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 |
5873b8a9 PM |
9 | |
10 | //////////////////////////////////////////////////////////////////////// | |
11 | // | |
12 | // Generic data structures. | |
13 | ||
14 | struct rcu_tasks; | |
15 | typedef void (*rcu_tasks_gp_func_t)(struct rcu_tasks *rtp); | |
e4fe5dd6 PM |
16 | typedef void (*pregp_func_t)(void); |
17 | typedef void (*pertask_func_t)(struct task_struct *t, struct list_head *hop); | |
9796e1ae | 18 | typedef void (*postscan_func_t)(struct list_head *hop); |
e4fe5dd6 | 19 | typedef void (*holdouts_func_t)(struct list_head *hop, bool ndrpt, bool *frptp); |
af051ca4 | 20 | typedef void (*postgp_func_t)(struct rcu_tasks *rtp); |
eacd6f04 | 21 | |
07e10515 | 22 | /** |
cafafd67 | 23 | * struct rcu_tasks_percpu - Per-CPU component of definition for a Tasks-RCU-like mechanism. |
07e10515 PM |
24 | * @cbs_head: Head of callback list. |
25 | * @cbs_tail: Tail pointer for callback list. | |
cafafd67 PM |
26 | * @cbs_pcpu_lock: Lock protecting per-CPU callback list. |
27 | */ | |
28 | struct rcu_tasks_percpu { | |
29 | struct rcu_head *cbs_head; | |
30 | struct rcu_head **cbs_tail; | |
31 | raw_spinlock_t cbs_pcpu_lock; | |
32 | }; | |
33 | ||
34 | /** | |
35 | * struct rcu_tasks - Definition for a Tasks-RCU-like mechanism. | |
a616aec9 | 36 | * @cbs_wq: Wait queue allowing new callback to get kthread's attention. |
cafafd67 | 37 | * @cbs_gbl_lock: Lock protecting callback list. |
07e10515 | 38 | * @kthread_ptr: This flavor's grace-period/callback-invocation kthread. |
5873b8a9 | 39 | * @gp_func: This flavor's grace-period-wait function. |
af051ca4 | 40 | * @gp_state: Grace period's most recent state transition (debugging). |
4fe192df | 41 | * @gp_sleep: Per-grace-period sleep to prevent CPU-bound looping. |
2393a613 | 42 | * @init_fract: Initial backoff sleep interval. |
af051ca4 PM |
43 | * @gp_jiffies: Time of last @gp_state transition. |
44 | * @gp_start: Most recent grace-period start in jiffies. | |
238dbce3 PM |
45 | * @n_gps: Number of grace periods completed since boot. |
46 | * @n_ipis: Number of IPIs sent to encourage grace periods to end. | |
7e0669c3 | 47 | * @n_ipis_fails: Number of IPI-send failures. |
e4fe5dd6 PM |
48 | * @pregp_func: This flavor's pre-grace-period function (optional). |
49 | * @pertask_func: This flavor's per-task scan function (optional). | |
50 | * @postscan_func: This flavor's post-task scan function (optional). | |
85b86994 | 51 | * @holdouts_func: This flavor's holdout-list scan function (optional). |
e4fe5dd6 | 52 | * @postgp_func: This flavor's post-grace-period function (optional). |
5873b8a9 | 53 | * @call_func: This flavor's call_rcu()-equivalent function. |
cafafd67 | 54 | * @rtpcpu: This flavor's rcu_tasks_percpu structure. |
7a30871b | 55 | * @percpu_enqueue_shift: Shift down CPU ID this much when enqueuing callbacks. |
c97d12a6 PM |
56 | * @name: This flavor's textual name. |
57 | * @kname: This flavor's kthread name. | |
07e10515 PM |
58 | */ |
59 | struct rcu_tasks { | |
07e10515 | 60 | struct wait_queue_head cbs_wq; |
cafafd67 | 61 | raw_spinlock_t cbs_gbl_lock; |
af051ca4 | 62 | int gp_state; |
4fe192df | 63 | int gp_sleep; |
2393a613 | 64 | int init_fract; |
af051ca4 | 65 | unsigned long gp_jiffies; |
88092d0c | 66 | unsigned long gp_start; |
238dbce3 PM |
67 | unsigned long n_gps; |
68 | unsigned long n_ipis; | |
7e0669c3 | 69 | unsigned long n_ipis_fails; |
07e10515 | 70 | struct task_struct *kthread_ptr; |
5873b8a9 | 71 | rcu_tasks_gp_func_t gp_func; |
e4fe5dd6 PM |
72 | pregp_func_t pregp_func; |
73 | pertask_func_t pertask_func; | |
74 | postscan_func_t postscan_func; | |
75 | holdouts_func_t holdouts_func; | |
76 | postgp_func_t postgp_func; | |
5873b8a9 | 77 | call_rcu_func_t call_func; |
cafafd67 | 78 | struct rcu_tasks_percpu __percpu *rtpcpu; |
7a30871b | 79 | int percpu_enqueue_shift; |
c97d12a6 PM |
80 | char *name; |
81 | char *kname; | |
07e10515 PM |
82 | }; |
83 | ||
cafafd67 PM |
84 | #define DEFINE_RCU_TASKS(rt_name, gp, call, n) \ |
85 | static DEFINE_PER_CPU(struct rcu_tasks_percpu, rt_name ## __percpu) = { \ | |
86 | .cbs_pcpu_lock = __RAW_SPIN_LOCK_UNLOCKED(rt_name ## __percpu.cbs_pcpu_lock), \ | |
87 | }; \ | |
88 | static struct rcu_tasks rt_name = \ | |
89 | { \ | |
90 | .cbs_wq = __WAIT_QUEUE_HEAD_INITIALIZER(rt_name.cbs_wq), \ | |
91 | .cbs_gbl_lock = __RAW_SPIN_LOCK_UNLOCKED(rt_name.cbs_gbl_lock), \ | |
92 | .gp_func = gp, \ | |
93 | .call_func = call, \ | |
94 | .rtpcpu = &rt_name ## __percpu, \ | |
95 | .name = n, \ | |
7a30871b | 96 | .percpu_enqueue_shift = ilog2(CONFIG_NR_CPUS), \ |
cafafd67 | 97 | .kname = #rt_name, \ |
07e10515 PM |
98 | } |
99 | ||
eacd6f04 PM |
100 | /* Track exiting tasks in order to allow them to be waited for. */ |
101 | DEFINE_STATIC_SRCU(tasks_rcu_exit_srcu); | |
102 | ||
b0afa0f0 | 103 | /* Avoid IPIing CPUs early in the grace period. */ |
574de876 | 104 | #define RCU_TASK_IPI_DELAY (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB) ? HZ / 2 : 0) |
b0afa0f0 PM |
105 | static int rcu_task_ipi_delay __read_mostly = RCU_TASK_IPI_DELAY; |
106 | module_param(rcu_task_ipi_delay, int, 0644); | |
107 | ||
eacd6f04 PM |
108 | /* Control stall timeouts. Disable with <= 0, otherwise jiffies till stall. */ |
109 | #define RCU_TASK_STALL_TIMEOUT (HZ * 60 * 10) | |
110 | static int rcu_task_stall_timeout __read_mostly = RCU_TASK_STALL_TIMEOUT; | |
111 | module_param(rcu_task_stall_timeout, int, 0644); | |
112 | ||
af051ca4 PM |
113 | /* RCU tasks grace-period state for debugging. */ |
114 | #define RTGS_INIT 0 | |
115 | #define RTGS_WAIT_WAIT_CBS 1 | |
116 | #define RTGS_WAIT_GP 2 | |
117 | #define RTGS_PRE_WAIT_GP 3 | |
118 | #define RTGS_SCAN_TASKLIST 4 | |
119 | #define RTGS_POST_SCAN_TASKLIST 5 | |
120 | #define RTGS_WAIT_SCAN_HOLDOUTS 6 | |
121 | #define RTGS_SCAN_HOLDOUTS 7 | |
122 | #define RTGS_POST_GP 8 | |
123 | #define RTGS_WAIT_READERS 9 | |
124 | #define RTGS_INVOKE_CBS 10 | |
125 | #define RTGS_WAIT_CBS 11 | |
8344496e | 126 | #ifndef CONFIG_TINY_RCU |
af051ca4 PM |
127 | static const char * const rcu_tasks_gp_state_names[] = { |
128 | "RTGS_INIT", | |
129 | "RTGS_WAIT_WAIT_CBS", | |
130 | "RTGS_WAIT_GP", | |
131 | "RTGS_PRE_WAIT_GP", | |
132 | "RTGS_SCAN_TASKLIST", | |
133 | "RTGS_POST_SCAN_TASKLIST", | |
134 | "RTGS_WAIT_SCAN_HOLDOUTS", | |
135 | "RTGS_SCAN_HOLDOUTS", | |
136 | "RTGS_POST_GP", | |
137 | "RTGS_WAIT_READERS", | |
138 | "RTGS_INVOKE_CBS", | |
139 | "RTGS_WAIT_CBS", | |
140 | }; | |
8344496e | 141 | #endif /* #ifndef CONFIG_TINY_RCU */ |
af051ca4 | 142 | |
5873b8a9 PM |
143 | //////////////////////////////////////////////////////////////////////// |
144 | // | |
145 | // Generic code. | |
146 | ||
af051ca4 PM |
147 | /* Record grace-period phase and time. */ |
148 | static void set_tasks_gp_state(struct rcu_tasks *rtp, int newstate) | |
149 | { | |
150 | rtp->gp_state = newstate; | |
151 | rtp->gp_jiffies = jiffies; | |
152 | } | |
153 | ||
8344496e | 154 | #ifndef CONFIG_TINY_RCU |
af051ca4 PM |
155 | /* Return state name. */ |
156 | static const char *tasks_gp_state_getname(struct rcu_tasks *rtp) | |
157 | { | |
158 | int i = data_race(rtp->gp_state); // Let KCSAN detect update races | |
159 | int j = READ_ONCE(i); // Prevent the compiler from reading twice | |
160 | ||
161 | if (j >= ARRAY_SIZE(rcu_tasks_gp_state_names)) | |
162 | return "???"; | |
163 | return rcu_tasks_gp_state_names[j]; | |
164 | } | |
8344496e | 165 | #endif /* #ifndef CONFIG_TINY_RCU */ |
af051ca4 | 166 | |
cafafd67 PM |
167 | // Initialize per-CPU callback lists for the specified flavor of |
168 | // Tasks RCU. | |
169 | static void cblist_init_generic(struct rcu_tasks *rtp) | |
170 | { | |
171 | int cpu; | |
172 | unsigned long flags; | |
173 | ||
174 | raw_spin_lock_irqsave(&rtp->cbs_gbl_lock, flags); | |
7a30871b | 175 | rtp->percpu_enqueue_shift = ilog2(nr_cpu_ids); |
cafafd67 PM |
176 | for_each_possible_cpu(cpu) { |
177 | struct rcu_tasks_percpu *rtpcp = per_cpu_ptr(rtp->rtpcpu, cpu); | |
178 | ||
179 | WARN_ON_ONCE(!rtpcp); | |
180 | if (cpu) | |
181 | raw_spin_lock_init(&rtpcp->cbs_pcpu_lock); | |
182 | raw_spin_lock(&rtpcp->cbs_pcpu_lock); // irqs already disabled. | |
183 | if (!WARN_ON_ONCE(rtpcp->cbs_tail)) | |
184 | rtpcp->cbs_tail = &rtpcp->cbs_head; | |
185 | raw_spin_unlock(&rtpcp->cbs_pcpu_lock); // irqs remain disabled. | |
186 | } | |
187 | raw_spin_unlock_irqrestore(&rtp->cbs_gbl_lock, flags); | |
188 | ||
189 | } | |
190 | ||
5873b8a9 PM |
