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 #ifdef CONFIG_CRASH_HOTPLUG 286 static ssize_t crash_hotplug_show(struct device *dev, 287 struct device_attribute *attr, 288 char *buf) 289 { 290 return sysfs_emit(buf, "%d\n", crash_hotplug_cpu_support()); 291 } 292 static DEVICE_ATTR_ADMIN_RO(crash_hotplug); 293 #endif 294 295 static void cpu_device_release(struct device *dev) 296 { 297 /* 298 * This is an empty function to prevent the driver core from spitting a 299 * warning at us. Yes, I know this is directly opposite of what the 300 * documentation for the driver core and kobjects say, and the author 301 * of this code has already been publically ridiculed for doing 302 * something as foolish as this. However, at this point in time, it is 303 * the only way to handle the issue of statically allocated cpu 304 * devices. The different architectures will have their cpu device 305 * code reworked to properly handle this in the near future, so this 306 * function will then be changed to correctly free up the memory held 307 * by the cpu device. 308 * 309 * Never copy this way of doing things, or you too will be made fun of 310 * on the linux-kernel list, you have been warned. 311 */ 312 } 313 314 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 315 static ssize_t print_cpu_modalias(struct device *dev, 316 struct device_attribute *attr, 317 char *buf) 318 { 319 int len = 0; 320 u32 i; 321 322 len += sysfs_emit_at(buf, len, 323 "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:", 324 CPU_FEATURE_TYPEVAL); 325 326 for (i = 0; i < MAX_CPU_FEATURES; i++) 327 if (cpu_have_feature(i)) { 328 if (len + sizeof(",XXXX\n") >= PAGE_SIZE) { 329 WARN(1, "CPU features overflow page\n"); 330 break; 331 } 332 len += sysfs_emit_at(buf, len, ",%04X", i); 333 } 334 len += sysfs_emit_at(buf, len, "\n"); 335 return len; 336 } 337 338 static int cpu_uevent(const struct device *dev, struct kobj_uevent_env *env) 339 { 340 char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL); 341 if (buf) { 342 print_cpu_modalias(NULL, NULL, buf); 343 add_uevent_var(env, "MODALIAS=%s", buf); 344 kfree(buf); 345 } 346 return 0; 347 } 348 #endif 349 350 struct bus_type cpu_subsys = { 351 .name = "cpu", 352 .dev_name = "cpu", 353 .match = cpu_subsys_match, 354 #ifdef CONFIG_HOTPLUG_CPU 355 .online = cpu_subsys_online, 356 .offline = cpu_subsys_offline, 357 #endif 358 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 359 .uevent = cpu_uevent, 360 #endif 361 }; 362 EXPORT_SYMBOL_GPL(cpu_subsys); 363 364 /* 365 * register_cpu - Setup a sysfs device for a CPU. 366 * @cpu - cpu->hotpluggable field set to 1 will generate a control file in 367 * sysfs for this CPU. 368 * @num - CPU number to use when creating the device. 369 * 370 * Initialize and register the CPU device. 371 */ 372 int register_cpu(struct cpu *cpu, int num) 373 { 374 int error; 375 376 cpu->node_id = cpu_to_node(num); 377 memset(&cpu->dev, 0x00, sizeof(struct device)); 378 cpu->dev.id = num; 379 cpu->dev.bus = &cpu_subsys; 380 cpu->dev.release = cpu_device_release; 381 cpu->dev.offline_disabled = !cpu->hotpluggable; 382 cpu->dev.offline = !cpu_online(num); 383 cpu->dev.of_node = of_get_cpu_node(num, NULL); 384 cpu->dev.groups = common_cpu_attr_groups; 385 if (cpu->hotpluggable) 386 cpu->dev.groups = hotplugable_cpu_attr_groups; 387 error = device_register(&cpu->dev); 388 if (error) { 389 put_device(&cpu->dev); 390 return error; 391 } 392 393 per_cpu(cpu_sys_devices, num) = &cpu->dev; 394 register_cpu_under_node(num, cpu_to_node(num)); 395 dev_pm_qos_expose_latency_limit(&cpu->dev, 396 PM_QOS_RESUME_LATENCY_NO_CONSTRAINT); 397 398 return 0; 399 } 400 401 struct device *get_cpu_device(unsigned int cpu) 402 { 403 if (cpu < nr_cpu_ids && cpu_possible(cpu)) 404 return per_cpu(cpu_sys_devices, cpu); 405 else 406 return NULL; 407 } 408 EXPORT_SYMBOL_GPL(get_cpu_device); 409 410 static void device_create_release(struct device *dev) 411 { 412 kfree(dev); 413 } 414 415 __printf(4, 0) 416 static struct device * 417 __cpu_device_create(struct device *parent, void *drvdata, 418 const struct attribute_group **groups, 419 const char *fmt, va_list args) 420 { 421 struct device *dev = NULL; 422 int retval = -ENOMEM; 423 424 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 425 if (!dev) 426 goto error; 427 428 device_initialize(dev); 429 dev->parent = parent; 430 dev->groups = groups; 431 dev->release = device_create_release; 432 device_set_pm_not_required(dev); 433 dev_set_drvdata(dev, drvdata); 434 435 retval = kobject_set_name_vargs(&dev->kobj, fmt, args); 436 if (retval) 437 goto error; 438 439 retval = device_add(dev); 440 if (retval) 441 goto error; 442 443 return dev; 444 445 error: 446 put_device(dev); 447 return ERR_PTR(retval); 448 } 449 450 struct device *cpu_device_create(struct device *parent, void *drvdata, 451 const struct attribute_group **groups, 452 const char *fmt, ...) 453 { 454 va_list vargs; 455 struct device *dev; 456 457 va_start(vargs, fmt); 458 dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs); 459 va_end(vargs); 460 return dev; 461 } 462 EXPORT_SYMBOL_GPL(cpu_device_create); 463 464 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 465 static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL); 466 #endif 467 468 static struct attribute *cpu_root_attrs[] = { 469 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE 470 &dev_attr_probe.attr, 471 &dev_attr_release.attr, 472 #endif 473 &cpu_attrs[0].attr.attr, 474 &cpu_attrs[1].attr.attr, 475 &cpu_attrs[2].attr.attr, 476 &dev_attr_kernel_max.attr, 477 &dev_attr_offline.attr, 478 &dev_attr_isolated.attr, 479 #ifdef CONFIG_NO_HZ_FULL 480 &dev_attr_nohz_full.attr, 481 #endif 482 #ifdef CONFIG_CRASH_HOTPLUG 483 &dev_attr_crash_hotplug.attr, 484 #endif 485 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 486 &dev_attr_modalias.attr, 487 #endif 488 NULL 489 }; 490 491 static const struct attribute_group cpu_root_attr_group = { 492 .attrs = cpu_root_attrs, 493 }; 494 495 static const struct attribute_group *cpu_root_attr_groups[] = { 496 &cpu_root_attr_group, 497 NULL, 498 }; 499 500 bool cpu_is_hotpluggable(unsigned int cpu) 501 { 502 struct device *dev = get_cpu_device(cpu); 503 return dev && container_of(dev, struct cpu, dev)->hotpluggable 504 && tick_nohz_cpu_hotpluggable(cpu); 505 } 506 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable); 507 508 #ifdef CONFIG_GENERIC_CPU_DEVICES 509 static DEFINE_PER_CPU(struct cpu, cpu_devices); 510 #endif 511 512 static void __init cpu_dev_register_generic(void) 513 { 514 #ifdef CONFIG_GENERIC_CPU_DEVICES 515 int i; 516 517 for_each_possible_cpu(i) { 518 if (register_cpu(&per_cpu(cpu_devices, i), i)) 519 panic("Failed to register CPU device"); 520 } 521 #endif 522 } 523 524 #ifdef CONFIG_GENERIC_CPU_VULNERABILITIES 525 static ssize_t cpu_show_not_affected(struct device *dev, 526 struct device_attribute *attr, char *buf) 527 { 528 return sysfs_emit(buf, "Not affected\n"); 529 } 530 531 #define CPU_SHOW_VULN_FALLBACK(func) \ 532 ssize_t cpu_show_##func(struct device *, \ 533 struct device_attribute *, char *) \ 534 __attribute__((weak, alias("cpu_show_not_affected"))) 535 536 CPU_SHOW_VULN_FALLBACK(meltdown); 537 CPU_SHOW_VULN_FALLBACK(spectre_v1); 538 CPU_SHOW_VULN_FALLBACK(spectre_v2); 539 CPU_SHOW_VULN_FALLBACK(spec_store_bypass); 540 CPU_SHOW_VULN_FALLBACK(l1tf); 541 CPU_SHOW_VULN_FALLBACK(mds); 542 CPU_SHOW_VULN_FALLBACK(tsx_async_abort); 543 CPU_SHOW_VULN_FALLBACK(itlb_multihit); 544 CPU_SHOW_VULN_FALLBACK(srbds); 545 CPU_SHOW_VULN_FALLBACK(mmio_stale_data); 546 CPU_SHOW_VULN_FALLBACK(retbleed); 547 CPU_SHOW_VULN_FALLBACK(spec_rstack_overflow); 548 CPU_SHOW_VULN_FALLBACK(gds); 549 550 static DEVICE_ATTR(meltdown, 0444, cpu_show_meltdown, NULL); 551 static DEVICE_ATTR(spectre_v1, 0444, cpu_show_spectre_v1, NULL); 552 static DEVICE_ATTR(spectre_v2, 0444, cpu_show_spectre_v2, NULL); 553 static DEVICE_ATTR(spec_store_bypass, 0444, cpu_show_spec_store_bypass, NULL); 554 static DEVICE_ATTR(l1tf, 0444, cpu_show_l1tf, NULL); 555 static DEVICE_ATTR(mds, 0444, cpu_show_mds, NULL); 556 static DEVICE_ATTR(tsx_async_abort, 0444, cpu_show_tsx_async_abort, NULL); 557 static DEVICE_ATTR(itlb_multihit, 0444, cpu_show_itlb_multihit, NULL); 558 static DEVICE_ATTR(srbds, 0444, cpu_show_srbds, NULL); 559 static DEVICE_ATTR(mmio_stale_data, 0444, cpu_show_mmio_stale_data, NULL); 560 static DEVICE_ATTR(retbleed, 0444, cpu_show_retbleed, NULL); 561 static DEVICE_ATTR(spec_rstack_overflow, 0444, cpu_show_spec_rstack_overflow, NULL); 562 static DEVICE_ATTR(gather_data_sampling, 0444, cpu_show_gds, NULL); 563 564 static struct attribute *cpu_root_vulnerabilities_attrs[] = { 565 &dev_attr_meltdown.attr, 566 &dev_attr_spectre_v1.attr, 567 &dev_attr_spectre_v2.attr, 568 &dev_attr_spec_store_bypass.attr, 569 &dev_attr_l1tf.attr, 570 &dev_attr_mds.attr, 571 &dev_attr_tsx_async_abort.attr, 572 &dev_attr_itlb_multihit.attr, 573 &dev_attr_srbds.attr, 574 &dev_attr_mmio_stale_data.attr, 575 &dev_attr_retbleed.attr, 576 &dev_attr_spec_rstack_overflow.attr, 577 &dev_attr_gather_data_sampling.attr, 578 NULL 579 }; 580 581 static const struct attribute_group cpu_root_vulnerabilities_group = { 582 .name = "vulnerabilities", 583 .attrs = cpu_root_vulnerabilities_attrs, 584 }; 585 586 static void __init cpu_register_vulnerabilities(void) 587 { 588 struct device *dev = bus_get_dev_root(&cpu_subsys); 589 590 if (dev) { 591 if (sysfs_create_group(&dev->kobj, &cpu_root_vulnerabilities_group)) 592 pr_err("Unable to register CPU vulnerabilities\n"); 593 put_device(dev); 594 } 595 } 596 597 #else 598 static inline void cpu_register_vulnerabilities(void) { } 599 #endif 600 601 void __init cpu_dev_init(void) 602 { 603 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups)) 604 panic("Failed to register CPU subsystem"); 605 606 cpu_dev_register_generic(); 607 cpu_register_vulnerabilities(); 608 } 609