1 /* 2 * CPU subsystem support 3 */ 4 5 #include <linux/kernel.h> 6 #include <linux/module.h> 7 #include <linux/init.h> 8 #include <linux/sched.h> 9 #include <linux/cpu.h> 10 #include <linux/topology.h> 11 #include <linux/device.h> 12 #include <linux/node.h> 13 #include <linux/gfp.h> 14 #include <linux/slab.h> 15 #include <linux/percpu.h> 16 #include <linux/acpi.h> 17 #include <linux/of.h> 18 #include <linux/cpufeature.h> 19 #include <linux/tick.h> 20 #include <linux/pm_qos.h> 21 #include <linux/sched/isolation.h> 22 23 #include "base.h" 24 25 static DEFINE_PER_CPU(struct device *, cpu_sys_devices); 26 27 static int cpu_subsys_match(struct device *dev, struct device_driver *drv) 28 { 29 /* ACPI style match is the only one that may succeed. */ 30 if (acpi_driver_match_device(dev, drv)) 31 return 1; 32 33 return 0; 34 } 35 36 #ifdef CONFIG_HOTPLUG_CPU 37 static void change_cpu_under_node(struct cpu *cpu, 38 unsigned int from_nid, unsigned int to_nid) 39 { 40 int cpuid = cpu->dev.id; 41 unregister_cpu_under_node(cpuid, from_nid); 42 register_cpu_under_node(cpuid, to_nid); 43 cpu->node_id = to_nid; 44 } 45 46 static int cpu_subsys_online(struct device *dev) 47 { 48 struct cpu *cpu = container_of(dev, struct cpu, dev); 49 int cpuid = dev->id; 50 int from_nid, to_nid; 51 int ret; 52 53 from_nid = cpu_to_node(cpuid); 54 if (from_nid == NUMA_NO_NODE) 55 return -ENODEV; 56 57 ret = cpu_up(cpuid); 58 /* 59 * When hot adding memory to memoryless node and enabling a cpu 60 * on the node, node number of the cpu may internally change. 61 */ 62 to_nid = cpu_to_node(cpuid); 63 if (from_nid != to_nid) 64 change_cpu_under_node(cpu, from_nid, to_nid); 65 66 return ret; 67 } 68 69 static int cpu_subsys_offline(struct device *dev) 70 { 71 return cpu_down(dev->id); 72 } 73 74 void unregister_cpu(struct cpu *cpu) 75 { 76 int logical_cpu = cpu->dev.id; 77 78 unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu)); 79 80 device_unregister(&cpu->dev); 81 per_cpu(cpu_sys_devices, logical_cpu) = NULL; 82 return; 83 } 84 85 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE 86 static ssize_t cpu_probe_store(struct device *dev, 87 struct device_attribute *attr, 88 const char *buf, 89 size_t count) 90 { 91 ssize_t cnt; 92 int ret; 93 94 ret = lock_device_hotplug_sysfs(); 95 if (ret) 96 return ret; 97 98 cnt = arch_cpu_probe(buf, count); 99 100 unlock_device_hotplug(); 101 return cnt; 102 } 103 104 static ssize_t cpu_release_store(struct device *dev, 105 struct device_attribute *attr, 106 const char *buf, 107 size_t count) 108 { 109 ssize_t cnt; 110 int ret; 111 112 ret = lock_device_hotplug_sysfs(); 113 if (ret) 114 return ret; 115 116 cnt = arch_cpu_release(buf, count); 117 118 unlock_device_hotplug(); 119 return cnt; 120 } 121 122 static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store); 123 static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store); 124 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ 125 #endif /* CONFIG_HOTPLUG_CPU */ 126 127 struct bus_type cpu_subsys = { 128 .name = "cpu", 129 .dev_name = "cpu", 130 .match = cpu_subsys_match, 131 #ifdef CONFIG_HOTPLUG_CPU 132 .online = cpu_subsys_online, 133 .offline = cpu_subsys_offline, 134 #endif 135 }; 136 EXPORT_SYMBOL_GPL(cpu_subsys); 137 138 #ifdef CONFIG_KEXEC 139 #include <linux/kexec.h> 140 141 static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr, 142 char *buf) 143 { 144 struct cpu *cpu = container_of(dev, struct cpu, dev); 145 ssize_t rc; 146 unsigned long long addr; 147 int cpunum; 148 149 cpunum = cpu->dev.id; 150 151 /* 152 * Might be reading other cpu's data based on which cpu read thread 153 * has been scheduled. But cpu data (memory) is allocated once during 154 * boot up and this data does not change there after. Hence this 155 * operation should be safe. No locking required. 156 */ 157 addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum)); 158 rc = sprintf(buf, "%Lx\n", addr); 159 return rc; 160 } 161 static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL); 162 163 static ssize_t show_crash_notes_size(struct device *dev, 164 struct device_attribute *attr, 165 char *buf) 166 { 167 ssize_t rc; 168 169 rc = sprintf(buf, "%zu\n", sizeof(note_buf_t)); 170 return rc; 171 } 172 static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL); 173 174 static struct attribute *crash_note_cpu_attrs[] = { 175 &dev_attr_crash_notes.attr, 176 &dev_attr_crash_notes_size.attr, 177 NULL 178 }; 179 180 static struct attribute_group crash_note_cpu_attr_group = { 181 .attrs = crash_note_cpu_attrs, 182 }; 183 #endif 184 185 static const struct attribute_group *common_cpu_attr_groups[] = { 186 #ifdef CONFIG_KEXEC 187 &crash_note_cpu_attr_group, 188 #endif 189 NULL 190 }; 191 192 static const struct attribute_group *hotplugable_cpu_attr_groups[] = { 193 #ifdef CONFIG_KEXEC 194 &crash_note_cpu_attr_group, 195 #endif 196 NULL 197 }; 198 199 /* 200 * Print cpu online, possible, present, and system maps 201 */ 202 203 struct cpu_attr { 204 struct device_attribute attr; 205 const struct cpumask *const map; 206 }; 207 208 static ssize_t show_cpus_attr(struct device *dev, 209 struct device_attribute *attr, 210 char *buf) 211 { 212 struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr); 213 214 return cpumap_print_to_pagebuf(true, buf, ca->map); 215 } 216 217 #define _CPU_ATTR(name, map) \ 218 { __ATTR(name, 0444, show_cpus_attr, NULL), map } 219 220 /* Keep in sync with cpu_subsys_attrs */ 221 static struct cpu_attr cpu_attrs[] = { 222 _CPU_ATTR(online, &__cpu_online_mask), 223 _CPU_ATTR(possible, &__cpu_possible_mask), 224 _CPU_ATTR(present, &__cpu_present_mask), 225 }; 226 227 /* 228 * Print values for NR_CPUS and offlined cpus 229 */ 230 static ssize_t print_cpus_kernel_max(struct device *dev, 231 struct device_attribute *attr, char *buf) 232 { 233 int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1); 234 return n; 235 } 236 static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL); 237 238 /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */ 239 unsigned int total_cpus; 240 241 static ssize_t print_cpus_offline(struct device *dev, 242 struct device_attribute *attr, char *buf) 243 { 244 int n = 0, len = PAGE_SIZE-2; 245 cpumask_var_t offline; 246 247 /* display offline cpus < nr_cpu_ids */ 248 if (!alloc_cpumask_var(&offline, GFP_KERNEL)) 249 return -ENOMEM; 250 cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask); 251 n = scnprintf(buf, len, "%*pbl", cpumask_pr_args(offline)); 252 free_cpumask_var(offline); 253 254 /* display offline cpus >= nr_cpu_ids */ 255 if (total_cpus && nr_cpu_ids < total_cpus) { 256 if (n && n < len) 257 buf[n++] = ','; 258 259 if (nr_cpu_ids == total_cpus-1) 260 n += snprintf(&buf[n], len - n, "%u", nr_cpu_ids); 261 else 262 n += snprintf(&buf[n], len - n, "%u-%d", 263 nr_cpu_ids, total_cpus-1); 264 } 265 266 n += snprintf(&buf[n], len - n, "\n"); 267 return n; 268 } 269 static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL); 270 271 static ssize_t print_cpus_isolated(struct device *dev, 272 struct device_attribute *attr, char *buf) 273 { 274 int n = 0, len = PAGE_SIZE-2; 275 cpumask_var_t isolated; 276 277 if (!alloc_cpumask_var(&isolated, GFP_KERNEL)) 278 return -ENOMEM; 279 280 cpumask_andnot(isolated, cpu_possible_mask, 281 housekeeping_cpumask(HK_FLAG_DOMAIN)); 282 n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(isolated)); 283 284 free_cpumask_var(isolated); 285 286 return n; 287 } 288 static DEVICE_ATTR(isolated, 0444, print_cpus_isolated, NULL); 289 290 #ifdef CONFIG_NO_HZ_FULL 291 static ssize_t print_cpus_nohz_full(struct device *dev, 292 struct device_attribute *attr, char *buf) 293 { 294 int n = 0, len = PAGE_SIZE-2; 295 296 n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(tick_nohz_full_mask)); 297 298 return n; 299 } 300 static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL); 301 #endif 302 303 static void cpu_device_release(struct device *dev) 304 { 305 /* 306 * This is an empty function to prevent the driver core from spitting a 307 * warning at us. Yes, I know this is directly opposite of what the 308 * documentation for the driver core and kobjects say, and the author 309 * of this code has already been publically ridiculed for doing 310 * something as foolish as this. However, at this point in time, it is 311 * the only way to handle the issue of statically allocated cpu 312 * devices. The different architectures will have their cpu device 313 * code reworked to properly handle this in the near future, so this 314 * function will then be changed to correctly free up the memory held 315 * by the cpu device. 316 * 317 * Never copy this way of doing things, or you too will be made fun of 318 * on the linux-kernel list, you have been warned. 319 */ 320 } 321 322 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 323 static ssize_t print_cpu_modalias(struct device *dev, 324 struct device_attribute *attr, 325 char *buf) 326 { 327 ssize_t n; 328 u32 i; 329 330 n = sprintf(buf, "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:", 331 CPU_FEATURE_TYPEVAL); 332 333 for (i = 0; i < MAX_CPU_FEATURES; i++) 334 if (cpu_have_feature(i)) { 335 if (PAGE_SIZE < n + sizeof(",XXXX\n")) { 336 WARN(1, "CPU features overflow page\n"); 337 break; 338 } 339 n += sprintf(&buf[n], ",%04X", i); 340 } 341 buf[n++] = '\n'; 342 return n; 343 } 344 345 static int cpu_uevent(struct device *dev, struct kobj_uevent_env *env) 346 { 347 char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL); 348 if (buf) { 349 print_cpu_modalias(NULL, NULL, buf); 350 add_uevent_var(env, "MODALIAS=%s", buf); 351 kfree(buf); 352 } 353 return 0; 354 } 355 #endif 356 357 /* 358 * register_cpu - Setup a sysfs device for a CPU. 359 * @cpu - cpu->hotpluggable field set to 1 will generate a control file in 360 * sysfs for this CPU. 361 * @num - CPU number to use when creating the device. 362 * 363 * Initialize and register the CPU device. 364 */ 365 int register_cpu(struct cpu *cpu, int num) 366 { 367 int error; 368 369 cpu->node_id = cpu_to_node(num); 370 memset(&cpu->dev, 0x00, sizeof(struct device)); 371 cpu->dev.id = num; 372 cpu->dev.bus = &cpu_subsys; 373 cpu->dev.release = cpu_device_release; 374 cpu->dev.offline_disabled = !cpu->hotpluggable; 375 cpu->dev.offline = !cpu_online(num); 376 cpu->dev.of_node = of_get_cpu_node(num, NULL); 377 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 378 cpu->dev.bus->uevent = cpu_uevent; 379 #endif 380 cpu->dev.groups = common_cpu_attr_groups; 381 if (cpu->hotpluggable) 382 cpu->dev.groups = hotplugable_cpu_attr_groups; 383 error = device_register(&cpu->dev); 384 if (error) 385 return error; 386 387 per_cpu(cpu_sys_devices, num) = &cpu->dev; 388 register_cpu_under_node(num, cpu_to_node(num)); 389 dev_pm_qos_expose_latency_limit(&cpu->dev, 390 PM_QOS_RESUME_LATENCY_NO_CONSTRAINT); 391 392 return 0; 393 } 394 395 struct device *get_cpu_device(unsigned cpu) 396 { 397 if (cpu < nr_cpu_ids && cpu_possible(cpu)) 398 return per_cpu(cpu_sys_devices, cpu); 399 else 400 return NULL; 401 } 402 EXPORT_SYMBOL_GPL(get_cpu_device); 403 404 static void device_create_release(struct device *dev) 405 { 406 kfree(dev); 407 } 408 409 static struct device * 410 __cpu_device_create(struct device *parent, void *drvdata, 411 const struct attribute_group **groups, 412 const char *fmt, va_list args) 413 { 414 struct device *dev = NULL; 415 int retval = -ENODEV; 416 417 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 418 if (!dev) { 419 retval = -ENOMEM; 420 goto error; 421 } 422 423 device_initialize(dev); 424 dev->parent = parent; 425 dev->groups = groups; 426 dev->release = device_create_release; 427 dev_set_drvdata(dev, drvdata); 428 429 retval = kobject_set_name_vargs(&dev->kobj, fmt, args); 430 if (retval) 431 goto error; 432 433 retval = device_add(dev); 434 if (retval) 435 goto error; 436 437 return dev; 438 439 error: 440 put_device(dev); 441 return ERR_PTR(retval); 442 } 443 444 struct device *cpu_device_create(struct device *parent, void *drvdata, 445 const struct attribute_group **groups, 446 const char *fmt, ...) 447 { 448 va_list vargs; 449 struct device *dev; 450 451 va_start(vargs, fmt); 452 dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs); 453 va_end(vargs); 454 return dev; 455 } 456 EXPORT_SYMBOL_GPL(cpu_device_create); 457 458 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 459 static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL); 460 #endif 461 462 static struct attribute *cpu_root_attrs[] = { 463 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE 464 &dev_attr_probe.attr, 465 &dev_attr_release.attr, 466 #endif 467 &cpu_attrs[0].attr.attr, 468 &cpu_attrs[1].attr.attr, 469 &cpu_attrs[2].attr.attr, 470 &dev_attr_kernel_max.attr, 471 &dev_attr_offline.attr, 472 &dev_attr_isolated.attr, 473 #ifdef CONFIG_NO_HZ_FULL 474 &dev_attr_nohz_full.attr, 475 #endif 476 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE 477 &dev_attr_modalias.attr, 478 #endif 479 NULL 480 }; 481 482 static struct attribute_group cpu_root_attr_group = { 483 .attrs = cpu_root_attrs, 484 }; 485 486 static const struct attribute_group *cpu_root_attr_groups[] = { 487 &cpu_root_attr_group, 488 NULL, 489 }; 490 491 bool cpu_is_hotpluggable(unsigned cpu) 492 { 493 struct device *dev = get_cpu_device(cpu); 494 return dev && container_of(dev, struct cpu, dev)->hotpluggable; 495 } 496 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable); 497 498 #ifdef CONFIG_GENERIC_CPU_DEVICES 499 static DEFINE_PER_CPU(struct cpu, cpu_devices); 500 #endif 501 502 static void __init cpu_dev_register_generic(void) 503 { 504 #ifdef CONFIG_GENERIC_CPU_DEVICES 505 int i; 506 507 for_each_possible_cpu(i) { 508 if (register_cpu(&per_cpu(cpu_devices, i), i)) 509 panic("Failed to register CPU device"); 510 } 511 #endif 512 } 513 514 void __init cpu_dev_init(void) 515 { 516 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups)) 517 panic("Failed to register CPU subsystem"); 518 519 cpu_dev_register_generic(); 520 } 521