Loading...
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * Read-Copy Update mechanism for mutual exclusion, the Bloatwatch edition.
4 *
5 * Copyright IBM Corporation, 2008
6 *
7 * Author: Paul E. McKenney <paulmck@linux.ibm.com>
8 *
9 * For detailed explanation of Read-Copy Update mechanism see -
10 * Documentation/RCU
11 */
12#include <linux/completion.h>
13#include <linux/interrupt.h>
14#include <linux/notifier.h>
15#include <linux/rcupdate_wait.h>
16#include <linux/kernel.h>
17#include <linux/export.h>
18#include <linux/mutex.h>
19#include <linux/sched.h>
20#include <linux/types.h>
21#include <linux/init.h>
22#include <linux/time.h>
23#include <linux/cpu.h>
24#include <linux/prefetch.h>
25#include <linux/slab.h>
26#include <linux/mm.h>
27
28#include "rcu.h"
29
30/* Global control variables for rcupdate callback mechanism. */
31struct rcu_ctrlblk {
32 struct rcu_head *rcucblist; /* List of pending callbacks (CBs). */
33 struct rcu_head **donetail; /* ->next pointer of last "done" CB. */
34 struct rcu_head **curtail; /* ->next pointer of last CB. */
35 unsigned long gp_seq; /* Grace-period counter. */
36};
37
38/* Definition for rcupdate control block. */
39static struct rcu_ctrlblk rcu_ctrlblk = {
40 .donetail = &rcu_ctrlblk.rcucblist,
41 .curtail = &rcu_ctrlblk.rcucblist,
42 .gp_seq = 0 - 300UL,
43};
44
45void rcu_barrier(void)
46{
47 wait_rcu_gp(call_rcu_hurry);
48}
49EXPORT_SYMBOL(rcu_barrier);
50
51/* Record an rcu quiescent state. */
52void rcu_qs(void)
53{
54 unsigned long flags;
55
56 local_irq_save(flags);
57 if (rcu_ctrlblk.donetail != rcu_ctrlblk.curtail) {
58 rcu_ctrlblk.donetail = rcu_ctrlblk.curtail;
59 raise_softirq_irqoff(RCU_SOFTIRQ);
60 }
61 WRITE_ONCE(rcu_ctrlblk.gp_seq, rcu_ctrlblk.gp_seq + 2);
62 local_irq_restore(flags);
63}
64
65/*
66 * Check to see if the scheduling-clock interrupt came from an extended
67 * quiescent state, and, if so, tell RCU about it. This function must
68 * be called from hardirq context. It is normally called from the
69 * scheduling-clock interrupt.
70 */
71void rcu_sched_clock_irq(int user)
72{
73 if (user) {
74 rcu_qs();
75 } else if (rcu_ctrlblk.donetail != rcu_ctrlblk.curtail) {
76 set_tsk_need_resched(current);
77 set_preempt_need_resched();
78 }
79}
80
81/*
82 * Reclaim the specified callback, either by invoking it for non-kfree cases or
83 * freeing it directly (for kfree). Return true if kfreeing, false otherwise.
84 */
85static inline bool rcu_reclaim_tiny(struct rcu_head *head)
86{
87 rcu_callback_t f;
88 unsigned long offset = (unsigned long)head->func;
89
90 rcu_lock_acquire(&rcu_callback_map);
91 if (__is_kvfree_rcu_offset(offset)) {
92 trace_rcu_invoke_kvfree_callback("", head, offset);
93 kvfree((void *)head - offset);
94 rcu_lock_release(&rcu_callback_map);
95 return true;
96 }
97
98 trace_rcu_invoke_callback("", head);
99 f = head->func;
100 WRITE_ONCE(head->func, (rcu_callback_t)0L);
101 f(head);
102 rcu_lock_release(&rcu_callback_map);
103 return false;
104}
105
106/* Invoke the RCU callbacks whose grace period has elapsed. */
107static __latent_entropy void rcu_process_callbacks(struct softirq_action *unused)
108{
109 struct rcu_head *next, *list;
110 unsigned long flags;
111
112 /* Move the ready-to-invoke callbacks to a local list. */
113 local_irq_save(flags);
114 if (rcu_ctrlblk.donetail == &rcu_ctrlblk.rcucblist) {
115 /* No callbacks ready, so just leave. */
116 local_irq_restore(flags);
117 return;
118 }
119 list = rcu_ctrlblk.rcucblist;
120 rcu_ctrlblk.rcucblist = *rcu_ctrlblk.donetail;
121 *rcu_ctrlblk.donetail = NULL;
122 if (rcu_ctrlblk.curtail == rcu_ctrlblk.donetail)
123 rcu_ctrlblk.curtail = &rcu_ctrlblk.rcucblist;
124 rcu_ctrlblk.donetail = &rcu_ctrlblk.rcucblist;
125 local_irq_restore(flags);
126
127 /* Invoke the callbacks on the local list. */
128 while (list) {
129 next = list->next;
130 prefetch(next);
131 debug_rcu_head_unqueue(list);
132 local_bh_disable();
133 rcu_reclaim_tiny(list);
134 local_bh_enable();
135 list = next;
136 }
137}
138
139/*
140 * Wait for a grace period to elapse. But it is illegal to invoke
141 * synchronize_rcu() from within an RCU read-side critical section.
