Commit | Line | Data |
---|---|---|
3d442233 JA |
1 | /* |
2 | * Generic helpers for smp ipi calls | |
3 | * | |
4 | * (C) Jens Axboe <jens.axboe@oracle.com> 2008 | |
3d442233 | 5 | */ |
ca7dfdbb ME |
6 | |
7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
8 | ||
47885016 | 9 | #include <linux/irq_work.h> |
3d442233 | 10 | #include <linux/rcupdate.h> |
59190f42 | 11 | #include <linux/rculist.h> |
641cd4cf | 12 | #include <linux/kernel.h> |
9984de1a | 13 | #include <linux/export.h> |
0b13fda1 IM |
14 | #include <linux/percpu.h> |
15 | #include <linux/init.h> | |
5a0e3ad6 | 16 | #include <linux/gfp.h> |
3d442233 | 17 | #include <linux/smp.h> |
8969a5ed | 18 | #include <linux/cpu.h> |
c6f4459f | 19 | #include <linux/sched.h> |
4c822698 | 20 | #include <linux/sched/idle.h> |
47ae4b05 | 21 | #include <linux/hypervisor.h> |
3d442233 | 22 | |
3bb5d2ee SS |
23 | #include "smpboot.h" |
24 | ||
3d442233 | 25 | enum { |
6e275637 | 26 | CSD_FLAG_LOCK = 0x01, |
8053871d | 27 | CSD_FLAG_SYNCHRONOUS = 0x02, |
3d442233 JA |
28 | }; |
29 | ||
30 | struct call_function_data { | |
966a9671 | 31 | call_single_data_t __percpu *csd; |
0b13fda1 | 32 | cpumask_var_t cpumask; |
3fc5b3b6 | 33 | cpumask_var_t cpumask_ipi; |
3d442233 JA |
34 | }; |
35 | ||
e03bcb68 MM |
36 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct call_function_data, cfd_data); |
37 | ||
6897fc22 | 38 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct llist_head, call_single_queue); |
8969a5ed | 39 | |
8d056c48 SB |
40 | static void flush_smp_call_function_queue(bool warn_cpu_offline); |
41 | ||
31487f83 | 42 | int smpcfd_prepare_cpu(unsigned int cpu) |
8969a5ed | 43 | { |
8969a5ed PZ |
44 | struct call_function_data *cfd = &per_cpu(cfd_data, cpu); |
45 | ||
31487f83 RW |
46 | if (!zalloc_cpumask_var_node(&cfd->cpumask, GFP_KERNEL, |
47 | cpu_to_node(cpu))) | |
48 | return -ENOMEM; | |
3fc5b3b6 AL |
49 | if (!zalloc_cpumask_var_node(&cfd->cpumask_ipi, GFP_KERNEL, |
50 | cpu_to_node(cpu))) { | |
51 | free_cpumask_var(cfd->cpumask); | |
52 | return -ENOMEM; | |
53 | } | |
966a9671 | 54 | cfd->csd = alloc_percpu(call_single_data_t); |
31487f83 | 55 | if (!cfd->csd) { |
8969a5ed | 56 | free_cpumask_var(cfd->cpumask); |
3fc5b3b6 | 57 | free_cpumask_var(cfd->cpumask_ipi); |
31487f83 RW |
58 | return -ENOMEM; |
59 | } | |
60 | ||
61 | return 0; | |
8969a5ed PZ |
62 | } |
63 | ||
31487f83 RW |
64 | int smpcfd_dead_cpu(unsigned int cpu) |
65 | { | |
66 | struct call_function_data *cfd = &per_cpu(cfd_data, cpu); | |
67 | ||
68 | free_cpumask_var(cfd->cpumask); | |
3fc5b3b6 | 69 | free_cpumask_var(cfd->cpumask_ipi); |
31487f83 RW |
70 | free_percpu(cfd->csd); |
71 | return 0; | |
72 | } | |
73 | ||
74 | int smpcfd_dying_cpu(unsigned int cpu) | |
75 | { | |
76 | /* | |
77 | * The IPIs for the smp-call-function callbacks queued by other | |
78 | * CPUs might arrive late, either due to hardware latencies or | |
79 | * because this CPU disabled interrupts (inside stop-machine) | |
80 | * before the IPIs were sent. So flush out any pending callbacks | |
81 | * explicitly (without waiting for the IPIs to arrive), to | |
82 | * ensure that the outgoing CPU doesn't go offline with work | |
83 | * still pending. | |
84 | */ | |
85 | flush_smp_call_function_queue(false); | |
86 | return 0; | |
87 | } | |
8969a5ed | 88 | |
d8ad7d11 | 89 | void __init call_function_init(void) |
3d442233 JA |
90 | { |
91 | int i; | |
92 | ||
6897fc22 CH |
93 | for_each_possible_cpu(i) |
94 | init_llist_head(&per_cpu(call_single_queue, i)); | |
8969a5ed | 95 | |
31487f83 | 96 | smpcfd_prepare_cpu(smp_processor_id()); |
3d442233 JA |
97 | } |
98 | ||
8969a5ed PZ |
99 | /* |
100 | * csd_lock/csd_unlock used to serialize access to per-cpu csd resources | |
101 | * | |
0b13fda1 IM |
102 | * For non-synchronous ipi calls the csd can still be in use by the |
103 | * previous function call. For multi-cpu calls its even more interesting | |
104 | * as we'll have to ensure no other cpu is observing our csd. | |
