~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/rcu/tree.c

Version: ~ [ linux-4.18 ] ~ [ linux-4.17.14 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.62 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.119 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.147 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.118 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.57 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~ [ linux-next-20180810 ] ~ [ linux-next-20180813 ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Read-Copy Update mechanism for mutual exclusion
  3  *
  4  * This program is free software; you can redistribute it and/or modify
  5  * it under the terms of the GNU General Public License as published by
  6  * the Free Software Foundation; either version 2 of the License, or
  7  * (at your option) any later version.
  8  *
  9  * This program is distributed in the hope that it will be useful,
 10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 12  * GNU General Public License for more details.
 13  *
 14  * You should have received a copy of the GNU General Public License
 15  * along with this program; if not, you can access it online at
 16  * http://www.gnu.org/licenses/gpl-2.0.html.
 17  *
 18  * Copyright IBM Corporation, 2008
 19  *
 20  * Authors: Dipankar Sarma <dipankar@in.ibm.com>
 21  *          Manfred Spraul <manfred@colorfullife.com>
 22  *          Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
 23  *
 24  * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
 25  * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
 26  *
 27  * For detailed explanation of Read-Copy Update mechanism see -
 28  *      Documentation/RCU
 29  */
 30 #include <linux/types.h>
 31 #include <linux/kernel.h>
 32 #include <linux/init.h>
 33 #include <linux/spinlock.h>
 34 #include <linux/smp.h>
 35 #include <linux/rcupdate_wait.h>
 36 #include <linux/interrupt.h>
 37 #include <linux/sched.h>
 38 #include <linux/sched/debug.h>
 39 #include <linux/nmi.h>
 40 #include <linux/atomic.h>
 41 #include <linux/bitops.h>
 42 #include <linux/export.h>
 43 #include <linux/completion.h>
 44 #include <linux/moduleparam.h>
 45 #include <linux/percpu.h>
 46 #include <linux/notifier.h>
 47 #include <linux/cpu.h>
 48 #include <linux/mutex.h>
 49 #include <linux/time.h>
 50 #include <linux/kernel_stat.h>
 51 #include <linux/wait.h>
 52 #include <linux/kthread.h>
 53 #include <uapi/linux/sched/types.h>
 54 #include <linux/prefetch.h>
 55 #include <linux/delay.h>
 56 #include <linux/stop_machine.h>
 57 #include <linux/random.h>
 58 #include <linux/trace_events.h>
 59 #include <linux/suspend.h>
 60 #include <linux/ftrace.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  * In order to export the rcu_state name to the tracing tools, it
 74  * needs to be added in the __tracepoint_string section.
 75  * This requires defining a separate variable tp_<sname>_varname
 76  * that points to the string being used, and this will allow
 77  * the tracing userspace tools to be able to decipher the string
 78  * address to the matching string.
 79  */
 80 #ifdef CONFIG_TRACING
 81 # define DEFINE_RCU_TPS(sname) \
 82 static char sname##_varname[] = #sname; \
 83 static const char *tp_##sname##_varname __used __tracepoint_string = sname##_varname;
 84 # define RCU_STATE_NAME(sname) sname##_varname
 85 #else
 86 # define DEFINE_RCU_TPS(sname)
 87 # define RCU_STATE_NAME(sname) __stringify(sname)
 88 #endif
 89 
 90 #define RCU_STATE_INITIALIZER(sname, sabbr, cr) \
 91 DEFINE_RCU_TPS(sname) \
 92 static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, sname##_data); \
 93 struct rcu_state sname##_state = { \
 94         .level = { &sname##_state.node[0] }, \
 95         .rda = &sname##_data, \
 96         .call = cr, \
 97         .gp_state = RCU_GP_IDLE, \
 98         .gpnum = 0UL - 300UL, \
 99         .completed = 0UL - 300UL, \
100         .barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \
101         .name = RCU_STATE_NAME(sname), \
102         .abbr = sabbr, \
103         .exp_mutex = __MUTEX_INITIALIZER(sname##_state.exp_mutex), \
104         .exp_wake_mutex = __MUTEX_INITIALIZER(sname##_state.exp_wake_mutex), \
105 }
106 
107 RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched);
108 RCU_STATE_INITIALIZER(rcu_bh, 'b', call_rcu_bh);
109 
110 static struct rcu_state *const rcu_state_p;
111 LIST_HEAD(rcu_struct_flavors);
112 
113 /* Dump rcu_node combining tree at boot to verify correct setup. */
114 static bool dump_tree;
115 module_param(dump_tree, bool, 0444);
116 /* Control rcu_node-tree auto-balancing at boot time. */
117 static bool rcu_fanout_exact;
118 module_param(rcu_fanout_exact, bool, 0444);
119 /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */
120 static int rcu_fanout_leaf = RCU_FANOUT_LEAF;
121 module_param(rcu_fanout_leaf, int, 0444);
122 int rcu_num_lvls __read_mostly = RCU_NUM_LVLS;
123 /* Number of rcu_nodes at specified level. */
124 int num_rcu_lvl[] = NUM_RCU_LVL_INIT;
125 int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */
126 /* panic() on RCU Stall sysctl. */
127 int sysctl_panic_on_rcu_stall __read_mostly;
128 
129 /*
130  * The rcu_scheduler_active variable is initialized to the value
131  * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the
132  * first task is spawned.  So when this variable is RCU_SCHEDULER_INACTIVE,
133  * RCU can assume that there is but one task, allowing RCU to (for example)
134  * optimize synchronize_rcu() to a simple barrier().  When this variable
135  * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required
136  * to detect real grace periods.  This variable is also used to suppress
137  * boot-time false positives from lockdep-RCU error checking.  Finally, it
138  * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU
139  * is fully initialized, including all of its kthreads having been spawned.
140  */
141 int rcu_scheduler_active __read_mostly;
142 EXPORT_SYMBOL_GPL(rcu_scheduler_active);
143 
144 /*
145  * The rcu_scheduler_fully_active variable transitions from zero to one
146  * during the early_initcall() processing, which is after the scheduler
147  * is capable of creating new tasks.  So RCU processing (for example,
148  * creating tasks for RCU priority boosting) must be delayed until after
149  * rcu_scheduler_fully_active transitions from zero to one.  We also
150  * currently delay invocation of any RCU callbacks until after this point.
151  *
152  * It might later prove better for people registering RCU callbacks during
153  * early boot to take responsibility for these callbacks, but one step at
154  * a time.
155  */
156 static int rcu_scheduler_fully_active __read_mostly;
157 
158 static void rcu_init_new_rnp(struct rcu_node *rnp_leaf);
159 static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf);
160 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
161 static void invoke_rcu_core(void);
162 static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp);
163 static void rcu_report_exp_rdp(struct rcu_state *rsp,
164                                struct rcu_data *rdp, bool wake);
165 static void sync_sched_exp_online_cleanup(int cpu);
166 
167 /* rcuc/rcub kthread realtime priority */
168 static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0;
169 module_param(kthread_prio, int, 0644);
170 
171 /* Delay in jiffies for grace-period initialization delays, debug only. */
172 
173 static int gp_preinit_delay;
174 module_param(gp_preinit_delay, int, 0444);
175 static int gp_init_delay;
176 module_param(gp_init_delay, int, 0444);
177 static int gp_cleanup_delay;
178 module_param(gp_cleanup_delay, int, 0444);
179 
180 /*
181  * Number of grace periods between delays, normalized by the duration of
182  * the delay.  The longer the delay, the more the grace periods between
183  * each delay.  The reason for this normalization is that it means that,
184  * for non-zero delays, the overall slowdown of grace periods is constant
185  * regardless of the duration of the delay.  This arrangement balances
186  * the need for long delays to increase some race probabilities with the
187  * need for fast grace periods to increase other race probabilities.
188  */
189 #define PER_RCU_NODE_PERIOD 3   /* Number of grace periods between delays. */
190 
191 /*
192  * Track the rcutorture test sequence number and the update version
193  * number within a given test.  The rcutorture_testseq is incremented
194  * on every rcutorture module load and unload, so has an odd value
195  * when a test is running.  The rcutorture_vernum is set to zero
196  * when rcutorture starts and is incremented on each rcutorture update.
197  * These variables enable correlating rcutorture output with the
198  * RCU tracing information.
199  */
200 unsigned long rcutorture_testseq;
201 unsigned long rcutorture_vernum;
202 
203 /*
204  * Compute the mask of online CPUs for the specified rcu_node structure.
205  * This will not be stable unless the rcu_node structure's ->lock is
206  * held, but the bit corresponding to the current CPU will be stable
207  * in most contexts.
208  */
209 unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp)
210 {
211         return READ_ONCE(rnp->qsmaskinitnext);
212 }
213 
214 /*
215  * Return true if an RCU grace period is in progress.  The READ_ONCE()s
216  * permit this function to be invoked without holding the root rcu_node
217  * structure's ->lock, but of course results can be subject to change.
218  */
219 static int rcu_gp_in_progress(struct rcu_state *rsp)
220 {
221         return READ_ONCE(rsp->completed) != READ_ONCE(rsp->gpnum);
222 }
223 
224 /*
225  * Note a quiescent state.  Because we do not need to know
226  * how many quiescent states passed, just if there was at least
227  * one since the start of the grace period, this just sets a flag.
228  * The caller must have disabled preemption.
229  */
230 void rcu_sched_qs(void)
231 {
232         RCU_LOCKDEP_WARN(preemptible(), "rcu_sched_qs() invoked with preemption enabled!!!");
233         if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.s))
234                 return;
235         trace_rcu_grace_period(TPS("rcu_sched"),
236                                __this_cpu_read(rcu_sched_data.gpnum),
237                                TPS("cpuqs"));
238         __this_cpu_write(rcu_sched_data.cpu_no_qs.b.norm, false);
239         if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.b.exp))
240                 return;
241         __this_cpu_write(rcu_sched_data.cpu_no_qs.b.exp, false);
242         rcu_report_exp_rdp(&rcu_sched_state,
243                            this_cpu_ptr(&rcu_sched_data), true);
244 }
245 
246 void rcu_bh_qs(void)
247 {
248         RCU_LOCKDEP_WARN(preemptible(), "rcu_bh_qs() invoked with preemption enabled!!!");
249         if (__this_cpu_read(rcu_bh_data.cpu_no_qs.s)) {
250                 trace_rcu_grace_period(TPS("rcu_bh"),
251                                        __this_cpu_read(rcu_bh_data.gpnum),
252                                        TPS("cpuqs"));
253                 __this_cpu_write(rcu_bh_data.cpu_no_qs.b.norm, false);
254         }
255 }
256 
257 /*
258  * Steal a bit from the bottom of ->dynticks for idle entry/exit
259  * control.  Initially this is for TLB flushing.
260  */
261 #define RCU_DYNTICK_CTRL_MASK 0x1
262 #define RCU_DYNTICK_CTRL_CTR  (RCU_DYNTICK_CTRL_MASK + 1)
263 #ifndef rcu_eqs_special_exit
264 #define rcu_eqs_special_exit() do { } while (0)
265 #endif
266 
267 static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
268         .dynticks_nesting = 1,
269         .dynticks_nmi_nesting = DYNTICK_IRQ_NONIDLE,
270         .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR),
271 };
272 
273 /*
274  * Record entry into an extended quiescent state.  This is only to be
275  * called when not already in an extended quiescent state.
276  */
277 static void rcu_dynticks_eqs_enter(void)
278 {
279         struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
280         int seq;
281 
282         /*
283          * CPUs seeing atomic_add_return() must see prior RCU read-side
284          * critical sections, and we also must force ordering with the
285          * next idle sojourn.
286          */
287         seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks);
288         /* Better be in an extended quiescent state! */
289         WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
290                      (seq & RCU_DYNTICK_CTRL_CTR));
291         /* Better not have special action (TLB flush) pending! */
292         WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
293                      (seq & RCU_DYNTICK_CTRL_MASK));
294 }
295 
296 /*
297  * Record exit from an extended quiescent state.  This is only to be
298  * called from an extended quiescent state.
299  */
300 static void rcu_dynticks_eqs_exit(void)
301 {
302         struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
303         int seq;
304 
305         /*
306          * CPUs seeing atomic_add_return() must see prior idle sojourns,
307          * and we also must force ordering with the next RCU read-side
308          * critical section.
309          */
310         seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks);
311         WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
312                      !(seq & RCU_DYNTICK_CTRL_CTR));
313         if (seq & RCU_DYNTICK_CTRL_MASK) {
314                 atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdtp->dynticks);
315                 smp_mb__after_atomic(); /* _exit after clearing mask. */
316                 /* Prefer duplicate flushes to losing a flush. */
317                 rcu_eqs_special_exit();
318         }
319 }
320 
321 /*
322  * Reset the current CPU's ->dynticks counter to indicate that the
323  * newly onlined CPU is no longer in an extended quiescent state.
324  * This will either leave the counter unchanged, or increment it
325  * to the next non-quiescent value.
326  *
327  * The non-atomic test/increment sequence works because the upper bits
328  * of the ->dynticks counter are manipulated only by the corresponding CPU,
329  * or when the corresponding CPU is offline.
330  */
331 static void rcu_dynticks_eqs_online(void)
332 {
333         struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
334 
335         if (atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR)
336                 return;
337         atomic_add(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks);
338 }
339 
340 /*
341  * Is the current CPU in an extended quiescent state?
342  *
343  * No ordering, as we are sampling CPU-local information.
344  */
345 bool rcu_dynticks_curr_cpu_in_eqs(void)
346 {
347         struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
348 
349         return !(atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR);
350 }
351 
352 /*
353  * Snapshot the ->dynticks counter with full ordering so as to allow
354  * stable comparison of this counter with past and future snapshots.
355  */
356 int rcu_dynticks_snap(struct rcu_dynticks *rdtp)
357 {
358         int snap = atomic_add_return(0, &rdtp->dynticks);
359 
360         return snap & ~RCU_DYNTICK_CTRL_MASK;
361 }
362 
363 /*
364  * Return true if the snapshot returned from rcu_dynticks_snap()
365  * indicates that RCU is in an extended quiescent state.
366  */
367 static bool rcu_dynticks_in_eqs(int snap)
368 {
369         return !(snap & RCU_DYNTICK_CTRL_CTR);
370 }
371 
372 /*
373  * Return true if the CPU corresponding to the specified rcu_dynticks
374  * structure has spent some time in an extended quiescent state since
375  * rcu_dynticks_snap() returned the specified snapshot.
376  */
377 static bool rcu_dynticks_in_eqs_since(struct rcu_dynticks *rdtp, int snap)
378 {
379         return snap != rcu_dynticks_snap(rdtp);
380 }
381 
382 /*
383  * Do a double-increment of the ->dynticks counter to emulate a
384  * momentary idle-CPU quiescent state.
385  */
386 static void rcu_dynticks_momentary_idle(void)
387 {
388         struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
389         int special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR,
390                                         &rdtp->dynticks);
391 
392         /* It is illegal to call this from idle state. */
393         WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR));
394 }
395 
396 /*
397  * Set the special (bottom) bit of the specified CPU so that it
398  * will take special action (such as flushing its TLB) on the
399  * next exit from an extended quiescent state.  Returns true if
400  * the bit was successfully set, or false if the CPU was not in
401  * an extended quiescent state.
402  */
403 bool rcu_eqs_special_set(int cpu)
404 {
405         int old;
406         int new;
407         struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
408 
409         do {
410                 old = atomic_read(&rdtp->dynticks);
411                 if (old & RCU_DYNTICK_CTRL_CTR)
412                         return false;
413                 new = old | RCU_DYNTICK_CTRL_MASK;
414         } while (atomic_cmpxchg(&rdtp->dynticks, old, new) != old);
415         return true;
416 }
417 
418 /*
419  * Let the RCU core know that this CPU has gone through the scheduler,
420  * which is a quiescent state.  This is called when the need for a
421  * quiescent state is urgent, so we burn an atomic operation and full
422  * memory barriers to let the RCU core know about it, regardless of what
423  * this CPU might (or might not) do in the near future.
424  *
425  * We inform the RCU core by emulating a zero-duration dyntick-idle period.
426  *
427  * The caller must have disabled interrupts.
428  */
429 static void rcu_momentary_dyntick_idle(void)
430 {
431         raw_cpu_write(rcu_dynticks.rcu_need_heavy_qs, false);
432         rcu_dynticks_momentary_idle();
433 }
434 
435 /*
436  * Note a context switch.  This is a quiescent state for RCU-sched,
437  * and requires special handling for preemptible RCU.
438  * The caller must have disabled interrupts.
439  */
440 void rcu_note_context_switch(bool preempt)
441 {
442         barrier(); /* Avoid RCU read-side critical sections leaking down. */
443         trace_rcu_utilization(TPS("Start context switch"));
444         rcu_sched_qs();
445         rcu_preempt_note_context_switch(preempt);
446         /* Load rcu_urgent_qs before other flags. */
447         if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs)))
448                 goto out;
449         this_cpu_write(rcu_dynticks.rcu_urgent_qs, false);
450         if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs)))
451                 rcu_momentary_dyntick_idle();
452         this_cpu_inc(rcu_dynticks.rcu_qs_ctr);
453         if (!preempt)
454                 rcu_note_voluntary_context_switch_lite(current);
455 out:
456         trace_rcu_utilization(TPS("End context switch"));
457         barrier(); /* Avoid RCU read-side critical sections leaking up. */
458 }
459 EXPORT_SYMBOL_GPL(rcu_note_context_switch);
460 
461 /*
462  * Register a quiescent state for all RCU flavors.  If there is an
463  * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight
464  * dyntick-idle quiescent state visible to other CPUs (but only for those
465  * RCU flavors in desperate need of a quiescent state, which will normally
466  * be none of them).  Either way, do a lightweight quiescent state for
467  * all RCU flavors.
468  *
469  * The barrier() calls are redundant in the common case when this is
470  * called externally, but just in case this is called from within this
471  * file.
472  *
473  */
474 void rcu_all_qs(void)
475 {
476         unsigned long flags;
477 
478         if (!raw_cpu_read(rcu_dynticks.rcu_urgent_qs))
479                 return;
480         preempt_disable();
481         /* Load rcu_urgent_qs before other flags. */
482         if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs))) {
483                 preempt_enable();
484                 return;
485         }
486         this_cpu_write(rcu_dynticks.rcu_urgent_qs, false);
487         barrier(); /* Avoid RCU read-side critical sections leaking down. */
488         if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) {
489                 local_irq_save(flags);
490                 rcu_momentary_dyntick_idle();
491                 local_irq_restore(flags);
492         }
493         if (unlikely(raw_cpu_read(rcu_sched_data.cpu_no_qs.b.exp)))
494                 rcu_sched_qs();
495         this_cpu_inc(rcu_dynticks.rcu_qs_ctr);
496         barrier(); /* Avoid RCU read-side critical sections leaking up. */
497         preempt_enable();
498 }
499 EXPORT_SYMBOL_GPL(rcu_all_qs);
500 
501 #define DEFAULT_RCU_BLIMIT 10     /* Maximum callbacks per rcu_do_batch. */
502 static long blimit = DEFAULT_RCU_BLIMIT;
503 #define DEFAULT_RCU_QHIMARK 10000 /* If this many pending, ignore blimit. */
504 static long qhimark = DEFAULT_RCU_QHIMARK;
505 #define DEFAULT_RCU_QLOMARK 100   /* Once only this many pending, use blimit. */
506 static long qlowmark = DEFAULT_RCU_QLOMARK;
507 
508 module_param(blimit, long, 0444);
509 module_param(qhimark, long, 0444);
510 module_param(qlowmark, long, 0444);
511 
512 static ulong jiffies_till_first_fqs = ULONG_MAX;
513 static ulong jiffies_till_next_fqs = ULONG_MAX;
514 static bool rcu_kick_kthreads;
515 
516 module_param(jiffies_till_first_fqs, ulong, 0644);
517 module_param(jiffies_till_next_fqs, ulong, 0644);
518 module_param(rcu_kick_kthreads, bool, 0644);
519 
520 /*
521  * How long the grace period must be before we start recruiting
522  * quiescent-state help from rcu_note_context_switch().
