2 * CPU subsystem support
5 #include <linux/kernel.h>
6 #include <linux/module.h>
7 #include <linux/init.h>
8 #include <linux/sched.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>
18 #include <linux/cpufeature.h>
19 #include <linux/tick.h>
23 static DEFINE_PER_CPU(struct device *, cpu_sys_devices);
25 static int cpu_subsys_match(struct device *dev, struct device_driver *drv)
27 /* ACPI style match is the only one that may succeed. */
28 if (acpi_driver_match_device(dev, drv))
34 #ifdef CONFIG_HOTPLUG_CPU
35 static void change_cpu_under_node(struct cpu *cpu,
36 unsigned int from_nid, unsigned int to_nid)
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;
44 static int cpu_subsys_online(struct device *dev)
46 struct cpu *cpu = container_of(dev, struct cpu, dev);
51 from_nid = cpu_to_node(cpuid);
52 if (from_nid == NUMA_NO_NODE)
57 * When hot adding memory to memoryless node and enabling a cpu
58 * on the node, node number of the cpu may internally change.
60 to_nid = cpu_to_node(cpuid);
61 if (from_nid != to_nid)
62 change_cpu_under_node(cpu, from_nid, to_nid);
67 static int cpu_subsys_offline(struct device *dev)
69 return cpu_down(dev->id);
72 void unregister_cpu(struct cpu *cpu)
74 int logical_cpu = cpu->dev.id;
76 unregister_cpu_under_node(logical_cpu, cpu_to_node(logical_cpu));
78 device_unregister(&cpu->dev);
79 per_cpu(cpu_sys_devices, logical_cpu) = NULL;
83 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
84 static ssize_t cpu_probe_store(struct device *dev,
85 struct device_attribute *attr,
92 ret = lock_device_hotplug_sysfs();
96 cnt = arch_cpu_probe(buf, count);
98 unlock_device_hotplug();
102 static ssize_t cpu_release_store(struct device *dev,
103 struct device_attribute *attr,
110 ret = lock_device_hotplug_sysfs();
114 cnt = arch_cpu_release(buf, count);
116 unlock_device_hotplug();
120 static DEVICE_ATTR(probe, S_IWUSR, NULL, cpu_probe_store);
121 static DEVICE_ATTR(release, S_IWUSR, NULL, cpu_release_store);
122 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
123 #endif /* CONFIG_HOTPLUG_CPU */
125 struct bus_type cpu_subsys = {
128 .match = cpu_subsys_match,
129 #ifdef CONFIG_HOTPLUG_CPU
130 .online = cpu_subsys_online,
131 .offline = cpu_subsys_offline,
134 EXPORT_SYMBOL_GPL(cpu_subsys);
137 #include <linux/kexec.h>
139 static ssize_t show_crash_notes(struct device *dev, struct device_attribute *attr,
142 struct cpu *cpu = container_of(dev, struct cpu, dev);
144 unsigned long long addr;
147 cpunum = cpu->dev.id;
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.
155 addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpunum));
156 rc = sprintf(buf, "%Lx\n", addr);
159 static DEVICE_ATTR(crash_notes, 0400, show_crash_notes, NULL);
161 static ssize_t show_crash_notes_size(struct device *dev,
162 struct device_attribute *attr,
167 rc = sprintf(buf, "%zu\n", sizeof(note_buf_t));
170 static DEVICE_ATTR(crash_notes_size, 0400, show_crash_notes_size, NULL);
172 static struct attribute *crash_note_cpu_attrs[] = {
173 &dev_attr_crash_notes.attr,
174 &dev_attr_crash_notes_size.attr,
178 static struct attribute_group crash_note_cpu_attr_group = {
179 .attrs = crash_note_cpu_attrs,
183 static const struct attribute_group *common_cpu_attr_groups[] = {
185 &crash_note_cpu_attr_group,
190 static const struct attribute_group *hotplugable_cpu_attr_groups[] = {
192 &crash_note_cpu_attr_group,
198 * Print cpu online, possible, present, and system maps
202 struct device_attribute attr;
203 const struct cpumask *const map;
206 static ssize_t show_cpus_attr(struct device *dev,
207 struct device_attribute *attr,
210 struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr);
212 return cpumap_print_to_pagebuf(true, buf, ca->map);
215 #define _CPU_ATTR(name, map) \
216 { __ATTR(name, 0444, show_cpus_attr, NULL), map }
