1 /* CPU control. 1 /* CPU control.
2 * (C) 2001, 2002, 2003, 2004 Rusty Russell 2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
3 * 3 *
4 * This code is licenced under the GPL. 4 * This code is licenced under the GPL.
5 */ 5 */
6 #include <linux/sched/mm.h> <<
7 #include <linux/proc_fs.h> 6 #include <linux/proc_fs.h>
8 #include <linux/smp.h> 7 #include <linux/smp.h>
9 #include <linux/init.h> 8 #include <linux/init.h>
10 #include <linux/notifier.h> 9 #include <linux/notifier.h>
11 #include <linux/sched/signal.h> !! 10 #include <linux/sched.h>
12 #include <linux/sched/hotplug.h> <<
13 #include <linux/sched/isolation.h> <<
14 #include <linux/sched/task.h> <<
15 #include <linux/sched/smt.h> <<
16 #include <linux/unistd.h> 11 #include <linux/unistd.h>
17 #include <linux/cpu.h> 12 #include <linux/cpu.h>
18 #include <linux/oom.h> 13 #include <linux/oom.h>
19 #include <linux/rcupdate.h> 14 #include <linux/rcupdate.h>
20 #include <linux/delay.h> <<
21 #include <linux/export.h> 15 #include <linux/export.h>
22 #include <linux/bug.h> 16 #include <linux/bug.h>
23 #include <linux/kthread.h> 17 #include <linux/kthread.h>
24 #include <linux/stop_machine.h> 18 #include <linux/stop_machine.h>
25 #include <linux/mutex.h> 19 #include <linux/mutex.h>
26 #include <linux/gfp.h> 20 #include <linux/gfp.h>
27 #include <linux/suspend.h> 21 #include <linux/suspend.h>
28 #include <linux/lockdep.h> <<
29 #include <linux/tick.h> <<
30 #include <linux/irq.h> <<
31 #include <linux/nmi.h> <<
32 #include <linux/smpboot.h> <<
33 #include <linux/relay.h> <<
34 #include <linux/slab.h> <<
35 #include <linux/scs.h> <<
36 #include <linux/percpu-rwsem.h> <<
37 #include <linux/cpuset.h> <<
38 #include <linux/random.h> <<
39 #include <linux/cc_platform.h> <<
40 <<
41 #include <trace/events/power.h> <<
42 #define CREATE_TRACE_POINTS <<
43 #include <trace/events/cpuhp.h> <<
44 22
45 #include "smpboot.h" 23 #include "smpboot.h"
46 24
47 /** <<
48 * struct cpuhp_cpu_state - Per cpu hotplug st <<
49 * @state: The current cpu state <<
50 * @target: The target state <<
51 * @fail: Current CPU hotplug callback s <<
52 * @thread: Pointer to the hotplug thread <<
53 * @should_run: Thread should execute <<
54 * @rollback: Perform a rollback <<
55 * @single: Single callback invocation <<
56 * @bringup: Single callback bringup or tea <<
57 * @node: Remote CPU node; for multi-ins <<
58 * single entry callback for inst <<
59 * @last: For multi-instance rollback, r <<
60 * @cb_state: The state for a single callbac <<
61 * @result: Result of the operation <<
62 * @ap_sync_state: State for AP synchroni <<
63 * @done_up: Signal completion to the issue <<
64 * @done_down: Signal completion to the issue <<
65 */ <<
66 struct cpuhp_cpu_state { <<
67 enum cpuhp_state state; <<
68 enum cpuhp_state target; <<
69 enum cpuhp_state fail; <<
70 #ifdef CONFIG_SMP 25 #ifdef CONFIG_SMP
71 struct task_struct *thread; <<
72 bool should_run; <<
73 bool rollback; <<
74 bool single; <<
75 bool bringup; <<
76 struct hlist_node *node; <<
77 struct hlist_node *last; <<
78 enum cpuhp_state cb_state; <<
79 int result; <<
80 atomic_t ap_sync_state; <<
81 struct completion done_up; <<
82 struct completion done_down; <<
83 #endif <<
84 }; <<
85 <<
86 static DEFINE_PER_CPU(struct cpuhp_cpu_state, <<
87 .fail = CPUHP_INVALID, <<
88 }; <<
89 <<
90 #ifdef CONFIG_SMP <<
91 cpumask_t cpus_booted_once_mask; <<
92 #endif <<
93 <<
94 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_ <<
95 static struct lockdep_map cpuhp_state_up_map = <<
96 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-u <<
97 static struct lockdep_map cpuhp_state_down_map <<
98 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-d <<
99 <<
100 <<
101 static inline void cpuhp_lock_acquire(bool bri <<
102 { <<
103 lock_map_acquire(bringup ? &cpuhp_stat <<
104 } <<
105 <<
106 static inline void cpuhp_lock_release(bool bri <<
107 { <<
108 lock_map_release(bringup ? &cpuhp_stat <<
109 } <<
110 #else <<
111 <<
112 static inline void cpuhp_lock_acquire(bool bri <<
113 static inline void cpuhp_lock_release(bool bri <<
114 <<
115 #endif <<
116 <<
117 /** <<
118 * struct cpuhp_step - Hotplug state machine s <<
119 * @name: Name of the step <<
120 * @startup: Startup function of the step <<
121 * @teardown: Teardown function of the step <<
122 * @cant_stop: Bringup/teardown can't be stop <<
123 * @multi_instance: State has multiple ins <<
124 */ <<
125 struct cpuhp_step { <<
126 const char *name; <<
127 union { <<
128 int (*single)(unsi <<
129 int (*multi)(unsig <<
130 struc <<
131 } startup; <<
132 union { <<
133 int (*single)(unsi <<
134 int (*multi)(unsig <<
135 struc <<
136 } teardown; <<
137 /* private: */ <<
138 struct hlist_head list; <<
139 /* public: */ <<
140 bool cant_stop; <<
141 bool multi_instance <<
142 }; <<
143 <<
144 static DEFINE_MUTEX(cpuhp_state_mutex); <<
145 static struct cpuhp_step cpuhp_hp_states[]; <<
146 <<
147 static struct cpuhp_step *cpuhp_get_step(enum <<
148 { <<
149 return cpuhp_hp_states + state; <<
150 } <<
151 <<
152 static bool cpuhp_step_empty(bool bringup, str <<
153 { <<
154 return bringup ? !step->startup.single <<
155 } <<
156 <<
157 /** <<
158 * cpuhp_invoke_callback - Invoke the callback <<
159 * @cpu: The cpu for which the callback <<
160 * @state: The state to do callbacks for <<
161 * @bringup: True if the bringup callback s <<
162 * @node: For multi-instance, do a singl <<
163 * @lastp: For multi-instance rollback, r <<
164 * <<
165 * Called from cpu hotplug and from the state <<
166 * <<
167 * Return: %0 on success or a negative errno c <<
168 */ <<
169 static int cpuhp_invoke_callback(unsigned int <<
170 bool bringup, <<
171 struct hlist_ <<
172 { <<
173 struct cpuhp_cpu_state *st = per_cpu_p <<
174 struct cpuhp_step *step = cpuhp_get_st <<
175 int (*cbm)(unsigned int cpu, struct hl <<
176 int (*cb)(unsigned int cpu); <<
177 int ret, cnt; <<
178 <<
179 if (st->fail == state) { <<
180 st->fail = CPUHP_INVALID; <<
181 return -EAGAIN; <<
182 } <<
183 <<
184 if (cpuhp_step_empty(bringup, step)) { <<
185 WARN_ON_ONCE(1); <<
186 return 0; <<
187 } <<
188 <<
189 if (!step->multi_instance) { <<
190 WARN_ON_ONCE(lastp && *lastp); <<
191 cb = bringup ? step->startup.s <<
192 <<
193 trace_cpuhp_enter(cpu, st->tar <<
194 ret = cb(cpu); <<
195 trace_cpuhp_exit(cpu, st->stat <<
196 return ret; <<
197 } <<
198 cbm = bringup ? step->startup.multi : <<
199 <<
200 /* Single invocation for instance add/ <<
201 if (node) { <<
202 WARN_ON_ONCE(lastp && *lastp); <<
203 trace_cpuhp_multi_enter(cpu, s <<
204 ret = cbm(cpu, node); <<
205 trace_cpuhp_exit(cpu, st->stat <<
206 return ret; <<
207 } <<
208 <<
209 /* State transition. Invoke on all ins <<
210 cnt = 0; <<
211 hlist_for_each(node, &step->list) { <<
212 if (lastp && node == *lastp) <<
213 break; <<
214 <<
215 trace_cpuhp_multi_enter(cpu, s <<
216 ret = cbm(cpu, node); <<
217 trace_cpuhp_exit(cpu, st->stat <<
218 if (ret) { <<
219 if (!lastp) <<
220 goto err; <<
221 <<
222 *lastp = node; <<
223 return ret; <<
224 } <<
225 cnt++; <<
226 } <<
227 if (lastp) <<
228 *lastp = NULL; <<
229 return 0; <<
230 err: <<
231 /* Rollback the instances if one faile <<
232 cbm = !bringup ? step->startup.multi : <<
233 if (!cbm) <<
234 return ret; <<
235 <<
236 hlist_for_each(node, &step->list) { <<
237 if (!cnt--) <<
238 break; <<
239 <<
240 trace_cpuhp_multi_enter(cpu, s <<
241 ret = cbm(cpu, node); <<
242 trace_cpuhp_exit(cpu, st->stat <<
243 /* <<
244 * Rollback must not fail, <<
245 */ <<
246 WARN_ON_ONCE(ret); <<
247 } <<
248 return ret; <<
249 } <<
250 <<
251 #ifdef CONFIG_SMP <<
252 static bool cpuhp_is_ap_state(enum cpuhp_state <<
253 { <<
254 /* <<
255 * The extra check for CPUHP_TEARDOWN_ <<
256 * purposes as that state is handled e <<
257 */ <<
258 return state > CPUHP_BRINGUP_CPU && st <<
259 } <<
260 <<
261 static inline void wait_for_ap_thread(struct c <<
262 { <<
263 struct completion *done = bringup ? &s <<
264 wait_for_completion(done); <<
265 } <<
266 <<
267 static inline void complete_ap_thread(struct c <<
268 { <<
269 struct completion *done = bringup ? &s <<
270 complete(done); <<
271 } <<
272 <<
273 /* <<
274 * The former STARTING/DYING states, ran with <<
275 */ <<
276 static bool cpuhp_is_atomic_state(enum cpuhp_s <<
277 { <<
278 return CPUHP_AP_IDLE_DEAD <= state && <<
279 } <<
280 <<
281 /* Synchronization state management */ <<
282 enum cpuhp_sync_state { <<
283 SYNC_STATE_DEAD, <<
284 SYNC_STATE_KICKED, <<
285 SYNC_STATE_SHOULD_DIE, <<
286 SYNC_STATE_ALIVE, <<
287 SYNC_STATE_SHOULD_ONLINE, <<
288 SYNC_STATE_ONLINE, <<
289 }; <<
290 <<
291 #ifdef CONFIG_HOTPLUG_CORE_SYNC <<
292 /** <<
293 * cpuhp_ap_update_sync_state - Update synchro <<
294 * @state: The synchronization state to s <<
295 * <<
296 * No synchronization point. Just update of th <<
297 * a full barrier so that the AP changes are v <<
298 */ <<
299 static inline void cpuhp_ap_update_sync_state( <<
300 { <<
301 atomic_t *st = this_cpu_ptr(&cpuhp_sta <<
302 <<
303 (void)atomic_xchg(st, state); <<
304 } <<
305 <<
306 void __weak arch_cpuhp_sync_state_poll(void) { <<
307 <<
308 static bool cpuhp_wait_for_sync_state(unsigned <<
309 enum cpu <<
310 { <<
311 atomic_t *st = per_cpu_ptr(&cpuhp_stat <<
312 ktime_t now, end, start = ktime_get(); <<
313 int sync; <<
314 <<
315 end = start + 10ULL * NSEC_PER_SEC; <<
316 <<
317 sync = atomic_read(st); <<
318 while (1) { <<
319 if (sync == state) { <<
320 if (!atomic_try_cmpxch <<
321 continue; <<
322 return true; <<
323 } <<
324 <<
325 now = ktime_get(); <<
326 if (now > end) { <<
327 /* Timeout. Leave the <<
328 return false; <<
329 } else if (now - start < NSEC_ <<
330 /* Poll for one millis <<
331 arch_cpuhp_sync_state_ <<
332 } else { <<
333 usleep_range(USEC_PER_ <<
334 } <<
335 sync = atomic_read(st); <<
336 } <<
337 return true; <<
338 } <<
339 #else /* CONFIG_HOTPLUG_CORE_SYNC */ <<
340 static inline void cpuhp_ap_update_sync_state( <<
341 #endif /* !CONFIG_HOTPLUG_CORE_SYNC */ <<
342 <<
343 #ifdef CONFIG_HOTPLUG_CORE_SYNC_DEAD <<
344 /** <<
345 * cpuhp_ap_report_dead - Update synchronizati <<
346 * <<
347 * No synchronization point. Just update of th <<
348 */ <<
349 void cpuhp_ap_report_dead(void) <<
350 { <<
351 cpuhp_ap_update_sync_state(SYNC_STATE_ <<
352 } <<
353 <<
354 void __weak arch_cpuhp_cleanup_dead_cpu(unsign <<
355 <<
356 /* <<
357 * Late CPU shutdown synchronization point. Ca <<
358 * because the AP cannot issue complete() at t <<
359 */ <<
360 static void cpuhp_bp_sync_dead(unsigned int cp <<
361 { <<
362 atomic_t *st = per_cpu_ptr(&cpuhp_stat <<
363 int sync = atomic_read(st); <<
364 <<
365 do { <<
366 /* CPU can have reported dead <<
367 if (sync == SYNC_STATE_DEAD) <<
368 break; <<
369 } while (!atomic_try_cmpxchg(st, &sync <<
370 <<
371 if (cpuhp_wait_for_sync_state(cpu, SYN <<
372 /* CPU reached dead state. Inv <<
373 arch_cpuhp_cleanup_dead_cpu(cp <<
374 return; <<
375 } <<
376 <<
377 /* No further action possible. Emit me <<
378 pr_err("CPU%u failed to report dead st <<
379 } <<
380 #else /* CONFIG_HOTPLUG_CORE_SYNC_DEAD */ <<
381 static inline void cpuhp_bp_sync_dead(unsigned <<
382 #endif /* !CONFIG_HOTPLUG_CORE_SYNC_DEAD */ <<
383 <<
384 #ifdef CONFIG_HOTPLUG_CORE_SYNC_FULL <<
385 /** <<
386 * cpuhp_ap_sync_alive - Synchronize AP with t <<
387 * <<
388 * Updates the AP synchronization state to SYN <<
389 * for the BP to release it. <<
390 */ <<
391 void cpuhp_ap_sync_alive(void) <<
392 { <<
393 atomic_t *st = this_cpu_ptr(&cpuhp_sta <<
394 <<
395 cpuhp_ap_update_sync_state(SYNC_STATE_ <<
396 <<
397 /* Wait for the control CPU to release <<
398 while (atomic_read(st) != SYNC_STATE_S <<
399 cpu_relax(); <<
400 } <<
401 <<
402 static bool cpuhp_can_boot_ap(unsigned int cpu <<
403 { <<
404 atomic_t *st = per_cpu_ptr(&cpuhp_stat <<
405 int sync = atomic_read(st); <<
406 <<
407 again: <<
408 switch (sync) { <<
409 case SYNC_STATE_DEAD: <<
410 /* CPU is properly dead */ <<
411 break; <<
412 case SYNC_STATE_KICKED: <<
413 /* CPU did not come up in prev <<
414 break; <<
415 case SYNC_STATE_ALIVE: <<
416 /* CPU is stuck cpuhp_ap_sync_ <<
417 break; <<
418 default: <<
419 /* CPU failed to report online <<
420 return false; <<
421 } <<
422 <<
423 /* Prepare for booting */ <<
424 if (!atomic_try_cmpxchg(st, &sync, SYN <<
425 goto again; <<
426 <<
427 return true; <<
428 } <<
429 <<
430 void __weak arch_cpuhp_cleanup_kick_cpu(unsign <<
431 <<
432 /* <<
433 * Early CPU bringup synchronization point. Ca <<
434 * because the AP cannot issue complete() so e <<
435 */ <<
436 static int cpuhp_bp_sync_alive(unsigned int cp <<
437 { <<
438 int ret = 0; <<
439 <<
440 if (!IS_ENABLED(CONFIG_HOTPLUG_CORE_SY <<
441 return 0; <<
442 <<
443 if (!cpuhp_wait_for_sync_state(cpu, SY <<
444 pr_err("CPU%u failed to report <<
445 ret = -EIO; <<
446 } <<
447 <<
448 /* Let the architecture cleanup the ki <<
449 arch_cpuhp_cleanup_kick_cpu(cpu); <<
450 return ret; <<
451 } <<
452 #else /* CONFIG_HOTPLUG_CORE_SYNC_FULL */ <<
453 static inline int cpuhp_bp_sync_alive(unsigned <<
454 static inline bool cpuhp_can_boot_ap(unsigned <<
455 #endif /* !CONFIG_HOTPLUG_CORE_SYNC_FULL */ <<
456 <<
457 /* Serializes the updates to cpu_online_mask, 26 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
458 static DEFINE_MUTEX(cpu_add_remove_lock); 27 static DEFINE_MUTEX(cpu_add_remove_lock);
459 bool cpuhp_tasks_frozen; <<
460 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); <<
461 28
462 /* 29 /*
463 * The following two APIs (cpu_maps_update_beg !! 30 * The following two API's must be used when attempting
464 * attempting to serialize the updates to cpu_ !! 31 * to serialize the updates to cpu_online_mask, cpu_present_mask.
465 */ 32 */
466 void cpu_maps_update_begin(void) 33 void cpu_maps_update_begin(void)
467 { 34 {
468 mutex_lock(&cpu_add_remove_lock); 35 mutex_lock(&cpu_add_remove_lock);
469 } 36 }
470 37
471 void cpu_maps_update_done(void) 38 void cpu_maps_update_done(void)
472 { 39 {
473 mutex_unlock(&cpu_add_remove_lock); 40 mutex_unlock(&cpu_add_remove_lock);
474 } 41 }
475 42
476 /* !! 43 static RAW_NOTIFIER_HEAD(cpu_chain);
477 * If set, cpu_up and cpu_down will return -EB !! 44
>> 45 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
478 * Should always be manipulated under cpu_add_ 46 * Should always be manipulated under cpu_add_remove_lock
479 */ 47 */
480 static int cpu_hotplug_disabled; 48 static int cpu_hotplug_disabled;
481 49
482 #ifdef CONFIG_HOTPLUG_CPU 50 #ifdef CONFIG_HOTPLUG_CPU
483 51
484 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); !! 52 static struct {
485 !! 53 struct task_struct *active_writer;
486 static bool cpu_hotplug_offline_disabled __ro_ !! 54 struct mutex lock; /* Synchronizes accesses to refcount, */
487 !! 55 /*
488 void cpus_read_lock(void) !! 56 * Also blocks the new readers during
489 { !! 57 * an ongoing cpu hotplug operation.
490 percpu_down_read(&cpu_hotplug_lock); !! 58 */
491 } !! 59 int refcount;
492 EXPORT_SYMBOL_GPL(cpus_read_lock); !! 60 } cpu_hotplug = {
493 !! 61 .active_writer = NULL,
494 int cpus_read_trylock(void) !! 62 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
495 { !! 63 .refcount = 0,
496 return percpu_down_read_trylock(&cpu_h !! 64 };
497 } <<
498 EXPORT_SYMBOL_GPL(cpus_read_trylock); <<
499 <<
500 void cpus_read_unlock(void) <<
501 { <<
502 percpu_up_read(&cpu_hotplug_lock); <<
503 } <<
504 EXPORT_SYMBOL_GPL(cpus_read_unlock); <<
505 65
506 void cpus_write_lock(void) !! 66 void get_online_cpus(void)
507 { 67 {
508 percpu_down_write(&cpu_hotplug_lock); !! 68 might_sleep();
509 } !! 69 if (cpu_hotplug.active_writer == current)
>> 70 return;
>> 71 mutex_lock(&cpu_hotplug.lock);
>> 72 cpu_hotplug.refcount++;
>> 73 mutex_unlock(&cpu_hotplug.lock);
510 74
511 void cpus_write_unlock(void) <<
512 { <<
513 percpu_up_write(&cpu_hotplug_lock); <<
514 } 75 }
>> 76 EXPORT_SYMBOL_GPL(get_online_cpus);
515 77
516 void lockdep_assert_cpus_held(void) !! 78 void put_online_cpus(void)
517 { 79 {
518 /* !! 80 if (cpu_hotplug.active_writer == current)
519 * We can't have hotplug operations be <<
520 * and some init codepaths will knowin <<
521 * This is all valid, so mute lockdep <<
522 * unheld locks. <<
523 */ <<
524 if (system_state < SYSTEM_RUNNING) <<
525 return; 81 return;
>> 82 mutex_lock(&cpu_hotplug.lock);
526 83
527 percpu_rwsem_assert_held(&cpu_hotplug_ !! 84 if (WARN_ON(!cpu_hotplug.refcount))
528 } !! 85 cpu_hotplug.refcount++; /* try to fix things up */
529 86
530 #ifdef CONFIG_LOCKDEP !! 87 if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
531 int lockdep_is_cpus_held(void) !! 88 wake_up_process(cpu_hotplug.active_writer);
532 { !! 89 mutex_unlock(&cpu_hotplug.lock);
533 return percpu_rwsem_is_held(&cpu_hotpl <<
534 } <<
535 #endif <<
536 90
537 static void lockdep_acquire_cpus_lock(void) <<
538 { <<
539 rwsem_acquire(&cpu_hotplug_lock.dep_ma <<
540 } 91 }
>> 92 EXPORT_SYMBOL_GPL(put_online_cpus);
541 93
542 static void lockdep_release_cpus_lock(void) !! 94 /*
>> 95 * This ensures that the hotplug operation can begin only when the
>> 96 * refcount goes to zero.
>> 97 *
>> 98 * Note that during a cpu-hotplug operation, the new readers, if any,
>> 99 * will be blocked by the cpu_hotplug.lock
>> 100 *
>> 101 * Since cpu_hotplug_begin() is always called after invoking
>> 102 * cpu_maps_update_begin(), we can be sure that only one writer is active.
>> 103 *
>> 104 * Note that theoretically, there is a possibility of a livelock:
>> 105 * - Refcount goes to zero, last reader wakes up the sleeping
>> 106 * writer.
>> 107 * - Last reader unlocks the cpu_hotplug.lock.
>> 108 * - A new reader arrives at this moment, bumps up the refcount.
>> 109 * - The writer acquires the cpu_hotplug.lock finds the refcount
>> 110 * non zero and goes to sleep again.
>> 111 *
>> 112 * However, this is very difficult to achieve in practice since
>> 113 * get_online_cpus() not an api which is called all that often.
>> 114 *
>> 115 */
>> 116 static void cpu_hotplug_begin(void)
543 { 117 {
544 rwsem_release(&cpu_hotplug_lock.dep_ma !! 118 cpu_hotplug.active_writer = current;
>> 119
>> 120 for (;;) {
>> 121 mutex_lock(&cpu_hotplug.lock);
>> 122 if (likely(!cpu_hotplug.refcount))
>> 123 break;
>> 124 __set_current_state(TASK_UNINTERRUPTIBLE);
>> 125 mutex_unlock(&cpu_hotplug.lock);
>> 126 schedule();
>> 127 }
545 } 128 }
546 129
547 /* Declare CPU offlining not supported */ !! 130 static void cpu_hotplug_done(void)
548 void cpu_hotplug_disable_offlining(void) <<
549 { 131 {
550 cpu_maps_update_begin(); !! 132 cpu_hotplug.active_writer = NULL;
551 cpu_hotplug_offline_disabled = true; !! 133 mutex_unlock(&cpu_hotplug.lock);
552 cpu_maps_update_done(); <<
553 } 134 }
554 135
555 /* 136 /*
556 * Wait for currently running CPU hotplug oper 137 * Wait for currently running CPU hotplug operations to complete (if any) and
557 * disable future CPU hotplug (from sysfs). Th 138 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
558 * the 'cpu_hotplug_disabled' flag. The same l 139 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
559 * hotplug path before performing hotplug oper 140 * hotplug path before performing hotplug operations. So acquiring that lock
560 * guarantees mutual exclusion from any curren 141 * guarantees mutual exclusion from any currently running hotplug operations.
561 */ 142 */
562 void cpu_hotplug_disable(void) 143 void cpu_hotplug_disable(void)
563 { 144 {
564 cpu_maps_update_begin(); 145 cpu_maps_update_begin();
565 cpu_hotplug_disabled++; !! 146 cpu_hotplug_disabled = 1;
566 cpu_maps_update_done(); 147 cpu_maps_update_done();
567 } 148 }
568 EXPORT_SYMBOL_GPL(cpu_hotplug_disable); <<
569 <<
570 static void __cpu_hotplug_enable(void) <<
571 { <<
572 if (WARN_ONCE(!cpu_hotplug_disabled, " <<
573 return; <<
574 cpu_hotplug_disabled--; <<
575 } <<
576 149
577 void cpu_hotplug_enable(void) 150 void cpu_hotplug_enable(void)
578 { 151 {
579 cpu_maps_update_begin(); 152 cpu_maps_update_begin();
580 __cpu_hotplug_enable(); !! 153 cpu_hotplug_disabled = 0;
581 cpu_maps_update_done(); 154 cpu_maps_update_done();
582 } 155 }
583 EXPORT_SYMBOL_GPL(cpu_hotplug_enable); <<
584 <<
585 #else <<
586 <<
587 static void lockdep_acquire_cpus_lock(void) <<
588 { <<
589 } <<
590 <<
591 static void lockdep_release_cpus_lock(void) <<
592 { <<
593 } <<
594 <<
595 #endif /* CONFIG_HOTPLUG_CPU */ <<
596 <<
597 /* <<
598 * Architectures that need SMT-specific errata <<
599 * should override this. <<
600 */ <<
601 void __weak arch_smt_update(void) { } <<
602 <<
603 #ifdef CONFIG_HOTPLUG_SMT <<
604 <<
605 enum cpuhp_smt_control cpu_smt_control __read_ <<
606 static unsigned int cpu_smt_max_threads __ro_a <<
607 unsigned int cpu_smt_num_threads __read_mostly <<
608 <<
609 void __init cpu_smt_disable(bool force) <<
610 { <<
611 if (!cpu_smt_possible()) <<
612 return; <<
613 <<
614 if (force) { <<
615 pr_info("SMT: Force disabled\n <<
616 cpu_smt_control = CPU_SMT_FORC <<
617 } else { <<
618 pr_info("SMT: disabled\n"); <<
619 cpu_smt_control = CPU_SMT_DISA <<
620 } <<
621 cpu_smt_num_threads = 1; <<
622 } <<
623 <<
624 /* <<
625 * The decision whether SMT is supported can o <<
626 * CPU identification. Called from architectur <<
627 */ <<
628 void __init cpu_smt_set_num_threads(unsigned i <<
629 unsigned i <<
630 { <<
631 WARN_ON(!num_threads || (num_threads > <<
632 <<
633 if (max_threads == 1) <<
634 cpu_smt_control = CPU_SMT_NOT_ <<
635 <<
636 cpu_smt_max_threads = max_threads; <<
637 <<
638 /* <<
639 * If SMT has been disabled via the ke <<
640 * not supported, set cpu_smt_num_thre <<
641 * If enabled, take the architecture r <<
642 * to bring up into account. <<
643 */ <<
644 if (cpu_smt_control != CPU_SMT_ENABLED <<
645 cpu_smt_num_threads = 1; <<
646 else if (num_threads < cpu_smt_num_thr <<
647 cpu_smt_num_threads = num_thre <<
648 } <<
649 <<
650 static int __init smt_cmdline_disable(char *st <<
651 { <<
652 cpu_smt_disable(str && !strcmp(str, "f <<
653 return 0; <<
654 } <<
655 early_param("nosmt", smt_cmdline_disable); <<
656 <<
657 /* <<
658 * For Archicture supporting partial SMT state <<
659 * Otherwise this has already been checked thr <<
660 * setting the SMT level. <<
661 */ <<
662 static inline bool cpu_smt_thread_allowed(unsi <<
663 { <<
664 #ifdef CONFIG_SMT_NUM_THREADS_DYNAMIC <<
665 return topology_smt_thread_allowed(cpu <<
666 #else <<
667 return true; <<
668 #endif <<
669 } <<
670 <<
671 static inline bool cpu_bootable(unsigned int c <<
672 { <<
673 if (cpu_smt_control == CPU_SMT_ENABLED <<
674 return true; <<
675 <<
676 /* All CPUs are bootable if controls a <<
677 if (cpu_smt_control == CPU_SMT_NOT_IMP <<
678 return true; <<
679 <<
680 /* All CPUs are bootable if CPU is not <<
681 if (cpu_smt_control == CPU_SMT_NOT_SUP <<
682 return true; <<
683 <<
684 if (topology_is_primary_thread(cpu)) <<
685 return true; <<
686 <<
687 /* <<
688 * On x86 it's required to boot all lo <<
689 * that the init code can get a chance <<
690 * CPU. Otherwise, a broadcasted MCE o <<
691 * core will shutdown the machine. <<
692 */ <<
693 return !cpumask_test_cpu(cpu, &cpus_bo <<
694 } <<
695 <<
696 /* Returns true if SMT is supported and not fo <<
697 bool cpu_smt_possible(void) <<
698 { <<
699 return cpu_smt_control != CPU_SMT_FORC <<
700 cpu_smt_control != CPU_SMT_NOT <<
701 } <<
702 EXPORT_SYMBOL_GPL(cpu_smt_possible); <<
703 <<
704 #else <<
705 static inline bool cpu_bootable(unsigned int c <<
706 #endif <<
707 <<
708 static inline enum cpuhp_state <<
709 cpuhp_set_state(int cpu, struct cpuhp_cpu_stat <<
710 { <<
711 enum cpuhp_state prev_state = st->stat <<
712 bool bringup = st->state < target; <<
713 <<
714 st->rollback = false; <<
715 st->last = NULL; <<
716 <<
717 st->target = target; <<
718 st->single = false; <<
719 st->bringup = bringup; <<
720 if (cpu_dying(cpu) != !bringup) <<
721 set_cpu_dying(cpu, !bringup); <<
722 <<
723 return prev_state; <<
724 } <<
725 <<
726 static inline void <<
727 cpuhp_reset_state(int cpu, struct cpuhp_cpu_st <<
728 enum cpuhp_state prev_state) <<
729 { <<
730 bool bringup = !st->bringup; <<
731 <<
732 st->target = prev_state; <<
733 <<
734 /* <<
735 * Already rolling back. No need inver <<
736 * the current state. <<
737 */ <<
738 if (st->rollback) <<
739 return; <<
740 <<
741 st->rollback = true; <<
742 156
743 /* !! 157 #else /* #if CONFIG_HOTPLUG_CPU */
744 * If we have st->last we need to undo !! 158 static void cpu_hotplug_begin(void) {}
745 * state first. Otherwise start undo a !! 159 static void cpu_hotplug_done(void) {}
746 */ !! 160 #endif /* #else #if CONFIG_HOTPLUG_CPU */
747 if (!st->last) { <<
748 if (st->bringup) <<
749 st->state--; <<
750 else <<
751 st->state++; <<
752 } <<
753 161
754 st->bringup = bringup; !! 162 /* Need to know about CPUs going up/down? */
755 if (cpu_dying(cpu) != !bringup) !! 163 int __ref register_cpu_notifier(struct notifier_block *nb)
756 set_cpu_dying(cpu, !bringup); <<
757 } <<
758 <<
759 /* Regular hotplug invocation of the AP hotplu <<
760 static void __cpuhp_kick_ap(struct cpuhp_cpu_s <<
761 { 164 {
762 if (!st->single && st->state == st->ta <<
763 return; <<
764 <<
765 st->result = 0; <<
766 /* <<
767 * Make sure the above stores are visi <<
768 * true. Paired with the mb() above in <<
769 */ <<
770 smp_mb(); <<
771 st->should_run = true; <<
772 wake_up_process(st->thread); <<
773 wait_for_ap_thread(st, st->bringup); <<
774 } <<
775 <<
776 static int cpuhp_kick_ap(int cpu, struct cpuhp <<
777 enum cpuhp_state targ <<
778 { <<
779 enum cpuhp_state prev_state; <<
780 int ret; 165 int ret;
781 !! 166 cpu_maps_update_begin();
782 prev_state = cpuhp_set_state(cpu, st, !! 167 ret = raw_notifier_chain_register(&cpu_chain, nb);
783 __cpuhp_kick_ap(st); !! 168 cpu_maps_update_done();
784 if ((ret = st->result)) { <<
785 cpuhp_reset_state(cpu, st, pre <<
786 __cpuhp_kick_ap(st); <<
787 } <<
788 <<
789 return ret; <<
790 } <<
791 <<
792 static int bringup_wait_for_ap_online(unsigned <<
793 { <<
794 struct cpuhp_cpu_state *st = per_cpu_p <<
795 <<
796 /* Wait for the CPU to reach CPUHP_AP_ <<
797 wait_for_ap_thread(st, true); <<
798 if (WARN_ON_ONCE((!cpu_online(cpu)))) <<
799 return -ECANCELED; <<
800 <<
801 /* Unpark the hotplug thread of the ta <<
802 kthread_unpark(st->thread); <<
803 <<
804 /* <<
805 * SMT soft disabling on X86 requires <<
806 * BIOS 'wait for SIPI' state in order <<
807 * CPU marked itself as booted_once in <<
808 * cpu_bootable() check will now retur <<
809 * primary sibling. <<
810 */ <<
811 if (!cpu_bootable(cpu)) <<
812 return -ECANCELED; <<
813 return 0; <<
814 } <<
815 <<
816 #ifdef CONFIG_HOTPLUG_SPLIT_STARTUP <<
817 static int cpuhp_kick_ap_alive(unsigned int cp <<
818 { <<
819 if (!cpuhp_can_boot_ap(cpu)) <<
820 return -EAGAIN; <<
821 <<
822 return arch_cpuhp_kick_ap_alive(cpu, i <<
823 } <<
824 <<
825 static int cpuhp_bringup_ap(unsigned int cpu) <<
826 { <<
827 struct cpuhp_cpu_state *st = per_cpu_p <<
828 int ret; <<
829 <<
830 /* <<
831 * Some architectures have to walk the <<
832 * setup the vector space for the cpu <<
833 * Prevent irq alloc/free across the b <<
834 */ <<
835 irq_lock_sparse(); <<
836 <<
837 ret = cpuhp_bp_sync_alive(cpu); <<
838 if (ret) <<
839 goto out_unlock; <<
840 <<
841 ret = bringup_wait_for_ap_online(cpu); <<
842 if (ret) <<
843 goto out_unlock; <<
844 <<
845 irq_unlock_sparse(); <<
846 <<
847 if (st->target <= CPUHP_AP_ONLINE_IDLE <<
848 return 0; <<
849 <<
850 return cpuhp_kick_ap(cpu, st, st->targ <<
851 <<
852 out_unlock: <<
853 irq_unlock_sparse(); <<
854 return ret; <<
855 } <<
856 #else <<
857 static int bringup_cpu(unsigned int cpu) <<
858 { <<
859 struct cpuhp_cpu_state *st = per_cpu_p <<
860 struct task_struct *idle = idle_thread <<
861 int ret; <<
862 <<
863 if (!cpuhp_can_boot_ap(cpu)) <<
864 return -EAGAIN; <<
865 <<
866 /* <<
867 * Some architectures have to walk the <<
868 * setup the vector space for the cpu <<
869 * <<
870 * Prevent irq alloc/free across the b <<
871 * sparse irq lock. Hold it until the <<
872 * startup in cpuhp_online_idle() whic <<
873 * intermediate synchronization points <<
874 */ <<
875 irq_lock_sparse(); <<
876 <<
877 ret = __cpu_up(cpu, idle); <<
878 if (ret) <<
879 goto out_unlock; <<
880 <<
881 ret = cpuhp_bp_sync_alive(cpu); <<
882 if (ret) <<
883 goto out_unlock; <<
884 <<
885 ret = bringup_wait_for_ap_online(cpu); <<
886 if (ret) <<
887 goto out_unlock; <<
888 <<
889 irq_unlock_sparse(); <<
890 <<
891 if (st->target <= CPUHP_AP_ONLINE_IDLE <<
892 return 0; <<
893 <<
894 return cpuhp_kick_ap(cpu, st, st->targ <<
895 <<
896 out_unlock: <<
897 irq_unlock_sparse(); <<
898 return ret; <<
899 } <<
900 #endif <<
901 <<
902 static int finish_cpu(unsigned int cpu) <<
903 { <<
904 struct task_struct *idle = idle_thread <<
905 struct mm_struct *mm = idle->active_mm <<
906 <<
907 /* <<
908 * idle_task_exit() will have switched <<
909 * clean up any remaining active_mm st <<
910 */ <<
911 if (mm != &init_mm) <<
912 idle->active_mm = &init_mm; <<
913 mmdrop_lazy_tlb(mm); <<
914 return 0; <<
915 } <<
916 <<
917 /* <<
918 * Hotplug state machine related functions <<
919 */ <<
920 <<
921 /* <<
922 * Get the next state to run. Empty ones will <<
923 * state must be run. <<
924 * <<
925 * st->state will be modified ahead of time, t <<
926 * has already ran. <<
927 */ <<
928 static bool cpuhp_next_state(bool bringup, <<
929 enum cpuhp_state <<
930 struct cpuhp_cpu_ <<
931 enum cpuhp_state <<
932 { <<
933 do { <<
934 if (bringup) { <<
935 if (st->state >= targe <<
936 return false; <<
937 <<
938 *state_to_run = ++st-> <<
939 } else { <<
940 if (st->state <= targe <<
941 return false; <<
942 <<
943 *state_to_run = st->st <<
944 } <<
945 <<
946 if (!cpuhp_step_empty(bringup, <<
947 break; <<
948 } while (true); <<
949 <<
950 return true; <<
951 } <<
952 <<
953 static int __cpuhp_invoke_callback_range(bool <<
954 unsig <<
955 struc <<
956 enum <<
957 bool <<
958 { <<
959 enum cpuhp_state state; <<
960 int ret = 0; <<
961 <<
962 while (cpuhp_next_state(bringup, &stat <<
963 int err; <<
964 <<
965 err = cpuhp_invoke_callback(cp <<
966 if (!err) <<
967 continue; <<
968 <<
969 if (nofail) { <<
970 pr_warn("CPU %u %s sta <<
971 cpu, bringup ? <<
972 cpuhp_get_step <<
973 st->state, err <<
974 ret = -1; <<
975 } else { <<
976 ret = err; <<
977 break; <<
978 } <<
979 } <<
980 <<
981 return ret; <<
982 } <<
983 <<
984 static inline int cpuhp_invoke_callback_range( <<
985 <<
986 <<
987 <<
988 { <<
989 return __cpuhp_invoke_callback_range(b <<
990 } <<
991 <<
992 static inline void cpuhp_invoke_callback_range <<
993 <<
994 <<
995 <<
996 { <<
997 __cpuhp_invoke_callback_range(bringup, <<
998 } <<
999 <<
1000 static inline bool can_rollback_cpu(struct cp <<
1001 { <<
1002 if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) <<
1003 return true; <<
1004 /* <<
1005 * When CPU hotplug is disabled, then <<
1006 * possible because takedown_cpu() an <<
1007 * subsystem specific mechanisms are <<
1008 * which would be completely unplugge <<
1009 * in the current state. <<
1010 */ <<
1011 return st->state <= CPUHP_BRINGUP_CPU <<
1012 } <<
1013 <<
1014 static int cpuhp_up_callbacks(unsigned int cp <<
1015 enum cpuhp_stat <<
1016 { <<
1017 enum cpuhp_state prev_state = st->sta <<
1018 int ret = 0; <<
1019 <<
1020 ret = cpuhp_invoke_callback_range(tru <<
1021 if (ret) { <<
1022 pr_debug("CPU UP failed (%d) <<
1023 ret, cpu, cpuhp_get_ <<
1024 st->state); <<
1025 <<
1026 cpuhp_reset_state(cpu, st, pr <<
1027 if (can_rollback_cpu(st)) <<
1028 WARN_ON(cpuhp_invoke_ <<
1029 <<
1030 } <<
1031 return ret; 169 return ret;
1032 } 170 }
1033 171
1034 /* !! 172 static int __cpu_notify(unsigned long val, void *v, int nr_to_call,
1035 * The cpu hotplug threads manage the bringup !! 173 int *nr_calls)
1036 */ <<
1037 static int cpuhp_should_run(unsigned int cpu) <<
1038 { <<
1039 struct cpuhp_cpu_state *st = this_cpu <<
1040 <<
1041 return st->should_run; <<
1042 } <<
1043 <<
1044 /* <<
1045 * Execute teardown/startup callbacks on the <<
1046 * callbacks when a state gets [un]installed <<
1047 * <<
1048 * Each invocation of this function by the sm <<
1049 * state callback. <<
1050 * <<
1051 * It has 3 modes of operation: <<
1052 * - single: runs st->cb_state <<
1053 * - up: runs ++st->state, while st->sta <<
1054 * - down: runs st->state--, while st->sta <<
1055 * <<
1056 * When complete or on error, should_run is c <<
1057 */ <<
1058 static void cpuhp_thread_fun(unsigned int cpu <<
1059 { <<
1060 struct cpuhp_cpu_state *st = this_cpu <<
1061 bool bringup = st->bringup; <<
1062 enum cpuhp_state state; <<
1063 <<
1064 if (WARN_ON_ONCE(!st->should_run)) <<
1065 return; <<
1066 <<
1067 /* <<
1068 * ACQUIRE for the cpuhp_should_run() <<
1069 * that if we see ->should_run we als <<
1070 */ <<
1071 smp_mb(); <<
1072 <<
1073 /* <<
1074 * The BP holds the hotplug lock, but <<
1075 * ensure that anybody asserting the <<
1076 * it so. <<
1077 */ <<
1078 lockdep_acquire_cpus_lock(); <<
1079 cpuhp_lock_acquire(bringup); <<
1080 <<
1081 if (st->single) { <<
1082 state = st->cb_state; <<
1083 st->should_run = false; <<
1084 } else { <<
1085 st->should_run = cpuhp_next_s <<
1086 if (!st->should_run) <<
1087 goto end; <<
1088 } <<
1089 <<
1090 WARN_ON_ONCE(!cpuhp_is_ap_state(state <<
1091 <<
1092 if (cpuhp_is_atomic_state(state)) { <<
1093 local_irq_disable(); <<
1094 st->result = cpuhp_invoke_cal <<
1095 local_irq_enable(); <<
1096 <<
1097 /* <<
1098 * STARTING/DYING must not fa <<
1099 */ <<
1100 WARN_ON_ONCE(st->result); <<
1101 } else { <<
1102 st->result = cpuhp_invoke_cal <<
1103 } <<
1104 <<
1105 if (st->result) { <<
1106 /* <<
1107 * If we fail on a rollback, <<
1108 * paddle, no way forward, no <<
1109 * playing. <<
1110 */ <<
1111 WARN_ON_ONCE(st->rollback); <<
1112 st->should_run = false; <<
1113 } <<
1114 <<
1115 end: <<
1116 cpuhp_lock_release(bringup); <<
1117 lockdep_release_cpus_lock(); <<
1118 <<
1119 if (!st->should_run) <<
1120 complete_ap_thread(st, bringu <<
1121 } <<
1122 <<
1123 /* Invoke a single callback on a remote cpu * <<
1124 static int <<
1125 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_ <<
1126 struct hlist_node *n <<
1127 { 174 {
1128 struct cpuhp_cpu_state *st = per_cpu_ <<
1129 int ret; 175 int ret;
1130 176
1131 if (!cpu_online(cpu)) !! 177 ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call,
1132 return 0; !! 178 nr_calls);
1133 <<
1134 cpuhp_lock_acquire(false); <<
1135 cpuhp_lock_release(false); <<
1136 <<
1137 cpuhp_lock_acquire(true); <<
1138 cpuhp_lock_release(true); <<
1139 <<
1140 /* <<
1141 * If we are up and running, use the <<
1142 * we invoke the thread function dire <<
1143 */ <<
1144 if (!st->thread) <<
1145 return cpuhp_invoke_callback( <<
1146 <<
1147 st->rollback = false; <<
1148 st->last = NULL; <<
1149 179
1150 st->node = node; !! 180 return notifier_to_errno(ret);
1151 st->bringup = bringup; <<
1152 st->cb_state = state; <<
1153 st->single = true; <<
1154 <<
1155 __cpuhp_kick_ap(st); <<
1156 <<
1157 /* <<
1158 * If we failed and did a partial, do <<
1159 */ <<
1160 if ((ret = st->result) && st->last) { <<
1161 st->rollback = true; <<
1162 st->bringup = !bringup; <<
1163 <<
1164 __cpuhp_kick_ap(st); <<
1165 } <<
1166 <<
1167 /* <<
1168 * Clean up the leftovers so the next <<
1169 * data. <<
1170 */ <<
1171 st->node = st->last = NULL; <<
1172 return ret; <<
1173 } 181 }
1174 182
1175 static int cpuhp_kick_ap_work(unsigned int cp !! 183 static int cpu_notify(unsigned long val, void *v)
1176 { 184 {
1177 struct cpuhp_cpu_state *st = per_cpu_ !! 185 return __cpu_notify(val, v, -1, NULL);
1178 enum cpuhp_state prev_state = st->sta <<
1179 int ret; <<
1180 <<
1181 cpuhp_lock_acquire(false); <<
1182 cpuhp_lock_release(false); <<
1183 <<
1184 cpuhp_lock_acquire(true); <<
1185 cpuhp_lock_release(true); <<
1186 <<
1187 trace_cpuhp_enter(cpu, st->target, pr <<
1188 ret = cpuhp_kick_ap(cpu, st, st->targ <<
1189 trace_cpuhp_exit(cpu, st->state, prev <<
1190 <<
1191 return ret; <<
1192 } 186 }
1193 187
1194 static struct smp_hotplug_thread cpuhp_thread !! 188 static void cpu_notify_nofail(unsigned long val, void *v)
1195 .store = &cpuhp_stat <<
1196 .thread_should_run = cpuhp_shoul <<
1197 .thread_fn = cpuhp_threa <<
1198 .thread_comm = "cpuhp/%u", <<
1199 .selfparking = true, <<
1200 }; <<
1201 <<
1202 static __init void cpuhp_init_state(void) <<
1203 { 189 {
1204 struct cpuhp_cpu_state *st; !! 190 BUG_ON(cpu_notify(val, v));
1205 int cpu; <<
1206 <<
1207 for_each_possible_cpu(cpu) { <<
1208 st = per_cpu_ptr(&cpuhp_state <<
1209 init_completion(&st->done_up) <<
1210 init_completion(&st->done_dow <<
1211 } <<
1212 } 191 }
>> 192 EXPORT_SYMBOL(register_cpu_notifier);
1213 193
1214 void __init cpuhp_threads_init(void) !! 194 void __ref unregister_cpu_notifier(struct notifier_block *nb)
1215 { 195 {
1216 cpuhp_init_state(); !! 196 cpu_maps_update_begin();
1217 BUG_ON(smpboot_register_percpu_thread !! 197 raw_notifier_chain_unregister(&cpu_chain, nb);
1218 kthread_unpark(this_cpu_read(cpuhp_st !! 198 cpu_maps_update_done();
1219 } 199 }
>> 200 EXPORT_SYMBOL(unregister_cpu_notifier);
1220 201
1221 #ifdef CONFIG_HOTPLUG_CPU 202 #ifdef CONFIG_HOTPLUG_CPU
1222 #ifndef arch_clear_mm_cpumask_cpu <<
1223 #define arch_clear_mm_cpumask_cpu(cpu, mm) cp <<
1224 #endif <<
1225 <<
1226 /** 203 /**
1227 * clear_tasks_mm_cpumask - Safely clear task 204 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
1228 * @cpu: a CPU id 205 * @cpu: a CPU id
1229 * 206 *
1230 * This function walks all processes, finds a 207 * This function walks all processes, finds a valid mm struct for each one and
1231 * then clears a corresponding bit in mm's cp 208 * then clears a corresponding bit in mm's cpumask. While this all sounds
1232 * trivial, there are various non-obvious cor 209 * trivial, there are various non-obvious corner cases, which this function
1233 * tries to solve in a safe manner. 210 * tries to solve in a safe manner.
1234 * 211 *
1235 * Also note that the function uses a somewha 212 * Also note that the function uses a somewhat relaxed locking scheme, so it may
1236 * be called only for an already offlined CPU 213 * be called only for an already offlined CPU.
1237 */ 214 */
1238 void clear_tasks_mm_cpumask(int cpu) 215 void clear_tasks_mm_cpumask(int cpu)
1239 { 216 {
1240 struct task_struct *p; 217 struct task_struct *p;
1241 218
1242 /* 219 /*
1243 * This function is called after the 220 * This function is called after the cpu is taken down and marked
1244 * offline, so its not like new tasks 221 * offline, so its not like new tasks will ever get this cpu set in
1245 * their mm mask. -- Peter Zijlstra 222 * their mm mask. -- Peter Zijlstra
1246 * Thus, we may use rcu_read_lock() h 223 * Thus, we may use rcu_read_lock() here, instead of grabbing
1247 * full-fledged tasklist_lock. 224 * full-fledged tasklist_lock.
1248 */ 225 */
1249 WARN_ON(cpu_online(cpu)); 226 WARN_ON(cpu_online(cpu));
1250 rcu_read_lock(); 227 rcu_read_lock();
1251 for_each_process(p) { 228 for_each_process(p) {
1252 struct task_struct *t; 229 struct task_struct *t;
1253 230
1254 /* 231 /*
1255 * Main thread might exit, bu 232 * Main thread might exit, but other threads may still have
1256 * a valid mm. Find one. 233 * a valid mm. Find one.
1257 */ 234 */
1258 t = find_lock_task_mm(p); 235 t = find_lock_task_mm(p);
1259 if (!t) 236 if (!t)
1260 continue; 237 continue;
1261 arch_clear_mm_cpumask_cpu(cpu !! 238 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
1262 task_unlock(t); 239 task_unlock(t);
1263 } 240 }
1264 rcu_read_unlock(); 241 rcu_read_unlock();
1265 } 242 }
1266 243
>> 244 static inline void check_for_tasks(int cpu)
>> 245 {
>> 246 struct task_struct *p;
>> 247 cputime_t utime, stime;
>> 248
>> 249 write_lock_irq(&tasklist_lock);
>> 250 for_each_process(p) {
>> 251 task_cputime(p, &utime, &stime);
>> 252 if (task_cpu(p) == cpu && p->state == TASK_RUNNING &&
>> 253 (utime || stime))
>> 254 printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d "
>> 255 "(state = %ld, flags = %x)\n",
>> 256 p->comm, task_pid_nr(p), cpu,
>> 257 p->state, p->flags);
>> 258 }
>> 259 write_unlock_irq(&tasklist_lock);
>> 260 }
>> 261
>> 262 struct take_cpu_down_param {
>> 263 unsigned long mod;
>> 264 void *hcpu;
>> 265 };
>> 266
1267 /* Take this CPU down. */ 267 /* Take this CPU down. */
1268 static int take_cpu_down(void *_param) !! 268 static int __ref take_cpu_down(void *_param)
1269 { 269 {
1270 struct cpuhp_cpu_state *st = this_cpu !! 270 struct take_cpu_down_param *param = _param;
1271 enum cpuhp_state target = max((int)st !! 271 int err;
1272 int err, cpu = smp_processor_id(); <<
1273 272
1274 /* Ensure this CPU doesn't handle any 273 /* Ensure this CPU doesn't handle any more interrupts. */
1275 err = __cpu_disable(); 274 err = __cpu_disable();
1276 if (err < 0) 275 if (err < 0)
1277 return err; 276 return err;
1278 277
1279 /* !! 278 cpu_notify(CPU_DYING | param->mod, param->hcpu);
1280 * Must be called from CPUHP_TEARDOWN <<
1281 * down, that the current state is CP <<
1282 */ <<
1283 WARN_ON(st->state != (CPUHP_TEARDOWN_ <<
1284 <<
1285 /* <<
1286 * Invoke the former CPU_DYING callba <<
1287 */ <<
1288 cpuhp_invoke_callback_range_nofail(fa <<
1289 <<
1290 /* Park the stopper thread */ 279 /* Park the stopper thread */
1291 stop_machine_park(cpu); !! 280 kthread_park(current);
1292 return 0; 281 return 0;
1293 } 282 }
1294 283
1295 static int takedown_cpu(unsigned int cpu) !! 284 /* Requires cpu_add_remove_lock to be held */
>> 285 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
1296 { 286 {
1297 struct cpuhp_cpu_state *st = per_cpu_ !! 287 int err, nr_calls = 0;
1298 int err; !! 288 void *hcpu = (void *)(long)cpu;
>> 289 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
>> 290 struct take_cpu_down_param tcd_param = {
>> 291 .mod = mod,
>> 292 .hcpu = hcpu,
>> 293 };
1299 294
1300 /* Park the smpboot threads */ !! 295 if (num_online_cpus() == 1)
1301 kthread_park(st->thread); !! 296 return -EBUSY;
1302 297
1303 /* !! 298 if (!cpu_online(cpu))
1304 * Prevent irq alloc/free while the d !! 299 return -EINVAL;
1305 * interrupt affinities. <<
1306 */ <<
1307 irq_lock_sparse(); <<
1308 300
1309 /* !! 301 cpu_hotplug_begin();
1310 * So now all preempt/rcu users must !! 302
1311 */ !! 303 err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
1312 err = stop_machine_cpuslocked(take_cp <<
1313 if (err) { 304 if (err) {
1314 /* CPU refused to die */ !! 305 nr_calls--;
1315 irq_unlock_sparse(); !! 306 __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL);
1316 /* Unpark the hotplug thread !! 307 printk("%s: attempt to take down CPU %u failed\n",
1317 kthread_unpark(st->thread); !! 308 __func__, cpu);
1318 return err; !! 309 goto out_release;
>> 310 }
>> 311 smpboot_park_threads(cpu);
>> 312
>> 313 err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
>> 314 if (err) {
>> 315 /* CPU didn't die: tell everyone. Can't complain. */
>> 316 smpboot_unpark_threads(cpu);
>> 317 cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
>> 318 goto out_release;
1319 } 319 }
1320 BUG_ON(cpu_online(cpu)); 320 BUG_ON(cpu_online(cpu));
1321 321
1322 /* 322 /*
1323 * The teardown callback for CPUHP_AP !! 323 * The migration_call() CPU_DYING callback will have removed all
1324 * all runnable tasks from the CPU, t !! 324 * runnable tasks from the cpu, there's only the idle task left now
1325 * that the migration thread is done 325 * that the migration thread is done doing the stop_machine thing.
1326 * 326 *
1327 * Wait for the stop thread to go awa 327 * Wait for the stop thread to go away.
1328 */ 328 */
1329 wait_for_ap_thread(st, false); !! 329 while (!idle_cpu(cpu))
1330 BUG_ON(st->state != CPUHP_AP_IDLE_DEA !! 330 cpu_relax();
1331 <<
1332 /* Interrupts are moved away from the <<
1333 irq_unlock_sparse(); <<
1334 331
1335 hotplug_cpu__broadcast_tick_pull(cpu) <<
1336 /* This actually kills the CPU. */ 332 /* This actually kills the CPU. */
1337 __cpu_die(cpu); 333 __cpu_die(cpu);
1338 334
1339 cpuhp_bp_sync_dead(cpu); !! 335 /* CPU is completely dead: tell everyone. Too late to complain. */
1340 !! 336 cpu_notify_nofail(CPU_DEAD | mod, hcpu);
1341 tick_cleanup_dead_cpu(cpu); <<
1342 <<
1343 /* <<
1344 * Callbacks must be re-integrated ri <<
1345 * Otherwise an RCU callback could bl <<
1346 * waiting for its completion. <<
1347 */ <<
1348 rcutree_migrate_callbacks(cpu); <<
1349 <<
1350 return 0; <<
1351 } <<
1352 <<
1353 static void cpuhp_complete_idle_dead(void *ar <<
1354 { <<
1355 struct cpuhp_cpu_state *st = arg; <<
1356 <<
1357 complete_ap_thread(st, false); <<
1358 } <<
1359 <<
1360 void cpuhp_report_idle_dead(void) <<
1361 { <<
1362 struct cpuhp_cpu_state *st = this_cpu <<
1363 <<
1364 BUG_ON(st->state != CPUHP_AP_OFFLINE) <<
1365 tick_assert_timekeeping_handover(); <<
1366 rcutree_report_cpu_dead(); <<
1367 st->state = CPUHP_AP_IDLE_DEAD; <<
1368 /* <<
1369 * We cannot call complete after rcut <<
1370 * to an online cpu. <<
1371 */ <<
1372 smp_call_function_single(cpumask_firs <<
1373 cpuhp_comple <<
1374 } <<
1375 <<
1376 static int cpuhp_down_callbacks(unsigned int <<
1377 enum cpuhp_st <<
1378 { <<
1379 enum cpuhp_state prev_state = st->sta <<
1380 int ret = 0; <<
1381 <<
1382 ret = cpuhp_invoke_callback_range(fal <<
1383 if (ret) { <<
1384 pr_debug("CPU DOWN failed (%d <<
1385 ret, cpu, cpuhp_get_ <<
1386 st->state); <<
1387 <<
1388 cpuhp_reset_state(cpu, st, pr <<
1389 <<
1390 if (st->state < prev_state) <<
1391 WARN_ON(cpuhp_invoke_ <<
1392 <<
1393 } <<
1394 <<
1395 return ret; <<
1396 } <<
1397 <<
1398 /* Requires cpu_add_remove_lock to be held */ <<
1399 static int __ref _cpu_down(unsigned int cpu, <<
1400 enum cpuhp_state t <<
1401 { <<
1402 struct cpuhp_cpu_state *st = per_cpu_ <<
1403 int prev_state, ret = 0; <<
1404 <<
1405 if (num_online_cpus() == 1) <<
1406 return -EBUSY; <<
1407 <<
1408 if (!cpu_present(cpu)) <<
1409 return -EINVAL; <<
1410 <<
1411 cpus_write_lock(); <<
1412 <<
1413 cpuhp_tasks_frozen = tasks_frozen; <<
1414 <<
1415 prev_state = cpuhp_set_state(cpu, st, <<
1416 /* <<
1417 * If the current CPU state is in the <<
1418 * then we need to kick the thread. <<
1419 */ <<
1420 if (st->state > CPUHP_TEARDOWN_CPU) { <<
1421 st->target = max((int)target, <<
1422 ret = cpuhp_kick_ap_work(cpu) <<
1423 /* <<
1424 * The AP side has done the e <<
1425 * return the error code.. <<
1426 */ <<
1427 if (ret) <<
1428 goto out; <<
1429 <<
1430 /* <<
1431 * We might have stopped stil <<
1432 * thread. Nothing to do anym <<
1433 */ <<
1434 if (st->state > CPUHP_TEARDOW <<
1435 goto out; <<
1436 <<
1437 st->target = target; <<
1438 } <<
1439 /* <<
1440 * The AP brought itself down to CPUH <<
1441 * to do the further cleanups. <<
1442 */ <<
1443 ret = cpuhp_down_callbacks(cpu, st, t <<
1444 if (ret && st->state < prev_state) { <<
1445 if (st->state == CPUHP_TEARDO <<
1446 cpuhp_reset_state(cpu <<
1447 __cpuhp_kick_ap(st); <<
1448 } else { <<
1449 WARN(1, "DEAD callbac <<
1450 } <<
1451 } <<
1452 337
1453 out: !! 338 check_for_tasks(cpu);
1454 cpus_write_unlock(); <<
1455 /* <<
1456 * Do post unplug cleanup. This is st <<
1457 * concurrent CPU hotplug via cpu_add <<
1458 */ <<
1459 lockup_detector_cleanup(); <<
1460 arch_smt_update(); <<
1461 return ret; <<
1462 } <<
1463 <<
1464 struct cpu_down_work { <<
1465 unsigned int cpu; <<
1466 enum cpuhp_state target; <<
1467 }; <<
1468 <<
1469 static long __cpu_down_maps_locked(void *arg) <<
1470 { <<
1471 struct cpu_down_work *work = arg; <<
1472 339
1473 return _cpu_down(work->cpu, 0, work-> !! 340 out_release:
1474 } !! 341 cpu_hotplug_done();
1475 !! 342 if (!err)
1476 static int cpu_down_maps_locked(unsigned int !! 343 cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
1477 { <<
1478 struct cpu_down_work work = { .cpu = <<
1479 <<
1480 /* <<
1481 * If the platform does not support h <<
1482 * differentiate it from a transient <<
1483 */ <<
1484 if (cpu_hotplug_offline_disabled) <<
1485 return -EOPNOTSUPP; <<
1486 if (cpu_hotplug_disabled) <<
1487 return -EBUSY; <<
1488 <<
1489 /* <<
1490 * Ensure that the control task does <<
1491 * CPU to prevent a deadlock against <<
1492 * Also keep at least one housekeepin <<
1493 * an empty sched_domain span. <<
1494 */ <<
1495 for_each_cpu_and(cpu, cpu_online_mask <<
1496 if (cpu != work.cpu) <<
1497 return work_on_cpu(cp <<
1498 } <<
1499 return -EBUSY; <<
1500 } <<
1501 <<
1502 static int cpu_down(unsigned int cpu, enum cp <<
1503 { <<
1504 int err; <<
1505 <<
1506 cpu_maps_update_begin(); <<
1507 err = cpu_down_maps_locked(cpu, targe <<
1508 cpu_maps_update_done(); <<
1509 return err; 344 return err;
1510 } 345 }
1511 346
1512 /** !! 347 int __ref cpu_down(unsigned int cpu)
1513 * cpu_device_down - Bring down a cpu device <<
1514 * @dev: Pointer to the cpu device to offline <<
1515 * <<
1516 * This function is meant to be used by devic <<
1517 * <<
1518 * Other subsystems should use remove_cpu() i <<
1519 * <<
1520 * Return: %0 on success or a negative errno <<
1521 */ <<
1522 int cpu_device_down(struct device *dev) <<
1523 { <<
1524 return cpu_down(dev->id, CPUHP_OFFLIN <<
1525 } <<
1526 <<
1527 int remove_cpu(unsigned int cpu) <<
1528 { 348 {
1529 int ret; !! 349 int err;
1530 <<
1531 lock_device_hotplug(); <<
1532 ret = device_offline(get_cpu_device(c <<
1533 unlock_device_hotplug(); <<
1534 <<
1535 return ret; <<
1536 } <<
1537 EXPORT_SYMBOL_GPL(remove_cpu); <<
1538 <<
1539 void smp_shutdown_nonboot_cpus(unsigned int p <<
1540 { <<
1541 unsigned int cpu; <<
1542 int error; <<
1543 350
1544 cpu_maps_update_begin(); 351 cpu_maps_update_begin();
1545 352
1546 /* !! 353 if (cpu_hotplug_disabled) {
1547 * Make certain the cpu I'm about to !! 354 err = -EBUSY;
1548 * !! 355 goto out;
1549 * This is inline to what migrate_to_ <<
1550 */ <<
1551 if (!cpu_online(primary_cpu)) <<
1552 primary_cpu = cpumask_first(c <<
1553 <<
1554 for_each_online_cpu(cpu) { <<
1555 if (cpu == primary_cpu) <<
1556 continue; <<
1557 <<
1558 error = cpu_down_maps_locked( <<
1559 if (error) { <<
1560 pr_err("Failed to off <<
1561 cpu, error); <<
1562 break; <<
1563 } <<
1564 } 356 }
1565 357
1566 /* !! 358 err = _cpu_down(cpu, 0);
1567 * Ensure all but the reboot CPU are <<
1568 */ <<
1569 BUG_ON(num_online_cpus() > 1); <<
1570 <<
1571 /* <<
1572 * Make sure the CPUs won't be enable <<
1573 * point. Kexec will reboot to a new <<
1574 * everything along the way. <<
1575 */ <<
1576 cpu_hotplug_disabled++; <<
1577 359
>> 360 out:
1578 cpu_maps_update_done(); 361 cpu_maps_update_done();
>> 362 return err;
1579 } 363 }
1580 !! 364 EXPORT_SYMBOL(cpu_down);
1581 #else <<
1582 #define takedown_cpu NULL <<
1583 #endif /*CONFIG_HOTPLUG_CPU*/ 365 #endif /*CONFIG_HOTPLUG_CPU*/
1584 366
1585 /** <<
1586 * notify_cpu_starting(cpu) - Invoke the call <<
1587 * @cpu: cpu that just started <<
1588 * <<
1589 * It must be called by the arch code on the <<
1590 * enables interrupts and before the "boot" c <<
1591 */ <<
1592 void notify_cpu_starting(unsigned int cpu) <<
1593 { <<
1594 struct cpuhp_cpu_state *st = per_cpu_ <<
1595 enum cpuhp_state target = min((int)st <<
1596 <<
1597 rcutree_report_cpu_starting(cpu); <<
1598 cpumask_set_cpu(cpu, &cpus_booted_onc <<
1599 <<
1600 /* <<
1601 * STARTING must not fail! <<
1602 */ <<
1603 cpuhp_invoke_callback_range_nofail(tr <<
1604 } <<
1605 <<
1606 /* <<
1607 * Called from the idle task. Wake up the con <<
1608 * hotplug thread of the upcoming CPU up and <<
1609 * online bringup to the hotplug thread. <<
1610 */ <<
1611 void cpuhp_online_idle(enum cpuhp_state state <<
1612 { <<
1613 struct cpuhp_cpu_state *st = this_cpu <<
1614 <<
1615 /* Happens for the boot cpu */ <<
1616 if (state != CPUHP_AP_ONLINE_IDLE) <<
1617 return; <<
1618 <<
1619 cpuhp_ap_update_sync_state(SYNC_STATE <<
1620 <<
1621 /* <<
1622 * Unpark the stopper thread before w <<
1623 * scheduling); this ensures the stop <<
1624 */ <<
1625 stop_machine_unpark(smp_processor_id( <<
1626 <<
1627 st->state = CPUHP_AP_ONLINE_IDLE; <<
1628 complete_ap_thread(st, true); <<
1629 } <<
1630 <<
1631 /* Requires cpu_add_remove_lock to be held */ 367 /* Requires cpu_add_remove_lock to be held */
1632 static int _cpu_up(unsigned int cpu, int task !! 368 static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
1633 { 369 {
1634 struct cpuhp_cpu_state *st = per_cpu_ !! 370 int ret, nr_calls = 0;
>> 371 void *hcpu = (void *)(long)cpu;
>> 372 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
1635 struct task_struct *idle; 373 struct task_struct *idle;
1636 int ret = 0; <<
1637 374
1638 cpus_write_lock(); !! 375 cpu_hotplug_begin();
1639 376
1640 if (!cpu_present(cpu)) { !! 377 if (cpu_online(cpu) || !cpu_present(cpu)) {
1641 ret = -EINVAL; 378 ret = -EINVAL;
1642 goto out; 379 goto out;
1643 } 380 }
1644 381
1645 /* !! 382 idle = idle_thread_get(cpu);
1646 * The caller of cpu_up() might have !! 383 if (IS_ERR(idle)) {
1647 * caller. Nothing to do. !! 384 ret = PTR_ERR(idle);
1648 */ <<
1649 if (st->state >= target) <<
1650 goto out; 385 goto out;
1651 <<
1652 if (st->state == CPUHP_OFFLINE) { <<
1653 /* Let it fail before we try <<
1654 idle = idle_thread_get(cpu); <<
1655 if (IS_ERR(idle)) { <<
1656 ret = PTR_ERR(idle); <<
1657 goto out; <<
1658 } <<
1659 <<
1660 /* <<
1661 * Reset stale stack state fr <<
1662 */ <<
1663 scs_task_reset(idle); <<
1664 kasan_unpoison_task_stack(idl <<
1665 } 386 }
1666 387
1667 cpuhp_tasks_frozen = tasks_frozen; !! 388 ret = smpboot_create_threads(cpu);
>> 389 if (ret)
>> 390 goto out;
1668 391
1669 cpuhp_set_state(cpu, st, target); !! 392 ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls);
1670 /* !! 393 if (ret) {
1671 * If the current CPU state is in the !! 394 nr_calls--;
1672 * then we need to kick the thread on !! 395 printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
1673 */ !! 396 __func__, cpu);
1674 if (st->state > CPUHP_BRINGUP_CPU) { !! 397 goto out_notify;
1675 ret = cpuhp_kick_ap_work(cpu) <<
1676 /* <<
1677 * The AP side has done the e <<
1678 * return the error code.. <<
1679 */ <<
1680 if (ret) <<
1681 goto out; <<
1682 } 398 }
1683 399
1684 /* !! 400 /* Arch-specific enabling code. */
1685 * Try to reach the target state. We !! 401 ret = __cpu_up(cpu, idle);
1686 * CPUHP_BRINGUP_CPU. After that the !! 402 if (ret != 0)
1687 * responsible for bringing it up to !! 403 goto out_notify;
1688 */ !! 404 BUG_ON(!cpu_online(cpu));
1689 target = min((int)target, CPUHP_BRING !! 405
1690 ret = cpuhp_up_callbacks(cpu, st, tar !! 406 /* Wake the per cpu threads */
>> 407 smpboot_unpark_threads(cpu);
>> 408
>> 409 /* Now call notifier in preparation. */
>> 410 cpu_notify(CPU_ONLINE | mod, hcpu);
>> 411
>> 412 out_notify:
>> 413 if (ret != 0)
>> 414 __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
1691 out: 415 out:
1692 cpus_write_unlock(); !! 416 cpu_hotplug_done();
1693 arch_smt_update(); !! 417
1694 return ret; 418 return ret;
1695 } 419 }
1696 420
1697 static int cpu_up(unsigned int cpu, enum cpuh !! 421 int __cpuinit cpu_up(unsigned int cpu)
1698 { 422 {
1699 int err = 0; 423 int err = 0;
1700 424
>> 425 #ifdef CONFIG_MEMORY_HOTPLUG
>> 426 int nid;
>> 427 pg_data_t *pgdat;
>> 428 #endif
>> 429
1701 if (!cpu_possible(cpu)) { 430 if (!cpu_possible(cpu)) {
1702 pr_err("can't online cpu %d b !! 431 printk(KERN_ERR "can't online cpu %d because it is not "
1703 cpu); !! 432 "configured as may-hotadd at boot time\n", cpu);
>> 433 #if defined(CONFIG_IA64)
>> 434 printk(KERN_ERR "please check additional_cpus= boot "
>> 435 "parameter\n");
>> 436 #endif
1704 return -EINVAL; 437 return -EINVAL;
1705 } 438 }
1706 439
1707 err = try_online_node(cpu_to_node(cpu !! 440 #ifdef CONFIG_MEMORY_HOTPLUG
1708 if (err) !! 441 nid = cpu_to_node(cpu);
1709 return err; !! 442 if (!node_online(nid)) {
1710 !! 443 err = mem_online_node(nid);
1711 cpu_maps_update_begin(); !! 444 if (err)
1712 !! 445 return err;
1713 if (cpu_hotplug_disabled) { <<
1714 err = -EBUSY; <<
1715 goto out; <<
1716 } 446 }
1717 if (!cpu_bootable(cpu)) { <<
1718 err = -EPERM; <<
1719 goto out; <<
1720 } <<
1721 <<
1722 err = _cpu_up(cpu, 0, target); <<
1723 out: <<
1724 cpu_maps_update_done(); <<
1725 return err; <<
1726 } <<
1727 <<
1728 /** <<
1729 * cpu_device_up - Bring up a cpu device <<
1730 * @dev: Pointer to the cpu device to online <<
1731 * <<
1732 * This function is meant to be used by devic <<
1733 * <<
1734 * Other subsystems should use add_cpu() inst <<
1735 * <<
1736 * Return: %0 on success or a negative errno <<
1737 */ <<
1738 int cpu_device_up(struct device *dev) <<
1739 { <<
1740 return cpu_up(dev->id, CPUHP_ONLINE); <<
1741 } <<
1742 <<
1743 int add_cpu(unsigned int cpu) <<
1744 { <<
1745 int ret; <<
1746 <<
1747 lock_device_hotplug(); <<
1748 ret = device_online(get_cpu_device(cp <<
1749 unlock_device_hotplug(); <<
1750 <<
1751 return ret; <<
1752 } <<
1753 EXPORT_SYMBOL_GPL(add_cpu); <<
1754 447
1755 /** !! 448 pgdat = NODE_DATA(nid);
1756 * bringup_hibernate_cpu - Bring up the CPU t !! 449 if (!pgdat) {
1757 * @sleep_cpu: The cpu we hibernated on and s !! 450 printk(KERN_ERR
1758 * !! 451 "Can't online cpu %d due to NULL pgdat\n", cpu);
1759 * On some architectures like arm64, we can h !! 452 return -ENOMEM;
1760 * wake up the CPU we hibernated on might be <<
1761 * using maxcpus= for example. <<
1762 * <<
1763 * Return: %0 on success or a negative errno <<
1764 */ <<
1765 int bringup_hibernate_cpu(unsigned int sleep_ <<
1766 { <<
1767 int ret; <<
1768 <<
1769 if (!cpu_online(sleep_cpu)) { <<
1770 pr_info("Hibernated on a CPU <<
1771 ret = cpu_up(sleep_cpu, CPUHP <<
1772 if (ret) { <<
1773 pr_err("Failed to bri <<
1774 return ret; <<
1775 } <<
1776 } 453 }
1777 return 0; <<
1778 } <<
1779 <<
1780 static void __init cpuhp_bringup_mask(const s <<
1781 enum cp <<
1782 { <<
1783 unsigned int cpu; <<
1784 <<
1785 for_each_cpu(cpu, mask) { <<
1786 struct cpuhp_cpu_state *st = <<
1787 454
1788 if (cpu_up(cpu, target) && ca !! 455 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1789 /* !! 456 mutex_lock(&zonelists_mutex);
1790 * If this failed the !! 457 build_all_zonelists(NULL, NULL);
1791 * rolled back to CPU !! 458 mutex_unlock(&zonelists_mutex);
1792 * online. Clean it u <<
1793 */ <<
1794 WARN_ON(cpuhp_invoke_ <<
1795 } <<
1796 <<
1797 if (!--ncpus) <<
1798 break; <<
1799 } 459 }
1800 } <<
1801 <<
1802 #ifdef CONFIG_HOTPLUG_PARALLEL <<
1803 static bool __cpuhp_parallel_bringup __ro_aft <<
1804 <<
1805 static int __init parallel_bringup_parse_para <<
1806 { <<
1807 return kstrtobool(arg, &__cpuhp_paral <<
1808 } <<
1809 early_param("cpuhp.parallel", parallel_bringu <<
1810 <<
1811 #ifdef CONFIG_HOTPLUG_SMT <<
1812 static inline bool cpuhp_smt_aware(void) <<
1813 { <<
1814 return cpu_smt_max_threads > 1; <<
1815 } <<
1816 <<
1817 static inline const struct cpumask *cpuhp_get <<
1818 { <<
1819 return cpu_primary_thread_mask; <<
1820 } <<
1821 #else <<
1822 static inline bool cpuhp_smt_aware(void) <<
1823 { <<
1824 return false; <<
1825 } <<
1826 static inline const struct cpumask *cpuhp_get <<
1827 { <<
1828 return cpu_none_mask; <<
1829 } <<
1830 #endif 460 #endif
1831 461
1832 bool __weak arch_cpuhp_init_parallel_bringup( !! 462 cpu_maps_update_begin();
1833 { <<
1834 return true; <<
1835 } <<
1836 <<
1837 /* <<
1838 * On architectures which have enabled parall <<
1839 * prepare states for each of the to be onlin <<
1840 * sends the startup IPI to the APs. The APs <<
1841 * bringup code in parallel and then wait for <<
1842 * them one by one for the final onlining pro <<
1843 * <<
1844 * This avoids waiting for each AP to respond <<
1845 * CPUHP_BRINGUP_CPU. <<
1846 */ <<
1847 static bool __init cpuhp_bringup_cpus_paralle <<
1848 { <<
1849 const struct cpumask *mask = cpu_pres <<
1850 <<
1851 if (__cpuhp_parallel_bringup) <<
1852 __cpuhp_parallel_bringup = ar <<
1853 if (!__cpuhp_parallel_bringup) <<
1854 return false; <<
1855 <<
1856 if (cpuhp_smt_aware()) { <<
1857 const struct cpumask *pmask = <<
1858 static struct cpumask tmp_mas <<
1859 463
1860 /* !! 464 if (cpu_hotplug_disabled) {
1861 * X86 requires to prevent th !! 465 err = -EBUSY;
1862 * the primary thread does a !! 466 goto out;
1863 * reasons. Bring the primary <<
1864 */ <<
1865 cpumask_and(&tmp_mask, mask, <<
1866 cpuhp_bringup_mask(&tmp_mask, <<
1867 cpuhp_bringup_mask(&tmp_mask, <<
1868 /* Account for the online CPU <<
1869 ncpus -= num_online_cpus(); <<
1870 if (!ncpus) <<
1871 return true; <<
1872 /* Create the mask for second <<
1873 cpumask_andnot(&tmp_mask, mas <<
1874 mask = &tmp_mask; <<
1875 } 467 }
1876 468
1877 /* Bring the not-yet started CPUs up !! 469 err = _cpu_up(cpu, 0);
1878 cpuhp_bringup_mask(mask, ncpus, CPUHP <<
1879 cpuhp_bringup_mask(mask, ncpus, CPUHP <<
1880 return true; <<
1881 } <<
1882 #else <<
1883 static inline bool cpuhp_bringup_cpus_paralle <<
1884 #endif /* CONFIG_HOTPLUG_PARALLEL */ <<
1885 470
1886 void __init bringup_nonboot_cpus(unsigned int !! 471 out:
1887 { !! 472 cpu_maps_update_done();
1888 if (!max_cpus) !! 473 return err;
1889 return; <<
1890 <<
1891 /* Try parallel bringup optimization <<
1892 if (cpuhp_bringup_cpus_parallel(max_c <<
1893 return; <<
1894 <<
1895 /* Full per CPU serialized bringup */ <<
1896 cpuhp_bringup_mask(cpu_present_mask, <<
1897 } 474 }
>> 475 EXPORT_SYMBOL_GPL(cpu_up);
1898 476
1899 #ifdef CONFIG_PM_SLEEP_SMP 477 #ifdef CONFIG_PM_SLEEP_SMP
1900 static cpumask_var_t frozen_cpus; 478 static cpumask_var_t frozen_cpus;
1901 479
1902 int freeze_secondary_cpus(int primary) !! 480 int disable_nonboot_cpus(void)
1903 { 481 {
1904 int cpu, error = 0; !! 482 int cpu, first_cpu, error = 0;
1905 483
1906 cpu_maps_update_begin(); 484 cpu_maps_update_begin();
1907 if (primary == -1) { !! 485 first_cpu = cpumask_first(cpu_online_mask);
1908 primary = cpumask_first(cpu_o <<
1909 if (!housekeeping_cpu(primary <<
1910 primary = housekeepin <<
1911 } else { <<
1912 if (!cpu_online(primary)) <<
1913 primary = cpumask_fir <<
1914 } <<
1915 <<
1916 /* 486 /*
1917 * We take down all of the non-boot C 487 * We take down all of the non-boot CPUs in one shot to avoid races
1918 * with the userspace trying to use t 488 * with the userspace trying to use the CPU hotplug at the same time
1919 */ 489 */
1920 cpumask_clear(frozen_cpus); 490 cpumask_clear(frozen_cpus);
1921 491
1922 pr_info("Disabling non-boot CPUs ...\ !! 492 printk("Disabling non-boot CPUs ...\n");
1923 for (cpu = nr_cpu_ids - 1; cpu >= 0; !! 493 for_each_online_cpu(cpu) {
1924 if (!cpu_online(cpu) || cpu = !! 494 if (cpu == first_cpu)
1925 continue; 495 continue;
1926 !! 496 error = _cpu_down(cpu, 1);
1927 if (pm_wakeup_pending()) { <<
1928 pr_info("Wakeup pendi <<
1929 error = -EBUSY; <<
1930 break; <<
1931 } <<
1932 <<
1933 trace_suspend_resume(TPS("CPU <<
1934 error = _cpu_down(cpu, 1, CPU <<
1935 trace_suspend_resume(TPS("CPU <<
1936 if (!error) 497 if (!error)
1937 cpumask_set_cpu(cpu, 498 cpumask_set_cpu(cpu, frozen_cpus);
1938 else { 499 else {
1939 pr_err("Error taking !! 500 printk(KERN_ERR "Error taking CPU%d down: %d\n",
>> 501 cpu, error);
1940 break; 502 break;
1941 } 503 }
1942 } 504 }
1943 505
1944 if (!error) !! 506 if (!error) {
1945 BUG_ON(num_online_cpus() > 1) 507 BUG_ON(num_online_cpus() > 1);
1946 else !! 508 /* Make sure the CPUs won't be enabled by someone else */
1947 pr_err("Non-boot CPUs are not !! 509 cpu_hotplug_disabled = 1;
1948 !! 510 } else {
1949 /* !! 511 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
1950 * Make sure the CPUs won't be enable !! 512 }
1951 * this even in case of failure as al <<
1952 * supposed to do thaw_secondary_cpus <<
1953 */ <<
1954 cpu_hotplug_disabled++; <<
1955 <<
1956 cpu_maps_update_done(); 513 cpu_maps_update_done();
1957 return error; 514 return error;
1958 } 515 }
1959 516
1960 void __weak arch_thaw_secondary_cpus_begin(vo !! 517 void __weak arch_enable_nonboot_cpus_begin(void)
1961 { 518 {
1962 } 519 }
1963 520
1964 void __weak arch_thaw_secondary_cpus_end(void !! 521 void __weak arch_enable_nonboot_cpus_end(void)
1965 { 522 {
1966 } 523 }
1967 524
1968 void thaw_secondary_cpus(void) !! 525 void __ref enable_nonboot_cpus(void)
1969 { 526 {
1970 int cpu, error; 527 int cpu, error;
1971 528
1972 /* Allow everyone to use the CPU hotp 529 /* Allow everyone to use the CPU hotplug again */
1973 cpu_maps_update_begin(); 530 cpu_maps_update_begin();
1974 __cpu_hotplug_enable(); !! 531 cpu_hotplug_disabled = 0;
1975 if (cpumask_empty(frozen_cpus)) 532 if (cpumask_empty(frozen_cpus))
1976 goto out; 533 goto out;
1977 534
1978 pr_info("Enabling non-boot CPUs ...\n !! 535 printk(KERN_INFO "Enabling non-boot CPUs ...\n");
1979 536
1980 arch_thaw_secondary_cpus_begin(); !! 537 arch_enable_nonboot_cpus_begin();
1981 538
1982 for_each_cpu(cpu, frozen_cpus) { 539 for_each_cpu(cpu, frozen_cpus) {
1983 trace_suspend_resume(TPS("CPU !! 540 error = _cpu_up(cpu, 1);
1984 error = _cpu_up(cpu, 1, CPUHP <<
1985 trace_suspend_resume(TPS("CPU <<
1986 if (!error) { 541 if (!error) {
1987 pr_info("CPU%d is up\ !! 542 printk(KERN_INFO "CPU%d is up\n", cpu);
1988 continue; 543 continue;
1989 } 544 }
1990 pr_warn("Error taking CPU%d u !! 545 printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
1991 } 546 }
1992 547
1993 arch_thaw_secondary_cpus_end(); !! 548 arch_enable_nonboot_cpus_end();
1994 549
1995 cpumask_clear(frozen_cpus); 550 cpumask_clear(frozen_cpus);
1996 out: 551 out:
1997 cpu_maps_update_done(); 552 cpu_maps_update_done();
1998 } 553 }
1999 554
2000 static int __init alloc_frozen_cpus(void) 555 static int __init alloc_frozen_cpus(void)
2001 { 556 {
2002 if (!alloc_cpumask_var(&frozen_cpus, 557 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
2003 return -ENOMEM; 558 return -ENOMEM;
2004 return 0; 559 return 0;
2005 } 560 }
2006 core_initcall(alloc_frozen_cpus); 561 core_initcall(alloc_frozen_cpus);
2007 562
2008 /* 563 /*
2009 * When callbacks for CPU hotplug notificatio 564 * When callbacks for CPU hotplug notifications are being executed, we must
2010 * ensure that the state of the system with r 565 * ensure that the state of the system with respect to the tasks being frozen
2011 * or not, as reported by the notification, r 566 * or not, as reported by the notification, remains unchanged *throughout the
2012 * duration* of the execution of the callback 567 * duration* of the execution of the callbacks.
2013 * Hence we need to prevent the freezer from 568 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
2014 * 569 *
2015 * This synchronization is implemented by mut 570 * This synchronization is implemented by mutually excluding regular CPU
2016 * hotplug and Suspend/Hibernate call paths b 571 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
2017 * Hibernate notifications. 572 * Hibernate notifications.
2018 */ 573 */
2019 static int 574 static int
2020 cpu_hotplug_pm_callback(struct notifier_block 575 cpu_hotplug_pm_callback(struct notifier_block *nb,
2021 unsigned long action, 576 unsigned long action, void *ptr)
2022 { 577 {
2023 switch (action) { 578 switch (action) {
2024 579
2025 case PM_SUSPEND_PREPARE: 580 case PM_SUSPEND_PREPARE:
2026 case PM_HIBERNATION_PREPARE: 581 case PM_HIBERNATION_PREPARE:
2027 cpu_hotplug_disable(); 582 cpu_hotplug_disable();
2028 break; 583 break;
2029 584
2030 case PM_POST_SUSPEND: 585 case PM_POST_SUSPEND:
2031 case PM_POST_HIBERNATION: 586 case PM_POST_HIBERNATION:
2032 cpu_hotplug_enable(); 587 cpu_hotplug_enable();
2033 break; 588 break;
2034 589
2035 default: 590 default:
2036 return NOTIFY_DONE; 591 return NOTIFY_DONE;
2037 } 592 }
2038 593
2039 return NOTIFY_OK; 594 return NOTIFY_OK;
2040 } 595 }
2041 596
2042 597
2043 static int __init cpu_hotplug_pm_sync_init(vo 598 static int __init cpu_hotplug_pm_sync_init(void)
2044 { 599 {
2045 /* 600 /*
2046 * cpu_hotplug_pm_callback has higher 601 * cpu_hotplug_pm_callback has higher priority than x86
2047 * bsp_pm_callback which depends on c 602 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
2048 * to disable cpu hotplug to avoid cp 603 * to disable cpu hotplug to avoid cpu hotplug race.
2049 */ 604 */
2050 pm_notifier(cpu_hotplug_pm_callback, 605 pm_notifier(cpu_hotplug_pm_callback, 0);
2051 return 0; 606 return 0;
2052 } 607 }
2053 core_initcall(cpu_hotplug_pm_sync_init); 608 core_initcall(cpu_hotplug_pm_sync_init);
2054 609
2055 #endif /* CONFIG_PM_SLEEP_SMP */ 610 #endif /* CONFIG_PM_SLEEP_SMP */
2056 611
2057 int __boot_cpu_id; <<
2058 <<
2059 #endif /* CONFIG_SMP */ <<
2060 <<
2061 /* Boot processor state steps */ <<
2062 static struct cpuhp_step cpuhp_hp_states[] = <<
2063 [CPUHP_OFFLINE] = { <<
2064 .name = "of <<
2065 .startup.single = NUL <<
2066 .teardown.single = NUL <<
2067 }, <<
2068 #ifdef CONFIG_SMP <<
2069 [CPUHP_CREATE_THREADS]= { <<
2070 .name = "th <<
2071 .startup.single = smp <<
2072 .teardown.single = NUL <<
2073 .cant_stop = tru <<
2074 }, <<
2075 [CPUHP_PERF_PREPARE] = { <<
2076 .name = "pe <<
2077 .startup.single = per <<
2078 .teardown.single = per <<
2079 }, <<
2080 [CPUHP_RANDOM_PREPARE] = { <<
2081 .name = "ra <<
2082 .startup.single = ran <<
2083 .teardown.single = NUL <<
2084 }, <<
2085 [CPUHP_WORKQUEUE_PREP] = { <<
2086 .name = "wo <<
2087 .startup.single = wor <<
2088 .teardown.single = NUL <<
2089 }, <<
2090 [CPUHP_HRTIMERS_PREPARE] = { <<
2091 .name = "hr <<
2092 .startup.single = hrt <<
2093 .teardown.single = NUL <<
2094 }, <<
2095 [CPUHP_SMPCFD_PREPARE] = { <<
2096 .name = "sm <<
2097 .startup.single = smp <<
2098 .teardown.single = smp <<
2099 }, <<
2100 [CPUHP_RELAY_PREPARE] = { <<
2101 .name = "re <<
2102 .startup.single = rel <<
2103 .teardown.single = NUL <<
2104 }, <<
2105 [CPUHP_RCUTREE_PREP] = { <<
2106 .name = "RC <<
2107 .startup.single = rcu <<
2108 .teardown.single = rcu <<
2109 }, <<
2110 /* <<
2111 * On the tear-down path, timers_dead <<
2112 * before blk_mq_queue_reinit_notify( <<
2113 * otherwise a RCU stall occurs. <<
2114 */ <<
2115 [CPUHP_TIMERS_PREPARE] = { <<
2116 .name = "ti <<
2117 .startup.single = tim <<
2118 .teardown.single = tim <<
2119 }, <<
2120 <<
2121 #ifdef CONFIG_HOTPLUG_SPLIT_STARTUP <<
2122 /* <<
2123 * Kicks the AP alive. AP will wait i <<
2124 * the next step will release it. <<
2125 */ <<
2126 [CPUHP_BP_KICK_AP] = { <<
2127 .name = "cp <<
2128 .startup.single = cpu <<
2129 }, <<
2130 <<
2131 /* <<
2132 * Waits for the AP to reach cpuhp_ap <<
2133 * releases it for the complete bring <<
2134 */ <<
2135 [CPUHP_BRINGUP_CPU] = { <<
2136 .name = "cp <<
2137 .startup.single = cpu <<
2138 .teardown.single = fin <<
2139 .cant_stop = tru <<
2140 }, <<
2141 #else <<
2142 /* <<
2143 * All-in-one CPU bringup state which <<
2144 */ <<
2145 [CPUHP_BRINGUP_CPU] = { <<
2146 .name = "cp <<
2147 .startup.single = bri <<
2148 .teardown.single = fin <<
2149 .cant_stop = tru <<
2150 }, <<
2151 #endif <<
2152 /* Final state before CPU kills itsel <<
2153 [CPUHP_AP_IDLE_DEAD] = { <<
2154 .name = "id <<
2155 }, <<
2156 /* <<
2157 * Last state before CPU enters the i <<
2158 * for synchronization. <<
2159 */ <<
2160 [CPUHP_AP_OFFLINE] = { <<
2161 .name = "ap <<
2162 .cant_stop = tru <<
2163 }, <<
2164 /* First state is scheduler control. <<
2165 [CPUHP_AP_SCHED_STARTING] = { <<
2166 .name = "sc <<
2167 .startup.single = sch <<
2168 .teardown.single = sch <<
2169 }, <<
2170 [CPUHP_AP_RCUTREE_DYING] = { <<
2171 .name = "RC <<
2172 .startup.single = NUL <<
2173 .teardown.single = rcu <<
2174 }, <<
2175 [CPUHP_AP_SMPCFD_DYING] = { <<
2176 .name = "sm <<
2177 .startup.single = NUL <<
2178 .teardown.single = smp <<
2179 }, <<
2180 [CPUHP_AP_HRTIMERS_DYING] = { <<
2181 .name = "hr <<
2182 .startup.single = NUL <<
2183 .teardown.single = hrt <<
2184 }, <<
2185 [CPUHP_AP_TICK_DYING] = { <<
2186 .name = "ti <<
2187 .startup.single = NUL <<
2188 .teardown.single = tic <<
2189 }, <<
2190 /* Entry state on starting. Interrupt <<
2191 * state for synchronsization */ <<
2192 [CPUHP_AP_ONLINE] = { <<
2193 .name = "ap <<
2194 }, <<
2195 /* <<
2196 * Handled on control processor until <<
2197 * this itself. <<
2198 */ <<
2199 [CPUHP_TEARDOWN_CPU] = { <<
2200 .name = "cp <<
2201 .startup.single = NUL <<
2202 .teardown.single = tak <<
2203 .cant_stop = tru <<
2204 }, <<
2205 <<
2206 [CPUHP_AP_SCHED_WAIT_EMPTY] = { <<
2207 .name = "sc <<
2208 .startup.single = NUL <<
2209 .teardown.single = sch <<
2210 }, <<
2211 <<
2212 /* Handle smpboot threads park/unpark <<
2213 [CPUHP_AP_SMPBOOT_THREADS] = { <<
2214 .name = "sm <<
2215 .startup.single = smp <<
2216 .teardown.single = smp <<
2217 }, <<
2218 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = { <<
2219 .name = "ir <<
2220 .startup.single = irq <<
2221 .teardown.single = NUL <<
2222 }, <<
2223 [CPUHP_AP_PERF_ONLINE] = { <<
2224 .name = "pe <<
2225 .startup.single = per <<
2226 .teardown.single = per <<
2227 }, <<
2228 [CPUHP_AP_WATCHDOG_ONLINE] = { <<
2229 .name = "lo <<
2230 .startup.single = loc <<
2231 .teardown.single = loc <<
2232 }, <<
2233 [CPUHP_AP_WORKQUEUE_ONLINE] = { <<
2234 .name = "wo <<
2235 .startup.single = wor <<
2236 .teardown.single = wor <<
2237 }, <<
2238 [CPUHP_AP_RANDOM_ONLINE] = { <<
2239 .name = "ra <<
2240 .startup.single = ran <<
2241 .teardown.single = NUL <<
2242 }, <<
2243 [CPUHP_AP_RCUTREE_ONLINE] = { <<
2244 .name = "RC <<
2245 .startup.single = rcu <<
2246 .teardown.single = rcu <<
2247 }, <<
2248 #endif <<
2249 /* <<
2250 * The dynamically registered state s <<
2251 */ <<
2252 <<
2253 #ifdef CONFIG_SMP <<
2254 /* Last state is scheduler control se <<
2255 [CPUHP_AP_ACTIVE] = { <<
2256 .name = "sc <<
2257 .startup.single = sch <<
2258 .teardown.single = sch <<
2259 }, <<
2260 #endif <<
2261 <<
2262 /* CPU is fully up and running. */ <<
2263 [CPUHP_ONLINE] = { <<
2264 .name = "on <<
2265 .startup.single = NUL <<
2266 .teardown.single = NUL <<
2267 }, <<
2268 }; <<
2269 <<
2270 /* Sanity check for callbacks */ <<
2271 static int cpuhp_cb_check(enum cpuhp_state st <<
2272 { <<
2273 if (state <= CPUHP_OFFLINE || state > <<
2274 return -EINVAL; <<
2275 return 0; <<
2276 } <<
2277 <<
2278 /* <<
2279 * Returns a free for dynamic slot assignment <<
2280 * are protected by the cpuhp_slot_states mut <<
2281 * by having no name assigned. <<
2282 */ <<
2283 static int cpuhp_reserve_state(enum cpuhp_sta <<
2284 { <<
2285 enum cpuhp_state i, end; <<
2286 struct cpuhp_step *step; <<
2287 <<
2288 switch (state) { <<
2289 case CPUHP_AP_ONLINE_DYN: <<
2290 step = cpuhp_hp_states + CPUH <<
2291 end = CPUHP_AP_ONLINE_DYN_END <<
2292 break; <<
2293 case CPUHP_BP_PREPARE_DYN: <<
2294 step = cpuhp_hp_states + CPUH <<
2295 end = CPUHP_BP_PREPARE_DYN_EN <<
2296 break; <<
2297 default: <<
2298 return -EINVAL; <<
2299 } <<
2300 <<
2301 for (i = state; i <= end; i++, step++ <<
2302 if (!step->name) <<
2303 return i; <<
2304 } <<
2305 WARN(1, "No more dynamic states avail <<
2306 return -ENOSPC; <<
2307 } <<
2308 <<
2309 static int cpuhp_store_callbacks(enum cpuhp_s <<
2310 int (*startu <<
2311 int (*teardo <<
2312 bool multi_i <<
2313 { <<
2314 /* (Un)Install the callbacks for furt <<
2315 struct cpuhp_step *sp; <<
2316 int ret = 0; <<
2317 <<
2318 /* <<
2319 * If name is NULL, then the state ge <<
2320 * <<
2321 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_P <<
2322 * the first allocation from these dy <<
2323 * would trigger a new allocation and <<
2324 * empty) state, leaving the callback <<
2325 * dangling, which causes wreckage on <<
2326 */ <<
2327 if (name && (state == CPUHP_AP_ONLINE <<
2328 state == CPUHP_BP_PREPAR <<
2329 ret = cpuhp_reserve_state(sta <<
2330 if (ret < 0) <<
2331 return ret; <<
2332 state = ret; <<
2333 } <<
2334 sp = cpuhp_get_step(state); <<
2335 if (name && sp->name) <<
2336 return -EBUSY; <<
2337 <<
2338 sp->startup.single = startup; <<
2339 sp->teardown.single = teardown; <<
2340 sp->name = name; <<
2341 sp->multi_instance = multi_instance; <<
2342 INIT_HLIST_HEAD(&sp->list); <<
2343 return ret; <<
2344 } <<
2345 <<
2346 static void *cpuhp_get_teardown_cb(enum cpuhp <<
2347 { <<
2348 return cpuhp_get_step(state)->teardow <<
2349 } <<
2350 <<
2351 /* <<
2352 * Call the startup/teardown function for a s <<
2353 * on the current CPU. <<
2354 */ <<
2355 static int cpuhp_issue_call(int cpu, enum cpu <<
2356 struct hlist_node <<
2357 { <<
2358 struct cpuhp_step *sp = cpuhp_get_ste <<
2359 int ret; <<
2360 <<
2361 /* <<
2362 * If there's nothing to do, we done. <<
2363 * Relies on the union for multi_inst <<
2364 */ <<
2365 if (cpuhp_step_empty(bringup, sp)) <<
2366 return 0; <<
2367 /* <<
2368 * The non AP bound callbacks can fai <<
2369 * e.g. module removal we crash for n <<
2370 */ <<
2371 #ifdef CONFIG_SMP <<
2372 if (cpuhp_is_ap_state(state)) <<
2373 ret = cpuhp_invoke_ap_callbac <<
2374 else <<
2375 ret = cpuhp_invoke_callback(c <<
2376 #else <<
2377 ret = cpuhp_invoke_callback(cpu, stat <<
2378 #endif <<
2379 BUG_ON(ret && !bringup); <<
2380 return ret; <<
2381 } <<
2382 <<
2383 /* <<
2384 * Called from __cpuhp_setup_state on a recov <<
2385 * <<
2386 * Note: The teardown callbacks for rollback <<
2387 */ <<
2388 static void cpuhp_rollback_install(int failed <<
2389 struct hli <<
2390 { <<
2391 int cpu; <<
2392 <<
2393 /* Roll back the already executed ste <<
2394 for_each_present_cpu(cpu) { <<
2395 struct cpuhp_cpu_state *st = <<
2396 int cpustate = st->state; <<
2397 <<
2398 if (cpu >= failedcpu) <<
2399 break; <<
2400 <<
2401 /* Did we invoke the startup <<
2402 if (cpustate >= state) <<
2403 cpuhp_issue_call(cpu, <<
2404 } <<
2405 } <<
2406 <<
2407 int __cpuhp_state_add_instance_cpuslocked(enu <<
2408 str <<
2409 boo <<
2410 { <<
2411 struct cpuhp_step *sp; <<
2412 int cpu; <<
2413 int ret; <<
2414 <<
2415 lockdep_assert_cpus_held(); <<
2416 <<
2417 sp = cpuhp_get_step(state); <<
2418 if (sp->multi_instance == false) <<
2419 return -EINVAL; <<
2420 <<
2421 mutex_lock(&cpuhp_state_mutex); <<
2422 <<
2423 if (!invoke || !sp->startup.multi) <<
2424 goto add_node; <<
2425 <<
2426 /* <<
2427 * Try to call the startup callback f <<
2428 * depending on the hotplug state of <<
2429 */ <<
2430 for_each_present_cpu(cpu) { <<
2431 struct cpuhp_cpu_state *st = <<
2432 int cpustate = st->state; <<
2433 <<
2434 if (cpustate < state) <<
2435 continue; <<
2436 <<
2437 ret = cpuhp_issue_call(cpu, s <<
2438 if (ret) { <<
2439 if (sp->teardown.mult <<
2440 cpuhp_rollbac <<
2441 goto unlock; <<
2442 } <<
2443 } <<
2444 add_node: <<
2445 ret = 0; <<
2446 hlist_add_head(node, &sp->list); <<
2447 unlock: <<
2448 mutex_unlock(&cpuhp_state_mutex); <<
2449 return ret; <<
2450 } <<
2451 <<
2452 int __cpuhp_state_add_instance(enum cpuhp_sta <<
2453 bool invoke) <<
2454 { <<
2455 int ret; <<
2456 <<
2457 cpus_read_lock(); <<
2458 ret = __cpuhp_state_add_instance_cpus <<
2459 cpus_read_unlock(); <<
2460 return ret; <<
2461 } <<
2462 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance) <<
2463 <<
2464 /** <<
2465 * __cpuhp_setup_state_cpuslocked - Setup the <<
2466 * @state: The state to setup <<
2467 * @name: Name of the step <<
2468 * @invoke: If true, the startup <<
2469 * cpu state >= @state <<
2470 * @startup: startup callback func <<
2471 * @teardown: teardown callback fun <<
2472 * @multi_instance: State is set up for m <<
2473 * added afterwards. <<
2474 * <<
2475 * The caller needs to hold cpus read locked <<
2476 * Return: <<
2477 * On success: <<
2478 * Positive state number if @state is CP <<
2479 * 0 for all other states <<
2480 * On failure: proper (negative) error code <<
2481 */ <<
2482 int __cpuhp_setup_state_cpuslocked(enum cpuhp <<
2483 const char <<
2484 int (*star <<
2485 int (*tear <<
2486 bool multi <<
2487 { <<
2488 int cpu, ret = 0; <<
2489 bool dynstate; <<
2490 <<
2491 lockdep_assert_cpus_held(); <<
2492 <<
2493 if (cpuhp_cb_check(state) || !name) <<
2494 return -EINVAL; <<
2495 <<
2496 mutex_lock(&cpuhp_state_mutex); <<
2497 <<
2498 ret = cpuhp_store_callbacks(state, na <<
2499 multi_ins <<
2500 <<
2501 dynstate = state == CPUHP_AP_ONLINE_D <<
2502 if (ret > 0 && dynstate) { <<
2503 state = ret; <<
2504 ret = 0; <<
2505 } <<
2506 <<
2507 if (ret || !invoke || !startup) <<
2508 goto out; <<
2509 <<
2510 /* <<
2511 * Try to call the startup callback f <<
2512 * depending on the hotplug state of <<
2513 */ <<
2514 for_each_present_cpu(cpu) { <<
2515 struct cpuhp_cpu_state *st = <<
2516 int cpustate = st->state; <<
2517 <<
2518 if (cpustate < state) <<
2519 continue; <<
2520 <<
2521 ret = cpuhp_issue_call(cpu, s <<
2522 if (ret) { <<
2523 if (teardown) <<
2524 cpuhp_rollbac <<
2525 cpuhp_store_callbacks <<
2526 goto out; <<
2527 } <<
2528 } <<
2529 out: <<
2530 mutex_unlock(&cpuhp_state_mutex); <<
2531 /* <<
2532 * If the requested state is CPUHP_AP <<
2533 * return the dynamically allocated s <<
2534 */ <<
2535 if (!ret && dynstate) <<
2536 return state; <<
2537 return ret; <<
2538 } <<
2539 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked) <<
2540 <<
2541 int __cpuhp_setup_state(enum cpuhp_state stat <<
2542 const char *name, boo <<
2543 int (*startup)(unsign <<
2544 int (*teardown)(unsig <<
2545 bool multi_instance) <<
2546 { <<
2547 int ret; <<
2548 <<
2549 cpus_read_lock(); <<
2550 ret = __cpuhp_setup_state_cpuslocked( <<
2551 <<
2552 cpus_read_unlock(); <<
2553 return ret; <<
2554 } <<
2555 EXPORT_SYMBOL(__cpuhp_setup_state); <<
2556 <<
2557 int __cpuhp_state_remove_instance(enum cpuhp_ <<
2558 struct hlis <<
2559 { <<
2560 struct cpuhp_step *sp = cpuhp_get_ste <<
2561 int cpu; <<
2562 <<
2563 BUG_ON(cpuhp_cb_check(state)); <<
2564 <<
2565 if (!sp->multi_instance) <<
2566 return -EINVAL; <<
2567 <<
2568 cpus_read_lock(); <<
2569 mutex_lock(&cpuhp_state_mutex); <<
2570 <<
2571 if (!invoke || !cpuhp_get_teardown_cb <<
2572 goto remove; <<
2573 /* <<
2574 * Call the teardown callback for eac <<
2575 * on the hotplug state of the cpu. T <<
2576 * allowed to fail currently! <<
2577 */ <<
2578 for_each_present_cpu(cpu) { <<
2579 struct cpuhp_cpu_state *st = <<
2580 int cpustate = st->state; <<
2581 <<
2582 if (cpustate >= state) <<
2583 cpuhp_issue_call(cpu, <<
2584 } <<
2585 <<
2586 remove: <<
2587 hlist_del(node); <<
2588 mutex_unlock(&cpuhp_state_mutex); <<
2589 cpus_read_unlock(); <<
2590 <<
2591 return 0; <<
2592 } <<
2593 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instan <<
2594 <<
2595 /** 612 /**
2596 * __cpuhp_remove_state_cpuslocked - Remove t !! 613 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
2597 * @state: The state to remove !! 614 * @cpu: cpu that just started
2598 * @invoke: If true, the teardown functio <<
2599 * cpu state >= @state <<
2600 * 615 *
2601 * The caller needs to hold cpus read locked !! 616 * This function calls the cpu_chain notifiers with CPU_STARTING.
2602 * The teardown callback is currently not all !! 617 * It must be called by the arch code on the new cpu, before the new cpu
2603 * about module removal! !! 618 * enables interrupts and before the "boot" cpu returns from __cpu_up().
2604 */ 619 */
2605 void __cpuhp_remove_state_cpuslocked(enum cpu !! 620 void __cpuinit notify_cpu_starting(unsigned int cpu)
2606 { <<
2607 struct cpuhp_step *sp = cpuhp_get_ste <<
2608 int cpu; <<
2609 <<
2610 BUG_ON(cpuhp_cb_check(state)); <<
2611 <<
2612 lockdep_assert_cpus_held(); <<
2613 <<
2614 mutex_lock(&cpuhp_state_mutex); <<
2615 if (sp->multi_instance) { <<
2616 WARN(!hlist_empty(&sp->list), <<
2617 "Error: Removing state % <<
2618 state); <<
2619 goto remove; <<
2620 } <<
2621 <<
2622 if (!invoke || !cpuhp_get_teardown_cb <<
2623 goto remove; <<
2624 <<
2625 /* <<
2626 * Call the teardown callback for eac <<
2627 * on the hotplug state of the cpu. T <<
2628 * allowed to fail currently! <<
2629 */ <<
2630 for_each_present_cpu(cpu) { <<
2631 struct cpuhp_cpu_state *st = <<
2632 int cpustate = st->state; <<
2633 <<
2634 if (cpustate >= state) <<
2635 cpuhp_issue_call(cpu, <<
2636 } <<
2637 remove: <<
2638 cpuhp_store_callbacks(state, NULL, NU <<
2639 mutex_unlock(&cpuhp_state_mutex); <<
2640 } <<
2641 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked <<
2642 <<
2643 void __cpuhp_remove_state(enum cpuhp_state st <<
2644 { <<
2645 cpus_read_lock(); <<
2646 __cpuhp_remove_state_cpuslocked(state <<
2647 cpus_read_unlock(); <<
2648 } <<
2649 EXPORT_SYMBOL(__cpuhp_remove_state); <<
2650 <<
2651 #ifdef CONFIG_HOTPLUG_SMT <<
2652 static void cpuhp_offline_cpu_device(unsigned <<
2653 { <<
2654 struct device *dev = get_cpu_device(c <<
2655 <<
2656 dev->offline = true; <<
2657 /* Tell user space about the state ch <<
2658 kobject_uevent(&dev->kobj, KOBJ_OFFLI <<
2659 } <<
2660 <<
2661 static void cpuhp_online_cpu_device(unsigned <<
2662 { <<
2663 struct device *dev = get_cpu_device(c <<
2664 <<
2665 dev->offline = false; <<
2666 /* Tell user space about the state ch <<
2667 kobject_uevent(&dev->kobj, KOBJ_ONLIN <<
2668 } <<
2669 <<
2670 int cpuhp_smt_disable(enum cpuhp_smt_control <<
2671 { <<
2672 int cpu, ret = 0; <<
2673 <<
2674 cpu_maps_update_begin(); <<
2675 for_each_online_cpu(cpu) { <<
2676 if (topology_is_primary_threa <<
2677 continue; <<
2678 /* <<
2679 * Disable can be called with <<
2680 * from a higher to lower num <<
2681 */ <<
2682 if (ctrlval == CPU_SMT_ENABLE <<
2683 continue; <<
2684 ret = cpu_down_maps_locked(cp <<
2685 if (ret) <<
2686 break; <<
2687 /* <<
2688 * As this needs to hold the <<
2689 * to call device_offline() b <<
2690 * cpu_down() which takes cpu <<
2691 * needs to be held as this m <<
2692 * abusers of the hotplug mac <<
2693 * <<
2694 * So nothing would update de <<
2695 * leave the sysfs entry stal <<
2696 * smt control has been chang <<
2697 * called under the sysfs hot <<
2698 * serialized against the reg <<
2699 */ <<
2700 cpuhp_offline_cpu_device(cpu) <<
2701 } <<
2702 if (!ret) <<
2703 cpu_smt_control = ctrlval; <<
2704 cpu_maps_update_done(); <<
2705 return ret; <<
2706 } <<
2707 <<
2708 /* Check if the core a CPU belongs to is onli <<
2709 #if !defined(topology_is_core_online) <<
2710 static inline bool topology_is_core_online(un <<
2711 { <<
2712 return true; <<
2713 } <<
2714 #endif <<
2715 <<
2716 int cpuhp_smt_enable(void) <<
2717 { <<
2718 int cpu, ret = 0; <<
2719 <<
2720 cpu_maps_update_begin(); <<
2721 cpu_smt_control = CPU_SMT_ENABLED; <<
2722 for_each_present_cpu(cpu) { <<
2723 /* Skip online CPUs and CPUs <<
2724 if (cpu_online(cpu) || !node_ <<
2725 continue; <<
2726 if (!cpu_smt_thread_allowed(c <<
2727 continue; <<
2728 ret = _cpu_up(cpu, 0, CPUHP_O <<
2729 if (ret) <<
2730 break; <<
2731 /* See comment in cpuhp_smt_d <<
2732 cpuhp_online_cpu_device(cpu); <<
2733 } <<
2734 cpu_maps_update_done(); <<
2735 return ret; <<
2736 } <<
2737 #endif <<
2738 <<
2739 #if defined(CONFIG_SYSFS) && defined(CONFIG_H <<
2740 static ssize_t state_show(struct device *dev, <<
2741 struct device_attri <<
2742 { <<
2743 struct cpuhp_cpu_state *st = per_cpu_ <<
2744 <<
2745 return sprintf(buf, "%d\n", st->state <<
2746 } <<
2747 static DEVICE_ATTR_RO(state); <<
2748 <<
2749 static ssize_t target_store(struct device *de <<
2750 const char *buf, <<
2751 { <<
2752 struct cpuhp_cpu_state *st = per_cpu_ <<
2753 struct cpuhp_step *sp; <<
2754 int target, ret; <<
2755 <<
2756 ret = kstrtoint(buf, 10, &target); <<
2757 if (ret) <<
2758 return ret; <<
2759 <<
2760 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL <<
2761 if (target < CPUHP_OFFLINE || target <<
2762 return -EINVAL; <<
2763 #else <<
2764 if (target != CPUHP_OFFLINE && target <<
2765 return -EINVAL; <<
2766 #endif <<
2767 <<
2768 ret = lock_device_hotplug_sysfs(); <<
2769 if (ret) <<
2770 return ret; <<
2771 <<
2772 mutex_lock(&cpuhp_state_mutex); <<
2773 sp = cpuhp_get_step(target); <<
2774 ret = !sp->name || sp->cant_stop ? -E <<
2775 mutex_unlock(&cpuhp_state_mutex); <<
2776 if (ret) <<
2777 goto out; <<
2778 <<
2779 if (st->state < target) <<
2780 ret = cpu_up(dev->id, target) <<
2781 else if (st->state > target) <<
2782 ret = cpu_down(dev->id, targe <<
2783 else if (WARN_ON(st->target != target <<
2784 st->target = target; <<
2785 out: <<
2786 unlock_device_hotplug(); <<
2787 return ret ? ret : count; <<
2788 } <<
2789 <<
2790 static ssize_t target_show(struct device *dev <<
2791 struct device_attr <<
2792 { <<
2793 struct cpuhp_cpu_state *st = per_cpu_ <<
2794 <<
2795 return sprintf(buf, "%d\n", st->targe <<
2796 } <<
2797 static DEVICE_ATTR_RW(target); <<
2798 <<
2799 static ssize_t fail_store(struct device *dev, <<
2800 const char *buf, si <<
2801 { <<
2802 struct cpuhp_cpu_state *st = per_cpu_ <<
2803 struct cpuhp_step *sp; <<
2804 int fail, ret; <<
2805 <<
2806 ret = kstrtoint(buf, 10, &fail); <<
2807 if (ret) <<
2808 return ret; <<
2809 <<
2810 if (fail == CPUHP_INVALID) { <<
2811 st->fail = fail; <<
2812 return count; <<
2813 } <<
2814 <<
2815 if (fail < CPUHP_OFFLINE || fail > CP <<
2816 return -EINVAL; <<
2817 <<
2818 /* <<
2819 * Cannot fail STARTING/DYING callbac <<
2820 */ <<
2821 if (cpuhp_is_atomic_state(fail)) <<
2822 return -EINVAL; <<
2823 <<
2824 /* <<
2825 * DEAD callbacks cannot fail... <<
2826 * ... neither can CPUHP_BRINGUP_CPU <<
2827 * triggering STARTING callbacks, a f <<
2828 * hinder rollback. <<
2829 */ <<
2830 if (fail <= CPUHP_BRINGUP_CPU && st-> <<
2831 return -EINVAL; <<
2832 <<
2833 /* <<
2834 * Cannot fail anything that doesn't <<
2835 */ <<
2836 mutex_lock(&cpuhp_state_mutex); <<
2837 sp = cpuhp_get_step(fail); <<
2838 if (!sp->startup.single && !sp->teard <<
2839 ret = -EINVAL; <<
2840 mutex_unlock(&cpuhp_state_mutex); <<
2841 if (ret) <<
2842 return ret; <<
2843 <<
2844 st->fail = fail; <<
2845 <<
2846 return count; <<
2847 } <<
2848 <<
2849 static ssize_t fail_show(struct device *dev, <<
2850 struct device_attrib <<
2851 { <<
2852 struct cpuhp_cpu_state *st = per_cpu_ <<
2853 <<
2854 return sprintf(buf, "%d\n", st->fail) <<
2855 } <<
2856 <<
2857 static DEVICE_ATTR_RW(fail); <<
2858 <<
2859 static struct attribute *cpuhp_cpu_attrs[] = <<
2860 &dev_attr_state.attr, <<
2861 &dev_attr_target.attr, <<
2862 &dev_attr_fail.attr, <<
2863 NULL <<
2864 }; <<
2865 <<
2866 static const struct attribute_group cpuhp_cpu <<
2867 .attrs = cpuhp_cpu_attrs, <<
2868 .name = "hotplug", <<
2869 NULL <<
2870 }; <<
2871 <<
2872 static ssize_t states_show(struct device *dev <<
2873 struct devic <<
2874 { <<
2875 ssize_t cur, res = 0; <<
2876 int i; <<
2877 <<
2878 mutex_lock(&cpuhp_state_mutex); <<
2879 for (i = CPUHP_OFFLINE; i <= CPUHP_ON <<
2880 struct cpuhp_step *sp = cpuhp <<
2881 <<
2882 if (sp->name) { <<
2883 cur = sprintf(buf, "% <<
2884 buf += cur; <<
2885 res += cur; <<
2886 } <<
2887 } <<
2888 mutex_unlock(&cpuhp_state_mutex); <<
2889 return res; <<
2890 } <<
2891 static DEVICE_ATTR_RO(states); <<
2892 <<
2893 static struct attribute *cpuhp_cpu_root_attrs <<
2894 &dev_attr_states.attr, <<
2895 NULL <<
2896 }; <<
2897 <<
2898 static const struct attribute_group cpuhp_cpu <<
2899 .attrs = cpuhp_cpu_root_attrs, <<
2900 .name = "hotplug", <<
2901 NULL <<
2902 }; <<
2903 <<
2904 #ifdef CONFIG_HOTPLUG_SMT <<
2905 <<
2906 static bool cpu_smt_num_threads_valid(unsigne <<
2907 { <<
2908 if (IS_ENABLED(CONFIG_SMT_NUM_THREADS <<
2909 return threads >= 1 && thread <<
2910 return threads == 1 || threads == cpu <<
2911 } <<
2912 <<
2913 static ssize_t <<
2914 __store_smt_control(struct device *dev, struc <<
2915 const char *buf, size_t c <<
2916 { <<
2917 int ctrlval, ret, num_threads, orig_t <<
2918 bool force_off; <<
