blob: 34e40ef2f6e0dc35d76ca76d497ccefb98aed2d3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* CPU control.
2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
3 *
4 * This code is licenced under the GPL.
5 */
6#include <linux/proc_fs.h>
7#include <linux/smp.h>
8#include <linux/init.h>
9#include <linux/notifier.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010010#include <linux/sched/signal.h>
Ingo Molnaref8bd772017-02-08 18:51:36 +010011#include <linux/sched/hotplug.h>
Ingo Molnar29930022017-02-08 18:51:36 +010012#include <linux/sched/task.h>
Thomas Gleixnera74cfff2018-11-25 19:33:39 +010013#include <linux/sched/smt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/unistd.h>
15#include <linux/cpu.h>
Anton Vorontsovcb792952012-05-31 16:26:22 -070016#include <linux/oom.h>
17#include <linux/rcupdate.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040018#include <linux/export.h>
Anton Vorontsove4cc2f82012-05-31 16:26:26 -070019#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/kthread.h>
21#include <linux/stop_machine.h>
Ingo Molnar81615b622006-06-26 00:24:32 -070022#include <linux/mutex.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/gfp.h>
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +010024#include <linux/suspend.h>
Gautham R. Shenoya19423b2014-03-11 02:04:03 +053025#include <linux/lockdep.h>
Preeti U Murthy345527b2015-03-30 14:59:19 +053026#include <linux/tick.h>
Thomas Gleixnera8994182015-07-05 17:12:30 +000027#include <linux/irq.h>
Thomas Gleixner941154b2017-09-12 21:37:04 +020028#include <linux/nmi.h>
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000029#include <linux/smpboot.h>
Richard Weinbergere6d49892016-08-18 14:57:17 +020030#include <linux/relay.h>
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +020031#include <linux/slab.h>
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +020032#include <linux/percpu-rwsem.h>
Thomas Gleixnercff7d372016-02-26 18:43:28 +000033
Todd E Brandtbb3632c2014-06-06 05:40:17 -070034#include <trace/events/power.h>
Thomas Gleixnercff7d372016-02-26 18:43:28 +000035#define CREATE_TRACE_POINTS
36#include <trace/events/cpuhp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Thomas Gleixner38498a62012-04-20 13:05:44 +000038#include "smpboot.h"
39
Thomas Gleixnercff7d372016-02-26 18:43:28 +000040/**
41 * cpuhp_cpu_state - Per cpu hotplug state storage
42 * @state: The current cpu state
43 * @target: The target state
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000044 * @thread: Pointer to the hotplug thread
45 * @should_run: Thread should execute
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +020046 * @rollback: Perform a rollback
Thomas Gleixnera7246322016-08-12 19:49:38 +020047 * @single: Single callback invocation
48 * @bringup: Single callback bringup or teardown selector
49 * @cb_state: The state for a single callback (install/uninstall)
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000050 * @result: Result of the operation
Peter Zijlstra5ebe7742017-09-20 19:00:19 +020051 * @done_up: Signal completion to the issuer of the task for cpu-up
52 * @done_down: Signal completion to the issuer of the task for cpu-down
Thomas Gleixnercff7d372016-02-26 18:43:28 +000053 */
54struct cpuhp_cpu_state {
55 enum cpuhp_state state;
56 enum cpuhp_state target;
Peter Zijlstra1db49482017-09-20 19:00:21 +020057 enum cpuhp_state fail;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000058#ifdef CONFIG_SMP
59 struct task_struct *thread;
60 bool should_run;
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +020061 bool rollback;
Thomas Gleixnera7246322016-08-12 19:49:38 +020062 bool single;
63 bool bringup;
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +020064 bool booted_once;
Thomas Gleixnercf392d12016-08-12 19:49:39 +020065 struct hlist_node *node;
Peter Zijlstra4dddfb52017-09-20 19:00:17 +020066 struct hlist_node *last;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000067 enum cpuhp_state cb_state;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000068 int result;
Peter Zijlstra5ebe7742017-09-20 19:00:19 +020069 struct completion done_up;
70 struct completion done_down;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000071#endif
Thomas Gleixnercff7d372016-02-26 18:43:28 +000072};
73
Peter Zijlstra1db49482017-09-20 19:00:21 +020074static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
75 .fail = CPUHP_INVALID,
76};
Thomas Gleixnercff7d372016-02-26 18:43:28 +000077
Thomas Gleixner49dfe2a2017-05-24 10:15:43 +020078#if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +020079static struct lockdep_map cpuhp_state_up_map =
80 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
81static struct lockdep_map cpuhp_state_down_map =
82 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
83
84
Mathieu Malaterre76dc6c02017-12-26 15:08:53 +010085static inline void cpuhp_lock_acquire(bool bringup)
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +020086{
87 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
88}
89
Mathieu Malaterre76dc6c02017-12-26 15:08:53 +010090static inline void cpuhp_lock_release(bool bringup)
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +020091{
92 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
93}
94#else
95
Mathieu Malaterre76dc6c02017-12-26 15:08:53 +010096static inline void cpuhp_lock_acquire(bool bringup) { }
97static inline void cpuhp_lock_release(bool bringup) { }
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +020098
Thomas Gleixner49dfe2a2017-05-24 10:15:43 +020099#endif
100
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000101/**
102 * cpuhp_step - Hotplug state machine step
103 * @name: Name of the step
104 * @startup: Startup function of the step
105 * @teardown: Teardown function of the step
Thomas Gleixner757c9892016-02-26 18:43:32 +0000106 * @cant_stop: Bringup/teardown can't be stopped at this step
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000107 */
108struct cpuhp_step {
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200109 const char *name;
110 union {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200111 int (*single)(unsigned int cpu);
112 int (*multi)(unsigned int cpu,
113 struct hlist_node *node);
114 } startup;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200115 union {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200116 int (*single)(unsigned int cpu);
117 int (*multi)(unsigned int cpu,
118 struct hlist_node *node);
119 } teardown;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200120 struct hlist_head list;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200121 bool cant_stop;
122 bool multi_instance;
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000123};
124
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +0000125static DEFINE_MUTEX(cpuhp_state_mutex);
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +0800126static struct cpuhp_step cpuhp_hp_states[];
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000127
Thomas Gleixnera7246322016-08-12 19:49:38 +0200128static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
129{
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +0800130 return cpuhp_hp_states + state;
Thomas Gleixnera7246322016-08-12 19:49:38 +0200131}
132
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000133/**
134 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
135 * @cpu: The cpu for which the callback should be invoked
Peter Zijlstra96abb962017-09-20 19:00:16 +0200136 * @state: The state to do callbacks for
Thomas Gleixnera7246322016-08-12 19:49:38 +0200137 * @bringup: True if the bringup callback should be invoked
Peter Zijlstra96abb962017-09-20 19:00:16 +0200138 * @node: For multi-instance, do a single entry callback for install/remove
139 * @lastp: For multi-instance rollback, remember how far we got
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000140 *
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200141 * Called from cpu hotplug and from the state register machinery.
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000142 */
Thomas Gleixnera7246322016-08-12 19:49:38 +0200143static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
Peter Zijlstra96abb962017-09-20 19:00:16 +0200144 bool bringup, struct hlist_node *node,
145 struct hlist_node **lastp)
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000146{
147 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Thomas Gleixnera7246322016-08-12 19:49:38 +0200148 struct cpuhp_step *step = cpuhp_get_step(state);
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200149 int (*cbm)(unsigned int cpu, struct hlist_node *node);
150 int (*cb)(unsigned int cpu);
151 int ret, cnt;
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000152
Peter Zijlstra1db49482017-09-20 19:00:21 +0200153 if (st->fail == state) {
154 st->fail = CPUHP_INVALID;
155
156 if (!(bringup ? step->startup.single : step->teardown.single))
157 return 0;
158
159 return -EAGAIN;
160 }
161
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200162 if (!step->multi_instance) {
Peter Zijlstra96abb962017-09-20 19:00:16 +0200163 WARN_ON_ONCE(lastp && *lastp);
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200164 cb = bringup ? step->startup.single : step->teardown.single;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200165 if (!cb)
166 return 0;
Thomas Gleixnera7246322016-08-12 19:49:38 +0200167 trace_cpuhp_enter(cpu, st->target, state, cb);
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000168 ret = cb(cpu);
Thomas Gleixnera7246322016-08-12 19:49:38 +0200169 trace_cpuhp_exit(cpu, st->state, state, ret);
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200170 return ret;
171 }
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200172 cbm = bringup ? step->startup.multi : step->teardown.multi;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200173 if (!cbm)
174 return 0;
175
176 /* Single invocation for instance add/remove */
177 if (node) {
Peter Zijlstra96abb962017-09-20 19:00:16 +0200178 WARN_ON_ONCE(lastp && *lastp);
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200179 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
180 ret = cbm(cpu, node);
181 trace_cpuhp_exit(cpu, st->state, state, ret);
182 return ret;
183 }
184
185 /* State transition. Invoke on all instances */
186 cnt = 0;
187 hlist_for_each(node, &step->list) {
Peter Zijlstra96abb962017-09-20 19:00:16 +0200188 if (lastp && node == *lastp)
189 break;
190
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200191 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
192 ret = cbm(cpu, node);
193 trace_cpuhp_exit(cpu, st->state, state, ret);
Peter Zijlstra96abb962017-09-20 19:00:16 +0200194 if (ret) {
195 if (!lastp)
196 goto err;
197
198 *lastp = node;
199 return ret;
200 }
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200201 cnt++;
202 }
Peter Zijlstra96abb962017-09-20 19:00:16 +0200203 if (lastp)
204 *lastp = NULL;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200205 return 0;
206err:
207 /* Rollback the instances if one failed */
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200208 cbm = !bringup ? step->startup.multi : step->teardown.multi;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200209 if (!cbm)
210 return ret;
211
212 hlist_for_each(node, &step->list) {
213 if (!cnt--)
214 break;
Peter Zijlstra724a8682017-09-20 19:00:18 +0200215
216 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
217 ret = cbm(cpu, node);
218 trace_cpuhp_exit(cpu, st->state, state, ret);
219 /*
220 * Rollback must not fail,
221 */
222 WARN_ON_ONCE(ret);
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000223 }
224 return ret;
225}
226
Rusty Russell98a79d62008-12-13 21:19:41 +1030227#ifdef CONFIG_SMP
Arnd Bergmannfcb30292018-03-15 16:38:04 +0100228static bool cpuhp_is_ap_state(enum cpuhp_state state)
229{
230 /*
231 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
232 * purposes as that state is handled explicitly in cpu_down.
