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_up(cpuid); 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_down(dev->id); 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 struct bus_type cpu_subsys = { 129 .name = "cpu", 130 .dev_name = "cpu", 131 .match = cpu_subsys_match, 132 #ifdef CONFIG_HOTPLUG_CPU 133 .online = cpu_subsys_online, 134 .offline = cpu_subsys_offline, 135 #endif 136 }; 137 EXPORT_SYMBOL_GPL(cpu_subsys); 138 139 #ifdef CONFIG_KEXEC 140 #include <linux/kexec.h> 141 142 static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr, 143 char *buf) 144 { 145 struct cpu *cpu = container_of(dev, struct cpu, dev); 146 ssize_t rc; 147 unsigned long long addr; 148 int cpunum; 149 150 cpunum = cpu->dev.id; 151 152 /* 153 * Might be reading other cpu's data based on which cpu read thread 154 * has been scheduled. But cpu data (memory) is allocated once during 155 * boot up and this data does not change there after. Hence this 156 * operation should be safe. No locking required. 157 */ 158 addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum)); 159 rc = sprintf(buf, "%Lx\n", addr); 160 return rc; 161 } 162 static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL); 163 164 static ssize_t show_crash_notes_size(struct device *dev, 165 struct device_attribute *attr, 166 char *buf) 167 { 168 ssize_t rc; 169 170 rc = sprintf(buf, "%zu\n", sizeof(note_buf_t)); 171 return rc; 172 } 173 static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL); 174 175 static struct attribute *crash_note_cpu_attrs[] = { 176 &dev_attr_crash_notes.attr, 177 &dev_attr_crash_notes_size.attr, 178 NULL 179 }; 180 181 static struct attribute_group crash_note_cpu_attr_group = { 182 .attrs = crash_note_cpu_attrs, 183 }; 184 #endif 185 186 static const struct attribute_group *common_cpu_attr_groups[] = { 187 #ifdef CONFIG_KEXEC 188 &crash_note_cpu_attr_group, 189 #endif 190 NULL 191 }; 192 193 static const struct attribute_group *hotplugable_cpu_attr_groups[] = { 194 #ifdef CONFIG_KEXEC 195 &crash_note_cpu_attr_group, 196 #endif 197 NULL 198 }; 199 200 /* 201 * Print cpu online, possible, present, and system maps 202 */ 203 204 struct cpu_attr { 205 struct device_attribute attr; 206 const struct cpumask *const map; 207 }; 208 209 static ssize_t show_cpus_attr(struct device *dev, 210 struct device_attribute *attr, 211 char *buf) 212 { 213 struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr); 214 215 return cpumap_print_to_pagebuf(true, buf, ca->map); 216 } 217 218 #define _CPU_ATTR(name, map) \ 219 { __ATTR(name, 0444, show_cpus_attr, NULL), map } 220 221 /* Keep in sync with cpu_subsys_attrs */ 222 static struct cpu_attr cpu_attrs[] = { 223 _CPU_ATTR(online, &__cpu_online_mask), 224 _CPU_ATTR(possible, &__cpu_possible_mask), 225 _CPU_ATTR(present, &__cpu_present_mask), 226 }; 227 228 /* 229 * Print values for NR_CPUS and offlined cpus 230 */ 231 static ssize_t print_cpus_kernel_max(struct device *dev, 232 struct device_attribute *attr, char *buf) 233 { 234 int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1); 235 return n; 236 } 237 static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL); 238 239 /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */ 240 unsigned int total_cpus; 241 242 static ssize_t print_cpus_offline(struct device *dev, 243 struct device_attribute *attr, char *buf) 244 { 245 int n = 0, len = PAGE_SIZE-2; 246 cpumask_var_t offline; 247 248 /* display offline cpus < nr_cpu_ids */ 249 if (!alloc_cpumask_var(&offline, GFP_KERNEL)) 250 return -ENOMEM; 251 cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask); 252 n = scnprintf(buf, len, "%*pbl", cpumask_pr_args(offline)); 253 free_cpumask_var(offline); 254 255 /* display offline cpus >= nr_cpu_ids */ 256 if (total_cpus && nr_cpu_ids < total_cpus) { 257 if (n && n < len) 258 buf[n++] = ','; 259 260 if (nr_cpu_ids == total_cpus-1) 261 n += snprintf(&buf[n], len - n, "%u", nr_cpu_ids); 262 else 263 n += snprintf(&buf[n], len - n, "%u-%d", 264 nr_cpu_ids, total_cpus-1); 265 } 266 267 n += snprintf(&buf[n], len - n, "\n"); 268 return n; 269 } 270 static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL); 271 272 static ssize_t print_cpus_isolated(struct device *dev, 273 struct device_attribute *attr, char *buf) 274 { 275 int n = 0, len = PAGE_SIZE-2; 276 cpumask_var_t isolated; 277 278 if (!alloc_cpumask_var(&isolated, GFP_KERNEL)) 279 return -ENOMEM; 280 281 cpumask_andnot(isolated, cpu_possible_mask, 282 housekeeping_cpumask(HK_FLAG_DOMAIN)); 283 n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(isolated)); 284 285 free_cpumask_var(isolated); 286 287 return n; 288 } 289 static DEVICE_ATTR(isolated, 0444, print_cpus_isolated, NULL); 290 291 #ifdef CONFIG_NO_HZ_FULL 292 static ssize_t print_cpus_nohz_full(struct device *dev, 293 struct device_attribute *attr, char *buf) 294 { 295 int n = 0, len = PAGE_SIZE-2; 296 297 n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(tick_nohz_full_mask)); 298 299 return n; 300 } 301 static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL); 302 #endif 303 304 static void cpu_device_release(struct device *dev) 305 { 306 /* 307 * This is an empty function to prevent the driver core from spitting a 308 * warning at us. Yes, I know this is directly opposite of what the 309 * documentation for the driver core and kobjects say, and the author 310 * of this code has already been publically ridiculed for doing 311 * something as foolish as this. However, at this point in time, it is 312 * the only way to handle the issue of statically allocated cpu 313 * devices. The different architectures will have their cpu device 314 * code reworked to properly handle this in the near future, so this 315 * function will then be changed to correctly free up the memory held 316 * by the cpu device. 317 * 318 * Never copy this way of doing things, or you too will be made fun of 319 * on the linux-kernel list, you have been warned. 320 */ 321 } 322 323 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 324 static ssize_t print_cpu_modalias(struct device *dev, 325 struct device_attribute *attr, 326 char *buf) 327 { 328 ssize_t n; 329 u32 i; 330 331 n = sprintf(buf, "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:", 332 CPU_FEATURE_TYPEVAL); 333 334 for (i = 0; i < MAX_CPU_FEATURES; i++) 335 if (cpu_have_feature(i)) { 336 if (PAGE_SIZE < n + sizeof(",XXXX\n")) { 337 WARN(1, "CPU features overflow page\n"); 338 break; 339 } 340 n += sprintf(&buf[n], ",%04X", i); 341 } 342 buf[n++] = '\n'; 343 return n; 344 } 345 346 static int cpu_uevent(struct device *dev, struct kobj_uevent_env *env) 347 { 348 char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL); 349 if (buf) { 350 print_cpu_modalias(NULL, NULL, buf); 351 add_uevent_var(env, "MODALIAS=%s", buf); 352 kfree(buf); 353 } 354 return 0; 355 } 356 #endif 357 358 /* 359 * register_cpu - Setup a sysfs device for a CPU. 360 * @cpu - cpu->hotpluggable field set to 1 will generate a control file in 361 * sysfs for this CPU. 362 * @num - CPU number to use when creating the device. 363 * 364 * Initialize and register the CPU device. 365 */ 366 int register_cpu(struct cpu *cpu, int num) 367 { 368 int error; 369 370 cpu->node_id = cpu_to_node(num); 371 memset(&cpu->dev, 0x00, sizeof(struct device)); 372 cpu->dev.id = num; 373 cpu->dev.bus = &cpu_subsys; 374 cpu->dev.release = cpu_device_release; 375 cpu->dev.offline_disabled = !cpu->hotpluggable; 376 cpu->dev.offline = !cpu_online(num); 377 cpu->dev.of_node = of_get_cpu_node(num, NULL); 378 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 379 cpu->dev.bus->uevent = cpu_uevent; 380 #endif 381 cpu->dev.groups = common_cpu_attr_groups; 382 if (cpu->hotpluggable) 383 cpu->dev.groups = hotplugable_cpu_attr_groups; 384 error = device_register(&cpu->dev); 385 if (error) { 386 put_device(&cpu->dev); 387 return error; 388 } 389 390 per_cpu(cpu_sys_devices, num) = &cpu->dev; 391 register_cpu_under_node(num, cpu_to_node(num)); 392 dev_pm_qos_expose_latency_limit(&cpu->dev, 393 PM_QOS_RESUME_LATENCY_NO_CONSTRAINT); 394 395 return 0; 396 } 397 398 struct device *get_cpu_device(unsigned cpu) 399 { 400 if (cpu < nr_cpu_ids && cpu_possible(cpu)) 401 return per_cpu(cpu_sys_devices, cpu); 402 else 403 return NULL; 404 } 405 EXPORT_SYMBOL_GPL(get_cpu_device); 406 407 static void device_create_release(struct device *dev) 408 { 409 kfree(dev); 410 } 411 412 __printf(4, 0) 413 static struct device * 414 __cpu_device_create(struct device *parent, void *drvdata, 415 const struct attribute_group **groups, 416 const char *fmt, va_list args) 417 { 418 struct device *dev = NULL; 419 int retval = -ENODEV; 420 421 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 422 if (!dev) { 423 retval = -ENOMEM; 424 goto error; 425 } 426 427 device_initialize(dev); 428 dev->parent = parent; 429 dev->groups = groups; 430 dev->release = device_create_release; 431 device_set_pm_not_required(dev); 432 dev_set_drvdata(dev, drvdata); 433 434 retval = kobject_set_name_vargs(&dev->kobj, fmt, args); 435 if (retval) 436 goto error; 437 438 retval = device_add(dev); 439 if (retval) 440 goto error; 441 442 return dev; 443 444 error: 445 put_device(dev); 446 return ERR_PTR(retval); 447 } 448 449 struct device *cpu_device_create(struct device *parent, void *drvdata, 450 const struct attribute_group **groups, 451 const char *fmt, ...) 452 { 453 va_list vargs; 454 struct device *dev; 455 456 va_start(vargs, fmt); 457 dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs); 458 va_end(vargs); 459 return dev; 460 } 461 EXPORT_SYMBOL_GPL(cpu_device_create); 462 463 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 464 static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL); 465 #endif 466 467 static struct attribute *cpu_root_attrs[] = { 468 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE 469 &dev_attr_probe.attr, 470 &dev_attr_release.attr, 471 #endif 472 &cpu_attrs[0].attr.attr, 473 &cpu_attrs[1].attr.attr, 474 &cpu_attrs[2].attr.attr, 475 &dev_attr_kernel_max.attr, 476 &dev_attr_offline.attr, 477 &dev_attr_isolated.attr, 478 #ifdef CONFIG_NO_HZ_FULL 479 &dev_attr_nohz_full.attr, 480 #endif 481 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 482 &dev_attr_modalias.attr, 483 #endif 484 NULL 485 }; 486 487 static struct attribute_group cpu_root_attr_group = { 488 .attrs = cpu_root_attrs, 489 }; 490 491 static const struct attribute_group *cpu_root_attr_groups[] = { 492 &cpu_root_attr_group, 493 NULL, 494 }; 495 496 bool cpu_is_hotpluggable(unsigned cpu) 497 { 498 struct device *dev = get_cpu_device(cpu); 499 return dev && container_of(dev, struct cpu, dev)->hotpluggable; 500 } 501 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable); 502 503 #ifdef CONFIG_GENERIC_CPU_DEVICES 504 static DEFINE_PER_CPU(struct cpu, cpu_devices); 505 #endif 506 507 static void __init cpu_dev_register_generic(void) 508 { 509 #ifdef CONFIG_GENERIC_CPU_DEVICES 510 int i; 511 512 for_each_possible_cpu(i) { 513 if (register_cpu(&per_cpu(cpu_devices, i), i)) 514 panic("Failed to register CPU device"); 515 } 516 #endif 517 } 518 519 #ifdef CONFIG_GENERIC_CPU_VULNERABILITIES 520 521 ssize_t __weak cpu_show_meltdown(struct device *dev, 522 struct device_attribute *attr, char *buf) 523 { 524 return sprintf(buf, "Not affected\n"); 525 } 526 527 ssize_t __weak cpu_show_spectre_v1(struct device *dev, 528 struct device_attribute *attr, char *buf) 529 { 530 return sprintf(buf, "Not affected\n"); 531 } 532 533 ssize_t __weak cpu_show_spectre_v2(struct device *dev, 534 struct device_attribute *attr, char *buf) 535 { 536 return sprintf(buf, "Not affected\n"); 537 } 538 539 ssize_t __weak cpu_show_spec_store_bypass(struct device *dev, 540 struct device_attribute *attr, char *buf) 541 { 542 return sprintf(buf, "Not affected\n"); 543 } 544 545 ssize_t __weak cpu_show_l1tf(struct device *dev, 546 struct device_attribute *attr, char *buf) 547 { 548 return sprintf(buf, "Not affected\n"); 549 } 550 551 ssize_t __weak cpu_show_mds(struct device *dev, 552 struct device_attribute *attr, char *buf) 553 { 554 return sprintf(buf, "Not affected\n"); 555 } 556 557 ssize_t __weak cpu_show_tsx_async_abort(struct device *dev, 558 struct device_attribute *attr, 559 char *buf) 560 { 561 return sprintf(buf, "Not affected\n"); 562 } 563 564 ssize_t __weak cpu_show_itlb_multihit(struct device *dev, 565 struct device_attribute *attr, char *buf) 566 { 567 return sprintf(buf, "Not affected\n"); 568 } 569 570 static DEVICE_ATTR(meltdown, 0444, cpu_show_meltdown, NULL); 571 static DEVICE_ATTR(spectre_v1, 0444, cpu_show_spectre_v1, NULL); 572 static DEVICE_ATTR(spectre_v2, 0444, cpu_show_spectre_v2, NULL); 573 static DEVICE_ATTR(spec_store_bypass, 0444, cpu_show_spec_store_bypass, NULL); 574 static DEVICE_ATTR(l1tf, 0444, cpu_show_l1tf, NULL); 575 static DEVICE_ATTR(mds, 0444, cpu_show_mds, NULL); 576 static DEVICE_ATTR(tsx_async_abort, 0444, cpu_show_tsx_async_abort, NULL); 577 static DEVICE_ATTR(itlb_multihit, 0444, cpu_show_itlb_multihit, NULL); 578 579 static struct attribute *cpu_root_vulnerabilities_attrs[] = { 580 &dev_attr_meltdown.attr, 581 &dev_attr_spectre_v1.attr, 582 &dev_attr_spectre_v2.attr, 583 &dev_attr_spec_store_bypass.attr, 584 &dev_attr_l1tf.attr, 585 &dev_attr_mds.attr, 586 &dev_attr_tsx_async_abort.attr, 587 &dev_attr_itlb_multihit.attr, 588 NULL 589 }; 590 591 static const struct attribute_group cpu_root_vulnerabilities_group = { 592 .name = "vulnerabilities", 593 .attrs = cpu_root_vulnerabilities_attrs, 594 }; 595 596 static void __init cpu_register_vulnerabilities(void) 597 { 598 if (sysfs_create_group(&cpu_subsys.dev_root->kobj, 599 &cpu_root_vulnerabilities_group)) 600 pr_err("Unable to register CPU vulnerabilities\n"); 601 } 602 603 #else 604 static inline void cpu_register_vulnerabilities(void) { } 605 #endif 606 607 void __init cpu_dev_init(void) 608 { 609 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups)) 610 panic("Failed to register CPU subsystem"); 611 612 cpu_dev_register_generic(); 613 cpu_register_vulnerabilities(); 614 } 615