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 33#include <asm/processor.h> 34#include <asm/cpu_device_id.h> 35 36/* bitmasks for RAPL MSRs, used by primitive access functions */ 37#define ENERGY_STATUS_MASK 0xffffffff 38 39#define POWER_LIMIT1_MASK 0x7FFF 40#define POWER_LIMIT1_ENABLE BIT(15) 41#define POWER_LIMIT1_CLAMP BIT(16) 42 43#define POWER_LIMIT2_MASK (0x7FFFULL<<32) 44#define POWER_LIMIT2_ENABLE BIT_ULL(47) 45#define POWER_LIMIT2_CLAMP BIT_ULL(48) 46#define POWER_PACKAGE_LOCK BIT_ULL(63) 47#define POWER_PP_LOCK BIT(31) 48 49#define TIME_WINDOW1_MASK (0x7FULL<<17) 50#define TIME_WINDOW2_MASK (0x7FULL<<49) 51 52#define POWER_UNIT_OFFSET 0 53#define POWER_UNIT_MASK 0x0F 54 55#define ENERGY_UNIT_OFFSET 0x08 56#define ENERGY_UNIT_MASK 0x1F00 57 58#define TIME_UNIT_OFFSET 0x10 59#define TIME_UNIT_MASK 0xF0000 60 61#define POWER_INFO_MAX_MASK (0x7fffULL<<32) 62#define POWER_INFO_MIN_MASK (0x7fffULL<<16) 63#define POWER_INFO_MAX_TIME_WIN_MASK (0x3fULL<<48) 64#define POWER_INFO_THERMAL_SPEC_MASK 0x7fff 65 66#define PERF_STATUS_THROTTLE_TIME_MASK 0xffffffff 67#define PP_POLICY_MASK 0x1F 68 69/* Non HW constants */ 70#define RAPL_PRIMITIVE_DERIVED BIT(1) /* not from raw data */ 71#define RAPL_PRIMITIVE_DUMMY BIT(2) 72 73/* scale RAPL units to avoid floating point math inside kernel */ 74#define POWER_UNIT_SCALE (1000000) 75#define ENERGY_UNIT_SCALE (1000000) 76#define TIME_UNIT_SCALE (1000000) 77 78#define TIME_WINDOW_MAX_MSEC 40000 79#define TIME_WINDOW_MIN_MSEC 250 80 81enum unit_type { 82 ARBITRARY_UNIT, /* no translation */ 83 POWER_UNIT, 84 ENERGY_UNIT, 85 TIME_UNIT, 86}; 87 88enum 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_MAX, 94}; 95 96enum 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, 103}; 104 105/* per domain data, some are optional */ 106enum rapl_primitives { 107 ENERGY_COUNTER, 108 POWER_LIMIT1, 109 POWER_LIMIT2, 110 FW_LOCK, 111 112 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, 116 117 TIME_WINDOW1, /* long term */ 118 TIME_WINDOW2, /* short term */ 119 THERMAL_SPEC_POWER, 120 MAX_POWER, 121 122 MIN_POWER, 123 MAX_TIME_WINDOW, 124 THROTTLED_TIME, 125 PRIORITY_LEVEL, 126 127 /* below are not raw primitive data */ 128 AVERAGE_POWER, 129 NR_RAPL_PRIMITIVES, 130}; 131 132#define NR_RAW_PRIMITIVES (NR_RAPL_PRIMITIVES - 2) 133 134/* Can be expanded to include events, etc.*/ 135struct rapl_domain_data { 136 u64 primitives[NR_RAPL_PRIMITIVES]; 137 unsigned long timestamp; 138}; 139 140 141#define DOMAIN_STATE_INACTIVE BIT(0) 142#define DOMAIN_STATE_POWER_LIMIT_SET BIT(1) 143#define DOMAIN_STATE_BIOS_LOCKED BIT(2) 144 145#define NR_POWER_LIMITS (2) 146struct rapl_power_limit { 147 struct powercap_zone_constraint *constraint; 148 int prim_id; /* primitive ID used to enable */ 149 struct rapl_domain *domain; 150 const char *name; 151}; 152 153static const char pl1_name[] = "long_term"; 154static const char pl2_name[] = "short_term"; 155 156struct rapl_domain { 157 const char *name; 158 enum rapl_domain_type id; 159 int msrs[RAPL_DOMAIN_MSR_MAX]; 160 struct powercap_zone power_zone; 161 struct rapl_domain_data rdd; 162 struct rapl_power_limit rpl[NR_POWER_LIMITS]; 163 u64 attr_map; /* track capabilities */ 164 unsigned int state; 165 int package_id; 166}; 167#define power_zone_to_rapl_domain(_zone) \ 168 container_of(_zone, struct rapl_domain, power_zone) 169 170 171/* Each physical package contains multiple domains, these are the common 172 * data across RAPL domains within a package. 173 */ 174struct rapl_package { 175 unsigned int id; /* physical package/socket id */ 176 unsigned int nr_domains; 177 unsigned long domain_map; /* bit map of active domains */ 178 unsigned int power_unit_divisor; 179 unsigned int energy_unit_divisor; 180 unsigned int time_unit_divisor; 181 struct rapl_domain *domains; /* array of domains, sized at runtime */ 182 struct powercap_zone *power_zone; /* keep track of parent zone */ 183 int nr_cpus; /* active cpus on the package, topology info is lost during 184 * cpu hotplug. so we have to track ourselves. 185 */ 186 unsigned long power_limit_irq; /* keep track of package power limit 187 * notify interrupt enable status. 188 */ 189 struct list_head plist; 190}; 191#define PACKAGE_PLN_INT_SAVED BIT(0) 192#define MAX_PRIM_NAME (32) 193 194/* per domain data. used to describe individual knobs such that access function 195 * can be consolidated into one instead of many inline functions. 