Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* CPU control. |
2 | * (C) 2001, 2002, 2003, 2004 Rusty Russell | |
3 | * | |
4 | * This code is licenced under the GPL. | |
5 | */ | |
6 | #include <linux/proc_fs.h> | |
7 | #include <linux/smp.h> | |
8 | #include <linux/init.h> | |
9 | #include <linux/notifier.h> | |
3f07c014 | 10 | #include <linux/sched/signal.h> |
ef8bd77f | 11 | #include <linux/sched/hotplug.h> |
29930025 | 12 | #include <linux/sched/task.h> |
1da177e4 LT |
13 | #include <linux/unistd.h> |
14 | #include <linux/cpu.h> | |
cb79295e AV |
15 | #include <linux/oom.h> |
16 | #include <linux/rcupdate.h> | |
9984de1a | 17 | #include <linux/export.h> |
e4cc2f87 | 18 | #include <linux/bug.h> |
1da177e4 LT |
19 | #include <linux/kthread.h> |
20 | #include <linux/stop_machine.h> | |
81615b62 | 21 | #include <linux/mutex.h> |
5a0e3ad6 | 22 | #include <linux/gfp.h> |
79cfbdfa | 23 | #include <linux/suspend.h> |
a19423b9 | 24 | #include <linux/lockdep.h> |
345527b1 | 25 | #include <linux/tick.h> |
a8994181 | 26 | #include <linux/irq.h> |
941154bd | 27 | #include <linux/nmi.h> |
4cb28ced | 28 | #include <linux/smpboot.h> |
e6d4989a | 29 | #include <linux/relay.h> |
6731d4f1 | 30 | #include <linux/slab.h> |
fc8dffd3 | 31 | #include <linux/percpu-rwsem.h> |
cff7d378 | 32 | |
bb3632c6 | 33 | #include <trace/events/power.h> |
cff7d378 TG |
34 | #define CREATE_TRACE_POINTS |
35 | #include <trace/events/cpuhp.h> | |
1da177e4 | 36 | |
38498a67 TG |
37 | #include "smpboot.h" |
38 | ||
cff7d378 TG |
39 | /** |
40 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
41 | * @state: The current cpu state | |
42 | * @target: The target state | |
4cb28ced TG |
43 | * @thread: Pointer to the hotplug thread |
44 | * @should_run: Thread should execute | |
3b9d6da6 | 45 | * @rollback: Perform a rollback |
a724632c TG |
46 | * @single: Single callback invocation |
47 | * @bringup: Single callback bringup or teardown selector | |
48 | * @cb_state: The state for a single callback (install/uninstall) | |
4cb28ced | 49 | * @result: Result of the operation |
5ebe7742 PZ |
50 | * @done_up: Signal completion to the issuer of the task for cpu-up |
51 | * @done_down: Signal completion to the issuer of the task for cpu-down | |
cff7d378 TG |
52 | */ |
53 | struct cpuhp_cpu_state { | |
54 | enum cpuhp_state state; | |
55 | enum cpuhp_state target; | |
1db49484 | 56 | enum cpuhp_state fail; |
4cb28ced TG |
57 | #ifdef CONFIG_SMP |
58 | struct task_struct *thread; | |
59 | bool should_run; | |
3b9d6da6 | 60 | bool rollback; |
a724632c TG |
61 | bool single; |
62 | bool bringup; | |
0cc3cd21 | 63 | bool booted_once; |
cf392d10 | 64 | struct hlist_node *node; |
4dddfb5f | 65 | struct hlist_node *last; |
4cb28ced | 66 | enum cpuhp_state cb_state; |
4cb28ced | 67 | int result; |
5ebe7742 PZ |
68 | struct completion done_up; |
69 | struct completion done_down; | |
4cb28ced | 70 | #endif |
cff7d378 TG |
71 | }; |
72 | ||
1db49484 PZ |
73 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = { |
74 | .fail = CPUHP_INVALID, | |
75 | }; | |
cff7d378 | 76 | |
49dfe2a6 | 77 | #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP) |
5f4b55e1 PZ |
78 | static struct lockdep_map cpuhp_state_up_map = |
79 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map); | |
80 | static struct lockdep_map cpuhp_state_down_map = | |
81 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map); | |
82 | ||
83 | ||
76dc6c09 | 84 | static inline void cpuhp_lock_acquire(bool bringup) |
5f4b55e1 PZ |
85 | { |
86 | lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
87 | } | |
88 | ||
76dc6c09 | 89 | static inline void cpuhp_lock_release(bool bringup) |
5f4b55e1 PZ |
90 | { |
91 | lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
92 | } | |
93 | #else | |
94 | ||
76dc6c09 MM |
95 | static inline void cpuhp_lock_acquire(bool bringup) { } |
96 | static inline void cpuhp_lock_release(bool bringup) { } | |
5f4b55e1 | 97 | |
49dfe2a6 TG |
98 | #endif |
99 | ||
cff7d378 TG |
100 | /** |
101 | * cpuhp_step - Hotplug state machine step | |
102 | * @name: Name of the step | |
103 | * @startup: Startup function of the step | |
104 | * @teardown: Teardown function of the step | |
757c989b | 105 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
106 | */ |
107 | struct cpuhp_step { | |
cf392d10 TG |
108 | const char *name; |
109 | union { | |
3c1627e9 TG |
110 | int (*single)(unsigned int cpu); |
111 | int (*multi)(unsigned int cpu, | |
112 | struct hlist_node *node); | |
113 | } startup; | |
cf392d10 | 114 | union { |
3c1627e9 TG |
115 | int (*single)(unsigned int cpu); |
116 | int (*multi)(unsigned int cpu, | |
117 | struct hlist_node *node); | |
118 | } teardown; | |
cf392d10 | 119 | struct hlist_head list; |
cf392d10 TG |
120 | bool cant_stop; |
121 | bool multi_instance; | |
cff7d378 TG |
122 | }; |
123 | ||
98f8cdce | 124 | static DEFINE_MUTEX(cpuhp_state_mutex); |
17a2f1ce | 125 | static struct cpuhp_step cpuhp_hp_states[]; |
cff7d378 | 126 | |
a724632c TG |
127 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) |
128 | { | |
17a2f1ce | 129 | return cpuhp_hp_states + state; |
a724632c TG |
130 | } |
131 | ||
cff7d378 TG |
132 | /** |
133 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
134 | * @cpu: The cpu for which the callback should be invoked | |
96abb968 | 135 | * @state: The state to do callbacks for |
a724632c | 136 | * @bringup: True if the bringup callback should be invoked |
96abb968 PZ |
137 | * @node: For multi-instance, do a single entry callback for install/remove |
138 | * @lastp: For multi-instance rollback, remember how far we got | |
cff7d378 | 139 | * |
cf392d10 | 140 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 141 | */ |
a724632c | 142 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
96abb968 PZ |
143 | bool bringup, struct hlist_node *node, |
144 | struct hlist_node **lastp) | |
cff7d378 TG |
145 | { |
146 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 147 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
148 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
149 | int (*cb)(unsigned int cpu); | |
150 | int ret, cnt; | |
151 | ||
1db49484 PZ |
152 | if (st->fail == state) { |
153 | st->fail = CPUHP_INVALID; | |
154 | ||
155 | if (!(bringup ? step->startup.single : step->teardown.single)) | |
156 | return 0; | |
157 | ||
158 | return -EAGAIN; | |
159 | } | |
160 | ||
cf392d10 | 161 | if (!step->multi_instance) { |
96abb968 | 162 | WARN_ON_ONCE(lastp && *lastp); |
3c1627e9 | 163 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
164 | if (!cb) |
165 | return 0; | |
a724632c | 166 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 167 | ret = cb(cpu); |
a724632c | 168 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
169 | return ret; |
170 | } | |
3c1627e9 | 171 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
172 | if (!cbm) |
173 | return 0; | |
174 | ||
175 | /* Single invocation for instance add/remove */ | |
176 | if (node) { | |
96abb968 | 177 | WARN_ON_ONCE(lastp && *lastp); |
cf392d10 TG |
178 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
179 | ret = cbm(cpu, node); | |
180 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
181 | return ret; | |
182 | } | |
183 | ||
184 | /* State transition. Invoke on all instances */ | |
185 | cnt = 0; | |
186 | hlist_for_each(node, &step->list) { | |
96abb968 PZ |
187 | if (lastp && node == *lastp) |
188 | break; | |
189 | ||
cf392d10 TG |
190 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
191 | ret = cbm(cpu, node); | |
192 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
96abb968 PZ |
193 | if (ret) { |
194 | if (!lastp) | |
195 | goto err; | |
196 | ||
197 | *lastp = node; | |
198 | return ret; | |
199 | } | |
cf392d10 TG |
200 | cnt++; |
201 | } | |
96abb968 PZ |
202 | if (lastp) |
203 | *lastp = NULL; | |
cf392d10 TG |
204 | return 0; |
205 | err: | |
206 | /* Rollback the instances if one failed */ | |
3c1627e9 | 207 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
208 | if (!cbm) |
209 | return ret; | |
210 | ||
211 | hlist_for_each(node, &step->list) { | |
212 | if (!cnt--) | |
213 | break; | |
724a8688 PZ |
214 | |
215 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
216 | ret = cbm(cpu, node); | |
217 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
218 | /* | |
219 | * Rollback must not fail, | |
220 | */ | |
221 | WARN_ON_ONCE(ret); | |
cff7d378 TG |
222 | } |
223 | return ret; | |
224 | } | |
225 | ||
98a79d6a | 226 | #ifdef CONFIG_SMP |
fcb3029a AB |
227 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
228 | { | |
229 | /* | |
230 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
231 | * purposes as that state is handled explicitly in cpu_down. | |
232 | */ | |
233 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
234 | } | |
235 | ||
5ebe7742 PZ |
236 | static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup) |
237 | { | |
238 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
239 | wait_for_completion(done); | |
240 | } | |
241 | ||
242 | static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup) | |
243 | { | |
244 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
245 | complete(done); | |
246 | } | |
247 | ||
248 | /* | |
249 | * The former STARTING/DYING states, ran with IRQs disabled and must not fail. | |
250 | */ | |
251 | static bool cpuhp_is_atomic_state(enum cpuhp_state state) | |
252 | { | |
253 | return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE; | |
254 | } | |
255 | ||
b3199c02 | 256 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 257 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
258 | bool cpuhp_tasks_frozen; |
259 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 260 | |
79a6cdeb | 261 | /* |
93ae4f97 SB |
262 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
263 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
79a6cdeb LJ |
264 | */ |
265 | void cpu_maps_update_begin(void) | |
266 | { | |
267 | mutex_lock(&cpu_add_remove_lock); | |
268 | } | |
269 | ||
270 | void cpu_maps_update_done(void) | |
271 | { | |
272 | mutex_unlock(&cpu_add_remove_lock); | |
273 | } | |
1da177e4 | 274 | |
fc8dffd3 TG |
275 | /* |
276 | * If set, cpu_up and cpu_down will return -EBUSY and do nothing. | |