191 | // Enqueue a callback for the specified flavor of Tasks RCU. |
192 | static void call_rcu_tasks_generic(struct rcu_head *rhp, rcu_callback_t func, | |
193 | struct rcu_tasks *rtp) | |
eacd6f04 PM |
194 | { |
195 | unsigned long flags; | |
196 | bool needwake; | |
cafafd67 | 197 | struct rcu_tasks_percpu *rtpcp; |
eacd6f04 PM |
198 | |
199 | rhp->next = NULL; | |
200 | rhp->func = func; | |
cafafd67 | 201 | local_irq_save(flags); |
7a30871b PM |
202 | rtpcp = per_cpu_ptr(rtp->rtpcpu, |
203 | smp_processor_id() >> READ_ONCE(rtp->percpu_enqueue_shift)); | |
cafafd67 PM |
204 | raw_spin_lock(&rtpcp->cbs_pcpu_lock); |
205 | if (!rtpcp->cbs_tail) { | |
206 | raw_spin_unlock(&rtpcp->cbs_pcpu_lock); // irqs remain disabled. | |
207 | cblist_init_generic(rtp); | |
208 | raw_spin_lock(&rtpcp->cbs_pcpu_lock); // irqs already disabled. | |
209 | } | |
210 | needwake = !rtpcp->cbs_head; | |
211 | WRITE_ONCE(*rtpcp->cbs_tail, rhp); | |
212 | rtpcp->cbs_tail = &rhp->next; | |
213 | raw_spin_unlock_irqrestore(&rtpcp->cbs_pcpu_lock, flags); | |
eacd6f04 | 214 | /* We can't create the thread unless interrupts are enabled. */ |
07e10515 PM |
215 | if (needwake && READ_ONCE(rtp->kthread_ptr)) |
216 | wake_up(&rtp->cbs_wq); | |
eacd6f04 | 217 | } |
eacd6f04 | 218 | |
5873b8a9 PM |
219 | // Wait for a grace period for the specified flavor of Tasks RCU. |
220 | static void synchronize_rcu_tasks_generic(struct rcu_tasks *rtp) | |
eacd6f04 PM |
221 | { |
222 | /* Complain if the scheduler has not started. */ | |
223 | RCU_LOCKDEP_WARN(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE, | |
224 | "synchronize_rcu_tasks called too soon"); | |
225 | ||
226 | /* Wait for the grace period. */ | |
5873b8a9 | 227 | wait_rcu_gp(rtp->call_func); |
eacd6f04 PM |
228 | } |
229 | ||
230 | /* RCU-tasks kthread that detects grace periods and invokes callbacks. */ | |
231 | static int __noreturn rcu_tasks_kthread(void *arg) | |
232 | { | |
233 | unsigned long flags; | |
eacd6f04 PM |
234 | struct rcu_head *list; |
235 | struct rcu_head *next; | |
07e10515 | 236 | struct rcu_tasks *rtp = arg; |
eacd6f04 PM |
237 | |
238 | /* Run on housekeeping CPUs by default. Sysadm can move if desired. */ | |
239 | housekeeping_affine(current, HK_FLAG_RCU); | |
07e10515 | 240 | WRITE_ONCE(rtp->kthread_ptr, current); // Let GPs start! |
eacd6f04 PM |
241 | |
242 | /* | |
243 | * Each pass through the following loop makes one check for | |
244 | * newly arrived callbacks, and, if there are some, waits for | |
245 | * one RCU-tasks grace period and then invokes the callbacks. | |
246 | * This loop is terminated by the system going down. ;-) | |
247 | */ | |
248 | for (;;) { | |
cafafd67 PM |
249 | struct rcu_tasks_percpu *rtpcp = per_cpu_ptr(rtp->rtpcpu, 0); // for_each... |
250 | ||
0db7c32a | 251 | set_tasks_gp_state(rtp, RTGS_WAIT_CBS); |
eacd6f04 PM |
252 | |
253 | /* Pick up any new callbacks. */ | |
cafafd67 | 254 | raw_spin_lock_irqsave(&rtpcp->cbs_pcpu_lock, flags); |
43766c3e | 255 | smp_mb__after_spinlock(); // Order updates vs. GP. |
cafafd67 PM |
256 | list = rtpcp->cbs_head; |
257 | rtpcp->cbs_head = NULL; | |
258 | rtpcp->cbs_tail = &rtpcp->cbs_head; | |
259 | raw_spin_unlock_irqrestore(&rtpcp->cbs_pcpu_lock, flags); | |
eacd6f04 PM |
260 | |
261 | /* If there were none, wait a bit and start over. */ | |
262 | if (!list) { | |
07e10515 | 263 | wait_event_interruptible(rtp->cbs_wq, |
cafafd67 PM |
264 | READ_ONCE(rtpcp->cbs_head)); |
265 | if (!rtpcp->cbs_head) { | |
eacd6f04 | 266 | WARN_ON(signal_pending(current)); |
af051ca4 | 267 | set_tasks_gp_state(rtp, RTGS_WAIT_WAIT_CBS); |
ea6eed9f | 268 | schedule_timeout_idle(HZ/10); |
eacd6f04 PM |
269 | } |
270 | continue; | |
271 | } | |
272 | ||
5873b8a9 | 273 | // Wait for one grace period. |
af051ca4 | 274 | set_tasks_gp_state(rtp, RTGS_WAIT_GP); |
88092d0c | 275 | rtp->gp_start = jiffies; |
5873b8a9 | 276 | rtp->gp_func(rtp); |
238dbce3 | 277 | rtp->n_gps++; |
eacd6f04 PM |
278 | |
279 | /* Invoke the callbacks. */ | |
af051ca4 | 280 | set_tasks_gp_state(rtp, RTGS_INVOKE_CBS); |
eacd6f04 PM |
281 | while (list) { |
282 | next = list->next; | |
283 | local_bh_disable(); | |
284 | list->func(list); | |
285 | local_bh_enable(); | |
286 | list = next; | |
287 | cond_resched(); | |
288 | } | |
289 | /* Paranoid sleep to keep this from entering a tight loop */ | |
4fe192df | 290 | schedule_timeout_idle(rtp->gp_sleep); |
eacd6f04 PM |
291 | } |
292 | } | |
293 | ||
1b04fa99 | 294 | /* Spawn RCU-tasks grace-period kthread. */ |
5873b8a9 | 295 | static void __init rcu_spawn_tasks_kthread_generic(struct rcu_tasks *rtp) |
eacd6f04 PM |
296 | { |
297 | struct task_struct *t; | |
298 | ||
c97d12a6 PM |
299 | t = kthread_run(rcu_tasks_kthread, rtp, "%s_kthread", rtp->kname); |
300 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start %s grace-period kthread, OOM is now expected behavior\n", __func__, rtp->name)) | |
5873b8a9 | 301 | return; |
eacd6f04 | 302 | smp_mb(); /* Ensure others see full kthread. */ |
eacd6f04 | 303 | } |
eacd6f04 | 304 | |
eacd6f04 PM |
305 | #ifndef CONFIG_TINY_RCU |
306 | ||
307 | /* | |
308 | * Print any non-default Tasks RCU settings. | |
309 | */ | |
310 | static void __init rcu_tasks_bootup_oddness(void) | |
311 | { | |
d5f177d3 | 312 | #if defined(CONFIG_TASKS_RCU) || defined(CONFIG_TASKS_TRACE_RCU) |
eacd6f04 PM |
313 | if (rcu_task_stall_timeout != RCU_TASK_STALL_TIMEOUT) |
314 | pr_info("\tTasks-RCU CPU stall warnings timeout set to %d (rcu_task_stall_timeout).\n", rcu_task_stall_timeout); | |
d5f177d3 PM |
315 | #endif /* #ifdef CONFIG_TASKS_RCU */ |
316 | #ifdef CONFIG_TASKS_RCU | |
317 | pr_info("\tTrampoline variant of Tasks RCU enabled.\n"); | |
eacd6f04 | 318 | #endif /* #ifdef CONFIG_TASKS_RCU */ |
c84aad76 PM |
319 | #ifdef CONFIG_TASKS_RUDE_RCU |
320 | pr_info("\tRude variant of Tasks RCU enabled.\n"); | |
321 | #endif /* #ifdef CONFIG_TASKS_RUDE_RCU */ | |
d5f177d3 PM |
322 | #ifdef CONFIG_TASKS_TRACE_RCU |
323 | pr_info("\tTracing variant of Tasks RCU enabled.\n"); | |
324 | #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */ | |
eacd6f04 PM |
325 | } |
326 | ||
327 | #endif /* #ifndef CONFIG_TINY_RCU */ | |
5873b8a9 | 328 | |
8344496e | 329 | #ifndef CONFIG_TINY_RCU |
e21408ce PM |
330 | /* Dump out rcutorture-relevant state common to all RCU-tasks flavors. */ |
331 | static void show_rcu_tasks_generic_gp_kthread(struct rcu_tasks *rtp, char *s) | |
332 | { | |
cafafd67 | 333 | struct rcu_tasks_percpu *rtpcp = per_cpu_ptr(rtp->rtpcpu, 0); // for_each... |
7e0669c3 | 334 | pr_info("%s: %s(%d) since %lu g:%lu i:%lu/%lu %c%c %s\n", |
e21408ce | 335 | rtp->kname, |
7e0669c3 | 336 | tasks_gp_state_getname(rtp), data_race(rtp->gp_state), |
af051ca4 | 337 | jiffies - data_race(rtp->gp_jiffies), |
7e0669c3 PM |
338 | data_race(rtp->n_gps), |
339 | data_race(rtp->n_ipis_fails), data_race(rtp->n_ipis), | |
e21408ce | 340 | ".k"[!!data_race(rtp->kthread_ptr)], |
cafafd67 | 341 | ".C"[!!data_race(rtpcp->cbs_head)], |
e21408ce PM |
342 | s); |
343 | } | |
27c0f144 | 344 | #endif // #ifndef CONFIG_TINY_RCU |
e21408ce | 345 | |
25246fc8 PM |
346 | static void exit_tasks_rcu_finish_trace(struct task_struct *t); |
347 | ||
348 | #if defined(CONFIG_TASKS_RCU) || defined(CONFIG_TASKS_TRACE_RCU) | |
5873b8a9 | 349 | |
d01aa263 PM |
350 | //////////////////////////////////////////////////////////////////////// |
351 | // | |
352 | // Shared code between task-list-scanning variants of Tasks RCU. | |
353 | ||
354 | /* Wait for one RCU-tasks grace period. */ | |
355 | static void rcu_tasks_wait_gp(struct rcu_tasks *rtp) | |
356 | { | |
357 | struct task_struct *g, *t; | |
358 | unsigned long lastreport; | |
359 | LIST_HEAD(holdouts); | |
360 | int fract; | |
361 | ||
af051ca4 | 362 | set_tasks_gp_state(rtp, RTGS_PRE_WAIT_GP); |
d01aa263 PM |
363 | rtp->pregp_func(); |
364 | ||
365 | /* | |
366 | * There were callbacks, so we need to wait for an RCU-tasks | |
367 | * grace period. Start off by scanning the task list for tasks | |
368 | * that are not already voluntarily blocked. Mark these tasks | |
369 | * and make a list of them in holdouts. | |
370 | */ | |
af051ca4 | 371 | set_tasks_gp_state(rtp, RTGS_SCAN_TASKLIST); |
d01aa263 PM |
372 | rcu_read_lock(); |
373 | for_each_process_thread(g, t) | |
374 | rtp->pertask_func(t, &holdouts); | |
375 | rcu_read_unlock(); | |
376 | ||
af051ca4 | 377 | set_tasks_gp_state(rtp, RTGS_POST_SCAN_TASKLIST); |
9796e1ae | 378 | rtp->postscan_func(&holdouts); |
d01aa263 PM |
379 | |
380 | /* | |
381 | * Each pass through the following loop scans the list of holdout | |
382 | * tasks, removing any that are no longer holdouts. When the list | |
383 | * is empty, we are done. | |
384 | */ | |
385 | lastreport = jiffies; | |
386 | ||
2393a613 PM |
387 | // Start off with initial wait and slowly back off to 1 HZ wait. |
388 | fract = rtp->init_fract; | |
d01aa263 | 389 | |
77dc1741 | 390 | while (!list_empty(&holdouts)) { |
d01aa263 PM |
391 | bool firstreport; |
392 | bool needreport; | |
393 | int rtst; | |
394 | ||
d01aa263 | 395 | /* Slowly back off waiting for holdouts */ |
af051ca4 | 396 | set_tasks_gp_state(rtp, RTGS_WAIT_SCAN_HOLDOUTS); |
75dc2da5 | 397 | schedule_timeout_idle(fract); |
d01aa263 | 398 | |
75dc2da5 PM |
399 | if (fract < HZ) |
400 | fract++; | |
d01aa263 PM |
401 | |
402 | rtst = READ_ONCE(rcu_task_stall_timeout); | |
403 | needreport = rtst > 0 && time_after(jiffies, lastreport + rtst); | |
404 | if (needreport) | |
405 | lastreport = jiffies; | |
406 | firstreport = true; | |
407 | WARN_ON(signal_pending(current)); | |
af051ca4 | 408 | set_tasks_gp_state(rtp, RTGS_SCAN_HOLDOUTS); |
d01aa263 PM |
409 | rtp->holdouts_func(&holdouts, needreport, &firstreport); |
410 | } | |
411 | ||
af051ca4 PM |
412 | set_tasks_gp_state(rtp, RTGS_POST_GP); |
413 | rtp->postgp_func(rtp); | |
d01aa263 PM |
414 | } |
415 | ||
25246fc8 PM |
416 | #endif /* #if defined(CONFIG_TASKS_RCU) || defined(CONFIG_TASKS_TRACE_RCU) */ |
417 | ||
418 | #ifdef CONFIG_TASKS_RCU | |
419 | ||
5873b8a9 PM |
420 | //////////////////////////////////////////////////////////////////////// |
421 | // | |
422 | // Simple variant of RCU whose quiescent states are voluntary context | |
8af9e2c7 | 423 | // switch, cond_resched_tasks_rcu_qs(), user-space execution, and idle. |
5873b8a9 PM |
424 | // As such, grace periods can take one good long time. There are no |
425 | // read-side primitives similar to rcu_read_lock() and rcu_read_unlock() | |
426 | // because this implementation is intended to get the system into a safe | |
427 | // state for some of the manipulations involved in tracing and the like. | |
428 | // Finally, this implementation does not support high call_rcu_tasks() | |
429 | // rates from multiple CPUs. If this is required, per-CPU callback lists | |
430 | // will be needed. | |
06a3ec92 PM |
431 | // |
432 | // The implementation uses rcu_tasks_wait_gp(), which relies on function | |
433 | // pointers in the rcu_tasks structure. The rcu_spawn_tasks_kthread() | |
434 | // function sets these function pointers up so that rcu_tasks_wait_gp() | |
435 | // invokes these functions in this order: | |
436 | // | |
437 | // rcu_tasks_pregp_step(): | |
438 | // Invokes synchronize_rcu() in order to wait for all in-flight | |
439 | // t->on_rq and t->nvcsw transitions to complete. This works because | |
440 | // all such transitions are carried out with interrupts disabled. | |
441 | // rcu_tasks_pertask(), invoked on every non-idle task: | |
442 | // For every runnable non-idle task other than the current one, use | |
443 | // get_task_struct() to pin down that task, snapshot that task's | |
444 | // number of voluntary context switches, and add that task to the | |
445 | // holdout list. | |
446 | // rcu_tasks_postscan(): | |
447 | // Invoke synchronize_srcu() to ensure that all tasks that were | |
448 | // in the process of exiting (and which thus might not know to | |
449 | // synchronize with this RCU Tasks grace period) have completed | |
450 | // exiting. | |
451 | // check_all_holdout_tasks(), repeatedly until holdout list is empty: | |
452 | // Scans the holdout list, attempting to identify a quiescent state | |
453 | // for each task on the list. If there is a quiescent state, the | |
454 | // corresponding task is removed from the holdout list. | |
455 | // rcu_tasks_postgp(): | |
456 | // Invokes synchronize_rcu() in order to ensure that all prior | |
457 | // t->on_rq and t->nvcsw transitions are seen by all CPUs and tasks | |
458 | // to have happened before the end of this RCU Tasks grace period. | |
459 | // Again, this works because all such transitions are carried out | |
460 | // with interrupts disabled. | |
461 | // | |
462 | // For each exiting task, the exit_tasks_rcu_start() and | |
463 | // exit_tasks_rcu_finish() functions begin and end, respectively, the SRCU | |
464 | // read-side critical sections waited for by rcu_tasks_postscan(). | |
465 | // | |
466 | // Pre-grace-period update-side code is ordered before the grace via the | |
467 | // ->cbs_lock and the smp_mb__after_spinlock(). Pre-grace-period read-side | |
468 | // code is ordered before the grace period via synchronize_rcu() call | |
469 | // in rcu_tasks_pregp_step() and by the scheduler's locks and interrupt | |
470 | // disabling. | |
5873b8a9 | 471 | |
e4fe5dd6 PM |
472 | /* Pre-grace-period preparation. */ |
473 | static void rcu_tasks_pregp_step(void) | |
474 | { | |
475 | /* | |
476 | * Wait for all pre-existing t->on_rq and t->nvcsw transitions | |
477 | * to complete. Invoking synchronize_rcu() suffices because all | |
478 | * these transitions occur with interrupts disabled. Without this | |
479 | * synchronize_rcu(), a read-side critical section that started | |
480 | * before the grace period might be incorrectly seen as having | |
481 | * started after the grace period. | |
482 | * | |
483 | * This synchronize_rcu() also dispenses with the need for a | |
484 | * memory barrier on the first store to t->rcu_tasks_holdout, | |
485 | * as it forces the store to happen after the beginning of the | |
486 | * grace period. | |
487 | */ | |
488 | synchronize_rcu(); | |
489 | } | |
490 | ||
491 | /* Per-task initial processing. */ | |
492 | static void rcu_tasks_pertask(struct task_struct *t, struct list_head *hop) | |
493 | { | |
494 | if (t != current && READ_ONCE(t->on_rq) && !is_idle_task(t)) { | |
495 | get_task_struct(t); | |
496 | t->rcu_tasks_nvcsw = READ_ONCE(t->nvcsw); | |
497 | WRITE_ONCE(t->rcu_tasks_holdout, true); | |
498 | list_add(&t->rcu_tasks_holdout_list, hop); | |
499 | } | |
500 | } | |
501 | ||
502 | /* Processing between scanning taskslist and draining the holdout list. */ | |
04a3c5aa | 503 | static void rcu_tasks_postscan(struct list_head *hop) |
e4fe5dd6 PM |
504 | { |
505 | /* | |
506 | * Wait for tasks that are in the process of exiting. This | |
507 | * does only part of the job, ensuring that all tasks that were | |
508 | * previously exiting reach the point where they have disabled | |
509 | * preemption, allowing the later synchronize_rcu() to finish | |
510 | * the job. | |
511 | */ | |
512 | synchronize_srcu(&tasks_rcu_exit_srcu); | |
513 | } | |
514 | ||
5873b8a9 PM |
515 | /* See if tasks are still holding out, complain if so. */ |
516 | static void check_holdout_task(struct task_struct *t, | |
517 | bool needreport, bool *firstreport) | |
518 | { | |
519 | int cpu; | |
520 | ||
521 | if (!READ_ONCE(t->rcu_tasks_holdout) || | |
522 | t->rcu_tasks_nvcsw != READ_ONCE(t->nvcsw) || | |
523 | !READ_ONCE(t->on_rq) || | |
524 | (IS_ENABLED(CONFIG_NO_HZ_FULL) && | |
525 | !is_idle_task(t) && t->rcu_tasks_idle_cpu >= 0)) { | |
526 | WRITE_ONCE(t->rcu_tasks_holdout, false); | |
527 | list_del_init(&t->rcu_tasks_holdout_list); | |
528 | put_task_struct(t); | |
529 | return; | |
530 | } | |
531 | rcu_request_urgent_qs_task(t); | |
532 | if (!needreport) | |
533 | return; | |
534 | if (*firstreport) { | |
535 | pr_err("INFO: rcu_tasks detected stalls on tasks:\n"); | |
536 | *firstreport = false; | |
537 | } | |
538 | cpu = task_cpu(t); | |
539 | pr_alert("%p: %c%c nvcsw: %lu/%lu holdout: %d idle_cpu: %d/%d\n", | |
540 | t, ".I"[is_idle_task(t)], | |
541 | "N."[cpu < 0 || !tick_nohz_full_cpu(cpu)], | |
542 | t->rcu_tasks_nvcsw, t->nvcsw, t->rcu_tasks_holdout, | |
543 | t->rcu_tasks_idle_cpu, cpu); | |
544 | sched_show_task(t); | |
545 | } | |
546 | ||
e4fe5dd6 PM |
547 | /* Scan the holdout lists for tasks no longer holding out. */ |
548 | static void check_all_holdout_tasks(struct list_head *hop, | |
549 | bool needreport, bool *firstreport) | |
550 | { | |
551 | struct task_struct *t, *t1; | |
552 | ||
553 | list_for_each_entry_safe(t, t1, hop, rcu_tasks_holdout_list) { | |
554 | check_holdout_task(t, needreport, firstreport); | |
555 | cond_resched(); | |
556 | } | |
557 | } | |
558 | ||
559 | /* Finish off the Tasks-RCU grace period. */ | |
af051ca4 | 560 | static void rcu_tasks_postgp(struct rcu_tasks *rtp) |
e4fe5dd6 PM |
561 | { |
562 | /* | |
