Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * CPU subsystem support
4 */
5
6#include <linux/kernel.h>
7#include <linux/module.h>
8#include <linux/init.h>
9#include <linux/sched.h>
10#include <linux/cpu.h>
11#include <linux/topology.h>
12#include <linux/device.h>
13#include <linux/node.h>
14#include <linux/gfp.h>
15#include <linux/slab.h>
16#include <linux/percpu.h>
17#include <linux/acpi.h>
18#include <linux/of.h>
19#include <linux/cpufeature.h>
20#include <linux/tick.h>
21#include <linux/pm_qos.h>
22#include <linux/delay.h>
23#include <linux/sched/isolation.h>
24
25#include "base.h"
26
27static DEFINE_PER_CPU(struct device *, cpu_sys_devices);
28
29static int cpu_subsys_match(struct device *dev, struct device_driver *drv)
30{
31 /* ACPI style match is the only one that may succeed. */
32 if (acpi_driver_match_device(dev, drv))
33 return 1;
34
35 return 0;
36}
37
38#ifdef CONFIG_HOTPLUG_CPU
39static void change_cpu_under_node(struct cpu *cpu,
40 unsigned int from_nid, unsigned int to_nid)
41{
42 int cpuid = cpu->dev.id;
43 unregister_cpu_under_node(cpuid, from_nid);
44 register_cpu_under_node(cpuid, to_nid);
45 cpu->node_id = to_nid;
46}
47
48static int cpu_subsys_online(struct device *dev)
49{
50 struct cpu *cpu = container_of(dev, struct cpu, dev);
51 int cpuid = dev->id;
52 int from_nid, to_nid;
53 int ret;
54 int retries = 0;
55
56 from_nid = cpu_to_node(cpuid);
57 if (from_nid == NUMA_NO_NODE)
58 return -ENODEV;
59
60retry:
61 ret = cpu_device_up(dev);
62
63 /*
64 * If -EBUSY is returned, it is likely that hotplug is temporarily
65 * disabled when cpu_hotplug_disable() was called. This condition is
66 * transient. So we retry after waiting for an exponentially
67 * increasing delay up to a total of at least 620ms as some PCI
68 * device initialization can take quite a while.
69 */
70 if (ret == -EBUSY) {
71 retries++;
72 if (retries > 5)
73 return ret;
74 msleep(10 * (1 << retries));
75 goto retry;
76 }
77
78 /*
79 * When hot adding memory to memoryless node and enabling a cpu
80 * on the node, node number of the cpu may internally change.
81 */
82 to_nid = cpu_to_node(cpuid);
83 if (from_nid != to_nid)
84 change_cpu_under_node(cpu, from_nid, to_nid);
85
86 return ret;
87}
88
89static int cpu_subsys_offline(struct device *dev)
90{
91 return cpu_device_down(dev);
92}
93
94void unregister_cpu(struct cpu *cpu)
95{
96 int logical_cpu = cpu->dev.id;
97
98 unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu));
99
100 device_unregister(&cpu->dev);
101 per_cpu(cpu_sys_devices, logical_cpu) = NULL;
102 return;
103}
104
105#ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
106static ssize_t cpu_probe_store(struct device *dev,
107 struct device_attribute *attr,
108 const char *buf,
109 size_t count)
110{
111 ssize_t cnt;
112 int ret;
113
114 ret = lock_device_hotplug_sysfs();
115 if (ret)
116 return ret;
117
118 cnt = arch_cpu_probe(buf, count);
119
120 unlock_device_hotplug();
121 return cnt;
122}
123
124static ssize_t cpu_release_store(struct device *dev,
125 struct device_attribute *attr,
126 const char *buf,
127 size_t count)
128{
129 ssize_t cnt;
130 int ret;
131
132 ret = lock_device_hotplug_sysfs();
133 if (ret)
134 return ret;
135
136 cnt = arch_cpu_release(buf, count);
137
138 unlock_device_hotplug();
139 return cnt;
140}
141
142static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store);
143static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store);
144#endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
145#endif /* CONFIG_HOTPLUG_CPU */
146
147#ifdef CONFIG_KEXEC_CORE
148#include <linux/kexec.h>
149
150static ssize_t crash_notes_show(struct device *dev,
151 struct device_attribute *attr,
152 char *buf)
153{
154 struct cpu *cpu = container_of(dev, struct cpu, dev);
155 unsigned long long addr;
156 int cpunum;
157
158 cpunum = cpu->dev.id;
159
160 /*
161 * Might be reading other cpu's data based on which cpu read thread
162 * has been scheduled. But cpu data (memory) is allocated once during
163 * boot up and this data does not change there after. Hence this
164 * operation should be safe. No locking required.
