1 /* CPU control. 1 /* CPU control. 2 * (C) 2001, 2002, 2003, 2004 Rusty Russell 2 * (C) 2001, 2002, 2003, 2004 Rusty Russell 3 * 3 * 4 * This code is licenced under the GPL. 4 * This code is licenced under the GPL. 5 */ 5 */ 6 #include <linux/sched/mm.h> << 7 #include <linux/proc_fs.h> 6 #include <linux/proc_fs.h> 8 #include <linux/smp.h> 7 #include <linux/smp.h> 9 #include <linux/init.h> 8 #include <linux/init.h> 10 #include <linux/notifier.h> 9 #include <linux/notifier.h> 11 #include <linux/sched/signal.h> 10 #include <linux/sched/signal.h> 12 #include <linux/sched/hotplug.h> 11 #include <linux/sched/hotplug.h> 13 #include <linux/sched/isolation.h> << 14 #include <linux/sched/task.h> 12 #include <linux/sched/task.h> 15 #include <linux/sched/smt.h> << 16 #include <linux/unistd.h> 13 #include <linux/unistd.h> 17 #include <linux/cpu.h> 14 #include <linux/cpu.h> 18 #include <linux/oom.h> 15 #include <linux/oom.h> 19 #include <linux/rcupdate.h> 16 #include <linux/rcupdate.h> 20 #include <linux/delay.h> << 21 #include <linux/export.h> 17 #include <linux/export.h> 22 #include <linux/bug.h> 18 #include <linux/bug.h> 23 #include <linux/kthread.h> 19 #include <linux/kthread.h> 24 #include <linux/stop_machine.h> 20 #include <linux/stop_machine.h> 25 #include <linux/mutex.h> 21 #include <linux/mutex.h> 26 #include <linux/gfp.h> 22 #include <linux/gfp.h> 27 #include <linux/suspend.h> 23 #include <linux/suspend.h> 28 #include <linux/lockdep.h> 24 #include <linux/lockdep.h> 29 #include <linux/tick.h> 25 #include <linux/tick.h> 30 #include <linux/irq.h> 26 #include <linux/irq.h> 31 #include <linux/nmi.h> << 32 #include <linux/smpboot.h> 27 #include <linux/smpboot.h> 33 #include <linux/relay.h> 28 #include <linux/relay.h> 34 #include <linux/slab.h> 29 #include <linux/slab.h> 35 #include <linux/scs.h> << 36 #include <linux/percpu-rwsem.h> << 37 #include <linux/cpuset.h> << 38 #include <linux/random.h> << 39 #include <linux/cc_platform.h> << 40 30 41 #include <trace/events/power.h> 31 #include <trace/events/power.h> 42 #define CREATE_TRACE_POINTS 32 #define CREATE_TRACE_POINTS 43 #include <trace/events/cpuhp.h> 33 #include <trace/events/cpuhp.h> 44 34 45 #include "smpboot.h" 35 #include "smpboot.h" 46 36 47 /** 37 /** 48 * struct cpuhp_cpu_state - Per cpu hotplug st !! 38 * cpuhp_cpu_state - Per cpu hotplug state storage 49 * @state: The current cpu state 39 * @state: The current cpu state 50 * @target: The target state 40 * @target: The target state 51 * @fail: Current CPU hotplug callback s << 52 * @thread: Pointer to the hotplug thread 41 * @thread: Pointer to the hotplug thread 53 * @should_run: Thread should execute 42 * @should_run: Thread should execute 54 * @rollback: Perform a rollback 43 * @rollback: Perform a rollback 55 * @single: Single callback invocation 44 * @single: Single callback invocation 56 * @bringup: Single callback bringup or tea 45 * @bringup: Single callback bringup or teardown selector 57 * @node: Remote CPU node; for multi-ins << 58 * single entry callback for inst << 59 * @last: For multi-instance rollback, r << 60 * @cb_state: The state for a single callbac 46 * @cb_state: The state for a single callback (install/uninstall) 61 * @result: Result of the operation 47 * @result: Result of the operation 62 * @ap_sync_state: State for AP synchroni !! 48 * @done: Signal completion to the issuer of the task 63 * @done_up: Signal completion to the issue << 64 * @done_down: Signal completion to the issue << 65 */ 49 */ 66 struct cpuhp_cpu_state { 50 struct cpuhp_cpu_state { 67 enum cpuhp_state state; 51 enum cpuhp_state state; 68 enum cpuhp_state target; 52 enum cpuhp_state target; 69 enum cpuhp_state fail; << 70 #ifdef CONFIG_SMP 53 #ifdef CONFIG_SMP 71 struct task_struct *thread; 54 struct task_struct *thread; 72 bool should_run; 55 bool should_run; 73 bool rollback; 56 bool rollback; 74 bool single; 57 bool single; 75 bool bringup; 58 bool bringup; 76 struct hlist_node *node; 59 struct hlist_node *node; 77 struct hlist_node *last; << 78 enum cpuhp_state cb_state; 60 enum cpuhp_state cb_state; 79 int result; 61 int result; 80 atomic_t ap_sync_state; !! 62 struct completion done; 81 struct completion done_up; << 82 struct completion done_down; << 83 #endif 63 #endif 84 }; 64 }; 85 65 86 static DEFINE_PER_CPU(struct cpuhp_cpu_state, !! 66 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); 87 .fail = CPUHP_INVALID, << 88 }; << 89 << 90 #ifdef CONFIG_SMP << 91 cpumask_t cpus_booted_once_mask; << 92 #endif << 93 << 94 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_ << 95 static struct lockdep_map cpuhp_state_up_map = << 96 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-u << 97 static struct lockdep_map cpuhp_state_down_map << 98 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-d << 99 << 100 << 101 static inline void cpuhp_lock_acquire(bool bri << 102 { << 103 lock_map_acquire(bringup ? &cpuhp_stat << 104 } << 105 << 106 static inline void cpuhp_lock_release(bool bri << 107 { << 108 lock_map_release(bringup ? &cpuhp_stat << 109 } << 110 #else << 111 << 112 static inline void cpuhp_lock_acquire(bool bri << 113 static inline void cpuhp_lock_release(bool bri << 114 << 115 #endif << 116 67 117 /** 68 /** 118 * struct cpuhp_step - Hotplug state machine s !! 69 * cpuhp_step - Hotplug state machine step 119 * @name: Name of the step 70 * @name: Name of the step 120 * @startup: Startup function of the step 71 * @startup: Startup function of the step 121 * @teardown: Teardown function of the step 72 * @teardown: Teardown function of the step >> 73 * @skip_onerr: Do not invoke the functions on error rollback >> 74 * Will go away once the notifiers are gone 122 * @cant_stop: Bringup/teardown can't be stop 75 * @cant_stop: Bringup/teardown can't be stopped at this step 123 * @multi_instance: State has multiple ins << 124 */ 76 */ 125 struct cpuhp_step { 77 struct cpuhp_step { 126 const char *name; 78 const char *name; 127 union { 79 union { 128 int (*single)(unsi 80 int (*single)(unsigned int cpu); 129 int (*multi)(unsig 81 int (*multi)(unsigned int cpu, 130 struc 82 struct hlist_node *node); 131 } startup; 83 } startup; 132 union { 84 union { 133 int (*single)(unsi 85 int (*single)(unsigned int cpu); 134 int (*multi)(unsig 86 int (*multi)(unsigned int cpu, 135 struc 87 struct hlist_node *node); 136 } teardown; 88 } teardown; 137 /* private: */ << 138 struct hlist_head list; 89 struct hlist_head list; 139 /* public: */ !! 90 bool skip_onerr; 140 bool cant_stop; 91 bool cant_stop; 141 bool multi_instance 92 bool multi_instance; 142 }; 93 }; 143 94 144 static DEFINE_MUTEX(cpuhp_state_mutex); 95 static DEFINE_MUTEX(cpuhp_state_mutex); 145 static struct cpuhp_step cpuhp_hp_states[]; !! 96 static struct cpuhp_step cpuhp_bp_states[]; >> 97 static struct cpuhp_step cpuhp_ap_states[]; 146 98 147 static struct cpuhp_step *cpuhp_get_step(enum !! 99 static bool cpuhp_is_ap_state(enum cpuhp_state state) 148 { 100 { 149 return cpuhp_hp_states + state; !! 101 /* >> 102 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation >> 103 * purposes as that state is handled explicitly in cpu_down. >> 104 */ >> 105 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; 150 } 106 } 151 107 152 static bool cpuhp_step_empty(bool bringup, str !! 108 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) 153 { 109 { 154 return bringup ? !step->startup.single !! 110 struct cpuhp_step *sp; >> 111 >> 112 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; >> 113 return sp + state; 155 } 114 } 156 115 157 /** 116 /** 158 * cpuhp_invoke_callback - Invoke the callback !! 117 * cpuhp_invoke_callback _ Invoke the callbacks for a given state 159 * @cpu: The cpu for which the callback 118 * @cpu: The cpu for which the callback should be invoked 160 * @state: The state to do callbacks for !! 119 * @step: The step in the state machine 161 * @bringup: True if the bringup callback s 120 * @bringup: True if the bringup callback should be invoked 162 * @node: For multi-instance, do a singl << 163 * @lastp: For multi-instance rollback, r << 164 * 121 * 165 * Called from cpu hotplug and from the state 122 * Called from cpu hotplug and from the state register machinery. 166 * << 167 * Return: %0 on success or a negative errno c << 168 */ 123 */ 169 static int cpuhp_invoke_callback(unsigned int 124 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, 170 bool bringup, !! 125 bool bringup, struct hlist_node *node) 171 struct hlist_ << 172 { 126 { 173 struct cpuhp_cpu_state *st = per_cpu_p 127 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 174 struct cpuhp_step *step = cpuhp_get_st 128 struct cpuhp_step *step = cpuhp_get_step(state); 175 int (*cbm)(unsigned int cpu, struct hl 129 int (*cbm)(unsigned int cpu, struct hlist_node *node); 176 int (*cb)(unsigned int cpu); 130 int (*cb)(unsigned int cpu); 177 int ret, cnt; 131 int ret, cnt; 178 132 179 if (st->fail == state) { << 180 st->fail = CPUHP_INVALID; << 181 return -EAGAIN; << 182 } << 183 << 184 if (cpuhp_step_empty(bringup, step)) { << 185 WARN_ON_ONCE(1); << 186 return 0; << 187 } << 188 << 189 if (!step->multi_instance) { 133 if (!step->multi_instance) { 190 WARN_ON_ONCE(lastp && *lastp); << 191 cb = bringup ? step->startup.s 134 cb = bringup ? step->startup.single : step->teardown.single; 192 !! 135 if (!cb) >> 136 return 0; 193 trace_cpuhp_enter(cpu, st->tar 137 trace_cpuhp_enter(cpu, st->target, state, cb); 194 ret = cb(cpu); 138 ret = cb(cpu); 195 trace_cpuhp_exit(cpu, st->stat 139 trace_cpuhp_exit(cpu, st->state, state, ret); 196 return ret; 140 return ret; 197 } 141 } 198 cbm = bringup ? step->startup.multi : 142 cbm = bringup ? step->startup.multi : step->teardown.multi; >> 143 if (!cbm) >> 144 return 0; 199 145 200 /* Single invocation for instance add/ 146 /* Single invocation for instance add/remove */ 201 if (node) { 147 if (node) { 202 WARN_ON_ONCE(lastp && *lastp); << 203 trace_cpuhp_multi_enter(cpu, s 148 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); 204 ret = cbm(cpu, node); 149 ret = cbm(cpu, node); 205 trace_cpuhp_exit(cpu, st->stat 150 trace_cpuhp_exit(cpu, st->state, state, ret); 206 return ret; 151 return ret; 207 } 152 } 208 153 209 /* State transition. Invoke on all ins 154 /* State transition. Invoke on all instances */ 210 cnt = 0; 155 cnt = 0; 211 hlist_for_each(node, &step->list) { 156 hlist_for_each(node, &step->list) { 212 if (lastp && node == *lastp) << 213 break; << 214 << 215 trace_cpuhp_multi_enter(cpu, s 157 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); 216 ret = cbm(cpu, node); 158 ret = cbm(cpu, node); 217 trace_cpuhp_exit(cpu, st->stat 159 trace_cpuhp_exit(cpu, st->state, state, ret); 218 if (ret) { !! 160 if (ret) 219 if (!lastp) !! 161 goto err; 220 goto err; << 221 << 222 *lastp = node; << 223 return ret; << 224 } << 225 cnt++; 162 cnt++; 226 } 163 } 227 if (lastp) << 228 *lastp = NULL; << 229 return 0; 164 return 0; 230 err: 165 err: 231 /* Rollback the instances if one faile 166 /* Rollback the instances if one failed */ 232 cbm = !bringup ? step->startup.multi : 167 cbm = !bringup ? step->startup.multi : step->teardown.multi; 233 if (!cbm) 168 if (!cbm) 234 return ret; 169 return ret; 235 170 236 hlist_for_each(node, &step->list) { 171 hlist_for_each(node, &step->list) { 237 if (!cnt--) 172 if (!cnt--) 238 break; 173 break; 239 !! 174 cbm(cpu, node); 240 trace_cpuhp_multi_enter(cpu, s << 241 ret = cbm(cpu, node); << 242 trace_cpuhp_exit(cpu, st->stat << 243 /* << 244 * Rollback must not fail, << 245 */ << 246 WARN_ON_ONCE(ret); << 247 } 175 } 248 return ret; 176 return ret; 249 } 177 } 250 178 251 #ifdef CONFIG_SMP 179 #ifdef CONFIG_SMP 252 static bool cpuhp_is_ap_state(enum cpuhp_state << 253 { << 254 /* << 255 * The extra check for CPUHP_TEARDOWN_ << 256 * purposes as that state is handled e << 257 */ << 258 return state > CPUHP_BRINGUP_CPU && st << 259 } << 260 << 261 static inline void wait_for_ap_thread(struct c << 262 { << 263 struct completion *done = bringup ? &s << 264 wait_for_completion(done); << 265 } << 266 << 267 static inline void complete_ap_thread(struct c << 268 { << 269 struct completion *done = bringup ? &s << 270 complete(done); << 271 } << 272 << 273 /* << 274 * The former STARTING/DYING states, ran with << 275 */ << 276 static bool cpuhp_is_atomic_state(enum cpuhp_s << 277 { << 278 return CPUHP_AP_IDLE_DEAD <= state && << 279 } << 280 << 281 /* Synchronization state management */ << 282 enum cpuhp_sync_state { << 283 SYNC_STATE_DEAD, << 284 SYNC_STATE_KICKED, << 285 SYNC_STATE_SHOULD_DIE, << 286 SYNC_STATE_ALIVE, << 287 SYNC_STATE_SHOULD_ONLINE, << 288 SYNC_STATE_ONLINE, << 289 }; << 290 << 291 #ifdef CONFIG_HOTPLUG_CORE_SYNC << 292 /** << 293 * cpuhp_ap_update_sync_state - Update synchro << 294 * @state: The synchronization state to s << 295 * << 296 * No synchronization point. Just update of th << 297 * a full barrier so that the AP changes are v << 298 */ << 299 static inline void cpuhp_ap_update_sync_state( << 300 { << 301 atomic_t *st = this_cpu_ptr(&cpuhp_sta << 302 << 303 (void)atomic_xchg(st, state); << 304 } << 305 << 306 void __weak arch_cpuhp_sync_state_poll(void) { << 307 << 308 static bool cpuhp_wait_for_sync_state(unsigned << 309 enum cpu << 310 { << 311 atomic_t *st = per_cpu_ptr(&cpuhp_stat << 312 ktime_t now, end, start = ktime_get(); << 313 int sync; << 314 << 315 end = start + 10ULL * NSEC_PER_SEC; << 316 << 317 sync = atomic_read(st); << 318 while (1) { << 319 if (sync == state) { << 320 if (!atomic_try_cmpxch << 321 continue; << 322 return true; << 323 } << 324 << 325 now = ktime_get(); << 326 if (now > end) { << 327 /* Timeout. Leave the << 328 return false; << 329 } else if (now - start < NSEC_ << 330 /* Poll for one millis << 331 arch_cpuhp_sync_state_ << 332 } else { << 333 usleep_range(USEC_PER_ << 334 } << 335 sync = atomic_read(st); << 336 } << 337 return true; << 338 } << 339 #else /* CONFIG_HOTPLUG_CORE_SYNC */ << 340 static inline void cpuhp_ap_update_sync_state( << 341 #endif /* !CONFIG_HOTPLUG_CORE_SYNC */ << 342 << 343 #ifdef CONFIG_HOTPLUG_CORE_SYNC_DEAD << 344 /** << 345 * cpuhp_ap_report_dead - Update synchronizati << 346 * << 347 * No synchronization point. Just update of th << 348 */ << 349 void cpuhp_ap_report_dead(void) << 350 { << 351 cpuhp_ap_update_sync_state(SYNC_STATE_ << 352 } << 353 << 354 void __weak arch_cpuhp_cleanup_dead_cpu(unsign << 355 << 356 /* << 357 * Late CPU shutdown synchronization point. Ca << 358 * because the AP cannot issue complete() at t << 359 */ << 360 static void cpuhp_bp_sync_dead(unsigned int cp << 361 { << 362 atomic_t *st = per_cpu_ptr(&cpuhp_stat << 363 int sync = atomic_read(st); << 364 << 365 do { << 366 /* CPU can have reported dead << 367 if (sync == SYNC_STATE_DEAD) << 368 break; << 369 } while (!atomic_try_cmpxchg(st, &sync << 370 << 371 if (cpuhp_wait_for_sync_state(cpu, SYN << 372 /* CPU reached dead state. Inv << 373 arch_cpuhp_cleanup_dead_cpu(cp << 374 return; << 375 } << 376 << 377 /* No further action possible. Emit me << 378 pr_err("CPU%u failed to report dead st << 379 } << 380 #else /* CONFIG_HOTPLUG_CORE_SYNC_DEAD */ << 381 static inline void cpuhp_bp_sync_dead(unsigned << 382 #endif /* !CONFIG_HOTPLUG_CORE_SYNC_DEAD */ << 383 << 384 #ifdef CONFIG_HOTPLUG_CORE_SYNC_FULL << 385 /** << 386 * cpuhp_ap_sync_alive - Synchronize AP with t << 387 * << 388 * Updates the AP synchronization state to SYN << 389 * for the BP to release it. << 390 */ << 391 void cpuhp_ap_sync_alive(void) << 392 { << 393 atomic_t *st = this_cpu_ptr(&cpuhp_sta << 394 << 395 cpuhp_ap_update_sync_state(SYNC_STATE_ << 396 << 397 /* Wait for the control CPU to release << 398 while (atomic_read(st) != SYNC_STATE_S << 399 cpu_relax(); << 400 } << 401 << 402 static bool cpuhp_can_boot_ap(unsigned int cpu << 403 { << 404 atomic_t *st = per_cpu_ptr(&cpuhp_stat << 405 int sync = atomic_read(st); << 406 << 407 again: << 408 switch (sync) { << 409 case SYNC_STATE_DEAD: << 410 /* CPU is properly dead */ << 411 break; << 412 case SYNC_STATE_KICKED: << 413 /* CPU did not come up in prev << 414 break; << 415 case SYNC_STATE_ALIVE: << 416 /* CPU is stuck cpuhp_ap_sync_ << 417 break; << 418 default: << 419 /* CPU failed to report online << 420 return false; << 421 } << 422 << 423 /* Prepare for booting */ << 424 if (!atomic_try_cmpxchg(st, &sync, SYN << 425 goto again; << 426 << 427 return true; << 428 } << 429 << 430 void __weak arch_cpuhp_cleanup_kick_cpu(unsign << 431 << 432 /* << 433 * Early CPU bringup synchronization point. Ca << 434 * because the AP cannot issue complete() so e << 435 */ << 436 static int cpuhp_bp_sync_alive(unsigned int cp << 437 { << 438 int ret = 0; << 439 << 440 if (!IS_ENABLED(CONFIG_HOTPLUG_CORE_SY << 441 return 0; << 442 << 443 if (!cpuhp_wait_for_sync_state(cpu, SY << 444 pr_err("CPU%u failed to report << 445 ret = -EIO; << 446 } << 447 << 448 /* Let the architecture cleanup the ki << 449 arch_cpuhp_cleanup_kick_cpu(cpu); << 450 return ret; << 451 } << 452 #else /* CONFIG_HOTPLUG_CORE_SYNC_FULL */ << 453 static inline int cpuhp_bp_sync_alive(unsigned << 454 static inline bool cpuhp_can_boot_ap(unsigned << 455 #endif /* !CONFIG_HOTPLUG_CORE_SYNC_FULL */ << 456 << 457 /* Serializes the updates to cpu_online_mask, 180 /* Serializes the updates to cpu_online_mask, cpu_present_mask */ 458 static DEFINE_MUTEX(cpu_add_remove_lock); 181 static DEFINE_MUTEX(cpu_add_remove_lock); 459 bool cpuhp_tasks_frozen; 182 bool cpuhp_tasks_frozen; 460 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); 183 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); 461 184 462 /* 185 /* 463 * The following two APIs (cpu_maps_update_beg 186 * The following two APIs (cpu_maps_update_begin/done) must be used when 464 * attempting to serialize the updates to cpu_ 187 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. 465 */ 188 */ 466 void cpu_maps_update_begin(void) 189 void cpu_maps_update_begin(void) 467 { 190 { 468 mutex_lock(&cpu_add_remove_lock); 191 mutex_lock(&cpu_add_remove_lock); 469 } 192 } 470 193 471 void cpu_maps_update_done(void) 194 void cpu_maps_update_done(void) 472 { 195 { 473 mutex_unlock(&cpu_add_remove_lock); 196 mutex_unlock(&cpu_add_remove_lock); 474 } 197 } 475 198 476 /* !! 199 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. 477 * If set, cpu_up and cpu_down will return -EB << 478 * Should always be manipulated under cpu_add_ 200 * Should always be manipulated under cpu_add_remove_lock 479 */ 201 */ 480 static int cpu_hotplug_disabled; 202 static int cpu_hotplug_disabled; 481 203 482 #ifdef CONFIG_HOTPLUG_CPU 204 #ifdef CONFIG_HOTPLUG_CPU 483 205 484 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); !! 206 static struct { >> 207 struct task_struct *active_writer; >> 208 /* wait queue to wake up the active_writer */ >> 209 wait_queue_head_t wq; >> 210 /* verifies that no writer will get active while readers are active */ >> 211 struct mutex lock; >> 212 /* >> 213 * Also blocks the new readers during >> 214 * an ongoing cpu hotplug operation. >> 215 */ >> 216 atomic_t refcount; >> 217 >> 218 #ifdef CONFIG_DEBUG_LOCK_ALLOC >> 219 struct lockdep_map dep_map; >> 220 #endif >> 221 } cpu_hotplug = { >> 222 .active_writer = NULL, >> 223 .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), >> 224 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), >> 225 #ifdef CONFIG_DEBUG_LOCK_ALLOC >> 226 .dep_map = STATIC_LOCKDEP_MAP_INIT("cpu_hotplug.dep_map", &cpu_hotplug.dep_map), >> 227 #endif >> 228 }; 485 229 486 static bool cpu_hotplug_offline_disabled __ro_ !! 230 /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ >> 231 #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) >> 232 #define cpuhp_lock_acquire_tryread() \ >> 233 lock_map_acquire_tryread(&cpu_hotplug.dep_map) >> 234 #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) >> 235 #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) 487 236 488 void cpus_read_lock(void) << 489 { << 490 percpu_down_read(&cpu_hotplug_lock); << 491 } << 492 EXPORT_SYMBOL_GPL(cpus_read_lock); << 493 237 494 int cpus_read_trylock(void) !! 238 void get_online_cpus(void) 495 { 239 { 496 return percpu_down_read_trylock(&cpu_h !! 240 might_sleep(); >> 241 if (cpu_hotplug.active_writer == current) >> 242 return; >> 243 cpuhp_lock_acquire_read(); >> 244 mutex_lock(&cpu_hotplug.lock); >> 245 atomic_inc(&cpu_hotplug.refcount); >> 246 mutex_unlock(&cpu_hotplug.lock); 497 } 247 } 498 EXPORT_SYMBOL_GPL(cpus_read_trylock); !! 248 EXPORT_SYMBOL_GPL(get_online_cpus); 499 249 500 void cpus_read_unlock(void) !! 250 void put_online_cpus(void) 501 { 251 { 502 percpu_up_read(&cpu_hotplug_lock); !! 252 int refcount; 503 } << 504 EXPORT_SYMBOL_GPL(cpus_read_unlock); << 505 253 506 void cpus_write_lock(void) !! 254 if (cpu_hotplug.active_writer == current) 507 { !! 255 return; 508 percpu_down_write(&cpu_hotplug_lock); << 509 } << 510 256 511 void cpus_write_unlock(void) !! 257 refcount = atomic_dec_return(&cpu_hotplug.refcount); 512 { !! 258 if (WARN_ON(refcount < 0)) /* try to fix things up */ 513 percpu_up_write(&cpu_hotplug_lock); !! 259 atomic_inc(&cpu_hotplug.refcount); 514 } << 515 260 516 void lockdep_assert_cpus_held(void) !! 261 if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) 517 { !! 262 wake_up(&cpu_hotplug.wq); 518 /* << 519 * We can't have hotplug operations be << 520 * and some init codepaths will knowin << 521 * This is all valid, so mute lockdep << 522 * unheld locks. << 523 */ << 524 if (system_state < SYSTEM_RUNNING) << 525 return; << 526 263 527 percpu_rwsem_assert_held(&cpu_hotplug_ !! 264 cpuhp_lock_release(); 528 } << 529 265 530 #ifdef CONFIG_LOCKDEP << 531 int lockdep_is_cpus_held(void) << 532 { << 533 return percpu_rwsem_is_held(&cpu_hotpl << 534 } 266 } 535 #endif !! 267 EXPORT_SYMBOL_GPL(put_online_cpus); 536 268 537 static void lockdep_acquire_cpus_lock(void) !! 269 /* >> 270 * This ensures that the hotplug operation can begin only when the >> 271 * refcount goes to zero. >> 272 * >> 273 * Note that during a cpu-hotplug operation, the new readers, if any, >> 274 * will be blocked by the cpu_hotplug.lock >> 275 * >> 276 * Since cpu_hotplug_begin() is always called after invoking >> 277 * cpu_maps_update_begin(), we can be sure that only one writer is active. >> 278 * >> 279 * Note that theoretically, there is a possibility of a livelock: >> 280 * - Refcount goes to zero, last reader wakes up the sleeping >> 281 * writer. >> 282 * - Last reader unlocks the cpu_hotplug.lock. >> 283 * - A new reader arrives at this moment, bumps up the refcount. >> 284 * - The writer acquires the cpu_hotplug.lock finds the refcount >> 285 * non zero and goes to sleep again. >> 286 * >> 287 * However, this is very difficult to achieve in practice since >> 288 * get_online_cpus() not an api which is called all that often. >> 289 * >> 290 */ >> 291 void cpu_hotplug_begin(void) 538 { 292 { 539 rwsem_acquire(&cpu_hotplug_lock.dep_ma !! 293 DEFINE_WAIT(wait); 540 } << 541 294 542 static void lockdep_release_cpus_lock(void) !! 295 cpu_hotplug.active_writer = current; 543 { !! 296 cpuhp_lock_acquire(); 544 rwsem_release(&cpu_hotplug_lock.dep_ma !! 297 >> 298 for (;;) { >> 299 mutex_lock(&cpu_hotplug.lock); >> 300 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); >> 301 if (likely(!atomic_read(&cpu_hotplug.refcount))) >> 302 break; >> 303 mutex_unlock(&cpu_hotplug.lock); >> 304 schedule(); >> 305 } >> 306 finish_wait(&cpu_hotplug.wq, &wait); 545 } 307 } 546 308 547 /* Declare CPU offlining not supported */ !! 309 void cpu_hotplug_done(void) 548 void cpu_hotplug_disable_offlining(void) << 549 { 310 { 550 cpu_maps_update_begin(); !! 311 cpu_hotplug.active_writer = NULL; 551 cpu_hotplug_offline_disabled = true; !! 312 mutex_unlock(&cpu_hotplug.lock); 552 cpu_maps_update_done(); !! 313 cpuhp_lock_release(); 553 } 314 } 554 315 555 /* 316 /* 556 * Wait for currently running CPU hotplug oper 317 * Wait for currently running CPU hotplug operations to complete (if any) and 557 * disable future CPU hotplug (from sysfs). Th 318 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects 558 * the 'cpu_hotplug_disabled' flag. The same l 319 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the 559 * hotplug path before performing hotplug oper 320 * hotplug path before performing hotplug operations. So acquiring that lock 560 * guarantees mutual exclusion from any curren 321 * guarantees mutual exclusion from any currently running hotplug operations. 561 */ 322 */ 562 void cpu_hotplug_disable(void) 323 void cpu_hotplug_disable(void) 563 { 324 { 564 cpu_maps_update_begin(); 325 cpu_maps_update_begin(); 565 cpu_hotplug_disabled++; 326 cpu_hotplug_disabled++; 566 cpu_maps_update_done(); 327 cpu_maps_update_done(); 567 } 328 } 568 EXPORT_SYMBOL_GPL(cpu_hotplug_disable); 329 EXPORT_SYMBOL_GPL(cpu_hotplug_disable); 569 330 570 static void __cpu_hotplug_enable(void) 331 static void __cpu_hotplug_enable(void) 571 { 332 { 572 if (WARN_ONCE(!cpu_hotplug_disabled, " 333 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) 573 return; 334 return; 574 cpu_hotplug_disabled--; 335 cpu_hotplug_disabled--; 575 } 336 } 576 337 577 void cpu_hotplug_enable(void) 338 void cpu_hotplug_enable(void) 578 { 339 { 579 cpu_maps_update_begin(); 340 cpu_maps_update_begin(); 580 __cpu_hotplug_enable(); 341 __cpu_hotplug_enable(); 581 cpu_maps_update_done(); 342 cpu_maps_update_done(); 582 } 343 } 583 EXPORT_SYMBOL_GPL(cpu_hotplug_enable); 344 EXPORT_SYMBOL_GPL(cpu_hotplug_enable); 584 << 585 #else << 586 << 587 static void lockdep_acquire_cpus_lock(void) << 588 { << 589 } << 590 << 591 static void lockdep_release_cpus_lock(void) << 592 { << 593 } << 594 << 595 #endif /* CONFIG_HOTPLUG_CPU */ 345 #endif /* CONFIG_HOTPLUG_CPU */ 596 346 597 /* !! 347 /* Notifier wrappers for transitioning to state machine */ 598 * Architectures that need SMT-specific errata << 599 * should override this. << 600 */ << 601 void __weak arch_smt_update(void) { } << 602 << 603 #ifdef CONFIG_HOTPLUG_SMT << 604 348 605 enum cpuhp_smt_control cpu_smt_control __read_ !! 349 static int bringup_wait_for_ap(unsigned int cpu) 606 static unsigned int cpu_smt_max_threads __ro_a << 607 unsigned int cpu_smt_num_threads __read_mostly << 608 << 609 void __init cpu_smt_disable(bool force) << 610 { << 611 if (!cpu_smt_possible()) << 612 return; << 613 << 614 if (force) { << 615 pr_info("SMT: Force disabled\n << 616 cpu_smt_control = CPU_SMT_FORC << 617 } else { << 618 pr_info("SMT: disabled\n"); << 619 cpu_smt_control = CPU_SMT_DISA << 620 } << 621 cpu_smt_num_threads = 1; << 622 } << 623 << 624 /* << 625 * The decision whether SMT is supported can o << 626 * CPU identification. Called from architectur << 627 */ << 628 void __init cpu_smt_set_num_threads(unsigned i << 629 unsigned i << 630 { << 631 WARN_ON(!num_threads || (num_threads > << 632 << 633 if (max_threads == 1) << 634 cpu_smt_control = CPU_SMT_NOT_ << 635 << 636 cpu_smt_max_threads = max_threads; << 637 << 638 /* << 639 * If SMT has been disabled via the ke << 640 * not supported, set cpu_smt_num_thre << 641 * If enabled, take the architecture r << 642 * to bring up into account. << 643 */ << 644 if (cpu_smt_control != CPU_SMT_ENABLED << 645 cpu_smt_num_threads = 1; << 646 else if (num_threads < cpu_smt_num_thr << 647 cpu_smt_num_threads = num_thre << 648 } << 649 << 650 static int __init smt_cmdline_disable(char *st << 651 { << 652 cpu_smt_disable(str && !strcmp(str, "f << 653 return 0; << 654 } << 655 early_param("nosmt", smt_cmdline_disable); << 656 << 657 /* << 658 * For Archicture supporting partial SMT state << 659 * Otherwise this has already been checked thr << 660 * setting the SMT level. << 661 */ << 662 static inline bool cpu_smt_thread_allowed(unsi << 663 { << 664 #ifdef CONFIG_SMT_NUM_THREADS_DYNAMIC << 665 return topology_smt_thread_allowed(cpu << 666 #else << 667 return true; << 668 #endif << 669 } << 670 << 671 static inline bool cpu_bootable(unsigned int c << 672 { << 673 if (cpu_smt_control == CPU_SMT_ENABLED << 674 return true; << 675 << 676 /* All CPUs are bootable if controls a << 677 if (cpu_smt_control == CPU_SMT_NOT_IMP << 678 return true; << 679 << 680 /* All CPUs are bootable if CPU is not << 681 if (cpu_smt_control == CPU_SMT_NOT_SUP << 682 return true; << 683 << 684 if (topology_is_primary_thread(cpu)) << 685 return true; << 686 << 687 /* << 688 * On x86 it's required to boot all lo << 689 * that the init code can get a chance << 690 * CPU. Otherwise, a broadcasted MCE o << 691 * core will shutdown the machine. << 692 */ << 693 return !cpumask_test_cpu(cpu, &cpus_bo << 694 } << 695 << 696 /* Returns true if SMT is supported and not fo << 697 bool cpu_smt_possible(void) << 698 { << 699 return cpu_smt_control != CPU_SMT_FORC << 700 cpu_smt_control != CPU_SMT_NOT << 701 } << 702 EXPORT_SYMBOL_GPL(cpu_smt_possible); << 703 << 704 #else << 705 static inline bool cpu_bootable(unsigned int c << 706 #endif << 707 << 708 static inline enum cpuhp_state << 709 cpuhp_set_state(int cpu, struct cpuhp_cpu_stat << 710 { << 711 enum cpuhp_state prev_state = st->stat << 712 bool bringup = st->state < target; << 713 << 714 st->rollback = false; << 715 st->last = NULL; << 716 << 717 st->target = target; << 718 st->single = false; << 719 st->bringup = bringup; << 720 if (cpu_dying(cpu) != !bringup) << 721 set_cpu_dying(cpu, !bringup); << 722 << 723 return prev_state; << 724 } << 725 << 726 static inline void << 727 cpuhp_reset_state(int cpu, struct cpuhp_cpu_st << 728 enum cpuhp_state prev_state) << 729 { << 730 bool bringup = !st->bringup; << 731 << 732 st->target = prev_state; << 733 << 734 /* << 735 * Already rolling back. No need inver << 736 * the current state. << 737 */ << 738 if (st->rollback) << 739 return; << 740 << 741 st->rollback = true; << 742 << 743 /* << 744 * If we have st->last we need to undo << 745 * state first. Otherwise start undo a << 746 */ << 747 if (!st->last) { << 748 if (st->bringup) << 749 st->state--; << 750 else << 751 st->state++; << 752 } << 753 << 754 st->bringup = bringup; << 755 if (cpu_dying(cpu) != !bringup) << 756 set_cpu_dying(cpu, !bringup); << 757 } << 758 << 759 /* Regular hotplug invocation of the AP hotplu << 760 static void __cpuhp_kick_ap(struct cpuhp_cpu_s << 761 { << 762 if (!st->single && st->state == st->ta << 763 return; << 764 << 765 st->result = 0; << 766 /* << 767 * Make sure the above stores are visi << 768 * true. Paired with the mb() above in << 769 */ << 770 smp_mb(); << 771 st->should_run = true; << 772 wake_up_process(st->thread); << 773 wait_for_ap_thread(st, st->bringup); << 774 } << 775 << 776 static int cpuhp_kick_ap(int cpu, struct cpuhp << 777 enum cpuhp_state targ << 778 { << 779 enum cpuhp_state prev_state; << 780 int ret; << 781 << 782 prev_state = cpuhp_set_state(cpu, st, << 783 __cpuhp_kick_ap(st); << 784 if ((ret = st->result)) { << 785 cpuhp_reset_state(cpu, st, pre << 786 __cpuhp_kick_ap(st); << 787 } << 788 << 789 return ret; << 790 } << 791 << 792 static int bringup_wait_for_ap_online(unsigned << 793 { 350 { 794 struct cpuhp_cpu_state *st = per_cpu_p 351 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 795 352 796 /* Wait for the CPU to reach CPUHP_AP_ !! 353 wait_for_completion(&st->done); 797 wait_for_ap_thread(st, true); !! 354 return st->result; 798 if (WARN_ON_ONCE((!cpu_online(cpu)))) << 799 return -ECANCELED; << 800 << 801 /* Unpark the hotplug thread of the ta << 802 kthread_unpark(st->thread); << 803 << 804 /* << 805 * SMT soft disabling on X86 requires << 806 * BIOS 'wait for SIPI' state in order << 807 * CPU marked itself as booted_once in << 808 * cpu_bootable() check will now retur << 809 * primary sibling. << 810 */ << 811 if (!cpu_bootable(cpu)) << 812 return -ECANCELED; << 813 return 0; << 814 } << 815 << 816 #ifdef CONFIG_HOTPLUG_SPLIT_STARTUP << 817 static int cpuhp_kick_ap_alive(unsigned int cp << 818 { << 819 if (!cpuhp_can_boot_ap(cpu)) << 820 return -EAGAIN; << 821 << 822 return arch_cpuhp_kick_ap_alive(cpu, i << 823 } 355 } 824 356 825 static int cpuhp_bringup_ap(unsigned int cpu) << 826 { << 827 struct cpuhp_cpu_state *st = per_cpu_p << 828 int ret; << 829 << 830 /* << 831 * Some architectures have to walk the << 832 * setup the vector space for the cpu << 833 * Prevent irq alloc/free across the b << 834 */ << 835 irq_lock_sparse(); << 836 << 837 ret = cpuhp_bp_sync_alive(cpu); << 838 if (ret) << 839 goto out_unlock; << 840 << 841 ret = bringup_wait_for_ap_online(cpu); << 842 if (ret) << 843 goto out_unlock; << 844 << 845 irq_unlock_sparse(); << 846 << 847 if (st->target <= CPUHP_AP_ONLINE_IDLE << 848 return 0; << 849 << 850 return cpuhp_kick_ap(cpu, st, st->targ << 851 << 852 out_unlock: << 853 irq_unlock_sparse(); << 854 return ret; << 855 } << 856 #else << 857 static int bringup_cpu(unsigned int cpu) 357 static int bringup_cpu(unsigned int cpu) 858 { 358 { 859 struct cpuhp_cpu_state *st = per_cpu_p << 860 struct task_struct *idle = idle_thread 359 struct task_struct *idle = idle_thread_get(cpu); 861 int ret; 360 int ret; 862 361 863 if (!cpuhp_can_boot_ap(cpu)) << 864 return -EAGAIN; << 865 << 866 /* 362 /* 867 * Some architectures have to walk the 363 * Some architectures have to walk the irq descriptors to 868 * setup the vector space for the cpu 364 * setup the vector space for the cpu which comes online. 