e3920fb4 RW |
277 | * Should always be manipulated under cpu_add_remove_lock |
278 | */ | |
279 | static int cpu_hotplug_disabled; | |
280 | ||
79a6cdeb LJ |
281 | #ifdef CONFIG_HOTPLUG_CPU |
282 | ||
fc8dffd3 | 283 | DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); |
a19423b9 | 284 | |
8f553c49 | 285 | void cpus_read_lock(void) |
a9d9baa1 | 286 | { |
fc8dffd3 | 287 | percpu_down_read(&cpu_hotplug_lock); |
a9d9baa1 | 288 | } |
8f553c49 | 289 | EXPORT_SYMBOL_GPL(cpus_read_lock); |
90d45d17 | 290 | |
6f4ceee9 WL |
291 | int cpus_read_trylock(void) |
292 | { | |
293 | return percpu_down_read_trylock(&cpu_hotplug_lock); | |
294 | } | |
295 | EXPORT_SYMBOL_GPL(cpus_read_trylock); | |
296 | ||
8f553c49 | 297 | void cpus_read_unlock(void) |
a9d9baa1 | 298 | { |
fc8dffd3 | 299 | percpu_up_read(&cpu_hotplug_lock); |
a9d9baa1 | 300 | } |
8f553c49 | 301 | EXPORT_SYMBOL_GPL(cpus_read_unlock); |
a9d9baa1 | 302 | |
8f553c49 | 303 | void cpus_write_lock(void) |
d221938c | 304 | { |
fc8dffd3 | 305 | percpu_down_write(&cpu_hotplug_lock); |
d221938c | 306 | } |
87af9e7f | 307 | |
8f553c49 | 308 | void cpus_write_unlock(void) |
d221938c | 309 | { |
fc8dffd3 | 310 | percpu_up_write(&cpu_hotplug_lock); |
d221938c GS |
311 | } |
312 | ||
fc8dffd3 | 313 | void lockdep_assert_cpus_held(void) |
d221938c | 314 | { |
fc8dffd3 | 315 | percpu_rwsem_assert_held(&cpu_hotplug_lock); |
d221938c | 316 | } |
79a6cdeb | 317 | |
16e53dbf SB |
318 | /* |
319 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
320 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
321 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
322 | * hotplug path before performing hotplug operations. So acquiring that lock | |
323 | * guarantees mutual exclusion from any currently running hotplug operations. | |
324 | */ | |
325 | void cpu_hotplug_disable(void) | |
326 | { | |
327 | cpu_maps_update_begin(); | |
89af7ba5 | 328 | cpu_hotplug_disabled++; |
16e53dbf SB |
329 | cpu_maps_update_done(); |
330 | } | |
32145c46 | 331 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 332 | |
01b41159 LW |
333 | static void __cpu_hotplug_enable(void) |
334 | { | |
335 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
336 | return; | |
337 | cpu_hotplug_disabled--; | |
338 | } | |
339 | ||
16e53dbf SB |
340 | void cpu_hotplug_enable(void) |
341 | { | |
342 | cpu_maps_update_begin(); | |
01b41159 | 343 | __cpu_hotplug_enable(); |
16e53dbf SB |
344 | cpu_maps_update_done(); |
345 | } | |
32145c46 | 346 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 347 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 348 | |
0cc3cd21 TG |
349 | #ifdef CONFIG_HOTPLUG_SMT |
350 | enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED; | |
26acfb66 | 351 | EXPORT_SYMBOL_GPL(cpu_smt_control); |
0cc3cd21 | 352 | |
bc2d8d26 TG |
353 | static bool cpu_smt_available __read_mostly; |
354 | ||
8e1b706b | 355 | void __init cpu_smt_disable(bool force) |
0cc3cd21 | 356 | { |
8e1b706b JK |
357 | if (cpu_smt_control == CPU_SMT_FORCE_DISABLED || |
358 | cpu_smt_control == CPU_SMT_NOT_SUPPORTED) | |
359 | return; | |
360 | ||
361 | if (force) { | |
0cc3cd21 TG |
362 | pr_info("SMT: Force disabled\n"); |
363 | cpu_smt_control = CPU_SMT_FORCE_DISABLED; | |
8e1b706b JK |
364 | } else { |
365 | cpu_smt_control = CPU_SMT_DISABLED; | |
0cc3cd21 | 366 | } |
8e1b706b JK |
367 | } |
368 | ||
fee0aede TG |
369 | /* |
370 | * The decision whether SMT is supported can only be done after the full | |
bc2d8d26 TG |
371 | * CPU identification. Called from architecture code before non boot CPUs |
372 | * are brought up. | |
fee0aede | 373 | */ |
bc2d8d26 | 374 | void __init cpu_smt_check_topology_early(void) |
fee0aede TG |
375 | { |
376 | if (!topology_smt_supported()) | |
377 | cpu_smt_control = CPU_SMT_NOT_SUPPORTED; | |
378 | } | |
379 | ||
bc2d8d26 TG |
380 | /* |
381 | * If SMT was disabled by BIOS, detect it here, after the CPUs have been | |
382 | * brought online. This ensures the smt/l1tf sysfs entries are consistent | |
383 | * with reality. cpu_smt_available is set to true during the bringup of non | |
384 | * boot CPUs when a SMT sibling is detected. Note, this may overwrite | |
385 | * cpu_smt_control's previous setting. | |
386 | */ | |
387 | void __init cpu_smt_check_topology(void) | |
388 | { | |
389 | if (!cpu_smt_available) | |
390 | cpu_smt_control = CPU_SMT_NOT_SUPPORTED; | |
391 | } | |
392 | ||
8e1b706b JK |
393 | static int __init smt_cmdline_disable(char *str) |
394 | { | |
395 | cpu_smt_disable(str && !strcmp(str, "force")); | |
0cc3cd21 TG |
396 | return 0; |
397 | } | |
398 | early_param("nosmt", smt_cmdline_disable); | |
399 | ||
400 | static inline bool cpu_smt_allowed(unsigned int cpu) | |
401 | { | |
bc2d8d26 | 402 | if (topology_is_primary_thread(cpu)) |
0cc3cd21 TG |
403 | return true; |
404 | ||
bc2d8d26 TG |
405 | /* |
406 | * If the CPU is not a 'primary' thread and the booted_once bit is | |
407 | * set then the processor has SMT support. Store this information | |
408 | * for the late check of SMT support in cpu_smt_check_topology(). | |
409 | */ | |
410 | if (per_cpu(cpuhp_state, cpu).booted_once) | |
411 | cpu_smt_available = true; | |
412 | ||
413 | if (cpu_smt_control == CPU_SMT_ENABLED) | |
0cc3cd21 TG |
414 | return true; |
415 | ||
416 | /* | |
417 | * On x86 it's required to boot all logical CPUs at least once so | |
418 | * that the init code can get a chance to set CR4.MCE on each | |
419 | * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any | |
420 | * core will shutdown the machine. | |
421 | */ | |
422 | return !per_cpu(cpuhp_state, cpu).booted_once; | |
423 | } | |
424 | #else | |
425 | static inline bool cpu_smt_allowed(unsigned int cpu) { return true; } | |
426 | #endif | |
427 | ||
4dddfb5f PZ |
428 | static inline enum cpuhp_state |
429 | cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
430 | { | |
431 | enum cpuhp_state prev_state = st->state; | |
432 | ||
433 | st->rollback = false; | |
434 | st->last = NULL; | |
435 | ||
436 | st->target = target; | |
437 | st->single = false; | |
438 | st->bringup = st->state < target; | |
439 | ||
440 | return prev_state; | |
441 | } | |
442 | ||
443 | static inline void | |
444 | cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state) | |
445 | { | |
446 | st->rollback = true; | |
447 | ||
448 | /* | |
449 | * If we have st->last we need to undo partial multi_instance of this | |
450 | * state first. Otherwise start undo at the previous state. | |
451 | */ | |
452 | if (!st->last) { | |
453 | if (st->bringup) | |
454 | st->state--; | |
455 | else | |
456 | st->state++; | |
457 | } | |
458 | ||
459 | st->target = prev_state; | |
460 | st->bringup = !st->bringup; | |
461 | } | |
462 | ||
463 | /* Regular hotplug invocation of the AP hotplug thread */ | |
464 | static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st) | |
465 | { | |
466 | if (!st->single && st->state == st->target) | |
467 | return; | |
468 | ||
469 | st->result = 0; | |
470 | /* | |
471 | * Make sure the above stores are visible before should_run becomes | |
472 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
473 | */ | |
474 | smp_mb(); | |
475 | st->should_run = true; | |
476 | wake_up_process(st->thread); | |
5ebe7742 | 477 | wait_for_ap_thread(st, st->bringup); |
4dddfb5f PZ |
478 | } |
479 | ||
480 | static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
481 | { | |
482 | enum cpuhp_state prev_state; | |
483 | int ret; | |
484 | ||
485 | prev_state = cpuhp_set_state(st, target); | |
486 | __cpuhp_kick_ap(st); | |
487 | if ((ret = st->result)) { | |
488 | cpuhp_reset_state(st, prev_state); | |
489 | __cpuhp_kick_ap(st); | |
490 | } | |
491 | ||
492 | return ret; | |
493 | } | |
9cd4f1a4 | 494 | |
8df3e07e TG |
495 | static int bringup_wait_for_ap(unsigned int cpu) |
496 | { | |
497 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
498 | ||
9cd4f1a4 | 499 | /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */ |
5ebe7742 | 500 | wait_for_ap_thread(st, true); |
dea1d0f5 TG |
501 | if (WARN_ON_ONCE((!cpu_online(cpu)))) |
502 | return -ECANCELED; | |
9cd4f1a4 TG |
503 | |
504 | /* Unpark the stopper thread and the hotplug thread of the target cpu */ | |
505 | stop_machine_unpark(cpu); | |
506 | kthread_unpark(st->thread); | |
507 | ||
0cc3cd21 TG |
508 | /* |
509 | * SMT soft disabling on X86 requires to bring the CPU out of the | |
510 | * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The | |
511 | * CPU marked itself as booted_once in cpu_notify_starting() so the | |
512 | * cpu_smt_allowed() check will now return false if this is not the | |
513 | * primary sibling. | |
514 | */ | |
515 | if (!cpu_smt_allowed(cpu)) | |
516 | return -ECANCELED; | |
517 | ||
4dddfb5f PZ |
518 | if (st->target <= CPUHP_AP_ONLINE_IDLE) |
519 | return 0; | |
520 | ||
521 | return cpuhp_kick_ap(st, st->target); | |
8df3e07e TG |
522 | } |
523 | ||
ba997462 TG |
524 | static int bringup_cpu(unsigned int cpu) |
525 | { | |
526 | struct task_struct *idle = idle_thread_get(cpu); | |
527 | int ret; | |
528 | ||
aa877175 BO |
529 | /* |
530 | * Some architectures have to walk the irq descriptors to | |
531 | * setup the vector space for the cpu which comes online. | |
532 | * Prevent irq alloc/free across the bringup. | |
533 | */ | |
534 | irq_lock_sparse(); | |
535 | ||
ba997462 TG |
536 | /* Arch-specific enabling code. */ |
537 | ret = __cpu_up(cpu, idle); | |
aa877175 | 538 | irq_unlock_sparse(); |
530e9b76 | 539 | if (ret) |
ba997462 | 540 | return ret; |
9cd4f1a4 | 541 | return bringup_wait_for_ap(cpu); |
ba997462 TG |
542 | } |
543 | ||
2e1a3483 TG |
544 | /* |
545 | * Hotplug state machine related functions | |
546 | */ | |
2e1a3483 | 547 | |
a724632c | 548 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 | 549 | { |
6fb86d97 MO |
550 | for (st->state--; st->state > st->target; st->state--) |
551 | cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
2e1a3483 TG |
552 | } |
553 | ||
554 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 555 | enum cpuhp_state target) |
2e1a3483 TG |
556 | { |
557 | enum cpuhp_state prev_state = st->state; | |
558 | int ret = 0; | |
559 | ||
560 | while (st->state < target) { | |
2e1a3483 | 561 | st->state++; |