233 */
234 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
235}
236
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200237static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
238{
239 struct completion *done = bringup ? &st->done_up : &st->done_down;
240 wait_for_completion(done);
241}
242
243static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
244{
245 struct completion *done = bringup ? &st->done_up : &st->done_down;
246 complete(done);
247}
248
249/*
250 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
251 */
252static bool cpuhp_is_atomic_state(enum cpuhp_state state)
253{
254 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
255}
256
Rusty Russellb3199c02008-12-30 09:05:14 +1030257/* Serializes the updates to cpu_online_mask, cpu_present_mask */
Linus Torvaldsaa953872006-07-23 12:12:16 -0700258static DEFINE_MUTEX(cpu_add_remove_lock);
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000259bool cpuhp_tasks_frozen;
260EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700262/*
Srivatsa S. Bhat93ae4f92014-03-11 02:04:14 +0530263 * The following two APIs (cpu_maps_update_begin/done) must be used when
264 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700265 */
266void cpu_maps_update_begin(void)
267{
268 mutex_lock(&cpu_add_remove_lock);
269}
270
271void cpu_maps_update_done(void)
272{
273 mutex_unlock(&cpu_add_remove_lock);
274}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200276/*
277 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700278 * Should always be manipulated under cpu_add_remove_lock
279 */
280static int cpu_hotplug_disabled;
281
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700282#ifdef CONFIG_HOTPLUG_CPU
283
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200284DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530285
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200286void cpus_read_lock(void)
Ashok Raja9d9baa2005-11-28 13:43:46 -0800287{
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200288 percpu_down_read(&cpu_hotplug_lock);
Ashok Raja9d9baa2005-11-28 13:43:46 -0800289}
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200290EXPORT_SYMBOL_GPL(cpus_read_lock);
Ashok Raj90d45d12005-11-08 21:34:24 -0800291
Waiman Long6f4ceee2018-07-24 14:26:04 -0400292int cpus_read_trylock(void)
293{
294 return percpu_down_read_trylock(&cpu_hotplug_lock);
295}
296EXPORT_SYMBOL_GPL(cpus_read_trylock);
297
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200298void cpus_read_unlock(void)
Ashok Raja9d9baa2005-11-28 13:43:46 -0800299{
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200300 percpu_up_read(&cpu_hotplug_lock);
Ashok Raja9d9baa2005-11-28 13:43:46 -0800301}
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200302EXPORT_SYMBOL_GPL(cpus_read_unlock);
Ashok Raja9d9baa2005-11-28 13:43:46 -0800303
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200304void cpus_write_lock(void)
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100305{
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200306 percpu_down_write(&cpu_hotplug_lock);
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100307}
308
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200309void cpus_write_unlock(void)
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100310{
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200311 percpu_up_write(&cpu_hotplug_lock);
312}
313
314void lockdep_assert_cpus_held(void)
315{
Valentin Schneiderce48c4572018-12-19 18:23:15 +0000316 /*
317 * We can't have hotplug operations before userspace starts running,
318 * and some init codepaths will knowingly not take the hotplug lock.
319 * This is all valid, so mute lockdep until it makes sense to report
320 * unheld locks.
321 */
322 if (system_state < SYSTEM_RUNNING)
323 return;
324
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200325 percpu_rwsem_assert_held(&cpu_hotplug_lock);
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100326}
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700327
Peter Zijlstracb921732018-09-11 11:51:27 +0200328static void lockdep_acquire_cpus_lock(void)
329{
330 rwsem_acquire(&cpu_hotplug_lock.rw_sem.dep_map, 0, 0, _THIS_IP_);
331}
332
333static void lockdep_release_cpus_lock(void)
334{
335 rwsem_release(&cpu_hotplug_lock.rw_sem.dep_map, 1, _THIS_IP_);
336}
337
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700338/*
339 * Wait for currently running CPU hotplug operations to complete (if any) and
340 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
341 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
342 * hotplug path before performing hotplug operations. So acquiring that lock
343 * guarantees mutual exclusion from any currently running hotplug operations.
344 */
345void cpu_hotplug_disable(void)
346{
347 cpu_maps_update_begin();
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -0700348 cpu_hotplug_disabled++;
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700349 cpu_maps_update_done();
350}
Vitaly Kuznetsov32145c42015-08-05 00:52:47 -0700351EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700352
Lianwei Wang01b41152016-06-09 23:43:28 -0700353static void __cpu_hotplug_enable(void)
354{
355 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
356 return;
357 cpu_hotplug_disabled--;
358}
359
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700360void cpu_hotplug_enable(void)
361{
362 cpu_maps_update_begin();
Lianwei Wang01b41152016-06-09 23:43:28 -0700363 __cpu_hotplug_enable();
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700364 cpu_maps_update_done();
365}
Vitaly Kuznetsov32145c42015-08-05 00:52:47 -0700366EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
Peter Zijlstracb921732018-09-11 11:51:27 +0200367
368#else
369
370static void lockdep_acquire_cpus_lock(void)
371{
372}
373
374static void lockdep_release_cpus_lock(void)
375{
376}
377
Toshi Kanib9d10be2013-08-12 09:45:53 -0600378#endif /* CONFIG_HOTPLUG_CPU */
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700379
Thomas Gleixnera74cfff2018-11-25 19:33:39 +0100380/*
381 * Architectures that need SMT-specific errata handling during SMT hotplug
382 * should override this.
383 */
384void __weak arch_smt_update(void) { }
385
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200386#ifdef CONFIG_HOTPLUG_SMT
387enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
Konrad Rzeszutek Wilk26acfb62018-06-20 11:29:53 -0400388EXPORT_SYMBOL_GPL(cpu_smt_control);
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200389
Thomas Gleixnerbc2d8d262018-08-07 08:19:57 +0200390static bool cpu_smt_available __read_mostly;
391
Jiri Kosina8e1b7062018-07-13 16:23:23 +0200392void __init cpu_smt_disable(bool force)
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200393{
Jiri Kosina8e1b7062018-07-13 16:23:23 +0200394 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
395 cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
396 return;
397
398 if (force) {
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200399 pr_info("SMT: Force disabled\n");
400 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
Jiri Kosina8e1b7062018-07-13 16:23:23 +0200401 } else {
Borislav Petkovd0e7d142018-10-04 19:22:27 +0200402 pr_info("SMT: disabled\n");
Jiri Kosina8e1b7062018-07-13 16:23:23 +0200403 cpu_smt_control = CPU_SMT_DISABLED;
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200404 }
Jiri Kosina8e1b7062018-07-13 16:23:23 +0200405}
406
Thomas Gleixnerfee0aed2018-07-13 16:23:24 +0200407/*
408 * The decision whether SMT is supported can only be done after the full
Thomas Gleixnerbc2d8d262018-08-07 08:19:57 +0200409 * CPU identification. Called from architecture code before non boot CPUs
410 * are brought up.
411 */
412void __init cpu_smt_check_topology_early(void)
413{
414 if (!topology_smt_supported())
415 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
416}
417
418/*
419 * If SMT was disabled by BIOS, detect it here, after the CPUs have been
420 * brought online. This ensures the smt/l1tf sysfs entries are consistent
421 * with reality. cpu_smt_available is set to true during the bringup of non
422 * boot CPUs when a SMT sibling is detected. Note, this may overwrite
423 * cpu_smt_control's previous setting.
Thomas Gleixnerfee0aed2018-07-13 16:23:24 +0200424 */
425void __init cpu_smt_check_topology(void)
426{
Thomas Gleixnerbc2d8d262018-08-07 08:19:57 +0200427 if (!cpu_smt_available)
Thomas Gleixnerfee0aed2018-07-13 16:23:24 +0200428 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
429}
430
Jiri Kosina8e1b7062018-07-13 16:23:23 +0200431static int __init smt_cmdline_disable(char *str)
432{
433 cpu_smt_disable(str && !strcmp(str, "force"));
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200434 return 0;
435}
436early_param("nosmt", smt_cmdline_disable);
437
438static inline bool cpu_smt_allowed(unsigned int cpu)
439{
Thomas Gleixnerbc2d8d262018-08-07 08:19:57 +0200440 if (topology_is_primary_thread(cpu))
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200441 return true;
442
Thomas Gleixnerbc2d8d262018-08-07 08:19:57 +0200443 /*
444 * If the CPU is not a 'primary' thread and the booted_once bit is
445 * set then the processor has SMT support. Store this information
446 * for the late check of SMT support in cpu_smt_check_topology().
447 */
448 if (per_cpu(cpuhp_state, cpu).booted_once)
449 cpu_smt_available = true;
450
451 if (cpu_smt_control == CPU_SMT_ENABLED)
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200452 return true;
453
454 /*
455 * On x86 it's required to boot all logical CPUs at least once so
456 * that the init code can get a chance to set CR4.MCE on each
457 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
458 * core will shutdown the machine.
459 */
460 return !per_cpu(cpuhp_state, cpu).booted_once;
461}
462#else
463static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
464#endif
465
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200466static inline enum cpuhp_state
467cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
468{
469 enum cpuhp_state prev_state = st->state;
470
471 st->rollback = false;
472 st->last = NULL;
473
474 st->target = target;
475 st->single = false;
476 st->bringup = st->state < target;
477
478 return prev_state;
479}
480
481static inline void
482cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
483{
484 st->rollback = true;
485
486 /*
487 * If we have st->last we need to undo partial multi_instance of this
488 * state first. Otherwise start undo at the previous state.
489 */
490 if (!st->last) {
491 if (st->bringup)
492 st->state--;
493 else
494 st->state++;
495 }
496
497 st->target = prev_state;
498 st->bringup = !st->bringup;
499}
500
501/* Regular hotplug invocation of the AP hotplug thread */
502static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
503{
504 if (!st->single && st->state == st->target)
505 return;
506
507 st->result = 0;
508 /*
509 * Make sure the above stores are visible before should_run becomes
510 * true. Paired with the mb() above in cpuhp_thread_fun()
511 */
512 smp_mb();
513 st->should_run = true;
514 wake_up_process(st->thread);
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200515 wait_for_ap_thread(st, st->bringup);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200516}
517
518static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
519{
520 enum cpuhp_state prev_state;
521 int ret;
522
523 prev_state = cpuhp_set_state(st, target);
524 __cpuhp_kick_ap(st);
525 if ((ret = st->result)) {
526 cpuhp_reset_state(st, prev_state);
527 __cpuhp_kick_ap(st);
528 }
529
530 return ret;
531}
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +0200532
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000533static int bringup_wait_for_ap(unsigned int cpu)
534{
535 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
536
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +0200537 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200538 wait_for_ap_thread(st, true);
Thomas Gleixnerdea1d0f2017-07-11 22:06:24 +0200539 if (WARN_ON_ONCE((!cpu_online(cpu))))
540 return -ECANCELED;
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +0200541
542 /* Unpark the stopper thread and the hotplug thread of the target cpu */
543 stop_machine_unpark(cpu);
544 kthread_unpark(st->thread);
545
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200546 /*
547 * SMT soft disabling on X86 requires to bring the CPU out of the
548 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
549 * CPU marked itself as booted_once in cpu_notify_starting() so the
550 * cpu_smt_allowed() check will now return false if this is not the
551 * primary sibling.
552 */
553 if (!cpu_smt_allowed(cpu))
554 return -ECANCELED;
555
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200556 if (st->target <= CPUHP_AP_ONLINE_IDLE)
557 return 0;
558
559 return cpuhp_kick_ap(st, st->target);
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000560}
561
Thomas Gleixnerba997462016-02-26 18:43:24 +0000562static int bringup_cpu(unsigned int cpu)
563{
564 struct task_struct *idle = idle_thread_get(cpu);
565 int ret;
566
Boris Ostrovskyaa877172016-08-03 13:22:28 -0400567 /*
568 * Some architectures have to walk the irq descriptors to
569 * setup the vector space for the cpu which comes online.
570 * Prevent irq alloc/free across the bringup.