869 * !! 365 * Prevent irq alloc/free across the bringup. 870 * Prevent irq alloc/free across the b << 871 * sparse irq lock. Hold it until the << 872 * startup in cpuhp_online_idle() whic << 873 * intermediate synchronization points << 874 */ 366 */ 875 irq_lock_sparse(); 367 irq_lock_sparse(); 876 368 >> 369 /* Arch-specific enabling code. */ 877 ret = __cpu_up(cpu, idle); 370 ret = __cpu_up(cpu, idle); 878 if (ret) << 879 goto out_unlock; << 880 << 881 ret = cpuhp_bp_sync_alive(cpu); << 882 if (ret) << 883 goto out_unlock; << 884 << 885 ret = bringup_wait_for_ap_online(cpu); << 886 if (ret) << 887 goto out_unlock; << 888 << 889 irq_unlock_sparse(); << 890 << 891 if (st->target <= CPUHP_AP_ONLINE_IDLE << 892 return 0; << 893 << 894 return cpuhp_kick_ap(cpu, st, st->targ << 895 << 896 out_unlock: << 897 irq_unlock_sparse(); 371 irq_unlock_sparse(); >> 372 if (ret) >> 373 return ret; >> 374 ret = bringup_wait_for_ap(cpu); >> 375 BUG_ON(!cpu_online(cpu)); 898 return ret; 376 return ret; 899 } 377 } 900 #endif << 901 << 902 static int finish_cpu(unsigned int cpu) << 903 { << 904 struct task_struct *idle = idle_thread << 905 struct mm_struct *mm = idle->active_mm << 906 << 907 /* << 908 * idle_task_exit() will have switched << 909 * clean up any remaining active_mm st << 910 */ << 911 if (mm != &init_mm) << 912 idle->active_mm = &init_mm; << 913 mmdrop_lazy_tlb(mm); << 914 return 0; << 915 } << 916 378 917 /* 379 /* 918 * Hotplug state machine related functions 380 * Hotplug state machine related functions 919 */ 381 */ >> 382 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) >> 383 { >> 384 for (st->state++; st->state < st->target; st->state++) { >> 385 struct cpuhp_step *step = cpuhp_get_step(st->state); 920 386 921 /* !! 387 if (!step->skip_onerr) 922 * Get the next state to run. Empty ones will !! 388 cpuhp_invoke_callback(cpu, st->state, true, NULL); 923 * state must be run. !! 389 } 924 * << 925 * st->state will be modified ahead of time, t << 926 * has already ran. << 927 */ << 928 static bool cpuhp_next_state(bool bringup, << 929 enum cpuhp_state << 930 struct cpuhp_cpu_ << 931 enum cpuhp_state << 932 { << 933 do { << 934 if (bringup) { << 935 if (st->state >= targe << 936 return false; << 937 << 938 *state_to_run = ++st-> << 939 } else { << 940 if (st->state <= targe << 941 return false; << 942 << 943 *state_to_run = st->st << 944 } << 945 << 946 if (!cpuhp_step_empty(bringup, << 947 break; << 948 } while (true); << 949 << 950 return true; << 951 } 390 } 952 391 953 static int __cpuhp_invoke_callback_range(bool !! 392 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, 954 unsig !! 393 enum cpuhp_state target) 955 struc << 956 enum << 957 bool << 958 { 394 { 959 enum cpuhp_state state; !! 395 enum cpuhp_state prev_state = st->state; 960 int ret = 0; 396 int ret = 0; 961 397 962 while (cpuhp_next_state(bringup, &stat !! 398 for (; st->state > target; st->state--) { 963 int err; !! 399 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL); 964 !! 400 if (ret) { 965 err = cpuhp_invoke_callback(cp !! 401 st->target = prev_state; 966 if (!err) !! 402 undo_cpu_down(cpu, st); 967 continue; << 968 << 969 if (nofail) { << 970 pr_warn("CPU %u %s sta << 971 cpu, bringup ? << 972 cpuhp_get_step << 973 st->state, err << 974 ret = -1; << 975 } else { << 976 ret = err; << 977 break; 403 break; 978 } 404 } 979 } 405 } 980 << 981 return ret; 406 return ret; 982 } 407 } 983 408 984 static inline int cpuhp_invoke_callback_range( !! 409 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) 985 << 986 << 987 << 988 { 410 { 989 return __cpuhp_invoke_callback_range(b !! 411 for (st->state--; st->state > st->target; st->state--) { 990 } !! 412 struct cpuhp_step *step = cpuhp_get_step(st->state); 991 << 992 static inline void cpuhp_invoke_callback_range << 993 << 994 << 995 << 996 { << 997 __cpuhp_invoke_callback_range(bringup, << 998 } << 999 413 1000 static inline bool can_rollback_cpu(struct cp !! 414 if (!step->skip_onerr) 1001 { !! 415 cpuhp_invoke_callback(cpu, st->state, false, NULL); 1002 if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) !! 416 } 1003 return true; << 1004 /* << 1005 * When CPU hotplug is disabled, then << 1006 * possible because takedown_cpu() an << 1007 * subsystem specific mechanisms are << 1008 * which would be completely unplugge << 1009 * in the current state. << 1010 */ << 1011 return st->state <= CPUHP_BRINGUP_CPU << 1012 } 417 } 1013 418 1014 static int cpuhp_up_callbacks(unsigned int cp 419 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, 1015 enum cpuhp_stat 420 enum cpuhp_state target) 1016 { 421 { 1017 enum cpuhp_state prev_state = st->sta 422 enum cpuhp_state prev_state = st->state; 1018 int ret = 0; 423 int ret = 0; 1019 424 1020 ret = cpuhp_invoke_callback_range(tru !! 425 while (st->state < target) { 1021 if (ret) { !! 426 st->state++; 1022 pr_debug("CPU UP failed (%d) !! 427 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL); 1023 ret, cpu, cpuhp_get_ !! 428 if (ret) { 1024 st->state); !! 429 st->target = prev_state; 1025 !! 430 undo_cpu_up(cpu, st); 1026 cpuhp_reset_state(cpu, st, pr !! 431 break; 1027 if (can_rollback_cpu(st)) !! 432 } 1028 WARN_ON(cpuhp_invoke_ << 1029 << 1030 } 433 } 1031 return ret; 434 return ret; 1032 } 435 } 1033 436 1034 /* 437 /* 1035 * The cpu hotplug threads manage the bringup 438 * The cpu hotplug threads manage the bringup and teardown of the cpus 1036 */ 439 */ >> 440 static void cpuhp_create(unsigned int cpu) >> 441 { >> 442 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); >> 443 >> 444 init_completion(&st->done); >> 445 } >> 446 1037 static int cpuhp_should_run(unsigned int cpu) 447 static int cpuhp_should_run(unsigned int cpu) 1038 { 448 { 1039 struct cpuhp_cpu_state *st = this_cpu 449 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); 1040 450 1041 return st->should_run; 451 return st->should_run; 1042 } 452 } 1043 453 >> 454 /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */ >> 455 static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st) >> 456 { >> 457 enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU); >> 458 >> 459 return cpuhp_down_callbacks(cpu, st, target); >> 460 } >> 461 >> 462 /* Execute the online startup callbacks. Used to be CPU_ONLINE */ >> 463 static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st) >> 464 { >> 465 return cpuhp_up_callbacks(cpu, st, st->target); >> 466 } >> 467 1044 /* 468 /* 1045 * Execute teardown/startup callbacks on the 469 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke 1046 * callbacks when a state gets [un]installed 470 * callbacks when a state gets [un]installed at runtime. 1047 * << 1048 * Each invocation of this function by the sm << 1049 * state callback. << 1050 * << 1051 * It has 3 modes of operation: << 1052 * - single: runs st->cb_state << 1053 * - up: runs ++st->state, while st->sta << 1054 * - down: runs st->state--, while st->sta << 1055 * << 1056 * When complete or on error, should_run is c << 1057 */ 471 */ 1058 static void cpuhp_thread_fun(unsigned int cpu 472 static void cpuhp_thread_fun(unsigned int cpu) 1059 { 473 { 1060 struct cpuhp_cpu_state *st = this_cpu 474 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); 1061 bool bringup = st->bringup; !! 475 int ret = 0; 1062 enum cpuhp_state state; << 1063 << 1064 if (WARN_ON_ONCE(!st->should_run)) << 1065 return; << 1066 476 1067 /* 477 /* 1068 * ACQUIRE for the cpuhp_should_run() !! 478 * Paired with the mb() in cpuhp_kick_ap_work and 1069 * that if we see ->should_run we als !! 479 * cpuhp_invoke_ap_callback, so the work set is consistent visible. 1070 */ 480 */ 1071 smp_mb(); 481 smp_mb(); >> 482 if (!st->should_run) >> 483 return; 1072 484 1073 /* !! 485 st->should_run = false; 1074 * The BP holds the hotplug lock, but << 1075 * ensure that anybody asserting the << 1076 * it so. << 1077 */ << 1078 lockdep_acquire_cpus_lock(); << 1079 cpuhp_lock_acquire(bringup); << 1080 486 >> 487 /* Single callback invocation for [un]install ? */ 1081 if (st->single) { 488 if (st->single) { 1082 state = st->cb_state; !! 489 if (st->cb_state < CPUHP_AP_ONLINE) { 1083 st->should_run = false; !! 490 local_irq_disable(); 1084 } else { !! 491 ret = cpuhp_invoke_callback(cpu, st->cb_state, 1085 st->should_run = cpuhp_next_s !! 492 st->bringup, st->node); 1086 if (!st->should_run) !! 493 local_irq_enable(); 1087 goto end; !! 494 } else { 1088 } !! 495 ret = cpuhp_invoke_callback(cpu, st->cb_state, 1089 !! 496 st->bringup, st->node); 1090 WARN_ON_ONCE(!cpuhp_is_ap_state(state !! 497 } 1091 !! 498 } else if (st->rollback) { 1092 if (cpuhp_is_atomic_state(state)) { !! 499 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); 1093 local_irq_disable(); << 1094 st->result = cpuhp_invoke_cal << 1095 local_irq_enable(); << 1096 500 1097 /* !! 501 undo_cpu_down(cpu, st); 1098 * STARTING/DYING must not fa !! 502 st->rollback = false; 1099 */ << 1100 WARN_ON_ONCE(st->result); << 1101 } else { 503 } else { 1102 st->result = cpuhp_invoke_cal !! 504 /* Cannot happen .... */ 1103 } !! 505 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); 1104 506 1105 if (st->result) { !! 507 /* Regular hotplug work */ 1106 /* !! 508 if (st->state < st->target) 1107 * If we fail on a rollback, !! 509 ret = cpuhp_ap_online(cpu, st); 1108 * paddle, no way forward, no !! 510 else if (st->state > st->target) 1109 * playing. !! 511 ret = cpuhp_ap_offline(cpu, st); 1110 */ << 1111 WARN_ON_ONCE(st->rollback); << 1112 st->should_run = false; << 1113 } 512 } 1114 !! 513 st->result = ret; 1115 end: !! 514 complete(&st->done); 1116 cpuhp_lock_release(bringup); << 1117 lockdep_release_cpus_lock(); << 1118 << 1119 if (!st->should_run) << 1120 complete_ap_thread(st, bringu << 1121 } 515 } 1122 516 1123 /* Invoke a single callback on a remote cpu * 517 /* Invoke a single callback on a remote cpu */ 1124 static int 518 static int 1125 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_ 519 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, 1126 struct hlist_node *n 520 struct hlist_node *node) 1127 { 521 { 1128 struct cpuhp_cpu_state *st = per_cpu_ 522 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 1129 int ret; << 1130 523 1131 if (!cpu_online(cpu)) 524 if (!cpu_online(cpu)) 1132 return 0; 525 return 0; 1133 526 1134 cpuhp_lock_acquire(false); << 1135 cpuhp_lock_release(false); << 1136 << 1137 cpuhp_lock_acquire(true); << 1138 cpuhp_lock_release(true); << 1139 << 1140 /* 527 /* 1141 * If we are up and running, use the 528 * If we are up and running, use the hotplug thread. For early calls 1142 * we invoke the thread function dire 529 * we invoke the thread function directly. 1143 */ 530 */ 1144 if (!st->thread) 531 if (!st->thread) 1145 return cpuhp_invoke_callback( !! 532 return cpuhp_invoke_callback(cpu, state, bringup, node); 1146 533 1147 st->rollback = false; << 1148 st->last = NULL; << 1149 << 1150 st->node = node; << 1151 st->bringup = bringup; << 1152 st->cb_state = state; 534 st->cb_state = state; 1153 st->single = true; 535 st->single = true; 1154 !! 536 st->bringup = bringup; 1155 __cpuhp_kick_ap(st); !! 537 st->node = node; 1156 538 1157 /* 539 /* 1158 * If we failed and did a partial, do !! 540 * Make sure the above stores are visible before should_run becomes >> 541 * true. Paired with the mb() above in cpuhp_thread_fun() 1159 */ 542 */ 1160 if ((ret = st->result) && st->last) { !! 543 smp_mb(); 1161 st->rollback = true; !! 544 st->should_run = true; 1162 st->bringup = !bringup; !! 545 wake_up_process(st->thread); 1163 !! 546 wait_for_completion(&st->done); 1164 __cpuhp_kick_ap(st); !! 547 return st->result; 1165 } !! 548 } 1166 549 >> 550 /* Regular hotplug invocation of the AP hotplug thread */ >> 551 static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st) >> 552 { >> 553 st->result = 0; >> 554 st->single = false; 1167 /* 555 /* 1168 * Clean up the leftovers so the next !! 556 * Make sure the above stores are visible before should_run becomes 1169 * data. !! 557 * true. Paired with the mb() above in cpuhp_thread_fun() 1170 */ 558 */ 1171 st->node = st->last = NULL; !! 559 smp_mb(); 1172 return ret; !! 560 st->should_run = true; >> 561 wake_up_process(st->thread); 1173 } 562 } 1174 563 1175 static int cpuhp_kick_ap_work(unsigned int cp 564 static int cpuhp_kick_ap_work(unsigned int cpu) 1176 { 565 { 1177 struct cpuhp_cpu_state *st = per_cpu_ 566 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 1178 enum cpuhp_state prev_state = st->sta !! 567 enum cpuhp_state state = st->state; 1179 int ret; << 1180 << 1181 cpuhp_lock_acquire(false); << 1182 cpuhp_lock_release(false); << 1183 << 1184 cpuhp_lock_acquire(true); << 1185 cpuhp_lock_release(true); << 1186 568 1187 trace_cpuhp_enter(cpu, st->target, pr !! 569 trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work); 1188 ret = cpuhp_kick_ap(cpu, st, st->targ !! 570 __cpuhp_kick_ap_work(st); 1189 trace_cpuhp_exit(cpu, st->state, prev !! 571 wait_for_completion(&st->done); 1190 !! 572 trace_cpuhp_exit(cpu, st->state, state, st->result); 1191 return ret; !! 573 return st->result; 1192 } 574 } 1193 575 1194 static struct smp_hotplug_thread cpuhp_thread 576 static struct smp_hotplug_thread cpuhp_threads = { 1195 .store = &cpuhp_stat 577 .store = &cpuhp_state.thread, >> 578 .create = &cpuhp_create, 1196 .thread_should_run = cpuhp_shoul 579 .thread_should_run = cpuhp_should_run, 1197 .thread_fn = cpuhp_threa 580 .thread_fn = cpuhp_thread_fun, 1198 .thread_comm = "cpuhp/%u", 581 .thread_comm = "cpuhp/%u", 1199 .selfparking = true, 582 .selfparking = true, 1200 }; 583 }; 1201 584 1202 static __init void cpuhp_init_state(void) << 1203 { << 1204 struct cpuhp_cpu_state *st; << 1205 int cpu; << 1206 << 1207 for_each_possible_cpu(cpu) { << 1208 st = per_cpu_ptr(&cpuhp_state << 1209 init_completion(&st->done_up) << 1210 init_completion(&st->done_dow << 1211 } << 1212 } << 1213 << 1214 void __init cpuhp_threads_init(void) 585 void __init cpuhp_threads_init(void) 1215 { 586 { 1216 cpuhp_init_state(); << 1217 BUG_ON(smpboot_register_percpu_thread 587 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); 1218 kthread_unpark(this_cpu_read(cpuhp_st 588 kthread_unpark(this_cpu_read(cpuhp_state.thread)); 1219 } 589 } 1220 590 1221 #ifdef CONFIG_HOTPLUG_CPU 591 #ifdef CONFIG_HOTPLUG_CPU 1222 #ifndef arch_clear_mm_cpumask_cpu << 1223 #define arch_clear_mm_cpumask_cpu(cpu, mm) cp << 1224 #endif << 1225 << 1226 /** 592 /** 1227 * clear_tasks_mm_cpumask - Safely clear task 593 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU 1228 * @cpu: a CPU id 594 * @cpu: a CPU id 1229 * 595 * 1230 * This function walks all processes, finds a 596 * This function walks all processes, finds a valid mm struct for each one and 1231 * then clears a corresponding bit in mm's cp 597 * then clears a corresponding bit in mm's cpumask. While this all sounds 1232 * trivial, there are various non-obvious cor 598 * trivial, there are various non-obvious corner cases, which this function 1233 * tries to solve in a safe manner. 599 * tries to solve in a safe manner. 1234 * 600 * 1235 * Also note that the function uses a somewha 601 * Also note that the function uses a somewhat relaxed locking scheme, so it may 1236 * be called only for an already offlined CPU 602 * be called only for an already offlined CPU. 1237 */ 603 */ 1238 void clear_tasks_mm_cpumask(int cpu) 604 void clear_tasks_mm_cpumask(int cpu) 1239 { 605 { 1240 struct task_struct *p; 606 struct task_struct *p; 1241 607 1242 /* 608 /* 1243 * This function is called after the 609 * This function is called after the cpu is taken down and marked 1244 * offline, so its not like new tasks 610 * offline, so its not like new tasks will ever get this cpu set in 1245 * their mm mask. -- Peter Zijlstra 611 * their mm mask. -- Peter Zijlstra 1246 * Thus, we may use rcu_read_lock() h 612 * Thus, we may use rcu_read_lock() here, instead of grabbing 1247 * full-fledged tasklist_lock. 613 * full-fledged tasklist_lock. 1248 */ 614 */ 1249 WARN_ON(cpu_online(cpu)); 615 WARN_ON(cpu_online(cpu)); 1250 rcu_read_lock(); 616 rcu_read_lock(); 1251 for_each_process(p) { 617 for_each_process(p) { 1252 struct task_struct *t; 618 struct task_struct *t; 1253 619 1254 /* 620 /* 1255 * Main thread might exit, bu 621 * Main thread might exit, but other threads may still have 1256 * a valid mm. Find one. 622 * a valid mm. Find one. 1257 */ 623 */ 1258 t = find_lock_task_mm(p); 624 t = find_lock_task_mm(p); 1259 if (!t) 625 if (!t) 1260 continue; 626 continue; 1261 arch_clear_mm_cpumask_cpu(cpu !! 