523  */
524 static ulong jiffies_till_sched_qs = HZ / 10;
525 module_param(jiffies_till_sched_qs, ulong, 0444);
526 
527 static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *rsp));
528 static void force_quiescent_state(struct rcu_state *rsp);
529 static int rcu_pending(void);
530 
531 /*
532  * Return the number of RCU batches started thus far for debug & stats.
533  */
534 unsigned long rcu_batches_started(void)
535 {
536         return rcu_state_p->gpnum;
537 }
538 EXPORT_SYMBOL_GPL(rcu_batches_started);
539 
540 /*
541  * Return the number of RCU-sched batches started thus far for debug & stats.
542  */
543 unsigned long rcu_batches_started_sched(void)
544 {
545         return rcu_sched_state.gpnum;
546 }
547 EXPORT_SYMBOL_GPL(rcu_batches_started_sched);
548 
549 /*
550  * Return the number of RCU BH batches started thus far for debug & stats.
551  */
552 unsigned long rcu_batches_started_bh(void)
553 {
554         return rcu_bh_state.gpnum;
555 }
556 EXPORT_SYMBOL_GPL(rcu_batches_started_bh);
557 
558 /*
559  * Return the number of RCU batches completed thus far for debug & stats.
560  */
561 unsigned long rcu_batches_completed(void)
562 {
563         return rcu_state_p->completed;
564 }
565 EXPORT_SYMBOL_GPL(rcu_batches_completed);
566 
567 /*
568  * Return the number of RCU-sched batches completed thus far for debug & stats.
569  */
570 unsigned long rcu_batches_completed_sched(void)
571 {
572         return rcu_sched_state.completed;
573 }
574 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
575 
576 /*
577  * Return the number of RCU BH batches completed thus far for debug & stats.
578  */
579 unsigned long rcu_batches_completed_bh(void)
580 {
581         return rcu_bh_state.completed;
582 }
583 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
584 
585 /*
586  * Return the number of RCU expedited batches completed thus far for
587  * debug & stats.  Odd numbers mean that a batch is in progress, even
588  * numbers mean idle.  The value returned will thus be roughly double
589  * the cumulative batches since boot.
590  */
591 unsigned long rcu_exp_batches_completed(void)
592 {
593         return rcu_state_p->expedited_sequence;
594 }
595 EXPORT_SYMBOL_GPL(rcu_exp_batches_completed);
596 
597 /*
598  * Return the number of RCU-sched expedited batches completed thus far
599  * for debug & stats.  Similar to rcu_exp_batches_completed().
600  */
601 unsigned long rcu_exp_batches_completed_sched(void)
602 {
603         return rcu_sched_state.expedited_sequence;
604 }
605 EXPORT_SYMBOL_GPL(rcu_exp_batches_completed_sched);
606 
607 /*
608  * Force a quiescent state.
609  */
610 void rcu_force_quiescent_state(void)
611 {
612         force_quiescent_state(rcu_state_p);
613 }
614 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
615 
616 /*
617  * Force a quiescent state for RCU BH.
618  */
619 void rcu_bh_force_quiescent_state(void)
620 {
621         force_quiescent_state(&rcu_bh_state);
622 }
623 EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);
624 
625 /*
626  * Force a quiescent state for RCU-sched.
627  */
628 void rcu_sched_force_quiescent_state(void)
629 {
630         force_quiescent_state(&rcu_sched_state);
631 }
632 EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);
633 
634 /*
635  * Show the state of the grace-period kthreads.
636  */
637 void show_rcu_gp_kthreads(void)
638 {
639         struct rcu_state *rsp;
640 
641         for_each_rcu_flavor(rsp) {
642                 pr_info("%s: wait state: %d ->state: %#lx\n",
643                         rsp->name, rsp->gp_state, rsp->gp_kthread->state);
644                 /* sched_show_task(rsp->gp_kthread); */
645         }
646 }
647 EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads);
648 
649 /*
650  * Record the number of times rcutorture tests have been initiated and
651  * terminated.  This information allows the debugfs tracing stats to be
652  * correlated to the rcutorture messages, even when the rcutorture module
653  * is being repeatedly loaded and unloaded.  In other words, we cannot
654  * store this state in rcutorture itself.
655  */
656 void rcutorture_record_test_transition(void)
657 {
658         rcutorture_testseq++;
659         rcutorture_vernum = 0;
660 }
661 EXPORT_SYMBOL_GPL(rcutorture_record_test_transition);
662 
663 /*
664  * Send along grace-period-related data for rcutorture diagnostics.
665  */
666 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
667                             unsigned long *gpnum, unsigned long *completed)
668 {
669         struct rcu_state *rsp = NULL;
670 
671         switch (test_type) {
672         case RCU_FLAVOR:
673                 rsp = rcu_state_p;
674                 break;
675         case RCU_BH_FLAVOR:
676                 rsp = &rcu_bh_state;
677                 break;
678         case RCU_SCHED_FLAVOR:
679                 rsp = &rcu_sched_state;
680                 break;
681         default:
682                 break;
683         }
684         if (rsp == NULL)
685                 return;
686         *flags = READ_ONCE(rsp->gp_flags);
687         *gpnum = READ_ONCE(rsp->gpnum);
688         *completed = READ_ONCE(rsp->completed);
689 }
690 EXPORT_SYMBOL_GPL(rcutorture_get_gp_data);
691 
692 /*
693  * Record the number of writer passes through the current rcutorture test.
694  * This is also used to correlate debugfs tracing stats with the rcutorture
695  * messages.
696  */
697 void rcutorture_record_progress(unsigned long vernum)
698 {
699         rcutorture_vernum++;
700 }
701 EXPORT_SYMBOL_GPL(rcutorture_record_progress);
702 
703 /*
704  * Return the root node of the specified rcu_state structure.
705  */
706 static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
707 {
708         return &rsp->node[0];
709 }
710 
711 /*
712  * Enter an RCU extended quiescent state, which can be either the
713  * idle loop or adaptive-tickless usermode execution.
714  *
715  * We crowbar the ->dynticks_nmi_nesting field to zero to allow for
716  * the possibility of usermode upcalls having messed up our count
717  * of interrupt nesting level during the prior busy period.
718  */
719 static void rcu_eqs_enter(bool user)
720 {
721         struct rcu_state *rsp;
722         struct rcu_data *rdp;
723         struct rcu_dynticks *rdtp;
724 
725         rdtp = this_cpu_ptr(&rcu_dynticks);
726         WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0);
727         WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
728                      rdtp->dynticks_nesting == 0);
729         if (rdtp->dynticks_nesting != 1) {
730                 rdtp->dynticks_nesting--;
731                 return;
732         }
733 
734         lockdep_assert_irqs_disabled();
735         trace_rcu_dyntick(TPS("Start"), rdtp->dynticks_nesting, 0, rdtp->dynticks);
736         WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current));
737         for_each_rcu_flavor(rsp) {
738                 rdp = this_cpu_ptr(rsp->rda);
739                 do_nocb_deferred_wakeup(rdp);
740         }
741         rcu_prepare_for_idle();
742         WRITE_ONCE(rdtp->dynticks_nesting, 0); /* Avoid irq-access tearing. */
743         rcu_dynticks_eqs_enter();
744         rcu_dynticks_task_enter();
745 }
746 
747 /**
748  * rcu_idle_enter - inform RCU that current CPU is entering idle
749  *
750  * Enter idle mode, in other words, -leave- the mode in which RCU
751  * read-side critical sections can occur.  (Though RCU read-side
752  * critical sections can occur in irq handlers in idle, a possibility
753  * handled by irq_enter() and irq_exit().)
754  *
755  * If you add or remove a call to rcu_idle_enter(), be sure to test with
756  * CONFIG_RCU_EQS_DEBUG=y.
757  */
758 void rcu_idle_enter(void)
759 {
760         lockdep_assert_irqs_disabled();
761         rcu_eqs_enter(false);
762 }
763 
764 #ifdef CONFIG_NO_HZ_FULL
765 /**
766  * rcu_user_enter - inform RCU that we are resuming userspace.
767  *
768  * Enter RCU idle mode right before resuming userspace.  No use of RCU
769  * is permitted between this call and rcu_user_exit(). This way the
770  * CPU doesn't need to maintain the tick for RCU maintenance purposes
771  * when the CPU runs in userspace.
772  *
773  * If you add or remove a call to rcu_user_enter(), be sure to test with
774  * CONFIG_RCU_EQS_DEBUG=y.
775  */
776 void rcu_user_enter(void)
777 {
778         lockdep_assert_irqs_disabled();
779         rcu_eqs_enter(true);
780 }
781 #endif /* CONFIG_NO_HZ_FULL */
782 
783 /**
784  * rcu_nmi_exit - inform RCU of exit from NMI context
785  *
786  * If we are returning from the outermost NMI handler that interrupted an
787  * RCU-idle period, update rdtp->dynticks and rdtp->dynticks_nmi_nesting
788  * to let the RCU grace-period handling know that the CPU is back to
789  * being RCU-idle.
790  *
791  * If you add or remove a call to rcu_nmi_exit(), be sure to test
792  * with CONFIG_RCU_EQS_DEBUG=y.
793  */
794 void rcu_nmi_exit(void)
795 {
796         struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
797 
798         /*
799          * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks.
800          * (We are exiting an NMI handler, so RCU better be paying attention
801          * to us!)
802          */
803         WARN_ON_ONCE(rdtp->dynticks_nmi_nesting <= 0);
804         WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs());
805 
806         /*
807          * If the nesting level is not 1, the CPU wasn't RCU-idle, so
808          * leave it in non-RCU-idle state.
809          */
810         if (rdtp->dynticks_nmi_nesting != 1) {
811                 trace_rcu_dyntick(TPS("--="), rdtp->dynticks_nmi_nesting, rdtp->dynticks_nmi_nesting - 2, rdtp->dynticks);
812                 WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* No store tearing. */
813                            rdtp->dynticks_nmi_nesting - 2);
814                 return;
815         }
816 
817         /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */
818         trace_rcu_dyntick(TPS("Startirq"), rdtp->dynticks_nmi_nesting, 0, rdtp->dynticks);
819         WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */
820         rcu_dynticks_eqs_enter();
821 }
822 
823 /**
824  * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle
825  *
826  * Exit from an interrupt handler, which might possibly result in entering
827  * idle mode, in other words, leaving the mode in which read-side critical
828  * sections can occur.  The caller must have disabled interrupts.
829  *
830  * This code assumes that the idle loop never does anything that might
831  * result in unbalanced calls to irq_enter() and irq_exit().  If your
832  * architecture's idle loop violates this assumption, RCU will give you what
833  * you deserve, good and hard.  But very infrequently and irreproducibly.
834  *
835  * Use things like work queues to work around this limitation.
836  *
837  * You have been warned.
838  *
839  * If you add or remove a call to rcu_irq_exit(), be sure to test with
840  * CONFIG_RCU_EQS_DEBUG=y.
841  */
842 void rcu_irq_exit(void)
843 {
844         struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
845 
846         lockdep_assert_irqs_disabled();
847         if (rdtp->dynticks_nmi_nesting == 1)
848                 rcu_prepare_for_idle();
849         rcu_nmi_exit();
850         if (rdtp->dynticks_nmi_nesting == 0)
851                 rcu_dynticks_task_enter();
852 }
853 
854 /*
855  * Wrapper for rcu_irq_exit() where interrupts are enabled.
856  *
857  * If you add or remove a call to rcu_irq_exit_irqson(), be sure to test
858  * with CONFIG_RCU_EQS_DEBUG=y.
859  */
860 void rcu_irq_exit_irqson(void)
861 {
862         unsigned long flags;
863 
864         local_irq_save(flags);
865         rcu_irq_exit();
866         local_irq_restore(flags);
867 }
868 
869 /*
870  * Exit an RCU extended quiescent state, which can be either the
871  * idle loop or adaptive-tickless usermode execution.
872  *
873  * We crowbar the ->dynticks_nmi_nesting field to DYNTICK_IRQ_NONIDLE to
874  * allow for the possibility of usermode upcalls messing up our count of
875  * interrupt nesting level during the busy period that is just now starting.
876  */
877 static void rcu_eqs_exit(bool user)
878 {
879         struct rcu_dynticks *rdtp;
880         long oldval;
881 
882         lockdep_assert_irqs_disabled();
883         rdtp = this_cpu_ptr(&rcu_dynticks);
884         oldval = rdtp->dynticks_nesting;
885         WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0);
886         if (oldval) {
887                 rdtp->dynticks_nesting++;
888                 return;
889         }
890         rcu_dynticks_task_exit();
891         rcu_dynticks_eqs_exit();
892         rcu_cleanup_after_idle();
893         trace_rcu_dyntick(TPS("End"), rdtp->dynticks_nesting, 1, rdtp->dynticks);
894         WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current));
895         WRITE_ONCE(rdtp->dynticks_nesting, 1);
896         WRITE_ONCE(rdtp->dynticks_nmi_nesting, DYNTICK_IRQ_NONIDLE);
897 }
898 
899 /**
900  * rcu_idle_exit - inform RCU that current CPU is leaving idle
901  *
902  * Exit idle mode, in other words, -enter- the mode in which RCU
903  * read-side critical sections can occur.
904  *
905  * If you add or remove a call to rcu_idle_exit(), be sure to test with
906  * CONFIG_RCU_EQS_DEBUG=y.
907  */
908 void rcu_idle_exit(void)
909 {
910         unsigned long flags;
911 
912         local_irq_save(flags);
913         rcu_eqs_exit(false);
914         local_irq_restore(flags);
915 }
916 
917 #ifdef CONFIG_NO_HZ_FULL
918 /**
919  * rcu_user_exit - inform RCU that we are exiting userspace.
920  *
921  * Exit RCU idle mode while entering the kernel because it can
922  * run a RCU read side critical section anytime.
923  *
924  * If you add or remove a call to rcu_user_exit(), be sure to test with
925  * CONFIG_RCU_EQS_DEBUG=y.
926  */
927 void rcu_user_exit(void)
928 {
929         rcu_eqs_exit(1);
930 }
931 #endif /* CONFIG_NO_HZ_FULL */
932 
933 /**
934  * rcu_nmi_enter - inform RCU of entry to NMI context
935  *
936  * If the CPU was idle from RCU's viewpoint, update rdtp->dynticks and
937  * rdtp->dynticks_nmi_nesting to let the RCU grace-period handling know
938  * that the CPU is active.  This implementation permits nested NMIs, as
939  * long as the nesting level does not overflow an int.  (You will probably
940  * run out of stack space first.)
941  *
942  * If you add or remove a call to rcu_nmi_enter(), be sure to test
943  * with CONFIG_RCU_EQS_DEBUG=y.
944  */
945 void rcu_nmi_enter(void)
946 {
947         struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
948         long incby = 2;
949 
950         /* Complain about underflow. */
951         WARN_ON_ONCE(rdtp->dynticks_nmi_nesting < 0);
952 
953         /*
954          * If idle from RCU viewpoint, atomically increment ->dynticks
955          * to mark non-idle and increment ->dynticks_nmi_nesting by one.
956          * Otherwise, increment ->dynticks_nmi_nesting by two.  This means
957          * if ->dynticks_nmi_nesting is equal to one, we are guaranteed
958          * to be in the outermost NMI handler that interrupted an RCU-idle
959          * period (observation due to Andy Lutomirski).
960          */
961         if (rcu_dynticks_curr_cpu_in_eqs()) {
962                 rcu_dynticks_eqs_exit();
963                 incby = 1;
964         }
965         trace_rcu_dyntick(incby == 1 ? TPS("Endirq") : TPS("++="),
966                           rdtp->dynticks_nmi_nesting,
967                           rdtp->dynticks_nmi_nesting + incby, rdtp->dynticks);
968         WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* Prevent store tearing. */
969                    rdtp->dynticks_nmi_nesting + incby);
970         barrier();
971 }
972 
973 /**
974  * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle
975  *
976  * Enter an interrupt handler, which might possibly result in exiting
977  * idle mode, in other words, entering the mode in which read-side critical
978  * sections can occur.  The caller must have disabled interrupts.
979  *
980  * Note that the Linux kernel is fully capable of entering an interrupt
981  * handler that it never exits, for example when doing upcalls to user mode!
982  * This code assumes that the idle loop never does upcalls to user mode.
983  * If your architecture's idle loop does do upcalls to user mode (or does
984  * anything else that results in unbalanced calls to the irq_enter() and
985  * irq_exit() functions), RCU will give you what you deserve, good and hard.
986  * But very infrequently and irreproducibly.
987  *
988  * Use things like work queues to work around this limitation.
989  *
990  * You have been warned.
991  *
992  * If you add or remove a call to rcu_irq_enter(), be sure to test with
993  * CONFIG_RCU_EQS_DEBUG=y.
994  */
995 void rcu_irq_enter(void)
996 {
997         struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
998 
999         lockdep_assert_irqs_disabled();
1000         if (rdtp->dynticks_nmi_nesting == 0)
1001                 rcu_dynticks_task_exit();
1002         rcu_nmi_enter();
1003         if (rdtp->dynticks_nmi_nesting == 1)
1004                 rcu_cleanup_after_idle();
1005 }
1006 
1007 /*
1008  * Wrapper for rcu_irq_enter() where interrupts are enabled.
1009  *
1010  * If you add or remove a call to rcu_irq_enter_irqson(), be sure to test
1011  * with CONFIG_RCU_EQS_DEBUG=y.
1012  */
1013 void rcu_irq_enter_irqson(void)
1014 {
1015         unsigned long flags;
1016 
1017         local_irq_save(flags);
1018         rcu_irq_enter();
1019         local_irq_restore(flags);
1020 }
1021 
1022 /**
1023  * rcu_is_watching - see if RCU thinks that the current CPU is idle
1024  *
1025  * Return true if RCU is watching the running CPU, which means that this
1026  * CPU can safely enter RCU read-side critical sections.  In other words,
1027  * if the current CPU is in its idle loop and is neither in an interrupt
1028  * or NMI handler, return true.
1029  */
1030 bool notrace rcu_is_watching(void)
1031 {
1032         bool ret;
1033 
1034         preempt_disable_notrace();
1035         ret = !rcu_dynticks_curr_cpu_in_eqs();
1036         preempt_enable_notrace();
1037         return ret;
1038 }
1039 EXPORT_SYMBOL_GPL(rcu_is_watching);
1040 
1041 /*
1042  * If a holdout task is actually running, request an urgent quiescent
1043  * state from its CPU.  This is unsynchronized, so migrations can cause
1044  * the request to go to the wrong CPU.  Which is OK, all that will happen
1045  * is that the CPU's next context switch will be a bit slower and next
1046  * time around this task will generate another request.
1047  */
1048 void rcu_request_urgent_qs_task(struct task_struct *t)
1049 {
1050         int cpu;
1051 
1052         barrier();
1053         cpu = task_cpu(t);
1054         if (!task_curr(t))
1055                 return; /* This task is not running on that CPU. */
1056         smp_store_release(per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, cpu), true);
1057 }
1058 
1059 #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
1060 
1061 /*
1062  * Is the current CPU online?  Disable preemption to avoid false positives
1063  * that could otherwise happen due to the current CPU number being sampled,
1064  * this task being preempted, its old CPU being taken offline, resuming
1065  * on some other CPU, then determining that its old CPU is now offline.
1066  * It is OK to use RCU on an offline processor during initial boot, hence
1067  * the check for rcu_scheduler_fully_active.  Note also that it is OK
1068  * for a CPU coming online to use RCU for one jiffy prior to marking itself
1069  * online in the cpu_online_mask.  Similarly, it is OK for a CPU going
1070  * offline to continue to use RCU for one jiffy after marking itself
1071  * offline in the cpu_online_mask.  This leniency is necessary given the
1072  * non-atomic nature of the online and offline processing, for example,
1073  * the fact that a CPU enters the scheduler after completing the teardown
1074  * of the CPU.
1075  *
1076  * This is also why RCU internally marks CPUs online during in the
1077  * preparation phase and offline after the CPU has been taken down.
1078  *
1079  * Disable checking if in an NMI handler because we cannot safely report
1080  * errors from NMI handlers anyway.
1081  */
1082 bool rcu_lockdep_current_cpu_online(void)
1083 {
1084         struct rcu_data *rdp;
1085         struct rcu_node *rnp;
1086         bool ret;
1087 
1088         if (in_nmi())
1089                 return true;
1090         preempt_disable();
1091         rdp = this_cpu_ptr(&rcu_sched_data);
1092         rnp = rdp->mynode;
1093         ret = (rdp->grpmask & rcu_rnp_online_cpus(rnp)) ||
1094               !rcu_scheduler_fully_active;
1095         preempt_enable();
1096         return ret;
1097 }
1098 EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online);
1099 
1100 #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */
1101 
1102 /**
1103  * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle
1104  *
1105  * If the current CPU is idle or running at a first-level (not nested)
1106  * interrupt from idle, return true.  The caller must have at least
1107  * disabled preemption.
1108  */
1109 static int rcu_is_cpu_rrupt_from_idle(void)
1110 {
1111         return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 0 &&
1112                __this_cpu_read(rcu_dynticks.dynticks_nmi_nesting) <= 1;
1113 }
1114 
1115 /*
1116  * We are reporting a quiescent state on behalf of some other CPU, so
1117  * it is our responsibility to check for and handle potential overflow
1118  * of the rcu_node ->gpnum counter with respect to the rcu_data counters.
1119  * After all, the CPU might be in deep idle state, and thus executing no
1120  * code whatsoever.
1121  */
1122 static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp)
1123 {
1124         raw_lockdep_assert_held_rcu_node(rnp);
1125         if (ULONG_CMP_LT(READ_ONCE(rdp->gpnum) + ULONG_MAX / 4, rnp->gpnum))
1126                 WRITE_ONCE(rdp->gpwrap, true);
1127         if (ULONG_CMP_LT(rdp->rcu_iw_gpnum + ULONG_MAX / 4, rnp->gpnum))
1128                 rdp->rcu_iw_gpnum = rnp->gpnum + ULONG_MAX / 4;
1129 }
1130 
1131 /*
1132  * Snapshot the specified CPU's dynticks counter so that we can later
1133  * credit them with an implicit quiescent state.  Return 1 if this CPU
1134  * is in dynticks idle mode, which is an extended quiescent state.
1135  */
1136 static int dyntick_save_progress_counter(struct rcu_data *rdp)
1137 {
1138         rdp->dynticks_snap = rcu_dynticks_snap(rdp->dynticks);
1139         if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) {
1140                 trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
1141                 rcu_gpnum_ovf(rdp->mynode, rdp);
1142                 return 1;
1143         }
1144         return 0;
1145 }
1146 
1147 /*
1148  * Handler for the irq_work request posted when a grace period has
1149  * gone on for too long, but not yet long enough for an RCU CPU
1150  * stall warning.  Set state appropriately, but just complain if
1151  * there is unexpected state on entry.
1152  */
1153 static void rcu_iw_handler(struct irq_work *iwp)
1154 {
1155         struct rcu_data *rdp;
1156         struct rcu_node *rnp;
1157 
1158         rdp = container_of(iwp, struct rcu_data, rcu_iw);
1159         rnp = rdp->mynode;
1160         raw_spin_lock_rcu_node(rnp);
1161         if (!WARN_ON_ONCE(!rdp->rcu_iw_pending)) {
1162                 rdp->rcu_iw_gpnum = rnp->gpnum;
1163                 rdp->rcu_iw_pending = false;
1164         }
1165         raw_spin_unlock_rcu_node(rnp);
1166 }
1167 
1168 /*
1169  * Return true if the specified CPU has passed through a quiescent
1170  * state by virtue of being in or having passed through an dynticks
1171  * idle state since the last call to dyntick_save_progress_counter()
1172  * for this same CPU, or by virtue of having been offline.
1173  */
1174 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
1175 {
1176         unsigned long jtsq;
1177         bool *rnhqp;
1178         bool *ruqp;
1179         struct rcu_node *rnp = rdp->mynode;
1180 
1181         /*
1182          * If the CPU passed through or entered a dynticks idle phase with
1183          * no active irq/NMI handlers, then we can safely pretend that the CPU
1184          * already acknowledged the request to pass through a quiescent
1185          * state.  Either way, that CPU cannot possibly be in an RCU
1186          * read-side critical section that started before the beginning
1187          * of the current RCU grace period.
1188          */
1189         if (rcu_dynticks_in_eqs_since(rdp->dynticks, rdp->dynticks_snap)) {
1190                 trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
1191                 rdp->dynticks_fqs++;
1192                 rcu_gpnum_ovf(rnp, rdp);
1193                 return 1;
1194         }
1195 
1196         /*
1197          * Has this CPU encountered a cond_resched() since the beginning
1198          * of the grace period?  For this to be the case, the CPU has to
1199          * have noticed the current grace period.  This might not be the
1200          * case for nohz_full CPUs looping in the kernel.
1201          */
1202         jtsq = jiffies_till_sched_qs;
1203         ruqp = per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, rdp->cpu);
1204         if (time_after(jiffies, rdp->rsp->gp_start + jtsq) &&
1205             READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) &&
1206             READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) {
1207                 trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc"));
1208                 rcu_gpnum_ovf(rnp, rdp);
1209                 return 1;
1210         } else if (time_after(jiffies, rdp->rsp->gp_start + jtsq)) {
1211                 /* Load rcu_qs_ctr before store to rcu_urgent_qs. */
1212                 smp_store_release(ruqp, true);
1213         }
1214 
1215         /* Check for the CPU being offline. */
1216         if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp))) {
1217                 trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl"));
1218                 rdp->offline_fqs++;
1219                 rcu_gpnum_ovf(rnp, rdp);
1220                 return 1;
1221         }
1222 
1223         /*
1224          * A CPU running for an extended time within the kernel can
1225          * delay RCU grace periods.  When the CPU is in NO_HZ_FULL mode,
1226          * even context-switching back and forth between a pair of
1227          * in-kernel CPU-bound tasks cannot advance grace periods.
1228          * So if the grace period is old enough, make the CPU pay attention.
1229          * Note that the unsynchronized assignments to the per-CPU
1230          * rcu_need_heavy_qs variable are safe.  Yes, setting of
1231          * bits can be lost, but they will be set again on the next
1232          * force-quiescent-state pass.  So lost bit sets do not result
1233          * in incorrect behavior, merely in a grace period lasting
1234          * a few jiffies longer than it might otherwise.  Because
1235          * there are at most four threads involved, and because the
1236          * updates are only once every few jiffies, the probability of
1237          * lossage (and thus of slight grace-period extension) is
1238          * quite low.
1239          */
1240         rnhqp = &per_cpu(rcu_dynticks.rcu_need_heavy_qs, rdp->cpu);
1241         if (!READ_ONCE(*rnhqp) &&
1242             (time_after(jiffies, rdp->rsp->gp_start + jtsq) ||
1243              time_after(jiffies, rdp->rsp->jiffies_resched))) {
1244                 WRITE_ONCE(*rnhqp, true);
1245                 /* Store rcu_need_heavy_qs before rcu_urgent_qs. */
1246                 smp_store_release(ruqp, true);
1247                 rdp->rsp->jiffies_resched += jtsq; /* Re-enable beating. */
1248         }
1249 
1250         /*
1251          * If more than halfway to RCU CPU stall-warning time, do a
1252          * resched_cpu() to try to loosen things up a bit.  Also check to
1253          * see if the CPU is getting hammered with interrupts, but only
1254          * once per grace period, just to keep the IPIs down to a dull roar.
1255          */
1256         if (jiffies - rdp->rsp->gp_start > rcu_jiffies_till_stall_check() / 2) {
1257                 resched_cpu(rdp->cpu);
1258                 if (IS_ENABLED(CONFIG_IRQ_WORK) &&
1259                     !rdp->rcu_iw_pending && rdp->rcu_iw_gpnum != rnp->gpnum &&
1260                     (rnp->ffmask & rdp->grpmask)) {
1261                         init_irq_work(&rdp->rcu_iw, rcu_iw_handler);
1262                         rdp->rcu_iw_pending = true;
1263                         rdp->rcu_iw_gpnum = rnp->gpnum;
1264                         irq_work_queue_on(&rdp->rcu_iw, rdp->cpu);
1265                 }
1266         }
1267 
1268         return 0;
1269 }
1270 
1271 static void record_gp_stall_check_time(struct rcu_state *rsp)
1272 {
1273         unsigned long j = jiffies;
1274         unsigned long j1;
1275 
1276         rsp->gp_start = j;
1277         smp_wmb(); /* Record start time before stall time. */
1278         j1 = rcu_jiffies_till_stall_check();
1279         WRITE_ONCE(rsp->jiffies_stall, j + j1);
1280         rsp->jiffies_resched = j + j1 / 2;
1281         rsp->n_force_qs_gpstart = READ_ONCE(rsp->n_force_qs);
1282 }
1283 
1284 /*
1285  * Convert a ->gp_state value to a character string.
1286  */
1287 static const char *gp_state_getname(short gs)
1288 {
1289         if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names))
1290                 return "???";
1291         return gp_state_names[gs];
1292 }
1293 
1294 /*
1295  * Complain about starvation of grace-period kthread.
1296  */
1297 static void rcu_check_gp_kthread_starvation(struct rcu_state *rsp)
1298 {
1299         unsigned long gpa;
1300         unsigned long j;
1301 
1302         j = jiffies;
1303         gpa = READ_ONCE(rsp->gp_activity);
1304         if (j - gpa > 2 * HZ) {
1305                 pr_err("%s kthread starved for %ld jiffies! g%lu c%lu f%#x %s(%d) ->state=%#lx ->cpu=%d\n",
1306                        rsp->name, j - gpa,
1307                        rsp->gpnum, rsp->completed,
1308                        rsp->gp_flags,
1309                        gp_state_getname(rsp->gp_state), rsp->gp_state,
1310                        rsp->gp_kthread ? rsp->gp_kthread->state : ~0,
1311                        rsp->gp_kthread ? task_cpu(rsp->gp_kthread) : -1);
1312                 if (rsp->gp_kthread) {
1313                         pr_err("RCU grace-period kthread stack dump:\n");
1314                         sched_show_task(rsp->gp_kthread);
1315                         wake_up_process(rsp->gp_kthread);
1316                 }
1317         }
1318 }
1319 
1320 /*
1321  * Dump stacks of all tasks running on stalled CPUs.  First try using
1322  * NMIs, but fall back to manual remote stack tracing on architectures
1323  * that don't support NMI-based stack dumps.  The NMI-triggered stack
1324  * traces are more accurate because they are printed by the target CPU.
1325  */
1326 static void rcu_dump_cpu_stacks(struct rcu_state *rsp)
1327 {
1328         int cpu;
1329         unsigned long flags;
1330         struct rcu_node *rnp;
1331 
1332         rcu_for_each_leaf_node(rsp, rnp) {
1333                 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1334                 for_each_leaf_node_possible_cpu(rnp, cpu)
1335                         if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu))
1336                                 if (!trigger_single_cpu_backtrace(cpu))
1337                                         dump_cpu_task(cpu);
1338                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1339         }
1340 }
1341 
1342 /*
1343  * If too much time has passed in the current grace period, and if
1344  * so configured, go kick the relevant kthreads.
1345  */
1346 static void rcu_stall_kick_kthreads(struct rcu_state *rsp)
1347 {
1348         unsigned long j;
1349 
1350         if (!rcu_kick_kthreads)
1351                 return;
1352         j = READ_ONCE(rsp->jiffies_kick_kthreads);
1353         if (time_after(jiffies, j) && rsp->gp_kthread &&
1354             (rcu_gp_in_progress(rsp) || READ_ONCE(rsp->gp_flags))) {
1355                 WARN_ONCE(1, "Kicking %s grace-period kthread\n", rsp->name);
1356                 rcu_ftrace_dump(DUMP_ALL);
1357                 wake_up_process(rsp->gp_kthread);
1358                 WRITE_ONCE(rsp->jiffies_kick_kthreads, j + HZ);
1359         }
1360 }
1361 
1362 static inline void panic_on_rcu_stall(void)
1363 {
1364         if (sysctl_panic_on_rcu_stall)
1365                 panic("RCU Stall\n");
1366 }
1367 
1368 static void print_other_cpu_stall(struct rcu_state *rsp, unsigned long gpnum)
1369 {
1370         int cpu;
1371         long delta;
1372         unsigned long flags;
1373         unsigned long gpa;
1374         unsigned long j;
1375         int ndetected = 0;
1376         struct rcu_node *rnp = rcu_get_root(rsp);
1377         long totqlen = 0;
1378 
1379         /* Kick and suppress, if so configured. */
1380         rcu_stall_kick_kthreads(rsp);
1381         if (rcu_cpu_stall_suppress)
1382                 return;
1383 
1384         /* Only let one CPU complain about others per time interval. */
1385 
1386         raw_spin_lock_irqsave_rcu_node(rnp, flags);
1387         delta = jiffies - READ_ONCE(rsp->jiffies_stall);
1388         if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
1389                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1390                 return;
1391         }
1392         WRITE_ONCE(rsp->jiffies_stall,
1393                    jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
1394         raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1395 
1396         /*
1397          * OK, time to rat on our buddy...
1398          * See Documentation/RCU/stallwarn.txt for info on how to debug
1399          * RCU CPU stall warnings.
1400          */
1401         pr_err("INFO: %s detected stalls on CPUs/tasks:",
1402                rsp->name);
1403         print_cpu_stall_info_begin();
1404         rcu_for_each_leaf_node(rsp, rnp) {
1405                 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1406                 ndetected += rcu_print_task_stall(rnp);
1407                 if (rnp->qsmask != 0) {
1408                         for_each_leaf_node_possible_cpu(rnp, cpu)
1409                                 if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) {
1410                                         print_cpu_stall_info(rsp, cpu);
1411                                         ndetected++;
1412                                 }
1413                 }
1414                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1415         }
1416 
1417         print_cpu_stall_info_end();
1418         for_each_possible_cpu(cpu)
1419                 totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(rsp->rda,
1420                                                             cpu)->cblist);
1421         pr_cont("(detected by %d, t=%ld jiffies, g=%ld, c=%ld, q=%lu)\n",
1422                smp_processor_id(), (long)(jiffies - rsp->gp_start),
1423                (long)rsp->gpnum, (long)rsp->completed, totqlen);
1424         if (ndetected) {
1425                 rcu_dump_cpu_stacks(rsp);
1426 
1427                 /* Complain about tasks blocking the grace period. */
1428                 rcu_print_detail_task_stall(rsp);
1429         } else {
1430                 if (READ_ONCE(rsp->gpnum) != gpnum ||
1431                     READ_ONCE(rsp->completed) == gpnum) {
1432                         pr_err("INFO: Stall ended before state dump start\n");
1433                 } else {
1434                         j = jiffies;
1435                         gpa = READ_ONCE(rsp->gp_activity);
1436                         pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n",
1437                                rsp->name, j - gpa, j, gpa,
1438                                jiffies_till_next_fqs,
1439                                rcu_get_root(rsp)->qsmask);
1440                         /* In this case, the current CPU might be at fault. */
1441                         sched_show_task(current);
1442                 }
1443         }
1444 
1445         rcu_check_gp_kthread_starvation(rsp);
1446 
1447         panic_on_rcu_stall();
1448 
1449         force_quiescent_state(rsp);  /* Kick them all. */
1450 }
1451 
1452 static void print_cpu_stall(struct rcu_state *rsp)
1453 {
1454         int cpu;
1455         unsigned long flags;
1456         struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
1457         struct rcu_node *rnp = rcu_get_root(rsp);
1458         long totqlen = 0;
1459 
1460         /* Kick and suppress, if so configured. */
1461         rcu_stall_kick_kthreads(rsp);
1462         if (rcu_cpu_stall_suppress)
1463                 return;
1464 
1465         /*
1466          * OK, time to rat on ourselves...
1467          * See Documentation/RCU/stallwarn.txt for info on how to debug
1468          * RCU CPU stall warnings.
1469          */
1470         pr_err("INFO: %s self-detected stall on CPU", rsp->name);
1471         print_cpu_stall_info_begin();
1472         raw_spin_lock_irqsave_rcu_node(rdp->mynode, flags);
1473         print_cpu_stall_info(rsp, smp_processor_id());
1474         raw_spin_unlock_irqrestore_rcu_node(rdp->mynode, flags);
1475         print_cpu_stall_info_end();
1476         for_each_possible_cpu(cpu)
1477                 totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(rsp->rda,
1478                                                             cpu)->cblist);
1479         pr_cont(" (t=%lu jiffies g=%ld c=%ld q=%lu)\n",
1480                 jiffies - rsp->gp_start,
1481                 (long)rsp->gpnum, (long)rsp->completed, totqlen);
1482 
1483         rcu_check_gp_kthread_starvation(rsp);
1484 
1485         rcu_dump_cpu_stacks(rsp);
1486 
1487         raw_spin_lock_irqsave_rcu_node(rnp, flags);
1488         if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall)))
1489                 WRITE_ONCE(rsp->jiffies_stall,
1490                            jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
1491         raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1492 
1493         panic_on_rcu_stall();
1494 
1495         /*
1496          * Attempt to revive the RCU machinery by forcing a context switch.
1497          *
1498          * A context switch would normally allow the RCU state machine to make
1499          * progress and it could be we're stuck in kernel space without context
1500          * switches for an entirely unreasonable amount of time.
1501          */
1502         resched_cpu(smp_processor_id());
1503 }
1504 
1505 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
1506 {
1507         unsigned long completed;
1508         unsigned long gpnum;
1509         unsigned long gps;
1510         unsigned long j;
1511         unsigned long js;
1512         struct rcu_node *rnp;
1513 
1514         if ((rcu_cpu_stall_suppress && !rcu_kick_kthreads) ||
1515             !rcu_gp_in_progress(rsp))
1516                 return;
1517         rcu_stall_kick_kthreads(rsp);
1518         j = jiffies;
1519 
1520         /*
1521          * Lots of memory barriers to reject false positives.
1522          *
1523          * The idea is to pick up rsp->gpnum, then rsp->jiffies_stall,
1524          * then rsp->gp_start, and finally rsp->completed.  These values
1525          * are updated in the opposite order with memory barriers (or
1526          * equivalent) during grace-period initialization and cleanup.
1527          * Now, a false positive can occur if we get an new value of
1528          * rsp->gp_start and a old value of rsp->jiffies_stall.  But given
1529          * the memory barriers, the only way that this can happen is if one
1530          * grace period ends and another starts between these two fetches.
1531          * Detect this by comparing rsp->completed with the previous fetch
1532          * from rsp->gpnum.
1533          *
1534          * Given this check, comparisons of jiffies, rsp->jiffies_stall,
1535          * and rsp->gp_start suffice to forestall false positives.
1536          */
1537         gpnum = READ_ONCE(rsp->gpnum);
1538         smp_rmb(); /* Pick up ->gpnum first... */
1539         js = READ_ONCE(rsp->jiffies_stall);
1540         smp_rmb(); /* ...then ->jiffies_stall before the rest... */
1541         gps = READ_ONCE(rsp->gp_start);
1542         smp_rmb(); /* ...and finally ->gp_start before ->completed. */
1543         completed = READ_ONCE(rsp->completed);
1544         if (ULONG_CMP_GE(completed, gpnum) ||
1545             ULONG_CMP_LT(j, js) ||
1546             ULONG_CMP_GE(gps, js))
1547                 return; /* No stall or GP completed since entering function. */
1548         rnp = rdp->mynode;
1549         if (rcu_gp_in_progress(rsp) &&
1550             (READ_ONCE(rnp->qsmask) & rdp->grpmask)) {
1551 
1552                 /* We haven't checked in, so go dump stack. */
1553                 print_cpu_stall(rsp);
1554 
1555         } else if (rcu_gp_in_progress(rsp) &&
1556                    ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) {
1557 
1558                 /* They had a few time units to dump stack, so complain. */
1559                 print_other_cpu_stall(rsp, gpnum);
1560         }
1561 }
1562 
1563 /**
1564  * rcu_cpu_stall_reset - prevent further stall warnings in current grace period
1565  *
1566  * Set the stall-warning timeout way off into the future, thus preventing
1567  * any RCU CPU stall-warning messages from appearing in the current set of
1568  * RCU grace periods.
1569  *
1570  * The caller must disable hard irqs.