8969a5ed | 105 | */ |
966a9671 | 106 | static __always_inline void csd_lock_wait(call_single_data_t *csd) |
8969a5ed | 107 | { |
1f03e8d2 | 108 | smp_cond_load_acquire(&csd->flags, !(VAL & CSD_FLAG_LOCK)); |
6e275637 PZ |
109 | } |
110 | ||
966a9671 | 111 | static __always_inline void csd_lock(call_single_data_t *csd) |
6e275637 | 112 | { |
e1d12f32 AM |
113 | csd_lock_wait(csd); |
114 | csd->flags |= CSD_FLAG_LOCK; | |
8969a5ed PZ |
115 | |
116 | /* | |
0b13fda1 IM |
117 | * prevent CPU from reordering the above assignment |
118 | * to ->flags with any subsequent assignments to other | |
966a9671 | 119 | * fields of the specified call_single_data_t structure: |
8969a5ed | 120 | */ |
8053871d | 121 | smp_wmb(); |
8969a5ed PZ |
122 | } |
123 | ||
966a9671 | 124 | static __always_inline void csd_unlock(call_single_data_t *csd) |
8969a5ed | 125 | { |
8053871d | 126 | WARN_ON(!(csd->flags & CSD_FLAG_LOCK)); |
0b13fda1 | 127 | |
8969a5ed | 128 | /* |
0b13fda1 | 129 | * ensure we're all done before releasing data: |
8969a5ed | 130 | */ |
8053871d | 131 | smp_store_release(&csd->flags, 0); |
3d442233 JA |
132 | } |
133 | ||
966a9671 | 134 | static DEFINE_PER_CPU_SHARED_ALIGNED(call_single_data_t, csd_data); |
8b28499a | 135 | |
3d442233 | 136 | /* |
966a9671 | 137 | * Insert a previously allocated call_single_data_t element |
0b13fda1 IM |
138 | * for execution on the given CPU. data must already have |
139 | * ->func, ->info, and ->flags set. | |
3d442233 | 140 | */ |
966a9671 | 141 | static int generic_exec_single(int cpu, call_single_data_t *csd, |
8053871d | 142 | smp_call_func_t func, void *info) |
3d442233 | 143 | { |
8b28499a | 144 | if (cpu == smp_processor_id()) { |
8053871d LT |
145 | unsigned long flags; |
146 | ||
147 | /* | |
148 | * We can unlock early even for the synchronous on-stack case, | |
149 | * since we're doing this from the same CPU.. | |
150 | */ | |
151 | csd_unlock(csd); | |
8b28499a FW |
152 | local_irq_save(flags); |
153 | func(info); | |
154 | local_irq_restore(flags); | |
155 | return 0; | |
156 | } | |
157 | ||
158 | ||
5224b961 LT |
159 | if ((unsigned)cpu >= nr_cpu_ids || !cpu_online(cpu)) { |
160 | csd_unlock(csd); | |
8b28499a | 161 | return -ENXIO; |
5224b961 | 162 | } |
8b28499a | 163 | |
8b28499a FW |
164 | csd->func = func; |
165 | csd->info = info; | |
166 | ||
561920a0 | 167 | /* |
15d0d3b3 NP |
168 | * The list addition should be visible before sending the IPI |
169 | * handler locks the list to pull the entry off it because of | |
170 | * normal cache coherency rules implied by spinlocks. | |
171 | * | |
172 | * If IPIs can go out of order to the cache coherency protocol | |
173 | * in an architecture, sufficient synchronisation should be added | |
174 | * to arch code to make it appear to obey cache coherency WRT | |
0b13fda1 IM |
175 | * locking and barrier primitives. Generic code isn't really |
176 | * equipped to do the right thing... | |
561920a0 | 177 | */ |
6897fc22 | 178 | if (llist_add(&csd->llist, &per_cpu(call_single_queue, cpu))) |
3d442233 JA |
179 | arch_send_call_function_single_ipi(cpu); |
180 | ||
8b28499a | 181 | return 0; |
3d442233 JA |
182 | } |
183 | ||
8d056c48 SB |
184 | /** |
185 | * generic_smp_call_function_single_interrupt - Execute SMP IPI callbacks | |
186 | * | |
187 | * Invoked by arch to handle an IPI for call function single. | |
188 | * Must be called with interrupts disabled. | |
3d442233 JA |
189 | */ |
190 | void generic_smp_call_function_single_interrupt(void) | |
191 | { | |
8d056c48 SB |
192 | flush_smp_call_function_queue(true); |
193 | } | |
194 | ||
195 | /** | |
196 | * flush_smp_call_function_queue - Flush pending smp-call-function callbacks | |
197 | * | |
198 | * @warn_cpu_offline: If set to 'true', warn if callbacks were queued on an | |
199 | * offline CPU. Skip this check if set to 'false'. | |
200 | * | |
201 | * Flush any pending smp-call-function callbacks queued on this CPU. This is | |