563 | * Because ->on_rq and ->nvcsw are not guaranteed to have a full | |
564 | * memory barriers prior to them in the schedule() path, memory | |
565 | * reordering on other CPUs could cause their RCU-tasks read-side | |
566 | * critical sections to extend past the end of the grace period. | |
567 | * However, because these ->nvcsw updates are carried out with | |
568 | * interrupts disabled, we can use synchronize_rcu() to force the | |
569 | * needed ordering on all such CPUs. | |
570 | * | |
571 | * This synchronize_rcu() also confines all ->rcu_tasks_holdout | |
572 | * accesses to be within the grace period, avoiding the need for | |
573 | * memory barriers for ->rcu_tasks_holdout accesses. | |
574 | * | |
575 | * In addition, this synchronize_rcu() waits for exiting tasks | |
576 | * to complete their final preempt_disable() region of execution, | |
577 | * cleaning up after the synchronize_srcu() above. | |
578 | */ | |
579 | synchronize_rcu(); | |
580 | } | |
581 | ||
5873b8a9 | 582 | void call_rcu_tasks(struct rcu_head *rhp, rcu_callback_t func); |
c97d12a6 | 583 | DEFINE_RCU_TASKS(rcu_tasks, rcu_tasks_wait_gp, call_rcu_tasks, "RCU Tasks"); |
5873b8a9 PM |
584 | |
585 | /** | |
586 | * call_rcu_tasks() - Queue an RCU for invocation task-based grace period | |
587 | * @rhp: structure to be used for queueing the RCU updates. | |
588 | * @func: actual callback function to be invoked after the grace period | |
589 | * | |
590 | * The callback function will be invoked some time after a full grace | |
591 | * period elapses, in other words after all currently executing RCU | |
592 | * read-side critical sections have completed. call_rcu_tasks() assumes | |
593 | * that the read-side critical sections end at a voluntary context | |
8af9e2c7 | 594 | * switch (not a preemption!), cond_resched_tasks_rcu_qs(), entry into idle, |
5873b8a9 PM |
595 | * or transition to usermode execution. As such, there are no read-side |
596 | * primitives analogous to rcu_read_lock() and rcu_read_unlock() because | |
597 | * this primitive is intended to determine that all tasks have passed | |
a616aec9 | 598 | * through a safe state, not so much for data-structure synchronization. |
5873b8a9 PM |
599 | * |
600 | * See the description of call_rcu() for more detailed information on | |
601 | * memory ordering guarantees. | |
602 | */ | |
603 | void call_rcu_tasks(struct rcu_head *rhp, rcu_callback_t func) | |
604 | { | |
605 | call_rcu_tasks_generic(rhp, func, &rcu_tasks); | |
606 | } | |
607 | EXPORT_SYMBOL_GPL(call_rcu_tasks); | |
608 | ||
609 | /** | |
610 | * synchronize_rcu_tasks - wait until an rcu-tasks grace period has elapsed. | |
611 | * | |
612 | * Control will return to the caller some time after a full rcu-tasks | |
613 | * grace period has elapsed, in other words after all currently | |
614 | * executing rcu-tasks read-side critical sections have elapsed. These | |
615 | * read-side critical sections are delimited by calls to schedule(), | |
616 | * cond_resched_tasks_rcu_qs(), idle execution, userspace execution, calls | |
617 | * to synchronize_rcu_tasks(), and (in theory, anyway) cond_resched(). | |
618 | * | |
619 | * This is a very specialized primitive, intended only for a few uses in | |
620 | * tracing and other situations requiring manipulation of function | |
621 | * preambles and profiling hooks. The synchronize_rcu_tasks() function | |
622 | * is not (yet) intended for heavy use from multiple CPUs. | |
623 | * | |
624 | * See the description of synchronize_rcu() for more detailed information | |
625 | * on memory ordering guarantees. | |
626 | */ | |
627 | void synchronize_rcu_tasks(void) | |
628 | { | |
629 | synchronize_rcu_tasks_generic(&rcu_tasks); | |
630 | } | |
631 | EXPORT_SYMBOL_GPL(synchronize_rcu_tasks); | |
632 | ||
633 | /** | |
634 | * rcu_barrier_tasks - Wait for in-flight call_rcu_tasks() callbacks. | |
635 | * | |
636 | * Although the current implementation is guaranteed to wait, it is not | |
637 | * obligated to, for example, if there are no pending callbacks. | |
638 | */ | |
639 | void rcu_barrier_tasks(void) | |
640 | { | |
641 | /* There is only one callback queue, so this is easy. ;-) */ | |
642 | synchronize_rcu_tasks(); | |
643 | } | |
644 | EXPORT_SYMBOL_GPL(rcu_barrier_tasks); | |
645 | ||
646 | static int __init rcu_spawn_tasks_kthread(void) | |
647 | { | |
cafafd67 | 648 | cblist_init_generic(&rcu_tasks); |
4fe192df | 649 | rcu_tasks.gp_sleep = HZ / 10; |
75dc2da5 | 650 | rcu_tasks.init_fract = HZ / 10; |
e4fe5dd6 PM |
651 | rcu_tasks.pregp_func = rcu_tasks_pregp_step; |
652 | rcu_tasks.pertask_func = rcu_tasks_pertask; | |
653 | rcu_tasks.postscan_func = rcu_tasks_postscan; | |
654 | rcu_tasks.holdouts_func = check_all_holdout_tasks; | |
655 | rcu_tasks.postgp_func = rcu_tasks_postgp; | |
5873b8a9 PM |
656 | rcu_spawn_tasks_kthread_generic(&rcu_tasks); |
657 | return 0; | |
658 | } | |
5873b8a9 | 659 | |
27c0f144 PM |
660 | #if !defined(CONFIG_TINY_RCU) |
661 | void show_rcu_tasks_classic_gp_kthread(void) | |
e21408ce PM |
662 | { |
663 | show_rcu_tasks_generic_gp_kthread(&rcu_tasks, ""); | |
664 | } | |
27c0f144 PM |
665 | EXPORT_SYMBOL_GPL(show_rcu_tasks_classic_gp_kthread); |
666 | #endif // !defined(CONFIG_TINY_RCU) | |
e21408ce | 667 | |
25246fc8 PM |
668 | /* Do the srcu_read_lock() for the above synchronize_srcu(). */ |
669 | void exit_tasks_rcu_start(void) __acquires(&tasks_rcu_exit_srcu) | |
670 | { | |
671 | preempt_disable(); | |
672 | current->rcu_tasks_idx = __srcu_read_lock(&tasks_rcu_exit_srcu); | |
673 | preempt_enable(); | |
674 | } | |
675 | ||
676 | /* Do the srcu_read_unlock() for the above synchronize_srcu(). */ | |
677 | void exit_tasks_rcu_finish(void) __releases(&tasks_rcu_exit_srcu) | |
678 | { | |
679 | struct task_struct *t = current; | |
680 | ||
681 | preempt_disable(); | |
682 | __srcu_read_unlock(&tasks_rcu_exit_srcu, t->rcu_tasks_idx); | |
683 | preempt_enable(); | |
684 | exit_tasks_rcu_finish_trace(t); | |
685 | } | |
686 | ||
e21408ce | 687 | #else /* #ifdef CONFIG_TASKS_RCU */ |
25246fc8 PM |
688 | void exit_tasks_rcu_start(void) { } |
689 | void exit_tasks_rcu_finish(void) { exit_tasks_rcu_finish_trace(current); } | |
e21408ce | 690 | #endif /* #else #ifdef CONFIG_TASKS_RCU */ |
c84aad76 PM |
691 | |
692 | #ifdef CONFIG_TASKS_RUDE_RCU | |
693 | ||
694 | //////////////////////////////////////////////////////////////////////// | |
695 | // | |
696 | // "Rude" variant of Tasks RCU, inspired by Steve Rostedt's trick of | |
697 | // passing an empty function to schedule_on_each_cpu(). This approach | |
e4be1f44 PM |
698 | // provides an asynchronous call_rcu_tasks_rude() API and batching of |
699 | // concurrent calls to the synchronous synchronize_rcu_tasks_rude() API. | |
9fc98e31 PM |
700 | // This invokes schedule_on_each_cpu() in order to send IPIs far and wide |
701 | // and induces otherwise unnecessary context switches on all online CPUs, | |
702 | // whether idle or not. | |
703 | // | |
704 | // Callback handling is provided by the rcu_tasks_kthread() function. | |
705 | // | |
706 | // Ordering is provided by the scheduler's context-switch code. | |
c84aad76 PM |
707 | |
708 | // Empty function to allow workqueues to force a context switch. | |
709 | static void rcu_tasks_be_rude(struct work_struct *work) | |
710 | { | |
711 | } | |
712 | ||
713 | // Wait for one rude RCU-tasks grace period. | |
714 | static void rcu_tasks_rude_wait_gp(struct rcu_tasks *rtp) | |
715 | { | |
238dbce3 | 716 | rtp->n_ipis += cpumask_weight(cpu_online_mask); |
c84aad76 PM |
717 | schedule_on_each_cpu(rcu_tasks_be_rude); |
718 | } | |
719 | ||
720 | void call_rcu_tasks_rude(struct rcu_head *rhp, rcu_callback_t func); | |
c97d12a6 PM |
721 | DEFINE_RCU_TASKS(rcu_tasks_rude, rcu_tasks_rude_wait_gp, call_rcu_tasks_rude, |
722 | "RCU Tasks Rude"); | |
c84aad76 PM |
723 | |
724 | /** | |
725 | * call_rcu_tasks_rude() - Queue a callback rude task-based grace period | |
726 | * @rhp: structure to be used for queueing the RCU updates. | |
727 | * @func: actual callback function to be invoked after the grace period | |
728 | * | |
729 | * The callback function will be invoked some time after a full grace | |
730 | * period elapses, in other words after all currently executing RCU | |
731 | * read-side critical sections have completed. call_rcu_tasks_rude() | |
732 | * assumes that the read-side critical sections end at context switch, | |
8af9e2c7 | 733 | * cond_resched_tasks_rcu_qs(), or transition to usermode execution (as |
a6517e9c NU |
734 | * usermode execution is schedulable). As such, there are no read-side |
735 | * primitives analogous to rcu_read_lock() and rcu_read_unlock() because | |
736 | * this primitive is intended to determine that all tasks have passed | |
737 | * through a safe state, not so much for data-structure synchronization. | |
c84aad76 PM |
738 | * |
739 | * See the description of call_rcu() for more detailed information on | |
740 | * memory ordering guarantees. | |
741 | */ | |
742 | void call_rcu_tasks_rude(struct rcu_head *rhp, rcu_callback_t func) | |
743 | { | |
744 | call_rcu_tasks_generic(rhp, func, &rcu_tasks_rude); | |
745 | } | |
746 | EXPORT_SYMBOL_GPL(call_rcu_tasks_rude); | |
747 | ||
748 | /** | |
749 | * synchronize_rcu_tasks_rude - wait for a rude rcu-tasks grace period | |
750 | * | |
751 | * Control will return to the caller some time after a rude rcu-tasks | |
752 | * grace period has elapsed, in other words after all currently | |
753 | * executing rcu-tasks read-side critical sections have elapsed. These | |
754 | * read-side critical sections are delimited by calls to schedule(), | |
a6517e9c NU |
755 | * cond_resched_tasks_rcu_qs(), userspace execution (which is a schedulable |
756 | * context), and (in theory, anyway) cond_resched(). | |
c84aad76 PM |
757 | * |
758 | * This is a very specialized primitive, intended only for a few uses in | |
759 | * tracing and other situations requiring manipulation of function preambles | |
760 | * and profiling hooks. The synchronize_rcu_tasks_rude() function is not | |
761 | * (yet) intended for heavy use from multiple CPUs. | |
762 | * | |
763 | * See the description of synchronize_rcu() for more detailed information | |
764 | * on memory ordering guarantees. | |
765 | */ | |
766 | void synchronize_rcu_tasks_rude(void) | |
767 | { | |
768 | synchronize_rcu_tasks_generic(&rcu_tasks_rude); | |
769 | } | |
770 | EXPORT_SYMBOL_GPL(synchronize_rcu_tasks_rude); | |
771 | ||
772 | /** | |
773 | * rcu_barrier_tasks_rude - Wait for in-flight call_rcu_tasks_rude() callbacks. | |
774 | * | |
775 | * Although the current implementation is guaranteed to wait, it is not | |
776 | * obligated to, for example, if there are no pending callbacks. | |
777 | */ | |
778 | void rcu_barrier_tasks_rude(void) | |
779 | { | |
780 | /* There is only one callback queue, so this is easy. ;-) */ | |
781 | synchronize_rcu_tasks_rude(); | |
782 | } | |
783 | EXPORT_SYMBOL_GPL(rcu_barrier_tasks_rude); | |
784 | ||
785 | static int __init rcu_spawn_tasks_rude_kthread(void) | |
786 | { | |
cafafd67 | 787 | cblist_init_generic(&rcu_tasks_rude); |
4fe192df | 788 | rcu_tasks_rude.gp_sleep = HZ / 10; |
c84aad76 PM |
789 | rcu_spawn_tasks_kthread_generic(&rcu_tasks_rude); |
790 | return 0; | |
791 | } | |
c84aad76 | 792 | |
27c0f144 PM |
793 | #if !defined(CONFIG_TINY_RCU) |
794 | void show_rcu_tasks_rude_gp_kthread(void) | |
e21408ce PM |
795 | { |
796 | show_rcu_tasks_generic_gp_kthread(&rcu_tasks_rude, ""); | |
797 | } | |
27c0f144 PM |
798 | EXPORT_SYMBOL_GPL(show_rcu_tasks_rude_gp_kthread); |
799 | #endif // !defined(CONFIG_TINY_RCU) | |
800 | #endif /* #ifdef CONFIG_TASKS_RUDE_RCU */ | |
d5f177d3 PM |
801 | |
802 | //////////////////////////////////////////////////////////////////////// | |
803 | // | |
804 | // Tracing variant of Tasks RCU. This variant is designed to be used | |
805 | // to protect tracing hooks, including those of BPF. This variant | |
806 | // therefore: | |
807 | // | |
808 | // 1. Has explicit read-side markers to allow finite grace periods | |
809 | // in the face of in-kernel loops for PREEMPT=n builds. | |
810 | // | |
811 | // 2. Protects code in the idle loop, exception entry/exit, and | |
812 | // CPU-hotplug code paths, similar to the capabilities of SRCU. | |
813 | // | |
c4f113ac | 814 | // 3. Avoids expensive read-side instructions, having overhead similar |
d5f177d3 PM |
815 | // to that of Preemptible RCU. |
816 | // | |
817 | // There are of course downsides. The grace-period code can send IPIs to | |
818 | // CPUs, even when those CPUs are in the idle loop or in nohz_full userspace. | |
819 | // It is necessary to scan the full tasklist, much as for Tasks RCU. There | |
820 | // is a single callback queue guarded by a single lock, again, much as for | |
821 | // Tasks RCU. If needed, these downsides can be at least partially remedied. | |
822 | // | |
823 | // Perhaps most important, this variant of RCU does not affect the vanilla | |
824 | // flavors, rcu_preempt and rcu_sched. The fact that RCU Tasks Trace | |
825 | // readers can operate from idle, offline, and exception entry/exit in no | |
826 | // way allows rcu_preempt and rcu_sched readers to also do so. | |
a434dd10 PM |
827 | // |
828 | // The implementation uses rcu_tasks_wait_gp(), which relies on function | |
829 | // pointers in the rcu_tasks structure. The rcu_spawn_tasks_trace_kthread() | |
830 | // function sets these function pointers up so that rcu_tasks_wait_gp() | |
831 | // invokes these functions in this order: | |
832 | // | |
833 | // rcu_tasks_trace_pregp_step(): | |
834 | // Initialize the count of readers and block CPU-hotplug operations. | |
835 | // rcu_tasks_trace_pertask(), invoked on every non-idle task: | |
836 | // Initialize per-task state and attempt to identify an immediate | |
837 | // quiescent state for that task, or, failing that, attempt to | |
838 | // set that task's .need_qs flag so that task's next outermost | |
839 | // rcu_read_unlock_trace() will report the quiescent state (in which | |
840 | // case the count of readers is incremented). If both attempts fail, | |
45f4b4a2 PM |
841 | // the task is added to a "holdout" list. Note that IPIs are used |
842 | // to invoke trc_read_check_handler() in the context of running tasks | |
843 | // in order to avoid ordering overhead on common-case shared-variable | |
844 | // accessses. | |
a434dd10 PM |
845 | // rcu_tasks_trace_postscan(): |
846 | // Initialize state and attempt to identify an immediate quiescent | |
847 | // state as above (but only for idle tasks), unblock CPU-hotplug | |
848 | // operations, and wait for an RCU grace period to avoid races with | |
849 | // tasks that are in the process of exiting. | |
850 | // check_all_holdout_tasks_trace(), repeatedly until holdout list is empty: | |
851 | // Scans the holdout list, attempting to identify a quiescent state | |
852 | // for each task on the list. If there is a quiescent state, the | |
853 | // corresponding task is removed from the holdout list. | |
854 | // rcu_tasks_trace_postgp(): | |
855 | // Wait for the count of readers do drop to zero, reporting any stalls. | |
856 | // Also execute full memory barriers to maintain ordering with code | |
857 | // executing after the grace period. | |
858 | // | |
859 | // The exit_tasks_rcu_finish_trace() synchronizes with exiting tasks. | |
860 | // | |
861 | // Pre-grace-period update-side code is ordered before the grace | |
862 | // period via the ->cbs_lock and barriers in rcu_tasks_kthread(). | |
863 | // Pre-grace-period read-side code is ordered before the grace period by | |
864 | // atomic_dec_and_test() of the count of readers (for IPIed readers) and by | |
865 | // scheduler context-switch ordering (for locked-down non-running readers). | |
d5f177d3 PM |
866 | |
867 | // The lockdep state must be outside of #ifdef to be useful. | |
868 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
869 | static struct lock_class_key rcu_lock_trace_key; | |
870 | struct lockdep_map rcu_trace_lock_map = | |
871 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_trace", &rcu_lock_trace_key); | |
872 | EXPORT_SYMBOL_GPL(rcu_trace_lock_map); | |
873 | #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
874 | ||
875 | #ifdef CONFIG_TASKS_TRACE_RCU | |
876 | ||
30d8aa51 PM |
877 | static atomic_t trc_n_readers_need_end; // Number of waited-for readers. |
878 | static DECLARE_WAIT_QUEUE_HEAD(trc_wait); // List of holdout tasks. | |
d5f177d3 PM |
879 | |
880 | // Record outstanding IPIs to each CPU. No point in sending two... | |
881 | static DEFINE_PER_CPU(bool, trc_ipi_to_cpu); | |
882 | ||
40471509 PM |
883 | // The number of detections of task quiescent state relying on |
884 | // heavyweight readers executing explicit memory barriers. | |
6731da9e PM |
885 | static unsigned long n_heavy_reader_attempts; |
886 | static unsigned long n_heavy_reader_updates; | |
887 | static unsigned long n_heavy_reader_ofl_updates; | |
40471509 | 888 | |
b0afa0f0 PM |
889 | void call_rcu_tasks_trace(struct rcu_head *rhp, rcu_callback_t func); |
890 | DEFINE_RCU_TASKS(rcu_tasks_trace, rcu_tasks_wait_gp, call_rcu_tasks_trace, | |
891 | "RCU Tasks Trace"); | |
892 | ||
b38f57c1 PM |
893 | /* |
894 | * This irq_work handler allows rcu_read_unlock_trace() to be invoked | |
895 | * while the scheduler locks are held. | |
896 | */ | |
897 | static void rcu_read_unlock_iw(struct irq_work *iwp) | |
898 | { | |
899 | wake_up(&trc_wait); | |
900 | } | |
901 | static DEFINE_IRQ_WORK(rcu_tasks_trace_iw, rcu_read_unlock_iw); | |
902 | ||
d5f177d3 | 903 | /* If we are the last reader, wake up the grace-period kthread. */ |
a5c071cc | 904 | void rcu_read_unlock_trace_special(struct task_struct *t) |
d5f177d3 | 905 | { |
f8ab3fad | 906 | int nq = READ_ONCE(t->trc_reader_special.b.need_qs); |
276c4104 | 907 | |
9ae58d7b PM |
908 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB) && |
909 | t->trc_reader_special.b.need_mb) | |
276c4104 PM |
910 | smp_mb(); // Pairs with update-side barriers. |
911 | // Update .need_qs before ->trc_reader_nesting for irq/NMI handlers. | |
912 | if (nq) | |
913 | WRITE_ONCE(t->trc_reader_special.b.need_qs, false); | |
a5c071cc | 914 | WRITE_ONCE(t->trc_reader_nesting, 0); |
276c4104 | 915 | if (nq && atomic_dec_and_test(&trc_n_readers_need_end)) |
b38f57c1 | 916 | irq_work_queue(&rcu_tasks_trace_iw); |
d5f177d3 PM |
917 | } |
918 | EXPORT_SYMBOL_GPL(rcu_read_unlock_trace_special); | |
919 | ||
920 | /* Add a task to the holdout list, if it is not already on the list. */ | |
921 | static void trc_add_holdout(struct task_struct *t, struct list_head *bhp) | |
922 | { | |
923 | if (list_empty(&t->trc_holdout_list)) { | |
924 | get_task_struct(t); | |
925 | list_add(&t->trc_holdout_list, bhp); | |
926 | } | |
927 | } | |
928 | ||
929 | /* Remove a task from the holdout list, if it is in fact present. */ | |
930 | static void trc_del_holdout(struct task_struct *t) | |
931 | { | |
932 | if (!list_empty(&t->trc_holdout_list)) { | |
933 | list_del_init(&t->trc_holdout_list); | |
934 | put_task_struct(t); | |
935 | } | |
936 | } | |
937 | ||
938 | /* IPI handler to check task state. */ | |
939 | static void trc_read_check_handler(void *t_in) | |
940 | { | |
941 | struct task_struct *t = current; | |
942 | struct task_struct *texp = t_in; | |
943 | ||
944 | // If the task is no longer running on this CPU, leave. | |
945 | if (unlikely(texp != t)) { | |
d5f177d3 PM |
946 | goto reset_ipi; // Already on holdout list, so will check later. |
947 | } | |
948 | ||
949 | // If the task is not in a read-side critical section, and | |
950 | // if this is the last reader, awaken the grace-period kthread. | |
bdb0cca0 | 951 | if (likely(!READ_ONCE(t->trc_reader_nesting))) { |
d5f177d3 PM |
952 | WRITE_ONCE(t->trc_reader_checked, true); |
953 | goto reset_ipi; | |
954 | } | |
ba3a86e4 | 955 | // If we are racing with an rcu_read_unlock_trace(), try again later. |
96017bf9 | 956 | if (unlikely(READ_ONCE(t->trc_reader_nesting) < 0)) |
ba3a86e4 | 957 | goto reset_ipi; |
d5f177d3 PM |
958 | WRITE_ONCE(t->trc_reader_checked, true); |
959 | ||
960 | // Get here if the task is in a read-side critical section. Set | |
961 | // its state so that it will awaken the grace-period kthread upon | |
962 | // exit from that critical section. | |
96017bf9 | 963 | atomic_inc(&trc_n_readers_need_end); // One more to wait on. |
f8ab3fad | 964 | WARN_ON_ONCE(READ_ONCE(t->trc_reader_special.b.need_qs)); |
276c4104 | 965 | WRITE_ONCE(t->trc_reader_special.b.need_qs, true); |
d5f177d3 PM |
966 | |
967 | reset_ipi: | |
968 | // Allow future IPIs to be sent on CPU and for task. | |
969 | // Also order this IPI handler against any later manipulations of | |
970 | // the intended task. | |
8211e922 | 971 | smp_store_release(per_cpu_ptr(&trc_ipi_to_cpu, smp_processor_id()), false); // ^^^ |
d5f177d3 PM |
972 | smp_store_release(&texp->trc_ipi_to_cpu, -1); // ^^^ |
973 | } | |
974 | ||
975 | /* Callback function for scheduler to check locked-down task. */ | |
9b3c4ab3 | 976 | static int trc_inspect_reader(struct task_struct *t, void *arg) |
d5f177d3 | 977 | { |
7d0c9c50 | 978 | int cpu = task_cpu(t); |
18f08e75 | 979 | int nesting; |
7e3b70e0 | 980 | bool ofl = cpu_is_offline(cpu); |
7d0c9c50 PM |
981 | |
982 | if (task_curr(t)) { | |
30d8aa51 | 983 | WARN_ON_ONCE(ofl && !is_idle_task(t)); |
7e3b70e0 | 984 | |
7d0c9c50 | 985 | // If no chance of heavyweight readers, do it the hard way. |
7e3b70e0 | 986 | if (!ofl && !IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) |
9b3c4ab3 | 987 | return -EINVAL; |
7d0c9c50 PM |
988 | |
989 | // If heavyweight readers are enabled on the remote task, | |
990 | // we can inspect its state despite its currently running. | |
991 | // However, we cannot safely change its state. | |
40471509 | 992 | n_heavy_reader_attempts++; |
7e3b70e0 PM |
993 | if (!ofl && // Check for "running" idle tasks on offline CPUs. |
994 | !rcu_dynticks_zero_in_eqs(cpu, &t->trc_reader_nesting)) | |
9b3c4ab3 | 995 | return -EINVAL; // No quiescent state, do it the hard way. |
40471509 | 996 | n_heavy_reader_updates++; |
edf3775f PM |
997 | if (ofl) |
998 | n_heavy_reader_ofl_updates++; | |
18f08e75 | 999 | nesting = 0; |
7d0c9c50 | 1000 | } else { |
bdb0cca0 | 1001 | // The task is not running, so C-language access is safe. |
18f08e75 | 1002 | nesting = t->trc_reader_nesting; |
7d0c9c50 | 1003 | } |
d5f177d3 | 1004 | |
18f08e75 PM |
1005 | // If not exiting a read-side critical section, mark as checked |
1006 | // so that the grace-period kthread will remove it from the | |
1007 | // holdout list. | |
1008 | t->trc_reader_checked = nesting >= 0; | |
1009 | if (nesting <= 0) | |
6fedc280 | 1010 | return nesting ? -EINVAL : 0; // If in QS, done, otherwise try again later. |
7d0c9c50 PM |
1011 | |
1012 | // The task is in a read-side critical section, so set up its | |
1013 | // state so that it will awaken the grace-period kthread upon exit | |
1014 | // from that critical section. | |
1015 | atomic_inc(&trc_n_readers_need_end); // One more to wait on. | |
f8ab3fad | 1016 | WARN_ON_ONCE(READ_ONCE(t->trc_reader_special.b.need_qs)); |
7d0c9c50 | 1017 | WRITE_ONCE(t->trc_reader_special.b.need_qs, true); |
9b3c4ab3 | 1018 | return 0; |
d5f177d3 PM |
1019 | } |
1020 | ||
1021 | /* Attempt to extract the state for the specified task. */ | |
1022 | static void trc_wait_for_one_reader(struct task_struct *t, | |
1023 | struct list_head *bhp) | |
1024 | { | |
1025 | int cpu; | |
1026 | ||
1027 | // If a previous IPI is still in flight, let it complete. | |
1028 | if (smp_load_acquire(&t->trc_ipi_to_cpu) != -1) // Order IPI | |
1029 | return; | |
1030 | ||
1031 | // The current task had better be in a quiescent state. | |
1032 | if (t == current) { | |
1033 | t->trc_reader_checked = true; | |
bdb0cca0 | 1034 | WARN_ON_ONCE(READ_ONCE(t->trc_reader_nesting)); |
d5f177d3 PM |
1035 | return; |
1036 | } | |
1037 | ||
1038 | // Attempt to nail down the task for inspection. | |
1039 | get_task_struct(t); | |
9b3c4ab3 | 1040 | if (!task_call_func(t, trc_inspect_reader, NULL)) { |
d5f177d3 PM |
1041 | put_task_struct(t); |
1042 | return; | |
1043 | } | |
1044 | put_task_struct(t); | |
1045 | ||
45f4b4a2 PM |
1046 | // If this task is not yet on the holdout list, then we are in |
1047 | // an RCU read-side critical section. Otherwise, the invocation of | |
d0a85858 | 1048 | // trc_add_holdout() that added it to the list did the necessary |
45f4b4a2 PM |
1049 | // get_task_struct(). Either way, the task cannot be freed out |
1050 | // from under this code. | |
1051 | ||
d5f177d3 PM |
1052 | // If currently running, send an IPI, either way, add to list. |
1053 | trc_add_holdout(t, bhp); | |
574de876 PM |
1054 | if (task_curr(t) && |
1055 | time_after(jiffies + 1, rcu_tasks_trace.gp_start + rcu_task_ipi_delay)) { | |
d5f177d3 PM |
1056 | // The task is currently running, so try IPIing it. |
1057 | cpu = task_cpu(t); | |
1058 | ||
1059 | // If there is already an IPI outstanding, let it happen. | |
1060 | if (per_cpu(trc_ipi_to_cpu, cpu) || t->trc_ipi_to_cpu >= 0) | |
1061 | return; | |
1062 | ||
d5f177d3 PM |
1063 | per_cpu(trc_ipi_to_cpu, cpu) = true; |
1064 | t->trc_ipi_to_cpu = cpu; | |
238dbce3 | 1065 | rcu_tasks_trace.n_ipis++; |
96017bf9 | 1066 | if (smp_call_function_single(cpu, trc_read_check_handler, t, 0)) { |
d5f177d3 PM |
1067 | // Just in case there is some other reason for |
1068 | // failure than the target CPU being offline. | |
46aa886c NU |
1069 | WARN_ONCE(1, "%s(): smp_call_function_single() failed for CPU: %d\n", |
1070 | __func__, cpu); | |
7e0669c3 | 1071 | rcu_tasks_trace.n_ipis_fails++; |
d5f177d3 | 1072 | per_cpu(trc_ipi_to_cpu, cpu) = false; |
46aa886c | 1073 | t->trc_ipi_to_cpu = -1; |
d5f177d3 PM |
1074 | } |
1075 | } | |
1076 | } | |
1077 | ||
1078 | /* Initialize for a new RCU-tasks-trace grace period. */ | |
1079 | static void rcu_tasks_trace_pregp_step(void) | |
1080 | { | |
1081 | int cpu; | |
1082 | ||
d5f177d3 PM |
1083 | // Allow for fast-acting IPIs. |
1084 | atomic_set(&trc_n_readers_need_end, 1); | |
1085 | ||
1086 | // There shouldn't be any old IPIs, but... | |
1087 | for_each_possible_cpu(cpu) | |
1088 | WARN_ON_ONCE(per_cpu(trc_ipi_to_cpu, cpu)); | |
81b4a7bc PM |
1089 | |
1090 | // Disable CPU hotplug across the tasklist scan. | |
1091 | // This also waits for all readers in CPU-hotplug code paths. | |
1092 | cpus_read_lock(); | |
d5f177d3 PM |
1093 | } |
1094 | ||
1095 | /* Do first-round processing for the specified task. */ | |
1096 | static void rcu_tasks_trace_pertask(struct task_struct *t, | |
1097 | struct list_head *hop) | |
1098 | { | |
1b04fa99 URS |
1099 | // During early boot when there is only the one boot CPU, there |
1100 | // is no idle task for the other CPUs. Just return. | |
1101 | if (unlikely(t == NULL)) | |
1102 | return; | |
1103 | ||
276c4104 | 1104 | WRITE_ONCE(t->trc_reader_special.b.need_qs, false); |
43766c3e | 1105 | WRITE_ONCE(t->trc_reader_checked, false); |
d5f177d3 PM |
1106 | t->trc_ipi_to_cpu = -1; |
1107 | trc_wait_for_one_reader(t, hop); | |
1108 | } | |
1109 | ||
9796e1ae PM |
1110 | /* |
1111 | * Do intermediate processing between task and holdout scans and | |
1112 | * pick up the idle tasks. | |
1113 | */ | |
1114 | static void rcu_tasks_trace_postscan(struct list_head *hop) | |
d5f177d3 | 1115 | { |
9796e1ae PM |
1116 | int cpu; |
1117 | ||
1118 | for_each_possible_cpu(cpu) | |
1119 | rcu_tasks_trace_pertask(idle_task(cpu), hop); | |
1120 | ||
81b4a7bc PM |
1121 | // Re-enable CPU hotplug now that the tasklist scan has completed. |
1122 | cpus_read_unlock(); | |
1123 | ||
d5f177d3 PM |
1124 | // Wait for late-stage exiting tasks to finish exiting. |
1125 | // These might have passed the call to exit_tasks_rcu_finish(). | |
1126 | synchronize_rcu(); | |
1127 | // Any tasks that exit after this point will set ->trc_reader_checked. | |
1128 | } | |
1129 | ||
4593e772 PM |
1130 | /* Show the state of a task stalling the current RCU tasks trace GP. */ |
1131 | static void show_stalled_task_trace(struct task_struct *t, bool *firstreport) | |
1132 | { | |
1133 | int cpu; | |
1134 | ||
1135 | if (*firstreport) { | |
1136 | pr_err("INFO: rcu_tasks_trace detected stalls on tasks:\n"); | |
1137 | *firstreport = false; | |
1138 | } | |
1139 | // FIXME: This should attempt to use try_invoke_on_nonrunning_task(). | |
1140 | cpu = task_cpu(t); | |
1141 | pr_alert("P%d: %c%c%c nesting: %d%c cpu: %d\n", | |
1142 | t->pid, | |
d39ec8f3 | 1143 | ".I"[READ_ONCE(t->trc_ipi_to_cpu) >= 0], |
4593e772 | 1144 | ".i"[is_idle_task(t)], |
d39ec8f3 | 1145 | ".N"[cpu >= 0 && tick_nohz_full_cpu(cpu)], |
bdb0cca0 | 1146 | READ_ONCE(t->trc_reader_nesting), |
f8ab3fad | 1147 | " N"[!!READ_ONCE(t->trc_reader_special.b.need_qs)], |
4593e772 PM |
1148 | cpu); |
1149 | sched_show_task(t); | |
1150 | } | |
1151 | ||
1152 | /* List stalled IPIs for RCU tasks trace. */ | |
1153 | static void show_stalled_ipi_trace(void) | |
1154 | { | |
1155 | int cpu; | |
1156 | ||
1157 | for_each_possible_cpu(cpu) | |
1158 | if (per_cpu(trc_ipi_to_cpu, cpu)) | |
1159 | pr_alert("\tIPI outstanding to CPU %d\n", cpu); | |
1160 | } | |
1161 | ||
d5f177d3 PM |
1162 | /* Do one scan of the holdout list. */ |
1163 | static void check_all_holdout_tasks_trace(struct list_head *hop, | |
4593e772 | 1164 | bool needreport, bool *firstreport) |
d5f177d3 PM |
1165 | { |
1166 | struct task_struct *g, *t; | |
1167 | ||
81b4a7bc PM |
1168 | // Disable CPU hotplug across the holdout list scan. |
1169 | cpus_read_lock(); | |
1170 | ||
d5f177d3 PM |
1171 | list_for_each_entry_safe(t, g, hop, trc_holdout_list) { |
1172 | // If safe and needed, try to check the current task. | |
1173 | if (READ_ONCE(t->trc_ipi_to_cpu) == -1 && | |
1174 | !READ_ONCE(t->trc_reader_checked)) | |
1175 | trc_wait_for_one_reader(t, hop); | |
1176 | ||
1177 | // If check succeeded, remove this task from the list. | |
f5dbc594 PM |
1178 | if (smp_load_acquire(&t->trc_ipi_to_cpu) == -1 && |
1179 | READ_ONCE(t->trc_reader_checked)) | |
d5f177d3 | 1180 | trc_del_holdout(t); |
4593e772 PM |
1181 | else if (needreport) |
1182 | show_stalled_task_trace(t, firstreport); | |
1183 | } | |
81b4a7bc PM |
1184 | |
1185 | // Re-enable CPU hotplug now that the holdout list scan has completed. | |
1186 | cpus_read_unlock(); | |
1187 | ||
4593e772 | 1188 | if (needreport) { |
89401176 | 1189 | if (*firstreport) |
4593e772 PM |
1190 | pr_err("INFO: rcu_tasks_trace detected stalls? (Late IPI?)\n"); |
1191 | show_stalled_ipi_trace(); | |
d5f177d3 PM |
1192 | } |
1193 | } | |
1194 | ||
cbe0d8d9 PM |
1195 | static void rcu_tasks_trace_empty_fn(void *unused) |
1196 | { | |
1197 | } | |
1198 | ||
d5f177d3 | 1199 | /* Wait for grace period to complete and provide ordering. */ |
af051ca4 | 1200 | static void rcu_tasks_trace_postgp(struct rcu_tasks *rtp) |
d5f177d3 | 1201 | { |
cbe0d8d9 | 1202 | int cpu; |
4593e772 PM |
1203 | bool firstreport; |
1204 | struct task_struct *g, *t; | |
1205 | LIST_HEAD(holdouts); | |
1206 | long ret; | |
1207 | ||
cbe0d8d9 PM |
1208 | // Wait for any lingering IPI handlers to complete. Note that |
1209 | // if a CPU has gone offline or transitioned to userspace in the | |
1210 | // meantime, all IPI handlers should have been drained beforehand. | |
1211 | // Yes, this assumes that CPUs process IPIs in order. If that ever | |
1212 | // changes, there will need to be a recheck and/or timed wait. | |
1213 | for_each_online_cpu(cpu) | |
f5dbc594 | 1214 | if (WARN_ON_ONCE(smp_load_acquire(per_cpu_ptr(&trc_ipi_to_cpu, cpu)))) |
cbe0d8d9 PM |
1215 | smp_call_function_single(cpu, rcu_tasks_trace_empty_fn, NULL, 1); |
1216 | ||
d5f177d3 PM |
1217 | // Remove the safety count. |
1218 | smp_mb__before_atomic(); // Order vs. earlier atomics | |
1219 | atomic_dec(&trc_n_readers_need_end); | |
1220 | smp_mb__after_atomic(); // Order vs. later atomics | |
1221 | ||
1222 | // Wait for readers. | |
af051ca4 | 1223 | set_tasks_gp_state(rtp, RTGS_WAIT_READERS); |
4593e772 PM |
1224 | for (;;) { |
1225 | ret = wait_event_idle_exclusive_timeout( | |
1226 | trc_wait, | |
1227 | atomic_read(&trc_n_readers_need_end) == 0, | |
1228 | READ_ONCE(rcu_task_stall_timeout)); | |
1229 | if (ret) | |
1230 | break; // Count reached zero. | |
af051ca4 | 1231 | // Stall warning time, so make a list of the offenders. |
f747c7e1 | 1232 | rcu_read_lock(); |
4593e772 | 1233 | for_each_process_thread(g, t) |
276c4104 | 1234 | if (READ_ONCE(t->trc_reader_special.b.need_qs)) |
4593e772 | 1235 | trc_add_holdout(t, &holdouts); |
f747c7e1 | 1236 | rcu_read_unlock(); |
4593e772 | 1237 | firstreport = true; |
592031cc PM |
1238 | list_for_each_entry_safe(t, g, &holdouts, trc_holdout_list) { |
1239 | if (READ_ONCE(t->trc_reader_special.b.need_qs)) | |
4593e772 | 1240 | show_stalled_task_trace(t, &firstreport); |
592031cc PM |
1241 | trc_del_holdout(t); // Release task_struct reference. |
1242 | } | |
4593e772 PM |
1243 | if (firstreport) |
1244 | pr_err("INFO: rcu_tasks_trace detected stalls? (Counter/taskslist mismatch?)\n"); | |
1245 | show_stalled_ipi_trace(); | |
1246 | pr_err("\t%d holdouts\n", atomic_read(&trc_n_readers_need_end)); | |
1247 | } | |
d5f177d3 | 1248 | smp_mb(); // Caller's code must be ordered after wakeup. |
43766c3e | 1249 | // Pairs with pretty much every ordering primitive. |
d5f177d3 PM |
1250 | } |
1251 | ||
1252 | /* Report any needed quiescent state for this exiting task. */ | |
25246fc8 | 1253 | static void exit_tasks_rcu_finish_trace(struct task_struct *t) |
d5f177d3 PM |
1254 | { |
1255 | WRITE_ONCE(t->trc_reader_checked, true); | |
bdb0cca0 | 1256 | WARN_ON_ONCE(READ_ONCE(t->trc_reader_nesting)); |
d5f177d3 | 1257 | WRITE_ONCE(t->trc_reader_nesting, 0); |
276c4104 | 1258 | if (WARN_ON_ONCE(READ_ONCE(t->trc_reader_special.b.need_qs))) |
a5c071cc | 1259 | rcu_read_unlock_trace_special(t); |
d5f177d3 PM |
1260 | } |
1261 | ||
d5f177d3 PM |
1262 | /** |
1263 | * call_rcu_tasks_trace() - Queue a callback trace task-based grace period | |
1264 | * @rhp: structure to be used for queueing the RCU updates. | |
1265 | * @func: actual callback function to be invoked after the grace period | |
1266 | * | |
ed42c380 NU |
1267 | * The callback function will be invoked some time after a trace rcu-tasks |
1268 | * grace period elapses, in other words after all currently executing | |
1269 | * trace rcu-tasks read-side critical sections have completed. These | |
1270 | * read-side critical sections are delimited by calls to rcu_read_lock_trace() | |
1271 | * and rcu_read_unlock_trace(). | |
d5f177d3 PM |
1272 | * |
1273 | * See the description of call_rcu() for more detailed information on | |
1274 | * memory ordering guarantees. | |
1275 | */ | |
1276 | void call_rcu_tasks_trace(struct rcu_head *rhp, rcu_callback_t func) | |
1277 | { | |
1278 | call_rcu_tasks_generic(rhp, func, &rcu_tasks_trace); | |
1279 | } | |
1280 | EXPORT_SYMBOL_GPL(call_rcu_tasks_trace); | |
1281 | ||
1282 | /** | |
1283 | * synchronize_rcu_tasks_trace - wait for a trace rcu-tasks grace period | |
1284 | * | |
1285 | * Control will return to the caller some time after a trace rcu-tasks | |
c7dcf810 | 1286 | * grace period has elapsed, in other words after all currently executing |
ed42c380 | 1287 | * trace rcu-tasks read-side critical sections have elapsed. These read-side |
c7dcf810 PM |
1288 | * critical sections are delimited by calls to rcu_read_lock_trace() |
1289 | * and rcu_read_unlock_trace(). | |
d5f177d3 PM |
1290 | * |
1291 | * This is a very specialized primitive, intended only for a few uses in | |
1292 | * tracing and other situations requiring manipulation of function preambles | |
1293 | * and profiling hooks. The synchronize_rcu_tasks_trace() function is not | |
1294 | * (yet) intended for heavy use from multiple CPUs. | |
1295 | * | |
1296 | * See the description of synchronize_rcu() for more detailed information | |
1297 | * on memory ordering guarantees. | |
1298 | */ | |
1299 | void synchronize_rcu_tasks_trace(void) | |
1300 | { | |
1301 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_trace_lock_map), "Illegal synchronize_rcu_tasks_trace() in RCU Tasks Trace read-side critical section"); | |
1302 | synchronize_rcu_tasks_generic(&rcu_tasks_trace); | |
1303 | } | |
1304 | EXPORT_SYMBOL_GPL(synchronize_rcu_tasks_trace); | |
1305 | ||
1306 | /** | |
1307 | * rcu_barrier_tasks_trace - Wait for in-flight call_rcu_tasks_trace() callbacks. | |
1308 | * | |
1309 | * Although the current implementation is guaranteed to wait, it is not | |
1310 | * obligated to, for example, if there are no pending callbacks. | |
1311 | */ | |
1312 | void rcu_barrier_tasks_trace(void) | |
1313 | { | |
1314 | /* There is only one callback queue, so this is easy. ;-) */ | |
1315 | synchronize_rcu_tasks_trace(); | |
1316 | } | |
1317 | EXPORT_SYMBOL_GPL(rcu_barrier_tasks_trace); | |
1318 | ||
1319 | static int __init rcu_spawn_tasks_trace_kthread(void) | |
1320 | { | |
cafafd67 | 1321 | cblist_init_generic(&rcu_tasks_trace); |
2393a613 | 1322 | if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB)) { |
4fe192df | 1323 | rcu_tasks_trace.gp_sleep = HZ / 10; |
75dc2da5 | 1324 | rcu_tasks_trace.init_fract = HZ / 10; |
2393a613 | 1325 | } else { |
4fe192df PM |
1326 | rcu_tasks_trace.gp_sleep = HZ / 200; |
1327 | if (rcu_tasks_trace.gp_sleep <= 0) | |
1328 | rcu_tasks_trace.gp_sleep = 1; | |
75dc2da5 | 1329 | rcu_tasks_trace.init_fract = HZ / 200; |
2393a613 PM |
1330 | if (rcu_tasks_trace.init_fract <= 0) |
1331 | rcu_tasks_trace.init_fract = 1; | |
1332 | } | |
d5f177d3 PM |
1333 | rcu_tasks_trace.pregp_func = rcu_tasks_trace_pregp_step; |
1334 | rcu_tasks_trace.pertask_func = rcu_tasks_trace_pertask; | |
1335 | rcu_tasks_trace.postscan_func = rcu_tasks_trace_postscan; | |
1336 | rcu_tasks_trace.holdouts_func = check_all_holdout_tasks_trace; | |
1337 | rcu_tasks_trace.postgp_func = rcu_tasks_trace_postgp; | |
1338 | rcu_spawn_tasks_kthread_generic(&rcu_tasks_trace); | |
1339 | return 0; | |
1340 | } | |
d5f177d3 | 1341 | |
27c0f144 PM |
1342 | #if !defined(CONFIG_TINY_RCU) |
1343 | void show_rcu_tasks_trace_gp_kthread(void) | |
e21408ce | 1344 | { |
40471509 | 1345 | char buf[64]; |
e21408ce | 1346 | |
edf3775f PM |
1347 | sprintf(buf, "N%d h:%lu/%lu/%lu", atomic_read(&trc_n_readers_need_end), |
1348 | data_race(n_heavy_reader_ofl_updates), | |
40471509 PM |
1349 | data_race(n_heavy_reader_updates), |
1350 | data_race(n_heavy_reader_attempts)); | |
e21408ce PM |
1351 | show_rcu_tasks_generic_gp_kthread(&rcu_tasks_trace, buf); |
1352 | } | |
27c0f144 PM |
1353 | EXPORT_SYMBOL_GPL(show_rcu_tasks_trace_gp_kthread); |
1354 | #endif // !defined(CONFIG_TINY_RCU) | |
e21408ce | 1355 | |
d5f177d3 | 1356 | #else /* #ifdef CONFIG_TASKS_TRACE_RCU */ |
25246fc8 | 1357 | static void exit_tasks_rcu_finish_trace(struct task_struct *t) { } |
d5f177d3 | 1358 | #endif /* #else #ifdef CONFIG_TASKS_TRACE_RCU */ |
8fd8ca38 | 1359 | |
8344496e | 1360 | #ifndef CONFIG_TINY_RCU |
e21408ce PM |
1361 | void show_rcu_tasks_gp_kthreads(void) |
1362 | { | |
1363 | show_rcu_tasks_classic_gp_kthread(); | |
1364 | show_rcu_tasks_rude_gp_kthread(); | |
1365 | show_rcu_tasks_trace_gp_kthread(); | |
1366 | } | |
8344496e | 1367 | #endif /* #ifndef CONFIG_TINY_RCU */ |
e21408ce | 1368 | |
bfba7ed0 URS |
1369 | #ifdef CONFIG_PROVE_RCU |
1370 | struct rcu_tasks_test_desc { | |
1371 | struct rcu_head rh; | |
1372 | const char *name; | |
1373 | bool notrun; | |
1374 | }; | |
1375 | ||
1376 | static struct rcu_tasks_test_desc tests[] = { | |
1377 | { | |
1378 | .name = "call_rcu_tasks()", | |
1379 | /* If not defined, the test is skipped. */ | |
1380 | .notrun = !IS_ENABLED(CONFIG_TASKS_RCU), | |
1381 | }, | |
1382 | { | |
1383 | .name = "call_rcu_tasks_rude()", | |
1384 | /* If not defined, the test is skipped. */ | |
1385 | .notrun = !IS_ENABLED(CONFIG_TASKS_RUDE_RCU), | |
1386 | }, | |
1387 | { | |
1388 | .name = "call_rcu_tasks_trace()", | |
1389 | /* If not defined, the test is skipped. */ | |
1390 | .notrun = !IS_ENABLED(CONFIG_TASKS_TRACE_RCU) | |
1391 | } | |
1392 | }; | |
1393 | ||
1394 | static void test_rcu_tasks_callback(struct rcu_head *rhp) | |
1395 | { | |
1396 | struct rcu_tasks_test_desc *rttd = | |
1397 | container_of(rhp, struct rcu_tasks_test_desc, rh); | |
1398 | ||
1399 | pr_info("Callback from %s invoked.\n", rttd->name); | |
1400 | ||
1401 | rttd->notrun = true; | |
1402 | } | |
1403 | ||
1404 | static void rcu_tasks_initiate_self_tests(void) | |
1405 | { | |
1406 | pr_info("Running RCU-tasks wait API self tests\n"); | |
1407 | #ifdef CONFIG_TASKS_RCU | |
1408 | synchronize_rcu_tasks(); | |
1409 | call_rcu_tasks(&tests[0].rh, test_rcu_tasks_callback); | |
1410 | #endif | |
1411 | ||
1412 | #ifdef CONFIG_TASKS_RUDE_RCU | |
1413 | synchronize_rcu_tasks_rude(); | |
1414 | call_rcu_tasks_rude(&tests[1].rh, test_rcu_tasks_callback); | |
1415 | #endif | |
1416 | ||
1417 | #ifdef CONFIG_TASKS_TRACE_RCU | |
1418 | synchronize_rcu_tasks_trace(); | |
1419 | call_rcu_tasks_trace(&tests[2].rh, test_rcu_tasks_callback); | |
1420 | #endif | |
1421 | } | |
1422 | ||
1423 | static int rcu_tasks_verify_self_tests(void) | |
1424 | { | |
1425 | int ret = 0; | |
1426 | int i; | |
1427 | ||
1428 | for (i = 0; i < ARRAY_SIZE(tests); i++) { | |
1429 | if (!tests[i].notrun) { // still hanging. | |
1430 | pr_err("%s has been failed.\n", tests[i].name); | |
1431 | ret = -1; | |
1432 | } | |
1433 | } | |
1434 | ||
1435 | if (ret) | |
1436 | WARN_ON(1); | |
1437 | ||
1438 | return ret; | |
1439 | } | |
1440 | late_initcall(rcu_tasks_verify_self_tests); | |
1441 | #else /* #ifdef CONFIG_PROVE_RCU */ | |
1442 | static void rcu_tasks_initiate_self_tests(void) { } | |
1443 | #endif /* #else #ifdef CONFIG_PROVE_RCU */ | |
1444 | ||
1b04fa99 URS |
1445 | void __init rcu_init_tasks_generic(void) |
1446 | { | |
1447 | #ifdef CONFIG_TASKS_RCU | |
1448 | rcu_spawn_tasks_kthread(); | |
1449 | #endif | |
1450 | ||
1451 | #ifdef CONFIG_TASKS_RUDE_RCU | |
1452 | rcu_spawn_tasks_rude_kthread(); | |
1453 | #endif | |
1454 | ||
1455 | #ifdef CONFIG_TASKS_TRACE_RCU | |
1456 | rcu_spawn_tasks_trace_kthread(); | |
1457 | #endif | |
bfba7ed0 URS |
1458 | |
1459 | // Run the self-tests. | |
1460 | rcu_tasks_initiate_self_tests(); | |
1b04fa99 URS |
1461 | } |
1462 | ||
8fd8ca38 PM |
1463 | #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */ |
1464 | static inline void rcu_tasks_bootup_oddness(void) {} | |
1465 | #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */ |