Branch data Line data Source code
1 : : // SPDX-License-Identifier: GPL-2.0+
2 : : /*
3 : : * Read-Copy Update mechanism for mutual exclusion
4 : : *
5 : : * Copyright IBM Corporation, 2008
6 : : *
7 : : * Authors: Dipankar Sarma <dipankar@in.ibm.com>
8 : : * Manfred Spraul <manfred@colorfullife.com>
9 : : * Paul E. McKenney <paulmck@linux.ibm.com> Hierarchical version
10 : : *
11 : : * Based on the original work by Paul McKenney <paulmck@linux.ibm.com>
12 : : * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
13 : : *
14 : : * For detailed explanation of Read-Copy Update mechanism see -
15 : : * Documentation/RCU
16 : : */
17 : :
18 : : #define pr_fmt(fmt) "rcu: " fmt
19 : :
20 : : #include <linux/types.h>
21 : : #include <linux/kernel.h>
22 : : #include <linux/init.h>
23 : : #include <linux/spinlock.h>
24 : : #include <linux/smp.h>
25 : : #include <linux/rcupdate_wait.h>
26 : : #include <linux/interrupt.h>
27 : : #include <linux/sched.h>
28 : : #include <linux/sched/debug.h>
29 : : #include <linux/nmi.h>
30 : : #include <linux/atomic.h>
31 : : #include <linux/bitops.h>
32 : : #include <linux/export.h>
33 : : #include <linux/completion.h>
34 : : #include <linux/moduleparam.h>
35 : : #include <linux/percpu.h>
36 : : #include <linux/notifier.h>
37 : : #include <linux/cpu.h>
38 : : #include <linux/mutex.h>
39 : : #include <linux/time.h>
40 : : #include <linux/kernel_stat.h>
41 : : #include <linux/wait.h>
42 : : #include <linux/kthread.h>
43 : : #include <uapi/linux/sched/types.h>
44 : : #include <linux/prefetch.h>
45 : : #include <linux/delay.h>
46 : : #include <linux/random.h>
47 : : #include <linux/trace_events.h>
48 : : #include <linux/suspend.h>
49 : : #include <linux/ftrace.h>
50 : : #include <linux/tick.h>
51 : : #include <linux/sysrq.h>
52 : : #include <linux/kprobes.h>
53 : : #include <linux/gfp.h>
54 : : #include <linux/oom.h>
55 : : #include <linux/smpboot.h>
56 : : #include <linux/jiffies.h>
57 : : #include <linux/slab.h>
58 : : #include <linux/sched/isolation.h>
59 : : #include <linux/sched/clock.h>
60 : : #include "../time/tick-internal.h"
61 : :
62 : : #include "tree.h"
63 : : #include "rcu.h"
64 : :
65 : : #ifdef MODULE_PARAM_PREFIX
66 : : #undef MODULE_PARAM_PREFIX
67 : : #endif
68 : : #define MODULE_PARAM_PREFIX "rcutree."
69 : :
70 : : /* Data structures. */
71 : :
72 : : /*
73 : : * Steal a bit from the bottom of ->dynticks for idle entry/exit
74 : : * control. Initially this is for TLB flushing.
75 : : */
76 : : #define RCU_DYNTICK_CTRL_MASK 0x1
77 : : #define RCU_DYNTICK_CTRL_CTR (RCU_DYNTICK_CTRL_MASK + 1)
78 : : #ifndef rcu_eqs_special_exit
79 : : #define rcu_eqs_special_exit() do { } while (0)
80 : : #endif
81 : :
82 : : static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, rcu_data) = {
83 : : .dynticks_nesting = 1,
84 : : .dynticks_nmi_nesting = DYNTICK_IRQ_NONIDLE,
85 : : .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR),
86 : : };
87 : : static struct rcu_state rcu_state = {
88 : : .level = { &rcu_state.node[0] },
89 : : .gp_state = RCU_GP_IDLE,
90 : : .gp_seq = (0UL - 300UL) << RCU_SEQ_CTR_SHIFT,
91 : : .barrier_mutex = __MUTEX_INITIALIZER(rcu_state.barrier_mutex),
92 : : .name = RCU_NAME,
93 : : .abbr = RCU_ABBR,
94 : : .exp_mutex = __MUTEX_INITIALIZER(rcu_state.exp_mutex),
95 : : .exp_wake_mutex = __MUTEX_INITIALIZER(rcu_state.exp_wake_mutex),
96 : : .ofl_lock = __RAW_SPIN_LOCK_UNLOCKED(rcu_state.ofl_lock),
97 : : };
98 : :
99 : : /* Dump rcu_node combining tree at boot to verify correct setup. */
100 : : static bool dump_tree;
101 : : module_param(dump_tree, bool, 0444);
102 : : /* By default, use RCU_SOFTIRQ instead of rcuc kthreads. */
103 : : static bool use_softirq = 1;
104 : : module_param(use_softirq, bool, 0444);
105 : : /* Control rcu_node-tree auto-balancing at boot time. */
106 : : static bool rcu_fanout_exact;
107 : : module_param(rcu_fanout_exact, bool, 0444);
108 : : /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */
109 : : static int rcu_fanout_leaf = RCU_FANOUT_LEAF;
110 : : module_param(rcu_fanout_leaf, int, 0444);
111 : : int rcu_num_lvls __read_mostly = RCU_NUM_LVLS;
112 : : /* Number of rcu_nodes at specified level. */
113 : : int num_rcu_lvl[] = NUM_RCU_LVL_INIT;
114 : : int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */
115 : :
116 : : /*
117 : : * The rcu_scheduler_active variable is initialized to the value
118 : : * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the
119 : : * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE,
120 : : * RCU can assume that there is but one task, allowing RCU to (for example)
121 : : * optimize synchronize_rcu() to a simple barrier(). When this variable
122 : : * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required
123 : : * to detect real grace periods. This variable is also used to suppress
124 : : * boot-time false positives from lockdep-RCU error checking. Finally, it
125 : : * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU
126 : : * is fully initialized, including all of its kthreads having been spawned.
127 : : */
128 : : int rcu_scheduler_active __read_mostly;
129 : : EXPORT_SYMBOL_GPL(rcu_scheduler_active);
130 : :
131 : : /*
132 : : * The rcu_scheduler_fully_active variable transitions from zero to one
133 : : * during the early_initcall() processing, which is after the scheduler
134 : : * is capable of creating new tasks. So RCU processing (for example,
135 : : * creating tasks for RCU priority boosting) must be delayed until after
136 : : * rcu_scheduler_fully_active transitions from zero to one. We also
137 : : * currently delay invocation of any RCU callbacks until after this point.
138 : : *
139 : : * It might later prove better for people registering RCU callbacks during
140 : : * early boot to take responsibility for these callbacks, but one step at
141 : : * a time.
142 : : */
143 : : static int rcu_scheduler_fully_active __read_mostly;
144 : :
145 : : static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp,
146 : : unsigned long gps, unsigned long flags);
147 : : static void rcu_init_new_rnp(struct rcu_node *rnp_leaf);
148 : : static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf);
149 : : static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
150 : : static void invoke_rcu_core(void);
151 : : static void rcu_report_exp_rdp(struct rcu_data *rdp);
152 : : static void sync_sched_exp_online_cleanup(int cpu);
153 : :
154 : : /* rcuc/rcub kthread realtime priority */
155 : : static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0;
156 : : module_param(kthread_prio, int, 0444);
157 : :
158 : : /* Delay in jiffies for grace-period initialization delays, debug only. */
159 : :
160 : : static int gp_preinit_delay;
161 : : module_param(gp_preinit_delay, int, 0444);
162 : : static int gp_init_delay;
163 : : module_param(gp_init_delay, int, 0444);
164 : : static int gp_cleanup_delay;
165 : : module_param(gp_cleanup_delay, int, 0444);
166 : :
167 : : /* Retrieve RCU kthreads priority for rcutorture */
168 : 0 : int rcu_get_gp_kthreads_prio(void)
169 : : {
170 : 0 : return kthread_prio;
171 : : }
172 : : EXPORT_SYMBOL_GPL(rcu_get_gp_kthreads_prio);
173 : :
174 : : /*
175 : : * Number of grace periods between delays, normalized by the duration of
176 : : * the delay. The longer the delay, the more the grace periods between
177 : : * each delay. The reason for this normalization is that it means that,
178 : : * for non-zero delays, the overall slowdown of grace periods is constant
179 : : * regardless of the duration of the delay. This arrangement balances
180 : : * the need for long delays to increase some race probabilities with the
181 : : * need for fast grace periods to increase other race probabilities.
182 : : */
183 : : #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */
184 : :
185 : : /*
186 : : * Compute the mask of online CPUs for the specified rcu_node structure.
187 : : * This will not be stable unless the rcu_node structure's ->lock is
188 : : * held, but the bit corresponding to the current CPU will be stable
189 : : * in most contexts.
190 : : */
191 : 0 : static unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp)
192 : : {
193 : 0 : return READ_ONCE(rnp->qsmaskinitnext);
194 : : }
195 : :
196 : : /*
197 : : * Return true if an RCU grace period is in progress. The READ_ONCE()s
198 : : * permit this function to be invoked without holding the root rcu_node
199 : : * structure's ->lock, but of course results can be subject to change.
200 : : */
201 : 426616 : static int rcu_gp_in_progress(void)
202 : : {
203 [ # # # # ]: 0 : return rcu_seq_state(rcu_seq_current(&rcu_state.gp_seq));
204 : : }
205 : :
206 : : /*
207 : : * Return the number of callbacks queued on the specified CPU.
208 : : * Handles both the nocbs and normal cases.
209 : : */
210 : 0 : static long rcu_get_n_cbs_cpu(int cpu)
211 : : {
212 : 0 : struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
213 : :
214 [ # # # # : 0 : if (rcu_segcblist_is_enabled(&rdp->cblist))
# # ]
215 : 0 : return rcu_segcblist_n_cbs(&rdp->cblist);
216 : : return 0;
217 : : }
218 : :
219 : 178698 : void rcu_softirq_qs(void)
220 : : {
221 : 178698 : rcu_qs();
222 : 178698 : rcu_preempt_deferred_qs(current);
223 : 178698 : }
224 : :
225 : : /*
226 : : * Record entry into an extended quiescent state. This is only to be
227 : : * called when not already in an extended quiescent state.
228 : : */
229 : 21899 : static void rcu_dynticks_eqs_enter(void)
230 : : {
231 : 43798 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
232 : 21899 : int seq;
233 : :
234 : : /*
235 : : * CPUs seeing atomic_add_return() must see prior RCU read-side
236 : : * critical sections, and we also must force ordering with the
237 : : * next idle sojourn.
238 : : */
239 : 21899 : seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks);
240 : : /* Better be in an extended quiescent state! */
241 : 21899 : WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
242 : : (seq & RCU_DYNTICK_CTRL_CTR));
243 : : /* Better not have special action (TLB flush) pending! */
244 : 21899 : WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
245 : : (seq & RCU_DYNTICK_CTRL_MASK));
246 : 21899 : }
247 : :
248 : : /*
249 : : * Record exit from an extended quiescent state. This is only to be
250 : : * called from an extended quiescent state.
251 : : */
252 : 21899 : static void rcu_dynticks_eqs_exit(void)
253 : : {
254 : 21899 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
255 : 21899 : int seq;
256 : :
257 : : /*
258 : : * CPUs seeing atomic_add_return() must see prior idle sojourns,
259 : : * and we also must force ordering with the next RCU read-side
260 : : * critical section.
261 : : */
262 : 21899 : seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks);
263 : 21899 : WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
264 : : !(seq & RCU_DYNTICK_CTRL_CTR));
265 [ - + ]: 21899 : if (seq & RCU_DYNTICK_CTRL_MASK) {
266 : 0 : atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdp->dynticks);
267 : 21899 : smp_mb__after_atomic(); /* _exit after clearing mask. */
268 : : /* Prefer duplicate flushes to losing a flush. */
269 : 21899 : rcu_eqs_special_exit();
270 : : }
271 : 21899 : }
272 : :
273 : : /*
274 : : * Reset the current CPU's ->dynticks counter to indicate that the
275 : : * newly onlined CPU is no longer in an extended quiescent state.
276 : : * This will either leave the counter unchanged, or increment it
277 : : * to the next non-quiescent value.
278 : : *
279 : : * The non-atomic test/increment sequence works because the upper bits
280 : : * of the ->dynticks counter are manipulated only by the corresponding CPU,
281 : : * or when the corresponding CPU is offline.
282 : : */
283 : 28 : static void rcu_dynticks_eqs_online(void)
284 : : {
285 : 28 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
286 : :
287 [ - + ]: 28 : if (atomic_read(&rdp->dynticks) & RCU_DYNTICK_CTRL_CTR)
288 : : return;
289 : 0 : atomic_add(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks);
290 : : }
291 : :
292 : : /*
293 : : * Is the current CPU in an extended quiescent state?
294 : : *
295 : : * No ordering, as we are sampling CPU-local information.
296 : : */
297 : 2342810 : static bool rcu_dynticks_curr_cpu_in_eqs(void)
298 : : {
299 : 28 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
300 : :
301 : 2342810 : return !(atomic_read(&rdp->dynticks) & RCU_DYNTICK_CTRL_CTR);
302 : : }
303 : :
304 : : /*
305 : : * Snapshot the ->dynticks counter with full ordering so as to allow
306 : : * stable comparison of this counter with past and future snapshots.
307 : : */
308 : 28 : static int rcu_dynticks_snap(struct rcu_data *rdp)
309 : : {
310 : 0 : int snap = atomic_add_return(0, &rdp->dynticks);
311 : :
312 [ # # # # : 0 : return snap & ~RCU_DYNTICK_CTRL_MASK;
# # ]
313 : : }
314 : :
315 : : /*
316 : : * Return true if the snapshot returned from rcu_dynticks_snap()
317 : : * indicates that RCU is in an extended quiescent state.
318 : : */
319 : 28 : static bool rcu_dynticks_in_eqs(int snap)
320 : : {
321 [ # # ]: 0 : return !(snap & RCU_DYNTICK_CTRL_CTR);
322 : : }
323 : :
324 : : /*
325 : : * Return true if the CPU corresponding to the specified rcu_data
326 : : * structure has spent some time in an extended quiescent state since
327 : : * rcu_dynticks_snap() returned the specified snapshot.
328 : : */
329 : 0 : static bool rcu_dynticks_in_eqs_since(struct rcu_data *rdp, int snap)
330 : : {
331 : 0 : return snap != rcu_dynticks_snap(rdp);
332 : : }
333 : :
334 : : /*
335 : : * Set the special (bottom) bit of the specified CPU so that it
336 : : * will take special action (such as flushing its TLB) on the
337 : : * next exit from an extended quiescent state. Returns true if
338 : : * the bit was successfully set, or false if the CPU was not in
339 : : * an extended quiescent state.
340 : : */
341 : 0 : bool rcu_eqs_special_set(int cpu)
342 : : {
343 : 0 : int old;
344 : 0 : int new;
345 : 0 : struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
346 : :
347 : 0 : do {
348 : 0 : old = atomic_read(&rdp->dynticks);
349 [ # # ]: 0 : if (old & RCU_DYNTICK_CTRL_CTR)
350 : : return false;
351 : 0 : new = old | RCU_DYNTICK_CTRL_MASK;
352 [ # # ]: 0 : } while (atomic_cmpxchg(&rdp->dynticks, old, new) != old);
353 : : return true;
354 : : }
355 : :
356 : : /*
357 : : * Let the RCU core know that this CPU has gone through the scheduler,
358 : : * which is a quiescent state. This is called when the need for a
359 : : * quiescent state is urgent, so we burn an atomic operation and full
360 : : * memory barriers to let the RCU core know about it, regardless of what
361 : : * this CPU might (or might not) do in the near future.
362 : : *
363 : : * We inform the RCU core by emulating a zero-duration dyntick-idle period.
364 : : *
365 : : * The caller must have disabled interrupts and must not be idle.
366 : : */
367 : 224 : void rcu_momentary_dyntick_idle(void)
368 : : {
369 : 224 : int special;
370 : :
371 : 224 : raw_cpu_write(rcu_data.rcu_need_heavy_qs, false);
372 : 448 : special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR,
373 : 224 : &this_cpu_ptr(&rcu_data)->dynticks);
374 : : /* It is illegal to call this from idle state. */
375 [ - + ]: 224 : WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR));
376 : 224 : rcu_preempt_deferred_qs(current);
377 : 224 : }
378 : : EXPORT_SYMBOL_GPL(rcu_momentary_dyntick_idle);
379 : :
380 : : /**
381 : : * rcu_is_cpu_rrupt_from_idle - see if interrupted from idle
382 : : *
383 : : * If the current CPU is idle and running at a first-level (not nested)
384 : : * interrupt from idle, return true. The caller must have at least
385 : : * disabled preemption.
386 : : */
387 : 129410 : static int rcu_is_cpu_rrupt_from_idle(void)
388 : : {
389 : : /* Called only from within the scheduling-clock interrupt */
390 : 129410 : lockdep_assert_in_irq();
391 : :
392 : : /* Check for counter underflows */
393 : 129410 : RCU_LOCKDEP_WARN(__this_cpu_read(rcu_data.dynticks_nesting) < 0,
394 : : "RCU dynticks_nesting counter underflow!");
395 : 129410 : RCU_LOCKDEP_WARN(__this_cpu_read(rcu_data.dynticks_nmi_nesting) <= 0,
396 : : "RCU dynticks_nmi_nesting counter underflow/zero!");
397 : :
398 : : /* Are we at first interrupt nesting level? */
399 [ + + - - : 129410 : if (__this_cpu_read(rcu_data.dynticks_nmi_nesting) != 1)
+ + ]
400 : : return false;
401 : :
402 : : /* Does CPU appear to be idle from an RCU standpoint? */
403 [ + - - - : 4722 : return __this_cpu_read(rcu_data.dynticks_nesting) == 0;
- - ]
404 : : }
405 : :
406 : : #define DEFAULT_RCU_BLIMIT 10 /* Maximum callbacks per rcu_do_batch ... */
407 : : #define DEFAULT_MAX_RCU_BLIMIT 10000 /* ... even during callback flood. */
408 : : static long blimit = DEFAULT_RCU_BLIMIT;
409 : : #define DEFAULT_RCU_QHIMARK 10000 /* If this many pending, ignore blimit. */
410 : : static long qhimark = DEFAULT_RCU_QHIMARK;
411 : : #define DEFAULT_RCU_QLOMARK 100 /* Once only this many pending, use blimit. */
412 : : static long qlowmark = DEFAULT_RCU_QLOMARK;
413 : :
414 : : module_param(blimit, long, 0444);
415 : : module_param(qhimark, long, 0444);
416 : : module_param(qlowmark, long, 0444);
417 : :
418 : : static ulong jiffies_till_first_fqs = ULONG_MAX;
419 : : static ulong jiffies_till_next_fqs = ULONG_MAX;
420 : : static bool rcu_kick_kthreads;
421 : : static int rcu_divisor = 7;
422 : : module_param(rcu_divisor, int, 0644);
423 : :
424 : : /* Force an exit from rcu_do_batch() after 3 milliseconds. */
425 : : static long rcu_resched_ns = 3 * NSEC_PER_MSEC;
426 : : module_param(rcu_resched_ns, long, 0644);
427 : :
428 : : /*
429 : : * How long the grace period must be before we start recruiting
430 : : * quiescent-state help from rcu_note_context_switch().
431 : : */
432 : : static ulong jiffies_till_sched_qs = ULONG_MAX;
433 : : module_param(jiffies_till_sched_qs, ulong, 0444);
434 : : static ulong jiffies_to_sched_qs; /* See adjust_jiffies_till_sched_qs(). */
435 : : module_param(jiffies_to_sched_qs, ulong, 0444); /* Display only! */
436 : :
437 : : /*
438 : : * Make sure that we give the grace-period kthread time to detect any
439 : : * idle CPUs before taking active measures to force quiescent states.
440 : : * However, don't go below 100 milliseconds, adjusted upwards for really
441 : : * large systems.
442 : : */
443 : 28 : static void adjust_jiffies_till_sched_qs(void)
444 : : {
445 : 28 : unsigned long j;
446 : :
447 : : /* If jiffies_till_sched_qs was specified, respect the request. */
448 [ - + ]: 28 : if (jiffies_till_sched_qs != ULONG_MAX) {
449 : 0 : WRITE_ONCE(jiffies_to_sched_qs, jiffies_till_sched_qs);
450 : 0 : return;
451 : : }
452 : : /* Otherwise, set to third fqs scan, but bound below on large system. */
453 : 28 : j = READ_ONCE(jiffies_till_first_fqs) +
454 : 28 : 2 * READ_ONCE(jiffies_till_next_fqs);
455 : 28 : if (j < HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV)
456 : : j = HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV;
457 : 28 : pr_info("RCU calculated value of scheduler-enlistment delay is %ld jiffies.\n", j);
458 : 28 : WRITE_ONCE(jiffies_to_sched_qs, j);
459 : : }
460 : :
461 : 0 : static int param_set_first_fqs_jiffies(const char *val, const struct kernel_param *kp)
462 : : {
463 : 0 : ulong j;
464 : 0 : int ret = kstrtoul(val, 0, &j);
465 : :
466 [ # # ]: 0 : if (!ret) {
467 : 0 : WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : j);
468 : 0 : adjust_jiffies_till_sched_qs();
469 : : }
470 : 0 : return ret;
471 : : }
472 : :
473 : 0 : static int param_set_next_fqs_jiffies(const char *val, const struct kernel_param *kp)
474 : : {
475 : 0 : ulong j;
476 : 0 : int ret = kstrtoul(val, 0, &j);
477 : :
478 [ # # ]: 0 : if (!ret) {
479 : 0 : WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : (j ?: 1));
480 : 0 : adjust_jiffies_till_sched_qs();
481 : : }
482 : 0 : return ret;
483 : : }
484 : :
485 : : static struct kernel_param_ops first_fqs_jiffies_ops = {
486 : : .set = param_set_first_fqs_jiffies,
487 : : .get = param_get_ulong,
488 : : };
489 : :
490 : : static struct kernel_param_ops next_fqs_jiffies_ops = {
491 : : .set = param_set_next_fqs_jiffies,
492 : : .get = param_get_ulong,
493 : : };
494 : :
495 : : module_param_cb(jiffies_till_first_fqs, &first_fqs_jiffies_ops, &jiffies_till_first_fqs, 0644);
496 : : module_param_cb(jiffies_till_next_fqs, &next_fqs_jiffies_ops, &jiffies_till_next_fqs, 0644);
497 : : module_param(rcu_kick_kthreads, bool, 0644);
498 : :
499 : : static void force_qs_rnp(int (*f)(struct rcu_data *rdp));
500 : : static int rcu_pending(int user);
501 : :
502 : : /*
503 : : * Return the number of RCU GPs completed thus far for debug & stats.
504 : : */
505 : 0 : unsigned long rcu_get_gp_seq(void)
506 : : {
507 : 0 : return READ_ONCE(rcu_state.gp_seq);
508 : : }
509 : : EXPORT_SYMBOL_GPL(rcu_get_gp_seq);
510 : :
511 : : /*
512 : : * Return the number of RCU expedited batches completed thus far for
513 : : * debug & stats. Odd numbers mean that a batch is in progress, even
514 : : * numbers mean idle. The value returned will thus be roughly double
515 : : * the cumulative batches since boot.
516 : : */
517 : 0 : unsigned long rcu_exp_batches_completed(void)
518 : : {
519 : 0 : return rcu_state.expedited_sequence;
520 : : }
521 : : EXPORT_SYMBOL_GPL(rcu_exp_batches_completed);
522 : :
523 : : /*
524 : : * Return the root node of the rcu_state structure.
525 : : */
526 : 789647 : static struct rcu_node *rcu_get_root(void)
527 : : {
528 [ - - - - ]: 230048 : return &rcu_state.node[0];
529 : : }
530 : :
531 : : /*
532 : : * Send along grace-period-related data for rcutorture diagnostics.
533 : : */
534 : 0 : void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
535 : : unsigned long *gp_seq)
536 : : {
537 [ # # ]: 0 : switch (test_type) {
538 : : case RCU_FLAVOR:
539 : 0 : *flags = READ_ONCE(rcu_state.gp_flags);
540 : 0 : *gp_seq = rcu_seq_current(&rcu_state.gp_seq);
541 : 0 : break;
542 : : default:
543 : : break;
544 : : }
545 : 0 : }
546 : : EXPORT_SYMBOL_GPL(rcutorture_get_gp_data);
547 : :
548 : : /*
549 : : * Enter an RCU extended quiescent state, which can be either the
550 : : * idle loop or adaptive-tickless usermode execution.
551 : : *
552 : : * We crowbar the ->dynticks_nmi_nesting field to zero to allow for
553 : : * the possibility of usermode upcalls having messed up our count
554 : : * of interrupt nesting level during the prior busy period.
555 : : */
556 : 10947 : static void rcu_eqs_enter(bool user)
557 : : {
558 : 10947 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
559 : :
560 [ - + ]: 10947 : WARN_ON_ONCE(rdp->dynticks_nmi_nesting != DYNTICK_IRQ_NONIDLE);
561 [ - + ]: 10947 : WRITE_ONCE(rdp->dynticks_nmi_nesting, 0);
562 : 10947 : WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
563 : : rdp->dynticks_nesting == 0);
564 [ - + ]: 10947 : if (rdp->dynticks_nesting != 1) {
565 : 0 : rdp->dynticks_nesting--;
566 : 0 : return;
567 : : }
568 : :
569 : 10947 : lockdep_assert_irqs_disabled();
570 : 10947 : trace_rcu_dyntick(TPS("Start"), rdp->dynticks_nesting, 0, atomic_read(&rdp->dynticks));
571 : 10947 : WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current));
572 : 10947 : rdp = this_cpu_ptr(&rcu_data);
573 : 10947 : do_nocb_deferred_wakeup(rdp);
574 : 10947 : rcu_prepare_for_idle();
575 : 10947 : rcu_preempt_deferred_qs(current);
576 : 10947 : WRITE_ONCE(rdp->dynticks_nesting, 0); /* Avoid irq-access tearing. */
577 : 10947 : rcu_dynticks_eqs_enter();
578 : : rcu_dynticks_task_enter();
579 : : }
580 : :
581 : : /**
582 : : * rcu_idle_enter - inform RCU that current CPU is entering idle
583 : : *
584 : : * Enter idle mode, in other words, -leave- the mode in which RCU
585 : : * read-side critical sections can occur. (Though RCU read-side
586 : : * critical sections can occur in irq handlers in idle, a possibility
587 : : * handled by irq_enter() and irq_exit().)
588 : : *
589 : : * If you add or remove a call to rcu_idle_enter(), be sure to test with
590 : : * CONFIG_RCU_EQS_DEBUG=y.
591 : : */
592 : 10947 : void rcu_idle_enter(void)
593 : : {
594 : 10947 : lockdep_assert_irqs_disabled();
595 : 10947 : rcu_eqs_enter(false);
596 : 10947 : }
597 : :
598 : : #ifdef CONFIG_NO_HZ_FULL
599 : : /**
600 : : * rcu_user_enter - inform RCU that we are resuming userspace.
601 : : *
602 : : * Enter RCU idle mode right before resuming userspace. No use of RCU
603 : : * is permitted between this call and rcu_user_exit(). This way the
604 : : * CPU doesn't need to maintain the tick for RCU maintenance purposes
605 : : * when the CPU runs in userspace.
606 : : *
607 : : * If you add or remove a call to rcu_user_enter(), be sure to test with
608 : : * CONFIG_RCU_EQS_DEBUG=y.
609 : : */
610 : : void rcu_user_enter(void)
611 : : {
612 : : lockdep_assert_irqs_disabled();
613 : : rcu_eqs_enter(true);
614 : : }
615 : : #endif /* CONFIG_NO_HZ_FULL */
616 : :
617 : : /*
618 : : * If we are returning from the outermost NMI handler that interrupted an
619 : : * RCU-idle period, update rdp->dynticks and rdp->dynticks_nmi_nesting
620 : : * to let the RCU grace-period handling know that the CPU is back to
621 : : * being RCU-idle.
622 : : *
623 : : * If you add or remove a call to rcu_nmi_exit_common(), be sure to test
624 : : * with CONFIG_RCU_EQS_DEBUG=y.
625 : : */
626 : 141689 : static __always_inline void rcu_nmi_exit_common(bool irq)
627 : : {
628 : 283378 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
629 : :
630 : : /*
631 : : * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks.
632 : : * (We are exiting an NMI handler, so RCU better be paying attention
633 : : * to us!)
634 : : */
635 [ - + - + ]: 141689 : WARN_ON_ONCE(rdp->dynticks_nmi_nesting <= 0);
636 [ - + - + ]: 283378 : WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs());
637 : :
638 : : /*
639 : : * If the nesting level is not 1, the CPU wasn't RCU-idle, so
640 : : * leave it in non-RCU-idle state.
641 : : */
642 [ + + + - ]: 141689 : if (rdp->dynticks_nmi_nesting != 1) {
643 : 130737 : trace_rcu_dyntick(TPS("--="), rdp->dynticks_nmi_nesting, rdp->dynticks_nmi_nesting - 2,
644 : 130737 : atomic_read(&rdp->dynticks));
645 : 130737 : WRITE_ONCE(rdp->dynticks_nmi_nesting, /* No store tearing. */
646 : : rdp->dynticks_nmi_nesting - 2);
647 : 130737 : return;
648 : : }
649 : :
650 : : /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */
651 : 10952 : trace_rcu_dyntick(TPS("Startirq"), rdp->dynticks_nmi_nesting, 0, atomic_read(&rdp->dynticks));
652 : 10952 : WRITE_ONCE(rdp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */
653 : :
654 : 10952 : if (irq)
655 : 10952 : rcu_prepare_for_idle();
656 : :
657 : 10952 : rcu_dynticks_eqs_enter();
658 : :
659 : : if (irq)
660 : : rcu_dynticks_task_enter();
661 : : }
662 : :
663 : : /**
664 : : * rcu_nmi_exit - inform RCU of exit from NMI context
665 : : *
666 : : * If you add or remove a call to rcu_nmi_exit(), be sure to test
667 : : * with CONFIG_RCU_EQS_DEBUG=y.
668 : : */
669 : 28 : void rcu_nmi_exit(void)
670 : : {
671 : 28 : rcu_nmi_exit_common(false);
672 : 28 : }
673 : :
674 : : /**
675 : : * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle
676 : : *
677 : : * Exit from an interrupt handler, which might possibly result in entering
678 : : * idle mode, in other words, leaving the mode in which read-side critical
679 : : * sections can occur. The caller must have disabled interrupts.
680 : : *
681 : : * This code assumes that the idle loop never does anything that might
682 : : * result in unbalanced calls to irq_enter() and irq_exit(). If your
683 : : * architecture's idle loop violates this assumption, RCU will give you what
684 : : * you deserve, good and hard. But very infrequently and irreproducibly.
685 : : *
686 : : * Use things like work queues to work around this limitation.
687 : : *
688 : : * You have been warned.
689 : : *
690 : : * If you add or remove a call to rcu_irq_exit(), be sure to test with
691 : : * CONFIG_RCU_EQS_DEBUG=y.
692 : : */
693 : 141661 : void rcu_irq_exit(void)
694 : : {
695 : 141661 : lockdep_assert_irqs_disabled();
696 : 141661 : rcu_nmi_exit_common(true);
697 : 141661 : }
698 : :
699 : : /*
700 : : * Wrapper for rcu_irq_exit() where interrupts are enabled.
701 : : *
702 : : * If you add or remove a call to rcu_irq_exit_irqson(), be sure to test
703 : : * with CONFIG_RCU_EQS_DEBUG=y.
704 : : */
705 : 0 : void rcu_irq_exit_irqson(void)
706 : : {
707 : 0 : unsigned long flags;
708 : :
709 : 0 : local_irq_save(flags);
710 : 0 : rcu_irq_exit();
711 : 0 : local_irq_restore(flags);
712 : 0 : }
713 : :
714 : : /*
715 : : * Exit an RCU extended quiescent state, which can be either the
716 : : * idle loop or adaptive-tickless usermode execution.
717 : : *
718 : : * We crowbar the ->dynticks_nmi_nesting field to DYNTICK_IRQ_NONIDLE to
719 : : * allow for the possibility of usermode upcalls messing up our count of
720 : : * interrupt nesting level during the busy period that is just now starting.
721 : : */
722 : 10947 : static void rcu_eqs_exit(bool user)
723 : : {
724 : 10947 : struct rcu_data *rdp;
725 : 10947 : long oldval;
726 : :
727 : 10947 : lockdep_assert_irqs_disabled();
728 : 10947 : rdp = this_cpu_ptr(&rcu_data);
729 : 10947 : oldval = rdp->dynticks_nesting;
730 : 10947 : WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0);
731 [ - + ]: 10947 : if (oldval) {
732 : 0 : rdp->dynticks_nesting++;
733 : 0 : return;
734 : : }
735 : 10947 : rcu_dynticks_task_exit();
736 : 10947 : rcu_dynticks_eqs_exit();
737 : 10947 : rcu_cleanup_after_idle();
738 : 10947 : trace_rcu_dyntick(TPS("End"), rdp->dynticks_nesting, 1, atomic_read(&rdp->dynticks));
739 : 10947 : WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current));
740 [ - + ]: 10947 : WRITE_ONCE(rdp->dynticks_nesting, 1);
741 [ - + ]: 10947 : WARN_ON_ONCE(rdp->dynticks_nmi_nesting);
742 : 10947 : WRITE_ONCE(rdp->dynticks_nmi_nesting, DYNTICK_IRQ_NONIDLE);
743 : : }
744 : :
745 : : /**
746 : : * rcu_idle_exit - inform RCU that current CPU is leaving idle
747 : : *
748 : : * Exit idle mode, in other words, -enter- the mode in which RCU
749 : : * read-side critical sections can occur.
750 : : *
751 : : * If you add or remove a call to rcu_idle_exit(), be sure to test with
752 : : * CONFIG_RCU_EQS_DEBUG=y.
753 : : */
754 : 10947 : void rcu_idle_exit(void)
755 : : {
756 : 10947 : unsigned long flags;
757 : :
758 : 10947 : local_irq_save(flags);
759 : 10947 : rcu_eqs_exit(false);
760 : 10947 : local_irq_restore(flags);
761 : 10947 : }
762 : :
763 : : #ifdef CONFIG_NO_HZ_FULL
764 : : /**
765 : : * rcu_user_exit - inform RCU that we are exiting userspace.
766 : : *
767 : : * Exit RCU idle mode while entering the kernel because it can
768 : : * run a RCU read side critical section anytime.
769 : : *
770 : : * If you add or remove a call to rcu_user_exit(), be sure to test with
771 : : * CONFIG_RCU_EQS_DEBUG=y.
772 : : */
773 : : void rcu_user_exit(void)
774 : : {
775 : : rcu_eqs_exit(1);
776 : : }
777 : : #endif /* CONFIG_NO_HZ_FULL */
778 : :
779 : : /**
780 : : * rcu_nmi_enter_common - inform RCU of entry to NMI context
781 : : * @irq: Is this call from rcu_irq_enter?
782 : : *
783 : : * If the CPU was idle from RCU's viewpoint, update rdp->dynticks and
784 : : * rdp->dynticks_nmi_nesting to let the RCU grace-period handling know
785 : : * that the CPU is active. This implementation permits nested NMIs, as
786 : : * long as the nesting level does not overflow an int. (You will probably
787 : : * run out of stack space first.)
788 : : *
789 : : * If you add or remove a call to rcu_nmi_enter_common(), be sure to test
790 : : * with CONFIG_RCU_EQS_DEBUG=y.
791 : : */
792 : 141689 : static __always_inline void rcu_nmi_enter_common(bool irq)
793 : : {
794 : 141689 : long incby = 2;
795 : 283378 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
796 : :
797 : : /* Complain about underflow. */
798 [ - + - + ]: 141689 : WARN_ON_ONCE(rdp->dynticks_nmi_nesting < 0);
799 : :
800 : : /*
801 : : * If idle from RCU viewpoint, atomically increment ->dynticks
802 : : * to mark non-idle and increment ->dynticks_nmi_nesting by one.
803 : : * Otherwise, increment ->dynticks_nmi_nesting by two. This means
804 : : * if ->dynticks_nmi_nesting is equal to one, we are guaranteed
805 : : * to be in the outermost NMI handler that interrupted an RCU-idle
806 : : * period (observation due to Andy Lutomirski).
807 : : */
808 [ + + - + ]: 283350 : if (rcu_dynticks_curr_cpu_in_eqs()) {
809 : :
810 : 10952 : if (irq)
811 : 10952 : rcu_dynticks_task_exit();
812 : :
813 : 10952 : rcu_dynticks_eqs_exit();
814 : :
815 : 10952 : if (irq)
816 : : rcu_cleanup_after_idle();
817 : :
818 : : incby = 1;
819 : : } else if (tick_nohz_full_cpu(rdp->cpu) &&
820 : : rdp->dynticks_nmi_nesting == DYNTICK_IRQ_NONIDLE &&
821 : : READ_ONCE(rdp->rcu_urgent_qs) && !rdp->rcu_forced_tick) {
822 : : raw_spin_lock_rcu_node(rdp->mynode);
823 : : // Recheck under lock.
824 : : if (rdp->rcu_urgent_qs && !rdp->rcu_forced_tick) {
825 : : rdp->rcu_forced_tick = true;
826 : : tick_dep_set_cpu(rdp->cpu, TICK_DEP_BIT_RCU);
827 : : }
828 : : raw_spin_unlock_rcu_node(rdp->mynode);
829 : : }
830 : 141689 : trace_rcu_dyntick(incby == 1 ? TPS("Endirq") : TPS("++="),
831 : : rdp->dynticks_nmi_nesting,
832 [ + + - + ]: 141689 : rdp->dynticks_nmi_nesting + incby, atomic_read(&rdp->dynticks));
833 : 141689 : WRITE_ONCE(rdp->dynticks_nmi_nesting, /* Prevent store tearing. */
834 : : rdp->dynticks_nmi_nesting + incby);
835 : 141689 : barrier();
836 : : }
837 : :
838 : : /**
839 : : * rcu_nmi_enter - inform RCU of entry to NMI context
840 : : */
841 : 28 : void rcu_nmi_enter(void)
842 : : {
843 : 28 : rcu_nmi_enter_common(false);
844 : 28 : }
845 : : NOKPROBE_SYMBOL(rcu_nmi_enter);
846 : :
847 : : /**
848 : : * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle
849 : : *
850 : : * Enter an interrupt handler, which might possibly result in exiting
851 : : * idle mode, in other words, entering the mode in which read-side critical
852 : : * sections can occur. The caller must have disabled interrupts.
853 : : *
854 : : * Note that the Linux kernel is fully capable of entering an interrupt
855 : : * handler that it never exits, for example when doing upcalls to user mode!
856 : : * This code assumes that the idle loop never does upcalls to user mode.
857 : : * If your architecture's idle loop does do upcalls to user mode (or does
858 : : * anything else that results in unbalanced calls to the irq_enter() and
859 : : * irq_exit() functions), RCU will give you what you deserve, good and hard.
860 : : * But very infrequently and irreproducibly.
861 : : *
862 : : * Use things like work queues to work around this limitation.
863 : : *
864 : : * You have been warned.
865 : : *
866 : : * If you add or remove a call to rcu_irq_enter(), be sure to test with
867 : : * CONFIG_RCU_EQS_DEBUG=y.
868 : : */
869 : 141661 : void rcu_irq_enter(void)
870 : : {
871 : 141661 : lockdep_assert_irqs_disabled();
872 : 141661 : rcu_nmi_enter_common(true);
873 : 141661 : }
874 : :
875 : : /*
876 : : * Wrapper for rcu_irq_enter() where interrupts are enabled.
877 : : *
878 : : * If you add or remove a call to rcu_irq_enter_irqson(), be sure to test
879 : : * with CONFIG_RCU_EQS_DEBUG=y.
880 : : */
881 : 0 : void rcu_irq_enter_irqson(void)
882 : : {
883 : 0 : unsigned long flags;
884 : :
885 : 0 : local_irq_save(flags);
886 : 0 : rcu_irq_enter();
887 : 0 : local_irq_restore(flags);
888 : 0 : }
889 : :
890 : : /*
891 : : * If any sort of urgency was applied to the current CPU (for example,
892 : : * the scheduler-clock interrupt was enabled on a nohz_full CPU) in order
893 : : * to get to a quiescent state, disable it.
894 : : */
895 : 0 : static void rcu_disable_urgency_upon_qs(struct rcu_data *rdp)
896 : : {
897 : 0 : raw_lockdep_assert_held_rcu_node(rdp->mynode);
898 : 0 : WRITE_ONCE(rdp->rcu_urgent_qs, false);
899 : 0 : WRITE_ONCE(rdp->rcu_need_heavy_qs, false);
900 : 0 : if (tick_nohz_full_cpu(rdp->cpu) && rdp->rcu_forced_tick) {
901 : : tick_dep_clear_cpu(rdp->cpu, TICK_DEP_BIT_RCU);
902 : : rdp->rcu_forced_tick = false;
903 : : }
904 : 0 : }
905 : :
906 : : /**
907 : : * rcu_is_watching - see if RCU thinks that the current CPU is not idle
908 : : *
909 : : * Return true if RCU is watching the running CPU, which means that this
910 : : * CPU can safely enter RCU read-side critical sections. In other words,
911 : : * if the current CPU is not in its idle loop or is in an interrupt or
912 : : * NMI handler, return true.
913 : : */
914 : 2059432 : bool notrace rcu_is_watching(void)
915 : : {
916 : 2059432 : bool ret;
917 : :
918 : 5450 : preempt_disable_notrace();
919 : 2059432 : ret = !rcu_dynticks_curr_cpu_in_eqs();
920 : 2059432 : preempt_enable_notrace();
921 : 2059432 : return ret;
922 : : }
923 : : EXPORT_SYMBOL_GPL(rcu_is_watching);
924 : :
925 : : /*
926 : : * If a holdout task is actually running, request an urgent quiescent
927 : : * state from its CPU. This is unsynchronized, so migrations can cause
928 : : * the request to go to the wrong CPU. Which is OK, all that will happen
929 : : * is that the CPU's next context switch will be a bit slower and next
930 : : * time around this task will generate another request.
931 : : */
932 : 0 : void rcu_request_urgent_qs_task(struct task_struct *t)
933 : : {
934 : 0 : int cpu;
935 : :
936 : 0 : barrier();
937 : 0 : cpu = task_cpu(t);
938 [ # # ]: 0 : if (!task_curr(t))
939 : : return; /* This task is not running on that CPU. */
940 : 0 : smp_store_release(per_cpu_ptr(&rcu_data.rcu_urgent_qs, cpu), true);
941 : : }
942 : :
943 : : #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
944 : :
945 : : /*
946 : : * Is the current CPU online as far as RCU is concerned?
947 : : *
948 : : * Disable preemption to avoid false positives that could otherwise
949 : : * happen due to the current CPU number being sampled, this task being
950 : : * preempted, its old CPU being taken offline, resuming on some other CPU,
951 : : * then determining that its old CPU is now offline.
952 : : *
953 : : * Disable checking if in an NMI handler because we cannot safely
954 : : * report errors from NMI handlers anyway. In addition, it is OK to use
955 : : * RCU on an offline processor during initial boot, hence the check for
956 : : * rcu_scheduler_fully_active.
957 : : */
958 : : bool rcu_lockdep_current_cpu_online(void)
959 : : {
960 : : struct rcu_data *rdp;
961 : : struct rcu_node *rnp;
962 : : bool ret = false;
963 : :
964 : : if (in_nmi() || !rcu_scheduler_fully_active)
965 : : return true;
966 : : preempt_disable();
967 : : rdp = this_cpu_ptr(&rcu_data);
968 : : rnp = rdp->mynode;
969 : : if (rdp->grpmask & rcu_rnp_online_cpus(rnp))
970 : : ret = true;
971 : : preempt_enable();
972 : : return ret;
973 : : }
974 : : EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online);
975 : :
976 : : #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */
977 : :
978 : : /*
979 : : * We are reporting a quiescent state on behalf of some other CPU, so
980 : : * it is our responsibility to check for and handle potential overflow
981 : : * of the rcu_node ->gp_seq counter with respect to the rcu_data counters.
982 : : * After all, the CPU might be in deep idle state, and thus executing no
983 : : * code whatsoever.
984 : : */
985 : 124378 : static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp)
986 : : {
987 : 124378 : raw_lockdep_assert_held_rcu_node(rnp);
988 : 124378 : if (ULONG_CMP_LT(rcu_seq_current(&rdp->gp_seq) + ULONG_MAX / 4,
989 : : rnp->gp_seq))
990 : 0 : WRITE_ONCE(rdp->gpwrap, true);
991 [ - + - + : 124378 : if (ULONG_CMP_LT(rdp->rcu_iw_gp_seq + ULONG_MAX / 4, rnp->gp_seq))
- - - - ]
992 : 0 : rdp->rcu_iw_gp_seq = rnp->gp_seq + ULONG_MAX / 4;
993 : : }
994 : :
995 : : /*
996 : : * Snapshot the specified CPU's dynticks counter so that we can later
997 : : * credit them with an implicit quiescent state. Return 1 if this CPU
998 : : * is in dynticks idle mode, which is an extended quiescent state.
999 : : */
1000 : 0 : static int dyntick_save_progress_counter(struct rcu_data *rdp)
1001 : : {
1002 : 0 : rdp->dynticks_snap = rcu_dynticks_snap(rdp);
1003 [ # # ]: 0 : if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) {
1004 : 0 : trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti"));
1005 [ # # ]: 0 : rcu_gpnum_ovf(rdp->mynode, rdp);
1006 : 0 : return 1;
1007 : : }
1008 : : return 0;
1009 : : }
1010 : :
1011 : : /*
1012 : : * Return true if the specified CPU has passed through a quiescent
1013 : : * state by virtue of being in or having passed through an dynticks
1014 : : * idle state since the last call to dyntick_save_progress_counter()
1015 : : * for this same CPU, or by virtue of having been offline.
1016 : : */
1017 : 0 : static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
1018 : : {
1019 : 0 : unsigned long jtsq;
1020 : 0 : bool *rnhqp;
1021 : 0 : bool *ruqp;
1022 : 0 : struct rcu_node *rnp = rdp->mynode;
1023 : :
1024 : : /*
1025 : : * If the CPU passed through or entered a dynticks idle phase with
1026 : : * no active irq/NMI handlers, then we can safely pretend that the CPU
1027 : : * already acknowledged the request to pass through a quiescent
1028 : : * state. Either way, that CPU cannot possibly be in an RCU
1029 : : * read-side critical section that started before the beginning
1030 : : * of the current RCU grace period.
1031 : : */
1032 [ # # ]: 0 : if (rcu_dynticks_in_eqs_since(rdp, rdp->dynticks_snap)) {
1033 : 0 : trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti"));
1034 [ # # ]: 0 : rcu_gpnum_ovf(rnp, rdp);
1035 : 0 : return 1;
1036 : : }
1037 : :
1038 : : /* If waiting too long on an offline CPU, complain. */
1039 [ # # ]: 0 : if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp)) &&
1040 [ # # ]: 0 : time_after(jiffies, rcu_state.gp_start + HZ)) {
1041 : 0 : bool onl;
1042 : 0 : struct rcu_node *rnp1;
1043 : :
1044 : 0 : WARN_ON(1); /* Offline CPUs are supposed to report QS! */
1045 : 0 : pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n",
1046 : : __func__, rnp->grplo, rnp->grphi, rnp->level,
1047 : : (long)rnp->gp_seq, (long)rnp->completedqs);
1048 [ # # ]: 0 : for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent)
1049 : 0 : pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx ->rcu_gp_init_mask %#lx\n",
1050 : : __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext, rnp1->rcu_gp_init_mask);
1051 : 0 : onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp));
1052 : 0 : pr_info("%s %d: %c online: %ld(%d) offline: %ld(%d)\n",
1053 : : __func__, rdp->cpu, ".o"[onl],
1054 : : (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags,
1055 : : (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags);
1056 : 0 : return 1; /* Break things loose after complaining. */
1057 : : }
1058 : :
1059 : : /*
1060 : : * A CPU running for an extended time within the kernel can
1061 : : * delay RCU grace periods: (1) At age jiffies_to_sched_qs,
1062 : : * set .rcu_urgent_qs, (2) At age 2*jiffies_to_sched_qs, set
1063 : : * both .rcu_need_heavy_qs and .rcu_urgent_qs. Note that the
1064 : : * unsynchronized assignments to the per-CPU rcu_need_heavy_qs
1065 : : * variable are safe because the assignments are repeated if this
1066 : : * CPU failed to pass through a quiescent state. This code
1067 : : * also checks .jiffies_resched in case jiffies_to_sched_qs
1068 : : * is set way high.
1069 : : */
1070 [ # # ]: 0 : jtsq = READ_ONCE(jiffies_to_sched_qs);
1071 : 0 : ruqp = per_cpu_ptr(&rcu_data.rcu_urgent_qs, rdp->cpu);
1072 : 0 : rnhqp = &per_cpu(rcu_data.rcu_need_heavy_qs, rdp->cpu);
1073 [ # # ]: 0 : if (!READ_ONCE(*rnhqp) &&
1074 [ # # ]: 0 : (time_after(jiffies, rcu_state.gp_start + jtsq * 2) ||
1075 [ # # ]: 0 : time_after(jiffies, rcu_state.jiffies_resched))) {
1076 : 0 : WRITE_ONCE(*rnhqp, true);
1077 : : /* Store rcu_need_heavy_qs before rcu_urgent_qs. */
1078 : 0 : smp_store_release(ruqp, true);
1079 [ # # ]: 0 : } else if (time_after(jiffies, rcu_state.gp_start + jtsq)) {
1080 : 0 : WRITE_ONCE(*ruqp, true);
1081 : : }
1082 : :
1083 : : /*
1084 : : * NO_HZ_FULL CPUs can run in-kernel without rcu_sched_clock_irq!
1085 : : * The above code handles this, but only for straight cond_resched().
1086 : : * And some in-kernel loops check need_resched() before calling
1087 : : * cond_resched(), which defeats the above code for CPUs that are
1088 : : * running in-kernel with scheduling-clock interrupts disabled.
1089 : : * So hit them over the head with the resched_cpu() hammer!
1090 : : */
1091 [ # # ]: 0 : if (tick_nohz_full_cpu(rdp->cpu) &&
1092 : : time_after(jiffies,
1093 : : READ_ONCE(rdp->last_fqs_resched) + jtsq * 3)) {
1094 : : WRITE_ONCE(*ruqp, true);
1095 : : resched_cpu(rdp->cpu);
1096 : : WRITE_ONCE(rdp->last_fqs_resched, jiffies);
1097 : : }
1098 : :
1099 : : /*
1100 : : * If more than halfway to RCU CPU stall-warning time, invoke
1101 : : * resched_cpu() more frequently to try to loosen things up a bit.
1102 : : * Also check to see if the CPU is getting hammered with interrupts,
1103 : : * but only once per grace period, just to keep the IPIs down to
1104 : : * a dull roar.
1105 : : */
1106 [ # # ]: 0 : if (time_after(jiffies, rcu_state.jiffies_resched)) {
1107 [ # # ]: 0 : if (time_after(jiffies,
1108 : : READ_ONCE(rdp->last_fqs_resched) + jtsq)) {
1109 : 0 : resched_cpu(rdp->cpu);
1110 : 0 : WRITE_ONCE(rdp->last_fqs_resched, jiffies);
1111 : : }
1112 : 0 : if (IS_ENABLED(CONFIG_IRQ_WORK) &&
1113 [ # # # # ]: 0 : !rdp->rcu_iw_pending && rdp->rcu_iw_gp_seq != rnp->gp_seq &&
1114 [ # # ]: 0 : (rnp->ffmask & rdp->grpmask)) {
1115 : 0 : init_irq_work(&rdp->rcu_iw, rcu_iw_handler);
1116 : 0 : rdp->rcu_iw_pending = true;
1117 : 0 : rdp->rcu_iw_gp_seq = rnp->gp_seq;
1118 : 0 : irq_work_queue_on(&rdp->rcu_iw, rdp->cpu);
1119 : : }
1120 : : }
1121 : :
1122 : : return 0;
1123 : : }
1124 : :
1125 : : /* Trace-event wrapper function for trace_rcu_future_grace_period. */
1126 : 62161 : static void trace_rcu_this_gp(struct rcu_node *rnp, struct rcu_data *rdp,
1127 : : unsigned long gp_seq_req, const char *s)
1128 : : {
1129 : 62161 : trace_rcu_future_grace_period(rcu_state.name, rnp->gp_seq, gp_seq_req,
1130 : 62161 : rnp->level, rnp->grplo, rnp->grphi, s);
1131 : : }
1132 : :
1133 : : /*
1134 : : * rcu_start_this_gp - Request the start of a particular grace period
1135 : : * @rnp_start: The leaf node of the CPU from which to start.
1136 : : * @rdp: The rcu_data corresponding to the CPU from which to start.
1137 : : * @gp_seq_req: The gp_seq of the grace period to start.
1138 : : *
1139 : : * Start the specified grace period, as needed to handle newly arrived
1140 : : * callbacks. The required future grace periods are recorded in each
1141 : : * rcu_node structure's ->gp_seq_needed field. Returns true if there
1142 : : * is reason to awaken the grace-period kthread.
1143 : : *
1144 : : * The caller must hold the specified rcu_node structure's ->lock, which
1145 : : * is why the caller is responsible for waking the grace-period kthread.
1146 : : *
1147 : : * Returns true if the GP thread needs to be awakened else false.
1148 : : */
1149 : : static bool rcu_start_this_gp(struct rcu_node *rnp_start, struct rcu_data *rdp,
1150 : : unsigned long gp_seq_req)
1151 : : {
1152 : : bool ret = false;
1153 : : struct rcu_node *rnp;
1154 : :
1155 : : /*
1156 : : * Use funnel locking to either acquire the root rcu_node
1157 : : * structure's lock or bail out if the need for this grace period
1158 : : * has already been recorded -- or if that grace period has in
1159 : : * fact already started. If there is already a grace period in
1160 : : * progress in a non-leaf node, no recording is needed because the
1161 : : * end of the grace period will scan the leaf rcu_node structures.
1162 : : * Note that rnp_start->lock must not be released.
1163 : : */
1164 : : raw_lockdep_assert_held_rcu_node(rnp_start);
1165 : : trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, TPS("Startleaf"));
1166 : : for (rnp = rnp_start; 1; rnp = rnp->parent) {
1167 : : if (rnp != rnp_start)
1168 : : raw_spin_lock_rcu_node(rnp);
1169 : : if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req) ||
1170 : : rcu_seq_started(&rnp->gp_seq, gp_seq_req) ||
1171 : : (rnp != rnp_start &&
1172 : : rcu_seq_state(rcu_seq_current(&rnp->gp_seq)))) {
1173 : : trace_rcu_this_gp(rnp, rdp, gp_seq_req,
1174 : : TPS("Prestarted"));
1175 : : goto unlock_out;
1176 : : }
1177 : : rnp->gp_seq_needed = gp_seq_req;
1178 : : if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) {
1179 : : /*
1180 : : * We just marked the leaf or internal node, and a
1181 : : * grace period is in progress, which means that
1182 : : * rcu_gp_cleanup() will see the marking. Bail to
1183 : : * reduce contention.
1184 : : */
1185 : : trace_rcu_this_gp(rnp_start, rdp, gp_seq_req,
1186 : : TPS("Startedleaf"));
1187 : : goto unlock_out;
1188 : : }
1189 : : if (rnp != rnp_start && rnp->parent != NULL)
1190 : : raw_spin_unlock_rcu_node(rnp);
1191 : : if (!rnp->parent)
1192 : : break; /* At root, and perhaps also leaf. */
1193 : : }
1194 : :
1195 : : /* If GP already in progress, just leave, otherwise start one. */
1196 : : if (rcu_gp_in_progress()) {
1197 : : trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedleafroot"));
1198 : : goto unlock_out;
1199 : : }
1200 : : trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedroot"));
1201 : : WRITE_ONCE(rcu_state.gp_flags, rcu_state.gp_flags | RCU_GP_FLAG_INIT);
1202 : : rcu_state.gp_req_activity = jiffies;
1203 : : if (!rcu_state.gp_kthread) {
1204 : : trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("NoGPkthread"));
1205 : : goto unlock_out;
1206 : : }
1207 : : trace_rcu_grace_period(rcu_state.name, READ_ONCE(rcu_state.gp_seq), TPS("newreq"));
1208 : : ret = true; /* Caller must wake GP kthread. */
1209 : : unlock_out:
1210 : : /* Push furthest requested GP to leaf node and rcu_data structure. */
1211 : : if (ULONG_CMP_LT(gp_seq_req, rnp->gp_seq_needed)) {
1212 : : rnp_start->gp_seq_needed = rnp->gp_seq_needed;
1213 : : rdp->gp_seq_needed = rnp->gp_seq_needed;
1214 : : }
1215 : : if (rnp != rnp_start)
1216 : : raw_spin_unlock_rcu_node(rnp);
1217 : : return ret;
1218 : : }
1219 : :
1220 : : /*
1221 : : * Clean up any old requests for the just-ended grace period. Also return
1222 : : * whether any additional grace periods have been requested.
1223 : : */
1224 : 62161 : static bool rcu_future_gp_cleanup(struct rcu_node *rnp)
1225 : : {
1226 : 62161 : bool needmore;
1227 : 62161 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
1228 : :
1229 : 62161 : needmore = ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed);
1230 [ + + ]: 62161 : if (!needmore)
1231 : 2033 : rnp->gp_seq_needed = rnp->gp_seq; /* Avoid counter wrap. */
1232 [ + + ]: 62161 : trace_rcu_this_gp(rnp, rdp, rnp->gp_seq,
1233 : 62161 : needmore ? TPS("CleanupMore") : TPS("Cleanup"));
1234 : 62161 : return needmore;
1235 : : }
1236 : :
1237 : : /*
1238 : : * Awaken the grace-period kthread. Don't do a self-awaken (unless in
1239 : : * an interrupt or softirq handler), and don't bother awakening when there
1240 : : * is nothing for the grace-period kthread to do (as in several CPUs raced
1241 : : * to awaken, and we lost), and finally don't try to awaken a kthread that
1242 : : * has not yet been created. If all those checks are passed, track some
1243 : : * debug information and awaken.
1244 : : *
1245 : : * So why do the self-wakeup when in an interrupt or softirq handler
1246 : : * in the grace-period kthread's context? Because the kthread might have
1247 : : * been interrupted just as it was going to sleep, and just after the final
1248 : : * pre-sleep check of the awaken condition. In this case, a wakeup really
1249 : : * is required, and is therefore supplied.
1250 : : */
1251 : 64194 : static void rcu_gp_kthread_wake(void)
1252 : : {
1253 [ - + - - ]: 64194 : if ((current == rcu_state.gp_kthread &&
1254 [ - - + - ]: 64194 : !in_irq() && !in_serving_softirq()) ||
1255 [ + - + - ]: 64194 : !READ_ONCE(rcu_state.gp_flags) ||
1256 : : !rcu_state.gp_kthread)
1257 : : return;
1258 : 64194 : WRITE_ONCE(rcu_state.gp_wake_time, jiffies);
1259 : 64194 : WRITE_ONCE(rcu_state.gp_wake_seq, READ_ONCE(rcu_state.gp_seq));
1260 : 64194 : swake_up_one(&rcu_state.gp_wq);
1261 : : }
1262 : :
1263 : : /*
1264 : : * If there is room, assign a ->gp_seq number to any callbacks on this
1265 : : * CPU that have not already been assigned. Also accelerate any callbacks
1266 : : * that were previously assigned a ->gp_seq number that has since proven
1267 : : * to be too conservative, which can happen if callbacks get assigned a
1268 : : * ->gp_seq number while RCU is idle, but with reference to a non-root
1269 : : * rcu_node structure. This function is idempotent, so it does not hurt
1270 : : * to call it repeatedly. Returns an flag saying that we should awaken
1271 : : * the RCU grace-period kthread.
1272 : : *
1273 : : * The caller must hold rnp->lock with interrupts disabled.
1274 : : */
1275 : 248786 : static bool rcu_accelerate_cbs(struct rcu_node *rnp, struct rcu_data *rdp)
1276 : : {
1277 : 248786 : unsigned long gp_seq_req;
1278 : 248786 : bool ret = false;
1279 : :
1280 : 248786 : rcu_lockdep_assert_cblist_protected(rdp);
1281 : 248786 : raw_lockdep_assert_held_rcu_node(rnp);
1282 : :
1283 : : /* If no pending (not yet ready to invoke) callbacks, nothing to do. */
1284 [ + + ]: 248786 : if (!rcu_segcblist_pend_cbs(&rdp->cblist))
1285 : : return false;
1286 : :
1287 : : /*
1288 : : * Callbacks are often registered with incomplete grace-period
1289 : : * information. Something about the fact that getting exact
1290 : : * information requires acquiring a global lock... RCU therefore
1291 : : * makes a conservative estimate of the grace period number at which
1292 : : * a given callback will become ready to invoke. The following
1293 : : * code checks this estimate and improves it when possible, thus
1294 : : * accelerating callback invocation to an earlier grace-period
1295 : : * number.
1296 : : */
1297 : 199462 : gp_seq_req = rcu_seq_snap(&rcu_state.gp_seq);
1298 [ + - ]: 199462 : if (rcu_segcblist_accelerate(&rdp->cblist, gp_seq_req))
1299 : 199462 : ret = rcu_start_this_gp(rnp, rdp, gp_seq_req);
1300 : :
1301 : : /* Trace depending on how much we were able to accelerate. */
1302 [ + + ]: 199462 : if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL))
1303 : 177582 : trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccWaitCB"));
1304 : : else
1305 : 21880 : trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccReadyCB"));
1306 : : return ret;
1307 : : }
1308 : :
1309 : : /*
1310 : : * Similar to rcu_accelerate_cbs(), but does not require that the leaf
1311 : : * rcu_node structure's ->lock be held. It consults the cached value
1312 : : * of ->gp_seq_needed in the rcu_data structure, and if that indicates
1313 : : * that a new grace-period request be made, invokes rcu_accelerate_cbs()
1314 : : * while holding the leaf rcu_node structure's ->lock.
1315 : : */
1316 : 2147 : static void rcu_accelerate_cbs_unlocked(struct rcu_node *rnp,
1317 : : struct rcu_data *rdp)
1318 : : {
1319 : 2147 : unsigned long c;
1320 : 2147 : bool needwake;
1321 : :
1322 : 2147 : rcu_lockdep_assert_cblist_protected(rdp);
1323 : 2147 : c = rcu_seq_snap(&rcu_state.gp_seq);
1324 [ + - - + ]: 2147 : if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) {
1325 : : /* Old request still live, so mark recent callbacks. */
1326 : 0 : (void)rcu_segcblist_accelerate(&rdp->cblist, c);
1327 : 0 : return;
1328 : : }
1329 : 2147 : raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
1330 : 2147 : needwake = rcu_accelerate_cbs(rnp, rdp);
1331 : 2147 : raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */
1332 [ + + ]: 2147 : if (needwake)
1333 : 2033 : rcu_gp_kthread_wake();
1334 : : }
1335 : :
1336 : : /*
1337 : : * Move any callbacks whose grace period has completed to the
1338 : : * RCU_DONE_TAIL sublist, then compact the remaining sublists and
1339 : : * assign ->gp_seq numbers to any callbacks in the RCU_NEXT_TAIL
1340 : : * sublist. This function is idempotent, so it does not hurt to
1341 : : * invoke it repeatedly. As long as it is not invoked -too- often...
1342 : : * Returns true if the RCU grace-period kthread needs to be awakened.
1343 : : *
1344 : : * The caller must hold rnp->lock with interrupts disabled.
1345 : : */
1346 : 62161 : static bool rcu_advance_cbs(struct rcu_node *rnp, struct rcu_data *rdp)
1347 : : {
1348 : 62161 : rcu_lockdep_assert_cblist_protected(rdp);
1349 : 62161 : raw_lockdep_assert_held_rcu_node(rnp);
1350 : :
1351 : : /* If no pending (not yet ready to invoke) callbacks, nothing to do. */
1352 [ + + ]: 62161 : if (!rcu_segcblist_pend_cbs(&rdp->cblist))
1353 : : return false;
1354 : :
1355 : : /*
1356 : : * Find all callbacks whose ->gp_seq numbers indicate that they
1357 : : * are ready to invoke, and put them into the RCU_DONE_TAIL sublist.
1358 : : */
1359 : 60128 : rcu_segcblist_advance(&rdp->cblist, rnp->gp_seq);
1360 : :
1361 : : /* Classify any remaining callbacks. */
1362 : 60128 : return rcu_accelerate_cbs(rnp, rdp);
1363 : : }
1364 : :
1365 : : /*
1366 : : * Move and classify callbacks, but only if doing so won't require
1367 : : * that the RCU grace-period kthread be awakened.
1368 : : */
1369 : : static void __maybe_unused rcu_advance_cbs_nowake(struct rcu_node *rnp,
1370 : : struct rcu_data *rdp)
1371 : : {
1372 : : rcu_lockdep_assert_cblist_protected(rdp);
1373 : : if (!rcu_seq_state(rcu_seq_current(&rnp->gp_seq)) ||
1374 : : !raw_spin_trylock_rcu_node(rnp))
1375 : : return;
1376 : : WARN_ON_ONCE(rcu_advance_cbs(rnp, rdp));
1377 : : raw_spin_unlock_rcu_node(rnp);
1378 : : }
1379 : :
1380 : : /*
1381 : : * Update CPU-local rcu_data state to record the beginnings and ends of
1382 : : * grace periods. The caller must hold the ->lock of the leaf rcu_node
1383 : : * structure corresponding to the current CPU, and must have irqs disabled.
1384 : : * Returns true if the grace-period kthread needs to be awakened.
1385 : : */
1386 : 124350 : static bool __note_gp_changes(struct rcu_node *rnp, struct rcu_data *rdp)
1387 : : {
1388 : 124350 : bool ret = false;
1389 : 124350 : bool need_gp;
1390 : 124350 : const bool offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) &&
1391 : : rcu_segcblist_is_offloaded(&rdp->cblist);
1392 : :
1393 : 124350 : raw_lockdep_assert_held_rcu_node(rnp);
1394 : :
1395 [ + - ]: 124350 : if (rdp->gp_seq == rnp->gp_seq)
1396 : : return false; /* Nothing to do. */
1397 : :
1398 : : /* Handle the ends of any preceding grace periods first. */
1399 [ + + - + ]: 124350 : if (rcu_seq_completed_gp(rdp->gp_seq, rnp->gp_seq) ||
1400 [ - + ]: 62189 : unlikely(READ_ONCE(rdp->gpwrap))) {
1401 : 62161 : if (!offloaded)
1402 : 62161 : ret = rcu_advance_cbs(rnp, rdp); /* Advance CBs. */
1403 : 62161 : trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuend"));
1404 : : } else {
1405 : 62189 : if (!offloaded)
1406 : 62189 : ret = rcu_accelerate_cbs(rnp, rdp); /* Recent CBs. */
1407 : : }
1408 : :
1409 : : /* Now handle the beginnings of any new-to-this-CPU grace periods. */
1410 [ + + - + ]: 124350 : if (rcu_seq_new_gp(rdp->gp_seq, rnp->gp_seq) ||
1411 [ - + ]: 62161 : unlikely(READ_ONCE(rdp->gpwrap))) {
1412 : : /*
1413 : : * If the current grace period is waiting for this CPU,
1414 : : * set up to detect a quiescent state, otherwise don't
1415 : : * go looking for one.
1416 : : */
1417 : 62189 : trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, TPS("cpustart"));
1418 : 62189 : need_gp = !!(rnp->qsmask & rdp->grpmask);
1419 : 62189 : rdp->cpu_no_qs.b.norm = need_gp;
1420 : 62189 : rdp->core_needs_qs = need_gp;
1421 : 62189 : zero_cpu_stall_ticks(rdp);
1422 : : }
1423 : 124350 : rdp->gp_seq = rnp->gp_seq; /* Remember new grace-period state. */
1424 [ + + - + ]: 124350 : if (ULONG_CMP_LT(rdp->gp_seq_needed, rnp->gp_seq_needed) || rdp->gpwrap)
1425 : 60156 : rdp->gp_seq_needed = rnp->gp_seq_needed;
1426 [ - + ]: 124350 : WRITE_ONCE(rdp->gpwrap, false);
1427 [ - + ]: 124350 : rcu_gpnum_ovf(rnp, rdp);
1428 : : return ret;
1429 : : }
1430 : :
1431 : 230055 : static void note_gp_changes(struct rcu_data *rdp)
1432 : : {
1433 : 230055 : unsigned long flags;
1434 : 230055 : bool needwake;
1435 : 230055 : struct rcu_node *rnp;
1436 : :
1437 : 230055 : local_irq_save(flags);
1438 : 230055 : rnp = rdp->mynode;
1439 [ + - - + ]: 230055 : if ((rdp->gp_seq == rcu_seq_current(&rnp->gp_seq) &&
1440 [ - + ]: 230055 : !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */
1441 [ # # ]: 0 : !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */
1442 : 230055 : local_irq_restore(flags);
1443 : 230055 : return;
1444 : : }
1445 : 0 : needwake = __note_gp_changes(rnp, rdp);
1446 : 0 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1447 [ # # ]: 0 : if (needwake)
1448 : 0 : rcu_gp_kthread_wake();
1449 : : }
1450 : :
1451 : 186539 : static void rcu_gp_slow(int delay)
1452 : : {
1453 [ - + ]: 186539 : if (delay > 0 &&
1454 [ # # ]: 0 : !(rcu_seq_ctr(rcu_state.gp_seq) %
1455 [ # # ]: 0 : (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay)))
1456 : 0 : schedule_timeout_uninterruptible(delay);
1457 : 186539 : }
1458 : :
1459 : : /*
1460 : : * Initialize a new grace period. Return false if no grace period required.
1461 : : */
1462 : 62187 : static bool rcu_gp_init(void)
1463 : : {
1464 : 62187 : unsigned long flags;
1465 : 62187 : unsigned long oldmask;
1466 : 62187 : unsigned long mask;
1467 : 62187 : struct rcu_data *rdp;
1468 : 62187 : struct rcu_node *rnp = rcu_get_root();
1469 : :
1470 : 62187 : WRITE_ONCE(rcu_state.gp_activity, jiffies);
1471 : 62187 : raw_spin_lock_irq_rcu_node(rnp);
1472 [ - + ]: 62187 : if (!READ_ONCE(rcu_state.gp_flags)) {
1473 : : /* Spurious wakeup, tell caller to go back to sleep. */
1474 : 0 : raw_spin_unlock_irq_rcu_node(rnp);
1475 : 0 : return false;
1476 : : }
1477 [ - + ]: 62187 : WRITE_ONCE(rcu_state.gp_flags, 0); /* Clear all flags: New GP. */
1478 : :
1479 [ - + - + ]: 62187 : if (WARN_ON_ONCE(rcu_gp_in_progress())) {
1480 : : /*
1481 : : * Grace period already in progress, don't start another.
1482 : : * Not supposed to be able to happen.