202 | * invoked by the generic IPI handler, as well as by a CPU about to go offline, | |
203 | * to ensure that all pending IPI callbacks are run before it goes completely | |
204 | * offline. | |
205 | * | |
206 | * Loop through the call_single_queue and run all the queued callbacks. | |
207 | * Must be called with interrupts disabled. | |
208 | */ | |
209 | static void flush_smp_call_function_queue(bool warn_cpu_offline) | |
210 | { | |
211 | struct llist_head *head; | |
5fd77595 | 212 | struct llist_node *entry; |
966a9671 | 213 | call_single_data_t *csd, *csd_next; |
a219ccf4 SB |
214 | static bool warned; |
215 | ||
8d056c48 SB |
216 | WARN_ON(!irqs_disabled()); |
217 | ||
bb964a92 | 218 | head = this_cpu_ptr(&call_single_queue); |
8d056c48 | 219 | entry = llist_del_all(head); |
a219ccf4 | 220 | entry = llist_reverse_order(entry); |
3d442233 | 221 | |
8d056c48 SB |
222 | /* There shouldn't be any pending callbacks on an offline CPU. */ |
223 | if (unlikely(warn_cpu_offline && !cpu_online(smp_processor_id()) && | |
224 | !warned && !llist_empty(head))) { | |
a219ccf4 SB |
225 | warned = true; |
226 | WARN(1, "IPI on offline CPU %d\n", smp_processor_id()); | |
227 | ||
228 | /* | |
229 | * We don't have to use the _safe() variant here | |
230 | * because we are not invoking the IPI handlers yet. | |
231 | */ | |
232 | llist_for_each_entry(csd, entry, llist) | |
233 | pr_warn("IPI callback %pS sent to offline CPU\n", | |
234 | csd->func); | |
235 | } | |
3d442233 | 236 | |
5fd77595 | 237 | llist_for_each_entry_safe(csd, csd_next, entry, llist) { |
8053871d LT |
238 | smp_call_func_t func = csd->func; |
239 | void *info = csd->info; | |
240 | ||
ddad4f4d | 241 | /* This should be called by handle_IPI in smp.c */ |
32feb52d | 242 | dbg_snapshot_irq(DSS_FLAG_SMP_CALL_FN, func, NULL, 0, DSS_FLAG_IN); |
ddad4f4d | 243 | |
8053871d LT |
244 | /* Do we wait until *after* callback? */ |
245 | if (csd->flags & CSD_FLAG_SYNCHRONOUS) { | |
246 | func(info); | |
247 | csd_unlock(csd); | |
248 | } else { | |
249 | csd_unlock(csd); | |
250 | func(info); | |
251 | } | |
32feb52d HK |
252 | |
253 | dbg_snapshot_irq(DSS_FLAG_SMP_CALL_FN, func, NULL, 0, DSS_FLAG_OUT); | |
3d442233 | 254 | } |
47885016 FW |
255 | |
256 | /* | |
257 | * Handle irq works queued remotely by irq_work_queue_on(). | |
258 | * Smp functions above are typically synchronous so they | |
259 | * better run first since some other CPUs may be busy waiting | |
260 | * for them. | |
261 | */ | |
262 | irq_work_run(); | |
3d442233 JA |
263 | } |
264 | ||
265 | /* | |
266 | * smp_call_function_single - Run a function on a specific CPU | |
267 | * @func: The function to run. This must be fast and non-blocking. | |
268 | * @info: An arbitrary pointer to pass to the function. | |
3d442233 JA |
269 | * @wait: If true, wait until function has completed on other CPUs. |
270 | * | |
72f279b2 | 271 | * Returns 0 on success, else a negative status code. |
3d442233 | 272 | */ |
3a5f65df | 273 | int smp_call_function_single(int cpu, smp_call_func_t func, void *info, |
8691e5a8 | 274 | int wait) |
3d442233 | 275 | { |
966a9671 YH |
276 | call_single_data_t *csd; |
277 | call_single_data_t csd_stack = { | |
278 | .flags = CSD_FLAG_LOCK | CSD_FLAG_SYNCHRONOUS, | |
279 | }; | |
0b13fda1 | 280 | int this_cpu; |
8b28499a | 281 | int err; |
3d442233 | 282 | |
0b13fda1 IM |
283 | /* |
284 | * prevent preemption and reschedule on another processor, | |
285 | * as well as CPU removal | |
286 | */ | |
287 | this_cpu = get_cpu(); | |
288 | ||
269c861b SS |
289 | /* |
290 | * Can deadlock when called with interrupts disabled. | |
291 | * We allow cpu's that are not yet online though, as no one else can | |
292 | * send smp call function interrupt to this cpu and as such deadlocks | |
293 | * can't happen. | |
294 | */ | |
295 | WARN_ON_ONCE(cpu_online(this_cpu) && irqs_disabled() | |
296 | && !oops_in_progress); | |
3d442233 | 297 | |
8053871d LT |
298 | csd = &csd_stack; |
299 | if (!wait) { | |
300 | csd = this_cpu_ptr(&csd_data); | |
301 | csd_lock(csd); | |
302 | } | |
303 | ||
304 | err = generic_exec_single(cpu, csd, func, info); | |
305 | ||
306 | if (wait) | |
307 | csd_lock_wait(csd); | |
3d442233 JA |
308 | |
309 | put_cpu(); | |
0b13fda1 | 310 | |
f73be6de | 311 | return err; |
3d442233 JA |
312 | } |
313 | EXPORT_SYMBOL(smp_call_function_single); | |
314 | ||
d7877c03 | 315 | /** |
c46fff2a FW |
316 | * smp_call_function_single_async(): Run an asynchronous function on a |
317 | * specific CPU. | |
d7877c03 FW |
318 | * @cpu: The CPU to run on. |
319 | * @csd: Pre-allocated and setup data structure | |
d7877c03 | 320 | * |
c46fff2a FW |
321 | * Like smp_call_function_single(), but the call is asynchonous and |
322 | * can thus be done from contexts with disabled interrupts. | |
323 | * | |
324 | * The caller passes his own pre-allocated data structure | |
325 | * (ie: embedded in an object) and is responsible for synchronizing it | |
326 | * such that the IPIs performed on the @csd are strictly serialized. | |
327 | * | |
328 | * NOTE: Be careful, there is unfortunately no current debugging facility to | |
329 | * validate the correctness of this serialization. | |
d7877c03 | 330 | */ |
966a9671 | 331 | int smp_call_function_single_async(int cpu, call_single_data_t *csd) |
d7877c03 FW |
332 | { |
333 | int err = 0; | |
d7877c03 | 334 | |
fce8ad15 | 335 | preempt_disable(); |
8053871d LT |
336 | |
337 | /* We could deadlock if we have to wait here with interrupts disabled! */ | |
338 | if (WARN_ON_ONCE(csd->flags & CSD_FLAG_LOCK)) | |
339 | csd_lock_wait(csd); | |
340 | ||
341 | csd->flags = CSD_FLAG_LOCK; | |
342 | smp_wmb(); | |
343 | ||
344 | err = generic_exec_single(cpu, csd, csd->func, csd->info); | |
fce8ad15 | 345 | preempt_enable(); |
d7877c03 FW |
346 | |
347 | return err; | |
348 | } | |
c46fff2a | 349 | EXPORT_SYMBOL_GPL(smp_call_function_single_async); |
d7877c03 | 350 | |
2ea6dec4 RR |
351 | /* |
352 | * smp_call_function_any - Run a function on any of the given cpus | |
353 | * @mask: The mask of cpus it can run on. | |
354 | * @func: The function to run. This must be fast and non-blocking. | |
355 | * @info: An arbitrary pointer to pass to the function. | |
356 | * @wait: If true, wait until function has completed. | |
357 | * | |
358 | * Returns 0 on success, else a negative status code (if no cpus were online). | |
2ea6dec4 RR |
359 | * |
360 | * Selection preference: | |
361 | * 1) current cpu if in @mask | |
362 | * 2) any cpu of current node if in @mask | |
363 | * 3) any other online cpu in @mask | |
364 | */ | |
365 | int smp_call_function_any(const struct cpumask *mask, | |
3a5f65df | 366 | smp_call_func_t func, void *info, int wait) |
2ea6dec4 RR |
367 | { |
368 | unsigned int cpu; | |
369 | const struct cpumask *nodemask; | |
370 | int ret; | |
371 | ||
372 | /* Try for same CPU (cheapest) */ | |
373 | cpu = get_cpu(); | |
374 | if (cpumask_test_cpu(cpu, mask)) | |
375 | goto call; | |
376 | ||
377 | /* Try for same node. */ | |
af2422c4 | 378 | nodemask = cpumask_of_node(cpu_to_node(cpu)); |
2ea6dec4 RR |
379 | for (cpu = cpumask_first_and(nodemask, mask); cpu < nr_cpu_ids; |
380 | cpu = cpumask_next_and(cpu, nodemask, mask)) { | |
381 | if (cpu_online(cpu)) | |
382 | goto call; | |
383 | } | |
384 | ||
385 | /* Any online will do: smp_call_function_single handles nr_cpu_ids. */ | |
386 | cpu = cpumask_any_and(mask, cpu_online_mask); | |
387 | call: | |
388 | ret = smp_call_function_single(cpu, func, info, wait); | |
389 | put_cpu(); | |
390 | return ret; | |
391 | } | |
392 | EXPORT_SYMBOL_GPL(smp_call_function_any); | |
393 | ||
3d442233 | 394 | /** |
54b11e6d RR |
395 | * smp_call_function_many(): Run a function on a set of other CPUs. |
396 | * @mask: The set of cpus to run on (only runs on online subset). | |
3d442233 JA |
397 | * @func: The function to run. This must be fast and non-blocking. |
398 | * @info: An arbitrary pointer to pass to the function. | |
0b13fda1 IM |
399 | * @wait: If true, wait (atomically) until function has completed |
400 | * on other CPUs. | |
3d442233 | 401 | * |
72f279b2 | 402 | * If @wait is true, then returns once @func has returned. |
3d442233 JA |
403 | * |
404 | * You must not call this function with disabled interrupts or from a | |
405 | * hardware interrupt handler or from a bottom half handler. Preemption | |
406 | * must be disabled when calling this function. | |
407 | */ | |
54b11e6d | 408 | void smp_call_function_many(const struct cpumask *mask, |
3a5f65df | 409 | smp_call_func_t func, void *info, bool wait) |
3d442233 | 410 | { |
e1d12f32 | 411 | struct call_function_data *cfd; |
9a46ad6d | 412 | int cpu, next_cpu, this_cpu = smp_processor_id(); |
3d442233 | 413 | |
269c861b SS |
414 | /* |
415 | * Can deadlock when called with interrupts disabled. | |
416 | * We allow cpu's that are not yet online though, as no one else can | |
417 | * send smp call function interrupt to this cpu and as such deadlocks | |
418 | * can't happen. | |
419 | */ | |
420 | WARN_ON_ONCE(cpu_online(this_cpu) && irqs_disabled() | |
bd924e8c | 421 | && !oops_in_progress && !early_boot_irqs_disabled); |
3d442233 | 422 | |
723aae25 | 423 | /* Try to fastpath. So, what's a CPU they want? Ignoring this one. */ |
54b11e6d | 424 | cpu = cpumask_first_and(mask, cpu_online_mask); |
0b13fda1 | 425 | if (cpu == this_cpu) |
54b11e6d | 426 | cpu = cpumask_next_and(cpu, mask, cpu_online_mask); |
0b13fda1 | 427 | |
54b11e6d RR |
428 | /* No online cpus? We're done. */ |
429 | if (cpu >= nr_cpu_ids) | |
430 | return; | |
431 | ||
432 | /* Do we have another CPU which isn't us? */ | |
433 | next_cpu = cpumask_next_and(cpu, mask, cpu_online_mask); | |
0b13fda1 | 434 | if (next_cpu == this_cpu) |
54b11e6d RR |
435 | next_cpu = cpumask_next_and(next_cpu, mask, cpu_online_mask); |
436 | ||
437 | /* Fastpath: do that cpu by itself. */ | |
438 | if (next_cpu >= nr_cpu_ids) { | |
439 | smp_call_function_single(cpu, func, info, wait); | |
440 | return; | |
3d442233 JA |
441 | } |
442 | ||
bb964a92 | 443 | cfd = this_cpu_ptr(&cfd_data); |
45a57919 | 444 | |
e1d12f32 | 445 | cpumask_and(cfd->cpumask, mask, cpu_online_mask); |
6c8557bd | 446 | __cpumask_clear_cpu(this_cpu, cfd->cpumask); |
723aae25 MM |
447 | |
448 | /* Some callers race with other cpus changing the passed mask */ | |
e1d12f32 | 449 | if (unlikely(!cpumask_weight(cfd->cpumask))) |
723aae25 | 450 | return; |
3d442233 | 451 | |
3fc5b3b6 | 452 | cpumask_clear(cfd->cpumask_ipi); |
e1d12f32 | 453 | for_each_cpu(cpu, cfd->cpumask) { |
966a9671 | 454 | call_single_data_t *csd = per_cpu_ptr(cfd->csd, cpu); |
9a46ad6d SL |
455 | |
456 | csd_lock(csd); | |
8053871d LT |
457 | if (wait) |
458 | csd->flags |= CSD_FLAG_SYNCHRONOUS; | |
9a46ad6d SL |
459 | csd->func = func; |
460 | csd->info = info; | |
3fc5b3b6 | 461 | if (llist_add(&csd->llist, &per_cpu(call_single_queue, cpu))) |
6c8557bd | 462 | __cpumask_set_cpu(cpu, cfd->cpumask_ipi); |
9a46ad6d | 463 | } |
561920a0 | 464 | |
3d442233 | 465 | /* Send a message to all CPUs in the map */ |
3fc5b3b6 | 466 | arch_send_call_function_ipi_mask(cfd->cpumask_ipi); |
3d442233 | 467 | |
9a46ad6d | 468 | if (wait) { |
e1d12f32 | 469 | for_each_cpu(cpu, cfd->cpumask) { |
966a9671 | 470 | call_single_data_t *csd; |
e1d12f32 AM |
471 | |
472 | csd = per_cpu_ptr(cfd->csd, cpu); | |
9a46ad6d SL |
473 | csd_lock_wait(csd); |
474 | } | |
475 | } | |
3d442233 | 476 | } |
54b11e6d | 477 | EXPORT_SYMBOL(smp_call_function_many); |
3d442233 JA |
478 | |
479 | /** | |
480 | * smp_call_function(): Run a function on all other CPUs. | |
481 | * @func: The function to run. This must be fast and non-blocking. | |
482 | * @info: An arbitrary pointer to pass to the function. | |
0b13fda1 IM |
483 | * @wait: If true, wait (atomically) until function has completed |
484 | * on other CPUs. | |
3d442233 | 485 | * |
54b11e6d | 486 | * Returns 0. |
3d442233 JA |
487 | * |
488 | * If @wait is true, then returns once @func has returned; otherwise | |
72f279b2 | 489 | * it returns just before the target cpu calls @func. |
3d442233 JA |
490 | * |
491 | * You must not call this function with disabled interrupts or from a | |
492 | * hardware interrupt handler or from a bottom half handler. | |
493 | */ | |
3a5f65df | 494 | int smp_call_function(smp_call_func_t func, void *info, int wait) |
3d442233 | 495 | { |
3d442233 | 496 | preempt_disable(); |
54b11e6d | 497 | smp_call_function_many(cpu_online_mask, func, info, wait); |
3d442233 | 498 | preempt_enable(); |
0b13fda1 | 499 | |
54b11e6d | 500 | return 0; |
3d442233 JA |
501 | } |
502 | EXPORT_SYMBOL(smp_call_function); | |
351f8f8e | 503 | |
34db18a0 AW |
504 | /* Setup configured maximum number of CPUs to activate */ |
505 | unsigned int setup_max_cpus = NR_CPUS; | |
506 | EXPORT_SYMBOL(setup_max_cpus); | |
4aab5595 YL |
507 | struct cpumask early_cpu_mask; |
508 | EXPORT_SYMBOL(early_cpu_mask); | |
34db18a0 AW |
509 | |
510 | ||
511 | /* | |
512 | * Setup routine for controlling SMP activation | |
513 | * | |
514 | * Command-line option of "nosmp" or "maxcpus=0" will disable SMP | |
515 | * activation entirely (the MPS table probe still happens, though). | |
516 | * | |
517 | * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer | |
518 | * greater than 0, limits the maximum number of CPUs activated in | |
519 | * SMP mode to <NUM>. | |
520 | */ | |
521 | ||
522 | void __weak arch_disable_smp_support(void) { } | |
523 | ||
524 | static int __init nosmp(char *str) | |
525 | { | |
526 | setup_max_cpus = 0; | |
527 | arch_disable_smp_support(); | |
528 | ||
529 | return 0; | |
530 | } | |
531 | ||
532 | early_param("nosmp", nosmp); | |
533 | ||
534 | /* this is hard limit */ | |
535 | static int __init nrcpus(char *str) | |
536 | { | |
537 | int nr_cpus; | |
538 | ||
539 | get_option(&str, &nr_cpus); | |
540 | if (nr_cpus > 0 && nr_cpus < nr_cpu_ids) | |
541 | nr_cpu_ids = nr_cpus; | |
542 | ||
543 | return 0; | |
544 | } | |
545 | ||
546 | early_param("nr_cpus", nrcpus); | |
547 | ||
548 | static int __init maxcpus(char *str) | |
549 | { | |
550 | get_option(&str, &setup_max_cpus); | |
551 | if (setup_max_cpus == 0) | |
552 | arch_disable_smp_support(); | |
553 | ||
554 | return 0; | |
555 | } | |
556 | ||
557 | early_param("maxcpus", maxcpus); | |
558 | ||
559 | /* Setup number of possible processor ids */ | |
9b130ad5 | 560 | unsigned int nr_cpu_ids __read_mostly = NR_CPUS; |
34db18a0 AW |
561 | EXPORT_SYMBOL(nr_cpu_ids); |
562 | ||
563 | /* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */ | |
564 | void __init setup_nr_cpu_ids(void) | |
565 | { | |
566 | nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1; | |
567 | } | |
568 | ||
569 | /* Called by boot processor to activate the rest. */ | |
570 | void __init smp_init(void) | |
571 | { | |
92b23278 | 572 | int num_nodes, num_cpus; |
34db18a0 AW |
573 | unsigned int cpu; |
574 | ||
3bb5d2ee | 575 | idle_threads_init(); |
4cb28ced | 576 | cpuhp_threads_init(); |
3bb5d2ee | 577 | |
51111dce ME |
578 | pr_info("Bringing up secondary CPUs ...\n"); |
579 | ||
4aab5595 YL |
580 | cpumask_clear(&early_cpu_mask); |
581 | cpumask_set_cpu(0, &early_cpu_mask); | |
34db18a0 AW |
582 | /* FIXME: This should be done in userspace --RR */ |
583 | for_each_present_cpu(cpu) { | |
584 | if (num_online_cpus() >= setup_max_cpus) | |
585 | break; | |
4aab5595 | 586 | if (!cpu_online(cpu)) { |
34db18a0 | 587 | cpu_up(cpu); |
4aab5595 YL |
588 | cpumask_set_cpu(cpu, &early_cpu_mask); |
589 | } | |
34db18a0 AW |
590 | } |
591 | ||
92b23278 ME |
592 | num_nodes = num_online_nodes(); |
593 | num_cpus = num_online_cpus(); | |
594 | pr_info("Brought up %d node%s, %d CPU%s\n", | |
595 | num_nodes, (num_nodes > 1 ? "s" : ""), | |
596 | num_cpus, (num_cpus > 1 ? "s" : "")); | |
597 | ||
34db18a0 | 598 | /* Any cleanup work */ |
34db18a0 AW |
599 | smp_cpus_done(setup_max_cpus); |
600 | } | |
601 | ||
351f8f8e | 602 | /* |
bd924e8c TH |
603 | * Call a function on all processors. May be used during early boot while |
604 | * early_boot_irqs_disabled is set. Use local_irq_save/restore() instead | |
605 | * of local_irq_disable/enable(). | |
351f8f8e AW |
606 | */ |
607 | int on_each_cpu(void (*func) (void *info), void *info, int wait) | |
608 | { | |
bd924e8c | 609 | unsigned long flags; |
351f8f8e AW |
610 | int ret = 0; |
611 | ||
612 | preempt_disable(); | |
613 | ret = smp_call_function(func, info, wait); | |
bd924e8c | 614 | local_irq_save(flags); |
351f8f8e | 615 | func(info); |
bd924e8c | 616 | local_irq_restore(flags); |
351f8f8e AW |
617 | preempt_enable(); |
618 | return ret; | |
619 | } | |
620 | EXPORT_SYMBOL(on_each_cpu); | |
3fc498f1 GBY |
621 | |
622 | /** | |
623 | * on_each_cpu_mask(): Run a function on processors specified by | |
624 | * cpumask, which may include the local processor. | |
625 | * @mask: The set of cpus to run on (only runs on online subset). | |
626 | * @func: The function to run. This must be fast and non-blocking. | |
627 | * @info: An arbitrary pointer to pass to the function. | |
628 | * @wait: If true, wait (atomically) until function has completed | |
629 | * on other CPUs. | |
630 | * | |
631 | * If @wait is true, then returns once @func has returned. | |
632 | * | |
202da400 DD |
633 | * You must not call this function with disabled interrupts or from a |
634 | * hardware interrupt handler or from a bottom half handler. The | |
635 | * exception is that it may be used during early boot while | |
636 | * early_boot_irqs_disabled is set. | |
3fc498f1 GBY |
637 | */ |
638 | void on_each_cpu_mask(const struct cpumask *mask, smp_call_func_t func, | |
639 | void *info, bool wait) | |
640 | { | |
641 | int cpu = get_cpu(); | |
642 | ||
643 | smp_call_function_many(mask, func, info, wait); | |
644 | if (cpumask_test_cpu(cpu, mask)) { | |
202da400 DD |
645 | unsigned long flags; |
646 | local_irq_save(flags); | |
3fc498f1 | 647 | func(info); |
202da400 | 648 | local_irq_restore(flags); |
3fc498f1 GBY |
649 | } |
650 | put_cpu(); | |
651 | } | |
652 | EXPORT_SYMBOL(on_each_cpu_mask); | |
b3a7e98e GBY |
653 | |
654 | /* | |
655 | * on_each_cpu_cond(): Call a function on each processor for which | |
656 | * the supplied function cond_func returns true, optionally waiting | |
657 | * for all the required CPUs to finish. This may include the local | |
658 | * processor. | |
659 | * @cond_func: A callback function that is passed a cpu id and | |
660 | * the the info parameter. The function is called | |
661 | * with preemption disabled. The function should | |
662 | * return a blooean value indicating whether to IPI | |
663 | * the specified CPU. | |
664 | * @func: The function to run on all applicable CPUs. | |
665 | * This must be fast and non-blocking. | |
666 | * @info: An arbitrary pointer to pass to both functions. | |
667 | * @wait: If true, wait (atomically) until function has | |
668 | * completed on other CPUs. | |
669 | * @gfp_flags: GFP flags to use when allocating the cpumask | |
670 | * used internally by the function. | |
671 | * | |
672 | * The function might sleep if the GFP flags indicates a non | |
673 | * atomic allocation is allowed. | |
674 | * | |
675 | * Preemption is disabled to protect against CPUs going offline but not online. | |
676 | * CPUs going online during the call will not be seen or sent an IPI. | |
677 | * | |
678 | * You must not call this function with disabled interrupts or | |
679 | * from a hardware interrupt handler or from a bottom half handler. | |
680 | */ | |
681 | void on_each_cpu_cond(bool (*cond_func)(int cpu, void *info), | |
682 | smp_call_func_t func, void *info, bool wait, | |
683 | gfp_t gfp_flags) | |
684 | { | |
685 | cpumask_var_t cpus; | |
686 | int cpu, ret; | |
687 | ||
d0164adc | 688 | might_sleep_if(gfpflags_allow_blocking(gfp_flags)); |
b3a7e98e GBY |
689 | |
690 | if (likely(zalloc_cpumask_var(&cpus, (gfp_flags|__GFP_NOWARN)))) { | |
691 | preempt_disable(); | |
692 | for_each_online_cpu(cpu) | |
693 | if (cond_func(cpu, info)) | |
694 | cpumask_set_cpu(cpu, cpus); | |
695 | on_each_cpu_mask(cpus, func, info, wait); | |
696 | preempt_enable(); | |
697 | free_cpumask_var(cpus); | |
698 | } else { | |
699 | /* | |
700 | * No free cpumask, bother. No matter, we'll | |
701 | * just have to IPI them one by one. | |
702 | */ | |
703 | preempt_disable(); | |
704 | for_each_online_cpu(cpu) | |
705 | if (cond_func(cpu, info)) { | |
706 | ret = smp_call_function_single(cpu, func, | |
707 | info, wait); | |
618fde87 | 708 | WARN_ON_ONCE(ret); |
b3a7e98e GBY |
709 | } |
710 | preempt_enable(); | |
711 | } | |
712 | } | |
713 | EXPORT_SYMBOL(on_each_cpu_cond); | |
f37f435f TG |
714 | |
715 | static void do_nothing(void *unused) | |
716 | { | |
717 | } | |
718 | ||
719 | /** | |
720 | * kick_all_cpus_sync - Force all cpus out of idle | |
721 | * | |
722 | * Used to synchronize the update of pm_idle function pointer. It's | |
723 | * called after the pointer is updated and returns after the dummy | |
724 | * callback function has been executed on all cpus. The execution of | |
725 | * the function can only happen on the remote cpus after they have | |
726 | * left the idle function which had been called via pm_idle function | |
727 | * pointer. So it's guaranteed that nothing uses the previous pointer | |
728 | * anymore. | |
729 | */ | |
730 | void kick_all_cpus_sync(void) | |
731 | { | |
732 | /* Make sure the change is visible before we kick the cpus */ | |
733 | smp_mb(); | |
734 | smp_call_function(do_nothing, NULL, 1); | |
735 | } | |
736 | EXPORT_SYMBOL_GPL(kick_all_cpus_sync); | |
c6f4459f CL |
737 | |
738 | /** | |
739 | * wake_up_all_idle_cpus - break all cpus out of idle | |
740 | * wake_up_all_idle_cpus try to break all cpus which is in idle state even | |
741 | * including idle polling cpus, for non-idle cpus, we will do nothing | |
742 | * for them. | |
743 | */ | |
744 | void wake_up_all_idle_cpus(void) | |
745 | { | |
746 | int cpu; | |
747 | ||
748 | preempt_disable(); | |
749 | for_each_online_cpu(cpu) { | |
750 | if (cpu == smp_processor_id()) | |
751 | continue; | |
752 | ||
753 | wake_up_if_idle(cpu); | |
754 | } | |
755 | preempt_enable(); | |
756 | } | |
757 | EXPORT_SYMBOL_GPL(wake_up_all_idle_cpus); | |
df8ce9d7 JG |
758 | |
759 | /** | |
760 | * smp_call_on_cpu - Call a function on a specific cpu | |
761 | * | |
762 | * Used to call a function on a specific cpu and wait for it to return. | |
763 | * Optionally make sure the call is done on a specified physical cpu via vcpu | |
764 | * pinning in order to support virtualized environments. | |
765 | */ | |
766 | struct smp_call_on_cpu_struct { | |
767 | struct work_struct work; | |
768 | struct completion done; | |
769 | int (*func)(void *); | |
770 | void *data; | |
771 | int ret; | |
772 | int cpu; | |
773 | }; | |
774 | ||
775 | static void smp_call_on_cpu_callback(struct work_struct *work) | |
776 | { | |
777 | struct smp_call_on_cpu_struct *sscs; | |
778 | ||
779 | sscs = container_of(work, struct smp_call_on_cpu_struct, work); | |
780 | if (sscs->cpu >= 0) | |
781 | hypervisor_pin_vcpu(sscs->cpu); | |
782 | sscs->ret = sscs->func(sscs->data); | |
783 | if (sscs->cpu >= 0) | |
784 | hypervisor_pin_vcpu(-1); | |
785 | ||
786 | complete(&sscs->done); | |
787 | } | |
788 | ||
789 | int smp_call_on_cpu(unsigned int cpu, int (*func)(void *), void *par, bool phys) | |
790 | { | |
791 | struct smp_call_on_cpu_struct sscs = { | |
df8ce9d7 JG |
792 | .done = COMPLETION_INITIALIZER_ONSTACK(sscs.done), |
793 | .func = func, | |
794 | .data = par, | |
795 | .cpu = phys ? cpu : -1, | |
796 | }; | |
797 | ||
8db54949 PZ |
798 | INIT_WORK_ONSTACK(&sscs.work, smp_call_on_cpu_callback); |
799 | ||
df8ce9d7 JG |
800 | if (cpu >= nr_cpu_ids || !cpu_online(cpu)) |
801 | return -ENXIO; | |
802 | ||
803 | queue_work_on(cpu, system_wq, &sscs.work); | |
804 | wait_for_completion(&sscs.done); | |
805 | ||
806 | return sscs.ret; | |
807 | } | |
808 | EXPORT_SYMBOL_GPL(smp_call_on_cpu); |