Commit | Line | Data |
---|---|---|
eacd6f04 PM |
1 | /* SPDX-License-Identifier: GPL-2.0+ */ |
2 | /* | |
3 | * Task-based RCU implementations. | |
4 | * | |
5 | * Copyright (C) 2020 Paul E. McKenney | |
6 | */ | |
7 | ||
8fd8ca38 | 8 | #ifdef CONFIG_TASKS_RCU_GENERIC |
9b073de1 | 9 | #include "rcu_segcblist.h" |
5873b8a9 PM |
10 | |
11 | //////////////////////////////////////////////////////////////////////// | |
12 | // | |
13 | // Generic data structures. | |
14 | ||
15 | struct rcu_tasks; | |
16 | typedef void (*rcu_tasks_gp_func_t)(struct rcu_tasks *rtp); | |
e4fe5dd6 PM |
17 | typedef void (*pregp_func_t)(void); |
18 | typedef void (*pertask_func_t)(struct task_struct *t, struct list_head *hop); | |
9796e1ae | 19 | typedef void (*postscan_func_t)(struct list_head *hop); |
e4fe5dd6 | 20 | typedef void (*holdouts_func_t)(struct list_head *hop, bool ndrpt, bool *frptp); |
af051ca4 | 21 | typedef void (*postgp_func_t)(struct rcu_tasks *rtp); |
eacd6f04 | 22 | |
07e10515 | 23 | /** |
cafafd67 | 24 | * struct rcu_tasks_percpu - Per-CPU component of definition for a Tasks-RCU-like mechanism. |
9b073de1 | 25 | * @cblist: Callback list. |
381a4f3b | 26 | * @lock: Lock protecting per-CPU callback list. |
7d13d30b PM |
27 | * @rtp_jiffies: Jiffies counter value for statistics. |
28 | * @rtp_n_lock_retries: Rough lock-contention statistic. | |
d363f833 | 29 | * @rtp_work: Work queue for invoking callbacks. |
3063b33a | 30 | * @rtp_irq_work: IRQ work queue for deferred wakeups. |
ce9b1c66 PM |
31 | * @barrier_q_head: RCU callback for barrier operation. |
32 | * @cpu: CPU number corresponding to this entry. | |
33 | * @rtpp: Pointer to the rcu_tasks structure. | |
cafafd67 PM |
34 | */ |
35 | struct rcu_tasks_percpu { | |
9b073de1 | 36 | struct rcu_segcblist cblist; |
381a4f3b | 37 | raw_spinlock_t __private lock; |
7d13d30b PM |
38 | unsigned long rtp_jiffies; |
39 | unsigned long rtp_n_lock_retries; | |
d363f833 | 40 | struct work_struct rtp_work; |
3063b33a | 41 | struct irq_work rtp_irq_work; |
ce9b1c66 | 42 | struct rcu_head barrier_q_head; |
d363f833 PM |
43 | int cpu; |
44 | struct rcu_tasks *rtpp; | |
cafafd67 PM |
45 | }; |
46 | ||
47 | /** | |
48 | * struct rcu_tasks - Definition for a Tasks-RCU-like mechanism. | |
a616aec9 | 49 | * @cbs_wq: Wait queue allowing new callback to get kthread's attention. |
cafafd67 | 50 | * @cbs_gbl_lock: Lock protecting callback list. |
07e10515 | 51 | * @kthread_ptr: This flavor's grace-period/callback-invocation kthread. |
5873b8a9 | 52 | * @gp_func: This flavor's grace-period-wait function. |
af051ca4 | 53 | * @gp_state: Grace period's most recent state transition (debugging). |
4fe192df | 54 | * @gp_sleep: Per-grace-period sleep to prevent CPU-bound looping. |
2393a613 | 55 | * @init_fract: Initial backoff sleep interval. |
af051ca4 PM |
56 | * @gp_jiffies: Time of last @gp_state transition. |
57 | * @gp_start: Most recent grace-period start in jiffies. | |
b14fb4fb | 58 | * @tasks_gp_seq: Number of grace periods completed since boot. |
238dbce3 | 59 | * @n_ipis: Number of IPIs sent to encourage grace periods to end. |
7e0669c3 | 60 | * @n_ipis_fails: Number of IPI-send failures. |
e4fe5dd6 PM |
61 | * @pregp_func: This flavor's pre-grace-period function (optional). |
62 | * @pertask_func: This flavor's per-task scan function (optional). | |
63 | * @postscan_func: This flavor's post-task scan function (optional). | |
85b86994 | 64 | * @holdouts_func: This flavor's holdout-list scan function (optional). |
e4fe5dd6 | 65 | * @postgp_func: This flavor's post-grace-period function (optional). |
5873b8a9 | 66 | * @call_func: This flavor's call_rcu()-equivalent function. |
cafafd67 | 67 | * @rtpcpu: This flavor's rcu_tasks_percpu structure. |
7a30871b | 68 | * @percpu_enqueue_shift: Shift down CPU ID this much when enqueuing callbacks. |
2cee0789 PM |
69 | * @percpu_enqueue_lim: Number of per-CPU callback queues in use for enqueuing. |
70 | * @percpu_dequeue_lim: Number of per-CPU callback queues in use for dequeuing. | |
ce9b1c66 PM |
71 | * @barrier_q_mutex: Serialize barrier operations. |
72 | * @barrier_q_count: Number of queues being waited on. | |
73 | * @barrier_q_completion: Barrier wait/wakeup mechanism. | |
74 | * @barrier_q_seq: Sequence number for barrier operations. | |
c97d12a6 PM |
75 | * @name: This flavor's textual name. |
76 | * @kname: This flavor's kthread name. | |
07e10515 PM |
77 | */ |
78 | struct rcu_tasks { | |
07e10515 | 79 | struct wait_queue_head cbs_wq; |
cafafd67 | 80 | raw_spinlock_t cbs_gbl_lock; |
af051ca4 | 81 | int gp_state; |
4fe192df | 82 | int gp_sleep; |
2393a613 | 83 | int init_fract; |
af051ca4 | 84 | unsigned long gp_jiffies; |
88092d0c | 85 | unsigned long gp_start; |
b14fb4fb | 86 | unsigned long tasks_gp_seq; |
238dbce3 | 87 | unsigned long n_ipis; |
7e0669c3 | 88 | unsigned long n_ipis_fails; |
07e10515 | 89 | struct task_struct *kthread_ptr; |
5873b8a9 | 90 | rcu_tasks_gp_func_t gp_func; |
e4fe5dd6 PM |
91 | pregp_func_t pregp_func; |
92 | pertask_func_t pertask_func; | |
93 | postscan_func_t postscan_func; | |
94 | holdouts_func_t holdouts_func; | |
95 | postgp_func_t postgp_func; | |
5873b8a9 | 96 | call_rcu_func_t call_func; |
cafafd67 | 97 | struct rcu_tasks_percpu __percpu *rtpcpu; |
7a30871b | 98 | int percpu_enqueue_shift; |
8dd593fd | 99 | int percpu_enqueue_lim; |
2cee0789 | 100 | int percpu_dequeue_lim; |
ce9b1c66 PM |
101 | struct mutex barrier_q_mutex; |
102 | atomic_t barrier_q_count; | |
103 | struct completion barrier_q_completion; | |
104 | unsigned long barrier_q_seq; | |
c97d12a6 PM |
105 | char *name; |
106 | char *kname; | |
07e10515 PM |
107 | }; |
108 | ||
3063b33a PM |
109 | static void call_rcu_tasks_iw_wakeup(struct irq_work *iwp); |
110 | ||
cafafd67 PM |
111 | #define DEFINE_RCU_TASKS(rt_name, gp, call, n) \ |
112 | static DEFINE_PER_CPU(struct rcu_tasks_percpu, rt_name ## __percpu) = { \ | |
381a4f3b | 113 | .lock = __RAW_SPIN_LOCK_UNLOCKED(rt_name ## __percpu.cbs_pcpu_lock), \ |
3063b33a | 114 | .rtp_irq_work = IRQ_WORK_INIT(call_rcu_tasks_iw_wakeup), \ |
cafafd67 PM |
115 | }; \ |
116 | static struct rcu_tasks rt_name = \ | |
117 | { \ | |
118 | .cbs_wq = __WAIT_QUEUE_HEAD_INITIALIZER(rt_name.cbs_wq), \ | |
119 | .cbs_gbl_lock = __RAW_SPIN_LOCK_UNLOCKED(rt_name.cbs_gbl_lock), \ | |
120 | .gp_func = gp, \ | |
121 | .call_func = call, \ | |
122 | .rtpcpu = &rt_name ## __percpu, \ | |
123 | .name = n, \ | |
7a30871b | 124 | .percpu_enqueue_shift = ilog2(CONFIG_NR_CPUS), \ |
8dd593fd | 125 | .percpu_enqueue_lim = 1, \ |
2cee0789 | 126 | .percpu_dequeue_lim = 1, \ |
ce9b1c66 PM |
127 | .barrier_q_mutex = __MUTEX_INITIALIZER(rt_name.barrier_q_mutex), \ |
128 | .barrier_q_seq = (0UL - 50UL) << RCU_SEQ_CTR_SHIFT, \ | |
cafafd67 | 129 | .kname = #rt_name, \ |
07e10515 PM |
130 | } |
131 | ||
eacd6f04 PM |
132 | /* Track exiting tasks in order to allow them to be waited for. */ |
133 | DEFINE_STATIC_SRCU(tasks_rcu_exit_srcu); | |
134 | ||
b0afa0f0 | 135 | /* Avoid IPIing CPUs early in the grace period. */ |
574de876 | 136 | #define RCU_TASK_IPI_DELAY (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB) ? HZ / 2 : 0) |
b0afa0f0 PM |
137 | static int rcu_task_ipi_delay __read_mostly = RCU_TASK_IPI_DELAY; |
138 | module_param(rcu_task_ipi_delay, int, 0644); | |
139 | ||
eacd6f04 PM |
140 | /* Control stall timeouts. Disable with <= 0, otherwise jiffies till stall. */ |
141 | #define RCU_TASK_STALL_TIMEOUT (HZ * 60 * 10) | |
142 | static int rcu_task_stall_timeout __read_mostly = RCU_TASK_STALL_TIMEOUT; | |
143 | module_param(rcu_task_stall_timeout, int, 0644); | |
144 | ||
8610b656 PM |
145 | static int rcu_task_enqueue_lim __read_mostly = -1; |
146 | module_param(rcu_task_enqueue_lim, int, 0444); | |
147 | ||
ab97152f PM |
148 | static bool rcu_task_cb_adjust; |
149 | static int rcu_task_contend_lim __read_mostly = 100; | |
150 | module_param(rcu_task_contend_lim, int, 0444); | |
151 | ||
af051ca4 PM |
152 | /* RCU tasks grace-period state for debugging. */ |
153 | #define RTGS_INIT 0 | |
154 | #define RTGS_WAIT_WAIT_CBS 1 | |
155 | #define RTGS_WAIT_GP 2 | |
156 | #define RTGS_PRE_WAIT_GP 3 | |
157 | #define RTGS_SCAN_TASKLIST 4 | |
158 | #define RTGS_POST_SCAN_TASKLIST 5 | |
159 | #define RTGS_WAIT_SCAN_HOLDOUTS 6 | |
160 | #define RTGS_SCAN_HOLDOUTS 7 | |
161 | #define RTGS_POST_GP 8 | |
162 | #define RTGS_WAIT_READERS 9 | |
163 | #define RTGS_INVOKE_CBS 10 | |
164 | #define RTGS_WAIT_CBS 11 | |
8344496e | 165 | #ifndef CONFIG_TINY_RCU |
af051ca4 PM |
166 | static const char * const rcu_tasks_gp_state_names[] = { |
167 | "RTGS_INIT", | |
168 | "RTGS_WAIT_WAIT_CBS", | |
169 | "RTGS_WAIT_GP", | |
170 | "RTGS_PRE_WAIT_GP", | |
171 | "RTGS_SCAN_TASKLIST", | |
172 | "RTGS_POST_SCAN_TASKLIST", | |
173 | "RTGS_WAIT_SCAN_HOLDOUTS", | |
174 | "RTGS_SCAN_HOLDOUTS", | |
175 | "RTGS_POST_GP", | |
176 | "RTGS_WAIT_READERS", | |
177 | "RTGS_INVOKE_CBS", | |
178 | "RTGS_WAIT_CBS", | |
179 | }; | |
8344496e | 180 | #endif /* #ifndef CONFIG_TINY_RCU */ |
af051ca4 | 181 | |
5873b8a9 PM |
182 | //////////////////////////////////////////////////////////////////////// |
183 | // | |
184 | // Generic code. | |
185 | ||
d363f833 PM |
186 | static void rcu_tasks_invoke_cbs_wq(struct work_struct *wp); |
187 | ||
af051ca4 PM |
188 | /* Record grace-period phase and time. */ |
189 | static void set_tasks_gp_state(struct rcu_tasks *rtp, int newstate) | |
190 | { | |
191 | rtp->gp_state = newstate; | |
192 | rtp->gp_jiffies = jiffies; | |
193 | } | |
194 | ||
8344496e | 195 | #ifndef CONFIG_TINY_RCU |
af051ca4 PM |
196 | /* Return state name. */ |
197 | static const char *tasks_gp_state_getname(struct rcu_tasks *rtp) | |
198 | { | |
199 | int i = data_race(rtp->gp_state); // Let KCSAN detect update races | |
200 | int j = READ_ONCE(i); // Prevent the compiler from reading twice | |
201 | ||
