dmaengine: mv_xor_v2: new driver
[linux/fpc-iii.git] / drivers / powercap / intel_rapl.c
blobb2766b867b0eb692dd30d7a1987567b5688077c9
1 /*
2 * Intel Running Average Power Limit (RAPL) Driver
3 * Copyright (c) 2013, Intel Corporation.
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.
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.
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.
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
34 #include <asm/processor.h>
35 #include <asm/cpu_device_id.h>
37 /* Local defines */
38 #define MSR_PLATFORM_POWER_LIMIT 0x0000065C
40 /* bitmasks for RAPL MSRs, used by primitive access functions */
41 #define ENERGY_STATUS_MASK 0xffffffff
43 #define POWER_LIMIT1_MASK 0x7FFF
44 #define POWER_LIMIT1_ENABLE BIT(15)
45 #define POWER_LIMIT1_CLAMP BIT(16)
47 #define POWER_LIMIT2_MASK (0x7FFFULL<<32)
48 #define POWER_LIMIT2_ENABLE BIT_ULL(47)
49 #define POWER_LIMIT2_CLAMP BIT_ULL(48)
50 #define POWER_PACKAGE_LOCK BIT_ULL(63)
51 #define POWER_PP_LOCK BIT(31)
53 #define TIME_WINDOW1_MASK (0x7FULL<<17)
54 #define TIME_WINDOW2_MASK (0x7FULL<<49)
56 #define POWER_UNIT_OFFSET 0
57 #define POWER_UNIT_MASK 0x0F
59 #define ENERGY_UNIT_OFFSET 0x08
60 #define ENERGY_UNIT_MASK 0x1F00
62 #define TIME_UNIT_OFFSET 0x10
63 #define TIME_UNIT_MASK 0xF0000
65 #define POWER_INFO_MAX_MASK (0x7fffULL<<32)
66 #define POWER_INFO_MIN_MASK (0x7fffULL<<16)
67 #define POWER_INFO_MAX_TIME_WIN_MASK (0x3fULL<<48)
68 #define POWER_INFO_THERMAL_SPEC_MASK 0x7fff
70 #define PERF_STATUS_THROTTLE_TIME_MASK 0xffffffff
71 #define PP_POLICY_MASK 0x1F
73 /* Non HW constants */
74 #define RAPL_PRIMITIVE_DERIVED BIT(1) /* not from raw data */
75 #define RAPL_PRIMITIVE_DUMMY BIT(2)
77 #define TIME_WINDOW_MAX_MSEC 40000
78 #define TIME_WINDOW_MIN_MSEC 250
79 #define ENERGY_UNIT_SCALE 1000 /* scale from driver unit to powercap unit */
80 enum unit_type {
81 ARBITRARY_UNIT, /* no translation */
82 POWER_UNIT,
83 ENERGY_UNIT,
84 TIME_UNIT,
87 enum rapl_domain_type {
88 RAPL_DOMAIN_PACKAGE, /* entire package/socket */
89 RAPL_DOMAIN_PP0, /* core power plane */
90 RAPL_DOMAIN_PP1, /* graphics uncore */
91 RAPL_DOMAIN_DRAM,/* DRAM control_type */
92 RAPL_DOMAIN_PLATFORM, /* PSys control_type */
93 RAPL_DOMAIN_MAX,
96 enum rapl_domain_msr_id {
97 RAPL_DOMAIN_MSR_LIMIT,
98 RAPL_DOMAIN_MSR_STATUS,
99 RAPL_DOMAIN_MSR_PERF,
100 RAPL_DOMAIN_MSR_POLICY,
101 RAPL_DOMAIN_MSR_INFO,
102 RAPL_DOMAIN_MSR_MAX,
105 /* per domain data, some are optional */
106 enum rapl_primitives {
107 ENERGY_COUNTER,
108 POWER_LIMIT1,
109 POWER_LIMIT2,
110 FW_LOCK,
112 PL1_ENABLE, /* power limit 1, aka long term */
113 PL1_CLAMP, /* allow frequency to go below OS request */
114 PL2_ENABLE, /* power limit 2, aka short term, instantaneous */
115 PL2_CLAMP,
117 TIME_WINDOW1, /* long term */
118 TIME_WINDOW2, /* short term */
119 THERMAL_SPEC_POWER,
120 MAX_POWER,
122 MIN_POWER,
123 MAX_TIME_WINDOW,
124 THROTTLED_TIME,
125 PRIORITY_LEVEL,
127 /* below are not raw primitive data */
128 AVERAGE_POWER,
129 NR_RAPL_PRIMITIVES,
132 #define NR_RAW_PRIMITIVES (NR_RAPL_PRIMITIVES - 2)
134 /* Can be expanded to include events, etc.*/
135 struct rapl_domain_data {
136 u64 primitives[NR_RAPL_PRIMITIVES];
137 unsigned long timestamp;
140 struct msrl_action {
141 u32 msr_no;
142 u64 clear_mask;
143 u64 set_mask;
144 int err;
147 #define DOMAIN_STATE_INACTIVE BIT(0)
148 #define DOMAIN_STATE_POWER_LIMIT_SET BIT(1)
149 #define DOMAIN_STATE_BIOS_LOCKED BIT(2)
151 #define NR_POWER_LIMITS (2)
152 struct rapl_power_limit {
153 struct powercap_zone_constraint *constraint;
154 int prim_id; /* primitive ID used to enable */
155 struct rapl_domain *domain;
156 const char *name;
159 static const char pl1_name[] = "long_term";
160 static const char pl2_name[] = "short_term";
162 struct rapl_package;
163 struct rapl_domain {
164 const char *name;
165 enum rapl_domain_type id;
166 int msrs[RAPL_DOMAIN_MSR_MAX];
167 struct powercap_zone power_zone;
168 struct rapl_domain_data rdd;
169 struct rapl_power_limit rpl[NR_POWER_LIMITS];
170 u64 attr_map; /* track capabilities */
171 unsigned int state;
172 unsigned int domain_energy_unit;
173 struct rapl_package *rp;
175 #define power_zone_to_rapl_domain(_zone) \
176 container_of(_zone, struct rapl_domain, power_zone)
179 /* Each physical package contains multiple domains, these are the common
180 * data across RAPL domains within a package.
182 struct rapl_package {
183 unsigned int id; /* physical package/socket id */
184 unsigned int nr_domains;
185 unsigned long domain_map; /* bit map of active domains */
186 unsigned int power_unit;
187 unsigned int energy_unit;
188 unsigned int time_unit;
189 struct rapl_domain *domains; /* array of domains, sized at runtime */
190 struct powercap_zone *power_zone; /* keep track of parent zone */
191 int nr_cpus; /* active cpus on the package, topology info is lost during
192 * cpu hotplug. so we have to track ourselves.
194 unsigned long power_limit_irq; /* keep track of package power limit
195 * notify interrupt enable status.
197 struct list_head plist;
198 int lead_cpu; /* one active cpu per package for access */
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;
209 static struct rapl_defaults *rapl_defaults;
211 /* Sideband MBI registers */
212 #define IOSF_CPU_POWER_BUDGET_CTL_BYT (0x2)
213 #define IOSF_CPU_POWER_BUDGET_CTL_TNG (0xdf)
215 #define PACKAGE_PLN_INT_SAVED BIT(0)
216 #define MAX_PRIM_NAME (32)
218 /* per domain data. used to describe individual knobs such that access function
219 * can be consolidated into one instead of many inline functions.