142 * Therefore, any legal call to synchronize_rcu() is a quiescent state,
143 * and so on a UP system, synchronize_rcu() need do nothing, other than
144 * let the polled APIs know that another grace period elapsed.
145 *
146 * (But Lai Jiangshan points out the benefits of doing might_sleep()
147 * to reduce latency.)
148 *
149 * Cool, huh? (Due to Josh Triplett.)
150 */
151void synchronize_rcu(void)
152{
153 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) ||
154 lock_is_held(&rcu_lock_map) ||
155 lock_is_held(&rcu_sched_lock_map),
156 "Illegal synchronize_rcu() in RCU read-side critical section");
157 WRITE_ONCE(rcu_ctrlblk.gp_seq, rcu_ctrlblk.gp_seq + 2);
158}
159EXPORT_SYMBOL_GPL(synchronize_rcu);
160
161static void tiny_rcu_leak_callback(struct rcu_head *rhp)
162{
163}
164
165/*
166 * Post an RCU callback to be invoked after the end of an RCU grace
167 * period. But since we have but one CPU, that would be after any
168 * quiescent state.
169 */
170void call_rcu(struct rcu_head *head, rcu_callback_t func)
171{
172 static atomic_t doublefrees;
173 unsigned long flags;
174
175 if (debug_rcu_head_queue(head)) {
176 if (atomic_inc_return(&doublefrees) < 4) {
177 pr_err("%s(): Double-freed CB %p->%pS()!!! ", __func__, head, head->func);
178 mem_dump_obj(head);
179 }
180
181 if (!__is_kvfree_rcu_offset((unsigned long)head->func))
182 WRITE_ONCE(head->func, tiny_rcu_leak_callback);
183 return;
184 }
185
186 head->func = func;
187 head->next = NULL;
188
189 local_irq_save(flags);
190 *rcu_ctrlblk.curtail = head;
191 rcu_ctrlblk.curtail = &head->next;
192 local_irq_restore(flags);
193
194 if (unlikely(is_idle_task(current))) {
195 /* force scheduling for rcu_qs() */
196 resched_cpu(0);
197 }
198}
199EXPORT_SYMBOL_GPL(call_rcu);
200
201/*
202 * Store a grace-period-counter "cookie". For more information,
203 * see the Tree RCU header comment.
204 */
205void get_completed_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp)
206{
207 rgosp->rgos_norm = RCU_GET_STATE_COMPLETED;
208}
209EXPORT_SYMBOL_GPL(get_completed_synchronize_rcu_full);
210
211/*
212 * Return a grace-period-counter "cookie". For more information,
213 * see the Tree RCU header comment.
214 */
215unsigned long get_state_synchronize_rcu(void)
216{
217 return READ_ONCE(rcu_ctrlblk.gp_seq);
218}
219EXPORT_SYMBOL_GPL(get_state_synchronize_rcu);
220
221/*
222 * Return a grace-period-counter "cookie" and ensure that a future grace
223 * period completes. For more information, see the Tree RCU header comment.
224 */
225unsigned long start_poll_synchronize_rcu(void)
226{
227 unsigned long gp_seq = get_state_synchronize_rcu();
228
229 if (unlikely(is_idle_task(current))) {
230 /* force scheduling for rcu_qs() */
231 resched_cpu(0);
232 }
233 return gp_seq;
234}
235EXPORT_SYMBOL_GPL(start_poll_synchronize_rcu);
236
237/*
238 * Return true if the grace period corresponding to oldstate has completed
239 * and false otherwise. For more information, see the Tree RCU header
240 * comment.
