projects/linux·blob v3.13 · refs

391 lines · 9.4 KiB

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
19#include "base.h"
20
21static DEFINE_PER_CPU(struct device *, cpu_sys_devices);
22
23static 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
32#ifdef CONFIG_HOTPLUG_CPU
33static 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
42static int __ref cpu_subsys_online(struct device *dev)
43{
44 struct cpu *cpu = container_of(dev, struct cpu, dev);
45 int cpuid = dev->id;
46 int from_nid, to_nid;
47 int ret;
48
49 from_nid = cpu_to_node(cpuid);
50 if (from_nid == NUMA_NO_NODE)
51 return -ENODEV;
52
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);
61
62 return ret;
63}
64
65static int cpu_subsys_offline(struct device *dev)
66{
67 return cpu_down(dev->id);
68}
69
70void unregister_cpu(struct cpu *cpu)
71{
72 int logical_cpu = cpu->dev.id;
73
74 unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu));
75
76 device_unregister(&cpu->dev);
77 per_cpu(cpu_sys_devices, logical_cpu) = NULL;
78 return;
79}
80
81#ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
82static ssize_t cpu_probe_store(struct device *dev,
83 struct device_attribute *attr,
84 const char *buf,
85 size_t count)
86{
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;
98}
99
100static ssize_t cpu_release_store(struct device *dev,
101 struct device_attribute *attr,
102 const char *buf,
103 size_t count)
104{
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;
116}
117
118static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store);
119static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store);
120#endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
121#endif /* CONFIG_HOTPLUG_CPU */
122
123struct bus_type cpu_subsys = {
124 .name = "cpu",
125 .dev_name = "cpu",
126 .match = cpu_subsys_match,
127#ifdef CONFIG_HOTPLUG_CPU
128 .online = cpu_subsys_online,
129 .offline = cpu_subsys_offline,
130#endif
131};
132EXPORT_SYMBOL_GPL(cpu_subsys);
133
134#ifdef CONFIG_KEXEC
135#include <linux/kexec.h>
136
137static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr,
138 char *buf)
139{
140 struct cpu *cpu = container_of(dev, struct cpu, dev);
141 ssize_t rc;
142 unsigned long long addr;
143 int cpunum;
144
145 cpunum = cpu->dev.id;
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 */
153 addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum));
154 rc = sprintf(buf, "%Lx\n", addr);
155 return rc;
156}
157static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL);
158
159static ssize_t show_crash_notes_size(struct device *dev,
160 struct device_attribute *attr,
161 char *buf)
162{
163 ssize_t rc;
164
165 rc = sprintf(buf, "%zu\n", sizeof(note_buf_t));
166 return rc;
167}
168static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL);
169
170static struct attribute *crash_note_cpu_attrs[] = {
171 &dev_attr_crash_notes.attr,
172 &dev_attr_crash_notes_size.attr,
173 NULL
174};
175
176static struct attribute_group crash_note_cpu_attr_group = {
177 .attrs = crash_note_cpu_attrs,
178};
179#endif
180
181static const struct attribute_group *common_cpu_attr_groups[] = {
182#ifdef CONFIG_KEXEC
183 &crash_note_cpu_attr_group,
184#endif
185 NULL
186};
187
188static const struct attribute_group *hotplugable_cpu_attr_groups[] = {
189#ifdef CONFIG_KEXEC
190 &crash_note_cpu_attr_group,
191#endif
192 NULL
193};
194
195/*
196 * Print cpu online, possible, present, and system maps
197 */
198
199struct cpu_attr {
200 struct device_attribute attr;
201 const struct cpumask *const * const map;
202};
203
204static ssize_t show_cpus_attr(struct device *dev,
205 struct device_attribute *attr,
206 char *buf)
207{
208 struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr);
209 int n = cpulist_scnprintf(buf, PAGE_SIZE-2, *(ca->map));
210
211 buf[n++] = '\n';
212 buf[n] = '\0';
213 return n;
214}
215
216#define _CPU_ATTR(name, map) \
217 { __ATTR(name, 0444, show_cpus_attr, NULL), map }
218
219/* Keep in sync with cpu_subsys_attrs */
220static 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};
225
226/*
227 * Print values for NR_CPUS and offlined cpus
228 */
229static ssize_t print_cpus_kernel_max(struct device *dev,
230 struct device_attribute *attr, char *buf)
231{
232 int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1);
233 return n;
234}
235static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL);
236
237/* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */
238unsigned int total_cpus;
239
240static ssize_t print_cpus_offline(struct device *dev,
241 struct device_attribute *attr, char *buf)
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;
249 cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask);
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}
268static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL);
269
270static 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
285 * on the linux-kernel list, you have been warned.
286 */
287}
288
289/*
290 * register_cpu - Setup a sysfs device for a CPU.
291 * @cpu - cpu->hotpluggable field set to 1 will generate a control file in
292 * sysfs for this CPU.
293 * @num - CPU number to use when creating the device.
294 *
295 * Initialize and register the CPU device.
296 */
297int register_cpu(struct cpu *cpu, int num)
298{
299 int error;
300
301 cpu->node_id = cpu_to_node(num);
302 memset(&cpu->dev, 0x00, sizeof(struct device));
303 cpu->dev.id = num;
304 cpu->dev.bus = &cpu_subsys;
305 cpu->dev.release = cpu_device_release;
306 cpu->dev.offline_disabled = !cpu->hotpluggable;
307 cpu->dev.offline = !cpu_online(num);
308 cpu->dev.of_node = of_get_cpu_node(num, NULL);
309#ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE
310 cpu->dev.bus->uevent = arch_cpu_uevent;
311#endif
312 cpu->dev.groups = common_cpu_attr_groups;
313 if (cpu->hotpluggable)
314 cpu->dev.groups = hotplugable_cpu_attr_groups;
315 error = device_register(&cpu->dev);
316 if (!error)
317 per_cpu(cpu_sys_devices, num) = &cpu->dev;
318 if (!error)
319 register_cpu_under_node(num, cpu_to_node(num));
320
321 return error;
322}
323
324struct device *get_cpu_device(unsigned cpu)
325{
326 if (cpu < nr_cpu_ids && cpu_possible(cpu))
327 return per_cpu(cpu_sys_devices, cpu);
328 else
329 return NULL;
330}
331EXPORT_SYMBOL_GPL(get_cpu_device);
332
333#ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE
334static DEVICE_ATTR(modalias, 0444, arch_print_cpu_modalias, NULL);
335#endif
336
337static 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,
347#ifdef CONFIG_ARCH_HAS_CPU_AUTOPROBE
348 &dev_attr_modalias.attr,
349#endif
350 NULL
351};
352
353static struct attribute_group cpu_root_attr_group = {
354 .attrs = cpu_root_attrs,
355};
356
357static const struct attribute_group *cpu_root_attr_groups[] = {
358 &cpu_root_attr_group,
359 NULL,
360};
361
362bool cpu_is_hotpluggable(unsigned cpu)
363{
364 struct device *dev = get_cpu_device(cpu);
365 return dev && container_of(dev, struct cpu, dev)->hotpluggable;
366}
367EXPORT_SYMBOL_GPL(cpu_is_hotpluggable);
368
369#ifdef CONFIG_GENERIC_CPU_DEVICES
370static DEFINE_PER_CPU(struct cpu, cpu_devices);
371#endif
372
373static 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
385void __init cpu_dev_init(void)
386{
387 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups))
388 panic("Failed to register CPU subsystem");
389
390 cpu_dev_register_generic();
391}