96abb968 | 562 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); |
2e1a3483 TG |
563 | if (ret) { |
564 | st->target = prev_state; | |
a724632c | 565 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
566 | break; |
567 | } | |
568 | } | |
569 | return ret; | |
570 | } | |
571 | ||
4cb28ced TG |
572 | /* |
573 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
574 | */ | |
575 | static void cpuhp_create(unsigned int cpu) | |
576 | { | |
577 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
578 | ||
5ebe7742 PZ |
579 | init_completion(&st->done_up); |
580 | init_completion(&st->done_down); | |
4cb28ced TG |
581 | } |
582 | ||
583 | static int cpuhp_should_run(unsigned int cpu) | |
584 | { | |
585 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
586 | ||
587 | return st->should_run; | |
588 | } | |
589 | ||
4cb28ced TG |
590 | /* |
591 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
592 | * callbacks when a state gets [un]installed at runtime. | |
4dddfb5f PZ |
593 | * |
594 | * Each invocation of this function by the smpboot thread does a single AP | |
595 | * state callback. | |
596 | * | |
597 | * It has 3 modes of operation: | |
598 | * - single: runs st->cb_state | |
599 | * - up: runs ++st->state, while st->state < st->target | |
600 | * - down: runs st->state--, while st->state > st->target | |
601 | * | |
602 | * When complete or on error, should_run is cleared and the completion is fired. | |
4cb28ced TG |
603 | */ |
604 | static void cpuhp_thread_fun(unsigned int cpu) | |
605 | { | |
606 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
4dddfb5f PZ |
607 | bool bringup = st->bringup; |
608 | enum cpuhp_state state; | |
4cb28ced | 609 | |
f8b7530a NU |
610 | if (WARN_ON_ONCE(!st->should_run)) |
611 | return; | |
612 | ||
4cb28ced | 613 | /* |
4dddfb5f PZ |
614 | * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures |
615 | * that if we see ->should_run we also see the rest of the state. | |
4cb28ced TG |
616 | */ |
617 | smp_mb(); | |
4cb28ced | 618 | |
5f4b55e1 | 619 | cpuhp_lock_acquire(bringup); |
4dddfb5f | 620 | |
a724632c | 621 | if (st->single) { |
4dddfb5f PZ |
622 | state = st->cb_state; |
623 | st->should_run = false; | |
624 | } else { | |
625 | if (bringup) { | |
626 | st->state++; | |
627 | state = st->state; | |
628 | st->should_run = (st->state < st->target); | |
629 | WARN_ON_ONCE(st->state > st->target); | |
4cb28ced | 630 | } else { |
4dddfb5f PZ |
631 | state = st->state; |
632 | st->state--; | |
633 | st->should_run = (st->state > st->target); | |
634 | WARN_ON_ONCE(st->state < st->target); | |
4cb28ced | 635 | } |
4dddfb5f PZ |
636 | } |
637 | ||
638 | WARN_ON_ONCE(!cpuhp_is_ap_state(state)); | |
639 | ||
4dddfb5f PZ |
640 | if (cpuhp_is_atomic_state(state)) { |
641 | local_irq_disable(); | |
642 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); | |
643 | local_irq_enable(); | |
3b9d6da6 | 644 | |
4dddfb5f PZ |
645 | /* |
646 | * STARTING/DYING must not fail! | |
647 | */ | |
648 | WARN_ON_ONCE(st->result); | |
4cb28ced | 649 | } else { |
4dddfb5f PZ |
650 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); |
651 | } | |
652 | ||
653 | if (st->result) { | |
654 | /* | |
655 | * If we fail on a rollback, we're up a creek without no | |
656 | * paddle, no way forward, no way back. We loose, thanks for | |
657 | * playing. | |
658 | */ | |
659 | WARN_ON_ONCE(st->rollback); | |
660 | st->should_run = false; | |
4cb28ced | 661 | } |
4dddfb5f | 662 | |
5f4b55e1 | 663 | cpuhp_lock_release(bringup); |
4dddfb5f PZ |
664 | |
665 | if (!st->should_run) | |
5ebe7742 | 666 | complete_ap_thread(st, bringup); |
4cb28ced TG |
667 | } |
668 | ||
669 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 670 | static int |
cf392d10 TG |
671 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
672 | struct hlist_node *node) | |
4cb28ced TG |
673 | { |
674 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f | 675 | int ret; |
4cb28ced TG |
676 | |
677 | if (!cpu_online(cpu)) | |
678 | return 0; | |
679 | ||
5f4b55e1 PZ |
680 | cpuhp_lock_acquire(false); |
681 | cpuhp_lock_release(false); | |
682 | ||
683 | cpuhp_lock_acquire(true); | |
684 | cpuhp_lock_release(true); | |
49dfe2a6 | 685 | |
6a4e2451 TG |
686 | /* |
687 | * If we are up and running, use the hotplug thread. For early calls | |
688 | * we invoke the thread function directly. | |
689 | */ | |
690 | if (!st->thread) | |
96abb968 | 691 | return cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
6a4e2451 | 692 | |
4dddfb5f PZ |
693 | st->rollback = false; |
694 | st->last = NULL; | |
695 | ||
696 | st->node = node; | |
697 | st->bringup = bringup; | |
4cb28ced | 698 | st->cb_state = state; |
a724632c | 699 | st->single = true; |
a724632c | 700 | |
4dddfb5f | 701 | __cpuhp_kick_ap(st); |
4cb28ced | 702 | |
4cb28ced | 703 | /* |
4dddfb5f | 704 | * If we failed and did a partial, do a rollback. |
4cb28ced | 705 | */ |
4dddfb5f PZ |
706 | if ((ret = st->result) && st->last) { |
707 | st->rollback = true; | |
708 | st->bringup = !bringup; | |
709 | ||
710 | __cpuhp_kick_ap(st); | |
711 | } | |
712 | ||
1f7c70d6 TG |
713 | /* |
714 | * Clean up the leftovers so the next hotplug operation wont use stale | |
715 | * data. | |
716 | */ | |
717 | st->node = st->last = NULL; | |
4dddfb5f | 718 | return ret; |
1cf4f629 TG |
719 | } |
720 | ||
721 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
722 | { | |
723 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f PZ |
724 | enum cpuhp_state prev_state = st->state; |
725 | int ret; | |
1cf4f629 | 726 | |
5f4b55e1 PZ |
727 | cpuhp_lock_acquire(false); |
728 | cpuhp_lock_release(false); | |
729 | ||
730 | cpuhp_lock_acquire(true); | |
731 | cpuhp_lock_release(true); | |
4dddfb5f PZ |
732 | |
733 | trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work); | |
734 | ret = cpuhp_kick_ap(st, st->target); | |
735 | trace_cpuhp_exit(cpu, st->state, prev_state, ret); | |
736 | ||
737 | return ret; | |
4cb28ced TG |
738 | } |
739 | ||
740 | static struct smp_hotplug_thread cpuhp_threads = { | |
741 | .store = &cpuhp_state.thread, | |
742 | .create = &cpuhp_create, | |
743 | .thread_should_run = cpuhp_should_run, | |
744 | .thread_fn = cpuhp_thread_fun, | |
745 | .thread_comm = "cpuhp/%u", | |
746 | .selfparking = true, | |
747 | }; | |
748 | ||
749 | void __init cpuhp_threads_init(void) | |
750 | { | |
751 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
752 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
753 | } | |
754 | ||
777c6e0d | 755 | #ifdef CONFIG_HOTPLUG_CPU |
e4cc2f87 AV |
756 | /** |
757 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
758 | * @cpu: a CPU id | |
759 | * | |
760 | * This function walks all processes, finds a valid mm struct for each one and | |
761 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
762 | * trivial, there are various non-obvious corner cases, which this function | |
763 | * tries to solve in a safe manner. | |
764 | * | |
765 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
766 | * be called only for an already offlined CPU. | |
767 | */ | |
cb79295e AV |
768 | void clear_tasks_mm_cpumask(int cpu) |
769 | { | |
770 | struct task_struct *p; | |
771 | ||
772 | /* | |
773 | * This function is called after the cpu is taken down and marked | |
774 | * offline, so its not like new tasks will ever get this cpu set in | |
775 | * their mm mask. -- Peter Zijlstra | |
776 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
777 | * full-fledged tasklist_lock. | |
778 | */ | |
e4cc2f87 | 779 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
780 | rcu_read_lock(); |
781 | for_each_process(p) { | |
782 | struct task_struct *t; | |
783 | ||
e4cc2f87 AV |
784 | /* |
785 | * Main thread might exit, but other threads may still have | |
786 | * a valid mm. Find one. | |
787 | */ | |
cb79295e AV |
788 | t = find_lock_task_mm(p); |
789 | if (!t) | |
790 | continue; | |
791 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
792 | task_unlock(t); | |
793 | } | |
794 | rcu_read_unlock(); | |
795 | } | |
796 | ||
1da177e4 | 797 | /* Take this CPU down. */ |
71cf5aee | 798 | static int take_cpu_down(void *_param) |
1da177e4 | 799 | { |
4baa0afc TG |
800 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
801 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 802 | int err, cpu = smp_processor_id(); |
724a8688 | 803 | int ret; |
1da177e4 | 804 | |
1da177e4 LT |
805 | /* Ensure this CPU doesn't handle any more interrupts. */ |
806 | err = __cpu_disable(); | |
807 | if (err < 0) | |
f3705136 | 808 | return err; |
1da177e4 | 809 | |
a724632c TG |
810 | /* |
811 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
812 | * do this step again. | |
813 | */ | |
814 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
815 | st->state--; | |
4baa0afc | 816 | /* Invoke the former CPU_DYING callbacks */ |
724a8688 PZ |
817 | for (; st->state > target; st->state--) { |
818 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
819 | /* | |
820 | * DYING must not fail! | |
821 | */ | |
822 | WARN_ON_ONCE(ret); | |
823 | } | |
4baa0afc | 824 | |
52c063d1 TG |
825 | /* Give up timekeeping duties */ |
826 | tick_handover_do_timer(); | |
14e568e7 | 827 | /* Park the stopper thread */ |
090e77c3 | 828 | stop_machine_park(cpu); |
f3705136 | 829 | return 0; |
1da177e4 LT |
830 | } |
831 | ||
98458172 | 832 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 833 | { |
e69aab13 | 834 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 835 | int err; |
1da177e4 | 836 | |
2a58c527 | 837 | /* Park the smpboot threads */ |
1cf4f629 TG |
838 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
839 | ||
6acce3ef | 840 | /* |
a8994181 TG |
841 | * Prevent irq alloc/free while the dying cpu reorganizes the |
842 | * interrupt affinities. | |
6acce3ef | 843 | */ |
a8994181 | 844 | irq_lock_sparse(); |
6acce3ef | 845 | |
a8994181 TG |
846 | /* |
847 | * So now all preempt/rcu users must observe !cpu_active(). | |
848 | */ | |
210e2133 | 849 | err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 850 | if (err) { |
3b9d6da6 | 851 | /* CPU refused to die */ |
a8994181 | 852 | irq_unlock_sparse(); |
3b9d6da6 SAS |
853 | /* Unpark the hotplug thread so we can rollback there */ |
854 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 855 | return err; |
8fa1d7d3 | 856 | } |
04321587 | 857 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 858 | |
48c5ccae | 859 | /* |
5b1ead68 BJ |
860 | * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed |
861 | * all runnable tasks from the CPU, there's only the idle task left now | |
48c5ccae | 862 | * that the migration thread is done doing the stop_machine thing. |
51a96c77 PZ |
863 | * |
864 | * Wait for the stop thread to go away. | |
48c5ccae | 865 | */ |
5ebe7742 | 866 | wait_for_ap_thread(st, false); |
e69aab13 | 867 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); |
1da177e4 | 868 | |
a8994181 TG |
869 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
870 | irq_unlock_sparse(); | |
871 | ||
345527b1 | 872 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
873 | /* This actually kills the CPU. */ |
874 | __cpu_die(cpu); | |
875 | ||
a49b116d | 876 | tick_cleanup_dead_cpu(cpu); |
a58163d8 | 877 | rcutree_migrate_callbacks(cpu); |
98458172 TG |
878 | return 0; |
879 | } | |
1da177e4 | 880 | |
71f87b2f TG |
881 | static void cpuhp_complete_idle_dead(void *arg) |
882 | { | |
883 | struct cpuhp_cpu_state *st = arg; | |
884 | ||
5ebe7742 | 885 | complete_ap_thread(st, false); |
71f87b2f TG |
886 | } |
887 | ||
e69aab13 TG |
888 | void cpuhp_report_idle_dead(void) |
889 | { | |
890 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
891 | ||
892 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 893 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
894 | st->state = CPUHP_AP_IDLE_DEAD; |
895 | /* | |
896 | * We cannot call complete after rcu_report_dead() so we delegate it | |
897 | * to an online cpu. | |
898 | */ | |
899 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
900 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
901 | } |
902 | ||
4dddfb5f PZ |
903 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
904 | { | |
6fb86d97 MO |
905 | for (st->state++; st->state < st->target; st->state++) |
906 | cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); | |
4dddfb5f PZ |
907 | } |
908 | ||
909 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
910 | enum cpuhp_state target) | |
911 | { | |
912 | enum cpuhp_state prev_state = st->state; | |
913 | int ret = 0; | |
914 | ||
915 | for (; st->state > target; st->state--) { | |
916 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
917 | if (ret) { | |
918 | st->target = prev_state; | |
69fa6eb7 TG |
919 | if (st->state < prev_state) |
920 | undo_cpu_down(cpu, st); | |
4dddfb5f PZ |
921 | break; |
922 | } | |
923 | } | |
924 | return ret; | |
925 | } | |
cff7d378 | 926 | |
98458172 | 927 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
928 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
929 | enum cpuhp_state target) | |
98458172 | 930 | { |
cff7d378 TG |
931 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
932 | int prev_state, ret = 0; | |
98458172 TG |
933 | |
934 | if (num_online_cpus() == 1) | |
935 | return -EBUSY; | |
936 | ||
757c989b | 937 | if (!cpu_present(cpu)) |
98458172 TG |
938 | return -EINVAL; |
939 | ||
8f553c49 | 940 | cpus_write_lock(); |
98458172 TG |
941 | |
942 | cpuhp_tasks_frozen = tasks_frozen; | |
943 | ||
4dddfb5f | 944 | prev_state = cpuhp_set_state(st, target); |
1cf4f629 TG |
945 | /* |
946 | * If the current CPU state is in the range of the AP hotplug thread, | |
947 | * then we need to kick the thread. | |
948 | */ | |
8df3e07e | 949 | if (st->state > CPUHP_TEARDOWN_CPU) { |
4dddfb5f | 950 | st->target = max((int)target, CPUHP_TEARDOWN_CPU); |
1cf4f629 TG |
951 | ret = cpuhp_kick_ap_work(cpu); |
952 | /* | |
953 | * The AP side has done the error rollback already. Just | |
954 | * return the error code.. | |
955 | */ | |
956 | if (ret) | |
957 | goto out; | |
958 | ||
959 | /* | |
960 | * We might have stopped still in the range of the AP hotplug | |
961 | * thread. Nothing to do anymore. | |
962 | */ | |
8df3e07e | 963 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 | 964 | goto out; |
4dddfb5f PZ |
965 | |
966 | st->target = target; | |
1cf4f629 TG |
967 | } |
968 | /* | |
8df3e07e | 969 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
970 | * to do the further cleanups. |
971 | */ | |
a724632c | 972 | ret = cpuhp_down_callbacks(cpu, st, target); |
69fa6eb7 | 973 | if (ret && st->state == CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
4dddfb5f PZ |
974 | cpuhp_reset_state(st, prev_state); |
975 | __cpuhp_kick_ap(st); | |
3b9d6da6 | 976 | } |
98458172 | 977 | |
1cf4f629 | 978 | out: |
8f553c49 | 979 | cpus_write_unlock(); |
941154bd TG |
980 | /* |
981 | * Do post unplug cleanup. This is still protected against | |
982 | * concurrent CPU hotplug via cpu_add_remove_lock. | |
983 | */ | |
984 | lockup_detector_cleanup(); | |
cff7d378 | 985 | return ret; |
e3920fb4 RW |
986 | } |
987 | ||
cc1fe215 TG |
988 | static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target) |
989 | { | |
990 | if (cpu_hotplug_disabled) | |
991 | return -EBUSY; | |
992 | return _cpu_down(cpu, 0, target); | |
993 | } | |
994 | ||
af1f4045 | 995 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 996 | { |
9ea09af3 | 997 | int err; |
e3920fb4 | 998 | |
d221938c | 999 | cpu_maps_update_begin(); |
cc1fe215 | 1000 | err = cpu_down_maps_locked(cpu, target); |
d221938c | 1001 | cpu_maps_update_done(); |
1da177e4 LT |
1002 | return err; |
1003 | } | |
4dddfb5f | 1004 | |
af1f4045 TG |
1005 | int cpu_down(unsigned int cpu) |
1006 | { | |
1007 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
1008 | } | |
b62b8ef9 | 1009 | EXPORT_SYMBOL(cpu_down); |
4dddfb5f PZ |
1010 | |
1011 | #else | |
1012 | #define takedown_cpu NULL | |
1da177e4 LT |
1013 | #endif /*CONFIG_HOTPLUG_CPU*/ |
1014 | ||
4baa0afc | 1015 | /** |
ee1e714b | 1016 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
1017 | * @cpu: cpu that just started |
1018 | * | |
4baa0afc TG |
1019 | * It must be called by the arch code on the new cpu, before the new cpu |
1020 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
1021 | */ | |
1022 | void notify_cpu_starting(unsigned int cpu) | |
1023 | { | |
1024 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1025 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
724a8688 | 1026 | int ret; |
4baa0afc | 1027 | |
0c6d4576 | 1028 | rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
0cc3cd21 | 1029 | st->booted_once = true; |
4baa0afc | 1030 | while (st->state < target) { |
4baa0afc | 1031 | st->state++; |
724a8688 PZ |
1032 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); |
1033 | /* | |
1034 | * STARTING must not fail! | |
1035 | */ | |
1036 | WARN_ON_ONCE(ret); | |
4baa0afc TG |
1037 | } |
1038 | } | |
1039 | ||
949338e3 | 1040 | /* |
9cd4f1a4 TG |
1041 | * Called from the idle task. Wake up the controlling task which brings the |
1042 | * stopper and the hotplug thread of the upcoming CPU up and then delegates | |
1043 | * the rest of the online bringup to the hotplug thread. | |
949338e3 | 1044 | */ |
8df3e07e | 1045 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 1046 | { |
8df3e07e | 1047 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
8df3e07e TG |
1048 | |
1049 | /* Happens for the boot cpu */ | |
1050 | if (state != CPUHP_AP_ONLINE_IDLE) | |
1051 | return; | |
1052 | ||
1053 | st->state = CPUHP_AP_ONLINE_IDLE; | |
5ebe7742 | 1054 | complete_ap_thread(st, true); |
949338e3 TG |
1055 | } |
1056 | ||
e3920fb4 | 1057 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 1058 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 1059 | { |
cff7d378 | 1060 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 1061 | struct task_struct *idle; |
2e1a3483 | 1062 | int ret = 0; |
1da177e4 | 1063 | |
8f553c49 | 1064 | cpus_write_lock(); |
38498a67 | 1065 | |
757c989b | 1066 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
1067 | ret = -EINVAL; |
1068 | goto out; | |
1069 | } | |
1070 | ||
757c989b TG |
1071 | /* |
1072 | * The caller of do_cpu_up might have raced with another | |
1073 | * caller. Ignore it for now. | |
1074 | */ | |
1075 | if (st->state >= target) | |
38498a67 | 1076 | goto out; |
757c989b TG |
1077 | |
1078 | if (st->state == CPUHP_OFFLINE) { | |
1079 | /* Let it fail before we try to bring the cpu up */ | |
1080 | idle = idle_thread_get(cpu); | |
1081 | if (IS_ERR(idle)) { | |
1082 | ret = PTR_ERR(idle); | |
1083 | goto out; | |
1084 | } | |
3bb5d2ee | 1085 | } |
38498a67 | 1086 | |
ba997462 TG |
1087 | cpuhp_tasks_frozen = tasks_frozen; |
1088 | ||
4dddfb5f | 1089 | cpuhp_set_state(st, target); |
1cf4f629 TG |
1090 | /* |
1091 | * If the current CPU state is in the range of the AP hotplug thread, | |
1092 | * then we need to kick the thread once more. | |
1093 | */ | |
8df3e07e | 1094 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
1095 | ret = cpuhp_kick_ap_work(cpu); |
1096 | /* | |
1097 | * The AP side has done the error rollback already. Just | |
1098 | * return the error code.. | |
1099 | */ | |
1100 | if (ret) | |
1101 | goto out; | |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 1106 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
1107 | * responsible for bringing it up to the target state. |
1108 | */ | |
8df3e07e | 1109 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 1110 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 1111 | out: |
8f553c49 | 1112 | cpus_write_unlock(); |
e3920fb4 RW |
1113 | return ret; |
1114 | } | |
1115 | ||
af1f4045 | 1116 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1117 | { |
1118 | int err = 0; | |
cf23422b | 1119 | |
e0b582ec | 1120 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1121 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1122 | cpu); | |
87d5e023 | 1123 | #if defined(CONFIG_IA64) |
84117da5 | 1124 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
1125 | #endif |
1126 | return -EINVAL; | |
1127 | } | |
e3920fb4 | 1128 | |
01b0f197 TK |
1129 | err = try_online_node(cpu_to_node(cpu)); |
1130 | if (err) | |
1131 | return err; | |
cf23422b | 1132 | |
d221938c | 1133 | cpu_maps_update_begin(); |
e761b772 MK |
1134 | |
1135 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1136 | err = -EBUSY; |
e761b772 MK |
1137 | goto out; |
1138 | } | |