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;
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 \
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);
251 static LIST_HEAD(rapl_packages); /* guarded by CPU hotplug lock */
253 static const char * const rapl_domain_names[] = {
254 "package",
255 "core",
256 "uncore",
257 "dram",
258 "psys",
261 static struct powercap_control_type *control_type; /* PowerCap Controller */
262 static struct rapl_domain *platform_rapl_domain; /* Platform (PSys) domain */
264 /* caller to ensure CPU hotplug lock is held */
265 static struct rapl_package *find_package_by_id(int id)
267 struct rapl_package *rp;
269 list_for_each_entry(rp, &rapl_packages, plist) {
270 if (rp->id == id)
271 return rp;
274 return NULL;
277 /* caller must hold cpu hotplug lock */
278 static void rapl_cleanup_data(void)
280 struct rapl_package *p, *tmp;
282 list_for_each_entry_safe(p, tmp, &rapl_packages, plist) {
283 kfree(p->domains);
284 list_del(&p->plist);
285 kfree(p);
289 static int get_energy_counter(struct powercap_zone *power_zone, u64 *energy_raw)
291 struct rapl_domain *rd;
292 u64 energy_now;
294 /* prevent CPU hotplug, make sure the RAPL domain does not go
295 * away while reading the counter.
297 get_online_cpus();
298 rd = power_zone_to_rapl_domain(power_zone);
300 if (!rapl_read_data_raw(rd, ENERGY_COUNTER, true, &energy_now)) {
301 *energy_raw = energy_now;
302 put_online_cpus();
304 return 0;
306 put_online_cpus();
308 return -EIO;
311 static int get_max_energy_counter(struct powercap_zone *pcd_dev, u64 *energy)
313 struct rapl_domain *rd = power_zone_to_rapl_domain(pcd_dev);
315 *energy = rapl_unit_xlate(rd, ENERGY_UNIT, ENERGY_STATUS_MASK, 0);
316 return 0;
319 static int release_zone(struct powercap_zone *power_zone)
321 struct rapl_domain *rd = power_zone_to_rapl_domain(power_zone);
322 struct rapl_package *rp = rd->rp;
324 /* package zone is the last zone of a package, we can free
325 * memory here since all children has been unregistered.
327 if (rd->id == RAPL_DOMAIN_PACKAGE) {
328 kfree(rd);
329 rp->domains = NULL;
332 return 0;
336 static int find_nr_power_limit(struct rapl_domain *rd)
338 int i;
340 for (i = 0; i < NR_POWER_LIMITS; i++) {
341 if (rd->rpl[i].name == NULL)
342 break;
345 return i;
348 static int set_domain_enable(struct powercap_zone *power_zone, bool mode)
350 struct rapl_domain *rd = power_zone_to_rapl_domain(power_zone);
352 if (rd->state & DOMAIN_STATE_BIOS_LOCKED)
353 return -EACCES;
355 get_online_cpus();
356 rapl_write_data_raw(rd, PL1_ENABLE, mode);
357 if (rapl_defaults->set_floor_freq)
358 rapl_defaults->set_floor_freq(rd, mode);
359 put_online_cpus();
361 return 0;
364 static int get_domain_enable(struct powercap_zone *power_zone, bool *mode)
366 struct rapl_domain *rd = power_zone_to_rapl_domain(power_zone);
367 u64 val;
369 if (rd->state & DOMAIN_STATE_BIOS_LOCKED) {
370 *mode = false;
371 return 0;
373 get_online_cpus();
374 if (rapl_read_data_raw(rd, PL1_ENABLE, true, &val)) {
375 put_online_cpus();
376 return -EIO;
378 *mode = val;
379 put_online_cpus();
381 return 0;
384 /* per RAPL domain ops, in the order of rapl_domain_type */
385 static const struct powercap_zone_ops zone_ops[] = {
386 /* RAPL_DOMAIN_PACKAGE */
388 .get_energy_uj = get_energy_counter,
389 .get_max_energy_range_uj = get_max_energy_counter,
390 .release = release_zone,
391 .set_enable = set_domain_enable,
392 .get_enable = get_domain_enable,
394 /* RAPL_DOMAIN_PP0 */
396 .get_energy_uj = get_energy_counter,
397 .get_max_energy_range_uj = get_max_energy_counter,
398 .release = release_zone,
399 .set_enable = set_domain_enable,
400 .get_enable = get_domain_enable,
402 /* RAPL_DOMAIN_PP1 */
404 .get_energy_uj = get_energy_counter,
405 .get_max_energy_range_uj = get_max_energy_counter,
406 .release = release_zone,
407 .set_enable = set_domain_enable,
408 .get_enable = get_domain_enable,
410 /* RAPL_DOMAIN_DRAM */
412 .get_energy_uj = get_energy_counter,
413 .get_max_energy_range_uj = get_max_energy_counter,
414 .release = release_zone,
415 .set_enable = set_domain_enable,
416 .get_enable = get_domain_enable,
418 /* RAPL_DOMAIN_PLATFORM */
420 .get_energy_uj = get_energy_counter,
421 .get_max_energy_range_uj = get_max_energy_counter,
422 .release = release_zone,
423 .set_enable = set_domain_enable,
424 .get_enable = get_domain_enable,
428 static int set_power_limit(struct powercap_zone *power_zone, int id,
429 u64 power_limit)
431 struct rapl_domain *rd;
432 struct rapl_package *rp;
433 int ret = 0;
435 get_online_cpus();
436 rd = power_zone_to_rapl_domain(power_zone);
437 rp = rd->rp;
439 if (rd->state & DOMAIN_STATE_BIOS_LOCKED) {
440 dev_warn(&power_zone->dev, "%s locked by BIOS, monitoring only\n",
441 rd->name);
442 ret = -EACCES;
443 goto set_exit;
446 switch (rd->rpl[id].prim_id) {
447 case PL1_ENABLE:
448 rapl_write_data_raw(rd, POWER_LIMIT1, power_limit);
449 break;
450 case PL2_ENABLE:
451 rapl_write_data_raw(rd, POWER_LIMIT2, power_limit);
452 break;
453 default:
454 ret = -EINVAL;
456 if (!ret)
457 package_power_limit_irq_save(rp);
458 set_exit:
459 put_online_cpus();
460 return ret;
463 static int get_current_power_limit(struct powercap_zone *power_zone, int id,
464 u64 *data)
466 struct rapl_domain *rd;
467 u64 val;
468 int prim;
469 int ret = 0;
471 get_online_cpus();
472 rd = power_zone_to_rapl_domain(power_zone);
473 switch (rd->rpl[id].prim_id) {
474 case PL1_ENABLE:
475 prim = POWER_LIMIT1;
476 break;
477 case PL2_ENABLE:
478 prim = POWER_LIMIT2;
479 break;
480 default:
481 put_online_cpus();
482 return -EINVAL;
484 if (rapl_read_data_raw(rd, prim, true, &val))
485 ret = -EIO;
486 else
487 *data = val;
489 put_online_cpus();
491 return ret;
494 static int set_time_window(struct powercap_zone *power_zone, int id,
495 u64 window)
497 struct rapl_domain *rd;
498 int ret = 0;
500 get_online_cpus();
501 rd = power_zone_to_rapl_domain(power_zone);
502 switch (rd->rpl[id].prim_id) {
503 case PL1_ENABLE:
504 rapl_write_data_raw(rd, TIME_WINDOW1, window);
505 break;
506 case PL2_ENABLE:
507 rapl_write_data_raw(rd, TIME_WINDOW2, window);
508 break;
509 default:
510 ret = -EINVAL;
512 put_online_cpus();
513 return ret;
516 static int get_time_window(struct powercap_zone *power_zone, int id, u64 *data)
518 struct rapl_domain *rd;
519 u64 val;
520 int ret = 0;
522 get_online_cpus();
523 rd = power_zone_to_rapl_domain(power_zone);
524 switch (rd->rpl[id].prim_id) {
525 case PL1_ENABLE:
526 ret = rapl_read_data_raw(rd, TIME_WINDOW1, true, &val);
527 break;
528 case PL2_ENABLE:
529 ret = rapl_read_data_raw(rd, TIME_WINDOW2, true, &val);
530 break;
531 default:
532 put_online_cpus();
533 return -EINVAL;
535 if (!ret)
536 *data = val;
537 put_online_cpus();
539 return ret;
542 static const char *get_constraint_name(struct powercap_zone *power_zone, int id)
544 struct rapl_power_limit *rpl;
545 struct rapl_domain *rd;
547 rd = power_zone_to_rapl_domain(power_zone);
548 rpl = (struct rapl_power_limit *) &rd->rpl[id];
550 return rpl->name;
554 static int get_max_power(struct powercap_zone *power_zone, int id,
555 u64 *data)
557 struct rapl_domain *rd;
558 u64 val;
559 int prim;
560 int ret = 0;
562 get_online_cpus();
563 rd = power_zone_to_rapl_domain(power_zone);
564 switch (rd->rpl[id].prim_id) {
565 case PL1_ENABLE:
566 prim = THERMAL_SPEC_POWER;
567 break;
568 case PL2_ENABLE:
569 prim = MAX_POWER;
570 break;
571 default:
572 put_online_cpus();
573 return -EINVAL;
575 if (rapl_read_data_raw(rd, prim, true, &val))
576 ret = -EIO;
577 else
578 *data = val;
580 put_online_cpus();
582 return ret;
585 static const struct powercap_zone_constraint_ops constraint_ops = {
586 .set_power_limit_uw = set_power_limit,
587 .get_power_limit_uw = get_current_power_limit,
588 .set_time_window_us = set_time_window,
589 .get_time_window_us = get_time_window,
590 .get_max_power_uw = get_max_power,
591 .get_name = get_constraint_name,
594 /* called after domain detection and package level data are set */
595 static void rapl_init_domains(struct rapl_package *rp)
597 int i;
598 struct rapl_domain *rd = rp->domains;
600 for (i = 0; i < RAPL_DOMAIN_MAX; i++) {
601 unsigned int mask = rp->domain_map & (1 << i);
602 switch (mask) {
603 case BIT(RAPL_DOMAIN_PACKAGE):
604 rd->name = rapl_domain_names[RAPL_DOMAIN_PACKAGE];
605 rd->id = RAPL_DOMAIN_PACKAGE;
606 rd->msrs[0] = MSR_PKG_POWER_LIMIT;
607 rd->msrs[1] = MSR_PKG_ENERGY_STATUS;
608 rd->msrs[2] = MSR_PKG_PERF_STATUS;
609 rd->msrs[3] = 0;
610 rd->msrs[4] = MSR_PKG_POWER_INFO;
611 rd->rpl[0].prim_id = PL1_ENABLE;
612 rd->rpl[0].name = pl1_name;
613 rd->rpl[1].prim_id = PL2_ENABLE;
614 rd->rpl[1].name = pl2_name;
615 break;
616 case BIT(RAPL_DOMAIN_PP0):
617 rd->name = rapl_domain_names[RAPL_DOMAIN_PP0];
618 rd->id = RAPL_DOMAIN_PP0;
619 rd->msrs[0] = MSR_PP0_POWER_LIMIT;
620 rd->msrs[1] = MSR_PP0_ENERGY_STATUS;
621 rd->msrs[2] = 0;
622 rd->msrs[3] = MSR_PP0_POLICY;
623 rd->msrs[4] = 0;
624 rd->rpl[0].prim_id = PL1_ENABLE;
625 rd->rpl[0].name = pl1_name;
626 break;
627 case BIT(RAPL_DOMAIN_PP1):
628 rd->name = rapl_domain_names[RAPL_DOMAIN_PP1];
629 rd->id = RAPL_DOMAIN_PP1;
630 rd->msrs[0] = MSR_PP1_POWER_LIMIT;
631 rd->msrs[1] = MSR_PP1_ENERGY_STATUS;
632 rd->msrs[2] = 0;
633 rd->msrs[3] = MSR_PP1_POLICY;
634 rd->msrs[4] = 0;
635 rd->rpl[0].prim_id = PL1_ENABLE;
636 rd->rpl[0].name = pl1_name;
637 break;
638 case BIT(RAPL_DOMAIN_DRAM):
639 rd->name = rapl_domain_names[RAPL_DOMAIN_DRAM];
640 rd->id = RAPL_DOMAIN_DRAM;
641 rd->msrs[0] = MSR_DRAM_POWER_LIMIT;
642 rd->msrs[1] = MSR_DRAM_ENERGY_STATUS;
643 rd->msrs[2] = MSR_DRAM_PERF_STATUS;
644 rd->msrs[3] = 0;
645 rd->msrs[4] = MSR_DRAM_POWER_INFO;
646 rd->rpl[0].prim_id = PL1_ENABLE;
647 rd->rpl[0].name = pl1_name;
648 rd->domain_energy_unit =
649 rapl_defaults->dram_domain_energy_unit;
650 if (rd->domain_energy_unit)
651 pr_info("DRAM domain energy unit %dpj\n",
652 rd->domain_energy_unit);
653 break;
655 if (mask) {
656 rd->rp = rp;
657 rd++;
662 static u64 rapl_unit_xlate(struct rapl_domain *rd, enum unit_type type,
663 u64 value, int to_raw)
665 u64 units = 1;
666 struct rapl_package *rp = rd->rp;
667 u64 scale = 1;
669 switch (type) {
670 case POWER_UNIT:
671 units = rp->power_unit;
672 break;
673 case ENERGY_UNIT:
674 scale = ENERGY_UNIT_SCALE;
675 /* per domain unit takes precedence */
676 if (rd && rd->domain_energy_unit)
677 units = rd->domain_energy_unit;
678 else
679 units = rp->energy_unit;
680 break;
681 case TIME_UNIT:
682 return rapl_defaults->compute_time_window(rp, value, to_raw);
683 case ARBITRARY_UNIT:
684 default:
685 return value;
688 if (to_raw)
689 return div64_u64(value, units) * scale;
