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 #ifdef CONFIG_HOTPLUG_CPU
185 static ssize_t isolate_show(struct device *dev,
186 struct device_attribute *attr, char *buf)
188 struct cpu *cpu = container_of(dev, struct cpu, dev);
190 int cpuid = cpu->dev.id;
191 unsigned int isolated = cpu_isolated(cpuid);
193 rc = snprintf(buf, PAGE_SIZE-2, "%d\n", isolated);
198 static DEVICE_ATTR_RO(isolate);
200 static struct attribute *cpu_isolated_attrs[] = {
201 &dev_attr_isolate.attr,
205 static struct attribute_group cpu_isolated_attr_group = {
206 .attrs = cpu_isolated_attrs,
211 #ifdef CONFIG_SCHED_HMP
213 static ssize_t show_sched_static_cpu_pwr_cost(struct device *dev,
214 struct device_attribute *attr, char *buf)
216 struct cpu *cpu = container_of(dev, struct cpu, dev);
218 int cpuid = cpu->dev.id;
219 unsigned int pwr_cost;
221 pwr_cost = sched_get_static_cpu_pwr_cost(cpuid);
223 rc = snprintf(buf, PAGE_SIZE-2, "%d\n", pwr_cost);
228 static ssize_t __ref store_sched_static_cpu_pwr_cost(struct device *dev,
229 struct device_attribute *attr,
230 const char *buf, size_t count)
232 struct cpu *cpu = container_of(dev, struct cpu, dev);
234 int cpuid = cpu->dev.id;
235 unsigned int pwr_cost;
237 err = kstrtouint(strstrip((char *)buf), 0, &pwr_cost);
241 err = sched_set_static_cpu_pwr_cost(cpuid, pwr_cost);
249 static ssize_t show_sched_static_cluster_pwr_cost(struct device *dev,
250 struct device_attribute *attr, char *buf)
252 struct cpu *cpu = container_of(dev, struct cpu, dev);
254 int cpuid = cpu->dev.id;
255 unsigned int pwr_cost;
257 pwr_cost = sched_get_static_cluster_pwr_cost(cpuid);
259 rc = snprintf(buf, PAGE_SIZE-2, "%d\n", pwr_cost);
264 static ssize_t __ref store_sched_static_cluster_pwr_cost(struct device *dev,
265 struct device_attribute *attr,
266 const char *buf, size_t count)
268 struct cpu *cpu = container_of(dev, struct cpu, dev);
270 int cpuid = cpu->dev.id;
271 unsigned int pwr_cost;
273 err = kstrtouint(strstrip((char *)buf), 0, &pwr_cost);
277 err = sched_set_static_cluster_pwr_cost(cpuid, pwr_cost);
285 static ssize_t show_sched_cluser_wake_idle(struct device *dev,
286 struct device_attribute *attr, char *buf)
288 struct cpu *cpu = container_of(dev, struct cpu, dev);
290 int cpuid = cpu->dev.id;
291 unsigned int wake_up_idle;
293 wake_up_idle = sched_get_cluster_wake_idle(cpuid);
295 rc = scnprintf(buf, PAGE_SIZE-2, "%d\n", wake_up_idle);
300 static ssize_t __ref store_sched_cluster_wake_idle(struct device *dev,
301 struct device_attribute *attr,
302 const char *buf, size_t count)
304 struct cpu *cpu = container_of(dev, struct cpu, dev);
306 int cpuid = cpu->dev.id;
307 unsigned int wake_up_idle;
309 err = kstrtouint(strstrip((char *)buf), 0, &wake_up_idle);
313 err = sched_set_cluster_wake_idle(cpuid, wake_up_idle);
321 static DEVICE_ATTR(sched_static_cpu_pwr_cost, 0644,
322 show_sched_static_cpu_pwr_cost,
323 store_sched_static_cpu_pwr_cost);
324 static DEVICE_ATTR(sched_static_cluster_pwr_cost, 0644,
325 show_sched_static_cluster_pwr_cost,
326 store_sched_static_cluster_pwr_cost);
327 static DEVICE_ATTR(sched_cluster_wake_up_idle, 0644,
328 show_sched_cluser_wake_idle,
329 store_sched_cluster_wake_idle);
331 static struct attribute *hmp_sched_cpu_attrs[] = {
332 &dev_attr_sched_static_cpu_pwr_cost.attr,
333 &dev_attr_sched_static_cluster_pwr_cost.attr,
334 &dev_attr_sched_cluster_wake_up_idle.attr,
338 static struct attribute_group sched_hmp_cpu_attr_group = {
339 .attrs = hmp_sched_cpu_attrs,
342 #endif /* CONFIG_SCHED_HMP */
344 static const struct attribute_group *common_cpu_attr_groups[] = {
346 &crash_note_cpu_attr_group,
348 #ifdef CONFIG_SCHED_HMP