1571  */
1572 void rcu_cpu_stall_reset(void)
1573 {
1574         struct rcu_state *rsp;
1575 
1576         for_each_rcu_flavor(rsp)
1577                 WRITE_ONCE(rsp->jiffies_stall, jiffies + ULONG_MAX / 2);
1578 }
1579 
1580 /*
1581  * Determine the value that ->completed will have at the end of the
1582  * next subsequent grace period.  This is used to tag callbacks so that
1583  * a CPU can invoke callbacks in a timely fashion even if that CPU has
1584  * been dyntick-idle for an extended period with callbacks under the
1585  * influence of RCU_FAST_NO_HZ.
1586  *
1587  * The caller must hold rnp->lock with interrupts disabled.
1588  */
1589 static unsigned long rcu_cbs_completed(struct rcu_state *rsp,
1590                                        struct rcu_node *rnp)
1591 {
1592         raw_lockdep_assert_held_rcu_node(rnp);
1593 
1594         /*
1595          * If RCU is idle, we just wait for the next grace period.
1596          * But we can only be sure that RCU is idle if we are looking
1597          * at the root rcu_node structure -- otherwise, a new grace
1598          * period might have started, but just not yet gotten around
1599          * to initializing the current non-root rcu_node structure.
1600          */
1601         if (rcu_get_root(rsp) == rnp && rnp->gpnum == rnp->completed)
1602                 return rnp->completed + 1;
1603 
1604         /*
1605          * If the current rcu_node structure believes that RCU is
1606          * idle, and if the rcu_state structure does not yet reflect
1607          * the start of a new grace period, then the next grace period
1608          * will suffice.  The memory barrier is needed to accurately
1609          * sample the rsp->gpnum, and pairs with the second lock
1610          * acquisition in rcu_gp_init(), which is augmented with
1611          * smp_mb__after_unlock_lock() for this purpose.
1612          */
1613         if (rnp->gpnum == rnp->completed) {
1614                 smp_mb(); /* See above block comment. */
1615                 if (READ_ONCE(rsp->gpnum) == rnp->completed)
1616                         return rnp->completed + 1;
1617         }
1618 
1619         /*
1620          * Otherwise, wait for a possible partial grace period and
1621          * then the subsequent full grace period.
1622          */
1623         return rnp->completed + 2;
1624 }
1625 
1626 /* Trace-event wrapper function for trace_rcu_future_grace_period.  */
1627 static void trace_rcu_this_gp(struct rcu_node *rnp, struct rcu_data *rdp,
1628                               unsigned long c, const char *s)
1629 {
1630         trace_rcu_future_grace_period(rdp->rsp->name, rnp->gpnum,
1631                                       rnp->completed, c, rnp->level,
1632                                       rnp->grplo, rnp->grphi, s);
1633 }
1634 
1635 /*
1636  * Start the specified grace period, as needed to handle newly arrived
1637  * callbacks.  The required future grace periods are recorded in each
1638  * rcu_node structure's ->need_future_gp[] field.  Returns true if there
1639  * is reason to awaken the grace-period kthread.
1640  *
1641  * The caller must hold the specified rcu_node structure's ->lock, which
1642  * is why the caller is responsible for waking the grace-period kthread.
1643  */
1644 static bool rcu_start_this_gp(struct rcu_node *rnp, struct rcu_data *rdp,
1645                               unsigned long c)
1646 {
1647         bool ret = false;
1648         struct rcu_state *rsp = rdp->rsp;
1649         struct rcu_node *rnp_root;
1650 
1651         /*
1652          * Use funnel locking to either acquire the root rcu_node
1653          * structure's lock or bail out if the need for this grace period
1654          * has already been recorded -- or has already started.  If there
1655          * is already a grace period in progress in a non-leaf node, no
1656          * recording is needed because the end of the grace period will
1657          * scan the leaf rcu_node structures.  Note that rnp->lock must
1658          * not be released.
1659          */
1660         raw_lockdep_assert_held_rcu_node(rnp);
1661         trace_rcu_this_gp(rnp, rdp, c, TPS("Startleaf"));
1662         for (rnp_root = rnp; 1; rnp_root = rnp_root->parent) {
1663                 if (rnp_root != rnp)
1664                         raw_spin_lock_rcu_node(rnp_root);
1665                 WARN_ON_ONCE(ULONG_CMP_LT(rnp_root->gpnum +
1666                                           need_future_gp_mask(), c));
1667                 if (need_future_gp_element(rnp_root, c) ||
1668                     ULONG_CMP_GE(rnp_root->gpnum, c) ||
1669                     (rnp != rnp_root &&
1670                      rnp_root->gpnum != rnp_root->completed)) {
1671                         trace_rcu_this_gp(rnp_root, rdp, c, TPS("Prestarted"));
1672                         goto unlock_out;
1673                 }
1674                 need_future_gp_element(rnp_root, c) = true;
1675                 if (rnp_root != rnp && rnp_root->parent != NULL)
1676                         raw_spin_unlock_rcu_node(rnp_root);
1677                 if (!rnp_root->parent)
1678                         break;  /* At root, and perhaps also leaf. */
1679         }
1680 
1681         /* If GP already in progress, just leave, otherwise start one. */
1682         if (rnp_root->gpnum != rnp_root->completed) {
1683                 trace_rcu_this_gp(rnp_root, rdp, c, TPS("Startedleafroot"));
1684                 goto unlock_out;
1685         }
1686         trace_rcu_this_gp(rnp_root, rdp, c, TPS("Startedroot"));
1687         WRITE_ONCE(rsp->gp_flags, rsp->gp_flags | RCU_GP_FLAG_INIT);
1688         if (!rsp->gp_kthread) {
1689                 trace_rcu_this_gp(rnp_root, rdp, c, TPS("NoGPkthread"));
1690                 goto unlock_out;
1691         }
1692         trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gpnum), TPS("newreq"));
1693         ret = true;  /* Caller must wake GP kthread. */
1694 unlock_out:
1695         if (rnp != rnp_root)
1696                 raw_spin_unlock_rcu_node(rnp_root);
1697         return ret;
1698 }
1699 
1700 /*
1701  * Clean up any old requests for the just-ended grace period.  Also return
1702  * whether any additional grace periods have been requested.
1703  */
1704 static bool rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp)
1705 {
1706         unsigned long c = rnp->completed;
1707         bool needmore;
1708         struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
1709 
1710         need_future_gp_element(rnp, c) = false;
1711         needmore = need_any_future_gp(rnp);
1712         trace_rcu_this_gp(rnp, rdp, c,
1713                           needmore ? TPS("CleanupMore") : TPS("Cleanup"));
1714         return needmore;
1715 }
1716 
1717 /*
1718  * Awaken the grace-period kthread for the specified flavor of RCU.
1719  * Don't do a self-awaken, and don't bother awakening when there is
1720  * nothing for the grace-period kthread to do (as in several CPUs
1721  * raced to awaken, and we lost), and finally don't try to awaken
1722  * a kthread that has not yet been created.
1723  */
1724 static void rcu_gp_kthread_wake(struct rcu_state *rsp)
1725 {
1726         if (current == rsp->gp_kthread ||
1727             !READ_ONCE(rsp->gp_flags) ||
1728             !rsp->gp_kthread)
1729                 return;
1730         swake_up(&rsp->gp_wq);
1731 }
1732 
1733 /*
1734  * If there is room, assign a ->completed number to any callbacks on
1735  * this CPU that have not already been assigned.  Also accelerate any
1736  * callbacks that were previously assigned a ->completed number that has
1737  * since proven to be too conservative, which can happen if callbacks get
1738  * assigned a ->completed number while RCU is idle, but with reference to
1739  * a non-root rcu_node structure.  This function is idempotent, so it does
1740  * not hurt to call it repeatedly.  Returns an flag saying that we should
1741  * awaken the RCU grace-period kthread.
1742  *
1743  * The caller must hold rnp->lock with interrupts disabled.
1744  */
1745 static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1746                                struct rcu_data *rdp)
1747 {
1748         unsigned long c;
1749         bool ret = false;
1750 
1751         raw_lockdep_assert_held_rcu_node(rnp);
1752 
1753         /* If no pending (not yet ready to invoke) callbacks, nothing to do. */
1754         if (!rcu_segcblist_pend_cbs(&rdp->cblist))
1755                 return false;
1756 
1757         /*
1758          * Callbacks are often registered with incomplete grace-period
1759          * information.  Something about the fact that getting exact
1760          * information requires acquiring a global lock...  RCU therefore
1761          * makes a conservative estimate of the grace period number at which
1762          * a given callback will become ready to invoke.        The following
1763          * code checks this estimate and improves it when possible, thus
1764          * accelerating callback invocation to an earlier grace-period
1765          * number.
1766          */
1767         c = rcu_cbs_completed(rsp, rnp);
1768         if (rcu_segcblist_accelerate(&rdp->cblist, c))
1769                 ret = rcu_start_this_gp(rnp, rdp, c);
1770 
1771         /* Trace depending on how much we were able to accelerate. */
1772         if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL))
1773                 trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccWaitCB"));
1774         else
1775                 trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccReadyCB"));
1776         return ret;
1777 }
1778 
1779 /*
1780  * Move any callbacks whose grace period has completed to the
1781  * RCU_DONE_TAIL sublist, then compact the remaining sublists and
1782  * assign ->completed numbers to any callbacks in the RCU_NEXT_TAIL
1783  * sublist.  This function is idempotent, so it does not hurt to
1784  * invoke it repeatedly.  As long as it is not invoked -too- often...
1785  * Returns true if the RCU grace-period kthread needs to be awakened.
1786  *
1787  * The caller must hold rnp->lock with interrupts disabled.
1788  */
1789 static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1790                             struct rcu_data *rdp)
1791 {
1792         raw_lockdep_assert_held_rcu_node(rnp);
1793 
1794         /* If no pending (not yet ready to invoke) callbacks, nothing to do. */
1795         if (!rcu_segcblist_pend_cbs(&rdp->cblist))
1796                 return false;
1797 
1798         /*
1799          * Find all callbacks whose ->completed numbers indicate that they
1800          * are ready to invoke, and put them into the RCU_DONE_TAIL sublist.
1801          */
1802         rcu_segcblist_advance(&rdp->cblist, rnp->completed);
1803 
1804         /* Classify any remaining callbacks. */
1805         return rcu_accelerate_cbs(rsp, rnp, rdp);
1806 }
1807 
1808 /*
1809  * Update CPU-local rcu_data state to record the beginnings and ends of
1810  * grace periods.  The caller must hold the ->lock of the leaf rcu_node
1811  * structure corresponding to the current CPU, and must have irqs disabled.
1812  * Returns true if the grace-period kthread needs to be awakened.
1813  */
1814 static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp,
1815                               struct rcu_data *rdp)
1816 {
1817         bool ret;
1818         bool need_gp;
1819 
1820         raw_lockdep_assert_held_rcu_node(rnp);
1821 
1822         /* Handle the ends of any preceding grace periods first. */
1823         if (rdp->completed == rnp->completed &&
1824             !unlikely(READ_ONCE(rdp->gpwrap))) {
1825 
1826                 /* No grace period end, so just accelerate recent callbacks. */
1827                 ret = rcu_accelerate_cbs(rsp, rnp, rdp);
1828 
1829         } else {
1830 
1831                 /* Advance callbacks. */
1832                 ret = rcu_advance_cbs(rsp, rnp, rdp);
1833 
1834                 /* Remember that we saw this grace-period completion. */
1835                 rdp->completed = rnp->completed;
1836                 trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuend"));
1837         }
1838 
1839         if (rdp->gpnum != rnp->gpnum || unlikely(READ_ONCE(rdp->gpwrap))) {
1840                 /*
1841                  * If the current grace period is waiting for this CPU,
1842                  * set up to detect a quiescent state, otherwise don't
1843                  * go looking for one.
1844                  */
1845                 rdp->gpnum = rnp->gpnum;
1846                 trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart"));
1847                 need_gp = !!(rnp->qsmask & rdp->grpmask);
1848                 rdp->cpu_no_qs.b.norm = need_gp;
1849                 rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr);
1850                 rdp->core_needs_qs = need_gp;
1851                 zero_cpu_stall_ticks(rdp);
1852                 WRITE_ONCE(rdp->gpwrap, false);
1853                 rcu_gpnum_ovf(rnp, rdp);
1854         }
1855         return ret;
1856 }
1857 
1858 static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp)
1859 {
1860         unsigned long flags;
1861         bool needwake;
1862         struct rcu_node *rnp;
1863 
1864         local_irq_save(flags);
1865         rnp = rdp->mynode;
1866         if ((rdp->gpnum == READ_ONCE(rnp->gpnum) &&
1867              rdp->completed == READ_ONCE(rnp->completed) &&
1868              !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */
1869             !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */
1870                 local_irq_restore(flags);
1871                 return;
1872         }
1873         needwake = __note_gp_changes(rsp, rnp, rdp);
1874         raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1875         if (needwake)
1876                 rcu_gp_kthread_wake(rsp);
1877 }
1878 
1879 static void rcu_gp_slow(struct rcu_state *rsp, int delay)
1880 {
1881         if (delay > 0 &&
1882             !(rsp->gpnum % (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay)))
1883                 schedule_timeout_uninterruptible(delay);
1884 }
1885 
1886 /*
1887  * Initialize a new grace period.  Return false if no grace period required.
1888  */
1889 static bool rcu_gp_init(struct rcu_state *rsp)
1890 {
1891         unsigned long oldmask;
1892         struct rcu_data *rdp;
1893         struct rcu_node *rnp = rcu_get_root(rsp);
1894 
1895         WRITE_ONCE(rsp->gp_activity, jiffies);
1896         raw_spin_lock_irq_rcu_node(rnp);
1897         if (!READ_ONCE(rsp->gp_flags)) {
1898                 /* Spurious wakeup, tell caller to go back to sleep.  */
1899                 raw_spin_unlock_irq_rcu_node(rnp);
1900                 return false;
1901         }
1902         WRITE_ONCE(rsp->gp_flags, 0); /* Clear all flags: New grace period. */
1903 
1904         if (WARN_ON_ONCE(rcu_gp_in_progress(rsp))) {
1905                 /*
1906                  * Grace period already in progress, don't start another.
1907                  * Not supposed to be able to happen.
1908                  */
1909                 raw_spin_unlock_irq_rcu_node(rnp);
1910                 return false;
1911         }
1912 
1913         /* Advance to a new grace period and initialize state. */
1914         record_gp_stall_check_time(rsp);
1915         /* Record GP times before starting GP, hence smp_store_release(). */
1916         smp_store_release(&rsp->gpnum, rsp->gpnum + 1);
1917         trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start"));
1918         raw_spin_unlock_irq_rcu_node(rnp);
1919 
1920         /*
1921          * Apply per-leaf buffered online and offline operations to the
1922          * rcu_node tree.  Note that this new grace period need not wait
1923          * for subsequent online CPUs, and that quiescent-state forcing
1924          * will handle subsequent offline CPUs.
1925          */
1926         rcu_for_each_leaf_node(rsp, rnp) {
1927                 rcu_gp_slow(rsp, gp_preinit_delay);
1928                 raw_spin_lock_irq_rcu_node(rnp);
1929                 if (rnp->qsmaskinit == rnp->qsmaskinitnext &&
1930                     !rnp->wait_blkd_tasks) {
1931                         /* Nothing to do on this leaf rcu_node structure. */
1932                         raw_spin_unlock_irq_rcu_node(rnp);
1933                         continue;
1934                 }
1935 
1936                 /* Record old state, apply changes to ->qsmaskinit field. */
1937                 oldmask = rnp->qsmaskinit;
1938                 rnp->qsmaskinit = rnp->qsmaskinitnext;
1939 
1940                 /* If zero-ness of ->qsmaskinit changed, propagate up tree. */
1941                 if (!oldmask != !rnp->qsmaskinit) {
1942                         if (!oldmask) /* First online CPU for this rcu_node. */
1943                                 rcu_init_new_rnp(rnp);
1944                         else if (rcu_preempt_has_tasks(rnp)) /* blocked tasks */
1945                                 rnp->wait_blkd_tasks = true;
1946                         else /* Last offline CPU and can propagate. */
1947                                 rcu_cleanup_dead_rnp(rnp);
1948                 }
1949 
1950                 /*
1951                  * If all waited-on tasks from prior grace period are
1952                  * done, and if all this rcu_node structure's CPUs are
1953                  * still offline, propagate up the rcu_node tree and
1954                  * clear ->wait_blkd_tasks.  Otherwise, if one of this
1955                  * rcu_node structure's CPUs has since come back online,
1956                  * simply clear ->wait_blkd_tasks (but rcu_cleanup_dead_rnp()
1957                  * checks for this, so just call it unconditionally).
1958                  */
1959                 if (rnp->wait_blkd_tasks &&
1960                     (!rcu_preempt_has_tasks(rnp) ||
1961                      rnp->qsmaskinit)) {
1962                         rnp->wait_blkd_tasks = false;
1963                         rcu_cleanup_dead_rnp(rnp);
1964                 }
1965 
1966                 raw_spin_unlock_irq_rcu_node(rnp);
1967         }
1968 
1969         /*
1970          * Set the quiescent-state-needed bits in all the rcu_node
1971          * structures for all currently online CPUs in breadth-first order,
1972          * starting from the root rcu_node structure, relying on the layout
1973          * of the tree within the rsp->node[] array.  Note that other CPUs
1974          * will access only the leaves of the hierarchy, thus seeing that no
1975          * grace period is in progress, at least until the corresponding
1976          * leaf node has been initialized.
1977          *
1978          * The grace period cannot complete until the initialization
1979          * process finishes, because this kthread handles both.
1980          */
1981         rcu_for_each_node_breadth_first(rsp, rnp) {
1982                 rcu_gp_slow(rsp, gp_init_delay);
1983                 raw_spin_lock_irq_rcu_node(rnp);
1984                 rdp = this_cpu_ptr(rsp->rda);
1985                 rcu_preempt_check_blocked_tasks(rnp);
1986                 rnp->qsmask = rnp->qsmaskinit;
1987                 WRITE_ONCE(rnp->gpnum, rsp->gpnum);
1988                 if (WARN_ON_ONCE(rnp->completed != rsp->completed))
1989                         WRITE_ONCE(rnp->completed, rsp->completed);
1990                 if (rnp == rdp->mynode)
1991                         (void)__note_gp_changes(rsp, rnp, rdp);
1992                 rcu_preempt_boost_start_gp(rnp);
1993                 trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
1994                                             rnp->level, rnp->grplo,
1995                                             rnp->grphi, rnp->qsmask);
1996                 raw_spin_unlock_irq_rcu_node(rnp);
1997                 cond_resched_tasks_rcu_qs();
1998                 WRITE_ONCE(rsp->gp_activity, jiffies);
1999         }
2000 
2001         return true;
2002 }
2003 
2004 /*
2005  * Helper function for swait_event_idle() wakeup at force-quiescent-state
2006  * time.
2007  */
2008 static bool rcu_gp_fqs_check_wake(struct rcu_state *rsp, int *gfp)
2009 {
2010         struct rcu_node *rnp = rcu_get_root(rsp);
2011 
2012         /* Someone like call_rcu() requested a force-quiescent-state scan. */
2013         *gfp = READ_ONCE(rsp->gp_flags);
2014         if (*gfp & RCU_GP_FLAG_FQS)
2015                 return true;
2016 
2017         /* The current grace period has completed. */
2018         if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp))
2019                 return true;
2020 
2021         return false;
2022 }
2023 
2024 /*
2025  * Do one round of quiescent-state forcing.
2026  */
2027 static void rcu_gp_fqs(struct rcu_state *rsp, bool first_time)
2028 {
2029         struct rcu_node *rnp = rcu_get_root(rsp);
2030 
2031         WRITE_ONCE(rsp->gp_activity, jiffies);
2032         rsp->n_force_qs++;
2033         if (first_time) {
2034                 /* Collect dyntick-idle snapshots. */
2035                 force_qs_rnp(rsp, dyntick_save_progress_counter);
2036         } else {
2037                 /* Handle dyntick-idle and offline CPUs. */
2038                 force_qs_rnp(rsp, rcu_implicit_dynticks_qs);
2039         }
2040         /* Clear flag to prevent immediate re-entry. */
2041         if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
2042                 raw_spin_lock_irq_rcu_node(rnp);
2043                 WRITE_ONCE(rsp->gp_flags,
2044                            READ_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS);
2045                 raw_spin_unlock_irq_rcu_node(rnp);
2046         }
2047 }
2048 
2049 /*
2050  * Clean up after the old grace period.