691 value *= units;
693 return div64_u64(value, scale);
696 /* in the order of enum rapl_primitives */
697 static struct rapl_primitive_info rpi[] = {
698 /* name, mask, shift, msr index, unit divisor */
699 PRIMITIVE_INFO_INIT(ENERGY_COUNTER, ENERGY_STATUS_MASK, 0,
700 RAPL_DOMAIN_MSR_STATUS, ENERGY_UNIT, 0),
701 PRIMITIVE_INFO_INIT(POWER_LIMIT1, POWER_LIMIT1_MASK, 0,
702 RAPL_DOMAIN_MSR_LIMIT, POWER_UNIT, 0),
703 PRIMITIVE_INFO_INIT(POWER_LIMIT2, POWER_LIMIT2_MASK, 32,
704 RAPL_DOMAIN_MSR_LIMIT, POWER_UNIT, 0),
705 PRIMITIVE_INFO_INIT(FW_LOCK, POWER_PP_LOCK, 31,
706 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
707 PRIMITIVE_INFO_INIT(PL1_ENABLE, POWER_LIMIT1_ENABLE, 15,
708 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
709 PRIMITIVE_INFO_INIT(PL1_CLAMP, POWER_LIMIT1_CLAMP, 16,
710 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
711 PRIMITIVE_INFO_INIT(PL2_ENABLE, POWER_LIMIT2_ENABLE, 47,
712 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
713 PRIMITIVE_INFO_INIT(PL2_CLAMP, POWER_LIMIT2_CLAMP, 48,
714 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0),
715 PRIMITIVE_INFO_INIT(TIME_WINDOW1, TIME_WINDOW1_MASK, 17,
716 RAPL_DOMAIN_MSR_LIMIT, TIME_UNIT, 0),
717 PRIMITIVE_INFO_INIT(TIME_WINDOW2, TIME_WINDOW2_MASK, 49,
718 RAPL_DOMAIN_MSR_LIMIT, TIME_UNIT, 0),
719 PRIMITIVE_INFO_INIT(THERMAL_SPEC_POWER, POWER_INFO_THERMAL_SPEC_MASK,
720 0, RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0),
721 PRIMITIVE_INFO_INIT(MAX_POWER, POWER_INFO_MAX_MASK, 32,
722 RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0),
723 PRIMITIVE_INFO_INIT(MIN_POWER, POWER_INFO_MIN_MASK, 16,
724 RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0),
725 PRIMITIVE_INFO_INIT(MAX_TIME_WINDOW, POWER_INFO_MAX_TIME_WIN_MASK, 48,
726 RAPL_DOMAIN_MSR_INFO, TIME_UNIT, 0),
727 PRIMITIVE_INFO_INIT(THROTTLED_TIME, PERF_STATUS_THROTTLE_TIME_MASK, 0,
728 RAPL_DOMAIN_MSR_PERF, TIME_UNIT, 0),
729 PRIMITIVE_INFO_INIT(PRIORITY_LEVEL, PP_POLICY_MASK, 0,
730 RAPL_DOMAIN_MSR_POLICY, ARBITRARY_UNIT, 0),
731 /* non-hardware */
732 PRIMITIVE_INFO_INIT(AVERAGE_POWER, 0, 0, 0, POWER_UNIT,
733 RAPL_PRIMITIVE_DERIVED),
734 {NULL, 0, 0, 0},
737 /* Read primitive data based on its related struct rapl_primitive_info.
738 * if xlate flag is set, return translated data based on data units, i.e.
739 * time, energy, and power.
740 * RAPL MSRs are non-architectual and are laid out not consistently across
741 * domains. Here we use primitive info to allow writing consolidated access
742 * functions.
743 * For a given primitive, it is processed by MSR mask and shift. Unit conversion
744 * is pre-assigned based on RAPL unit MSRs read at init time.
745 * 63-------------------------- 31--------------------------- 0
746 * | xxxxx (mask) |
747 * | |<- shift ----------------|
748 * 63-------------------------- 31--------------------------- 0
750 static int rapl_read_data_raw(struct rapl_domain *rd,
751 enum rapl_primitives prim,
752 bool xlate, u64 *data)
754 u64 value, final;
755 u32 msr;
756 struct rapl_primitive_info *rp = &rpi[prim];
757 int cpu;
759 if (!rp->name || rp->flag & RAPL_PRIMITIVE_DUMMY)
760 return -EINVAL;
762 msr = rd->msrs[rp->id];
763 if (!msr)
764 return -EINVAL;
766 cpu = rd->rp->lead_cpu;
768 /* special-case package domain, which uses a different bit*/
769 if (prim == FW_LOCK && rd->id == RAPL_DOMAIN_PACKAGE) {
770 rp->mask = POWER_PACKAGE_LOCK;
771 rp->shift = 63;
773 /* non-hardware data are collected by the polling thread */
774 if (rp->flag & RAPL_PRIMITIVE_DERIVED) {
775 *data = rd->rdd.primitives[prim];
776 return 0;
779 if (rdmsrl_safe_on_cpu(cpu, msr, &value)) {
780 pr_debug("failed to read msr 0x%x on cpu %d\n", msr, cpu);
781 return -EIO;
784 final = value & rp->mask;
785 final = final >> rp->shift;
786 if (xlate)
787 *data = rapl_unit_xlate(rd, rp->unit, final, 0);
788 else
789 *data = final;
791 return 0;
795 static int msrl_update_safe(u32 msr_no, u64 clear_mask, u64 set_mask)
797 int err;
798 u64 val;
800 err = rdmsrl_safe(msr_no, &val);
801 if (err)
802 goto out;
804 val &= ~clear_mask;
805 val |= set_mask;
807 err = wrmsrl_safe(msr_no, val);
809 out:
810 return err;
813 static void msrl_update_func(void *info)
815 struct msrl_action *ma = info;
817 ma->err = msrl_update_safe(ma->msr_no, ma->clear_mask, ma->set_mask);
820 /* Similar use of primitive info in the read counterpart */
821 static int rapl_write_data_raw(struct rapl_domain *rd,
822 enum rapl_primitives prim,
823 unsigned long long value)
825 struct rapl_primitive_info *rp = &rpi[prim];
826 int cpu;
827 u64 bits;
828 struct msrl_action ma;
829 int ret;
831 cpu = rd->rp->lead_cpu;
832 bits = rapl_unit_xlate(rd, rp->unit, value, 1);
833 bits |= bits << rp->shift;
834 memset(&ma, 0, sizeof(ma));
836 ma.msr_no = rd->msrs[rp->id];
837 ma.clear_mask = rp->mask;
838 ma.set_mask = bits;
840 ret = smp_call_function_single(cpu, msrl_update_func, &ma, 1);
841 if (ret)
842 WARN_ON_ONCE(ret);
843 else
844 ret = ma.err;
846 return ret;
850 * Raw RAPL data stored in MSRs are in certain scales. We need to
851 * convert them into standard units based on the units reported in
852 * the RAPL unit MSRs. This is specific to CPUs as the method to
853 * calculate units differ on different CPUs.
854 * We convert the units to below format based on CPUs.
855 * i.e.
856 * energy unit: picoJoules : Represented in picoJoules by default
857 * power unit : microWatts : Represented in milliWatts by default
858 * time unit : microseconds: Represented in seconds by default
860 static int rapl_check_unit_core(struct rapl_package *rp, int cpu)
862 u64 msr_val;
863 u32 value;
865 if (rdmsrl_safe_on_cpu(cpu, MSR_RAPL_POWER_UNIT, &msr_val)) {
866 pr_err("Failed to read power unit MSR 0x%x on CPU %d, exit.\n",
867 MSR_RAPL_POWER_UNIT, cpu);
868 return -ENODEV;
871 value = (msr_val & ENERGY_UNIT_MASK) >> ENERGY_UNIT_OFFSET;
872 rp->energy_unit = ENERGY_UNIT_SCALE * 1000000 / (1 << value);
874 value = (msr_val & POWER_UNIT_MASK) >> POWER_UNIT_OFFSET;
875 rp->power_unit = 1000000 / (1 << value);
877 value = (msr_val & TIME_UNIT_MASK) >> TIME_UNIT_OFFSET;
878 rp->time_unit = 1000000 / (1 << value);
880 pr_debug("Core CPU package %d energy=%dpJ, time=%dus, power=%duW\n",
881 rp->id, rp->energy_unit, rp->time_unit, rp->power_unit);
883 return 0;
886 static int rapl_check_unit_atom(struct rapl_package *rp, int cpu)
888 u64 msr_val;
889 u32 value;
891 if (rdmsrl_safe_on_cpu(cpu, MSR_RAPL_POWER_UNIT, &msr_val)) {
892 pr_err("Failed to read power unit MSR 0x%x on CPU %d, exit.\n",
893 MSR_RAPL_POWER_UNIT, cpu);
894 return -ENODEV;
896 value = (msr_val & ENERGY_UNIT_MASK) >> ENERGY_UNIT_OFFSET;
897 rp->energy_unit = ENERGY_UNIT_SCALE * 1 << value;
899 value = (msr_val & POWER_UNIT_MASK) >> POWER_UNIT_OFFSET;
900 rp->power_unit = (1 << value) * 1000;
902 value = (msr_val & TIME_UNIT_MASK) >> TIME_UNIT_OFFSET;
903 rp->time_unit = 1000000 / (1 << value);
905 pr_debug("Atom package %d energy=%dpJ, time=%dus, power=%duW\n",
906 rp->id, rp->energy_unit, rp->time_unit, rp->power_unit);
908 return 0;
911 static void power_limit_irq_save_cpu(void *info)
913 u32 l, h = 0;
914 struct rapl_package *rp = (struct rapl_package *)info;
916 /* save the state of PLN irq mask bit before disabling it */
917 rdmsr_safe(MSR_IA32_PACKAGE_THERM_INTERRUPT, &l, &h);
918 if (!(rp->power_limit_irq & PACKAGE_PLN_INT_SAVED)) {
919 rp->power_limit_irq = l & PACKAGE_THERM_INT_PLN_ENABLE;
920 rp->power_limit_irq |= PACKAGE_PLN_INT_SAVED;
922 l &= ~PACKAGE_THERM_INT_PLN_ENABLE;
923 wrmsr_safe(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
927 /* REVISIT:
928 * When package power limit is set artificially low by RAPL, LVT
929 * thermal interrupt for package power limit should be ignored
930 * since we are not really exceeding the real limit. The intention
931 * is to avoid excessive interrupts while we are trying to save power.
932 * A useful feature might be routing the package_power_limit interrupt
933 * to userspace via eventfd. once we have a usecase, this is simple
934 * to do by adding an atomic notifier.
937 static void package_power_limit_irq_save(struct rapl_package *rp)
939 if (!boot_cpu_has(X86_FEATURE_PTS) || !boot_cpu_has(X86_FEATURE_PLN))
940 return;
942 smp_call_function_single(rp->lead_cpu, power_limit_irq_save_cpu, rp, 1);
945 static void power_limit_irq_restore_cpu(void *info)
947 u32 l, h = 0;
948 struct rapl_package *rp = (struct rapl_package *)info;
950 rdmsr_safe(MSR_IA32_PACKAGE_THERM_INTERRUPT, &l, &h);
952 if (rp->power_limit_irq & PACKAGE_THERM_INT_PLN_ENABLE)
953 l |= PACKAGE_THERM_INT_PLN_ENABLE;
954 else
955 l &= ~PACKAGE_THERM_INT_PLN_ENABLE;
957 wrmsr_safe(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
960 /* restore per package power limit interrupt enable state */
961 static void package_power_limit_irq_restore(struct rapl_package *rp)
963 if (!boot_cpu_has(X86_FEATURE_PTS) || !boot_cpu_has(X86_FEATURE_PLN))
964 return;
966 /* irq enable state not saved, nothing to restore */
967 if (!(rp->power_limit_irq & PACKAGE_PLN_INT_SAVED))
968 return;
970 smp_call_function_single(rp->lead_cpu, power_limit_irq_restore_cpu, rp, 1);
973 static void set_floor_freq_default(struct rapl_domain *rd, bool mode)
975 int nr_powerlimit = find_nr_power_limit(rd);
977 /* always enable clamp such that p-state can go below OS requested
978 * range. power capping priority over guranteed frequency.
980 rapl_write_data_raw(rd, PL1_CLAMP, mode);
982 /* some domains have pl2 */
983 if (nr_powerlimit > 1) {
984 rapl_write_data_raw(rd, PL2_ENABLE, mode);
985 rapl_write_data_raw(rd, PL2_CLAMP, mode);
989 static void set_floor_freq_atom(struct rapl_domain *rd, bool enable)
991 static u32 power_ctrl_orig_val;
992 u32 mdata;
994 if (!rapl_defaults->floor_freq_reg_addr) {
995 pr_err("Invalid floor frequency config register\n");
996 return;
999 if (!power_ctrl_orig_val)
1000 iosf_mbi_read(BT_MBI_UNIT_PMC, MBI_CR_READ,
1001 rapl_defaults->floor_freq_reg_addr,
1002 &power_ctrl_orig_val);
1003 mdata = power_ctrl_orig_val;
1004 if (enable) {
1005 mdata &= ~(0x7f << 8);
1006 mdata |= 1 << 8;
1008 iosf_mbi_write(BT_MBI_UNIT_PMC, MBI_CR_WRITE,
1009 rapl_defaults->floor_freq_reg_addr, mdata);
1012 static u64 rapl_compute_time_window_core(struct rapl_package *rp, u64 value,
1013 bool to_raw)
1015 u64 f, y; /* fraction and exp. used for time unit */
1018 * Special processing based on 2^Y*(1+F/4), refer
1019 * to Intel Software Developer's manual Vol.3B: CH 14.9.3.
1021 if (!to_raw) {
1022 f = (value & 0x60) >> 5;
1023 y = value & 0x1f;
1024 value = (1 << y) * (4 + f) * rp->time_unit / 4;
1025 } else {
1026 do_div(value, rp->time_unit);
1027 y = ilog2(value);
1028 f = div64_u64(4 * (value - (1 << y)), 1 << y);
1029 value = (y & 0x1f) | ((f & 0x3) << 5);
1031 return value;
1034 static u64 rapl_compute_time_window_atom(struct rapl_package *rp, u64 value,
1035 bool to_raw)
1038 * Atom time unit encoding is straight forward val * time_unit,
1039 * where time_unit is default to 1 sec. Never 0.
1041 if (!to_raw)
1042 return (value) ? value *= rp->time_unit : rp->time_unit;
1043 else
1044 value = div64_u64(value, rp->time_unit);
1046 return value;
1049 static const struct rapl_defaults rapl_defaults_core = {
1050 .floor_freq_reg_addr = 0,
1051 .check_unit = rapl_check_unit_core,
1052 .set_floor_freq = set_floor_freq_default,
1053 .compute_time_window = rapl_compute_time_window_core,
1056 static const struct rapl_defaults rapl_defaults_hsw_server = {
1057 .check_unit = rapl_check_unit_core,
1058 .set_floor_freq = set_floor_freq_default,
1059 .compute_time_window = rapl_compute_time_window_core,
1060 .dram_domain_energy_unit = 15300,
1063 static const struct rapl_defaults rapl_defaults_byt = {
1064 .floor_freq_reg_addr = IOSF_CPU_POWER_BUDGET_CTL_BYT,
1065 .check_unit = rapl_check_unit_atom,
1066 .set_floor_freq = set_floor_freq_atom,
1067 .compute_time_window = rapl_compute_time_window_atom,
1070 static const struct rapl_defaults rapl_defaults_tng = {
1071 .floor_freq_reg_addr = IOSF_CPU_POWER_BUDGET_CTL_TNG,
1072 .check_unit = rapl_check_unit_atom,
1073 .set_floor_freq = set_floor_freq_atom,
1074 .compute_time_window = rapl_compute_time_window_atom,
1077 static const struct rapl_defaults rapl_defaults_ann = {
1078 .floor_freq_reg_addr = 0,
1079 .check_unit = rapl_check_unit_atom,
1080 .set_floor_freq = NULL,
1081 .compute_time_window = rapl_compute_time_window_atom,
1084 static const struct rapl_defaults rapl_defaults_cht = {
1085 .floor_freq_reg_addr = 0,
1086 .check_unit = rapl_check_unit_atom,
1087 .set_floor_freq = NULL,
1088 .compute_time_window = rapl_compute_time_window_atom,
1091 #define RAPL_CPU(_model, _ops) { \
1092 .vendor = X86_VENDOR_INTEL, \
1093 .family = 6, \
1094 .model = _model, \
1095 .driver_data = (kernel_ulong_t)&_ops, \
1098 static const struct x86_cpu_id rapl_ids[] __initconst = {
1099 RAPL_CPU(0x2a, rapl_defaults_core),/* Sandy Bridge */
1100 RAPL_CPU(0x2d, rapl_defaults_core),/* Sandy Bridge EP */
1101 RAPL_CPU(0x37, rapl_defaults_byt),/* Valleyview */
1102 RAPL_CPU(0x3a, rapl_defaults_core),/* Ivy Bridge */
1103 RAPL_CPU(0x3c, rapl_defaults_core),/* Haswell */
1104 RAPL_CPU(0x3d, rapl_defaults_core),/* Broadwell */
1105 RAPL_CPU(0x3f, rapl_defaults_hsw_server),/* Haswell servers */
1106 RAPL_CPU(0x4f, rapl_defaults_hsw_server),/* Broadwell servers */
1107 RAPL_CPU(0x45, rapl_defaults_core),/* Haswell ULT */
1108 RAPL_CPU(0x46, rapl_defaults_core),/* Haswell */
1109 RAPL_CPU(0x47, rapl_defaults_core),/* Broadwell-H */
1110 RAPL_CPU(0x4E, rapl_defaults_core),/* Skylake */
1111 RAPL_CPU(0x4C, rapl_defaults_cht),/* Braswell/Cherryview */
1112 RAPL_CPU(0x4A, rapl_defaults_tng),/* Tangier */
1113 RAPL_CPU(0x56, rapl_defaults_core),/* Future Xeon */
1114 RAPL_CPU(0x5A, rapl_defaults_ann),/* Annidale */
1115 RAPL_CPU(0X5C, rapl_defaults_core),/* Broxton */
1116 RAPL_CPU(0x5E, rapl_defaults_core),/* Skylake-H/S */
1117 RAPL_CPU(0x57, rapl_defaults_hsw_server),/* Knights Landing */
1118 RAPL_CPU(0x8E, rapl_defaults_core),/* Kabylake */
1119 RAPL_CPU(0x9E, rapl_defaults_core),/* Kabylake */
1122 MODULE_DEVICE_TABLE(x86cpu, rapl_ids);
1124 /* read once for all raw primitive data for all packages, domains */
1125 static void rapl_update_domain_data(void)
1127 int dmn, prim;
1128 u64 val;
1129 struct rapl_package *rp;
1131 list_for_each_entry(rp, &rapl_packages, plist) {
1132 for (dmn = 0; dmn < rp->nr_domains; dmn++) {
1133 pr_debug("update package %d domain %s data\n", rp->id,
1134 rp->domains[dmn].name);
1135 /* exclude non-raw primitives */
1136 for (prim = 0; prim < NR_RAW_PRIMITIVES; prim++)
1137 if (!rapl_read_data_raw(&rp->domains[dmn], prim,
1138 rpi[prim].unit,
1139 &val))
1140 rp->domains[dmn].rdd.primitives[prim] =
1141 val;
1147 static int rapl_unregister_powercap(void)
1149 struct rapl_package *rp;
1150 struct rapl_domain *rd, *rd_package = NULL;
1152 /* unregister all active rapl packages from the powercap layer,
1153 * hotplug lock held
1155 list_for_each_entry(rp, &rapl_packages, plist) {
1156 package_power_limit_irq_restore(rp);
1158 for (rd = rp->domains; rd < rp->domains + rp->nr_domains;
1159 rd++) {
1160 pr_debug("remove package, undo power limit on %d: %s\n",
1161 rp->id, rd->name);
1162 rapl_write_data_raw(rd, PL1_ENABLE, 0);
1163 rapl_write_data_raw(rd, PL1_CLAMP, 0);
1164 if (find_nr_power_limit(rd) > 1) {
1165 rapl_write_data_raw(rd, PL2_ENABLE, 0);
1166 rapl_write_data_raw(rd, PL2_CLAMP, 0);
1168 if (rd->id == RAPL_DOMAIN_PACKAGE) {
1169 rd_package = rd;
1170 continue;
1172 powercap_unregister_zone(control_type, &rd->power_zone);
1174 /* do the package zone last */
1175 if (rd_package)
1176 powercap_unregister_zone(control_type,
1177 &rd_package->power_zone);
1180 if (platform_rapl_domain) {
1181 powercap_unregister_zone(control_type,
1182 &platform_rapl_domain->power_zone);
1183 kfree(platform_rapl_domain);
1186 powercap_unregister_control_type(control_type);
1188 return 0;
1191 static int rapl_package_register_powercap(struct rapl_package *rp)
1193 struct rapl_domain *rd;
1194 int ret = 0;
1195 char dev_name[17]; /* max domain name = 7 + 1 + 8 for int + 1 for null*/
1196 struct powercap_zone *power_zone = NULL;
1197 int nr_pl;
1199 /* first we register package domain as the parent zone*/
1200 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) {
1201 if (rd->id == RAPL_DOMAIN_PACKAGE) {
1202 nr_pl = find_nr_power_limit(rd);
1203 pr_debug("register socket %d package domain %s\n",
1204 rp->id, rd->name);
1205 memset(dev_name, 0, sizeof(dev_name));
1206 snprintf(dev_name, sizeof(dev_name), "%s-%d",
1207 rd->name, rp->id);
1208 power_zone = powercap_register_zone(&rd->power_zone,
1209 control_type,
1210 dev_name, NULL,
1211 &zone_ops[rd->id],
1212 nr_pl,
1213 &constraint_ops);
1214 if (IS_ERR(power_zone)) {
1215 pr_debug("failed to register package, %d\n",
1216 rp->id);
1217 ret = PTR_ERR(power_zone);
1218 goto exit_package;
1220 /* track parent zone in per package/socket data */
1221 rp->power_zone = power_zone;
1222 /* done, only one package domain per socket */
1223 break;
1226 if (!power_zone) {
1227 pr_err("no package domain found, unknown topology!\n");
1228 ret = -ENODEV;
1229 goto exit_package;
1231 /* now register domains as children of the socket/package*/
1232 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) {
1233 if (rd->id == RAPL_DOMAIN_PACKAGE)
1234 continue;
1235 /* number of power limits per domain varies */
1236 nr_pl = find_nr_power_limit(rd);
1237 power_zone = powercap_register_zone(&rd->power_zone,
1238 control_type, rd->name,
1239 rp->power_zone,
1240 &zone_ops[rd->id], nr_pl,
1241 &constraint_ops);
1243 if (IS_ERR(power_zone)) {
1244 pr_debug("failed to register power_zone, %d:%s:%s\n",
1245 rp->id, rd->name, dev_name);
1246 ret = PTR_ERR(power_zone);
1247 goto err_cleanup;
1251 exit_package:
1252 return ret;
1253 err_cleanup:
1254 /* clean up previously initialized domains within the package if we
1255 * failed after the first domain setup.
1257 while (--rd >= rp->domains) {
1258 pr_debug("unregister package %d domain %s\n", rp->id, rd->name);
1259 powercap_unregister_zone(control_type, &rd->power_zone);
1262 return ret;
1265 static int rapl_register_psys(void)
1267 struct rapl_domain *rd;
1268 struct powercap_zone *power_zone;
1269 u64 val;
1271 if (rdmsrl_safe_on_cpu(0, MSR_PLATFORM_ENERGY_STATUS, &val) || !val)
1272 return -ENODEV;
1274 if (rdmsrl_safe_on_cpu(0, MSR_PLATFORM_POWER_LIMIT, &val) || !val)
1275 return -ENODEV;
1277 rd = kzalloc(sizeof(*rd), GFP_KERNEL);
1278 if (!rd)
1279 return -ENOMEM;
1281 rd->name = rapl_domain_names[RAPL_DOMAIN_PLATFORM];
1282 rd->id = RAPL_DOMAIN_PLATFORM;
1283 rd->msrs[0] = MSR_PLATFORM_POWER_LIMIT;
1284 rd->msrs[1] = MSR_PLATFORM_ENERGY_STATUS;
1285 rd->rpl[0].prim_id = PL1_ENABLE;
1286 rd->rpl[0].name = pl1_name;
1287 rd->rpl[1].prim_id = PL2_ENABLE;
1288 rd->rpl[1].name = pl2_name;
1289 rd->rp = find_package_by_id(0);
1291 power_zone = powercap_register_zone(&rd->power_zone, control_type,
1292 "psys", NULL,
1293 &zone_ops[RAPL_DOMAIN_PLATFORM],
1294 2, &constraint_ops);
1296 if (IS_ERR(power_zone)) {
1297 kfree(rd);
1298 return PTR_ERR(power_zone);
1301 platform_rapl_domain = rd;
1303 return 0;
1306 static int rapl_register_powercap(void)
1308 struct rapl_domain *rd;
1309 struct rapl_package *rp;
1310 int ret = 0;
1312 control_type = powercap_register_control_type(NULL, "intel-rapl", NULL);
1313 if (IS_ERR(control_type)) {
1314 pr_debug("failed to register powercap control_type.\n");
1315 return PTR_ERR(control_type);
1317 /* read the initial data */
1318 rapl_update_domain_data();
1319 list_for_each_entry(rp, &rapl_packages, plist)
1320 if (rapl_package_register_powercap(rp))
1321 goto err_cleanup_package;
1323 /* Don't bail out if PSys is not supported */
1324 rapl_register_psys();
1326 return ret;
1328 err_cleanup_package:
1329 /* clean up previously initialized packages */
1330 list_for_each_entry_continue_reverse(rp, &rapl_packages, plist) {
1331 for (rd = rp->domains; rd < rp->domains + rp->nr_domains;
1332 rd++) {
1333 pr_debug("unregister zone/package %d, %s domain\n",
1334 rp->id, rd->name);
1335 powercap_unregister_zone(control_type, &rd->power_zone);
1339 return ret;
1342 static int rapl_check_domain(int cpu, int domain)
1344 unsigned msr;
1345 u64 val = 0;
1347 switch (domain) {
1348 case RAPL_DOMAIN_PACKAGE:
1349 msr = MSR_PKG_ENERGY_STATUS;
1350 break;
1351 case RAPL_DOMAIN_PP0:
1352 msr = MSR_PP0_ENERGY_STATUS;
1353 break;
1354 case RAPL_DOMAIN_PP1:
1355 msr = MSR_PP1_ENERGY_STATUS;
1356 break;
1357 case RAPL_DOMAIN_DRAM:
1358 msr = MSR_DRAM_ENERGY_STATUS;
1359 break;
1360 case RAPL_DOMAIN_PLATFORM:
1361 /* PSYS(PLATFORM) is not a CPU domain, so avoid printng error */
1362 return -EINVAL;
1363 default:
1364 pr_err("invalid domain id %d\n", domain);
1365 return -EINVAL;
1367 /* make sure domain counters are available and contains non-zero
1368 * values, otherwise skip it.
1370 if (rdmsrl_safe_on_cpu(cpu, msr, &val) || !val)
1371 return -ENODEV;
1373 return 0;
1376 /* Detect active and valid domains for the given CPU, caller must
1377 * ensure the CPU belongs to the targeted package and CPU hotlug is disabled.
1379 static int rapl_detect_domains(struct rapl_package *rp, int cpu)
1381 int i;
1382 int ret = 0;
1383 struct rapl_domain *rd;
1384 u64 locked;
1386 for (i = 0; i < RAPL_DOMAIN_MAX; i++) {
1387 /* use physical package id to read counters */
1388 if (!rapl_check_domain(cpu, i)) {
1389 rp->domain_map |= 1 << i;
1390 pr_info("Found RAPL domain %s\n", rapl_domain_names[i]);
1393 rp->nr_domains = bitmap_weight(&rp->domain_map, RAPL_DOMAIN_MAX);
1394 if (!rp->nr_domains) {
1395 pr_err("no valid rapl domains found in package %d\n", rp->id);
1396 ret = -ENODEV;
1397 goto done;
1399 pr_debug("found %d domains on package %d\n", rp->nr_domains, rp->id);
1401 rp->domains = kcalloc(rp->nr_domains + 1, sizeof(struct rapl_domain),
1402 GFP_KERNEL);
1403 if (!rp->domains) {
1404 ret = -ENOMEM;
1405 goto done;
1407 rapl_init_domains(rp);
1409 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) {
1410 /* check if the domain is locked by BIOS */
1411 ret = rapl_read_data_raw(rd, FW_LOCK, false, &locked);
1412 if (ret)
1413 return ret;
1414 if (locked) {
1415 pr_info("RAPL package %d domain %s locked by BIOS\n",
1416 rp->id, rd->name);
1417 rd->state |= DOMAIN_STATE_BIOS_LOCKED;
1422 done:
1423 return ret;
1426 static bool is_package_new(int package)
1428 struct rapl_package *rp;
1430 /* caller prevents cpu hotplug, there will be no new packages added
1431 * or deleted while traversing the package list, no need for locking.
1433 list_for_each_entry(rp, &rapl_packages, plist)
1434 if (package == rp->id)
1435 return false;
1437 return true;
1440 /* RAPL interface can be made of a two-level hierarchy: package level and domain
1441 * level. We first detect the number of packages then domains of each package.
1442 * We have to consider the possiblity of CPU online/offline due to hotplug and
1443 * other scenarios.
1445 static int rapl_detect_topology(void)
1447 int i;
1448 int phy_package_id;
1449 struct rapl_package *new_package, *rp;
1451 for_each_online_cpu(i) {
1452 phy_package_id = topology_physical_package_id(i);
1453 if (is_package_new(phy_package_id)) {
1454 new_package = kzalloc(sizeof(*rp), GFP_KERNEL);
1455 if (!new_package) {
1456 rapl_cleanup_data();
1457 return -ENOMEM;
1459 /* add the new package to the list */
1460 new_package->id = phy_package_id;
1461 new_package->nr_cpus = 1;
1462 /* use the first active cpu of the package to access */
1463 new_package->lead_cpu = i;
1464 /* check if the package contains valid domains */
1465 if (rapl_detect_domains(new_package, i) ||
1466 rapl_defaults->check_unit(new_package, i)) {
1467 kfree(new_package->domains);
1468 kfree(new_package);
1469 /* free up the packages already initialized */
1470 rapl_cleanup_data();
1471 return -ENODEV;
1473 INIT_LIST_HEAD(&new_package->plist);
1474 list_add(&new_package->plist, &rapl_packages);
1475 } else {
1476 rp = find_package_by_id(phy_package_id);
1477 if (rp)
1478 ++rp->nr_cpus;
1482 return 0;
1485 /* called from CPU hotplug notifier, hotplug lock held */
1486 static void rapl_remove_package(struct rapl_package *rp)
1488 struct rapl_domain *rd, *rd_package = NULL;
1490 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) {
1491 if (rd->id == RAPL_DOMAIN_PACKAGE) {
1492 rd_package = rd;
1493 continue;
1495 pr_debug("remove package %d, %s domain\n", rp->id, rd->name);
1496 powercap_unregister_zone(control_type, &rd->power_zone);
1498 /* do parent zone last */
1499 powercap_unregister_zone(control_type, &rd_package->power_zone);
1500 list_del(&rp->plist);
1501 kfree(rp);
1504 /* called from CPU hotplug notifier, hotplug lock held */
1505 static int rapl_add_package(int cpu)
1507 int ret = 0;
1508 int phy_package_id;
1509 struct rapl_package *rp;
1511 phy_package_id = topology_physical_package_id(cpu);
1512 rp = kzalloc(sizeof(struct rapl_package), GFP_KERNEL);
1513 if (!rp)
1514 return -ENOMEM;
1516 /* add the new package to the list */
1517 rp->id = phy_package_id;
1518 rp->nr_cpus = 1;
1519 rp->lead_cpu = cpu;
1521 /* check if the package contains valid domains */
1522 if (rapl_detect_domains(rp, cpu) ||
1523 rapl_defaults->check_unit(rp, cpu)) {
1524 ret = -ENODEV;
1525 goto err_free_package;
1527 if (!rapl_package_register_powercap(rp)) {
1528 INIT_LIST_HEAD(&rp->plist);
1529 list_add(&rp->plist, &rapl_packages);
1530 return ret;
1533 err_free_package:
1534 kfree(rp->domains);
1535 kfree(rp);
1537 return ret;
1540 /* Handles CPU hotplug on multi-socket systems.
1541 * If a CPU goes online as the first CPU of the physical package
1542 * we add the RAPL package to the system. Similarly, when the last
1543 * CPU of the package is removed, we remove the RAPL package and its
1544 * associated domains. Cooling devices are handled accordingly at
1545 * per-domain level.
1547 static int rapl_cpu_callback(struct notifier_block *nfb,
1548 unsigned long action, void *hcpu)
1550 unsigned long cpu = (unsigned long)hcpu;
1551 int phy_package_id;
1552 struct rapl_package *rp;
1553 int lead_cpu;
1555 phy_package_id = topology_physical_package_id(cpu);
1556 switch (action) {
1557 case CPU_ONLINE:
1558 case CPU_ONLINE_FROZEN:
1559 case CPU_DOWN_FAILED:
1560 case CPU_DOWN_FAILED_FROZEN:
1561 rp = find_package_by_id(phy_package_id);
1562 if (rp)
1563 ++rp->nr_cpus;
1564 else
1565 rapl_add_package(cpu);
1566 break;
1567 case CPU_DOWN_PREPARE:
1568 case CPU_DOWN_PREPARE_FROZEN:
1569 rp = find_package_by_id(phy_package_id);
1570 if (!rp)
1571 break;
1572 if (--rp->nr_cpus == 0)
1573 rapl_remove_package(rp);
1574 else if (cpu == rp->lead_cpu) {
1575 /* choose another active cpu in the package */
1576 lead_cpu = cpumask_any_but(topology_core_cpumask(cpu), cpu);
1577 if (lead_cpu < nr_cpu_ids)
1578 rp->lead_cpu = lead_cpu;
1579 else /* should never go here */
1580 pr_err("no active cpu available for package %d\n",
1581 phy_package_id);
1585 return NOTIFY_OK;
1588 static struct notifier_block rapl_cpu_notifier = {
1589 .notifier_call = rapl_cpu_callback,
1592 static int __init rapl_init(void)
1594 int ret = 0;
1595 const struct x86_cpu_id *id;
1597 id = x86_match_cpu(rapl_ids);
1598 if (!id) {
1599 pr_err("driver does not support CPU family %d model %d\n",
1600 boot_cpu_data.x86, boot_cpu_data.x86_model);
1602 return -ENODEV;
1605 rapl_defaults = (struct rapl_defaults *)id->driver_data;
1607 cpu_notifier_register_begin();
1609 /* prevent CPU hotplug during detection */
1610 get_online_cpus();
1611 ret = rapl_detect_topology();
1612 if (ret)
1613 goto done;
1615 if (rapl_register_powercap()) {
1616 rapl_cleanup_data();
1617 ret = -ENODEV;
1618 goto done;
1620 __register_hotcpu_notifier(&rapl_cpu_notifier);
1621 done:
1622 put_online_cpus();
1623 cpu_notifier_register_done();
1625 return ret;
1628 static void __exit rapl_exit(void)
1630 cpu_notifier_register_begin();
1631 get_online_cpus();
1632 __unregister_hotcpu_notifier(&rapl_cpu_notifier);
1633 rapl_unregister_powercap();
1634 rapl_cleanup_data();
1635 put_online_cpus();
1636 cpu_notifier_register_done();
1639 module_init(rapl_init);
1640 module_exit(rapl_exit);
1642 MODULE_DESCRIPTION("Driver for Intel RAPL (Running Average Power Limit)");
1643 MODULE_AUTHOR("Jacob Pan <jacob.jun.pan@intel.com>");
1644 MODULE_LICENSE("GPL v2");