349 &sched_hmp_cpu_attr_group,
351 #ifdef CONFIG_HOTPLUG_CPU
352 &cpu_isolated_attr_group,
357 static const struct attribute_group *hotplugable_cpu_attr_groups[] = {
359 &crash_note_cpu_attr_group,
361 #ifdef CONFIG_SCHED_HMP
362 &sched_hmp_cpu_attr_group,
364 #ifdef CONFIG_HOTPLUG_CPU
365 &cpu_isolated_attr_group,
371 * Print cpu online, possible, present, and system maps
375 struct device_attribute attr;
376 const struct cpumask *const * const map;
379 static ssize_t show_cpus_attr(struct device *dev,
380 struct device_attribute *attr,
383 struct cpu_attr *ca = container_of(attr, struct cpu_attr, attr);
385 return cpumap_print_to_pagebuf(true, buf, *ca->map);
388 #define _CPU_ATTR(name, map) \
389 { __ATTR(name, 0444, show_cpus_attr, NULL), map }
391 /* Keep in sync with cpu_subsys_attrs */
392 static struct cpu_attr cpu_attrs[] = {
393 _CPU_ATTR(online, &cpu_online_mask),
394 _CPU_ATTR(possible, &cpu_possible_mask),
395 _CPU_ATTR(present, &cpu_present_mask),
396 _CPU_ATTR(core_ctl_isolated, &cpu_isolated_mask),
400 * Print values for NR_CPUS and offlined cpus
402 static ssize_t print_cpus_kernel_max(struct device *dev,
403 struct device_attribute *attr, char *buf)
405 int n = snprintf(buf, PAGE_SIZE-2, "%d\n", NR_CPUS - 1);
408 static DEVICE_ATTR(kernel_max, 0444, print_cpus_kernel_max, NULL);
410 /* arch-optional setting to enable display of offline cpus >= nr_cpu_ids */
411 unsigned int total_cpus;
413 static ssize_t print_cpus_offline(struct device *dev,
414 struct device_attribute *attr, char *buf)
416 int n = 0, len = PAGE_SIZE-2;
417 cpumask_var_t offline;
419 /* display offline cpus < nr_cpu_ids */
420 if (!alloc_cpumask_var(&offline, GFP_KERNEL))
422 cpumask_andnot(offline, cpu_possible_mask, cpu_online_mask);
423 n = scnprintf(buf, len, "%*pbl", cpumask_pr_args(offline));
424 free_cpumask_var(offline);
426 /* display offline cpus >= nr_cpu_ids */
427 if (total_cpus && nr_cpu_ids < total_cpus) {
431 if (nr_cpu_ids == total_cpus-1)
432 n += snprintf(&buf[n], len - n, "%d", nr_cpu_ids);
434 n += snprintf(&buf[n], len - n, "%d-%d",
435 nr_cpu_ids, total_cpus-1);
438 n += snprintf(&buf[n], len - n, "\n");
441 static DEVICE_ATTR(offline, 0444, print_cpus_offline, NULL);
443 static ssize_t print_cpus_isolated(struct device *dev,
444 struct device_attribute *attr, char *buf)
446 int n = 0, len = PAGE_SIZE-2;
448 n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(cpu_isolated_map));
452 static DEVICE_ATTR(isolated, 0444, print_cpus_isolated, NULL);
454 #ifdef CONFIG_NO_HZ_FULL
455 static ssize_t print_cpus_nohz_full(struct device *dev,
456 struct device_attribute *attr, char *buf)
458 int n = 0, len = PAGE_SIZE-2;
460 n = scnprintf(buf, len, "%*pbl\n", cpumask_pr_args(tick_nohz_full_mask));
464 static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL);
467 static void cpu_device_release(struct device *dev)
470 * This is an empty function to prevent the driver core from spitting a
471 * warning at us. Yes, I know this is directly opposite of what the
472 * documentation for the driver core and kobjects say, and the author
473 * of this code has already been publically ridiculed for doing
474 * something as foolish as this. However, at this point in time, it is
475 * the only way to handle the issue of statically allocated cpu
476 * devices. The different architectures will have their cpu device
477 * code reworked to properly handle this in the near future, so this
478 * function will then be changed to correctly free up the memory held
481 * Never copy this way of doing things, or you too will be made fun of
482 * on the linux-kernel list, you have been warned.