202 | if (j >= ARRAY_SIZE(rcu_tasks_gp_state_names)) | |
203 | return "???"; | |
204 | return rcu_tasks_gp_state_names[j]; | |
205 | } | |
8344496e | 206 | #endif /* #ifndef CONFIG_TINY_RCU */ |
af051ca4 | 207 | |
cafafd67 PM |
208 | // Initialize per-CPU callback lists for the specified flavor of |
209 | // Tasks RCU. | |
210 | static void cblist_init_generic(struct rcu_tasks *rtp) | |
211 | { | |
212 | int cpu; | |
213 | unsigned long flags; | |
8610b656 | 214 | int lim; |
cafafd67 PM |
215 | |
216 | raw_spin_lock_irqsave(&rtp->cbs_gbl_lock, flags); | |
ab97152f PM |
217 | if (rcu_task_enqueue_lim < 0) { |
218 | rcu_task_enqueue_lim = 1; | |
219 | rcu_task_cb_adjust = true; | |
220 | pr_info("%s: Setting adjustable number of callback queues.\n", __func__); | |
221 | } else if (rcu_task_enqueue_lim == 0) { | |
8610b656 | 222 | rcu_task_enqueue_lim = 1; |
ab97152f | 223 | } |
8610b656 PM |
224 | lim = rcu_task_enqueue_lim; |
225 | ||
226 | if (lim > nr_cpu_ids) | |
227 | lim = nr_cpu_ids; | |
228 | WRITE_ONCE(rtp->percpu_enqueue_shift, ilog2(nr_cpu_ids / lim)); | |
2cee0789 | 229 | WRITE_ONCE(rtp->percpu_dequeue_lim, lim); |
8610b656 | 230 | smp_store_release(&rtp->percpu_enqueue_lim, lim); |
cafafd67 PM |
231 | for_each_possible_cpu(cpu) { |
232 | struct rcu_tasks_percpu *rtpcp = per_cpu_ptr(rtp->rtpcpu, cpu); | |
233 | ||
234 | WARN_ON_ONCE(!rtpcp); | |
235 | if (cpu) | |
381a4f3b PM |
236 | raw_spin_lock_init(&ACCESS_PRIVATE(rtpcp, lock)); |
237 | raw_spin_lock_rcu_node(rtpcp); // irqs already disabled. | |
9b073de1 PM |
238 | if (rcu_segcblist_empty(&rtpcp->cblist)) |
239 | rcu_segcblist_init(&rtpcp->cblist); | |
d363f833 PM |
240 | INIT_WORK(&rtpcp->rtp_work, rcu_tasks_invoke_cbs_wq); |
241 | rtpcp->cpu = cpu; | |
242 | rtpcp->rtpp = rtp; | |
381a4f3b | 243 | raw_spin_unlock_rcu_node(rtpcp); // irqs remain disabled. |
cafafd67 PM |
244 | } |
245 | raw_spin_unlock_irqrestore(&rtp->cbs_gbl_lock, flags); | |
8610b656 | 246 | pr_info("%s: Setting shift to %d and lim to %d.\n", __func__, data_race(rtp->percpu_enqueue_shift), data_race(rtp->percpu_enqueue_lim)); |
cafafd67 PM |
247 | } |
248 | ||
3063b33a PM |
249 | // IRQ-work handler that does deferred wakeup for call_rcu_tasks_generic(). |
250 | static void call_rcu_tasks_iw_wakeup(struct irq_work *iwp) | |
251 | { | |
252 | struct rcu_tasks *rtp; | |
253 | struct rcu_tasks_percpu *rtpcp = container_of(iwp, struct rcu_tasks_percpu, rtp_irq_work); | |
254 | ||
255 | rtp = rtpcp->rtpp; | |
256 | wake_up(&rtp->cbs_wq); | |
257 | } | |
258 | ||
5873b8a9 PM |
259 | // Enqueue a callback for the specified flavor of Tasks RCU. |
260 | static void call_rcu_tasks_generic(struct rcu_head *rhp, rcu_callback_t func, | |
261 | struct rcu_tasks *rtp) | |
eacd6f04 PM |
262 | { |
263 | unsigned long flags; | |
7d13d30b | 264 | unsigned long j; |
ab97152f | 265 | bool needadjust = false; |
eacd6f04 | 266 | bool needwake; |
cafafd67 | 267 | struct rcu_tasks_percpu *rtpcp; |
eacd6f04 PM |
268 | |
269 | rhp->next = NULL; | |
270 | rhp->func = func; | |
cafafd67 | 271 | local_irq_save(flags); |
7a30871b PM |
272 | rtpcp = per_cpu_ptr(rtp->rtpcpu, |
273 | smp_processor_id() >> READ_ONCE(rtp->percpu_enqueue_shift)); | |
7d13d30b PM |
274 | if (!raw_spin_trylock_rcu_node(rtpcp)) { // irqs already disabled. |
275 | raw_spin_lock_rcu_node(rtpcp); // irqs already disabled. | |
276 | j = jiffies; | |
277 | if (rtpcp->rtp_jiffies != j) { | |
278 | rtpcp->rtp_jiffies = j; | |
279 | rtpcp->rtp_n_lock_retries = 0; | |
280 | } | |
ab97152f PM |
281 | if (rcu_task_cb_adjust && ++rtpcp->rtp_n_lock_retries > rcu_task_contend_lim && |
282 | READ_ONCE(rtp->percpu_enqueue_lim) != nr_cpu_ids) | |
283 | needadjust = true; // Defer adjustment to avoid deadlock. | |
7d13d30b | 284 | } |
9b073de1 | 285 | if (!rcu_segcblist_is_enabled(&rtpcp->cblist)) { |
381a4f3b | 286 | raw_spin_unlock_rcu_node(rtpcp); // irqs remain disabled. |
cafafd67 | 287 | cblist_init_generic(rtp); |
381a4f3b | 288 | raw_spin_lock_rcu_node(rtpcp); // irqs already disabled. |
cafafd67 | 289 | } |
9b073de1 PM |
290 | needwake = rcu_segcblist_empty(&rtpcp->cblist); |
291 | rcu_segcblist_enqueue(&rtpcp->cblist, rhp); | |
381a4f3b | 292 | raw_spin_unlock_irqrestore_rcu_node(rtpcp, flags); |
ab97152f PM |
293 | if (unlikely(needadjust)) { |
294 | raw_spin_lock_irqsave(&rtp->cbs_gbl_lock, flags); | |
295 | if (rtp->percpu_enqueue_lim != nr_cpu_ids) { | |
296 | WRITE_ONCE(rtp->percpu_enqueue_shift, ilog2(nr_cpu_ids)); | |
2cee0789 | 297 | WRITE_ONCE(rtp->percpu_enqueue_lim, nr_cpu_ids); |
ab97152f PM |
298 | smp_store_release(&rtp->percpu_enqueue_lim, nr_cpu_ids); |
299 | pr_info("Switching %s to per-CPU callback queuing.\n", rtp->name); | |
300 | } | |
301 | raw_spin_unlock_irqrestore(&rtp->cbs_gbl_lock, flags); | |
302 | } | |
eacd6f04 | 303 | /* We can't create the thread unless interrupts are enabled. */ |
07e10515 | 304 | if (needwake && READ_ONCE(rtp->kthread_ptr)) |
3063b33a | 305 | irq_work_queue(&rtpcp->rtp_irq_work); |
eacd6f04 | 306 | } |
eacd6f04 | 307 | |
5873b8a9 PM |
308 | // Wait for a grace period for the specified flavor of Tasks RCU. |
309 | static void synchronize_rcu_tasks_generic(struct rcu_tasks *rtp) | |
eacd6f04 PM |
310 | { |
311 | /* Complain if the scheduler has not started. */ | |
312 | RCU_LOCKDEP_WARN(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE, | |
313 | "synchronize_rcu_tasks called too soon"); | |
314 | ||
315 | /* Wait for the grace period. */ | |
5873b8a9 | 316 | wait_rcu_gp(rtp->call_func); |
eacd6f04 PM |
317 | } |
318 | ||
ce9b1c66 PM |
319 | // RCU callback function for rcu_barrier_tasks_generic(). |
320 | static void rcu_barrier_tasks_generic_cb(struct rcu_head *rhp) | |
321 | { | |
322 | struct rcu_tasks *rtp; | |
323 | struct rcu_tasks_percpu *rtpcp; | |
324 | ||
325 | rtpcp = container_of(rhp, struct rcu_tasks_percpu, barrier_q_head); | |
326 | rtp = rtpcp->rtpp; | |
327 | if (atomic_dec_and_test(&rtp->barrier_q_count)) | |
328 | complete(&rtp->barrier_q_completion); | |
329 | } | |
330 | ||
331 | // Wait for all in-flight callbacks for the specified RCU Tasks flavor. | |
332 | // Operates in a manner similar to rcu_barrier(). | |
333 | static void rcu_barrier_tasks_generic(struct rcu_tasks *rtp) | |
334 | { | |
335 | int cpu; | |
336 | unsigned long flags; | |
337 | struct rcu_tasks_percpu *rtpcp; | |
338 | unsigned long s = rcu_seq_snap(&rtp->barrier_q_seq); | |
339 | ||
340 | mutex_lock(&rtp->barrier_q_mutex); | |
341 | if (rcu_seq_done(&rtp->barrier_q_seq, s)) { | |
342 | smp_mb(); | |
343 | mutex_unlock(&rtp->barrier_q_mutex); | |
344 | return; | |
345 | } | |
346 | rcu_seq_start(&rtp->barrier_q_seq); | |
347 | init_completion(&rtp->barrier_q_completion); | |
348 | atomic_set(&rtp->barrier_q_count, 2); | |
349 | for_each_possible_cpu(cpu) { | |
2cee0789 | 350 | if (cpu >= smp_load_acquire(&rtp->percpu_dequeue_lim)) |
ce9b1c66 PM |
351 | break; |
352 | rtpcp = per_cpu_ptr(rtp->rtpcpu, cpu); | |
353 | rtpcp->barrier_q_head.func = rcu_barrier_tasks_generic_cb; | |
354 | raw_spin_lock_irqsave_rcu_node(rtpcp, flags); | |
355 | if (rcu_segcblist_entrain(&rtpcp->cblist, &rtpcp->barrier_q_head)) | |
356 | atomic_inc(&rtp->barrier_q_count); | |
357 | raw_spin_unlock_irqrestore_rcu_node(rtpcp, flags); | |
358 | } | |
359 | if (atomic_sub_and_test(2, &rtp->barrier_q_count)) | |
360 | complete(&rtp->barrier_q_completion); | |
361 | wait_for_completion(&rtp->barrier_q_completion); | |
362 | rcu_seq_end(&rtp->barrier_q_seq); | |
363 | mutex_unlock(&rtp->barrier_q_mutex); | |
364 | } | |
365 | ||
4d1114c0 PM |
366 | // Advance callbacks and indicate whether either a grace period or |
367 | // callback invocation is needed. | |
368 | static int rcu_tasks_need_gpcb(struct rcu_tasks *rtp) | |
369 | { | |
370 | int cpu; | |
371 | unsigned long flags; | |
372 | int needgpcb = 0; | |
373 | ||
2cee0789 | 374 | for (cpu = 0; cpu < smp_load_acquire(&rtp->percpu_dequeue_lim); cpu++) { |
4d1114c0 PM |
375 | struct rcu_tasks_percpu *rtpcp = per_cpu_ptr(rtp->rtpcpu, cpu); |
376 | ||
377 | /* Advance and accelerate any new callbacks. */ | |
378 | if (rcu_segcblist_empty(&rtpcp->cblist)) | |
379 | continue; | |
380 | raw_spin_lock_irqsave_rcu_node(rtpcp, flags); | |
381 | rcu_segcblist_advance(&rtpcp->cblist, rcu_seq_current(&rtp->tasks_gp_seq)); | |
382 | (void)rcu_segcblist_accelerate(&rtpcp->cblist, rcu_seq_snap(&rtp->tasks_gp_seq)); | |
383 | if (rcu_segcblist_pend_cbs(&rtpcp->cblist)) | |
384 | needgpcb |= 0x3; | |
385 | if (!rcu_segcblist_empty(&rtpcp->cblist)) | |
386 | needgpcb |= 0x1; | |
387 | raw_spin_unlock_irqrestore_rcu_node(rtpcp, flags); | |
388 | } | |
389 | return needgpcb; | |
390 | } | |
391 | ||
57881863 | 392 | // Advance callbacks and invoke any that are ready. |
d363f833 | 393 | static void rcu_tasks_invoke_cbs(struct rcu_tasks *rtp, struct rcu_tasks_percpu *rtpcp) |
eacd6f04 | 394 | { |
57881863 | 395 | int cpu; |
d363f833 | 396 | int cpunext; |
eacd6f04 | 397 | unsigned long flags; |
9b073de1 | 398 | int len; |
9b073de1 | 399 | struct rcu_head *rhp; |
d363f833 PM |
400 | struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl); |
401 | struct rcu_tasks_percpu *rtpcp_next; | |
402 | ||
403 | cpu = rtpcp->cpu; | |
404 | cpunext = cpu * 2 + 1; | |
2cee0789 | 405 | if (cpunext < smp_load_acquire(&rtp->percpu_dequeue_lim)) { |
d363f833 PM |
406 | rtpcp_next = per_cpu_ptr(rtp->rtpcpu, cpunext); |
407 | queue_work_on(cpunext, system_wq, &rtpcp_next->rtp_work); | |
408 | cpunext++; | |
2cee0789 | 409 | if (cpunext < smp_load_acquire(&rtp->percpu_dequeue_lim)) { |
d363f833 PM |
410 | rtpcp_next = per_cpu_ptr(rtp->rtpcpu, cpunext); |
411 | queue_work_on(cpunext, system_wq, &rtpcp_next->rtp_work); | |
57881863 | 412 | } |
57881863 | 413 | } |
d363f833 PM |
414 | |
415 | if (rcu_segcblist_empty(&rtpcp->cblist)) | |
416 | return; | |
417 | raw_spin_lock_irqsave_rcu_node(rtpcp, flags); | |
418 | rcu_segcblist_advance(&rtpcp->cblist, rcu_seq_current(&rtp->tasks_gp_seq)); | |
419 | rcu_segcblist_extract_done_cbs(&rtpcp->cblist, &rcl); | |
420 | raw_spin_unlock_irqrestore_rcu_node(rtpcp, flags); | |
421 | len = rcl.len; | |
422 | for (rhp = rcu_cblist_dequeue(&rcl); rhp; rhp = rcu_cblist_dequeue(&rcl)) { | |
423 | local_bh_disable(); | |
424 | rhp->func(rhp); | |
425 | local_bh_enable(); | |
426 | cond_resched(); | |
427 | } | |
428 | raw_spin_lock_irqsave_rcu_node(rtpcp, flags); | |
429 | rcu_segcblist_add_len(&rtpcp->cblist, -len); | |
430 | (void)rcu_segcblist_accelerate(&rtpcp->cblist, rcu_seq_snap(&rtp->tasks_gp_seq)); | |
431 | raw_spin_unlock_irqrestore_rcu_node(rtpcp, flags); | |
432 | } | |
433 | ||
434 | // Workqueue flood to advance callbacks and invoke any that are ready. | |
435 | static void rcu_tasks_invoke_cbs_wq(struct work_struct *wp) | |
436 | { | |
437 | struct rcu_tasks *rtp; | |
438 | struct rcu_tasks_percpu *rtpcp = container_of(wp, struct rcu_tasks_percpu, rtp_work); | |
439 | ||
440 | rtp = rtpcp->rtpp; | |
441 | rcu_tasks_invoke_cbs(rtp, rtpcp); | |
57881863 PM |
442 | } |
443 | ||
444 | /* RCU-tasks kthread that detects grace periods and invokes callbacks. */ | |
445 | static int __noreturn rcu_tasks_kthread(void *arg) | |
446 | { | |
447 | int needgpcb; | |
07e10515 | 448 | struct rcu_tasks *rtp = arg; |
eacd6f04 PM |
449 | |
450 | /* Run on housekeeping CPUs by default. Sysadm can move if desired. */ | |
451 | housekeeping_affine(current, HK_FLAG_RCU); | |
07e10515 | 452 | WRITE_ONCE(rtp->kthread_ptr, current); // Let GPs start! |
eacd6f04 PM |
453 | |
454 | /* | |
455 | * Each pass through the following loop makes one check for | |
456 | * newly arrived callbacks, and, if there are some, waits for | |
457 | * one RCU-tasks grace period and then invokes the callbacks. | |
458 | * This loop is terminated by the system going down. ;-) | |
459 | */ | |
460 | for (;;) { | |
0db7c32a | 461 | set_tasks_gp_state(rtp, RTGS_WAIT_CBS); |
eacd6f04 | 462 | |
eacd6f04 | 463 | /* If there were none, wait a bit and start over. */ |
4d1114c0 | 464 | wait_event_idle(rtp->cbs_wq, (needgpcb = rcu_tasks_need_gpcb(rtp))); |
eacd6f04 | 465 | |
4d1114c0 PM |
466 | if (needgpcb & 0x2) { |
467 | // Wait for one grace period. | |
468 | set_tasks_gp_state(rtp, RTGS_WAIT_GP); | |
469 | rtp->gp_start = jiffies; | |
470 | rcu_seq_start(&rtp->tasks_gp_seq); | |
471 | rtp->gp_func(rtp); | |
472 | rcu_seq_end(&rtp->tasks_gp_seq); | |
473 | } | |
eacd6f04 | 474 | |
57881863 | 475 | /* Invoke callbacks. */ |
af051ca4 | 476 | set_tasks_gp_state(rtp, RTGS_INVOKE_CBS); |
d363f833 | 477 | rcu_tasks_invoke_cbs(rtp, per_cpu_ptr(rtp->rtpcpu, 0)); |
57881863 | 478 | |
eacd6f04 | 479 | /* Paranoid sleep to keep this from entering a tight loop */ |
4fe192df | 480 | schedule_timeout_idle(rtp->gp_sleep); |
eacd6f04 PM |
481 | } |
482 | } | |
483 | ||
1b04fa99 | 484 | /* Spawn RCU-tasks grace-period kthread. */ |
5873b8a9 | 485 | static void __init rcu_spawn_tasks_kthread_generic(struct rcu_tasks *rtp) |
eacd6f04 PM |
486 | { |
487 | struct task_struct *t; | |
488 | ||
c97d12a6 PM |
489 | t = kthread_run(rcu_tasks_kthread, rtp, "%s_kthread", rtp->kname); |
490 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start %s grace-period kthread, OOM is now expected behavior\n", __func__, rtp->name)) | |
5873b8a9 | 491 | return; |
eacd6f04 | 492 | smp_mb(); /* Ensure others see full kthread. */ |
eacd6f04 | 493 | } |
eacd6f04 | 494 | |
eacd6f04 PM |
495 | #ifndef CONFIG_TINY_RCU |
496 | ||
497 | /* | |
498 | * Print any non-default Tasks RCU settings. | |
499 | */ | |
500 | static void __init rcu_tasks_bootup_oddness(void) | |
501 | { | |
d5f177d3 | 502 | #if defined(CONFIG_TASKS_RCU) || defined(CONFIG_TASKS_TRACE_RCU) |
eacd6f04 PM |
503 | if (rcu_task_stall_timeout != RCU_TASK_STALL_TIMEOUT) |
504 | pr_info("\tTasks-RCU CPU stall warnings timeout set to %d (rcu_task_stall_timeout).\n", rcu_task_stall_timeout); | |
d5f177d3 PM |
505 | #endif /* #ifdef CONFIG_TASKS_RCU */ |
506 | #ifdef CONFIG_TASKS_RCU | |
507 | pr_info("\tTrampoline variant of Tasks RCU enabled.\n"); | |
eacd6f04 | 508 | #endif /* #ifdef CONFIG_TASKS_RCU */ |
c84aad76 PM |
509 | #ifdef CONFIG_TASKS_RUDE_RCU |
510 | pr_info("\tRude variant of Tasks RCU enabled.\n"); | |
511 | #endif /* #ifdef CONFIG_TASKS_RUDE_RCU */ | |
d5f177d3 PM |
512 | #ifdef CONFIG_TASKS_TRACE_RCU |
513 | pr_info("\tTracing variant of Tasks RCU enabled.\n"); | |
514 | #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */ | |
eacd6f04 PM |
515 | } |
516 | ||
517 | #endif /* #ifndef CONFIG_TINY_RCU */ | |
5873b8a9 | 518 | |
8344496e | 519 | #ifndef CONFIG_TINY_RCU |
e21408ce PM |
520 | /* Dump out rcutorture-relevant state common to all RCU-tasks flavors. */ |
521 | static void show_rcu_tasks_generic_gp_kthread(struct rcu_tasks *rtp, char *s) | |
522 | { | |
cafafd67 | 523 | struct rcu_tasks_percpu *rtpcp = per_cpu_ptr(rtp->rtpcpu, 0); // for_each... |
7e0669c3 | 524 | pr_info("%s: %s(%d) since %lu g:%lu i:%lu/%lu %c%c %s\n", |
e21408ce | 525 | rtp->kname, |
7e0669c3 | 526 | tasks_gp_state_getname(rtp), data_race(rtp->gp_state), |
af051ca4 | 527 | jiffies - data_race(rtp->gp_jiffies), |
b14fb4fb | 528 | data_race(rcu_seq_current(&rtp->tasks_gp_seq)), |
7e0669c3 | 529 | data_race(rtp->n_ipis_fails), data_race(rtp->n_ipis), |
e21408ce | 530 | ".k"[!!data_race(rtp->kthread_ptr)], |
9b073de1 | 531 | ".C"[!data_race(rcu_segcblist_empty(&rtpcp->cblist))], |
e21408ce PM |
532 | s); |
533 | } | |
27c0f144 | 534 | #endif // #ifndef CONFIG_TINY_RCU |
e21408ce | 535 | |
25246fc8 PM |
536 | static void exit_tasks_rcu_finish_trace(struct task_struct *t); |
537 | ||
538 | #if defined(CONFIG_TASKS_RCU) || defined(CONFIG_TASKS_TRACE_RCU) | |
5873b8a9 | 539 | |
d01aa263 PM |
540 | //////////////////////////////////////////////////////////////////////// |
541 | // | |
542 | // Shared code between task-list-scanning variants of Tasks RCU. | |
543 | ||
544 | /* Wait for one RCU-tasks grace period. */ | |
545 | static void rcu_tasks_wait_gp(struct rcu_tasks *rtp) | |
546 | { | |
547 | struct task_struct *g, *t; | |
548 | unsigned long lastreport; | |
549 | LIST_HEAD(holdouts); | |
550 | int fract; | |
551 | ||
af051ca4 | 552 | set_tasks_gp_state(rtp, RTGS_PRE_WAIT_GP); |
d01aa263 PM |
553 | rtp->pregp_func(); |
554 | ||
555 | /* | |
556 | * There were callbacks, so we need to wait for an RCU-tasks | |
557 | * grace period. Start off by scanning the task list for tasks | |
558 | * that are not already voluntarily blocked. Mark these tasks | |
559 | * and make a list of them in holdouts. | |
560 | */ | |
af051ca4 | 561 | set_tasks_gp_state(rtp, RTGS_SCAN_TASKLIST); |
d01aa263 PM |
562 | rcu_read_lock(); |
563 | for_each_process_thread(g, t) | |
564 | rtp->pertask_func(t, &holdouts); | |
565 | rcu_read_unlock(); | |
566 | ||
af051ca4 | 567 | set_tasks_gp_state(rtp, RTGS_POST_SCAN_TASKLIST); |
9796e1ae | 568 | rtp->postscan_func(&holdouts); |
d01aa263 PM |
569 | |
570 | /* | |
571 | * Each pass through the following loop scans the list of holdout | |
572 | * tasks, removing any that are no longer holdouts. When the list | |
573 | * is empty, we are done. | |
574 | */ | |
575 | lastreport = jiffies; | |
576 | ||
2393a613 PM |
577 | // Start off with initial wait and slowly back off to 1 HZ wait. |
578 | fract = rtp->init_fract; | |
d01aa263 | 579 | |
77dc1741 | 580 | while (!list_empty(&holdouts)) { |
d01aa263 PM |
581 | bool firstreport; |
582 | bool needreport; | |
583 | int rtst; | |
584 | ||
d01aa263 | 585 | /* Slowly back off waiting for holdouts */ |
af051ca4 | 586 | set_tasks_gp_state(rtp, RTGS_WAIT_SCAN_HOLDOUTS); |
75dc2da5 | 587 | schedule_timeout_idle(fract); |
d01aa263 | 588 | |
75dc2da5 PM |
589 | if (fract < HZ) |
590 | fract++; | |
d01aa263 PM |
591 | |
592 | rtst = READ_ONCE(rcu_task_stall_timeout); | |
593 | needreport = rtst > 0 && time_after(jiffies, lastreport + rtst); | |
594 | if (needreport) | |
595 | lastreport = jiffies; | |
596 | firstreport = true; | |
597 | WARN_ON(signal_pending(current)); | |
af051ca4 | 598 | set_tasks_gp_state(rtp, RTGS_SCAN_HOLDOUTS); |
d01aa263 PM |
599 | rtp->holdouts_func(&holdouts, needreport, &firstreport); |
600 | } | |
601 | ||
af051ca4 PM |
602 | set_tasks_gp_state(rtp, RTGS_POST_GP); |
603 | rtp->postgp_func(rtp); | |
d01aa263 PM |
604 | } |
605 | ||
25246fc8 PM |
606 | #endif /* #if defined(CONFIG_TASKS_RCU) || defined(CONFIG_TASKS_TRACE_RCU) */ |
607 | ||
608 | #ifdef CONFIG_TASKS_RCU | |
609 | ||
5873b8a9 PM |
610 | //////////////////////////////////////////////////////////////////////// |
611 | // | |
612 | // Simple variant of RCU whose quiescent states are voluntary context | |
8af9e2c7 | 613 | // switch, cond_resched_tasks_rcu_qs(), user-space execution, and idle. |
5873b8a9 PM |
614 | // As such, grace periods can take one good long time. There are no |
615 | // read-side primitives similar to rcu_read_lock() and rcu_read_unlock() | |
616 | // because this implementation is intended to get the system into a safe | |
617 | // state for some of the manipulations involved in tracing and the like. | |
618 | // Finally, this implementation does not support high call_rcu_tasks() | |
619 | // rates from multiple CPUs. If this is required, per-CPU callback lists | |
620 | // will be needed. | |
06a3ec92 PM |
621 | // |
622 | // The implementation uses rcu_tasks_wait_gp(), which relies on function | |
623 | // pointers in the rcu_tasks structure. The rcu_spawn_tasks_kthread() | |
624 | // function sets these function pointers up so that rcu_tasks_wait_gp() | |
625 | // invokes these functions in this order: | |
626 | // | |
627 | // rcu_tasks_pregp_step(): | |
628 | // Invokes synchronize_rcu() in order to wait for all in-flight | |
629 | // t->on_rq and t->nvcsw transitions to complete. This works because | |
630 | // all such transitions are carried out with interrupts disabled. | |
631 | // rcu_tasks_pertask(), invoked on every non-idle task: | |
632 | // For every runnable non-idle task other than the current one, use | |
633 | // get_task_struct() to pin down that task, snapshot that task's | |
634 | // number of voluntary context switches, and add that task to the | |
635 | // holdout list. | |
636 | // rcu_tasks_postscan(): | |
637 | // Invoke synchronize_srcu() to ensure that all tasks that were | |
638 | // in the process of exiting (and which thus might not know to | |
639 | // synchronize with this RCU Tasks grace period) have completed | |
640 | // exiting. | |
641 | // check_all_holdout_tasks(), repeatedly until holdout list is empty: | |
642 | // Scans the holdout list, attempting to identify a quiescent state | |
643 | // for each task on the list. If there is a quiescent state, the | |
644 | // corresponding task is removed from the holdout list. | |
645 | // rcu_tasks_postgp(): | |
646 | // Invokes synchronize_rcu() in order to ensure that all prior | |
647 | // t->on_rq and t->nvcsw transitions are seen by all CPUs and tasks | |
648 | // to have happened before the end of this RCU Tasks grace period. | |
649 | // Again, this works because all such transitions are carried out | |
650 | // with interrupts disabled. | |
651 | // | |
652 | // For each exiting task, the exit_tasks_rcu_start() and | |
653 | // exit_tasks_rcu_finish() functions begin and end, respectively, the SRCU | |
654 | // read-side critical sections waited for by rcu_tasks_postscan(). | |
655 | // | |
381a4f3b PM |
656 | // Pre-grace-period update-side code is ordered before the grace |
657 | // via the raw_spin_lock.*rcu_node(). Pre-grace-period read-side code | |
658 | // is ordered before the grace period via synchronize_rcu() call in | |
659 | // rcu_tasks_pregp_step() and by the scheduler's locks and interrupt | |
06a3ec92 | 660 | // disabling. |
5873b8a9 | 661 | |
e4fe5dd6 PM |
662 | /* Pre-grace-period preparation. */ |
663 | static void rcu_tasks_pregp_step(void) | |
664 | { | |
665 | /* | |
666 | * Wait for all pre-existing t->on_rq and t->nvcsw transitions | |
667 | * to complete. Invoking synchronize_rcu() suffices because all | |
668 | * these transitions occur with interrupts disabled. Without this | |
669 | * synchronize_rcu(), a read-side critical section that started | |
670 | * before the grace period might be incorrectly seen as having | |
671 | * started after the grace period. | |
672 | * | |
673 | * This synchronize_rcu() also dispenses with the need for a | |
674 | * memory barrier on the first store to t->rcu_tasks_holdout, | |
675 | * as it forces the store to happen after the beginning of the | |
676 | * grace period. | |
677 | */ | |
678 | synchronize_rcu(); | |
679 | } | |
680 | ||
681 | /* Per-task initial processing. */ | |
682 | static void rcu_tasks_pertask(struct task_struct *t, struct list_head *hop) | |
683 | { | |
684 | if (t != current && READ_ONCE(t->on_rq) && !is_idle_task(t)) { | |
685 | get_task_struct(t); | |
686 | t->rcu_tasks_nvcsw = READ_ONCE(t->nvcsw); | |
687 | WRITE_ONCE(t->rcu_tasks_holdout, true); | |
688 | list_add(&t->rcu_tasks_holdout_list, hop); | |
689 | } | |
690 | } | |
691 | ||
692 | /* Processing between scanning taskslist and draining the holdout list. */ | |
04a3c5aa | 693 | static void rcu_tasks_postscan(struct list_head *hop) |
e4fe5dd6 PM |
694 | { |
695 | /* | |
696 | * Wait for tasks that are in the process of exiting. This | |
697 | * does only part of the job, ensuring that all tasks that were | |
698 | * previously exiting reach the point where they have disabled | |
699 | * preemption, allowing the later synchronize_rcu() to finish | |
700 | * the job. | |
701 | */ | |
702 | synchronize_srcu(&tasks_rcu_exit_srcu); | |
703 | } | |
704 | ||
5873b8a9 PM |
705 | /* See if tasks are still holding out, complain if so. */ |
706 | static void check_holdout_task(struct task_struct *t, | |
707 | bool needreport, bool *firstreport) | |
708 | { | |
709 | int cpu; | |
710 | ||
711 | if (!READ_ONCE(t->rcu_tasks_holdout) || | |
712 | t->rcu_tasks_nvcsw != READ_ONCE(t->nvcsw) || | |
713 | !READ_ONCE(t->on_rq) || | |
714 | (IS_ENABLED(CONFIG_NO_HZ_FULL) && | |
715 | !is_idle_task(t) && t->rcu_tasks_idle_cpu >= 0)) { | |
716 | WRITE_ONCE(t->rcu_tasks_holdout, false); | |
717 | list_del_init(&t->rcu_tasks_holdout_list); | |
718 | put_task_struct(t); | |
719 | return; | |
720 | } | |
721 | rcu_request_urgent_qs_task(t); | |
722 | if (!needreport) | |
723 | return; | |
724 | if (*firstreport) { | |
725 | pr_err("INFO: rcu_tasks detected stalls on tasks:\n"); | |
726 | *firstreport = false; | |
727 | } | |
728 | cpu = task_cpu(t); | |
729 | pr_alert("%p: %c%c nvcsw: %lu/%lu holdout: %d idle_cpu: %d/%d\n", | |
730 | t, ".I"[is_idle_task(t)], | |
731 | "N."[cpu < 0 || !tick_nohz_full_cpu(cpu)], | |
732 | t->rcu_tasks_nvcsw, t->nvcsw, t->rcu_tasks_holdout, | |
733 | t->rcu_tasks_idle_cpu, cpu); | |
734 | sched_show_task(t); | |
735 | } | |
736 | ||
e4fe5dd6 PM |
737 | /* Scan the holdout lists for tasks no longer holding out. */ |
738 | static void check_all_holdout_tasks(struct list_head *hop, | |
739 | bool needreport, bool *firstreport) | |
740 | { | |
741 | struct task_struct *t, *t1; | |
742 | ||
743 | list_for_each_entry_safe(t, t1, hop, rcu_tasks_holdout_list) { | |
744 | check_holdout_task(t, needreport, firstreport); | |
745 | cond_resched(); | |
746 | } | |
747 | } | |
748 | ||
749 | /* Finish off the Tasks-RCU grace period. */ | |
af051ca4 | 750 | static void rcu_tasks_postgp(struct rcu_tasks *rtp) |
e4fe5dd6 PM |
751 | { |
752 | /* | |
753 | * Because ->on_rq and ->nvcsw are not guaranteed to have a full | |
754 | * memory barriers prior to them in the schedule() path, memory | |
755 | * reordering on other CPUs could cause their RCU-tasks read-side | |
756 | * critical sections to extend past the end of the grace period. | |
757 | * However, because these ->nvcsw updates are carried out with | |
758 | * interrupts disabled, we can use synchronize_rcu() to force the | |
759 | * needed ordering on all such CPUs. | |
760 | * | |
761 | * This synchronize_rcu() also confines all ->rcu_tasks_holdout | |
762 | * accesses to be within the grace period, avoiding the need for | |
763 | * memory barriers for ->rcu_tasks_holdout accesses. | |
764 | * | |
765 | * In addition, this synchronize_rcu() waits for exiting tasks | |
766 | * to complete their final preempt_disable() region of execution, | |
767 | * cleaning up after the synchronize_srcu() above. | |
768 | */ | |
769 | synchronize_rcu(); | |
770 | } | |
771 | ||
5873b8a9 | 772 | void call_rcu_tasks(struct rcu_head *rhp, rcu_callback_t func); |
c97d12a6 | 773 | DEFINE_RCU_TASKS(rcu_tasks, rcu_tasks_wait_gp, call_rcu_tasks, "RCU Tasks"); |
5873b8a9 PM |
774 | |
775 | /** | |
776 | * call_rcu_tasks() - Queue an RCU for invocation task-based grace period | |
777 | * @rhp: structure to be used for queueing the RCU updates. | |
778 | * @func: actual callback function to be invoked after the grace period | |
779 | * | |
780 | * The callback function will be invoked some time after a full grace | |
781 | * period elapses, in other words after all currently executing RCU | |
782 | * read-side critical sections have completed. call_rcu_tasks() assumes | |
783 | * that the read-side critical sections end at a voluntary context | |
8af9e2c7 | 784 | * switch (not a preemption!), cond_resched_tasks_rcu_qs(), entry into idle, |
5873b8a9 PM |
785 | * or transition to usermode execution. As such, there are no read-side |
786 | * primitives analogous to rcu_read_lock() and rcu_read_unlock() because | |
787 | * this primitive is intended to determine that all tasks have passed | |
a616aec9 | 788 | * through a safe state, not so much for data-structure synchronization. |
5873b8a9 PM |
789 | * |
790 | * See the description of call_rcu() for more detailed information on | |
791 | * memory ordering guarantees. | |
792 | */ | |
793 | void call_rcu_tasks(struct rcu_head *rhp, rcu_callback_t func) | |
794 | { | |
795 | call_rcu_tasks_generic(rhp, func, &rcu_tasks); | |
796 | } | |
797 | EXPORT_SYMBOL_GPL(call_rcu_tasks); | |
798 | ||
799 | /** | |
800 | * synchronize_rcu_tasks - wait until an rcu-tasks grace period has elapsed. | |
801 | * | |
802 | * Control will return to the caller some time after a full rcu-tasks | |
803 | * grace period has elapsed, in other words after all currently | |
804 | * executing rcu-tasks read-side critical sections have elapsed. These | |
805 | * read-side critical sections are delimited by calls to schedule(), | |
806 | * cond_resched_tasks_rcu_qs(), idle execution, userspace execution, calls | |
807 | * to synchronize_rcu_tasks(), and (in theory, anyway) cond_resched(). | |
808 | * | |
809 | * This is a very specialized primitive, intended only for a few uses in | |
810 | * tracing and other situations requiring manipulation of function | |
811 | * preambles and profiling hooks. The synchronize_rcu_tasks() function | |
812 | * is not (yet) intended for heavy use from multiple CPUs. | |
813 | * | |
814 | * See the description of synchronize_rcu() for more detailed information | |
815 | * on memory ordering guarantees. | |
816 | */ | |
817 | void synchronize_rcu_tasks(void) | |
818 | { | |
819 | synchronize_rcu_tasks_generic(&rcu_tasks); | |
820 | } | |
821 | EXPORT_SYMBOL_GPL(synchronize_rcu_tasks); | |
822 | ||
823 | /** | |
824 | * rcu_barrier_tasks - Wait for in-flight call_rcu_tasks() callbacks. | |
825 | * | |
826 | * Although the current implementation is guaranteed to wait, it is not | |
827 | * obligated to, for example, if there are no pending callbacks. | |
828 | */ | |
829 | void rcu_barrier_tasks(void) | |
830 | { | |
ce9b1c66 | 831 | rcu_barrier_tasks_generic(&rcu_tasks); |
5873b8a9 PM |
832 | } |
833 | EXPORT_SYMBOL_GPL(rcu_barrier_tasks); | |
834 | ||
835 | static int __init rcu_spawn_tasks_kthread(void) | |
836 | { | |
cafafd67 | 837 | cblist_init_generic(&rcu_tasks); |
4fe192df | 838 | rcu_tasks.gp_sleep = HZ / 10; |
75dc2da5 | 839 | rcu_tasks.init_fract = HZ / 10; |
e4fe5dd6 PM |
840 | rcu_tasks.pregp_func = rcu_tasks_pregp_step; |
841 | rcu_tasks.pertask_func = rcu_tasks_pertask; | |
842 | rcu_tasks.postscan_func = rcu_tasks_postscan; | |
843 | rcu_tasks.holdouts_func = check_all_holdout_tasks; | |
844 | rcu_tasks.postgp_func = rcu_tasks_postgp; | |
5873b8a9 PM |
845 | rcu_spawn_tasks_kthread_generic(&rcu_tasks); |
846 | return 0; | |
847 | } | |
5873b8a9 | 848 | |
27c0f144 PM |
849 | #if !defined(CONFIG_TINY_RCU) |
850 | void show_rcu_tasks_classic_gp_kthread(void) | |
e21408ce PM |
851 | { |
852 | show_rcu_tasks_generic_gp_kthread(&rcu_tasks, ""); | |
853 | } | |
27c0f144 PM |
854 | EXPORT_SYMBOL_GPL(show_rcu_tasks_classic_gp_kthread); |
855 | #endif // !defined(CONFIG_TINY_RCU) | |
e21408ce | 856 | |
25246fc8 PM |
857 | /* Do the srcu_read_lock() for the above synchronize_srcu(). */ |
858 | void exit_tasks_rcu_start(void) __acquires(&tasks_rcu_exit_srcu) | |
859 | { | |
860 | preempt_disable(); | |
861 | current->rcu_tasks_idx = __srcu_read_lock(&tasks_rcu_exit_srcu); | |
862 | preempt_enable(); | |
863 | } | |
864 | ||
865 | /* Do the srcu_read_unlock() for the above synchronize_srcu(). */ | |
866 | void exit_tasks_rcu_finish(void) __releases(&tasks_rcu_exit_srcu) | |
867 | { | |
868 | struct task_struct *t = current; | |
869 | ||
870 | preempt_disable(); | |
871 | __srcu_read_unlock(&tasks_rcu_exit_srcu, t->rcu_tasks_idx); | |
872 | preempt_enable(); | |
873 | exit_tasks_rcu_finish_trace(t); | |
874 | } | |
875 | ||
e21408ce | 876 | #else /* #ifdef CONFIG_TASKS_RCU */ |
25246fc8 PM |
877 | void exit_tasks_rcu_start(void) { } |
878 | void exit_tasks_rcu_finish(void) { exit_tasks_rcu_finish_trace(current); } | |
e21408ce | 879 | #endif /* #else #ifdef CONFIG_TASKS_RCU */ |
c84aad76 PM |
880 | |
881 | #ifdef CONFIG_TASKS_RUDE_RCU | |
882 | ||
883 | //////////////////////////////////////////////////////////////////////// | |
884 | // | |
885 | // "Rude" variant of Tasks RCU, inspired by Steve Rostedt's trick of | |
886 | // passing an empty function to schedule_on_each_cpu(). This approach | |
e4be1f44 PM |
887 | // provides an asynchronous call_rcu_tasks_rude() API and batching of |
888 | // concurrent calls to the synchronous synchronize_rcu_tasks_rude() API. | |
9fc98e31 PM |
889 | // This invokes schedule_on_each_cpu() in order to send IPIs far and wide |
890 | // and induces otherwise unnecessary context switches on all online CPUs, | |
891 | // whether idle or not. | |
892 | // | |
893 | // Callback handling is provided by the rcu_tasks_kthread() function. | |
894 | // | |
895 | // Ordering is provided by the scheduler's context-switch code. | |
c84aad76 PM |
896 | |
897 | // Empty function to allow workqueues to force a context switch. | |
898 | static void rcu_tasks_be_rude(struct work_struct *work) | |
899 | { | |
900 | } | |
901 | ||
902 | // Wait for one rude RCU-tasks grace period. | |
903 | static void rcu_tasks_rude_wait_gp(struct rcu_tasks *rtp) | |
904 | { | |
238dbce3 | 905 | rtp->n_ipis += cpumask_weight(cpu_online_mask); |
c84aad76 PM |
906 | schedule_on_each_cpu(rcu_tasks_be_rude); |
907 | } | |
908 | ||
909 | void call_rcu_tasks_rude(struct rcu_head *rhp, rcu_callback_t func); | |
c97d12a6 PM |
910 | DEFINE_RCU_TASKS(rcu_tasks_rude, rcu_tasks_rude_wait_gp, call_rcu_tasks_rude, |
911 | "RCU Tasks Rude"); | |
c84aad76 PM |
912 | |
913 | /** | |
914 | * call_rcu_tasks_rude() - Queue a callback rude task-based grace period | |
915 | * @rhp: structure to be used for queueing the RCU updates. | |
916 | * @func: actual callback function to be invoked after the grace period | |
917 | * | |
918 | * The callback function will be invoked some time after a full grace | |
919 | * period elapses, in other words after all currently executing RCU | |
920 | * read-side critical sections have completed. call_rcu_tasks_rude() | |
921 | * assumes that the read-side critical sections end at context switch, | |
8af9e2c7 | 922 | * cond_resched_tasks_rcu_qs(), or transition to usermode execution (as |
a6517e9c NU |
923 | * usermode execution is schedulable). As such, there are no read-side |
924 | * primitives analogous to rcu_read_lock() and rcu_read_unlock() because | |
925 | * this primitive is intended to determine that all tasks have passed | |
926 | * through a safe state, not so much for data-structure synchronization. | |
c84aad76 PM |
927 | * |
928 | * See the description of call_rcu() for more detailed information on | |
929 | * memory ordering guarantees. | |
930 | */ | |
931 | void call_rcu_tasks_rude(struct rcu_head *rhp, rcu_callback_t func) | |
932 | { | |
933 | call_rcu_tasks_generic(rhp, func, &rcu_tasks_rude); | |
934 | } | |
935 | EXPORT_SYMBOL_GPL(call_rcu_tasks_rude); | |
936 | ||
937 | /** | |
938 | * synchronize_rcu_tasks_rude - wait for a rude rcu-tasks grace period | |
939 | * | |
940 | * Control will return to the caller some time after a rude rcu-tasks | |
941 | * grace period has elapsed, in other words after all currently | |
942 | * executing rcu-tasks read-side critical sections have elapsed. These | |
943 | * read-side critical sections are delimited by calls to schedule(), | |
a6517e9c NU |
944 | * cond_resched_tasks_rcu_qs(), userspace execution (which is a schedulable |
945 | * context), and (in theory, anyway) cond_resched(). | |
c84aad76 PM |
946 | * |
947 | * This is a very specialized primitive, intended only for a few uses in | |
948 | * tracing and other situations requiring manipulation of function preambles | |
949 | * and profiling hooks. The synchronize_rcu_tasks_rude() function is not | |
950 | * (yet) intended for heavy use from multiple CPUs. | |
951 | * | |
952 | * See the description of synchronize_rcu() for more detailed information | |
953 | * on memory ordering guarantees. | |
954 | */ | |
955 | void synchronize_rcu_tasks_rude(void) | |
956 | { | |
957 | synchronize_rcu_tasks_generic(&rcu_tasks_rude); | |
958 | } | |
959 | EXPORT_SYMBOL_GPL(synchronize_rcu_tasks_rude); | |
960 | ||
961 | /** | |
962 | * rcu_barrier_tasks_rude - Wait for in-flight call_rcu_tasks_rude() callbacks. | |
963 | * | |
964 | * Although the current implementation is guaranteed to wait, it is not | |
965 | * obligated to, for example, if there are no pending callbacks. | |
966 | */ | |
967 | void rcu_barrier_tasks_rude(void) | |
968 | { | |
ce9b1c66 | 969 | rcu_barrier_tasks_generic(&rcu_tasks_rude); |
c84aad76 PM |
970 | } |
971 | EXPORT_SYMBOL_GPL(rcu_barrier_tasks_rude); | |
972 | ||
973 | static int __init rcu_spawn_tasks_rude_kthread(void) | |
974 | { | |
cafafd67 | 975 | cblist_init_generic(&rcu_tasks_rude); |
4fe192df | 976 | rcu_tasks_rude.gp_sleep = HZ / 10; |
c84aad76 PM |
977 | rcu_spawn_tasks_kthread_generic(&rcu_tasks_rude); |
978 | return 0; | |
979 | } | |
c84aad76 | 980 | |
27c0f144 PM |
981 | #if !defined(CONFIG_TINY_RCU) |
982 | void show_rcu_tasks_rude_gp_kthread(void) | |
e21408ce PM |
983 | { |
984 | show_rcu_tasks_generic_gp_kthread(&rcu_tasks_rude, ""); | |
985 | } | |
27c0f144 PM |
986 | EXPORT_SYMBOL_GPL(show_rcu_tasks_rude_gp_kthread); |
987 | #endif // !defined(CONFIG_TINY_RCU) | |
988 | #endif /* #ifdef CONFIG_TASKS_RUDE_RCU */ | |
d5f177d3 PM |
989 | |
990 | //////////////////////////////////////////////////////////////////////// | |
991 | // | |
992 | // Tracing variant of Tasks RCU. This variant is designed to be used | |
993 | // to protect tracing hooks, including those of BPF. This variant | |
994 | // therefore: | |
995 | // | |
996 | // 1. Has explicit read-side markers to allow finite grace periods | |
997 | // in the face of in-kernel loops for PREEMPT=n builds. | |
998 | // | |
999 | // 2. Protects code in the idle loop, exception entry/exit, and | |
1000 | // CPU-hotplug code paths, similar to the capabilities of SRCU. | |
1001 | // | |
c4f113ac | 1002 | // 3. Avoids expensive read-side instructions, having overhead similar |
d5f177d3 PM |
1003 | // to that of Preemptible RCU. |
1004 | // | |
1005 | // There are of course downsides. The grace-period code can send IPIs to | |
1006 | // CPUs, even when those CPUs are in the idle loop or in nohz_full userspace. | |
1007 | // It is necessary to scan the full tasklist, much as for Tasks RCU. There | |
1008 | // is a single callback queue guarded by a single lock, again, much as for | |
1009 | // Tasks RCU. If needed, these downsides can be at least partially remedied. | |
1010 | // | |
1011 | // Perhaps most important, this variant of RCU does not affect the vanilla | |
1012 | // flavors, rcu_preempt and rcu_sched. The fact that RCU Tasks Trace | |
1013 | // readers can operate from idle, offline, and exception entry/exit in no | |
1014 | // way allows rcu_preempt and rcu_sched readers to also do so. | |
a434dd10 PM |
1015 | // |
1016 | // The implementation uses rcu_tasks_wait_gp(), which relies on function | |
1017 | // pointers in the rcu_tasks structure. The rcu_spawn_tasks_trace_kthread() | |
1018 | // function sets these function pointers up so that rcu_tasks_wait_gp() | |
1019 | // invokes these functions in this order: | |
1020 | // | |
1021 | // rcu_tasks_trace_pregp_step(): | |
1022 | // Initialize the count of readers and block CPU-hotplug operations. | |
1023 | // rcu_tasks_trace_pertask(), invoked on every non-idle task: | |
1024 | // Initialize per-task state and attempt to identify an immediate | |
1025 | // quiescent state for that task, or, failing that, attempt to | |
1026 | // set that task's .need_qs flag so that task's next outermost | |
1027 | // rcu_read_unlock_trace() will report the quiescent state (in which | |
1028 | // case the count of readers is incremented). If both attempts fail, | |
45f4b4a2 PM |
1029 | // the task is added to a "holdout" list. Note that IPIs are used |
1030 | // to invoke trc_read_check_handler() in the context of running tasks | |
1031 | // in order to avoid ordering overhead on common-case shared-variable | |
1032 | // accessses. | |
a434dd10 PM |
1033 | // rcu_tasks_trace_postscan(): |
1034 | // Initialize state and attempt to identify an immediate quiescent | |
1035 | // state as above (but only for idle tasks), unblock CPU-hotplug | |
1036 | // operations, and wait for an RCU grace period to avoid races with | |
1037 | // tasks that are in the process of exiting. | |
1038 | // check_all_holdout_tasks_trace(), repeatedly until holdout list is empty: | |
1039 | // Scans the holdout list, attempting to identify a quiescent state | |
1040 | // for each task on the list. If there is a quiescent state, the | |
1041 | // corresponding task is removed from the holdout list. | |
1042 | // rcu_tasks_trace_postgp(): | |
1043 | // Wait for the count of readers do drop to zero, reporting any stalls. | |
1044 | // Also execute full memory barriers to maintain ordering with code | |
1045 | // executing after the grace period. | |
1046 | // | |
1047 | // The exit_tasks_rcu_finish_trace() synchronizes with exiting tasks. | |
1048 | // | |
1049 | // Pre-grace-period update-side code is ordered before the grace | |
1050 | // period via the ->cbs_lock and barriers in rcu_tasks_kthread(). | |
1051 | // Pre-grace-period read-side code is ordered before the grace period by | |
1052 | // atomic_dec_and_test() of the count of readers (for IPIed readers) and by | |
1053 | // scheduler context-switch ordering (for locked-down non-running readers). | |
d5f177d3 PM |
1054 | |
1055 | // The lockdep state must be outside of #ifdef to be useful. | |
1056 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
1057 | static struct lock_class_key rcu_lock_trace_key; | |
1058 | struct lockdep_map rcu_trace_lock_map = | |
1059 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_trace", &rcu_lock_trace_key); | |
1060 | EXPORT_SYMBOL_GPL(rcu_trace_lock_map); | |
1061 | #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
1062 | ||
1063 | #ifdef CONFIG_TASKS_TRACE_RCU | |
1064 | ||
30d8aa51 PM |
1065 | static atomic_t trc_n_readers_need_end; // Number of waited-for readers. |
1066 | static DECLARE_WAIT_QUEUE_HEAD(trc_wait); // List of holdout tasks. | |
d5f177d3 PM |
1067 | |
1068 | // Record outstanding IPIs to each CPU. No point in sending two... | |
1069 | static DEFINE_PER_CPU(bool, trc_ipi_to_cpu); | |
1070 | ||
40471509 PM |
1071 | // The number of detections of task quiescent state relying on |
1072 | // heavyweight readers executing explicit memory barriers. | |
6731da9e PM |
1073 | static unsigned long n_heavy_reader_attempts; |
1074 | static unsigned long n_heavy_reader_updates; | |
1075 | static unsigned long n_heavy_reader_ofl_updates; | |
40471509 | 1076 | |
b0afa0f0 PM |
1077 | void call_rcu_tasks_trace(struct rcu_head *rhp, rcu_callback_t func); |
1078 | DEFINE_RCU_TASKS(rcu_tasks_trace, rcu_tasks_wait_gp, call_rcu_tasks_trace, | |
1079 | "RCU Tasks Trace"); | |
1080 | ||
b38f57c1 PM |
1081 | /* |
1082 | * This irq_work handler allows rcu_read_unlock_trace() to be invoked | |
1083 | * while the scheduler locks are held. | |
1084 | */ | |
1085 | static void rcu_read_unlock_iw(struct irq_work *iwp) | |
1086 | { | |
1087 | wake_up(&trc_wait); | |
1088 | } | |
1089 | static DEFINE_IRQ_WORK(rcu_tasks_trace_iw, rcu_read_unlock_iw); | |
1090 | ||
d5f177d3 | 1091 | /* If we are the last reader, wake up the grace-period kthread. */ |
a5c071cc | 1092 | void rcu_read_unlock_trace_special(struct task_struct *t) |
d5f177d3 | 1093 | { |
f8ab3fad | 1094 | int nq = READ_ONCE(t->trc_reader_special.b.need_qs); |
276c4104 | 1095 | |
9ae58d7b PM |
1096 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB) && |
1097 | t->trc_reader_special.b.need_mb) | |
276c4104 PM |
1098 | smp_mb(); // Pairs with update-side barriers. |
1099 | // Update .need_qs before ->trc_reader_nesting for irq/NMI handlers. | |
1100 | if (nq) | |
1101 | WRITE_ONCE(t->trc_reader_special.b.need_qs, false); | |
a5c071cc | 1102 | WRITE_ONCE(t->trc_reader_nesting, 0); |
276c4104 | 1103 | if (nq && atomic_dec_and_test(&trc_n_readers_need_end)) |
b38f57c1 | 1104 | irq_work_queue(&rcu_tasks_trace_iw); |
d5f177d3 PM |
1105 | } |
1106 | EXPORT_SYMBOL_GPL(rcu_read_unlock_trace_special); | |
1107 | ||
1108 | /* Add a task to the holdout list, if it is not already on the list. */ | |
1109 | static void trc_add_holdout(struct task_struct *t, struct list_head *bhp) | |
1110 | { | |
1111 | if (list_empty(&t->trc_holdout_list)) { | |
1112 | get_task_struct(t); | |
1113 | list_add(&t->trc_holdout_list, bhp); | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | /* Remove a task from the holdout list, if it is in fact present. */ | |
1118 | static void trc_del_holdout(struct task_struct *t) | |
1119 | { | |
1120 | if (!list_empty(&t->trc_holdout_list)) { | |
1121 | list_del_init(&t->trc_holdout_list); | |
1122 | put_task_struct(t); | |
1123 | } | |
1124 | } | |
1125 | ||
1126 | /* IPI handler to check task state. */ | |
1127 | static void trc_read_check_handler(void *t_in) | |
1128 | { | |
1129 | struct task_struct *t = current; | |
1130 | struct task_struct *texp = t_in; | |
1131 | ||
1132 | // If the task is no longer running on this CPU, leave. | |
1133 | if (unlikely(texp != t)) { | |
d5f177d3 PM |
1134 | goto reset_ipi; // Already on holdout list, so will check later. |
1135 | } | |
1136 | ||
1137 | // If the task is not in a read-side critical section, and | |
1138 | // if this is the last reader, awaken the grace-period kthread. | |
bdb0cca0 | 1139 | if (likely(!READ_ONCE(t->trc_reader_nesting))) { |
d5f177d3 PM |
1140 | WRITE_ONCE(t->trc_reader_checked, true); |
1141 | goto reset_ipi; | |
1142 | } | |
ba3a86e4 | 1143 | // If we are racing with an rcu_read_unlock_trace(), try again later. |
96017bf9 | 1144 | if (unlikely(READ_ONCE(t->trc_reader_nesting) < 0)) |
ba3a86e4 | 1145 | goto reset_ipi; |
d5f177d3 PM |
1146 | WRITE_ONCE(t->trc_reader_checked, true); |
1147 | ||
1148 | // Get here if the task is in a read-side critical section. Set | |
1149 | // its state so that it will awaken the grace-period kthread upon | |
1150 | // exit from that critical section. | |
96017bf9 | 1151 | atomic_inc(&trc_n_readers_need_end); // One more to wait on. |
f8ab3fad | 1152 | WARN_ON_ONCE(READ_ONCE(t->trc_reader_special.b.need_qs)); |
276c4104 | 1153 | WRITE_ONCE(t->trc_reader_special.b.need_qs, true); |
d5f177d3 PM |
1154 | |
1155 | reset_ipi: | |
1156 | // Allow future IPIs to be sent on CPU and for task. | |
1157 | // Also order this IPI handler against any later manipulations of | |
1158 | // the intended task. | |
8211e922 | 1159 | smp_store_release(per_cpu_ptr(&trc_ipi_to_cpu, smp_processor_id()), false); // ^^^ |
d5f177d3 PM |
1160 | smp_store_release(&texp->trc_ipi_to_cpu, -1); // ^^^ |
1161 | } | |
1162 | ||
1163 | /* Callback function for scheduler to check locked-down task. */ | |
9b3c4ab3 | 1164 | static int trc_inspect_reader(struct task_struct *t, void *arg) |
d5f177d3 | 1165 | { |
7d0c9c50 | 1166 | int cpu = task_cpu(t); |
18f08e75 | 1167 | int nesting; |
7e3b70e0 | 1168 | bool ofl = cpu_is_offline(cpu); |
7d0c9c50 PM |
1169 | |
1170 | if (task_curr(t)) { | |
30d8aa51 | 1171 | WARN_ON_ONCE(ofl && !is_idle_task(t)); |
7e3b70e0 | 1172 | |
7d0c9c50 | 1173 | // If no chance of heavyweight readers, do it the hard way. |
7e3b70e0 | 1174 | if (!ofl && !IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) |
9b3c4ab3 | 1175 | return -EINVAL; |
7d0c9c50 PM |
1176 | |
1177 | // If heavyweight readers are enabled on the remote task, | |
1178 | // we can inspect its state despite its currently running. | |
1179 | // However, we cannot safely change its state. | |
40471509 | 1180 | n_heavy_reader_attempts++; |
7e3b70e0 PM |
1181 | if (!ofl && // Check for "running" idle tasks on offline CPUs. |
1182 | !rcu_dynticks_zero_in_eqs(cpu, &t->trc_reader_nesting)) | |
9b3c4ab3 | 1183 | return -EINVAL; // No quiescent state, do it the hard way. |
40471509 | 1184 | n_heavy_reader_updates++; |
edf3775f PM |
1185 | if (ofl) |
1186 | n_heavy_reader_ofl_updates++; | |
18f08e75 | 1187 | nesting = 0; |
7d0c9c50 | 1188 | } else { |
bdb0cca0 | 1189 | // The task is not running, so C-language access is safe. |
18f08e75 | 1190 | nesting = t->trc_reader_nesting; |
7d0c9c50 | 1191 | } |
d5f177d3 | 1192 | |
18f08e75 PM |
1193 | // If not exiting a read-side critical section, mark as checked |
1194 | // so that the grace-period kthread will remove it from the | |
1195 | // holdout list. | |
1196 | t->trc_reader_checked = nesting >= 0; | |
1197 | if (nesting <= 0) | |
6fedc280 | 1198 | return nesting ? -EINVAL : 0; // If in QS, done, otherwise try again later. |
7d0c9c50 PM |
1199 | |
1200 | // The task is in a read-side critical section, so set up its | |
1201 | // state so that it will awaken the grace-period kthread upon exit | |
1202 | // from that critical section. | |
1203 | atomic_inc(&trc_n_readers_need_end); // One more to wait on. | |
f8ab3fad | 1204 | WARN_ON_ONCE(READ_ONCE(t->trc_reader_special.b.need_qs)); |
7d0c9c50 | 1205 | WRITE_ONCE(t->trc_reader_special.b.need_qs, true); |
9b3c4ab3 | 1206 | return 0; |
d5f177d3 PM |
1207 | } |
1208 | ||
1209 | /* Attempt to extract the state for the specified task. */ | |
1210 | static void trc_wait_for_one_reader(struct task_struct *t, | |
1211 | struct list_head *bhp) | |
1212 | { | |
1213 | int cpu; | |
1214 | ||
1215 | // If a previous IPI is still in flight, let it complete. | |
1216 | if (smp_load_acquire(&t->trc_ipi_to_cpu) != -1) // Order IPI | |
1217 | return; | |
1218 | ||
1219 | // The current task had better be in a quiescent state. | |
1220 | if (t == current) { | |
1221 | t->trc_reader_checked = true; | |
bdb0cca0 | 1222 | WARN_ON_ONCE(READ_ONCE(t->trc_reader_nesting)); |
d5f177d3 PM |
1223 | return; |
1224 | } | |
1225 | ||
1226 | // Attempt to nail down the task for inspection. | |
1227 | get_task_struct(t); | |
9b3c4ab3 | 1228 | if (!task_call_func(t, trc_inspect_reader, NULL)) { |
d5f177d3 PM |
1229 | put_task_struct(t); |
1230 | return; | |
1231 | } | |
1232 | put_task_struct(t); | |
1233 | ||
45f4b4a2 PM |
1234 | // If this task is not yet on the holdout list, then we are in |
1235 | // an RCU read-side critical section. Otherwise, the invocation of | |
d0a85858 | 1236 | // trc_add_holdout() that added it to the list did the necessary |
45f4b4a2 PM |
1237 | // get_task_struct(). Either way, the task cannot be freed out |
1238 | // from under this code. | |
1239 | ||
d5f177d3 PM |
1240 | // If currently running, send an IPI, either way, add to list. |
1241 | trc_add_holdout(t, bhp); | |
574de876 PM |
1242 | if (task_curr(t) && |
1243 | time_after(jiffies + 1, rcu_tasks_trace.gp_start + rcu_task_ipi_delay)) { | |
d5f177d3 PM |
1244 | // The task is currently running, so try IPIing it. |
1245 | cpu = task_cpu(t); | |
1246 | ||
1247 | // If there is already an IPI outstanding, let it happen. | |
1248 | if (per_cpu(trc_ipi_to_cpu, cpu) || t->trc_ipi_to_cpu >= 0) | |
1249 | return; | |
1250 | ||
d5f177d3 PM |
1251 | per_cpu(trc_ipi_to_cpu, cpu) = true; |
1252 | t->trc_ipi_to_cpu = cpu; | |
238dbce3 | 1253 | rcu_tasks_trace.n_ipis++; |
96017bf9 | 1254 | if (smp_call_function_single(cpu, trc_read_check_handler, t, 0)) { |
d5f177d3 PM |
1255 | // Just in case there is some other reason for |
1256 | // failure than the target CPU being offline. | |
46aa886c NU |
1257 | WARN_ONCE(1, "%s(): smp_call_function_single() failed for CPU: %d\n", |
1258 | __func__, cpu); | |
7e0669c3 | 1259 | rcu_tasks_trace.n_ipis_fails++; |
d5f177d3 | 1260 | per_cpu(trc_ipi_to_cpu, cpu) = false; |
46aa886c | 1261 | t->trc_ipi_to_cpu = -1; |
d5f177d3 PM |
1262 | } |
1263 | } | |
1264 | } | |
1265 | ||
1266 | /* Initialize for a new RCU-tasks-trace grace period. */ | |
1267 | static void rcu_tasks_trace_pregp_step(void) | |
1268 | { | |
1269 | int cpu; | |
1270 | ||
d5f177d3 PM |
1271 | // Allow for fast-acting IPIs. |
1272 | atomic_set(&trc_n_readers_need_end, 1); | |
1273 | ||
1274 | // There shouldn't be any old IPIs, but... | |
1275 | for_each_possible_cpu(cpu) | |
1276 | WARN_ON_ONCE(per_cpu(trc_ipi_to_cpu, cpu)); | |
81b4a7bc PM |
1277 | |
1278 | // Disable CPU hotplug across the tasklist scan. | |
1279 | // This also waits for all readers in CPU-hotplug code paths. | |
1280 | cpus_read_lock(); | |
d5f177d3 PM |
1281 | } |
1282 | ||
1283 | /* Do first-round processing for the specified task. */ | |
1284 | static void rcu_tasks_trace_pertask(struct task_struct *t, | |
1285 | struct list_head *hop) | |
1286 | { | |
1b04fa99 URS |
1287 | // During early boot when there is only the one boot CPU, there |
1288 | // is no idle task for the other CPUs. Just return. | |
1289 | if (unlikely(t == NULL)) | |
1290 | return; | |
1291 | ||
276c4104 | 1292 | WRITE_ONCE(t->trc_reader_special.b.need_qs, false); |
43766c3e | 1293 | WRITE_ONCE(t->trc_reader_checked, false); |
d5f177d3 PM |
1294 | t->trc_ipi_to_cpu = -1; |
1295 | trc_wait_for_one_reader(t, hop); | |
1296 | } | |
1297 | ||
9796e1ae PM |
1298 | /* |
1299 | * Do intermediate processing between task and holdout scans and | |
1300 | * pick up the idle tasks. | |
1301 | */ | |
1302 | static void rcu_tasks_trace_postscan(struct list_head *hop) | |
d5f177d3 | 1303 | { |
9796e1ae PM |
1304 | int cpu; |
1305 | ||
1306 | for_each_possible_cpu(cpu) | |
1307 | rcu_tasks_trace_pertask(idle_task(cpu), hop); | |
1308 | ||
81b4a7bc PM |
1309 | // Re-enable CPU hotplug now that the tasklist scan has completed. |
1310 | cpus_read_unlock(); | |
1311 | ||
d5f177d3 PM |
1312 | // Wait for late-stage exiting tasks to finish exiting. |
1313 | // These might have passed the call to exit_tasks_rcu_finish(). | |
1314 | synchronize_rcu(); | |
1315 | // Any tasks that exit after this point will set ->trc_reader_checked. | |
1316 | } | |
1317 | ||
65b629e7 NU |
1318 | /* Communicate task state back to the RCU tasks trace stall warning request. */ |
1319 | struct trc_stall_chk_rdr { | |
1320 | int nesting; | |
1321 | int ipi_to_cpu; | |
1322 | u8 needqs; | |
1323 | }; | |
1324 | ||
1325 | static int trc_check_slow_task(struct task_struct *t, void *arg) | |
1326 | { | |
1327 | struct trc_stall_chk_rdr *trc_rdrp = arg; | |
1328 | ||
1329 | if (task_curr(t)) | |
1330 | return false; // It is running, so decline to inspect it. | |
1331 | trc_rdrp->nesting = READ_ONCE(t->trc_reader_nesting); | |
1332 | trc_rdrp->ipi_to_cpu = READ_ONCE(t->trc_ipi_to_cpu); | |
1333 | trc_rdrp->needqs = READ_ONCE(t->trc_reader_special.b.need_qs); | |
1334 | return true; | |
1335 | } | |
1336 | ||
4593e772 PM |
1337 | /* Show the state of a task stalling the current RCU tasks trace GP. */ |
1338 | static void show_stalled_task_trace(struct task_struct *t, bool *firstreport) | |
1339 | { | |
1340 | int cpu; | |
65b629e7 NU |
1341 | struct trc_stall_chk_rdr trc_rdr; |
1342 | bool is_idle_tsk = is_idle_task(t); | |
4593e772 PM |
1343 | |
1344 | if (*firstreport) { | |
1345 | pr_err("INFO: rcu_tasks_trace detected stalls on tasks:\n"); | |
1346 | *firstreport = false; | |
1347 | } | |
4593e772 | 1348 | cpu = task_cpu(t); |
65b629e7 NU |
1349 | if (!task_call_func(t, trc_check_slow_task, &trc_rdr)) |
1350 | pr_alert("P%d: %c\n", | |
1351 | t->pid, | |
1352 | ".i"[is_idle_tsk]); | |
1353 | else | |
1354 | pr_alert("P%d: %c%c%c nesting: %d%c cpu: %d\n", | |
1355 | t->pid, | |
1356 | ".I"[trc_rdr.ipi_to_cpu >= 0], | |
1357 | ".i"[is_idle_tsk], | |
1358 | ".N"[cpu >= 0 && tick_nohz_full_cpu(cpu)], | |
1359 | trc_rdr.nesting, | |
1360 | " N"[!!trc_rdr.needqs], | |
1361 | cpu); | |
4593e772 PM |
1362 | sched_show_task(t); |
1363 | } | |
1364 | ||
1365 | /* List stalled IPIs for RCU tasks trace. */ | |
1366 | static void show_stalled_ipi_trace(void) | |
1367 | { | |
1368 | int cpu; | |
1369 | ||
1370 | for_each_possible_cpu(cpu) | |
1371 | if (per_cpu(trc_ipi_to_cpu, cpu)) | |
1372 | pr_alert("\tIPI outstanding to CPU %d\n", cpu); | |
1373 | } | |
1374 | ||
d5f177d3 PM |
1375 | /* Do one scan of the holdout list. */ |
1376 | static void check_all_holdout_tasks_trace(struct list_head *hop, | |
4593e772 | 1377 | bool needreport, bool *firstreport) |
d5f177d3 PM |
1378 | { |
1379 | struct task_struct *g, *t; | |
1380 | ||
81b4a7bc PM |
1381 | // Disable CPU hotplug across the holdout list scan. |
1382 | cpus_read_lock(); | |
1383 | ||
d5f177d3 PM |
1384 | list_for_each_entry_safe(t, g, hop, trc_holdout_list) { |
1385 | // If safe and needed, try to check the current task. | |
1386 | if (READ_ONCE(t->trc_ipi_to_cpu) == -1 && | |
1387 | !READ_ONCE(t->trc_reader_checked)) | |
1388 | trc_wait_for_one_reader(t, hop); | |
1389 | ||
1390 | // If check succeeded, remove this task from the list. | |
f5dbc594 PM |
1391 | if (smp_load_acquire(&t->trc_ipi_to_cpu) == -1 && |
1392 | READ_ONCE(t->trc_reader_checked)) | |
d5f177d3 | 1393 | trc_del_holdout(t); |
4593e772 PM |
1394 | else if (needreport) |
1395 | show_stalled_task_trace(t, firstreport); | |
1396 | } | |
81b4a7bc PM |
1397 | |
1398 | // Re-enable CPU hotplug now that the holdout list scan has completed. | |
1399 | cpus_read_unlock(); | |
1400 | ||
4593e772 | 1401 | if (needreport) { |
89401176 | 1402 | if (*firstreport) |
4593e772 PM |
1403 | pr_err("INFO: rcu_tasks_trace detected stalls? (Late IPI?)\n"); |
1404 | show_stalled_ipi_trace(); | |
d5f177d3 PM |
1405 | } |
1406 | } | |
1407 | ||
cbe0d8d9 PM |
1408 | static void rcu_tasks_trace_empty_fn(void *unused) |
1409 | { | |
1410 | } | |
1411 | ||
d5f177d3 | 1412 | /* Wait for grace period to complete and provide ordering. */ |
af051ca4 | 1413 | static void rcu_tasks_trace_postgp(struct rcu_tasks *rtp) |
d5f177d3 | 1414 | { |
cbe0d8d9 | 1415 | int cpu; |
4593e772 PM |
1416 | bool firstreport; |
1417 | struct task_struct *g, *t; | |
1418 | LIST_HEAD(holdouts); | |
1419 | long ret; | |
1420 | ||
cbe0d8d9 PM |
1421 | // Wait for any lingering IPI handlers to complete. Note that |
1422 | // if a CPU has gone offline or transitioned to userspace in the | |
1423 | // meantime, all IPI handlers should have been drained beforehand. | |
1424 | // Yes, this assumes that CPUs process IPIs in order. If that ever | |
1425 | // changes, there will need to be a recheck and/or timed wait. | |
1426 | for_each_online_cpu(cpu) | |
f5dbc594 | 1427 | if (WARN_ON_ONCE(smp_load_acquire(per_cpu_ptr(&trc_ipi_to_cpu, cpu)))) |
cbe0d8d9 PM |
1428 | smp_call_function_single(cpu, rcu_tasks_trace_empty_fn, NULL, 1); |
1429 | ||
d5f177d3 PM |
1430 | // Remove the safety count. |
1431 | smp_mb__before_atomic(); // Order vs. earlier atomics | |
1432 | atomic_dec(&trc_n_readers_need_end); | |
1433 | smp_mb__after_atomic(); // Order vs. later atomics | |
1434 | ||
1435 | // Wait for readers. | |
af051ca4 | 1436 | set_tasks_gp_state(rtp, RTGS_WAIT_READERS); |
4593e772 PM |
1437 | for (;;) { |
1438 | ret = wait_event_idle_exclusive_timeout( | |
1439 | trc_wait, | |
1440 | atomic_read(&trc_n_readers_need_end) == 0, | |
1441 | READ_ONCE(rcu_task_stall_timeout)); | |
1442 | if (ret) | |
1443 | break; // Count reached zero. | |
af051ca4 | 1444 | // Stall warning time, so make a list of the offenders. |
f747c7e1 | 1445 | rcu_read_lock(); |
4593e772 | 1446 | for_each_process_thread(g, t) |
276c4104 | 1447 | if (READ_ONCE(t->trc_reader_special.b.need_qs)) |
4593e772 | 1448 | trc_add_holdout(t, &holdouts); |
f747c7e1 | 1449 | rcu_read_unlock(); |
4593e772 | 1450 | firstreport = true; |
592031cc PM |
1451 | list_for_each_entry_safe(t, g, &holdouts, trc_holdout_list) { |
1452 | if (READ_ONCE(t->trc_reader_special.b.need_qs)) | |
4593e772 | 1453 | show_stalled_task_trace(t, &firstreport); |
592031cc PM |
1454 | trc_del_holdout(t); // Release task_struct reference. |
1455 | } | |
4593e772 PM |
1456 | if (firstreport) |
1457 | pr_err("INFO: rcu_tasks_trace detected stalls? (Counter/taskslist mismatch?)\n"); | |
1458 | show_stalled_ipi_trace(); | |
1459 | pr_err("\t%d holdouts\n", atomic_read(&trc_n_readers_need_end)); | |
1460 | } | |
d5f177d3 | 1461 | smp_mb(); // Caller's code must be ordered after wakeup. |
43766c3e | 1462 | // Pairs with pretty much every ordering primitive. |
d5f177d3 PM |
1463 | } |
1464 | ||
1465 | /* Report any needed quiescent state for this exiting task. */ | |
25246fc8 | 1466 | static void exit_tasks_rcu_finish_trace(struct task_struct *t) |
d5f177d3 PM |
1467 | { |
1468 | WRITE_ONCE(t->trc_reader_checked, true); | |
bdb0cca0 | 1469 | WARN_ON_ONCE(READ_ONCE(t->trc_reader_nesting)); |
d5f177d3 | 1470 | WRITE_ONCE(t->trc_reader_nesting, 0); |
276c4104 | 1471 | if (WARN_ON_ONCE(READ_ONCE(t->trc_reader_special.b.need_qs))) |
a5c071cc | 1472 | rcu_read_unlock_trace_special(t); |
d5f177d3 PM |
1473 | } |
1474 | ||
d5f177d3 PM |
1475 | /** |
1476 | * call_rcu_tasks_trace() - Queue a callback trace task-based grace period | |
1477 | * @rhp: structure to be used for queueing the RCU updates. | |
1478 | * @func: actual callback function to be invoked after the grace period | |
1479 | * | |
ed42c380 NU |
1480 | * The callback function will be invoked some time after a trace rcu-tasks |
1481 | * grace period elapses, in other words after all currently executing | |
1482 | * trace rcu-tasks read-side critical sections have completed. These | |
1483 | * read-side critical sections are delimited by calls to rcu_read_lock_trace() | |
1484 | * and rcu_read_unlock_trace(). | |
d5f177d3 PM |
1485 | * |
1486 | * See the description of call_rcu() for more detailed information on | |
1487 | * memory ordering guarantees. | |
1488 | */ | |
1489 | void call_rcu_tasks_trace(struct rcu_head *rhp, rcu_callback_t func) | |
1490 | { | |
1491 | call_rcu_tasks_generic(rhp, func, &rcu_tasks_trace); | |
1492 | } | |
1493 | EXPORT_SYMBOL_GPL(call_rcu_tasks_trace); | |
1494 | ||
1495 | /** | |
1496 | * synchronize_rcu_tasks_trace - wait for a trace rcu-tasks grace period | |
1497 | * | |
1498 | * Control will return to the caller some time after a trace rcu-tasks | |
c7dcf810 | 1499 | * grace period has elapsed, in other words after all currently executing |
ed42c380 | 1500 | * trace rcu-tasks read-side critical sections have elapsed. These read-side |
c7dcf810 PM |
1501 | * critical sections are delimited by calls to rcu_read_lock_trace() |
1502 | * and rcu_read_unlock_trace(). | |
d5f177d3 PM |
1503 | * |
1504 | * This is a very specialized primitive, intended only for a few uses in | |
1505 | * tracing and other situations requiring manipulation of function preambles | |
1506 | * and profiling hooks. The synchronize_rcu_tasks_trace() function is not | |
1507 | * (yet) intended for heavy use from multiple CPUs. | |
1508 | * | |
1509 | * See the description of synchronize_rcu() for more detailed information | |
1510 | * on memory ordering guarantees. | |
1511 | */ | |
1512 | void synchronize_rcu_tasks_trace(void) | |
1513 | { | |
1514 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_trace_lock_map), "Illegal synchronize_rcu_tasks_trace() in RCU Tasks Trace read-side critical section"); | |
1515 | synchronize_rcu_tasks_generic(&rcu_tasks_trace); | |
1516 | } | |
1517 | EXPORT_SYMBOL_GPL(synchronize_rcu_tasks_trace); | |
1518 | ||
1519 | /** | |
1520 | * rcu_barrier_tasks_trace - Wait for in-flight call_rcu_tasks_trace() callbacks. | |
1521 | * | |
1522 | * Although the current implementation is guaranteed to wait, it is not | |
1523 | * obligated to, for example, if there are no pending callbacks. | |
1524 | */ | |
1525 | void rcu_barrier_tasks_trace(void) | |
1526 | { | |
ce9b1c66 | 1527 | rcu_barrier_tasks_generic(&rcu_tasks_trace); |
d5f177d3 PM |
1528 | } |
1529 | EXPORT_SYMBOL_GPL(rcu_barrier_tasks_trace); | |
1530 | ||
1531 | static int __init rcu_spawn_tasks_trace_kthread(void) | |
1532 | { | |
cafafd67 | 1533 | cblist_init_generic(&rcu_tasks_trace); |
2393a613 | 1534 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) { |
4fe192df | 1535 | rcu_tasks_trace.gp_sleep = HZ / 10; |
75dc2da5 | 1536 | rcu_tasks_trace.init_fract = HZ / 10; |
2393a613 | 1537 | } else { |
4fe192df PM |
1538 | rcu_tasks_trace.gp_sleep = HZ / 200; |
1539 | if (rcu_tasks_trace.gp_sleep <= 0) | |
1540 | rcu_tasks_trace.gp_sleep = 1; | |
75dc2da5 | 1541 | rcu_tasks_trace.init_fract = HZ / 200; |
2393a613 PM |
1542 | if (rcu_tasks_trace.init_fract <= 0) |
1543 | rcu_tasks_trace.init_fract = 1; | |
1544 | } | |
d5f177d3 PM |
1545 | rcu_tasks_trace.pregp_func = rcu_tasks_trace_pregp_step; |
1546 | rcu_tasks_trace.pertask_func = rcu_tasks_trace_pertask; | |
1547 | rcu_tasks_trace.postscan_func = rcu_tasks_trace_postscan; | |
1548 | rcu_tasks_trace.holdouts_func = check_all_holdout_tasks_trace; | |
1549 | rcu_tasks_trace.postgp_func = rcu_tasks_trace_postgp; | |
1550 | rcu_spawn_tasks_kthread_generic(&rcu_tasks_trace); | |
1551 | return 0; | |
1552 | } | |
d5f177d3 | 1553 | |
27c0f144 PM |
1554 | #if !defined(CONFIG_TINY_RCU) |
1555 | void show_rcu_tasks_trace_gp_kthread(void) | |
e21408ce | 1556 | { |
40471509 | 1557 | char buf[64]; |
e21408ce | 1558 | |
edf3775f PM |
1559 | sprintf(buf, "N%d h:%lu/%lu/%lu", atomic_read(&trc_n_readers_need_end), |
1560 | data_race(n_heavy_reader_ofl_updates), | |
40471509 PM |
1561 | data_race(n_heavy_reader_updates), |
1562 | data_race(n_heavy_reader_attempts)); | |
e21408ce PM |
1563 | show_rcu_tasks_generic_gp_kthread(&rcu_tasks_trace, buf); |
1564 | } | |
27c0f144 PM |
1565 | EXPORT_SYMBOL_GPL(show_rcu_tasks_trace_gp_kthread); |
1566 | #endif // !defined(CONFIG_TINY_RCU) | |
e21408ce | 1567 | |
d5f177d3 | 1568 | #else /* #ifdef CONFIG_TASKS_TRACE_RCU */ |
25246fc8 | 1569 | static void exit_tasks_rcu_finish_trace(struct task_struct *t) { } |
d5f177d3 | 1570 | #endif /* #else #ifdef CONFIG_TASKS_TRACE_RCU */ |
8fd8ca38 | 1571 | |
8344496e | 1572 | #ifndef CONFIG_TINY_RCU |
e21408ce PM |
1573 | void show_rcu_tasks_gp_kthreads(void) |
1574 | { | |
1575 | show_rcu_tasks_classic_gp_kthread(); | |
1576 | show_rcu_tasks_rude_gp_kthread(); | |
1577 | show_rcu_tasks_trace_gp_kthread(); | |
1578 | } | |
8344496e | 1579 | #endif /* #ifndef CONFIG_TINY_RCU */ |
e21408ce | 1580 | |
bfba7ed0 URS |
1581 | #ifdef CONFIG_PROVE_RCU |
1582 | struct rcu_tasks_test_desc { | |
1583 | struct rcu_head rh; | |
1584 | const char *name; | |
1585 | bool notrun; | |
1586 | }; | |
1587 | ||
1588 | static struct rcu_tasks_test_desc tests[] = { | |
1589 | { | |
1590 | .name = "call_rcu_tasks()", | |
1591 | /* If not defined, the test is skipped. */ | |
1592 | .notrun = !IS_ENABLED(CONFIG_TASKS_RCU), | |
1593 | }, | |
1594 | { | |
1595 | .name = "call_rcu_tasks_rude()", | |
1596 | /* If not defined, the test is skipped. */ | |
1597 | .notrun = !IS_ENABLED(CONFIG_TASKS_RUDE_RCU), | |
1598 | }, | |
1599 | { | |
1600 | .name = "call_rcu_tasks_trace()", | |
1601 | /* If not defined, the test is skipped. */ | |
1602 | .notrun = !IS_ENABLED(CONFIG_TASKS_TRACE_RCU) | |
1603 | } | |
1604 | }; | |
1605 | ||
1606 | static void test_rcu_tasks_callback(struct rcu_head *rhp) | |
1607 | { | |
1608 | struct rcu_tasks_test_desc *rttd = | |
1609 | container_of(rhp, struct rcu_tasks_test_desc, rh); | |
1610 | ||
1611 | pr_info("Callback from %s invoked.\n", rttd->name); | |
1612 | ||
1613 | rttd->notrun = true; | |
1614 | } | |
1615 | ||
1616 | static void rcu_tasks_initiate_self_tests(void) | |
1617 | { | |
1618 | pr_info("Running RCU-tasks wait API self tests\n"); | |
1619 | #ifdef CONFIG_TASKS_RCU | |
1620 | synchronize_rcu_tasks(); | |
1621 | call_rcu_tasks(&tests[0].rh, test_rcu_tasks_callback); | |
1622 | #endif | |
1623 | ||
1624 | #ifdef CONFIG_TASKS_RUDE_RCU | |
1625 | synchronize_rcu_tasks_rude(); | |
1626 | call_rcu_tasks_rude(&tests[1].rh, test_rcu_tasks_callback); | |
1627 | #endif | |
1628 | ||
1629 | #ifdef CONFIG_TASKS_TRACE_RCU | |
1630 | synchronize_rcu_tasks_trace(); | |
1631 | call_rcu_tasks_trace(&tests[2].rh, test_rcu_tasks_callback); | |
1632 | #endif | |
1633 | } | |
1634 | ||
1635 | static int rcu_tasks_verify_self_tests(void) | |
1636 | { | |
1637 | int ret = 0; | |
1638 | int i; | |
1639 | ||
1640 | for (i = 0; i < ARRAY_SIZE(tests); i++) { | |
1641 | if (!tests[i].notrun) { // still hanging. | |
1642 | pr_err("%s has been failed.\n", tests[i].name); | |
1643 | ret = -1; | |
1644 | } | |
1645 | } | |
1646 | ||
1647 | if (ret) | |
1648 | WARN_ON(1); | |
1649 | ||
1650 | return ret; | |
1651 | } | |
1652 | late_initcall(rcu_tasks_verify_self_tests); | |
1653 | #else /* #ifdef CONFIG_PROVE_RCU */ | |
1654 | static void rcu_tasks_initiate_self_tests(void) { } | |
1655 | #endif /* #else #ifdef CONFIG_PROVE_RCU */ | |
1656 | ||
1b04fa99 URS |
1657 | void __init rcu_init_tasks_generic(void) |
1658 | { | |
1659 | #ifdef CONFIG_TASKS_RCU | |
1660 | rcu_spawn_tasks_kthread(); | |
1661 | #endif | |
1662 | ||
1663 | #ifdef CONFIG_TASKS_RUDE_RCU | |
1664 | rcu_spawn_tasks_rude_kthread(); | |
1665 | #endif | |
1666 | ||
1667 | #ifdef CONFIG_TASKS_TRACE_RCU | |
1668 | rcu_spawn_tasks_trace_kthread(); | |
1669 | #endif | |
bfba7ed0 URS |
1670 | |
1671 | // Run the self-tests. | |
1672 | rcu_tasks_initiate_self_tests(); | |
1b04fa99 URS |
1673 | } |
1674 | ||
8fd8ca38 PM |
1675 | #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */ |
1676 | static inline void rcu_tasks_bootup_oddness(void) {} | |
1677 | #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */ |