]>
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> |
1da177e4 | 18 | |
a1bdc7aa | 19 | #include "base.h" |
1da177e4 | 20 | |
8a25a2fd | 21 | static DEFINE_PER_CPU(struct device *, cpu_sys_devices); |
ad74557a | 22 | |
ac212b69 RW |
23 | static int cpu_subsys_match(struct device *dev, struct device_driver *drv) |
24 | { | |
25 | /* ACPI style match is the only one that may succeed. */ | |
26 | if (acpi_driver_match_device(dev, drv)) | |
27 | return 1; | |
28 | ||
29 | return 0; | |
30 | } | |
31 | ||
1da177e4 | 32 | #ifdef CONFIG_HOTPLUG_CPU |
34640468 YI |
33 | static void change_cpu_under_node(struct cpu *cpu, |
34 | unsigned int from_nid, unsigned int to_nid) | |
35 | { | |
36 | int cpuid = cpu->dev.id; | |
37 | unregister_cpu_under_node(cpuid, from_nid); | |
38 | register_cpu_under_node(cpuid, to_nid); | |
39 | cpu->node_id = to_nid; | |
40 | } | |
41 | ||
0902a904 | 42 | static int __ref cpu_subsys_online(struct device *dev) |
1da177e4 | 43 | { |
8a25a2fd | 44 | struct cpu *cpu = container_of(dev, struct cpu, dev); |
0902a904 RW |
45 | int cpuid = dev->id; |
46 | int from_nid, to_nid; | |
6dedcca6 | 47 | int ret; |
1da177e4 | 48 | |
0902a904 | 49 | from_nid = cpu_to_node(cpuid); |
c7991b0b | 50 | if (from_nid == NUMA_NO_NODE) |
6dedcca6 | 51 | return -ENODEV; |
c7991b0b | 52 | |
0902a904 RW |
53 | ret = cpu_up(cpuid); |
54 | /* | |
55 | * When hot adding memory to memoryless node and enabling a cpu | |
56 | * on the node, node number of the cpu may internally change. | |
57 | */ | |
58 | to_nid = cpu_to_node(cpuid); | |
59 | if (from_nid != to_nid) | |
60 | change_cpu_under_node(cpu, from_nid, to_nid); | |
1da177e4 | 61 | |
0902a904 | 62 | return ret; |
1da177e4 LT |
63 | } |
64 | ||
0902a904 | 65 | static int cpu_subsys_offline(struct device *dev) |
1da177e4 | 66 | { |
6dedcca6 | 67 | return cpu_down(dev->id); |
1da177e4 | 68 | } |
1c4e2d70 | 69 | |
76b67ed9 | 70 | void unregister_cpu(struct cpu *cpu) |
1da177e4 | 71 | { |
8a25a2fd | 72 | int logical_cpu = cpu->dev.id; |
1da177e4 | 73 | |
76b67ed9 KH |
74 | unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu)); |
75 | ||
8a25a2fd | 76 | device_unregister(&cpu->dev); |
e37d05da | 77 | per_cpu(cpu_sys_devices, logical_cpu) = NULL; |
1da177e4 LT |
78 | return; |
79 | } | |
12633e80 NF |
80 | |
81 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE | |
8a25a2fd KS |
82 | static ssize_t cpu_probe_store(struct device *dev, |
83 | struct device_attribute *attr, | |
67fc233f | 84 | const char *buf, |
12633e80 NF |
85 | size_t count) |
86 | { | |
574b851e TK |
87 | ssize_t cnt; |
88 | int ret; | |
89 | ||
90 | ret = lock_device_hotplug_sysfs(); | |
91 | if (ret) | |
92 | return ret; | |
93 | ||
94 | cnt = arch_cpu_probe(buf, count); | |
95 | ||
96 | unlock_device_hotplug(); | |
97 | return cnt; | |
12633e80 NF |
98 | } |
99 | ||
8a25a2fd KS |
100 | static ssize_t cpu_release_store(struct device *dev, |
101 | struct device_attribute *attr, | |
67fc233f | 102 | const char *buf, |
12633e80 NF |
103 | size_t count) |
104 | { | |
574b851e TK |
105 | ssize_t cnt; |
106 | int ret; | |
107 | ||
108 | ret = lock_device_hotplug_sysfs(); | |
109 | if (ret) | |
110 | return ret; | |
111 | ||
112 | cnt = arch_cpu_release(buf, count); | |
113 | ||
114 | unlock_device_hotplug(); | |
115 | return cnt; | |
12633e80 NF |
116 | } |
117 | ||
8a25a2fd KS |