241 */
242bool poll_state_synchronize_rcu(unsigned long oldstate)
243{
244 return oldstate == RCU_GET_STATE_COMPLETED || READ_ONCE(rcu_ctrlblk.gp_seq) != oldstate;
245}
246EXPORT_SYMBOL_GPL(poll_state_synchronize_rcu);
247
248#ifdef CONFIG_KASAN_GENERIC
249void kvfree_call_rcu(struct rcu_head *head, rcu_callback_t func)
250{
251 if (head) {
252 void *ptr = (void *) head - (unsigned long) func;
253
254 kasan_record_aux_stack_noalloc(ptr);
255 }
256
257 __kvfree_call_rcu(head, func);
258}
259EXPORT_SYMBOL_GPL(kvfree_call_rcu);
260#endif
261
262void __init rcu_init(void)
263{
264 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
265 rcu_early_boot_tests();
266}
1/*
2 * Read-Copy Update mechanism for mutual exclusion, the Bloatwatch edition.
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, you can access it online at
16 * http://www.gnu.org/licenses/gpl-2.0.html.
17 *
18 * Copyright IBM Corporation, 2008
19 *
20 * Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
21 *
22 * For detailed explanation of Read-Copy Update mechanism see -
23 * Documentation/RCU
24 */
25#include <linux/completion.h>
26#include <linux/interrupt.h>
27#include <linux/notifier.h>
28#include <linux/rcupdate.h>
29#include <linux/kernel.h>
30#include <linux/export.h>
31#include <linux/mutex.h>
32#include <linux/sched.h>
33#include <linux/types.h>
34#include <linux/init.h>
35#include <linux/time.h>
36#include <linux/cpu.h>
37#include <linux/prefetch.h>
38#include <linux/ftrace_event.h>
39
40#include "rcu.h"
41
42/* Forward declarations for tiny_plugin.h. */
43struct rcu_ctrlblk;
44static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp);
45static void rcu_process_callbacks(struct softirq_action *unused);
46static void __call_rcu(struct rcu_head *head,
47 void (*func)(struct rcu_head *rcu),
48 struct rcu_ctrlblk *rcp);
49
50static long long rcu_dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
51
52#include "tiny_plugin.h"
53
54/* Common code for rcu_idle_enter() and rcu_irq_exit(), see kernel/rcutree.c. */
55static void rcu_idle_enter_common(long long newval)
56{
57 if (newval) {
58 RCU_TRACE(trace_rcu_dyntick(TPS("--="),
59 rcu_dynticks_nesting, newval));
60 rcu_dynticks_nesting = newval;
61 return;
62 }
63 RCU_TRACE(trace_rcu_dyntick(TPS("Start"),
64 rcu_dynticks_nesting, newval));
65 if (!is_idle_task(current)) {
66 struct task_struct *idle __maybe_unused = idle_task(smp_processor_id());
67
68 RCU_TRACE(trace_rcu_dyntick(TPS("Entry error: not idle task"),
69 rcu_dynticks_nesting, newval));
70 ftrace_dump(DUMP_ALL);
71 WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s",
72 current->pid, current->comm,
73 idle->pid, idle->comm); /* must be idle task! */
74 }
75 rcu_sched_qs(0); /* implies rcu_bh_qsctr_inc(0) */
76 barrier();
77 rcu_dynticks_nesting = newval;
78}
79
80/*
81 * Enter idle, which is an extended quiescent state if we have fully
82 * entered that mode (i.e., if the new value of dynticks_nesting is zero).
83 */
84void rcu_idle_enter(void)
85{
86 unsigned long flags;
87 long long newval;
88
89 local_irq_save(flags);
90 WARN_ON_ONCE((rcu_dynticks_nesting & DYNTICK_TASK_NEST_MASK) == 0);
91 if ((rcu_dynticks_nesting & DYNTICK_TASK_NEST_MASK) ==
92 DYNTICK_TASK_NEST_VALUE)
93 newval = 0;
94 else
95 newval = rcu_dynticks_nesting - DYNTICK_TASK_NEST_VALUE;
96 rcu_idle_enter_common(newval);
97 local_irq_restore(flags);
98}
99EXPORT_SYMBOL_GPL(rcu_idle_enter);
100
101/*
102 * Exit an interrupt handler towards idle.
