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 18 #include "base.h" 19 20 static DEFINE_PER_CPU(struct device *, cpu_sys_devices); 21 22 static int cpu_subsys_match(struct device *dev, struct device_driver *drv) 23 { 24 /* ACPI style match is the only one that may succeed. */ 25 if (acpi_driver_match_device(dev, drv)) 26 return 1; 27 28 return 0; 29 } 30 31 #ifdef CONFIG_HOTPLUG_CPU 32 static void change_cpu_under_node(struct cpu *cpu, 33 unsigned int from_nid, unsigned int to_nid) 34 { 35 int cpuid = cpu->dev.id; 36 unregister_cpu_under_node(cpuid, from_nid); 37 register_cpu_under_node(cpuid, to_nid); 38 cpu->node_id = to_nid; 39 } 40 41 static int __ref cpu_subsys_online(struct device *dev) 42 { 43 struct cpu *cpu = container_of(dev, struct cpu, dev); 44 int cpuid = dev->id; 45 int from_nid, to_nid; 46 int ret; 47 48 cpu_hotplug_driver_lock(); 49 50 from_nid = cpu_to_node(cpuid); 51 ret = cpu_up(cpuid); 52 /* 53 * When hot adding memory to memoryless node and enabling a cpu 54 * on the node, node number of the cpu may internally change. 55 */ 56 to_nid = cpu_to_node(cpuid); 57 if (from_nid != to_nid) 58 change_cpu_under_node(cpu, from_nid, to_nid); 59 60 cpu_hotplug_driver_unlock(); 61 return ret; 62 } 63 64 static int cpu_subsys_offline(struct device *dev) 65 { 66 int ret; 67 68 cpu_hotplug_driver_lock(); 69 ret = cpu_down(dev->id); 70 cpu_hotplug_driver_unlock(); 71 return ret; 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 return arch_cpu_probe(buf, count); 92 } 93 94 static ssize_t cpu_release_store(struct device *dev, 95 struct device_attribute *attr, 96 const char *buf, 97 size_t count) 98 { 99 return arch_cpu_release(buf, count); 100 } 101 102 static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store); 103 static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store); 104 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ 105 #endif /* CONFIG_HOTPLUG_CPU */ 106 107 struct bus_type cpu_subsys = { 108 .name = "cpu", 109 .dev_name = "cpu", 110 .match = cpu_subsys_match, 111 #ifdef CONFIG_HOTPLUG_CPU 112 .online = cpu_subsys_online, 113 .offline = cpu_subsys_offline, 114 #endif 115 }; 116 EXPORT_SYMBOL_GPL(cpu_subsys); 117 118 #ifdef CONFIG_KEXEC 119 #include <linux/kexec.h> 120 121 static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr, 122 char *buf) 123 { 124 struct cpu *cpu = container_of(dev, struct cpu, dev); 125 ssize_t rc; 126 unsigned long long addr; 127 int cpunum; 128 129 cpunum = cpu->dev.id; 130 131 /* 132 * Might be reading other cpu's data based on which cpu read thread 133 * has been scheduled. But cpu data (memory) is allocated once during 134 * boot up and this data does not change there after. Hence this 135 * operation should be safe. No locking required. 136 */ 137 addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum)); 138 rc = sprintf(buf, "%Lx\n", addr); 139 return rc; 140 } 141 static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL); 142 143 static ssize_t show_crash_notes_size(struct device *dev, 144 struct device_attribute *attr, 145 char *buf) 146 { 147 ssize_t rc; 148 149 rc = sprintf(buf, "%zu\n", sizeof(note_buf_t)); 150 return rc; 151 } 152 static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL); 153 154 static struct attribute *crash_note_cpu_attrs[] = { 155 &dev_attr_crash_notes.attr, 156 &dev_attr_crash_notes_size.attr, 157 NULL 158 }; 159 160 static struct attribute_group crash_note_cpu_attr_group = { 161 .attrs = crash_note_cpu_attrs, 162 }; 163 #endif 164 165 static const struct attribute_group *common_cpu_attr_groups[] = { 166 #ifdef CONFIG_KEXEC 167 &crash_note_cpu_attr_group, 168 #endif 169 NULL 170 }; 171 172 static const struct attribute_group *hotplugable_cpu_attr_groups[] = { 173 #ifdef CONFIG_KEXEC 174 &crash_note_cpu_attr_group, 175 #endif 176 NULL 177 }; 178 179 /* 180 * Print cpu online, possible, present, and system maps 181 */ 182 183 struct cpu_attr { 184 struct device_attribute attr; 185 const struct cpumask *const * const map; 186 }; 187 188 static ssize_t show_cpus_attr(struct device *dev, 189 struct device_attribute *attr, 190 char *buf) 191 { 192 struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr); 193 int n = cpulist_scnprintf(buf, PAGE_SIZE-2, *(ca->map)); 194 195 buf[n++] = '\n'; 196 buf[n] = '\0'; 197 return n; 198 } 199 200 #define _CPU_ATTR(name, map) \ 201 { __ATTR(name, 0444, show_cpus_attr, NULL), map } 202 203 /* Keep in sync with cpu_subsys_attrs */ 204 static struct cpu_attr cpu_attrs[] = { 205 _CPU_ATTR(online, &cpu_online_mask), 206 _CPU_ATTR(possible, &cpu_possible_mask), 207 _CPU_ATTR(present, &cpu_present_mask), 208 }; 209 210 /* 211 * Print values for NR_CPUS and offlined cpus 212 */ 213 static ssize_t print_cpus_kernel_max(struct device *dev, 214 struct device_attribute *attr, char *buf) 215 { 216 int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1); 217 return n; 218 } 219 static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL); 220 221 /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */ 222 unsigned int total_cpus; 223 224 static ssize_t print_cpus_offline(struct device *dev, 225 struct device_attribute *attr, char *buf) 226 { 227 int n = 0, len = PAGE_SIZE-2; 228 cpumask_var_t offline; 229 230 /* display offline cpus < nr_cpu_ids */ 231 if (!alloc_cpumask_var(&offline, GFP_KERNEL)) 232 return -ENOMEM; 233 cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask); 234 n = cpulist_scnprintf(buf, len, offline); 235 free_cpumask_var(offline); 236 237 /* display offline cpus >= nr_cpu_ids */ 238 if (total_cpus && nr_cpu_ids < total_cpus) { 239 if (n && n < len) 240 buf[n++] = ','; 241 242 if (nr_cpu_ids == total_cpus-1) 243 n += snprintf(&buf[n], len - n, "%d", nr_cpu_ids); 244 else 245 n += snprintf(&buf[n], len - n, "%d-%d", 246 nr_cpu_ids, total_cpus-1); 247 } 248 249 n += snprintf(&buf[n], len - n, "\n"); 250 return n; 251 } 252 static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL); 253 254 static void cpu_device_release(struct device *dev) 255 { 256 /* 257 * This is an empty function to prevent the driver core from spitting a 258 * warning at us. Yes, I know this is directly opposite of what the 259 * documentation for the driver core and kobjects say, and the author 260 * of this code has already been publically ridiculed for doing 261 * something as foolish as this. However, at this point in time, it is 262 * the only way to handle the issue of statically allocated cpu 263 * devices. The different architectures will have their cpu device 264 * code reworked to properly handle this in the near future, so this 265 * function will then be changed to correctly free up the memory held 266 * by the cpu device. 267 * 268 * Never copy this way of doing things, or you too will be made fun of 269 * on the linux-kernel list, you have been warned. 270 */ 271 } 272 273 /* 274 * register_cpu - Setup a sysfs device for a CPU. 275 * @cpu - cpu->hotpluggable field set to 1 will generate a control file in 276 * sysfs for this CPU. 277 * @num - CPU number to use when creating the device. 278 * 279 * Initialize and register the CPU device. 280 */ 281 int __cpuinit register_cpu(struct cpu *cpu, int num) 282 { 283 int error; 284 285 cpu->node_id = cpu_to_node(num); 286 memset(&cpu->dev, 0x00, sizeof(struct device)); 287 cpu->dev.id = num; 288 cpu->dev.bus = &cpu_subsys; 289 cpu->dev.release = cpu_device_release; 290 cpu->dev.offline_disabled = !cpu->hotpluggable; 291 cpu->dev.offline = !cpu_online(num); 292 #ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE 293 cpu->dev.bus->uevent = arch_cpu_uevent; 294 #endif 295 cpu->dev.groups = common_cpu_attr_groups; 296 if (cpu->hotpluggable) 297 cpu->dev.groups = hotplugable_cpu_attr_groups; 298 error = device_register(&cpu->dev); 299 if (!error) 300 per_cpu(cpu_sys_devices, num) = &cpu->dev; 301 if (!error) 302 register_cpu_under_node(num, cpu_to_node(num)); 303 304 return error; 305 } 306 307 struct device *get_cpu_device(unsigned cpu) 308 { 309 if (cpu < nr_cpu_ids && cpu_possible(cpu)) 310 return per_cpu(cpu_sys_devices, cpu); 311 else 312 return NULL; 313 } 314 EXPORT_SYMBOL_GPL(get_cpu_device); 315 316 #ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE 317 static DEVICE_ATTR(modalias, 0444, arch_print_cpu_modalias, NULL); 318 #endif 319 320 static struct attribute *cpu_root_attrs[] = { 321 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE 322 &dev_attr_probe.attr, 323 &dev_attr_release.attr, 324 #endif 325 &cpu_attrs[0].attr.attr, 326 &cpu_attrs[1].attr.attr, 327 &cpu_attrs[2].attr.attr, 328 &dev_attr_kernel_max.attr, 329 &dev_attr_offline.attr, 330 #ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE 331 &dev_attr_modalias.attr, 332 #endif 333 NULL 334 }; 335 336 static struct attribute_group cpu_root_attr_group = { 337 .attrs = cpu_root_attrs, 338 }; 339 340 static const struct attribute_group *cpu_root_attr_groups[] = { 341 &cpu_root_attr_group, 342 NULL, 343 }; 344 345 bool cpu_is_hotpluggable(unsigned cpu) 346 { 347 struct device *dev = get_cpu_device(cpu); 348 return dev && container_of(dev, struct cpu, dev)->hotpluggable; 349 } 350 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable); 351 352 #ifdef CONFIG_GENERIC_CPU_DEVICES 353 static DEFINE_PER_CPU(struct cpu, cpu_devices); 354 #endif 355 356 static void __init cpu_dev_register_generic(void) 357 { 358 #ifdef CONFIG_GENERIC_CPU_DEVICES 359 int i; 360 361 for_each_possible_cpu(i) { 362 if (register_cpu(&per_cpu(cpu_devices, i), i)) 363 panic("Failed to register CPU device"); 364 } 365 #endif 366 } 367 368 void __init cpu_dev_init(void) 369 { 370 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups)) 371 panic("Failed to register CPU subsystem"); 372 373 cpu_dev_register_generic(); 374 } 375