196 */ 197struct rapl_primitive_info { 198 const char *name; 199 u64 mask; 200 int shift; 201 enum rapl_domain_msr_id id; 202 enum unit_type unit; 203 u32 flag; 204}; 205 206#define PRIMITIVE_INFO_INIT(p, m, s, i, u, f) { \ 207 .name = #p, \ 208 .mask = m, \ 209 .shift = s, \ 210 .id = i, \ 211 .unit = u, \ 212 .flag = f \ 213 } 214 215static void rapl_init_domains(struct rapl_package *rp); 216static int rapl_read_data_raw(struct rapl_domain *rd, 217 enum rapl_primitives prim, 218 bool xlate, u64 *data); 219static int rapl_write_data_raw(struct rapl_domain *rd, 220 enum rapl_primitives prim, 221 unsigned long long value); 222static u64 rapl_unit_xlate(int package, enum unit_type type, u64 value, 223 int to_raw); 224static void package_power_limit_irq_save(int package_id); 225 226static LIST_HEAD(rapl_packages); /* guarded by CPU hotplug lock */ 227 228static const char * const rapl_domain_names[] = { 229 "package", 230 "core", 231 "uncore", 232 "dram", 233}; 234 235static struct powercap_control_type *control_type; /* PowerCap Controller */ 236 237/* caller to ensure CPU hotplug lock is held */ 238static struct rapl_package *find_package_by_id(int id) 239{ 240 struct rapl_package *rp; 241 242 list_for_each_entry(rp, &rapl_packages, plist) { 243 if (rp->id == id) 244 return rp; 245 } 246 247 return NULL; 248} 249 250/* caller to ensure CPU hotplug lock is held */ 251static int find_active_cpu_on_package(int package_id) 252{ 253 int i; 254 255 for_each_online_cpu(i) { 256 if (topology_physical_package_id(i) == package_id) 257 return i; 258 } 259 /* all CPUs on this package are offline */ 260 261 return -ENODEV; 262} 263 264/* caller must hold cpu hotplug lock */ 265static void rapl_cleanup_data(void) 266{ 267 struct rapl_package *p, *tmp; 268 269 list_for_each_entry_safe(p, tmp, &rapl_packages, plist) { 270 kfree(p->domains); 271 list_del(&p->plist); 272 kfree(p); 273 } 274} 275 276static int get_energy_counter(struct powercap_zone *power_zone, u64 *energy_raw) 277{ 278 struct rapl_domain *rd; 279 u64 energy_now; 280 281 /* prevent CPU hotplug, make sure the RAPL domain does not go 282 * away while reading the counter. 283 */ 284 get_online_cpus(); 285 rd = power_zone_to_rapl_domain(power_zone); 286 287 if (!rapl_read_data_raw(rd, ENERGY_COUNTER, true, &energy_now)) { 288 *energy_raw = energy_now; 289 put_online_cpus(); 290 291 return 0; 292 } 293 put_online_cpus(); 294 295 return -EIO; 296} 297 298static int get_max_energy_counter(struct powercap_zone *pcd_dev, u64 *energy) 299{ 300 *energy = rapl_unit_xlate(0, ENERGY_UNIT, ENERGY_STATUS_MASK, 0); 301 return 0; 302} 303 304static int release_zone(struct powercap_zone *power_zone) 305{ 306 struct rapl_domain *rd = power_zone_to_rapl_domain(power_zone); 307 struct rapl_package *rp; 308 309 /* package zone is the last zone of a package, we can free 310 * memory here since all children has been unregistered. 311 */ 312 if (rd->id == RAPL_DOMAIN_PACKAGE) { 313 rp = find_package_by_id(rd->package_id); 314 if (!rp) { 315 dev_warn(&power_zone->dev, "no package id %s\n", 316 rd->name); 317 return -ENODEV; 318 } 319 kfree(rd); 320 rp->domains = NULL; 321 } 322 323 return 0; 324 325} 326 327static int find_nr_power_limit(struct rapl_domain *rd) 328{ 329 int i; 330 331 for (i = 0; i < NR_POWER_LIMITS; i++) { 332 if (rd->rpl[i].name == NULL) 333 break; 334 } 335 336 return i; 337} 338 339static int set_domain_enable(struct powercap_zone *power_zone, bool mode) 340{ 341 struct rapl_domain *rd = power_zone_to_rapl_domain(power_zone); 342 int nr_powerlimit; 343 344 if (rd->state & DOMAIN_STATE_BIOS_LOCKED) 345 return -EACCES; 346 get_online_cpus(); 347 nr_powerlimit = find_nr_power_limit(rd); 348 /* here we activate/deactivate the hardware for power limiting */ 349 rapl_write_data_raw(rd, PL1_ENABLE, mode); 350 /* always enable clamp such that p-state can go below OS requested 351 * range. power capping priority over guranteed frequency. 352 */ 353 rapl_write_data_raw(rd, PL1_CLAMP, mode); 354 /* some domains have pl2 */ 355 if (nr_powerlimit > 1) { 356 rapl_write_data_raw(rd, PL2_ENABLE, mode); 357 rapl_write_data_raw(rd, PL2_CLAMP, mode); 358 } 359 put_online_cpus(); 360 361 return 0; 362} 363 364static int get_domain_enable(struct powercap_zone *power_zone, bool *mode) 365{ 366 struct rapl_domain *rd = power_zone_to_rapl_domain(power_zone); 367 u64 val; 368 369 if (rd->state & DOMAIN_STATE_BIOS_LOCKED) { 370 *mode = false; 371 return 0; 372 } 373 get_online_cpus(); 374 if (rapl_read_data_raw(rd, PL1_ENABLE, true, &val)) { 375 put_online_cpus(); 376 return -EIO; 377 } 378 *mode = val; 379 put_online_cpus(); 380 381 return 0; 382} 383 384/* per RAPL domain ops, in the order of rapl_domain_type */ 385static struct powercap_zone_ops zone_ops[] = { 386 /* RAPL_DOMAIN_PACKAGE */ 387 { 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, 393 }, 394 /* RAPL_DOMAIN_PP0 */ 395 { 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, 401 }, 402 /* RAPL_DOMAIN_PP1 */ 403 { 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, 409 }, 410 /* RAPL_DOMAIN_DRAM */ 411 { 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, 417 }, 418}; 419 420static int set_power_limit(struct powercap_zone *power_zone, int id, 421 u64 power_limit) 422{ 423 struct rapl_domain *rd; 424 struct rapl_package *rp; 425 int ret = 0; 426 427 get_online_cpus(); 428 rd = power_zone_to_rapl_domain(power_zone); 429 rp = find_package_by_id(rd->package_id); 430 if (!rp) { 431 ret = -ENODEV; 432 goto set_exit; 433 } 434 435 if (rd->state & DOMAIN_STATE_BIOS_LOCKED) { 436 dev_warn(&power_zone->dev, "%s locked by BIOS, monitoring only\n", 437 rd->name); 438 ret = -EACCES; 439 goto set_exit; 440 } 441 442 switch (rd->rpl[id].prim_id) { 443 case PL1_ENABLE: 444 rapl_write_data_raw(rd, POWER_LIMIT1, power_limit); 445 break; 446 case PL2_ENABLE: 447 rapl_write_data_raw(rd, POWER_LIMIT2, power_limit); 448 break; 449 default: 450 ret = -EINVAL; 451 } 452 if (!ret) 453 package_power_limit_irq_save(rd->package_id); 454set_exit: 455 put_online_cpus(); 456 return ret; 457} 458 459static int get_current_power_limit(struct powercap_zone *power_zone, int id, 460 u64 *data) 461{ 462 struct rapl_domain *rd; 463 u64 val; 464 int prim; 465 int ret = 0; 466 467 get_online_cpus(); 468 rd = power_zone_to_rapl_domain(power_zone); 469 switch (rd->rpl[id].prim_id) { 470 case PL1_ENABLE: 471 prim = POWER_LIMIT1; 472 break; 473 case PL2_ENABLE: 474 prim = POWER_LIMIT2; 475 break; 476 default: 477 put_online_cpus(); 478 return -EINVAL; 479 } 480 if (rapl_read_data_raw(rd, prim, true, &val)) 481 ret = -EIO; 482 else 483 *data = val; 484 485 put_online_cpus(); 486 487 return ret; 488} 489 490static int set_time_window(struct powercap_zone *power_zone, int id, 491 u64 window) 492{ 493 struct rapl_domain *rd; 494 int ret = 0; 495 496 get_online_cpus(); 497 rd = power_zone_to_rapl_domain(power_zone); 498 switch (rd->rpl[id].prim_id) { 499 case PL1_ENABLE: 500 rapl_write_data_raw(rd, TIME_WINDOW1, window); 501 break; 502 case PL2_ENABLE: 503 rapl_write_data_raw(rd, TIME_WINDOW2, window); 504 break; 505 default: 506 ret = -EINVAL; 507 } 508 put_online_cpus(); 509 return ret; 510} 511 512static int get_time_window(struct powercap_zone *power_zone, int id, u64 *data) 513{ 514 struct rapl_domain *rd; 515 u64 val; 516 int ret = 0; 517 518 get_online_cpus(); 519 rd = power_zone_to_rapl_domain(power_zone); 520 switch (rd->rpl[id].prim_id) { 521 case PL1_ENABLE: 522 ret = rapl_read_data_raw(rd, TIME_WINDOW1, true, &val); 523 break; 524 case PL2_ENABLE: 525 ret = rapl_read_data_raw(rd, TIME_WINDOW2, true, &val); 526 break; 527 default: 528 put_online_cpus(); 529 return -EINVAL; 530 } 531 if (!ret) 532 *data = val; 533 put_online_cpus(); 534 535 return ret; 536} 537 538static const char *get_constraint_name(struct powercap_zone *power_zone, int id) 539{ 540 struct rapl_power_limit *rpl; 541 struct rapl_domain *rd; 542 543 rd = power_zone_to_rapl_domain(power_zone); 544 rpl = (struct rapl_power_limit *) &rd->rpl[id]; 545 546 return rpl->name; 547} 548 549 550static int get_max_power(struct powercap_zone *power_zone, int id, 551 u64 *data) 552{ 553 struct rapl_domain *rd; 554 u64 val; 555 int prim; 556 int ret = 0; 557 558 get_online_cpus(); 559 rd = power_zone_to_rapl_domain(power_zone); 560 switch (rd->rpl[id].prim_id) { 561 case PL1_ENABLE: 562 prim = THERMAL_SPEC_POWER; 563 break; 564 case PL2_ENABLE: 565 prim = MAX_POWER; 566 break; 567 default: 568 put_online_cpus(); 569 return -EINVAL; 570 } 571 if (rapl_read_data_raw(rd, prim, true, &val)) 572 ret = -EIO; 573 else 574 *data = val; 575 576 put_online_cpus(); 577 578 return ret; 579} 580 581static struct powercap_zone_constraint_ops constraint_ops = { 582 .set_power_limit_uw = set_power_limit, 583 .get_power_limit_uw = get_current_power_limit, 584 .set_time_window_us = set_time_window, 585 .get_time_window_us = get_time_window, 586 .get_max_power_uw = get_max_power, 587 .get_name = get_constraint_name, 588}; 589 590/* called after domain detection and package level data are set */ 591static void rapl_init_domains(struct rapl_package *rp) 592{ 593 int i; 594 struct rapl_domain *rd = rp->domains; 595 596 for (i = 0; i < RAPL_DOMAIN_MAX; i++) { 597 unsigned int mask = rp->domain_map & (1 << i); 598 switch (mask) { 599 case BIT(RAPL_DOMAIN_PACKAGE): 600 rd->name = rapl_domain_names[RAPL_DOMAIN_PACKAGE]; 601 rd->id = RAPL_DOMAIN_PACKAGE; 602 rd->msrs[0] = MSR_PKG_POWER_LIMIT; 603 rd->msrs[1] = MSR_PKG_ENERGY_STATUS; 604 rd->msrs[2] = MSR_PKG_PERF_STATUS; 605 rd->msrs[3] = 0; 606 rd->msrs[4] = MSR_PKG_POWER_INFO; 607 rd->rpl[0].prim_id = PL1_ENABLE; 608 rd->rpl[0].name = pl1_name; 609 rd->rpl[1].prim_id = PL2_ENABLE; 610 rd->rpl[1].name = pl2_name; 611 break; 612 case BIT(RAPL_DOMAIN_PP0): 613 rd->name = rapl_domain_names[RAPL_DOMAIN_PP0]; 614 rd->id = RAPL_DOMAIN_PP0; 615 rd->msrs[0] = MSR_PP0_POWER_LIMIT; 616 rd->msrs[1] = MSR_PP0_ENERGY_STATUS; 617 rd->msrs[2] = 0; 618 rd->msrs[3] = MSR_PP0_POLICY; 619 rd->msrs[4] = 0; 620 rd->rpl[0].prim_id = PL1_ENABLE; 621 rd->rpl[0].name = pl1_name; 622 break; 623 case BIT(RAPL_DOMAIN_PP1): 624 rd->name = rapl_domain_names[RAPL_DOMAIN_PP1]; 625 rd->id = RAPL_DOMAIN_PP1; 626 rd->msrs[0] = MSR_PP1_POWER_LIMIT; 627 rd->msrs[1] = MSR_PP1_ENERGY_STATUS; 628 rd->msrs[2] = 0; 629 rd->msrs[3] = MSR_PP1_POLICY; 630 rd->msrs[4] = 0; 631 rd->rpl[0].prim_id = PL1_ENABLE; 632 rd->rpl[0].name = pl1_name; 633 break; 634 case BIT(RAPL_DOMAIN_DRAM): 635 rd->name = rapl_domain_names[RAPL_DOMAIN_DRAM]; 636 rd->id = RAPL_DOMAIN_DRAM; 637 rd->msrs[0] = MSR_DRAM_POWER_LIMIT; 638 rd->msrs[1] = MSR_DRAM_ENERGY_STATUS; 639 rd->msrs[2] = MSR_DRAM_PERF_STATUS; 640 rd->msrs[3] = 0; 641 rd->msrs[4] = MSR_DRAM_POWER_INFO; 642 rd->rpl[0].prim_id = PL1_ENABLE; 643 rd->rpl[0].name = pl1_name; 644 break; 645 } 646 if (mask) { 647 rd->package_id = rp->id; 648 rd++; 649 } 650 } 651} 652 653static u64 rapl_unit_xlate(int package, enum unit_type type, u64 value, 654 int to_raw) 655{ 656 u64 divisor = 1; 657 int scale = 1; /* scale to user friendly data without floating point */ 658 u64 f, y; /* fraction and exp. used for time unit */ 659 struct rapl_package *rp; 660 661 rp = find_package_by_id(package); 662 if (!rp) 663 return value; 664 665 switch (type) { 666 case POWER_UNIT: 667 divisor = rp->power_unit_divisor; 668 scale = POWER_UNIT_SCALE; 669 break; 670 case ENERGY_UNIT: 671 scale = ENERGY_UNIT_SCALE; 672 divisor = rp->energy_unit_divisor; 673 break; 674 case TIME_UNIT: 675 divisor = rp->time_unit_divisor; 676 scale = TIME_UNIT_SCALE; 677 /* special processing based on 2^Y*(1+F)/4 = val/divisor, refer 678 * to Intel Software Developer's manual Vol. 3a, CH 14.7.4. 679 */ 680 if (!to_raw) { 681 f = (value & 0x60) >> 5; 682 y = value & 0x1f; 683 value = (1 << y) * (4 + f) * scale / 4; 684 return div64_u64(value, divisor); 685 } else { 686 do_div(value, scale); 687 value *= divisor; 688 y = ilog2(value); 689 f = div64_u64(4 * (value - (1 << y)), 1 << y); 690 value = (y & 0x1f) | ((f & 0x3) << 5); 691 return value; 692 } 693 break; 694 case ARBITRARY_UNIT: 695 default: 696 return value; 697 }; 698 699 if (to_raw) 700 return div64_u64(value * divisor, scale); 701 else 702 return div64_u64(value * scale, divisor); 703} 704 705/* in the order of enum rapl_primitives */ 706static struct rapl_primitive_info rpi[] = { 707 /* name, mask, shift, msr index, unit divisor */ 708 PRIMITIVE_INFO_INIT(ENERGY_COUNTER, ENERGY_STATUS_MASK, 0, 709 RAPL_DOMAIN_MSR_STATUS, ENERGY_UNIT, 0), 710 PRIMITIVE_INFO_INIT(POWER_LIMIT1, POWER_LIMIT1_MASK, 0, 711 RAPL_DOMAIN_MSR_LIMIT, POWER_UNIT, 0), 712 PRIMITIVE_INFO_INIT(POWER_LIMIT2, POWER_LIMIT2_MASK, 32, 713 RAPL_DOMAIN_MSR_LIMIT, POWER_UNIT, 0), 714 PRIMITIVE_INFO_INIT(FW_LOCK, POWER_PP_LOCK, 31, 715 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0), 716 PRIMITIVE_INFO_INIT(PL1_ENABLE, POWER_LIMIT1_ENABLE, 15, 717 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0), 718 PRIMITIVE_INFO_INIT(PL1_CLAMP, POWER_LIMIT1_CLAMP, 16, 719 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0), 720 PRIMITIVE_INFO_INIT(PL2_ENABLE, POWER_LIMIT2_ENABLE, 47, 721 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0), 722 PRIMITIVE_INFO_INIT(PL2_CLAMP, POWER_LIMIT2_CLAMP, 48, 723 RAPL_DOMAIN_MSR_LIMIT, ARBITRARY_UNIT, 0), 724 PRIMITIVE_INFO_INIT(TIME_WINDOW1, TIME_WINDOW1_MASK, 17, 725 RAPL_DOMAIN_MSR_LIMIT, TIME_UNIT, 0), 726 PRIMITIVE_INFO_INIT(TIME_WINDOW2, TIME_WINDOW2_MASK, 49, 727 RAPL_DOMAIN_MSR_LIMIT, TIME_UNIT, 0), 728 PRIMITIVE_INFO_INIT(THERMAL_SPEC_POWER, POWER_INFO_THERMAL_SPEC_MASK, 729 0, RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0), 730 PRIMITIVE_INFO_INIT(MAX_POWER, POWER_INFO_MAX_MASK, 32, 731 RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0), 732 PRIMITIVE_INFO_INIT(MIN_POWER, POWER_INFO_MIN_MASK, 16, 733 RAPL_DOMAIN_MSR_INFO, POWER_UNIT, 0), 734 PRIMITIVE_INFO_INIT(MAX_TIME_WINDOW, POWER_INFO_MAX_TIME_WIN_MASK, 48, 735 RAPL_DOMAIN_MSR_INFO, TIME_UNIT, 0), 736 PRIMITIVE_INFO_INIT(THROTTLED_TIME, PERF_STATUS_THROTTLE_TIME_MASK, 0, 737 RAPL_DOMAIN_MSR_PERF, TIME_UNIT, 0), 738 PRIMITIVE_INFO_INIT(PRIORITY_LEVEL, PP_POLICY_MASK, 0, 739 RAPL_DOMAIN_MSR_POLICY, ARBITRARY_UNIT, 0), 740 /* non-hardware */ 741 PRIMITIVE_INFO_INIT(AVERAGE_POWER, 0, 0, 0, POWER_UNIT, 742 RAPL_PRIMITIVE_DERIVED), 743 {NULL, 0, 0, 0}, 744}; 745 746/* Read primitive data based on its related struct rapl_primitive_info. 747 * if xlate flag is set, return translated data based on data units, i.e. 748 * time, energy, and power. 749 * RAPL MSRs are non-architectual and are laid out not consistently across 750 * domains. Here we use primitive info to allow writing consolidated access 751 * functions. 752 * For a given primitive, it is processed by MSR mask and shift. Unit conversion 753 * is pre-assigned based on RAPL unit MSRs read at init time. 754 * 63-------------------------- 31--------------------------- 0 755 * | xxxxx (mask) | 756 * | |<- shift ----------------| 757 * 63-------------------------- 31--------------------------- 0 758 */ 759static int rapl_read_data_raw(struct rapl_domain *rd, 760 enum rapl_primitives prim, 761 bool xlate, u64 *data) 762{ 763 u64 value, final; 764 u32 msr; 765 struct rapl_primitive_info *rp = &rpi[prim]; 766 int cpu; 767 768 if (!rp->name || rp->flag & RAPL_PRIMITIVE_DUMMY) 769 return -EINVAL; 770 771 msr = rd->msrs[rp->id]; 772 if (!msr) 773 return -EINVAL; 774 /* use physical package id to look up active cpus */ 775 cpu = find_active_cpu_on_package(rd->package_id); 776 if (cpu < 0) 777 return cpu; 778 779 /* special-case package domain, which uses a different bit*/ 780 if (prim == FW_LOCK && rd->id == RAPL_DOMAIN_PACKAGE) { 781 rp->mask = POWER_PACKAGE_LOCK; 782 rp->shift = 63; 783 } 784 /* non-hardware data are collected by the polling thread */ 785 if (rp->flag & RAPL_PRIMITIVE_DERIVED) { 786 *data = rd->rdd.primitives[prim]; 787 return 0; 788 } 789 790 if (rdmsrl_safe_on_cpu(cpu, msr, &value)) { 791 pr_debug("failed to read msr 0x%x on cpu %d\n", msr, cpu); 792 return -EIO; 793 } 794 795 final = value & rp->mask; 796 final = final >> rp->shift; 797 if (xlate) 798 *data = rapl_unit_xlate(rd->package_id, rp->unit, final, 0); 799 else 800 *data = final; 801 802 return 0; 803} 804 805/* Similar use of primitive info in the read counterpart */ 806static int rapl_write_data_raw(struct rapl_domain *rd, 807 enum rapl_primitives prim, 808 unsigned long long value) 809{ 810 u64 msr_val; 811 u32 msr; 812 struct rapl_primitive_info *rp = &rpi[prim]; 813 int cpu; 814 815 cpu = find_active_cpu_on_package(rd->package_id); 816 if (cpu < 0) 817 return cpu; 818 msr = rd->msrs[rp->id]; 819 if (rdmsrl_safe_on_cpu(cpu, msr, &msr_val)) { 820 dev_dbg(&rd->power_zone.dev, 821 "failed to read msr 0x%x on cpu %d\n", msr, cpu); 822 return -EIO; 823 } 824 value = rapl_unit_xlate(rd->package_id, rp->unit, value, 1); 825 msr_val &= ~rp->mask; 826 msr_val |= value << rp->shift; 827 if (wrmsrl_safe_on_cpu(cpu, msr, msr_val)) { 828 dev_dbg(&rd->power_zone.dev, 829 "failed to write msr 0x%x on cpu %d\n", msr, cpu); 830 return -EIO; 831 } 832 833 return 0; 834} 835 836static const struct x86_cpu_id energy_unit_quirk_ids[] = { 837 { X86_VENDOR_INTEL, 6, 0x37},/* Valleyview */ 838 {} 839}; 840 841static int rapl_check_unit(struct rapl_package *rp, int cpu) 842{ 843 u64 msr_val; 844 u32 value; 845 846 if (rdmsrl_safe_on_cpu(cpu, MSR_RAPL_POWER_UNIT, &msr_val)) { 847 pr_err("Failed to read power unit MSR 0x%x on CPU %d, exit.\n", 848 MSR_RAPL_POWER_UNIT, cpu); 849 return -ENODEV; 850 } 851 852 /* Raw RAPL data stored in MSRs are in certain scales. We need to 853 * convert them into standard units based on the divisors reported in 854 * the RAPL unit MSRs. 855 * i.e. 856 * energy unit: 1/enery_unit_divisor Joules 857 * power unit: 1/power_unit_divisor Watts 858 * time unit: 1/time_unit_divisor Seconds 859 */ 860 value = (msr_val & ENERGY_UNIT_MASK) >> ENERGY_UNIT_OFFSET; 861 /* some CPUs have different way to calculate energy unit */ 862 if (x86_match_cpu(energy_unit_quirk_ids)) 863 rp->energy_unit_divisor = 1000000 / (1 << value); 864 else 865 rp->energy_unit_divisor = 1 << value; 866 867 value = (msr_val & POWER_UNIT_MASK) >> POWER_UNIT_OFFSET; 868 rp->power_unit_divisor = 1 << value; 869 870 value = (msr_val & TIME_UNIT_MASK) >> TIME_UNIT_OFFSET; 871 rp->time_unit_divisor = 1 << value; 872 873 pr_debug("Physical package %d units: energy=%d, time=%d, power=%d\n", 874 rp->id, 875 rp->energy_unit_divisor, 876 rp->time_unit_divisor, 877 rp->power_unit_divisor); 878 879 return 0; 880} 881 882/* REVISIT: 883 * When package power limit is set artificially low by RAPL, LVT 884 * thermal interrupt for package power limit should be ignored 885 * since we are not really exceeding the real limit. The intention 886 * is to avoid excessive interrupts while we are trying to save power. 887 * A useful feature might be routing the package_power_limit interrupt 888 * to userspace via eventfd. once we have a usecase, this is simple 889 * to do by adding an atomic notifier. 890 */ 891 892static void package_power_limit_irq_save(int package_id) 893{ 894 u32 l, h = 0; 895 int cpu; 896 struct rapl_package *rp; 897 898 rp = find_package_by_id(package_id); 899 if (!rp) 900 return; 901 902 if (!boot_cpu_has(X86_FEATURE_PTS) || !boot_cpu_has(X86_FEATURE_PLN)) 903 return; 904 905 cpu = find_active_cpu_on_package(package_id); 906 if (cpu < 0) 907 return; 908 /* save the state of PLN irq mask bit before disabling it */ 909 rdmsr_safe_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &l, &h); 910 if (!(rp->power_limit_irq & PACKAGE_PLN_INT_SAVED)) { 911 rp->power_limit_irq = l & PACKAGE_THERM_INT_PLN_ENABLE; 912 rp->power_limit_irq |= PACKAGE_PLN_INT_SAVED; 913 } 914 l &= ~PACKAGE_THERM_INT_PLN_ENABLE; 915 wrmsr_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h); 916} 917 918/* restore per package power limit interrupt enable state */ 919static void package_power_limit_irq_restore(int package_id) 920{ 921 u32 l, h; 922 int cpu; 923 struct rapl_package *rp; 924 925 rp = find_package_by_id(package_id); 926 if (!rp) 927 return; 928 929 if (!boot_cpu_has(X86_FEATURE_PTS) || !boot_cpu_has(X86_FEATURE_PLN)) 930 return; 931 932 cpu = find_active_cpu_on_package(package_id); 933 if (cpu < 0) 934 return; 935 936 /* irq enable state not saved, nothing to restore */ 937 if (!(rp->power_limit_irq & PACKAGE_PLN_INT_SAVED)) 938 return; 939 rdmsr_safe_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, &l, &h); 940 941 if (rp->power_limit_irq & PACKAGE_THERM_INT_PLN_ENABLE) 942 l |= PACKAGE_THERM_INT_PLN_ENABLE; 943 else 944 l &= ~PACKAGE_THERM_INT_PLN_ENABLE; 945 946 wrmsr_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h); 947} 948 949static const struct x86_cpu_id rapl_ids[] = { 950 { X86_VENDOR_INTEL, 6, 0x2a},/* Sandy Bridge */ 951 { X86_VENDOR_INTEL, 6, 0x2d},/* Sandy Bridge EP */ 952 { X86_VENDOR_INTEL, 6, 0x37},/* Valleyview */ 953 { X86_VENDOR_INTEL, 6, 0x3a},/* Ivy Bridge */ 954 { X86_VENDOR_INTEL, 6, 0x3c},/* Haswell */ 955 { X86_VENDOR_INTEL, 6, 0x3d},/* Broadwell */ 956 { X86_VENDOR_INTEL, 6, 0x3f},/* Haswell */ 957 { X86_VENDOR_INTEL, 6, 0x45},/* Haswell ULT */ 958 /* TODO: Add more CPU IDs after testing */ 959 {} 960}; 961MODULE_DEVICE_TABLE(x86cpu, rapl_ids); 962 963/* read once for all raw primitive data for all packages, domains */ 964static void rapl_update_domain_data(void) 965{ 966 int dmn, prim; 967 u64 val; 968 struct rapl_package *rp; 969 970 list_for_each_entry(rp, &rapl_packages, plist) { 971 for (dmn = 0; dmn < rp->nr_domains; dmn++) { 972 pr_debug("update package %d domain %s data\n", rp->id, 973 rp->domains[dmn].name); 974 /* exclude non-raw primitives */ 975 for (prim = 0; prim < NR_RAW_PRIMITIVES; prim++) 976 if (!rapl_read_data_raw(&rp->domains[dmn], prim, 977 rpi[prim].unit, 978 &val)) 979 rp->domains[dmn].rdd.primitives[prim] = 980 val; 981 } 982 } 983 984} 985 986static int rapl_unregister_powercap(void) 987{ 988 struct rapl_package *rp; 989 struct rapl_domain *rd, *rd_package = NULL; 990 991 /* unregister all active rapl packages from the powercap layer, 992 * hotplug lock held 993 */ 994 list_for_each_entry(rp, &rapl_packages, plist) { 995 package_power_limit_irq_restore(rp->id); 996 997 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; 998 rd++) { 999 pr_debug("remove package, undo power limit on %d: %s\n", 1000 rp->id, rd->name); 1001 rapl_write_data_raw(rd, PL1_ENABLE, 0); 1002 rapl_write_data_raw(rd, PL2_ENABLE, 0); 1003 rapl_write_data_raw(rd, PL1_CLAMP, 0); 1004 rapl_write_data_raw(rd, PL2_CLAMP, 0); 1005 if (rd->id == RAPL_DOMAIN_PACKAGE) { 1006 rd_package = rd; 1007 continue; 1008 } 1009 powercap_unregister_zone(control_type, &rd->power_zone); 1010 } 1011 /* do the package zone last */ 1012 if (rd_package) 1013 powercap_unregister_zone(control_type, 1014 &rd_package->power_zone); 1015 } 1016 powercap_unregister_control_type(control_type); 1017 1018 return 0; 1019} 1020 1021static int rapl_package_register_powercap(struct rapl_package *rp) 1022{ 1023 struct rapl_domain *rd; 1024 int ret = 0; 1025 char dev_name[17]; /* max domain name = 7 + 1 + 8 for int + 1 for null*/ 1026 struct powercap_zone *power_zone = NULL; 1027 int nr_pl; 1028 1029 /* first we register package domain as the parent zone*/ 1030 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) { 1031 if (rd->id == RAPL_DOMAIN_PACKAGE) { 1032 nr_pl = find_nr_power_limit(rd); 1033 pr_debug("register socket %d package domain %s\n", 1034 rp->id, rd->name); 1035 memset(dev_name, 0, sizeof(dev_name)); 1036 snprintf(dev_name, sizeof(dev_name), "%s-%d", 1037 rd->name, rp->id); 1038 power_zone = powercap_register_zone(&rd->power_zone, 1039 control_type, 1040 dev_name, NULL, 1041 &zone_ops[rd->id], 1042 nr_pl, 1043 &constraint_ops); 1044 if (IS_ERR(power_zone)) { 1045 pr_debug("failed to register package, %d\n", 1046 rp->id); 1047 ret = PTR_ERR(power_zone); 1048 goto exit_package; 1049 } 1050 /* track parent zone in per package/socket data */ 1051 rp->power_zone = power_zone; 1052 /* done, only one package domain per socket */ 1053 break; 1054 } 1055 } 1056 if (!power_zone) { 1057 pr_err("no package domain found, unknown topology!\n"); 1058 ret = -ENODEV; 1059 goto exit_package; 1060 } 1061 /* now register domains as children of the socket/package*/ 1062 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) { 1063 if (rd->id == RAPL_DOMAIN_PACKAGE) 1064 continue; 1065 /* number of power limits per domain varies */ 1066 nr_pl = find_nr_power_limit(rd); 1067 power_zone = powercap_register_zone(&rd->power_zone, 1068 control_type, rd->name, 1069 rp->power_zone, 1070 &zone_ops[rd->id], nr_pl, 1071 &constraint_ops); 1072 1073 if (IS_ERR(power_zone)) { 1074 pr_debug("failed to register power_zone, %d:%s:%s\n", 1075 rp->id, rd->name, dev_name); 1076 ret = PTR_ERR(power_zone); 1077 goto err_cleanup; 1078 } 1079 } 1080 1081exit_package: 1082 return ret; 1083err_cleanup: 1084 /* clean up previously initialized domains within the package if we 1085 * failed after the first domain setup. 1086 */ 1087 while (--rd >= rp->domains) { 1088 pr_debug("unregister package %d domain %s\n", rp->id, rd->name); 1089 powercap_unregister_zone(control_type, &rd->power_zone); 1090 } 1091 1092 return ret; 1093} 1094 1095static int rapl_register_powercap(void) 1096{ 1097 struct rapl_domain *rd; 1098 struct rapl_package *rp; 1099 int ret = 0; 1100 1101 control_type = powercap_register_control_type(NULL, "intel-rapl", NULL); 1102 if (IS_ERR(control_type)) { 1103 pr_debug("failed to register powercap control_type.\n"); 1104 return PTR_ERR(control_type); 1105 } 1106 /* read the initial data */ 1107 rapl_update_domain_data(); 1108 list_for_each_entry(rp, &rapl_packages, plist) 1109 if (rapl_package_register_powercap(rp)) 1110 goto err_cleanup_package; 1111 return ret; 1112 1113err_cleanup_package: 1114 /* clean up previously initialized packages */ 1115 list_for_each_entry_continue_reverse(rp, &rapl_packages, plist) { 1116 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; 1117 rd++) { 1118 pr_debug("unregister zone/package %d, %s domain\n", 1119 rp->id, rd->name); 1120 powercap_unregister_zone(control_type, &rd->power_zone); 1121 } 1122 } 1123 1124 return ret; 1125} 1126 1127static int rapl_check_domain(int cpu, int domain) 1128{ 1129 unsigned msr; 1130 u64 val = 0; 1131 1132 switch (domain) { 1133 case RAPL_DOMAIN_PACKAGE: 1134 msr = MSR_PKG_ENERGY_STATUS; 1135 break; 1136 case RAPL_DOMAIN_PP0: 1137 msr = MSR_PP0_ENERGY_STATUS; 1138 break; 1139 case RAPL_DOMAIN_PP1: 1140 msr = MSR_PP1_ENERGY_STATUS; 1141 break; 1142 case RAPL_DOMAIN_DRAM: 1143 msr = MSR_DRAM_ENERGY_STATUS; 1144 break; 1145 default: 1146 pr_err("invalid domain id %d\n", domain); 1147 return -EINVAL; 1148 } 1149 /* make sure domain counters are available and contains non-zero 1150 * values, otherwise skip it. 1151 */ 1152 if (rdmsrl_safe_on_cpu(cpu, msr, &val) || !val) 1153 return -ENODEV; 1154 1155 return 0; 1156} 1157 1158/* Detect active and valid domains for the given CPU, caller must 1159 * ensure the CPU belongs to the targeted package and CPU hotlug is disabled. 1160 */ 1161static int rapl_detect_domains(struct rapl_package *rp, int cpu) 1162{ 1163 int i; 1164 int ret = 0; 1165 struct rapl_domain *rd; 1166 u64 locked; 1167 1168 for (i = 0; i < RAPL_DOMAIN_MAX; i++) { 1169 /* use physical package id to read counters */ 1170 if (!rapl_check_domain(cpu, i)) { 1171 rp->domain_map |= 1 << i; 1172 pr_info("Found RAPL domain %s\n", rapl_domain_names[i]); 1173 } 1174 } 1175 rp->nr_domains = bitmap_weight(&rp->domain_map, RAPL_DOMAIN_MAX); 1176 if (!rp->nr_domains) { 1177 pr_err("no valid rapl domains found in package %d\n", rp->id); 1178 ret = -ENODEV; 1179 goto done; 1180 } 1181 pr_debug("found %d domains on package %d\n", rp->nr_domains, rp->id); 1182 1183 rp->domains = kcalloc(rp->nr_domains + 1, sizeof(struct rapl_domain), 1184 GFP_KERNEL); 1185 if (!rp->domains) { 1186 ret = -ENOMEM; 1187 goto done; 1188 } 1189 rapl_init_domains(rp); 1190 1191 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) { 1192 /* check if the domain is locked by BIOS */ 1193 if (rapl_read_data_raw(rd, FW_LOCK, false, &locked)) { 1194 pr_info("RAPL package %d domain %s locked by BIOS\n", 1195 rp->id, rd->name); 1196 rd->state |= DOMAIN_STATE_BIOS_LOCKED; 1197 } 1198 } 1199 1200 1201done: 1202 return ret; 1203} 1204 1205static bool is_package_new(int package) 1206{ 1207 struct rapl_package *rp; 1208 1209 /* caller prevents cpu hotplug, there will be no new packages added 1210 * or deleted while traversing the package list, no need for locking. 1211 */ 1212 list_for_each_entry(rp, &rapl_packages, plist) 1213 if (package == rp->id) 1214 return false; 1215 1216 return true; 1217} 1218 1219/* RAPL interface can be made of a two-level hierarchy: package level and domain 1220 * level. We first detect the number of packages then domains of each package. 1221 * We have to consider the possiblity of CPU online/offline due to hotplug and 1222 * other scenarios. 1223 */ 1224static int rapl_detect_topology(void) 1225{ 1226 int i; 1227 int phy_package_id; 1228 struct rapl_package *new_package, *rp; 1229 1230 for_each_online_cpu(i) { 1231 phy_package_id = topology_physical_package_id(i); 1232 if (is_package_new(phy_package_id)) { 1233 new_package = kzalloc(sizeof(*rp), GFP_KERNEL); 1234 if (!new_package) { 1235 rapl_cleanup_data(); 1236 return -ENOMEM; 1237 } 1238 /* add the new package to the list */ 1239 new_package->id = phy_package_id; 1240 new_package->nr_cpus = 1; 1241 1242 /* check if the package contains valid domains */ 1243 if (rapl_detect_domains(new_package, i) || 1244 rapl_check_unit(new_package, i)) { 1245 kfree(new_package->domains); 1246 kfree(new_package); 1247 /* free up the packages already initialized */ 1248 rapl_cleanup_data(); 1249 return -ENODEV; 1250 } 1251 INIT_LIST_HEAD(&new_package->plist); 1252 list_add(&new_package->plist, &rapl_packages); 1253 } else { 1254 rp = find_package_by_id(phy_package_id); 1255 if (rp) 1256 ++rp->nr_cpus; 1257 } 1258 } 1259 1260 return 0; 1261} 1262 1263/* called from CPU hotplug notifier, hotplug lock held */ 1264static void rapl_remove_package(struct rapl_package *rp) 1265{ 1266 struct rapl_domain *rd, *rd_package = NULL; 1267 1268 for (rd = rp->domains; rd < rp->domains + rp->nr_domains; rd++) { 1269 if (rd->id == RAPL_DOMAIN_PACKAGE) { 1270 rd_package = rd; 1271 continue; 1272 } 1273 pr_debug("remove package %d, %s domain\n", rp->id, rd->name); 1274 powercap_unregister_zone(control_type, &rd->power_zone); 1275 } 1276 /* do parent zone last */ 1277 powercap_unregister_zone(control_type, &rd_package->power_zone); 1278 list_del(&rp->plist); 1279 kfree(rp); 1280} 1281 1282/* called from CPU hotplug notifier, hotplug lock held */ 1283static int rapl_add_package(int cpu) 1284{ 1285 int ret = 0; 1286 int phy_package_id; 1287 struct rapl_package *rp; 1288 1289 phy_package_id = topology_physical_package_id(cpu); 1290 rp = kzalloc(sizeof(struct rapl_package), GFP_KERNEL); 1291 if (!rp) 1292 return -ENOMEM; 1293 1294 /* add the new package to the list */ 1295 rp->id = phy_package_id; 1296 rp->nr_cpus = 1; 1297 /* check if the package contains valid domains */ 1298 if (rapl_detect_domains(rp, cpu) || 1299 rapl_check_unit(rp, cpu)) { 1300 ret = -ENODEV; 1301 goto err_free_package; 1302 } 1303 if (!rapl_package_register_powercap(rp)) { 1304 INIT_LIST_HEAD(&rp->plist); 1305 list_add(&rp->plist, &rapl_packages); 1306 return ret; 1307 } 1308 1309err_free_package: 1310 kfree(rp->domains); 1311 kfree(rp); 1312 1313 return ret; 1314} 1315 1316/* Handles CPU hotplug on multi-socket systems. 1317 * If a CPU goes online as the first CPU of the physical package 1318 * we add the RAPL package to the system. Similarly, when the last 1319 * CPU of the package is removed, we remove the RAPL package and its 1320 * associated domains. Cooling devices are handled accordingly at 1321 * per-domain level. 1322 */ 1323static int rapl_cpu_callback(struct notifier_block *nfb, 1324 unsigned long action, void *hcpu) 1325{ 1326 unsigned long cpu = (unsigned long)hcpu; 1327 int phy_package_id; 1328 struct rapl_package *rp; 1329 1330 phy_package_id = topology_physical_package_id(cpu); 1331 switch (action) { 1332 case CPU_ONLINE: 1333 case CPU_ONLINE_FROZEN: 1334 case CPU_DOWN_FAILED: 1335 case CPU_DOWN_FAILED_FROZEN: 1336 rp = find_package_by_id(phy_package_id); 1337 if (rp) 1338 ++rp->nr_cpus; 1339 else 1340 rapl_add_package(cpu); 1341 break; 1342 case CPU_DOWN_PREPARE: 1343 case CPU_DOWN_PREPARE_FROZEN: 1344 rp = find_package_by_id(phy_package_id); 1345 if (!rp) 1346 break; 1347 if (--rp->nr_cpus == 0) 1348 rapl_remove_package(rp); 1349 } 1350 1351 return NOTIFY_OK; 1352} 1353 1354static struct notifier_block rapl_cpu_notifier = { 1355 .notifier_call = rapl_cpu_callback, 1356}; 1357 1358static int __init rapl_init(void) 1359{ 1360 int ret = 0; 1361 1362 if (!x86_match_cpu(rapl_ids)) { 1363 pr_err("driver does not support CPU family %d model %d\n", 1364 boot_cpu_data.x86, boot_cpu_data.x86_model); 1365 1366 return -ENODEV; 1367 } 1368 1369 cpu_notifier_register_begin(); 1370 1371 /* prevent CPU hotplug during detection */ 1372 get_online_cpus(); 1373 ret = rapl_detect_topology(); 1374 if (ret) 1375 goto done; 1376 1377 if (rapl_register_powercap()) { 1378 rapl_cleanup_data(); 1379 ret = -ENODEV; 1380 goto done; 1381 } 1382 __register_hotcpu_notifier(&rapl_cpu_notifier); 1383done: 1384 put_online_cpus(); 1385 cpu_notifier_register_done(); 1386 1387 return ret; 1388} 1389 1390static void __exit rapl_exit(void) 1391{ 1392 cpu_notifier_register_begin(); 1393 get_online_cpus(); 1394 __unregister_hotcpu_notifier(&rapl_cpu_notifier); 1395 rapl_unregister_powercap(); 1396 rapl_cleanup_data(); 1397 put_online_cpus(); 1398 cpu_notifier_register_done(); 1399} 1400 1401module_init(rapl_init); 1402module_exit(rapl_exit); 1403 1404MODULE_DESCRIPTION("Driver for Intel RAPL (Running Average Power Limit)"); 1405MODULE_AUTHOR("Jacob Pan <jacob.jun.pan@intel.com>"); 1406MODULE_LICENSE("GPL v2"); 1407