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> |
1da177e4 | 19 | |
a1bdc7aa | 20 | #include "base.h" |
1da177e4 | 21 | |
8a25a2fd | 22 | static DEFINE_PER_CPU(struct device *, cpu_sys_devices); |
ad74557a | 23 | |
ac212b69 RW |
24 | static int cpu_subsys_match(struct device *dev, struct device_driver *drv) |
25 | { | |
26 | /* ACPI style match is the only one that may succeed. */ | |
27 | if (acpi_driver_match_device(dev, drv)) | |
28 | return 1; | |
29 | ||
30 | return 0; | |
31 | } | |
32 | ||
1da177e4 | 33 | #ifdef CONFIG_HOTPLUG_CPU |
34640468 YI |
34 | static void change_cpu_under_node(struct cpu *cpu, |
35 | unsigned int from_nid, unsigned int to_nid) | |
36 | { | |
37 | int cpuid = cpu->dev.id; | |
38 | unregister_cpu_under_node(cpuid, from_nid); | |
39 | register_cpu_under_node(cpuid, to_nid); | |
40 | cpu->node_id = to_nid; | |
41 | } | |
42 | ||
0902a904 | 43 | static int __ref cpu_subsys_online(struct device *dev) |
1da177e4 | 44 | { |
8a25a2fd | 45 | struct cpu *cpu = container_of(dev, struct cpu, dev); |
0902a904 RW |
46 | int cpuid = dev->id; |
47 | int from_nid, to_nid; | |
6dedcca6 | 48 | int ret; |
1da177e4 | 49 | |
0902a904 | 50 | from_nid = cpu_to_node(cpuid); |
c7991b0b | 51 | if (from_nid == NUMA_NO_NODE) |
6dedcca6 | 52 | return -ENODEV; |
c7991b0b | 53 | |
0902a904 RW |
54 | ret = cpu_up(cpuid); |
55 | /* | |
56 | * When hot adding memory to memoryless node and enabling a cpu | |
57 | * on the node, node number of the cpu may internally change. | |
58 | */ | |
59 | to_nid = cpu_to_node(cpuid); | |
60 | if (from_nid != to_nid) | |
61 | change_cpu_under_node(cpu, from_nid, to_nid); | |
1da177e4 | 62 | |
0902a904 | 63 | return ret; |
1da177e4 LT |
64 | } |
65 | ||
0902a904 | 66 | static int cpu_subsys_offline(struct device *dev) |
1da177e4 | 67 | { |
6dedcca6 | 68 | return cpu_down(dev->id); |
1da177e4 | 69 | } |
1c4e2d70 | 70 | |
76b67ed9 | 71 | void unregister_cpu(struct cpu *cpu) |
1da177e4 | 72 | { |
8a25a2fd | 73 | int logical_cpu = cpu->dev.id; |
1da177e4 | 74 | |
76b67ed9 KH |
75 | unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu)); |
76 | ||
8a25a2fd | 77 | device_unregister(&cpu->dev); |
e37d05da | 78 | per_cpu(cpu_sys_devices, logical_cpu) = NULL; |
1da177e4 LT |
79 | return; |
80 | } | |
12633e80 NF |
81 | |
82 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE | |
8a25a2fd KS |
83 | static ssize_t cpu_probe_store(struct device *dev, |
84 | struct device_attribute *attr, | |
67fc233f | 85 | const char *buf, |
12633e80 NF |
86 | size_t count) |
87 | { | |
574b851e TK |
88 | ssize_t cnt; |
89 | int ret; | |
90 | ||
91 | ret = lock_device_hotplug_sysfs(); | |
92 | if (ret) | |
93 | return ret; | |
94 | ||
95 | cnt = arch_cpu_probe(buf, count); | |
96 | ||
97 | unlock_device_hotplug(); | |
98 | return cnt; | |
12633e80 NF |
99 | } |
100 | ||
8a25a2fd KS |
101 | static ssize_t cpu_release_store(struct device *dev, |
102 | struct device_attribute *attr, | |
67fc233f | 103 | const char *buf, |
12633e80 NF |
104 | size_t count) |
105 | { | |
574b851e TK |
106 | ssize_t cnt; |
107 | int ret; | |
108 | ||
109 | ret = lock_device_hotplug_sysfs(); | |
110 | if (ret) | |
111 | return ret; | |
112 | ||
113 | cnt = arch_cpu_release(buf, count); | |
114 | ||
115 | unlock_device_hotplug(); | |
116 | return cnt; | |
12633e80 NF |
117 | } |
118 | ||
8a25a2fd KS |
119 | static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store); |
120 | static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store); | |
12633e80 | 121 | #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ |
1da177e4 LT |
122 | #endif /* CONFIG_HOTPLUG_CPU */ |
123 | ||
0902a904 RW |
124 | struct bus_type cpu_subsys = { |
125 | .name = "cpu", | |
126 | .dev_name = "cpu", | |
ac212b69 | 127 | .match = cpu_subsys_match, |
0902a904 RW |
128 | #ifdef CONFIG_HOTPLUG_CPU |
129 | .online = cpu_subsys_online, | |
130 | .offline = cpu_subsys_offline, | |
131 | #endif | |
132 | }; | |
133 | EXPORT_SYMBOL_GPL(cpu_subsys); | |
134 | ||
51be5606 VG |
135 | #ifdef CONFIG_KEXEC |
136 | #include <linux/kexec.h> | |
137 | ||
8a25a2fd | 138 | static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr, |
4a0b2b4d | 139 | char *buf) |
51be5606 | 140 | { |
8a25a2fd | 141 | struct cpu *cpu = container_of(dev, struct cpu, dev); |
51be5606 VG |
142 | ssize_t rc; |
143 | unsigned long long addr; | |
144 | int cpunum; | |
145 | ||
8a25a2fd | 146 | cpunum = cpu->dev.id; |
51be5606 VG |
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 | */ | |
3b034b0d | 154 | addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum)); |
51be5606 | 155 | rc = sprintf(buf, "%Lx\n", addr); |
51be5606 VG |
156 | return rc; |
157 | } | |
8a25a2fd | 158 | static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL); |
eca4549f ZY |
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 | ||
bcfb87fb | 166 | rc = sprintf(buf, "%zu\n", sizeof(note_buf_t)); |
eca4549f ZY |
167 | return rc; |
168 | } | |
169 | static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL); | |
c055da9f IM |
170 | |
171 | static struct attribute *crash_note_cpu_attrs[] = { | |
172 | &dev_attr_crash_notes.attr, | |
173 | &dev_attr_crash_notes_size.attr, | |
174 | NULL | |
175 | }; | |
176 | ||
177 | static struct attribute_group crash_note_cpu_attr_group = { | |
178 | .attrs = crash_note_cpu_attrs, | |
179 | }; | |
180 | #endif | |
181 | ||
182 | static const struct attribute_group *common_cpu_attr_groups[] = { | |
183 | #ifdef CONFIG_KEXEC | |
184 | &crash_note_cpu_attr_group, | |
51be5606 | 185 | #endif |
c055da9f IM |
186 | NULL |
187 | }; | |
51be5606 | 188 | |
1c4e2d70 IM |
189 | static const struct attribute_group *hotplugable_cpu_attr_groups[] = { |
190 | #ifdef CONFIG_KEXEC | |
191 | &crash_note_cpu_attr_group, | |
1c4e2d70 IM |
192 | #endif |
193 | NULL | |
194 | }; | |
195 | ||
9d1fe323 MT |
196 | /* |
197 | * Print cpu online, possible, present, and system maps | |
198 | */ | |
265d2e2e AK |
199 | |
200 | struct cpu_attr { | |
8a25a2fd | 201 | struct device_attribute attr; |
265d2e2e AK |
202 | const struct cpumask *const * const map; |
203 | }; | |
204 | ||
8a25a2fd KS |
205 | static ssize_t show_cpus_attr(struct device *dev, |
206 | struct device_attribute *attr, | |
265d2e2e | 207 | char *buf) |
9d1fe323 | 208 | { |
265d2e2e AK |
209 | struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr); |
210 | int n = cpulist_scnprintf(buf, PAGE_SIZE-2, *(ca->map)); | |
9d1fe323 MT |
211 | |
212 | buf[n++] = '\n'; | |
213 | buf[n] = '\0'; | |
214 | return n; | |
215 | } | |
216 | ||
8a25a2fd KS |
217 | #define _CPU_ATTR(name, map) \ |
218 | { __ATTR(name, 0444, show_cpus_attr, NULL), map } | |
9d1fe323 | 219 | |
8a25a2fd | 220 | /* Keep in sync with cpu_subsys_attrs */ |
265d2e2e AK |
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 | }; | |
9d1fe323 | 226 | |
e057d7ae MT |
227 | /* |
228 | * Print values for NR_CPUS and offlined cpus | |
229 | */ | |
8a25a2fd KS |
230 | static ssize_t print_cpus_kernel_max(struct device *dev, |
231 | struct device_attribute *attr, char *buf) | |
e057d7ae | 232 | { |
8fd2d2d5 | 233 | int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1); |
e057d7ae MT |
234 | return n; |
235 | } | |
8a25a2fd | 236 | static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL); |
e057d7ae MT |
237 | |
238 | /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */ | |
239 | unsigned int total_cpus; | |
240 | ||
8a25a2fd KS |
241 | static ssize_t print_cpus_offline(struct device *dev, |
242 | struct device_attribute *attr, char *buf) | |
e057d7ae MT |
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; | |
cdc6e3d3 | 250 | cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask); |
e057d7ae MT |
251 | n = cpulist_scnprintf(buf, len, 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, "%d", nr_cpu_ids); | |
261 | else | |
262 | n += snprintf(&buf[n], len - n, "%d-%d", | |
263 | nr_cpu_ids, total_cpus-1); | |
264 | } | |
265 | ||
266 | n += snprintf(&buf[n], len - n, "\n"); | |
267 | return n; | |
268 | } | |
8a25a2fd | 269 | static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL); |
e057d7ae | 270 | |
2885e25c GKH |
271 | static void cpu_device_release(struct device *dev) |
272 | { | |
273 | /* | |
274 | * This is an empty function to prevent the driver core from spitting a | |
275 | * warning at us. Yes, I know this is directly opposite of what the | |
276 | * documentation for the driver core and kobjects say, and the author | |
277 | * of this code has already been publically ridiculed for doing | |
278 | * something as foolish as this. However, at this point in time, it is | |
279 | * the only way to handle the issue of statically allocated cpu | |
280 | * devices. The different architectures will have their cpu device | |
281 | * code reworked to properly handle this in the near future, so this | |
282 | * function will then be changed to correctly free up the memory held | |
283 | * by the cpu device. | |
284 | * | |
285 | * Never copy this way of doing things, or you too will be made fun of | |
30a4840a | 286 | * on the linux-kernel list, you have been warned. |
2885e25c GKH |
287 | */ |
288 | } | |
289 | ||
67bad2fd AB |
290 | #ifdef CONFIG_GENERIC_CPU_AUTOPROBE |
291 | static ssize_t print_cpu_modalias(struct device *dev, | |
292 | struct device_attribute *attr, | |
293 | char *buf) | |
294 | { | |
295 | ssize_t n; | |
296 | u32 i; | |
297 | ||
298 | n = sprintf(buf, "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:", | |
299 | CPU_FEATURE_TYPEVAL); | |
300 | ||
301 | for (i = 0; i < MAX_CPU_FEATURES; i++) | |
302 | if (cpu_have_feature(i)) { | |
303 | if (PAGE_SIZE < n + sizeof(",XXXX\n")) { | |
304 | WARN(1, "CPU features overflow page\n"); | |
305 | break; | |
306 | } | |
307 | n += sprintf(&buf[n], ",%04X", i); | |
308 | } | |
309 | buf[n++] = '\n'; | |
310 | return n; | |
311 | } | |
67bad2fd AB |
312 | |
313 | static int cpu_uevent(struct device *dev, struct kobj_uevent_env *env) | |
314 | { | |
315 | char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
316 | if (buf) { | |
317 | print_cpu_modalias(NULL, NULL, buf); | |
318 | add_uevent_var(env, "MODALIAS=%s", buf); | |
319 | kfree(buf); | |
320 | } | |
321 | return 0; | |
322 | } | |
323 | #endif | |
324 | ||
1da177e4 | 325 | /* |
405ae7d3 | 326 | * register_cpu - Setup a sysfs device for a CPU. |
72486f1f SS |
327 | * @cpu - cpu->hotpluggable field set to 1 will generate a control file in |
328 | * sysfs for this CPU. | |
1da177e4 LT |
329 | * @num - CPU number to use when creating the device. |
330 | * | |
331 | * Initialize and register the CPU device. | |
332 | */ | |
a83048eb | 333 | int register_cpu(struct cpu *cpu, int num) |
1da177e4 LT |
334 | { |
335 | int error; | |
76b67ed9 | 336 | |
8a25a2fd | 337 | cpu->node_id = cpu_to_node(num); |
29bb5d4f | 338 | memset(&cpu->dev, 0x00, sizeof(struct device)); |
8a25a2fd KS |
339 | cpu->dev.id = num; |
340 | cpu->dev.bus = &cpu_subsys; | |
2885e25c | 341 | cpu->dev.release = cpu_device_release; |
0902a904 | 342 | cpu->dev.offline_disabled = !cpu->hotpluggable; |
1001b4d4 | 343 | cpu->dev.offline = !cpu_online(num); |
f86e4718 | 344 | cpu->dev.of_node = of_get_cpu_node(num, NULL); |
2b9c1f03 | 345 | #ifdef CONFIG_GENERIC_CPU_AUTOPROBE |
67bad2fd | 346 | cpu->dev.bus->uevent = cpu_uevent; |
fad12ac8 | 347 | #endif |
c055da9f | 348 | cpu->dev.groups = common_cpu_attr_groups; |
1c4e2d70 IM |
349 | if (cpu->hotpluggable) |
350 | cpu->dev.groups = hotplugable_cpu_attr_groups; | |
8a25a2fd | 351 | error = device_register(&cpu->dev); |
ad74557a | 352 | if (!error) |
8a25a2fd | 353 | per_cpu(cpu_sys_devices, num) = &cpu->dev; |
76b67ed9 KH |
354 | if (!error) |
355 | register_cpu_under_node(num, cpu_to_node(num)); | |
51be5606 | 356 | |
1da177e4 LT |
357 | return error; |
358 | } | |
359 | ||
8a25a2fd | 360 | struct device *get_cpu_device(unsigned cpu) |
ad74557a | 361 | { |
e37d05da MT |
362 | if (cpu < nr_cpu_ids && cpu_possible(cpu)) |
363 | return per_cpu(cpu_sys_devices, cpu); | |
ad74557a AR |
364 | else |
365 | return NULL; | |
366 | } | |
8a25a2fd KS |
367 | EXPORT_SYMBOL_GPL(get_cpu_device); |
368 | ||
2b9c1f03 | 369 | #ifdef CONFIG_GENERIC_CPU_AUTOPROBE |
67bad2fd | 370 | static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL); |
fad12ac8 TR |
371 | #endif |
372 | ||
8a25a2fd KS |
373 | static struct attribute *cpu_root_attrs[] = { |
374 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE | |
375 | &dev_attr_probe.attr, | |
376 | &dev_attr_release.attr, | |
377 | #endif | |
378 | &cpu_attrs[0].attr.attr, | |
379 | &cpu_attrs[1].attr.attr, | |
380 | &cpu_attrs[2].attr.attr, | |
381 | &dev_attr_kernel_max.attr, | |
382 | &dev_attr_offline.attr, | |
2b9c1f03 | 383 | #ifdef CONFIG_GENERIC_CPU_AUTOPROBE |
fad12ac8 TR |
384 | &dev_attr_modalias.attr, |
385 | #endif | |
8a25a2fd KS |
386 | NULL |
387 | }; | |
388 | ||
389 | static struct attribute_group cpu_root_attr_group = { | |
390 | .attrs = cpu_root_attrs, | |
391 | }; | |
392 | ||
393 | static const struct attribute_group *cpu_root_attr_groups[] = { | |
394 | &cpu_root_attr_group, | |
395 | NULL, | |
396 | }; | |
1da177e4 | 397 | |
2987557f JT |
398 | bool cpu_is_hotpluggable(unsigned cpu) |
399 | { | |
7affca35 LT |
400 | struct device *dev = get_cpu_device(cpu); |
401 | return dev && container_of(dev, struct cpu, dev)->hotpluggable; | |
2987557f JT |
402 | } |
403 | EXPORT_SYMBOL_GPL(cpu_is_hotpluggable); | |
404 | ||
9f13a1fd BH |
405 | #ifdef CONFIG_GENERIC_CPU_DEVICES |
406 | static DEFINE_PER_CPU(struct cpu, cpu_devices); | |
407 | #endif | |
408 | ||
409 | static void __init cpu_dev_register_generic(void) | |
410 | { | |
411 | #ifdef CONFIG_GENERIC_CPU_DEVICES | |
412 | int i; | |
413 | ||
414 | for_each_possible_cpu(i) { | |
415 | if (register_cpu(&per_cpu(cpu_devices, i), i)) | |
416 | panic("Failed to register CPU device"); | |
417 | } | |
418 | #endif | |
419 | } | |
420 | ||
024f7846 | 421 | void __init cpu_dev_init(void) |
1da177e4 | 422 | { |
024f7846 BH |
423 | if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups)) |
424 | panic("Failed to register CPU subsystem"); | |
8a25a2fd | 425 | |
9f13a1fd | 426 | cpu_dev_register_generic(); |
1da177e4 | 427 | } |