165 */
166 addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum));
167
168 return sysfs_emit(buf, "%llx\n", addr);
169}
170static DEVICE_ATTR_ADMIN_RO(crash_notes);
171
172static ssize_t crash_notes_size_show(struct device *dev,
173 struct device_attribute *attr,
174 char *buf)
175{
176 return sysfs_emit(buf, "%zu\n", sizeof(note_buf_t));
177}
178static DEVICE_ATTR_ADMIN_RO(crash_notes_size);
179
180static struct attribute *crash_note_cpu_attrs[] = {
181 &dev_attr_crash_notes.attr,
182 &dev_attr_crash_notes_size.attr,
183 NULL
184};
185
186static const struct attribute_group crash_note_cpu_attr_group = {
187 .attrs = crash_note_cpu_attrs,
188};
189#endif
190
191static const struct attribute_group *common_cpu_attr_groups[] = {
192#ifdef CONFIG_KEXEC_CORE
193 &crash_note_cpu_attr_group,
194#endif
195 NULL
196};
197
198static const struct attribute_group *hotplugable_cpu_attr_groups[] = {
199#ifdef CONFIG_KEXEC_CORE
200 &crash_note_cpu_attr_group,
201#endif
202 NULL
203};
204
205/*
206 * Print cpu online, possible, present, and system maps
207 */
208
209struct cpu_attr {
210 struct device_attribute attr;
211 const struct cpumask *const map;
212};
213
214static ssize_t show_cpus_attr(struct device *dev,
215 struct device_attribute *attr,
216 char *buf)
217{
218 struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr);
219
220 return cpumap_print_to_pagebuf(true, buf, ca->map);
221}
222
223#define _CPU_ATTR(name, map) \
224 { __ATTR(name, 0444, show_cpus_attr, NULL), map }
225
226/* Keep in sync with cpu_subsys_attrs */
227static struct cpu_attr cpu_attrs[] = {
228 _CPU_ATTR(online, &__cpu_online_mask),
229 _CPU_ATTR(possible, &__cpu_possible_mask),
230 _CPU_ATTR(present, &__cpu_present_mask),
231};
232
233/*
234 * Print values for NR_CPUS and offlined cpus
235 */
236static ssize_t print_cpus_kernel_max(struct device *dev,
237 struct device_attribute *attr, char *buf)
238{
239 return sysfs_emit(buf, "%d\n", NR_CPUS - 1);
240}
241static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL);
242
243/* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */
244unsigned int total_cpus;
245
246static ssize_t print_cpus_offline(struct device *dev,
247 struct device_attribute *attr, char *buf)
248{
249 int len = 0;
250 cpumask_var_t offline;
251
252 /* display offline cpus < nr_cpu_ids */
253 if (!alloc_cpumask_var(&offline, GFP_KERNEL))
254 return -ENOMEM;
255 cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask);
256 len += sysfs_emit_at(buf, len, "%*pbl", cpumask_pr_args(offline));
257 free_cpumask_var(offline);
258
259 /* display offline cpus >= nr_cpu_ids */
260 if (total_cpus && nr_cpu_ids < total_cpus) {
261 len += sysfs_emit_at(buf, len, ",");
262
263 if (nr_cpu_ids == total_cpus-1)
264 len += sysfs_emit_at(buf, len, "%u", nr_cpu_ids);
265 else
266 len += sysfs_emit_at(buf, len, "%u-%d",
267 nr_cpu_ids, total_cpus - 1);
268 }
269
270 len += sysfs_emit_at(buf, len, "\n");
271
272 return len;
273}
274static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL);
275
276static ssize_t print_cpus_isolated(struct device *dev,
277 struct device_attribute *attr, char *buf)
278{
279 int len;
280 cpumask_var_t isolated;
281
282 if (!alloc_cpumask_var(&isolated, GFP_KERNEL))
283 return -ENOMEM;
284
285 cpumask_andnot(isolated, cpu_possible_mask,
286 housekeeping_cpumask(HK_TYPE_DOMAIN));
287 len = sysfs_emit(buf, "%*pbl\n", cpumask_pr_args(isolated));
288
289 free_cpumask_var(isolated);
290
291 return len;
292}
293static DEVICE_ATTR(isolated, 0444, print_cpus_isolated, NULL);
294
295#ifdef CONFIG_NO_HZ_FULL
296static ssize_t print_cpus_nohz_full(struct device *dev,
297 struct device_attribute *attr, char *buf)
298{
299 return sysfs_emit(buf, "%*pbl\n", cpumask_pr_args(tick_nohz_full_mask));
300}
301static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL);
302#endif
303
304#ifdef CONFIG_CRASH_HOTPLUG
305static ssize_t crash_hotplug_show(struct device *dev,
306 struct device_attribute *attr,
307 char *buf)
308{
309 return sysfs_emit(buf, "%d\n", crash_hotplug_cpu_support());
310}
311static DEVICE_ATTR_ADMIN_RO(crash_hotplug);
312#endif
313
314static void cpu_device_release(struct device *dev)
315{
316 /*
317 * This is an empty function to prevent the driver core from spitting a
318 * warning at us. Yes, I know this is directly opposite of what the
319 * documentation for the driver core and kobjects say, and the author
320 * of this code has already been publically ridiculed for doing
321 * something as foolish as this. However, at this point in time, it is
322 * the only way to handle the issue of statically allocated cpu
323 * devices. The different architectures will have their cpu device
324 * code reworked to properly handle this in the near future, so this
325 * function will then be changed to correctly free up the memory held
326 * by the cpu device.
327 *
328 * Never copy this way of doing things, or you too will be made fun of
329 * on the linux-kernel list, you have been warned.
330 */
331}
332
333#ifdef CONFIG_GENERIC_CPU_AUTOPROBE
334static ssize_t print_cpu_modalias(struct device *dev,
335 struct device_attribute *attr,
336 char *buf)
337{
338 int len = 0;
339 u32 i;
340
341 len += sysfs_emit_at(buf, len,
342 "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:",
343 CPU_FEATURE_TYPEVAL);
344
345 for (i = 0; i < MAX_CPU_FEATURES; i++)
346 if (cpu_have_feature(i)) {
347 if (len + sizeof(",XXXX\n") >= PAGE_SIZE) {
348 WARN(1, "CPU features overflow page\n");
349 break;
350 }
351 len += sysfs_emit_at(buf, len, ",%04X", i);
352 }
353 len += sysfs_emit_at(buf, len, "\n");
354 return len;
355}
356
357static int cpu_uevent(const struct device *dev, struct kobj_uevent_env *env)
358{
359 char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
360 if (buf) {
361 print_cpu_modalias(NULL, NULL, buf);
362 add_uevent_var(env, "MODALIAS=%s", buf);
363 kfree(buf);
364 }
365 return 0;
366}
367#endif
368
369struct bus_type cpu_subsys = {
370 .name = "cpu",
371 .dev_name = "cpu",
372 .match = cpu_subsys_match,
373#ifdef CONFIG_HOTPLUG_CPU
374 .online = cpu_subsys_online,
375 .offline = cpu_subsys_offline,
376#endif
377#ifdef CONFIG_GENERIC_CPU_AUTOPROBE
378 .uevent = cpu_uevent,
379#endif
380};
381EXPORT_SYMBOL_GPL(cpu_subsys);
382
383/*
384 * register_cpu - Setup a sysfs device for a CPU.
385 * @cpu - cpu->hotpluggable field set to 1 will generate a control file in
386 * sysfs for this CPU.
387 * @num - CPU number to use when creating the device.
388 *
389 * Initialize and register the CPU device.
390 */
391int register_cpu(struct cpu *cpu, int num)
392{
393 int error;
394
395 cpu->node_id = cpu_to_node(num);
396 memset(&cpu->dev, 0x00, sizeof(struct device));
397 cpu->dev.id = num;
398 cpu->dev.bus = &cpu_subsys;
399 cpu->dev.release = cpu_device_release;
400 cpu->dev.offline_disabled = !cpu->hotpluggable;
401 cpu->dev.offline = !cpu_online(num);
402 cpu->dev.of_node = of_get_cpu_node(num, NULL);
403 cpu->dev.groups = common_cpu_attr_groups;
404 if (cpu->hotpluggable)
405 cpu->dev.groups = hotplugable_cpu_attr_groups;
406 error = device_register(&cpu->dev);
407 if (error) {
408 put_device(&cpu->dev);
409 return error;
410 }
411
412 per_cpu(cpu_sys_devices, num) = &cpu->dev;
413 register_cpu_under_node(num, cpu_to_node(num));
414 dev_pm_qos_expose_latency_limit(&cpu->dev,
415 PM_QOS_RESUME_LATENCY_NO_CONSTRAINT);
416
417 return 0;
418}
419
420struct device *get_cpu_device(unsigned int cpu)
421{
422 if (cpu < nr_cpu_ids && cpu_possible(cpu))
423 return per_cpu(cpu_sys_devices, cpu);
424 else
425 return NULL;
426}
427EXPORT_SYMBOL_GPL(get_cpu_device);
428
429static void device_create_release(struct device *dev)
430{
431 kfree(dev);
432}
433
434__printf(4, 0)
435static struct device *
436__cpu_device_create(struct device *parent, void *drvdata,
437 const struct attribute_group **groups,
438 const char *fmt, va_list args)
439{
440 struct device *dev = NULL;
441 int retval = -ENOMEM;
442
443 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
444 if (!dev)
445 goto error;
446
447 device_initialize(dev);
448 dev->parent = parent;
449 dev->groups = groups;
450 dev->release = device_create_release;
451 device_set_pm_not_required(dev);
452 dev_set_drvdata(dev, drvdata);
453
454 retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
455 if (retval)
456 goto error;
457
458 retval = device_add(dev);
459 if (retval)
460 goto error;
461
462 return dev;
463
464error:
465 put_device(dev);
466 return ERR_PTR(retval);
467}
468
469struct device *cpu_device_create(struct device *parent, void *drvdata,
470 const struct attribute_group **groups,
471 const char *fmt, ...)
472{
473 va_list vargs;
474 struct device *dev;
475
476 va_start(vargs, fmt);
477 dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs);
478 va_end(vargs);
479 return dev;
480}
481EXPORT_SYMBOL_GPL(cpu_device_create);
482
483#ifdef CONFIG_GENERIC_CPU_AUTOPROBE
484static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL);
485#endif
486
487static struct attribute *cpu_root_attrs[] = {
488#ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
489 &dev_attr_probe.attr,
490 &dev_attr_release.attr,
491#endif
492 &cpu_attrs[0].attr.attr,
493 &cpu_attrs[1].attr.attr,
494 &cpu_attrs[2].attr.attr,
495 &dev_attr_kernel_max.attr,
496 &dev_attr_offline.attr,
497 &dev_attr_isolated.attr,
498#ifdef CONFIG_NO_HZ_FULL
499 &dev_attr_nohz_full.attr,
500#endif
501#ifdef CONFIG_CRASH_HOTPLUG
502 &dev_attr_crash_hotplug.attr,
503#endif
504#ifdef CONFIG_GENERIC_CPU_AUTOPROBE
505 &dev_attr_modalias.attr,
506#endif
507 NULL
508};
509
510static const struct attribute_group cpu_root_attr_group = {
511 .attrs = cpu_root_attrs,
512};
513
514static const struct attribute_group *cpu_root_attr_groups[] = {
515 &cpu_root_attr_group,
516 NULL,
517};
518
519bool cpu_is_hotpluggable(unsigned int cpu)
520{
521 struct device *dev = get_cpu_device(cpu);
522 return dev && container_of(dev, struct cpu, dev)->hotpluggable
523 && tick_nohz_cpu_hotpluggable(cpu);
524}
525EXPORT_SYMBOL_GPL(cpu_is_hotpluggable);
526
527#ifdef CONFIG_GENERIC_CPU_DEVICES
528DEFINE_PER_CPU(struct cpu, cpu_devices);
529
530bool __weak arch_cpu_is_hotpluggable(int cpu)
531{
532 return false;
533}
534
535int __weak arch_register_cpu(int cpu)
536{
537 struct cpu *c = &per_cpu(cpu_devices, cpu);
538
539 c->hotpluggable = arch_cpu_is_hotpluggable(cpu);
540
541 return register_cpu(c, cpu);
542}
543
544#ifdef CONFIG_HOTPLUG_CPU
545void __weak arch_unregister_cpu(int num)
546{
547 unregister_cpu(&per_cpu(cpu_devices, num));
548}
549#endif /* CONFIG_HOTPLUG_CPU */
550#endif /* CONFIG_GENERIC_CPU_DEVICES */
551
552static void __init cpu_dev_register_generic(void)
553{
554 int i, ret;
555
556 if (!IS_ENABLED(CONFIG_GENERIC_CPU_DEVICES))
557 return;
558
559 for_each_present_cpu(i) {
560 ret = arch_register_cpu(i);
561 if (ret)
562 pr_warn("register_cpu %d failed (%d)\n", i, ret);
563 }
564}
565
566#ifdef CONFIG_GENERIC_CPU_VULNERABILITIES
567static ssize_t cpu_show_not_affected(struct device *dev,
568 struct device_attribute *attr, char *buf)
569{
570 return sysfs_emit(buf, "Not affected\n");
571}
572
573#define CPU_SHOW_VULN_FALLBACK(func) \
574 ssize_t cpu_show_##func(struct device *, \
575 struct device_attribute *, char *) \
576 __attribute__((weak, alias("cpu_show_not_affected")))
577
578CPU_SHOW_VULN_FALLBACK(meltdown);
579CPU_SHOW_VULN_FALLBACK(spectre_v1);
580CPU_SHOW_VULN_FALLBACK(spectre_v2);
581CPU_SHOW_VULN_FALLBACK(spec_store_bypass);
582CPU_SHOW_VULN_FALLBACK(l1tf);
583CPU_SHOW_VULN_FALLBACK(mds);
584CPU_SHOW_VULN_FALLBACK(tsx_async_abort);
585CPU_SHOW_VULN_FALLBACK(itlb_multihit);
586CPU_SHOW_VULN_FALLBACK(srbds);
587CPU_SHOW_VULN_FALLBACK(mmio_stale_data);
588CPU_SHOW_VULN_FALLBACK(retbleed);
589CPU_SHOW_VULN_FALLBACK(spec_rstack_overflow);
590CPU_SHOW_VULN_FALLBACK(gds);
591
592static DEVICE_ATTR(meltdown, 0444, cpu_show_meltdown, NULL);
593static DEVICE_ATTR(spectre_v1, 0444, cpu_show_spectre_v1, NULL);
594static DEVICE_ATTR(spectre_v2, 0444, cpu_show_spectre_v2, NULL);
595static DEVICE_ATTR(spec_store_bypass, 0444, cpu_show_spec_store_bypass, NULL);
596static DEVICE_ATTR(l1tf, 0444, cpu_show_l1tf, NULL);
597static DEVICE_ATTR(mds, 0444, cpu_show_mds, NULL);
598static DEVICE_ATTR(tsx_async_abort, 0444, cpu_show_tsx_async_abort, NULL);
599static DEVICE_ATTR(itlb_multihit, 0444, cpu_show_itlb_multihit, NULL);
600static DEVICE_ATTR(srbds, 0444, cpu_show_srbds, NULL);
601static DEVICE_ATTR(mmio_stale_data, 0444, cpu_show_mmio_stale_data, NULL);
602static DEVICE_ATTR(retbleed, 0444, cpu_show_retbleed, NULL);
603static DEVICE_ATTR(spec_rstack_overflow, 0444, cpu_show_spec_rstack_overflow, NULL);
604static DEVICE_ATTR(gather_data_sampling, 0444, cpu_show_gds, NULL);
605
606static struct attribute *cpu_root_vulnerabilities_attrs[] = {
607 &dev_attr_meltdown.attr,
608 &dev_attr_spectre_v1.attr,
609 &dev_attr_spectre_v2.attr,
610 &dev_attr_spec_store_bypass.attr,
611 &dev_attr_l1tf.attr,
612 &dev_attr_mds.attr,
613 &dev_attr_tsx_async_abort.attr,
614 &dev_attr_itlb_multihit.attr,
615 &dev_attr_srbds.attr,
616 &dev_attr_mmio_stale_data.attr,
617 &dev_attr_retbleed.attr,
618 &dev_attr_spec_rstack_overflow.attr,
619 &dev_attr_gather_data_sampling.attr,
620 NULL
621};
622
623static const struct attribute_group cpu_root_vulnerabilities_group = {
624 .name = "vulnerabilities",
625 .attrs = cpu_root_vulnerabilities_attrs,
626};
627
628static void __init cpu_register_vulnerabilities(void)
629{
630 struct device *dev = bus_get_dev_root(&cpu_subsys);
631
632 if (dev) {
633 if (sysfs_create_group(&dev->kobj, &cpu_root_vulnerabilities_group))
634 pr_err("Unable to register CPU vulnerabilities\n");
635 put_device(dev);
636 }
637}
638
639#else
640static inline void cpu_register_vulnerabilities(void) { }
641#endif
642
643void __init cpu_dev_init(void)
644{
645 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups))
646 panic("Failed to register CPU subsystem");
647
648 cpu_dev_register_generic();
649 cpu_register_vulnerabilities();
650}
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#include <linux/cpufeature.h>
19#include <linux/tick.h>
20
21#include "base.h"
22
23static DEFINE_PER_CPU(struct device *, cpu_sys_devices);
24
25static 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
34#ifdef CONFIG_HOTPLUG_CPU
35static 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
44static int cpu_subsys_online(struct device *dev)
45{
46 struct cpu *cpu = container_of(dev, struct cpu, dev);
47 int cpuid = dev->id;
48 int from_nid, to_nid;
49 int ret;
50
51 from_nid = cpu_to_node(cpuid);
52 if (from_nid == NUMA_NO_NODE)
53 return -ENODEV;
54
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);
63
64 return ret;
65}
66
67static int cpu_subsys_offline(struct device *dev)
68{
69 return cpu_down(dev->id);
70}
71
72void unregister_cpu(struct cpu *cpu)
73{
74 int logical_cpu = cpu->dev.id;
75
76 unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu));
77
78 device_unregister(&cpu->dev);
79 per_cpu(cpu_sys_devices, logical_cpu) = NULL;
80 return;
81}
82
83#ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
84static ssize_t cpu_probe_store(struct device *dev,
85 struct device_attribute *attr,
86 const char *buf,
87 size_t count)
88{
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;
100}
101
102static ssize_t cpu_release_store(struct device *dev,
103 struct device_attribute *attr,
104 const char *buf,
105 size_t count)
106{
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;
118}
119
120static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store);
121static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store);
122#endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
123#endif /* CONFIG_HOTPLUG_CPU */
124
125struct bus_type cpu_subsys = {
126 .name = "cpu",
127 .dev_name = "cpu",
128 .match = cpu_subsys_match,
129#ifdef CONFIG_HOTPLUG_CPU
130 .online = cpu_subsys_online,
131 .offline = cpu_subsys_offline,
132#endif
133};
134EXPORT_SYMBOL_GPL(cpu_subsys);
135
136#ifdef CONFIG_KEXEC
137#include <linux/kexec.h>
138
139static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr,
140 char *buf)
141{
142 struct cpu *cpu = container_of(dev, struct cpu, dev);
143 ssize_t rc;
144 unsigned long long addr;
145 int cpunum;
146
147 cpunum = cpu->dev.id;
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 */
155 addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum));
156 rc = sprintf(buf, "%Lx\n", addr);
157 return rc;
158}
159static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL);
160
161static ssize_t show_crash_notes_size(struct device *dev,
162 struct device_attribute *attr,
163 char *buf)
164{
165 ssize_t rc;
166
167 rc = sprintf(buf, "%zu\n", sizeof(note_buf_t));
168 return rc;
169}
170static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL);
171
172static struct attribute *crash_note_cpu_attrs[] = {
173 &dev_attr_crash_notes.attr,
174 &dev_attr_crash_notes_size.attr,
175 NULL
176};
177
178static struct attribute_group crash_note_cpu_attr_group = {
179 .attrs = crash_note_cpu_attrs,
180};
181#endif
182
183static const struct attribute_group *common_cpu_attr_groups[] = {
184#ifdef CONFIG_KEXEC
185 &crash_note_cpu_attr_group,
186#endif
187 NULL
188};
189
190static const struct attribute_group *hotplugable_cpu_attr_groups[] = {
191#ifdef CONFIG_KEXEC
192 &crash_note_cpu_attr_group,
193#endif
194 NULL
195};
196
197/*
198 * Print cpu online, possible, present, and system maps
199 */
200
201struct cpu_attr {
202 struct device_attribute attr;
203 const struct cpumask *const map;
204};
205
206static ssize_t show_cpus_attr(struct device *dev,
207 struct device_attribute *attr,
208 char *buf)
209{
210 struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr);
211
212 return cpumap_print_to_pagebuf(true, buf, ca->map);
213}
214
215#define _CPU_ATTR(name, map) \
216 { __ATTR(name, 0444, show_cpus_attr, NULL), map }
217
218/* Keep in sync with cpu_subsys_attrs */
219static struct cpu_attr cpu_attrs[] = {
220 _CPU_ATTR(online, &__cpu_online_mask),
221 _CPU_ATTR(possible, &__cpu_possible_mask),
222 _CPU_ATTR(present, &__cpu_present_mask),
223};
224
225/*
226 * Print values for NR_CPUS and offlined cpus
227 */
228static ssize_t print_cpus_kernel_max(struct device *dev,
229 struct device_attribute *attr, char *buf)
230{
231 int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1);
232 return n;
233}
234static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL);
235
236/* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */
237unsigned int total_cpus;
238
239static ssize_t print_cpus_offline(struct device *dev,
240 struct device_attribute *attr, char *buf)
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;
248 cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask);
249 n = scnprintf(buf, len, "%*pbl", cpumask_pr_args(offline));
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}
267static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL);
268
269static 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}
278static DEVICE_ATTR(isolated, 0444, print_cpus_isolated, NULL);
279
280#ifdef CONFIG_NO_HZ_FULL
281static 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}
290static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL);
291#endif
292
293static 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
308 * on the linux-kernel list, you have been warned.
309 */
310}
311
312#ifdef CONFIG_GENERIC_CPU_AUTOPROBE
313static 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}
334
335static 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
347/*
348 * register_cpu - Setup a sysfs device for a CPU.
349 * @cpu - cpu->hotpluggable field set to 1 will generate a control file in
350 * sysfs for this CPU.
351 * @num - CPU number to use when creating the device.
352 *
353 * Initialize and register the CPU device.
354 */
355int register_cpu(struct cpu *cpu, int num)
356{
357 int error;
358
359 cpu->node_id = cpu_to_node(num);
360 memset(&cpu->dev, 0x00, sizeof(struct device));
361 cpu->dev.id = num;
362 cpu->dev.bus = &cpu_subsys;
363 cpu->dev.release = cpu_device_release;
364 cpu->dev.offline_disabled = !cpu->hotpluggable;
365 cpu->dev.offline = !cpu_online(num);
366 cpu->dev.of_node = of_get_cpu_node(num, NULL);
367#ifdef CONFIG_GENERIC_CPU_AUTOPROBE
368 cpu->dev.bus->uevent = cpu_uevent;
369#endif
370 cpu->dev.groups = common_cpu_attr_groups;
371 if (cpu->hotpluggable)
372 cpu->dev.groups = hotplugable_cpu_attr_groups;
373 error = device_register(&cpu->dev);
374 if (!error)
375 per_cpu(cpu_sys_devices, num) = &cpu->dev;
376 if (!error)
377 register_cpu_under_node(num, cpu_to_node(num));
378
379 return error;
380}
381
382struct device *get_cpu_device(unsigned cpu)
383{
384 if (cpu < nr_cpu_ids && cpu_possible(cpu))
385 return per_cpu(cpu_sys_devices, cpu);
386 else
387 return NULL;
388}
389EXPORT_SYMBOL_GPL(get_cpu_device);
390
391static void device_create_release(struct device *dev)
392{
393 kfree(dev);
394}
395
396static struct device *
397__cpu_device_create(struct device *parent, void *drvdata,
398 const struct attribute_group **groups,
399 const char *fmt, va_list args)
400{
401 struct device *dev = NULL;
402 int retval = -ENODEV;
403
404 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
405 if (!dev) {
406 retval = -ENOMEM;
407 goto error;
408 }
409
410 device_initialize(dev);
411 dev->parent = parent;
412 dev->groups = groups;
413 dev->release = device_create_release;
414 dev_set_drvdata(dev, drvdata);
415
416 retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
417 if (retval)
418 goto error;
419
420 retval = device_add(dev);
421 if (retval)
422 goto error;
423
424 return dev;
425
426error:
427 put_device(dev);
428 return ERR_PTR(retval);
429}
430
431struct device *cpu_device_create(struct device *parent, void *drvdata,
432 const struct attribute_group **groups,
433 const char *fmt, ...)
434{
435 va_list vargs;
436 struct device *dev;
437
438 va_start(vargs, fmt);
439 dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs);
440 va_end(vargs);
441 return dev;
442}
443EXPORT_SYMBOL_GPL(cpu_device_create);
444
445#ifdef CONFIG_GENERIC_CPU_AUTOPROBE
446static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL);
447#endif
448
449static struct attribute *cpu_root_attrs[] = {
450#ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
451 &dev_attr_probe.attr,
452 &dev_attr_release.attr,
453#endif
454 &cpu_attrs[0].attr.attr,
455 &cpu_attrs[1].attr.attr,
456 &cpu_attrs[2].attr.attr,
457 &dev_attr_kernel_max.attr,
458 &dev_attr_offline.attr,
459 &dev_attr_isolated.attr,
460#ifdef CONFIG_NO_HZ_FULL
461 &dev_attr_nohz_full.attr,
462#endif
463#ifdef CONFIG_GENERIC_CPU_AUTOPROBE
464 &dev_attr_modalias.attr,
465#endif
466 NULL
467};
468
469static struct attribute_group cpu_root_attr_group = {
470 .attrs = cpu_root_attrs,
471};
472
473static const struct attribute_group *cpu_root_attr_groups[] = {
474 &cpu_root_attr_group,
475 NULL,
476};
477
478bool cpu_is_hotpluggable(unsigned cpu)
479{
480 struct device *dev = get_cpu_device(cpu);
481 return dev && container_of(dev, struct cpu, dev)->hotpluggable;
482}
483EXPORT_SYMBOL_GPL(cpu_is_hotpluggable);
484
485#ifdef CONFIG_GENERIC_CPU_DEVICES
486static DEFINE_PER_CPU(struct cpu, cpu_devices);
487#endif
488
489static void __init cpu_dev_register_generic(void)
490{
491#ifdef CONFIG_GENERIC_CPU_DEVICES
492 int i;
493
494 for_each_possible_cpu(i) {
495 if (register_cpu(&per_cpu(cpu_devices, i), i))
496 panic("Failed to register CPU device");
497 }
498#endif
499}
500
501void __init cpu_dev_init(void)
502{
503 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups))
504 panic("Failed to register CPU subsystem");
505
506 cpu_dev_register_generic();
507}