1483 : : */
1484 : 0 : raw_spin_unlock_irq_rcu_node(rnp);
1485 : 0 : return false;
1486 : : }
1487 : :
1488 : : /* Advance to a new grace period and initialize state. */
1489 : 62187 : record_gp_stall_check_time();
1490 : : /* Record GP times before starting GP, hence rcu_seq_start(). */
1491 : 62187 : rcu_seq_start(&rcu_state.gp_seq);
1492 : 62187 : trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("start"));
1493 : 62187 : raw_spin_unlock_irq_rcu_node(rnp);
1494 : :
1495 : : /*
1496 : : * Apply per-leaf buffered online and offline operations to the
1497 : : * rcu_node tree. Note that this new grace period need not wait
1498 : : * for subsequent online CPUs, and that quiescent-state forcing
1499 : : * will handle subsequent offline CPUs.
1500 : : */
1501 : 62187 : rcu_state.gp_state = RCU_GP_ONOFF;
1502 [ + + ]: 124374 : rcu_for_each_leaf_node(rnp) {
1503 : 62187 : raw_spin_lock(&rcu_state.ofl_lock);
1504 : 62187 : raw_spin_lock_irq_rcu_node(rnp);
1505 [ + + ]: 62187 : if (rnp->qsmaskinit == rnp->qsmaskinitnext &&
1506 [ + - ]: 62159 : !rnp->wait_blkd_tasks) {
1507 : : /* Nothing to do on this leaf rcu_node structure. */
1508 : 62159 : raw_spin_unlock_irq_rcu_node(rnp);
1509 : 62159 : raw_spin_unlock(&rcu_state.ofl_lock);
1510 : 62159 : continue;
1511 : : }
1512 : :
1513 : : /* Record old state, apply changes to ->qsmaskinit field. */
1514 : 28 : oldmask = rnp->qsmaskinit;
1515 : 28 : rnp->qsmaskinit = rnp->qsmaskinitnext;
1516 : :
1517 : : /* If zero-ness of ->qsmaskinit changed, propagate up tree. */
1518 [ + - ]: 28 : if (!oldmask != !rnp->qsmaskinit) {
1519 [ + - ]: 28 : if (!oldmask) { /* First online CPU for rcu_node. */
1520 [ + - ]: 28 : if (!rnp->wait_blkd_tasks) /* Ever offline? */
1521 : 28 : rcu_init_new_rnp(rnp);
1522 : 0 : } else if (rcu_preempt_has_tasks(rnp)) {
1523 : : rnp->wait_blkd_tasks = true; /* blocked tasks */
1524 : : } else { /* Last offline CPU and can propagate. */
1525 : 0 : rcu_cleanup_dead_rnp(rnp);
1526 : : }
1527 : : }
1528 : :
1529 : : /*
1530 : : * If all waited-on tasks from prior grace period are
1531 : : * done, and if all this rcu_node structure's CPUs are
1532 : : * still offline, propagate up the rcu_node tree and
1533 : : * clear ->wait_blkd_tasks. Otherwise, if one of this
1534 : : * rcu_node structure's CPUs has since come back online,
1535 : : * simply clear ->wait_blkd_tasks.
1536 : : */
1537 [ - + ]: 28 : if (rnp->wait_blkd_tasks &&
1538 : : (!rcu_preempt_has_tasks(rnp) || rnp->qsmaskinit)) {
1539 : 0 : rnp->wait_blkd_tasks = false;
1540 [ # # ]: 0 : if (!rnp->qsmaskinit)
1541 : 0 : rcu_cleanup_dead_rnp(rnp);
1542 : : }
1543 : :
1544 : 28 : raw_spin_unlock_irq_rcu_node(rnp);
1545 : 28 : raw_spin_unlock(&rcu_state.ofl_lock);
1546 : : }
1547 : 62187 : rcu_gp_slow(gp_preinit_delay); /* Races with CPU hotplug. */
1548 : :
1549 : : /*
1550 : : * Set the quiescent-state-needed bits in all the rcu_node
1551 : : * structures for all currently online CPUs in breadth-first
1552 : : * order, starting from the root rcu_node structure, relying on the
1553 : : * layout of the tree within the rcu_state.node[] array. Note that
1554 : : * other CPUs will access only the leaves of the hierarchy, thus
1555 : : * seeing that no grace period is in progress, at least until the
1556 : : * corresponding leaf node has been initialized.
1557 : : *
1558 : : * The grace period cannot complete until the initialization
1559 : : * process finishes, because this kthread handles both.
1560 : : */
1561 : 62187 : rcu_state.gp_state = RCU_GP_INIT;
1562 [ + + ]: 124374 : rcu_for_each_node_breadth_first(rnp) {
1563 : 62187 : rcu_gp_slow(gp_init_delay);
1564 : 62187 : raw_spin_lock_irqsave_rcu_node(rnp, flags);
1565 : 62187 : rdp = this_cpu_ptr(&rcu_data);
1566 [ - + ]: 62187 : rcu_preempt_check_blocked_tasks(rnp);
1567 : 62187 : rnp->qsmask = rnp->qsmaskinit;
1568 [ + - ]: 62187 : WRITE_ONCE(rnp->gp_seq, rcu_state.gp_seq);
1569 [ + - ]: 62187 : if (rnp == rdp->mynode)
1570 : 62187 : (void)__note_gp_changes(rnp, rdp);
1571 : 62187 : rcu_preempt_boost_start_gp(rnp);
1572 : 62187 : trace_rcu_grace_period_init(rcu_state.name, rnp->gp_seq,
1573 : 62187 : rnp->level, rnp->grplo,
1574 : : rnp->grphi, rnp->qsmask);
1575 : : /* Quiescent states for tasks on any now-offline CPUs. */
1576 : 62187 : mask = rnp->qsmask & ~rnp->qsmaskinitnext;
1577 : 62187 : rnp->rcu_gp_init_mask = mask;
1578 [ + - - + : 62187 : if ((mask || rnp->wait_blkd_tasks) && rcu_is_leaf_node(rnp))
- - ]
1579 : 0 : rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags);
1580 : : else
1581 : 62187 : raw_spin_unlock_irq_rcu_node(rnp);
1582 : 62187 : cond_resched_tasks_rcu_qs();
1583 : 62187 : WRITE_ONCE(rcu_state.gp_activity, jiffies);
1584 : : }
1585 : :
1586 : : return true;
1587 : : }
1588 : :
1589 : : /*
1590 : : * Helper function for swait_event_idle_exclusive() wakeup at force-quiescent-state
1591 : : * time.
1592 : : */
1593 : 186533 : static bool rcu_gp_fqs_check_wake(int *gfp)
1594 : : {
1595 : 186533 : struct rcu_node *rnp = rcu_get_root();
1596 : :
1597 : : /* Someone like call_rcu() requested a force-quiescent-state scan. */
1598 : 186533 : *gfp = READ_ONCE(rcu_state.gp_flags);
1599 [ + - + + ]: 186533 : if (*gfp & RCU_GP_FLAG_FQS)
1600 : : return true;
1601 : :
1602 : : /* The current grace period has completed. */
1603 [ - + - + ]: 124374 : if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp))
1604 : 0 : return true;
1605 : :
1606 : : return false;
1607 : : }
1608 : :
1609 : : /*
1610 : : * Do one round of quiescent-state forcing.
1611 : : */
1612 : 0 : static void rcu_gp_fqs(bool first_time)
1613 : : {
1614 : 0 : struct rcu_node *rnp = rcu_get_root();
1615 : :
1616 [ # # ]: 0 : WRITE_ONCE(rcu_state.gp_activity, jiffies);
1617 : 0 : rcu_state.n_force_qs++;
1618 [ # # ]: 0 : if (first_time) {
1619 : : /* Collect dyntick-idle snapshots. */
1620 : 0 : force_qs_rnp(dyntick_save_progress_counter);
1621 : : } else {
1622 : : /* Handle dyntick-idle and offline CPUs. */
1623 : 0 : force_qs_rnp(rcu_implicit_dynticks_qs);
1624 : : }
1625 : : /* Clear flag to prevent immediate re-entry. */
1626 [ # # ]: 0 : if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) {
1627 : 0 : raw_spin_lock_irq_rcu_node(rnp);
1628 : 0 : WRITE_ONCE(rcu_state.gp_flags,
1629 : : READ_ONCE(rcu_state.gp_flags) & ~RCU_GP_FLAG_FQS);
1630 : 0 : raw_spin_unlock_irq_rcu_node(rnp);
1631 : : }
1632 : 0 : }
1633 : :
1634 : : /*
1635 : : * Loop doing repeated quiescent-state forcing until the grace period ends.
1636 : : */
1637 : 62187 : static void rcu_gp_fqs_loop(void)
1638 : : {
1639 : 62187 : bool first_gp_fqs;
1640 : 62187 : int gf;
1641 : 62187 : unsigned long j;
1642 : 62187 : int ret;
1643 : 62187 : struct rcu_node *rnp = rcu_get_root();
1644 : :
1645 : 62187 : first_gp_fqs = true;
1646 : 62187 : j = READ_ONCE(jiffies_till_first_fqs);
1647 : 62187 : ret = 0;
1648 : 62187 : for (;;) {
1649 [ + - ]: 62187 : if (!ret) {
1650 : 62187 : rcu_state.jiffies_force_qs = jiffies + j;
1651 [ + - ]: 62187 : WRITE_ONCE(rcu_state.jiffies_kick_kthreads,
1652 : : jiffies + (j ? 3 * j : 2));
1653 : : }
1654 : 62187 : trace_rcu_grace_period(rcu_state.name,
1655 : 62187 : READ_ONCE(rcu_state.gp_seq),
1656 : 62187 : TPS("fqswait"));
1657 : 62187 : rcu_state.gp_state = RCU_GP_WAIT_FQS;
1658 [ + - - + : 248720 : ret = swait_event_idle_timeout_exclusive(
+ - + + +
+ + + ]
1659 : : rcu_state.gp_wq, rcu_gp_fqs_check_wake(&gf), j);
1660 : 62159 : rcu_state.gp_state = RCU_GP_DOING_FQS;
1661 : : /* Locking provides needed memory barriers. */
1662 : : /* If grace period done, leave loop. */
1663 [ - + ]: 62159 : if (!READ_ONCE(rnp->qsmask) &&
1664 : : !rcu_preempt_blocked_readers_cgp(rnp))
1665 : : break;
1666 : : /* If time for quiescent-state forcing, do it. */
1667 [ # # ]: 0 : if (ULONG_CMP_GE(jiffies, rcu_state.jiffies_force_qs) ||
1668 [ # # ]: 0 : (gf & RCU_GP_FLAG_FQS)) {
1669 : 0 : trace_rcu_grace_period(rcu_state.name,
1670 : 0 : READ_ONCE(rcu_state.gp_seq),
1671 : 0 : TPS("fqsstart"));
1672 : 0 : rcu_gp_fqs(first_gp_fqs);
1673 : 0 : first_gp_fqs = false;
1674 : 0 : trace_rcu_grace_period(rcu_state.name,
1675 : 0 : READ_ONCE(rcu_state.gp_seq),
1676 : 0 : TPS("fqsend"));
1677 : 0 : cond_resched_tasks_rcu_qs();
1678 : 0 : WRITE_ONCE(rcu_state.gp_activity, jiffies);
1679 : 0 : ret = 0; /* Force full wait till next FQS. */
1680 : 0 : j = READ_ONCE(jiffies_till_next_fqs);
1681 : : } else {
1682 : : /* Deal with stray signal. */
1683 : 0 : cond_resched_tasks_rcu_qs();
1684 : 0 : WRITE_ONCE(rcu_state.gp_activity, jiffies);
1685 [ # # ]: 0 : WARN_ON(signal_pending(current));
1686 : 0 : trace_rcu_grace_period(rcu_state.name,
1687 : 0 : READ_ONCE(rcu_state.gp_seq),
1688 : 0 : TPS("fqswaitsig"));
1689 : 0 : ret = 1; /* Keep old FQS timing. */
1690 : 0 : j = jiffies;
1691 [ # # ]: 0 : if (time_after(jiffies, rcu_state.jiffies_force_qs))
1692 : : j = 1;
1693 : : else
1694 : 0 : j = rcu_state.jiffies_force_qs - j;
1695 : : }
1696 : : }
1697 : 62159 : }
1698 : :
1699 : : /*
1700 : : * Clean up after the old grace period.
1701 : : */
1702 : 62159 : static void rcu_gp_cleanup(void)
1703 : : {
1704 : 62159 : unsigned long gp_duration;
1705 : 62159 : bool needgp = false;
1706 : 62159 : unsigned long new_gp_seq;
1707 : 62159 : bool offloaded;
1708 : 62159 : struct rcu_data *rdp;
1709 : 62159 : struct rcu_node *rnp = rcu_get_root();
1710 : 62159 : struct swait_queue_head *sq;
1711 : :
1712 : 62159 : WRITE_ONCE(rcu_state.gp_activity, jiffies);
1713 : 62159 : raw_spin_lock_irq_rcu_node(rnp);
1714 : 62159 : rcu_state.gp_end = jiffies;
1715 : 62159 : gp_duration = rcu_state.gp_end - rcu_state.gp_start;
1716 [ + + ]: 62159 : if (gp_duration > rcu_state.gp_max)
1717 : 102 : rcu_state.gp_max = gp_duration;
1718 : :
1719 : : /*
1720 : : * We know the grace period is complete, but to everyone else
1721 : : * it appears to still be ongoing. But it is also the case
1722 : : * that to everyone else it looks like there is nothing that
1723 : : * they can do to advance the grace period. It is therefore
1724 : : * safe for us to drop the lock in order to mark the grace
1725 : : * period as completed in all of the rcu_node structures.
1726 : : */
1727 : 62159 : raw_spin_unlock_irq_rcu_node(rnp);
1728 : :
1729 : : /*
1730 : : * Propagate new ->gp_seq value to rcu_node structures so that
1731 : : * other CPUs don't have to wait until the start of the next grace
1732 : : * period to process their callbacks. This also avoids some nasty
1733 : : * RCU grace-period initialization races by forcing the end of
1734 : : * the current grace period to be completely recorded in all of
1735 : : * the rcu_node structures before the beginning of the next grace
1736 : : * period is recorded in any of the rcu_node structures.
1737 : : */
1738 : 62159 : new_gp_seq = rcu_state.gp_seq;
1739 : 62159 : rcu_seq_end(&new_gp_seq);
1740 [ + + ]: 124318 : rcu_for_each_node_breadth_first(rnp) {
1741 : 62159 : raw_spin_lock_irq_rcu_node(rnp);
1742 [ - + ]: 62159 : if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)))
1743 : : dump_blkd_tasks(rnp, 10);
1744 [ - + ]: 62159 : WARN_ON_ONCE(rnp->qsmask);
1745 : 62159 : WRITE_ONCE(rnp->gp_seq, new_gp_seq);
1746 : 62159 : rdp = this_cpu_ptr(&rcu_data);
1747 [ + - ]: 62159 : if (rnp == rdp->mynode)
1748 [ + - + - ]: 124318 : needgp = __note_gp_changes(rnp, rdp) || needgp;
1749 : : /* smp_mb() provided by prior unlock-lock pair. */
1750 [ + + + - ]: 62159 : needgp = rcu_future_gp_cleanup(rnp) || needgp;
1751 : 62159 : sq = rcu_nocb_gp_get(rnp);
1752 : 62159 : raw_spin_unlock_irq_rcu_node(rnp);
1753 : 62159 : rcu_nocb_gp_cleanup(sq);
1754 : 62159 : cond_resched_tasks_rcu_qs();
1755 : 62159 : WRITE_ONCE(rcu_state.gp_activity, jiffies);
1756 : 62159 : rcu_gp_slow(gp_cleanup_delay);
1757 : : }
1758 : 62159 : rnp = rcu_get_root();
1759 : 62159 : raw_spin_lock_irq_rcu_node(rnp); /* GP before ->gp_seq update. */
1760 : :
1761 : : /* Declare grace period done, trace first to use old GP number. */
1762 : 62159 : trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("end"));
1763 : 62159 : rcu_seq_end(&rcu_state.gp_seq);
1764 : 62159 : rcu_state.gp_state = RCU_GP_IDLE;
1765 : : /* Check for GP requests since above loop. */
1766 : 62159 : rdp = this_cpu_ptr(&rcu_data);
1767 [ + + - + ]: 62159 : if (!needgp && ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed)) {
1768 : 0 : trace_rcu_this_gp(rnp, rdp, rnp->gp_seq_needed,
1769 : 0 : TPS("CleanupMore"));
1770 : 0 : needgp = true;
1771 : : }
1772 : : /* Advance CBs to reduce false positives below. */
1773 : 62159 : offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) &&
1774 : : rcu_segcblist_is_offloaded(&rdp->cblist);
1775 [ + - + + ]: 62159 : if ((offloaded || !rcu_accelerate_cbs(rnp, rdp)) && needgp) {
1776 : 60126 : WRITE_ONCE(rcu_state.gp_flags, RCU_GP_FLAG_INIT);
1777 : 60126 : rcu_state.gp_req_activity = jiffies;
1778 : 60126 : trace_rcu_grace_period(rcu_state.name,
1779 : 60126 : READ_ONCE(rcu_state.gp_seq),
1780 : 60126 : TPS("newreq"));
1781 : : } else {
1782 : 2033 : WRITE_ONCE(rcu_state.gp_flags,
1783 : : rcu_state.gp_flags & RCU_GP_FLAG_INIT);
1784 : : }
1785 : 62159 : raw_spin_unlock_irq_rcu_node(rnp);
1786 : 62159 : }
1787 : :
1788 : : /*
1789 : : * Body of kthread that handles grace periods.
1790 : : */
1791 : 28 : static int __noreturn rcu_gp_kthread(void *unused)
1792 : : {
1793 : 28 : rcu_bind_gp_kthread();
1794 : 124346 : for (;;) {
1795 : :
1796 : : /* Handle grace-period start. */
1797 : 0 : for (;;) {
1798 : 0 : trace_rcu_grace_period(rcu_state.name,
1799 : 62187 : READ_ONCE(rcu_state.gp_seq),
1800 : 62187 : TPS("reqwait"));
1801 : 62187 : rcu_state.gp_state = RCU_GP_WAIT_GPS;
1802 [ + + + + ]: 64220 : swait_event_idle_exclusive(rcu_state.gp_wq,
1803 : : READ_ONCE(rcu_state.gp_flags) &
1804 : : RCU_GP_FLAG_INIT);
1805 : 62187 : rcu_state.gp_state = RCU_GP_DONE_GPS;
1806 : : /* Locking provides needed memory barrier. */
1807 [ - + ]: 62187 : if (rcu_gp_init())
1808 : : break;
1809 : 0 : cond_resched_tasks_rcu_qs();
1810 : 0 : WRITE_ONCE(rcu_state.gp_activity, jiffies);
1811 [ # # ]: 0 : WARN_ON(signal_pending(current));
1812 : 0 : trace_rcu_grace_period(rcu_state.name,
1813 : 0 : READ_ONCE(rcu_state.gp_seq),
1814 : 0 : TPS("reqwaitsig"));
1815 : : }
1816 : :
1817 : : /* Handle quiescent-state forcing. */
1818 : 62187 : rcu_gp_fqs_loop();
1819 : :
1820 : : /* Handle grace-period end. */
1821 : 62159 : rcu_state.gp_state = RCU_GP_CLEANUP;
1822 : 62159 : rcu_gp_cleanup();
1823 : 62159 : rcu_state.gp_state = RCU_GP_CLEANED;
1824 : : }
1825 : : }
1826 : :
1827 : : /*
1828 : : * Report a full set of quiescent states to the rcu_state data structure.
1829 : : * Invoke rcu_gp_kthread_wake() to awaken the grace-period kthread if
1830 : : * another grace period is required. Whether we wake the grace-period
1831 : : * kthread or it awakens itself for the next round of quiescent-state
1832 : : * forcing, that kthread will clean up after the just-completed grace
1833 : : * period. Note that the caller must hold rnp->lock, which is released
1834 : : * before return.
1835 : : */
1836 : 62159 : static void rcu_report_qs_rsp(unsigned long flags)
1837 : : __releases(rcu_get_root()->lock)
1838 : : {
1839 : 62159 : raw_lockdep_assert_held_rcu_node(rcu_get_root());
1840 [ - + ]: 62159 : WARN_ON_ONCE(!rcu_gp_in_progress());
1841 : 62159 : WRITE_ONCE(rcu_state.gp_flags,
1842 : : READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS);
1843 : 62159 : raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(), flags);
1844 : 62159 : rcu_gp_kthread_wake();
1845 : 62159 : }
1846 : :
1847 : : /*
1848 : : * Similar to rcu_report_qs_rdp(), for which it is a helper function.
1849 : : * Allows quiescent states for a group of CPUs to be reported at one go
1850 : : * to the specified rcu_node structure, though all the CPUs in the group
1851 : : * must be represented by the same rcu_node structure (which need not be a
1852 : : * leaf rcu_node structure, though it often will be). The gps parameter
1853 : : * is the grace-period snapshot, which means that the quiescent states
1854 : : * are valid only if rnp->gp_seq is equal to gps. That structure's lock
1855 : : * must be held upon entry, and it is released before return.
1856 : : *
1857 : : * As a special case, if mask is zero, the bit-already-cleared check is
1858 : : * disabled. This allows propagating quiescent state due to resumed tasks
1859 : : * during grace-period initialization.
1860 : : */
1861 : 62159 : static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp,
1862 : : unsigned long gps, unsigned long flags)
1863 : : __releases(rnp->lock)
1864 : : {
1865 : 62159 : unsigned long oldmask = 0;
1866 : 62159 : struct rcu_node *rnp_c;
1867 : :
1868 : 62159 : raw_lockdep_assert_held_rcu_node(rnp);
1869 : :
1870 : : /* Walk up the rcu_node hierarchy. */
1871 : 62159 : for (;;) {
1872 [ - + - - : 62159 : if ((!(rnp->qsmask & mask) && mask) || rnp->gp_seq != gps) {
- + ]
1873 : :
1874 : : /*
1875 : : * Our bit has already been cleared, or the
1876 : : * relevant grace period is already over, so done.
1877 : : */
1878 : 0 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1879 : 0 : return;
1880 : : }
1881 [ - + ]: 62159 : WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */
1882 : 62159 : WARN_ON_ONCE(!rcu_is_leaf_node(rnp) &&
1883 : : rcu_preempt_blocked_readers_cgp(rnp));
1884 : 62159 : rnp->qsmask &= ~mask;
1885 : 62159 : trace_rcu_quiescent_state_report(rcu_state.name, rnp->gp_seq,
1886 : : mask, rnp->qsmask, rnp->level,
1887 : : rnp->grplo, rnp->grphi,
1888 : 62159 : !!rnp->gp_tasks);
1889 [ - + ]: 62159 : if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
1890 : :
1891 : : /* Other bits still set at this level, so done. */
1892 : 0 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1893 : 0 : return;
1894 : : }
1895 : 62159 : rnp->completedqs = rnp->gp_seq;
1896 : 62159 : mask = rnp->grpmask;
1897 [ - + ]: 62159 : if (rnp->parent == NULL) {
1898 : :
1899 : : /* No more levels. Exit loop holding root lock. */
1900 : :
1901 : : break;
1902 : : }
1903 : 0 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1904 : 0 : rnp_c = rnp;
1905 : 0 : rnp = rnp->parent;
1906 : 0 : raw_spin_lock_irqsave_rcu_node(rnp, flags);
1907 : 0 : oldmask = rnp_c->qsmask;
1908 : : }
1909 : :
1910 : : /*
1911 : : * Get here if we are the last CPU to pass through a quiescent
1912 : : * state for this grace period. Invoke rcu_report_qs_rsp()
1913 : : * to clean up and start the next grace period if one is needed.
1914 : : */
1915 : 62159 : rcu_report_qs_rsp(flags); /* releases rnp->lock. */
1916 : : }
1917 : :
1918 : : /*
1919 : : * Record a quiescent state for all tasks that were previously queued
1920 : : * on the specified rcu_node structure and that were blocking the current
1921 : : * RCU grace period. The caller must hold the corresponding rnp->lock with
1922 : : * irqs disabled, and this lock is released upon return, but irqs remain
1923 : : * disabled.
1924 : : */
1925 : : static void __maybe_unused
1926 : : rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags)
1927 : : __releases(rnp->lock)
1928 : : {
1929 : : unsigned long gps;
1930 : : unsigned long mask;
1931 : : struct rcu_node *rnp_p;
1932 : :
1933 : : raw_lockdep_assert_held_rcu_node(rnp);
1934 : : if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPT_RCU)) ||
1935 : : WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)) ||
1936 : : rnp->qsmask != 0) {
1937 : : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1938 : : return; /* Still need more quiescent states! */
1939 : : }
1940 : :
1941 : : rnp->completedqs = rnp->gp_seq;
1942 : : rnp_p = rnp->parent;
1943 : : if (rnp_p == NULL) {
1944 : : /*
1945 : : * Only one rcu_node structure in the tree, so don't
1946 : : * try to report up to its nonexistent parent!
1947 : : */
1948 : : rcu_report_qs_rsp(flags);
1949 : : return;
1950 : : }
1951 : :
1952 : : /* Report up the rest of the hierarchy, tracking current ->gp_seq. */
1953 : : gps = rnp->gp_seq;
1954 : : mask = rnp->grpmask;
1955 : : raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */
1956 : : raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */
1957 : : rcu_report_qs_rnp(mask, rnp_p, gps, flags);
1958 : : }
1959 : :
1960 : : /*
1961 : : * Record a quiescent state for the specified CPU to that CPU's rcu_data
1962 : : * structure. This must be called from the specified CPU.
1963 : : */
1964 : : static void
1965 : : rcu_report_qs_rdp(int cpu, struct rcu_data *rdp)
1966 : : {
1967 : : unsigned long flags;
1968 : : unsigned long mask;
1969 : : bool needwake = false;
1970 : : const bool offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) &&
1971 : : rcu_segcblist_is_offloaded(&rdp->cblist);
1972 : : struct rcu_node *rnp;
1973 : :
1974 : : rnp = rdp->mynode;
1975 : : raw_spin_lock_irqsave_rcu_node(rnp, flags);
1976 : : if (rdp->cpu_no_qs.b.norm || rdp->gp_seq != rnp->gp_seq ||
1977 : : rdp->gpwrap) {
1978 : :
1979 : : /*
1980 : : * The grace period in which this quiescent state was
1981 : : * recorded has ended, so don't report it upwards.
1982 : : * We will instead need a new quiescent state that lies
1983 : : * within the current grace period.
1984 : : */
1985 : : rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */
1986 : : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1987 : : return;
1988 : : }
1989 : : mask = rdp->grpmask;
1990 : : if ((rnp->qsmask & mask) == 0) {
1991 : : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1992 : : } else {
1993 : : /*
1994 : : * This GP can't end until cpu checks in, so all of our
1995 : : * callbacks can be processed during the next GP.
1996 : : */
1997 : : if (!offloaded)
1998 : : needwake = rcu_accelerate_cbs(rnp, rdp);
1999 : :
2000 : : rcu_disable_urgency_upon_qs(rdp);
2001 : : rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags);
2002 : : /* ^^^ Released rnp->lock */
2003 : : if (needwake)
2004 : : rcu_gp_kthread_wake();
2005 : : }
2006 : : }
2007 : :
2008 : : /*
2009 : : * Check to see if there is a new grace period of which this CPU
2010 : : * is not yet aware, and if so, set up local rcu_data state for it.
2011 : : * Otherwise, see if this CPU has just passed through its first
2012 : : * quiescent state for this grace period, and record that fact if so.
2013 : : */
2014 : : static void
2015 : 230048 : rcu_check_quiescent_state(struct rcu_data *rdp)
2016 : : {
2017 : : /* Check for grace-period ends and beginnings. */
2018 : 230048 : note_gp_changes(rdp);
2019 : :
2020 : : /*
2021 : : * Does this CPU still need to do its part for current grace period?
2022 : : * If no, return and let the other CPUs do their part as well.
2023 : : */
2024 [ + + ]: 230048 : if (!rdp->core_needs_qs)
2025 : : return;
2026 : :
2027 : : /*
2028 : : * Was there a quiescent state since the beginning of the grace
2029 : : * period? If no, then exit and wait for the next call.
2030 : : */
2031 [ + + ]: 230020 : if (rdp->cpu_no_qs.b.norm)
2032 : : return;
2033 : :
2034 : : /*
2035 : : * Tell RCU we are done (but rcu_report_qs_rdp() will be the
2036 : : * judge of that).
2037 : : */
2038 : 230010 : rcu_report_qs_rdp(rdp->cpu, rdp);
2039 : : }
2040 : :
2041 : : /*
2042 : : * Near the end of the offline process. Trace the fact that this CPU
2043 : : * is going offline.
2044 : : */
2045 : 0 : int rcutree_dying_cpu(unsigned int cpu)
2046 : : {
2047 : 0 : bool blkd;
2048 : 0 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
2049 : 0 : struct rcu_node *rnp = rdp->mynode;
2050 : :
2051 : 0 : if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
2052 : : return 0;
2053 : :
2054 : 0 : blkd = !!(rnp->qsmask & rdp->grpmask);
2055 [ # # ]: 0 : trace_rcu_grace_period(rcu_state.name, rnp->gp_seq,
2056 : 0 : blkd ? TPS("cpuofl") : TPS("cpuofl-bgp"));
2057 : 0 : return 0;
2058 : : }
2059 : :
2060 : : /*
2061 : : * All CPUs for the specified rcu_node structure have gone offline,
2062 : : * and all tasks that were preempted within an RCU read-side critical
2063 : : * section while running on one of those CPUs have since exited their RCU
2064 : : * read-side critical section. Some other CPU is reporting this fact with
2065 : : * the specified rcu_node structure's ->lock held and interrupts disabled.
2066 : : * This function therefore goes up the tree of rcu_node structures,
2067 : : * clearing the corresponding bits in the ->qsmaskinit fields. Note that
2068 : : * the leaf rcu_node structure's ->qsmaskinit field has already been
2069 : : * updated.
2070 : : *
2071 : : * This function does check that the specified rcu_node structure has
2072 : : * all CPUs offline and no blocked tasks, so it is OK to invoke it
2073 : : * prematurely. That said, invoking it after the fact will cost you
2074 : : * a needless lock acquisition. So once it has done its work, don't
2075 : : * invoke it again.
2076 : : */
2077 : 0 : static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf)
2078 : : {
2079 : 0 : long mask;
2080 : 0 : struct rcu_node *rnp = rnp_leaf;
2081 : :
2082 : 0 : raw_lockdep_assert_held_rcu_node(rnp_leaf);
2083 : 0 : if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) ||
2084 [ # # # # ]: 0 : WARN_ON_ONCE(rnp_leaf->qsmaskinit) ||
2085 : : WARN_ON_ONCE(rcu_preempt_has_tasks(rnp_leaf)))
2086 : : return;
2087 : 0 : for (;;) {
2088 : 0 : mask = rnp->grpmask;
2089 : 0 : rnp = rnp->parent;
2090 [ # # ]: 0 : if (!rnp)
2091 : : break;
2092 : 0 : raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
2093 : 0 : rnp->qsmaskinit &= ~mask;
2094 : : /* Between grace periods, so better already be zero! */
2095 [ # # ]: 0 : WARN_ON_ONCE(rnp->qsmask);
2096 [ # # ]: 0 : if (rnp->qsmaskinit) {
2097 : 0 : raw_spin_unlock_rcu_node(rnp);
2098 : : /* irqs remain disabled. */
2099 : 0 : return;
2100 : : }
2101 : 0 : raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */
2102 : : }
2103 : : }
2104 : :
2105 : : /*
2106 : : * The CPU has been completely removed, and some other CPU is reporting
2107 : : * this fact from process context. Do the remainder of the cleanup.
2108 : : * There can only be one CPU hotplug operation at a time, so no need for
2109 : : * explicit locking.
2110 : : */
2111 : 0 : int rcutree_dead_cpu(unsigned int cpu)
2112 : : {
2113 : 0 : struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
2114 : 0 : struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */
2115 : :
2116 : 0 : if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
2117 : : return 0;
2118 : :
2119 : : /* Adjust any no-longer-needed kthreads. */
2120 : 0 : rcu_boost_kthread_setaffinity(rnp, -1);
2121 : : /* Do any needed no-CB deferred wakeups from this CPU. */
2122 : 0 : do_nocb_deferred_wakeup(per_cpu_ptr(&rcu_data, cpu));
2123 : :
2124 : : // Stop-machine done, so allow nohz_full to disable tick.
2125 : 0 : tick_dep_clear(TICK_DEP_BIT_RCU);
2126 : 0 : return 0;
2127 : : }
2128 : :
2129 : : /*
2130 : : * Invoke any RCU callbacks that have made it to the end of their grace
2131 : : * period. Thottle as specified by rdp->blimit.
2132 : : */
2133 : 221185 : static void rcu_do_batch(struct rcu_data *rdp)
2134 : : {
2135 : 221185 : unsigned long flags;
2136 : 221185 : const bool offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) &&
2137 : : rcu_segcblist_is_offloaded(&rdp->cblist);
2138 : 221185 : struct rcu_head *rhp;
2139 : 221185 : struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl);
2140 : 221185 : long bl, count;
2141 : 221185 : long pending, tlimit = 0;
2142 : :
2143 : : /* If no callbacks are ready, just return. */
2144 [ - + ]: 221185 : if (!rcu_segcblist_ready_cbs(&rdp->cblist)) {
2145 : 0 : trace_rcu_batch_start(rcu_state.name,
2146 : : rcu_segcblist_n_cbs(&rdp->cblist), 0);
2147 : 0 : trace_rcu_batch_end(rcu_state.name, 0,
2148 : 0 : !rcu_segcblist_empty(&rdp->cblist),
2149 : : need_resched(), is_idle_task(current),
2150 : : rcu_is_callbacks_kthread());
2151 : 0 : return;
2152 : : }
2153 : :
2154 : : /*
2155 : : * Extract the list of ready callbacks, disabling to prevent
2156 : : * races with call_rcu() from interrupt handlers. Leave the
2157 : : * callback counts, as rcu_barrier() needs to be conservative.
2158 : : */
2159 : 221185 : local_irq_save(flags);
2160 : 221185 : rcu_nocb_lock(rdp);
2161 [ - + ]: 221185 : WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
2162 [ - + ]: 221185 : pending = rcu_segcblist_n_cbs(&rdp->cblist);
2163 : 221185 : bl = max(rdp->blimit, pending >> rcu_divisor);
2164 [ - + ]: 221185 : if (unlikely(bl > 100))
2165 : 0 : tlimit = local_clock() + rcu_resched_ns;
2166 : 221185 : trace_rcu_batch_start(rcu_state.name,
2167 : : rcu_segcblist_n_cbs(&rdp->cblist), bl);
2168 : 221185 : rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl);
2169 : 221185 : if (offloaded)
2170 : : rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist);
2171 : 221185 : rcu_nocb_unlock_irqrestore(rdp, flags);
2172 : :
2173 : : /* Invoke callbacks. */
2174 : 221185 : tick_dep_set_task(current, TICK_DEP_BIT_RCU);
2175 : 221185 : rhp = rcu_cblist_dequeue(&rcl);
2176 [ + + ]: 2305066 : for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) {
2177 : 2052350 : rcu_callback_t f;
2178 : :
2179 : 2052350 : debug_rcu_head_unqueue(rhp);
2180 : :
2181 : 2052350 : rcu_lock_acquire(&rcu_callback_map);
2182 : 2052350 : trace_rcu_invoke_callback(rcu_state.name, rhp);
2183 : :
2184 : 2052350 : f = rhp->func;
2185 : 2052350 : WRITE_ONCE(rhp->func, (rcu_callback_t)0L);
2186 : 2052350 : f(rhp);
2187 : :
2188 : 2052350 : rcu_lock_release(&rcu_callback_map);
2189 : :
2190 : : /*
2191 : : * Stop only if limit reached and CPU has something to do.
2192 : : * Note: The rcl structure counts down from zero.
2193 : : */
2194 [ + + + + ]: 2242004 : if (-rcl.len >= bl && !offloaded &&
2195 [ - + ]: 166328 : (need_resched() ||
2196 : : (!is_idle_task(current) && !rcu_is_callbacks_kthread())))
2197 : : break;
2198 [ - + ]: 1862696 : if (unlikely(tlimit)) {
2199 : : /* only call local_clock() every 32 callbacks */
2200 [ # # # # ]: 0 : if (likely((-rcl.len & 31) || local_clock() < tlimit))
2201 : 0 : continue;
2202 : : /* Exceeded the time limit, so leave. */
2203 : : break;
2204 : : }
2205 : : if (offloaded) {
2206 : : WARN_ON_ONCE(in_serving_softirq());
2207 : : local_bh_enable();
2208 : : lockdep_assert_irqs_enabled();
2209 : : cond_resched_tasks_rcu_qs();
2210 : : lockdep_assert_irqs_enabled();
2211 : : local_bh_disable();
2212 : : }
2213 : : }
2214 : :
2215 : 221185 : local_irq_save(flags);
2216 : 221185 : rcu_nocb_lock(rdp);
2217 : 221185 : count = -rcl.len;
2218 : 221185 : trace_rcu_batch_end(rcu_state.name, count, !!rcl.head, need_resched(),
2219 : : is_idle_task(current), rcu_is_callbacks_kthread());
2220 : :
2221 : : /* Update counts and requeue any remaining callbacks. */
2222 : 221185 : rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl);
2223 : 221185 : smp_mb(); /* List handling before counting for rcu_barrier(). */
2224 : 221185 : rcu_segcblist_insert_count(&rdp->cblist, &rcl);
2225 : :
2226 : : /* Reinstate batch limit if we have worked down the excess. */
2227 [ - + ]: 221185 : count = rcu_segcblist_n_cbs(&rdp->cblist);
2228 [ - + - - ]: 221185 : if (rdp->blimit >= DEFAULT_MAX_RCU_BLIMIT && count <= qlowmark)
2229 : 0 : rdp->blimit = blimit;
2230 : :
2231 : : /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
2232 [ + + - + ]: 221185 : if (count == 0 && rdp->qlen_last_fqs_check != 0) {
2233 : 0 : rdp->qlen_last_fqs_check = 0;
2234 : 0 : rdp->n_force_qs_snap = rcu_state.n_force_qs;
2235 [ - + ]: 221185 : } else if (count < rdp->qlen_last_fqs_check - qhimark)
2236 : 0 : rdp->qlen_last_fqs_check = count;
2237 : :
2238 : : /*
2239 : : * The following usually indicates a double call_rcu(). To track
2240 : : * this down, try building with CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.
2241 : : */
2242 [ + + + - : 223245 : WARN_ON_ONCE(count == 0 && !rcu_segcblist_empty(&rdp->cblist));
- + ]
2243 [ + + + - : 440310 : WARN_ON_ONCE(!IS_ENABLED(CONFIG_RCU_NOCB_CPU) &&
- + ]
2244 : : count != 0 && rcu_segcblist_empty(&rdp->cblist));
2245 : :
2246 : 221185 : rcu_nocb_unlock_irqrestore(rdp, flags);
2247 : :
2248 : : /* Re-invoke RCU core processing if there are callbacks remaining. */
2249 [ + + ]: 221185 : if (!offloaded && rcu_segcblist_ready_cbs(&rdp->cblist))
2250 : 186613 : invoke_rcu_core();
2251 : 221185 : tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
2252 : : }
2253 : :
2254 : : /*
2255 : : * This function is invoked from each scheduling-clock interrupt,
2256 : : * and checks to see if this CPU is in a non-context-switch quiescent
2257 : : * state, for example, user mode or idle loop. It also schedules RCU
2258 : : * core processing. If the current grace period has gone on too long,
2259 : : * it will ask the scheduler to manufacture a context switch for the sole
2260 : : * purpose of providing a providing the needed quiescent state.
2261 : : */
2262 : 72222 : void rcu_sched_clock_irq(int user)
2263 : : {
2264 : 72222 : trace_rcu_utilization(TPS("Start scheduler-tick"));
2265 : 72222 : raw_cpu_inc(rcu_data.ticks_this_gp);
2266 : : /* The load-acquire pairs with the store-release setting to true. */
2267 [ + - ]: 72222 : if (smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) {
2268 : : /* Idle and userspace execution already are quiescent states. */
2269 [ # # # # : 0 : if (!rcu_is_cpu_rrupt_from_idle() && !user) {
# # ]
2270 : 0 : set_tsk_need_resched(current);
2271 : 0 : set_preempt_need_resched();
2272 : : }
2273 : 72222 : __this_cpu_write(rcu_data.rcu_urgent_qs, false);
2274 : : }
2275 : 72222 : rcu_flavor_sched_clock_irq(user);
2276 [ + + ]: 72222 : if (rcu_pending(user))
2277 : 43742 : invoke_rcu_core();
2278 : :
2279 : 72222 : trace_rcu_utilization(TPS("End scheduler-tick"));
2280 : 72222 : }
2281 : :
2282 : : /*
2283 : : * Scan the leaf rcu_node structures. For each structure on which all
2284 : : * CPUs have reported a quiescent state and on which there are tasks
2285 : : * blocking the current grace period, initiate RCU priority boosting.
2286 : : * Otherwise, invoke the specified function to check dyntick state for
2287 : : * each CPU that has not yet reported a quiescent state.
2288 : : */
2289 : 0 : static void force_qs_rnp(int (*f)(struct rcu_data *rdp))
2290 : : {
2291 : 0 : int cpu;
2292 : 0 : unsigned long flags;
2293 : 0 : unsigned long mask;
2294 : 0 : struct rcu_data *rdp;
2295 : 0 : struct rcu_node *rnp;
2296 : :
2297 [ # # ]: 0 : rcu_for_each_leaf_node(rnp) {
2298 : 0 : cond_resched_tasks_rcu_qs();
2299 : 0 : mask = 0;
2300 : 0 : raw_spin_lock_irqsave_rcu_node(rnp, flags);
2301 [ # # ]: 0 : if (rnp->qsmask == 0) {
2302 : 0 : if (!IS_ENABLED(CONFIG_PREEMPT_RCU) ||
2303 : : rcu_preempt_blocked_readers_cgp(rnp)) {
2304 : : /*
2305 : : * No point in scanning bits because they
2306 : : * are all zero. But we might need to
2307 : : * priority-boost blocked readers.
2308 : : */
2309 : 0 : rcu_initiate_boost(rnp, flags);
2310 : : /* rcu_initiate_boost() releases rnp->lock */
2311 : 0 : continue;
2312 : : }
2313 : : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2314 : : continue;
2315 : : }
2316 [ # # ]: 0 : for_each_leaf_node_cpu_mask(rnp, cpu, rnp->qsmask) {
2317 : 0 : rdp = per_cpu_ptr(&rcu_data, cpu);
2318 [ # # ]: 0 : if (f(rdp)) {
2319 : 0 : mask |= rdp->grpmask;
2320 : 0 : rcu_disable_urgency_upon_qs(rdp);
2321 : : }
2322 : : }
2323 [ # # ]: 0 : if (mask != 0) {
2324 : : /* Idle/offline CPUs, report (releases rnp->lock). */
2325 : 0 : rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags);
2326 : : } else {
2327 : : /* Nothing to do here, so just drop the lock. */
2328 : 0 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2329 : : }
2330 : : }
2331 : 0 : }
2332 : :
2333 : : /*
2334 : : * Force quiescent states on reluctant CPUs, and also detect which
2335 : : * CPUs are in dyntick-idle mode.
2336 : : */
2337 : 0 : void rcu_force_quiescent_state(void)
2338 : : {
2339 : 0 : unsigned long flags;
2340 : 0 : bool ret;
2341 : 0 : struct rcu_node *rnp;
2342 : 0 : struct rcu_node *rnp_old = NULL;
2343 : :
2344 : : /* Funnel through hierarchy to reduce memory contention. */
2345 : 0 : rnp = __this_cpu_read(rcu_data.mynode);
2346 [ # # ]: 0 : for (; rnp != NULL; rnp = rnp->parent) {
2347 [ # # # # ]: 0 : ret = (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) ||
2348 : 0 : !raw_spin_trylock(&rnp->fqslock);
2349 [ # # ]: 0 : if (rnp_old != NULL)
2350 : 0 : raw_spin_unlock(&rnp_old->fqslock);
2351 [ # # ]: 0 : if (ret)
2352 : : return;
2353 : 0 : rnp_old = rnp;
2354 : : }
2355 : : /* rnp_old == rcu_get_root(), rnp == NULL. */
2356 : :
2357 : : /* Reached the root of the rcu_node tree, acquire lock. */
2358 : 0 : raw_spin_lock_irqsave_rcu_node(rnp_old, flags);
2359 : 0 : raw_spin_unlock(&rnp_old->fqslock);
2360 [ # # ]: 0 : if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) {
2361 : 0 : raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags);
2362 : 0 : return; /* Someone beat us to it. */
2363 : : }
2364 : 0 : WRITE_ONCE(rcu_state.gp_flags,
2365 : : READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS);
2366 : 0 : raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags);
2367 : 0 : rcu_gp_kthread_wake();
2368 : : }
2369 : : EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
2370 : :
2371 : : /* Perform RCU core processing work for the current CPU. */
2372 : 230048 : static __latent_entropy void rcu_core(void)
2373 : : {
2374 : 230048 : unsigned long flags;
2375 : 230048 : struct rcu_data *rdp = raw_cpu_ptr(&rcu_data);
2376 : 230048 : struct rcu_node *rnp = rdp->mynode;
2377 : 230048 : const bool offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) &&
2378 : : rcu_segcblist_is_offloaded(&rdp->cblist);
2379 : :
2380 [ + - ]: 230048 : if (cpu_is_offline(smp_processor_id()))
2381 : : return;
2382 : 230048 : trace_rcu_utilization(TPS("Start RCU core"));
2383 [ - + ]: 230048 : WARN_ON_ONCE(!rdp->beenonline);
2384 : :
2385 : : /* Report any deferred quiescent states if preemption enabled. */
2386 [ + + ]: 230048 : if (!(preempt_count() & PREEMPT_MASK)) {
2387 : 228774 : rcu_preempt_deferred_qs(current);
2388 : 1274 : } else if (rcu_preempt_need_deferred_qs(current)) {
2389 : : set_tsk_need_resched(current);
2390 : : set_preempt_need_resched();
2391 : : }
2392 : :
2393 : : /* Update RCU state based on any recent quiescent states. */
2394 : 230048 : rcu_check_quiescent_state(rdp);
2395 : :
2396 : : /* No grace period and unregistered callbacks? */
2397 [ + + + - ]: 230048 : if (!rcu_gp_in_progress() &&
2398 [ + - ]: 5616 : rcu_segcblist_is_enabled(&rdp->cblist) && !offloaded) {
2399 : 5616 : local_irq_save(flags);
2400 [ + + ]: 5616 : if (!rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL))
2401 : 2147 : rcu_accelerate_cbs_unlocked(rnp, rdp);
2402 : 5616 : local_irq_restore(flags);
2403 : : }
2404 : :
2405 [ - + ]: 230048 : rcu_check_gp_start_stall(rnp, rdp, rcu_jiffies_till_stall_check());
2406 : :
2407 : : /* If there are callbacks ready, invoke them. */
2408 [ + + + - ]: 230048 : if (!offloaded && rcu_segcblist_ready_cbs(&rdp->cblist) &&
2409 [ + - ]: 221185 : likely(READ_ONCE(rcu_scheduler_fully_active)))
2410 : 221185 : rcu_do_batch(rdp);
2411 : :
2412 : : /* Do any needed deferred wakeups of rcuo kthreads. */
2413 : 230048 : do_nocb_deferred_wakeup(rdp);
2414 : 230048 : trace_rcu_utilization(TPS("End RCU core"));
2415 : : }
2416 : :
2417 : 230048 : static void rcu_core_si(struct softirq_action *h)
2418 : : {
2419 : 230048 : rcu_core();
2420 : 230048 : }
2421 : :
2422 : 0 : static void rcu_wake_cond(struct task_struct *t, int status)
2423 : : {
2424 : : /*
2425 : : * If the thread is yielding, only wake it when this
2426 : : * is invoked from idle
2427 : : */
2428 [ # # # # : 0 : if (t && (status != RCU_KTHREAD_YIELDING || is_idle_task(current)))
# # ]
2429 : 0 : wake_up_process(t);
2430 : 0 : }
2431 : :
2432 : 0 : static void invoke_rcu_core_kthread(void)
2433 : : {
2434 : 0 : struct task_struct *t;
2435 : 0 : unsigned long flags;
2436 : :
2437 : 0 : local_irq_save(flags);
2438 [ # # ]: 0 : __this_cpu_write(rcu_data.rcu_cpu_has_work, 1);
2439 : 0 : t = __this_cpu_read(rcu_data.rcu_cpu_kthread_task);
2440 [ # # # # ]: 0 : if (t != NULL && t != current)
2441 : 0 : rcu_wake_cond(t, __this_cpu_read(rcu_data.rcu_cpu_kthread_status));
2442 : 0 : local_irq_restore(flags);
2443 : 0 : }
2444 : :
2445 : : /*
2446 : : * Wake up this CPU's rcuc kthread to do RCU core processing.
2447 : : */
2448 : 230355 : static void invoke_rcu_core(void)
2449 : : {
2450 [ + - ]: 230355 : if (!cpu_online(smp_processor_id()))
2451 : : return;
2452 [ + - ]: 230355 : if (use_softirq)
2453 : 230355 : raise_softirq(RCU_SOFTIRQ);
2454 : : else
2455 : 0 : invoke_rcu_core_kthread();
2456 : : }
2457 : :
2458 : 0 : static void rcu_cpu_kthread_park(unsigned int cpu)
2459 : : {
2460 : 0 : per_cpu(rcu_data.rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
2461 : 0 : }
2462 : :
2463 : 0 : static int rcu_cpu_kthread_should_run(unsigned int cpu)
2464 : : {
2465 : 0 : return __this_cpu_read(rcu_data.rcu_cpu_has_work);
2466 : : }
2467 : :
2468 : : /*
2469 : : * Per-CPU kernel thread that invokes RCU callbacks. This replaces
2470 : : * the RCU softirq used in configurations of RCU that do not support RCU
2471 : : * priority boosting.
2472 : : */
2473 : 0 : static void rcu_cpu_kthread(unsigned int cpu)
2474 : : {
2475 : 0 : unsigned int *statusp = this_cpu_ptr(&rcu_data.rcu_cpu_kthread_status);
2476 : 0 : char work, *workp = this_cpu_ptr(&rcu_data.rcu_cpu_has_work);
2477 : 0 : int spincnt;
2478 : :
2479 : 0 : trace_rcu_utilization(TPS("Start CPU kthread@rcu_run"));
2480 [ # # ]: 0 : for (spincnt = 0; spincnt < 10; spincnt++) {
2481 : 0 : local_bh_disable();
2482 : 0 : *statusp = RCU_KTHREAD_RUNNING;
2483 : 0 : local_irq_disable();
2484 : 0 : work = *workp;
2485 : 0 : *workp = 0;
2486 : 0 : local_irq_enable();
2487 [ # # ]: 0 : if (work)
2488 : 0 : rcu_core();
2489 : 0 : local_bh_enable();
2490 [ # # ]: 0 : if (*workp == 0) {
2491 : 0 : trace_rcu_utilization(TPS("End CPU kthread@rcu_wait"));
2492 : 0 : *statusp = RCU_KTHREAD_WAITING;
2493 : 0 : return;
2494 : : }
2495 : : }
2496 : 0 : *statusp = RCU_KTHREAD_YIELDING;
2497 : 0 : trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield"));
2498 : 0 : schedule_timeout_interruptible(2);
2499 : 0 : trace_rcu_utilization(TPS("End CPU kthread@rcu_yield"));
2500 : 0 : *statusp = RCU_KTHREAD_WAITING;
2501 : : }
2502 : :
2503 : : static struct smp_hotplug_thread rcu_cpu_thread_spec = {
2504 : : .store = &rcu_data.rcu_cpu_kthread_task,
2505 : : .thread_should_run = rcu_cpu_kthread_should_run,
2506 : : .thread_fn = rcu_cpu_kthread,
2507 : : .thread_comm = "rcuc/%u",
2508 : : .setup = rcu_cpu_kthread_setup,
2509 : : .park = rcu_cpu_kthread_park,
2510 : : };
2511 : :
2512 : : /*
2513 : : * Spawn per-CPU RCU core processing kthreads.
2514 : : */
2515 : 28 : static int __init rcu_spawn_core_kthreads(void)
2516 : : {
2517 : 28 : int cpu;
2518 : :
2519 [ + + ]: 56 : for_each_possible_cpu(cpu)
2520 : 28 : per_cpu(rcu_data.rcu_cpu_has_work, cpu) = 0;
2521 [ - + ]: 28 : if (!IS_ENABLED(CONFIG_RCU_BOOST) && use_softirq)
2522 : : return 0;
2523 [ # # # # ]: 0 : WARN_ONCE(smpboot_register_percpu_thread(&rcu_cpu_thread_spec),
2524 : : "%s: Could not start rcuc kthread, OOM is now expected behavior\n", __func__);
2525 : : return 0;
2526 : : }
2527 : : early_initcall(rcu_spawn_core_kthreads);
2528 : :
2529 : : /*
2530 : : * Handle any core-RCU processing required by a call_rcu() invocation.
2531 : : */
2532 : 2053982 : static void __call_rcu_core(struct rcu_data *rdp, struct rcu_head *head,
2533 : : unsigned long flags)
2534 : : {
2535 : : /*
2536 : : * If called from an extended quiescent state, invoke the RCU
2537 : : * core in order to force a re-evaluation of RCU's idleness.
2538 : : */
2539 [ - + ]: 4107964 : if (!rcu_is_watching())
2540 : 0 : invoke_rcu_core();
2541 : :
2542 : : /* If interrupts were disabled or CPU offline, don't invoke RCU core. */
2543 [ + + - + ]: 2053982 : if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id()))
2544 : 218976 : return;
2545 : :
2546 : : /*
2547 : : * Force the grace period if too many callbacks or too long waiting.
2548 : : * Enforce hysteresis, and don't invoke rcu_force_quiescent_state()
2549 : : * if some other CPU has recently done so. Also, don't bother
2550 : : * invoking rcu_force_quiescent_state() if the newly enqueued callback
2551 : : * is the only one waiting for a grace period to complete.
2552 : : */
2553 [ - + ]: 1835006 : if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) >
2554 : : rdp->qlen_last_fqs_check + qhimark)) {
2555 : :
2556 : : /* Are we ignoring a completed grace period? */
2557 : 0 : note_gp_changes(rdp);
2558 : :
2559 : : /* Start a new grace period if one not already started. */
2560 [ # # ]: 0 : if (!rcu_gp_in_progress()) {
2561 : 0 : rcu_accelerate_cbs_unlocked(rdp->mynode, rdp);
2562 : : } else {
2563 : : /* Give the grace period a kick. */
2564 : 0 : rdp->blimit = DEFAULT_MAX_RCU_BLIMIT;
2565 [ # # # # ]: 0 : if (rcu_state.n_force_qs == rdp->n_force_qs_snap &&
2566 : 0 : rcu_segcblist_first_pend_cb(&rdp->cblist) != head)
2567 : 0 : rcu_force_quiescent_state();
2568 : 0 : rdp->n_force_qs_snap = rcu_state.n_force_qs;
2569 : 0 : rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist);
2570 : : }
2571 : : }
2572 : : }
2573 : :
2574 : : /*
2575 : : * RCU callback function to leak a callback.
2576 : : */
2577 : : static void rcu_leak_callback(struct rcu_head *rhp)
2578 : : {
2579 : : }
2580 : :
2581 : : /*
2582 : : * Helper function for call_rcu() and friends. The cpu argument will
2583 : : * normally be -1, indicating "currently running CPU". It may specify
2584 : : * a CPU only if that CPU is a no-CBs CPU. Currently, only rcu_barrier()
2585 : : * is expected to specify a CPU.
2586 : : */
2587 : : static void
2588 : 2053982 : __call_rcu(struct rcu_head *head, rcu_callback_t func)
2589 : : {
2590 : 2053982 : unsigned long flags;
2591 : 2053982 : struct rcu_data *rdp;
2592 : 2053982 : bool was_alldone;
2593 : :
2594 : : /* Misaligned rcu_head! */
2595 [ - + ]: 2053982 : WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1));
2596 : :
2597 : 2053982 : if (debug_rcu_head_queue(head)) {
2598 : : /*
2599 : : * Probable double call_rcu(), so leak the callback.
2600 : : * Use rcu:rcu_callback trace event to find the previous
2601 : : * time callback was passed to __call_rcu().
2602 : : */
2603 : : WARN_ONCE(1, "__call_rcu(): Double-freed CB %p->%pS()!!!\n",
2604 : : head, head->func);
2605 : : WRITE_ONCE(head->func, rcu_leak_callback);
2606 : : return;
2607 : : }
2608 : 2053982 : head->func = func;
2609 : 2053982 : head->next = NULL;
2610 : 2053982 : local_irq_save(flags);
2611 : 2053982 : rdp = this_cpu_ptr(&rcu_data);
2612 : :
2613 : : /* Add the callback to our list. */
2614 [ - + ]: 2053982 : if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist))) {
2615 : : // This can trigger due to call_rcu() from offline CPU:
2616 [ # # ]: 0 : WARN_ON_ONCE(rcu_scheduler_active != RCU_SCHEDULER_INACTIVE);
2617 [ # # ]: 0 : WARN_ON_ONCE(!rcu_is_watching());
2618 : : // Very early boot, before rcu_init(). Initialize if needed
2619 : : // and then drop through to queue the callback.
2620 [ # # ]: 0 : if (rcu_segcblist_empty(&rdp->cblist))
2621 : 0 : rcu_segcblist_init(&rdp->cblist);
2622 : : }
2623 : :
2624 : 2053982 : if (rcu_nocb_try_bypass(rdp, head, &was_alldone, flags))
2625 : : return; // Enqueued onto ->nocb_bypass, so just leave.
2626 : : /* If we get here, rcu_nocb_try_bypass() acquired ->nocb_lock. */
2627 : 2053982 : rcu_segcblist_enqueue(&rdp->cblist, head);
2628 [ - + ]: 2053982 : if (__is_kfree_rcu_offset((unsigned long)func))
2629 : 0 : trace_rcu_kfree_callback(rcu_state.name, head,
2630 : : (unsigned long)func,
2631 : : rcu_segcblist_n_cbs(&rdp->cblist));
2632 : : else
2633 : 2053982 : trace_rcu_callback(rcu_state.name, head,
2634 : : rcu_segcblist_n_cbs(&rdp->cblist));
2635 : :
2636 : : /* Go handle any RCU core processing required. */
2637 : 2053982 : if (IS_ENABLED(CONFIG_RCU_NOCB_CPU) &&
2638 : : unlikely(rcu_segcblist_is_offloaded(&rdp->cblist))) {
2639 : : __call_rcu_nocb_wake(rdp, was_alldone, flags); /* unlocks */
2640 : : } else {
2641 : 2053982 : __call_rcu_core(rdp, head, flags);
2642 : 2053982 : local_irq_restore(flags);
2643 : : }
2644 : : }
2645 : :
2646 : : /**
2647 : : * call_rcu() - Queue an RCU callback for invocation after a grace period.
2648 : : * @head: structure to be used for queueing the RCU updates.
2649 : : * @func: actual callback function to be invoked after the grace period
2650 : : *
2651 : : * The callback function will be invoked some time after a full grace
2652 : : * period elapses, in other words after all pre-existing RCU read-side
2653 : : * critical sections have completed. However, the callback function
2654 : : * might well execute concurrently with RCU read-side critical sections
2655 : : * that started after call_rcu() was invoked. RCU read-side critical
2656 : : * sections are delimited by rcu_read_lock() and rcu_read_unlock(), and
2657 : : * may be nested. In addition, regions of code across which interrupts,
2658 : : * preemption, or softirqs have been disabled also serve as RCU read-side
2659 : : * critical sections. This includes hardware interrupt handlers, softirq
2660 : : * handlers, and NMI handlers.
2661 : : *
2662 : : * Note that all CPUs must agree that the grace period extended beyond
2663 : : * all pre-existing RCU read-side critical section. On systems with more
2664 : : * than one CPU, this means that when "func()" is invoked, each CPU is
2665 : : * guaranteed to have executed a full memory barrier since the end of its
2666 : : * last RCU read-side critical section whose beginning preceded the call
2667 : : * to call_rcu(). It also means that each CPU executing an RCU read-side
2668 : : * critical section that continues beyond the start of "func()" must have
2669 : : * executed a memory barrier after the call_rcu() but before the beginning
2670 : : * of that RCU read-side critical section. Note that these guarantees
2671 : : * include CPUs that are offline, idle, or executing in user mode, as
2672 : : * well as CPUs that are executing in the kernel.
2673 : : *
2674 : : * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
2675 : : * resulting RCU callback function "func()", then both CPU A and CPU B are
2676 : : * guaranteed to execute a full memory barrier during the time interval
2677 : : * between the call to call_rcu() and the invocation of "func()" -- even
2678 : : * if CPU A and CPU B are the same CPU (but again only if the system has
2679 : : * more than one CPU).
2680 : : */
2681 : 2053982 : void call_rcu(struct rcu_head *head, rcu_callback_t func)
2682 : : {
2683 : 2053982 : __call_rcu(head, func);
2684 : 2053982 : }
2685 : : EXPORT_SYMBOL_GPL(call_rcu);
2686 : :
2687 : :
2688 : : /* Maximum number of jiffies to wait before draining a batch. */
2689 : : #define KFREE_DRAIN_JIFFIES (HZ / 50)
2690 : : #define KFREE_N_BATCHES 2
2691 : :
2692 : : /**
2693 : : * struct kfree_rcu_cpu_work - single batch of kfree_rcu() requests
2694 : : * @rcu_work: Let queue_rcu_work() invoke workqueue handler after grace period
2695 : : * @head_free: List of kfree_rcu() objects waiting for a grace period
2696 : : * @krcp: Pointer to @kfree_rcu_cpu structure
2697 : : */
2698 : :
2699 : : struct kfree_rcu_cpu_work {
2700 : : struct rcu_work rcu_work;
2701 : : struct rcu_head *head_free;
2702 : : struct kfree_rcu_cpu *krcp;
2703 : : };
2704 : :
2705 : : /**
2706 : : * struct kfree_rcu_cpu - batch up kfree_rcu() requests for RCU grace period
2707 : : * @head: List of kfree_rcu() objects not yet waiting for a grace period
2708 : : * @krw_arr: Array of batches of kfree_rcu() objects waiting for a grace period
2709 : : * @lock: Synchronize access to this structure
2710 : : * @monitor_work: Promote @head to @head_free after KFREE_DRAIN_JIFFIES
2711 : : * @monitor_todo: Tracks whether a @monitor_work delayed work is pending
2712 : : * @initialized: The @lock and @rcu_work fields have been initialized
2713 : : *
2714 : : * This is a per-CPU structure. The reason that it is not included in
2715 : : * the rcu_data structure is to permit this code to be extracted from
2716 : : * the RCU files. Such extraction could allow further optimization of
2717 : : * the interactions with the slab allocators.
2718 : : */
2719 : : struct kfree_rcu_cpu {
2720 : : struct rcu_head *head;
2721 : : struct kfree_rcu_cpu_work krw_arr[KFREE_N_BATCHES];
2722 : : spinlock_t lock;
2723 : : struct delayed_work monitor_work;
2724 : : bool monitor_todo;
2725 : : bool initialized;
2726 : : };
2727 : :
2728 : : static DEFINE_PER_CPU(struct kfree_rcu_cpu, krc);
2729 : :
2730 : : /*
2731 : : * This function is invoked in workqueue context after a grace period.
2732 : : * It frees all the objects queued on ->head_free.
2733 : : */
2734 : 695 : static void kfree_rcu_work(struct work_struct *work)
2735 : : {
2736 : 695 : unsigned long flags;
2737 : 695 : struct rcu_head *head, *next;
2738 : 695 : struct kfree_rcu_cpu *krcp;
2739 : 695 : struct kfree_rcu_cpu_work *krwp;
2740 : :
2741 : 695 : krwp = container_of(to_rcu_work(work),
2742 : : struct kfree_rcu_cpu_work, rcu_work);
2743 : 695 : krcp = krwp->krcp;
2744 : 695 : spin_lock_irqsave(&krcp->lock, flags);
2745 : 695 : head = krwp->head_free;
2746 : 695 : krwp->head_free = NULL;
2747 : 695 : spin_unlock_irqrestore(&krcp->lock, flags);
2748 : :
2749 : : // List "head" is now private, so traverse locklessly.
2750 [ + + ]: 3318 : for (; head; head = next) {
2751 : 2623 : unsigned long offset = (unsigned long)head->func;
2752 : :
2753 : 2623 : next = head->next;
2754 : : // Potentially optimize with kfree_bulk in future.
2755 : 2623 : debug_rcu_head_unqueue(head);
2756 : 2623 : rcu_lock_acquire(&rcu_callback_map);
2757 : 2623 : trace_rcu_invoke_kfree_callback(rcu_state.name, head, offset);
2758 : :
2759 [ - + + - ]: 2623 : if (!WARN_ON_ONCE(!__is_kfree_rcu_offset(offset))) {
2760 : : /* Could be optimized with kfree_bulk() in future. */
2761 : 2623 : kfree((void *)head - offset);
2762 : : }
2763 : :
2764 : 2623 : rcu_lock_release(&rcu_callback_map);
2765 : 2623 : cond_resched_tasks_rcu_qs();
2766 : : }
2767 : 695 : }
2768 : :
2769 : : /*
2770 : : * Schedule the kfree batch RCU work to run in workqueue context after a GP.
2771 : : *
2772 : : * This function is invoked by kfree_rcu_monitor() when the KFREE_DRAIN_JIFFIES
2773 : : * timeout has been reached.
2774 : : */
2775 : 695 : static inline bool queue_kfree_rcu_work(struct kfree_rcu_cpu *krcp)
2776 : : {
2777 : 695 : int i;
2778 : 695 : struct kfree_rcu_cpu_work *krwp = NULL;
2779 : :
2780 : 695 : lockdep_assert_held(&krcp->lock);
2781 [ + - ]: 696 : for (i = 0; i < KFREE_N_BATCHES; i++)
2782 [ + + ]: 696 : if (!krcp->krw_arr[i].head_free) {
2783 : 695 : krwp = &(krcp->krw_arr[i]);
2784 : 695 : break;
2785 : : }
2786 : :
2787 : : // If a previous RCU batch is in progress, we cannot immediately
2788 : : // queue another one, so return false to tell caller to retry.
2789 [ + - ]: 695 : if (!krwp)
2790 : : return false;
2791 : :
2792 : 695 : krwp->head_free = krcp->head;
2793 : 695 : krcp->head = NULL;
2794 : 695 : INIT_RCU_WORK(&krwp->rcu_work, kfree_rcu_work);
2795 : 695 : queue_rcu_work(system_wq, &krwp->rcu_work);
2796 : 695 : return true;
2797 : : }
2798 : :
2799 : 695 : static inline void kfree_rcu_drain_unlock(struct kfree_rcu_cpu *krcp,
2800 : : unsigned long flags)
2801 : : {
2802 : : // Attempt to start a new batch.
2803 : 695 : krcp->monitor_todo = false;
2804 [ + - ]: 695 : if (queue_kfree_rcu_work(krcp)) {
2805 : : // Success! Our job is done here.
2806 : 695 : spin_unlock_irqrestore(&krcp->lock, flags);
2807 : 695 : return;
2808 : : }
2809 : :
2810 : : // Previous RCU batch still in progress, try again later.
2811 : 0 : krcp->monitor_todo = true;
2812 : 0 : schedule_delayed_work(&krcp->monitor_work, KFREE_DRAIN_JIFFIES);
2813 : 0 : spin_unlock_irqrestore(&krcp->lock, flags);
2814 : : }
2815 : :
2816 : : /*
2817 : : * This function is invoked after the KFREE_DRAIN_JIFFIES timeout.
2818 : : * It invokes kfree_rcu_drain_unlock() to attempt to start another batch.
2819 : : */
2820 : 695 : static void kfree_rcu_monitor(struct work_struct *work)
2821 : : {
2822 : 695 : unsigned long flags;
2823 : 695 : struct kfree_rcu_cpu *krcp = container_of(work, struct kfree_rcu_cpu,
2824 : : monitor_work.work);
2825 : :
2826 : 695 : spin_lock_irqsave(&krcp->lock, flags);
2827 [ + - ]: 695 : if (krcp->monitor_todo)
2828 : 695 : kfree_rcu_drain_unlock(krcp, flags);
2829 : : else
2830 : 0 : spin_unlock_irqrestore(&krcp->lock, flags);
2831 : 695 : }
2832 : :
2833 : : /*
2834 : : * Queue a request for lazy invocation of kfree() after a grace period.
2835 : : *
2836 : : * Each kfree_call_rcu() request is added to a batch. The batch will be drained
2837 : : * every KFREE_DRAIN_JIFFIES number of jiffies. All the objects in the batch
2838 : : * will be kfree'd in workqueue context. This allows us to:
2839 : : *
2840 : : * 1. Batch requests together to reduce the number of grace periods during
2841 : : * heavy kfree_rcu() load.
2842 : : *
2843 : : * 2. It makes it possible to use kfree_bulk() on a large number of
2844 : : * kfree_rcu() requests thus reducing cache misses and the per-object
2845 : : * overhead of kfree().
2846 : : */
2847 : 2623 : void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func)
2848 : : {
2849 : 2623 : unsigned long flags;
2850 : 2623 : struct kfree_rcu_cpu *krcp;
2851 : :
2852 : 2623 : local_irq_save(flags); // For safely calling this_cpu_ptr().
2853 : 2623 : krcp = this_cpu_ptr(&krc);
2854 [ + - ]: 2623 : if (krcp->initialized)
2855 : 2623 : spin_lock(&krcp->lock);
2856 : :
2857 : : // Queue the object but don't yet schedule the batch.
2858 [ - + ]: 2623 : if (debug_rcu_head_queue(head)) {
2859 : : // Probable double kfree_rcu(), just leak.
2860 : : WARN_ONCE(1, "%s(): Double-freed call. rcu_head %p\n",
2861 : : __func__, head);
2862 : : goto unlock_return;
2863 : : }
2864 : 2623 : head->func = func;
2865 : 2623 : head->next = krcp->head;
2866 : 2623 : krcp->head = head;
2867 : :
2868 : : // Set timer to drain after KFREE_DRAIN_JIFFIES.
2869 [ - + ]: 2623 : if (rcu_scheduler_active == RCU_SCHEDULER_RUNNING &&
2870 [ + + ]: 2623 : !krcp->monitor_todo) {
2871 : 695 : krcp->monitor_todo = true;
2872 : 695 : schedule_delayed_work(&krcp->monitor_work, KFREE_DRAIN_JIFFIES);
2873 : : }
2874 : :
2875 : 1928 : unlock_return:
2876 [ + - ]: 2623 : if (krcp->initialized)
2877 : 2623 : spin_unlock(&krcp->lock);
2878 : 2623 : local_irq_restore(flags);
2879 : 2623 : }
2880 : : EXPORT_SYMBOL_GPL(kfree_call_rcu);
2881 : :
2882 : 28 : void __init kfree_rcu_scheduler_running(void)
2883 : : {
2884 : 28 : int cpu;
2885 : 28 : unsigned long flags;
2886 : :
2887 [ + + ]: 56 : for_each_online_cpu(cpu) {
2888 : 28 : struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
2889 : :
2890 : 28 : spin_lock_irqsave(&krcp->lock, flags);
2891 [ - + - - ]: 28 : if (!krcp->head || krcp->monitor_todo) {
2892 : 28 : spin_unlock_irqrestore(&krcp->lock, flags);
2893 : 28 : continue;
2894 : : }
2895 : 0 : krcp->monitor_todo = true;
2896 : 0 : schedule_delayed_work_on(cpu, &krcp->monitor_work,
2897 : : KFREE_DRAIN_JIFFIES);
2898 : 0 : spin_unlock_irqrestore(&krcp->lock, flags);
2899 : : }
2900 : 28 : }
2901 : :
2902 : : /*
2903 : : * During early boot, any blocking grace-period wait automatically
2904 : : * implies a grace period. Later on, this is never the case for PREEMPTION.
2905 : : *
2906 : : * Howevr, because a context switch is a grace period for !PREEMPTION, any
2907 : : * blocking grace-period wait automatically implies a grace period if
2908 : : * there is only one CPU online at any point time during execution of
2909 : : * either synchronize_rcu() or synchronize_rcu_expedited(). It is OK to
2910 : : * occasionally incorrectly indicate that there are multiple CPUs online
2911 : : * when there was in fact only one the whole time, as this just adds some
2912 : : * overhead: RCU still operates correctly.
2913 : : */
2914 : 2012 : static int rcu_blocking_is_gp(void)
2915 : : {
2916 : 2012 : int ret;
2917 : :
2918 : 2012 : if (IS_ENABLED(CONFIG_PREEMPTION))
2919 : : return rcu_scheduler_active == RCU_SCHEDULER_INACTIVE;
2920 : 2012 : might_sleep(); /* Check for RCU read-side critical section. */
2921 : 2012 : preempt_disable();
2922 : 2012 : ret = num_online_cpus() <= 1;
2923 : 2012 : preempt_enable();
2924 : 2012 : return ret;
2925 : : }
2926 : :
2927 : : /**
2928 : : * synchronize_rcu - wait until a grace period has elapsed.
2929 : : *
2930 : : * Control will return to the caller some time after a full grace
2931 : : * period has elapsed, in other words after all currently executing RCU
2932 : : * read-side critical sections have completed. Note, however, that
2933 : : * upon return from synchronize_rcu(), the caller might well be executing
2934 : : * concurrently with new RCU read-side critical sections that began while
2935 : : * synchronize_rcu() was waiting. RCU read-side critical sections are
2936 : : * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
2937 : : * In addition, regions of code across which interrupts, preemption, or
2938 : : * softirqs have been disabled also serve as RCU read-side critical
2939 : : * sections. This includes hardware interrupt handlers, softirq handlers,
2940 : : * and NMI handlers.
2941 : : *
2942 : : * Note that this guarantee implies further memory-ordering guarantees.
2943 : : * On systems with more than one CPU, when synchronize_rcu() returns,
2944 : : * each CPU is guaranteed to have executed a full memory barrier since
2945 : : * the end of its last RCU read-side critical section whose beginning
2946 : : * preceded the call to synchronize_rcu(). In addition, each CPU having
2947 : : * an RCU read-side critical section that extends beyond the return from
2948 : : * synchronize_rcu() is guaranteed to have executed a full memory barrier
2949 : : * after the beginning of synchronize_rcu() and before the beginning of
2950 : : * that RCU read-side critical section. Note that these guarantees include
2951 : : * CPUs that are offline, idle, or executing in user mode, as well as CPUs
2952 : : * that are executing in the kernel.
2953 : : *
2954 : : * Furthermore, if CPU A invoked synchronize_rcu(), which returned
2955 : : * to its caller on CPU B, then both CPU A and CPU B are guaranteed
2956 : : * to have executed a full memory barrier during the execution of
2957 : : * synchronize_rcu() -- even if CPU A and CPU B are the same CPU (but
2958 : : * again only if the system has more than one CPU).
2959 : : */
2960 : 1872 : void synchronize_rcu(void)
2961 : : {
2962 : 1872 : RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) ||
2963 : : lock_is_held(&rcu_lock_map) ||
2964 : : lock_is_held(&rcu_sched_lock_map),
2965 : : "Illegal synchronize_rcu() in RCU read-side critical section");
2966 [ - + ]: 1872 : if (rcu_blocking_is_gp())
2967 : : return;
2968 [ # # ]: 0 : if (rcu_gp_is_expedited())
2969 : 0 : synchronize_rcu_expedited();
2970 : : else
2971 : 0 : wait_rcu_gp(call_rcu);
2972 : : }
2973 : : EXPORT_SYMBOL_GPL(synchronize_rcu);
2974 : :
2975 : : /**
2976 : : * get_state_synchronize_rcu - Snapshot current RCU state
2977 : : *
2978 : : * Returns a cookie that is used by a later call to cond_synchronize_rcu()
2979 : : * to determine whether or not a full grace period has elapsed in the
2980 : : * meantime.
2981 : : */
2982 : 0 : unsigned long get_state_synchronize_rcu(void)
2983 : : {
2984 : : /*
2985 : : * Any prior manipulation of RCU-protected data must happen
2986 : : * before the load from ->gp_seq.
2987 : : */
2988 : 0 : smp_mb(); /* ^^^ */
2989 : 0 : return rcu_seq_snap(&rcu_state.gp_seq);
2990 : : }
2991 : : EXPORT_SYMBOL_GPL(get_state_synchronize_rcu);
2992 : :
2993 : : /**
2994 : : * cond_synchronize_rcu - Conditionally wait for an RCU grace period
2995 : : *
2996 : : * @oldstate: return value from earlier call to get_state_synchronize_rcu()
2997 : : *
2998 : : * If a full RCU grace period has elapsed since the earlier call to
2999 : : * get_state_synchronize_rcu(), just return. Otherwise, invoke
3000 : : * synchronize_rcu() to wait for a full grace period.
3001 : : *
3002 : : * Yes, this function does not take counter wrap into account. But
3003 : : * counter wrap is harmless. If the counter wraps, we have waited for
3004 : : * more than 2 billion grace periods (and way more on a 64-bit system!),
3005 : : * so waiting for one additional grace period should be just fine.
3006 : : */
3007 : 0 : void cond_synchronize_rcu(unsigned long oldstate)
3008 : : {
3009 [ # # ]: 0 : if (!rcu_seq_done(&rcu_state.gp_seq, oldstate))
3010 : 0 : synchronize_rcu();
3011 : : else
3012 : 0 : smp_mb(); /* Ensure GP ends before subsequent accesses. */
3013 : 0 : }
3014 : : EXPORT_SYMBOL_GPL(cond_synchronize_rcu);
3015 : :
3016 : : /*
3017 : : * Check to see if there is any immediate RCU-related work to be done by
3018 : : * the current CPU, returning 1 if so and zero otherwise. The checks are
3019 : : * in order of increasing expense: checks that can be carried out against
3020 : : * CPU-local state are performed first. However, we must check for CPU
3021 : : * stalls first, else we might not get a chance.
3022 : : */
3023 : 72222 : static int rcu_pending(int user)
3024 : : {
3025 : 72222 : bool gp_in_progress;
3026 : 72222 : struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
3027 : 72222 : struct rcu_node *rnp = rdp->mynode;
3028 : :
3029 : : /* Check for CPU stalls, if enabled. */
3030 : 72222 : check_cpu_stall(rdp);
3031 : :
3032 : : /* Does this CPU need a deferred NOCB wakeup? */
3033 [ + + ]: 72222 : if (rcu_nocb_need_deferred_wakeup(rdp))
3034 : : return 1;
3035 : :
3036 : : /* Is this a nohz_full CPU in userspace or idle? (Ignore RCU if so.) */
3037 [ + + ]: 72222 : if ((user || rcu_is_cpu_rrupt_from_idle()) && rcu_nohz_full_cpu())
3038 : : return 0;
3039 : :
3040 : : /* Is the RCU core waiting for a quiescent state from this CPU? */
3041 [ + + ]: 72222 : gp_in_progress = rcu_gp_in_progress();
3042 [ + + + + : 72222 : if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm && gp_in_progress)
+ + ]
3043 : : return 1;
3044 : :
3045 : : /* Does this CPU have callbacks ready to invoke? */
3046 [ + + ]: 30595 : if (rcu_segcblist_ready_cbs(&rdp->cblist))
3047 : : return 1;
3048 : :
3049 : : /* Has RCU gone idle with this CPU needing another grace period? */
3050 [ + + + - ]: 30592 : if (!gp_in_progress && rcu_segcblist_is_enabled(&rdp->cblist) &&
3051 : : (!IS_ENABLED(CONFIG_RCU_NOCB_CPU) ||
3052 [ + + ]: 30590 : !rcu_segcblist_is_offloaded(&rdp->cblist)) &&
3053 : : !rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL))
3054 : : return 1;
3055 : :
3056 : : /* Have RCU grace period completed or started? */
3057 [ + - - + ]: 28480 : if (rcu_seq_current(&rnp->gp_seq) != rdp->gp_seq ||
3058 [ - + ]: 28480 : unlikely(READ_ONCE(rdp->gpwrap))) /* outside lock */
3059 : 0 : return 1;
3060 : :
3061 : : /* nothing to do */
3062 : : return 0;
3063 : : }
3064 : :
3065 : : /*
3066 : : * Helper function for rcu_barrier() tracing. If tracing is disabled,
3067 : : * the compiler is expected to optimize this away.
3068 : : */
3069 : 168 : static void rcu_barrier_trace(const char *s, int cpu, unsigned long done)
3070 : : {
3071 : 168 : trace_rcu_barrier(rcu_state.name, s, cpu,
3072 : : atomic_read(&rcu_state.barrier_cpu_count), done);
3073 : 168 : }
3074 : :
3075 : : /*
3076 : : * RCU callback function for rcu_barrier(). If we are last, wake
3077 : : * up the task executing rcu_barrier().
3078 : : */
3079 : 28 : static void rcu_barrier_callback(struct rcu_head *rhp)
3080 : : {
3081 [ + - ]: 28 : if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) {
3082 : 28 : rcu_barrier_trace(TPS("LastCB"), -1,
3083 : : rcu_state.barrier_sequence);
3084 : 28 : complete(&rcu_state.barrier_completion);
3085 : : } else {
3086 : 0 : rcu_barrier_trace(TPS("CB"), -1, rcu_state.barrier_sequence);
3087 : : }
3088 : 28 : }
3089 : :
3090 : : /*
3091 : : * Called with preemption disabled, and from cross-cpu IRQ context.
3092 : : */
3093 : 28 : static void rcu_barrier_func(void *unused)
3094 : : {
3095 : 28 : struct rcu_data *rdp = raw_cpu_ptr(&rcu_data);
3096 : :
3097 : 28 : rcu_barrier_trace(TPS("IRQ"), -1, rcu_state.barrier_sequence);
3098 : 28 : rdp->barrier_head.func = rcu_barrier_callback;
3099 : 28 : debug_rcu_head_queue(&rdp->barrier_head);
3100 : 28 : rcu_nocb_lock(rdp);
3101 : 28 : WARN_ON_ONCE(!rcu_nocb_flush_bypass(rdp, NULL, jiffies));
3102 [ + - ]: 28 : if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head)) {
3103 : 28 : atomic_inc(&rcu_state.barrier_cpu_count);
3104 : : } else {
3105 : 0 : debug_rcu_head_unqueue(&rdp->barrier_head);
3106 : 0 : rcu_barrier_trace(TPS("IRQNQ"), -1,
3107 : : rcu_state.barrier_sequence);
3108 : : }
3109 : 28 : rcu_nocb_unlock(rdp);
3110 : 28 : }
3111 : :
3112 : : /**
3113 : : * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
3114 : : *
3115 : : * Note that this primitive does not necessarily wait for an RCU grace period
3116 : : * to complete. For example, if there are no RCU callbacks queued anywhere
3117 : : * in the system, then rcu_barrier() is within its rights to return
3118 : : * immediately, without waiting for anything, much less an RCU grace period.
3119 : : */
3120 : 28 : void rcu_barrier(void)
3121 : : {
3122 : 28 : int cpu;
3123 : 28 : struct rcu_data *rdp;
3124 : 28 : unsigned long s = rcu_seq_snap(&rcu_state.barrier_sequence);
3125 : :
3126 : 28 : rcu_barrier_trace(TPS("Begin"), -1, s);
3127 : :
3128 : : /* Take mutex to serialize concurrent rcu_barrier() requests. */
3129 : 28 : mutex_lock(&rcu_state.barrier_mutex);
3130 : :
3131 : : /* Did someone else do our work for us? */
3132 [ - + ]: 28 : if (rcu_seq_done(&rcu_state.barrier_sequence, s)) {
3133 : 0 : rcu_barrier_trace(TPS("EarlyExit"), -1,
3134 : : rcu_state.barrier_sequence);
3135 : 0 : smp_mb(); /* caller's subsequent code after above check. */
3136 : 0 : mutex_unlock(&rcu_state.barrier_mutex);
3137 : 0 : return;
3138 : : }
3139 : :
3140 : : /* Mark the start of the barrier operation. */
3141 : 28 : rcu_seq_start(&rcu_state.barrier_sequence);
3142 : 28 : rcu_barrier_trace(TPS("Inc1"), -1, rcu_state.barrier_sequence);
3143 : :
3144 : : /*
3145 : : * Initialize the count to one rather than to zero in order to
3146 : : * avoid a too-soon return to zero in case of a short grace period
3147 : : * (or preemption of this task). Exclude CPU-hotplug operations
3148 : : * to ensure that no offline CPU has callbacks queued.
3149 : : */
3150 : 28 : init_completion(&rcu_state.barrier_completion);
3151 : 28 : atomic_set(&rcu_state.barrier_cpu_count, 1);
3152 : 28 : get_online_cpus();
3153 : :
3154 : : /*
3155 : : * Force each CPU with callbacks to register a new callback.
3156 : : * When that callback is invoked, we will know that all of the
3157 : : * corresponding CPU's preceding callbacks have been invoked.
3158 : : */
3159 [ + + ]: 84 : for_each_possible_cpu(cpu) {
3160 : 28 : rdp = per_cpu_ptr(&rcu_data, cpu);
3161 [ - + - - ]: 28 : if (!cpu_online(cpu) &&
3162 [ # # ]: 0 : !rcu_segcblist_is_offloaded(&rdp->cblist))
3163 : 0 : continue;
3164 [ + - ]: 28 : if (rcu_segcblist_n_cbs(&rdp->cblist)) {
3165 : 28 : rcu_barrier_trace(TPS("OnlineQ"), cpu,
3166 : : rcu_state.barrier_sequence);
3167 : 28 : smp_call_function_single(cpu, rcu_barrier_func, NULL, 1);
3168 : : } else {
3169 : 0 : rcu_barrier_trace(TPS("OnlineNQ"), cpu,
3170 : : rcu_state.barrier_sequence);
3171 : : }
3172 : : }
3173 : 28 : put_online_cpus();
3174 : :
3175 : : /*
3176 : : * Now that we have an rcu_barrier_callback() callback on each
3177 : : * CPU, and thus each counted, remove the initial count.
3178 : : */
3179 [ - + ]: 28 : if (atomic_dec_and_test(&rcu_state.barrier_cpu_count))
3180 : 0 : complete(&rcu_state.barrier_completion);
3181 : :
3182 : : /* Wait for all rcu_barrier_callback() callbacks to be invoked. */
3183 : 28 : wait_for_completion(&rcu_state.barrier_completion);
3184 : :
3185 : : /* Mark the end of the barrier operation. */
3186 : 28 : rcu_barrier_trace(TPS("Inc2"), -1, rcu_state.barrier_sequence);
3187 : 28 : rcu_seq_end(&rcu_state.barrier_sequence);
3188 : :
3189 : : /* Other rcu_barrier() invocations can now safely proceed. */
3190 : 28 : mutex_unlock(&rcu_state.barrier_mutex);
3191 : : }
3192 : : EXPORT_SYMBOL_GPL(rcu_barrier);
3193 : :
3194 : : /*
3195 : : * Propagate ->qsinitmask bits up the rcu_node tree to account for the
3196 : : * first CPU in a given leaf rcu_node structure coming online. The caller
3197 : : * must hold the corresponding leaf rcu_node ->lock with interrrupts
3198 : : * disabled.
3199 : : */
3200 : 28 : static void rcu_init_new_rnp(struct rcu_node *rnp_leaf)
3201 : : {
3202 : 28 : long mask;
3203 : 28 : long oldmask;
3204 : 28 : struct rcu_node *rnp = rnp_leaf;
3205 : :
3206 : 28 : raw_lockdep_assert_held_rcu_node(rnp_leaf);
3207 [ - + ]: 28 : WARN_ON_ONCE(rnp->wait_blkd_tasks);
3208 : 28 : for (;;) {
3209 : 28 : mask = rnp->grpmask;
3210 : 28 : rnp = rnp->parent;
3211 [ - + ]: 28 : if (rnp == NULL)
3212 : : return;
3213 : 0 : raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */
3214 : 0 : oldmask = rnp->qsmaskinit;
3215 : 0 : rnp->qsmaskinit |= mask;
3216 : 0 : raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */
3217 [ # # ]: 0 : if (oldmask)
3218 : : return;
3219 : : }
3220 : : }
3221 : :
3222 : : /*
3223 : : * Do boot-time initialization of a CPU's per-CPU RCU data.
3224 : : */
3225 : : static void __init
3226 : 28 : rcu_boot_init_percpu_data(int cpu)
3227 : : {
3228 : 28 : struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
3229 : :
3230 : : /* Set up local state, ensuring consistent view of global state. */
3231 : 28 : rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu);
3232 [ - + ]: 28 : WARN_ON_ONCE(rdp->dynticks_nesting != 1);
3233 [ - + ]: 28 : WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp)));
3234 : 28 : rdp->rcu_ofl_gp_seq = rcu_state.gp_seq;
3235 : 28 : rdp->rcu_ofl_gp_flags = RCU_GP_CLEANED;
3236 : 28 : rdp->rcu_onl_gp_seq = rcu_state.gp_seq;
3237 : 28 : rdp->rcu_onl_gp_flags = RCU_GP_CLEANED;
3238 : 28 : rdp->cpu = cpu;
3239 : 28 : rcu_boot_init_nocb_percpu_data(rdp);
3240 : 28 : }
3241 : :
3242 : : /*
3243 : : * Invoked early in the CPU-online process, when pretty much all services
3244 : : * are available. The incoming CPU is not present.
3245 : : *
3246 : : * Initializes a CPU's per-CPU RCU data. Note that only one online or
3247 : : * offline event can be happening at a given time. Note also that we can
3248 : : * accept some slop in the rsp->gp_seq access due to the fact that this
3249 : : * CPU cannot possibly have any non-offloaded RCU callbacks in flight yet.
3250 : : * And any offloaded callbacks are being numbered elsewhere.
3251 : : */
3252 : 28 : int rcutree_prepare_cpu(unsigned int cpu)
3253 : : {
3254 : 28 : unsigned long flags;
3255 : 28 : struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
3256 : 28 : struct rcu_node *rnp = rcu_get_root();
3257 : :
3258 : : /* Set up local state, ensuring consistent view of global state. */
3259 : 28 : raw_spin_lock_irqsave_rcu_node(rnp, flags);
3260 : 28 : rdp->qlen_last_fqs_check = 0;
3261 : 28 : rdp->n_force_qs_snap = rcu_state.n_force_qs;
3262 : 28 : rdp->blimit = blimit;
3263 [ + - + - ]: 28 : if (rcu_segcblist_empty(&rdp->cblist) && /* No early-boot CBs? */
3264 [ + - ]: 28 : !rcu_segcblist_is_offloaded(&rdp->cblist))
3265 : 28 : rcu_segcblist_init(&rdp->cblist); /* Re-enable callbacks. */
3266 : 28 : rdp->dynticks_nesting = 1; /* CPU not up, no tearing. */
3267 : 28 : rcu_dynticks_eqs_online();
3268 : 28 : raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */
3269 : :
3270 : : /*
3271 : : * Add CPU to leaf rcu_node pending-online bitmask. Any needed
3272 : : * propagation up the rcu_node tree will happen at the beginning
3273 : : * of the next grace period.
3274 : : */
3275 : 28 : rnp = rdp->mynode;
3276 : 28 : raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
3277 : 28 : rdp->beenonline = true; /* We have now been online. */
3278 : 28 : rdp->gp_seq = rnp->gp_seq;
3279 : 28 : rdp->gp_seq_needed = rnp->gp_seq;
3280 : 28 : rdp->cpu_no_qs.b.norm = true;
3281 : 28 : rdp->core_needs_qs = false;
3282 : 28 : rdp->rcu_iw_pending = false;
3283 : 28 : rdp->rcu_iw_gp_seq = rnp->gp_seq - 1;
3284 : 28 : trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuonl"));
3285 : 28 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3286 : 28 : rcu_prepare_kthreads(cpu);
3287 : 28 : rcu_spawn_cpu_nocb_kthread(cpu);
3288 : :
3289 : 28 : return 0;
3290 : : }
3291 : :
3292 : : /*
3293 : : * Update RCU priority boot kthread affinity for CPU-hotplug changes.
3294 : : */
3295 : 0 : static void rcutree_affinity_setting(unsigned int cpu, int outgoing)
3296 : : {
3297 : 0 : struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
3298 : :
3299 : 0 : rcu_boost_kthread_setaffinity(rdp->mynode, outgoing);
3300 : : }
3301 : :
3302 : : /*
3303 : : * Near the end of the CPU-online process. Pretty much all services
3304 : : * enabled, and the CPU is now very much alive.
3305 : : */
3306 : 28 : int rcutree_online_cpu(unsigned int cpu)
3307 : : {
3308 : 28 : unsigned long flags;
3309 : 28 : struct rcu_data *rdp;
3310 : 28 : struct rcu_node *rnp;
3311 : :
3312 : 28 : rdp = per_cpu_ptr(&rcu_data, cpu);
3313 : 28 : rnp = rdp->mynode;
3314 : 28 : raw_spin_lock_irqsave_rcu_node(rnp, flags);
3315 : 28 : rnp->ffmask |= rdp->grpmask;
3316 : 28 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3317 [ - + ]: 28 : if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE)
3318 : : return 0; /* Too early in boot for scheduler work. */
3319 : 0 : sync_sched_exp_online_cleanup(cpu);
3320 : 0 : rcutree_affinity_setting(cpu, -1);
3321 : :
3322 : : // Stop-machine done, so allow nohz_full to disable tick.
3323 : 0 : tick_dep_clear(TICK_DEP_BIT_RCU);
3324 : 0 : return 0;
3325 : : }
3326 : :
3327 : : /*
3328 : : * Near the beginning of the process. The CPU is still very much alive
3329 : : * with pretty much all services enabled.
3330 : : */
3331 : 0 : int rcutree_offline_cpu(unsigned int cpu)
3332 : : {
3333 : 0 : unsigned long flags;
3334 : 0 : struct rcu_data *rdp;
3335 : 0 : struct rcu_node *rnp;
3336 : :
3337 : 0 : rdp = per_cpu_ptr(&rcu_data, cpu);
3338 : 0 : rnp = rdp->mynode;
3339 : 0 : raw_spin_lock_irqsave_rcu_node(rnp, flags);
3340 : 0 : rnp->ffmask &= ~rdp->grpmask;
3341 : 0 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3342 : :
3343 : 0 : rcutree_affinity_setting(cpu, cpu);
3344 : :
3345 : : // nohz_full CPUs need the tick for stop-machine to work quickly
3346 : 0 : tick_dep_set(TICK_DEP_BIT_RCU);
3347 : 0 : return 0;
3348 : : }
3349 : :
3350 : : static DEFINE_PER_CPU(int, rcu_cpu_started);
3351 : :
3352 : : /*
3353 : : * Mark the specified CPU as being online so that subsequent grace periods
3354 : : * (both expedited and normal) will wait on it. Note that this means that
3355 : : * incoming CPUs are not allowed to use RCU read-side critical sections
3356 : : * until this function is called. Failing to observe this restriction
3357 : : * will result in lockdep splats.