486 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
487 static ssize_t print_cpu_modalias(struct device *dev,
488 struct device_attribute *attr,
494 n = sprintf(buf, "cpu:type:" CPU_FEATURE_TYPEFMT ":feature:",
495 CPU_FEATURE_TYPEVAL);
497 for (i = 0; i < MAX_CPU_FEATURES; i++)
498 if (cpu_have_feature(i)) {
499 if (PAGE_SIZE < n + sizeof(",XXXX\n")) {
500 WARN(1, "CPU features overflow page\n");
503 n += sprintf(&buf[n], ",%04X", i);
509 static int cpu_uevent(struct device *dev, struct kobj_uevent_env *env)
511 char *buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
513 print_cpu_modalias(NULL, NULL, buf);
514 add_uevent_var(env, "MODALIAS=%s", buf);
522 * register_cpu - Setup a sysfs device for a CPU.
523 * @cpu - cpu->hotpluggable field set to 1 will generate a control file in
524 * sysfs for this CPU.
525 * @num - CPU number to use when creating the device.
527 * Initialize and register the CPU device.
529 int register_cpu(struct cpu *cpu, int num)
533 cpu->node_id = cpu_to_node(num);
534 memset(&cpu->dev, 0x00, sizeof(struct device));
536 cpu->dev.bus = &cpu_subsys;
537 cpu->dev.release = cpu_device_release;
538 cpu->dev.offline_disabled = !cpu->hotpluggable;
539 cpu->dev.offline = !cpu_online(num);
540 cpu->dev.of_node = of_get_cpu_node(num, NULL);
541 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
542 cpu->dev.bus->uevent = cpu_uevent;
544 cpu->dev.groups = common_cpu_attr_groups;
545 if (cpu->hotpluggable)
546 cpu->dev.groups = hotplugable_cpu_attr_groups;
547 error = device_register(&cpu->dev);
549 per_cpu(cpu_sys_devices, num) = &cpu->dev;
551 register_cpu_under_node(num, cpu_to_node(num));
556 struct device *get_cpu_device(unsigned cpu)
558 if (cpu < nr_cpu_ids && cpu_possible(cpu))
559 return per_cpu(cpu_sys_devices, cpu);
563 EXPORT_SYMBOL_GPL(get_cpu_device);
565 static void device_create_release(struct device *dev)
570 static struct device *
571 __cpu_device_create(struct device *parent, void *drvdata,
572 const struct attribute_group **groups,
573 const char *fmt, va_list args)
575 struct device *dev = NULL;
576 int retval = -ENODEV;
578 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
584 device_initialize(dev);
585 dev->parent = parent;
586 dev->groups = groups;
587 dev->release = device_create_release;
588 dev_set_drvdata(dev, drvdata);
590 retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
594 retval = device_add(dev);
602 return ERR_PTR(retval);
605 struct device *cpu_device_create(struct device *parent, void *drvdata,
606 const struct attribute_group **groups,
607 const char *fmt, ...)
