blob: 9c706af713fbc3c8ee6fa0daf5805b607a92faad [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>
Nicholas Piggin9ca12ac2019-04-11 13:34:46 +100012#include <linux/sched/isolation.h>
Ingo Molnar29930022017-02-08 18:51:36 +010013#include <linux/sched/task.h>
Thomas Gleixnera74cfff2018-11-25 19:33:39 +010014#include <linux/sched/smt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/unistd.h>
16#include <linux/cpu.h>
Anton Vorontsovcb792952012-05-31 16:26:22 -070017#include <linux/oom.h>
18#include <linux/rcupdate.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040019#include <linux/export.h>
Anton Vorontsove4cc2f82012-05-31 16:26:26 -070020#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/kthread.h>
22#include <linux/stop_machine.h>
Ingo Molnar81615b622006-06-26 00:24:32 -070023#include <linux/mutex.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090024#include <linux/gfp.h>
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +010025#include <linux/suspend.h>
Gautham R. Shenoya19423b2014-03-11 02:04:03 +053026#include <linux/lockdep.h>
Preeti U Murthy345527b2015-03-30 14:59:19 +053027#include <linux/tick.h>
Thomas Gleixnera8994182015-07-05 17:12:30 +000028#include <linux/irq.h>
Thomas Gleixner941154b2017-09-12 21:37:04 +020029#include <linux/nmi.h>
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000030#include <linux/smpboot.h>
Richard Weinbergere6d49892016-08-18 14:57:17 +020031#include <linux/relay.h>
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +020032#include <linux/slab.h>
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +020033#include <linux/percpu-rwsem.h>
Thomas Gleixnercff7d372016-02-26 18:43:28 +000034
Todd E Brandtbb3632c2014-06-06 05:40:17 -070035#include <trace/events/power.h>
Thomas Gleixnercff7d372016-02-26 18:43:28 +000036#define CREATE_TRACE_POINTS
37#include <trace/events/cpuhp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Thomas Gleixner38498a62012-04-20 13:05:44 +000039#include "smpboot.h"
40
Thomas Gleixnercff7d372016-02-26 18:43:28 +000041/**
42 * cpuhp_cpu_state - Per cpu hotplug state storage
43 * @state: The current cpu state
44 * @target: The target state
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000045 * @thread: Pointer to the hotplug thread
46 * @should_run: Thread should execute
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +020047 * @rollback: Perform a rollback
Thomas Gleixnera7246322016-08-12 19:49:38 +020048 * @single: Single callback invocation
49 * @bringup: Single callback bringup or teardown selector
50 * @cb_state: The state for a single callback (install/uninstall)
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000051 * @result: Result of the operation
Peter Zijlstra5ebe7742017-09-20 19:00:19 +020052 * @done_up: Signal completion to the issuer of the task for cpu-up
53 * @done_down: Signal completion to the issuer of the task for cpu-down
Thomas Gleixnercff7d372016-02-26 18:43:28 +000054 */
55struct cpuhp_cpu_state {
56 enum cpuhp_state state;
57 enum cpuhp_state target;
Peter Zijlstra1db49482017-09-20 19:00:21 +020058 enum cpuhp_state fail;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000059#ifdef CONFIG_SMP
60 struct task_struct *thread;
61 bool should_run;
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +020062 bool rollback;
Thomas Gleixnera7246322016-08-12 19:49:38 +020063 bool single;
64 bool bringup;
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 Gleixnere797bda2019-07-22 20:47:16 +020078#ifdef CONFIG_SMP
79cpumask_t cpus_booted_once_mask;
80#endif
81
Thomas Gleixner49dfe2a2017-05-24 10:15:43 +020082#if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +020083static struct lockdep_map cpuhp_state_up_map =
84 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
85static struct lockdep_map cpuhp_state_down_map =
86 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
87
88
Mathieu Malaterre76dc6c02017-12-26 15:08:53 +010089static inline void cpuhp_lock_acquire(bool bringup)
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +020090{
91 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
92}
93
Mathieu Malaterre76dc6c02017-12-26 15:08:53 +010094static inline void cpuhp_lock_release(bool bringup)
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +020095{
96 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
97}
98#else
99
Mathieu Malaterre76dc6c02017-12-26 15:08:53 +0100100static inline void cpuhp_lock_acquire(bool bringup) { }
101static inline void cpuhp_lock_release(bool bringup) { }
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +0200102
Thomas Gleixner49dfe2a2017-05-24 10:15:43 +0200103#endif
104
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000105/**
106 * cpuhp_step - Hotplug state machine step
107 * @name: Name of the step
108 * @startup: Startup function of the step
109 * @teardown: Teardown function of the step
Thomas Gleixner757c9892016-02-26 18:43:32 +0000110 * @cant_stop: Bringup/teardown can't be stopped at this step
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000111 */
112struct cpuhp_step {
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200113 const char *name;
114 union {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200115 int (*single)(unsigned int cpu);
116 int (*multi)(unsigned int cpu,
117 struct hlist_node *node);
118 } startup;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200119 union {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200120 int (*single)(unsigned int cpu);
121 int (*multi)(unsigned int cpu,
122 struct hlist_node *node);
123 } teardown;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200124 struct hlist_head list;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200125 bool cant_stop;
126 bool multi_instance;
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000127};
128
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +0000129static DEFINE_MUTEX(cpuhp_state_mutex);
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +0800130static struct cpuhp_step cpuhp_hp_states[];
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000131
Thomas Gleixnera7246322016-08-12 19:49:38 +0200132static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
133{
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +0800134 return cpuhp_hp_states + state;
Thomas Gleixnera7246322016-08-12 19:49:38 +0200135}
136
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000137/**
138 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
139 * @cpu: The cpu for which the callback should be invoked
Peter Zijlstra96abb962017-09-20 19:00:16 +0200140 * @state: The state to do callbacks for
Thomas Gleixnera7246322016-08-12 19:49:38 +0200141 * @bringup: True if the bringup callback should be invoked
Peter Zijlstra96abb962017-09-20 19:00:16 +0200142 * @node: For multi-instance, do a single entry callback for install/remove
143 * @lastp: For multi-instance rollback, remember how far we got
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000144 *
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200145 * Called from cpu hotplug and from the state register machinery.
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000146 */
Thomas Gleixnera7246322016-08-12 19:49:38 +0200147static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
Peter Zijlstra96abb962017-09-20 19:00:16 +0200148 bool bringup, struct hlist_node *node,
149 struct hlist_node **lastp)
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000150{
151 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Thomas Gleixnera7246322016-08-12 19:49:38 +0200152 struct cpuhp_step *step = cpuhp_get_step(state);
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200153 int (*cbm)(unsigned int cpu, struct hlist_node *node);
154 int (*cb)(unsigned int cpu);
155 int ret, cnt;
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000156
Peter Zijlstra1db49482017-09-20 19:00:21 +0200157 if (st->fail == state) {
158 st->fail = CPUHP_INVALID;
159
160 if (!(bringup ? step->startup.single : step->teardown.single))
161 return 0;
162
163 return -EAGAIN;
164 }
165
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200166 if (!step->multi_instance) {
Peter Zijlstra96abb962017-09-20 19:00:16 +0200167 WARN_ON_ONCE(lastp && *lastp);
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200168 cb = bringup ? step->startup.single : step->teardown.single;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200169 if (!cb)
170 return 0;
Thomas Gleixnera7246322016-08-12 19:49:38 +0200171 trace_cpuhp_enter(cpu, st->target, state, cb);
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000172 ret = cb(cpu);
Thomas Gleixnera7246322016-08-12 19:49:38 +0200173 trace_cpuhp_exit(cpu, st->state, state, ret);
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200174 return ret;
175 }
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200176 cbm = bringup ? step->startup.multi : step->teardown.multi;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200177 if (!cbm)
178 return 0;
179
180 /* Single invocation for instance add/remove */
181 if (node) {
Peter Zijlstra96abb962017-09-20 19:00:16 +0200182 WARN_ON_ONCE(lastp && *lastp);
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200183 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
184 ret = cbm(cpu, node);
185 trace_cpuhp_exit(cpu, st->state, state, ret);
186 return ret;
187 }
188
189 /* State transition. Invoke on all instances */
190 cnt = 0;
191 hlist_for_each(node, &step->list) {
Peter Zijlstra96abb962017-09-20 19:00:16 +0200192 if (lastp && node == *lastp)
193 break;
194
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200195 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
196 ret = cbm(cpu, node);
197 trace_cpuhp_exit(cpu, st->state, state, ret);
Peter Zijlstra96abb962017-09-20 19:00:16 +0200198 if (ret) {
199 if (!lastp)
200 goto err;
201
202 *lastp = node;
203 return ret;
204 }
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200205 cnt++;
206 }
Peter Zijlstra96abb962017-09-20 19:00:16 +0200207 if (lastp)
208 *lastp = NULL;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200209 return 0;
210err:
211 /* Rollback the instances if one failed */
Thomas Gleixner3c1627e2016-09-05 15:28:36 +0200212 cbm = !bringup ? step->startup.multi : step->teardown.multi;
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200213 if (!cbm)
214 return ret;
215
216 hlist_for_each(node, &step->list) {
217 if (!cnt--)
218 break;
Peter Zijlstra724a8682017-09-20 19:00:18 +0200219
220 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
221 ret = cbm(cpu, node);
222 trace_cpuhp_exit(cpu, st->state, state, ret);
223 /*
224 * Rollback must not fail,
225 */
226 WARN_ON_ONCE(ret);
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000227 }
228 return ret;
229}
230
Rusty Russell98a79d62008-12-13 21:19:41 +1030231#ifdef CONFIG_SMP
Arnd Bergmannfcb30292018-03-15 16:38:04 +0100232static bool cpuhp_is_ap_state(enum cpuhp_state state)
233{
234 /*
235 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
236 * purposes as that state is handled explicitly in cpu_down.
237 */
238 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
239}
240
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200241static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
242{
243 struct completion *done = bringup ? &st->done_up : &st->done_down;
244 wait_for_completion(done);
245}
246
247static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
248{
249 struct completion *done = bringup ? &st->done_up : &st->done_down;
250 complete(done);
251}
252
253/*
254 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
255 */
256static bool cpuhp_is_atomic_state(enum cpuhp_state state)
257{
258 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
259}
260
Rusty Russellb3199c02008-12-30 09:05:14 +1030261/* Serializes the updates to cpu_online_mask, cpu_present_mask */
Linus Torvaldsaa953872006-07-23 12:12:16 -0700262static DEFINE_MUTEX(cpu_add_remove_lock);
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000263bool cpuhp_tasks_frozen;
264EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700266/*
Srivatsa S. Bhat93ae4f92014-03-11 02:04:14 +0530267 * The following two APIs (cpu_maps_update_begin/done) must be used when
268 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700269 */
270void cpu_maps_update_begin(void)
271{
272 mutex_lock(&cpu_add_remove_lock);
273}
274
275void cpu_maps_update_done(void)
276{
277 mutex_unlock(&cpu_add_remove_lock);
278}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200280/*
281 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700282 * Should always be manipulated under cpu_add_remove_lock
283 */
284static int cpu_hotplug_disabled;
285
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700286#ifdef CONFIG_HOTPLUG_CPU
287
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200288DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530289
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200290void cpus_read_lock(void)
Ashok Raja9d9baa2005-11-28 13:43:46 -0800291{
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200292 percpu_down_read(&cpu_hotplug_lock);
Ashok Raja9d9baa2005-11-28 13:43:46 -0800293}
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200294EXPORT_SYMBOL_GPL(cpus_read_lock);
Ashok Raj90d45d12005-11-08 21:34:24 -0800295
Waiman Long6f4ceee2018-07-24 14:26:04 -0400296int cpus_read_trylock(void)
297{
298 return percpu_down_read_trylock(&cpu_hotplug_lock);
299}
300EXPORT_SYMBOL_GPL(cpus_read_trylock);
301
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200302void cpus_read_unlock(void)
Ashok Raja9d9baa2005-11-28 13:43:46 -0800303{
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200304 percpu_up_read(&cpu_hotplug_lock);
Ashok Raja9d9baa2005-11-28 13:43:46 -0800305}
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200306EXPORT_SYMBOL_GPL(cpus_read_unlock);
Ashok Raja9d9baa2005-11-28 13:43:46 -0800307
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200308void cpus_write_lock(void)
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100309{
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200310 percpu_down_write(&cpu_hotplug_lock);
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100311}
312
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200313void cpus_write_unlock(void)
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100314{
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200315 percpu_up_write(&cpu_hotplug_lock);
316}
317
318void lockdep_assert_cpus_held(void)
319{
Valentin Schneiderce48c4572018-12-19 18:23:15 +0000320 /*
321 * We can't have hotplug operations before userspace starts running,
322 * and some init codepaths will knowingly not take the hotplug lock.
323 * This is all valid, so mute lockdep until it makes sense to report
324 * unheld locks.
325 */
326 if (system_state < SYSTEM_RUNNING)
327 return;
328
Thomas Gleixnerfc8dffd2017-05-24 10:15:40 +0200329 percpu_rwsem_assert_held(&cpu_hotplug_lock);
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100330}
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700331
Peter Zijlstracb921732018-09-11 11:51:27 +0200332static void lockdep_acquire_cpus_lock(void)
333{
334 rwsem_acquire(&cpu_hotplug_lock.rw_sem.dep_map, 0, 0, _THIS_IP_);
335}
336
337static void lockdep_release_cpus_lock(void)
338{
Qian Cai5facae42019-09-19 12:09:40 -0400339 rwsem_release(&cpu_hotplug_lock.rw_sem.dep_map, _THIS_IP_);
Peter Zijlstracb921732018-09-11 11:51:27 +0200340}
341
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700342/*
343 * Wait for currently running CPU hotplug operations to complete (if any) and
344 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
345 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
346 * hotplug path before performing hotplug operations. So acquiring that lock
347 * guarantees mutual exclusion from any currently running hotplug operations.
348 */
349void cpu_hotplug_disable(void)
350{
351 cpu_maps_update_begin();
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -0700352 cpu_hotplug_disabled++;
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700353 cpu_maps_update_done();
354}
Vitaly Kuznetsov32145c42015-08-05 00:52:47 -0700355EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700356
Lianwei Wang01b41152016-06-09 23:43:28 -0700357static void __cpu_hotplug_enable(void)
358{
359 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
360 return;
361 cpu_hotplug_disabled--;
362}
363
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700364void cpu_hotplug_enable(void)
365{
366 cpu_maps_update_begin();
Lianwei Wang01b41152016-06-09 23:43:28 -0700367 __cpu_hotplug_enable();
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700368 cpu_maps_update_done();
369}
Vitaly Kuznetsov32145c42015-08-05 00:52:47 -0700370EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
Peter Zijlstracb921732018-09-11 11:51:27 +0200371
372#else
373
374static void lockdep_acquire_cpus_lock(void)
375{
376}
377
378static void lockdep_release_cpus_lock(void)
379{
380}
381
Toshi Kanib9d10be2013-08-12 09:45:53 -0600382#endif /* CONFIG_HOTPLUG_CPU */
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700383
Thomas Gleixnera74cfff2018-11-25 19:33:39 +0100384/*
385 * Architectures that need SMT-specific errata handling during SMT hotplug
386 * should override this.
387 */
388void __weak arch_smt_update(void) { }
389
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200390#ifdef CONFIG_HOTPLUG_SMT
391enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
Thomas Gleixnerbc2d8d262018-08-07 08:19:57 +0200392
Jiri Kosina8e1b7062018-07-13 16:23:23 +0200393void __init cpu_smt_disable(bool force)
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200394{
Vitaly Kuznetsove1572f12019-09-16 18:22:56 +0200395 if (!cpu_smt_possible())
Jiri Kosina8e1b7062018-07-13 16:23:23 +0200396 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
Josh Poimboeufb2849092019-01-30 07:13:58 -0600409 * CPU identification. Called from architecture code.
Thomas Gleixnerfee0aed2018-07-13 16:23:24 +0200410 */
411void __init cpu_smt_check_topology(void)
412{
Josh Poimboeufb2849092019-01-30 07:13:58 -0600413 if (!topology_smt_supported())
Thomas Gleixnerfee0aed2018-07-13 16:23:24 +0200414 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
415}
416
Jiri Kosina8e1b7062018-07-13 16:23:23 +0200417static int __init smt_cmdline_disable(char *str)
418{
419 cpu_smt_disable(str && !strcmp(str, "force"));
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200420 return 0;
421}
422early_param("nosmt", smt_cmdline_disable);
423
424static inline bool cpu_smt_allowed(unsigned int cpu)
425{
Josh Poimboeufb2849092019-01-30 07:13:58 -0600426 if (cpu_smt_control == CPU_SMT_ENABLED)
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200427 return true;
428
Josh Poimboeufb2849092019-01-30 07:13:58 -0600429 if (topology_is_primary_thread(cpu))
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200430 return true;
431
432 /*
433 * On x86 it's required to boot all logical CPUs at least once so
434 * that the init code can get a chance to set CR4.MCE on each
435 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
436 * core will shutdown the machine.
437 */
Thomas Gleixnere797bda2019-07-22 20:47:16 +0200438 return !cpumask_test_cpu(cpu, &cpus_booted_once_mask);
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200439}
Vitaly Kuznetsove1572f12019-09-16 18:22:56 +0200440
441/* Returns true if SMT is not supported of forcefully (irreversibly) disabled */
442bool cpu_smt_possible(void)
443{
444 return cpu_smt_control != CPU_SMT_FORCE_DISABLED &&
445 cpu_smt_control != CPU_SMT_NOT_SUPPORTED;
446}
447EXPORT_SYMBOL_GPL(cpu_smt_possible);
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200448#else
449static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
450#endif
451
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200452static inline enum cpuhp_state
453cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
454{
455 enum cpuhp_state prev_state = st->state;
456
457 st->rollback = false;
458 st->last = NULL;
459
460 st->target = target;
461 st->single = false;
462 st->bringup = st->state < target;
463
464 return prev_state;
465}
466
467static inline void
468cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
469{
470 st->rollback = true;
471
472 /*
473 * If we have st->last we need to undo partial multi_instance of this
474 * state first. Otherwise start undo at the previous state.
475 */
476 if (!st->last) {
477 if (st->bringup)
478 st->state--;
479 else
480 st->state++;
481 }
482
483 st->target = prev_state;
484 st->bringup = !st->bringup;
485}
486
487/* Regular hotplug invocation of the AP hotplug thread */
488static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
489{
490 if (!st->single && st->state == st->target)
491 return;
492
493 st->result = 0;
494 /*
495 * Make sure the above stores are visible before should_run becomes
496 * true. Paired with the mb() above in cpuhp_thread_fun()
497 */
498 smp_mb();
499 st->should_run = true;
500 wake_up_process(st->thread);
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200501 wait_for_ap_thread(st, st->bringup);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200502}
503
504static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
505{
506 enum cpuhp_state prev_state;
507 int ret;
508
509 prev_state = cpuhp_set_state(st, target);
510 __cpuhp_kick_ap(st);
511 if ((ret = st->result)) {
512 cpuhp_reset_state(st, prev_state);
513 __cpuhp_kick_ap(st);
514 }
515
516 return ret;
517}
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +0200518
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000519static int bringup_wait_for_ap(unsigned int cpu)
520{
521 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
522
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +0200523 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200524 wait_for_ap_thread(st, true);
Thomas Gleixnerdea1d0f2017-07-11 22:06:24 +0200525 if (WARN_ON_ONCE((!cpu_online(cpu))))
526 return -ECANCELED;
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +0200527
Peter Zijlstra45178ac2019-12-10 09:34:54 +0100528 /* Unpark the hotplug thread of the target cpu */
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +0200529 kthread_unpark(st->thread);
530
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200531 /*
532 * SMT soft disabling on X86 requires to bring the CPU out of the
533 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
Jiri Kosinaf5602012019-05-28 21:31:49 +0200534 * CPU marked itself as booted_once in notify_cpu_starting() so the
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +0200535 * cpu_smt_allowed() check will now return false if this is not the
536 * primary sibling.
537 */
538 if (!cpu_smt_allowed(cpu))
539 return -ECANCELED;
540
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200541 if (st->target <= CPUHP_AP_ONLINE_IDLE)
542 return 0;
543
544 return cpuhp_kick_ap(st, st->target);
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000545}
546
Thomas Gleixnerba997462016-02-26 18:43:24 +0000547static int bringup_cpu(unsigned int cpu)
548{
549 struct task_struct *idle = idle_thread_get(cpu);
550 int ret;
551
Boris Ostrovskyaa877172016-08-03 13:22:28 -0400552 /*
553 * Some architectures have to walk the irq descriptors to
554 * setup the vector space for the cpu which comes online.
555 * Prevent irq alloc/free across the bringup.
556 */
557 irq_lock_sparse();
558
Thomas Gleixnerba997462016-02-26 18:43:24 +0000559 /* Arch-specific enabling code. */
560 ret = __cpu_up(cpu, idle);
Boris Ostrovskyaa877172016-08-03 13:22:28 -0400561 irq_unlock_sparse();
Thomas Gleixner530e9b72016-12-21 20:19:53 +0100562 if (ret)
Thomas Gleixnerba997462016-02-26 18:43:24 +0000563 return ret;
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +0200564 return bringup_wait_for_ap(cpu);
Thomas Gleixnerba997462016-02-26 18:43:24 +0000565}
566
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000567/*
568 * Hotplug state machine related functions
569 */
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000570
Thomas Gleixnera7246322016-08-12 19:49:38 +0200571static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000572{
Mukesh Ojha6fb86d92018-08-28 12:24:54 +0530573 for (st->state--; st->state > st->target; st->state--)
574 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000575}
576
Thomas Gleixner206b9232019-03-26 17:36:05 +0100577static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st)
578{
579 if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
580 return true;
581 /*
582 * When CPU hotplug is disabled, then taking the CPU down is not
583 * possible because takedown_cpu() and the architecture and
584 * subsystem specific mechanisms are not available. So the CPU
585 * which would be completely unplugged again needs to stay around
586 * in the current state.
587 */
588 return st->state <= CPUHP_BRINGUP_CPU;
589}
590
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000591static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
Thomas Gleixnera7246322016-08-12 19:49:38 +0200592 enum cpuhp_state target)
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000593{
594 enum cpuhp_state prev_state = st->state;
595 int ret = 0;
596
597 while (st->state < target) {
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000598 st->state++;
Peter Zijlstra96abb962017-09-20 19:00:16 +0200599 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000600 if (ret) {
Thomas Gleixner206b9232019-03-26 17:36:05 +0100601 if (can_rollback_cpu(st)) {
602 st->target = prev_state;
603 undo_cpu_up(cpu, st);
604 }
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000605 break;
606 }
607 }
608 return ret;
609}
610
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000611/*
612 * The cpu hotplug threads manage the bringup and teardown of the cpus
613 */
614static void cpuhp_create(unsigned int cpu)
615{
616 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
617
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200618 init_completion(&st->done_up);
619 init_completion(&st->done_down);
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000620}
621
622static int cpuhp_should_run(unsigned int cpu)
623{
624 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
625
626 return st->should_run;
627}
628
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000629/*
630 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
631 * callbacks when a state gets [un]installed at runtime.
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200632 *
633 * Each invocation of this function by the smpboot thread does a single AP
634 * state callback.
635 *
636 * It has 3 modes of operation:
637 * - single: runs st->cb_state
638 * - up: runs ++st->state, while st->state < st->target
639 * - down: runs st->state--, while st->state > st->target
640 *
641 * When complete or on error, should_run is cleared and the completion is fired.
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000642 */
643static void cpuhp_thread_fun(unsigned int cpu)
644{
645 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200646 bool bringup = st->bringup;
647 enum cpuhp_state state;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000648
Neeraj Upadhyayf8b75302018-09-05 11:22:07 +0530649 if (WARN_ON_ONCE(!st->should_run))
650 return;
651
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000652 /*
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200653 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
654 * that if we see ->should_run we also see the rest of the state.
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000655 */
656 smp_mb();
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200657
Peter Zijlstracb921732018-09-11 11:51:27 +0200658 /*
659 * The BP holds the hotplug lock, but we're now running on the AP,
660 * ensure that anybody asserting the lock is held, will actually find
661 * it so.
662 */
663 lockdep_acquire_cpus_lock();
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +0200664 cpuhp_lock_acquire(bringup);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200665
Thomas Gleixnera7246322016-08-12 19:49:38 +0200666 if (st->single) {
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200667 state = st->cb_state;
668 st->should_run = false;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000669 } else {
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200670 if (bringup) {
671 st->state++;
672 state = st->state;
673 st->should_run = (st->state < st->target);
674 WARN_ON_ONCE(st->state > st->target);
675 } else {
676 state = st->state;
677 st->state--;
678 st->should_run = (st->state > st->target);
679 WARN_ON_ONCE(st->state < st->target);
680 }
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000681 }
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200682
683 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
684
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200685 if (cpuhp_is_atomic_state(state)) {
686 local_irq_disable();
687 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
688 local_irq_enable();
689
690 /*
691 * STARTING/DYING must not fail!
692 */
693 WARN_ON_ONCE(st->result);
694 } else {
695 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
696 }
697
698 if (st->result) {
699 /*
700 * If we fail on a rollback, we're up a creek without no
701 * paddle, no way forward, no way back. We loose, thanks for
702 * playing.
703 */
704 WARN_ON_ONCE(st->rollback);
705 st->should_run = false;
706 }
707
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +0200708 cpuhp_lock_release(bringup);
Peter Zijlstracb921732018-09-11 11:51:27 +0200709 lockdep_release_cpus_lock();
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200710
711 if (!st->should_run)
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200712 complete_ap_thread(st, bringup);
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000713}
714
715/* Invoke a single callback on a remote cpu */
Thomas Gleixnera7246322016-08-12 19:49:38 +0200716static int
Thomas Gleixnercf392d12016-08-12 19:49:39 +0200717cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
718 struct hlist_node *node)
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000719{
720 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200721 int ret;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000722
723 if (!cpu_online(cpu))
724 return 0;
725
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +0200726 cpuhp_lock_acquire(false);
727 cpuhp_lock_release(false);
728
729 cpuhp_lock_acquire(true);
730 cpuhp_lock_release(true);
Thomas Gleixner49dfe2a2017-05-24 10:15:43 +0200731
Thomas Gleixner6a4e2452016-07-13 17:16:03 +0000732 /*
733 * If we are up and running, use the hotplug thread. For early calls
734 * we invoke the thread function directly.
735 */
736 if (!st->thread)
Peter Zijlstra96abb962017-09-20 19:00:16 +0200737 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
Thomas Gleixner6a4e2452016-07-13 17:16:03 +0000738
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200739 st->rollback = false;
740 st->last = NULL;
741
742 st->node = node;
743 st->bringup = bringup;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000744 st->cb_state = state;
Thomas Gleixnera7246322016-08-12 19:49:38 +0200745 st->single = true;
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200746
747 __cpuhp_kick_ap(st);
Thomas Gleixnera7246322016-08-12 19:49:38 +0200748
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000749 /*
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200750 * If we failed and did a partial, do a rollback.
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000751 */
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200752 if ((ret = st->result) && st->last) {
753 st->rollback = true;
754 st->bringup = !bringup;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000755
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200756 __cpuhp_kick_ap(st);
757 }
758
Thomas Gleixner1f7c70d2017-10-21 16:06:52 +0200759 /*
760 * Clean up the leftovers so the next hotplug operation wont use stale
761 * data.
762 */
763 st->node = st->last = NULL;
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200764 return ret;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000765}
766
767static int cpuhp_kick_ap_work(unsigned int cpu)
768{
769 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200770 enum cpuhp_state prev_state = st->state;
771 int ret;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000772
Peter Zijlstra5f4b55e2017-09-20 19:00:20 +0200773 cpuhp_lock_acquire(false);
774 cpuhp_lock_release(false);
775
776 cpuhp_lock_acquire(true);
777 cpuhp_lock_release(true);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200778
779 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
780 ret = cpuhp_kick_ap(st, st->target);
781 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
782
783 return ret;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000784}
785
786static struct smp_hotplug_thread cpuhp_threads = {
787 .store = &cpuhp_state.thread,
788 .create = &cpuhp_create,
789 .thread_should_run = cpuhp_should_run,
790 .thread_fn = cpuhp_thread_fun,
791 .thread_comm = "cpuhp/%u",
792 .selfparking = true,
793};
794
795void __init cpuhp_threads_init(void)
796{
797 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
798 kthread_unpark(this_cpu_read(cpuhp_state.thread));
799}
800
Michal Hocko777c6e02016-12-07 14:54:38 +0100801#ifdef CONFIG_HOTPLUG_CPU
Anton Vorontsove4cc2f82012-05-31 16:26:26 -0700802/**
803 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
804 * @cpu: a CPU id
805 *
806 * This function walks all processes, finds a valid mm struct for each one and
807 * then clears a corresponding bit in mm's cpumask. While this all sounds
808 * trivial, there are various non-obvious corner cases, which this function
809 * tries to solve in a safe manner.
810 *
811 * Also note that the function uses a somewhat relaxed locking scheme, so it may
812 * be called only for an already offlined CPU.
813 */
Anton Vorontsovcb792952012-05-31 16:26:22 -0700814void clear_tasks_mm_cpumask(int cpu)
815{
816 struct task_struct *p;
817
818 /*
819 * This function is called after the cpu is taken down and marked
820 * offline, so its not like new tasks will ever get this cpu set in
821 * their mm mask. -- Peter Zijlstra
822 * Thus, we may use rcu_read_lock() here, instead of grabbing
823 * full-fledged tasklist_lock.
824 */
Anton Vorontsove4cc2f82012-05-31 16:26:26 -0700825 WARN_ON(cpu_online(cpu));
Anton Vorontsovcb792952012-05-31 16:26:22 -0700826 rcu_read_lock();
827 for_each_process(p) {
828 struct task_struct *t;
829
Anton Vorontsove4cc2f82012-05-31 16:26:26 -0700830 /*
831 * Main thread might exit, but other threads may still have
832 * a valid mm. Find one.
833 */
Anton Vorontsovcb792952012-05-31 16:26:22 -0700834 t = find_lock_task_mm(p);
835 if (!t)
836 continue;
837 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
838 task_unlock(t);
839 }
840 rcu_read_unlock();
841}
842
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843/* Take this CPU down. */
Mathias Krause71cf5ae2015-07-19 20:06:22 +0200844static int take_cpu_down(void *_param)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845{
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000846 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
847 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000848 int err, cpu = smp_processor_id();
Peter Zijlstra724a8682017-09-20 19:00:18 +0200849 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 /* Ensure this CPU doesn't handle any more interrupts. */
852 err = __cpu_disable();
853 if (err < 0)
Zwane Mwaikambof3705132005-06-25 14:54:50 -0700854 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Thomas Gleixnera7246322016-08-12 19:49:38 +0200856 /*
857 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
858 * do this step again.
859 */
860 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
861 st->state--;
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000862 /* Invoke the former CPU_DYING callbacks */
Peter Zijlstra724a8682017-09-20 19:00:18 +0200863 for (; st->state > target; st->state--) {
864 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
865 /*
866 * DYING must not fail!
867 */
868 WARN_ON_ONCE(ret);
869 }
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000870
Thomas Gleixner52c063d2015-04-03 02:37:24 +0200871 /* Give up timekeeping duties */
872 tick_handover_do_timer();
Thomas Gleixner1b72d432019-03-21 16:39:20 +0100873 /* Remove CPU from timer broadcasting */
874 tick_offline_cpu(cpu);
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000875 /* Park the stopper thread */
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000876 stop_machine_park(cpu);
Zwane Mwaikambof3705132005-06-25 14:54:50 -0700877 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878}
879
Thomas Gleixner98458172016-02-26 18:43:25 +0000880static int takedown_cpu(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881{
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000882 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Thomas Gleixner98458172016-02-26 18:43:25 +0000883 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Thomas Gleixner2a58c522016-03-10 20:42:08 +0100885 /* Park the smpboot threads */
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000886 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
887
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200888 /*
Thomas Gleixnera8994182015-07-05 17:12:30 +0000889 * Prevent irq alloc/free while the dying cpu reorganizes the
890 * interrupt affinities.
891 */
892 irq_lock_sparse();
893
894 /*
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200895 * So now all preempt/rcu users must observe !cpu_active().
896 */
Sebastian Andrzej Siewior210e2132017-05-24 10:15:28 +0200897 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
Rusty Russell04321582008-07-28 12:16:29 -0500898 if (err) {
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +0200899 /* CPU refused to die */
Thomas Gleixnera8994182015-07-05 17:12:30 +0000900 irq_unlock_sparse();
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +0200901 /* Unpark the hotplug thread so we can rollback there */
902 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
Thomas Gleixner98458172016-02-26 18:43:25 +0000903 return err;
Satoru Takeuchi8fa1d7d2006-10-28 10:38:57 -0700904 }
Rusty Russell04321582008-07-28 12:16:29 -0500905 BUG_ON(cpu_online(cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Peter Zijlstra48c5ccae2010-11-13 19:32:29 +0100907 /*
Brendan Jackman5b1ead62017-12-06 10:59:11 +0000908 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
909 * all runnable tasks from the CPU, there's only the idle task left now
Peter Zijlstra48c5ccae2010-11-13 19:32:29 +0100910 * that the migration thread is done doing the stop_machine thing.
Peter Zijlstra51a96c72010-11-19 20:37:53 +0100911 *
912 * Wait for the stop thread to go away.
Peter Zijlstra48c5ccae2010-11-13 19:32:29 +0100913 */
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200914 wait_for_ap_thread(st, false);
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000915 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Thomas Gleixnera8994182015-07-05 17:12:30 +0000917 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
918 irq_unlock_sparse();
919
Preeti U Murthy345527b2015-03-30 14:59:19 +0530920 hotplug_cpu__broadcast_tick_pull(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 /* This actually kills the CPU. */
922 __cpu_die(cpu);
923
Thomas Gleixnera49b1162015-04-03 02:38:05 +0200924 tick_cleanup_dead_cpu(cpu);
Paul E. McKenneya58163d2017-06-20 12:11:34 -0700925 rcutree_migrate_callbacks(cpu);
Thomas Gleixner98458172016-02-26 18:43:25 +0000926 return 0;
927}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
Thomas Gleixner71f87b22016-03-03 10:52:10 +0100929static void cpuhp_complete_idle_dead(void *arg)
930{
931 struct cpuhp_cpu_state *st = arg;
932
Peter Zijlstra5ebe7742017-09-20 19:00:19 +0200933 complete_ap_thread(st, false);
Thomas Gleixner71f87b22016-03-03 10:52:10 +0100934}
935
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000936void cpuhp_report_idle_dead(void)
937{
938 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
939
940 BUG_ON(st->state != CPUHP_AP_OFFLINE);
Thomas Gleixner27d50c72016-02-26 18:43:44 +0000941 rcu_report_dead(smp_processor_id());
Thomas Gleixner71f87b22016-03-03 10:52:10 +0100942 st->state = CPUHP_AP_IDLE_DEAD;
943 /*
944 * We cannot call complete after rcu_report_dead() so we delegate it
945 * to an online cpu.
946 */
947 smp_call_function_single(cpumask_first(cpu_online_mask),
948 cpuhp_complete_idle_dead, st, 0);
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000949}
950
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200951static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
952{
Mukesh Ojha6fb86d92018-08-28 12:24:54 +0530953 for (st->state++; st->state < st->target; st->state++)
954 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200955}
956
957static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
958 enum cpuhp_state target)
959{
960 enum cpuhp_state prev_state = st->state;
961 int ret = 0;
962
963 for (; st->state > target; st->state--) {
964 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
965 if (ret) {
966 st->target = prev_state;
Thomas Gleixner69fa6eb2018-09-06 15:21:38 +0200967 if (st->state < prev_state)
968 undo_cpu_down(cpu, st);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200969 break;
970 }
971 }
972 return ret;
973}
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000974
Thomas Gleixner98458172016-02-26 18:43:25 +0000975/* Requires cpu_add_remove_lock to be held */
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000976static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
977 enum cpuhp_state target)
Thomas Gleixner98458172016-02-26 18:43:25 +0000978{
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000979 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
980 int prev_state, ret = 0;
Thomas Gleixner98458172016-02-26 18:43:25 +0000981
982 if (num_online_cpus() == 1)
983 return -EBUSY;
984
Thomas Gleixner757c9892016-02-26 18:43:32 +0000985 if (!cpu_present(cpu))
Thomas Gleixner98458172016-02-26 18:43:25 +0000986 return -EINVAL;
987
Thomas Gleixner8f553c42017-05-24 10:15:12 +0200988 cpus_write_lock();
Thomas Gleixner98458172016-02-26 18:43:25 +0000989
990 cpuhp_tasks_frozen = tasks_frozen;
991
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200992 prev_state = cpuhp_set_state(st, target);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000993 /*
994 * If the current CPU state is in the range of the AP hotplug thread,
995 * then we need to kick the thread.
996 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000997 if (st->state > CPUHP_TEARDOWN_CPU) {
Peter Zijlstra4dddfb52017-09-20 19:00:17 +0200998 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000999 ret = cpuhp_kick_ap_work(cpu);
1000 /*
1001 * The AP side has done the error rollback already. Just
1002 * return the error code..
1003 */
1004 if (ret)
1005 goto out;
1006
1007 /*
1008 * We might have stopped still in the range of the AP hotplug
1009 * thread. Nothing to do anymore.
1010 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001011 if (st->state > CPUHP_TEARDOWN_CPU)
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001012 goto out;
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001013
1014 st->target = target;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001015 }
1016 /*
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001017 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001018 * to do the further cleanups.
1019 */
Thomas Gleixnera7246322016-08-12 19:49:38 +02001020 ret = cpuhp_down_callbacks(cpu, st, target);
Thomas Gleixner69fa6eb2018-09-06 15:21:38 +02001021 if (ret && st->state == CPUHP_TEARDOWN_CPU && st->state < prev_state) {
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001022 cpuhp_reset_state(st, prev_state);
1023 __cpuhp_kick_ap(st);
Sebastian Andrzej Siewior3b9d6da2016-04-08 14:40:15 +02001024 }
Thomas Gleixner98458172016-02-26 18:43:25 +00001025
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001026out:
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001027 cpus_write_unlock();
Thomas Gleixner941154b2017-09-12 21:37:04 +02001028 /*
1029 * Do post unplug cleanup. This is still protected against
1030 * concurrent CPU hotplug via cpu_add_remove_lock.
1031 */
1032 lockup_detector_cleanup();
Thomas Gleixnera74cfff2018-11-25 19:33:39 +01001033 arch_smt_update();
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001034 return ret;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001035}
1036
Thomas Gleixnercc1fe212018-05-29 17:49:05 +02001037static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1038{
1039 if (cpu_hotplug_disabled)
1040 return -EBUSY;
1041 return _cpu_down(cpu, 0, target);
1042}
1043
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001044static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001045{
Heiko Carstens9ea09af2008-12-22 12:36:30 +01001046 int err;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001047
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001048 cpu_maps_update_begin();
Thomas Gleixnercc1fe212018-05-29 17:49:05 +02001049 err = cpu_down_maps_locked(cpu, target);
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001050 cpu_maps_update_done();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 return err;
1052}
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001053
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001054int cpu_down(unsigned int cpu)
1055{
1056 return do_cpu_down(cpu, CPUHP_OFFLINE);
1057}
Zhang Ruib62b8ef2008-04-29 02:35:56 -04001058EXPORT_SYMBOL(cpu_down);
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001059
1060#else
1061#define takedown_cpu NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062#endif /*CONFIG_HOTPLUG_CPU*/
1063
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001064/**
Thomas Gleixneree1e7142016-08-18 14:57:16 +02001065 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001066 * @cpu: cpu that just started
1067 *
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001068 * It must be called by the arch code on the new cpu, before the new cpu
1069 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1070 */
1071void notify_cpu_starting(unsigned int cpu)
1072{
1073 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1074 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
Peter Zijlstra724a8682017-09-20 19:00:18 +02001075 int ret;
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001076
Sebastian Andrzej Siewior0c6d4572016-08-17 14:21:04 +02001077 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
Thomas Gleixnere797bda2019-07-22 20:47:16 +02001078 cpumask_set_cpu(cpu, &cpus_booted_once_mask);
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001079 while (st->state < target) {
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001080 st->state++;
Peter Zijlstra724a8682017-09-20 19:00:18 +02001081 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1082 /*
1083 * STARTING must not fail!
1084 */
1085 WARN_ON_ONCE(ret);
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001086 }
1087}
1088
Thomas Gleixner949338e2016-02-26 18:43:35 +00001089/*
Thomas Gleixner9cd4f1a2017-07-04 22:20:23 +02001090 * Called from the idle task. Wake up the controlling task which brings the
Peter Zijlstra45178ac2019-12-10 09:34:54 +01001091 * hotplug thread of the upcoming CPU up and then delegates the rest of the
1092 * online bringup to the hotplug thread.
Thomas Gleixner949338e2016-02-26 18:43:35 +00001093 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001094void cpuhp_online_idle(enum cpuhp_state state)
Thomas Gleixner949338e2016-02-26 18:43:35 +00001095{
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001096 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001097
1098 /* Happens for the boot cpu */
1099 if (state != CPUHP_AP_ONLINE_IDLE)
1100 return;
1101
Peter Zijlstra45178ac2019-12-10 09:34:54 +01001102 /*
1103 * Unpart the stopper thread before we start the idle loop (and start
1104 * scheduling); this ensures the stopper task is always available.
1105 */
1106 stop_machine_unpark(smp_processor_id());
1107
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001108 st->state = CPUHP_AP_ONLINE_IDLE;
Peter Zijlstra5ebe7742017-09-20 19:00:19 +02001109 complete_ap_thread(st, true);
Thomas Gleixner949338e2016-02-26 18:43:35 +00001110}
1111
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001112/* Requires cpu_add_remove_lock to be held */
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001113static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114{
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001115 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Suresh Siddha3bb5d2e2012-04-20 17:08:50 -07001116 struct task_struct *idle;
Thomas Gleixner2e1a3482016-02-26 18:43:37 +00001117 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001119 cpus_write_lock();
Thomas Gleixner38498a62012-04-20 13:05:44 +00001120
Thomas Gleixner757c9892016-02-26 18:43:32 +00001121 if (!cpu_present(cpu)) {
Yasuaki Ishimatsu5e5041f2012-10-23 01:30:54 +02001122 ret = -EINVAL;
1123 goto out;
1124 }
1125
Thomas Gleixner757c9892016-02-26 18:43:32 +00001126 /*
1127 * The caller of do_cpu_up might have raced with another
1128 * caller. Ignore it for now.
1129 */
1130 if (st->state >= target)
Thomas Gleixner38498a62012-04-20 13:05:44 +00001131 goto out;
Thomas Gleixner757c9892016-02-26 18:43:32 +00001132
1133 if (st->state == CPUHP_OFFLINE) {
1134 /* Let it fail before we try to bring the cpu up */
1135 idle = idle_thread_get(cpu);
1136 if (IS_ERR(idle)) {
1137 ret = PTR_ERR(idle);
1138 goto out;
1139 }
Suresh Siddha3bb5d2e2012-04-20 17:08:50 -07001140 }
Thomas Gleixner38498a62012-04-20 13:05:44 +00001141
Thomas Gleixnerba997462016-02-26 18:43:24 +00001142 cpuhp_tasks_frozen = tasks_frozen;
1143
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001144 cpuhp_set_state(st, target);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001145 /*
1146 * If the current CPU state is in the range of the AP hotplug thread,
1147 * then we need to kick the thread once more.
1148 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001149 if (st->state > CPUHP_BRINGUP_CPU) {
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001150 ret = cpuhp_kick_ap_work(cpu);
1151 /*
1152 * The AP side has done the error rollback already. Just
1153 * return the error code..
1154 */
1155 if (ret)
1156 goto out;
1157 }
1158
1159 /*
1160 * Try to reach the target state. We max out on the BP at
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001161 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001162 * responsible for bringing it up to the target state.
1163 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001164 target = min((int)target, CPUHP_BRINGUP_CPU);
Thomas Gleixnera7246322016-08-12 19:49:38 +02001165 ret = cpuhp_up_callbacks(cpu, st, target);
Thomas Gleixner38498a62012-04-20 13:05:44 +00001166out:
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001167 cpus_write_unlock();
Thomas Gleixnera74cfff2018-11-25 19:33:39 +01001168 arch_smt_update();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 return ret;
1170}
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001171
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001172static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001173{
1174 int err = 0;
minskey guocf234222010-05-24 14:32:41 -07001175
Rusty Russelle0b582e2009-01-01 10:12:28 +10301176 if (!cpu_possible(cpu)) {
Fabian Frederick84117da2014-06-04 16:11:17 -07001177 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1178 cpu);
Chen Gong87d5e0232010-03-05 13:42:38 -08001179#if defined(CONFIG_IA64)
Fabian Frederick84117da2014-06-04 16:11:17 -07001180 pr_err("please check additional_cpus= boot parameter\n");
KAMEZAWA Hiroyuki73e753a2007-10-18 23:40:47 -07001181#endif
1182 return -EINVAL;
1183 }
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001184
Toshi Kani01b0f192013-11-12 15:07:25 -08001185 err = try_online_node(cpu_to_node(cpu));
1186 if (err)
1187 return err;
minskey guocf234222010-05-24 14:32:41 -07001188
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001189 cpu_maps_update_begin();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001190
Max Krasnyanskye761b772008-07-15 04:43:49 -07001191 if (cpu_hotplug_disabled) {
1192 err = -EBUSY;
1193 goto out;
1194 }
Thomas Gleixner05736e42018-05-29 17:48:27 +02001195 if (!cpu_smt_allowed(cpu)) {
1196 err = -EPERM;
1197 goto out;
1198 }
Max Krasnyanskye761b772008-07-15 04:43:49 -07001199
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001200 err = _cpu_up(cpu, 0, target);
Max Krasnyanskye761b772008-07-15 04:43:49 -07001201out:
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001202 cpu_maps_update_done();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001203 return err;
1204}
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001205
1206int cpu_up(unsigned int cpu)
1207{
1208 return do_cpu_up(cpu, CPUHP_ONLINE);
1209}
Paul E. McKenneya513f6b2011-12-11 21:54:45 -08001210EXPORT_SYMBOL_GPL(cpu_up);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001211
Rafael J. Wysockif3de4be2007-08-30 23:56:29 -07001212#ifdef CONFIG_PM_SLEEP_SMP
Rusty Russelle0b582e2009-01-01 10:12:28 +10301213static cpumask_var_t frozen_cpus;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001214
James Morsed391e552016-08-17 13:50:25 +01001215int freeze_secondary_cpus(int primary)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001216{
James Morsed391e552016-08-17 13:50:25 +01001217 int cpu, error = 0;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001218
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001219 cpu_maps_update_begin();
Nicholas Piggin9ca12ac2019-04-11 13:34:46 +10001220 if (primary == -1) {
James Morsed391e552016-08-17 13:50:25 +01001221 primary = cpumask_first(cpu_online_mask);
Nicholas Piggin9ca12ac2019-04-11 13:34:46 +10001222 if (!housekeeping_cpu(primary, HK_FLAG_TIMER))
1223 primary = housekeeping_any_cpu(HK_FLAG_TIMER);
1224 } else {
1225 if (!cpu_online(primary))
1226 primary = cpumask_first(cpu_online_mask);
1227 }
1228
Xiaotian Feng9ee349a2009-12-16 18:04:32 +01001229 /*
1230 * We take down all of the non-boot CPUs in one shot to avoid races
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001231 * with the userspace trying to use the CPU hotplug at the same time
1232 */
Rusty Russelle0b582e2009-01-01 10:12:28 +10301233 cpumask_clear(frozen_cpus);
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01001234
Fabian Frederick84117da2014-06-04 16:11:17 -07001235 pr_info("Disabling non-boot CPUs ...\n");
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001236 for_each_online_cpu(cpu) {
James Morsed391e552016-08-17 13:50:25 +01001237 if (cpu == primary)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001238 continue;
Pavankumar Kondetia66d9552019-06-03 10:01:03 +05301239
1240 if (pm_wakeup_pending()) {
1241 pr_info("Wakeup pending. Abort CPU freeze\n");
1242 error = -EBUSY;
1243 break;
1244 }
1245
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001246 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001247 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001248 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
Mike Travisfeae3202009-11-17 18:22:13 -06001249 if (!error)
Rusty Russelle0b582e2009-01-01 10:12:28 +10301250 cpumask_set_cpu(cpu, frozen_cpus);
Mike Travisfeae3202009-11-17 18:22:13 -06001251 else {
Fabian Frederick84117da2014-06-04 16:11:17 -07001252 pr_err("Error taking CPU%d down: %d\n", cpu, error);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001253 break;
1254 }
1255 }
Joseph Cihula86886e52009-06-30 19:31:07 -07001256
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001257 if (!error)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001258 BUG_ON(num_online_cpus() > 1);
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001259 else
Fabian Frederick84117da2014-06-04 16:11:17 -07001260 pr_err("Non-boot CPUs are not disabled\n");
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001261
1262 /*
1263 * Make sure the CPUs won't be enabled by someone else. We need to do
1264 * this even in case of failure as all disable_nonboot_cpus() users are
1265 * supposed to do enable_nonboot_cpus() on the failure path.
1266 */
1267 cpu_hotplug_disabled++;
1268
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001269 cpu_maps_update_done();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001270 return error;
1271}
1272
Suresh Siddhad0af9ee2009-08-19 18:05:36 -07001273void __weak arch_enable_nonboot_cpus_begin(void)
1274{
1275}
1276
1277void __weak arch_enable_nonboot_cpus_end(void)
1278{
1279}
1280
Mathias Krause71cf5ae2015-07-19 20:06:22 +02001281void enable_nonboot_cpus(void)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001282{
1283 int cpu, error;
1284
1285 /* Allow everyone to use the CPU hotplug again */
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001286 cpu_maps_update_begin();
Lianwei Wang01b41152016-06-09 23:43:28 -07001287 __cpu_hotplug_enable();
Rusty Russelle0b582e2009-01-01 10:12:28 +10301288 if (cpumask_empty(frozen_cpus))
Rafael J. Wysocki1d64b9c2007-04-01 23:49:49 -07001289 goto out;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001290
Fabian Frederick84117da2014-06-04 16:11:17 -07001291 pr_info("Enabling non-boot CPUs ...\n");
Suresh Siddhad0af9ee2009-08-19 18:05:36 -07001292
1293 arch_enable_nonboot_cpus_begin();
1294
Rusty Russelle0b582e2009-01-01 10:12:28 +10301295 for_each_cpu(cpu, frozen_cpus) {
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001296 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001297 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001298 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001299 if (!error) {
Fabian Frederick84117da2014-06-04 16:11:17 -07001300 pr_info("CPU%d is up\n", cpu);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001301 continue;
1302 }
Fabian Frederick84117da2014-06-04 16:11:17 -07001303 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001304 }
Suresh Siddhad0af9ee2009-08-19 18:05:36 -07001305
1306 arch_enable_nonboot_cpus_end();
1307
Rusty Russelle0b582e2009-01-01 10:12:28 +10301308 cpumask_clear(frozen_cpus);
Rafael J. Wysocki1d64b9c2007-04-01 23:49:49 -07001309out:
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001310 cpu_maps_update_done();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001311}
Rusty Russelle0b582e2009-01-01 10:12:28 +10301312
Fenghua Yud7268a32011-11-15 21:59:31 +01001313static int __init alloc_frozen_cpus(void)
Rusty Russelle0b582e2009-01-01 10:12:28 +10301314{
1315 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1316 return -ENOMEM;
1317 return 0;
1318}
1319core_initcall(alloc_frozen_cpus);
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001320
1321/*
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001322 * When callbacks for CPU hotplug notifications are being executed, we must
1323 * ensure that the state of the system with respect to the tasks being frozen
1324 * or not, as reported by the notification, remains unchanged *throughout the
1325 * duration* of the execution of the callbacks.
1326 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1327 *
1328 * This synchronization is implemented by mutually excluding regular CPU
1329 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1330 * Hibernate notifications.
1331 */
1332static int
1333cpu_hotplug_pm_callback(struct notifier_block *nb,
1334 unsigned long action, void *ptr)
1335{
1336 switch (action) {
1337
1338 case PM_SUSPEND_PREPARE:
1339 case PM_HIBERNATION_PREPARE:
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -07001340 cpu_hotplug_disable();
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001341 break;
1342
1343 case PM_POST_SUSPEND:
1344 case PM_POST_HIBERNATION:
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -07001345 cpu_hotplug_enable();
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001346 break;
1347
1348 default:
1349 return NOTIFY_DONE;
1350 }
1351
1352 return NOTIFY_OK;
1353}
1354
1355
Fenghua Yud7268a32011-11-15 21:59:31 +01001356static int __init cpu_hotplug_pm_sync_init(void)
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001357{
Fenghua Yu6e32d472012-11-13 11:32:43 -08001358 /*
1359 * cpu_hotplug_pm_callback has higher priority than x86
1360 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1361 * to disable cpu hotplug to avoid cpu hotplug race.
1362 */
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001363 pm_notifier(cpu_hotplug_pm_callback, 0);
1364 return 0;
1365}
1366core_initcall(cpu_hotplug_pm_sync_init);
1367
Rafael J. Wysockif3de4be2007-08-30 23:56:29 -07001368#endif /* CONFIG_PM_SLEEP_SMP */
Max Krasnyansky68f4f1e2008-05-29 11:17:02 -07001369
Peter Zijlstra8ce371f2017-03-20 12:26:55 +01001370int __boot_cpu_id;
1371
Max Krasnyansky68f4f1e2008-05-29 11:17:02 -07001372#endif /* CONFIG_SMP */
Mike Travisb8d317d2008-07-24 18:21:29 -07001373
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001374/* Boot processor state steps */
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +08001375static struct cpuhp_step cpuhp_hp_states[] = {
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001376 [CPUHP_OFFLINE] = {
1377 .name = "offline",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001378 .startup.single = NULL,
1379 .teardown.single = NULL,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001380 },
1381#ifdef CONFIG_SMP
1382 [CPUHP_CREATE_THREADS]= {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001383 .name = "threads:prepare",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001384 .startup.single = smpboot_create_threads,
1385 .teardown.single = NULL,
Thomas Gleixner757c9892016-02-26 18:43:32 +00001386 .cant_stop = true,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001387 },
Thomas Gleixner00e16c32016-07-13 17:16:09 +00001388 [CPUHP_PERF_PREPARE] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001389 .name = "perf:prepare",
1390 .startup.single = perf_event_init_cpu,
1391 .teardown.single = perf_event_exit_cpu,
Thomas Gleixner00e16c32016-07-13 17:16:09 +00001392 },
Thomas Gleixner7ee681b2016-07-13 17:16:29 +00001393 [CPUHP_WORKQUEUE_PREP] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001394 .name = "workqueue:prepare",
1395 .startup.single = workqueue_prepare_cpu,
1396 .teardown.single = NULL,
Thomas Gleixner7ee681b2016-07-13 17:16:29 +00001397 },
Thomas Gleixner27590dc2016-07-15 10:41:04 +02001398 [CPUHP_HRTIMERS_PREPARE] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001399 .name = "hrtimers:prepare",
1400 .startup.single = hrtimers_prepare_cpu,
1401 .teardown.single = hrtimers_dead_cpu,
Thomas Gleixner27590dc2016-07-15 10:41:04 +02001402 },
Richard Weinberger31487f82016-07-13 17:17:01 +00001403 [CPUHP_SMPCFD_PREPARE] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001404 .name = "smpcfd:prepare",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001405 .startup.single = smpcfd_prepare_cpu,
1406 .teardown.single = smpcfd_dead_cpu,
Richard Weinberger31487f82016-07-13 17:17:01 +00001407 },
Richard Weinbergere6d49892016-08-18 14:57:17 +02001408 [CPUHP_RELAY_PREPARE] = {
1409 .name = "relay:prepare",
1410 .startup.single = relay_prepare_cpu,
1411 .teardown.single = NULL,
1412 },
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001413 [CPUHP_SLAB_PREPARE] = {
1414 .name = "slab:prepare",
1415 .startup.single = slab_prepare_cpu,
1416 .teardown.single = slab_dead_cpu,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001417 },
Thomas Gleixner4df83742016-07-13 17:17:03 +00001418 [CPUHP_RCUTREE_PREP] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001419 .name = "RCU/tree:prepare",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001420 .startup.single = rcutree_prepare_cpu,
1421 .teardown.single = rcutree_dead_cpu,
Thomas Gleixner4df83742016-07-13 17:17:03 +00001422 },
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001423 /*
Richard Cochran4fae16d2016-07-27 11:08:18 +02001424 * On the tear-down path, timers_dead_cpu() must be invoked
1425 * before blk_mq_queue_reinit_notify() from notify_dead(),
1426 * otherwise a RCU stall occurs.
1427 */
Thomas Gleixner26456f82017-12-27 21:37:25 +01001428 [CPUHP_TIMERS_PREPARE] = {
Mukesh Ojhad0180312018-07-24 20:17:48 +05301429 .name = "timers:prepare",
Thomas Gleixner26456f82017-12-27 21:37:25 +01001430 .startup.single = timers_prepare_cpu,
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001431 .teardown.single = timers_dead_cpu,
Richard Cochran4fae16d2016-07-27 11:08:18 +02001432 },
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001433 /* Kicks the plugged cpu into life */
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001434 [CPUHP_BRINGUP_CPU] = {
1435 .name = "cpu:bringup",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001436 .startup.single = bringup_cpu,
1437 .teardown.single = NULL,
Thomas Gleixner757c9892016-02-26 18:43:32 +00001438 .cant_stop = true,
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001439 },
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001440 /* Final state before CPU kills itself */
1441 [CPUHP_AP_IDLE_DEAD] = {
1442 .name = "idle:dead",
1443 },
1444 /*
1445 * Last state before CPU enters the idle loop to die. Transient state
1446 * for synchronization.
1447 */
1448 [CPUHP_AP_OFFLINE] = {
1449 .name = "ap:offline",
1450 .cant_stop = true,
1451 },
Thomas Gleixner9cf72432016-03-10 12:54:09 +01001452 /* First state is scheduler control. Interrupts are disabled */
1453 [CPUHP_AP_SCHED_STARTING] = {
1454 .name = "sched:starting",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001455 .startup.single = sched_cpu_starting,
1456 .teardown.single = sched_cpu_dying,
Thomas Gleixner9cf72432016-03-10 12:54:09 +01001457 },
Thomas Gleixner4df83742016-07-13 17:17:03 +00001458 [CPUHP_AP_RCUTREE_DYING] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001459 .name = "RCU/tree:dying",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001460 .startup.single = NULL,
1461 .teardown.single = rcutree_dying_cpu,
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001462 },
Lai Jiangshan46febd32017-11-28 21:19:53 +08001463 [CPUHP_AP_SMPCFD_DYING] = {
1464 .name = "smpcfd:dying",
1465 .startup.single = NULL,
1466 .teardown.single = smpcfd_dying_cpu,
1467 },
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001468 /* Entry state on starting. Interrupts enabled from here on. Transient
1469 * state for synchronsization */
1470 [CPUHP_AP_ONLINE] = {
1471 .name = "ap:online",
1472 },
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +08001473 /*
1474 * Handled on controll processor until the plugged processor manages
1475 * this itself.
1476 */
1477 [CPUHP_TEARDOWN_CPU] = {
1478 .name = "cpu:teardown",
1479 .startup.single = NULL,
1480 .teardown.single = takedown_cpu,
1481 .cant_stop = true,
1482 },
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001483 /* Handle smpboot threads park/unpark */
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001484 [CPUHP_AP_SMPBOOT_THREADS] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001485 .name = "smpboot/threads:online",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001486 .startup.single = smpboot_unpark_threads,
Thomas Gleixnerc4de6562018-05-29 19:05:25 +02001487 .teardown.single = smpboot_park_threads,
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001488 },
Thomas Gleixnerc5cb83b2017-06-20 01:37:51 +02001489 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1490 .name = "irq/affinity:online",
1491 .startup.single = irq_affinity_online_cpu,
1492 .teardown.single = NULL,
1493 },
Thomas Gleixner00e16c32016-07-13 17:16:09 +00001494 [CPUHP_AP_PERF_ONLINE] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001495 .name = "perf:online",
1496 .startup.single = perf_event_init_cpu,
1497 .teardown.single = perf_event_exit_cpu,
Thomas Gleixner00e16c32016-07-13 17:16:09 +00001498 },
Peter Zijlstra9cf57732018-06-07 10:52:03 +02001499 [CPUHP_AP_WATCHDOG_ONLINE] = {
1500 .name = "lockup_detector:online",
1501 .startup.single = lockup_detector_online_cpu,
1502 .teardown.single = lockup_detector_offline_cpu,
1503 },
Thomas Gleixner7ee681b2016-07-13 17:16:29 +00001504 [CPUHP_AP_WORKQUEUE_ONLINE] = {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001505 .name = "workqueue:online",
1506 .startup.single = workqueue_online_cpu,
1507 .teardown.single = workqueue_offline_cpu,
Thomas Gleixner7ee681b2016-07-13 17:16:29 +00001508 },
Thomas Gleixner4df83742016-07-13 17:17:03 +00001509 [CPUHP_AP_RCUTREE_ONLINE] = {
Thomas Gleixner677f6642016-09-06 16:13:48 +02001510 .name = "RCU/tree:online",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001511 .startup.single = rcutree_online_cpu,
1512 .teardown.single = rcutree_offline_cpu,
Thomas Gleixner4df83742016-07-13 17:17:03 +00001513 },
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001514#endif
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001515 /*
1516 * The dynamically registered state space is here
1517 */
1518
Thomas Gleixneraaddd7d2016-03-10 12:54:19 +01001519#ifdef CONFIG_SMP
1520 /* Last state is scheduler control setting the cpu active */
1521 [CPUHP_AP_ACTIVE] = {
1522 .name = "sched:active",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001523 .startup.single = sched_cpu_activate,
1524 .teardown.single = sched_cpu_deactivate,
Thomas Gleixneraaddd7d2016-03-10 12:54:19 +01001525 },
1526#endif
1527
Thomas Gleixnerd10ef6f2016-03-08 10:36:13 +01001528 /* CPU is fully up and running. */
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001529 [CPUHP_ONLINE] = {
1530 .name = "online",
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001531 .startup.single = NULL,
1532 .teardown.single = NULL,
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001533 },
1534};
1535
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001536/* Sanity check for callbacks */
1537static int cpuhp_cb_check(enum cpuhp_state state)
1538{
1539 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1540 return -EINVAL;
1541 return 0;
1542}
1543
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001544/*
1545 * Returns a free for dynamic slot assignment of the Online state. The states
1546 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1547 * by having no name assigned.
1548 */
1549static int cpuhp_reserve_state(enum cpuhp_state state)
1550{
Thomas Gleixner4205e472017-01-10 14:01:05 +01001551 enum cpuhp_state i, end;
1552 struct cpuhp_step *step;
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001553
Thomas Gleixner4205e472017-01-10 14:01:05 +01001554 switch (state) {
1555 case CPUHP_AP_ONLINE_DYN:
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +08001556 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
Thomas Gleixner4205e472017-01-10 14:01:05 +01001557 end = CPUHP_AP_ONLINE_DYN_END;
1558 break;
1559 case CPUHP_BP_PREPARE_DYN:
Lai Jiangshan17a2f1c2017-12-01 21:50:05 +08001560 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
Thomas Gleixner4205e472017-01-10 14:01:05 +01001561 end = CPUHP_BP_PREPARE_DYN_END;
1562 break;
1563 default:
1564 return -EINVAL;
1565 }
1566
1567 for (i = state; i <= end; i++, step++) {
1568 if (!step->name)
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001569 return i;
1570 }
1571 WARN(1, "No more dynamic states available for CPU hotplug\n");
1572 return -ENOSPC;
1573}
1574
1575static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1576 int (*startup)(unsigned int cpu),
1577 int (*teardown)(unsigned int cpu),
1578 bool multi_instance)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001579{
1580 /* (Un)Install the callbacks for further cpu hotplug operations */
1581 struct cpuhp_step *sp;
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001582 int ret = 0;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001583
Ethan Barnes0c96b272017-07-19 22:36:00 +00001584 /*
1585 * If name is NULL, then the state gets removed.
1586 *
1587 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1588 * the first allocation from these dynamic ranges, so the removal
1589 * would trigger a new allocation and clear the wrong (already
1590 * empty) state, leaving the callbacks of the to be cleared state
1591 * dangling, which causes wreckage on the next hotplug operation.
1592 */
1593 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1594 state == CPUHP_BP_PREPARE_DYN)) {
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001595 ret = cpuhp_reserve_state(state);
1596 if (ret < 0)
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001597 return ret;
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001598 state = ret;
1599 }
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001600 sp = cpuhp_get_step(state);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001601 if (name && sp->name)
1602 return -EBUSY;
1603
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001604 sp->startup.single = startup;
1605 sp->teardown.single = teardown;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001606 sp->name = name;
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001607 sp->multi_instance = multi_instance;
1608 INIT_HLIST_HEAD(&sp->list);
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001609 return ret;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001610}
1611
1612static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1613{
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001614 return cpuhp_get_step(state)->teardown.single;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001615}
1616
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001617/*
1618 * Call the startup/teardown function for a step either on the AP or
1619 * on the current CPU.
1620 */
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001621static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1622 struct hlist_node *node)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001623{
Thomas Gleixnera7246322016-08-12 19:49:38 +02001624 struct cpuhp_step *sp = cpuhp_get_step(state);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001625 int ret;
1626
Peter Zijlstra4dddfb52017-09-20 19:00:17 +02001627 /*
1628 * If there's nothing to do, we done.
1629 * Relies on the union for multi_instance.
1630 */
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001631 if ((bringup && !sp->startup.single) ||
1632 (!bringup && !sp->teardown.single))
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001633 return 0;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001634 /*
1635 * The non AP bound callbacks can fail on bringup. On teardown
1636 * e.g. module removal we crash for now.
1637 */
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001638#ifdef CONFIG_SMP
1639 if (cpuhp_is_ap_state(state))
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001640 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001641 else
Peter Zijlstra96abb962017-09-20 19:00:16 +02001642 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001643#else
Peter Zijlstra96abb962017-09-20 19:00:16 +02001644 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001645#endif
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001646 BUG_ON(ret && !bringup);
1647 return ret;
1648}
1649
1650/*
1651 * Called from __cpuhp_setup_state on a recoverable failure.
1652 *
1653 * Note: The teardown callbacks for rollback are not allowed to fail!
1654 */
1655static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001656 struct hlist_node *node)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001657{
1658 int cpu;
1659
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001660 /* Roll back the already executed steps on the other cpus */
1661 for_each_present_cpu(cpu) {
1662 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1663 int cpustate = st->state;
1664
1665 if (cpu >= failedcpu)
1666 break;
1667
1668 /* Did we invoke the startup call on that cpu ? */
1669 if (cpustate >= state)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001670 cpuhp_issue_call(cpu, state, false, node);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001671 }
1672}
1673
Thomas Gleixner9805c672017-05-24 10:15:15 +02001674int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1675 struct hlist_node *node,
1676 bool invoke)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001677{
1678 struct cpuhp_step *sp;
1679 int cpu;
1680 int ret;
1681
Thomas Gleixner9805c672017-05-24 10:15:15 +02001682 lockdep_assert_cpus_held();
1683
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001684 sp = cpuhp_get_step(state);
1685 if (sp->multi_instance == false)
1686 return -EINVAL;
1687
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001688 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001689
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001690 if (!invoke || !sp->startup.multi)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001691 goto add_node;
1692
1693 /*
1694 * Try to call the startup callback for each present cpu
1695 * depending on the hotplug state of the cpu.
1696 */
1697 for_each_present_cpu(cpu) {
1698 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1699 int cpustate = st->state;
1700
1701 if (cpustate < state)
1702 continue;
1703
1704 ret = cpuhp_issue_call(cpu, state, true, node);
1705 if (ret) {
Thomas Gleixner3c1627e2016-09-05 15:28:36 +02001706 if (sp->teardown.multi)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001707 cpuhp_rollback_install(cpu, state, node);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001708 goto unlock;
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001709 }
1710 }
1711add_node:
1712 ret = 0;
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001713 hlist_add_head(node, &sp->list);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001714unlock:
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001715 mutex_unlock(&cpuhp_state_mutex);
Thomas Gleixner9805c672017-05-24 10:15:15 +02001716 return ret;
1717}
1718
1719int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1720 bool invoke)
1721{
1722 int ret;
1723
1724 cpus_read_lock();
1725 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001726 cpus_read_unlock();
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001727 return ret;
1728}
1729EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1730
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001731/**
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001732 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001733 * @state: The state to setup
1734 * @invoke: If true, the startup function is invoked for cpus where
1735 * cpu state >= @state
1736 * @startup: startup callback function
1737 * @teardown: teardown callback function
1738 * @multi_instance: State is set up for multiple instances which get
1739 * added afterwards.
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001740 *
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001741 * The caller needs to hold cpus read locked while calling this function.
Boris Ostrovsky512f0982016-12-15 10:00:57 -05001742 * Returns:
1743 * On success:
1744 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1745 * 0 for all other states
1746 * On failure: proper (negative) error code
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001747 */
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001748int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1749 const char *name, bool invoke,
1750 int (*startup)(unsigned int cpu),
1751 int (*teardown)(unsigned int cpu),
1752 bool multi_instance)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001753{
1754 int cpu, ret = 0;
Thomas Gleixnerb9d9d692016-12-26 22:58:19 +01001755 bool dynstate;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001756
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001757 lockdep_assert_cpus_held();
1758
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001759 if (cpuhp_cb_check(state) || !name)
1760 return -EINVAL;
1761
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001762 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001763
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001764 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1765 multi_instance);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001766
Thomas Gleixnerb9d9d692016-12-26 22:58:19 +01001767 dynstate = state == CPUHP_AP_ONLINE_DYN;
1768 if (ret > 0 && dynstate) {
1769 state = ret;
1770 ret = 0;
1771 }
1772
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001773 if (ret || !invoke || !startup)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001774 goto out;
1775
1776 /*
1777 * Try to call the startup callback for each present cpu
1778 * depending on the hotplug state of the cpu.
1779 */
1780 for_each_present_cpu(cpu) {
1781 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1782 int cpustate = st->state;
1783
1784 if (cpustate < state)
1785 continue;
1786
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001787 ret = cpuhp_issue_call(cpu, state, true, NULL);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001788 if (ret) {
Thomas Gleixnera7246322016-08-12 19:49:38 +02001789 if (teardown)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001790 cpuhp_rollback_install(cpu, state, NULL);
1791 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001792 goto out;
1793 }
1794 }
1795out:
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001796 mutex_unlock(&cpuhp_state_mutex);
Thomas Gleixnerdc280d932016-12-21 20:19:49 +01001797 /*
1798 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1799 * dynamically allocated state in case of success.
1800 */
Thomas Gleixnerb9d9d692016-12-26 22:58:19 +01001801 if (!ret && dynstate)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001802 return state;
1803 return ret;
1804}
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001805EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1806
1807int __cpuhp_setup_state(enum cpuhp_state state,
1808 const char *name, bool invoke,
1809 int (*startup)(unsigned int cpu),
1810 int (*teardown)(unsigned int cpu),
1811 bool multi_instance)
1812{
1813 int ret;
1814
1815 cpus_read_lock();
1816 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1817 teardown, multi_instance);
1818 cpus_read_unlock();
1819 return ret;
1820}
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001821EXPORT_SYMBOL(__cpuhp_setup_state);
1822
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001823int __cpuhp_state_remove_instance(enum cpuhp_state state,
1824 struct hlist_node *node, bool invoke)
1825{
1826 struct cpuhp_step *sp = cpuhp_get_step(state);
1827 int cpu;
1828
1829 BUG_ON(cpuhp_cb_check(state));
1830
1831 if (!sp->multi_instance)
1832 return -EINVAL;
1833
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001834 cpus_read_lock();
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001835 mutex_lock(&cpuhp_state_mutex);
1836
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001837 if (!invoke || !cpuhp_get_teardown_cb(state))
1838 goto remove;
1839 /*
1840 * Call the teardown callback for each present cpu depending
1841 * on the hotplug state of the cpu. This function is not
1842 * allowed to fail currently!
1843 */
1844 for_each_present_cpu(cpu) {
1845 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1846 int cpustate = st->state;
1847
1848 if (cpustate >= state)
1849 cpuhp_issue_call(cpu, state, false, node);
1850 }
1851
1852remove:
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001853 hlist_del(node);
1854 mutex_unlock(&cpuhp_state_mutex);
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001855 cpus_read_unlock();
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001856
1857 return 0;
1858}
1859EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001860
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001861/**
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001862 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001863 * @state: The state to remove
1864 * @invoke: If true, the teardown function is invoked for cpus where
1865 * cpu state >= @state
1866 *
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001867 * The caller needs to hold cpus read locked while calling this function.
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001868 * The teardown callback is currently not allowed to fail. Think
1869 * about module removal!
1870 */
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001871void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001872{
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001873 struct cpuhp_step *sp = cpuhp_get_step(state);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001874 int cpu;
1875
1876 BUG_ON(cpuhp_cb_check(state));
1877
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001878 lockdep_assert_cpus_held();
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001879
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001880 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001881 if (sp->multi_instance) {
1882 WARN(!hlist_empty(&sp->list),
1883 "Error: Removing state %d which has instances left.\n",
1884 state);
1885 goto remove;
1886 }
1887
Thomas Gleixnera7246322016-08-12 19:49:38 +02001888 if (!invoke || !cpuhp_get_teardown_cb(state))
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001889 goto remove;
1890
1891 /*
1892 * Call the teardown callback for each present cpu depending
1893 * on the hotplug state of the cpu. This function is not
1894 * allowed to fail currently!
1895 */
1896 for_each_present_cpu(cpu) {
1897 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1898 int cpustate = st->state;
1899
1900 if (cpustate >= state)
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001901 cpuhp_issue_call(cpu, state, false, NULL);
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001902 }
1903remove:
Thomas Gleixnercf392d12016-08-12 19:49:39 +02001904 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
Sebastian Andrzej Siewiordc434e052017-03-14 16:06:45 +01001905 mutex_unlock(&cpuhp_state_mutex);
Sebastian Andrzej Siewior71def422017-05-24 10:15:14 +02001906}
1907EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1908
1909void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1910{
1911 cpus_read_lock();
1912 __cpuhp_remove_state_cpuslocked(state, invoke);
Thomas Gleixner8f553c42017-05-24 10:15:12 +02001913 cpus_read_unlock();
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001914}
1915EXPORT_SYMBOL(__cpuhp_remove_state);
1916
Arnd Bergmanndc8d37e2019-12-10 20:56:04 +01001917#ifdef CONFIG_HOTPLUG_SMT
1918static void cpuhp_offline_cpu_device(unsigned int cpu)
1919{
1920 struct device *dev = get_cpu_device(cpu);
1921
1922 dev->offline = true;
1923 /* Tell user space about the state change */
1924 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
1925}
1926
1927static void cpuhp_online_cpu_device(unsigned int cpu)
1928{
1929 struct device *dev = get_cpu_device(cpu);
1930
1931 dev->offline = false;
1932 /* Tell user space about the state change */
1933 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
1934}
1935
1936int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
1937{
1938 int cpu, ret = 0;
1939
1940 cpu_maps_update_begin();
1941 for_each_online_cpu(cpu) {
1942 if (topology_is_primary_thread(cpu))
1943 continue;
1944 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
1945 if (ret)
1946 break;
1947 /*
1948 * As this needs to hold the cpu maps lock it's impossible
1949 * to call device_offline() because that ends up calling
1950 * cpu_down() which takes cpu maps lock. cpu maps lock
1951 * needs to be held as this might race against in kernel
1952 * abusers of the hotplug machinery (thermal management).
1953 *
1954 * So nothing would update device:offline state. That would
1955 * leave the sysfs entry stale and prevent onlining after
1956 * smt control has been changed to 'off' again. This is
1957 * called under the sysfs hotplug lock, so it is properly
1958 * serialized against the regular offline usage.
1959 */
1960 cpuhp_offline_cpu_device(cpu);
1961 }
1962 if (!ret)
1963 cpu_smt_control = ctrlval;
1964 cpu_maps_update_done();
1965 return ret;
1966}
1967
1968int cpuhp_smt_enable(void)
1969{
1970 int cpu, ret = 0;
1971
1972 cpu_maps_update_begin();
1973 cpu_smt_control = CPU_SMT_ENABLED;
1974 for_each_present_cpu(cpu) {
1975 /* Skip online CPUs and CPUs on offline nodes */
1976 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
1977 continue;
1978 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
1979 if (ret)
1980 break;
1981 /* See comment in cpuhp_smt_disable() */
1982 cpuhp_online_cpu_device(cpu);
1983 }
1984 cpu_maps_update_done();
1985 return ret;
1986}
1987#endif
1988
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001989#if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1990static ssize_t show_cpuhp_state(struct device *dev,
1991 struct device_attribute *attr, char *buf)
1992{
1993 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1994
1995 return sprintf(buf, "%d\n", st->state);
1996}
1997static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1998
Thomas Gleixner757c9892016-02-26 18:43:32 +00001999static ssize_t write_cpuhp_target(struct device *dev,
2000 struct device_attribute *attr,
2001 const char *buf, size_t count)
2002{
2003 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2004 struct cpuhp_step *sp;
2005 int target, ret;
2006
2007 ret = kstrtoint(buf, 10, &target);
2008 if (ret)
2009 return ret;
2010
2011#ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
2012 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
2013 return -EINVAL;
2014#else
2015 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
2016 return -EINVAL;
2017#endif
2018
2019 ret = lock_device_hotplug_sysfs();
2020 if (ret)
2021 return ret;
2022
2023 mutex_lock(&cpuhp_state_mutex);
2024 sp = cpuhp_get_step(target);
2025 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
2026 mutex_unlock(&cpuhp_state_mutex);
2027 if (ret)
Sebastian Andrzej Siewior40da1b12017-06-02 16:27:14 +02002028 goto out;
Thomas Gleixner757c9892016-02-26 18:43:32 +00002029
2030 if (st->state < target)
2031 ret = do_cpu_up(dev->id, target);
2032 else
2033 ret = do_cpu_down(dev->id, target);
Sebastian Andrzej Siewior40da1b12017-06-02 16:27:14 +02002034out:
Thomas Gleixner757c9892016-02-26 18:43:32 +00002035 unlock_device_hotplug();
2036 return ret ? ret : count;
2037}
2038
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002039static ssize_t show_cpuhp_target(struct device *dev,
2040 struct device_attribute *attr, char *buf)
2041{
2042 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2043
2044 return sprintf(buf, "%d\n", st->target);
2045}
Thomas Gleixner757c9892016-02-26 18:43:32 +00002046static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002047
Peter Zijlstra1db49482017-09-20 19:00:21 +02002048
2049static ssize_t write_cpuhp_fail(struct device *dev,
2050 struct device_attribute *attr,
2051 const char *buf, size_t count)
2052{
2053 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2054 struct cpuhp_step *sp;
2055 int fail, ret;
2056
2057 ret = kstrtoint(buf, 10, &fail);
2058 if (ret)
2059 return ret;
2060
Eiichi Tsukata33d4a5a2019-06-27 11:47:32 +09002061 if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE)
2062 return -EINVAL;
2063
Peter Zijlstra1db49482017-09-20 19:00:21 +02002064 /*
2065 * Cannot fail STARTING/DYING callbacks.
2066 */
2067 if (cpuhp_is_atomic_state(fail))
2068 return -EINVAL;
2069
2070 /*
2071 * Cannot fail anything that doesn't have callbacks.
2072 */
2073 mutex_lock(&cpuhp_state_mutex);
2074 sp = cpuhp_get_step(fail);
2075 if (!sp->startup.single && !sp->teardown.single)
2076 ret = -EINVAL;
2077 mutex_unlock(&cpuhp_state_mutex);
2078 if (ret)
2079 return ret;
2080
2081 st->fail = fail;
2082
2083 return count;
2084}
2085
2086static ssize_t show_cpuhp_fail(struct device *dev,
2087 struct device_attribute *attr, char *buf)
2088{
2089 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2090
2091 return sprintf(buf, "%d\n", st->fail);
2092}
2093
2094static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
2095
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002096static struct attribute *cpuhp_cpu_attrs[] = {
2097 &dev_attr_state.attr,
2098 &dev_attr_target.attr,
Peter Zijlstra1db49482017-09-20 19:00:21 +02002099 &dev_attr_fail.attr,
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002100 NULL
2101};
2102
Arvind Yadav993647a2017-06-29 17:40:47 +05302103static const struct attribute_group cpuhp_cpu_attr_group = {
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002104 .attrs = cpuhp_cpu_attrs,
2105 .name = "hotplug",
2106 NULL
2107};
2108
2109static ssize_t show_cpuhp_states(struct device *dev,
2110 struct device_attribute *attr, char *buf)
2111{
2112 ssize_t cur, res = 0;
2113 int i;
2114
2115 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixner757c9892016-02-26 18:43:32 +00002116 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002117 struct cpuhp_step *sp = cpuhp_get_step(i);
2118
2119 if (sp->name) {
2120 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2121 buf += cur;
2122 res += cur;
2123 }
2124 }
2125 mutex_unlock(&cpuhp_state_mutex);
2126 return res;
2127}
2128static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
2129
2130static struct attribute *cpuhp_cpu_root_attrs[] = {
2131 &dev_attr_states.attr,
2132 NULL
2133};
2134
Arvind Yadav993647a2017-06-29 17:40:47 +05302135static const struct attribute_group cpuhp_cpu_root_attr_group = {
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002136 .attrs = cpuhp_cpu_root_attrs,
2137 .name = "hotplug",
2138 NULL
2139};
2140
Thomas Gleixner05736e42018-05-29 17:48:27 +02002141#ifdef CONFIG_HOTPLUG_SMT
2142
Thomas Gleixner05736e42018-05-29 17:48:27 +02002143static ssize_t
Josh Poimboeufde7b77e2019-03-27 07:00:29 -05002144__store_smt_control(struct device *dev, struct device_attribute *attr,
2145 const char *buf, size_t count)
Thomas Gleixner05736e42018-05-29 17:48:27 +02002146{
2147 int ctrlval, ret;
2148
2149 if (sysfs_streq(buf, "on"))
2150 ctrlval = CPU_SMT_ENABLED;
2151 else if (sysfs_streq(buf, "off"))
2152 ctrlval = CPU_SMT_DISABLED;
2153 else if (sysfs_streq(buf, "forceoff"))
2154 ctrlval = CPU_SMT_FORCE_DISABLED;
2155 else
2156 return -EINVAL;
2157
2158 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2159 return -EPERM;
2160
2161 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2162 return -ENODEV;
2163
2164 ret = lock_device_hotplug_sysfs();
2165 if (ret)
2166 return ret;
2167
2168 if (ctrlval != cpu_smt_control) {
2169 switch (ctrlval) {
2170 case CPU_SMT_ENABLED:
Thomas Gleixner215af542018-07-07 11:40:18 +02002171 ret = cpuhp_smt_enable();
Thomas Gleixner05736e42018-05-29 17:48:27 +02002172 break;
2173 case CPU_SMT_DISABLED:
2174 case CPU_SMT_FORCE_DISABLED:
2175 ret = cpuhp_smt_disable(ctrlval);
2176 break;
2177 }
2178 }
2179
2180 unlock_device_hotplug();
2181 return ret ? ret : count;
2182}
Josh Poimboeufde7b77e2019-03-27 07:00:29 -05002183
2184#else /* !CONFIG_HOTPLUG_SMT */
2185static ssize_t
2186__store_smt_control(struct device *dev, struct device_attribute *attr,
2187 const char *buf, size_t count)
2188{
2189 return -ENODEV;
2190}
2191#endif /* CONFIG_HOTPLUG_SMT */
2192
2193static const char *smt_states[] = {
2194 [CPU_SMT_ENABLED] = "on",
2195 [CPU_SMT_DISABLED] = "off",
2196 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2197 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2198 [CPU_SMT_NOT_IMPLEMENTED] = "notimplemented",
2199};
2200
2201static ssize_t
2202show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2203{
2204 const char *state = smt_states[cpu_smt_control];
2205
2206 return snprintf(buf, PAGE_SIZE - 2, "%s\n", state);
2207}
2208
2209static ssize_t
2210store_smt_control(struct device *dev, struct device_attribute *attr,
2211 const char *buf, size_t count)
2212{
2213 return __store_smt_control(dev, attr, buf, count);
2214}
Thomas Gleixner05736e42018-05-29 17:48:27 +02002215static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2216
2217static ssize_t
2218show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2219{
Josh Poimboeufde7b77e2019-03-27 07:00:29 -05002220 return snprintf(buf, PAGE_SIZE - 2, "%d\n", sched_smt_active());
Thomas Gleixner05736e42018-05-29 17:48:27 +02002221}
2222static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2223
2224static struct attribute *cpuhp_smt_attrs[] = {
2225 &dev_attr_control.attr,
2226 &dev_attr_active.attr,
2227 NULL
2228};
2229
2230static const struct attribute_group cpuhp_smt_attr_group = {
2231 .attrs = cpuhp_smt_attrs,
2232 .name = "smt",
2233 NULL
2234};
2235
Josh Poimboeufde7b77e2019-03-27 07:00:29 -05002236static int __init cpu_smt_sysfs_init(void)
Thomas Gleixner05736e42018-05-29 17:48:27 +02002237{
Thomas Gleixner05736e42018-05-29 17:48:27 +02002238 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2239 &cpuhp_smt_attr_group);
2240}
2241
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002242static int __init cpuhp_sysfs_init(void)
2243{
2244 int cpu, ret;
2245
Josh Poimboeufde7b77e2019-03-27 07:00:29 -05002246 ret = cpu_smt_sysfs_init();
Thomas Gleixner05736e42018-05-29 17:48:27 +02002247 if (ret)
2248 return ret;
2249
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002250 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2251 &cpuhp_cpu_root_attr_group);
2252 if (ret)
2253 return ret;
2254
2255 for_each_possible_cpu(cpu) {
2256 struct device *dev = get_cpu_device(cpu);
2257
2258 if (!dev)
2259 continue;
2260 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2261 if (ret)
2262 return ret;
2263 }
2264 return 0;
2265}
2266device_initcall(cpuhp_sysfs_init);
Josh Poimboeufde7b77e2019-03-27 07:00:29 -05002267#endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00002268
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002269/*
2270 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2271 * represents all NR_CPUS bits binary values of 1<<nr.
2272 *
Rusty Russelle0b582e2009-01-01 10:12:28 +10302273 * It is used by cpumask_of() to get a constant address to a CPU
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002274 * mask value that has a single bit set only.
2275 */
Mike Travisb8d317d2008-07-24 18:21:29 -07002276
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002277/* cpu_bit_bitmap[0] is empty - so we can back into it */
Michael Rodriguez4d519852011-03-22 16:34:07 -07002278#define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002279#define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2280#define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2281#define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
Mike Travisb8d317d2008-07-24 18:21:29 -07002282
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002283const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
Mike Travisb8d317d2008-07-24 18:21:29 -07002284
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002285 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2286 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2287#if BITS_PER_LONG > 32
2288 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2289 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
Mike Travisb8d317d2008-07-24 18:21:29 -07002290#endif
2291};
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07002292EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
Rusty Russell2d3854a2008-11-05 13:39:10 +11002293
2294const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2295EXPORT_SYMBOL(cpu_all_bits);
Rusty Russellb3199c02008-12-30 09:05:14 +10302296
2297#ifdef CONFIG_INIT_ALL_POSSIBLE
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002298struct cpumask __cpu_possible_mask __read_mostly
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08002299 = {CPU_BITS_ALL};
Rusty Russellb3199c02008-12-30 09:05:14 +10302300#else
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002301struct cpumask __cpu_possible_mask __read_mostly;
Rusty Russellb3199c02008-12-30 09:05:14 +10302302#endif
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002303EXPORT_SYMBOL(__cpu_possible_mask);
Rusty Russellb3199c02008-12-30 09:05:14 +10302304
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002305struct cpumask __cpu_online_mask __read_mostly;
2306EXPORT_SYMBOL(__cpu_online_mask);
Rusty Russellb3199c02008-12-30 09:05:14 +10302307
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002308struct cpumask __cpu_present_mask __read_mostly;
2309EXPORT_SYMBOL(__cpu_present_mask);
Rusty Russellb3199c02008-12-30 09:05:14 +10302310
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08002311struct cpumask __cpu_active_mask __read_mostly;
2312EXPORT_SYMBOL(__cpu_active_mask);
Rusty Russell3fa41522008-12-30 09:05:16 +10302313
Thomas Gleixner0c09ab92019-07-09 16:23:40 +02002314atomic_t __num_online_cpus __read_mostly;
2315EXPORT_SYMBOL(__num_online_cpus);
2316
Rusty Russell3fa41522008-12-30 09:05:16 +10302317void init_cpu_present(const struct cpumask *src)
2318{
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08002319 cpumask_copy(&__cpu_present_mask, src);
Rusty Russell3fa41522008-12-30 09:05:16 +10302320}
2321
2322void init_cpu_possible(const struct cpumask *src)
2323{
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08002324 cpumask_copy(&__cpu_possible_mask, src);
Rusty Russell3fa41522008-12-30 09:05:16 +10302325}
2326
2327void init_cpu_online(const struct cpumask *src)
2328{
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08002329 cpumask_copy(&__cpu_online_mask, src);
Rusty Russell3fa41522008-12-30 09:05:16 +10302330}
Thomas Gleixnercff7d372016-02-26 18:43:28 +00002331
Thomas Gleixner0c09ab92019-07-09 16:23:40 +02002332void set_cpu_online(unsigned int cpu, bool online)
2333{
2334 /*
2335 * atomic_inc/dec() is required to handle the horrid abuse of this
2336 * function by the reboot and kexec code which invoke it from
2337 * IPI/NMI broadcasts when shutting down CPUs. Invocation from
2338 * regular CPU hotplug is properly serialized.
2339 *
2340 * Note, that the fact that __num_online_cpus is of type atomic_t
2341 * does not protect readers which are not serialized against
2342 * concurrent hotplug operations.
2343 */
2344 if (online) {
2345 if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask))
2346 atomic_inc(&__num_online_cpus);
2347 } else {
2348 if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask))
2349 atomic_dec(&__num_online_cpus);
2350 }
2351}
2352
Thomas Gleixnercff7d372016-02-26 18:43:28 +00002353/*
2354 * Activate the first processor.
2355 */
2356void __init boot_cpu_init(void)
2357{
2358 int cpu = smp_processor_id();
2359
2360 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2361 set_cpu_online(cpu, true);
2362 set_cpu_active(cpu, true);
2363 set_cpu_present(cpu, true);
2364 set_cpu_possible(cpu, true);
Peter Zijlstra8ce371f2017-03-20 12:26:55 +01002365
2366#ifdef CONFIG_SMP
2367 __boot_cpu_id = cpu;
2368#endif
Thomas Gleixnercff7d372016-02-26 18:43:28 +00002369}
2370
2371/*
2372 * Must be called _AFTER_ setting up the per_cpu areas
2373 */
Linus Torvaldsb5b14042018-08-12 12:19:42 -07002374void __init boot_cpu_hotplug_init(void)
Thomas Gleixnercff7d372016-02-26 18:43:28 +00002375{
Abel Vesa269777a2018-08-15 00:26:00 +03002376#ifdef CONFIG_SMP
Thomas Gleixnere797bda2019-07-22 20:47:16 +02002377 cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask);
Abel Vesa269777a2018-08-15 00:26:00 +03002378#endif
Thomas Gleixner0cc3cd22018-06-29 16:05:48 +02002379 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
Thomas Gleixnercff7d372016-02-26 18:43:28 +00002380}
Josh Poimboeuf98af8452019-04-12 15:39:28 -05002381
Tyler Hicks731dc9d2019-11-04 12:22:02 +01002382/*
2383 * These are used for a global "mitigations=" cmdline option for toggling
2384 * optional CPU mitigations.
2385 */
2386enum cpu_mitigations {
2387 CPU_MITIGATIONS_OFF,
2388 CPU_MITIGATIONS_AUTO,
2389 CPU_MITIGATIONS_AUTO_NOSMT,
2390};
2391
2392static enum cpu_mitigations cpu_mitigations __ro_after_init =
2393 CPU_MITIGATIONS_AUTO;
Josh Poimboeuf98af8452019-04-12 15:39:28 -05002394
2395static int __init mitigations_parse_cmdline(char *arg)
2396{
2397 if (!strcmp(arg, "off"))
2398 cpu_mitigations = CPU_MITIGATIONS_OFF;
2399 else if (!strcmp(arg, "auto"))
2400 cpu_mitigations = CPU_MITIGATIONS_AUTO;
2401 else if (!strcmp(arg, "auto,nosmt"))
2402 cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT;
Geert Uytterhoeven1bf72722019-05-16 09:09:35 +02002403 else
2404 pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
2405 arg);
Josh Poimboeuf98af8452019-04-12 15:39:28 -05002406
2407 return 0;
2408}
2409early_param("mitigations", mitigations_parse_cmdline);
Tyler Hicks731dc9d2019-11-04 12:22:02 +01002410
2411/* mitigations=off */
2412bool cpu_mitigations_off(void)
2413{
2414 return cpu_mitigations == CPU_MITIGATIONS_OFF;
2415}
2416EXPORT_SYMBOL_GPL(cpu_mitigations_off);
2417
2418/* mitigations=auto,nosmt */
2419bool cpu_mitigations_auto_nosmt(void)
2420{
2421 return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT;
2422}
2423EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt);