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