2919 <<
2920 if (cpu_smt_control == CPU_SMT_FORCE_ <<
2921 return -EPERM; <<
2922 <<
2923 if (cpu_smt_control == CPU_SMT_NOT_SU <<
2924 return -ENODEV; <<
2925 <<
2926 if (sysfs_streq(buf, "on")) { <<
2927 ctrlval = CPU_SMT_ENABLED; <<
2928 num_threads = cpu_smt_max_thr <<
2929 } else if (sysfs_streq(buf, "off")) { <<
2930 ctrlval = CPU_SMT_DISABLED; <<
2931 num_threads = 1; <<
2932 } else if (sysfs_streq(buf, "forceoff <<
2933 ctrlval = CPU_SMT_FORCE_DISAB <<
2934 num_threads = 1; <<
2935 } else if (kstrtoint(buf, 10, &num_th <<
2936 if (num_threads == 1) <<
2937 ctrlval = CPU_SMT_DIS <<
2938 else if (cpu_smt_num_threads_ <<
2939 ctrlval = CPU_SMT_ENA <<
2940 else <<
2941 return -EINVAL; <<
2942 } else { <<
2943 return -EINVAL; <<
2944 } <<
2945 <<
2946 ret = lock_device_hotplug_sysfs(); <<
2947 if (ret) <<
2948 return ret; <<
2949 <<
2950 orig_threads = cpu_smt_num_threads; <<
2951 cpu_smt_num_threads = num_threads; <<
2952 <<
2953 force_off = ctrlval != cpu_smt_contro <<
2954 <<
2955 if (num_threads > orig_threads) <<
2956 ret = cpuhp_smt_enable(); <<
2957 else if (num_threads < orig_threads | <<
2958 ret = cpuhp_smt_disable(ctrlv <<
2959 <<
2960 unlock_device_hotplug(); <<
2961 return ret ? ret : count; <<
2962 } <<
2963 <<
2964 #else /* !CONFIG_HOTPLUG_SMT */ <<
2965 static ssize_t <<
2966 __store_smt_control(struct device *dev, struc <<
2967 const char *buf, size_t c <<
2968 { <<
2969 return -ENODEV; <<
2970 } <<
2971 #endif /* CONFIG_HOTPLUG_SMT */ <<
2972 <<
2973 static const char *smt_states[] = { <<
2974 [CPU_SMT_ENABLED] = "on <<
2975 [CPU_SMT_DISABLED] = "of <<
2976 [CPU_SMT_FORCE_DISABLED] = "fo <<
2977 [CPU_SMT_NOT_SUPPORTED] = "no <<
2978 [CPU_SMT_NOT_IMPLEMENTED] = "no <<
2979 }; <<
2980 <<
2981 static ssize_t control_show(struct device *de <<
2982 struct device_att <<
2983 { <<
2984 const char *state = smt_states[cpu_sm <<
2985 <<
2986 #ifdef CONFIG_HOTPLUG_SMT <<
2987 /* <<
2988 * If SMT is enabled but not all thre <<
2989 * number of threads. If all threads <<
2990 * show the state name. <<
2991 */ <<
2992 if (cpu_smt_control == CPU_SMT_ENABLE <<
2993 cpu_smt_num_threads != cpu_smt_ma <<
2994 return sysfs_emit(buf, "%d\n" <<
2995 #endif <<
2996 <<
2997 return sysfs_emit(buf, "%s\n", state) <<
2998 } <<
2999 <<
3000 static ssize_t control_store(struct device *d <<
3001 const char *buf, <<
3002 { 621 {
3003 return __store_smt_control(dev, attr, !! 622 unsigned long val = CPU_STARTING;
3004 } <<
3005 static DEVICE_ATTR_RW(control); <<
3006 <<
3007 static ssize_t active_show(struct device *dev <<
3008 struct device_attr <<
3009 { <<
3010 return sysfs_emit(buf, "%d\n", sched_ <<
3011 } <<
3012 static DEVICE_ATTR_RO(active); <<
3013 <<
3014 static struct attribute *cpuhp_smt_attrs[] = <<
3015 &dev_attr_control.attr, <<
3016 &dev_attr_active.attr, <<
3017 NULL <<
3018 }; <<
3019 <<
3020 static const struct attribute_group cpuhp_smt <<
3021 .attrs = cpuhp_smt_attrs, <<
3022 .name = "smt", <<
3023 NULL <<
3024 }; <<
3025 <<
3026 static int __init cpu_smt_sysfs_init(void) <<
3027 { <<
3028 struct device *dev_root; <<
3029 int ret = -ENODEV; <<
3030 623
3031 dev_root = bus_get_dev_root(&cpu_subs !! 624 #ifdef CONFIG_PM_SLEEP_SMP
3032 if (dev_root) { !! 625 if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus))
3033 ret = sysfs_create_group(&dev !! 626 val = CPU_STARTING_FROZEN;
3034 put_device(dev_root); !! 627 #endif /* CONFIG_PM_SLEEP_SMP */
3035 } !! 628 cpu_notify(val, (void *)(long)cpu);
3036 return ret; <<
3037 } 629 }
3038 630
3039 static int __init cpuhp_sysfs_init(void) !! 631 #endif /* CONFIG_SMP */
3040 { <<
3041 struct device *dev_root; <<
3042 int cpu, ret; <<
3043 <<
3044 ret = cpu_smt_sysfs_init(); <<
3045 if (ret) <<
3046 return ret; <<
3047 <<
3048 dev_root = bus_get_dev_root(&cpu_subs <<
3049 if (dev_root) { <<
3050 ret = sysfs_create_group(&dev <<
3051 put_device(dev_root); <<
3052 if (ret) <<
3053 return ret; <<
3054 } <<
3055 <<
3056 for_each_possible_cpu(cpu) { <<
3057 struct device *dev = get_cpu_ <<
3058 <<
3059 if (!dev) <<
3060 continue; <<
3061 ret = sysfs_create_group(&dev <<
3062 if (ret) <<
3063 return ret; <<
3064 } <<
3065 return 0; <<
3066 } <<
3067 device_initcall(cpuhp_sysfs_init); <<
3068 #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU <<
3069 632
3070 /* 633 /*
3071 * cpu_bit_bitmap[] is a special, "compressed 634 * cpu_bit_bitmap[] is a special, "compressed" data structure that
3072 * represents all NR_CPUS bits binary values 635 * represents all NR_CPUS bits binary values of 1<<nr.
3073 * 636 *
3074 * It is used by cpumask_of() to get a consta 637 * It is used by cpumask_of() to get a constant address to a CPU
3075 * mask value that has a single bit set only. 638 * mask value that has a single bit set only.
3076 */ 639 */
3077 640
3078 /* cpu_bit_bitmap[0] is empty - so we can bac 641 /* cpu_bit_bitmap[0] is empty - so we can back into it */
3079 #define MASK_DECLARE_1(x) [x+1][0] = (1 642 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
3080 #define MASK_DECLARE_2(x) MASK_DECLARE_ 643 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
3081 #define MASK_DECLARE_4(x) MASK_DECLARE_ 644 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
3082 #define MASK_DECLARE_8(x) MASK_DECLARE_ 645 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
3083 646
3084 const unsigned long cpu_bit_bitmap[BITS_PER_L 647 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
3085 648
3086 MASK_DECLARE_8(0), MASK_DECLARE_ 649 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
3087 MASK_DECLARE_8(16), MASK_DECLARE_ 650 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
3088 #if BITS_PER_LONG > 32 651 #if BITS_PER_LONG > 32
3089 MASK_DECLARE_8(32), MASK_DECLARE_ 652 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
3090 MASK_DECLARE_8(48), MASK_DECLARE_ 653 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
3091 #endif 654 #endif
3092 }; 655 };
3093 EXPORT_SYMBOL_GPL(cpu_bit_bitmap); 656 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
3094 657
3095 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = 658 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
3096 EXPORT_SYMBOL(cpu_all_bits); 659 EXPORT_SYMBOL(cpu_all_bits);
3097 660
3098 #ifdef CONFIG_INIT_ALL_POSSIBLE 661 #ifdef CONFIG_INIT_ALL_POSSIBLE
3099 struct cpumask __cpu_possible_mask __ro_after !! 662 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly
3100 = {CPU_BITS_ALL}; !! 663 = CPU_BITS_ALL;
3101 #else 664 #else
3102 struct cpumask __cpu_possible_mask __ro_after !! 665 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly;
3103 #endif 666 #endif
3104 EXPORT_SYMBOL(__cpu_possible_mask); !! 667 const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits);
3105 !! 668 EXPORT_SYMBOL(cpu_possible_mask);
3106 struct cpumask __cpu_online_mask __read_mostl <<
3107 EXPORT_SYMBOL(__cpu_online_mask); <<
3108 669
3109 struct cpumask __cpu_enabled_mask __read_most !! 670 static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly;
3110 EXPORT_SYMBOL(__cpu_enabled_mask); !! 671 const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits);
>> 672 EXPORT_SYMBOL(cpu_online_mask);
3111 673
3112 struct cpumask __cpu_present_mask __read_most !! 674 static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly;
3113 EXPORT_SYMBOL(__cpu_present_mask); !! 675 const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits);
>> 676 EXPORT_SYMBOL(cpu_present_mask);
3114 677
3115 struct cpumask __cpu_active_mask __read_mostl !! 678 static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly;
3116 EXPORT_SYMBOL(__cpu_active_mask); !! 679 const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits);
>> 680 EXPORT_SYMBOL(cpu_active_mask);
3117 681
3118 struct cpumask __cpu_dying_mask __read_mostly !! 682 void set_cpu_possible(unsigned int cpu, bool possible)
3119 EXPORT_SYMBOL(__cpu_dying_mask); <<
3120 <<
3121 atomic_t __num_online_cpus __read_mostly; <<
3122 EXPORT_SYMBOL(__num_online_cpus); <<
3123 <<
3124 void init_cpu_present(const struct cpumask *s <<
3125 { <<
3126 cpumask_copy(&__cpu_present_mask, src <<
3127 } <<
3128 <<
3129 void init_cpu_possible(const struct cpumask * <<
3130 { 683 {
3131 cpumask_copy(&__cpu_possible_mask, sr !! 684 if (possible)
>> 685 cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
>> 686 else
>> 687 cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits));
3132 } 688 }
3133 689
3134 void init_cpu_online(const struct cpumask *sr !! 690 void set_cpu_present(unsigned int cpu, bool present)
3135 { 691 {
3136 cpumask_copy(&__cpu_online_mask, src) !! 692 if (present)
>> 693 cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits));
>> 694 else
>> 695 cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits));
3137 } 696 }
3138 697
3139 void set_cpu_online(unsigned int cpu, bool on 698 void set_cpu_online(unsigned int cpu, bool online)
3140 { 699 {
3141 /* <<
3142 * atomic_inc/dec() is required to ha <<
3143 * function by the reboot and kexec c <<
3144 * IPI/NMI broadcasts when shutting d <<
3145 * regular CPU hotplug is properly se <<
3146 * <<
3147 * Note, that the fact that __num_onl <<
3148 * does not protect readers which are <<
3149 * concurrent hotplug operations. <<
3150 */ <<
3151 if (online) { 700 if (online) {
3152 if (!cpumask_test_and_set_cpu !! 701 cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits));
3153 atomic_inc(&__num_onl !! 702 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
3154 } else { 703 } else {
3155 if (cpumask_test_and_clear_cp !! 704 cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits));
3156 atomic_dec(&__num_onl <<
3157 } 705 }
3158 } 706 }
3159 707
3160 /* !! 708 void set_cpu_active(unsigned int cpu, bool active)
3161 * Activate the first processor. <<
3162 */ <<
3163 void __init boot_cpu_init(void) <<
3164 { <<
3165 int cpu = smp_processor_id(); <<
3166 <<
3167 /* Mark the boot cpu "present", "onli <<
3168 set_cpu_online(cpu, true); <<
3169 set_cpu_active(cpu, true); <<
3170 set_cpu_present(cpu, true); <<
3171 set_cpu_possible(cpu, true); <<
3172 <<
3173 #ifdef CONFIG_SMP <<
3174 __boot_cpu_id = cpu; <<
3175 #endif <<
3176 } <<
3177 <<
3178 /* <<
3179 * Must be called _AFTER_ setting up the per_ <<
3180 */ <<
3181 void __init boot_cpu_hotplug_init(void) <<
3182 { 709 {
3183 #ifdef CONFIG_SMP !! 710 if (active)
3184 cpumask_set_cpu(smp_processor_id(), & !! 711 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
3185 atomic_set(this_cpu_ptr(&cpuhp_state. <<
3186 #endif <<
3187 this_cpu_write(cpuhp_state.state, CPU <<
3188 this_cpu_write(cpuhp_state.target, CP <<
3189 } <<
3190 <<
3191 #ifdef CONFIG_CPU_MITIGATIONS <<
3192 /* <<
3193 * These are used for a global "mitigations=" <<
3194 * optional CPU mitigations. <<
3195 */ <<
3196 enum cpu_mitigations { <<
3197 CPU_MITIGATIONS_OFF, <<
3198 CPU_MITIGATIONS_AUTO, <<
3199 CPU_MITIGATIONS_AUTO_NOSMT, <<
3200 }; <<
3201 <<
3202 static enum cpu_mitigations cpu_mitigations _ <<
3203 <<
3204 static int __init mitigations_parse_cmdline(c <<
3205 { <<
3206 if (!strcmp(arg, "off")) <<
3207 cpu_mitigations = CPU_MITIGAT <<
3208 else if (!strcmp(arg, "auto")) <<
3209 cpu_mitigations = CPU_MITIGAT <<
3210 else if (!strcmp(arg, "auto,nosmt")) <<
3211 cpu_mitigations = CPU_MITIGAT <<
3212 else 712 else
3213 pr_crit("Unsupported mitigati !! 713 cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits));
3214 arg); <<
3215 <<
3216 return 0; <<
3217 } 714 }
3218 715
3219 /* mitigations=off */ !! 716 void init_cpu_present(const struct cpumask *src)
3220 bool cpu_mitigations_off(void) <<
3221 { 717 {
3222 return cpu_mitigations == CPU_MITIGAT !! 718 cpumask_copy(to_cpumask(cpu_present_bits), src);
3223 } 719 }
3224 EXPORT_SYMBOL_GPL(cpu_mitigations_off); <<
3225 720
3226 /* mitigations=auto,nosmt */ !! 721 void init_cpu_possible(const struct cpumask *src)
3227 bool cpu_mitigations_auto_nosmt(void) <<
3228 { 722 {
3229 return cpu_mitigations == CPU_MITIGAT !! 723 cpumask_copy(to_cpumask(cpu_possible_bits), src);
3230 } 724 }
3231 EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt) !! 725
3232 #else !! 726 void init_cpu_online(const struct cpumask *src)
3233 static int __init mitigations_parse_cmdline(c <<
3234 { 727 {
3235 pr_crit("Kernel compiled without miti !! 728 cpumask_copy(to_cpumask(cpu_online_bits), src);
3236 return 0; <<
3237 } 729 }
3238 #endif <<
3239 early_param("mitigations", mitigations_parse_ <<
3240 730
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.