?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'pm-cpufreq'
[linux-2.6/btrfs-unstable.git] / drivers / powercap / intel_rapl.c
blob9ddad0815ba909a7c77de8f93db06e7bc712d246
1 /*
2 * Intel Running Average Power Limit (RAPL) Driver
3 * Copyright (c) 2013, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc.
16 *
17 */
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list.h>
23 #include <linux/types.h>
24 #include <linux/device.h>
25 #include <linux/slab.h>
26 #include <linux/log2.h>
27 #include <linux/bitmap.h>
28 #include <linux/delay.h>
29 #include <linux/sysfs.h>
30 #include <linux/cpu.h>
31 #include <linux/powercap.h>
32 #include <asm/iosf_mbi.h>
33
34 #include <asm/processor.h>
35 #include <asm/cpu_device_id.h>
36 #include <asm/intel-family.h>
37
38 /* Local defines */
39 #define MSR_PLATFORM_POWER_LIMIT 0x0000065C
40
41 /* bitmasks for RAPL MSRs, used by primitive access functions */
42 #define ENERGY_STATUS_MASK 0xffffffff
43
44 #define POWER_LIMIT1_MASK 0x7FFF
45 #define POWER_LIMIT1_ENABLE BIT(15)
46 #define POWER_LIMIT1_CLAMP BIT(16)
47
48 #define POWER_LIMIT2_MASK (0x7FFFULL<<32)
49 #define POWER_LIMIT2_ENABLE BIT_ULL(47)
50 #define POWER_LIMIT2_CLAMP BIT_ULL(48)
51 #define POWER_PACKAGE_LOCK BIT_ULL(63)
52 #define POWER_PP_LOCK BIT(31)
53
54 #define TIME_WINDOW1_MASK (0x7FULL<<17)
55 #define TIME_WINDOW2_MASK (0x7FULL<<49)
56
57 #define POWER_UNIT_OFFSET 0
58 #define POWER_UNIT_MASK 0x0F
59
60 #define ENERGY_UNIT_OFFSET 0x08
61 #define ENERGY_UNIT_MASK 0x1F00
62
63 #define TIME_UNIT_OFFSET 0x10
64 #define TIME_UNIT_MASK 0xF0000
65
66 #define POWER_INFO_MAX_MASK (0x7fffULL<<32)
67 #define POWER_INFO_MIN_MASK (0x7fffULL<<16)
68 #define POWER_INFO_MAX_TIME_WIN_MASK (0x3fULL<<48)
69 #define POWER_INFO_THERMAL_SPEC_MASK 0x7fff
70
71 #define PERF_STATUS_THROTTLE_TIME_MASK 0xffffffff
72 #define PP_POLICY_MASK 0x1F
73
74 /* Non HW constants */
75 #define RAPL_PRIMITIVE_DERIVED BIT(1) /* not from raw data */
76 #define RAPL_PRIMITIVE_DUMMY BIT(2)
77
78 #define TIME_WINDOW_MAX_MSEC 40000
79 #define TIME_WINDOW_MIN_MSEC 250
80 #define ENERGY_UNIT_SCALE 1000 /* scale from driver unit to powercap unit */
81 enum unit_type {
82 ARBITRARY_UNIT, /* no translation */
83 POWER_UNIT,
84 ENERGY_UNIT,
85 TIME_UNIT,
86 };
87
88 enum rapl_domain_type {
89 RAPL_DOMAIN_PACKAGE, /* entire package/socket */
90 RAPL_DOMAIN_PP0, /* core power plane */
91 RAPL_DOMAIN_PP1, /* graphics uncore */
92 RAPL_DOMAIN_DRAM,/* DRAM control_type */
93 RAPL_DOMAIN_PLATFORM, /* PSys control_type */
94 RAPL_DOMAIN_MAX,
95 };
96
97 enum rapl_domain_msr_id {
98 RAPL_DOMAIN_MSR_LIMIT,
99 RAPL_DOMAIN_MSR_STATUS,
100 RAPL_DOMAIN_MSR_PERF,
101 RAPL_DOMAIN_MSR_POLICY,
102 RAPL_DOMAIN_MSR_INFO,
103 RAPL_DOMAIN_MSR_MAX,
104 };
105
106 /* per domain data, some are optional */
107 enum rapl_primitives {
108 ENERGY_COUNTER,
109 POWER_LIMIT1,
110 POWER_LIMIT2,
111 FW_LOCK,
112
113 PL1_ENABLE, /* power limit 1, aka long term */
114 PL1_CLAMP, /* allow frequency to go below OS request */
115 PL2_ENABLE, /* power limit 2, aka short term, instantaneous */
116 PL2_CLAMP,
117
118 TIME_WINDOW1, /* long term */
119 TIME_WINDOW2, /* short term */
120 THERMAL_SPEC_POWER,
121 MAX_POWER,
122
123 MIN_POWER,
124 MAX_TIME_WINDOW,
125 THROTTLED_TIME,
126 PRIORITY_LEVEL,
127
128 /* below are not raw primitive data */
129 AVERAGE_POWER,
130 NR_RAPL_PRIMITIVES,
131 };
132
133 #define NR_RAW_PRIMITIVES (NR_RAPL_PRIMITIVES - 2)
134
135 /* Can be expanded to include events, etc.*/
136 struct rapl_domain_data {
137 u64 primitives[NR_RAPL_PRIMITIVES];
138 unsigned long timestamp;
139 };
140
141 struct msrl_action {
142 u32 msr_no;
143 u64 clear_mask;
144 u64 set_mask;
145 int err;
146 };
147
148 #define DOMAIN_STATE_INACTIVE BIT(0)
149 #define DOMAIN_STATE_POWER_LIMIT_SET BIT(1)
150 #define DOMAIN_STATE_BIOS_LOCKED BIT(2)
151
152 #define NR_POWER_LIMITS (2)
153 struct rapl_power_limit {
154 struct powercap_zone_constraint *constraint;
155 int prim_id; /* primitive ID used to enable */
156 struct rapl_domain *domain;
157 const char *name;
158 };
159
160 static const char pl1_name[] = "long_term";
161 static const char pl2_name[] = "short_term";
162
163 struct rapl_package;
164 struct rapl_domain {
165 const char *name;
166 enum rapl_domain_type id;
167 int msrs[RAPL_DOMAIN_MSR_MAX];
168 struct powercap_zone power_zone;
169 struct rapl_domain_data rdd;
170 struct rapl_power_limit rpl[NR_POWER_LIMITS];
171 u64 attr_map; /* track capabilities */
172 unsigned int state;
173 unsigned int domain_energy_unit;
174 struct rapl_package *rp;
175 };
176 #define power_zone_to_rapl_domain(_zone) \
177 container_of(_zone, struct rapl_domain, power_zone)
178
179
180 /* Each physical package contains multiple domains, these are the common
181 * data across RAPL domains within a package.
182 */
183 struct rapl_package {
184 unsigned int id; /* physical package/socket id */
185 unsigned int nr_domains;
186 unsigned long domain_map; /* bit map of active domains */
187 unsigned int power_unit;
188 unsigned int energy_unit;
189 unsigned int time_unit;
190 struct rapl_domain *domains; /* array of domains, sized at runtime */
191 struct powercap_zone *power_zone; /* keep track of parent zone */
192 unsigned long power_limit_irq; /* keep track of package power limit
193 * notify interrupt enable status.