2051  */
2052 static void rcu_gp_cleanup(struct rcu_state *rsp)
2053 {
2054         unsigned long gp_duration;
2055         bool needgp = false;
2056         struct rcu_data *rdp;
2057         struct rcu_node *rnp = rcu_get_root(rsp);
2058         struct swait_queue_head *sq;
2059 
2060         WRITE_ONCE(rsp->gp_activity, jiffies);
2061         raw_spin_lock_irq_rcu_node(rnp);
2062         gp_duration = jiffies - rsp->gp_start;
2063         if (gp_duration > rsp->gp_max)
2064                 rsp->gp_max = gp_duration;
2065 
2066         /*
2067          * We know the grace period is complete, but to everyone else
2068          * it appears to still be ongoing.  But it is also the case
2069          * that to everyone else it looks like there is nothing that
2070          * they can do to advance the grace period.  It is therefore
2071          * safe for us to drop the lock in order to mark the grace
2072          * period as completed in all of the rcu_node structures.
2073          */
2074         raw_spin_unlock_irq_rcu_node(rnp);
2075 
2076         /*
2077          * Propagate new ->completed value to rcu_node structures so
2078          * that other CPUs don't have to wait until the start of the next
2079          * grace period to process their callbacks.  This also avoids
2080          * some nasty RCU grace-period initialization races by forcing
2081          * the end of the current grace period to be completely recorded in
2082          * all of the rcu_node structures before the beginning of the next
2083          * grace period is recorded in any of the rcu_node structures.
2084          */
2085         rcu_for_each_node_breadth_first(rsp, rnp) {
2086                 raw_spin_lock_irq_rcu_node(rnp);
2087                 WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp));
2088                 WARN_ON_ONCE(rnp->qsmask);
2089                 WRITE_ONCE(rnp->completed, rsp->gpnum);
2090                 rdp = this_cpu_ptr(rsp->rda);
2091                 if (rnp == rdp->mynode)
2092                         needgp = __note_gp_changes(rsp, rnp, rdp) || needgp;
2093                 /* smp_mb() provided by prior unlock-lock pair. */
2094                 needgp = rcu_future_gp_cleanup(rsp, rnp) || needgp;
2095                 sq = rcu_nocb_gp_get(rnp);
2096                 raw_spin_unlock_irq_rcu_node(rnp);
2097                 rcu_nocb_gp_cleanup(sq);
2098                 cond_resched_tasks_rcu_qs();
2099                 WRITE_ONCE(rsp->gp_activity, jiffies);
2100                 rcu_gp_slow(rsp, gp_cleanup_delay);
2101         }
2102         rnp = rcu_get_root(rsp);
2103         raw_spin_lock_irq_rcu_node(rnp); /* Order GP before ->completed update. */
2104 
2105         /* Declare grace period done. */
2106         WRITE_ONCE(rsp->completed, rsp->gpnum);
2107         trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end"));
2108         rsp->gp_state = RCU_GP_IDLE;
2109         /* Check for GP requests since above loop. */
2110         rdp = this_cpu_ptr(rsp->rda);
2111         if (need_any_future_gp(rnp)) {
2112                 trace_rcu_this_gp(rnp, rdp, rsp->completed - 1,
2113                                   TPS("CleanupMore"));
2114                 needgp = true;
2115         }
2116         /* Advance CBs to reduce false positives below. */
2117         if (!rcu_accelerate_cbs(rsp, rnp, rdp) && needgp) {
2118                 WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT);
2119                 trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gpnum),
2120                                        TPS("newreq"));
2121         }
2122         WRITE_ONCE(rsp->gp_flags, rsp->gp_flags & RCU_GP_FLAG_INIT);
2123         raw_spin_unlock_irq_rcu_node(rnp);
2124 }
2125 
2126 /*
2127  * Body of kthread that handles grace periods.
2128  */
2129 static int __noreturn rcu_gp_kthread(void *arg)
2130 {
2131         bool first_gp_fqs;
2132         int gf;
2133         unsigned long j;
2134         int ret;
2135         struct rcu_state *rsp = arg;
2136         struct rcu_node *rnp = rcu_get_root(rsp);
2137 
2138         rcu_bind_gp_kthread();
2139         for (;;) {
2140 
2141                 /* Handle grace-period start. */
2142                 for (;;) {
2143                         trace_rcu_grace_period(rsp->name,
2144                                                READ_ONCE(rsp->gpnum),
2145                                                TPS("reqwait"));
2146                         rsp->gp_state = RCU_GP_WAIT_GPS;
2147                         swait_event_idle(rsp->gp_wq, READ_ONCE(rsp->gp_flags) &
2148                                                      RCU_GP_FLAG_INIT);
2149                         rsp->gp_state = RCU_GP_DONE_GPS;
2150                         /* Locking provides needed memory barrier. */
2151                         if (rcu_gp_init(rsp))
2152                                 break;
2153                         cond_resched_tasks_rcu_qs();
2154                         WRITE_ONCE(rsp->gp_activity, jiffies);
2155                         WARN_ON(signal_pending(current));
2156                         trace_rcu_grace_period(rsp->name,
2157                                                READ_ONCE(rsp->gpnum),
2158                                                TPS("reqwaitsig"));
2159                 }
2160 
2161                 /* Handle quiescent-state forcing. */
2162                 first_gp_fqs = true;
2163                 j = jiffies_till_first_fqs;
2164                 if (j > HZ) {
2165                         j = HZ;
2166                         jiffies_till_first_fqs = HZ;
2167                 }
2168                 ret = 0;
2169                 for (;;) {
2170                         if (!ret) {
2171                                 rsp->jiffies_force_qs = jiffies + j;
2172                                 WRITE_ONCE(rsp->jiffies_kick_kthreads,
2173                                            jiffies + 3 * j);
2174                         }
2175                         trace_rcu_grace_period(rsp->name,
2176                                                READ_ONCE(rsp->gpnum),
2177                                                TPS("fqswait"));
2178                         rsp->gp_state = RCU_GP_WAIT_FQS;
2179                         ret = swait_event_idle_timeout(rsp->gp_wq,
2180                                         rcu_gp_fqs_check_wake(rsp, &gf), j);
2181                         rsp->gp_state = RCU_GP_DOING_FQS;
2182                         /* Locking provides needed memory barriers. */
2183                         /* If grace period done, leave loop. */
2184                         if (!READ_ONCE(rnp->qsmask) &&
2185                             !rcu_preempt_blocked_readers_cgp(rnp))
2186                                 break;
2187                         /* If time for quiescent-state forcing, do it. */
2188                         if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) ||
2189                             (gf & RCU_GP_FLAG_FQS)) {
2190                                 trace_rcu_grace_period(rsp->name,
2191                                                        READ_ONCE(rsp->gpnum),
2192                                                        TPS("fqsstart"));
2193                                 rcu_gp_fqs(rsp, first_gp_fqs);
2194                                 first_gp_fqs = false;
2195                                 trace_rcu_grace_period(rsp->name,
2196                                                        READ_ONCE(rsp->gpnum),
2197                                                        TPS("fqsend"));
2198                                 cond_resched_tasks_rcu_qs();
2199                                 WRITE_ONCE(rsp->gp_activity, jiffies);
2200                                 ret = 0; /* Force full wait till next FQS. */
2201                                 j = jiffies_till_next_fqs;
2202                                 if (j > HZ) {
2203                                         j = HZ;
2204                                         jiffies_till_next_fqs = HZ;
2205                                 } else if (j < 1) {
2206                                         j = 1;
2207                                         jiffies_till_next_fqs = 1;
2208                                 }
2209                         } else {
2210                                 /* Deal with stray signal. */
2211                                 cond_resched_tasks_rcu_qs();
2212                                 WRITE_ONCE(rsp->gp_activity, jiffies);
2213                                 WARN_ON(signal_pending(current));
2214                                 trace_rcu_grace_period(rsp->name,
2215                                                        READ_ONCE(rsp->gpnum),
2216                                                        TPS("fqswaitsig"));
2217                                 ret = 1; /* Keep old FQS timing. */
2218                                 j = jiffies;
2219                                 if (time_after(jiffies, rsp->jiffies_force_qs))
2220                                         j = 1;
2221                                 else
2222                                         j = rsp->jiffies_force_qs - j;
2223                         }
2224                 }
2225 
2226                 /* Handle grace-period end. */
2227                 rsp->gp_state = RCU_GP_CLEANUP;
2228                 rcu_gp_cleanup(rsp);
2229                 rsp->gp_state = RCU_GP_CLEANED;
2230         }
2231 }
2232 
2233 /*
2234  * Report a full set of quiescent states to the specified rcu_state data
2235  * structure.  Invoke rcu_gp_kthread_wake() to awaken the grace-period
2236  * kthread if another grace period is required.  Whether we wake
2237  * the grace-period kthread or it awakens itself for the next round
2238  * of quiescent-state forcing, that kthread will clean up after the
2239  * just-completed grace period.  Note that the caller must hold rnp->lock,
2240  * which is released before return.
2241  */
2242 static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
2243         __releases(rcu_get_root(rsp)->lock)
2244 {
2245         raw_lockdep_assert_held_rcu_node(rcu_get_root(rsp));
2246         WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
2247         WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS);
2248         raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags);
2249         rcu_gp_kthread_wake(rsp);
2250 }
2251 
2252 /*
2253  * Similar to rcu_report_qs_rdp(), for which it is a helper function.
2254  * Allows quiescent states for a group of CPUs to be reported at one go
2255  * to the specified rcu_node structure, though all the CPUs in the group
2256  * must be represented by the same rcu_node structure (which need not be a
2257  * leaf rcu_node structure, though it often will be).  The gps parameter
2258  * is the grace-period snapshot, which means that the quiescent states
2259  * are valid only if rnp->gpnum is equal to gps.  That structure's lock
2260  * must be held upon entry, and it is released before return.
2261  */
2262 static void
2263 rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
2264                   struct rcu_node *rnp, unsigned long gps, unsigned long flags)
2265         __releases(rnp->lock)
2266 {
2267         unsigned long oldmask = 0;
2268         struct rcu_node *rnp_c;
2269 
2270         raw_lockdep_assert_held_rcu_node(rnp);
2271 
2272         /* Walk up the rcu_node hierarchy. */
2273         for (;;) {
2274                 if (!(rnp->qsmask & mask) || rnp->gpnum != gps) {
2275 
2276                         /*
2277                          * Our bit has already been cleared, or the
2278                          * relevant grace period is already over, so done.
2279                          */
2280                         raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2281                         return;
2282                 }
2283                 WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */
2284                 WARN_ON_ONCE(!rcu_is_leaf_node(rnp) &&
2285                              rcu_preempt_blocked_readers_cgp(rnp));
2286                 rnp->qsmask &= ~mask;
2287                 trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum,
2288                                                  mask, rnp->qsmask, rnp->level,
2289                                                  rnp->grplo, rnp->grphi,
2290                                                  !!rnp->gp_tasks);
2291                 if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
2292 
2293                         /* Other bits still set at this level, so done. */
2294                         raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2295                         return;
2296                 }
2297                 mask = rnp->grpmask;
2298                 if (rnp->parent == NULL) {
2299 
2300                         /* No more levels.  Exit loop holding root lock. */
2301 
2302                         break;
2303                 }
2304                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2305                 rnp_c = rnp;
2306                 rnp = rnp->parent;
2307                 raw_spin_lock_irqsave_rcu_node(rnp, flags);
2308                 oldmask = rnp_c->qsmask;
2309         }
2310 
2311         /*
2312          * Get here if we are the last CPU to pass through a quiescent
2313          * state for this grace period.  Invoke rcu_report_qs_rsp()
2314          * to clean up and start the next grace period if one is needed.
2315          */
2316         rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
2317 }
2318 
2319 /*
2320  * Record a quiescent state for all tasks that were previously queued
2321  * on the specified rcu_node structure and that were blocking the current
2322  * RCU grace period.  The caller must hold the specified rnp->lock with
2323  * irqs disabled, and this lock is released upon return, but irqs remain
2324  * disabled.
2325  */
2326 static void rcu_report_unblock_qs_rnp(struct rcu_state *rsp,
2327                                       struct rcu_node *rnp, unsigned long flags)
2328         __releases(rnp->lock)
2329 {
2330         unsigned long gps;
2331         unsigned long mask;
2332         struct rcu_node *rnp_p;
2333 
2334         raw_lockdep_assert_held_rcu_node(rnp);
2335         if (rcu_state_p == &rcu_sched_state || rsp != rcu_state_p ||
2336             rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
2337                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2338                 return;  /* Still need more quiescent states! */
2339         }
2340 
2341         rnp_p = rnp->parent;
2342         if (rnp_p == NULL) {
2343                 /*
2344                  * Only one rcu_node structure in the tree, so don't
2345                  * try to report up to its nonexistent parent!
2346                  */
2347                 rcu_report_qs_rsp(rsp, flags);
2348                 return;
2349         }
2350 
2351         /* Report up the rest of the hierarchy, tracking current ->gpnum. */
2352         gps = rnp->gpnum;
2353         mask = rnp->grpmask;
2354         raw_spin_unlock_rcu_node(rnp);  /* irqs remain disabled. */
2355         raw_spin_lock_rcu_node(rnp_p);  /* irqs already disabled. */
2356         rcu_report_qs_rnp(mask, rsp, rnp_p, gps, flags);
2357 }
2358 
2359 /*
2360  * Record a quiescent state for the specified CPU to that CPU's rcu_data
2361  * structure.  This must be called from the specified CPU.
2362  */
2363 static void
2364 rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
2365 {
2366         unsigned long flags;
2367         unsigned long mask;
2368         bool needwake;
2369         struct rcu_node *rnp;
2370 
2371         rnp = rdp->mynode;
2372         raw_spin_lock_irqsave_rcu_node(rnp, flags);
2373         if (rdp->cpu_no_qs.b.norm || rdp->gpnum != rnp->gpnum ||
2374             rnp->completed == rnp->gpnum || rdp->gpwrap) {
2375 
2376                 /*
2377                  * The grace period in which this quiescent state was
2378                  * recorded has ended, so don't report it upwards.
2379                  * We will instead need a new quiescent state that lies
2380                  * within the current grace period.
2381                  */
2382                 rdp->cpu_no_qs.b.norm = true;   /* need qs for new gp. */
2383                 rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr);
2384                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2385                 return;
2386         }
2387         mask = rdp->grpmask;
2388         if ((rnp->qsmask & mask) == 0) {
2389                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2390         } else {
2391                 rdp->core_needs_qs = false;
2392 
2393                 /*
2394                  * This GP can't end until cpu checks in, so all of our
2395                  * callbacks can be processed during the next GP.
2396                  */
2397                 needwake = rcu_accelerate_cbs(rsp, rnp, rdp);
2398 
2399                 rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags);
2400                 /* ^^^ Released rnp->lock */
2401                 if (needwake)
2402                         rcu_gp_kthread_wake(rsp);
2403         }
2404 }
2405 
2406 /*
2407  * Check to see if there is a new grace period of which this CPU
2408  * is not yet aware, and if so, set up local rcu_data state for it.
2409  * Otherwise, see if this CPU has just passed through its first
2410  * quiescent state for this grace period, and record that fact if so.
2411  */
2412 static void
2413 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
2414 {
2415         /* Check for grace-period ends and beginnings. */
2416         note_gp_changes(rsp, rdp);
2417 
2418         /*
2419          * Does this CPU still need to do its part for current grace period?
2420          * If no, return and let the other CPUs do their part as well.
2421          */
2422         if (!rdp->core_needs_qs)
2423                 return;
2424 
2425         /*
2426          * Was there a quiescent state since the beginning of the grace
2427          * period? If no, then exit and wait for the next call.
2428          */
2429         if (rdp->cpu_no_qs.b.norm)
2430                 return;
2431 
2432         /*
2433          * Tell RCU we are done (but rcu_report_qs_rdp() will be the
2434          * judge of that).
2435          */
2436         rcu_report_qs_rdp(rdp->cpu, rsp, rdp);
2437 }
2438 
2439 /*
2440  * Trace the fact that this CPU is going offline.
2441  */
2442 static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
2443 {
2444         RCU_TRACE(unsigned long mask;)
2445         RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda);)
2446         RCU_TRACE(struct rcu_node *rnp = rdp->mynode;)
2447 
2448         if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
2449                 return;
2450 
2451         RCU_TRACE(mask = rdp->grpmask;)
2452         trace_rcu_grace_period(rsp->name,
2453                                rnp->gpnum + 1 - !!(rnp->qsmask & mask),
2454                                TPS("cpuofl"));
2455 }
2456 
2457 /*
2458  * All CPUs for the specified rcu_node structure have gone offline,
2459  * and all tasks that were preempted within an RCU read-side critical
2460  * section while running on one of those CPUs have since exited their RCU
2461  * read-side critical section.  Some other CPU is reporting this fact with
2462  * the specified rcu_node structure's ->lock held and interrupts disabled.
2463  * This function therefore goes up the tree of rcu_node structures,
2464  * clearing the corresponding bits in the ->qsmaskinit fields.  Note that
2465  * the leaf rcu_node structure's ->qsmaskinit field has already been
2466  * updated
2467  *
2468  * This function does check that the specified rcu_node structure has
2469  * all CPUs offline and no blocked tasks, so it is OK to invoke it
2470  * prematurely.  That said, invoking it after the fact will cost you
2471  * a needless lock acquisition.  So once it has done its work, don't
2472  * invoke it again.
2473  */
2474 static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf)
2475 {
2476         long mask;
2477         struct rcu_node *rnp = rnp_leaf;
2478 
2479         raw_lockdep_assert_held_rcu_node(rnp);
2480         if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) ||
2481             rnp->qsmaskinit || rcu_preempt_has_tasks(rnp))
2482                 return;
2483         for (;;) {
2484                 mask = rnp->grpmask;
2485                 rnp = rnp->parent;
2486                 if (!rnp)
2487                         break;
2488                 raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
2489                 rnp->qsmaskinit &= ~mask;
2490                 rnp->qsmask &= ~mask;
2491                 if (rnp->qsmaskinit) {
2492                         raw_spin_unlock_rcu_node(rnp);
2493                         /* irqs remain disabled. */
2494                         return;
2495                 }
2496                 raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */
2497         }
2498 }
2499 
2500 /*
2501  * The CPU has been completely removed, and some other CPU is reporting
2502  * this fact from process context.  Do the remainder of the cleanup.
2503  * There can only be one CPU hotplug operation at a time, so no need for
2504  * explicit locking.
2505  */
2506 static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2507 {
2508         struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2509         struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
2510 
2511         if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
2512                 return;
2513 
2514         /* Adjust any no-longer-needed kthreads. */
2515         rcu_boost_kthread_setaffinity(rnp, -1);
2516 }
2517 
2518 /*
2519  * Invoke any RCU callbacks that have made it to the end of their grace
2520  * period.  Thottle as specified by rdp->blimit.
2521  */
2522 static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
2523 {
2524         unsigned long flags;
2525         struct rcu_head *rhp;
2526         struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl);
2527         long bl, count;
2528 
2529         /* If no callbacks are ready, just return. */
2530         if (!rcu_segcblist_ready_cbs(&rdp->cblist)) {
2531                 trace_rcu_batch_start(rsp->name,
2532                                       rcu_segcblist_n_lazy_cbs(&rdp->cblist),
2533                                       rcu_segcblist_n_cbs(&rdp->cblist), 0);
2534                 trace_rcu_batch_end(rsp->name, 0,
2535                                     !rcu_segcblist_empty(&rdp->cblist),
2536                                     need_resched(), is_idle_task(current),
2537                                     rcu_is_callbacks_kthread());
2538                 return;
2539         }
2540 
2541         /*
2542          * Extract the list of ready callbacks, disabling to prevent
2543          * races with call_rcu() from interrupt handlers.  Leave the
2544          * callback counts, as rcu_barrier() needs to be conservative.
2545          */
2546         local_irq_save(flags);
2547         WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
2548         bl = rdp->blimit;
2549         trace_rcu_batch_start(rsp->name, rcu_segcblist_n_lazy_cbs(&rdp->cblist),
2550                               rcu_segcblist_n_cbs(&rdp->cblist), bl);
2551         rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl);
2552         local_irq_restore(flags);
2553 
2554         /* Invoke callbacks. */
2555         rhp = rcu_cblist_dequeue(&rcl);
2556         for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) {
2557                 debug_rcu_head_unqueue(rhp);
2558                 if (__rcu_reclaim(rsp->name, rhp))
2559                         rcu_cblist_dequeued_lazy(&rcl);
2560                 /*
2561                  * Stop only if limit reached and CPU has something to do.
2562                  * Note: The rcl structure counts down from zero.
2563                  */
2564                 if (-rcl.len >= bl &&
2565                     (need_resched() ||
2566                      (!is_idle_task(current) && !rcu_is_callbacks_kthread())))
2567                         break;
2568         }
2569 
2570         local_irq_save(flags);
2571         count = -rcl.len;
2572         trace_rcu_batch_end(rsp->name, count, !!rcl.head, need_resched(),
2573                             is_idle_task(current), rcu_is_callbacks_kthread());
2574 
2575         /* Update counts and requeue any remaining callbacks. */
2576         rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl);
2577         smp_mb(); /* List handling before counting for rcu_barrier(). */
2578         rcu_segcblist_insert_count(&rdp->cblist, &rcl);
2579 
2580         /* Reinstate batch limit if we have worked down the excess. */
2581         count = rcu_segcblist_n_cbs(&rdp->cblist);
2582         if (rdp->blimit == LONG_MAX && count <= qlowmark)
2583                 rdp->blimit = blimit;
2584 
2585         /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
2586         if (count == 0 && rdp->qlen_last_fqs_check != 0) {
2587                 rdp->qlen_last_fqs_check = 0;
2588                 rdp->n_force_qs_snap = rsp->n_force_qs;
2589         } else if (count < rdp->qlen_last_fqs_check - qhimark)
2590                 rdp->qlen_last_fqs_check = count;
2591 
2592         /*
2593          * The following usually indicates a double call_rcu().  To track
2594          * this down, try building with CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.
2595          */
2596         WARN_ON_ONCE(rcu_segcblist_empty(&rdp->cblist) != (count == 0));
2597 
2598         local_irq_restore(flags);
2599 
2600         /* Re-invoke RCU core processing if there are callbacks remaining. */
2601         if (rcu_segcblist_ready_cbs(&rdp->cblist))
2602                 invoke_rcu_core();
2603 }
2604 
2605 /*
2606  * Check to see if this CPU is in a non-context-switch quiescent state
2607  * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
2608  * Also schedule RCU core processing.
2609  *
2610  * This function must be called from hardirq context.  It is normally
2611  * invoked from the scheduling-clock interrupt.
2612  */
2613 void rcu_check_callbacks(int user)
2614 {
2615         trace_rcu_utilization(TPS("Start scheduler-tick"));
2616         increment_cpu_stall_ticks();
2617         if (user || rcu_is_cpu_rrupt_from_idle()) {
2618 
2619                 /*
2620                  * Get here if this CPU took its interrupt from user
2621                  * mode or from the idle loop, and if this is not a
2622                  * nested interrupt.  In this case, the CPU is in
2623                  * a quiescent state, so note it.
2624                  *
2625                  * No memory barrier is required here because both
2626                  * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
2627                  * variables that other CPUs neither access nor modify,
2628                  * at least not while the corresponding CPU is online.
2629                  */
2630 
2631                 rcu_sched_qs();
2632                 rcu_bh_qs();
2633 
2634         } else if (!in_softirq()) {
2635 
2636                 /*
2637                  * Get here if this CPU did not take its interrupt from
2638                  * softirq, in other words, if it is not interrupting
2639                  * a rcu_bh read-side critical section.  This is an _bh
2640                  * critical section, so note it.
2641                  */
2642 
2643                 rcu_bh_qs();
2644         }
2645         rcu_preempt_check_callbacks();
2646         if (rcu_pending())
2647                 invoke_rcu_core();
2648         if (user)
2649                 rcu_note_voluntary_context_switch(current);
2650         trace_rcu_utilization(TPS("End scheduler-tick"));
2651 }
2652 
2653 /*
2654  * Scan the leaf rcu_node structures, processing dyntick state for any that
2655  * have not yet encountered a quiescent state, using the function specified.
2656  * Also initiate boosting for any threads blocked on the root rcu_node.
2657  *
2658  * The caller must have suppressed start of new grace periods.
2659  */
2660 static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *rsp))
2661 {
2662         int cpu;
2663         unsigned long flags;
2664         unsigned long mask;
2665         struct rcu_node *rnp;
2666 
2667         rcu_for_each_leaf_node(rsp, rnp) {
2668                 cond_resched_tasks_rcu_qs();
2669                 mask = 0;
2670                 raw_spin_lock_irqsave_rcu_node(rnp, flags);
2671                 if (rnp->qsmask == 0) {
2672                         if (rcu_state_p == &rcu_sched_state ||
2673                             rsp != rcu_state_p ||
2674                             rcu_preempt_blocked_readers_cgp(rnp)) {
2675                                 /*
2676                                  * No point in scanning bits because they
2677                                  * are all zero.  But we might need to
2678                                  * priority-boost blocked readers.
2679                                  */
2680                                 rcu_initiate_boost(rnp, flags);
2681                                 /* rcu_initiate_boost() releases rnp->lock */
2682                                 continue;
2683                         }
2684                         if (rnp->parent &&
2685                             (rnp->parent->qsmask & rnp->grpmask)) {
2686                                 /*
2687                                  * Race between grace-period
2688                                  * initialization and task exiting RCU
2689                                  * read-side critical section: Report.
2690                                  */
2691                                 rcu_report_unblock_qs_rnp(rsp, rnp, flags);
2692                                 /* rcu_report_unblock_qs_rnp() rlses ->lock */
2693                                 continue;
2694                         }
2695                 }
2696                 for_each_leaf_node_possible_cpu(rnp, cpu) {
2697                         unsigned long bit = leaf_node_cpu_bit(rnp, cpu);
2698                         if ((rnp->qsmask & bit) != 0) {
2699                                 if (f(per_cpu_ptr(rsp->rda, cpu)))
2700                                         mask |= bit;
2701                         }
2702                 }
2703                 if (mask != 0) {
2704                         /* Idle/offline CPUs, report (releases rnp->lock. */
2705                         rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags);
2706                 } else {
2707                         /* Nothing to do here, so just drop the lock. */
2708                         raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2709                 }
2710         }
2711 }
2712 
2713 /*
2714  * Force quiescent states on reluctant CPUs, and also detect which
2715  * CPUs are in dyntick-idle mode.
2716  */
2717 static void force_quiescent_state(struct rcu_state *rsp)
2718 {
2719         unsigned long flags;
2720         bool ret;
2721         struct rcu_node *rnp;
2722         struct rcu_node *rnp_old = NULL;
2723 
2724         /* Funnel through hierarchy to reduce memory contention. */
2725         rnp = __this_cpu_read(rsp->rda->mynode);
2726         for (; rnp != NULL; rnp = rnp->parent) {
2727                 ret = (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) ||
2728                       !raw_spin_trylock(&rnp->fqslock);
2729                 if (rnp_old != NULL)
2730                         raw_spin_unlock(&rnp_old->fqslock);
2731                 if (ret)
2732                         return;
2733                 rnp_old = rnp;
2734         }
2735         /* rnp_old == rcu_get_root(rsp), rnp == NULL. */
2736 
2737         /* Reached the root of the rcu_node tree, acquire lock. */
2738         raw_spin_lock_irqsave_rcu_node(rnp_old, flags);
2739         raw_spin_unlock(&rnp_old->fqslock);
2740         if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
2741                 raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags);
2742                 return;  /* Someone beat us to it. */
2743         }
2744         WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS);
2745         raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags);
2746         rcu_gp_kthread_wake(rsp);
2747 }
2748 
2749 /*
2750  * This does the RCU core processing work for the specified rcu_state
2751  * and rcu_data structures.  This may be called only from the CPU to
2752  * whom the rdp belongs.
2753  */
2754 static void
2755 __rcu_process_callbacks(struct rcu_state *rsp)
2756 {
2757         unsigned long flags;
2758         bool needwake;
2759         struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
2760         struct rcu_node *rnp;
2761 
2762         WARN_ON_ONCE(!rdp->beenonline);
2763 
2764         /* Update RCU state based on any recent quiescent states. */
2765         rcu_check_quiescent_state(rsp, rdp);
2766 
2767         /* No grace period and unregistered callbacks? */
2768         if (!rcu_gp_in_progress(rsp) &&
2769             rcu_segcblist_is_enabled(&rdp->cblist)) {
2770                 local_irq_save(flags);
2771                 if (rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) {
2772                         local_irq_restore(flags);
2773                 } else {
2774                         rnp = rdp->mynode;
2775                         raw_spin_lock_rcu_node(rnp); /* irqs disabled. */
2776                         needwake = rcu_accelerate_cbs(rsp, rnp, rdp);
2777                         raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
2778                         if (needwake)
2779                                 rcu_gp_kthread_wake(rsp);
2780                 }
2781         }
2782 
2783         /* If there are callbacks ready, invoke them. */
2784         if (rcu_segcblist_ready_cbs(&rdp->cblist))
2785                 invoke_rcu_callbacks(rsp, rdp);
2786 
2787         /* Do any needed deferred wakeups of rcuo kthreads. */
2788         do_nocb_deferred_wakeup(rdp);
2789 }
2790 
2791 /*
2792  * Do RCU core processing for the current CPU.
2793  */
2794 static __latent_entropy void rcu_process_callbacks(struct softirq_action *unused)
2795 {
2796         struct rcu_state *rsp;
2797 
2798         if (cpu_is_offline(smp_processor_id()))
2799                 return;
2800         trace_rcu_utilization(TPS("Start RCU core"));
2801         for_each_rcu_flavor(rsp)
2802                 __rcu_process_callbacks(rsp);
2803         trace_rcu_utilization(TPS("End RCU core"));
2804 }
2805 
2806 /*
2807  * Schedule RCU callback invocation.  If the specified type of RCU
2808  * does not support RCU priority boosting, just do a direct call,
2809  * otherwise wake up the per-CPU kernel kthread.  Note that because we
2810  * are running on the current CPU with softirqs disabled, the
2811  * rcu_cpu_kthread_task cannot disappear out from under us.
2812  */
2813 static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
2814 {
2815         if (unlikely(!READ_ONCE(rcu_scheduler_fully_active)))
2816                 return;
2817         if (likely(!rsp->boost)) {
2818                 rcu_do_batch(rsp, rdp);
2819                 return;
2820         }
2821         invoke_rcu_callbacks_kthread();
2822 }
2823 
2824 static void invoke_rcu_core(void)
2825 {
2826         if (cpu_online(smp_processor_id()))
2827                 raise_softirq(RCU_SOFTIRQ);
2828 }
2829 
2830 /*
2831  * Handle any core-RCU processing required by a call_rcu() invocation.
2832  */
2833 static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp,
2834                             struct rcu_head *head, unsigned long flags)
2835 {
2836         bool needwake;
2837 
2838         /*
2839          * If called from an extended quiescent state, invoke the RCU
2840          * core in order to force a re-evaluation of RCU's idleness.
2841          */
2842         if (!rcu_is_watching())
2843                 invoke_rcu_core();
2844 
2845         /* If interrupts were disabled or CPU offline, don't invoke RCU core. */
2846         if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id()))
2847                 return;
2848 
2849         /*
2850          * Force the grace period if too many callbacks or too long waiting.
2851          * Enforce hysteresis, and don't invoke force_quiescent_state()
2852          * if some other CPU has recently done so.  Also, don't bother
2853          * invoking force_quiescent_state() if the newly enqueued callback
2854          * is the only one waiting for a grace period to complete.
2855          */
2856         if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) >
2857                      rdp->qlen_last_fqs_check + qhimark)) {
2858 
2859                 /* Are we ignoring a completed grace period? */
2860                 note_gp_changes(rsp, rdp);
2861 
2862                 /* Start a new grace period if one not already started. */
2863                 if (!rcu_gp_in_progress(rsp)) {
2864                         struct rcu_node *rnp = rdp->mynode;
2865 
2866                         raw_spin_lock_rcu_node(rnp);
2867                         needwake = rcu_accelerate_cbs(rsp, rnp, rdp);
2868                         raw_spin_unlock_rcu_node(rnp);
2869                         if (needwake)
2870                                 rcu_gp_kthread_wake(rsp);
2871                 } else {
2872                         /* Give the grace period a kick. */
2873                         rdp->blimit = LONG_MAX;
2874                         if (rsp->n_force_qs == rdp->n_force_qs_snap &&
2875                             rcu_segcblist_first_pend_cb(&rdp->cblist) != head)
2876                                 force_quiescent_state(rsp);
2877                         rdp->n_force_qs_snap = rsp->n_force_qs;
2878                         rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist);
2879                 }
2880         }
2881 }
2882 
2883 /*
2884  * RCU callback function to leak a callback.
2885  */
2886 static void rcu_leak_callback(struct rcu_head *rhp)
2887 {
2888 }
2889 
2890 /*
2891  * Helper function for call_rcu() and friends.  The cpu argument will
2892  * normally be -1, indicating "currently running CPU".  It may specify
2893  * a CPU only if that CPU is a no-CBs CPU.  Currently, only _rcu_barrier()
2894  * is expected to specify a CPU.
2895  */
2896 static void
2897 __call_rcu(struct rcu_head *head, rcu_callback_t func,
2898            struct rcu_state *rsp, int cpu, bool lazy)
2899 {
2900         unsigned long flags;
2901         struct rcu_data *rdp;
2902 
2903         /* Misaligned rcu_head! */
2904         WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1));
2905 
2906         if (debug_rcu_head_queue(head)) {
2907                 /*
2908                  * Probable double call_rcu(), so leak the callback.
2909                  * Use rcu:rcu_callback trace event to find the previous
2910                  * time callback was passed to __call_rcu().
2911                  */
2912                 WARN_ONCE(1, "__call_rcu(): Double-freed CB %p->%pF()!!!\n",
2913                           head, head->func);
2914                 WRITE_ONCE(head->func, rcu_leak_callback);
2915                 return;
2916         }
2917         head->func = func;
2918         head->next = NULL;
2919         local_irq_save(flags);
2920         rdp = this_cpu_ptr(rsp->rda);
2921 
2922         /* Add the callback to our list. */
2923         if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist)) || cpu != -1) {
2924                 int offline;
2925 
2926                 if (cpu != -1)
2927                         rdp = per_cpu_ptr(rsp->rda, cpu);
2928                 if (likely(rdp->mynode)) {
2929                         /* Post-boot, so this should be for a no-CBs CPU. */
2930                         offline = !__call_rcu_nocb(rdp, head, lazy, flags);
2931                         WARN_ON_ONCE(offline);
2932                         /* Offline CPU, _call_rcu() illegal, leak callback.  */
2933                         local_irq_restore(flags);
2934                         return;
2935                 }
2936                 /*
2937                  * Very early boot, before rcu_init().  Initialize if needed
2938                  * and then drop through to queue the callback.
2939                  */
2940                 BUG_ON(cpu != -1);
2941                 WARN_ON_ONCE(!rcu_is_watching());
2942                 if (rcu_segcblist_empty(&rdp->cblist))
2943                         rcu_segcblist_init(&rdp->cblist);
2944         }
2945         rcu_segcblist_enqueue(&rdp->cblist, head, lazy);
2946         if (!lazy)
2947                 rcu_idle_count_callbacks_posted();
2948 
2949         if (__is_kfree_rcu_offset((unsigned long)func))
2950                 trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func,
2951                                          rcu_segcblist_n_lazy_cbs(&rdp->cblist),
2952                                          rcu_segcblist_n_cbs(&rdp->cblist));
2953         else
2954                 trace_rcu_callback(rsp->name, head,
2955                                    rcu_segcblist_n_lazy_cbs(&rdp->cblist),
2956                                    rcu_segcblist_n_cbs(&rdp->cblist));
2957 
2958         /* Go handle any RCU core processing required. */
2959         __call_rcu_core(rsp, rdp, head, flags);
2960         local_irq_restore(flags);
2961 }
2962 
2963 /**
2964  * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
2965  * @head: structure to be used for queueing the RCU updates.
2966  * @func: actual callback function to be invoked after the grace period
2967  *
2968  * The callback function will be invoked some time after a full grace
2969  * period elapses, in other words after all currently executing RCU
2970  * read-side critical sections have completed. call_rcu_sched() assumes
2971  * that the read-side critical sections end on enabling of preemption
2972  * or on voluntary preemption.
2973  * RCU read-side critical sections are delimited by:
2974  *
2975  * - rcu_read_lock_sched() and rcu_read_unlock_sched(), OR
2976  * - anything that disables preemption.
2977  *
2978  *  These may be nested.
2979  *
2980  * See the description of call_rcu() for more detailed information on
2981  * memory ordering guarantees.
2982  */
2983 void call_rcu_sched(struct rcu_head *head, rcu_callback_t func)
2984 {
2985         __call_rcu(head, func, &rcu_sched_state, -1, 0);
2986 }
2987 EXPORT_SYMBOL_GPL(call_rcu_sched);
2988 
2989 /**
2990  * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
2991  * @head: structure to be used for queueing the RCU updates.
2992  * @func: actual callback function to be invoked after the grace period
2993  *
2994  * The callback function will be invoked some time after a full grace
2995  * period elapses, in other words after all currently executing RCU
2996  * read-side critical sections have completed. call_rcu_bh() assumes
2997  * that the read-side critical sections end on completion of a softirq
2998  * handler. This means that read-side critical sections in process
2999  * context must not be interrupted by softirqs. This interface is to be
3000  * used when most of the read-side critical sections are in softirq context.
3001  * RCU read-side critical sections are delimited by:
3002  *
3003  * - rcu_read_lock() and  rcu_read_unlock(), if in interrupt context, OR
3004  * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
3005  *
3006  * These may be nested.
3007  *
3008  * See the description of call_rcu() for more detailed information on
3009  * memory ordering guarantees.
3010  */
3011 void call_rcu_bh(struct rcu_head *head, rcu_callback_t func)
3012 {
3013         __call_rcu(head, func, &rcu_bh_state, -1, 0);
3014 }
3015 EXPORT_SYMBOL_GPL(call_rcu_bh);
3016 
3017 /*
3018  * Queue an RCU callback for lazy invocation after a grace period.
3019  * This will likely be later named something like "call_rcu_lazy()",
3020  * but this change will require some way of tagging the lazy RCU
3021  * callbacks in the list of pending callbacks. Until then, this
3022  * function may only be called from __kfree_rcu().
3023  */
3024 void kfree_call_rcu(struct rcu_head *head,
3025                     rcu_callback_t func)
3026 {
3027         __call_rcu(head, func, rcu_state_p, -1, 1);
3028 }
3029 EXPORT_SYMBOL_GPL(kfree_call_rcu);
3030 
3031 /*
3032  * Because a context switch is a grace period for RCU-sched and RCU-bh,
3033  * any blocking grace-period wait automatically implies a grace period
3034  * if there is only one CPU online at any point time during execution
3035  * of either synchronize_sched() or synchronize_rcu_bh().  It is OK to
3036  * occasionally incorrectly indicate that there are multiple CPUs online
3037  * when there was in fact only one the whole time, as this just adds
3038  * some overhead: RCU still operates correctly.
3039  */
3040 static inline int rcu_blocking_is_gp(void)
3041 {
3042         int ret;
3043 
3044         might_sleep();  /* Check for RCU read-side critical section. */
3045         preempt_disable();
3046         ret = num_online_cpus() <= 1;
3047         preempt_enable();
3048         return ret;
3049 }
3050 
3051 /**
3052  * synchronize_sched - wait until an rcu-sched grace period has elapsed.
3053  *
3054  * Control will return to the caller some time after a full rcu-sched
3055  * grace period has elapsed, in other words after all currently executing
3056  * rcu-sched read-side critical sections have completed.   These read-side
3057  * critical sections are delimited by rcu_read_lock_sched() and
3058  * rcu_read_unlock_sched(), and may be nested.  Note that preempt_disable(),
3059  * local_irq_disable(), and so on may be used in place of
3060  * rcu_read_lock_sched().
3061  *
3062  * This means that all preempt_disable code sequences, including NMI and
3063  * non-threaded hardware-interrupt handlers, in progress on entry will
3064  * have completed before this primitive returns.  However, this does not
3065  * guarantee that softirq handlers will have completed, since in some
3066  * kernels, these handlers can run in process context, and can block.
3067  *
3068  * Note that this guarantee implies further memory-ordering guarantees.
3069  * On systems with more than one CPU, when synchronize_sched() returns,
3070  * each CPU is guaranteed to have executed a full memory barrier since the
3071  * end of its last RCU-sched read-side critical section whose beginning
3072  * preceded the call to synchronize_sched().  In addition, each CPU having
3073  * an RCU read-side critical section that extends beyond the return from
3074  * synchronize_sched() is guaranteed to have executed a full memory barrier
3075  * after the beginning of synchronize_sched() and before the beginning of
3076  * that RCU read-side critical section.  Note that these guarantees include
3077  * CPUs that are offline, idle, or executing in user mode, as well as CPUs
3078  * that are executing in the kernel.
3079  *
3080  * Furthermore, if CPU A invoked synchronize_sched(), which returned
3081  * to its caller on CPU B, then both CPU A and CPU B are guaranteed
3082  * to have executed a full memory barrier during the execution of
3083  * synchronize_sched() -- even if CPU A and CPU B are the same CPU (but
3084  * again only if the system has more than one CPU).
3085  */
3086 void synchronize_sched(void)
3087 {
3088         RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) ||
3089                          lock_is_held(&rcu_lock_map) ||
3090                          lock_is_held(&rcu_sched_lock_map),
3091                          "Illegal synchronize_sched() in RCU-sched read-side critical section");
3092         if (rcu_blocking_is_gp())
3093                 return;
3094         if (rcu_gp_is_expedited())
3095                 synchronize_sched_expedited();
3096         else
3097                 wait_rcu_gp(call_rcu_sched);
3098 }
3099 EXPORT_SYMBOL_GPL(synchronize_sched);
3100 
3101 /**
3102  * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
3103  *
3104  * Control will return to the caller some time after a full rcu_bh grace
3105  * period has elapsed, in other words after all currently executing rcu_bh
3106  * read-side critical sections have completed.  RCU read-side critical
3107  * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
3108  * and may be nested.
3109  *
3110  * See the description of synchronize_sched() for more detailed information
3111  * on memory ordering guarantees.
3112  */
3113 void synchronize_rcu_bh(void)
3114 {
3115         RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) ||
3116                          lock_is_held(&rcu_lock_map) ||
3117                          lock_is_held(&rcu_sched_lock_map),
3118                          "Illegal synchronize_rcu_bh() in RCU-bh read-side critical section");
3119         if (rcu_blocking_is_gp())
3120                 return;
3121         if (rcu_gp_is_expedited())
3122                 synchronize_rcu_bh_expedited();
3123         else
3124                 wait_rcu_gp(call_rcu_bh);
3125 }
3126 EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
3127 
3128 /**
3129  * get_state_synchronize_rcu - Snapshot current RCU state
3130  *
3131  * Returns a cookie that is used by a later call to cond_synchronize_rcu()
3132  * to determine whether or not a full grace period has elapsed in the
3133  * meantime.
3134  */
3135 unsigned long get_state_synchronize_rcu(void)
3136 {
3137         /*
3138          * Any prior manipulation of RCU-protected data must happen
3139          * before the load from ->gpnum.
3140          */
3141         smp_mb();  /* ^^^ */
3142 
3143         /*
3144          * Make sure this load happens before the purportedly
3145          * time-consuming work between get_state_synchronize_rcu()
3146          * and cond_synchronize_rcu().
3147          */
3148         return smp_load_acquire(&rcu_state_p->gpnum);
3149 }
3150 EXPORT_SYMBOL_GPL(get_state_synchronize_rcu);
3151 
3152 /**
3153  * cond_synchronize_rcu - Conditionally wait for an RCU grace period
3154  *
3155  * @oldstate: return value from earlier call to get_state_synchronize_rcu()
3156  *
3157  * If a full RCU grace period has elapsed since the earlier call to
3158  * get_state_synchronize_rcu(), just return.  Otherwise, invoke
3159  * synchronize_rcu() to wait for a full grace period.
3160  *
3161  * Yes, this function does not take counter wrap into account.  But
3162  * counter wrap is harmless.  If the counter wraps, we have waited for
3163  * more than 2 billion grace periods (and way more on a 64-bit system!),
3164  * so waiting for one additional grace period should be just fine.
3165  */
3166 void cond_synchronize_rcu(unsigned long oldstate)
3167 {
3168         unsigned long newstate;
3169 
3170         /*
3171          * Ensure that this load happens before any RCU-destructive
3172          * actions the caller might carry out after we return.
3173          */
3174         newstate = smp_load_acquire(&rcu_state_p->completed);
3175         if (ULONG_CMP_GE(oldstate, newstate))
3176                 synchronize_rcu();
3177 }
3178 EXPORT_SYMBOL_GPL(cond_synchronize_rcu);
3179 
3180 /**
3181  * get_state_synchronize_sched - Snapshot current RCU-sched state
3182  *
3183  * Returns a cookie that is used by a later call to cond_synchronize_sched()
3184  * to determine whether or not a full grace period has elapsed in the
3185  * meantime.
3186  */
3187 unsigned long get_state_synchronize_sched(void)
3188 {
3189         /*
3190          * Any prior manipulation of RCU-protected data must happen
3191          * before the load from ->gpnum.
3192          */
3193         smp_mb();  /* ^^^ */
3194 
3195         /*
3196          * Make sure this load happens before the purportedly
3197          * time-consuming work between get_state_synchronize_sched()
3198          * and cond_synchronize_sched().
3199          */
3200         return smp_load_acquire(&rcu_sched_state.gpnum);
3201 }
3202 EXPORT_SYMBOL_GPL(get_state_synchronize_sched);
3203 
3204 /**
3205  * cond_synchronize_sched - Conditionally wait for an RCU-sched grace period
3206  *
3207  * @oldstate: return value from earlier call to get_state_synchronize_sched()
3208  *
3209  * If a full RCU-sched grace period has elapsed since the earlier call to
3210  * get_state_synchronize_sched(), just return.  Otherwise, invoke
3211  * synchronize_sched() to wait for a full grace period.
3212  *
3213  * Yes, this function does not take counter wrap into account.  But
3214  * counter wrap is harmless.  If the counter wraps, we have waited for
3215  * more than 2 billion grace periods (and way more on a 64-bit system!),
3216  * so waiting for one additional grace period should be just fine.
3217  */
3218 void cond_synchronize_sched(unsigned long oldstate)
3219 {
3220         unsigned long newstate;
3221 
3222         /*
3223          * Ensure that this load happens before any RCU-destructive
3224          * actions the caller might carry out after we return.
3225          */
3226         newstate = smp_load_acquire(&rcu_sched_state.completed);
3227         if (ULONG_CMP_GE(oldstate, newstate))
3228                 synchronize_sched();
3229 }
3230 EXPORT_SYMBOL_GPL(cond_synchronize_sched);
3231 
3232 /*
3233  * Check to see if there is any immediate RCU-related work to be done
3234  * by the current CPU, for the specified type of RCU, returning 1 if so.
3235  * The checks are in order of increasing expense: checks that can be
3236  * carried out against CPU-local state are performed first.  However,
3237  * we must check for CPU stalls first, else we might not get a chance.
3238  */
3239 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
3240 {
3241         struct rcu_node *rnp = rdp->mynode;
3242 
3243         /* Check for CPU stalls, if enabled. */
3244         check_cpu_stall(rsp, rdp);
3245 
3246         /* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */
3247         if (rcu_nohz_full_cpu(rsp))
3248                 return 0;
3249 
3250         /* Is the RCU core waiting for a quiescent state from this CPU? */
3251         if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm)
3252                 return 1;
3253 
3254         /* Does this CPU have callbacks ready to invoke? */
3255         if (rcu_segcblist_ready_cbs(&rdp->cblist))
3256                 return 1;
3257 
3258         /* Has RCU gone idle with this CPU needing another grace period? */
3259         if (!rcu_gp_in_progress(rsp) &&
3260             rcu_segcblist_is_enabled(&rdp->cblist) &&
3261             !rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL))
3262                 return 1;
3263 
3264         /* Has another RCU grace period completed?  */
3265         if (READ_ONCE(rnp->completed) != rdp->completed) /* outside lock */
3266                 return 1;
3267 
3268         /* Has a new RCU grace period started? */
3269         if (READ_ONCE(rnp->gpnum) != rdp->gpnum ||
3270             unlikely(READ_ONCE(rdp->gpwrap))) /* outside lock */
3271                 return 1;
3272 
3273         /* Does this CPU need a deferred NOCB wakeup? */
3274         if (rcu_nocb_need_deferred_wakeup(rdp))
3275                 return 1;
3276 
3277         /* nothing to do */
3278         return 0;
3279 }
3280 
3281 /*
3282  * Check to see if there is any immediate RCU-related work to be done
3283  * by the current CPU, returning 1 if so.  This function is part of the
3284  * RCU implementation; it is -not- an exported member of the RCU API.
3285  */
3286 static int rcu_pending(void)
3287 {
3288         struct rcu_state *rsp;
3289 
3290         for_each_rcu_flavor(rsp)
3291                 if (__rcu_pending(rsp, this_cpu_ptr(rsp->rda)))
3292                         return 1;
3293         return 0;
3294 }
3295 
3296 /*
3297  * Return true if the specified CPU has any callback.  If all_lazy is
3298  * non-NULL, store an indication of whether all callbacks are lazy.
3299  * (If there are no callbacks, all of them are deemed to be lazy.)
3300  */
3301 static bool __maybe_unused rcu_cpu_has_callbacks(bool *all_lazy)
3302 {
3303         bool al = true;
3304         bool hc = false;
3305         struct rcu_data *rdp;
3306         struct rcu_state *rsp;
3307 
3308         for_each_rcu_flavor(rsp) {
3309                 rdp = this_cpu_ptr(rsp->rda);
3310                 if (rcu_segcblist_empty(&rdp->cblist))
3311                         continue;
3312                 hc = true;
3313                 if (rcu_segcblist_n_nonlazy_cbs(&rdp->cblist) || !all_lazy) {
3314                         al = false;
3315                         break;
3316                 }
3317         }
3318         if (all_lazy)
3319                 *all_lazy = al;
3320         return hc;
3321 }
3322 
3323 /*
3324  * Helper function for _rcu_barrier() tracing.  If tracing is disabled,
3325  * the compiler is expected to optimize this away.
3326  */
3327 static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s,
3328                                int cpu, unsigned long done)
3329 {
3330         trace_rcu_barrier(rsp->name, s, cpu,
3331                           atomic_read(&rsp->barrier_cpu_count), done);
3332 }
3333 
3334 /*
3335  * RCU callback function for _rcu_barrier().  If we are last, wake
3336  * up the task executing _rcu_barrier().
3337  */
3338 static void rcu_barrier_callback(struct rcu_head *rhp)
3339 {
3340         struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head);
3341         struct rcu_state *rsp = rdp->rsp;
3342 
3343         if (atomic_dec_and_test(&rsp->barrier_cpu_count)) {
3344                 _rcu_barrier_trace(rsp, TPS("LastCB"), -1,
3345                                    rsp->barrier_sequence);
3346                 complete(&rsp->barrier_completion);
3347         } else {
3348                 _rcu_barrier_trace(rsp, TPS("CB"), -1, rsp->barrier_sequence);
3349         }
3350 }
3351 
3352 /*
3353  * Called with preemption disabled, and from cross-cpu IRQ context.
3354  */
3355 static void rcu_barrier_func(void *type)
3356 {
3357         struct rcu_state *rsp = type;
3358         struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
3359 
3360         _rcu_barrier_trace(rsp, TPS("IRQ"), -1, rsp->barrier_sequence);
3361         rdp->barrier_head.func = rcu_barrier_callback;
3362         debug_rcu_head_queue(&rdp->barrier_head);
3363         if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head, 0)) {
3364                 atomic_inc(&rsp->barrier_cpu_count);
3365         } else {
3366                 debug_rcu_head_unqueue(&rdp->barrier_head);
3367                 _rcu_barrier_trace(rsp, TPS("IRQNQ"), -1,
3368                                    rsp->barrier_sequence);
3369         }
3370 }
3371 
3372 /*
3373  * Orchestrate the specified type of RCU barrier, waiting for all
3374  * RCU callbacks of the specified type to complete.
3375  */
3376 static void _rcu_barrier(struct rcu_state *rsp)
3377 {
3378         int cpu;
3379         struct rcu_data *rdp;
3380         unsigned long s = rcu_seq_snap(&rsp->barrier_sequence);
3381 
3382         _rcu_barrier_trace(rsp, TPS("Begin"), -1, s);
3383 
3384         /* Take mutex to serialize concurrent rcu_barrier() requests. */
3385         mutex_lock(&rsp->barrier_mutex);
3386 
3387         /* Did someone else do our work for us? */
3388         if (rcu_seq_done(&rsp->barrier_sequence, s)) {
3389                 _rcu_barrier_trace(rsp, TPS("EarlyExit"), -1,
3390                                    rsp->barrier_sequence);
3391                 smp_mb(); /* caller's subsequent code after above check. */
3392                 mutex_unlock(&rsp->barrier_mutex);
3393                 return;
3394         }
3395 
3396         /* Mark the start of the barrier operation. */
3397         rcu_seq_start(&rsp->barrier_sequence);
3398         _rcu_barrier_trace(rsp, TPS("Inc1"), -1, rsp->barrier_sequence);
3399 
3400         /*
3401          * Initialize the count to one rather than to zero in order to
3402          * avoid a too-soon return to zero in case of a short grace period
3403          * (or preemption of this task).  Exclude CPU-hotplug operations
3404          * to ensure that no offline CPU has callbacks queued.
3405          */
3406         init_completion(&rsp->barrier_completion);
3407         atomic_set(&rsp->barrier_cpu_count, 1);
3408         get_online_cpus();
3409 
3410         /*
3411          * Force each CPU with callbacks to register a new callback.
3412          * When that callback is invoked, we will know that all of the
3413          * corresponding CPU's preceding callbacks have been invoked.
3414          */
3415         for_each_possible_cpu(cpu) {
3416                 if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu))
3417                         continue;
3418                 rdp = per_cpu_ptr(rsp->rda, cpu);
3419                 if (rcu_is_nocb_cpu(cpu)) {
3420                         if (!rcu_nocb_cpu_needs_barrier(rsp, cpu)) {
3421                                 _rcu_barrier_trace(rsp, TPS("OfflineNoCB"), cpu,
3422                                                    rsp->barrier_sequence);
3423                         } else {
3424                                 _rcu_barrier_trace(rsp, TPS("OnlineNoCB"), cpu,
3425                                                    rsp->barrier_sequence);
3426                                 smp_mb__before_atomic();
3427                                 atomic_inc(&rsp->barrier_cpu_count);
3428                                 __call_rcu(&rdp->barrier_head,
3429                                            rcu_barrier_callback, rsp, cpu, 0);
3430                         }
3431                 } else if (rcu_segcblist_n_cbs(&rdp->cblist)) {
3432                         _rcu_barrier_trace(rsp, TPS("OnlineQ"), cpu,
3433                                            rsp->barrier_sequence);
3434                         smp_call_function_single(cpu, rcu_barrier_func, rsp, 1);
3435                 } else {
3436                         _rcu_barrier_trace(rsp, TPS("OnlineNQ"), cpu,
3437                                            rsp->barrier_sequence);
3438                 }
3439         }
3440         put_online_cpus();
3441 
3442         /*
3443          * Now that we have an rcu_barrier_callback() callback on each
3444          * CPU, and thus each counted, remove the initial count.
3445          */
3446         if (atomic_dec_and_test(&rsp->barrier_cpu_count))
3447                 complete(&rsp->barrier_completion);
3448 
3449         /* Wait for all rcu_barrier_callback() callbacks to be invoked. */
3450         wait_for_completion(&rsp->barrier_completion);
3451 
3452         /* Mark the end of the barrier operation. */
3453         _rcu_barrier_trace(rsp, TPS("Inc2"), -1, rsp->barrier_sequence);
3454         rcu_seq_end(&rsp->barrier_sequence);
3455 
3456         /* Other rcu_barrier() invocations can now safely proceed. */
3457         mutex_unlock(&rsp->barrier_mutex);
3458 }
3459 
3460 /**
3461  * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
3462  */
3463 void rcu_barrier_bh(void)
3464 {
3465         _rcu_barrier(&rcu_bh_state);
3466 }
3467 EXPORT_SYMBOL_GPL(rcu_barrier_bh);
3468 
3469 /**
3470  * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
3471  */
3472 void rcu_barrier_sched(void)
3473 {
3474         _rcu_barrier(&rcu_sched_state);
3475 }
3476 EXPORT_SYMBOL_GPL(rcu_barrier_sched);
3477 
3478 /*
3479  * Propagate ->qsinitmask bits up the rcu_node tree to account for the
3480  * first CPU in a given leaf rcu_node structure coming online.  The caller
3481  * must hold the corresponding leaf rcu_node ->lock with interrrupts
3482  * disabled.
3483  */
3484 static void rcu_init_new_rnp(struct rcu_node *rnp_leaf)
3485 {
3486         long mask;
3487         struct rcu_node *rnp = rnp_leaf;
3488 
3489         raw_lockdep_assert_held_rcu_node(rnp);
3490         for (;;) {
3491                 mask = rnp->grpmask;
3492                 rnp = rnp->parent;
3493                 if (rnp == NULL)
3494                         return;
3495                 raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */
3496                 rnp->qsmaskinit |= mask;
3497                 raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */
3498         }
3499 }
3500 
3501 /*
3502  * Do boot-time initialization of a CPU's per-CPU RCU data.
3503  */
3504 static void __init
3505 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
3506 {
3507         struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3508 
3509         /* Set up local state, ensuring consistent view of global state. */
3510         rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu);
3511         rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
3512         WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != 1);
3513         WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp->dynticks)));
3514         rdp->cpu = cpu;
3515         rdp->rsp = rsp;
3516         rcu_boot_init_nocb_percpu_data(rdp);
3517 }
3518 
3519 /*
3520  * Initialize a CPU's per-CPU RCU data.  Note that only one online or
3521  * offline event can be happening at a given time.  Note also that we
3522  * can accept some slop in the rsp->completed access due to the fact
3523  * that this CPU cannot possibly have any RCU callbacks in flight yet.
3524  */
3525 static void
3526 rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
3527 {
3528         unsigned long flags;
3529         struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3530         struct rcu_node *rnp = rcu_get_root(rsp);
3531 
3532         /* Set up local state, ensuring consistent view of global state. */
3533         raw_spin_lock_irqsave_rcu_node(rnp, flags);
3534         rdp->qlen_last_fqs_check = 0;
3535         rdp->n_force_qs_snap = rsp->n_force_qs;
3536         rdp->blimit = blimit;
3537         if (rcu_segcblist_empty(&rdp->cblist) && /* No early-boot CBs? */
3538             !init_nocb_callback_list(rdp))
3539                 rcu_segcblist_init(&rdp->cblist);  /* Re-enable callbacks. */
3540         rdp->dynticks->dynticks_nesting = 1;    /* CPU not up, no tearing. */
3541         rcu_dynticks_eqs_online();
3542         raw_spin_unlock_rcu_node(rnp);          /* irqs remain disabled. */
3543 
3544         /*
3545          * Add CPU to leaf rcu_node pending-online bitmask.  Any needed
3546          * propagation up the rcu_node tree will happen at the beginning
3547          * of the next grace period.
