1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright IBM Corp. 2007, 2011 4 */ 5 6 #define KMSG_COMPONENT "cpu" 7 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 8 9 #include <linux/workqueue.h> 10 #include <linux/memblock.h> 11 #include <linux/uaccess.h> 12 #include <linux/sysctl.h> 13 #include <linux/cpuset.h> 14 #include <linux/device.h> 15 #include <linux/export.h> 16 #include <linux/kernel.h> 17 #include <linux/sched.h> 18 #include <linux/sched/topology.h> 19 #include <linux/delay.h> 20 #include <linux/init.h> 21 #include <linux/slab.h> 22 #include <linux/cpu.h> 23 #include <linux/smp.h> 24 #include <linux/mm.h> 25 #include <linux/nodemask.h> 26 #include <linux/node.h> 27 #include <asm/hiperdispatch.h> 28 #include <asm/sysinfo.h> 29 #include <asm/asm.h> 30 31 #define PTF_HORIZONTAL (0UL) 32 #define PTF_VERTICAL (1UL) 33 #define PTF_CHECK (2UL) 34 35 enum { 36 TOPOLOGY_MODE_HW, 37 TOPOLOGY_MODE_SINGLE, 38 TOPOLOGY_MODE_PACKAGE, 39 TOPOLOGY_MODE_UNINITIALIZED 40 }; 41 42 struct mask_info { 43 struct mask_info *next; 44 unsigned char id; 45 cpumask_t mask; 46 }; 47 48 static int topology_mode = TOPOLOGY_MODE_UNINITIALIZED; 49 static void set_topology_timer(void); 50 static void topology_work_fn(struct work_struct *work); 51 static struct sysinfo_15_1_x *tl_info; 52 static int cpu_management; 53 54 static DECLARE_WORK(topology_work, topology_work_fn); 55 56 /* 57 * Socket/Book linked lists and cpu_topology updates are 58 * protected by "sched_domains_mutex". 59 */ 60 static struct mask_info socket_info; 61 static struct mask_info book_info; 62 static struct mask_info drawer_info; 63 64 struct cpu_topology_s390 cpu_topology[NR_CPUS]; 65 EXPORT_SYMBOL_GPL(cpu_topology); 66 67 static void cpu_group_map(cpumask_t *dst, struct mask_info *info, unsigned int cpu) 68 { 69 static cpumask_t mask; 70 71 cpumask_clear(&mask); 72 if (!cpumask_test_cpu(cpu, &cpu_setup_mask)) 73 goto out; 74 cpumask_set_cpu(cpu, &mask); 75 switch (topology_mode) { 76 case TOPOLOGY_MODE_HW: 77 while (info) { 78 if (cpumask_test_cpu(cpu, &info->mask)) { 79 cpumask_copy(&mask, &info->mask); 80 break; 81 } 82 info = info->next; 83 } 84 break; 85 case TOPOLOGY_MODE_PACKAGE: 86 cpumask_copy(&mask, cpu_present_mask); 87 break; 88 default: 89 fallthrough; 90 case TOPOLOGY_MODE_SINGLE: 91 break; 92 } 93 cpumask_and(&mask, &mask, &cpu_setup_mask); 94 out: 95 cpumask_copy(dst, &mask); 96 } 97 98 static void cpu_thread_map(cpumask_t *dst, unsigned int cpu) 99 { 100 static cpumask_t mask; 101 unsigned int max_cpu; 102 103 cpumask_clear(&mask); 104 if (!cpumask_test_cpu(cpu, &cpu_setup_mask)) 105 goto out; 106 cpumask_set_cpu(cpu, &mask); 107 if (topology_mode != TOPOLOGY_MODE_HW) 108 goto out; 109 cpu -= cpu % (smp_cpu_mtid + 1); 110 max_cpu = min(cpu + smp_cpu_mtid, nr_cpu_ids - 1); 111 for (; cpu <= max_cpu; cpu++) { 112 if (cpumask_test_cpu(cpu, &cpu_setup_mask)) 113 cpumask_set_cpu(cpu, &mask); 114 } 115 out: 116 cpumask_copy(dst, &mask); 117 } 118 119 #define TOPOLOGY_CORE_BITS 64 120 121 static void add_cpus_to_mask(struct topology_core *tl_core, 122 struct mask_info *drawer, 123 struct mask_info *book, 124 struct mask_info *socket) 125 { 126 struct cpu_topology_s390 *topo; 127 unsigned int core; 128 129 for_each_set_bit(core, &tl_core->mask, TOPOLOGY_CORE_BITS) { 130 unsigned int max_cpu, rcore; 131 int cpu; 132 133 rcore = TOPOLOGY_CORE_BITS - 1 - core + tl_core->origin; 134 cpu = smp_find_processor_id(rcore << smp_cpu_mt_shift); 135 if (cpu < 0) 136 continue; 137 max_cpu = min(cpu + smp_cpu_mtid, nr_cpu_ids - 1); 138 for (; cpu <= max_cpu; cpu++) { 139 topo = &cpu_topology[cpu]; 140 topo->drawer_id = drawer->id; 141 topo->book_id = book->id; 142 topo->socket_id = socket->id; 143 topo->core_id = rcore; 144 topo->thread_id = cpu; 145 topo->dedicated = tl_core->d; 146 cpumask_set_cpu(cpu, &drawer->mask); 147 cpumask_set_cpu(cpu, &book->mask); 148 cpumask_set_cpu(cpu, &socket->mask); 149 smp_cpu_set_polarization(cpu, tl_core->pp); 150 smp_cpu_set_capacity(cpu, CPU_CAPACITY_HIGH); 151 } 152 } 153 } 154 155 static void clear_masks(void) 156 { 157 struct mask_info *info; 158 159 info = &socket_info; 160 while (info) { 161 cpumask_clear(&info->mask); 162 info = info->next; 163 } 164 info = &book_info; 165 while (info) { 166 cpumask_clear(&info->mask); 167 info = info->next; 168 } 169 info = &drawer_info; 170 while (info) { 171 cpumask_clear(&info->mask); 172 info = info->next; 173 } 174 } 175 176 static union topology_entry *next_tle(union topology_entry *tle) 177 { 178 if (!tle->nl) 179 return (union topology_entry *)((struct topology_core *)tle + 1); 180 return (union topology_entry *)((struct topology_container *)tle + 1); 181 } 182 183 static void tl_to_masks(struct sysinfo_15_1_x *info) 184 { 185 struct mask_info *socket = &socket_info; 186 struct mask_info *book = &book_info; 187 struct mask_info *drawer = &drawer_info; 188 union topology_entry *tle, *end; 189 190 clear_masks(); 191 tle = info->tle; 192 end = (union topology_entry *)((unsigned long)info + info->length); 193 while (tle < end) { 194 switch (tle->nl) { 195 case 3: 196 drawer = drawer->next; 197 drawer->id = tle->container.id; 198 break; 199 case 2: 200 book = book->next; 201 book->id = tle->container.id; 202 break; 203 case 1: 204 socket = socket->next; 205 socket->id = tle->container.id; 206 break; 207 case 0: 208 add_cpus_to_mask(&tle->cpu, drawer, book, socket); 209 break; 210 default: 211 clear_masks(); 212 return; 213 } 214 tle = next_tle(tle); 215 } 216 } 217 218 static void topology_update_polarization_simple(void) 219 { 220 int cpu; 221 222 for_each_possible_cpu(cpu) 223 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ); 224 } 225 226 static int ptf(unsigned long fc) 227 { 228 int cc; 229 230 asm volatile( 231 " .insn rre,0xb9a20000,%[fc],%[fc]\n" 232 CC_IPM(cc) 233 : CC_OUT(cc, cc) 234 : [fc] "d" (fc) 235 : CC_CLOBBER); 236 return CC_TRANSFORM(cc); 237 } 238 239 int topology_set_cpu_management(int fc) 240 { 241 int cpu, rc; 242 243 if (!MACHINE_HAS_TOPOLOGY) 244 return -EOPNOTSUPP; 245 if (fc) 246 rc = ptf(PTF_VERTICAL); 247 else 248 rc = ptf(PTF_HORIZONTAL); 249 if (rc) 250 return -EBUSY; 251 for_each_possible_cpu(cpu) 252 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN); 253 return rc; 254 } 255 256 void update_cpu_masks(void) 257 { 258 struct cpu_topology_s390 *topo, *topo_package, *topo_sibling; 259 int cpu, sibling, pkg_first, smt_first, id; 260 261 for_each_possible_cpu(cpu) { 262 topo = &cpu_topology[cpu]; 263 cpu_thread_map(&topo->thread_mask, cpu); 264 cpu_group_map(&topo->core_mask, &socket_info, cpu); 265 cpu_group_map(&topo->book_mask, &book_info, cpu); 266 cpu_group_map(&topo->drawer_mask, &drawer_info, cpu); 267 topo->booted_cores = 0; 268 if (topology_mode != TOPOLOGY_MODE_HW) { 269 id = topology_mode == TOPOLOGY_MODE_PACKAGE ? 0 : cpu; 270 topo->thread_id = cpu; 271 topo->core_id = cpu; 272 topo->socket_id = id; 273 topo->book_id = id; 274 topo->drawer_id = id; 275 } 276 } 277 hd_reset_state(); 278 for_each_online_cpu(cpu) { 279 topo = &cpu_topology[cpu]; 280 pkg_first = cpumask_first(&topo->core_mask); 281 topo_package = &cpu_topology[pkg_first]; 282 if (cpu == pkg_first) { 283 for_each_cpu(sibling, &topo->core_mask) { 284 topo_sibling = &cpu_topology[sibling]; 285 smt_first = cpumask_first(&topo_sibling->thread_mask); 286 if (sibling == smt_first) { 287 topo_package->booted_cores++; 288 hd_add_core(sibling); 289 } 290 } 291 } else { 292 topo->booted_cores = topo_package->booted_cores; 293 } 294 } 295 } 296 297 void store_topology(struct sysinfo_15_1_x *info) 298 { 299 stsi(info, 15, 1, topology_mnest_limit()); 300 } 301 302 static void __arch_update_dedicated_flag(void *arg) 303 { 304 if (topology_cpu_dedicated(smp_processor_id())) 305 set_cpu_flag(CIF_DEDICATED_CPU); 306 else 307 clear_cpu_flag(CIF_DEDICATED_CPU); 308 } 309 310 static int __arch_update_cpu_topology(void) 311 { 312 struct sysinfo_15_1_x *info = tl_info; 313 int rc, hd_status; 314 315 hd_status = 0; 316 rc = 0; 317 mutex_lock(&smp_cpu_state_mutex); 318 if (MACHINE_HAS_TOPOLOGY) { 319 rc = 1; 320 store_topology(info); 321 tl_to_masks(info); 322 } 323 update_cpu_masks(); 324 if (!MACHINE_HAS_TOPOLOGY) 325 topology_update_polarization_simple(); 326 if (cpu_management == 1) 327 hd_status = hd_enable_hiperdispatch(); 328 mutex_unlock(&smp_cpu_state_mutex); 329 if (hd_status == 0) 330 hd_disable_hiperdispatch(); 331 return rc; 332 } 333 334 int arch_update_cpu_topology(void) 335 { 336 int rc; 337 338 rc = __arch_update_cpu_topology(); 339 on_each_cpu(__arch_update_dedicated_flag, NULL, 0); 340 return rc; 341 } 342 343 static void topology_work_fn(struct work_struct *work) 344 { 345 rebuild_sched_domains(); 346 } 347 348 void topology_schedule_update(void) 349 { 350 schedule_work(&topology_work); 351 } 352 353 static void topology_flush_work(void) 354 { 355 flush_work(&topology_work); 356 } 357 358 static void topology_timer_fn(struct timer_list *unused) 359 { 360 if (ptf(PTF_CHECK)) 361 topology_schedule_update(); 362 set_topology_timer(); 363 } 364 365 static struct timer_list topology_timer; 366 367 static atomic_t topology_poll = ATOMIC_INIT(0); 368 369 static void set_topology_timer(void) 370 { 371 if (atomic_add_unless(&topology_poll, -1, 0)) 372 mod_timer(&topology_timer, jiffies + msecs_to_jiffies(100)); 373 else 374 mod_timer(&topology_timer, jiffies + msecs_to_jiffies(60 * MSEC_PER_SEC)); 375 } 376 377 void topology_expect_change(void) 378 { 379 if (!MACHINE_HAS_TOPOLOGY) 380 return; 381 /* This is racy, but it doesn't matter since it is just a heuristic. 382 * Worst case is that we poll in a higher frequency for a bit longer. 383 */ 384 if (atomic_read(&topology_poll) > 60) 385 return; 386 atomic_add(60, &topology_poll); 387 set_topology_timer(); 388 } 389 390 static int set_polarization(int polarization) 391 { 392 int rc = 0; 393 394 cpus_read_lock(); 395 mutex_lock(&smp_cpu_state_mutex); 396 if (cpu_management == polarization) 397 goto out; 398 rc = topology_set_cpu_management(polarization); 399 if (rc) 400 goto out; 401 cpu_management = polarization; 402 topology_expect_change(); 403 out: 404 mutex_unlock(&smp_cpu_state_mutex); 405 cpus_read_unlock(); 406 return rc; 407 } 408 409 static ssize_t dispatching_show(struct device *dev, 410 struct device_attribute *attr, 411 char *buf) 412 { 413 ssize_t count; 414 415 mutex_lock(&smp_cpu_state_mutex); 416 count = sysfs_emit(buf, "%d\n", cpu_management); 417 mutex_unlock(&smp_cpu_state_mutex); 418 return count; 419 } 420 421 static ssize_t dispatching_store(struct device *dev, 422 struct device_attribute *attr, 423 const char *buf, 424 size_t count) 425 { 426 int val, rc; 427 char delim; 428 429 if (sscanf(buf, "%d %c", &val, &delim) != 1) 430 return -EINVAL; 431 if (val != 0 && val != 1) 432 return -EINVAL; 433 rc = set_polarization(val); 434 return rc ? rc : count; 435 } 436 static DEVICE_ATTR_RW(dispatching); 437 438 static ssize_t cpu_polarization_show(struct device *dev, 439 struct device_attribute *attr, char *buf) 440 { 441 int cpu = dev->id; 442 ssize_t count; 443 444 mutex_lock(&smp_cpu_state_mutex); 445 switch (smp_cpu_get_polarization(cpu)) { 446 case POLARIZATION_HRZ: 447 count = sysfs_emit(buf, "horizontal\n"); 448 break; 449 case POLARIZATION_VL: 450 count = sysfs_emit(buf, "vertical:low\n"); 451 break; 452 case POLARIZATION_VM: 453 count = sysfs_emit(buf, "vertical:medium\n"); 454 break; 455 case POLARIZATION_VH: 456 count = sysfs_emit(buf, "vertical:high\n"); 457 break; 458 default: 459 count = sysfs_emit(buf, "unknown\n"); 460 break; 461 } 462 mutex_unlock(&smp_cpu_state_mutex); 463 return count; 464 } 465 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL); 466 467 static struct attribute *topology_cpu_attrs[] = { 468 &dev_attr_polarization.attr, 469 NULL, 470 }; 471 472 static struct attribute_group topology_cpu_attr_group = { 473 .attrs = topology_cpu_attrs, 474 }; 475 476 static ssize_t cpu_dedicated_show(struct device *dev, 477 struct device_attribute *attr, char *buf) 478 { 479 int cpu = dev->id; 480 ssize_t count; 481 482 mutex_lock(&smp_cpu_state_mutex); 483 count = sysfs_emit(buf, "%d\n", topology_cpu_dedicated(cpu)); 484 mutex_unlock(&smp_cpu_state_mutex); 485 return count; 486 } 487 static DEVICE_ATTR(dedicated, 0444, cpu_dedicated_show, NULL); 488 489 static struct attribute *topology_extra_cpu_attrs[] = { 490 &dev_attr_dedicated.attr, 491 NULL, 492 }; 493 494 static struct attribute_group topology_extra_cpu_attr_group = { 495 .attrs = topology_extra_cpu_attrs, 496 }; 497 498 int topology_cpu_init(struct cpu *cpu) 499 { 500 int rc; 501 502 rc = sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group); 503 if (rc || !MACHINE_HAS_TOPOLOGY) 504 return rc; 505 rc = sysfs_create_group(&cpu->dev.kobj, &topology_extra_cpu_attr_group); 506 if (rc) 507 sysfs_remove_group(&cpu->dev.kobj, &topology_cpu_attr_group); 508 return rc; 509 } 510 511 static const struct cpumask *cpu_thread_mask(int cpu) 512 { 513 return &cpu_topology[cpu].thread_mask; 514 } 515 516 517 const struct cpumask *cpu_coregroup_mask(int cpu) 518 { 519 return &cpu_topology[cpu].core_mask; 520 } 521 522 static const struct cpumask *cpu_book_mask(int cpu) 523 { 524 return &cpu_topology[cpu].book_mask; 525 } 526 527 static const struct cpumask *cpu_drawer_mask(int cpu) 528 { 529 return &cpu_topology[cpu].drawer_mask; 530 } 531 532 static struct sched_domain_topology_level s390_topology[] = { 533 { cpu_thread_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, 534 { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, 535 { cpu_book_mask, SD_INIT_NAME(BOOK) }, 536 { cpu_drawer_mask, SD_INIT_NAME(DRAWER) }, 537 { cpu_cpu_mask, SD_INIT_NAME(PKG) }, 538 { NULL, }, 539 }; 540 541 static void __init alloc_masks(struct sysinfo_15_1_x *info, 542 struct mask_info *mask, int offset) 543 { 544 int i, nr_masks; 545 546 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset]; 547 for (i = 0; i < info->mnest - offset; i++) 548 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i]; 549 nr_masks = max(nr_masks, 1); 550 for (i = 0; i < nr_masks; i++) { 551 mask->next = memblock_alloc(sizeof(*mask->next), 8); 552 if (!mask->next) 553 panic("%s: Failed to allocate %zu bytes align=0x%x\n", 554 __func__, sizeof(*mask->next), 8); 555 mask = mask->next; 556 } 557 } 558 559 void __init topology_init_early(void) 560 { 561 struct sysinfo_15_1_x *info; 562 563 set_sched_topology(s390_topology); 564 if (topology_mode == TOPOLOGY_MODE_UNINITIALIZED) { 565 if (MACHINE_HAS_TOPOLOGY) 566 topology_mode = TOPOLOGY_MODE_HW; 567 else 568 topology_mode = TOPOLOGY_MODE_SINGLE; 569 } 570 if (!MACHINE_HAS_TOPOLOGY) 571 goto out; 572 tl_info = memblock_alloc(PAGE_SIZE, PAGE_SIZE); 573 if (!tl_info) 574 panic("%s: Failed to allocate %lu bytes align=0x%lx\n", 575 __func__, PAGE_SIZE, PAGE_SIZE); 576 info = tl_info; 577 store_topology(info); 578 pr_info("The CPU configuration topology of the machine is: %d %d %d %d %d %d / %d\n", 579 info->mag[0], info->mag[1], info->mag[2], info->mag[3], 580 info->mag[4], info->mag[5], info->mnest); 581 alloc_masks(info, &socket_info, 1); 582 alloc_masks(info, &book_info, 2); 583 alloc_masks(info, &drawer_info, 3); 584 out: 585 cpumask_set_cpu(0, &cpu_setup_mask); 586 __arch_update_cpu_topology(); 587 __arch_update_dedicated_flag(NULL); 588 } 589 590 static inline int topology_get_mode(int enabled) 591 { 592 if (!enabled) 593 return TOPOLOGY_MODE_SINGLE; 594 return MACHINE_HAS_TOPOLOGY ? TOPOLOGY_MODE_HW : TOPOLOGY_MODE_PACKAGE; 595 } 596 597 static inline int topology_is_enabled(void) 598 { 599 return topology_mode != TOPOLOGY_MODE_SINGLE; 600 } 601 602 static int __init topology_setup(char *str) 603 { 604 bool enabled; 605 int rc; 606 607 rc = kstrtobool(str, &enabled); 608 if (rc) 609 return rc; 610 topology_mode = topology_get_mode(enabled); 611 return 0; 612 } 613 early_param("topology", topology_setup); 614 615 static int topology_ctl_handler(const struct ctl_table *ctl, int write, 616 void *buffer, size_t *lenp, loff_t *ppos) 617 { 618 int enabled = topology_is_enabled(); 619 int new_mode; 620 int rc; 621 struct ctl_table ctl_entry = { 622 .procname = ctl->procname, 623 .data = &enabled, 624 .maxlen = sizeof(int), 625 .extra1 = SYSCTL_ZERO, 626 .extra2 = SYSCTL_ONE, 627 }; 628 629 rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos); 630 if (rc < 0 || !write) 631 return rc; 632 633 mutex_lock(&smp_cpu_state_mutex); 634 new_mode = topology_get_mode(enabled); 635 if (topology_mode != new_mode) { 636 topology_mode = new_mode; 637 topology_schedule_update(); 638 } 639 mutex_unlock(&smp_cpu_state_mutex); 640 topology_flush_work(); 641 642 return rc; 643 } 644 645 static int polarization_ctl_handler(const struct ctl_table *ctl, int write, 646 void *buffer, size_t *lenp, loff_t *ppos) 647 { 648 int polarization; 649 int rc; 650 struct ctl_table ctl_entry = { 651 .procname = ctl->procname, 652 .data = &polarization, 653 .maxlen = sizeof(int), 654 .extra1 = SYSCTL_ZERO, 655 .extra2 = SYSCTL_ONE, 656 }; 657 658 polarization = cpu_management; 659 rc = proc_douintvec_minmax(&ctl_entry, write, buffer, lenp, ppos); 660 if (rc < 0 || !write) 661 return rc; 662 return set_polarization(polarization); 663 } 664 665 static struct ctl_table topology_ctl_table[] = { 666 { 667 .procname = "topology", 668 .mode = 0644, 669 .proc_handler = topology_ctl_handler, 670 }, 671 { 672 .procname = "polarization", 673 .mode = 0644, 674 .proc_handler = polarization_ctl_handler, 675 }, 676 }; 677 678 static int __init topology_init(void) 679 { 680 struct device *dev_root; 681 int rc = 0; 682 683 timer_setup(&topology_timer, topology_timer_fn, TIMER_DEFERRABLE); 684 if (MACHINE_HAS_TOPOLOGY) 685 set_topology_timer(); 686 else 687 topology_update_polarization_simple(); 688 if (IS_ENABLED(CONFIG_SCHED_TOPOLOGY_VERTICAL)) 689 set_polarization(1); 690 register_sysctl("s390", topology_ctl_table); 691 692 dev_root = bus_get_dev_root(&cpu_subsys); 693 if (dev_root) { 694 rc = device_create_file(dev_root, &dev_attr_dispatching); 695 put_device(dev_root); 696 } 697 return rc; 698 } 699 device_initcall(topology_init); 700