05736e4a TG |
1139 | if (!cpu_smt_allowed(cpu)) { |
1140 | err = -EPERM; | |
1141 | goto out; | |
1142 | } | |
e761b772 | 1143 | |
af1f4045 | 1144 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1145 | out: |
d221938c | 1146 | cpu_maps_update_done(); |
e3920fb4 RW |
1147 | return err; |
1148 | } | |
af1f4045 TG |
1149 | |
1150 | int cpu_up(unsigned int cpu) | |
1151 | { | |
1152 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
1153 | } | |
a513f6ba | 1154 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 1155 | |
f3de4be9 | 1156 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1157 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 1158 | |
d391e552 | 1159 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 1160 | { |
d391e552 | 1161 | int cpu, error = 0; |
e3920fb4 | 1162 | |
d221938c | 1163 | cpu_maps_update_begin(); |
d391e552 JM |
1164 | if (!cpu_online(primary)) |
1165 | primary = cpumask_first(cpu_online_mask); | |
9ee349ad XF |
1166 | /* |
1167 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1168 | * with the userspace trying to use the CPU hotplug at the same time |
1169 | */ | |
e0b582ec | 1170 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1171 | |
84117da5 | 1172 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 1173 | for_each_online_cpu(cpu) { |
d391e552 | 1174 | if (cpu == primary) |
e3920fb4 | 1175 | continue; |
bb3632c6 | 1176 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1177 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1178 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1179 | if (!error) |
e0b582ec | 1180 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1181 | else { |
84117da5 | 1182 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1183 | break; |
1184 | } | |
1185 | } | |
86886e55 | 1186 | |
89af7ba5 | 1187 | if (!error) |
e3920fb4 | 1188 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1189 | else |
84117da5 | 1190 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1191 | |
1192 | /* | |
1193 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
1194 | * this even in case of failure as all disable_nonboot_cpus() users are | |
1195 | * supposed to do enable_nonboot_cpus() on the failure path. | |
1196 | */ | |
1197 | cpu_hotplug_disabled++; | |
1198 | ||
d221938c | 1199 | cpu_maps_update_done(); |
e3920fb4 RW |
1200 | return error; |
1201 | } | |
1202 | ||
d0af9eed SS |
1203 | void __weak arch_enable_nonboot_cpus_begin(void) |
1204 | { | |
1205 | } | |
1206 | ||
1207 | void __weak arch_enable_nonboot_cpus_end(void) | |
1208 | { | |
1209 | } | |
1210 | ||
71cf5aee | 1211 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
1212 | { |
1213 | int cpu, error; | |
1214 | ||
1215 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1216 | cpu_maps_update_begin(); |
01b41159 | 1217 | __cpu_hotplug_enable(); |
e0b582ec | 1218 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1219 | goto out; |
e3920fb4 | 1220 | |
84117da5 | 1221 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
1222 | |
1223 | arch_enable_nonboot_cpus_begin(); | |
1224 | ||
e0b582ec | 1225 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1226 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1227 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1228 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1229 | if (!error) { |
84117da5 | 1230 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1231 | continue; |
1232 | } | |
84117da5 | 1233 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1234 | } |
d0af9eed SS |
1235 | |
1236 | arch_enable_nonboot_cpus_end(); | |
1237 | ||
e0b582ec | 1238 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1239 | out: |
d221938c | 1240 | cpu_maps_update_done(); |
1da177e4 | 1241 | } |
e0b582ec | 1242 | |
d7268a31 | 1243 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1244 | { |
1245 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1246 | return -ENOMEM; | |
1247 | return 0; | |
1248 | } | |
1249 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1250 | |
79cfbdfa SB |
1251 | /* |
1252 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1253 | * ensure that the state of the system with respect to the tasks being frozen | |
1254 | * or not, as reported by the notification, remains unchanged *throughout the | |
1255 | * duration* of the execution of the callbacks. | |
1256 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1257 | * | |
1258 | * This synchronization is implemented by mutually excluding regular CPU | |
1259 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1260 | * Hibernate notifications. | |
1261 | */ | |
1262 | static int | |
1263 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1264 | unsigned long action, void *ptr) | |
1265 | { | |
1266 | switch (action) { | |
1267 | ||
1268 | case PM_SUSPEND_PREPARE: | |
1269 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1270 | cpu_hotplug_disable(); |
79cfbdfa SB |
1271 | break; |
1272 | ||
1273 | case PM_POST_SUSPEND: | |
1274 | case PM_POST_HIBERNATION: | |
16e53dbf | 1275 | cpu_hotplug_enable(); |
79cfbdfa SB |
1276 | break; |
1277 | ||
1278 | default: | |
1279 | return NOTIFY_DONE; | |
1280 | } | |
1281 | ||
1282 | return NOTIFY_OK; | |
1283 | } | |
1284 | ||
1285 | ||
d7268a31 | 1286 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1287 | { |
6e32d479 FY |
1288 | /* |
1289 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1290 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1291 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1292 | */ | |
79cfbdfa SB |
1293 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1294 | return 0; | |
1295 | } | |
1296 | core_initcall(cpu_hotplug_pm_sync_init); | |
1297 | ||
f3de4be9 | 1298 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 1299 | |
8ce371f9 PZ |
1300 | int __boot_cpu_id; |
1301 | ||
68f4f1ec | 1302 | #endif /* CONFIG_SMP */ |
b8d317d1 | 1303 | |
cff7d378 | 1304 | /* Boot processor state steps */ |
17a2f1ce | 1305 | static struct cpuhp_step cpuhp_hp_states[] = { |
cff7d378 TG |
1306 | [CPUHP_OFFLINE] = { |
1307 | .name = "offline", | |
3c1627e9 TG |
1308 | .startup.single = NULL, |
1309 | .teardown.single = NULL, | |
cff7d378 TG |
1310 | }, |
1311 | #ifdef CONFIG_SMP | |
1312 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1313 | .name = "threads:prepare", |
3c1627e9 TG |
1314 | .startup.single = smpboot_create_threads, |
1315 | .teardown.single = NULL, | |
757c989b | 1316 | .cant_stop = true, |
cff7d378 | 1317 | }, |
00e16c3d | 1318 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1319 | .name = "perf:prepare", |
1320 | .startup.single = perf_event_init_cpu, | |
1321 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1322 | }, |
7ee681b2 | 1323 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1324 | .name = "workqueue:prepare", |
1325 | .startup.single = workqueue_prepare_cpu, | |
1326 | .teardown.single = NULL, | |
7ee681b2 | 1327 | }, |
27590dc1 | 1328 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1329 | .name = "hrtimers:prepare", |
1330 | .startup.single = hrtimers_prepare_cpu, | |
1331 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1332 | }, |
31487f83 | 1333 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1334 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1335 | .startup.single = smpcfd_prepare_cpu, |
1336 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1337 | }, |
e6d4989a RW |
1338 | [CPUHP_RELAY_PREPARE] = { |
1339 | .name = "relay:prepare", | |
1340 | .startup.single = relay_prepare_cpu, | |
1341 | .teardown.single = NULL, | |
1342 | }, | |
6731d4f1 SAS |
1343 | [CPUHP_SLAB_PREPARE] = { |
1344 | .name = "slab:prepare", | |
1345 | .startup.single = slab_prepare_cpu, | |
1346 | .teardown.single = slab_dead_cpu, | |
31487f83 | 1347 | }, |
4df83742 | 1348 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1349 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1350 | .startup.single = rcutree_prepare_cpu, |
1351 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1352 | }, |
4fae16df RC |
1353 | /* |
1354 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1355 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1356 | * otherwise a RCU stall occurs. | |
1357 | */ | |
26456f87 | 1358 | [CPUHP_TIMERS_PREPARE] = { |
d018031f | 1359 | .name = "timers:prepare", |
26456f87 | 1360 | .startup.single = timers_prepare_cpu, |
3c1627e9 | 1361 | .teardown.single = timers_dead_cpu, |
4fae16df | 1362 | }, |
d10ef6f9 | 1363 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1364 | [CPUHP_BRINGUP_CPU] = { |
1365 | .name = "cpu:bringup", | |
3c1627e9 TG |
1366 | .startup.single = bringup_cpu, |
1367 | .teardown.single = NULL, | |
757c989b | 1368 | .cant_stop = true, |
4baa0afc | 1369 | }, |
d10ef6f9 TG |
1370 | /* Final state before CPU kills itself */ |
1371 | [CPUHP_AP_IDLE_DEAD] = { | |
1372 | .name = "idle:dead", | |
1373 | }, | |
1374 | /* | |
1375 | * Last state before CPU enters the idle loop to die. Transient state | |
1376 | * for synchronization. | |
1377 | */ | |
1378 | [CPUHP_AP_OFFLINE] = { | |
1379 | .name = "ap:offline", | |
1380 | .cant_stop = true, | |
1381 | }, | |
9cf7243d TG |
1382 | /* First state is scheduler control. Interrupts are disabled */ |
1383 | [CPUHP_AP_SCHED_STARTING] = { | |
1384 | .name = "sched:starting", | |
3c1627e9 TG |
1385 | .startup.single = sched_cpu_starting, |
1386 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1387 | }, |
4df83742 | 1388 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1389 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1390 | .startup.single = NULL, |
1391 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 1392 | }, |
46febd37 LJ |
1393 | [CPUHP_AP_SMPCFD_DYING] = { |
1394 | .name = "smpcfd:dying", | |
1395 | .startup.single = NULL, | |
1396 | .teardown.single = smpcfd_dying_cpu, | |
1397 | }, | |
d10ef6f9 TG |
1398 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1399 | * state for synchronsization */ | |
1400 | [CPUHP_AP_ONLINE] = { | |
1401 | .name = "ap:online", | |
1402 | }, | |
17a2f1ce LJ |
1403 | /* |
1404 | * Handled on controll processor until the plugged processor manages | |
1405 | * this itself. | |
1406 | */ | |
1407 | [CPUHP_TEARDOWN_CPU] = { | |
1408 | .name = "cpu:teardown", | |
1409 | .startup.single = NULL, | |
1410 | .teardown.single = takedown_cpu, | |
1411 | .cant_stop = true, | |
1412 | }, | |
d10ef6f9 | 1413 | /* Handle smpboot threads park/unpark */ |
1cf4f629 | 1414 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1415 | .name = "smpboot/threads:online", |
3c1627e9 | 1416 | .startup.single = smpboot_unpark_threads, |
c4de6569 | 1417 | .teardown.single = smpboot_park_threads, |
1cf4f629 | 1418 | }, |
c5cb83bb TG |
1419 | [CPUHP_AP_IRQ_AFFINITY_ONLINE] = { |
1420 | .name = "irq/affinity:online", | |
1421 | .startup.single = irq_affinity_online_cpu, | |
1422 | .teardown.single = NULL, | |
1423 | }, | |
00e16c3d | 1424 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1425 | .name = "perf:online", |
1426 | .startup.single = perf_event_init_cpu, | |
1427 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1428 | }, |
9cf57731 PZ |
1429 | [CPUHP_AP_WATCHDOG_ONLINE] = { |
1430 | .name = "lockup_detector:online", | |
1431 | .startup.single = lockup_detector_online_cpu, | |
1432 | .teardown.single = lockup_detector_offline_cpu, | |
1433 | }, | |
7ee681b2 | 1434 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1435 | .name = "workqueue:online", |
1436 | .startup.single = workqueue_online_cpu, | |
1437 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1438 | }, |
4df83742 | 1439 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1440 | .name = "RCU/tree:online", |
3c1627e9 TG |
1441 | .startup.single = rcutree_online_cpu, |
1442 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1443 | }, |
4baa0afc | 1444 | #endif |
d10ef6f9 TG |
1445 | /* |
1446 | * The dynamically registered state space is here | |
1447 | */ | |
1448 | ||
aaddd7d1 TG |
1449 | #ifdef CONFIG_SMP |
1450 | /* Last state is scheduler control setting the cpu active */ | |
1451 | [CPUHP_AP_ACTIVE] = { | |
1452 | .name = "sched:active", | |
3c1627e9 TG |
1453 | .startup.single = sched_cpu_activate, |
1454 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1455 | }, |
1456 | #endif | |
1457 | ||
d10ef6f9 | 1458 | /* CPU is fully up and running. */ |
4baa0afc TG |
1459 | [CPUHP_ONLINE] = { |
1460 | .name = "online", | |
3c1627e9 TG |
1461 | .startup.single = NULL, |
1462 | .teardown.single = NULL, | |
4baa0afc TG |
1463 | }, |
1464 | }; | |
1465 | ||
5b7aa87e TG |
1466 | /* Sanity check for callbacks */ |
1467 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1468 | { | |
1469 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1470 | return -EINVAL; | |
1471 | return 0; | |
1472 | } | |
1473 | ||
dc280d93 TG |
1474 | /* |
1475 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1476 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1477 | * by having no name assigned. | |
1478 | */ | |
1479 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1480 | { | |
4205e478 TG |
1481 | enum cpuhp_state i, end; |
1482 | struct cpuhp_step *step; | |
dc280d93 | 1483 | |
4205e478 TG |
1484 | switch (state) { |
1485 | case CPUHP_AP_ONLINE_DYN: | |
17a2f1ce | 1486 | step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN; |
4205e478 TG |
1487 | end = CPUHP_AP_ONLINE_DYN_END; |
1488 | break; | |
1489 | case CPUHP_BP_PREPARE_DYN: | |
17a2f1ce | 1490 | step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN; |
4205e478 TG |
1491 | end = CPUHP_BP_PREPARE_DYN_END; |
1492 | break; | |
1493 | default: | |
1494 | return -EINVAL; | |
1495 | } | |
1496 | ||
1497 | for (i = state; i <= end; i++, step++) { | |
1498 | if (!step->name) | |
dc280d93 TG |
1499 | return i; |
1500 | } | |
1501 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1502 | return -ENOSPC; | |
1503 | } | |
1504 | ||
1505 | static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, | |
1506 | int (*startup)(unsigned int cpu), | |
1507 | int (*teardown)(unsigned int cpu), | |
1508 | bool multi_instance) | |
5b7aa87e TG |
1509 | { |
1510 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1511 | struct cpuhp_step *sp; | |
dc280d93 | 1512 | int ret = 0; |
5b7aa87e | 1513 | |
0c96b273 EB |
1514 | /* |
1515 | * If name is NULL, then the state gets removed. | |
1516 | * | |
1517 | * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on | |
1518 | * the first allocation from these dynamic ranges, so the removal | |
1519 | * would trigger a new allocation and clear the wrong (already | |
1520 | * empty) state, leaving the callbacks of the to be cleared state | |
1521 | * dangling, which causes wreckage on the next hotplug operation. | |
1522 | */ | |
1523 | if (name && (state == CPUHP_AP_ONLINE_DYN || | |
1524 | state == CPUHP_BP_PREPARE_DYN)) { | |
dc280d93 TG |
1525 | ret = cpuhp_reserve_state(state); |
1526 | if (ret < 0) | |
dc434e05 | 1527 | return ret; |
dc280d93 TG |
1528 | state = ret; |
1529 | } | |
5b7aa87e | 1530 | sp = cpuhp_get_step(state); |
dc434e05 SAS |
1531 | if (name && sp->name) |
1532 | return -EBUSY; | |
1533 | ||
3c1627e9 TG |
1534 | sp->startup.single = startup; |
1535 | sp->teardown.single = teardown; | |
5b7aa87e | 1536 | sp->name = name; |
cf392d10 TG |
1537 | sp->multi_instance = multi_instance; |
1538 | INIT_HLIST_HEAD(&sp->list); | |
dc280d93 | 1539 | return ret; |
5b7aa87e TG |
1540 | } |
1541 | ||
1542 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1543 | { | |
3c1627e9 | 1544 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1545 | } |
1546 | ||
5b7aa87e TG |
1547 | /* |
1548 | * Call the startup/teardown function for a step either on the AP or | |
1549 | * on the current CPU. | |
1550 | */ | |
cf392d10 TG |
1551 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1552 | struct hlist_node *node) | |
5b7aa87e | 1553 | { |
a724632c | 1554 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1555 | int ret; |
1556 | ||
4dddfb5f PZ |
1557 | /* |
1558 | * If there's nothing to do, we done. | |
1559 | * Relies on the union for multi_instance. | |
1560 | */ | |
3c1627e9 TG |
1561 | if ((bringup && !sp->startup.single) || |
1562 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1563 | return 0; |
5b7aa87e TG |
1564 | /* |
1565 | * The non AP bound callbacks can fail on bringup. On teardown | |
1566 | * e.g. module removal we crash for now. | |
1567 | */ | |
1cf4f629 TG |
1568 | #ifdef CONFIG_SMP |
1569 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1570 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1571 | else |
96abb968 | 1572 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1573 | #else |
96abb968 | 1574 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1575 | #endif |
5b7aa87e TG |
1576 | BUG_ON(ret && !bringup); |
1577 | return ret; | |
1578 | } | |
1579 | ||
1580 | /* | |
1581 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1582 | * | |
1583 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1584 | */ | |
1585 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 1586 | struct hlist_node *node) |
5b7aa87e TG |
1587 | { |
1588 | int cpu; | |
1589 | ||
5b7aa87e TG |
1590 | /* Roll back the already executed steps on the other cpus */ |
1591 | for_each_present_cpu(cpu) { | |
1592 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1593 | int cpustate = st->state; | |
1594 | ||
1595 | if (cpu >= failedcpu) | |
1596 | break; | |
1597 | ||
1598 | /* Did we invoke the startup call on that cpu ? */ | |
1599 | if (cpustate >= state) | |
cf392d10 | 1600 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1601 | } |
1602 | } | |
1603 | ||
9805c673 TG |
1604 | int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state, |
1605 | struct hlist_node *node, | |
1606 | bool invoke) | |
cf392d10 TG |
1607 | { |
1608 | struct cpuhp_step *sp; | |
1609 | int cpu; | |
1610 | int ret; | |
1611 | ||
9805c673 TG |
1612 | lockdep_assert_cpus_held(); |
1613 | ||
cf392d10 TG |
1614 | sp = cpuhp_get_step(state); |
1615 | if (sp->multi_instance == false) | |
1616 | return -EINVAL; | |
1617 | ||
dc434e05 | 1618 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 | 1619 | |
3c1627e9 | 1620 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1621 | goto add_node; |
1622 | ||
1623 | /* | |
1624 | * Try to call the startup callback for each present cpu | |
1625 | * depending on the hotplug state of the cpu. | |
1626 | */ | |
1627 | for_each_present_cpu(cpu) { | |
1628 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1629 | int cpustate = st->state; | |
1630 | ||
1631 | if (cpustate < state) | |
1632 | continue; | |
1633 | ||
1634 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1635 | if (ret) { | |
3c1627e9 | 1636 | if (sp->teardown.multi) |
cf392d10 | 1637 | cpuhp_rollback_install(cpu, state, node); |
dc434e05 | 1638 | goto unlock; |
cf392d10 TG |
1639 | } |
1640 | } | |
1641 | add_node: | |
1642 | ret = 0; | |
cf392d10 | 1643 | hlist_add_head(node, &sp->list); |
dc434e05 | 1644 | unlock: |
cf392d10 | 1645 | mutex_unlock(&cpuhp_state_mutex); |
9805c673 TG |
1646 | return ret; |
1647 | } | |
1648 | ||
1649 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, | |
1650 | bool invoke) | |
1651 | { | |
1652 | int ret; | |
1653 | ||
1654 | cpus_read_lock(); | |
1655 | ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke); | |
8f553c49 | 1656 | cpus_read_unlock(); |
cf392d10 TG |
1657 | return ret; |
1658 | } | |
1659 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1660 | ||
5b7aa87e | 1661 | /** |
71def423 | 1662 | * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state |
dc280d93 TG |
1663 | * @state: The state to setup |
1664 | * @invoke: If true, the startup function is invoked for cpus where | |
1665 | * cpu state >= @state | |
1666 | * @startup: startup callback function | |
1667 | * @teardown: teardown callback function | |
1668 | * @multi_instance: State is set up for multiple instances which get | |
1669 | * added afterwards. | |
5b7aa87e | 1670 | * |
71def423 | 1671 | * The caller needs to hold cpus read locked while calling this function. |
512f0980 BO |
1672 | * Returns: |
1673 | * On success: | |
1674 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN | |
1675 | * 0 for all other states | |
1676 | * On failure: proper (negative) error code | |
5b7aa87e | 1677 | */ |
71def423 SAS |
1678 | int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state, |
1679 | const char *name, bool invoke, | |
1680 | int (*startup)(unsigned int cpu), | |
1681 | int (*teardown)(unsigned int cpu), | |
1682 | bool multi_instance) | |
5b7aa87e TG |
1683 | { |
1684 | int cpu, ret = 0; | |
b9d9d691 | 1685 | bool dynstate; |
5b7aa87e | 1686 | |
71def423 SAS |
1687 | lockdep_assert_cpus_held(); |
1688 | ||
5b7aa87e TG |
1689 | if (cpuhp_cb_check(state) || !name) |
1690 | return -EINVAL; | |
1691 | ||
dc434e05 | 1692 | mutex_lock(&cpuhp_state_mutex); |
5b7aa87e | 1693 | |
dc280d93 TG |
1694 | ret = cpuhp_store_callbacks(state, name, startup, teardown, |
1695 | multi_instance); | |
5b7aa87e | 1696 | |
b9d9d691 TG |
1697 | dynstate = state == CPUHP_AP_ONLINE_DYN; |
1698 | if (ret > 0 && dynstate) { | |
1699 | state = ret; | |
1700 | ret = 0; | |
1701 | } | |
1702 | ||
dc280d93 | 1703 | if (ret || !invoke || !startup) |
5b7aa87e TG |
1704 | goto out; |
1705 | ||
1706 | /* | |
1707 | * Try to call the startup callback for each present cpu | |
1708 | * depending on the hotplug state of the cpu. | |
1709 | */ | |
1710 | for_each_present_cpu(cpu) { | |
1711 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1712 | int cpustate = st->state; | |
1713 | ||
1714 | if (cpustate < state) | |
1715 | continue; | |
1716 | ||
cf392d10 | 1717 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1718 | if (ret) { |
a724632c | 1719 | if (teardown) |
cf392d10 TG |
1720 | cpuhp_rollback_install(cpu, state, NULL); |
1721 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1722 | goto out; |
1723 | } | |
1724 | } | |
1725 | out: | |
dc434e05 | 1726 | mutex_unlock(&cpuhp_state_mutex); |
dc280d93 TG |
1727 | /* |
1728 | * If the requested state is CPUHP_AP_ONLINE_DYN, return the | |
1729 | * dynamically allocated state in case of success. | |
1730 | */ | |
b9d9d691 | 1731 | if (!ret && dynstate) |
5b7aa87e TG |
1732 | return state; |
1733 | return ret; | |
1734 | } | |
71def423 SAS |
1735 | EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked); |
1736 | ||
1737 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1738 | const char *name, bool invoke, | |
1739 | int (*startup)(unsigned int cpu), | |
1740 | int (*teardown)(unsigned int cpu), | |
1741 | bool multi_instance) | |
1742 | { | |
1743 | int ret; | |
1744 | ||
1745 | cpus_read_lock(); | |
1746 | ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup, | |
1747 | teardown, multi_instance); | |
1748 | cpus_read_unlock(); | |
1749 | return ret; | |
1750 | } | |
5b7aa87e TG |
1751 | EXPORT_SYMBOL(__cpuhp_setup_state); |
1752 | ||
cf392d10 TG |
1753 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1754 | struct hlist_node *node, bool invoke) | |
1755 | { | |
1756 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1757 | int cpu; | |
1758 | ||
1759 | BUG_ON(cpuhp_cb_check(state)); | |
1760 | ||
1761 | if (!sp->multi_instance) | |
1762 | return -EINVAL; | |
1763 | ||
8f553c49 | 1764 | cpus_read_lock(); |
dc434e05 SAS |
1765 | mutex_lock(&cpuhp_state_mutex); |
1766 | ||
cf392d10 TG |
1767 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
1768 | goto remove; | |
1769 | /* | |
1770 | * Call the teardown callback for each present cpu depending | |
1771 | * on the hotplug state of the cpu. This function is not | |
1772 | * allowed to fail currently! | |
1773 | */ | |
1774 | for_each_present_cpu(cpu) { | |
1775 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1776 | int cpustate = st->state; | |
1777 | ||
1778 | if (cpustate >= state) | |
1779 | cpuhp_issue_call(cpu, state, false, node); | |
1780 | } | |
1781 | ||
1782 | remove: | |
cf392d10 TG |
1783 | hlist_del(node); |
1784 | mutex_unlock(&cpuhp_state_mutex); | |
8f553c49 | 1785 | cpus_read_unlock(); |
cf392d10 TG |
1786 | |
1787 | return 0; | |
1788 | } | |
1789 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
dc434e05 | 1790 | |
5b7aa87e | 1791 | /** |
71def423 | 1792 | * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state |
5b7aa87e TG |
1793 | * @state: The state to remove |
1794 | * @invoke: If true, the teardown function is invoked for cpus where | |
1795 | * cpu state >= @state | |
1796 | * | |
71def423 | 1797 | * The caller needs to hold cpus read locked while calling this function. |
5b7aa87e TG |
1798 | * The teardown callback is currently not allowed to fail. Think |
1799 | * about module removal! | |
1800 | */ | |
71def423 | 1801 | void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke) |
5b7aa87e | 1802 | { |
cf392d10 | 1803 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1804 | int cpu; |
1805 | ||
1806 | BUG_ON(cpuhp_cb_check(state)); | |
1807 | ||
71def423 | 1808 | lockdep_assert_cpus_held(); |
5b7aa87e | 1809 | |
dc434e05 | 1810 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 TG |
1811 | if (sp->multi_instance) { |
1812 | WARN(!hlist_empty(&sp->list), | |
1813 | "Error: Removing state %d which has instances left.\n", | |
1814 | state); | |
1815 | goto remove; | |
1816 | } | |
1817 | ||
a724632c | 1818 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1819 | goto remove; |
1820 | ||
1821 | /* | |
1822 | * Call the teardown callback for each present cpu depending | |
1823 | * on the hotplug state of the cpu. This function is not | |
1824 | * allowed to fail currently! | |
1825 | */ | |
1826 | for_each_present_cpu(cpu) { | |
1827 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1828 | int cpustate = st->state; | |
1829 | ||
1830 | if (cpustate >= state) | |
cf392d10 | 1831 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1832 | } |
1833 | remove: | |
cf392d10 | 1834 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
dc434e05 | 1835 | mutex_unlock(&cpuhp_state_mutex); |
71def423 SAS |
1836 | } |
1837 | EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked); | |
1838 | ||
1839 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1840 | { | |
1841 | cpus_read_lock(); | |
1842 | __cpuhp_remove_state_cpuslocked(state, invoke); | |
8f553c49 | 1843 | cpus_read_unlock(); |
5b7aa87e TG |
1844 | } |
1845 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1846 | ||
98f8cdce TG |
1847 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1848 | static ssize_t show_cpuhp_state(struct device *dev, | |
1849 | struct device_attribute *attr, char *buf) | |
1850 | { | |
1851 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1852 | ||
1853 | return sprintf(buf, "%d\n", st->state); | |
1854 | } | |
1855 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1856 | ||
757c989b TG |
1857 | static ssize_t write_cpuhp_target(struct device *dev, |
1858 | struct device_attribute *attr, | |
1859 | const char *buf, size_t count) | |
1860 | { | |
1861 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1862 | struct cpuhp_step *sp; | |
1863 | int target, ret; | |
1864 | ||
1865 | ret = kstrtoint(buf, 10, &target); | |
1866 | if (ret) | |
1867 | return ret; | |
1868 | ||
1869 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1870 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1871 | return -EINVAL; | |
1872 | #else | |
1873 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1874 | return -EINVAL; | |
1875 | #endif | |
1876 | ||
1877 | ret = lock_device_hotplug_sysfs(); | |
1878 | if (ret) | |
1879 | return ret; | |
1880 | ||
1881 | mutex_lock(&cpuhp_state_mutex); | |
1882 | sp = cpuhp_get_step(target); | |
1883 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1884 | mutex_unlock(&cpuhp_state_mutex); | |
1885 | if (ret) | |
40da1b11 | 1886 | goto out; |
757c989b TG |
1887 | |
1888 | if (st->state < target) | |
1889 | ret = do_cpu_up(dev->id, target); | |
1890 | else | |
1891 | ret = do_cpu_down(dev->id, target); | |
40da1b11 | 1892 | out: |
757c989b TG |
1893 | unlock_device_hotplug(); |
1894 | return ret ? ret : count; | |
1895 | } | |
1896 | ||
98f8cdce TG |
1897 | static ssize_t show_cpuhp_target(struct device *dev, |
1898 | struct device_attribute *attr, char *buf) | |
1899 | { | |
1900 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1901 | ||
1902 | return sprintf(buf, "%d\n", st->target); | |
1903 | } | |
757c989b | 1904 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce | 1905 | |
1db49484 PZ |
1906 | |
1907 | static ssize_t write_cpuhp_fail(struct device *dev, | |
1908 | struct device_attribute *attr, | |
1909 | const char *buf, size_t count) | |
1910 | { | |
1911 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1912 | struct cpuhp_step *sp; | |
1913 | int fail, ret; | |
1914 | ||
1915 | ret = kstrtoint(buf, 10, &fail); | |
1916 | if (ret) | |
1917 | return ret; | |
1918 | ||
1919 | /* | |
1920 | * Cannot fail STARTING/DYING callbacks. | |
1921 | */ | |
1922 | if (cpuhp_is_atomic_state(fail)) | |
1923 | return -EINVAL; | |
1924 | ||
1925 | /* | |
1926 | * Cannot fail anything that doesn't have callbacks. | |
1927 | */ | |
1928 | mutex_lock(&cpuhp_state_mutex); | |
1929 | sp = cpuhp_get_step(fail); | |
1930 | if (!sp->startup.single && !sp->teardown.single) | |
1931 | ret = -EINVAL; | |
1932 | mutex_unlock(&cpuhp_state_mutex); | |
1933 | if (ret) | |
1934 | return ret; | |
1935 | ||
1936 | st->fail = fail; | |
1937 | ||
1938 | return count; | |
1939 | } | |
1940 | ||
1941 | static ssize_t show_cpuhp_fail(struct device *dev, | |
1942 | struct device_attribute *attr, char *buf) | |
1943 | { | |
1944 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1945 | ||
1946 | return sprintf(buf, "%d\n", st->fail); | |
1947 | } | |
1948 | ||
1949 | static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail); | |
1950 | ||
98f8cdce TG |
1951 | static struct attribute *cpuhp_cpu_attrs[] = { |
1952 | &dev_attr_state.attr, | |
1953 | &dev_attr_target.attr, | |
1db49484 | 1954 | &dev_attr_fail.attr, |
98f8cdce TG |
1955 | NULL |
1956 | }; | |
1957 | ||
993647a2 | 1958 | static const struct attribute_group cpuhp_cpu_attr_group = { |
98f8cdce TG |
1959 | .attrs = cpuhp_cpu_attrs, |
1960 | .name = "hotplug", | |
1961 | NULL | |
1962 | }; | |
1963 | ||
1964 | static ssize_t show_cpuhp_states(struct device *dev, | |
1965 | struct device_attribute *attr, char *buf) | |
1966 | { | |
1967 | ssize_t cur, res = 0; | |
1968 | int i; | |
1969 | ||
1970 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1971 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1972 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1973 | ||
1974 | if (sp->name) { | |
1975 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1976 | buf += cur; | |
1977 | res += cur; | |
1978 | } | |
1979 | } | |
1980 | mutex_unlock(&cpuhp_state_mutex); | |
1981 | return res; | |
1982 | } | |
1983 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1984 | ||
1985 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1986 | &dev_attr_states.attr, | |
1987 | NULL | |
1988 | }; | |
1989 | ||
993647a2 | 1990 | static const struct attribute_group cpuhp_cpu_root_attr_group = { |
98f8cdce TG |
1991 | .attrs = cpuhp_cpu_root_attrs, |
1992 | .name = "hotplug", | |
1993 | NULL | |
1994 | }; | |
1995 | ||
05736e4a TG |
1996 | #ifdef CONFIG_HOTPLUG_SMT |
1997 | ||
1998 | static const char *smt_states[] = { | |
1999 | [CPU_SMT_ENABLED] = "on", | |
2000 | [CPU_SMT_DISABLED] = "off", | |
2001 | [CPU_SMT_FORCE_DISABLED] = "forceoff", | |
2002 | [CPU_SMT_NOT_SUPPORTED] = "notsupported", | |
2003 | }; | |
2004 | ||
2005 | static ssize_t | |
2006 | show_smt_control(struct device *dev, struct device_attribute *attr, char *buf) | |
2007 | { | |
2008 | return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]); | |
2009 | } | |
2010 | ||
2011 | static void cpuhp_offline_cpu_device(unsigned int cpu) | |
2012 | { | |
2013 | struct device *dev = get_cpu_device(cpu); | |
2014 | ||
2015 | dev->offline = true; | |
2016 | /* Tell user space about the state change */ | |
2017 | kobject_uevent(&dev->kobj, KOBJ_OFFLINE); | |
2018 | } | |
2019 | ||
215af549 TG |
2020 | static void cpuhp_online_cpu_device(unsigned int cpu) |
2021 | { | |
2022 | struct device *dev = get_cpu_device(cpu); | |
2023 | ||
2024 | dev->offline = false; | |
2025 | /* Tell user space about the state change */ | |
2026 | kobject_uevent(&dev->kobj, KOBJ_ONLINE); | |
2027 | } | |
2028 | ||
05736e4a TG |
2029 | static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval) |
2030 | { | |
2031 | int cpu, ret = 0; | |
2032 | ||
2033 | cpu_maps_update_begin(); | |
2034 | for_each_online_cpu(cpu) { | |
2035 | if (topology_is_primary_thread(cpu)) | |
2036 | continue; | |
2037 | ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE); | |
2038 | if (ret) | |
2039 | break; | |
2040 | /* | |
2041 | * As this needs to hold the cpu maps lock it's impossible | |
2042 | * to call device_offline() because that ends up calling | |
2043 | * cpu_down() which takes cpu maps lock. cpu maps lock | |
2044 | * needs to be held as this might race against in kernel | |
2045 | * abusers of the hotplug machinery (thermal management). | |
2046 | * | |
2047 | * So nothing would update device:offline state. That would | |
2048 | * leave the sysfs entry stale and prevent onlining after | |
2049 | * smt control has been changed to 'off' again. This is | |
2050 | * called under the sysfs hotplug lock, so it is properly | |
2051 | * serialized against the regular offline usage. | |
2052 | */ | |
2053 | cpuhp_offline_cpu_device(cpu); | |
2054 | } | |
2055 | if (!ret) | |
2056 | cpu_smt_control = ctrlval; | |
2057 | cpu_maps_update_done(); | |
2058 | return ret; | |
2059 | } | |
2060 | ||
215af549 | 2061 | static int cpuhp_smt_enable(void) |
05736e4a | 2062 | { |
215af549 TG |
2063 | int cpu, ret = 0; |
2064 | ||
05736e4a TG |
2065 | cpu_maps_update_begin(); |
2066 | cpu_smt_control = CPU_SMT_ENABLED; | |
215af549 TG |
2067 | for_each_present_cpu(cpu) { |
2068 | /* Skip online CPUs and CPUs on offline nodes */ | |
2069 | if (cpu_online(cpu) || !node_online(cpu_to_node(cpu))) | |
2070 | continue; | |
2071 | ret = _cpu_up(cpu, 0, CPUHP_ONLINE); | |
2072 | if (ret) | |
2073 | break; | |
2074 | /* See comment in cpuhp_smt_disable() */ | |
2075 | cpuhp_online_cpu_device(cpu); | |
2076 | } | |
05736e4a | 2077 | cpu_maps_update_done(); |
215af549 | 2078 | return ret; |
05736e4a TG |
2079 | } |
2080 | ||
2081 | static ssize_t | |
2082 | store_smt_control(struct device *dev, struct device_attribute *attr, | |
2083 | const char *buf, size_t count) | |
2084 | { | |
2085 | int ctrlval, ret; | |
2086 | ||
2087 | if (sysfs_streq(buf, "on")) | |
2088 | ctrlval = CPU_SMT_ENABLED; | |
2089 | else if (sysfs_streq(buf, "off")) | |
2090 | ctrlval = CPU_SMT_DISABLED; | |
2091 | else if (sysfs_streq(buf, "forceoff")) | |
2092 | ctrlval = CPU_SMT_FORCE_DISABLED; | |
2093 | else | |
2094 | return -EINVAL; | |
2095 | ||
2096 | if (cpu_smt_control == CPU_SMT_FORCE_DISABLED) | |
2097 | return -EPERM; | |
2098 | ||
2099 | if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED) | |
2100 | return -ENODEV; | |
2101 | ||
2102 | ret = lock_device_hotplug_sysfs(); | |
2103 | if (ret) | |
2104 | return ret; | |
2105 | ||
2106 | if (ctrlval != cpu_smt_control) { | |
2107 | switch (ctrlval) { | |
2108 | case CPU_SMT_ENABLED: | |
215af549 | 2109 | ret = cpuhp_smt_enable(); |
05736e4a TG |
2110 | break; |
2111 | case CPU_SMT_DISABLED: | |
2112 | case CPU_SMT_FORCE_DISABLED: | |
2113 | ret = cpuhp_smt_disable(ctrlval); | |
2114 | break; | |
2115 | } | |
2116 | } | |
2117 | ||
2118 | unlock_device_hotplug(); | |
2119 | return ret ? ret : count; | |
2120 | } | |
2121 | static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control); | |
2122 | ||
2123 | static ssize_t | |
2124 | show_smt_active(struct device *dev, struct device_attribute *attr, char *buf) | |
2125 | { | |
2126 | bool active = topology_max_smt_threads() > 1; | |
2127 | ||
2128 | return snprintf(buf, PAGE_SIZE - 2, "%d\n", active); | |
2129 | } | |
2130 | static DEVICE_ATTR(active, 0444, show_smt_active, NULL); | |
2131 | ||
2132 | static struct attribute *cpuhp_smt_attrs[] = { | |
2133 | &dev_attr_control.attr, | |
2134 | &dev_attr_active.attr, | |
2135 | NULL | |
2136 | }; | |
2137 | ||
2138 | static const struct attribute_group cpuhp_smt_attr_group = { | |
2139 | .attrs = cpuhp_smt_attrs, | |
2140 | .name = "smt", | |
2141 | NULL | |
2142 | }; | |
2143 | ||
2144 | static int __init cpu_smt_state_init(void) | |
2145 | { | |
05736e4a TG |
2146 | return sysfs_create_group(&cpu_subsys.dev_root->kobj, |
2147 | &cpuhp_smt_attr_group); | |
2148 | } | |
2149 | ||
2150 | #else | |
2151 | static inline int cpu_smt_state_init(void) { return 0; } | |
2152 | #endif | |
2153 | ||
98f8cdce TG |
2154 | static int __init cpuhp_sysfs_init(void) |
2155 | { | |
2156 | int cpu, ret; | |
2157 | ||
05736e4a TG |
2158 | ret = cpu_smt_state_init(); |
2159 | if (ret) | |
2160 | return ret; | |
2161 | ||
98f8cdce TG |
2162 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, |
2163 | &cpuhp_cpu_root_attr_group); | |
2164 | if (ret) | |
2165 | return ret; | |
2166 | ||
2167 | for_each_possible_cpu(cpu) { | |
2168 | struct device *dev = get_cpu_device(cpu); | |
2169 | ||
2170 | if (!dev) | |
2171 | continue; | |
2172 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
2173 | if (ret) | |
2174 | return ret; | |
2175 | } | |
2176 | return 0; | |
2177 | } | |
2178 | device_initcall(cpuhp_sysfs_init); | |
2179 | #endif | |
2180 | ||
e56b3bc7 LT |
2181 | /* |
2182 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
2183 | * represents all NR_CPUS bits binary values of 1<<nr. | |
2184 | * | |
e0b582ec | 2185 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
2186 | * mask value that has a single bit set only. |
2187 | */ | |
b8d317d1 | 2188 | |
e56b3bc7 | 2189 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 2190 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
2191 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
2192 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
2193 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 2194 | |
e56b3bc7 LT |
2195 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
2196 | ||
2197 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
2198 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
2199 | #if BITS_PER_LONG > 32 | |
2200 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
2201 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
2202 | #endif |
2203 | }; | |
e56b3bc7 | 2204 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
2205 | |
2206 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
2207 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
2208 | |
2209 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 2210 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 2211 | = {CPU_BITS_ALL}; |
b3199c02 | 2212 | #else |
4b804c85 | 2213 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 2214 | #endif |
4b804c85 | 2215 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 2216 | |
4b804c85 RV |
2217 | struct cpumask __cpu_online_mask __read_mostly; |
2218 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 2219 | |
4b804c85 RV |
2220 | struct cpumask __cpu_present_mask __read_mostly; |
2221 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 2222 | |
4b804c85 RV |
2223 | struct cpumask __cpu_active_mask __read_mostly; |
2224 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 2225 | |
3fa41520 RR |
2226 | void init_cpu_present(const struct cpumask *src) |
2227 | { | |
c4c54dd1 | 2228 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
2229 | } |
2230 | ||
2231 | void init_cpu_possible(const struct cpumask *src) | |
2232 | { | |
c4c54dd1 | 2233 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
2234 | } |
2235 | ||
2236 | void init_cpu_online(const struct cpumask *src) | |
2237 | { | |
c4c54dd1 | 2238 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 2239 | } |
cff7d378 TG |
2240 | |
2241 | /* | |
2242 | * Activate the first processor. | |
2243 | */ | |
2244 | void __init boot_cpu_init(void) | |
2245 | { | |
2246 | int cpu = smp_processor_id(); | |
2247 | ||
2248 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
2249 | set_cpu_online(cpu, true); | |
2250 | set_cpu_active(cpu, true); | |
2251 | set_cpu_present(cpu, true); | |
2252 | set_cpu_possible(cpu, true); | |
8ce371f9 PZ |
2253 | |
2254 | #ifdef CONFIG_SMP | |
2255 | __boot_cpu_id = cpu; | |
2256 | #endif | |
cff7d378 TG |
2257 | } |
2258 | ||
2259 | /* | |
2260 | * Must be called _AFTER_ setting up the per_cpu areas | |
2261 | */ | |
b5b1404d | 2262 | void __init boot_cpu_hotplug_init(void) |
cff7d378 | 2263 | { |
269777aa | 2264 | #ifdef CONFIG_SMP |
0cc3cd21 | 2265 | this_cpu_write(cpuhp_state.booted_once, true); |
269777aa | 2266 | #endif |
0cc3cd21 | 2267 | this_cpu_write(cpuhp_state.state, CPUHP_ONLINE); |
cff7d378 | 2268 | } |