194 */
195 struct list_head plist;
196 int lead_cpu; /* one active cpu per package for access */
197 /* Track active cpus */
198 struct cpumask cpumask;
199 };
200
201 struct rapl_defaults {
202 u8 floor_freq_reg_addr;
203 int (*check_unit)(struct rapl_package *rp, int cpu);
204 void (*set_floor_freq)(struct rapl_domain *rd, bool mode);
205 u64 (*compute_time_window)(struct rapl_package *rp, u64 val,
206 bool to_raw);
207 unsigned int dram_domain_energy_unit;
208 };
209 static struct rapl_defaults *rapl_defaults;
210
211 /* Sideband MBI registers */
212 #define IOSF_CPU_POWER_BUDGET_CTL_BYT (0x2)
213 #define IOSF_CPU_POWER_BUDGET_CTL_TNG (0xdf)
214
215 #define PACKAGE_PLN_INT_SAVED BIT(0)
216 #define MAX_PRIM_NAME (32)
217
218 /* per domain data. used to describe individual knobs such that access function
219 * can be consolidated into one instead of many inline functions.
220 */
221 struct rapl_primitive_info {
222 const char *name;
223 u64 mask;
224 int shift;
225 enum rapl_domain_msr_id id;
226 enum unit_type unit;
227 u32 flag;
228 };
229
230 #define PRIMITIVE_INFO_INIT(p, m, s, i, u, f) { \
231 .name = #p, \
232 .mask = m, \
233 .shift = s, \
234 .id = i, \
235 .unit = u, \
236 .flag = f \
237 }
238
239 static void rapl_init_domains(struct rapl_package *rp);
240 static int rapl_read_data_raw(struct rapl_domain *rd,
241 enum rapl_primitives prim,
242 bool xlate, u64 *data);
243 static int rapl_write_data_raw(struct rapl_domain *rd,
244 enum rapl_primitives prim,
245 unsigned long long value);
246 static u64 rapl_unit_xlate(struct rapl_domain *rd,
247 enum unit_type type, u64 value,
248 int to_raw);
249 static void package_power_limit_irq_save(struct rapl_package *rp);
250
251 static LIST_HEAD(rapl_packages); /* guarded by CPU hotplug lock */
252
253 static const char * const rapl_domain_names[] = {
254 "package",
255 "core",
256 "uncore",
257 "dram",
258 "psys",
259 };
260
261 static struct powercap_control_type *control_type; /* PowerCap Controller */
262 static struct rapl_domain *platform_rapl_domain; /* Platform (PSys) domain */
263
264 /* caller to ensure CPU hotplug lock is held */
265 static struct rapl_package *find_package_by_id(int id)
266 {
267 struct rapl_package *rp;
268
269 list_for_each_entry(rp, &rapl_packages, plist) {
270 if (rp->id == id)
271 return rp;
272 }
273
274 return NULL;
275 }
276
277 static int get_energy_counter(struct powercap_zone *power_zone, u64 *energy_raw)
278 {
279 struct rapl_domain *rd;
280 u64 energy_now;
281
282 /* prevent CPU hotplug, make sure the RAPL domain does not go
283 * away while reading the counter.
284 */
285 get_online_cpus();
286 rd = power_zone_to_rapl_domain(power_zone);
287
288 if (!rapl_read_data_raw(rd, ENERGY_COUNTER, true, &energy_now)) {
289 *energy_raw = energy_now;
290 put_online_cpus();
291
292 return 0;
293 }
294 put_online_cpus();
295
296 return -EIO;
297 }
298
299 static int get_max_energy_counter(struct powercap_zone *pcd_dev, u64 *energy)
300 {
301 struct rapl_domain *rd = power_zone_to_rapl_domain(pcd_dev);
302
303 *energy = rapl_unit_xlate(rd, ENERGY_UNIT, ENERGY_STATUS_MASK, 0);
304 return 0;
305 }
306
307 static int release_zone(struct powercap_zone *power_zone)
308 {
309 struct rapl_domain *rd = power_zone_to_rapl_domain(power_zone);
310 struct rapl_package *rp = rd->rp;
311
312 /* package zone is the last zone of a package, we can free
313 * memory here since all children has been unregistered.
314 */
315 if (rd->id == RAPL_DOMAIN_PACKAGE) {
316 kfree(rd);
317 rp->domains = NULL;
318 }
319
320 return 0;
321
322 }
323
324 static int find_nr_power_limit(struct rapl_domain *rd)
325 {
326 int i, nr_pl = 0;
327
328 for (i = 0; i < NR_POWER_LIMITS; i++) {
329 if (rd->rpl[i].name)
330 nr_pl++;
331 }
332
333 return nr_pl;
334 }
335
336 static int set_domain_enable(struct powercap_zone *power_zone, bool mode)
337 {
338 struct rapl_domain *rd = power_zone_to_rapl_domain(power_zone);
339
340 if (rd->state & DOMAIN_STATE_BIOS_LOCKED)
341 return -EACCES;
342
343 get_online_cpus();
344 rapl_write_data_raw(rd, PL1_ENABLE, mode);
345 if (rapl_defaults->set_floor_freq)
346 rapl_defaults->set_floor_freq(rd, mode);
347 put_online_cpus();
348
349 return 0;
350 }
351
352 static int get_domain_enable(struct powercap_zone *power_zone, bool *mode)
353 {
354 struct rapl_domain *rd = power_zone_to_rapl_domain(power_zone);
355 u64 val;
356
357 if (rd->state & DOMAIN_STATE_BIOS_LOCKED) {
358 *mode = false;
359 return 0;
360 }
361 get_online_cpus();
362 if (rapl_read_data_raw(rd, PL1_ENABLE, true, &val)) {
363 put_online_cpus();
364 return -EIO;
365 }
366 *mode = val;
367 put_online_cpus();
368
369 return 0;
370 }
371
372 /* per RAPL domain ops, in the order of rapl_domain_type */
373 static const struct powercap_zone_ops zone_ops[] = {
374 /* RAPL_DOMAIN_PACKAGE */
375 {
376 .get_energy_uj = get_energy_counter,
377 .get_max_energy_range_uj = get_max_energy_counter,
378 .release = release_zone,
379 .set_enable = set_domain_enable,
380 .get_enable = get_domain_enable,
381 },
382 /* RAPL_DOMAIN_PP0 */
383 {
384 .get_energy_uj = get_energy_counter,
385 .get_max_energy_range_uj = get_max_energy_counter,
386 .release = release_zone,
387 .set_enable = set_domain_enable,
388 .get_enable = get_domain_enable,
389 },
390 /* RAPL_DOMAIN_PP1 */
391 {
392 .get_energy_uj = get_energy_counter,
393 .get_max_energy_range_uj = get_max_energy_counter,
394 .release = release_zone,
395 .set_enable = set_domain_enable,
396 .get_enable = get_domain_enable,
397 },
398 /* RAPL_DOMAIN_DRAM */
399 {
400 .get_energy_uj = get_energy_counter,
401 .get_max_energy_range_uj = get_max_energy_counter,
402 .release = release_zone,
403 .set_enable = set_domain_enable,
404 .get_enable = get_domain_enable,
405 },
406 /* RAPL_DOMAIN_PLATFORM */
407 {
408 .get_energy_uj = get_energy_counter,
409 .get_max_energy_range_uj = get_max_energy_counter,
410 .release = release_zone,
411 .set_enable = set_domain_enable,
412 .get_enable = get_domain_enable,
413 },
414 };
415
416
417 /*
418 * Constraint index used by powercap can be different than power limit (PL)
419 * index in that some PLs maybe missing due to non-existant MSRs. So we
420 * need to convert here by finding the valid PLs only (name populated).
421 */
422 static int contraint_to_pl(struct rapl_domain *rd, int cid)
423 {
424 int i, j;
425
426 for (i = 0, j = 0; i < NR_POWER_LIMITS; i++) {
427 if ((rd->rpl[i].name) && j++ == cid) {
428 pr_debug("%s: index %d\n", __func__, i);
429 return i;
430 }
431 }
432 pr_err("Cannot find matching power limit for constraint %d\n", cid);
433
434 return -EINVAL;
435 }
436
437 static int set_power_limit(struct powercap_zone *power_zone, int cid,
438 u64 power_limit)
439 {
440 struct rapl_domain *rd;
441 struct rapl_package *rp;
442 int ret = 0;
443 int id;
444
445 get_online_cpus();
446 rd = power_zone_to_rapl_domain(power_zone);
447 id = contraint_to_pl(rd, cid);
448 if (id < 0) {
449 ret = id;
450 goto set_exit;
451 }
452
453 rp = rd->rp;
454
455 if (rd->state & DOMAIN_STATE_BIOS_LOCKED) {
456 dev_warn(&power_zone->dev, "%s locked by BIOS, monitoring only\n",
457 rd->name);
458 ret = -EACCES;
459 goto set_exit;
460 }
461
462 switch (rd->rpl[id].prim_id) {
463 case PL1_ENABLE:
464 rapl_write_data_raw(rd, POWER_LIMIT1, power_limit);
465 break;
466 case PL2_ENABLE:
467 rapl_write_data_raw(rd, POWER_LIMIT2, power_limit);
468 break;
469 default:
470 ret = -EINVAL;
471 }
472 if (!ret)
473 package_power_limit_irq_save(rp);
474 set_exit:
475 put_online_cpus();
476 return ret;
477 }
478
479 static int get_current_power_limit(struct powercap_zone *power_zone, int cid,
480 u64 *data)
481 {
482 struct rapl_domain *rd;
483 u64 val;
484 int prim;
485 int ret = 0;
486 int id;
487
488 get_online_cpus();
489 rd = power_zone_to_rapl_domain(power_zone);
490 id = contraint_to_pl(rd, cid);
491 if (id < 0) {
492 ret = id;
493 goto get_exit;
494 }
495
496 switch (rd->rpl[id].prim_id) {
497 case PL1_ENABLE:
498 prim = POWER_LIMIT1;
499 break;
500 case PL2_ENABLE:
501 prim = POWER_LIMIT2;
502 break;
503 default:
504 put_online_cpus();
505 return -EINVAL;
506 }
507 if (rapl_read_data_raw(rd, prim, true, &val))
508 ret = -EIO;
509 else
510 *data = val;
511
512 get_exit:
513 put_online_cpus();
514
515 return ret;
516 }
517
518 static int set_time_window(struct powercap_zone *power_zone, int cid,
519 u64 window)
520 {
521 struct rapl_domain *rd;
522 int ret = 0;
523 int id;
524
525 get_online_cpus();
526 rd = power_zone_to_rapl_domain(power_zone);
527 id = contraint_to_pl(rd, cid);
528 if (id < 0) {
529 ret = id;
530 goto set_time_exit;
531 }
532
533 switch (rd->rpl[id].prim_id) {
534 case PL1_ENABLE:
535 rapl_write_data_raw(rd, TIME_WINDOW1, window);
536 break;
537 case PL2_ENABLE:
538 rapl_write_data_raw(rd, TIME_WINDOW2, window);
539 break;
540 default:
541 ret = -EINVAL;
542 }
543
544 set_time_exit:
545 put_online_cpus();
546 return ret;
547 }
548
549 static int get_time_window(struct powercap_zone *power_zone, int cid, u64 *data)
550 {
551 struct rapl_domain *rd;
552 u64 val;
553 int ret = 0;
554 int id;
555
556 get_online_cpus();
557 rd = power_zone_to_rapl_domain(power_zone);
558 id = contraint_to_pl(rd, cid);
559 if (id < 0) {
560 ret = id;
561 goto get_time_exit;
562 }
563
564 switch (rd->rpl[id].prim_id) {
565 case PL1_ENABLE:
566 ret = rapl_read_data_raw(rd, TIME_WINDOW1, true, &val);
567 break;
568 case PL2_ENABLE:
569 ret = rapl_read_data_raw(rd, TIME_WINDOW2, true, &val);
570 break;
571 default:
572 put_online_cpus();
573 return -EINVAL;
574 }
575 if (!ret)
576 *data = val;
577
578 get_time_exit:
579 put_online_cpus();
580
581 return ret;
582 }
583
584 static const char *get_constraint_name(struct powercap_zone *power_zone, int cid)
585 {
586 struct rapl_domain *rd;
587 int id;
588
589 rd = power_zone_to_rapl_domain(power_zone);
590 id = contraint_to_pl(rd, cid);
591 if (id >= 0)
592 return rd->rpl[id].name;
593
594 return NULL;
595 }
596
597
598 static int get_max_power(struct powercap_zone *power_zone, int id,
599 u64 *data)
600 {
601 struct rapl_domain *rd;
602 u64 val;
603 int prim;
604 int ret = 0;
605
606 get_online_cpus();
607 rd = power_zone_to_rapl_domain(power_zone);
608 switch (rd->rpl[id].prim_id) {
609 case PL1_ENABLE:
610 prim = THERMAL_SPEC_POWER;
611 break;
612 case PL2_ENABLE:
613 prim = MAX_POWER;
614 break;
615 default:
616 put_online_cpus();
617 return -EINVAL;
618 }
619 if (rapl_read_data_raw(rd, prim, true, &val))
620 ret = -EIO;
621 else
622 *data = val;
623
624 put_online_cpus();
625
626 return ret;
627 }
628
629 static const struct powercap_zone_constraint_ops constraint_ops = {
630 .set_power_limit_uw = set_power_limit,
631 .get_power_limit_uw = get_current_power_limit,
632 .set_time_window_us = set_time_window,
633 .get_time_window_us = get_time_window,
634 .get_max_power_uw = get_max_power,
635 .get_name = get_constraint_name,
636 };
637
638 /* called after domain detection and package level data are set */
639 static void rapl_init_domains(struct rapl_package *rp)
640 {
641 int i;
642 struct rapl_domain *rd = rp->domains;
643
644 for (i = 0; i < RAPL_DOMAIN_MAX; i++) {
645 unsigned int mask = rp->domain_map & (1 << i);
646 switch (mask) {
647 case BIT(RAPL_DOMAIN_PACKAGE):
648 rd->name = rapl_domain_names[RAPL_DOMAIN_PACKAGE];
649 rd->id = RAPL_DOMAIN_PACKAGE;
650 rd->msrs[0] = MSR_PKG_POWER_LIMIT;
651 rd->msrs[1] = MSR_PKG_ENERGY_STATUS;
652 rd->msrs[2] = MSR_PKG_PERF_STATUS;
653 rd->msrs[3] = 0;
654 rd->msrs[4] = MSR_PKG_POWER_INFO;
655 rd->rpl[0].prim_id = PL1_ENABLE;
656 rd->rpl[0].name = pl1_name;
657 rd->rpl[1].prim_id = PL2_ENABLE;
658 rd->rpl[1].name = pl2_name;
659 break;
660 case BIT(RAPL_DOMAIN_PP0):
661 rd->name = rapl_domain_names[RAPL_DOMAIN_PP0];
662 rd->id = RAPL_DOMAIN_PP0;
663 rd->msrs[0] = MSR_PP0_POWER_LIMIT;
664 rd->msrs[1] = MSR_PP0_ENERGY_STATUS;
665 rd->msrs[2] = 0;
666 rd->msrs[3] = MSR_PP0_POLICY;
667 rd->msrs[4] = 0;
668 rd->rpl[0].prim_id = PL1_ENABLE;
669 rd->rpl[0].name = pl1_name;
670 break;
671 case BIT(RAPL_DOMAIN_PP1):
672 rd->name = rapl_domain_names[RAPL_DOMAIN_PP1];
673 rd->id = RAPL_DOMAIN_PP1;
674 rd->msrs[0] = MSR_PP1_POWER_LIMIT;
675 rd->msrs[1] = MSR_PP1_ENERGY_STATUS;
676 rd->msrs[2] = 0;
677 rd->msrs[3] = MSR_PP1_POLICY;
678 rd->msrs[4] = 0;
679 rd->rpl[0].prim_id = PL1_ENABLE;
680 rd->rpl[0].name = pl1_name;
681 break;
682 case BIT(RAPL_DOMAIN_DRAM):
683 rd->name = rapl_domain_names[RAPL_DOMAIN_DRAM];
684 rd->id = RAPL_DOMAIN_DRAM;
685 rd->msrs[0] = MSR_DRAM_POWER_LIMIT;
686 rd->msrs[1] = MSR_DRAM_ENERGY_STATUS;
687 rd->msrs[2] = MSR_DRAM_PERF_STATUS;
688 rd->msrs[3] = 0;
689 rd->msrs[4] = MSR_DRAM_POWER_INFO;
690 rd->rpl[0].prim_id = PL1_ENABLE;
691 rd->rpl[0].name = pl1_name;
692 rd->domain_energy_unit =
693 rapl_defaults->dram_domain_energy_unit;
694 if (rd->domain_energy_unit)
695 pr_info("DRAM domain energy unit %dpj\n",
696 rd->domain_energy_unit);
697 break;
698 }
699 if (mask) {
700 rd->rp = rp;
701 rd++;
702 }
703 }
704 }
705
706 static u64 rapl_unit_xlate(struct rapl_domain *rd, enum unit_type type,
707 u64 value, int to_raw)
708 {
709 u64 units = 1;
710 struct rapl_package *rp = rd->rp;
711 u64 scale = 1;
712
713 switch (type) {
714 case POWER_UNIT:
715 units = rp->power_unit;
716 break;
717 case ENERGY_UNIT:
718 scale = ENERGY_UNIT_SCALE;
719 /* per domain unit takes precedence */
720 if (rd->domain_energy_unit)
721 units = rd->domain_energy_unit;
722 else
723 units = rp->energy_unit;
724 break;
725 case TIME_UNIT:
726 return rapl_defaults->compute_time_window(rp, value, to_raw);
727 case ARBITRARY_UNIT:
728 default:
729 return value;
730 };
731
732 if (to_raw)
733 return div64_u64(value, units) * scale;
734
735 value *= units;
736
737 return div64_u64(value, scale);
738 }
739
740 /* in the order of enum rapl_primitives */
741 static struct rapl_primitive_info rpi[] = {
742 /* name, mask, shift, msr index, unit divisor */
743 PRIMITIVE_INFO_INIT(ENERGY_COUNTER, ENERGY_STATUS_MASK, 0,
744 RAPL_DOMAIN_MSR_STATUS, ENERGY_UNIT, 0),
745 PRIMITIVE_INFO_INIT(POWER_LIMIT1, POWER_LIMIT1_MASK, 0,
746 RAPL_DOMAIN_MSR_LIMIT, POWER_UNIT, 0),
747 PRIMITIVE_INFO_INIT(POWER_LIMIT2, POWER_LIMIT2_MASK, 32,
748 RAPL_DOMAIN_MSR_LIMIT, POWER_UNIT, 0),
749 PRIMITIVE_INFO_INIT(FW_LOCK, POWER_PP_LOCK, 31,
750 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
751 PRIMITIVE_INFO_INIT(PL1_ENABLE, POWER_LIMIT1_ENABLE, 15,
752 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
753 PRIMITIVE_INFO_INIT(PL1_CLAMP, POWER_LIMIT1_CLAMP, 16,
754 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
755 PRIMITIVE_INFO_INIT(PL2_ENABLE, POWER_LIMIT2_ENABLE, 47,
756 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
757 PRIMITIVE_INFO_INIT(PL2_CLAMP, POWER_LIMIT2_CLAMP, 48,
758 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
759 PRIMITIVE_INFO_INIT(TIME_WINDOW1, TIME_WINDOW1_MASK, 17,
760 RAPL_DOMAIN_MSR_LIMIT, TIME_UNIT, 0),
761 PRIMITIVE_INFO_INIT(TIME_WINDOW2, TIME_WINDOW2_MASK, 49,
762 RAPL_DOMAIN_MSR_LIMIT, TIME_UNIT, 0),
763 PRIMITIVE_INFO_INIT(THERMAL_SPEC_POWER, POWER_INFO_THERMAL_SPEC_MASK,
764 0, RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0),
765 PRIMITIVE_INFO_INIT(MAX_POWER, POWER_INFO_MAX_MASK, 32,
766 RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0),
767 PRIMITIVE_INFO_INIT(MIN_POWER, POWER_INFO_MIN_MASK, 16,
768 RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0),
769 PRIMITIVE_INFO_INIT(MAX_TIME_WINDOW, POWER_INFO_MAX_TIME_WIN_MASK, 48,
770 RAPL_DOMAIN_MSR_INFO, TIME_UNIT, 0),
771 PRIMITIVE_INFO_INIT(THROTTLED_TIME, PERF_STATUS_THROTTLE_TIME_MASK, 0,
772 RAPL_DOMAIN_MSR_PERF, TIME_UNIT, 0),
773 PRIMITIVE_INFO_INIT(PRIORITY_LEVEL, PP_POLICY_MASK, 0,
774 RAPL_DOMAIN_MSR_POLICY, ARBITRARY_UNIT, 0),
775 /* non-hardware */
776 PRIMITIVE_INFO_INIT(AVERAGE_POWER, 0, 0, 0, POWER_UNIT,
777 RAPL_PRIMITIVE_DERIVED),
778 {NULL, 0, 0, 0},
779 };
780
781 /* Read primitive data based on its related struct rapl_primitive_info.
782 * if xlate flag is set, return translated data based on data units, i.e.
783 * time, energy, and power.
784 * RAPL MSRs are non-architectual and are laid out not consistently across
785 * domains. Here we use primitive info to allow writing consolidated access
786 * functions.
787 * For a given primitive, it is processed by MSR mask and shift. Unit conversion
788 * is pre-assigned based on RAPL unit MSRs read at init time.
789 * 63-------------------------- 31--------------------------- 0
790 * | xxxxx (mask) |
791 * | |<- shift ----------------|
792 * 63-------------------------- 31--------------------------- 0
793 */
794 static int rapl_read_data_raw(struct rapl_domain *rd,
795 enum rapl_primitives prim,
796 bool xlate, u64 *data)
797 {
798 u64 value, final;
799 u32 msr;
800 struct rapl_primitive_info *rp = &rpi[prim];
801 int cpu;
802
803 if (!rp->name || rp->flag & RAPL_PRIMITIVE_DUMMY)
804 return -EINVAL;
805
806 msr = rd->msrs[rp->id];
807 if (!msr)
808 return -EINVAL;
809
810 cpu = rd->rp->lead_cpu;
811
812 /* special-case package domain, which uses a different bit*/
813 if (prim == FW_LOCK && rd->id == RAPL_DOMAIN_PACKAGE) {
814 rp->mask = POWER_PACKAGE_LOCK;
815 rp->shift = 63;
816 }
817 /* non-hardware data are collected by the polling thread */
818 if (rp->flag & RAPL_PRIMITIVE_DERIVED) {
819 *data = rd->rdd.primitives[prim];
820 return 0;
821 }
822
823 if (rdmsrl_safe_on_cpu(cpu, msr, &value)) {
824 pr_debug("failed to read msr 0x%x on cpu %d\n", msr, cpu);
825 return -EIO;
826 }
827
828 final = value & rp->mask;
829 final = final >> rp->shift;
830 if (xlate)
831 *data = rapl_unit_xlate(rd, rp->unit, final, 0);
832 else
833 *data = final;
834
835 return 0;
836 }
837
838
839 static int msrl_update_safe(u32 msr_no, u64 clear_mask, u64 set_mask)
840 {
841 int err;
842 u64 val;
843
844 err = rdmsrl_safe(msr_no, &val);
845 if (err)
846 goto out;
847
848 val &= ~clear_mask;
849 val |= set_mask;
850
851 err = wrmsrl_safe(msr_no, val);
852
853 out:
854 return err;
855 }
856
857 static void msrl_update_func(void *info)
858 {
859 struct msrl_action *ma = info;
860
861 ma->err = msrl_update_safe(ma->msr_no, ma->clear_mask, ma->set_mask);
862 }
863
864 /* Similar use of primitive info in the read counterpart */
865 static int rapl_write_data_raw(struct rapl_domain *rd,
866 enum rapl_primitives prim,
867 unsigned long long value)
868 {
869 struct rapl_primitive_info *rp = &rpi[prim];
870 int cpu;
871 u64 bits;
872 struct msrl_action ma;
873 int ret;
874
875 cpu = rd->rp->lead_cpu;
876 bits = rapl_unit_xlate(rd, rp->unit, value, 1);
877 bits |= bits << rp->shift;
878 memset(&ma, 0, sizeof(ma));
879
880 ma.msr_no = rd->msrs[rp->id];
881 ma.clear_mask = rp->mask;
882 ma.set_mask = bits;
883
884 ret = smp_call_function_single(cpu, msrl_update_func, &ma, 1);
885 if (ret)
886 WARN_ON_ONCE(ret);
887 else
888 ret = ma.err;
889
890 return ret;
891 }
892
893 /*
894 * Raw RAPL data stored in MSRs are in certain scales. We need to
895 * convert them into standard units based on the units reported in
896 * the RAPL unit MSRs. This is specific to CPUs as the method to
897 * calculate units differ on different CPUs.
898 * We convert the units to below format based on CPUs.
899 * i.e.
900 * energy unit: picoJoules : Represented in picoJoules by default
901 * power unit : microWatts : Represented in milliWatts by default
902 * time unit : microseconds: Represented in seconds by default
903 */
904 static int rapl_check_unit_core(struct rapl_package *rp, int cpu)
905 {
906 u64 msr_val;
907 u32 value;
908
909 if (rdmsrl_safe_on_cpu(cpu, MSR_RAPL_POWER_UNIT, &msr_val)) {
910 pr_err("Failed to read power unit MSR 0x%x on CPU %d, exit.\n",
911 MSR_RAPL_POWER_UNIT, cpu);
912 return -ENODEV;
913 }
914
915 value = (msr_val & ENERGY_UNIT_MASK) >> ENERGY_UNIT_OFFSET;
916 rp->energy_unit = ENERGY_UNIT_SCALE * 1000000 / (1 << value);
917
918 value = (msr_val & POWER_UNIT_MASK) >> POWER_UNIT_OFFSET;
919 rp->power_unit = 1000000 / (1 << value);
920
921 value = (msr_val & TIME_UNIT_MASK) >> TIME_UNIT_OFFSET;
922 rp->time_unit = 1000000 / (1 << value);
923
924 pr_debug("Core CPU package %d energy=%dpJ, time=%dus, power=%duW\n",
925 rp->id, rp->energy_unit, rp->time_unit, rp->power_unit);
926
927 return 0;
928 }
929
930 static int rapl_check_unit_atom(struct rapl_package *rp, int cpu)
931 {
932 u64 msr_val;
933 u32 value;
934
935 if (rdmsrl_safe_on_cpu(cpu, MSR_RAPL_POWER_UNIT, &msr_val)) {
936 pr_err("Failed to read power unit MSR 0x%x on CPU %d, exit.\n",
937 MSR_RAPL_POWER_UNIT, cpu);
938 return -ENODEV;
939 }
940 value = (msr_val & ENERGY_UNIT_MASK) >> ENERGY_UNIT_OFFSET;
941 rp->energy_unit = ENERGY_UNIT_SCALE * 1 << value;
942
943 value = (msr_val & POWER_UNIT_MASK) >> POWER_UNIT_OFFSET;
944 rp->power_unit = (1 << value) * 1000;
945
946 value = (msr_val & TIME_UNIT_MASK) >> TIME_UNIT_OFFSET;
947 rp->time_unit = 1000000 / (1 << value);
948
949 pr_debug("Atom package %d energy=%dpJ, time=%dus, power=%duW\n",
950 rp->id, rp->energy_unit, rp->time_unit, rp->power_unit);
951
952 return 0;
953 }
954
955 static void power_limit_irq_save_cpu(void *info)
956 {
957 u32 l, h = 0;
958 struct rapl_package *rp = (struct rapl_package *)info;
959
960 /* save the state of PLN irq mask bit before disabling it */
961 rdmsr_safe(MSR_IA32_PACKAGE_THERM_INTERRUPT, &l, &h);
962 if (!(rp->power_limit_irq & PACKAGE_PLN_INT_SAVED)) {
963 rp->power_limit_irq = l & PACKAGE_THERM_INT_PLN_ENABLE;
964 rp->power_limit_irq |= PACKAGE_PLN_INT_SAVED;
965 }
966 l &= ~PACKAGE_THERM_INT_PLN_ENABLE;
967 wrmsr_safe(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
968 }
969
970
971 /* REVISIT:
972 * When package power limit is set artificially low by RAPL, LVT
973 * thermal interrupt for package power limit should be ignored
974 * since we are not really exceeding the real limit. The intention
975 * is to avoid excessive interrupts while we are trying to save power.
976 * A useful feature might be routing the package_power_limit interrupt
977 * to userspace via eventfd. once we have a usecase, this is simple
978 * to do by adding an atomic notifier.
979 */
980
981 static void package_power_limit_irq_save(struct rapl_package *rp)
982 {
983 if (!boot_cpu_has(X86_FEATURE_PTS) || !boot_cpu_has(X86_FEATURE_PLN))
984 return;
985
986 smp_call_function_single(rp->lead_cpu, power_limit_irq_save_cpu, rp, 1);
987 }
988
989 /*
990 * Restore per package power limit interrupt enable state. Called from cpu
991 * hotplug code on package removal.
992 */
993 static void package_power_limit_irq_restore(struct rapl_package *rp)
994 {
995 u32 l, h;
996
997 if (!boot_cpu_has(X86_FEATURE_PTS) || !boot_cpu_has(X86_FEATURE_PLN))
998 return;
999
1000 /* irq enable state not saved, nothing to restore */
1001 if (!(rp->power_limit_irq & PACKAGE_PLN_INT_SAVED))
1002 return;
1003
1004 rdmsr_safe(MSR_IA32_PACKAGE_THERM_INTERRUPT, &l, &h);
1005
1006 if (rp->power_limit_irq & PACKAGE_THERM_INT_PLN_ENABLE)
1007 l |= PACKAGE_THERM_INT_PLN_ENABLE;
1008 else
1009 l &= ~PACKAGE_THERM_INT_PLN_ENABLE;
1010
1011 wrmsr_safe(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
1012 }
1013
1014 static void set_floor_freq_default(struct rapl_domain *rd, bool mode)
1015 {
1016 int nr_powerlimit = find_nr_power_limit(rd);
1017
1018 /* always enable clamp such that p-state can go below OS requested
1019 * range. power capping priority over guranteed frequency.
1020 */
1021 rapl_write_data_raw(rd, PL1_CLAMP, mode);
1022
1023 /* some domains have pl2 */
1024 if (nr_powerlimit > 1) {
1025 rapl_write_data_raw(rd, PL2_ENABLE, mode);
1026 rapl_write_data_raw(rd, PL2_CLAMP, mode);
1027 }
1028 }
1029
1030 static void set_floor_freq_atom(struct rapl_domain *rd, bool enable)
1031 {
1032 static u32 power_ctrl_orig_val;
1033 u32 mdata;
1034
1035 if (!rapl_defaults->floor_freq_reg_addr) {
1036 pr_err("Invalid floor frequency config register\n");
1037 return;
1038 }
1039
1040 if (!power_ctrl_orig_val)
1041 iosf_mbi_read(BT_MBI_UNIT_PMC, MBI_CR_READ,
1042 rapl_defaults->floor_freq_reg_addr,
1043 &power_ctrl_orig_val);
1044 mdata = power_ctrl_orig_val;
1045 if (enable) {
1046 mdata &= ~(0x7f << 8);
1047 mdata |= 1 << 8;
1048 }
1049 iosf_mbi_write(BT_MBI_UNIT_PMC, MBI_CR_WRITE,
1050 rapl_defaults->floor_freq_reg_addr, mdata);
1051 }
1052
1053 static u64 rapl_compute_time_window_core(struct rapl_package *rp, u64 value,
1054 bool to_raw)
1055 {
1056 u64 f, y; /* fraction and exp. used for time unit */
1057
1058 /*
1059 * Special processing based on 2^Y*(1+F/4), refer
1060 * to Intel Software Developer's manual Vol.3B: CH 14.9.3.
1061 */
1062 if (!to_raw) {
1063 f = (value & 0x60) >> 5;
1064 y = value & 0x1f;
1065 value = (1 << y) * (4 + f) * rp->time_unit / 4;
1066 } else {
1067 do_div(value, rp->time_unit);
1068 y = ilog2(value);
1069 f = div64_u64(4 * (value - (1 << y)), 1 << y);
1070 value = (y & 0x1f) | ((f & 0x3) << 5);
1071 }
1072 return value;
1073 }
1074
1075 static u64 rapl_compute_time_window_atom(struct rapl_package *rp, u64 value,
1076 bool to_raw)
1077 {
1078 /*
1079 * Atom time unit encoding is straight forward val * time_unit,
1080 * where time_unit is default to 1 sec. Never 0.
1081 */
1082 if (!to_raw)
1083 return (value) ? value *= rp->time_unit : rp->time_unit;
1084 else
1085 value = div64_u64(value, rp->time_unit);
1086
1087 return value;
1088 }
1089
1090 static const struct rapl_defaults rapl_defaults_core = {
1091 .floor_freq_reg_addr = 0,
1092 .check_unit = rapl_check_unit_core,
1093 .set_floor_freq = set_floor_freq_default,
1094 .compute_time_window = rapl_compute_time_window_core,
1095 };
1096
1097 static const struct rapl_defaults rapl_defaults_hsw_server = {
1098 .check_unit = rapl_check_unit_core,
1099 .set_floor_freq = set_floor_freq_default,
1100 .compute_time_window = rapl_compute_time_window_core,
1101 .dram_domain_energy_unit = 15300,
1102 };
1103
1104 static const struct rapl_defaults rapl_defaults_byt = {
1105 .floor_freq_reg_addr = IOSF_CPU_POWER_BUDGET_CTL_BYT,
1106 .check_unit = rapl_check_unit_atom,
1107 .set_floor_freq = set_floor_freq_atom,
1108 .compute_time_window = rapl_compute_time_window_atom,
1109 };
1110
1111 static const struct rapl_defaults rapl_defaults_tng = {
1112 .floor_freq_reg_addr = IOSF_CPU_POWER_BUDGET_CTL_TNG,
1113 .check_unit = rapl_check_unit_atom,
1114 .set_floor_freq = set_floor_freq_atom,
1115 .compute_time_window = rapl_compute_time_window_atom,
1116 };
1117
1118 static const struct rapl_defaults rapl_defaults_ann = {
1119 .floor_freq_reg_addr = 0,
1120 .check_unit = rapl_check_unit_atom,
1121 .set_floor_freq = NULL,
1122 .compute_time_window = rapl_compute_time_window_atom,
1123 };
1124
1125 static const struct rapl_defaults rapl_defaults_cht = {
1126 .floor_freq_reg_addr = 0,
1127 .check_unit = rapl_check_unit_atom,
1128 .set_floor_freq = NULL,
1129 .compute_time_window = rapl_compute_time_window_atom,
1130 };
1131
1132 #define RAPL_CPU(_model, _ops) { \
1133 .vendor = X86_VENDOR_INTEL, \
1134 .family = 6, \
1135 .model = _model, \
1136 .driver_data = (kernel_ulong_t)&_ops, \
1137 }
1138
1139 static const struct x86_cpu_id rapl_ids[] __initconst = {
1140 RAPL_CPU(INTEL_FAM6_SANDYBRIDGE, rapl_defaults_core),
1141 RAPL_CPU(INTEL_FAM6_SANDYBRIDGE_X, rapl_defaults_core),
1142
1143 RAPL_CPU(INTEL_FAM6_IVYBRIDGE, rapl_defaults_core),
1144 RAPL_CPU(INTEL_FAM6_IVYBRIDGE_X, rapl_defaults_core),
1145
1146 RAPL_CPU(INTEL_FAM6_HASWELL_CORE, rapl_defaults_core),
1147 RAPL_CPU(INTEL_FAM6_HASWELL_ULT, rapl_defaults_core),
1148 RAPL_CPU(INTEL_FAM6_HASWELL_GT3E, rapl_defaults_core),
1149 RAPL_CPU(INTEL_FAM6_HASWELL_X, rapl_defaults_hsw_server),
1150
1151 RAPL_CPU(INTEL_FAM6_BROADWELL_CORE, rapl_defaults_core),
1152 RAPL_CPU(INTEL_FAM6_BROADWELL_GT3E, rapl_defaults_core),
1153 RAPL_CPU(INTEL_FAM6_BROADWELL_XEON_D, rapl_defaults_core),
1154 RAPL_CPU(INTEL_FAM6_BROADWELL_X, rapl_defaults_hsw_server),
1155
1156 RAPL_CPU(INTEL_FAM6_SKYLAKE_DESKTOP, rapl_defaults_core),
1157 RAPL_CPU(INTEL_FAM6_SKYLAKE_MOBILE, rapl_defaults_core),
1158 RAPL_CPU(INTEL_FAM6_SKYLAKE_X, rapl_defaults_hsw_server),
1159 RAPL_CPU(INTEL_FAM6_KABYLAKE_MOBILE, rapl_defaults_core),
1160 RAPL_CPU(INTEL_FAM6_KABYLAKE_DESKTOP, rapl_defaults_core),
1161
1162 RAPL_CPU(INTEL_FAM6_ATOM_SILVERMONT1, rapl_defaults_byt),
1163 RAPL_CPU(INTEL_FAM6_ATOM_AIRMONT, rapl_defaults_cht),
1164 RAPL_CPU(INTEL_FAM6_ATOM_MERRIFIELD, rapl_defaults_tng),
1165 RAPL_CPU(INTEL_FAM6_ATOM_MOOREFIELD, rapl_defaults_ann),
1166 RAPL_CPU(INTEL_FAM6_ATOM_GOLDMONT, rapl_defaults_core),
1167 RAPL_CPU(INTEL_FAM6_ATOM_GEMINI_LAKE, rapl_defaults_core),
1168 RAPL_CPU(INTEL_FAM6_ATOM_DENVERTON, rapl_defaults_core),
1169
1170 RAPL_CPU(INTEL_FAM6_XEON_PHI_KNL, rapl_defaults_hsw_server),
1171 RAPL_CPU(INTEL_FAM6_XEON_PHI_KNM, rapl_defaults_hsw_server),
1172 {}
1173 };
1174 MODULE_DEVICE_TABLE(x86cpu, rapl_ids);
1175
1176 /* Read once for all raw primitive data for domains */
1177 static void rapl_update_domain_data(struct rapl_package *rp)
1178 {
1179 int dmn, prim;
1180 u64 val;
1181
1182 for (dmn = 0; dmn < rp->nr_domains; dmn++) {
1183 pr_debug("update package %d domain %s data\n", rp->id,
1184 rp->domains[dmn].name);
1185 /* exclude non-raw primitives */
1186 for (prim = 0; prim < NR_RAW_PRIMITIVES; prim++) {
1187 if (!rapl_read_data_raw(&rp->domains[dmn], prim,
1188 rpi[prim].unit, &val))
1189 rp->domains[dmn].rdd.primitives[prim] = val;
1190 }
1191 }
1192
1193 }
1194
1195 static void rapl_unregister_powercap(void)
1196 {
1197 if (platform_rapl_domain) {
1198 powercap_unregister_zone(control_type,
1199 &platform_rapl_domain->power_zone);
1200 kfree(platform_rapl_domain);
1201 }
1202 powercap_unregister_control_type(control_type);
1203 }
1204
1205 static int rapl_package_register_powercap(struct rapl_package *rp)
1206 {
1207 struct rapl_domain *rd;
1208 char dev_name[17]; /* max domain name = 7 + 1 + 8 for int + 1 for null*/
1209 struct powercap_zone *power_zone = NULL;
1210 int nr_pl, ret;;
1211
1212 /* Update the domain data of the new package */
1213 rapl_update_domain_data(rp);
1214
1215 /* first we register package domain as the parent zone*/
1216 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) {
1217 if (rd->id == RAPL_DOMAIN_PACKAGE) {
1218 nr_pl = find_nr_power_limit(rd);
1219 pr_debug("register socket %d package domain %s\n",
1220 rp->id, rd->name);
1221 memset(dev_name, 0, sizeof(dev_name));
1222 snprintf(dev_name, sizeof(dev_name), "%s-%d",
1223 rd->name, rp->id);
1224 power_zone = powercap_register_zone(&rd->power_zone,
1225 control_type,
1226 dev_name, NULL,
1227 &zone_ops[rd->id],
1228 nr_pl,
1229 &constraint_ops);
1230 if (IS_ERR(power_zone)) {
1231 pr_debug("failed to register package, %d\n",
1232 rp->id);
1233 return PTR_ERR(power_zone);
1234 }
1235 /* track parent zone in per package/socket data */
1236 rp->power_zone = power_zone;
1237 /* done, only one package domain per socket */
1238 break;
1239 }
1240 }
1241 if (!power_zone) {
1242 pr_err("no package domain found, unknown topology!\n");
1243 return -ENODEV;
1244 }
1245 /* now register domains as children of the socket/package*/
1246 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) {
1247 if (rd->id == RAPL_DOMAIN_PACKAGE)
1248 continue;
1249 /* number of power limits per domain varies */
1250 nr_pl = find_nr_power_limit(rd);
1251 power_zone = powercap_register_zone(&rd->power_zone,
1252 control_type, rd->name,
1253 rp->power_zone,
1254 &zone_ops[rd->id], nr_pl,
1255 &constraint_ops);
1256
1257 if (IS_ERR(power_zone)) {
1258 pr_debug("failed to register power_zone, %d:%s:%s\n",
1259 rp->id, rd->name, dev_name);
1260 ret = PTR_ERR(power_zone);
1261 goto err_cleanup;
1262 }
1263 }
1264 return 0;
1265
1266 err_cleanup:
1267 /*
1268 * Clean up previously initialized domains within the package if we
1269 * failed after the first domain setup.
1270 */
1271 while (--rd >= rp->domains) {
1272 pr_debug("unregister package %d domain %s\n", rp->id, rd->name);
1273 powercap_unregister_zone(control_type, &rd->power_zone);
1274 }
1275
1276 return ret;
1277 }
1278
1279 static int __init rapl_register_psys(void)
1280 {
1281 struct rapl_domain *rd;
1282 struct powercap_zone *power_zone;
1283 u64 val;
1284
1285 if (rdmsrl_safe_on_cpu(0, MSR_PLATFORM_ENERGY_STATUS, &val) || !val)
1286 return -ENODEV;
1287
1288 if (rdmsrl_safe_on_cpu(0, MSR_PLATFORM_POWER_LIMIT, &val) || !val)
1289 return -ENODEV;
1290
1291 rd = kzalloc(sizeof(*rd), GFP_KERNEL);
1292 if (!rd)
1293 return -ENOMEM;
1294
1295 rd->name = rapl_domain_names[RAPL_DOMAIN_PLATFORM];
1296 rd->id = RAPL_DOMAIN_PLATFORM;
1297 rd->msrs[0] = MSR_PLATFORM_POWER_LIMIT;
1298 rd->msrs[1] = MSR_PLATFORM_ENERGY_STATUS;
1299 rd->rpl[0].prim_id = PL1_ENABLE;
1300 rd->rpl[0].name = pl1_name;
1301 rd->rpl[1].prim_id = PL2_ENABLE;
1302 rd->rpl[1].name = pl2_name;
1303 rd->rp = find_package_by_id(0);
1304
1305 power_zone = powercap_register_zone(&rd->power_zone, control_type,
1306 "psys", NULL,
1307 &zone_ops[RAPL_DOMAIN_PLATFORM],
1308 2, &constraint_ops);
1309
1310 if (IS_ERR(power_zone)) {
1311 kfree(rd);
1312 return PTR_ERR(power_zone);
1313 }
1314
1315 platform_rapl_domain = rd;
1316
1317 return 0;
1318 }
1319
1320 static int __init rapl_register_powercap(void)
1321 {
1322 control_type = powercap_register_control_type(NULL, "intel-rapl", NULL);
1323 if (IS_ERR(control_type)) {
1324 pr_debug("failed to register powercap control_type.\n");
1325 return PTR_ERR(control_type);
1326 }
1327 return 0;
1328 }
1329
1330 static int rapl_check_domain(int cpu, int domain)
1331 {
1332 unsigned msr;
1333 u64 val = 0;
1334
1335 switch (domain) {
1336 case RAPL_DOMAIN_PACKAGE:
1337 msr = MSR_PKG_ENERGY_STATUS;
1338 break;
1339 case RAPL_DOMAIN_PP0:
1340 msr = MSR_PP0_ENERGY_STATUS;
1341 break;
1342 case RAPL_DOMAIN_PP1:
1343 msr = MSR_PP1_ENERGY_STATUS;
1344 break;
1345 case RAPL_DOMAIN_DRAM:
1346 msr = MSR_DRAM_ENERGY_STATUS;
1347 break;
1348 case RAPL_DOMAIN_PLATFORM:
1349 /* PSYS(PLATFORM) is not a CPU domain, so avoid printng error */
1350 return -EINVAL;
1351 default:
1352 pr_err("invalid domain id %d\n", domain);
1353 return -EINVAL;
1354 }
1355 /* make sure domain counters are available and contains non-zero
1356 * values, otherwise skip it.
1357 */
1358 if (rdmsrl_safe_on_cpu(cpu, msr, &val) || !val)
1359 return -ENODEV;
1360
1361 return 0;
1362 }
1363
1364
1365 /*
1366 * Check if power limits are available. Two cases when they are not available:
1367 * 1. Locked by BIOS, in this case we still provide read-only access so that
1368 * users can see what limit is set by the BIOS.
1369 * 2. Some CPUs make some domains monitoring only which means PLx MSRs may not
1370 * exist at all. In this case, we do not show the contraints in powercap.
1371 *
1372 * Called after domains are detected and initialized.
