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