103 */
104void rcu_irq_exit(void)
105{
106 unsigned long flags;
107 long long newval;
108
109 local_irq_save(flags);
110 newval = rcu_dynticks_nesting - 1;
111 WARN_ON_ONCE(newval < 0);
112 rcu_idle_enter_common(newval);
113 local_irq_restore(flags);
114}
115EXPORT_SYMBOL_GPL(rcu_irq_exit);
116
117/* Common code for rcu_idle_exit() and rcu_irq_enter(), see kernel/rcutree.c. */
118static void rcu_idle_exit_common(long long oldval)
119{
120 if (oldval) {
121 RCU_TRACE(trace_rcu_dyntick(TPS("++="),
122 oldval, rcu_dynticks_nesting));
123 return;
124 }
125 RCU_TRACE(trace_rcu_dyntick(TPS("End"), oldval, rcu_dynticks_nesting));
126 if (!is_idle_task(current)) {
127 struct task_struct *idle __maybe_unused = idle_task(smp_processor_id());
128
129 RCU_TRACE(trace_rcu_dyntick(TPS("Exit error: not idle task"),
130 oldval, rcu_dynticks_nesting));
131 ftrace_dump(DUMP_ALL);
132 WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s",
133 current->pid, current->comm,
134 idle->pid, idle->comm); /* must be idle task! */
135 }
136}
137
138/*
139 * Exit idle, so that we are no longer in an extended quiescent state.
140 */
141void rcu_idle_exit(void)
142{
143 unsigned long flags;
144 long long oldval;
145
146 local_irq_save(flags);
147 oldval = rcu_dynticks_nesting;
148 WARN_ON_ONCE(rcu_dynticks_nesting < 0);
149 if (rcu_dynticks_nesting & DYNTICK_TASK_NEST_MASK)
150 rcu_dynticks_nesting += DYNTICK_TASK_NEST_VALUE;
151 else
152 rcu_dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
153 rcu_idle_exit_common(oldval);
154 local_irq_restore(flags);
155}
156EXPORT_SYMBOL_GPL(rcu_idle_exit);
157
158/*
159 * Enter an interrupt handler, moving away from idle.
160 */
161void rcu_irq_enter(void)
162{
163 unsigned long flags;
164 long long oldval;
165
166 local_irq_save(flags);
167 oldval = rcu_dynticks_nesting;
168 rcu_dynticks_nesting++;
169 WARN_ON_ONCE(rcu_dynticks_nesting == 0);
170 rcu_idle_exit_common(oldval);
171 local_irq_restore(flags);
172}
173EXPORT_SYMBOL_GPL(rcu_irq_enter);
174
175#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE)
176
177/*
178 * Test whether RCU thinks that the current CPU is idle.
179 */
180bool notrace __rcu_is_watching(void)
181{
182 return rcu_dynticks_nesting;
183}
184EXPORT_SYMBOL(__rcu_is_watching);
185
186#endif /* defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) */
187
188/*
189 * Test whether the current CPU was interrupted from idle. Nested
190 * interrupts don't count, we must be running at the first interrupt
191 * level.
192 */
193static int rcu_is_cpu_rrupt_from_idle(void)
194{
195 return rcu_dynticks_nesting <= 1;
196}
197
198/*
199 * Helper function for rcu_sched_qs() and rcu_bh_qs().
200 * Also irqs are disabled to avoid confusion due to interrupt handlers
201 * invoking call_rcu().
202 */
203static int rcu_qsctr_help(struct rcu_ctrlblk *rcp)
204{
205 RCU_TRACE(reset_cpu_stall_ticks(rcp));
206 if (rcp->rcucblist != NULL &&
207 rcp->donetail != rcp->curtail) {
208 rcp->donetail = rcp->curtail;
209 return 1;
210 }
211
212 return 0;
213}
214
215/*
216 * Record an rcu quiescent state. And an rcu_bh quiescent state while we
217 * are at it, given that any rcu quiescent state is also an rcu_bh
218 * quiescent state. Use "+" instead of "||" to defeat short circuiting.
219 */
220void rcu_sched_qs(int cpu)
221{
222 unsigned long flags;
223
224 local_irq_save(flags);
225 if (rcu_qsctr_help(&rcu_sched_ctrlblk) +
226 rcu_qsctr_help(&rcu_bh_ctrlblk))
227 raise_softirq(RCU_SOFTIRQ);
228 local_irq_restore(flags);
229}
230
231/*
232 * Record an rcu_bh quiescent state.
233 */
234void rcu_bh_qs(int cpu)
235{
236 unsigned long flags;
237
238 local_irq_save(flags);
239 if (rcu_qsctr_help(&rcu_bh_ctrlblk))
240 raise_softirq(RCU_SOFTIRQ);
241 local_irq_restore(flags);
242}
243
244/*
245 * Check to see if the scheduling-clock interrupt came from an extended
246 * quiescent state, and, if so, tell RCU about it. This function must
247 * be called from hardirq context. It is normally called from the
248 * scheduling-clock interrupt.
249 */
250void rcu_check_callbacks(int cpu, int user)
251{
252 RCU_TRACE(check_cpu_stalls());
253 if (user || rcu_is_cpu_rrupt_from_idle())
254 rcu_sched_qs(cpu);
255 else if (!in_softirq())
256 rcu_bh_qs(cpu);
257}
258
259/*
260 * Invoke the RCU callbacks on the specified rcu_ctrlkblk structure
261 * whose grace period has elapsed.
262 */
263static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
264{
265 const char *rn = NULL;
266 struct rcu_head *next, *list;
267 unsigned long flags;
268 RCU_TRACE(int cb_count = 0);
269
270 /* If no RCU callbacks ready to invoke, just return. */
271 if (&rcp->rcucblist == rcp->donetail) {
272 RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, 0, -1));
273 RCU_TRACE(trace_rcu_batch_end(rcp->name, 0,
274 !!ACCESS_ONCE(rcp->rcucblist),
275 need_resched(),
276 is_idle_task(current),
277 false));
278 return;
279 }
280
281 /* Move the ready-to-invoke callbacks to a local list. */
282 local_irq_save(flags);
283 RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, rcp->qlen, -1));
284 list = rcp->rcucblist;
285 rcp->rcucblist = *rcp->donetail;
286 *rcp->donetail = NULL;
287 if (rcp->curtail == rcp->donetail)
288 rcp->curtail = &rcp->rcucblist;
289 rcp->donetail = &rcp->rcucblist;
290 local_irq_restore(flags);
291
292 /* Invoke the callbacks on the local list. */
293 RCU_TRACE(rn = rcp->name);
294 while (list) {
295 next = list->next;
296 prefetch(next);
297 debug_rcu_head_unqueue(list);
298 local_bh_disable();
299 __rcu_reclaim(rn, list);
300 local_bh_enable();
301 list = next;
302 RCU_TRACE(cb_count++);
303 }
304 RCU_TRACE(rcu_trace_sub_qlen(rcp, cb_count));
305 RCU_TRACE(trace_rcu_batch_end(rcp->name,
306 cb_count, 0, need_resched(),
307 is_idle_task(current),
308 false));
309}
310
311static void rcu_process_callbacks(struct softirq_action *unused)
312{
313 __rcu_process_callbacks(&rcu_sched_ctrlblk);
314 __rcu_process_callbacks(&rcu_bh_ctrlblk);
315}
316
317/*
318 * Wait for a grace period to elapse. But it is illegal to invoke
319 * synchronize_sched() from within an RCU read-side critical section.
320 * Therefore, any legal call to synchronize_sched() is a quiescent
321 * state, and so on a UP system, synchronize_sched() need do nothing.
322 * Ditto for synchronize_rcu_bh(). (But Lai Jiangshan points out the
323 * benefits of doing might_sleep() to reduce latency.)
324 *
325 * Cool, huh? (Due to Josh Triplett.)
326 *
327 * But we want to make this a static inline later. The cond_resched()
328 * currently makes this problematic.
329 */
330void synchronize_sched(void)
331{
332 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
333 !lock_is_held(&rcu_lock_map) &&
334 !lock_is_held(&rcu_sched_lock_map),
335 "Illegal synchronize_sched() in RCU read-side critical section");
336 cond_resched();
337}
338EXPORT_SYMBOL_GPL(synchronize_sched);
339
340/*
341 * Helper function for call_rcu() and call_rcu_bh().
342 */
343static void __call_rcu(struct rcu_head *head,
344 void (*func)(struct rcu_head *rcu),
345 struct rcu_ctrlblk *rcp)
346{
347 unsigned long flags;
348
349 debug_rcu_head_queue(head);
350 head->func = func;
351 head->next = NULL;
352
353 local_irq_save(flags);
354 *rcp->curtail = head;
355 rcp->curtail = &head->next;
356 RCU_TRACE(rcp->qlen++);
357 local_irq_restore(flags);
358}
359
360/*
361 * Post an RCU callback to be invoked after the end of an RCU-sched grace
362 * period. But since we have but one CPU, that would be after any
363 * quiescent state.
364 */
365void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
366{
367 __call_rcu(head, func, &rcu_sched_ctrlblk);
368}
369EXPORT_SYMBOL_GPL(call_rcu_sched);
370
371/*
372 * Post an RCU bottom-half callback to be invoked after any subsequent
373 * quiescent state.
374 */
375void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
376{
377 __call_rcu(head, func, &rcu_bh_ctrlblk);
378}
379EXPORT_SYMBOL_GPL(call_rcu_bh);
380
381void rcu_init(void)
382{
383 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
384}