218 /* Keep in sync with cpu_subsys_attrs */
219 static 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),
226 * Print values for NR_CPUS and offlined cpus
228 static ssize_t print_cpus_kernel_max(struct device *dev,
229 struct device_attribute *attr, char *buf)
231 int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1);
234 static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL);
236 /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */
237 unsigned int total_cpus;
239 static ssize_t print_cpus_offline(struct device *dev,
240 struct device_attribute *attr, char *buf)
242 int n = 0, len = PAGE_SIZE-2;
243 cpumask_var_t offline;
245 /* display offline cpus < nr_cpu_ids */
246 if (!alloc_cpumask_var(&offline, GFP_KERNEL))
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);
252 /* display offline cpus >= nr_cpu_ids */
253 if (total_cpus && nr_cpu_ids < total_cpus) {
257 if (nr_cpu_ids == total_cpus-1)
258 n += snprintf(&buf[n], len - n, "%d", nr_cpu_ids);
260 n += snprintf(&buf[n], len - n, "%d-%d",
261 nr_cpu_ids, total_cpus-1);
264 n += snprintf(&buf[n], len - n, "\n");
267 static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL);
269 static ssize_t print_cpus_isolated(struct device *dev,
270 struct device_attribute *attr, char *buf)
272 int n = 0, len = PAGE_SIZE-2;
274 n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(cpu_isolated_map));
278 static DEVICE_ATTR(isolated, 0444, print_cpus_isolated, NULL);
280 #ifdef CONFIG_NO_HZ_FULL
281 static ssize_t print_cpus_nohz_full(struct device *dev,
282 struct device_attribute *attr, char *buf)
284 int n = 0, len = PAGE_SIZE-2;
286 n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(tick_nohz_full_mask));
290 static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL);
293 static void cpu_device_release(struct device *dev)
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
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.
312 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
313 static ssize_t print_cpu_modalias(struct device *dev,
314 struct device_attribute *attr,
320 n = sprintf(buf, "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:",
321 CPU_FEATURE_TYPEVAL);
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");
329 n += sprintf(&buf[n], ",%04X", i);
335 static int cpu_uevent(struct device *dev, struct kobj_uevent_env *env)
337 char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
339 print_cpu_modalias(NULL, NULL, buf);
340 add_uevent_var(env, "MODALIAS=%s", buf);
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.
353 * Initialize and register the CPU device.
355 int register_cpu(struct cpu *cpu, int num)
359 cpu->node_id = cpu_to_node(num);
360 memset(&cpu->dev, 0x00, sizeof(struct device));
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;
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);
377 per_cpu(cpu_sys_devices, num) = &cpu->dev;
378 register_cpu_under_node(num, cpu_to_node(num));
383 struct device *get_cpu_device(unsigned cpu)
385 if (cpu < nr_cpu_ids && cpu_possible(cpu))
386 return per_cpu(cpu_sys_devices, cpu);
390 EXPORT_SYMBOL_GPL(get_cpu_device);
392 static void device_create_release(struct device *dev)
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)
402 struct device *dev = NULL;
403 int retval = -ENODEV;
405 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
411 device_initialize(dev);
412 dev->parent = parent;
413 dev->groups = groups;
414 dev->release = device_create_release;
415 dev_set_drvdata(dev, drvdata);
417 retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
421 retval = device_add(dev);
429 return ERR_PTR(retval);
432 struct device *cpu_device_create(struct device *parent, void *drvdata,
433 const struct attribute_group **groups,
434 const char *fmt, ...)
439 va_start(vargs, fmt);
440 dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs);
444 EXPORT_SYMBOL_GPL(cpu_device_create);
446 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
447 static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL);
450 static struct attribute *cpu_root_attrs[] = {
451 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
452 &dev_attr_probe.attr,
453 &dev_attr_release.attr,
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,
460 &dev_attr_isolated.attr,
461 #ifdef CONFIG_NO_HZ_FULL
462 &dev_attr_nohz_full.attr,
464 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
465 &dev_attr_modalias.attr,
470 static struct attribute_group cpu_root_attr_group = {
471 .attrs = cpu_root_attrs,
474 static const struct attribute_group *cpu_root_attr_groups[] = {
475 &cpu_root_attr_group,
479 bool cpu_is_hotpluggable(unsigned cpu)
481 struct device *dev = get_cpu_device(cpu);
482 return dev && container_of(dev, struct cpu, dev)->hotpluggable;
484 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable);
486 #ifdef CONFIG_GENERIC_CPU_DEVICES
487 static DEFINE_PER_CPU(struct cpu, cpu_devices);
490 static void __init cpu_dev_register_generic(void)
492 #ifdef CONFIG_GENERIC_CPU_DEVICES
495 for_each_possible_cpu(i) {
496 if (register_cpu(&per_cpu(cpu_devices, i), i))
497 panic("Failed to register CPU device");
502 #ifdef CONFIG_GENERIC_CPU_VULNERABILITIES
504 ssize_t __weak cpu_show_meltdown(struct device *dev,
505 struct device_attribute *attr, char *buf)
507 return sprintf(buf, "Not affected\n");
510 ssize_t __weak cpu_show_spectre_v1(struct device *dev,
511 struct device_attribute *attr, char *buf)
513 return sprintf(buf, "Not affected\n");
516 ssize_t __weak cpu_show_spectre_v2(struct device *dev,
517 struct device_attribute *attr, char *buf)
519 return sprintf(buf, "Not affected\n");
522 ssize_t __weak cpu_show_spec_store_bypass(struct device *dev,
523 struct device_attribute *attr, char *buf)
525 return sprintf(buf, "Not affected\n");
528 ssize_t __weak cpu_show_l1tf(struct device *dev,
529 struct device_attribute *attr, char *buf)
531 return sprintf(buf, "Not affected\n");
534 ssize_t __weak cpu_show_mds(struct device *dev,
535 struct device_attribute *attr, char *buf)
537 return sprintf(buf, "Not affected\n");
540 ssize_t __weak cpu_show_tsx_async_abort(struct device *dev,
541 struct device_attribute *attr,
544 return sprintf(buf, "Not affected\n");
547 ssize_t __weak cpu_show_itlb_multihit(struct device *dev,
548 struct device_attribute *attr, char *buf)
550 return sprintf(buf, "Not affected\n");
553 ssize_t __weak cpu_show_srbds(struct device *dev,
554 struct device_attribute *attr, char *buf)
556 return sprintf(buf, "Not affected\n");
559 static DEVICE_ATTR(meltdown, 0444, cpu_show_meltdown, NULL);
560 static DEVICE_ATTR(spectre_v1, 0444, cpu_show_spectre_v1, NULL);
561 static DEVICE_ATTR(spectre_v2, 0444, cpu_show_spectre_v2, NULL);
562 static DEVICE_ATTR(spec_store_bypass, 0444, cpu_show_spec_store_bypass, NULL);
563 static DEVICE_ATTR(l1tf, 0444, cpu_show_l1tf, NULL);
564 static DEVICE_ATTR(mds, 0444, cpu_show_mds, NULL);
565 static DEVICE_ATTR(tsx_async_abort, 0444, cpu_show_tsx_async_abort, NULL);
566 static DEVICE_ATTR(itlb_multihit, 0444, cpu_show_itlb_multihit, NULL);
567 static DEVICE_ATTR(srbds, 0444, cpu_show_srbds, NULL);
569 static struct attribute *cpu_root_vulnerabilities_attrs[] = {
570 &dev_attr_meltdown.attr,
571 &dev_attr_spectre_v1.attr,
572 &dev_attr_spectre_v2.attr,
573 &dev_attr_spec_store_bypass.attr,
576 &dev_attr_tsx_async_abort.attr,
577 &dev_attr_itlb_multihit.attr,
578 &dev_attr_srbds.attr,
582 static const struct attribute_group cpu_root_vulnerabilities_group = {
583 .name = "vulnerabilities",
584 .attrs = cpu_root_vulnerabilities_attrs,
587 static void __init cpu_register_vulnerabilities(void)
589 if (sysfs_create_group(&cpu_subsys.dev_root->kobj,
590 &cpu_root_vulnerabilities_group))
591 pr_err("Unable to register CPU vulnerabilities\n");
595 static inline void cpu_register_vulnerabilities(void) { }
598 void __init cpu_dev_init(void)
600 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups))
601 panic("Failed to register CPU subsystem");
603 cpu_dev_register_generic();
604 cpu_register_vulnerabilities();