571 */
572 irq_lock_sparse();
573
Thomas Gleixnerba997462016-02-26 18:43:24 +0000574 /* Arch-specific enabling code. */
575 ret = __cpu_up(cpu, idle);
Boris Ostrovskyaa877172016-08-03 13:22:28 -0400576 irq_unlock_sparse();
Thomas Gleixner530e9b72016-12-21 20:19:53 +0100577 if (ret)
Thomas Gleixnerba997462016-02-26 18:43:24 +0000578 return ret;
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +0200579 return bringup_wait_for_ap(cpu);
Thomas Gleixnerba997462016-02-26 18:43:24 +0000580}
581
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000582/*
583 * Hotplug state machine related functions
584 */
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000585
Thomas Gleixnera7246322016-08-12 19:49:38 +0200586static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000587{
Mukesh Ojha6fb86d92018-08-28 12:24:54 +0530588 for (st->state--; st->state > st->target; st->state--)
589 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000590}
591
592static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
Thomas Gleixnera7246322016-08-12 19:49:38 +0200593 enum cpuhp_state target)
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000594{
595 enum cpuhp_state prev_state = st->state;
596 int ret = 0;
597
598 while (st->state < target) {
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000599 st->state++;
Peter Zijlstra96abb962017-09-20 19:00:16 +0200600 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000601 if (ret) {
602 st->target = prev_state;
Thomas Gleixnera7246322016-08-12 19:49:38 +0200603 undo_cpu_up(cpu, st);
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000604 break;
605 }
606 }
607 return ret;
608}
609
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000610/*
611 * The cpu hotplug threads manage the bringup and teardown of the cpus
612 */
613static void cpuhp_create(unsigned int cpu)
614{
615 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
616
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200617 init_completion(&st->done_up);
618 init_completion(&st->done_down);
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000619}
620
621static int cpuhp_should_run(unsigned int cpu)
622{
623 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
624
625 return st->should_run;
626}
627
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000628/*
629 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
630 * callbacks when a state gets [un]installed at runtime.
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200631 *
632 * Each invocation of this function by the smpboot thread does a single AP
633 * state callback.
634 *
635 * It has 3 modes of operation:
636 * - single: runs st->cb_state
637 * - up: runs ++st->state, while st->state < st->target
638 * - down: runs st->state--, while st->state > st->target
639 *
640 * When complete or on error, should_run is cleared and the completion is fired.
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000641 */
642static void cpuhp_thread_fun(unsigned int cpu)
643{
644 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200645 bool bringup = st->bringup;
646 enum cpuhp_state state;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000647
Neeraj Upadhyayf8b75302018-09-05 11:22:07 +0530648 if (WARN_ON_ONCE(!st->should_run))
649 return;
650
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000651 /*
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200652 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
653 * that if we see ->should_run we also see the rest of the state.
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000654 */
655 smp_mb();
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200656
Peter Zijlstracb921732018-09-11 11:51:27 +0200657 /*
658 * The BP holds the hotplug lock, but we're now running on the AP,
659 * ensure that anybody asserting the lock is held, will actually find
660 * it so.
661 */
662 lockdep_acquire_cpus_lock();
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +0200663 cpuhp_lock_acquire(bringup);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200664
Thomas Gleixnera7246322016-08-12 19:49:38 +0200665 if (st->single) {
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200666 state = st->cb_state;
667 st->should_run = false;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000668 } else {
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200669 if (bringup) {
670 st->state++;
671 state = st->state;
672 st->should_run = (st->state < st->target);
673 WARN_ON_ONCE(st->state > st->target);
674 } else {
675 state = st->state;
676 st->state--;
677 st->should_run = (st->state > st->target);
678 WARN_ON_ONCE(st->state < st->target);
679 }
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000680 }
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200681
682 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
683
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200684 if (cpuhp_is_atomic_state(state)) {
685 local_irq_disable();
686 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
687 local_irq_enable();
688
689 /*
690 * STARTING/DYING must not fail!
691 */
692 WARN_ON_ONCE(st->result);
693 } else {
694 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
695 }
696
697 if (st->result) {
698 /*
699 * If we fail on a rollback, we're up a creek without no
700 * paddle, no way forward, no way back. We loose, thanks for
701 * playing.
702 */
703 WARN_ON_ONCE(st->rollback);
704 st->should_run = false;
705 }
706
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +0200707 cpuhp_lock_release(bringup);
Peter Zijlstracb921732018-09-11 11:51:27 +0200708 lockdep_release_cpus_lock();
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200709
710 if (!st->should_run)
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200711 complete_ap_thread(st, bringup);
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000712}
713
714/* Invoke a single callback on a remote cpu */
Thomas Gleixnera7246322016-08-12 19:49:38 +0200715static int
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200716cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
717 struct hlist_node *node)
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000718{
719 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200720 int ret;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000721
722 if (!cpu_online(cpu))
723 return 0;
724
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +0200725 cpuhp_lock_acquire(false);
726 cpuhp_lock_release(false);
727
728 cpuhp_lock_acquire(true);
729 cpuhp_lock_release(true);
Thomas Gleixner49dfe2a2017-05-24 10:15:43 +0200730
Thomas Gleixner6a4e2452016-07-13 17:16:03 +0000731 /*
732 * If we are up and running, use the hotplug thread. For early calls
733 * we invoke the thread function directly.
734 */
735 if (!st->thread)
Peter Zijlstra96abb962017-09-20 19:00:16 +0200736 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
Thomas Gleixner6a4e2452016-07-13 17:16:03 +0000737
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200738 st->rollback = false;
739 st->last = NULL;
740
741 st->node = node;
742 st->bringup = bringup;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000743 st->cb_state = state;
Thomas Gleixnera7246322016-08-12 19:49:38 +0200744 st->single = true;
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200745
746 __cpuhp_kick_ap(st);
Thomas Gleixnera7246322016-08-12 19:49:38 +0200747
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000748 /*
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200749 * If we failed and did a partial, do a rollback.
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000750 */
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200751 if ((ret = st->result) && st->last) {
752 st->rollback = true;
753 st->bringup = !bringup;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000754
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200755 __cpuhp_kick_ap(st);
756 }
757
Thomas Gleixner1f7c70d2017-10-21 16:06:52 +0200758 /*
759 * Clean up the leftovers so the next hotplug operation wont use stale
760 * data.
761 */
762 st->node = st->last = NULL;
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200763 return ret;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000764}
765
766static int cpuhp_kick_ap_work(unsigned int cpu)
767{
768 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200769 enum cpuhp_state prev_state = st->state;
770 int ret;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000771
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +0200772 cpuhp_lock_acquire(false);
773 cpuhp_lock_release(false);
774
775 cpuhp_lock_acquire(true);
776 cpuhp_lock_release(true);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200777
778 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
779 ret = cpuhp_kick_ap(st, st->target);
780 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
781
782 return ret;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000783}
784
785static struct smp_hotplug_thread cpuhp_threads = {
786 .store = &cpuhp_state.thread,
787 .create = &cpuhp_create,
788 .thread_should_run = cpuhp_should_run,
789 .thread_fn = cpuhp_thread_fun,
790 .thread_comm = "cpuhp/%u",
791 .selfparking = true,
792};
793
794void __init cpuhp_threads_init(void)
795{
796 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
797 kthread_unpark(this_cpu_read(cpuhp_state.thread));
798}
799
Michal Hocko777c6e02016-12-07 14:54:38 +0100800#ifdef CONFIG_HOTPLUG_CPU
Anton Vorontsove4cc2f82012-05-31 16:26:26 -0700801/**
802 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
803 * @cpu: a CPU id
804 *
805 * This function walks all processes, finds a valid mm struct for each one and
806 * then clears a corresponding bit in mm's cpumask. While this all sounds
807 * trivial, there are various non-obvious corner cases, which this function
808 * tries to solve in a safe manner.
809 *
810 * Also note that the function uses a somewhat relaxed locking scheme, so it may
811 * be called only for an already offlined CPU.
812 */
Anton Vorontsovcb792952012-05-31 16:26:22 -0700813void clear_tasks_mm_cpumask(int cpu)
814{
815 struct task_struct *p;
816
817 /*
818 * This function is called after the cpu is taken down and marked
819 * offline, so its not like new tasks will ever get this cpu set in
820 * their mm mask. -- Peter Zijlstra
821 * Thus, we may use rcu_read_lock() here, instead of grabbing
822 * full-fledged tasklist_lock.
823 */
Anton Vorontsove4cc2f82012-05-31 16:26:26 -0700824 WARN_ON(cpu_online(cpu));
Anton Vorontsovcb792952012-05-31 16:26:22 -0700825 rcu_read_lock();
826 for_each_process(p) {
827 struct task_struct *t;
828
Anton Vorontsove4cc2f82012-05-31 16:26:26 -0700829 /*
830 * Main thread might exit, but other threads may still have
831 * a valid mm. Find one.
832 */
Anton Vorontsovcb792952012-05-31 16:26:22 -0700833 t = find_lock_task_mm(p);
834 if (!t)
835 continue;
836 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
837 task_unlock(t);
838 }
839 rcu_read_unlock();
840}
841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842/* Take this CPU down. */
Mathias Krause71cf5ae2015-07-19 20:06:22 +0200843static int take_cpu_down(void *_param)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844{
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000845 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
846 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000847 int err, cpu = smp_processor_id();
Peter Zijlstra724a8682017-09-20 19:00:18 +0200848 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 /* Ensure this CPU doesn't handle any more interrupts. */
851 err = __cpu_disable();
852 if (err < 0)
Zwane Mwaikambof3705132005-06-25 14:54:50 -0700853 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
Thomas Gleixnera7246322016-08-12 19:49:38 +0200855 /*
856 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
857 * do this step again.
858 */
859 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
860 st->state--;
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000861 /* Invoke the former CPU_DYING callbacks */
Peter Zijlstra724a8682017-09-20 19:00:18 +0200862 for (; st->state > target; st->state--) {
863 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
864 /*
865 * DYING must not fail!
866 */
867 WARN_ON_ONCE(ret);
868 }
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000869
Thomas Gleixner52c063d2015-04-03 02:37:24 +0200870 /* Give up timekeeping duties */
871 tick_handover_do_timer();
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000872 /* Park the stopper thread */
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000873 stop_machine_park(cpu);
Zwane Mwaikambof3705132005-06-25 14:54:50 -0700874 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
Thomas Gleixner98458172016-02-26 18:43:25 +0000877static int takedown_cpu(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878{
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000879 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Thomas Gleixner98458172016-02-26 18:43:25 +0000880 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
Thomas Gleixner2a58c522016-03-10 20:42:08 +0100882 /* Park the smpboot threads */
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000883 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
884
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200885 /*
Thomas Gleixnera8994182015-07-05 17:12:30 +0000886 * Prevent irq alloc/free while the dying cpu reorganizes the
887 * interrupt affinities.
888 */
889 irq_lock_sparse();
890
891 /*
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200892 * So now all preempt/rcu users must observe !cpu_active().
893 */
Sebastian Andrzej Siewior210e2132017-05-24 10:15:28 +0200894 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
Rusty Russell04321582008-07-28 12:16:29 -0500895 if (err) {
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +0200896 /* CPU refused to die */
Thomas Gleixnera8994182015-07-05 17:12:30 +0000897 irq_unlock_sparse();
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +0200898 /* Unpark the hotplug thread so we can rollback there */
899 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
Thomas Gleixner98458172016-02-26 18:43:25 +0000900 return err;
Satoru Takeuchi8fa1d7d2006-10-28 10:38:57 -0700901 }
Rusty Russell04321582008-07-28 12:16:29 -0500902 BUG_ON(cpu_online(cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Peter Zijlstra48c5ccae2010-11-13 19:32:29 +0100904 /*
Brendan Jackman5b1ead62017-12-06 10:59:11 +0000905 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
906 * all runnable tasks from the CPU, there's only the idle task left now
Peter Zijlstra48c5ccae2010-11-13 19:32:29 +0100907 * that the migration thread is done doing the stop_machine thing.
Peter Zijlstra51a96c72010-11-19 20:37:53 +0100908 *
909 * Wait for the stop thread to go away.
Peter Zijlstra48c5ccae2010-11-13 19:32:29 +0100910 */
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200911 wait_for_ap_thread(st, false);
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000912 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
Thomas Gleixnera8994182015-07-05 17:12:30 +0000914 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
915 irq_unlock_sparse();
916
Preeti U Murthy345527b2015-03-30 14:59:19 +0530917 hotplug_cpu__broadcast_tick_pull(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 /* This actually kills the CPU. */
919 __cpu_die(cpu);
920
Thomas Gleixnera49b1162015-04-03 02:38:05 +0200921 tick_cleanup_dead_cpu(cpu);
Paul E. McKenneya58163d2017-06-20 12:11:34 -0700922 rcutree_migrate_callbacks(cpu);
Thomas Gleixner98458172016-02-26 18:43:25 +0000923 return 0;
924}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
Thomas Gleixner71f87b22016-03-03 10:52:10 +0100926static void cpuhp_complete_idle_dead(void *arg)
927{
928 struct cpuhp_cpu_state *st = arg;
929
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200930 complete_ap_thread(st, false);
Thomas Gleixner71f87b22016-03-03 10:52:10 +0100931}
932
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000933void cpuhp_report_idle_dead(void)
934{
935 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
936
937 BUG_ON(st->state != CPUHP_AP_OFFLINE);
Thomas Gleixner27d50c72016-02-26 18:43:44 +0000938 rcu_report_dead(smp_processor_id());
Thomas Gleixner71f87b22016-03-03 10:52:10 +0100939 st->state = CPUHP_AP_IDLE_DEAD;
940 /*
941 * We cannot call complete after rcu_report_dead() so we delegate it
942 * to an online cpu.
943 */
944 smp_call_function_single(cpumask_first(cpu_online_mask),
945 cpuhp_complete_idle_dead, st, 0);
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000946}
947
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200948static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
949{
Mukesh Ojha6fb86d92018-08-28 12:24:54 +0530950 for (st->state++; st->state < st->target; st->state++)
951 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200952}
953
954static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
955 enum cpuhp_state target)
956{
957 enum cpuhp_state prev_state = st->state;
958 int ret = 0;
959
960 for (; st->state > target; st->state--) {
961 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
962 if (ret) {
963 st->target = prev_state;
Thomas Gleixner69fa6eb2018-09-06 15:21:38 +0200964 if (st->state < prev_state)
965 undo_cpu_down(cpu, st);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200966 break;
967 }
968 }
969 return ret;
970}
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000971
Thomas Gleixner98458172016-02-26 18:43:25 +0000972/* Requires cpu_add_remove_lock to be held */
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000973static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
974 enum cpuhp_state target)
Thomas Gleixner98458172016-02-26 18:43:25 +0000975{
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000976 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
977 int prev_state, ret = 0;
Thomas Gleixner98458172016-02-26 18:43:25 +0000978
979 if (num_online_cpus() == 1)
980 return -EBUSY;
981
Thomas Gleixner757c9892016-02-26 18:43:32 +0000982 if (!cpu_present(cpu))
Thomas Gleixner98458172016-02-26 18:43:25 +0000983 return -EINVAL;
984
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200985 cpus_write_lock();
Thomas Gleixner98458172016-02-26 18:43:25 +0000986
987 cpuhp_tasks_frozen = tasks_frozen;
988
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200989 prev_state = cpuhp_set_state(st, target);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000990 /*
991 * If the current CPU state is in the range of the AP hotplug thread,
992 * then we need to kick the thread.
993 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000994 if (st->state > CPUHP_TEARDOWN_CPU) {
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200995 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000996 ret = cpuhp_kick_ap_work(cpu);
997 /*
998 * The AP side has done the error rollback already. Just
999 * return the error code..
1000 */
1001 if (ret)
1002 goto out;
1003
1004 /*
1005 * We might have stopped still in the range of the AP hotplug
1006 * thread. Nothing to do anymore.
1007 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001008 if (st->state > CPUHP_TEARDOWN_CPU)
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001009 goto out;
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001010
1011 st->target = target;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001012 }
1013 /*
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001014 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001015 * to do the further cleanups.
1016 */
Thomas Gleixnera7246322016-08-12 19:49:38 +02001017 ret = cpuhp_down_callbacks(cpu, st, target);
Thomas Gleixner69fa6eb2018-09-06 15:21:38 +02001018 if (ret && st->state == CPUHP_TEARDOWN_CPU && st->state < prev_state) {
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001019 cpuhp_reset_state(st, prev_state);
1020 __cpuhp_kick_ap(st);
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +02001021 }
Thomas Gleixner98458172016-02-26 18:43:25 +00001022
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001023out:
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001024 cpus_write_unlock();
Thomas Gleixner941154b2017-09-12 21:37:04 +02001025 /*
1026 * Do post unplug cleanup. This is still protected against
1027 * concurrent CPU hotplug via cpu_add_remove_lock.
1028 */
1029 lockup_detector_cleanup();
Thomas Gleixnera74cfff2018-11-25 19:33:39 +01001030 arch_smt_update();
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001031 return ret;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001032}
1033
Thomas Gleixnercc1fe212018-05-29 17:49:05 +02001034static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1035{
1036 if (cpu_hotplug_disabled)
1037 return -EBUSY;
1038 return _cpu_down(cpu, 0, target);
1039}
1040
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001041static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001042{
Heiko Carstens9ea09af2008-12-22 12:36:30 +01001043 int err;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001044
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001045 cpu_maps_update_begin();
Thomas Gleixnercc1fe212018-05-29 17:49:05 +02001046 err = cpu_down_maps_locked(cpu, target);
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001047 cpu_maps_update_done();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 return err;
1049}
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001050
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001051int cpu_down(unsigned int cpu)
1052{
1053 return do_cpu_down(cpu, CPUHP_OFFLINE);
1054}
Zhang Ruib62b8ef2008-04-29 02:35:56 -04001055EXPORT_SYMBOL(cpu_down);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001056
1057#else
1058#define takedown_cpu NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059#endif /*CONFIG_HOTPLUG_CPU*/
1060
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001061/**
Thomas Gleixneree1e7142016-08-18 14:57:16 +02001062 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001063 * @cpu: cpu that just started
1064 *
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001065 * It must be called by the arch code on the new cpu, before the new cpu
1066 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1067 */
1068void notify_cpu_starting(unsigned int cpu)
1069{
1070 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1071 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
Peter Zijlstra724a8682017-09-20 19:00:18 +02001072 int ret;
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001073
Sebastian Andrzej Siewior0c6d4572016-08-17 14:21:04 +02001074 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +02001075 st->booted_once = true;
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001076 while (st->state < target) {
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001077 st->state++;
Peter Zijlstra724a8682017-09-20 19:00:18 +02001078 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1079 /*
1080 * STARTING must not fail!
1081 */
1082 WARN_ON_ONCE(ret);
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001083 }
1084}
1085
Thomas Gleixner949338e2016-02-26 18:43:35 +00001086/*
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +02001087 * Called from the idle task. Wake up the controlling task which brings the
1088 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1089 * the rest of the online bringup to the hotplug thread.
Thomas Gleixner949338e2016-02-26 18:43:35 +00001090 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001091void cpuhp_online_idle(enum cpuhp_state state)
Thomas Gleixner949338e2016-02-26 18:43:35 +00001092{
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001093 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001094
1095 /* Happens for the boot cpu */
1096 if (state != CPUHP_AP_ONLINE_IDLE)
1097 return;
1098
1099 st->state = CPUHP_AP_ONLINE_IDLE;
Peter Zijlstra5ebe7742017-09-20 19:00:19 +02001100 complete_ap_thread(st, true);
Thomas Gleixner949338e2016-02-26 18:43:35 +00001101}
1102
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001103/* Requires cpu_add_remove_lock to be held */
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001104static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001106 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Suresh Siddha3bb5d2e2012-04-20 17:08:50 -07001107 struct task_struct *idle;
Thomas Gleixner2e1a3482016-02-26 18:43:37 +00001108 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001110 cpus_write_lock();
Thomas Gleixner38498a62012-04-20 13:05:44 +00001111
Thomas Gleixner757c9892016-02-26 18:43:32 +00001112 if (!cpu_present(cpu)) {
Yasuaki Ishimatsu5e5041f2012-10-23 01:30:54 +02001113 ret = -EINVAL;
1114 goto out;
1115 }
1116
Thomas Gleixner757c9892016-02-26 18:43:32 +00001117 /*
1118 * The caller of do_cpu_up might have raced with another
1119 * caller. Ignore it for now.
1120 */
1121 if (st->state >= target)
Thomas Gleixner38498a62012-04-20 13:05:44 +00001122 goto out;
Thomas Gleixner757c9892016-02-26 18:43:32 +00001123
1124 if (st->state == CPUHP_OFFLINE) {
1125 /* Let it fail before we try to bring the cpu up */
1126 idle = idle_thread_get(cpu);
1127 if (IS_ERR(idle)) {
1128 ret = PTR_ERR(idle);
1129 goto out;
1130 }
Suresh Siddha3bb5d2e2012-04-20 17:08:50 -07001131 }
Thomas Gleixner38498a62012-04-20 13:05:44 +00001132
Thomas Gleixnerba997462016-02-26 18:43:24 +00001133 cpuhp_tasks_frozen = tasks_frozen;
1134
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001135 cpuhp_set_state(st, target);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001136 /*
1137 * If the current CPU state is in the range of the AP hotplug thread,
1138 * then we need to kick the thread once more.
1139 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001140 if (st->state > CPUHP_BRINGUP_CPU) {
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001141 ret = cpuhp_kick_ap_work(cpu);
1142 /*
1143 * The AP side has done the error rollback already. Just
1144 * return the error code..
1145 */
1146 if (ret)
1147 goto out;
1148 }
1149
1150 /*
1151 * Try to reach the target state. We max out on the BP at
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001152 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001153 * responsible for bringing it up to the target state.
1154 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001155 target = min((int)target, CPUHP_BRINGUP_CPU);
Thomas Gleixnera7246322016-08-12 19:49:38 +02001156 ret = cpuhp_up_callbacks(cpu, st, target);
Thomas Gleixner38498a62012-04-20 13:05:44 +00001157out:
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001158 cpus_write_unlock();
Thomas Gleixnera74cfff2018-11-25 19:33:39 +01001159 arch_smt_update();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 return ret;
1161}
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001162
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001163static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001164{
1165 int err = 0;
minskey guocf234222010-05-24 14:32:41 -07001166
Rusty Russelle0b582e2009-01-01 10:12:28 +10301167 if (!cpu_possible(cpu)) {
Fabian Frederick84117da2014-06-04 16:11:17 -07001168 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1169 cpu);
Chen Gong87d5e0232010-03-05 13:42:38 -08001170#if defined(CONFIG_IA64)
Fabian Frederick84117da2014-06-04 16:11:17 -07001171 pr_err("please check additional_cpus= boot parameter\n");
KAMEZAWA Hiroyuki73e753a2007-10-18 23:40:47 -07001172#endif
1173 return -EINVAL;
1174 }
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001175
Toshi Kani01b0f192013-11-12 15:07:25 -08001176 err = try_online_node(cpu_to_node(cpu));
1177 if (err)
1178 return err;
minskey guocf234222010-05-24 14:32:41 -07001179
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001180 cpu_maps_update_begin();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001181
Max Krasnyanskye761b772008-07-15 04:43:49 -07001182 if (cpu_hotplug_disabled) {
1183 err = -EBUSY;
1184 goto out;
1185 }
Thomas Gleixner05736e42018-05-29 17:48:27 +02001186 if (!cpu_smt_allowed(cpu)) {
1187 err = -EPERM;
1188 goto out;
1189 }
Max Krasnyanskye761b772008-07-15 04:43:49 -07001190
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001191 err = _cpu_up(cpu, 0, target);
Max Krasnyanskye761b772008-07-15 04:43:49 -07001192out:
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001193 cpu_maps_update_done();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001194 return err;
1195}
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001196
1197int cpu_up(unsigned int cpu)
1198{
1199 return do_cpu_up(cpu, CPUHP_ONLINE);
1200}
Paul E. McKenneya513f6b2011-12-11 21:54:45 -08001201EXPORT_SYMBOL_GPL(cpu_up);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001202
Rafael J. Wysockif3de4be2007-08-30 23:56:29 -07001203#ifdef CONFIG_PM_SLEEP_SMP
Rusty Russelle0b582e2009-01-01 10:12:28 +10301204static cpumask_var_t frozen_cpus;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001205
James Morsed391e552016-08-17 13:50:25 +01001206int freeze_secondary_cpus(int primary)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001207{
James Morsed391e552016-08-17 13:50:25 +01001208 int cpu, error = 0;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001209
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001210 cpu_maps_update_begin();
James Morsed391e552016-08-17 13:50:25 +01001211 if (!cpu_online(primary))
1212 primary = cpumask_first(cpu_online_mask);
Xiaotian Feng9ee349a2009-12-16 18:04:32 +01001213 /*
1214 * We take down all of the non-boot CPUs in one shot to avoid races
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001215 * with the userspace trying to use the CPU hotplug at the same time
1216 */
Rusty Russelle0b582e2009-01-01 10:12:28 +10301217 cpumask_clear(frozen_cpus);
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01001218
Fabian Frederick84117da2014-06-04 16:11:17 -07001219 pr_info("Disabling non-boot CPUs ...\n");
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001220 for_each_online_cpu(cpu) {
James Morsed391e552016-08-17 13:50:25 +01001221 if (cpu == primary)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001222 continue;
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001223 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001224 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001225 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
Mike Travisfeae3202009-11-17 18:22:13 -06001226 if (!error)
Rusty Russelle0b582e2009-01-01 10:12:28 +10301227 cpumask_set_cpu(cpu, frozen_cpus);
Mike Travisfeae3202009-11-17 18:22:13 -06001228 else {
Fabian Frederick84117da2014-06-04 16:11:17 -07001229 pr_err("Error taking CPU%d down: %d\n", cpu, error);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001230 break;
1231 }
1232 }
Joseph Cihula86886e52009-06-30 19:31:07 -07001233
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001234 if (!error)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001235 BUG_ON(num_online_cpus() > 1);
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001236 else
Fabian Frederick84117da2014-06-04 16:11:17 -07001237 pr_err("Non-boot CPUs are not disabled\n");
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001238
1239 /*
1240 * Make sure the CPUs won't be enabled by someone else. We need to do
1241 * this even in case of failure as all disable_nonboot_cpus() users are
1242 * supposed to do enable_nonboot_cpus() on the failure path.
1243 */
1244 cpu_hotplug_disabled++;
1245
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001246 cpu_maps_update_done();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001247 return error;
1248}
1249
Suresh Siddhad0af9ee2009-08-19 18:05:36 -07001250void __weak arch_enable_nonboot_cpus_begin(void)
1251{
1252}
1253
1254void __weak arch_enable_nonboot_cpus_end(void)
1255{
1256}
1257
Mathias Krause71cf5ae2015-07-19 20:06:22 +02001258void enable_nonboot_cpus(void)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001259{
1260 int cpu, error;
1261
1262 /* Allow everyone to use the CPU hotplug again */
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001263 cpu_maps_update_begin();
Lianwei Wang01b41152016-06-09 23:43:28 -07001264 __cpu_hotplug_enable();
Rusty Russelle0b582e2009-01-01 10:12:28 +10301265 if (cpumask_empty(frozen_cpus))
Rafael J. Wysocki1d64b9c2007-04-01 23:49:49 -07001266 goto out;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001267
Fabian Frederick84117da2014-06-04 16:11:17 -07001268 pr_info("Enabling non-boot CPUs ...\n");
Suresh Siddhad0af9ee2009-08-19 18:05:36 -07001269
1270 arch_enable_nonboot_cpus_begin();
1271
Rusty Russelle0b582e2009-01-01 10:12:28 +10301272 for_each_cpu(cpu, frozen_cpus) {
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001273 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001274 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001275 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001276 if (!error) {
Fabian Frederick84117da2014-06-04 16:11:17 -07001277 pr_info("CPU%d is up\n", cpu);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001278 continue;
1279 }
Fabian Frederick84117da2014-06-04 16:11:17 -07001280 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001281 }
Suresh Siddhad0af9ee2009-08-19 18:05:36 -07001282
1283 arch_enable_nonboot_cpus_end();
1284
Rusty Russelle0b582e2009-01-01 10:12:28 +10301285 cpumask_clear(frozen_cpus);
Rafael J. Wysocki1d64b9c2007-04-01 23:49:49 -07001286out:
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001287 cpu_maps_update_done();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001288}
Rusty Russelle0b582e2009-01-01 10:12:28 +10301289
Fenghua Yud7268a32011-11-15 21:59:31 +01001290static int __init alloc_frozen_cpus(void)
Rusty Russelle0b582e2009-01-01 10:12:28 +10301291{
1292 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1293 return -ENOMEM;
1294 return 0;
1295}
1296core_initcall(alloc_frozen_cpus);
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001297
1298/*
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001299 * When callbacks for CPU hotplug notifications are being executed, we must
1300 * ensure that the state of the system with respect to the tasks being frozen
1301 * or not, as reported by the notification, remains unchanged *throughout the
1302 * duration* of the execution of the callbacks.
1303 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1304 *
1305 * This synchronization is implemented by mutually excluding regular CPU
1306 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1307 * Hibernate notifications.
1308 */
1309static int
1310cpu_hotplug_pm_callback(struct notifier_block *nb,
1311 unsigned long action, void *ptr)
1312{
1313 switch (action) {
1314
1315 case PM_SUSPEND_PREPARE:
1316 case PM_HIBERNATION_PREPARE:
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -07001317 cpu_hotplug_disable();
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001318 break;
1319
1320 case PM_POST_SUSPEND:
1321 case PM_POST_HIBERNATION:
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -07001322 cpu_hotplug_enable();
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001323 break;
1324
1325 default:
1326 return NOTIFY_DONE;
1327 }
1328
1329 return NOTIFY_OK;
1330}
1331
1332
Fenghua Yud7268a32011-11-15 21:59:31 +01001333static int __init cpu_hotplug_pm_sync_init(void)
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001334{
Fenghua Yu6e32d472012-11-13 11:32:43 -08001335 /*
1336 * cpu_hotplug_pm_callback has higher priority than x86
1337 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1338 * to disable cpu hotplug to avoid cpu hotplug race.
1339 */
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001340 pm_notifier(cpu_hotplug_pm_callback, 0);
1341 return 0;
1342}
1343core_initcall(cpu_hotplug_pm_sync_init);
1344
Rafael J. Wysockif3de4be2007-08-30 23:56:29 -07001345#endif /* CONFIG_PM_SLEEP_SMP */
Max Krasnyansky68f4f1e2008-05-29 11:17:02 -07001346
Peter Zijlstra8ce371f2017-03-20 12:26:55 +01001347int __boot_cpu_id;
1348
Max Krasnyansky68f4f1e2008-05-29 11:17:02 -07001349#endif /* CONFIG_SMP */
Mike Travisb8d317d2008-07-24 18:21:29 -07001350
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001351/* Boot processor state steps */
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +08001352static struct cpuhp_step cpuhp_hp_states[] = {
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001353 [CPUHP_OFFLINE] = {
1354 .name = "offline",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001355 .startup.single = NULL,
1356 .teardown.single = NULL,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001357 },
1358#ifdef CONFIG_SMP
1359 [CPUHP_CREATE_THREADS]= {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001360 .name = "threads:prepare",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001361 .startup.single = smpboot_create_threads,
1362 .teardown.single = NULL,
Thomas Gleixner757c9892016-02-26 18:43:32 +00001363 .cant_stop = true,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001364 },
Thomas Gleixner00e16c32016-07-13 17:16:09 +00001365 [CPUHP_PERF_PREPARE] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001366 .name = "perf:prepare",
1367 .startup.single = perf_event_init_cpu,
1368 .teardown.single = perf_event_exit_cpu,
Thomas Gleixner00e16c32016-07-13 17:16:09 +00001369 },
Thomas Gleixner7ee681b2016-07-13 17:16:29 +00001370 [CPUHP_WORKQUEUE_PREP] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001371 .name = "workqueue:prepare",
1372 .startup.single = workqueue_prepare_cpu,
1373 .teardown.single = NULL,
Thomas Gleixner7ee681b2016-07-13 17:16:29 +00001374 },
Thomas Gleixner27590dc2016-07-15 10:41:04 +02001375 [CPUHP_HRTIMERS_PREPARE] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001376 .name = "hrtimers:prepare",
1377 .startup.single = hrtimers_prepare_cpu,
1378 .teardown.single = hrtimers_dead_cpu,
Thomas Gleixner27590dc2016-07-15 10:41:04 +02001379 },
Richard Weinberger31487f82016-07-13 17:17:01 +00001380 [CPUHP_SMPCFD_PREPARE] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001381 .name = "smpcfd:prepare",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001382 .startup.single = smpcfd_prepare_cpu,
1383 .teardown.single = smpcfd_dead_cpu,
Richard Weinberger31487f82016-07-13 17:17:01 +00001384 },
Richard Weinbergere6d49892016-08-18 14:57:17 +02001385 [CPUHP_RELAY_PREPARE] = {
1386 .name = "relay:prepare",
1387 .startup.single = relay_prepare_cpu,
1388 .teardown.single = NULL,
1389 },
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001390 [CPUHP_SLAB_PREPARE] = {
1391 .name = "slab:prepare",
1392 .startup.single = slab_prepare_cpu,
1393 .teardown.single = slab_dead_cpu,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001394 },
Thomas Gleixner4df83742016-07-13 17:17:03 +00001395 [CPUHP_RCUTREE_PREP] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001396 .name = "RCU/tree:prepare",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001397 .startup.single = rcutree_prepare_cpu,
1398 .teardown.single = rcutree_dead_cpu,
Thomas Gleixner4df83742016-07-13 17:17:03 +00001399 },
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001400 /*
Richard Cochran4fae16d2016-07-27 11:08:18 +02001401 * On the tear-down path, timers_dead_cpu() must be invoked
1402 * before blk_mq_queue_reinit_notify() from notify_dead(),
1403 * otherwise a RCU stall occurs.
1404 */
Thomas Gleixner26456f82017-12-27 21:37:25 +01001405 [CPUHP_TIMERS_PREPARE] = {
Mukesh Ojhad0180312018-07-24 20:17:48 +05301406 .name = "timers:prepare",
Thomas Gleixner26456f82017-12-27 21:37:25 +01001407 .startup.single = timers_prepare_cpu,
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001408 .teardown.single = timers_dead_cpu,
Richard Cochran4fae16d2016-07-27 11:08:18 +02001409 },
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001410 /* Kicks the plugged cpu into life */
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001411 [CPUHP_BRINGUP_CPU] = {
1412 .name = "cpu:bringup",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001413 .startup.single = bringup_cpu,
1414 .teardown.single = NULL,
Thomas Gleixner757c9892016-02-26 18:43:32 +00001415 .cant_stop = true,
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001416 },
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001417 /* Final state before CPU kills itself */
1418 [CPUHP_AP_IDLE_DEAD] = {
1419 .name = "idle:dead",
1420 },
1421 /*
1422 * Last state before CPU enters the idle loop to die. Transient state
1423 * for synchronization.
1424 */
1425 [CPUHP_AP_OFFLINE] = {
1426 .name = "ap:offline",
1427 .cant_stop = true,
1428 },
Thomas Gleixner9cf72432016-03-10 12:54:09 +01001429 /* First state is scheduler control. Interrupts are disabled */
1430 [CPUHP_AP_SCHED_STARTING] = {
1431 .name = "sched:starting",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001432 .startup.single = sched_cpu_starting,
1433 .teardown.single = sched_cpu_dying,
Thomas Gleixner9cf72432016-03-10 12:54:09 +01001434 },
Thomas Gleixner4df83742016-07-13 17:17:03 +00001435 [CPUHP_AP_RCUTREE_DYING] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001436 .name = "RCU/tree:dying",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001437 .startup.single = NULL,
1438 .teardown.single = rcutree_dying_cpu,
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001439 },
Lai Jiangshan46febd32017-11-28 21:19:53 +08001440 [CPUHP_AP_SMPCFD_DYING] = {
1441 .name = "smpcfd:dying",
1442 .startup.single = NULL,
1443 .teardown.single = smpcfd_dying_cpu,
1444 },
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001445 /* Entry state on starting. Interrupts enabled from here on. Transient
1446 * state for synchronsization */
1447 [CPUHP_AP_ONLINE] = {
1448 .name = "ap:online",
1449 },
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +08001450 /*
1451 * Handled on controll processor until the plugged processor manages
1452 * this itself.
1453 */
1454 [CPUHP_TEARDOWN_CPU] = {
1455 .name = "cpu:teardown",
1456 .startup.single = NULL,
1457 .teardown.single = takedown_cpu,
1458 .cant_stop = true,
1459 },
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001460 /* Handle smpboot threads park/unpark */
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001461 [CPUHP_AP_SMPBOOT_THREADS] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001462 .name = "smpboot/threads:online",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001463 .startup.single = smpboot_unpark_threads,
Thomas Gleixnerc4de6562018-05-29 19:05:25 +02001464 .teardown.single = smpboot_park_threads,
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001465 },
Thomas Gleixnerc5cb83b2017-06-20 01:37:51 +02001466 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1467 .name = "irq/affinity:online",
1468 .startup.single = irq_affinity_online_cpu,
1469 .teardown.single = NULL,
1470 },
Thomas Gleixner00e16c32016-07-13 17:16:09 +00001471 [CPUHP_AP_PERF_ONLINE] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001472 .name = "perf:online",
1473 .startup.single = perf_event_init_cpu,
1474 .teardown.single = perf_event_exit_cpu,
Thomas Gleixner00e16c32016-07-13 17:16:09 +00001475 },
Peter Zijlstra9cf57732018-06-07 10:52:03 +02001476 [CPUHP_AP_WATCHDOG_ONLINE] = {
1477 .name = "lockup_detector:online",
1478 .startup.single = lockup_detector_online_cpu,
1479 .teardown.single = lockup_detector_offline_cpu,
1480 },
Thomas Gleixner7ee681b2016-07-13 17:16:29 +00001481 [CPUHP_AP_WORKQUEUE_ONLINE] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001482 .name = "workqueue:online",
1483 .startup.single = workqueue_online_cpu,
1484 .teardown.single = workqueue_offline_cpu,
Thomas Gleixner7ee681b2016-07-13 17:16:29 +00001485 },
Thomas Gleixner4df83742016-07-13 17:17:03 +00001486 [CPUHP_AP_RCUTREE_ONLINE] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001487 .name = "RCU/tree:online",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001488 .startup.single = rcutree_online_cpu,
1489 .teardown.single = rcutree_offline_cpu,
Thomas Gleixner4df83742016-07-13 17:17:03 +00001490 },
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001491#endif
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001492 /*
1493 * The dynamically registered state space is here
1494 */
1495
Thomas Gleixneraaddd7d2016-03-10 12:54:19 +01001496#ifdef CONFIG_SMP
1497 /* Last state is scheduler control setting the cpu active */
1498 [CPUHP_AP_ACTIVE] = {
1499 .name = "sched:active",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001500 .startup.single = sched_cpu_activate,
1501 .teardown.single = sched_cpu_deactivate,
Thomas Gleixneraaddd7d2016-03-10 12:54:19 +01001502 },
1503#endif
1504
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001505 /* CPU is fully up and running. */
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001506 [CPUHP_ONLINE] = {
1507 .name = "online",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001508 .startup.single = NULL,
1509 .teardown.single = NULL,
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001510 },
1511};
1512
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001513/* Sanity check for callbacks */
1514static int cpuhp_cb_check(enum cpuhp_state state)
1515{
1516 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1517 return -EINVAL;
1518 return 0;
1519}
1520
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001521/*
1522 * Returns a free for dynamic slot assignment of the Online state. The states
1523 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1524 * by having no name assigned.
1525 */
1526static int cpuhp_reserve_state(enum cpuhp_state state)
1527{
Thomas Gleixner4205e472017-01-10 14:01:05 +01001528 enum cpuhp_state i, end;
1529 struct cpuhp_step *step;
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001530
Thomas Gleixner4205e472017-01-10 14:01:05 +01001531 switch (state) {
1532 case CPUHP_AP_ONLINE_DYN:
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +08001533 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
Thomas Gleixner4205e472017-01-10 14:01:05 +01001534 end = CPUHP_AP_ONLINE_DYN_END;
1535 break;
1536 case CPUHP_BP_PREPARE_DYN:
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +08001537 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
Thomas Gleixner4205e472017-01-10 14:01:05 +01001538 end = CPUHP_BP_PREPARE_DYN_END;
1539 break;
1540 default:
1541 return -EINVAL;
1542 }
1543
1544 for (i = state; i <= end; i++, step++) {
1545 if (!step->name)
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001546 return i;
1547 }
1548 WARN(1, "No more dynamic states available for CPU hotplug\n");
1549 return -ENOSPC;
1550}
1551
1552static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1553 int (*startup)(unsigned int cpu),
1554 int (*teardown)(unsigned int cpu),
1555 bool multi_instance)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001556{
1557 /* (Un)Install the callbacks for further cpu hotplug operations */
1558 struct cpuhp_step *sp;
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001559 int ret = 0;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001560
Ethan Barnes0c96b272017-07-19 22:36:00 +00001561 /*
1562 * If name is NULL, then the state gets removed.
1563 *
1564 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1565 * the first allocation from these dynamic ranges, so the removal
1566 * would trigger a new allocation and clear the wrong (already
1567 * empty) state, leaving the callbacks of the to be cleared state
1568 * dangling, which causes wreckage on the next hotplug operation.
1569 */
1570 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1571 state == CPUHP_BP_PREPARE_DYN)) {
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001572 ret = cpuhp_reserve_state(state);
1573 if (ret < 0)
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001574 return ret;
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001575 state = ret;
1576 }
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001577 sp = cpuhp_get_step(state);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001578 if (name && sp->name)
1579 return -EBUSY;
1580
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001581 sp->startup.single = startup;
1582 sp->teardown.single = teardown;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001583 sp->name = name;
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001584 sp->multi_instance = multi_instance;
1585 INIT_HLIST_HEAD(&sp->list);
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001586 return ret;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001587}
1588
1589static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1590{
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001591 return cpuhp_get_step(state)->teardown.single;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001592}
1593
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001594/*
1595 * Call the startup/teardown function for a step either on the AP or
1596 * on the current CPU.
1597 */
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001598static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1599 struct hlist_node *node)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001600{
Thomas Gleixnera7246322016-08-12 19:49:38 +02001601 struct cpuhp_step *sp = cpuhp_get_step(state);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001602 int ret;
1603
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001604 /*
1605 * If there's nothing to do, we done.
1606 * Relies on the union for multi_instance.
1607 */
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001608 if ((bringup && !sp->startup.single) ||
1609 (!bringup && !sp->teardown.single))
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001610 return 0;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001611 /*
1612 * The non AP bound callbacks can fail on bringup. On teardown
1613 * e.g. module removal we crash for now.
1614 */
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001615#ifdef CONFIG_SMP
1616 if (cpuhp_is_ap_state(state))
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001617 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001618 else
Peter Zijlstra96abb962017-09-20 19:00:16 +02001619 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001620#else
Peter Zijlstra96abb962017-09-20 19:00:16 +02001621 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001622#endif
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001623 BUG_ON(ret && !bringup);
1624 return ret;
1625}
1626
1627/*
1628 * Called from __cpuhp_setup_state on a recoverable failure.
1629 *
1630 * Note: The teardown callbacks for rollback are not allowed to fail!
1631 */
1632static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001633 struct hlist_node *node)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001634{
1635 int cpu;
1636
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001637 /* Roll back the already executed steps on the other cpus */
1638 for_each_present_cpu(cpu) {
1639 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1640 int cpustate = st->state;
1641
1642 if (cpu >= failedcpu)
1643 break;
1644
1645 /* Did we invoke the startup call on that cpu ? */
1646 if (cpustate >= state)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001647 cpuhp_issue_call(cpu, state, false, node);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001648 }
1649}
1650
Thomas Gleixner9805c672017-05-24 10:15:15 +02001651int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1652 struct hlist_node *node,
1653 bool invoke)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001654{
1655 struct cpuhp_step *sp;
1656 int cpu;
1657 int ret;
1658
Thomas Gleixner9805c672017-05-24 10:15:15 +02001659 lockdep_assert_cpus_held();
1660
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001661 sp = cpuhp_get_step(state);
1662 if (sp->multi_instance == false)
1663 return -EINVAL;
1664
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001665 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001666
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001667 if (!invoke || !sp->startup.multi)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001668 goto add_node;
1669
1670 /*
1671 * Try to call the startup callback for each present cpu
1672 * depending on the hotplug state of the cpu.
1673 */
1674 for_each_present_cpu(cpu) {
1675 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1676 int cpustate = st->state;
1677
1678 if (cpustate < state)
1679 continue;
1680
1681 ret = cpuhp_issue_call(cpu, state, true, node);
1682 if (ret) {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001683 if (sp->teardown.multi)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001684 cpuhp_rollback_install(cpu, state, node);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001685 goto unlock;
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001686 }
1687 }
1688add_node:
1689 ret = 0;
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001690 hlist_add_head(node, &sp->list);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001691unlock:
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001692 mutex_unlock(&cpuhp_state_mutex);
Thomas Gleixner9805c672017-05-24 10:15:15 +02001693 return ret;
1694}
1695
1696int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1697 bool invoke)
1698{
1699 int ret;
1700
1701 cpus_read_lock();
1702 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001703 cpus_read_unlock();
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001704 return ret;
1705}
1706EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1707
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001708/**
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001709 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001710 * @state: The state to setup
1711 * @invoke: If true, the startup function is invoked for cpus where
1712 * cpu state >= @state
1713 * @startup: startup callback function
1714 * @teardown: teardown callback function
1715 * @multi_instance: State is set up for multiple instances which get
1716 * added afterwards.
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001717 *
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001718 * The caller needs to hold cpus read locked while calling this function.
Boris Ostrovsky512f0982016-12-15 10:00:57 -05001719 * Returns:
1720 * On success:
1721 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1722 * 0 for all other states
1723 * On failure: proper (negative) error code
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001724 */
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001725int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1726 const char *name, bool invoke,
1727 int (*startup)(unsigned int cpu),
1728 int (*teardown)(unsigned int cpu),
1729 bool multi_instance)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001730{
1731 int cpu, ret = 0;
Thomas Gleixnerb9d9d692016-12-26 22:58:19 +01001732 bool dynstate;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001733
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001734 lockdep_assert_cpus_held();
1735
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001736 if (cpuhp_cb_check(state) || !name)
1737 return -EINVAL;
1738
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001739 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001740
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001741 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1742 multi_instance);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001743
Thomas Gleixnerb9d9d692016-12-26 22:58:19 +01001744 dynstate = state == CPUHP_AP_ONLINE_DYN;
1745 if (ret > 0 && dynstate) {
1746 state = ret;
1747 ret = 0;
1748 }
1749
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001750 if (ret || !invoke || !startup)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001751 goto out;
1752
1753 /*
1754 * Try to call the startup callback for each present cpu
1755 * depending on the hotplug state of the cpu.
1756 */
1757 for_each_present_cpu(cpu) {
1758 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1759 int cpustate = st->state;
1760
1761 if (cpustate < state)
1762 continue;
1763
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001764 ret = cpuhp_issue_call(cpu, state, true, NULL);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001765 if (ret) {
Thomas Gleixnera7246322016-08-12 19:49:38 +02001766 if (teardown)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001767 cpuhp_rollback_install(cpu, state, NULL);
1768 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001769 goto out;
1770 }
1771 }
1772out:
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001773 mutex_unlock(&cpuhp_state_mutex);
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001774 /*
1775 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1776 * dynamically allocated state in case of success.
1777 */
Thomas Gleixnerb9d9d692016-12-26 22:58:19 +01001778 if (!ret && dynstate)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001779 return state;
1780 return ret;
1781}
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001782EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1783
1784int __cpuhp_setup_state(enum cpuhp_state state,
1785 const char *name, bool invoke,
1786 int (*startup)(unsigned int cpu),
1787 int (*teardown)(unsigned int cpu),
1788 bool multi_instance)
1789{
1790 int ret;
1791
1792 cpus_read_lock();
1793 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1794 teardown, multi_instance);
1795 cpus_read_unlock();
1796 return ret;
1797}
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001798EXPORT_SYMBOL(__cpuhp_setup_state);
1799
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001800int __cpuhp_state_remove_instance(enum cpuhp_state state,
1801 struct hlist_node *node, bool invoke)
1802{
1803 struct cpuhp_step *sp = cpuhp_get_step(state);
1804 int cpu;
1805
1806 BUG_ON(cpuhp_cb_check(state));
1807
1808 if (!sp->multi_instance)
1809 return -EINVAL;
1810
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001811 cpus_read_lock();
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001812 mutex_lock(&cpuhp_state_mutex);
1813
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001814 if (!invoke || !cpuhp_get_teardown_cb(state))
1815 goto remove;
1816 /*
1817 * Call the teardown callback for each present cpu depending
1818 * on the hotplug state of the cpu. This function is not
1819 * allowed to fail currently!
1820 */
1821 for_each_present_cpu(cpu) {
1822 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1823 int cpustate = st->state;
1824
1825 if (cpustate >= state)
1826 cpuhp_issue_call(cpu, state, false, node);
1827 }
1828
1829remove:
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001830 hlist_del(node);
1831 mutex_unlock(&cpuhp_state_mutex);
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001832 cpus_read_unlock();
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001833
1834 return 0;
1835}
1836EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001837
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001838/**
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001839 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001840 * @state: The state to remove
1841 * @invoke: If true, the teardown function is invoked for cpus where
1842 * cpu state >= @state
1843 *
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001844 * The caller needs to hold cpus read locked while calling this function.
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001845 * The teardown callback is currently not allowed to fail. Think
1846 * about module removal!
1847 */
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001848void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001849{
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001850 struct cpuhp_step *sp = cpuhp_get_step(state);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001851 int cpu;
1852
1853 BUG_ON(cpuhp_cb_check(state));
1854
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001855 lockdep_assert_cpus_held();
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001856
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001857 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001858 if (sp->multi_instance) {
1859 WARN(!hlist_empty(&sp->list),
1860 "Error: Removing state %d which has instances left.\n",
1861 state);
1862 goto remove;
1863 }
1864
Thomas Gleixnera7246322016-08-12 19:49:38 +02001865 if (!invoke || !cpuhp_get_teardown_cb(state))
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001866 goto remove;
1867
1868 /*
1869 * Call the teardown callback for each present cpu depending
1870 * on the hotplug state of the cpu. This function is not
1871 * allowed to fail currently!
1872 */
1873 for_each_present_cpu(cpu) {
1874 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1875 int cpustate = st->state;
1876
1877 if (cpustate >= state)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001878 cpuhp_issue_call(cpu, state, false, NULL);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001879 }
1880remove:
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001881 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001882 mutex_unlock(&cpuhp_state_mutex);
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001883}
1884EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1885
1886void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1887{
1888 cpus_read_lock();
1889 __cpuhp_remove_state_cpuslocked(state, invoke);
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001890 cpus_read_unlock();
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001891}
1892EXPORT_SYMBOL(__cpuhp_remove_state);
1893
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001894#if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1895static ssize_t show_cpuhp_state(struct device *dev,
1896 struct device_attribute *attr, char *buf)
1897{
1898 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1899
1900 return sprintf(buf, "%d\n", st->state);
1901}
1902static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1903
Thomas Gleixner757c9892016-02-26 18:43:32 +00001904static ssize_t write_cpuhp_target(struct device *dev,
1905 struct device_attribute *attr,
1906 const char *buf, size_t count)
1907{
1908 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1909 struct cpuhp_step *sp;
1910 int target, ret;
1911
1912 ret = kstrtoint(buf, 10, &target);
1913 if (ret)
1914 return ret;
1915
1916#ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1917 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1918 return -EINVAL;
1919#else
1920 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1921 return -EINVAL;
1922#endif
1923
1924 ret = lock_device_hotplug_sysfs();
1925 if (ret)
1926 return ret;
1927
1928 mutex_lock(&cpuhp_state_mutex);
1929 sp = cpuhp_get_step(target);
1930 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1931 mutex_unlock(&cpuhp_state_mutex);
1932 if (ret)
Sebastian Andrzej Siewior40da1b12017-06-02 16:27:14 +02001933 goto out;
Thomas Gleixner757c9892016-02-26 18:43:32 +00001934
1935 if (st->state < target)
1936 ret = do_cpu_up(dev->id, target);
1937 else
1938 ret = do_cpu_down(dev->id, target);
Sebastian Andrzej Siewior40da1b12017-06-02 16:27:14 +02001939out:
Thomas Gleixner757c9892016-02-26 18:43:32 +00001940 unlock_device_hotplug();
1941 return ret ? ret : count;
1942}
1943
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001944static ssize_t show_cpuhp_target(struct device *dev,
1945 struct device_attribute *attr, char *buf)
1946{
1947 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1948
1949 return sprintf(buf, "%d\n", st->target);
1950}
Thomas Gleixner757c9892016-02-26 18:43:32 +00001951static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001952
Peter Zijlstra1db49482017-09-20 19:00:21 +02001953
1954static ssize_t write_cpuhp_fail(struct device *dev,
1955 struct device_attribute *attr,
1956 const char *buf, size_t count)
1957{
1958 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1959 struct cpuhp_step *sp;
1960 int fail, ret;
1961
1962 ret = kstrtoint(buf, 10, &fail);
1963 if (ret)
1964 return ret;
1965
1966 /*
1967 * Cannot fail STARTING/DYING callbacks.
1968 */
1969 if (cpuhp_is_atomic_state(fail))
1970 return -EINVAL;
1971
1972 /*
1973 * Cannot fail anything that doesn't have callbacks.
1974 */
1975 mutex_lock(&cpuhp_state_mutex);
1976 sp = cpuhp_get_step(fail);
1977 if (!sp->startup.single && !sp->teardown.single)
1978 ret = -EINVAL;
1979 mutex_unlock(&cpuhp_state_mutex);
1980 if (ret)
1981 return ret;
1982
1983 st->fail = fail;
1984
1985 return count;
1986}
1987
1988static ssize_t show_cpuhp_fail(struct device *dev,
1989 struct device_attribute *attr, char *buf)
1990{
1991 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1992
1993 return sprintf(buf, "%d\n", st->fail);
1994}
1995
1996static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1997
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001998static struct attribute *cpuhp_cpu_attrs[] = {
1999 &dev_attr_state.attr,
2000 &dev_attr_target.attr,
Peter Zijlstra1db49482017-09-20 19:00:21 +02002001 &dev_attr_fail.attr,
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002002 NULL
2003};
2004
Arvind Yadav993647a2017-06-29 17:40:47 +05302005static const struct attribute_group cpuhp_cpu_attr_group = {
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002006 .attrs = cpuhp_cpu_attrs,
2007 .name = "hotplug",
2008 NULL
2009};
2010
2011static ssize_t show_cpuhp_states(struct device *dev,
2012 struct device_attribute *attr, char *buf)
2013{
2014 ssize_t cur, res = 0;
2015 int i;
2016
2017 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixner757c9892016-02-26 18:43:32 +00002018 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002019 struct cpuhp_step *sp = cpuhp_get_step(i);
2020
2021 if (sp->name) {
2022 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2023 buf += cur;
2024 res += cur;
2025 }
2026 }
2027 mutex_unlock(&cpuhp_state_mutex);
2028 return res;
2029}
2030static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
2031
2032static struct attribute *cpuhp_cpu_root_attrs[] = {
2033 &dev_attr_states.attr,
2034 NULL
2035};
2036
Arvind Yadav993647a2017-06-29 17:40:47 +05302037static const struct attribute_group cpuhp_cpu_root_attr_group = {
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002038 .attrs = cpuhp_cpu_root_attrs,
2039 .name = "hotplug",
2040 NULL
2041};
2042
Thomas Gleixner05736e42018-05-29 17:48:27 +02002043#ifdef CONFIG_HOTPLUG_SMT
2044
2045static const char *smt_states[] = {
2046 [CPU_SMT_ENABLED] = "on",
2047 [CPU_SMT_DISABLED] = "off",
2048 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2049 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2050};
2051
2052static ssize_t
2053show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2054{
2055 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
2056}
2057
2058static void cpuhp_offline_cpu_device(unsigned int cpu)
2059{
2060 struct device *dev = get_cpu_device(cpu);
2061
2062 dev->offline = true;
2063 /* Tell user space about the state change */
2064 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2065}
2066
Thomas Gleixner215af542018-07-07 11:40:18 +02002067static void cpuhp_online_cpu_device(unsigned int cpu)
2068{
2069 struct device *dev = get_cpu_device(cpu);
2070
2071 dev->offline = false;
2072 /* Tell user space about the state change */
2073 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2074}
2075
Thomas Gleixner05736e42018-05-29 17:48:27 +02002076static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2077{
2078 int cpu, ret = 0;
2079
2080 cpu_maps_update_begin();
2081 for_each_online_cpu(cpu) {
2082 if (topology_is_primary_thread(cpu))
2083 continue;
2084 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2085 if (ret)
2086 break;
2087 /*
2088 * As this needs to hold the cpu maps lock it's impossible
2089 * to call device_offline() because that ends up calling
2090 * cpu_down() which takes cpu maps lock. cpu maps lock
2091 * needs to be held as this might race against in kernel
2092 * abusers of the hotplug machinery (thermal management).
2093 *
2094 * So nothing would update device:offline state. That would
2095 * leave the sysfs entry stale and prevent onlining after
2096 * smt control has been changed to 'off' again. This is
2097 * called under the sysfs hotplug lock, so it is properly
2098 * serialized against the regular offline usage.
2099 */
2100 cpuhp_offline_cpu_device(cpu);
2101 }
Jiri Kosina53c613f2018-09-25 14:38:55 +02002102 if (!ret) {
Thomas Gleixner05736e42018-05-29 17:48:27 +02002103 cpu_smt_control = ctrlval;
Jiri Kosina53c613f2018-09-25 14:38:55 +02002104 arch_smt_update();
2105 }
Thomas Gleixner05736e42018-05-29 17:48:27 +02002106 cpu_maps_update_done();
2107 return ret;
2108}
2109
Thomas Gleixner215af542018-07-07 11:40:18 +02002110static int cpuhp_smt_enable(void)
Thomas Gleixner05736e42018-05-29 17:48:27 +02002111{
Thomas Gleixner215af542018-07-07 11:40:18 +02002112 int cpu, ret = 0;
2113
Thomas Gleixner05736e42018-05-29 17:48:27 +02002114 cpu_maps_update_begin();
2115 cpu_smt_control = CPU_SMT_ENABLED;
Jiri Kosina53c613f2018-09-25 14:38:55 +02002116 arch_smt_update();
Thomas Gleixner215af542018-07-07 11:40:18 +02002117 for_each_present_cpu(cpu) {
2118 /* Skip online CPUs and CPUs on offline nodes */
2119 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2120 continue;
2121 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2122 if (ret)
2123 break;
2124 /* See comment in cpuhp_smt_disable() */
2125 cpuhp_online_cpu_device(cpu);
2126 }
Thomas Gleixner05736e42018-05-29 17:48:27 +02002127 cpu_maps_update_done();
Thomas Gleixner215af542018-07-07 11:40:18 +02002128 return ret;
Thomas Gleixner05736e42018-05-29 17:48:27 +02002129}
2130
2131static ssize_t
2132store_smt_control(struct device *dev, struct device_attribute *attr,
2133 const char *buf, size_t count)
2134{
2135 int ctrlval, ret;
2136
2137 if (sysfs_streq(buf, "on"))
2138 ctrlval = CPU_SMT_ENABLED;
2139 else if (sysfs_streq(buf, "off"))
2140 ctrlval = CPU_SMT_DISABLED;
2141 else if (sysfs_streq(buf, "forceoff"))
2142 ctrlval = CPU_SMT_FORCE_DISABLED;
2143 else
2144 return -EINVAL;
2145
2146 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2147 return -EPERM;
2148
2149 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2150 return -ENODEV;
2151
2152 ret = lock_device_hotplug_sysfs();
2153 if (ret)
2154 return ret;
2155
2156 if (ctrlval != cpu_smt_control) {
2157 switch (ctrlval) {
2158 case CPU_SMT_ENABLED:
Thomas Gleixner215af542018-07-07 11:40:18 +02002159 ret = cpuhp_smt_enable();
Thomas Gleixner05736e42018-05-29 17:48:27 +02002160 break;
2161 case CPU_SMT_DISABLED:
2162 case CPU_SMT_FORCE_DISABLED:
2163 ret = cpuhp_smt_disable(ctrlval);
2164 break;
2165 }
2166 }
2167
2168 unlock_device_hotplug();
2169 return ret ? ret : count;
2170}
2171static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2172
2173static ssize_t
2174show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2175{
2176 bool active = topology_max_smt_threads() > 1;
2177
2178 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2179}
2180static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2181
2182static struct attribute *cpuhp_smt_attrs[] = {
2183 &dev_attr_control.attr,
2184 &dev_attr_active.attr,
2185 NULL
2186};
2187
2188static const struct attribute_group cpuhp_smt_attr_group = {
2189 .attrs = cpuhp_smt_attrs,
2190 .name = "smt",
2191 NULL
2192};
2193
2194static int __init cpu_smt_state_init(void)
2195{
Thomas Gleixner05736e42018-05-29 17:48:27 +02002196 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2197 &cpuhp_smt_attr_group);
2198}
2199
2200#else
2201static inline int cpu_smt_state_init(void) { return 0; }
2202#endif
2203
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002204static int __init cpuhp_sysfs_init(void)
2205{
2206 int cpu, ret;
2207
Thomas Gleixner05736e42018-05-29 17:48:27 +02002208 ret = cpu_smt_state_init();
2209 if (ret)
2210 return ret;
2211
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002212 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2213 &cpuhp_cpu_root_attr_group);
2214 if (ret)
2215 return ret;
2216
2217 for_each_possible_cpu(cpu) {
2218 struct device *dev = get_cpu_device(cpu);
2219
2220 if (!dev)
2221 continue;
2222 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2223 if (ret)
2224 return ret;
2225 }
2226 return 0;
2227}
2228device_initcall(cpuhp_sysfs_init);
2229#endif
2230
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002231/*
2232 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2233 * represents all NR_CPUS bits binary values of 1<<nr.
2234 *
Rusty Russelle0b582e2009-01-01 10:12:28 +10302235 * It is used by cpumask_of() to get a constant address to a CPU
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002236 * mask value that has a single bit set only.
2237 */
Mike Travisb8d317d2008-07-24 18:21:29 -07002238
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002239/* cpu_bit_bitmap[0] is empty - so we can back into it */
Michael Rodriguez4d519852011-03-22 16:34:07 -07002240#define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002241#define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2242#define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2243#define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
Mike Travisb8d317d2008-07-24 18:21:29 -07002244
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002245const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
Mike Travisb8d317d2008-07-24 18:21:29 -07002246
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002247 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2248 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2249#if BITS_PER_LONG > 32
2250 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2251 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
Mike Travisb8d317d2008-07-24 18:21:29 -07002252#endif
2253};
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002254EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
Rusty Russell2d3854a2008-11-05 13:39:10 +11002255
2256const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2257EXPORT_SYMBOL(cpu_all_bits);
Rusty Russellb3199c02008-12-30 09:05:14 +10302258
2259#ifdef CONFIG_INIT_ALL_POSSIBLE
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002260struct cpumask __cpu_possible_mask __read_mostly
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08002261 = {CPU_BITS_ALL};
Rusty Russellb3199c02008-12-30 09:05:14 +10302262#else
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002263struct cpumask __cpu_possible_mask __read_mostly;
Rusty Russellb3199c02008-12-30 09:05:14 +10302264#endif
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002265EXPORT_SYMBOL(__cpu_possible_mask);
Rusty Russellb3199c02008-12-30 09:05:14 +10302266
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002267struct cpumask __cpu_online_mask __read_mostly;
2268EXPORT_SYMBOL(__cpu_online_mask);
Rusty Russellb3199c02008-12-30 09:05:14 +10302269
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002270struct cpumask __cpu_present_mask __read_mostly;
2271EXPORT_SYMBOL(__cpu_present_mask);
Rusty Russellb3199c02008-12-30 09:05:14 +10302272
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002273struct cpumask __cpu_active_mask __read_mostly;
2274EXPORT_SYMBOL(__cpu_active_mask);
Rusty Russell3fa41522008-12-30 09:05:16 +10302275
Rusty Russell3fa41522008-12-30 09:05:16 +10302276void init_cpu_present(const struct cpumask *src)
2277{
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08002278 cpumask_copy(&__cpu_present_mask, src);
Rusty Russell3fa41522008-12-30 09:05:16 +10302279}
2280
2281void init_cpu_possible(const struct cpumask *src)
2282{
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08002283 cpumask_copy(&__cpu_possible_mask, src);
Rusty Russell3fa41522008-12-30 09:05:16 +10302284}
2285
2286void init_cpu_online(const struct cpumask *src)
2287{
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08002288 cpumask_copy(&__cpu_online_mask, src);
Rusty Russell3fa41522008-12-30 09:05:16 +10302289}
Thomas Gleixnercff7d372016-02-26 18:43:28 +00002290
2291/*
2292 * Activate the first processor.
2293 */
2294void __init boot_cpu_init(void)
2295{
2296 int cpu = smp_processor_id();
2297
2298 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2299 set_cpu_online(cpu, true);
2300 set_cpu_active(cpu, true);
2301 set_cpu_present(cpu, true);
2302 set_cpu_possible(cpu, true);
Peter Zijlstra8ce371f2017-03-20 12:26:55 +01002303
2304#ifdef CONFIG_SMP
2305 __boot_cpu_id = cpu;
2306#endif
Thomas Gleixnercff7d372016-02-26 18:43:28 +00002307}
2308
2309/*
2310 * Must be called _AFTER_ setting up the per_cpu areas
2311 */
Linus Torvaldsb5b14042018-08-12 12:19:42 -07002312void __init boot_cpu_hotplug_init(void)
Thomas Gleixnercff7d372016-02-26 18:43:28 +00002313{
Abel Vesa269777a2018-08-15 00:26:00 +03002314#ifdef CONFIG_SMP
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +02002315 this_cpu_write(cpuhp_state.booted_once, true);
Abel Vesa269777a2018-08-15 00:26:00 +03002316#endif
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +02002317 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
Thomas Gleixnercff7d372016-02-26 18:43:28 +00002318}