118 | static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store); |
119 | static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store); | |
12633e80 | 120 | #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ |
1da177e4 LT |
121 | #endif /* CONFIG_HOTPLUG_CPU */ |
122 | ||
0902a904 RW |
123 | struct bus_type cpu_subsys = { |
124 | .name = "cpu", | |
125 | .dev_name = "cpu", | |
ac212b69 | 126 | .match = cpu_subsys_match, |
0902a904 RW |
127 | #ifdef CONFIG_HOTPLUG_CPU |
128 | .online = cpu_subsys_online, | |
129 | .offline = cpu_subsys_offline, | |
130 | #endif | |
131 | }; | |
132 | EXPORT_SYMBOL_GPL(cpu_subsys); | |
133 | ||
51be5606 VG |
134 | #ifdef CONFIG_KEXEC |
135 | #include <linux/kexec.h> | |
136 | ||
8a25a2fd | 137 | static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr, |
4a0b2b4d | 138 | char *buf) |
51be5606 | 139 | { |
8a25a2fd | 140 | struct cpu *cpu = container_of(dev, struct cpu, dev); |
51be5606 VG |
141 | ssize_t rc; |
142 | unsigned long long addr; | |
143 | int cpunum; | |
144 | ||
8a25a2fd | 145 | cpunum = cpu->dev.id; |
51be5606 VG |
146 | |
147 | /* | |
148 | * Might be reading other cpu's data based on which cpu read thread | |
149 | * has been scheduled. But cpu data (memory) is allocated once during | |
150 | * boot up and this data does not change there after. Hence this | |
151 | * operation should be safe. No locking required. | |
152 | */ | |
3b034b0d | 153 | addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum)); |
51be5606 | 154 | rc = sprintf(buf, "%Lx\n", addr); |
51be5606 VG |
155 | return rc; |
156 | } | |
8a25a2fd | 157 | static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL); |
eca4549f ZY |
158 | |
159 | static ssize_t show_crash_notes_size(struct device *dev, | |
160 | struct device_attribute *attr, | |
161 | char *buf) | |
162 | { | |
163 | ssize_t rc; | |
164 | ||
bcfb87fb | 165 | rc = sprintf(buf, "%zu\n", sizeof(note_buf_t)); |
eca4549f ZY |
166 | return rc; |
167 | } | |
168 | static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL); | |
c055da9f IM |
169 | |
170 | static struct attribute *crash_note_cpu_attrs[] = { | |
171 | &dev_attr_crash_notes.attr, | |
172 | &dev_attr_crash_notes_size.attr, | |
173 | NULL | |
174 | }; | |
175 | ||
176 | static struct attribute_group crash_note_cpu_attr_group = { | |
177 | .attrs = crash_note_cpu_attrs, | |
178 | }; | |
179 | #endif | |
180 | ||
181 | static const struct attribute_group *common_cpu_attr_groups[] = { | |
182 | #ifdef CONFIG_KEXEC | |
183 | &crash_note_cpu_attr_group, | |
51be5606 | 184 | #endif |
c055da9f IM |
185 | NULL |
186 | }; | |
51be5606 | 187 | |
1c4e2d70 IM |
188 | static const struct attribute_group *hotplugable_cpu_attr_groups[] = { |
189 | #ifdef CONFIG_KEXEC | |
190 | &crash_note_cpu_attr_group, | |
1c4e2d70 IM |
191 | #endif |
192 | NULL | |
193 | }; | |
194 | ||
9d1fe323 MT |
195 | /* |
196 | * Print cpu online, possible, present, and system maps | |
197 | */ | |
265d2e2e AK |
198 | |
199 | struct cpu_attr { | |
8a25a2fd | 200 | struct device_attribute attr; |
265d2e2e AK |
201 | const struct cpumask *const * const map; |
202 | }; | |
203 | ||
8a25a2fd KS |
204 | static ssize_t show_cpus_attr(struct device *dev, |
205 | struct device_attribute *attr, | |
265d2e2e | 206 | char *buf) |
9d1fe323 | 207 | { |
265d2e2e AK |
208 | struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr); |
209 | int n = cpulist_scnprintf(buf, PAGE_SIZE-2, *(ca->map)); | |
9d1fe323 MT |
210 | |
211 | buf[n++] = '\n'; | |
212 | buf[n] = '\0'; | |
213 | return n; | |
214 | } | |
215 | ||
8a25a2fd KS |
216 | #define _CPU_ATTR(name, map) \ |
217 | { __ATTR(name, 0444, show_cpus_attr, NULL), map } | |
9d1fe323 | 218 | |
8a25a2fd | 219 | /* Keep in sync with cpu_subsys_attrs */ |
265d2e2e AK |
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 | }; | |
9d1fe323 | 225 | |
e057d7ae MT |
226 | /* |
227 | * Print values for NR_CPUS and offlined cpus | |
228 | */ | |
8a25a2fd KS |
229 | static ssize_t print_cpus_kernel_max(struct device *dev, |
230 | struct device_attribute *attr, char *buf) | |
e057d7ae | 231 | { |
8fd2d2d5 | 232 | int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1); |
e057d7ae MT |
233 | return n; |
234 | } | |
8a25a2fd | 235 | static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL); |
e057d7ae MT |
236 | |
237 | /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */ | |
238 | unsigned int total_cpus; | |
239 | ||
8a25a2fd KS |
240 | static ssize_t print_cpus_offline(struct device *dev, |
241 | struct device_attribute *attr, char *buf) | |
e057d7ae MT |
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; | |
cdc6e3d3 | 249 | cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask); |
e057d7ae MT |
250 | n = cpulist_scnprintf(buf, len, 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, "%d", nr_cpu_ids); | |
260 | else | |
261 | n += snprintf(&buf[n], len - n, "%d-%d", | |
262 | nr_cpu_ids, total_cpus-1); | |
263 | } | |
264 | ||
265 | n += snprintf(&buf[n], len - n, "\n"); | |
266 | return n; | |
267 | } | |
8a25a2fd | 268 | static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL); |
e057d7ae | 269 | |
2885e25c GKH |
270 | static void cpu_device_release(struct device *dev) |
271 | { | |
272 | /* | |
273 | * This is an empty function to prevent the driver core from spitting a | |
274 | * warning at us. Yes, I know this is directly opposite of what the | |
275 | * documentation for the driver core and kobjects say, and the author | |
276 | * of this code has already been publically ridiculed for doing | |
277 | * something as foolish as this. However, at this point in time, it is | |
278 | * the only way to handle the issue of statically allocated cpu | |
279 | * devices. The different architectures will have their cpu device | |
280 | * code reworked to properly handle this in the near future, so this | |
281 | * function will then be changed to correctly free up the memory held | |
282 | * by the cpu device. | |
283 | * | |
284 | * Never copy this way of doing things, or you too will be made fun of | |
30a4840a | 285 | * on the linux-kernel list, you have been warned. |
2885e25c GKH |
286 | */ |
287 | } | |
288 | ||
1da177e4 | 289 | /* |
405ae7d3 | 290 | * register_cpu - Setup a sysfs device for a CPU. |
72486f1f SS |
291 | * @cpu - cpu->hotpluggable field set to 1 will generate a control file in |
292 | * sysfs for this CPU. | |
1da177e4 LT |
293 | * @num - CPU number to use when creating the device. |
294 | * | |
295 | * Initialize and register the CPU device. | |
296 | */ | |
a83048eb | 297 | int register_cpu(struct cpu *cpu, int num) |
1da177e4 LT |
298 | { |
299 | int error; | |
76b67ed9 | 300 | |
8a25a2fd | 301 | cpu->node_id = cpu_to_node(num); |
29bb5d4f | 302 | memset(&cpu->dev, 0x00, sizeof(struct device)); |
8a25a2fd KS |
303 | cpu->dev.id = num; |
304 | cpu->dev.bus = &cpu_subsys; | |
2885e25c | 305 | cpu->dev.release = cpu_device_release; |
0902a904 | 306 | cpu->dev.offline_disabled = !cpu->hotpluggable; |
1001b4d4 | 307 | cpu->dev.offline = !cpu_online(num); |
f86e4718 | 308 | cpu->dev.of_node = of_get_cpu_node(num, NULL); |
fad12ac8 TR |
309 | #ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE |
310 | cpu->dev.bus->uevent = arch_cpu_uevent; | |
311 | #endif | |
c055da9f | 312 | cpu->dev.groups = common_cpu_attr_groups; |
1c4e2d70 IM |
313 | if (cpu->hotpluggable) |
314 | cpu->dev.groups = hotplugable_cpu_attr_groups; | |
8a25a2fd | 315 | error = device_register(&cpu->dev); |
ad74557a | 316 | if (!error) |
8a25a2fd | 317 | per_cpu(cpu_sys_devices, num) = &cpu->dev; |
76b67ed9 KH |
318 | if (!error) |
319 | register_cpu_under_node(num, cpu_to_node(num)); | |
51be5606 | 320 | |
1da177e4 LT |
321 | return error; |
322 | } | |
323 | ||
8a25a2fd | 324 | struct device *get_cpu_device(unsigned cpu) |
ad74557a | 325 | { |
e37d05da MT |
326 | if (cpu < nr_cpu_ids && cpu_possible(cpu)) |
327 | return per_cpu(cpu_sys_devices, cpu); | |
ad74557a AR |
328 | else |
329 | return NULL; | |
330 | } | |
8a25a2fd KS |
331 | EXPORT_SYMBOL_GPL(get_cpu_device); |
332 | ||
fad12ac8 TR |
333 | #ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE |
334 | static DEVICE_ATTR(modalias, 0444, arch_print_cpu_modalias, NULL); | |
335 | #endif | |
336 | ||
8a25a2fd KS |
337 | static struct attribute *cpu_root_attrs[] = { |
338 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE | |
339 | &dev_attr_probe.attr, | |
340 | &dev_attr_release.attr, | |
341 | #endif | |
342 | &cpu_attrs[0].attr.attr, | |
343 | &cpu_attrs[1].attr.attr, | |
344 | &cpu_attrs[2].attr.attr, | |
345 | &dev_attr_kernel_max.attr, | |
346 | &dev_attr_offline.attr, | |
fad12ac8 TR |
347 | #ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE |
348 | &dev_attr_modalias.attr, | |
349 | #endif | |
8a25a2fd KS |
350 | NULL |
351 | }; | |
352 | ||
353 | static struct attribute_group cpu_root_attr_group = { | |
354 | .attrs = cpu_root_attrs, | |
355 | }; | |
356 | ||
357 | static const struct attribute_group *cpu_root_attr_groups[] = { | |
358 | &cpu_root_attr_group, | |
359 | NULL, | |
360 | }; | |
1da177e4 | 361 | |
2987557f JT |
362 | bool cpu_is_hotpluggable(unsigned cpu) |
363 | { | |
7affca35 LT |
364 | struct device *dev = get_cpu_device(cpu); |
365 | return dev && container_of(dev, struct cpu, dev)->hotpluggable; | |
2987557f JT |
366 | } |
367 | EXPORT_SYMBOL_GPL(cpu_is_hotpluggable); | |
368 | ||
9f13a1fd BH |
369 | #ifdef CONFIG_GENERIC_CPU_DEVICES |
370 | static DEFINE_PER_CPU(struct cpu, cpu_devices); | |
371 | #endif | |
372 | ||
373 | static void __init cpu_dev_register_generic(void) | |
374 | { | |
375 | #ifdef CONFIG_GENERIC_CPU_DEVICES | |
376 | int i; | |
377 | ||
378 | for_each_possible_cpu(i) { | |
379 | if (register_cpu(&per_cpu(cpu_devices, i), i)) | |
380 | panic("Failed to register CPU device"); | |
381 | } | |
382 | #endif | |
383 | } | |
384 | ||
024f7846 | 385 | void __init cpu_dev_init(void) |
1da177e4 | 386 | { |
024f7846 BH |
387 | if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups)) |
388 | panic("Failed to register CPU subsystem"); | |
8a25a2fd | 389 | |
9f13a1fd | 390 | cpu_dev_register_generic(); |
1da177e4 | 391 | } |