612 va_start(vargs, fmt);
613 dev = __cpu_device_create(parent, drvdata, groups, fmt, vargs);
617 EXPORT_SYMBOL_GPL(cpu_device_create);
619 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
620 static DEVICE_ATTR(modalias, 0444, print_cpu_modalias, NULL);
623 static struct attribute *cpu_root_attrs[] = {
624 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
625 &dev_attr_probe.attr,
626 &dev_attr_release.attr,
628 &cpu_attrs[0].attr.attr,
629 &cpu_attrs[1].attr.attr,
630 &cpu_attrs[2].attr.attr,
631 &cpu_attrs[3].attr.attr,
632 &dev_attr_kernel_max.attr,
633 &dev_attr_offline.attr,
634 &dev_attr_isolated.attr,
635 #ifdef CONFIG_NO_HZ_FULL
636 &dev_attr_nohz_full.attr,
638 #ifdef CONFIG_GENERIC_CPU_AUTOPROBE
639 &dev_attr_modalias.attr,
644 static struct attribute_group cpu_root_attr_group = {
645 .attrs = cpu_root_attrs,
648 static const struct attribute_group *cpu_root_attr_groups[] = {
649 &cpu_root_attr_group,
653 bool cpu_is_hotpluggable(unsigned cpu)
655 struct device *dev = get_cpu_device(cpu);
656 return dev && container_of(dev, struct cpu, dev)->hotpluggable;
658 EXPORT_SYMBOL_GPL(cpu_is_hotpluggable);
660 #ifdef CONFIG_GENERIC_CPU_DEVICES
661 static DEFINE_PER_CPU(struct cpu, cpu_devices);
664 static void __init cpu_dev_register_generic(void)
666 #ifdef CONFIG_GENERIC_CPU_DEVICES
669 for_each_possible_cpu(i) {
670 if (register_cpu(&per_cpu(cpu_devices, i), i))
671 panic("Failed to register CPU device");
676 #ifdef CONFIG_GENERIC_CPU_VULNERABILITIES
678 ssize_t __weak cpu_show_meltdown(struct device *dev,
679 struct device_attribute *attr, char *buf)
681 return sprintf(buf, "Not affected\n");
684 ssize_t __weak cpu_show_spectre_v1(struct device *dev,
685 struct device_attribute *attr, char *buf)
687 return sprintf(buf, "Not affected\n");
690 ssize_t __weak cpu_show_spectre_v2(struct device *dev,
691 struct device_attribute *attr, char *buf)
693 return sprintf(buf, "Not affected\n");
696 ssize_t __weak cpu_show_spec_store_bypass(struct device *dev,
697 struct device_attribute *attr, char *buf)
699 return sprintf(buf, "Not affected\n");
702 ssize_t __weak cpu_show_l1tf(struct device *dev,
703 struct device_attribute *attr, char *buf)
705 return sprintf(buf, "Not affected\n");
708 ssize_t __weak cpu_show_mds(struct device *dev,
709 struct device_attribute *attr, char *buf)
711 return sprintf(buf, "Not affected\n");
714 static DEVICE_ATTR(meltdown, 0444, cpu_show_meltdown, NULL);
715 static DEVICE_ATTR(spectre_v1, 0444, cpu_show_spectre_v1, NULL);
716 static DEVICE_ATTR(spectre_v2, 0444, cpu_show_spectre_v2, NULL);
717 static DEVICE_ATTR(spec_store_bypass, 0444, cpu_show_spec_store_bypass, NULL);
718 static DEVICE_ATTR(l1tf, 0444, cpu_show_l1tf, NULL);
719 static DEVICE_ATTR(mds, 0444, cpu_show_mds, NULL);
721 static struct attribute *cpu_root_vulnerabilities_attrs[] = {
722 &dev_attr_meltdown.attr,
723 &dev_attr_spectre_v1.attr,
724 &dev_attr_spectre_v2.attr,
725 &dev_attr_spec_store_bypass.attr,
731 static const struct attribute_group cpu_root_vulnerabilities_group = {
732 .name = "vulnerabilities",
733 .attrs = cpu_root_vulnerabilities_attrs,
736 static void __init cpu_register_vulnerabilities(void)
738 if (sysfs_create_group(&cpu_subsys.dev_root->kobj,
739 &cpu_root_vulnerabilities_group))
740 pr_err("Unable to register CPU vulnerabilities\n");
744 static inline void cpu_register_vulnerabilities(void) { }
747 void __init cpu_dev_init(void)
749 if (subsys_system_register(&cpu_subsys, cpu_root_attr_groups))
750 panic("Failed to register CPU subsystem");
752 cpu_dev_register_generic();
753 cpu_register_vulnerabilities();