1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * CPU subsystem support 4 */ 5 6 #include <linux/kernel.h> 7 #include <linux/module.h> 8 #include <linux/init.h> 9 #include <linux/sched.h> 10 #include <linux/cpu.h> 11 #include <linux/topology.h> 12 #include <linux/device.h> 13 #include <linux/node.h> 14 #include <linux/gfp.h> 15 #include <linux/slab.h> 16 #include <linux/percpu.h> 17 #include <linux/acpi.h> 18 #include <linux/of.h> 19 #include <linux/cpufeature.h> 20 #include <linux/tick.h> 21 #include <linux/pm_qos.h> 22 #include <linux/delay.h> 23 #include <linux/sched/isolation.h> 24 25 #include "base.h" 26 27 static DEFINE_PER_CPU(struct device *, cpu_sys_devices); 28 29 static int cpu_subsys_match(struct device *dev, struct device_driver *drv) 30 { 31 /* ACPI style match is the only one that may succeed. */ 32 if (acpi_driver_match_device(dev, drv)) 33 return 1; 34 35 return 0; 36 } 37 38 #ifdef CONFIG_HOTPLUG_CPU 39 static void change_cpu_under_node(struct cpu *cpu, 40 unsigned int from_nid, unsigned int to_nid) 41 { 42 int cpuid = cpu->dev.id; 43 unregister_cpu_under_node(cpuid, from_nid); 44 register_cpu_under_node(cpuid, to_nid); 45 cpu->node_id = to_nid; 46 } 47 48 static int cpu_subsys_online(struct device *dev) 49 { 50 struct cpu *cpu = container_of(dev, struct cpu, dev); 51 int cpuid = dev->id; 52 int from_nid, to_nid; 53 int ret; 54 int retries = 0; 55 56 from_nid = cpu_to_node(cpuid); 57 if (from_nid == NUMA_NO_NODE) 58 return -ENODEV; 59 60 retry: 61 ret = cpu_device_up(dev); 62 63 /* 64 * If -EBUSY is returned, it is likely that hotplug is temporarily 65 * disabled when cpu_hotplug_disable() was called. This condition is 66 * transient. So we retry after waiting for an exponentially 67 * increasing delay up to a total of at least 620ms as some PCI 68 * device initialization can take quite a while. 69 */ 70 if (ret == -EBUSY) { 71 retries++; 72 if (retries > 5) 73 return ret; 74 msleep(10 * (1 << retries)); 75 goto retry; 76 } 77 78 /* 79 * When hot adding memory to memoryless node and enabling a cpu 80 * on the node, node number of the cpu may internally change. 81 */ 82 to_nid = cpu_to_node(cpuid); 83 if (from_nid != to_nid) 84 change_cpu_under_node(cpu, from_nid, to_nid); 85 86 return ret; 87 } 88 89 static int cpu_subsys_offline(struct device *dev) 90 { 91 return cpu_device_down(dev); 92 } 93 94 void unregister_cpu(struct cpu *cpu) 95 { 96 int logical_cpu = cpu->dev.id; 97 98 unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu)); 99 100 device_unregister(&cpu->dev); 101 per_cpu(cpu_sys_devices, logical_cpu) = NULL; 102 return; 103 } 104 105 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE 106 static ssize_t cpu_probe_store(struct device *dev, 107 struct device_attribute *attr, 108 const char *buf, 109 size_t count) 110 { 111 ssize_t cnt; 112 int ret; 113 114 ret = lock_device_hotplug_sysfs(); 115 if (ret) 116 return ret; 117 118 cnt = arch_cpu_probe(buf, count); 119 120 unlock_device_hotplug(); 121 return cnt; 122 } 123 124 static ssize_t cpu_release_store(struct device *dev, 125 struct device_attribute *attr, 126 const char *buf, 127 size_t count) 128 { 129 ssize_t cnt; 130 int ret; 131 132 ret = lock_device_hotplug_sysfs(); 133 if (ret) 134 return ret; 135 136 cnt = arch_cpu_release(buf, count); 137 138 unlock_device_hotplug(); 139 return cnt; 140 } 141 142 static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store); 143 static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store); 144 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ 145 #endif /* CONFIG_HOTPLUG_CPU */ 146 147 #ifdef CONFIG_CRASH_DUMP 148 #include <linux/kexec.h> 149 150 static ssize_t crash_notes_show(struct device *dev, 151 struct device_attribute *attr, 152 char *buf) 153 { 154 struct cpu *cpu = container_of(dev, struct cpu, dev); 155 unsigned long long addr; 156 int cpunum; 157 158 cpunum = cpu->dev.id; 159 160 /* 161 * Might be reading other cpu's data based on which cpu read thread 162 * has been scheduled. But cpu data (memory) is allocated once during 163 * boot up and this data does not change there after. Hence this 164 * operation should be safe. No locking required. 165 */ 166 addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum)); 167 168 return sysfs_emit(buf, "%llx\n", addr); 169 } 170 static DEVICE_ATTR_ADMIN_RO(crash_notes); 171 172 static ssize_t crash_notes_size_show(struct device *dev, 173 struct device_attribute *attr, 174 char *buf) 175 { 176 return sysfs_emit(buf, "%zu\n", sizeof(note_buf_t)); 177 } 178 static DEVICE_ATTR_ADMIN_RO(crash_notes_size); 179 180 static struct attribute *crash_note_cpu_attrs[] = { 181 &dev_attr_crash_notes.attr, 182 &dev_attr_crash_notes_size.attr, 183 NULL 184 }; 185 186 static const struct attribute_group crash_note_cpu_attr_group = { 187 .attrs = crash_note_cpu_attrs, 188 }; 189 #endif 190 191 static const struct attribute_group *common_cpu_attr_groups[] = { 192 #ifdef CONFIG_CRASH_DUMP 193 &crash_note_cpu_attr_group, 194 #endif 195 NULL 196 }; 197 198 static const struct attribute_group *hotplugable_cpu_attr_groups[] = { 199 #ifdef CONFIG_CRASH_DUMP 200 &crash_note_cpu_attr_group, 201 #endif 202 NULL 203 }; 204 205 /* 206 * Print cpu online, possible, present, and system maps 207 */ 208 209 struct cpu_attr { 210 struct device_attribute attr; 211 const struct cpumask *const map; 212 }; 213 214 static ssize_t show_cpus_attr(struct device *dev, 215 struct device_attribute *attr, 216 char *buf) 217 { 218 struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr); 219 220 return cpumap_print_to_pagebuf(true, buf, ca->map); 221 } 222 223 #define _CPU_ATTR(name, map) \ 224 { __ATTR(name, 0444, show_cpus_attr, NULL), map } 225 226 /* Keep in sync with cpu_subsys_attrs */ 227 static struct cpu_attr cpu_attrs[] = { 228 _CPU_ATTR(online, &__cpu_online_mask), 229 _CPU_ATTR(possible, &__cpu_possible_mask), 230 _CPU_ATTR(present, &__cpu_present_mask), 231 }; 232 233 /* 234 * Print values for NR_CPUS and offlined cpus 235 */ 236 static ssize_t print_cpus_kernel_max(struct device *dev, 237 struct device_attribute *attr, char *buf) 238 { 239 return sysfs_emit(buf, "%d\n", NR_CPUS - 1); 240 } 241 static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL); 242 243 /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */ 244 unsigned int total_cpus; 245 246 static ssize_t print_cpus_offline(struct device *dev, 247 struct device_attribute *attr, char *buf) 248 { 249 int len = 0; 250 cpumask_var_t offline; 251 252 /* display offline cpus < nr_cpu_ids */ 253 if (!alloc_cpumask_var(&offline, GFP_KERNEL)) 254 return -ENOMEM; 255 cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask); 256 len += sysfs_emit_at(buf, len, "%*pbl", cpumask_pr_args(offline)); 257 free_cpumask_var(offline); 258 259 /* display offline cpus >= nr_cpu_ids */ 260 if (total_cpus && nr_cpu_ids < total_cpus) { 261 len += sysfs_emit_at(buf, len, ","); 262 263 if (nr_cpu_ids == total_cpus-1) 264 len += sysfs_emit_at(buf, len, "%u", nr_cpu_ids); 265 else 266 len += sysfs_emit_at(buf, len, "%u-%d", 267 nr_cpu_ids, total_cpus - 1); 268 } 269 270 len += sysfs_emit_at(buf, len, "\n"); 271 272 return len; 273 } 274 static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL); 275 276 static ssize_t print_cpus_isolated(struct device *dev, 277 struct device_attribute *attr, char *buf) 278 { 279 int len; 280 cpumask_var_t isolated; 281 282 if (!alloc_cpumask_var(&isolated, GFP_KERNEL)) 