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
  20 #include "base.h"
  21
  22 struct bus_type cpu_subsys = {
  23         .name = "cpu",
  24         .dev_name = "cpu",
  25 };
  26 EXPORT_SYMBOL_GPL(cpu_subsys);
  27
  28 static DEFINE_PER_CPU(struct device *, cpu_sys_devices);
  29
  30 #ifdef CONFIG_HOTPLUG_CPU
  31 static void change_cpu_under_node(struct cpu *cpu,
  32                         unsigned int from_nid, unsigned int to_nid)
  33 {
  34         int cpuid = cpu->dev.id;
  35         unregister_cpu_under_node(cpuid, from_nid);
  36         register_cpu_under_node(cpuid, to_nid);
  37         cpu->node_id = to_nid;
  38 }
  39
  40 static ssize_t show_online(struct device *dev,
  41                            struct device_attribute *attr,
  42                            char *buf)
  43 {
  44         struct cpu *cpu = container_of(dev, struct cpu, dev);
  45
  46         return sprintf(buf, "%u\n", !!cpu_online(cpu->dev.id));
  47 }
  48
  49 static ssize_t __ref store_online(struct device *dev,
  50                                   struct device_attribute *attr,
  51                                   const char *buf, size_t count)
  52 {
  53         struct cpu *cpu = container_of(dev, struct cpu, dev);
  54         int cpuid = cpu->dev.id;
  55         int from_nid, to_nid;
  56         ssize_t ret;
  57
  58         cpu_hotplug_driver_lock();
  59         switch (buf[0]) {
  60         case '0':
  61                 ret = cpu_down(cpuid);
  62                 if (!ret)
  63                         kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
  64                 break;
  65         case '1':
  66                 from_nid = cpu_to_node(cpuid);
  67                 ret = cpu_up(cpuid);
  68
  69                 /*
  70                  * When hot adding memory to memoryless node and enabling a cpu
  71                  * on the node, node number of the cpu may internally change.
  72                  */
  73                 to_nid = cpu_to_node(cpuid);
  74                 if (from_nid != to_nid)
  75                         change_cpu_under_node(cpu, from_nid, to_nid);
  76
  77                 if (!ret)
  78                         kobject_uevent(&dev->kobj, KOBJ_ONLINE);
  79                 break;
  80         default:
  81                 ret = -EINVAL;
  82         }
  83         cpu_hotplug_driver_unlock();
  84
  85         if (ret >= 0)
  86                 ret = count;
  87         return ret;
  88 }
  89 static DEVICE_ATTR(online, 0644, show_online, store_online);
  90
  91 static void __cpuinit register_cpu_control(struct cpu *cpu)
  92 {
  93         device_create_file(&cpu->dev, &dev_attr_online);
  94 }
  95 void unregister_cpu(struct cpu *cpu)
  96 {
  97         int logical_cpu = cpu->dev.id;
  98
  99         unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu));
 100
 101         device_remove_file(&cpu->dev, &dev_attr_online);
 102
 103         device_unregister(&cpu->dev);
 104         per_cpu(cpu_sys_devices, logical_cpu) = NULL;
 105         return;
 106 }
 107
 108 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 109 static ssize_t cpu_probe_store(struct device *dev,
 110                                struct device_attribute *attr,
 111                                const char *buf,
 112                                size_t count)
 113 {
 114         return arch_cpu_probe(buf, count);
 115 }
 116
 117 static ssize_t cpu_release_store(struct device *dev,
 118                                  struct device_attribute *attr,
 119                                  const char *buf,
 120                                  size_t count)
 121 {
 122         return arch_cpu_release(buf, count);
 123 }
 124
 125 static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store);
 126 static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store);
 127 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 128
 129 #else /* ... !CONFIG_HOTPLUG_CPU */
 130 static inline void register_cpu_control(struct cpu *cpu)
 131 {
 132 }
 133 #endif /* CONFIG_HOTPLUG_CPU */
 134
 135 #ifdef CONFIG_KEXEC
 136 #include <linux/kexec.h>
 137
 138 static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr,
 139                                 char *buf)
 140 {
 141         struct cpu *cpu = container_of(dev, struct cpu, dev);
 142         ssize_t rc;
 143         unsigned long long addr;
 144         int cpunum;
 145
 146         cpunum = cpu->dev.id;
 147
 148         /*
 149          * Might be reading other cpu's data based on which cpu read thread
 150          * has been scheduled. But cpu data (memory) is allocated once during
 151          * boot up and this data does not change there after. Hence this
 152          * operation should be safe. No locking required.
 153          */
 154         addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum));
 155         rc = sprintf(buf, "%Lx\n", addr);
 156         return rc;
 157 }
 158 static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL);
 159
 160 static ssize_t show_crash_notes_size(struct device *dev,
 161                                      struct device_attribute *attr,
 162                                      char *buf)
 163 {
 164         ssize_t rc;
 165
 166         rc = sprintf(buf, "%zu\n", sizeof(note_buf_t));
 167         return rc;
 168 }
 169 static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL);
 170 #endif
 171
 172 /*
 173  * Print cpu online, possible, present, and system maps
 174  */
 175
 176 struct cpu_attr {
 177         struct device_attribute attr;
 178         const struct cpumask *const * const map;
 179 };
 180
 181 static ssize_t show_cpus_attr(struct device *dev,
 182                               struct device_attribute *attr,
 183                               char *buf)
 184 {
 185         struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr);
 186         int n = cpulist_scnprintf(buf, PAGE_SIZE-2, *(ca->map));
 187
 188         buf[n++] = '\n';
 189         buf[n] = '\0';
 190         return n;
 191 }
 192
 193 #define _CPU_ATTR(name, map) \
 194         { __ATTR(name, 0444, show_cpus_attr, NULL), map }
 195
 196 /* Keep in sync with cpu_subsys_attrs */
 197 static struct cpu_attr cpu_attrs[] = {
 198         _CPU_ATTR(online, &cpu_online_mask),
 199         _CPU_ATTR(possible, &cpu_possible_mask),
 200         _CPU_ATTR(present, &cpu_present_mask),
 201 };
 202
 203 /*
 204  * Print values for NR_CPUS and offlined cpus
 205  */
 206 static ssize_t print_cpus_kernel_max(struct device *dev,
 207                                      struct device_attribute *attr, char *buf)
 208 {
 209         int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1);
 210         return n;
 211 }
 212 static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL);
 213
 214 /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */
 215 unsigned int total_cpus;
 216
 217 static ssize_t print_cpus_offline(struct device *dev,
 218                                   struct device_attribute *attr, char *buf)
 219 {
 220         int n = 0, len = PAGE_SIZE-2;
 221         cpumask_var_t offline;
 222
 223         /* display offline cpus < nr_cpu_ids */
 224         if (!alloc_cpumask_var(&offline, GFP_KERNEL))
 225                 return -ENOMEM;
 226         cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask);
 227         n = cpulist_scnprintf(buf, len, offline);
 228         free_cpumask_var(offline);
 229
 230         /* display offline cpus >= nr_cpu_ids */
 231         if (total_cpus && nr_cpu_ids < total_cpus) {
 232                 if (n && n < len)
 233                         buf[n++] = ',';
 234
 235                 if (nr_cpu_ids == total_cpus-1)
 236                         n += snprintf(&buf[n], len - n, "%d", nr_cpu_ids);
 237                 else
 238                         n += snprintf(&buf[n], len - n, "%d-%d",
 239                                                       nr_cpu_ids, total_cpus-1);
 240         }
 241
 242         n += snprintf(&buf[n], len - n, "\n");
 243         return n;
 244 }
 245 static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL);
 246
 247 static void cpu_device_release(struct device *dev)
 248 {
 249         /*
 250          * This is an empty function to prevent the driver core from spitting a
 251          * warning at us.  Yes, I know this is directly opposite of what the
 252          * documentation for the driver core and kobjects say, and the author
 253          * of this code has already been publically ridiculed for doing
 254          * something as foolish as this.  However, at this point in time, it is
 255          * the only way to handle the issue of statically allocated cpu
 256          * devices.  The different architectures will have their cpu device
 257          * code reworked to properly handle this in the near future, so this
 258          * function will then be changed to correctly free up the memory held
 259          * by the cpu device.
 260          *
 261          * Never copy this way of doing things, or you too will be made fun of
 262          * on the linux-kernel list, you have been warned.
 263          */
 264 }
 265
 266 #ifdef CONFIG_HAVE_CPU_AUTOPROBE
 267 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
 268 static ssize_t print_cpu_modalias(struct device *dev,
 269                                   struct device_attribute *attr,
 270                                   char *buf)
 271 {
 272         ssize_t n;
 273         u32 i;
 274
 275         n = sprintf(buf, "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:",
 276                     CPU_FEATURE_TYPEVAL);
 277
 278         for (i = 0; i < MAX_CPU_FEATURES; i++)
 279                 if (cpu_have_feature(i)) {
 280                         if (PAGE_SIZE < n + sizeof(",XXXX\n")) {
 281                                 WARN(1, "CPU features overflow page\n");
 282                                 break;
 283                         }
 284                         n += sprintf(&buf[n], ",%04X", i);
 285                 }
 286         buf[n++] = '\n';
 287         return n;
 288 }
 289 #else
 290 #define print_cpu_modalias      arch_print_cpu_modalias
 291 #endif
 292
 293 static int cpu_uevent(struct device *dev, struct kobj_uevent_env *env)
 294 {
 295         char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
 296         if (buf) {
 297                 print_cpu_modalias(NULL, NULL, buf);
 298                 add_uevent_var(env, "MODALIAS=%s", buf);
 299                 kfree(buf);
 300         }
 301         return 0;
 302 }
 303 #endif
 304
 305 /*
 306  * register_cpu - Setup a sysfs device for a CPU.
 307  * @cpu - cpu->hotpluggable field set to 1 will generate a control file in
 308  *        sysfs for this CPU.
 309  * @num - CPU number to use when creating the device.
 310  *
 311  * Initialize and register the CPU device.
 312  */
 313 int __cpuinit register_cpu(struct cpu *cpu, int num)
 314 {
 315         int error;
 316
 317         cpu->node_id = cpu_to_node(num);
 318         memset(&cpu->dev, 0x00, sizeof(struct device));
 319         cpu->dev.id = num;
 320         cpu->dev.bus = &cpu_subsys;
 321         cpu->dev.release = cpu_device_release;
 322 #ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE
 323         cpu->dev.bus->uevent = cpu_uevent;
 324 #endif
 325         error = device_register(&cpu->dev);
 326         if (!error && cpu->hotpluggable)
 327                 register_cpu_control(cpu);
 328         if (!error)
 329                 per_cpu(cpu_sys_devices, num) = &cpu->dev;
 330         if (!error)
 331                 register_cpu_under_node(num, cpu_to_node(num));
 332
 333 #ifdef CONFIG_KEXEC
 334         if (!error)
 335                 error = device_create_file(&cpu->dev, &dev_attr_crash_notes);
 336         if (!error)
 337                 error = device_create_file(&cpu->dev,
 338                                            &dev_attr_crash_notes_size);
 339 #endif
 340         return error;
 341 }
 342
 343 struct device *get_cpu_device(unsigned cpu)
 344 {
 345         if (cpu < nr_cpu_ids && cpu_possible(cpu))
 346                 return per_cpu(cpu_sys_devices, cpu);
 347         else
 348                 return NULL;
 349 }
 350 EXPORT_SYMBOL_GPL(get_cpu_device);
 351
 352 #ifdef CONFIG_HAVE_CPU_AUTOPROBE
 353 static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL);
 354 #endif
 355
 356 static struct attribute *cpu_root_attrs[] = {
 357 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 358         &dev_attr_probe.attr,
 359         &dev_attr_release.attr,
 360 #endif
 361         &cpu_attrs[0].attr.attr,
 362         &cpu_attrs[1].attr.attr,
 363         &cpu_attrs[2].attr.attr,
 364         &dev_attr_kernel_max.attr,
 365         &dev_attr_offline.attr,
 366 #ifdef CONFIG_HAVE_CPU_AUTOPROBE
 367         &dev_attr_modalias.attr,
 368 #endif
 369         NULL
 370 };
 371
 372 static struct attribute_group cpu_root_attr_group = {
 373         .attrs = cpu_root_attrs,
 374 };
 375
 376 static const struct attribute_group *cpu_root_attr_groups[] = {
 377         &cpu_root_attr_group,
 378         NULL,
 379 };
 380
 381 bool cpu_is_hotpluggable(unsigned cpu)
 382 {
 383         struct device *dev = get_cpu_device(cpu);
 384         return dev && container_of(dev, struct cpu, dev)->hotpluggable;
 385 }
 386 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable);
 387
 388 #ifdef CONFIG_GENERIC_CPU_DEVICES
 389 static DEFINE_PER_CPU(struct cpu, cpu_devices);
 390 #endif
 391
 392 static void __init cpu_dev_register_generic(void)
 393 {
 394 #ifdef CONFIG_GENERIC_CPU_DEVICES
 395         int i;
 396
 397         for_each_possible_cpu(i) {
 398                 if (register_cpu(&per_cpu(cpu_devices, i), i))
 399                         panic("Failed to register CPU device");
 400         }
 401 #endif
 402 }
 403
 404 void __init cpu_dev_init(void)
 405 {
 406         if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups))
 407                 panic("Failed to register CPU subsystem");
 408
 409         cpu_dev_register_generic();
 410 }