627 cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); 1262 task_unlock(t); 628 task_unlock(t); 1263 } 629 } 1264 rcu_read_unlock(); 630 rcu_read_unlock(); 1265 } 631 } 1266 632 >> 633 static inline void check_for_tasks(int dead_cpu) >> 634 { >> 635 struct task_struct *g, *p; >> 636 >> 637 read_lock(&tasklist_lock); >> 638 for_each_process_thread(g, p) { >> 639 if (!p->on_rq) >> 640 continue; >> 641 /* >> 642 * We do the check with unlocked task_rq(p)->lock. >> 643 * Order the reading to do not warn about a task, >> 644 * which was running on this cpu in the past, and >> 645 * it's just been woken on another cpu. >> 646 */ >> 647 rmb(); >> 648 if (task_cpu(p) != dead_cpu) >> 649 continue; >> 650 >> 651 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", >> 652 p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); >> 653 } >> 654 read_unlock(&tasklist_lock); >> 655 } >> 656 1267 /* Take this CPU down. */ 657 /* Take this CPU down. */ 1268 static int take_cpu_down(void *_param) 658 static int take_cpu_down(void *_param) 1269 { 659 { 1270 struct cpuhp_cpu_state *st = this_cpu 660 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); 1271 enum cpuhp_state target = max((int)st 661 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); 1272 int err, cpu = smp_processor_id(); 662 int err, cpu = smp_processor_id(); 1273 663 1274 /* Ensure this CPU doesn't handle any 664 /* Ensure this CPU doesn't handle any more interrupts. */ 1275 err = __cpu_disable(); 665 err = __cpu_disable(); 1276 if (err < 0) 666 if (err < 0) 1277 return err; 667 return err; 1278 668 1279 /* 669 /* 1280 * Must be called from CPUHP_TEARDOWN !! 670 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not 1281 * down, that the current state is CP !! 671 * do this step again. 1282 */ << 1283 WARN_ON(st->state != (CPUHP_TEARDOWN_ << 1284 << 1285 /* << 1286 * Invoke the former CPU_DYING callba << 1287 */ 672 */ 1288 cpuhp_invoke_callback_range_nofail(fa !! 673 WARN_ON(st->state != CPUHP_TEARDOWN_CPU); >> 674 st->state--; >> 675 /* Invoke the former CPU_DYING callbacks */ >> 676 for (; st->state > target; st->state--) >> 677 cpuhp_invoke_callback(cpu, st->state, false, NULL); 1289 678 >> 679 /* Give up timekeeping duties */ >> 680 tick_handover_do_timer(); 1290 /* Park the stopper thread */ 681 /* Park the stopper thread */ 1291 stop_machine_park(cpu); 682 stop_machine_park(cpu); 1292 return 0; 683 return 0; 1293 } 684 } 1294 685 1295 static int takedown_cpu(unsigned int cpu) 686 static int takedown_cpu(unsigned int cpu) 1296 { 687 { 1297 struct cpuhp_cpu_state *st = per_cpu_ 688 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 1298 int err; 689 int err; 1299 690 1300 /* Park the smpboot threads */ 691 /* Park the smpboot threads */ 1301 kthread_park(st->thread); !! 692 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); >> 693 smpboot_park_threads(cpu); 1302 694 1303 /* 695 /* 1304 * Prevent irq alloc/free while the d 696 * Prevent irq alloc/free while the dying cpu reorganizes the 1305 * interrupt affinities. 697 * interrupt affinities. 1306 */ 698 */ 1307 irq_lock_sparse(); 699 irq_lock_sparse(); 1308 700 1309 /* 701 /* 1310 * So now all preempt/rcu users must 702 * So now all preempt/rcu users must observe !cpu_active(). 1311 */ 703 */ 1312 err = stop_machine_cpuslocked(take_cp !! 704 err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu)); 1313 if (err) { 705 if (err) { 1314 /* CPU refused to die */ 706 /* CPU refused to die */ 1315 irq_unlock_sparse(); 707 irq_unlock_sparse(); 1316 /* Unpark the hotplug thread 708 /* Unpark the hotplug thread so we can rollback there */ 1317 kthread_unpark(st->thread); !! 709 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); 1318 return err; 710 return err; 1319 } 711 } 1320 BUG_ON(cpu_online(cpu)); 712 BUG_ON(cpu_online(cpu)); 1321 713 1322 /* 714 /* 1323 * The teardown callback for CPUHP_AP !! 715 * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all 1324 * all runnable tasks from the CPU, t !! 716 * runnable tasks from the cpu, there's only the idle task left now 1325 * that the migration thread is done 717 * that the migration thread is done doing the stop_machine thing. 1326 * 718 * 1327 * Wait for the stop thread to go awa 719 * Wait for the stop thread to go away. 1328 */ 720 */ 1329 wait_for_ap_thread(st, false); !! 721 wait_for_completion(&st->done); 1330 BUG_ON(st->state != CPUHP_AP_IDLE_DEA 722 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); 1331 723 1332 /* Interrupts are moved away from the 724 /* Interrupts are moved away from the dying cpu, reenable alloc/free */ 1333 irq_unlock_sparse(); 725 irq_unlock_sparse(); 1334 726 1335 hotplug_cpu__broadcast_tick_pull(cpu) 727 hotplug_cpu__broadcast_tick_pull(cpu); 1336 /* This actually kills the CPU. */ 728 /* This actually kills the CPU. */ 1337 __cpu_die(cpu); 729 __cpu_die(cpu); 1338 730 1339 cpuhp_bp_sync_dead(cpu); << 1340 << 1341 tick_cleanup_dead_cpu(cpu); 731 tick_cleanup_dead_cpu(cpu); 1342 << 1343 /* << 1344 * Callbacks must be re-integrated ri << 1345 * Otherwise an RCU callback could bl << 1346 * waiting for its completion. << 1347 */ << 1348 rcutree_migrate_callbacks(cpu); << 1349 << 1350 return 0; 732 return 0; 1351 } 733 } 1352 734 1353 static void cpuhp_complete_idle_dead(void *ar 735 static void cpuhp_complete_idle_dead(void *arg) 1354 { 736 { 1355 struct cpuhp_cpu_state *st = arg; 737 struct cpuhp_cpu_state *st = arg; 1356 738 1357 complete_ap_thread(st, false); !! 739 complete(&st->done); 1358 } 740 } 1359 741 1360 void cpuhp_report_idle_dead(void) 742 void cpuhp_report_idle_dead(void) 1361 { 743 { 1362 struct cpuhp_cpu_state *st = this_cpu 744 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); 1363 745 1364 BUG_ON(st->state != CPUHP_AP_OFFLINE) 746 BUG_ON(st->state != CPUHP_AP_OFFLINE); 1365 tick_assert_timekeeping_handover(); !! 747 rcu_report_dead(smp_processor_id()); 1366 rcutree_report_cpu_dead(); << 1367 st->state = CPUHP_AP_IDLE_DEAD; 748 st->state = CPUHP_AP_IDLE_DEAD; 1368 /* 749 /* 1369 * We cannot call complete after rcut !! 750 * We cannot call complete after rcu_report_dead() so we delegate it 1370 * to an online cpu. 751 * to an online cpu. 1371 */ 752 */ 1372 smp_call_function_single(cpumask_firs 753 smp_call_function_single(cpumask_first(cpu_online_mask), 1373 cpuhp_comple 754 cpuhp_complete_idle_dead, st, 0); 1374 } 755 } 1375 756 1376 static int cpuhp_down_callbacks(unsigned int !! 757 #else 1377 enum cpuhp_st !! 758 #define takedown_cpu NULL 1378 { !! 759 #endif 1379 enum cpuhp_state prev_state = st->sta << 1380 int ret = 0; << 1381 << 1382 ret = cpuhp_invoke_callback_range(fal << 1383 if (ret) { << 1384 pr_debug("CPU DOWN failed (%d << 1385 ret, cpu, cpuhp_get_ << 1386 st->state); << 1387 << 1388 cpuhp_reset_state(cpu, st, pr << 1389 << 1390 if (st->state < prev_state) << 1391 WARN_ON(cpuhp_invoke_ << 1392 << 1393 } << 1394 760 1395 return ret; !! 761 #ifdef CONFIG_HOTPLUG_CPU 1396 } << 1397 762 1398 /* Requires cpu_add_remove_lock to be held */ 763 /* Requires cpu_add_remove_lock to be held */ 1399 static int __ref _cpu_down(unsigned int cpu, 764 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, 1400 enum cpuhp_state t 765 enum cpuhp_state target) 1401 { 766 { 1402 struct cpuhp_cpu_state *st = per_cpu_ 767 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 1403 int prev_state, ret = 0; 768 int prev_state, ret = 0; 1404 769 1405 if (num_online_cpus() == 1) 770 if (num_online_cpus() == 1) 1406 return -EBUSY; 771 return -EBUSY; 1407 772 1408 if (!cpu_present(cpu)) 773 if (!cpu_present(cpu)) 1409 return -EINVAL; 774 return -EINVAL; 1410 775 1411 cpus_write_lock(); !! 776 cpu_hotplug_begin(); 1412 777 1413 cpuhp_tasks_frozen = tasks_frozen; 778 cpuhp_tasks_frozen = tasks_frozen; 1414 779 1415 prev_state = cpuhp_set_state(cpu, st, !! 780 prev_state = st->state; >> 781 st->target = target; 1416 /* 782 /* 1417 * If the current CPU state is in the 783 * If the current CPU state is in the range of the AP hotplug thread, 1418 * then we need to kick the thread. 784 * then we need to kick the thread. 1419 */ 785 */ 1420 if (st->state > CPUHP_TEARDOWN_CPU) { 786 if (st->state > CPUHP_TEARDOWN_CPU) { 1421 st->target = max((int)target, << 1422 ret = cpuhp_kick_ap_work(cpu) 787 ret = cpuhp_kick_ap_work(cpu); 1423 /* 788 /* 1424 * The AP side has done the e 789 * The AP side has done the error rollback already. Just 1425 * return the error code.. 790 * return the error code.. 1426 */ 791 */ 1427 if (ret) 792 if (ret) 1428 goto out; 793 goto out; 1429 794 1430 /* 795 /* 1431 * We might have stopped stil 796 * We might have stopped still in the range of the AP hotplug 1432 * thread. Nothing to do anym 797 * thread. Nothing to do anymore. 1433 */ 798 */ 1434 if (st->state > CPUHP_TEARDOW 799 if (st->state > CPUHP_TEARDOWN_CPU) 1435 goto out; 800 goto out; 1436 << 1437 st->target = target; << 1438 } 801 } 1439 /* 802 /* 1440 * The AP brought itself down to CPUH 803 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need 1441 * to do the further cleanups. 804 * to do the further cleanups. 1442 */ 805 */ 1443 ret = cpuhp_down_callbacks(cpu, st, t 806 ret = cpuhp_down_callbacks(cpu, st, target); 1444 if (ret && st->state < prev_state) { !! 807 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { 1445 if (st->state == CPUHP_TEARDO !! 808 st->target = prev_state; 1446 cpuhp_reset_state(cpu !! 809 st->rollback = true; 1447 __cpuhp_kick_ap(st); !! 810 cpuhp_kick_ap_work(cpu); 1448 } else { << 1449 WARN(1, "DEAD callbac << 1450 } << 1451 } 811 } 1452 812 1453 out: 813 out: 1454 cpus_write_unlock(); !! 814 cpu_hotplug_done(); 1455 /* << 1456 * Do post unplug cleanup. This is st << 1457 * concurrent CPU hotplug via cpu_add << 1458 */ << 1459 lockup_detector_cleanup(); << 1460 arch_smt_update(); << 1461 return ret; 815 return ret; 1462 } 816 } 1463 817 1464 struct cpu_down_work { !! 818 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) 1465 unsigned int cpu; << 1466 enum cpuhp_state target; << 1467 }; << 1468 << 1469 static long __cpu_down_maps_locked(void *arg) << 1470 { 819 { 1471 struct cpu_down_work *work = arg; !! 820 int err; 1472 << 1473 return _cpu_down(work->cpu, 0, work-> << 1474 } << 1475 << 1476 static int cpu_down_maps_locked(unsigned int << 1477 { << 1478 struct cpu_down_work work = { .cpu = << 1479 821 1480 /* !! 822 cpu_maps_update_begin(); 1481 * If the platform does not support h << 1482 * differentiate it from a transient << 1483 */ << 1484 if (cpu_hotplug_offline_disabled) << 1485 return -EOPNOTSUPP; << 1486 if (cpu_hotplug_disabled) << 1487 return -EBUSY; << 1488 823 1489 /* !! 824 if (cpu_hotplug_disabled) { 1490 * Ensure that the control task does !! 825 err = -EBUSY; 1491 * CPU to prevent a deadlock against !! 826 goto out; 1492 * Also keep at least one housekeepin << 1493 * an empty sched_domain span. << 1494 */ << 1495 for_each_cpu_and(cpu, cpu_online_mask << 1496 if (cpu != work.cpu) << 1497 return work_on_cpu(cp << 1498 } 827 } 1499 return -EBUSY; << 1500 } << 1501 828 1502 static int cpu_down(unsigned int cpu, enum cp !! 829 err = _cpu_down(cpu, 0, target); 1503 { << 1504 int err; << 1505 830 1506 cpu_maps_update_begin(); !! 831 out: 1507 err = cpu_down_maps_locked(cpu, targe << 1508 cpu_maps_update_done(); 832 cpu_maps_update_done(); 1509 return err; 833 return err; 1510 } 834 } 1511 !! 835 int cpu_down(unsigned int cpu) 1512 /** << 1513 * cpu_device_down - Bring down a cpu device << 1514 * @dev: Pointer to the cpu device to offline << 1515 * << 1516 * This function is meant to be used by devic << 1517 * << 1518 * Other subsystems should use remove_cpu() i << 1519 * << 1520 * Return: %0 on success or a negative errno << 1521 */ << 1522 int cpu_device_down(struct device *dev) << 1523 { << 1524 return cpu_down(dev->id, CPUHP_OFFLIN << 1525 } << 1526 << 1527 int remove_cpu(unsigned int cpu) << 1528 { << 1529 int ret; << 1530 << 1531 lock_device_hotplug(); << 1532 ret = device_offline(get_cpu_device(c << 1533 unlock_device_hotplug(); << 1534 << 1535 return ret; << 1536 } << 1537 EXPORT_SYMBOL_GPL(remove_cpu); << 1538 << 1539 void smp_shutdown_nonboot_cpus(unsigned int p << 1540 { 836 { 1541 unsigned int cpu; !! 837 return do_cpu_down(cpu, CPUHP_OFFLINE); 1542 int error; << 1543 << 1544 cpu_maps_update_begin(); << 1545 << 1546 /* << 1547 * Make certain the cpu I'm about to << 1548 * << 1549 * This is inline to what migrate_to_ << 1550 */ << 1551 if (!cpu_online(primary_cpu)) << 1552 primary_cpu = cpumask_first(c << 1553 << 1554 for_each_online_cpu(cpu) { << 1555 if (cpu == primary_cpu) << 1556 continue; << 1557 << 1558 error = cpu_down_maps_locked( << 1559 if (error) { << 1560 pr_err("Failed to off << 1561 cpu, error); << 1562 break; << 1563 } << 1564 } << 1565 << 1566 /* << 1567 * Ensure all but the reboot CPU are << 1568 */ << 1569 BUG_ON(num_online_cpus() > 1); << 1570 << 1571 /* << 1572 * Make sure the CPUs won't be enable << 1573 * point. Kexec will reboot to a new << 1574 * everything along the way. << 1575 */ << 1576 cpu_hotplug_disabled++; << 1577 << 1578 cpu_maps_update_done(); << 1579 } 838 } 1580 !! 839 EXPORT_SYMBOL(cpu_down); 1581 #else << 1582 #define takedown_cpu NULL << 1583 #endif /*CONFIG_HOTPLUG_CPU*/ 840 #endif /*CONFIG_HOTPLUG_CPU*/ 1584 841 1585 /** 842 /** 1586 * notify_cpu_starting(cpu) - Invoke the call 843 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU 1587 * @cpu: cpu that just started 844 * @cpu: cpu that just started 1588 * 845 * 1589 * It must be called by the arch code on the 846 * It must be called by the arch code on the new cpu, before the new cpu 1590 * enables interrupts and before the "boot" c 847 * enables interrupts and before the "boot" cpu returns from __cpu_up(). 1591 */ 848 */ 1592 void notify_cpu_starting(unsigned int cpu) 849 void notify_cpu_starting(unsigned int cpu) 1593 { 850 { 1594 struct cpuhp_cpu_state *st = per_cpu_ 851 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 1595 enum cpuhp_state target = min((int)st 852 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); 1596 853 1597 rcutree_report_cpu_starting(cpu); !! 854 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ 1598 cpumask_set_cpu(cpu, &cpus_booted_onc !! 855 while (st->state < target) { 1599 !! 856 st->state++; 1600 /* !! 857 cpuhp_invoke_callback(cpu, st->state, true, NULL); 1601 * STARTING must not fail! !! 858 } 1602 */ << 1603 cpuhp_invoke_callback_range_nofail(tr << 1604 } 859 } 1605 860 1606 /* 861 /* 1607 * Called from the idle task. Wake up the con !! 862 * Called from the idle task. We need to set active here, so we can kick off 1608 * hotplug thread of the upcoming CPU up and !! 863 * the stopper thread and unpark the smpboot threads. If the target state is 1609 * online bringup to the hotplug thread. !! 864 * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the >> 865 * cpu further. 1610 */ 866 */ 1611 void cpuhp_online_idle(enum cpuhp_state state 867 void cpuhp_online_idle(enum cpuhp_state state) 1612 { 868 { 1613 struct cpuhp_cpu_state *st = this_cpu 869 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); >> 870 unsigned int cpu = smp_processor_id(); 1614 871 1615 /* Happens for the boot cpu */ 872 /* Happens for the boot cpu */ 1616 if (state != CPUHP_AP_ONLINE_IDLE) 873 if (state != CPUHP_AP_ONLINE_IDLE) 1617 return; 874 return; 1618 875 1619 cpuhp_ap_update_sync_state(SYNC_STATE !! 876 st->state = CPUHP_AP_ONLINE_IDLE; 1620 877 1621 /* !! 878 /* Unpark the stopper thread and the hotplug thread of this cpu */ 1622 * Unpark the stopper thread before w !! 879 stop_machine_unpark(cpu); 1623 * scheduling); this ensures the stop !! 880 kthread_unpark(st->thread); 1624 */ << 1625 stop_machine_unpark(smp_processor_id( << 1626 881 1627 st->state = CPUHP_AP_ONLINE_IDLE; !! 882 /* Should we go further up ? */ 1628 complete_ap_thread(st, true); !! 883 if (st->target > CPUHP_AP_ONLINE_IDLE) >> 884 __cpuhp_kick_ap_work(st); >> 885 else >> 886 complete(&st->done); 1629 } 887 } 1630 888 1631 /* Requires cpu_add_remove_lock to be held */ 889 /* Requires cpu_add_remove_lock to be held */ 1632 static int _cpu_up(unsigned int cpu, int task 890 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) 1633 { 891 { 1634 struct cpuhp_cpu_state *st = per_cpu_ 892 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 1635 struct task_struct *idle; 893 struct task_struct *idle; 1636 int ret = 0; 894 int ret = 0; 1637 895 1638 cpus_write_lock(); !! 896 cpu_hotplug_begin(); 1639 897 1640 if (!cpu_present(cpu)) { 898 if (!cpu_present(cpu)) { 1641 ret = -EINVAL; 899 ret = -EINVAL; 1642 goto out; 900 goto out; 1643 } 901 } 1644 902 1645 /* 903 /* 1646 * The caller of cpu_up() might have !! 904 * The caller of do_cpu_up might have raced with another 1647 * caller. Nothing to do. !! 905 * caller. Ignore it for now. 1648 */ 906 */ 1649 if (st->state >= target) 907 if (st->state >= target) 1650 goto out; 908 goto out; 1651 909 1652 if (st->state == CPUHP_OFFLINE) { 910 if (st->state == CPUHP_OFFLINE) { 1653 /* Let it fail before we try 911 /* Let it fail before we try to bring the cpu up */ 1654 idle = idle_thread_get(cpu); 912 idle = idle_thread_get(cpu); 1655 if (IS_ERR(idle)) { 913 if (IS_ERR(idle)) { 1656 ret = PTR_ERR(idle); 914 ret = PTR_ERR(idle); 1657 goto out; 915 goto out; 1658 } 916 } 1659 << 1660 /* << 1661 * Reset stale stack state fr << 1662 */ << 1663 scs_task_reset(idle); << 1664 kasan_unpoison_task_stack(idl << 1665 } 917 } 1666 918 1667 cpuhp_tasks_frozen = tasks_frozen; 919 cpuhp_tasks_frozen = tasks_frozen; 1668 920 1669 cpuhp_set_state(cpu, st, target); !! 921 st->target = target; 1670 /* 922 /* 1671 * If the current CPU state is in the 923 * If the current CPU state is in the range of the AP hotplug thread, 1672 * then we need to kick the thread on 924 * then we need to kick the thread once more. 1673 */ 925 */ 1674 if (st->state > CPUHP_BRINGUP_CPU) { 926 if (st->state > CPUHP_BRINGUP_CPU) { 1675 ret = cpuhp_kick_ap_work(cpu) 927 ret = cpuhp_kick_ap_work(cpu); 1676 /* 928 /* 1677 * The AP side has done the e 929 * The AP side has done the error rollback already. Just 1678 * return the error code.. 930 * return the error code.. 1679 */ 931 */ 1680 if (ret) 932 if (ret) 1681 goto out; 933 goto out; 1682 } 934 } 1683 935 1684 /* 936 /* 1685 * Try to reach the target state. We 937 * Try to reach the target state. We max out on the BP at 1686 * CPUHP_BRINGUP_CPU. After that the 938 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is 1687 * responsible for bringing it up to 939 * responsible for bringing it up to the target state. 1688 */ 940 */ 1689 target = min((int)target, CPUHP_BRING 941 target = min((int)target, CPUHP_BRINGUP_CPU); 1690 ret = cpuhp_up_callbacks(cpu, st, tar 942 ret = cpuhp_up_callbacks(cpu, st, target); 1691 out: 943 out: 1692 cpus_write_unlock(); !! 944 cpu_hotplug_done(); 1693 arch_smt_update(); << 1694 return ret; 945 return ret; 1695 } 946 } 1696 947 1697 static int cpu_up(unsigned int cpu, enum cpuh !! 948 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) 1698 { 949 { 1699 int err = 0; 950 int err = 0; 1700 951 1701 if (!cpu_possible(cpu)) { 952 if (!cpu_possible(cpu)) { 1702 pr_err("can't online cpu %d b 953 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", 1703 cpu); 954 cpu); >> 955 #if defined(CONFIG_IA64) >> 956 pr_err("please check additional_cpus= boot parameter\n"); >> 957 #endif 1704 return -EINVAL; 958 return -EINVAL; 1705 } 959 } 1706 960 1707 err = try_online_node(cpu_to_node(cpu 961 err = try_online_node(cpu_to_node(cpu)); 1708 if (err) 962 if (err) 1709 return err; 963 return err; 1710 964 1711 cpu_maps_update_begin(); 965 cpu_maps_update_begin(); 1712 966 1713 if (cpu_hotplug_disabled) { 967 if (cpu_hotplug_disabled) { 1714 err = -EBUSY; 968 err = -EBUSY; 1715 goto out; 969 goto out; 1716 } 970 } 1717 if (!cpu_bootable(cpu)) { << 1718 err = -EPERM; << 1719 goto out; << 1720 } << 1721 971 1722 err = _cpu_up(cpu, 0, target); 972 err = _cpu_up(cpu, 0, target); 1723 out: 973 out: 1724 cpu_maps_update_done(); 974 cpu_maps_update_done(); 1725 return err; 975 return err; 1726 } 976 } 1727 977 1728 /** !! 978 int cpu_up(unsigned int cpu) 1729 * cpu_device_up - Bring up a cpu device << 1730 * @dev: Pointer to the cpu device to online << 1731 * << 1732 * This function is meant to be used by devic << 1733 * << 1734 * Other subsystems should use add_cpu() inst << 1735 * << 1736 * Return: %0 on success or a negative errno << 1737 */ << 1738 int cpu_device_up(struct device *dev) << 1739 { << 1740 return cpu_up(dev->id, CPUHP_ONLINE); << 1741 } << 1742 << 1743 int add_cpu(unsigned int cpu) << 1744 { 979 { 1745 int ret; !! 980 return do_cpu_up(cpu, CPUHP_ONLINE); 1746 << 1747 lock_device_hotplug(); << 1748 ret = device_online(get_cpu_device(cp << 1749 unlock_device_hotplug(); << 1750 << 1751 return ret; << 1752 } << 1753 EXPORT_SYMBOL_GPL(add_cpu); << 1754 << 1755 /** << 1756 * bringup_hibernate_cpu - Bring up the CPU t << 1757 * @sleep_cpu: The cpu we hibernated on and s << 1758 * << 1759 * On some architectures like arm64, we can h << 1760 * wake up the CPU we hibernated on might be << 1761 * using maxcpus= for example. << 1762 * << 1763 * Return: %0 on success or a negative errno << 1764 */ << 1765 int bringup_hibernate_cpu(unsigned int sleep_ << 1766 { << 1767 int ret; << 1768 << 1769 if (!cpu_online(sleep_cpu)) { << 1770 pr_info("Hibernated on a CPU << 1771 ret = cpu_up(sleep_cpu, CPUHP << 1772 if (ret) { << 1773 pr_err("Failed to bri << 1774 return ret; << 1775 } << 1776 } << 1777 return 0; << 1778 } << 1779 << 1780 static void __init cpuhp_bringup_mask(const s << 1781 enum cp << 1782 { << 1783 unsigned int cpu; << 1784 << 1785 for_each_cpu(cpu, mask) { << 1786 struct cpuhp_cpu_state *st = << 1787 << 1788 if (cpu_up(cpu, target) && ca << 1789 /* << 1790 * If this failed the << 1791 * rolled back to CPU << 1792 * online. Clean it u << 1793 */ << 1794 WARN_ON(cpuhp_invoke_ << 1795 } << 1796 << 1797 if (!--ncpus) << 1798 break; << 1799 } << 1800 } << 1801 << 1802 #ifdef CONFIG_HOTPLUG_PARALLEL << 1803 static bool __cpuhp_parallel_bringup __ro_aft << 1804 << 1805 static int __init parallel_bringup_parse_para << 1806 { << 1807 return kstrtobool(arg, &__cpuhp_paral << 1808 } << 1809 early_param("cpuhp.parallel", parallel_bringu << 1810 << 1811 #ifdef CONFIG_HOTPLUG_SMT << 1812 static inline bool cpuhp_smt_aware(void) << 1813 { << 1814 return cpu_smt_max_threads > 1; << 1815 } << 1816 << 1817 static inline const struct cpumask *cpuhp_get << 1818 { << 1819 return cpu_primary_thread_mask; << 1820 } << 1821 #else << 1822 static inline bool cpuhp_smt_aware(void) << 1823 { << 1824 return false; << 1825 } << 1826 static inline const struct cpumask *cpuhp_get << 1827 { << 1828 return cpu_none_mask; << 1829 } << 1830 #endif << 1831 << 1832 bool __weak arch_cpuhp_init_parallel_bringup( << 1833 { << 1834 return true; << 1835 } << 1836 << 1837 /* << 1838 * On architectures which have enabled parall << 1839 * prepare states for each of the to be onlin << 1840 * sends the startup IPI to the APs. The APs << 1841 * bringup code in parallel and then wait for << 1842 * them one by one for the final onlining pro << 1843 * << 1844 * This avoids waiting for each AP to respond << 1845 * CPUHP_BRINGUP_CPU. << 1846 */ << 1847 static bool __init cpuhp_bringup_cpus_paralle << 1848 { << 1849 const struct cpumask *mask = cpu_pres << 1850 << 1851 if (__cpuhp_parallel_bringup) << 1852 __cpuhp_parallel_bringup = ar << 1853 if (!__cpuhp_parallel_bringup) << 1854 return false; << 1855 << 1856 if (cpuhp_smt_aware()) { << 1857 const struct cpumask *pmask = << 1858 static struct cpumask tmp_mas << 1859 << 1860 /* << 1861 * X86 requires to prevent th << 1862 * the primary thread does a << 1863 * reasons. Bring the primary << 1864 */ << 1865 cpumask_and(&tmp_mask, mask, << 1866 cpuhp_bringup_mask(&tmp_mask, << 1867 cpuhp_bringup_mask(&tmp_mask, << 1868 /* Account for the online CPU << 1869 ncpus -= num_online_cpus(); << 1870 if (!ncpus) << 1871 return true; << 1872 /* Create the mask for second << 1873 cpumask_andnot(&tmp_mask, mas << 1874 mask = &tmp_mask; << 1875 } << 1876 << 1877 /* Bring the not-yet started CPUs up << 1878 cpuhp_bringup_mask(mask, ncpus, CPUHP << 1879 cpuhp_bringup_mask(mask, ncpus, CPUHP << 1880 return true; << 1881 } << 1882 #else << 1883 static inline bool cpuhp_bringup_cpus_paralle << 1884 #endif /* CONFIG_HOTPLUG_PARALLEL */ << 1885 << 1886 void __init bringup_nonboot_cpus(unsigned int << 1887 { << 1888 if (!max_cpus) << 1889 return; << 1890 << 1891 /* Try parallel bringup optimization << 1892 if (cpuhp_bringup_cpus_parallel(max_c << 1893 return; << 1894 << 1895 /* Full per CPU serialized bringup */ << 1896 cpuhp_bringup_mask(cpu_present_mask, << 1897 } 981 } >> 982 EXPORT_SYMBOL_GPL(cpu_up); 1898 983 1899 #ifdef CONFIG_PM_SLEEP_SMP 984 #ifdef CONFIG_PM_SLEEP_SMP 1900 static cpumask_var_t frozen_cpus; 985 static cpumask_var_t frozen_cpus; 1901 986 1902 int freeze_secondary_cpus(int primary) 987 int freeze_secondary_cpus(int primary) 1903 { 988 { 1904 int cpu, error = 0; 989 int cpu, error = 0; 1905 990 1906 cpu_maps_update_begin(); 991 cpu_maps_update_begin(); 1907 if (primary == -1) { !! 992 if (!cpu_online(primary)) 1908 primary = cpumask_first(cpu_o 993 primary = cpumask_first(cpu_online_mask); 1909 if (!housekeeping_cpu(primary << 1910 primary = housekeepin << 1911 } else { << 1912 if (!cpu_online(primary)) << 1913 primary = cpumask_fir << 1914 } << 1915 << 1916 /* 994 /* 1917 * We take down all of the non-boot C 995 * We take down all of the non-boot CPUs in one shot to avoid races 1918 * with the userspace trying to use t 996 * with the userspace trying to use the CPU hotplug at the same time 1919 */ 997 */ 1920 cpumask_clear(frozen_cpus); 998 cpumask_clear(frozen_cpus); 1921 999 1922 pr_info("Disabling non-boot CPUs ...\ 1000 pr_info("Disabling non-boot CPUs ...\n"); 1923 for (cpu = nr_cpu_ids - 1; cpu >= 0; !! 1001 for_each_online_cpu(cpu) { 1924 if (!cpu_online(cpu) || cpu = !! 1002 if (cpu == primary) 1925 continue; 1003 continue; 1926 << 1927 if (pm_wakeup_pending()) { << 1928 pr_info("Wakeup pendi << 1929 error = -EBUSY; << 1930 break; << 1931 } << 1932 << 1933 trace_suspend_resume(TPS("CPU 1004 trace_suspend_resume(TPS("CPU_OFF"), cpu, true); 1934 error = _cpu_down(cpu, 1, CPU 1005 error = _cpu_down(cpu, 1, CPUHP_OFFLINE); 1935 trace_suspend_resume(TPS("CPU 1006 trace_suspend_resume(TPS("CPU_OFF"), cpu, false); 1936 if (!error) 1007 if (!error) 1937 cpumask_set_cpu(cpu, 1008 cpumask_set_cpu(cpu, frozen_cpus); 1938 else { 1009 else { 1939 pr_err("Error taking 1010 pr_err("Error taking CPU%d down: %d\n", cpu, error); 1940 break; 1011 break; 1941 } 1012 } 1942 } 1013 } 1943 1014 1944 if (!error) 1015 if (!error) 1945 BUG_ON(num_online_cpus() > 1) 1016 BUG_ON(num_online_cpus() > 1); 1946 else 1017 else 1947 pr_err("Non-boot CPUs are not 1018 pr_err("Non-boot CPUs are not disabled\n"); 1948 1019 1949 /* 1020 /* 1950 * Make sure the CPUs won't be enable 1021 * Make sure the CPUs won't be enabled by someone else. We need to do 1951 * this even in case of failure as al !! 1022 * this even in case of failure as all disable_nonboot_cpus() users are 1952 * supposed to do thaw_secondary_cpus !! 1023 * supposed to do enable_nonboot_cpus() on the failure path. 1953 */ 1024 */ 1954 cpu_hotplug_disabled++; 1025 cpu_hotplug_disabled++; 1955 1026 1956 cpu_maps_update_done(); 1027 cpu_maps_update_done(); 1957 return error; 1028 return error; 1958 } 1029 } 1959 1030 1960 void __weak arch_thaw_secondary_cpus_begin(vo !! 1031 void __weak arch_enable_nonboot_cpus_begin(void) 1961 { 1032 { 1962 } 1033 } 1963 1034 1964 void __weak arch_thaw_secondary_cpus_end(void !! 1035 void __weak arch_enable_nonboot_cpus_end(void) 1965 { 1036 { 1966 } 1037 } 1967 1038 1968 void thaw_secondary_cpus(void) !! 1039 void enable_nonboot_cpus(void) 1969 { 1040 { 1970 int cpu, error; 1041 int cpu, error; 1971 1042 1972 /* Allow everyone to use the CPU hotp 1043 /* Allow everyone to use the CPU hotplug again */ 1973 cpu_maps_update_begin(); 1044 cpu_maps_update_begin(); 1974 __cpu_hotplug_enable(); 1045 __cpu_hotplug_enable(); 1975 if (cpumask_empty(frozen_cpus)) 1046 if (cpumask_empty(frozen_cpus)) 1976 goto out; 1047 goto out; 1977 1048 1978 pr_info("Enabling non-boot CPUs ...\n 1049 pr_info("Enabling non-boot CPUs ...\n"); 1979 1050 1980 arch_thaw_secondary_cpus_begin(); !! 1051 arch_enable_nonboot_cpus_begin(); 1981 1052 1982 for_each_cpu(cpu, frozen_cpus) { 1053 for_each_cpu(cpu, frozen_cpus) { 1983 trace_suspend_resume(TPS("CPU 1054 trace_suspend_resume(TPS("CPU_ON"), cpu, true); 1984 error = _cpu_up(cpu, 1, CPUHP 1055 error = _cpu_up(cpu, 1, CPUHP_ONLINE); 1985 trace_suspend_resume(TPS("CPU 1056 trace_suspend_resume(TPS("CPU_ON"), cpu, false); 1986 if (!error) { 1057 if (!error) { 1987 pr_info("CPU%d is up\ 1058 pr_info("CPU%d is up\n", cpu); 1988 continue; 1059 continue; 1989 } 1060 } 1990 pr_warn("Error taking CPU%d u 1061 pr_warn("Error taking CPU%d up: %d\n", cpu, error); 1991 } 1062 } 1992 1063 1993 arch_thaw_secondary_cpus_end(); !! 1064 arch_enable_nonboot_cpus_end(); 1994 1065 1995 cpumask_clear(frozen_cpus); 1066 cpumask_clear(frozen_cpus); 1996 out: 1067 out: 1997 cpu_maps_update_done(); 1068 cpu_maps_update_done(); 1998 } 1069 } 1999 1070 2000 static int __init alloc_frozen_cpus(void) 1071 static int __init alloc_frozen_cpus(void) 2001 { 1072 { 2002 if (!alloc_cpumask_var(&frozen_cpus, 1073 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) 2003 return -ENOMEM; 1074 return -ENOMEM; 2004 return 0; 1075 return 0; 2005 } 1076 } 2006 core_initcall(alloc_frozen_cpus); 1077 core_initcall(alloc_frozen_cpus); 2007 1078 2008 /* 1079 /* 2009 * When callbacks for CPU hotplug notificatio 1080 * When callbacks for CPU hotplug notifications are being executed, we must 2010 * ensure that the state of the system with r 1081 * ensure that the state of the system with respect to the tasks being frozen 2011 * or not, as reported by the notification, r 1082 * or not, as reported by the notification, remains unchanged *throughout the 2012 * duration* of the execution of the callback 1083 * duration* of the execution of the callbacks. 2013 * Hence we need to prevent the freezer from 1084 * Hence we need to prevent the freezer from racing with regular CPU hotplug. 2014 * 1085 * 2015 * This synchronization is implemented by mut 1086 * This synchronization is implemented by mutually excluding regular CPU 2016 * hotplug and Suspend/Hibernate call paths b 1087 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ 2017 * Hibernate notifications. 1088 * Hibernate notifications. 2018 */ 1089 */ 2019 static int 1090 static int 2020 cpu_hotplug_pm_callback(struct notifier_block 1091 cpu_hotplug_pm_callback(struct notifier_block *nb, 2021 unsigned long action, 1092 unsigned long action, void *ptr) 2022 { 1093 { 2023 switch (action) { 1094 switch (action) { 2024 1095 2025 case PM_SUSPEND_PREPARE: 1096 case PM_SUSPEND_PREPARE: 2026 case PM_HIBERNATION_PREPARE: 1097 case PM_HIBERNATION_PREPARE: 2027 cpu_hotplug_disable(); 1098 cpu_hotplug_disable(); 2028 break; 1099 break; 2029 1100 2030 case PM_POST_SUSPEND: 1101 case PM_POST_SUSPEND: 2031 case PM_POST_HIBERNATION: 1102 case PM_POST_HIBERNATION: 2032 cpu_hotplug_enable(); 1103 cpu_hotplug_enable(); 2033 break; 1104 break; 2034 1105 2035 default: 1106 default: 2036 return NOTIFY_DONE; 1107 return NOTIFY_DONE; 2037 } 1108 } 2038 1109 2039 return NOTIFY_OK; 1110 return NOTIFY_OK; 2040 } 1111 } 2041 1112 2042 1113 2043 static int __init cpu_hotplug_pm_sync_init(vo 1114 static int __init cpu_hotplug_pm_sync_init(void) 2044 { 1115 { 2045 /* 1116 /* 2046 * cpu_hotplug_pm_callback has higher 1117 * cpu_hotplug_pm_callback has higher priority than x86 2047 * bsp_pm_callback which depends on c 1118 * bsp_pm_callback which depends on cpu_hotplug_pm_callback 2048 * to disable cpu hotplug to avoid cp 1119 * to disable cpu hotplug to avoid cpu hotplug race. 2049 */ 1120 */ 2050 pm_notifier(cpu_hotplug_pm_callback, 1121 pm_notifier(cpu_hotplug_pm_callback, 0); 2051 return 0; 1122 return 0; 2052 } 1123 } 2053 core_initcall(cpu_hotplug_pm_sync_init); 1124 core_initcall(cpu_hotplug_pm_sync_init); 2054 1125 2055 #endif /* CONFIG_PM_SLEEP_SMP */ 1126 #endif /* CONFIG_PM_SLEEP_SMP */ 2056 1127 2057 int __boot_cpu_id; << 2058 << 2059 #endif /* CONFIG_SMP */ 1128 #endif /* CONFIG_SMP */ 2060 1129 2061 /* Boot processor state steps */ 1130 /* Boot processor state steps */ 2062 static struct cpuhp_step cpuhp_hp_states[] = !! 1131 static struct cpuhp_step cpuhp_bp_states[] = { 2063 [CPUHP_OFFLINE] = { 1132 [CPUHP_OFFLINE] = { 2064 .name = "of 1133 .name = "offline", 2065 .startup.single = NUL 1134 .startup.single = NULL, 2066 .teardown.single = NUL 1135 .teardown.single = NULL, 2067 }, 1136 }, 2068 #ifdef CONFIG_SMP 1137 #ifdef CONFIG_SMP 2069 [CPUHP_CREATE_THREADS]= { 1138 [CPUHP_CREATE_THREADS]= { 2070 .name = "th 1139 .name = "threads:prepare", 2071 .startup.single = smp 1140 .startup.single = smpboot_create_threads, 2072 .teardown.single = NUL 1141 .teardown.single = NULL, 2073 .cant_stop = tru 1142 .cant_stop = true, 2074 }, 1143 }, 2075 [CPUHP_PERF_PREPARE] = { 1144 [CPUHP_PERF_PREPARE] = { 2076 .name = "pe 1145 .name = "perf:prepare", 2077 .startup.single = per 1146 .startup.single = perf_event_init_cpu, 2078 .teardown.single = per 1147 .teardown.single = perf_event_exit_cpu, 2079 }, 1148 }, 2080 [CPUHP_RANDOM_PREPARE] = { << 2081 .name = "ra << 2082 .startup.single = ran << 2083 .teardown.single = NUL << 2084 }, << 2085 [CPUHP_WORKQUEUE_PREP] = { 1149 [CPUHP_WORKQUEUE_PREP] = { 2086 .name = "wo 1150 .name = "workqueue:prepare", 2087 .startup.single = wor 1151 .startup.single = workqueue_prepare_cpu, 2088 .teardown.single = NUL 1152 .teardown.single = NULL, 2089 }, 1153 }, 2090 [CPUHP_HRTIMERS_PREPARE] = { 1154 [CPUHP_HRTIMERS_PREPARE] = { 2091 .name = "hr 1155 .name = "hrtimers:prepare", 2092 .startup.single = hrt 1156 .startup.single = hrtimers_prepare_cpu, 2093 .teardown.single = NUL !! 1157 .teardown.single = hrtimers_dead_cpu, 2094 }, 1158 }, 2095 [CPUHP_SMPCFD_PREPARE] = { 1159 [CPUHP_SMPCFD_PREPARE] = { 2096 .name = "sm 1160 .name = "smpcfd:prepare", 2097 .startup.single = smp 1161 .startup.single = smpcfd_prepare_cpu, 2098 .teardown.single = smp 1162 .teardown.single = smpcfd_dead_cpu, 2099 }, 1163 }, 2100 [CPUHP_RELAY_PREPARE] = { 1164 [CPUHP_RELAY_PREPARE] = { 2101 .name = "re 1165 .name = "relay:prepare", 2102 .startup.single = rel 1166 .startup.single = relay_prepare_cpu, 2103 .teardown.single = NUL 1167 .teardown.single = NULL, 2104 }, 1168 }, >> 1169 [CPUHP_SLAB_PREPARE] = { >> 1170 .name = "slab:prepare", >> 1171 .startup.single = slab_prepare_cpu, >> 1172 .teardown.single = slab_dead_cpu, >> 1173 }, 2105 [CPUHP_RCUTREE_PREP] = { 1174 [CPUHP_RCUTREE_PREP] = { 2106 .name = "RC 1175 .name = "RCU/tree:prepare", 2107 .startup.single = rcu 1176 .startup.single = rcutree_prepare_cpu, 2108 .teardown.single = rcu 1177 .teardown.single = rcutree_dead_cpu, 2109 }, 1178 }, 2110 /* 1179 /* 2111 * On the tear-down path, timers_dead 1180 * On the tear-down path, timers_dead_cpu() must be invoked 2112 * before blk_mq_queue_reinit_notify( 1181 * before blk_mq_queue_reinit_notify() from notify_dead(), 2113 * otherwise a RCU stall occurs. 1182 * otherwise a RCU stall occurs. 2114 */ 1183 */ 2115 [CPUHP_TIMERS_PREPARE] = { !! 1184 [CPUHP_TIMERS_DEAD] = { 2116 .name = "ti !! 1185 .name = "timers:dead", 2117 .startup.single = tim !! 1186 .startup.single = NULL, 2118 .teardown.single = tim 1187 .teardown.single = timers_dead_cpu, 2119 }, 1188 }, 2120 !! 1189 /* Kicks the plugged cpu into life */ 2121 #ifdef CONFIG_HOTPLUG_SPLIT_STARTUP << 2122 /* << 2123 * Kicks the AP alive. AP will wait i << 2124 * the next step will release it. << 2125 */ << 2126 [CPUHP_BP_KICK_AP] = { << 2127 .name = "cp << 2128 .startup.single = cpu << 2129 }, << 2130 << 2131 /* << 2132 * Waits for the AP to reach cpuhp_ap << 2133 * releases it for the complete bring << 2134 */ << 2135 [CPUHP_BRINGUP_CPU] = { 1190 [CPUHP_BRINGUP_CPU] = { 2136 .name = "cp 1191 .name = "cpu:bringup", 2137 .startup.single = cpu !! 1192 .startup.single = bringup_cpu, 2138 .teardown.single = fin !! 1193 .teardown.single = NULL, 2139 .cant_stop = tru 1194 .cant_stop = true, 2140 }, 1195 }, 2141 #else !! 1196 [CPUHP_AP_SMPCFD_DYING] = { >> 1197 .name = "smpcfd:dying", >> 1198 .startup.single = NULL, >> 1199 .teardown.single = smpcfd_dying_cpu, >> 1200 }, 2142 /* 1201 /* 2143 * All-in-one CPU bringup state which !! 1202 * Handled on controll processor until the plugged processor manages >> 1203 * this itself. 2144 */ 1204 */ 2145 [CPUHP_BRINGUP_CPU] = { !! 1205 [CPUHP_TEARDOWN_CPU] = { 2146 .name = "cp !! 1206 .name = "cpu:teardown", 2147 .startup.single = bri !! 1207 .startup.single = NULL, 2148 .teardown.single = fin !! 1208 .teardown.single = takedown_cpu, 2149 .cant_stop = tru 1209 .cant_stop = true, 2150 }, 1210 }, >> 1211 #else >> 1212 [CPUHP_BRINGUP_CPU] = { }, 2151 #endif 1213 #endif >> 1214 }; >> 1215 >> 1216 /* Application processor state steps */ >> 1217 static struct cpuhp_step cpuhp_ap_states[] = { >> 1218 #ifdef CONFIG_SMP 2152 /* Final state before CPU kills itsel 1219 /* Final state before CPU kills itself */ 2153 [CPUHP_AP_IDLE_DEAD] = { 1220 [CPUHP_AP_IDLE_DEAD] = { 2154 .name = "id 1221 .name = "idle:dead", 2155 }, 1222 }, 2156 /* 1223 /* 2157 * Last state before CPU enters the i 1224 * Last state before CPU enters the idle loop to die. Transient state 2158 * for synchronization. 1225 * for synchronization. 2159 */ 1226 */ 2160 [CPUHP_AP_OFFLINE] = { 1227 [CPUHP_AP_OFFLINE] = { 2161 .name = "ap 1228 .name = "ap:offline", 2162 .cant_stop = tru 1229 .cant_stop = true, 2163 }, 1230 }, 2164 /* First state is scheduler control. 1231 /* First state is scheduler control. Interrupts are disabled */ 2165 [CPUHP_AP_SCHED_STARTING] = { 1232 [CPUHP_AP_SCHED_STARTING] = { 2166 .name = "sc 1233 .name = "sched:starting", 2167 .startup.single = sch 1234 .startup.single = sched_cpu_starting, 2168 .teardown.single = sch 1235 .teardown.single = sched_cpu_dying, 2169 }, 1236 }, 2170 [CPUHP_AP_RCUTREE_DYING] = { 1237 [CPUHP_AP_RCUTREE_DYING] = { 2171 .name = "RC 1238 .name = "RCU/tree:dying", 2172 .startup.single = NUL 1239 .startup.single = NULL, 2173 .teardown.single = rcu 1240 .teardown.single = rcutree_dying_cpu, 2174 }, 1241 }, 2175 [CPUHP_AP_SMPCFD_DYING] = { << 2176 .name = "sm << 2177 .startup.single = NUL << 2178 .teardown.single = smp << 2179 }, << 2180 [CPUHP_AP_HRTIMERS_DYING] = { << 2181 .name = "hr << 2182 .startup.single = NUL << 2183 .teardown.single = hrt << 2184 }, << 2185 [CPUHP_AP_TICK_DYING] = { << 2186 .name = "ti << 2187 .startup.single = NUL << 2188 .teardown.single = tic << 2189 }, << 2190 /* Entry state on starting. Interrupt 1242 /* Entry state on starting. Interrupts enabled from here on. Transient 2191 * state for synchronsization */ 1243 * state for synchronsization */ 2192 [CPUHP_AP_ONLINE] = { 1244 [CPUHP_AP_ONLINE] = { 2193 .name = "ap 1245 .name = "ap:online", 2194 }, 1246 }, 2195 /* << 2196 * Handled on control processor until << 2197 * this itself. << 2198 */ << 2199 [CPUHP_TEARDOWN_CPU] = { << 2200 .name = "cp << 2201 .startup.single = NUL << 2202 .teardown.single = tak << 2203 .cant_stop = tru << 2204 }, << 2205 << 2206 [CPUHP_AP_SCHED_WAIT_EMPTY] = { << 2207 .name = "sc << 2208 .startup.single = NUL << 2209 .teardown.single = sch << 2210 }, << 2211 << 2212 /* Handle smpboot threads park/unpark 1247 /* Handle smpboot threads park/unpark */ 2213 [CPUHP_AP_SMPBOOT_THREADS] = { 1248 [CPUHP_AP_SMPBOOT_THREADS] = { 2214 .name = "sm 1249 .name = "smpboot/threads:online", 2215 .startup.single = smp 1250 .startup.single = smpboot_unpark_threads, 2216 .teardown.single = smp << 2217 }, << 2218 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = { << 2219 .name = "ir << 2220 .startup.single = irq << 2221 .teardown.single = NUL 1251 .teardown.single = NULL, 2222 }, 1252 }, 2223 [CPUHP_AP_PERF_ONLINE] = { 1253 [CPUHP_AP_PERF_ONLINE] = { 2224 .name = "pe 1254 .name = "perf:online", 2225 .startup.single = per 1255 .startup.single = perf_event_init_cpu, 2226 .teardown.single = per 1256 .teardown.single = perf_event_exit_cpu, 2227 }, 1257 }, 2228 [CPUHP_AP_WATCHDOG_ONLINE] = { << 2229 .name = "lo << 2230 .startup.single = loc << 2231 .teardown.single = loc << 2232 }, << 2233 [CPUHP_AP_WORKQUEUE_ONLINE] = { 1258 [CPUHP_AP_WORKQUEUE_ONLINE] = { 2234 .name = "wo 1259 .name = "workqueue:online", 2235 .startup.single = wor 1260 .startup.single = workqueue_online_cpu, 2236 .teardown.single = wor 1261 .teardown.single = workqueue_offline_cpu, 2237 }, 1262 }, 2238 [CPUHP_AP_RANDOM_ONLINE] = { << 2239 .name = "ra << 2240 .startup.single = ran << 2241 .teardown.single = NUL << 2242 }, << 2243 [CPUHP_AP_RCUTREE_ONLINE] = { 1263 [CPUHP_AP_RCUTREE_ONLINE] = { 2244 .name = "RC 1264 .name = "RCU/tree:online", 2245 .startup.single = rcu 1265 .startup.single = rcutree_online_cpu, 2246 .teardown.single = rcu 1266 .teardown.single = rcutree_offline_cpu, 2247 }, 1267 }, 2248 #endif 1268 #endif 2249 /* 1269 /* 2250 * The dynamically registered state s 1270 * The dynamically registered state space is here 2251 */ 1271 */ 2252 1272 2253 #ifdef CONFIG_SMP 1273 #ifdef CONFIG_SMP 2254 /* Last state is scheduler control se 1274 /* Last state is scheduler control setting the cpu active */ 2255 [CPUHP_AP_ACTIVE] = { 1275 [CPUHP_AP_ACTIVE] = { 2256 .name = "sc 1276 .name = "sched:active", 2257 .startup.single = sch 1277 .startup.single = sched_cpu_activate, 2258 .teardown.single = sch 1278 .teardown.single = sched_cpu_deactivate, 2259 }, 1279 }, 2260 #endif 1280 #endif 2261 1281 2262 /* CPU is fully up and running. */ 1282 /* CPU is fully up and running. */ 2263 [CPUHP_ONLINE] = { 1283 [CPUHP_ONLINE] = { 2264 .name = "on 1284 .name = "online", 2265 .startup.single = NUL 1285 .startup.single = NULL, 2266 .teardown.single = NUL 1286 .teardown.single = NULL, 2267 }, 1287 }, 2268 }; 1288 }; 2269 1289 2270 /* Sanity check for callbacks */ 1290 /* Sanity check for callbacks */ 2271 static int cpuhp_cb_check(enum cpuhp_state st 1291 static int cpuhp_cb_check(enum cpuhp_state state) 2272 { 1292 { 2273 if (state <= CPUHP_OFFLINE || state > 1293 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) 2274 return -EINVAL; 1294 return -EINVAL; 2275 return 0; 1295 return 0; 2276 } 1296 } 2277 1297 2278 /* 1298 /* 2279 * Returns a free for dynamic slot assignment 1299 * Returns a free for dynamic slot assignment of the Online state. The states 2280 * are protected by the cpuhp_slot_states mut 1300 * are protected by the cpuhp_slot_states mutex and an empty slot is identified 2281 * by having no name assigned. 1301 * by having no name assigned. 2282 */ 1302 */ 2283 static int cpuhp_reserve_state(enum cpuhp_sta 1303 static int cpuhp_reserve_state(enum cpuhp_state state) 2284 { 1304 { 2285 enum cpuhp_state i, end; 1305 enum cpuhp_state i, end; 2286 struct cpuhp_step *step; 1306 struct cpuhp_step *step; 2287 1307 2288 switch (state) { 1308 switch (state) { 2289 case CPUHP_AP_ONLINE_DYN: 1309 case CPUHP_AP_ONLINE_DYN: 2290 step = cpuhp_hp_states + CPUH !! 1310 step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN; 2291 end = CPUHP_AP_ONLINE_DYN_END 1311 end = CPUHP_AP_ONLINE_DYN_END; 2292 break; 1312 break; 2293 case CPUHP_BP_PREPARE_DYN: 1313 case CPUHP_BP_PREPARE_DYN: 2294 step = cpuhp_hp_states + CPUH !! 1314 step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN; 2295 end = CPUHP_BP_PREPARE_DYN_EN 1315 end = CPUHP_BP_PREPARE_DYN_END; 2296 break; 1316 break; 2297 default: 1317 default: 2298 return -EINVAL; 1318 return -EINVAL; 2299 } 1319 } 2300 1320 2301 for (i = state; i <= end; i++, step++ 1321 for (i = state; i <= end; i++, step++) { 2302 if (!step->name) 1322 if (!step->name) 2303 return i; 1323 return i; 2304 } 1324 } 2305 WARN(1, "No more dynamic states avail 1325 WARN(1, "No more dynamic states available for CPU hotplug\n"); 2306 return -ENOSPC; 1326 return -ENOSPC; 2307 } 1327 } 2308 1328 2309 static int cpuhp_store_callbacks(enum cpuhp_s 1329 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, 2310 int (*startu 1330 int (*startup)(unsigned int cpu), 2311 int (*teardo 1331 int (*teardown)(unsigned int cpu), 2312 bool multi_i 1332 bool multi_instance) 2313 { 1333 { 2314 /* (Un)Install the callbacks for furt 1334 /* (Un)Install the callbacks for further cpu hotplug operations */ 2315 struct cpuhp_step *sp; 1335 struct cpuhp_step *sp; 2316 int ret = 0; 1336 int ret = 0; 2317 1337 2318 /* !! 1338 if (state == CPUHP_AP_ONLINE_DYN || state == CPUHP_BP_PREPARE_DYN) { 2319 * If name is NULL, then the state ge << 2320 * << 2321 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_P << 2322 * the first allocation from these dy << 2323 * would trigger a new allocation and << 2324 * empty) state, leaving the callback << 2325 * dangling, which causes wreckage on << 2326 */ << 2327 if (name && (state == CPUHP_AP_ONLINE << 2328 state == CPUHP_BP_PREPAR << 2329 ret = cpuhp_reserve_state(sta 1339 ret = cpuhp_reserve_state(state); 2330 if (ret < 0) 1340 if (ret < 0) 2331 return ret; 1341 return ret; 2332 state = ret; 1342 state = ret; 2333 } 1343 } 2334 sp = cpuhp_get_step(state); 1344 sp = cpuhp_get_step(state); 2335 if (name && sp->name) 1345 if (name && sp->name) 2336 return -EBUSY; 1346 return -EBUSY; 2337 1347 2338 sp->startup.single = startup; 1348 sp->startup.single = startup; 2339 sp->teardown.single = teardown; 1349 sp->teardown.single = teardown; 2340 sp->name = name; 1350 sp->name = name; 2341 sp->multi_instance = multi_instance; 1351 sp->multi_instance = multi_instance; 2342 INIT_HLIST_HEAD(&sp->list); 1352 INIT_HLIST_HEAD(&sp->list); 2343 return ret; 1353 return ret; 2344 } 1354 } 2345 1355 2346 static void *cpuhp_get_teardown_cb(enum cpuhp 1356 static void *cpuhp_get_teardown_cb(enum cpuhp_state state) 2347 { 1357 { 2348 return cpuhp_get_step(state)->teardow 1358 return cpuhp_get_step(state)->teardown.single; 2349 } 1359 } 2350 1360 2351 /* 1361 /* 2352 * Call the startup/teardown function for a s 1362 * Call the startup/teardown function for a step either on the AP or 2353 * on the current CPU. 1363 * on the current CPU. 2354 */ 1364 */ 2355 static int cpuhp_issue_call(int cpu, enum cpu 1365 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, 2356 struct hlist_node 1366 struct hlist_node *node) 2357 { 1367 { 2358 struct cpuhp_step *sp = cpuhp_get_ste 1368 struct cpuhp_step *sp = cpuhp_get_step(state); 2359 int ret; 1369 int ret; 2360 1370 2361 /* !! 1371 if ((bringup && !sp->startup.single) || 2362 * If there's nothing to do, we done. !! 1372 (!bringup && !sp->teardown.single)) 2363 * Relies on the union for multi_inst << 2364 */ << 2365 if (cpuhp_step_empty(bringup, sp)) << 2366 return 0; 1373 return 0; 2367 /* 1374 /* 2368 * The non AP bound callbacks can fai 1375 * The non AP bound callbacks can fail on bringup. On teardown 2369 * e.g. module removal we crash for n 1376 * e.g. module removal we crash for now. 2370 */ 1377 */ 2371 #ifdef CONFIG_SMP 1378 #ifdef CONFIG_SMP 2372 if (cpuhp_is_ap_state(state)) 1379 if (cpuhp_is_ap_state(state)) 2373 ret = cpuhp_invoke_ap_callbac 1380 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); 2374 else 1381 else 2375 ret = cpuhp_invoke_callback(c !! 1382 ret = cpuhp_invoke_callback(cpu, state, bringup, node); 2376 #else 1383 #else 2377 ret = cpuhp_invoke_callback(cpu, stat !! 1384 ret = cpuhp_invoke_callback(cpu, state, bringup, node); 2378 #endif 1385 #endif 2379 BUG_ON(ret && !bringup); 1386 BUG_ON(ret && !bringup); 2380 return ret; 1387 return ret; 2381 } 1388 } 2382 1389 2383 /* 1390 /* 2384 * Called from __cpuhp_setup_state on a recov 1391 * Called from __cpuhp_setup_state on a recoverable failure. 2385 * 1392 * 2386 * Note: The teardown callbacks for rollback 1393 * Note: The teardown callbacks for rollback are not allowed to fail! 2387 */ 1394 */ 2388 static void cpuhp_rollback_install(int failed 1395 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, 2389 struct hli 1396 struct hlist_node *node) 2390 { 1397 { 2391 int cpu; 1398 int cpu; 2392 1399 2393 /* Roll back the already executed ste 1400 /* Roll back the already executed steps on the other cpus */ 2394 for_each_present_cpu(cpu) { 1401 for_each_present_cpu(cpu) { 2395 struct cpuhp_cpu_state *st = 1402 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 2396 int cpustate = st->state; 1403 int cpustate = st->state; 2397 1404 2398 if (cpu >= failedcpu) 1405 if (cpu >= failedcpu) 2399 break; 1406 break; 2400 1407 2401 /* Did we invoke the startup 1408 /* Did we invoke the startup call on that cpu ? */ 2402 if (cpustate >= state) 1409 if (cpustate >= state) 2403 cpuhp_issue_call(cpu, 1410 cpuhp_issue_call(cpu, state, false, node); 2404 } 1411 } 2405 } 1412 } 2406 1413 2407 int __cpuhp_state_add_instance_cpuslocked(enu !! 1414 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, 2408 str !! 1415 bool invoke) 2409 boo << 2410 { 1416 { 2411 struct cpuhp_step *sp; 1417 struct cpuhp_step *sp; 2412 int cpu; 1418 int cpu; 2413 int ret; 1419 int ret; 2414 1420 2415 lockdep_assert_cpus_held(); << 2416 << 2417 sp = cpuhp_get_step(state); 1421 sp = cpuhp_get_step(state); 2418 if (sp->multi_instance == false) 1422 if (sp->multi_instance == false) 2419 return -EINVAL; 1423 return -EINVAL; 2420 1424 >> 1425 get_online_cpus(); 2421 mutex_lock(&cpuhp_state_mutex); 1426 mutex_lock(&cpuhp_state_mutex); 2422 1427 2423 if (!invoke || !sp->startup.multi) 1428 if (!invoke || !sp->startup.multi) 2424 goto add_node; 1429 goto add_node; 2425 1430 2426 /* 1431 /* 2427 * Try to call the startup callback f 1432 * Try to call the startup callback for each present cpu 2428 * depending on the hotplug state of 1433 * depending on the hotplug state of the cpu. 2429 */ 1434 */ 2430 for_each_present_cpu(cpu) { 1435 for_each_present_cpu(cpu) { 2431 struct cpuhp_cpu_state *st = 1436 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 2432 int cpustate = st->state; 1437 int cpustate = st->state; 2433 1438 2434 if (cpustate < state) 1439 if (cpustate < state) 2435 continue; 1440 continue; 2436 1441 2437 ret = cpuhp_issue_call(cpu, s 1442 ret = cpuhp_issue_call(cpu, state, true, node); 2438 if (ret) { 1443 if (ret) { 2439 if (sp->teardown.mult 1444 if (sp->teardown.multi) 2440 cpuhp_rollbac 1445 cpuhp_rollback_install(cpu, state, node); 2441 goto unlock; 1446 goto unlock; 2442 } 1447 } 2443 } 1448 } 2444 add_node: 1449 add_node: 2445 ret = 0; 1450 ret = 0; 2446 hlist_add_head(node, &sp->list); 1451 hlist_add_head(node, &sp->list); 2447 unlock: 1452 unlock: 2448 mutex_unlock(&cpuhp_state_mutex); 1453 mutex_unlock(&cpuhp_state_mutex); 2449 return ret; !! 1454 put_online_cpus(); 2450 } << 2451 << 2452 int __cpuhp_state_add_instance(enum cpuhp_sta << 2453 bool invoke) << 2454 { << 2455 int ret; << 2456 << 2457 cpus_read_lock(); << 2458 ret = __cpuhp_state_add_instance_cpus << 2459 cpus_read_unlock(); << 2460 return ret; 1455 return ret; 2461 } 1456 } 2462 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance) 1457 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); 2463 1458 2464 /** 1459 /** 2465 * __cpuhp_setup_state_cpuslocked - Setup the !! 1460 * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state 2466 * @state: The state to setup 1461 * @state: The state to setup 2467 * @name: Name of the step << 2468 * @invoke: If true, the startup 1462 * @invoke: If true, the startup function is invoked for cpus where 2469 * cpu state >= @state 1463 * cpu state >= @state 2470 * @startup: startup callback func 1464 * @startup: startup callback function 2471 * @teardown: teardown callback fun 1465 * @teardown: teardown callback function 2472 * @multi_instance: State is set up for m 1466 * @multi_instance: State is set up for multiple instances which get 2473 * added afterwards. 1467 * added afterwards. 2474 * 1468 * 2475 * The caller needs to hold cpus read locked !! 1469 * Returns: 2476 * Return: << 2477 * On success: 1470 * On success: 2478 * Positive state number if @state is CP !! 1471 * Positive state number if @state is CPUHP_AP_ONLINE_DYN 2479 * 0 for all other states 1472 * 0 for all other states 2480 * On failure: proper (negative) error code 1473 * On failure: proper (negative) error code 2481 */ 1474 */ 2482 int __cpuhp_setup_state_cpuslocked(enum cpuhp !! 1475 int __cpuhp_setup_state(enum cpuhp_state state, 2483 const char !! 1476 const char *name, bool invoke, 2484 int (*star !! 1477 int (*startup)(unsigned int cpu), 2485 int (*tear !! 1478 int (*teardown)(unsigned int cpu), 2486 bool multi !! 1479 bool multi_instance) 2487 { 1480 { 2488 int cpu, ret = 0; 1481 int cpu, ret = 0; 2489 bool dynstate; 1482 bool dynstate; 2490 1483 2491 lockdep_assert_cpus_held(); << 2492 << 2493 if (cpuhp_cb_check(state) || !name) 1484 if (cpuhp_cb_check(state) || !name) 2494 return -EINVAL; 1485 return -EINVAL; 2495 1486 >> 1487 get_online_cpus(); 2496 mutex_lock(&cpuhp_state_mutex); 1488 mutex_lock(&cpuhp_state_mutex); 2497 1489 2498 ret = cpuhp_store_callbacks(state, na 1490 ret = cpuhp_store_callbacks(state, name, startup, teardown, 2499 multi_ins 1491 multi_instance); 2500 1492 2501 dynstate = state == CPUHP_AP_ONLINE_D !! 1493 dynstate = state == CPUHP_AP_ONLINE_DYN; 2502 if (ret > 0 && dynstate) { 1494 if (ret > 0 && dynstate) { 2503 state = ret; 1495 state = ret; 2504 ret = 0; 1496 ret = 0; 2505 } 1497 } 2506 1498 2507 if (ret || !invoke || !startup) 1499 if (ret || !invoke || !startup) 2508 goto out; 1500 goto out; 2509 1501 2510 /* 1502 /* 2511 * Try to call the startup callback f 1503 * Try to call the startup callback for each present cpu 2512 * depending on the hotplug state of 1504 * depending on the hotplug state of the cpu. 2513 */ 1505 */ 2514 for_each_present_cpu(cpu) { 1506 for_each_present_cpu(cpu) { 2515 struct cpuhp_cpu_state *st = 1507 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 2516 int cpustate = st->state; 1508 int cpustate = st->state; 2517 1509 2518 if (cpustate < state) 1510 if (cpustate < state) 2519 continue; 1511 continue; 2520 1512 2521 ret = cpuhp_issue_call(cpu, s 1513 ret = cpuhp_issue_call(cpu, state, true, NULL); 2522 if (ret) { 1514 if (ret) { 2523 if (teardown) 1515 if (teardown) 2524 cpuhp_rollbac 1516 cpuhp_rollback_install(cpu, state, NULL); 2525 cpuhp_store_callbacks 1517 cpuhp_store_callbacks(state, NULL, NULL, NULL, false); 2526 goto out; 1518 goto out; 2527 } 1519 } 2528 } 1520 } 2529 out: 1521 out: 2530 mutex_unlock(&cpuhp_state_mutex); 1522 mutex_unlock(&cpuhp_state_mutex); >> 1523 put_online_cpus(); 2531 /* 1524 /* 2532 * If the requested state is CPUHP_AP !! 1525 * If the requested state is CPUHP_AP_ONLINE_DYN, return the 2533 * return the dynamically allocated s !! 1526 * dynamically allocated state in case of success. 2534 */ 1527 */ 2535 if (!ret && dynstate) 1528 if (!ret && dynstate) 2536 return state; 1529 return state; 2537 return ret; 1530 return ret; 2538 } 1531 } 2539 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked) << 2540 << 2541 int __cpuhp_setup_state(enum cpuhp_state stat << 2542 const char *name, boo << 2543 int (*startup)(unsign << 2544 int (*teardown)(unsig << 2545 bool multi_instance) << 2546 { << 2547 int ret; << 2548 << 2549 cpus_read_lock(); << 2550 ret = __cpuhp_setup_state_cpuslocked( << 2551 << 2552 cpus_read_unlock(); << 2553 return ret; << 2554 } << 2555 EXPORT_SYMBOL(__cpuhp_setup_state); 1532 EXPORT_SYMBOL(__cpuhp_setup_state); 2556 1533 2557 int __cpuhp_state_remove_instance(enum cpuhp_ 1534 int __cpuhp_state_remove_instance(enum cpuhp_state state, 2558 struct hlis 1535 struct hlist_node *node, bool invoke) 2559 { 1536 { 2560 struct cpuhp_step *sp = cpuhp_get_ste 1537 struct cpuhp_step *sp = cpuhp_get_step(state); 2561 int cpu; 1538 int cpu; 2562 1539 2563 BUG_ON(cpuhp_cb_check(state)); 1540 BUG_ON(cpuhp_cb_check(state)); 2564 1541 2565 if (!sp->multi_instance) 1542 if (!sp->multi_instance) 2566 return -EINVAL; 1543 return -EINVAL; 2567 1544 2568 cpus_read_lock(); !! 1545 get_online_cpus(); 2569 mutex_lock(&cpuhp_state_mutex); 1546 mutex_lock(&cpuhp_state_mutex); 2570 1547 2571 if (!invoke || !cpuhp_get_teardown_cb 1548 if (!invoke || !cpuhp_get_teardown_cb(state)) 2572 goto remove; 1549 goto remove; 2573 /* 1550 /* 2574 * Call the teardown callback for eac 1551 * Call the teardown callback for each present cpu depending 2575 * on the hotplug state of the cpu. T 1552 * on the hotplug state of the cpu. This function is not 2576 * allowed to fail currently! 1553 * allowed to fail currently! 2577 */ 1554 */ 2578 for_each_present_cpu(cpu) { 1555 for_each_present_cpu(cpu) { 2579 struct cpuhp_cpu_state *st = 1556 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 2580 int cpustate = st->state; 1557 int cpustate = st->state; 2581 1558 2582 if (cpustate >= state) 1559 if (cpustate >= state) 2583 cpuhp_issue_call(cpu, 1560 cpuhp_issue_call(cpu, state, false, node); 2584 } 1561 } 2585 1562 2586 remove: 1563 remove: 2587 hlist_del(node); 1564 hlist_del(node); 2588 mutex_unlock(&cpuhp_state_mutex); 1565 mutex_unlock(&cpuhp_state_mutex); 2589 cpus_read_unlock(); !! 1566 put_online_cpus(); 2590 1567 2591 return 0; 1568 return 0; 2592 } 1569 } 2593 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instan 1570 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); 2594 1571 2595 /** 1572 /** 2596 * __cpuhp_remove_state_cpuslocked - Remove t !! 1573 * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state 2597 * @state: The state to remove 1574 * @state: The state to remove 2598 * @invoke: If true, the teardown functio 1575 * @invoke: If true, the teardown function is invoked for cpus where 2599 * cpu state >= @state 1576 * cpu state >= @state 2600 * 1577 * 2601 * The caller needs to hold cpus read locked << 2602 * The teardown callback is currently not all 1578 * The teardown callback is currently not allowed to fail. Think 2603 * about module removal! 1579 * about module removal! 2604 */ 1580 */ 2605 void __cpuhp_remove_state_cpuslocked(enum cpu !! 1581 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) 2606 { 1582 { 2607 struct cpuhp_step *sp = cpuhp_get_ste 1583 struct cpuhp_step *sp = cpuhp_get_step(state); 2608 int cpu; 1584 int cpu; 2609 1585 2610 BUG_ON(cpuhp_cb_check(state)); 1586 BUG_ON(cpuhp_cb_check(state)); 2611 1587 2612 lockdep_assert_cpus_held(); !! 1588 get_online_cpus(); 2613 1589 2614 mutex_lock(&cpuhp_state_mutex); 1590 mutex_lock(&cpuhp_state_mutex); 2615 if (sp->multi_instance) { 1591 if (sp->multi_instance) { 2616 WARN(!hlist_empty(&sp->list), 1592 WARN(!hlist_empty(&sp->list), 2617 "Error: Removing state % 1593 "Error: Removing state %d which has instances left.\n", 2618 state); 1594 state); 2619 goto remove; 1595 goto remove; 2620 } 1596 } 2621 1597 2622 if (!invoke || !cpuhp_get_teardown_cb 1598 if (!invoke || !cpuhp_get_teardown_cb(state)) 2623 goto remove; 1599 goto remove; 2624 1600 2625 /* 1601 /* 2626 * Call the teardown callback for eac 1602 * Call the teardown callback for each present cpu depending 2627 * on the hotplug state of the cpu. T 1603 * on the hotplug state of the cpu. This function is not 2628 * allowed to fail currently! 1604 * allowed to fail currently! 2629 */ 1605 */ 2630 for_each_present_cpu(cpu) { 1606 for_each_present_cpu(cpu) { 2631 struct cpuhp_cpu_state *st = 1607 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); 2632 int cpustate = st->state; 1608 int cpustate = st->state; 2633 1609 2634 if (cpustate >= state) 1610 if (cpustate >= state) 2635 cpuhp_issue_call(cpu, 1611 cpuhp_issue_call(cpu, state, false, NULL); 2636 } 1612 } 2637 remove: 1613 remove: 2638 cpuhp_store_callbacks(state, NULL, NU 1614 cpuhp_store_callbacks(state, NULL, NULL, NULL, false); 2639 mutex_unlock(&cpuhp_state_mutex); 1615 mutex_unlock(&cpuhp_state_mutex); 2640 } !! 1616 put_online_cpus(); 2641 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked << 2642 << 2643 void __cpuhp_remove_state(enum cpuhp_state st << 2644 { << 2645 cpus_read_lock(); << 2646 __cpuhp_remove_state_cpuslocked(state << 2647 cpus_read_unlock(); << 2648 } 1617 } 2649 EXPORT_SYMBOL(__cpuhp_remove_state); 1618 EXPORT_SYMBOL(__cpuhp_remove_state); 2650 1619 2651 #ifdef CONFIG_HOTPLUG_SMT << 2652 static void cpuhp_offline_cpu_device(unsigned << 2653 { << 2654 struct device *dev = get_cpu_device(c << 2655 << 2656 dev->offline = true; << 2657 /* Tell user space about the state ch << 2658 kobject_uevent(&dev->kobj, KOBJ_OFFLI << 2659 } << 2660 << 2661 static void cpuhp_online_cpu_device(unsigned << 2662 { << 2663 struct device *dev = get_cpu_device(c << 2664 << 2665 dev->offline = false; << 2666 /* Tell user space about the state ch << 2667 kobject_uevent(&dev->kobj, KOBJ_ONLIN << 2668 } << 2669 << 2670 int cpuhp_smt_disable(enum cpuhp_smt_control << 2671 { << 2672 int cpu, ret = 0; << 2673 << 2674 cpu_maps_update_begin(); << 2675 for_each_online_cpu(cpu) { << 2676 if (topology_is_primary_threa << 2677 continue; << 2678 /* << 2679 * Disable can be called with << 2680 * from a higher to lower num << 2681 */ << 2682 if (ctrlval == CPU_SMT_ENABLE << 2683 continue; << 2684 ret = cpu_down_maps_locked(cp << 2685 if (ret) << 2686 break; << 2687 /* << 2688 * As this needs to hold the << 2689 * to call device_offline() b << 2690 * cpu_down() which takes cpu << 2691 * needs to be held as this m << 2692 * abusers of the hotplug mac << 2693 * << 2694 * So nothing would update de << 2695 * leave the sysfs entry stal << 2696 * smt control has been chang << 2697 * called under the sysfs hot << 2698 * serialized against the reg << 2699 */ << 2700 cpuhp_offline_cpu_device(cpu) << 2701 } << 2702 if (!ret) << 2703 cpu_smt_control = ctrlval; << 2704 cpu_maps_update_done(); << 2705 return ret; << 2706 } << 2707 << 2708 /* Check if the core a CPU belongs to is onli << 2709 #if !defined(topology_is_core_online) << 2710 static inline bool topology_is_core_online(un << 2711 { << 2712 return true; << 2713 } << 2714 #endif << 2715 << 2716 int cpuhp_smt_enable(void) << 2717 { << 2718 int cpu, ret = 0; << 2719 << 2720 cpu_maps_update_begin(); << 2721 cpu_smt_control = CPU_SMT_ENABLED; << 2722 for_each_present_cpu(cpu) { << 2723 /* Skip online CPUs and CPUs << 2724 if (cpu_online(cpu) || !node_ << 2725 continue; << 2726 if (!cpu_smt_thread_allowed(c << 2727 continue; << 2728 ret = _cpu_up(cpu, 0, CPUHP_O << 2729 if (ret) << 2730 break; << 2731 /* See comment in cpuhp_smt_d << 2732 cpuhp_online_cpu_device(cpu); << 2733 } << 2734 cpu_maps_update_done(); << 2735 return ret; << 2736 } << 2737 #endif << 2738 << 2739 #if defined(CONFIG_SYSFS) && defined(CONFIG_H 1620 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) 2740 static ssize_t state_show(struct device *dev, !! 1621 static ssize_t show_cpuhp_state(struct device *dev, 2741 struct device_attri !! 1622 struct device_attribute *attr, char *buf) 2742 { 1623 { 2743 struct cpuhp_cpu_state *st = per_cpu_ 1624 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); 2744 1625 2745 return sprintf(buf, "%d\n", st->state 1626 return sprintf(buf, "%d\n", st->state); 2746 } 1627 } 2747 static DEVICE_ATTR_RO(state); !! 1628 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); 2748 1629 2749 static ssize_t target_store(struct device *de !! 1630 static ssize_t write_cpuhp_target(struct device *dev, 2750 const char *buf, !! 1631 struct device_attribute *attr, >> 1632 const char *buf, size_t count) 2751 { 1633 { 2752 struct cpuhp_cpu_state *st = per_cpu_ 1634 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); 2753 struct cpuhp_step *sp; 1635 struct cpuhp_step *sp; 2754 int target, ret; 1636 int target, ret; 2755 1637 2756 ret = kstrtoint(buf, 10, &target); 1638 ret = kstrtoint(buf, 10, &target); 2757 if (ret) 1639 if (ret) 2758 return ret; 1640 return ret; 2759 1641 2760 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL 1642 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL 2761 if (target < CPUHP_OFFLINE || target 1643 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) 2762 return -EINVAL; 1644 return -EINVAL; 2763 #else 1645 #else 2764 if (target != CPUHP_OFFLINE && target 1646 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) 2765 return -EINVAL; 1647 return -EINVAL; 2766 #endif 1648 #endif 2767 1649 2768 ret = lock_device_hotplug_sysfs(); 1650 ret = lock_device_hotplug_sysfs(); 2769 if (ret) 1651 if (ret) 2770 return ret; 1652 return ret; 2771 1653 2772 mutex_lock(&cpuhp_state_mutex); 1654 mutex_lock(&cpuhp_state_mutex); 2773 sp = cpuhp_get_step(target); 1655 sp = cpuhp_get_step(target); 2774 ret = !sp->name || sp->cant_stop ? -E 1656 ret = !sp->name || sp->cant_stop ? -EINVAL : 0; 2775 mutex_unlock(&cpuhp_state_mutex); 1657 mutex_unlock(&cpuhp_state_mutex); 2776 if (ret) 1658 if (ret) 2777 goto out; 1659 goto out; 2778 1660 2779 if (st->state < target) 1661 if (st->state < target) 2780 ret = cpu_up(dev->id, target) !! 1662 ret = do_cpu_up(dev->id, target); 2781 else if (st->state > target) !! 1663 else 2782 ret = cpu_down(dev->id, targe !! 1664 ret = do_cpu_down(dev->id, target); 2783 else if (WARN_ON(st->target != target << 2784 st->target = target; << 2785 out: 1665 out: 2786 unlock_device_hotplug(); 1666 unlock_device_hotplug(); 2787 return ret ? ret : count; 1667 return ret ? ret : count; 2788 } 1668 } 2789 1669 2790 static ssize_t target_show(struct device *dev !! 1670 static ssize_t show_cpuhp_target(struct device *dev, 2791 struct device_attr !! 1671 struct device_attribute *attr, char *buf) 2792 { 1672 { 2793 struct cpuhp_cpu_state *st = per_cpu_ 1673 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); 2794 1674 2795 return sprintf(buf, "%d\n", st->targe 1675 return sprintf(buf, "%d\n", st->target); 2796 } 1676 } 2797 static DEVICE_ATTR_RW(target); !! 1677 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); 2798 << 2799 static ssize_t fail_store(struct device *dev, << 2800 const char *buf, si << 2801 { << 2802 struct cpuhp_cpu_state *st = per_cpu_ << 2803 struct cpuhp_step *sp; << 2804 int fail, ret; << 2805 << 2806 ret = kstrtoint(buf, 10, &fail); << 2807 if (ret) << 2808 return ret; << 2809 << 2810 if (fail == CPUHP_INVALID) { << 2811 st->fail = fail; << 2812 return count; << 2813 } << 2814 << 2815 if (fail < CPUHP_OFFLINE || fail > CP << 2816 return -EINVAL; << 2817 << 2818 /* << 2819 * Cannot fail STARTING/DYING callbac << 2820 */ << 2821 if (cpuhp_is_atomic_state(fail)) << 2822 return -EINVAL; << 2823 << 2824 /* << 2825 * DEAD callbacks cannot fail... << 2826 * ... neither can CPUHP_BRINGUP_CPU << 2827 * triggering STARTING callbacks, a f << 2828 * hinder rollback. << 2829 */ << 2830 if (fail <= CPUHP_BRINGUP_CPU && st-> << 2831 return -EINVAL; << 2832 << 2833 /* << 2834 * Cannot fail anything that doesn't << 2835 */ << 2836 mutex_lock(&cpuhp_state_mutex); << 2837 sp = cpuhp_get_step(fail); << 2838 if (!sp->startup.single && !sp->teard << 2839 ret = -EINVAL; << 2840 mutex_unlock(&cpuhp_state_mutex); << 2841 if (ret) << 2842 return ret; << 2843 << 2844 st->fail = fail; << 2845 << 2846 return count; << 2847 } << 2848 << 2849 static ssize_t fail_show(struct device *dev, << 2850 struct device_attrib << 2851 { << 2852 struct cpuhp_cpu_state *st = per_cpu_ << 2853 << 2854 return sprintf(buf, "%d\n", st->fail) << 2855 } << 2856 << 2857 static DEVICE_ATTR_RW(fail); << 2858 1678 2859 static struct attribute *cpuhp_cpu_attrs[] = 1679 static struct attribute *cpuhp_cpu_attrs[] = { 2860 &dev_attr_state.attr, 1680 &dev_attr_state.attr, 2861 &dev_attr_target.attr, 1681 &dev_attr_target.attr, 2862 &dev_attr_fail.attr, << 2863 NULL 1682 NULL 2864 }; 1683 }; 2865 1684 2866 static const struct attribute_group cpuhp_cpu !! 1685 static struct attribute_group cpuhp_cpu_attr_group = { 2867 .attrs = cpuhp_cpu_attrs, 1686 .attrs = cpuhp_cpu_attrs, 2868 .name = "hotplug", 1687 .name = "hotplug", 2869 NULL 1688 NULL 2870 }; 1689 }; 2871 1690 2872 static ssize_t states_show(struct device *dev !! 1691 static ssize_t show_cpuhp_states(struct device *dev, 2873 struct devic 1692 struct device_attribute *attr, char *buf) 2874 { 1693 { 2875 ssize_t cur, res = 0; 1694 ssize_t cur, res = 0; 2876 int i; 1695 int i; 2877 1696 2878 mutex_lock(&cpuhp_state_mutex); 1697 mutex_lock(&cpuhp_state_mutex); 2879 for (i = CPUHP_OFFLINE; i <= CPUHP_ON 1698 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { 2880 struct cpuhp_step *sp = cpuhp 1699 struct cpuhp_step *sp = cpuhp_get_step(i); 2881 1700 2882 if (sp->name) { 1701 if (sp->name) { 2883 cur = sprintf(buf, "% 1702 cur = sprintf(buf, "%3d: %s\n", i, sp->name); 2884 buf += cur; 1703 buf += cur; 2885 res += cur; 1704 res += cur; 2886 } 1705 } 2887 } 1706 } 2888 mutex_unlock(&cpuhp_state_mutex); 1707 mutex_unlock(&cpuhp_state_mutex); 2889 return res; 1708 return res; 2890 } 1709 } 2891 static DEVICE_ATTR_RO(states); !! 1710 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); 2892 1711 2893 static struct attribute *cpuhp_cpu_root_attrs 1712 static struct attribute *cpuhp_cpu_root_attrs[] = { 2894 &dev_attr_states.attr, 1713 &dev_attr_states.attr, 2895 NULL 1714 NULL 2896 }; 1715 }; 2897 1716 2898 static const struct attribute_group cpuhp_cpu !! 1717 static struct attribute_group cpuhp_cpu_root_attr_group = { 2899 .attrs = cpuhp_cpu_root_attrs, 1718 .attrs = cpuhp_cpu_root_attrs, 2900 .name = "hotplug", 1719 .name = "hotplug", 2901 NULL 1720 NULL 2902 }; 1721 }; 2903 1722 2904 #ifdef CONFIG_HOTPLUG_SMT << 2905 << 2906 static bool cpu_smt_num_threads_valid(unsigne << 2907 { << 2908 if (IS_ENABLED(CONFIG_SMT_NUM_THREADS << 2909 return threads >= 1 && thread << 2910 return threads == 1 || threads == cpu << 2911 } << 2912 << 2913 static ssize_t << 2914 __store_smt_control(struct device *dev, struc << 2915 const char *buf, size_t c << 2916 { << 2917 int ctrlval, ret, num_threads, orig_t << 2918 bool force_off; << 2919 << 2920 if (cpu_smt_control == CPU_SMT_FORCE_ << 2921 return -EPERM; << 2922 << 2923 if (cpu_smt_control == CPU_SMT_NOT_SU << 2924 return -ENODEV; << 2925 << 2926 if (sysfs_streq(buf, "on")) { << 2927 ctrlval = CPU_SMT_ENABLED; << 2928 num_threads = cpu_smt_max_thr << 2929 } else if (sysfs_streq(buf, "off")) { << 2930 ctrlval = CPU_SMT_DISABLED; << 2931 num_threads = 1; << 2932 } else if (sysfs_streq(buf, "forceoff << 2933 ctrlval = CPU_SMT_FORCE_DISAB << 2934 num_threads = 1; << 2935 } else if (kstrtoint(buf, 10, &num_th << 2936 if (num_threads == 1) << 2937 ctrlval = CPU_SMT_DIS << 2938 else if (cpu_smt_num_threads_ << 2939 ctrlval = CPU_SMT_ENA << 2940 else << 2941 return -EINVAL; << 2942 } else { << 2943 return -EINVAL; << 2944 } << 2945 << 2946 ret = lock_device_hotplug_sysfs(); << 2947 if (ret) << 2948 return ret; << 2949 << 2950 orig_threads = cpu_smt_num_threads; << 2951 cpu_smt_num_threads = num_threads; << 2952 << 2953 force_off = ctrlval != cpu_smt_contro << 2954 << 2955 if (num_threads > orig_threads) << 2956 ret = cpuhp_smt_enable(); << 2957 else if (num_threads < orig_threads | << 2958 ret = cpuhp_smt_disable(ctrlv << 2959 << 2960 unlock_device_hotplug(); << 2961 return ret ? ret : count; << 2962 } << 2963 << 2964 #else /* !CONFIG_HOTPLUG_SMT */ << 2965 static ssize_t << 2966 __store_smt_control(struct device *dev, struc << 2967 const char *buf, size_t c << 2968 { << 2969 return -ENODEV; << 2970 } << 2971 #endif /* CONFIG_HOTPLUG_SMT */ << 2972 << 2973 static const char *smt_states[] = { << 2974 [CPU_SMT_ENABLED] = "on << 2975 [CPU_SMT_DISABLED] = "of << 2976 [CPU_SMT_FORCE_DISABLED] = "fo << 2977 [CPU_SMT_NOT_SUPPORTED] = "no << 2978 [CPU_SMT_NOT_IMPLEMENTED] = "no << 2979 }; << 2980 << 2981 static ssize_t control_show(struct device *de << 2982 struct device_att << 2983 { << 2984 const char *state = smt_states[cpu_sm << 2985 << 2986 #ifdef CONFIG_HOTPLUG_SMT << 2987 /* << 2988 * If SMT is enabled but not all thre << 2989 * number of threads. If all threads << 2990 * show the state name. << 2991 */ << 2992 if (cpu_smt_control == CPU_SMT_ENABLE << 2993 cpu_smt_num_threads != cpu_smt_ma << 2994 return sysfs_emit(buf, "%d\n" << 2995 #endif << 2996 << 2997 return sysfs_emit(buf, "%s\n", state) << 2998 } << 2999 << 3000 static ssize_t control_store(struct device *d << 3001 const char *buf, << 3002 { << 3003 return __store_smt_control(dev, attr, << 3004 } << 3005 static DEVICE_ATTR_RW(control); << 3006 << 3007 static ssize_t active_show(struct device *dev << 3008 struct device_attr << 3009 { << 3010 return sysfs_emit(buf, "%d\n", sched_ << 3011 } << 3012 static DEVICE_ATTR_RO(active); << 3013 << 3014 static struct attribute *cpuhp_smt_attrs[] = << 3015 &dev_attr_control.attr, << 3016 &dev_attr_active.attr, << 3017 NULL << 3018 }; << 3019 << 3020 static const struct attribute_group cpuhp_smt << 3021 .attrs = cpuhp_smt_attrs, << 3022 .name = "smt", << 3023 NULL << 3024 }; << 3025 << 3026 static int __init cpu_smt_sysfs_init(void) << 3027 { << 3028 struct device *dev_root; << 3029 int ret = -ENODEV; << 3030 << 3031 dev_root = bus_get_dev_root(&cpu_subs << 3032 if (dev_root) { << 3033 ret = sysfs_create_group(&dev << 3034 put_device(dev_root); << 3035 } << 3036 return ret; << 3037 } << 3038 << 3039 static int __init cpuhp_sysfs_init(void) 1723 static int __init cpuhp_sysfs_init(void) 3040 { 1724 { 3041 struct device *dev_root; << 3042 int cpu, ret; 1725 int cpu, ret; 3043 1726 3044 ret = cpu_smt_sysfs_init(); !! 1727 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, >> 1728 &cpuhp_cpu_root_attr_group); 3045 if (ret) 1729 if (ret) 3046 return ret; 1730 return ret; 3047 1731 3048 dev_root = bus_get_dev_root(&cpu_subs << 3049 if (dev_root) { << 3050 ret = sysfs_create_group(&dev << 3051 put_device(dev_root); << 3052 if (ret) << 3053 return ret; << 3054 } << 3055 << 3056 for_each_possible_cpu(cpu) { 1732 for_each_possible_cpu(cpu) { 3057 struct device *dev = get_cpu_ 1733 struct device *dev = get_cpu_device(cpu); 3058 1734 3059 if (!dev) 1735 if (!dev) 3060 continue; 1736 continue; 3061 ret = sysfs_create_group(&dev 1737 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); 3062 if (ret) 1738 if (ret) 3063 return ret; 1739 return ret; 3064 } 1740 } 3065 return 0; 1741 return 0; 3066 } 1742 } 3067 device_initcall(cpuhp_sysfs_init); 1743 device_initcall(cpuhp_sysfs_init); 3068 #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU !! 1744 #endif 3069 1745 3070 /* 1746 /* 3071 * cpu_bit_bitmap[] is a special, "compressed 1747 * cpu_bit_bitmap[] is a special, "compressed" data structure that 3072 * represents all NR_CPUS bits binary values 1748 * represents all NR_CPUS bits binary values of 1<<nr. 3073 * 1749 * 3074 * It is used by cpumask_of() to get a consta 1750 * It is used by cpumask_of() to get a constant address to a CPU 3075 * mask value that has a single bit set only. 1751 * mask value that has a single bit set only. 3076 */ 1752 */ 3077 1753 3078 /* cpu_bit_bitmap[0] is empty - so we can bac 1754 /* cpu_bit_bitmap[0] is empty - so we can back into it */ 3079 #define MASK_DECLARE_1(x) [x+1][0] = (1 1755 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) 3080 #define MASK_DECLARE_2(x) MASK_DECLARE_ 1756 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) 3081 #define MASK_DECLARE_4(x) MASK_DECLARE_ 1757 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) 3082 #define MASK_DECLARE_8(x) MASK_DECLARE_ 1758 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) 3083 1759 3084 const unsigned long cpu_bit_bitmap[BITS_PER_L 1760 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { 3085 1761 3086 MASK_DECLARE_8(0), MASK_DECLARE_ 1762 MASK_DECLARE_8(0), MASK_DECLARE_8(8), 3087 MASK_DECLARE_8(16), MASK_DECLARE_ 1763 MASK_DECLARE_8(16), MASK_DECLARE_8(24), 3088 #if BITS_PER_LONG > 32 1764 #if BITS_PER_LONG > 32 3089 MASK_DECLARE_8(32), MASK_DECLARE_ 1765 MASK_DECLARE_8(32), MASK_DECLARE_8(40), 3090 MASK_DECLARE_8(48), MASK_DECLARE_ 1766 MASK_DECLARE_8(48), MASK_DECLARE_8(56), 3091 #endif 1767 #endif 3092 }; 1768 }; 3093 EXPORT_SYMBOL_GPL(cpu_bit_bitmap); 1769 EXPORT_SYMBOL_GPL(cpu_bit_bitmap); 3094 1770 3095 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = 1771 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; 3096 EXPORT_SYMBOL(cpu_all_bits); 1772 EXPORT_SYMBOL(cpu_all_bits); 3097 1773 3098 #ifdef CONFIG_INIT_ALL_POSSIBLE 1774 #ifdef CONFIG_INIT_ALL_POSSIBLE 3099 struct cpumask __cpu_possible_mask __ro_after !! 1775 struct cpumask __cpu_possible_mask __read_mostly 3100 = {CPU_BITS_ALL}; 1776 = {CPU_BITS_ALL}; 3101 #else 1777 #else 3102 struct cpumask __cpu_possible_mask __ro_after !! 1778 struct cpumask __cpu_possible_mask __read_mostly; 3103 #endif 1779 #endif 3104 EXPORT_SYMBOL(__cpu_possible_mask); 1780 EXPORT_SYMBOL(__cpu_possible_mask); 3105 1781 3106 struct cpumask __cpu_online_mask __read_mostl 1782 struct cpumask __cpu_online_mask __read_mostly; 3107 EXPORT_SYMBOL(__cpu_online_mask); 1783 EXPORT_SYMBOL(__cpu_online_mask); 3108 1784 3109 struct cpumask __cpu_enabled_mask __read_most << 3110 EXPORT_SYMBOL(__cpu_enabled_mask); << 3111 << 3112 struct cpumask __cpu_present_mask __read_most 1785 struct cpumask __cpu_present_mask __read_mostly; 3113 EXPORT_SYMBOL(__cpu_present_mask); 1786 EXPORT_SYMBOL(__cpu_present_mask); 3114 1787 3115 struct cpumask __cpu_active_mask __read_mostl 1788 struct cpumask __cpu_active_mask __read_mostly; 3116 EXPORT_SYMBOL(__cpu_active_mask); 1789 EXPORT_SYMBOL(__cpu_active_mask); 3117 1790 3118 struct cpumask __cpu_dying_mask __read_mostly << 3119 EXPORT_SYMBOL(__cpu_dying_mask); << 3120 << 3121 atomic_t __num_online_cpus __read_mostly; << 3122 EXPORT_SYMBOL(__num_online_cpus); << 3123 << 3124 void init_cpu_present(const struct cpumask *s 1791 void init_cpu_present(const struct cpumask *src) 3125 { 1792 { 3126 cpumask_copy(&__cpu_present_mask, src 1793 cpumask_copy(&__cpu_present_mask, src); 3127 } 1794 } 3128 1795 3129 void init_cpu_possible(const struct cpumask * 1796 void init_cpu_possible(const struct cpumask *src) 3130 { 1797 { 3131 cpumask_copy(&__cpu_possible_mask, sr 1798 cpumask_copy(&__cpu_possible_mask, src); 3132 } 1799 } 3133 1800 3134 void init_cpu_online(const struct cpumask *sr 1801 void init_cpu_online(const struct cpumask *src) 3135 { 1802 { 3136 cpumask_copy(&__cpu_online_mask, src) 1803 cpumask_copy(&__cpu_online_mask, src); 3137 } 1804 } 3138 1805 3139 void set_cpu_online(unsigned int cpu, bool on << 3140 { << 3141 /* << 3142 * atomic_inc/dec() is required to ha << 3143 * function by the reboot and kexec c << 3144 * IPI/NMI broadcasts when shutting d << 3145 * regular CPU hotplug is properly se << 3146 * << 3147 * Note, that the fact that __num_onl << 3148 * does not protect readers which are << 3149 * concurrent hotplug operations. << 3150 */ << 3151 if (online) { << 3152 if (!cpumask_test_and_set_cpu << 3153 atomic_inc(&__num_onl << 3154 } else { << 3155 if (cpumask_test_and_clear_cp << 3156 atomic_dec(&__num_onl << 3157 } << 3158 } << 3159 << 3160 /* 1806 /* 3161 * Activate the first processor. 1807 * Activate the first processor. 3162 */ 1808 */ 3163 void __init boot_cpu_init(void) 1809 void __init boot_cpu_init(void) 3164 { 1810 { 3165 int cpu = smp_processor_id(); 1811 int cpu = smp_processor_id(); 3166 1812 3167 /* Mark the boot cpu "present", "onli 1813 /* Mark the boot cpu "present", "online" etc for SMP and UP case */ 3168 set_cpu_online(cpu, true); 1814 set_cpu_online(cpu, true); 3169 set_cpu_active(cpu, true); 1815 set_cpu_active(cpu, true); 3170 set_cpu_present(cpu, true); 1816 set_cpu_present(cpu, true); 3171 set_cpu_possible(cpu, true); 1817 set_cpu_possible(cpu, true); 3172 << 3173 #ifdef CONFIG_SMP << 3174 __boot_cpu_id = cpu; << 3175 #endif << 3176 } 1818 } 3177 1819 3178 /* 1820 /* 3179 * Must be called _AFTER_ setting up the per_ 1821 * Must be called _AFTER_ setting up the per_cpu areas 3180 */ 1822 */ 3181 void __init boot_cpu_hotplug_init(void) !! 1823 void __init boot_cpu_state_init(void) 3182 { << 3183 #ifdef CONFIG_SMP << 3184 cpumask_set_cpu(smp_processor_id(), & << 3185 atomic_set(this_cpu_ptr(&cpuhp_state. << 3186 #endif << 3187 this_cpu_write(cpuhp_state.state, CPU << 3188 this_cpu_write(cpuhp_state.target, CP << 3189 } << 3190 << 3191 #ifdef CONFIG_CPU_MITIGATIONS << 3192 /* << 3193 * These are used for a global "mitigations=" << 3194 * optional CPU mitigations. << 3195 */ << 3196 enum cpu_mitigations { << 3197 CPU_MITIGATIONS_OFF, << 3198 CPU_MITIGATIONS_AUTO, << 3199 CPU_MITIGATIONS_AUTO_NOSMT, << 3200 }; << 3201 << 3202 static enum cpu_mitigations cpu_mitigations _ << 3203 << 3204 static int __init mitigations_parse_cmdline(c << 3205 { 1824 { 3206 if (!strcmp(arg, "off")) !! 1825 per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; 3207 cpu_mitigations = CPU_MITIGAT << 3208 else if (!strcmp(arg, "auto")) << 3209 cpu_mitigations = CPU_MITIGAT << 3210 else if (!strcmp(arg, "auto,nosmt")) << 3211 cpu_mitigations = CPU_MITIGAT << 3212 else << 3213 pr_crit("Unsupported mitigati << 3214 arg); << 3215 << 3216 return 0; << 3217 } << 3218 << 3219 /* mitigations=off */ << 3220 bool cpu_mitigations_off(void) << 3221 { << 3222 return cpu_mitigations == CPU_MITIGAT << 3223 } << 3224 EXPORT_SYMBOL_GPL(cpu_mitigations_off); << 3225 << 3226 /* mitigations=auto,nosmt */ << 3227 bool cpu_mitigations_auto_nosmt(void) << 3228 { << 3229 return cpu_mitigations == CPU_MITIGAT << 3230 } 1826 } 3231 EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt) << 3232 #else << 3233 static int __init mitigations_parse_cmdline(c << 3234 { << 3235 pr_crit("Kernel compiled without miti << 3236 return 0; << 3237 } << 3238 #endif << 3239 early_param("mitigations", mitigations_parse_ << 3240 1827
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.