283 return -ENOMEM; 284 285 cpumask_andnot(isolated, cpu_possible_mask, 286 housekeeping_cpumask(HK_TYPE_DOMAIN)); 287 len = sysfs_emit(buf, "%*pbl\n", cpumask_pr_args(isolated)); 288 289 free_cpumask_var(isolated); 290 291 return len; 292 } 293 static DEVICE_ATTR(isolated, 0444, print_cpus_isolated, NULL); 294 295 #ifdef CONFIG_NO_HZ_FULL 296 static ssize_t print_cpus_nohz_full(struct device *dev, 297 struct device_attribute *attr, char *buf) 298 { 299 return sysfs_emit(buf, "%*pbl\n", cpumask_pr_args(tick_nohz_full_mask)); 300 } 301 static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL); 302 #endif 303 304 #ifdef CONFIG_CRASH_HOTPLUG 305 static ssize_t crash_hotplug_show(struct device *dev, 306 struct device_attribute *attr, 307 char *buf) 308 { 309 return sysfs_emit(buf, "%d\n", crash_check_hotplug_support()); 310 } 311 static DEVICE_ATTR_ADMIN_RO(crash_hotplug); 312 #endif 313 314 static void cpu_device_release(struct device *dev) 315 { 316 /* 317 * This is an empty function to prevent the driver core from spitting a 318 * warning at us. Yes, I know this is directly opposite of what the 319 * documentation for the driver core and kobjects say, and the author 320 * of this code has already been publically ridiculed for doing 321 * something as foolish as this. However, at this point in time, it is 322 * the only way to handle the issue of statically allocated cpu 323 * devices. The different architectures will have their cpu device 324 * code reworked to properly handle this in the near future, so this 325 * function will then be changed to correctly free up the memory held 326 * by the cpu device. 327 * 328 * Never copy this way of doing things, or you too will be made fun of 329 * on the linux-kernel list, you have been warned. 330 */ 331 } 332 333 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 334 static ssize_t print_cpu_modalias(struct device *dev, 335 struct device_attribute *attr, 336 char *buf) 337 { 338 int len = 0; 339 u32 i; 340 341 len += sysfs_emit_at(buf, len, 342 "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:", 343 CPU_FEATURE_TYPEVAL); 344 345 for (i = 0; i < MAX_CPU_FEATURES; i++) 346 if (cpu_have_feature(i)) { 347 if (len + sizeof(",XXXX\n") >= PAGE_SIZE) { 348 WARN(1, "CPU features overflow page\n"); 349 break; 350 } 351 len += sysfs_emit_at(buf, len, ",%04X", i); 352 } 353 len += sysfs_emit_at(buf, len, "\n"); 354 return len; 355 } 356 357 static int cpu_uevent(const struct device *dev, struct kobj_uevent_env *env) 358 { 359 char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL); 360 if (buf) { 361 print_cpu_modalias(NULL, NULL, buf); 362 add_uevent_var(env, "MODALIAS=%s", buf); 363 kfree(buf); 364 } 365 return 0; 366 } 367 #endif 368 369 const struct bus_type cpu_subsys = { 370 .name = "cpu", 371 .dev_name = "cpu", 372 .match = cpu_subsys_match, 373 #ifdef CONFIG_HOTPLUG_CPU 374 .online = cpu_subsys_online, 375 .offline = cpu_subsys_offline, 376 #endif 377 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 378 .uevent = cpu_uevent, 379 #endif 380 }; 381 EXPORT_SYMBOL_GPL(cpu_subsys); 382 383 /* 384 * register_cpu - Setup a sysfs device for a CPU. 385 * @cpu - cpu->hotpluggable field set to 1 will generate a control file in 386 * sysfs for this CPU. 387 * @num - CPU number to use when creating the device. 388 * 389 * Initialize and register the CPU device. 390 */ 391 int register_cpu(struct cpu *cpu, int num) 392 { 393 int error; 394 395 cpu->node_id = cpu_to_node(num); 396 memset(&cpu->dev, 0x00, sizeof(struct device)); 397 cpu->dev.id = num; 398 cpu->dev.bus = &cpu_subsys; 399 cpu->dev.release = cpu_device_release; 400 cpu->dev.offline_disabled = !cpu->hotpluggable; 401 cpu->dev.offline = !cpu_online(num); 402 cpu->dev.of_node = of_get_cpu_node(num, NULL); 403 cpu->dev.groups = common_cpu_attr_groups; 404 if (cpu->hotpluggable) 405 cpu->dev.groups = hotplugable_cpu_attr_groups; 406 error = device_register(&cpu->dev); 407 if (error) { 408 put_device(&cpu->dev); 409 return error; 410 } 411 412 per_cpu(cpu_sys_devices, num) = &cpu->dev; 413 register_cpu_under_node(num, cpu_to_node(num)); 414 dev_pm_qos_expose_latency_limit(&cpu->dev, 415 PM_QOS_RESUME_LATENCY_NO_CONSTRAINT); 416 417 return 0; 418 } 419 420 struct device *get_cpu_device(unsigned int cpu) 421 { 422 if (cpu < nr_cpu_ids && cpu_possible(cpu)) 423 return per_cpu(cpu_sys_devices, cpu); 424 else 425 return NULL; 426 } 427 EXPORT_SYMBOL_GPL(get_cpu_device); 428 429 static void device_create_release(struct device *dev) 430 { 431 kfree(dev); 432 } 433 434 __printf(4, 0) 435 static struct device * 436 __cpu_device_create(struct device *parent, void *drvdata, 437 const struct attribute_group **groups, 438 const char *fmt, va_list args) 439 { 440 struct device *dev = NULL; 441 int retval = -ENOMEM; 442 443 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 444 if (!dev) 445 goto error; 446 447 device_initialize(dev); 448 dev->parent = parent; 449 dev->groups = groups; 450 dev->release = device_create_release; 451 device_set_pm_not_required(dev); 452 dev_set_drvdata(dev, drvdata); 453 454 retval = kobject_set_name_vargs(&dev->kobj, fmt, args); 455 if (retval) 456 goto error; 457 458 retval = device_add(dev); 459 if (retval) 460 goto error; 461 462 return dev; 463 464 error: 465 put_device(dev); 466 return ERR_PTR(retval); 467 } 468 469 struct device *cpu_device_create(struct device *parent, void *drvdata, 470 const struct attribute_group **groups, 471 const char *fmt, ...) 472 { 473 va_list vargs; 474 struct device *dev; 475 476 va_start(vargs, fmt); 477 dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs); 478 va_end(vargs); 479 return dev; 480 } 481 EXPORT_SYMBOL_GPL(cpu_device_create); 482 483 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 484 static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL); 485 #endif 486 487 static struct attribute *cpu_root_attrs[] = { 488 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE 489 &dev_attr_probe.attr, 490 &dev_attr_release.attr, 491 #endif 492 &cpu_attrs[0].attr.attr, 493 &cpu_attrs[1].attr.attr, 494 &cpu_attrs[2].attr.attr, 495 &dev_attr_kernel_max.attr, 496 &dev_attr_offline.attr, 497 &dev_attr_isolated.attr, 498 #ifdef CONFIG_NO_HZ_FULL 499 &dev_attr_nohz_full.attr, 500 #endif 501 #ifdef CONFIG_CRASH_HOTPLUG 502 &dev_attr_crash_hotplug.attr, 503 #endif 504 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 505 &dev_attr_modalias.attr, 506 #endif 507 NULL 508 }; 509 510 static const struct attribute_group cpu_root_attr_group = { 511 .attrs = cpu_root_attrs, 512 }; 513 514 static const struct attribute_group *cpu_root_attr_groups[] = { 515 &cpu_root_attr_group, 516 NULL, 517 }; 518 519 bool cpu_is_hotpluggable(unsigned int cpu) 520 { 521 struct device *dev = get_cpu_device(cpu); 522 return dev && container_of(dev, struct cpu, dev)->hotpluggable 523 && tick_nohz_cpu_hotpluggable(cpu); 524 } 525 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable); 526 527 #ifdef CONFIG_GENERIC_CPU_DEVICES 528 DEFINE_PER_CPU(struct cpu, cpu_devices); 529 530 bool __weak arch_cpu_is_hotpluggable(int cpu) 531 { 532 return false; 533 } 534 535 int __weak arch_register_cpu(int cpu) 536 { 537 struct cpu *c = &per_cpu(cpu_devices, cpu); 538 539 c->hotpluggable = arch_cpu_is_hotpluggable(cpu); 540 541 return register_cpu(c, cpu); 542 } 543 544 #ifdef CONFIG_HOTPLUG_CPU 545 void __weak arch_unregister_cpu(int num) 546 { 547 unregister_cpu(&per_cpu(cpu_devices, num)); 