODEL(INTEL_FAM6_ATOM_SILVERMONT2, slm_cstates),
548 X86_CSTATES_MODEL(INTEL_FAM6_ATOM_AIRMONT, slm_cstates),
549
550 X86_CSTATES_MODEL(INTEL_FAM6_BROADWELL_CORE, snb_cstates),
551 X86_CSTATES_MODEL(INTEL_FAM6_BROADWELL_XEON_D, snb_cstates),
552 X86_CSTATES_MODEL(INTEL_FAM6_BROADWELL_GT3E, snb_cstates),
553 X86_CSTATES_MODEL(INTEL_FAM6_BROADWELL_X, snb_cstates),
554
555 X86_CSTATES_MODEL(INTEL_FAM6_SKYLAKE_MOBILE, snb_cstates),
556 X86_CSTATES_MODEL(INTEL_FAM6_SKYLAKE_DESKTOP, snb_cstates),
557 X86_CSTATES_MODEL(INTEL_FAM6_SKYLAKE_X, snb_cstates),
558
559 X86_CSTATES_MODEL(INTEL_FAM6_KABYLAKE_MOBILE, snb_cstates),
560 X86_CSTATES_MODEL(INTEL_FAM6_KABYLAKE_DESKTOP, snb_cstates),
561
562 X86_CSTATES_MODEL(INTEL_FAM6_XEON_PHI_KNL, knl_cstates),
563 X86_CSTATES_MODEL(INTEL_FAM6_XEON_PHI_KNM, knl_cstates),
564
565 X86_CSTATES_MODEL(INTEL_FAM6_ATOM_GOLDMONT, glm_cstates),
566 X86_CSTATES_MODEL(INTEL_FAM6_ATOM_DENVERTON, glm_cstates),
567
568 X86_CSTATES_MODEL(INTEL_FAM6_ATOM_GEMINI_LAKE, glm_cstates),
569 { },
570 };
571 MODULE_DEVICE_TABLE(x86cpu, intel_cstates_match);
572
573 /*
574 * Probe the cstate events and insert the available one into sysfs attrs
575 * Return false if there are no available events.
576 */
577 static bool __init cstate_probe_msr(const unsigned long evmsk, int max,
578 struct perf_cstate_msr *msr,
579 struct attribute **attrs)
580 {
581 bool found = false;
582 unsigned int bit;
583 u64 val;
584
585 for (bit = 0; bit < max; bit++) {
586 if (test_bit(bit, &evmsk) && !rdmsrl_safe(msr[bit].msr, &val)) {
587 *attrs++ = &msr[bit].attr->attr.attr;
588 found = true;
589 } else {
590 msr[bit].attr = NULL;
591 }
592 }
593 *attrs = NULL;
594
595 return found;
596 }
597
598 static int __init cstate_probe(const struct cstate_model *cm)
599 {
600 /* SLM has different MSR for PKG C6 */
601 if (cm->quirks & SLM_PKG_C6_USE_C7_MSR)
602 pkg_msr[PERF_CSTATE_PKG_C6_RES].msr = MSR_PKG_C7_RESIDENCY;
603
604 /* KNL has different MSR for CORE C6 */
605 if (cm->quirks & KNL_CORE_C6_MSR)
606 pkg_msr[PERF_CSTATE_CORE_C6_RES].msr = MSR_KNL_CORE_C6_RESIDENCY;
607
608
609 has_cstate_core = cstate_probe_msr(cm->core_events,
610 PERF_CSTATE_CORE_EVENT_MAX,
611 core_msr, core_events_attrs);
612
613 has_cstate_pkg = cstate_probe_msr(cm->pkg_events,
614 PERF_CSTATE_PKG_EVENT_MAX,
615 pkg_msr, pkg_events_attrs);
616
617 return (has_cstate_core || has_cstate_pkg) ? 0 : -ENODEV;
618 }
619
620 static inline void cstate_cleanup(void)
621 {
622 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_X86_CSTATE_ONLINE);
623 cpuhp_remove_state_nocalls(CPUHP_AP_PERF_X86_CSTATE_STARTING);
624
625 if (has_cstate_core)
626 perf_pmu_unregister(&cstate_core_pmu);
627
628 if (has_cstate_pkg)
629 perf_pmu_unregister(&cstate_pkg_pmu);
630 }
631
632 static int __init cstate_init(void)
633 {
634 int err;
635
636 cpuhp_setup_state(CPUHP_AP_PERF_X86_CSTATE_STARTING,
637 "perf/x86/cstate:starting", cstate_cpu_init, NULL);
638 cpuhp_setup_state(CPUHP_AP_PERF_X86_CSTATE_ONLINE,
639 "perf/x86/cstate:online", NULL, cstate_cpu_exit);
640
641 if (has_cstate_core) {
642 err = perf_pmu_register(&cstate_core_pmu, cstate_core_pmu.name, -1);
643 if (err) {
644 has_cstate_core = false;
645 pr_info("Failed to register cstate core pmu\n");
646 cstate_cleanup();
647 return err;
648 }
649 }
650
651 if (has_cstate_pkg) {
652 err = perf_pmu_register(&cstate_pkg_pmu, cstate_pkg_pmu.name, -1);
653 if (err) {
654 has_cstate_pkg = false;
655 pr_info("Failed to register cstate pkg pmu\n");
656 cstate_cleanup();
657 return err;
658 }
659 }
660 return 0;
661 }
662
663 static int __init cstate_pmu_init(void)
664 {
665 const struct x86_cpu_id *id;
666 int err;
667
668 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
669 return -ENODEV;
670
671 id = x86_match_cpu(intel_cstates_match);
672 if (!id)
673 return -ENODEV;
674
675 err = cstate_probe((const struct cstate_model *) id->driver_data);
676 if (err)
677 return err;
678
679 return cstate_init();
680 }
681 module_init(cstate_pmu_init);
682
683 static void __exit cstate_pmu_exit(void)
684 {
685 cstate_cleanup();
686 }
687 module_exit(cstate_pmu_exit);