1373 */
1374 static void rapl_detect_powerlimit(struct rapl_domain *rd)
1375 {
1376 u64 val64;
1377 int i;
1378
1379 /* check if the domain is locked by BIOS, ignore if MSR doesn't exist */
1380 if (!rapl_read_data_raw(rd, FW_LOCK, false, &val64)) {
1381 if (val64) {
1382 pr_info("RAPL package %d domain %s locked by BIOS\n",
1383 rd->rp->id, rd->name);
1384 rd->state |= DOMAIN_STATE_BIOS_LOCKED;
1385 }
1386 }
1387 /* check if power limit MSRs exists, otherwise domain is monitoring only */
1388 for (i = 0; i < NR_POWER_LIMITS; i++) {
1389 int prim = rd->rpl[i].prim_id;
1390 if (rapl_read_data_raw(rd, prim, false, &val64))
1391 rd->rpl[i].name = NULL;
1392 }
1393 }
1394
1395 /* Detect active and valid domains for the given CPU, caller must
1396 * ensure the CPU belongs to the targeted package and CPU hotlug is disabled.
1397 */
1398 static int rapl_detect_domains(struct rapl_package *rp, int cpu)
1399 {
1400 struct rapl_domain *rd;
1401 int i;
1402
1403 for (i = 0; i < RAPL_DOMAIN_MAX; i++) {
1404 /* use physical package id to read counters */
1405 if (!rapl_check_domain(cpu, i)) {
1406 rp->domain_map |= 1 << i;
1407 pr_info("Found RAPL domain %s\n", rapl_domain_names[i]);
1408 }
1409 }
1410 rp->nr_domains = bitmap_weight(&rp->domain_map, RAPL_DOMAIN_MAX);
1411 if (!rp->nr_domains) {
1412 pr_debug("no valid rapl domains found in package %d\n", rp->id);
1413 return -ENODEV;
1414 }
1415 pr_debug("found %d domains on package %d\n", rp->nr_domains, rp->id);
1416
1417 rp->domains = kcalloc(rp->nr_domains + 1, sizeof(struct rapl_domain),
1418 GFP_KERNEL);
1419 if (!rp->domains)
1420 return -ENOMEM;
1421
1422 rapl_init_domains(rp);
1423
1424 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++)
1425 rapl_detect_powerlimit(rd);
1426
1427 return 0;
1428 }
1429
1430 /* called from CPU hotplug notifier, hotplug lock held */
1431 static void rapl_remove_package(struct rapl_package *rp)
1432 {
1433 struct rapl_domain *rd, *rd_package = NULL;
1434
1435 package_power_limit_irq_restore(rp);
1436
1437 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) {
1438 rapl_write_data_raw(rd, PL1_ENABLE, 0);
1439 rapl_write_data_raw(rd, PL1_CLAMP, 0);
1440 if (find_nr_power_limit(rd) > 1) {
1441 rapl_write_data_raw(rd, PL2_ENABLE, 0);
1442 rapl_write_data_raw(rd, PL2_CLAMP, 0);
1443 }
1444 if (rd->id == RAPL_DOMAIN_PACKAGE) {
1445 rd_package = rd;
1446 continue;
1447 }
1448 pr_debug("remove package, undo power limit on %d: %s\n",
1449 rp->id, rd->name);
1450 powercap_unregister_zone(control_type, &rd->power_zone);
1451 }
1452 /* do parent zone last */
1453 powercap_unregister_zone(control_type, &rd_package->power_zone);
1454 list_del(&rp->plist);
1455 kfree(rp);
1456 }
1457
1458 /* called from CPU hotplug notifier, hotplug lock held */
1459 static struct rapl_package *rapl_add_package(int cpu, int pkgid)
1460 {
1461 struct rapl_package *rp;
1462 int ret;
1463
1464 rp = kzalloc(sizeof(struct rapl_package), GFP_KERNEL);
1465 if (!rp)
1466 return ERR_PTR(-ENOMEM);
1467
1468 /* add the new package to the list */
1469 rp->id = pkgid;
1470 rp->lead_cpu = cpu;
1471
1472 /* check if the package contains valid domains */
1473 if (rapl_detect_domains(rp, cpu) ||
1474 rapl_defaults->check_unit(rp, cpu)) {
1475 ret = -ENODEV;
1476 goto err_free_package;
1477 }
1478 ret = rapl_package_register_powercap(rp);
1479 if (!ret) {
1480 INIT_LIST_HEAD(&rp->plist);
1481 list_add(&rp->plist, &rapl_packages);
1482 return rp;
1483 }
1484
1485 err_free_package:
1486 kfree(rp->domains);
1487 kfree(rp);
1488 return ERR_PTR(ret);
1489 }
1490
1491 /* Handles CPU hotplug on multi-socket systems.
1492 * If a CPU goes online as the first CPU of the physical package
1493 * we add the RAPL package to the system. Similarly, when the last
1494 * CPU of the package is removed, we remove the RAPL package and its
1495 * associated domains. Cooling devices are handled accordingly at
1496 * per-domain level.
1497 */
1498 static int rapl_cpu_online(unsigned int cpu)
1499 {
1500 int pkgid = topology_physical_package_id(cpu);
1501 struct rapl_package *rp;
1502
1503 rp = find_package_by_id(pkgid);
1504 if (!rp) {
1505 rp = rapl_add_package(cpu, pkgid);
1506 if (IS_ERR(rp))
1507 return PTR_ERR(rp);
1508 }
1509 cpumask_set_cpu(cpu, &rp->cpumask);
1510 return 0;
1511 }
1512
1513 static int rapl_cpu_down_prep(unsigned int cpu)
1514 {
1515 int pkgid = topology_physical_package_id(cpu);
1516 struct rapl_package *rp;
1517 int lead_cpu;
1518
1519 rp = find_package_by_id(pkgid);
1520 if (!rp)
1521 return 0;
1522
1523 cpumask_clear_cpu(cpu, &rp->cpumask);
1524 lead_cpu = cpumask_first(&rp->cpumask);
1525 if (lead_cpu >= nr_cpu_ids)
1526 rapl_remove_package(rp);
1527 else if (rp->lead_cpu == cpu)
1528 rp->lead_cpu = lead_cpu;
1529 return 0;
1530 }
1531
1532 static enum cpuhp_state pcap_rapl_online;
1533
1534 static int __init rapl_init(void)
1535 {
1536 const struct x86_cpu_id *id;
1537 int ret;
1538
1539 id = x86_match_cpu(rapl_ids);
1540 if (!id) {
1541 pr_err("driver does not support CPU family %d model %d\n",
1542 boot_cpu_data.x86, boot_cpu_data.x86_model);
1543
1544 return -ENODEV;
1545 }
1546
1547 rapl_defaults = (struct rapl_defaults *)id->driver_data;
1548
1549 ret = rapl_register_powercap();
1550 if (ret)
1551 return ret;
1552
1553 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "powercap/rapl:online",
1554 rapl_cpu_online, rapl_cpu_down_prep);
1555 if (ret < 0)
1556 goto err_unreg;
1557 pcap_rapl_online = ret;
1558
1559 /* Don't bail out if PSys is not supported */
1560 rapl_register_psys();
1561 return 0;
1562
1563 err_unreg:
1564 rapl_unregister_powercap();
1565 return ret;
1566 }
1567
1568 static void __exit rapl_exit(void)
1569 {
1570 cpuhp_remove_state(pcap_rapl_online);
1571 rapl_unregister_powercap();
1572 }
1573
1574 module_init(rapl_init);
1575 module_exit(rapl_exit);
1576
1577 MODULE_DESCRIPTION("Driver for Intel RAPL (Running Average Power Limit)");
1578 MODULE_AUTHOR("Jacob Pan <jacob.jun.pan@intel.com>");
1579 MODULE_LICENSE("GPL v2");
(deprecated) Btrfs devel tree