]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
8a25a2fd | 2 | * CPU subsystem support |
1da177e4 LT |
3 | */ |
4 | ||
024f7846 | 5 | #include <linux/kernel.h> |
1da177e4 LT |
6 | #include <linux/module.h> |
7 | #include <linux/init.h> | |
f6a57033 | 8 | #include <linux/sched.h> |
1da177e4 LT |
9 | #include <linux/cpu.h> |
10 | #include <linux/topology.h> | |
11 | #include <linux/device.h> | |
76b67ed9 | 12 | #include <linux/node.h> |
5a0e3ad6 | 13 | #include <linux/gfp.h> |
fad12ac8 | 14 | #include <linux/slab.h> |
9f13a1fd | 15 | #include <linux/percpu.h> |
ac212b69 | 16 | #include <linux/acpi.h> |
f86e4718 | 17 | #include <linux/of.h> |
67bad2fd | 18 | #include <linux/cpufeature.h> |
6570a9a1 | 19 | #include <linux/tick.h> |
1da177e4 | 20 | |
a1bdc7aa | 21 | #include "base.h" |
1da177e4 | 22 | |
8a25a2fd | 23 | static DEFINE_PER_CPU(struct device *, cpu_sys_devices); |
ad74557a | 24 | |
ac212b69 RW |
25 | static int cpu_subsys_match(struct device *dev, struct device_driver *drv) |
26 | { | |
27 | /* ACPI style match is the only one that may succeed. */ | |
28 | if (acpi_driver_match_device(dev, drv)) | |
29 | return 1; | |
30 | ||
31 | return 0; | |
32 | } | |
33 | ||
1da177e4 | 34 | #ifdef CONFIG_HOTPLUG_CPU |
34640468 YI |
35 | static void change_cpu_under_node(struct cpu *cpu, |
36 | unsigned int from_nid, unsigned int to_nid) | |
37 | { | |
38 | int cpuid = cpu->dev.id; | |
39 | unregister_cpu_under_node(cpuid, from_nid); | |
40 | register_cpu_under_node(cpuid, to_nid); | |
41 | cpu->node_id = to_nid; | |
42 | } | |
43 | ||
eda5867b | 44 | static int cpu_subsys_online(struct device *dev) |
1da177e4 | 45 | { |
8a25a2fd | 46 | struct cpu *cpu = container_of(dev, struct cpu, dev); |
0902a904 RW |
47 | int cpuid = dev->id; |
48 | int from_nid, to_nid; | |
6dedcca6 | 49 | int ret; |
1da177e4 | 50 | |
0902a904 | 51 | from_nid = cpu_to_node(cpuid); |
c7991b0b | 52 | if (from_nid == NUMA_NO_NODE) |
6dedcca6 | 53 | return -ENODEV; |
c7991b0b | 54 | |
0902a904 RW |
55 | ret = cpu_up(cpuid); |
56 | /* | |
57 | * When hot adding memory to memoryless node and enabling a cpu | |
58 | * on the node, node number of the cpu may internally change. | |
59 | */ | |
60 | to_nid = cpu_to_node(cpuid); | |
61 | if (from_nid != to_nid) | |
62 | change_cpu_under_node(cpu, from_nid, to_nid); | |
1da177e4 | 63 | |
0902a904 | 64 | return ret; |
1da177e4 LT |
65 | } |
66 | ||
0902a904 | 67 | static int cpu_subsys_offline(struct device *dev) |
1da177e4 | 68 | { |
6dedcca6 | 69 | return cpu_down(dev->id); |
1da177e4 | 70 | } |
1c4e2d70 | 71 | |
76b67ed9 | 72 | void unregister_cpu(struct cpu *cpu) |
1da177e4 | 73 | { |
8a25a2fd | 74 | int logical_cpu = cpu->dev.id; |
1da177e4 | 75 | |
76b67ed9 KH |
76 | unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu)); |
77 | ||
8a25a2fd | 78 | device_unregister(&cpu->dev); |
e37d05da | 79 | per_cpu(cpu_sys_devices, logical_cpu) = NULL; |
1da177e4 LT |
80 | return; |
81 | } | |
12633e80 NF |
82 | |
83 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE | |
8a25a2fd KS |
84 | static ssize_t cpu_probe_store(struct device *dev, |
85 | struct device_attribute *attr, | |
67fc233f | 86 | const char *buf, |
12633e80 NF |
87 | size_t count) |
88 | { | |
574b851e TK |
89 | ssize_t cnt; |
90 | int ret; | |
91 | ||
92 | ret = lock_device_hotplug_sysfs(); | |
93 | if (ret) | |
94 | return ret; | |
95 | ||
96 | cnt = arch_cpu_probe(buf, count); | |
97 | ||
98 | unlock_device_hotplug(); | |
99 | return cnt; | |
12633e80 NF |
100 | } |
101 | ||
8a25a2fd KS |
102 | static ssize_t cpu_release_store(struct device *dev, |
103 | struct device_attribute *attr, | |
67fc233f | 104 | const char *buf, |
12633e80 NF |
105 | size_t count) |
106 | { | |
574b851e TK |
107 | ssize_t cnt; |
108 | int ret; | |
109 | ||
110 | ret = lock_device_hotplug_sysfs(); | |
111 | if (ret) | |
112 | return ret; | |
113 | ||
114 | cnt = arch_cpu_release(buf, count); | |
115 | ||
116 | unlock_device_hotplug(); | |
117 | return cnt; | |
12633e80 NF |
118 | } |
119 | ||
8a25a2fd KS |
120 | static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store); |
121 | static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store); | |
12633e80 | 122 | #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ |
1da177e4 LT |
123 | #endif /* CONFIG_HOTPLUG_CPU */ |
124 | ||
0902a904 RW |
125 | struct bus_type cpu_subsys = { |
126 | .name = "cpu", | |
127 | .dev_name = "cpu", | |
ac212b69 | 128 | .match = cpu_subsys_match, |
0902a904 RW |
129 | #ifdef CONFIG_HOTPLUG_CPU |
130 | .online = cpu_subsys_online, | |
131 | .offline = cpu_subsys_offline, | |
132 | #endif | |
133 | }; | |
134 | EXPORT_SYMBOL_GPL(cpu_subsys); | |
135 | ||
51be5606 VG |
136 | #ifdef CONFIG_KEXEC |
137 | #include <linux/kexec.h> | |
138 | ||
8a25a2fd | 139 | static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr, |
4a0b2b4d | 140 | char *buf) |
51be5606 | 141 | { |
8a25a2fd | 142 | struct cpu *cpu = container_of(dev, struct cpu, dev); |
51be5606 VG |
143 | ssize_t rc; |
144 | unsigned long long addr; | |
145 | int cpunum; | |
146 | ||
8a25a2fd | 147 | cpunum = cpu->dev.id; |
51be5606 VG |
148 | |
149 | /* | |
150 | * Might be reading other cpu's data based on which cpu read thread | |
151 | * has been scheduled. But cpu data (memory) is allocated once during | |
152 | * boot up and this data does not change there after. Hence this | |
153 | * operation should be safe. No locking required. | |
154 | */ | |
3b034b0d | 155 | addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum)); |
51be5606 | 156 | rc = sprintf(buf, "%Lx\n", addr); |
51be5606 VG |
157 | return rc; |
158 | } | |
8a25a2fd | 159 | static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL); |
eca4549f ZY |
160 | |
161 | static ssize_t show_crash_notes_size(struct device *dev, | |
162 | struct device_attribute *attr, | |
163 | char *buf) | |
164 | { | |
165 | ssize_t rc; | |
166 | ||
bcfb87fb | 167 | rc = sprintf(buf, "%zu\n", sizeof(note_buf_t)); |
eca4549f ZY |
168 | return rc; |
169 | } | |
170 | static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL); | |
c055da9f IM |
171 | |
172 | static struct attribute *crash_note_cpu_attrs[] = { | |
173 | &dev_attr_crash_notes.attr, | |
174 | &dev_attr_crash_notes_size.attr, | |
175 | NULL | |
176 | }; | |
177 | ||
178 | static struct attribute_group crash_note_cpu_attr_group = { | |
179 | .attrs = crash_note_cpu_attrs, | |
180 | }; | |
181 | #endif | |
182 | ||
183 | static const struct attribute_group *common_cpu_attr_groups[] = { | |
184 | #ifdef CONFIG_KEXEC | |
185 | &crash_note_cpu_attr_group, | |
51be5606 | 186 | #endif |
c055da9f IM |
187 | NULL |
188 | }; | |
51be5606 | 189 | |
1c4e2d70 IM |
190 | static const struct attribute_group *hotplugable_cpu_attr_groups[] = { |
191 | #ifdef CONFIG_KEXEC | |
192 | &crash_note_cpu_attr_group, | |
1c4e2d70 IM |
193 | #endif |
194 | NULL | |
195 | }; | |
196 | ||
9d1fe323 MT |
197 | /* |
198 | * Print cpu online, possible, present, and system maps | |
199 | */ | |
265d2e2e AK |
200 | |
201 | struct cpu_attr { | |
8a25a2fd | 202 | struct device_attribute attr; |
848e2391 | 203 | const struct cpumask *const map; |
265d2e2e AK |
204 | }; |
205 | ||
8a25a2fd KS |
206 | static ssize_t show_cpus_attr(struct device *dev, |
207 | struct device_attribute *attr, | |
265d2e2e | 208 | char *buf) |
9d1fe323 | 209 | { |
265d2e2e | 210 | struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr); |
9d1fe323 | 211 | |
848e2391 | 212 | return cpumap_print_to_pagebuf(true, buf, ca->map); |
9d1fe323 MT |
213 | } |
214 | ||
8a25a2fd KS |
215 | #define _CPU_ATTR(name, map) \ |
216 | { __ATTR(name, 0444, show_cpus_attr, NULL), map } | |
9d1fe323 | 217 | |
8a25a2fd | 218 | /* Keep in sync with cpu_subsys_attrs */ |
265d2e2e | 219 | static struct cpu_attr cpu_attrs[] = { |
848e2391 RV |
220 | _CPU_ATTR(online, &__cpu_online_mask), |
221 | _CPU_ATTR(possible, &__cpu_possible_mask), | |
222 | _CPU_ATTR(present, &__cpu_present_mask), | |
265d2e2e | 223 | }; |
9d1fe323 | 224 | |
e057d7ae MT |
225 | /* |
226 | * Print values for NR_CPUS and offlined cpus | |
227 | */ | |
8a25a2fd KS |
228 | static ssize_t print_cpus_kernel_max(struct device *dev, |
229 | struct device_attribute *attr, char *buf) | |
e057d7ae | 230 | { |
8fd2d2d5 | 231 | int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1); |
e057d7ae MT |
232 | return n; |
233 | } | |
8a25a2fd | 234 | static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL); |
e057d7ae MT |
235 | |
236 | /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */ | |
237 | unsigned int total_cpus; | |
238 | ||
8a25a2fd KS |
239 | static ssize_t print_cpus_offline(struct device *dev, |
240 | struct device_attribute *attr, char *buf) | |
e057d7ae MT |
241 | { |
242 | int n = 0, len = PAGE_SIZE-2; | |
243 | cpumask_var_t offline; | |
244 | ||
245 | /* display offline cpus < nr_cpu_ids */ | |
246 | if (!alloc_cpumask_var(&offline, GFP_KERNEL)) | |
247 | return -ENOMEM; | |
cdc6e3d3 | 248 | cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask); |
f799b1a7 | 249 | n = scnprintf(buf, len, "%*pbl", cpumask_pr_args(offline)); |
e057d7ae MT |
250 | free_cpumask_var(offline); |
251 | ||
252 | /* display offline cpus >= nr_cpu_ids */ | |
253 | if (total_cpus && nr_cpu_ids < total_cpus) { | |
254 | if (n && n < len) | |
255 | buf[n++] = ','; | |
256 | ||
257 | if (nr_cpu_ids == total_cpus-1) | |
258 | n += snprintf(&buf[n], len - n, "%d", nr_cpu_ids); | |
259 | else | |
260 | n += snprintf(&buf[n], len - n, "%d-%d", | |
261 | nr_cpu_ids, total_cpus-1); | |
262 | } | |
263 | ||
264 | n += snprintf(&buf[n], len - n, "\n"); | |
265 | return n; | |
266 | } | |
8a25a2fd | 267 | static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL); |
e057d7ae | 268 | |
59f30abe RR |
269 | static ssize_t print_cpus_isolated(struct device *dev, |
270 | struct device_attribute *attr, char *buf) | |
271 | { | |
272 | int n = 0, len = PAGE_SIZE-2; | |
273 | ||
274 | n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(cpu_isolated_map)); | |
275 | ||
276 | return n; | |
277 | } | |
278 | static DEVICE_ATTR(isolated, 0444, print_cpus_isolated, NULL); | |
279 | ||
6570a9a1 RR |
280 | #ifdef CONFIG_NO_HZ_FULL |
281 | static ssize_t print_cpus_nohz_full(struct device *dev, | |
282 | struct device_attribute *attr, char *buf) | |
283 | { | |
284 | int n = 0, len = PAGE_SIZE-2; | |
285 | ||
286 | n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(tick_nohz_full_mask)); | |
287 | ||
288 | return n; | |
289 | } | |
290 | static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL); | |
291 | #endif | |
292 | ||
2885e25c GKH |
293 | static void cpu_device_release(struct device *dev) |
294 | { | |
295 | /* | |
296 | * This is an empty function to prevent the driver core from spitting a | |
297 | * warning at us. Yes, I know this is directly opposite of what the | |
298 | * documentation for the driver core and kobjects say, and the author | |
299 | * of this code has already been publically ridiculed for doing | |
300 | * something as foolish as this. However, at this point in time, it is | |
301 | * the only way to handle the issue of statically allocated cpu | |
302 | * devices. The different architectures will have their cpu device | |
303 | * code reworked to properly handle this in the near future, so this | |
304 | * function will then be changed to correctly free up the memory held | |
305 | * by the cpu device. | |
306 | * | |
307 | * Never copy this way of doing things, or you too will be made fun of | |
30a4840a | 308 | * on the linux-kernel list, you have been warned. |
2885e25c GKH |
309 | */ |
310 | } | |
311 | ||
67bad2fd AB |
312 | #ifdef CONFIG_GENERIC_CPU_AUTOPROBE |
313 | static ssize_t print_cpu_modalias(struct device *dev, | |
314 | struct device_attribute *attr, | |
315 | char *buf) | |
316 | { | |
317 | ssize_t n; | |
318 | u32 i; | |
319 | ||
320 | n = sprintf(buf, "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:", | |
321 | CPU_FEATURE_TYPEVAL); | |
322 | ||
323 | for (i = 0; i < MAX_CPU_FEATURES; i++) | |
324 | if (cpu_have_feature(i)) { | |
325 | if (PAGE_SIZE < n + sizeof(",XXXX\n")) { | |
326 | WARN(1, "CPU features overflow page\n"); | |
327 | break; | |
328 | } | |
329 | n += sprintf(&buf[n], ",%04X", i); | |
330 | } | |
331 | buf[n++] = '\n'; | |
332 | return n; | |
333 | } | |
67bad2fd AB |
334 | |
335 | static int cpu_uevent(struct device *dev, struct kobj_uevent_env *env) | |
336 | { | |
337 | char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
338 | if (buf) { | |
339 | print_cpu_modalias(NULL, NULL, buf); | |
340 | add_uevent_var(env, "MODALIAS=%s", buf); | |
341 | kfree(buf); | |
342 | } | |
343 | return 0; | |
344 | } | |
345 | #endif | |
346 | ||
1da177e4 | 347 | /* |
405ae7d3 | 348 | * register_cpu - Setup a sysfs device for a CPU. |
72486f1f SS |
349 | * @cpu - cpu->hotpluggable field set to 1 will generate a control file in |
350 | * sysfs for this CPU. | |
1da177e4 LT |
351 | * @num - CPU number to use when creating the device. |
352 | * | |
353 | * Initialize and register the CPU device. | |
354 | */ | |
a83048eb | 355 | int register_cpu(struct cpu *cpu, int num) |
1da177e4 LT |
356 | { |
357 | int error; | |
76b67ed9 | 358 | |
8a25a2fd | 359 | cpu->node_id = cpu_to_node(num); |
29bb5d4f | 360 | memset(&cpu->dev, 0x00, sizeof(struct device)); |
8a25a2fd KS |
361 | cpu->dev.id = num; |
362 | cpu->dev.bus = &cpu_subsys; | |
2885e25c | 363 | cpu->dev.release = cpu_device_release; |
0902a904 | 364 | cpu->dev.offline_disabled = !cpu->hotpluggable; |
1001b4d4 | 365 | cpu->dev.offline = !cpu_online(num); |
f86e4718 | 366 | cpu->dev.of_node = of_get_cpu_node(num, NULL); |
2b9c1f03 | 367 | #ifdef CONFIG_GENERIC_CPU_AUTOPROBE |
67bad2fd | 368 | cpu->dev.bus->uevent = cpu_uevent; |
fad12ac8 | 369 | #endif |
c055da9f | 370 | cpu->dev.groups = common_cpu_attr_groups; |
1c4e2d70 IM |
371 | if (cpu->hotpluggable) |
372 | cpu->dev.groups = hotplugable_cpu_attr_groups; | |
8a25a2fd | 373 | error = device_register(&cpu->dev); |
59fffa34 AS |
374 | if (error) |
375 | return error; | |
51be5606 | 376 | |
59fffa34 AS |
377 | per_cpu(cpu_sys_devices, num) = &cpu->dev; |
378 | register_cpu_under_node(num, cpu_to_node(num)); | |
379 | ||
380 | return 0; | |
1da177e4 LT |
381 | } |
382 | ||
8a25a2fd | 383 | struct device *get_cpu_device(unsigned cpu) |
ad74557a | 384 | { |
e37d05da MT |
385 | if (cpu < nr_cpu_ids && cpu_possible(cpu)) |
386 | return per_cpu(cpu_sys_devices, cpu); | |
ad74557a AR |
387 | else |
388 | return NULL; | |
389 | } | |
8a25a2fd KS |
390 | EXPORT_SYMBOL_GPL(get_cpu_device); |
391 | ||
3d52943b SH |
392 | static void device_create_release(struct device *dev) |
393 | { | |
394 | kfree(dev); | |
395 | } | |
396 | ||
397 | static struct device * | |
398 | __cpu_device_create(struct device *parent, void *drvdata, | |
399 | const struct attribute_group **groups, | |
400 | const char *fmt, va_list args) | |
401 | { | |
402 | struct device *dev = NULL; | |
403 | int retval = -ENODEV; | |
404 | ||
405 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
406 | if (!dev) { | |
407 | retval = -ENOMEM; | |
408 | goto error; | |
409 | } | |
410 | ||
411 | device_initialize(dev); | |
412 | dev->parent = parent; | |
413 | dev->groups = groups; | |
414 | dev->release = device_create_release; | |
415 | dev_set_drvdata(dev, drvdata); | |
416 | ||
417 | retval = kobject_set_name_vargs(&dev->kobj, fmt, args); | |
418 | if (retval) | |
419 | goto error; | |
420 | ||
421 | retval = device_add(dev); | |
422 | if (retval) | |
423 | goto error; | |
424 | ||
425 | return dev; | |
426 | ||
427 | error: | |
428 | put_device(dev); | |
429 | return ERR_PTR(retval); | |
430 | } | |
431 | ||
432 | struct device *cpu_device_create(struct device *parent, void *drvdata, | |
433 | const struct attribute_group **groups, | |
434 | const char *fmt, ...) | |
435 | { | |
436 | va_list vargs; | |
437 | struct device *dev; | |
438 | ||
439 | va_start(vargs, fmt); | |
440 | dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs); | |
441 | va_end(vargs); | |
442 | return dev; | |
443 | } | |
444 | EXPORT_SYMBOL_GPL(cpu_device_create); | |
445 | ||
2b9c1f03 | 446 | #ifdef CONFIG_GENERIC_CPU_AUTOPROBE |
67bad2fd | 447 | static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL); |
fad12ac8 TR |
448 | #endif |
449 | ||
8a25a2fd KS |
450 | static struct attribute *cpu_root_attrs[] = { |
451 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE | |
452 | &dev_attr_probe.attr, | |
453 | &dev_attr_release.attr, | |
454 | #endif | |
455 | &cpu_attrs[0].attr.attr, | |
456 | &cpu_attrs[1].attr.attr, | |
457 | &cpu_attrs[2].attr.attr, | |
458 | &dev_attr_kernel_max.attr, | |
459 | &dev_attr_offline.attr, | |
59f30abe | 460 | &dev_attr_isolated.attr, |
6570a9a1 RR |
461 | #ifdef CONFIG_NO_HZ_FULL |
462 | &dev_attr_nohz_full.attr, | |
463 | #endif | |
2b9c1f03 | 464 | #ifdef CONFIG_GENERIC_CPU_AUTOPROBE |
fad12ac8 TR |
465 | &dev_attr_modalias.attr, |
466 | #endif | |
8a25a2fd KS |
467 | NULL |
468 | }; | |
469 | ||
470 | static struct attribute_group cpu_root_attr_group = { | |
471 | .attrs = cpu_root_attrs, | |
472 | }; | |
473 | ||
474 | static const struct attribute_group *cpu_root_attr_groups[] = { | |
475 | &cpu_root_attr_group, | |
476 | NULL, | |
477 | }; | |
1da177e4 | 478 | |
2987557f JT |
479 | bool cpu_is_hotpluggable(unsigned cpu) |
480 | { | |
7affca35 LT |
481 | struct device *dev = get_cpu_device(cpu); |
482 | return dev && container_of(dev, struct cpu, dev)->hotpluggable; | |
2987557f JT |
483 | } |
484 | EXPORT_SYMBOL_GPL(cpu_is_hotpluggable); | |
485 | ||
9f13a1fd BH |
486 | #ifdef CONFIG_GENERIC_CPU_DEVICES |
487 | static DEFINE_PER_CPU(struct cpu, cpu_devices); | |
488 | #endif | |
489 | ||
490 | static void __init cpu_dev_register_generic(void) | |
491 | { | |
492 | #ifdef CONFIG_GENERIC_CPU_DEVICES | |
493 | int i; | |
494 | ||
495 | for_each_possible_cpu(i) { | |
496 | if (register_cpu(&per_cpu(cpu_devices, i), i)) | |
497 | panic("Failed to register CPU device"); | |
498 | } | |
499 | #endif | |
500 | } | |
501 | ||
024f7846 | 502 | void __init cpu_dev_init(void) |
1da177e4 | 503 | { |
024f7846 BH |
504 | if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups)) |
505 | panic("Failed to register CPU subsystem"); | |
8a25a2fd | 506 | |
9f13a1fd | 507 | cpu_dev_register_generic(); |
1da177e4 | 508 | } |