548 } 549 #endif /* CONFIG_HOTPLUG_CPU */ 550 #endif /* CONFIG_GENERIC_CPU_DEVICES */ 551 552 static void __init cpu_dev_register_generic(void) 553 { 554 int i, ret; 555 556 if (!IS_ENABLED(CONFIG_GENERIC_CPU_DEVICES)) 557 return; 558 559 for_each_present_cpu(i) { 560 ret = arch_register_cpu(i); 561 if (ret) 562 pr_warn("register_cpu %d failed (%d)\n", i, ret); 563 } 564 } 565 566 #ifdef CONFIG_GENERIC_CPU_VULNERABILITIES 567 static ssize_t cpu_show_not_affected(struct device *dev, 568 struct device_attribute *attr, char *buf) 569 { 570 return sysfs_emit(buf, "Not affected\n"); 571 } 572 573 #define CPU_SHOW_VULN_FALLBACK(func) \ 574 ssize_t cpu_show_##func(struct device *, \ 575 struct device_attribute *, char *) \ 576 __attribute__((weak, alias("cpu_show_not_affected"))) 577 578 CPU_SHOW_VULN_FALLBACK(meltdown); 579 CPU_SHOW_VULN_FALLBACK(spectre_v1); 580 CPU_SHOW_VULN_FALLBACK(spectre_v2); 581 CPU_SHOW_VULN_FALLBACK(spec_store_bypass); 582 CPU_SHOW_VULN_FALLBACK(l1tf); 583 CPU_SHOW_VULN_FALLBACK(mds); 584 CPU_SHOW_VULN_FALLBACK(tsx_async_abort); 585 CPU_SHOW_VULN_FALLBACK(itlb_multihit); 586 CPU_SHOW_VULN_FALLBACK(srbds); 587 CPU_SHOW_VULN_FALLBACK(mmio_stale_data); 588 CPU_SHOW_VULN_FALLBACK(retbleed); 589 CPU_SHOW_VULN_FALLBACK(spec_rstack_overflow); 590 CPU_SHOW_VULN_FALLBACK(gds); 591 CPU_SHOW_VULN_FALLBACK(reg_file_data_sampling); 592 593 static DEVICE_ATTR(meltdown, 0444, cpu_show_meltdown, NULL); 594 static DEVICE_ATTR(spectre_v1, 0444, cpu_show_spectre_v1, NULL); 595 static DEVICE_ATTR(spectre_v2, 0444, cpu_show_spectre_v2, NULL); 596 static DEVICE_ATTR(spec_store_bypass, 0444, cpu_show_spec_store_bypass, NULL); 597 static DEVICE_ATTR(l1tf, 0444, cpu_show_l1tf, NULL); 598 static DEVICE_ATTR(mds, 0444, cpu_show_mds, NULL); 599 static DEVICE_ATTR(tsx_async_abort, 0444, cpu_show_tsx_async_abort, NULL); 600 static DEVICE_ATTR(itlb_multihit, 0444, cpu_show_itlb_multihit, NULL); 601 static DEVICE_ATTR(srbds, 0444, cpu_show_srbds, NULL); 602 static DEVICE_ATTR(mmio_stale_data, 0444, cpu_show_mmio_stale_data, NULL); 603 static DEVICE_ATTR(retbleed, 0444, cpu_show_retbleed, NULL); 604 static DEVICE_ATTR(spec_rstack_overflow, 0444, cpu_show_spec_rstack_overflow, NULL); 605 static DEVICE_ATTR(gather_data_sampling, 0444, cpu_show_gds, NULL); 606 static DEVICE_ATTR(reg_file_data_sampling, 0444, cpu_show_reg_file_data_sampling, NULL); 607 608 static struct attribute *cpu_root_vulnerabilities_attrs[] = { 609 &dev_attr_meltdown.attr, 610 &dev_attr_spectre_v1.attr, 611 &dev_attr_spectre_v2.attr, 612 &dev_attr_spec_store_bypass.attr, 613 &dev_attr_l1tf.attr, 614 &dev_attr_mds.attr, 615 &dev_attr_tsx_async_abort.attr, 616 &dev_attr_itlb_multihit.attr, 617 &dev_attr_srbds.attr, 618 &dev_attr_mmio_stale_data.attr, 619 &dev_attr_retbleed.attr, 620 &dev_attr_spec_rstack_overflow.attr, 621 &dev_attr_gather_data_sampling.attr, 622 &dev_attr_reg_file_data_sampling.attr, 623 NULL 624 }; 625 626 static const struct attribute_group cpu_root_vulnerabilities_group = { 627 .name = "vulnerabilities", 628 .attrs = cpu_root_vulnerabilities_attrs, 629 }; 630 631 static void __init cpu_register_vulnerabilities(void) 632 { 633 struct device *dev = bus_get_dev_root(&cpu_subsys); 634 635 if (dev) { 636 if (sysfs_create_group(&dev->kobj, &cpu_root_vulnerabilities_group)) 637 pr_err("Unable to register CPU vulnerabilities\n"); 638 put_device(dev); 639 } 640 } 641 642 #else 643 static inline void cpu_register_vulnerabilities(void) { } 644 #endif 645 646 void __init cpu_dev_init(void) 647 { 648 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups)) 649 panic("Failed to register CPU subsystem"); 650 651 cpu_dev_register_generic(); 652 cpu_register_vulnerabilities(); 653 } 654