3358 : : *
3359 : : * Note that this function is special in that it is invoked directly
3360 : : * from the incoming CPU rather than from the cpuhp_step mechanism.
3361 : : * This is because this function must be invoked at a precise location.
3362 : : */
3363 : 28 : void rcu_cpu_starting(unsigned int cpu)
3364 : : {
3365 : 28 : unsigned long flags;
3366 : 28 : unsigned long mask;
3367 : 28 : int nbits;
3368 : 28 : unsigned long oldmask;
3369 : 28 : struct rcu_data *rdp;
3370 : 28 : struct rcu_node *rnp;
3371 : :
3372 [ + - ]: 28 : if (per_cpu(rcu_cpu_started, cpu))
3373 : : return;
3374 : :
3375 : 28 : per_cpu(rcu_cpu_started, cpu) = 1;
3376 : :
3377 : 28 : rdp = per_cpu_ptr(&rcu_data, cpu);
3378 : 28 : rnp = rdp->mynode;
3379 : 28 : mask = rdp->grpmask;
3380 : 28 : raw_spin_lock_irqsave_rcu_node(rnp, flags);
3381 : 28 : rnp->qsmaskinitnext |= mask;
3382 : 28 : oldmask = rnp->expmaskinitnext;
3383 : 28 : rnp->expmaskinitnext |= mask;
3384 : 28 : oldmask ^= rnp->expmaskinitnext;
3385 [ - + ]: 28 : nbits = bitmap_weight(&oldmask, BITS_PER_LONG);
3386 : : /* Allow lockless access for expedited grace periods. */
3387 [ - + ]: 28 : smp_store_release(&rcu_state.ncpus, rcu_state.ncpus + nbits); /* ^^^ */
3388 [ - + ]: 28 : rcu_gpnum_ovf(rnp, rdp); /* Offline-induced counter wrap? */
3389 [ - + ]: 28 : rdp->rcu_onl_gp_seq = READ_ONCE(rcu_state.gp_seq);
3390 : 28 : rdp->rcu_onl_gp_flags = READ_ONCE(rcu_state.gp_flags);
3391 [ - + ]: 28 : if (rnp->qsmask & mask) { /* RCU waiting on incoming CPU? */
3392 : 0 : rcu_disable_urgency_upon_qs(rdp);
3393 : : /* Report QS -after- changing ->qsmaskinitnext! */
3394 : 0 : rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags);
3395 : : } else {
3396 : 28 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3397 : : }
3398 : 28 : smp_mb(); /* Ensure RCU read-side usage follows above initialization. */
3399 : : }
3400 : :
3401 : : #ifdef CONFIG_HOTPLUG_CPU
3402 : : /*
3403 : : * The outgoing function has no further need of RCU, so remove it from
3404 : : * the rcu_node tree's ->qsmaskinitnext bit masks.
3405 : : *
3406 : : * Note that this function is special in that it is invoked directly
3407 : : * from the outgoing CPU rather than from the cpuhp_step mechanism.
3408 : : * This is because this function must be invoked at a precise location.
3409 : : */
3410 : 0 : void rcu_report_dead(unsigned int cpu)
3411 : : {
3412 : 0 : unsigned long flags;
3413 : 0 : unsigned long mask;
3414 : 0 : struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
3415 : 0 : struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */
3416 : :
3417 : : /* QS for any half-done expedited grace period. */
3418 : 0 : preempt_disable();
3419 : 0 : rcu_report_exp_rdp(this_cpu_ptr(&rcu_data));
3420 : 0 : preempt_enable();
3421 : 0 : rcu_preempt_deferred_qs(current);
3422 : :
3423 : : /* Remove outgoing CPU from mask in the leaf rcu_node structure. */
3424 : 0 : mask = rdp->grpmask;
3425 : 0 : raw_spin_lock(&rcu_state.ofl_lock);
3426 : 0 : raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */
3427 [ # # ]: 0 : rdp->rcu_ofl_gp_seq = READ_ONCE(rcu_state.gp_seq);
3428 : 0 : rdp->rcu_ofl_gp_flags = READ_ONCE(rcu_state.gp_flags);
3429 [ # # ]: 0 : if (rnp->qsmask & mask) { /* RCU waiting on outgoing CPU? */
3430 : : /* Report quiescent state -before- changing ->qsmaskinitnext! */
3431 : 0 : rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags);
3432 : 0 : raw_spin_lock_irqsave_rcu_node(rnp, flags);
3433 : : }
3434 : 0 : rnp->qsmaskinitnext &= ~mask;
3435 : 0 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3436 : 0 : raw_spin_unlock(&rcu_state.ofl_lock);
3437 : :
3438 : 0 : per_cpu(rcu_cpu_started, cpu) = 0;
3439 : 0 : }
3440 : :
3441 : : /*
3442 : : * The outgoing CPU has just passed through the dying-idle state, and we
3443 : : * are being invoked from the CPU that was IPIed to continue the offline
3444 : : * operation. Migrate the outgoing CPU's callbacks to the current CPU.
3445 : : */
3446 : 0 : void rcutree_migrate_callbacks(int cpu)
3447 : : {
3448 : 0 : unsigned long flags;
3449 : 0 : struct rcu_data *my_rdp;
3450 : 0 : struct rcu_node *my_rnp;
3451 : 0 : struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
3452 : 0 : bool needwake;
3453 : :
3454 [ # # # # ]: 0 : if (rcu_segcblist_is_offloaded(&rdp->cblist) ||
3455 : : rcu_segcblist_empty(&rdp->cblist))
3456 : : return; /* No callbacks to migrate. */
3457 : :
3458 : 0 : local_irq_save(flags);
3459 : 0 : my_rdp = this_cpu_ptr(&rcu_data);
3460 : 0 : my_rnp = my_rdp->mynode;
3461 : 0 : rcu_nocb_lock(my_rdp); /* irqs already disabled. */
3462 : 0 : WARN_ON_ONCE(!rcu_nocb_flush_bypass(my_rdp, NULL, jiffies));
3463 : 0 : raw_spin_lock_rcu_node(my_rnp); /* irqs already disabled. */
3464 : : /* Leverage recent GPs and set GP for new callbacks. */
3465 [ # # # # ]: 0 : needwake = rcu_advance_cbs(my_rnp, rdp) ||
3466 : 0 : rcu_advance_cbs(my_rnp, my_rdp);
3467 : 0 : rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist);
3468 [ # # # # ]: 0 : needwake = needwake || rcu_advance_cbs(my_rnp, my_rdp);
3469 : 0 : rcu_segcblist_disable(&rdp->cblist);
3470 [ # # ]: 0 : WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) !=
3471 : : !rcu_segcblist_n_cbs(&my_rdp->cblist));
3472 [ # # ]: 0 : if (rcu_segcblist_is_offloaded(&my_rdp->cblist)) {
3473 : 0 : raw_spin_unlock_rcu_node(my_rnp); /* irqs remain disabled. */
3474 : 0 : __call_rcu_nocb_wake(my_rdp, true, flags);
3475 : : } else {
3476 : 0 : rcu_nocb_unlock(my_rdp); /* irqs remain disabled. */
3477 : 0 : raw_spin_unlock_irqrestore_rcu_node(my_rnp, flags);
3478 : : }
3479 [ # # ]: 0 : if (needwake)
3480 : 0 : rcu_gp_kthread_wake();
3481 : 0 : lockdep_assert_irqs_enabled();
3482 [ # # # # : 0 : WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 ||
# # # # ]
3483 : : !rcu_segcblist_empty(&rdp->cblist),
3484 : : "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n",
3485 : : cpu, rcu_segcblist_n_cbs(&rdp->cblist),
3486 : : rcu_segcblist_first_cb(&rdp->cblist));
3487 : : }
3488 : : #endif
3489 : :
3490 : : /*
3491 : : * On non-huge systems, use expedited RCU grace periods to make suspend
3492 : : * and hibernation run faster.
3493 : : */
3494 : 0 : static int rcu_pm_notify(struct notifier_block *self,
3495 : : unsigned long action, void *hcpu)
3496 : : {
3497 [ # # # ]: 0 : switch (action) {
3498 : 0 : case PM_HIBERNATION_PREPARE:
3499 : : case PM_SUSPEND_PREPARE:
3500 : 0 : rcu_expedite_gp();
3501 : 0 : break;
3502 : 0 : case PM_POST_HIBERNATION:
3503 : : case PM_POST_SUSPEND:
3504 : 0 : rcu_unexpedite_gp();
3505 : 0 : break;
3506 : : default:
3507 : : break;
3508 : : }
3509 : 0 : return NOTIFY_OK;
3510 : : }
3511 : :
3512 : : /*
3513 : : * Spawn the kthreads that handle RCU's grace periods.
3514 : : */
3515 : 28 : static int __init rcu_spawn_gp_kthread(void)
3516 : : {
3517 : 28 : unsigned long flags;
3518 : 28 : int kthread_prio_in = kthread_prio;
3519 : 28 : struct rcu_node *rnp;
3520 : 28 : struct sched_param sp;
3521 : 28 : struct task_struct *t;
3522 : :
3523 : : /* Force priority into range. */
3524 : 28 : if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 2
3525 : : && IS_BUILTIN(CONFIG_RCU_TORTURE_TEST))
3526 : : kthread_prio = 2;
3527 : 28 : else if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1)
3528 : : kthread_prio = 1;
3529 [ - + ]: 28 : else if (kthread_prio < 0)
3530 : 0 : kthread_prio = 0;
3531 [ - + ]: 28 : else if (kthread_prio > 99)
3532 : 0 : kthread_prio = 99;
3533 : :
3534 [ - + ]: 28 : if (kthread_prio != kthread_prio_in)
3535 : 0 : pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n",
3536 : : kthread_prio, kthread_prio_in);
3537 : :
3538 : 28 : rcu_scheduler_fully_active = 1;
3539 : 28 : t = kthread_create(rcu_gp_kthread, NULL, "%s", rcu_state.name);
3540 [ - + - - : 28 : if (WARN_ONCE(IS_ERR(t), "%s: Could not start grace-period kthread, OOM is now expected behavior\n", __func__))
+ - ]
3541 : : return 0;
3542 [ - + ]: 28 : if (kthread_prio) {
3543 : 0 : sp.sched_priority = kthread_prio;
3544 : 0 : sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
3545 : : }
3546 : 28 : rnp = rcu_get_root();
3547 : 28 : raw_spin_lock_irqsave_rcu_node(rnp, flags);
3548 : 28 : rcu_state.gp_kthread = t;
3549 : 28 : raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3550 : 28 : wake_up_process(t);
3551 : 28 : rcu_spawn_nocb_kthreads();
3552 : 28 : rcu_spawn_boost_kthreads();
3553 : 28 : return 0;
3554 : : }
3555 : : early_initcall(rcu_spawn_gp_kthread);
3556 : :
3557 : : /*
3558 : : * This function is invoked towards the end of the scheduler's
3559 : : * initialization process. Before this is called, the idle task might
3560 : : * contain synchronous grace-period primitives (during which time, this idle
3561 : : * task is booting the system, and such primitives are no-ops). After this
3562 : : * function is called, any synchronous grace-period primitives are run as
3563 : : * expedited, with the requesting task driving the grace period forward.
3564 : : * A later core_initcall() rcu_set_runtime_mode() will switch to full
3565 : : * runtime RCU functionality.
3566 : : */
3567 : 28 : void rcu_scheduler_starting(void)
3568 : : {
3569 [ - + ]: 28 : WARN_ON(num_online_cpus() != 1);
3570 [ - + ]: 28 : WARN_ON(nr_context_switches() > 0);
3571 : 28 : rcu_test_sync_prims();
3572 : 28 : rcu_scheduler_active = RCU_SCHEDULER_INIT;
3573 : 28 : rcu_test_sync_prims();
3574 : 28 : }
3575 : :
3576 : : /*
3577 : : * Helper function for rcu_init() that initializes the rcu_state structure.
3578 : : */
3579 : 28 : static void __init rcu_init_one(void)
3580 : : {
3581 : 28 : static const char * const buf[] = RCU_NODE_NAME_INIT;
3582 : 28 : static const char * const fqs[] = RCU_FQS_NAME_INIT;
3583 : 28 : static struct lock_class_key rcu_node_class[RCU_NUM_LVLS];
3584 : 28 : static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS];
3585 : :
3586 : 28 : int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */
3587 : 28 : int cpustride = 1;
3588 : 28 : int i;
3589 : 28 : int j;
3590 : 28 : struct rcu_node *rnp;
3591 : :
3592 : 28 : BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */
3593 : :
3594 : : /* Silence gcc 4.8 false positive about array index out of range. */
3595 [ + - ]: 28 : if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS)
3596 : 0 : panic("rcu_init_one: rcu_num_lvls out of range");
3597 : :
3598 : : /* Initialize the level-tracking arrays. */
3599 : :
3600 [ - + ]: 28 : for (i = 1; i < rcu_num_lvls; i++)
3601 : 0 : rcu_state.level[i] =
3602 : 0 : rcu_state.level[i - 1] + num_rcu_lvl[i - 1];
3603 : 28 : rcu_init_levelspread(levelspread, num_rcu_lvl);
3604 : :
3605 : : /* Initialize the elements themselves, starting from the leaves. */
3606 : :
3607 [ + + ]: 56 : for (i = rcu_num_lvls - 1; i >= 0; i--) {
3608 : 28 : cpustride *= levelspread[i];
3609 : 28 : rnp = rcu_state.level[i];
3610 [ + + ]: 56 : for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) {
3611 : 28 : raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock));
3612 : 28 : lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock),
3613 : : &rcu_node_class[i], buf[i]);
3614 : 28 : raw_spin_lock_init(&rnp->fqslock);
3615 : 28 : lockdep_set_class_and_name(&rnp->fqslock,
3616 : : &rcu_fqs_class[i], fqs[i]);
3617 : 28 : rnp->gp_seq = rcu_state.gp_seq;
3618 : 28 : rnp->gp_seq_needed = rcu_state.gp_seq;
3619 : 28 : rnp->completedqs = rcu_state.gp_seq;
3620 : 28 : rnp->qsmask = 0;
3621 : 28 : rnp->qsmaskinit = 0;
3622 : 28 : rnp->grplo = j * cpustride;
3623 : 28 : rnp->grphi = (j + 1) * cpustride - 1;
3624 [ - + ]: 28 : if (rnp->grphi >= nr_cpu_ids)
3625 : 0 : rnp->grphi = nr_cpu_ids - 1;
3626 [ + - ]: 28 : if (i == 0) {
3627 : 28 : rnp->grpnum = 0;
3628 : 28 : rnp->grpmask = 0;
3629 : 28 : rnp->parent = NULL;
3630 : : } else {
3631 : 0 : rnp->grpnum = j % levelspread[i - 1];
3632 : 0 : rnp->grpmask = BIT(rnp->grpnum);
3633 : 0 : rnp->parent = rcu_state.level[i - 1] +
3634 : 0 : j / levelspread[i - 1];
3635 : : }
3636 : 28 : rnp->level = i;
3637 : 28 : INIT_LIST_HEAD(&rnp->blkd_tasks);
3638 : 28 : rcu_init_one_nocb(rnp);
3639 : 28 : init_waitqueue_head(&rnp->exp_wq[0]);
3640 : 28 : init_waitqueue_head(&rnp->exp_wq[1]);
3641 : 28 : init_waitqueue_head(&rnp->exp_wq[2]);
3642 : 28 : init_waitqueue_head(&rnp->exp_wq[3]);
3643 : 28 : spin_lock_init(&rnp->exp_lock);
3644 : : }
3645 : : }
3646 : :
3647 : 28 : init_swait_queue_head(&rcu_state.gp_wq);
3648 : 28 : init_swait_queue_head(&rcu_state.expedited_wq);
3649 : 28 : rnp = rcu_first_leaf_node();
3650 [ + + ]: 56 : for_each_possible_cpu(i) {
3651 [ - + ]: 28 : while (i > rnp->grphi)
3652 : 0 : rnp++;
3653 : 28 : per_cpu_ptr(&rcu_data, i)->mynode = rnp;
3654 : 28 : rcu_boot_init_percpu_data(i);
3655 : : }
3656 : 28 : }
3657 : :
3658 : : /*
3659 : : * Compute the rcu_node tree geometry from kernel parameters. This cannot
3660 : : * replace the definitions in tree.h because those are needed to size
3661 : : * the ->node array in the rcu_state structure.
3662 : : */
3663 : 28 : static void __init rcu_init_geometry(void)
3664 : : {
3665 : 28 : ulong d;
3666 : 28 : int i;
3667 : 28 : int rcu_capacity[RCU_NUM_LVLS];
3668 : :
3669 : : /*
3670 : : * Initialize any unspecified boot parameters.
3671 : : * The default values of jiffies_till_first_fqs and
3672 : : * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS
3673 : : * value, which is a function of HZ, then adding one for each
3674 : : * RCU_JIFFIES_FQS_DIV CPUs that might be on the system.
3675 : : */
3676 : 28 : d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV;
3677 [ + - ]: 28 : if (jiffies_till_first_fqs == ULONG_MAX)
3678 : 28 : jiffies_till_first_fqs = d;
3679 [ + - ]: 28 : if (jiffies_till_next_fqs == ULONG_MAX)
3680 : 28 : jiffies_till_next_fqs = d;
3681 : 28 : adjust_jiffies_till_sched_qs();
3682 : :
3683 : : /* If the compile-time values are accurate, just leave. */
3684 [ + - ]: 28 : if (rcu_fanout_leaf == RCU_FANOUT_LEAF &&
3685 [ + - ]: 28 : nr_cpu_ids == NR_CPUS)
3686 : 0 : return;
3687 : 28 : pr_info("Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n",
3688 : : rcu_fanout_leaf, nr_cpu_ids);
3689 : :
3690 : : /*
3691 : : * The boot-time rcu_fanout_leaf parameter must be at least two
3692 : : * and cannot exceed the number of bits in the rcu_node masks.
3693 : : * Complain and fall back to the compile-time values if this
3694 : : * limit is exceeded.
3695 : : */
3696 [ - + ]: 28 : if (rcu_fanout_leaf < 2 ||
3697 : : rcu_fanout_leaf > sizeof(unsigned long) * 8) {
3698 : 0 : rcu_fanout_leaf = RCU_FANOUT_LEAF;
3699 : 0 : WARN_ON(1);
3700 : 0 : return;
3701 : : }
3702 : :
3703 : : /*
3704 : : * Compute number of nodes that can be handled an rcu_node tree
3705 : : * with the given number of levels.
3706 : : */
3707 : 28 : rcu_capacity[0] = rcu_fanout_leaf;
3708 [ + + ]: 56 : for (i = 1; i < RCU_NUM_LVLS; i++)
3709 : 28 : rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT;
3710 : :
3711 : : /*
3712 : : * The tree must be able to accommodate the configured number of CPUs.
3713 : : * If this limit is exceeded, fall back to the compile-time values.
3714 : : */
3715 [ - + ]: 28 : if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) {
3716 : 0 : rcu_fanout_leaf = RCU_FANOUT_LEAF;
3717 : 0 : WARN_ON(1);
3718 : 0 : return;
3719 : : }
3720 : :
3721 : : /* Calculate the number of levels in the tree. */
3722 [ - + ]: 28 : for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) {
3723 : 0 : }
3724 : 28 : rcu_num_lvls = i + 1;
3725 : :
3726 : : /* Calculate the number of rcu_nodes at each level of the tree. */
3727 [ + + ]: 56 : for (i = 0; i < rcu_num_lvls; i++) {
3728 : 28 : int cap = rcu_capacity[(rcu_num_lvls - 1) - i];
3729 : 28 : num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap);
3730 : : }
3731 : :
3732 : : /* Calculate the total number of rcu_node structures. */
3733 : 28 : rcu_num_nodes = 0;
3734 [ + + ]: 56 : for (i = 0; i < rcu_num_lvls; i++)
3735 : 28 : rcu_num_nodes += num_rcu_lvl[i];
3736 : : }
3737 : :
3738 : : /*
3739 : : * Dump out the structure of the rcu_node combining tree associated
3740 : : * with the rcu_state structure.
3741 : : */
3742 : 0 : static void __init rcu_dump_rcu_node_tree(void)
3743 : : {
3744 : 0 : int level = 0;
3745 : 0 : struct rcu_node *rnp;
3746 : :
3747 : 0 : pr_info("rcu_node tree layout dump\n");
3748 : 0 : pr_info(" ");
3749 [ # # ]: 0 : rcu_for_each_node_breadth_first(rnp) {
3750 [ # # ]: 0 : if (rnp->level != level) {
3751 : 0 : pr_cont("\n");
3752 : 0 : pr_info(" ");
3753 : 0 : level = rnp->level;
3754 : : }
3755 : 0 : pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum);
3756 : : }
3757 : 0 : pr_cont("\n");
3758 : 0 : }
3759 : :
3760 : : struct workqueue_struct *rcu_gp_wq;
3761 : : struct workqueue_struct *rcu_par_gp_wq;
3762 : :
3763 : 28 : static void __init kfree_rcu_batch_init(void)
3764 : : {
3765 : 28 : int cpu;
3766 : 28 : int i;
3767 : :
3768 [ + + ]: 56 : for_each_possible_cpu(cpu) {
3769 : 28 : struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
3770 : :
3771 : 28 : spin_lock_init(&krcp->lock);
3772 [ + + ]: 84 : for (i = 0; i < KFREE_N_BATCHES; i++)
3773 : 56 : krcp->krw_arr[i].krcp = krcp;
3774 : 28 : INIT_DELAYED_WORK(&krcp->monitor_work, kfree_rcu_monitor);
3775 : 28 : krcp->initialized = true;
3776 : : }
3777 : 28 : }
3778 : :
3779 : 28 : void __init rcu_init(void)
3780 : : {
3781 : 28 : int cpu;
3782 : :
3783 : 28 : rcu_early_boot_tests();
3784 : :
3785 : 28 : kfree_rcu_batch_init();
3786 : 28 : rcu_bootup_announce();
3787 : 28 : rcu_init_geometry();
3788 : 28 : rcu_init_one();
3789 [ - + ]: 28 : if (dump_tree)
3790 : 0 : rcu_dump_rcu_node_tree();
3791 [ + - ]: 28 : if (use_softirq)
3792 : 28 : open_softirq(RCU_SOFTIRQ, rcu_core_si);
3793 : :
3794 : : /*
3795 : : * We don't need protection against CPU-hotplug here because
3796 : : * this is called early in boot, before either interrupts
3797 : : * or the scheduler are operational.
3798 : : */
3799 : 28 : pm_notifier(rcu_pm_notify, 0);
3800 [ + + ]: 84 : for_each_online_cpu(cpu) {
3801 : 28 : rcutree_prepare_cpu(cpu);
3802 : 28 : rcu_cpu_starting(cpu);
3803 : 28 : rcutree_online_cpu(cpu);
3804 : : }
3805 : :
3806 : : /* Create workqueue for expedited GPs and for Tree SRCU. */
3807 : 28 : rcu_gp_wq = alloc_workqueue("rcu_gp", WQ_MEM_RECLAIM, 0);
3808 [ - + ]: 28 : WARN_ON(!rcu_gp_wq);
3809 : 28 : rcu_par_gp_wq = alloc_workqueue("rcu_par_gp", WQ_MEM_RECLAIM, 0);
3810 [ - + ]: 28 : WARN_ON(!rcu_par_gp_wq);
3811 : 28 : srcu_init();
3812 : 28 : }
3813 : :
3814 : : #include "tree_stall.h"
3815 : : #include "tree_exp.h"
3816 : : #include "tree_plugin.h"
|
