drivers/base/cpu.c

   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
  22 #include "base.h"
  23
  24 static DEFINE_PER_CPU(struct device *, cpu_sys_devices);
  25
  26 static int cpu_subsys_match(struct device *dev, struct device_driver *drv)
  27 {
  28         /* ACPI style match is the only one that may succeed. */
  29         if (acpi_driver_match_device(dev, drv))
  30                 return 1;
  31
  32         return 0;
  33 }
  34
  35 #ifdef CONFIG_HOTPLUG_CPU
  36 static void change_cpu_under_node(struct cpu *cpu,
  37                         unsigned int from_nid, unsigned int to_nid)
  38 {
  39         int cpuid = cpu->dev.id;
  40         unregister_cpu_under_node(cpuid, from_nid);
  41         register_cpu_under_node(cpuid, to_nid);
  42         cpu->node_id = to_nid;
  43 }
  44
  45 static int cpu_subsys_online(struct device *dev)
  46 {
  47         struct cpu *cpu = container_of(dev, struct cpu, dev);
  48         int cpuid = dev->id;
  49         int from_nid, to_nid;
  50         int ret;
  51
  52         from_nid = cpu_to_node(cpuid);
  53         if (from_nid == NUMA_NO_NODE)
  54                 return -ENODEV;
  55
  56         ret = cpu_up(cpuid);
  57         /*
  58          * When hot adding memory to memoryless node and enabling a cpu
  59          * on the node, node number of the cpu may internally change.
  60          */
  61         to_nid = cpu_to_node(cpuid);
  62         if (from_nid != to_nid)
  63                 change_cpu_under_node(cpu, from_nid, to_nid);
  64
  65         return ret;
  66 }
  67
  68 static int cpu_subsys_offline(struct device *dev)
  69 {
  70         return cpu_down(dev->id);
  71 }
  72
  73 void unregister_cpu(struct cpu *cpu)
  74 {
  75         int logical_cpu = cpu->dev.id;
  76
  77         unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu));
  78
  79         device_unregister(&cpu->dev);
  80         per_cpu(cpu_sys_devices, logical_cpu) = NULL;
  81         return;
  82 }
  83
  84 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
  85 static ssize_t cpu_probe_store(struct device *dev,
  86                                struct device_attribute *attr,
  87                                const char *buf,
  88                                size_t count)
  89 {
  90         ssize_t cnt;
  91         int ret;
  92
  93         ret = lock_device_hotplug_sysfs();
  94         if (ret)
  95                 return ret;
  96
  97         cnt = arch_cpu_probe(buf, count);
  98
  99         unlock_device_hotplug();
 100         return cnt;
 101 }
 102
 103 static ssize_t cpu_release_store(struct device *dev,
 104                                  struct device_attribute *attr,
 105                                  const char *buf,
 106                                  size_t count)
 107 {
 108         ssize_t cnt;
 109         int ret;
 110
 111         ret = lock_device_hotplug_sysfs();
 112         if (ret)
 113                 return ret;
 114
 115         cnt = arch_cpu_release(buf, count);
 116
 117         unlock_device_hotplug();
 118         return cnt;
 119 }
 120
 121 static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store);
 122 static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store);
 123 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 124 #endif /* CONFIG_HOTPLUG_CPU */
 125
 126 struct bus_type cpu_subsys = {
 127         .name = "cpu",
 128         .dev_name = "cpu",
 129         .match = cpu_subsys_match,
 130 #ifdef CONFIG_HOTPLUG_CPU
 131         .online = cpu_subsys_online,
 132         .offline = cpu_subsys_offline,
 133 #endif
 134 };
 135 EXPORT_SYMBOL_GPL(cpu_subsys);
 136
 137 #ifdef CONFIG_KEXEC
 138 #include <linux/kexec.h>
 139
 140 static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr,
 141                                 char *buf)
 142 {
 143         struct cpu *cpu = container_of(dev, struct cpu, dev);
 144         ssize_t rc;
 145         unsigned long long addr;
 146         int cpunum;
 147
 148         cpunum = cpu->dev.id;
 149
 150         /*
 151          * Might be reading other cpu's data based on which cpu read thread
 152          * has been scheduled. But cpu data (memory) is allocated once during
 153          * boot up and this data does not change there after. Hence this
 154          * operation should be safe. No locking required.
 155          */
 156         addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum));
 157         rc = sprintf(buf, "%Lx\n", addr);
 158         return rc;
 159 }
 160 static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL);
 161
 162 static ssize_t show_crash_notes_size(struct device *dev,
 163                                      struct device_attribute *attr,
 164                                      char *buf)
 165 {
 166         ssize_t rc;
 167
 168         rc = sprintf(buf, "%zu\n", sizeof(note_buf_t));
 169         return rc;
 170 }
 171 static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL);
 172
 173 static struct attribute *crash_note_cpu_attrs[] = {
 174         &dev_attr_crash_notes.attr,
 175         &dev_attr_crash_notes_size.attr,
 176         NULL
 177 };
 178
 179 static struct attribute_group crash_note_cpu_attr_group = {
 180         .attrs = crash_note_cpu_attrs,
 181 };
 182 #endif
 183
 184 static const struct attribute_group *common_cpu_attr_groups[] = {
 185 #ifdef CONFIG_KEXEC
 186         &crash_note_cpu_attr_group,
 187 #endif
 188         NULL
 189 };
 190
 191 static const struct attribute_group *hotplugable_cpu_attr_groups[] = {
 192 #ifdef CONFIG_KEXEC
 193         &crash_note_cpu_attr_group,
 194 #endif
 195         NULL
 196 };
 197
 198 /*
 199  * Print cpu online, possible, present, and system maps
 200  */
 201
 202 struct cpu_attr {
 203         struct device_attribute attr;
 204         const struct cpumask *const map;
 205 };
 206
 207 static ssize_t show_cpus_attr(struct device *dev,
 208                               struct device_attribute *attr,
 209                               char *buf)
 210 {
 211         struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr);
 212
 213         return cpumap_print_to_pagebuf(true, buf, ca->map);
 214 }
 215
 216 #define _CPU_ATTR(name, map) \
 217         { __ATTR(name, 0444, show_cpus_attr, NULL), map }
 218
 219 /* Keep in sync with cpu_subsys_attrs */
 220 static struct cpu_attr cpu_attrs[] = {
 221         _CPU_ATTR(online, &__cpu_online_mask),
 222         _CPU_ATTR(possible, &__cpu_possible_mask),
 223         _CPU_ATTR(present, &__cpu_present_mask),
 224 };
 225
 226 /*
 227  * Print values for NR_CPUS and offlined cpus
 228  */
 229 static ssize_t print_cpus_kernel_max(struct device *dev,
 230                                      struct device_attribute *attr, char *buf)
 231 {
 232         int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1);
 233         return n;
 234 }
 235 static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL);
 236
 237 /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */
 238 unsigned int total_cpus;
 239
 240 static ssize_t print_cpus_offline(struct device *dev,
 241                                   struct device_attribute *attr, char *buf)
 242 {
 243         int n = 0, len = PAGE_SIZE-2;
 244         cpumask_var_t offline;
 245
 246         /* display offline cpus < nr_cpu_ids */
 247         if (!alloc_cpumask_var(&offline, GFP_KERNEL))
 248                 return -ENOMEM;
 249         cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask);
 250         n = scnprintf(buf, len, "%*pbl", cpumask_pr_args(offline));
 251         free_cpumask_var(offline);
 252
 253         /* display offline cpus >= nr_cpu_ids */
 254         if (total_cpus && nr_cpu_ids < total_cpus) {
 255                 if (n && n < len)
 256                         buf[n++] = ',';
 257
 258                 if (nr_cpu_ids == total_cpus-1)
 259                         n += snprintf(&buf[n], len - n, "%u", nr_cpu_ids);
 260                 else
 261                         n += snprintf(&buf[n], len - n, "%u-%d",
 262                                                       nr_cpu_ids, total_cpus-1);
 263         }
 264
 265         n += snprintf(&buf[n], len - n, "\n");
 266         return n;
 267 }
 268 static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL);
 269
 270 static ssize_t print_cpus_isolated(struct device *dev,
 271                                   struct device_attribute *attr, char *buf)
 272 {
 273         int n = 0, len = PAGE_SIZE-2;
 274
 275         n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(cpu_isolated_map));
 276
 277         return n;
 278 }
 279 static DEVICE_ATTR(isolated, 0444, print_cpus_isolated, NULL);
 280
 281 #ifdef CONFIG_NO_HZ_FULL
 282 static ssize_t print_cpus_nohz_full(struct device *dev,
 283                                   struct device_attribute *attr, char *buf)
 284 {
 285         int n = 0, len = PAGE_SIZE-2;
 286
 287         n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(tick_nohz_full_mask));
 288
 289         return n;
 290 }
 291 static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL);
 292 #endif
 293
 294 static void cpu_device_release(struct device *dev)
 295 {
 296         /*
 297          * This is an empty function to prevent the driver core from spitting a
 298          * warning at us.  Yes, I know this is directly opposite of what the
 299          * documentation for the driver core and kobjects say, and the author
 300          * of this code has already been publically ridiculed for doing
 301          * something as foolish as this.  However, at this point in time, it is
 302          * the only way to handle the issue of statically allocated cpu
 303          * devices.  The different architectures will have their cpu device
 304          * code reworked to properly handle this in the near future, so this
 305          * function will then be changed to correctly free up the memory held
 306          * by the cpu device.
 307          *
 308          * Never copy this way of doing things, or you too will be made fun of
 309          * on the linux-kernel list, you have been warned.
 310          */
 311 }
 312
 313 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
 314 static ssize_t print_cpu_modalias(struct device *dev,
 315                                   struct device_attribute *attr,
 316                                   char *buf)
 317 {
 318         ssize_t n;
 319         u32 i;
 320
 321         n = sprintf(buf, "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:",
 322                     CPU_FEATURE_TYPEVAL);
 323
 324         for (i = 0; i < MAX_CPU_FEATURES; i++)
 325                 if (cpu_have_feature(i)) {
 326                         if (PAGE_SIZE < n + sizeof(",XXXX\n")) {
 327                                 WARN(1, "CPU features overflow page\n");
 328                                 break;
 329                         }
 330                         n += sprintf(&buf[n], ",%04X", i);
 331                 }
 332         buf[n++] = '\n';
 333         return n;
 334 }
 335
 336 static int cpu_uevent(struct device *dev, struct kobj_uevent_env *env)
 337 {
 338         char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
 339         if (buf) {
 340                 print_cpu_modalias(NULL, NULL, buf);
 341                 add_uevent_var(env, "MODALIAS=%s", buf);
 342                 kfree(buf);
 343         }
 344         return 0;
 345 }
 346 #endif
 347
 348 /*
 349  * register_cpu - Setup a sysfs device for a CPU.
 350  * @cpu - cpu->hotpluggable field set to 1 will generate a control file in
 351  *        sysfs for this CPU.
 352  * @num - CPU number to use when creating the device.
 353  *
 354  * Initialize and register the CPU device.
 355  */
 356 int register_cpu(struct cpu *cpu, int num)
 357 {
 358         int error;
 359
 360         cpu->node_id = cpu_to_node(num);
 361         memset(&cpu->dev, 0x00, sizeof(struct device));
 362         cpu->dev.id = num;
 363         cpu->dev.bus = &cpu_subsys;
 364         cpu->dev.release = cpu_device_release;
 365         cpu->dev.offline_disabled = !cpu->hotpluggable;
 366         cpu->dev.offline = !cpu_online(num);
 367         cpu->dev.of_node = of_get_cpu_node(num, NULL);
 368 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
 369         cpu->dev.bus->uevent = cpu_uevent;
 370 #endif
 371         cpu->dev.groups = common_cpu_attr_groups;
 372         if (cpu->hotpluggable)
 373                 cpu->dev.groups = hotplugable_cpu_attr_groups;
 374         error = device_register(&cpu->dev);
 375         if (error)
 376                 return error;
 377
 378         per_cpu(cpu_sys_devices, num) = &cpu->dev;
 379         register_cpu_under_node(num, cpu_to_node(num));
 380         dev_pm_qos_expose_latency_limit(&cpu->dev, 0);
 381
 382         return 0;
 383 }
 384
 385 struct device *get_cpu_device(unsigned cpu)
 386 {
 387         if (cpu < nr_cpu_ids && cpu_possible(cpu))
 388                 return per_cpu(cpu_sys_devices, cpu);
 389         else
 390                 return NULL;
 391 }
 392 EXPORT_SYMBOL_GPL(get_cpu_device);
 393
 394 static void device_create_release(struct device *dev)
 395 {
 396         kfree(dev);
 397 }
 398
 399 static struct device *
 400 __cpu_device_create(struct device *parent, void *drvdata,
 401                     const struct attribute_group **groups,
 402                     const char *fmt, va_list args)
 403 {
 404         struct device *dev = NULL;
 405         int retval = -ENODEV;
 406
 407         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 408         if (!dev) {
 409                 retval = -ENOMEM;
 410                 goto error;
 411         }
 412
 413         device_initialize(dev);
 414         dev->parent = parent;
 415         dev->groups = groups;
 416         dev->release = device_create_release;
 417         dev_set_drvdata(dev, drvdata);
 418
 419         retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
 420         if (retval)
 421                 goto error;
 422
 423         retval = device_add(dev);
 424         if (retval)
 425                 goto error;
 426
 427         return dev;
 428
 429 error:
 430         put_device(dev);
 431         return ERR_PTR(retval);
 432 }
 433
 434 struct device *cpu_device_create(struct device *parent, void *drvdata,
 435                                  const struct attribute_group **groups,
 436                                  const char *fmt, ...)
 437 {
 438         va_list vargs;
 439         struct device *dev;
 440
 441         va_start(vargs, fmt);
 442         dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs);
 443         va_end(vargs);
 444         return dev;
 445 }
 446 EXPORT_SYMBOL_GPL(cpu_device_create);
 447
 448 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
 449 static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL);
 450 #endif
 451
 452 static struct attribute *cpu_root_attrs[] = {
 453 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 454         &dev_attr_probe.attr,
 455         &dev_attr_release.attr,
 456 #endif
 457         &cpu_attrs[0].attr.attr,
 458         &cpu_attrs[1].attr.attr,
 459         &cpu_attrs[2].attr.attr,
 460         &dev_attr_kernel_max.attr,
 461         &dev_attr_offline.attr,
 462         &dev_attr_isolated.attr,
 463 #ifdef CONFIG_NO_HZ_FULL
 464         &dev_attr_nohz_full.attr,
 465 #endif
 466 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
 467         &dev_attr_modalias.attr,
 468 #endif
 469         NULL
 470 };
 471
 472 static struct attribute_group cpu_root_attr_group = {
 473         .attrs = cpu_root_attrs,
 474 };
 475
 476 static const struct attribute_group *cpu_root_attr_groups[] = {
 477         &cpu_root_attr_group,
 478         NULL,
 479 };
 480
 481 bool cpu_is_hotpluggable(unsigned cpu)
 482 {
 483         struct device *dev = get_cpu_device(cpu);
 484         return dev && container_of(dev, struct cpu, dev)->hotpluggable;
 485 }
 486 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable);
 487
 488 #ifdef CONFIG_GENERIC_CPU_DEVICES
 489 static DEFINE_PER_CPU(struct cpu, cpu_devices);
 490 #endif
 491
 492 static void __init cpu_dev_register_generic(void)
 493 {
 494 #ifdef CONFIG_GENERIC_CPU_DEVICES
 495         int i;
 496
 497         for_each_possible_cpu(i) {
 498                 if (register_cpu(&per_cpu(cpu_devices, i), i))
 499                         panic("Failed to register CPU device");
 500         }
 501 #endif
 502 }
 503
 504 #ifdef CONFIG_GENERIC_CPU_VULNERABILITIES
 505
 506 ssize_t __weak cpu_show_meltdown(struct device *dev,
 507                                  struct device_attribute *attr, char *buf)
 508 {
 509         return sprintf(buf, "Not affected\n");
 510 }
 511
 512 ssize_t __weak cpu_show_spectre_v1(struct device *dev,
 513                                    struct device_attribute *attr, char *buf)
 514 {
 515         return sprintf(buf, "Not affected\n");
 516 }
 517
 518 ssize_t __weak cpu_show_spectre_v2(struct device *dev,
 519                                    struct device_attribute *attr, char *buf)
 520 {
 521         return sprintf(buf, "Not affected\n");
 522 }
 523
 524 ssize_t __weak cpu_show_spec_store_bypass(struct device *dev,
 525                                           struct device_attribute *attr, char *buf)
 526 {
 527         return sprintf(buf, "Not affected\n");
 528 }
 529
 530 ssize_t __weak cpu_show_l1tf(struct device *dev,
 531                              struct device_attribute *attr, char *buf)
 532 {
 533         return sprintf(buf, "Not affected\n");
 534 }
 535
 536 static DEVICE_ATTR(meltdown, 0444, cpu_show_meltdown, NULL);
 537 static DEVICE_ATTR(spectre_v1, 0444, cpu_show_spectre_v1, NULL);
 538 static DEVICE_ATTR(spectre_v2, 0444, cpu_show_spectre_v2, NULL);
 539 static DEVICE_ATTR(spec_store_bypass, 0444, cpu_show_spec_store_bypass, NULL);
 540 static DEVICE_ATTR(l1tf, 0444, cpu_show_l1tf, NULL);
 541
 542 static struct attribute *cpu_root_vulnerabilities_attrs[] = {
 543         &dev_attr_meltdown.attr,
 544         &dev_attr_spectre_v1.attr,
 545         &dev_attr_spectre_v2.attr,
 546         &dev_attr_spec_store_bypass.attr,
 547         &dev_attr_l1tf.attr,
 548         NULL
 549 };
 550
 551 static const struct attribute_group cpu_root_vulnerabilities_group = {
 552         .name  = "vulnerabilities",
 553         .attrs = cpu_root_vulnerabilities_attrs,
 554 };
 555
 556 static void __init cpu_register_vulnerabilities(void)
 557 {
 558         if (sysfs_create_group(&cpu_subsys.dev_root->kobj,
 559                                &cpu_root_vulnerabilities_group))
 560                 pr_err("Unable to register CPU vulnerabilities\n");
 561 }
 562
 563 #else
 564 static inline void cpu_register_vulnerabilities(void) { }
 565 #endif
 566
 567 void __init cpu_dev_init(void)
 568 {
 569         if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups))
 570                 panic("Failed to register CPU subsystem");
 571
 572         cpu_dev_register_generic();
 573         cpu_register_vulnerabilities();
 574 }