3548          */
3549         rnp = rdp->mynode;
3550         raw_spin_lock_rcu_node(rnp);            /* irqs already disabled. */
3551         rdp->beenonline = true;  /* We have now been online. */
3552         rdp->gpnum = rnp->completed; /* Make CPU later note any new GP. */
3553         rdp->completed = rnp->completed;
3554         rdp->cpu_no_qs.b.norm = true;
3555         rdp->rcu_qs_ctr_snap = per_cpu(rcu_dynticks.rcu_qs_ctr, cpu);
3556         rdp->core_needs_qs = false;
3557         rdp->rcu_iw_pending = false;
3558         rdp->rcu_iw_gpnum = rnp->gpnum - 1;
3559         trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
3560         raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3561 }
3562 
3563 /*
3564  * Invoked early in the CPU-online process, when pretty much all
3565  * services are available.  The incoming CPU is not present.
3566  */
3567 int rcutree_prepare_cpu(unsigned int cpu)
3568 {
3569         struct rcu_state *rsp;
3570 
3571         for_each_rcu_flavor(rsp)
3572                 rcu_init_percpu_data(cpu, rsp);
3573 
3574         rcu_prepare_kthreads(cpu);
3575         rcu_spawn_all_nocb_kthreads(cpu);
3576 
3577         return 0;
3578 }
3579 
3580 /*
3581  * Update RCU priority boot kthread affinity for CPU-hotplug changes.
3582  */
3583 static void rcutree_affinity_setting(unsigned int cpu, int outgoing)
3584 {
3585         struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu);
3586 
3587         rcu_boost_kthread_setaffinity(rdp->mynode, outgoing);
3588 }
3589 
3590 /*
3591  * Near the end of the CPU-online process.  Pretty much all services
3592  * enabled, and the CPU is now very much alive.
3593  */
3594 int rcutree_online_cpu(unsigned int cpu)
3595 {
3596         unsigned long flags;
3597         struct rcu_data *rdp;
3598         struct rcu_node *rnp;
3599         struct rcu_state *rsp;
3600 
3601         for_each_rcu_flavor(rsp) {
3602                 rdp = per_cpu_ptr(rsp->rda, cpu);
3603                 rnp = rdp->mynode;
3604                 raw_spin_lock_irqsave_rcu_node(rnp, flags);
3605                 rnp->ffmask |= rdp->grpmask;
3606                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3607         }
3608         if (IS_ENABLED(CONFIG_TREE_SRCU))
3609                 srcu_online_cpu(cpu);
3610         if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE)
3611                 return 0; /* Too early in boot for scheduler work. */
3612         sync_sched_exp_online_cleanup(cpu);
3613         rcutree_affinity_setting(cpu, -1);
3614         return 0;
3615 }
3616 
3617 /*
3618  * Near the beginning of the process.  The CPU is still very much alive
3619  * with pretty much all services enabled.
3620  */
3621 int rcutree_offline_cpu(unsigned int cpu)
3622 {
3623         unsigned long flags;
3624         struct rcu_data *rdp;
3625         struct rcu_node *rnp;
3626         struct rcu_state *rsp;
3627 
3628         for_each_rcu_flavor(rsp) {
3629                 rdp = per_cpu_ptr(rsp->rda, cpu);
3630                 rnp = rdp->mynode;
3631                 raw_spin_lock_irqsave_rcu_node(rnp, flags);
3632                 rnp->ffmask &= ~rdp->grpmask;
3633                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3634         }
3635 
3636         rcutree_affinity_setting(cpu, cpu);
3637         if (IS_ENABLED(CONFIG_TREE_SRCU))
3638                 srcu_offline_cpu(cpu);
3639         return 0;
3640 }
3641 
3642 /*
3643  * Near the end of the offline process.  We do only tracing here.
3644  */
3645 int rcutree_dying_cpu(unsigned int cpu)
3646 {
3647         struct rcu_state *rsp;
3648 
3649         for_each_rcu_flavor(rsp)
3650                 rcu_cleanup_dying_cpu(rsp);
3651         return 0;
3652 }
3653 
3654 /*
3655  * The outgoing CPU is gone and we are running elsewhere.
3656  */
3657 int rcutree_dead_cpu(unsigned int cpu)
3658 {
3659         struct rcu_state *rsp;
3660 
3661         for_each_rcu_flavor(rsp) {
3662                 rcu_cleanup_dead_cpu(cpu, rsp);
3663                 do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu));
3664         }
3665         return 0;
3666 }
3667 
3668 static DEFINE_PER_CPU(int, rcu_cpu_started);
3669 
3670 /*
3671  * Mark the specified CPU as being online so that subsequent grace periods
3672  * (both expedited and normal) will wait on it.  Note that this means that
3673  * incoming CPUs are not allowed to use RCU read-side critical sections
3674  * until this function is called.  Failing to observe this restriction
3675  * will result in lockdep splats.
3676  *
3677  * Note that this function is special in that it is invoked directly
3678  * from the incoming CPU rather than from the cpuhp_step mechanism.
3679  * This is because this function must be invoked at a precise location.
3680  */
3681 void rcu_cpu_starting(unsigned int cpu)
3682 {
3683         unsigned long flags;
3684         unsigned long mask;
3685         int nbits;
3686         unsigned long oldmask;
3687         struct rcu_data *rdp;
3688         struct rcu_node *rnp;
3689         struct rcu_state *rsp;
3690 
3691         if (per_cpu(rcu_cpu_started, cpu))
3692                 return;
3693 
3694         per_cpu(rcu_cpu_started, cpu) = 1;
3695 
3696         for_each_rcu_flavor(rsp) {
3697                 rdp = per_cpu_ptr(rsp->rda, cpu);
3698                 rnp = rdp->mynode;
3699                 mask = rdp->grpmask;
3700                 raw_spin_lock_irqsave_rcu_node(rnp, flags);
3701                 rnp->qsmaskinitnext |= mask;
3702                 oldmask = rnp->expmaskinitnext;
3703                 rnp->expmaskinitnext |= mask;
3704                 oldmask ^= rnp->expmaskinitnext;
3705                 nbits = bitmap_weight(&oldmask, BITS_PER_LONG);
3706                 /* Allow lockless access for expedited grace periods. */
3707                 smp_store_release(&rsp->ncpus, rsp->ncpus + nbits); /* ^^^ */
3708                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3709         }
3710         smp_mb(); /* Ensure RCU read-side usage follows above initialization. */
3711 }
3712 
3713 #ifdef CONFIG_HOTPLUG_CPU
3714 /*
3715  * The CPU is exiting the idle loop into the arch_cpu_idle_dead()
3716  * function.  We now remove it from the rcu_node tree's ->qsmaskinit
3717  * bit masks.
3718  */
3719 static void rcu_cleanup_dying_idle_cpu(int cpu, struct rcu_state *rsp)
3720 {
3721         unsigned long flags;
3722         unsigned long mask;
3723         struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3724         struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
3725 
3726         /* Remove outgoing CPU from mask in the leaf rcu_node structure. */
3727         mask = rdp->grpmask;
3728         raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */
3729         rnp->qsmaskinitnext &= ~mask;
3730         raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3731 }
3732 
3733 /*
3734  * The outgoing function has no further need of RCU, so remove it from
3735  * the list of CPUs that RCU must track.
3736  *
3737  * Note that this function is special in that it is invoked directly
3738  * from the outgoing CPU rather than from the cpuhp_step mechanism.
3739  * This is because this function must be invoked at a precise location.
3740  */
3741 void rcu_report_dead(unsigned int cpu)
3742 {
3743         struct rcu_state *rsp;
3744 
3745         /* QS for any half-done expedited RCU-sched GP. */
3746         preempt_disable();
3747         rcu_report_exp_rdp(&rcu_sched_state,
3748                            this_cpu_ptr(rcu_sched_state.rda), true);
3749         preempt_enable();
3750         for_each_rcu_flavor(rsp)
3751                 rcu_cleanup_dying_idle_cpu(cpu, rsp);
3752 
3753         per_cpu(rcu_cpu_started, cpu) = 0;
3754 }
3755 
3756 /* Migrate the dead CPU's callbacks to the current CPU. */
3757 static void rcu_migrate_callbacks(int cpu, struct rcu_state *rsp)
3758 {
3759         unsigned long flags;
3760         struct rcu_data *my_rdp;
3761         struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3762         struct rcu_node *rnp_root = rcu_get_root(rdp->rsp);
3763         bool needwake;
3764 
3765         if (rcu_is_nocb_cpu(cpu) || rcu_segcblist_empty(&rdp->cblist))
3766                 return;  /* No callbacks to migrate. */
3767 
3768         local_irq_save(flags);
3769         my_rdp = this_cpu_ptr(rsp->rda);
3770         if (rcu_nocb_adopt_orphan_cbs(my_rdp, rdp, flags)) {
3771                 local_irq_restore(flags);
3772                 return;
3773         }
3774         raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */
3775         /* Leverage recent GPs and set GP for new callbacks. */
3776         needwake = rcu_advance_cbs(rsp, rnp_root, rdp) ||
3777                    rcu_advance_cbs(rsp, rnp_root, my_rdp);
3778         rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist);
3779         WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) !=
3780                      !rcu_segcblist_n_cbs(&my_rdp->cblist));
3781         raw_spin_unlock_irqrestore_rcu_node(rnp_root, flags);
3782         if (needwake)
3783                 rcu_gp_kthread_wake(rsp);
3784         WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 ||
3785                   !rcu_segcblist_empty(&rdp->cblist),
3786                   "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n",
3787                   cpu, rcu_segcblist_n_cbs(&rdp->cblist),
3788                   rcu_segcblist_first_cb(&rdp->cblist));
3789 }
3790 
3791 /*
3792  * The outgoing CPU has just passed through the dying-idle state,
3793  * and we are being invoked from the CPU that was IPIed to continue the
3794  * offline operation.  We need to migrate the outgoing CPU's callbacks.
3795  */
3796 void rcutree_migrate_callbacks(int cpu)
3797 {
3798         struct rcu_state *rsp;
3799 
3800         for_each_rcu_flavor(rsp)
3801                 rcu_migrate_callbacks(cpu, rsp);
3802 }
3803 #endif
3804 
3805 /*
3806  * On non-huge systems, use expedited RCU grace periods to make suspend
3807  * and hibernation run faster.
3808  */
3809 static int rcu_pm_notify(struct notifier_block *self,
3810                          unsigned long action, void *hcpu)
3811 {
3812         switch (action) {
3813         case PM_HIBERNATION_PREPARE:
3814         case PM_SUSPEND_PREPARE:
3815                 if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */
3816                         rcu_expedite_gp();
3817                 break;
3818         case PM_POST_HIBERNATION:
3819         case PM_POST_SUSPEND:
3820                 if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */
3821                         rcu_unexpedite_gp();
3822                 break;
3823         default:
3824                 break;
3825         }
3826         return NOTIFY_OK;
3827 }
3828 
3829 /*
3830  * Spawn the kthreads that handle each RCU flavor's grace periods.
3831  */
3832 static int __init rcu_spawn_gp_kthread(void)
3833 {
3834         unsigned long flags;
3835         int kthread_prio_in = kthread_prio;
3836         struct rcu_node *rnp;
3837         struct rcu_state *rsp;
3838         struct sched_param sp;
3839         struct task_struct *t;
3840 
3841         /* Force priority into range. */
3842         if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1)
3843                 kthread_prio = 1;
3844         else if (kthread_prio < 0)
3845                 kthread_prio = 0;
3846         else if (kthread_prio > 99)
3847                 kthread_prio = 99;
3848         if (kthread_prio != kthread_prio_in)
3849                 pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n",
3850                          kthread_prio, kthread_prio_in);
3851 
3852         rcu_scheduler_fully_active = 1;
3853         for_each_rcu_flavor(rsp) {
3854                 t = kthread_create(rcu_gp_kthread, rsp, "%s", rsp->name);
3855                 BUG_ON(IS_ERR(t));
3856                 rnp = rcu_get_root(rsp);
3857                 raw_spin_lock_irqsave_rcu_node(rnp, flags);
3858                 rsp->gp_kthread = t;
3859                 if (kthread_prio) {
3860                         sp.sched_priority = kthread_prio;
3861                         sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
3862                 }
3863                 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
3864                 wake_up_process(t);
3865         }
3866         rcu_spawn_nocb_kthreads();
3867         rcu_spawn_boost_kthreads();
3868         return 0;
3869 }
3870 early_initcall(rcu_spawn_gp_kthread);
3871 
3872 /*
3873  * This function is invoked towards the end of the scheduler's
3874  * initialization process.  Before this is called, the idle task might
3875  * contain synchronous grace-period primitives (during which time, this idle
3876  * task is booting the system, and such primitives are no-ops).  After this
3877  * function is called, any synchronous grace-period primitives are run as
3878  * expedited, with the requesting task driving the grace period forward.
3879  * A later core_initcall() rcu_set_runtime_mode() will switch to full
3880  * runtime RCU functionality.
3881  */
3882 void rcu_scheduler_starting(void)
3883 {
3884         WARN_ON(num_online_cpus() != 1);
3885         WARN_ON(nr_context_switches() > 0);
3886         rcu_test_sync_prims();
3887         rcu_scheduler_active = RCU_SCHEDULER_INIT;
3888         rcu_test_sync_prims();
3889 }
3890 
3891 /*
3892  * Helper function for rcu_init() that initializes one rcu_state structure.
3893  */
3894 static void __init rcu_init_one(struct rcu_state *rsp)
3895 {
3896         static const char * const buf[] = RCU_NODE_NAME_INIT;
3897         static const char * const fqs[] = RCU_FQS_NAME_INIT;
3898         static struct lock_class_key rcu_node_class[RCU_NUM_LVLS];
3899         static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS];
3900 
3901         int levelspread[RCU_NUM_LVLS];          /* kids/node in each level. */
3902         int cpustride = 1;
3903         int i;
3904         int j;
3905         struct rcu_node *rnp;
3906 
3907         BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf));  /* Fix buf[] init! */
3908 
3909         /* Silence gcc 4.8 false positive about array index out of range. */
3910         if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS)
3911                 panic("rcu_init_one: rcu_num_lvls out of range");
3912 
3913         /* Initialize the level-tracking arrays. */
3914 
3915         for (i = 1; i < rcu_num_lvls; i++)
3916                 rsp->level[i] = rsp->level[i - 1] + num_rcu_lvl[i - 1];
3917         rcu_init_levelspread(levelspread, num_rcu_lvl);
3918 
3919         /* Initialize the elements themselves, starting from the leaves. */
3920 
3921         for (i = rcu_num_lvls - 1; i >= 0; i--) {
3922                 cpustride *= levelspread[i];
3923                 rnp = rsp->level[i];
3924                 for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) {
3925                         raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock));
3926                         lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock),
3927                                                    &rcu_node_class[i], buf[i]);
3928                         raw_spin_lock_init(&rnp->fqslock);
3929                         lockdep_set_class_and_name(&rnp->fqslock,
3930                                                    &rcu_fqs_class[i], fqs[i]);
3931                         rnp->gpnum = rsp->gpnum;
3932                         rnp->completed = rsp->completed;
3933                         rnp->qsmask = 0;
3934                         rnp->qsmaskinit = 0;
3935                         rnp->grplo = j * cpustride;
3936                         rnp->grphi = (j + 1) * cpustride - 1;
3937                         if (rnp->grphi >= nr_cpu_ids)
3938                                 rnp->grphi = nr_cpu_ids - 1;
3939                         if (i == 0) {
3940                                 rnp->grpnum = 0;
3941                                 rnp->grpmask = 0;
3942                                 rnp->parent = NULL;
3943                         } else {
3944                                 rnp->grpnum = j % levelspread[i - 1];
3945                                 rnp->grpmask = 1UL << rnp->grpnum;
3946                                 rnp->parent = rsp->level[i - 1] +
3947                                               j / levelspread[i - 1];
3948                         }
3949                         rnp->level = i;
3950                         INIT_LIST_HEAD(&rnp->blkd_tasks);
3951                         rcu_init_one_nocb(rnp);
3952                         init_waitqueue_head(&rnp->exp_wq[0]);
3953                         init_waitqueue_head(&rnp->exp_wq[1]);
3954                         init_waitqueue_head(&rnp->exp_wq[2]);
3955                         init_waitqueue_head(&rnp->exp_wq[3]);
3956                         spin_lock_init(&rnp->exp_lock);
3957                 }
3958         }
3959 
3960         init_swait_queue_head(&rsp->gp_wq);
3961         init_swait_queue_head(&rsp->expedited_wq);
3962         rnp = rcu_first_leaf_node(rsp);
3963         for_each_possible_cpu(i) {
3964                 while (i > rnp->grphi)
3965                         rnp++;
3966                 per_cpu_ptr(rsp->rda, i)->mynode = rnp;
3967                 rcu_boot_init_percpu_data(i, rsp);
3968         }
3969         list_add(&rsp->flavors, &rcu_struct_flavors);
3970 }
3971 
3972 /*
3973  * Compute the rcu_node tree geometry from kernel parameters.  This cannot
3974  * replace the definitions in tree.h because those are needed to size
3975  * the ->node array in the rcu_state structure.
3976  */
3977 static void __init rcu_init_geometry(void)
3978 {
3979         ulong d;
3980         int i;
3981         int rcu_capacity[RCU_NUM_LVLS];
3982 
3983         /*
3984          * Initialize any unspecified boot parameters.
3985          * The default values of jiffies_till_first_fqs and
3986          * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS
3987          * value, which is a function of HZ, then adding one for each
3988          * RCU_JIFFIES_FQS_DIV CPUs that might be on the system.
3989          */
3990         d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV;
3991         if (jiffies_till_first_fqs == ULONG_MAX)
3992                 jiffies_till_first_fqs = d;
3993         if (jiffies_till_next_fqs == ULONG_MAX)
3994                 jiffies_till_next_fqs = d;
3995 
3996         /* If the compile-time values are accurate, just leave. */
3997         if (rcu_fanout_leaf == RCU_FANOUT_LEAF &&
3998             nr_cpu_ids == NR_CPUS)
3999                 return;
4000         pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n",
4001                 rcu_fanout_leaf, nr_cpu_ids);
4002 
4003         /*
4004          * The boot-time rcu_fanout_leaf parameter must be at least two
4005          * and cannot exceed the number of bits in the rcu_node masks.
4006          * Complain and fall back to the compile-time values if this
4007          * limit is exceeded.
4008          */
4009         if (rcu_fanout_leaf < 2 ||
4010             rcu_fanout_leaf > sizeof(unsigned long) * 8) {
4011                 rcu_fanout_leaf = RCU_FANOUT_LEAF;
4012                 WARN_ON(1);
4013                 return;
4014         }
4015 
4016         /*
4017          * Compute number of nodes that can be handled an rcu_node tree
4018          * with the given number of levels.
4019          */
4020         rcu_capacity[0] = rcu_fanout_leaf;
4021         for (i = 1; i < RCU_NUM_LVLS; i++)
4022                 rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT;
4023 
4024         /*
4025          * The tree must be able to accommodate the configured number of CPUs.
4026          * If this limit is exceeded, fall back to the compile-time values.
4027          */
4028         if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) {
4029                 rcu_fanout_leaf = RCU_FANOUT_LEAF;
4030                 WARN_ON(1);
4031                 return;
4032         }
4033 
4034         /* Calculate the number of levels in the tree. */
4035         for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) {
4036         }
4037         rcu_num_lvls = i + 1;
4038 
4039         /* Calculate the number of rcu_nodes at each level of the tree. */
4040         for (i = 0; i < rcu_num_lvls; i++) {
4041                 int cap = rcu_capacity[(rcu_num_lvls - 1) - i];
4042                 num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap);
4043         }
4044 
4045         /* Calculate the total number of rcu_node structures. */
4046         rcu_num_nodes = 0;
4047         for (i = 0; i < rcu_num_lvls; i++)
4048                 rcu_num_nodes += num_rcu_lvl[i];
4049 }
4050 
4051 /*
4052  * Dump out the structure of the rcu_node combining tree associated
4053  * with the rcu_state structure referenced by rsp.
4054  */
4055 static void __init rcu_dump_rcu_node_tree(struct rcu_state *rsp)
4056 {
4057         int level = 0;
4058         struct rcu_node *rnp;
4059 
4060         pr_info("rcu_node tree layout dump\n");
4061         pr_info(" ");
4062         rcu_for_each_node_breadth_first(rsp, rnp) {
4063                 if (rnp->level != level) {
4064                         pr_cont("\n");
4065                         pr_info(" ");
4066                         level = rnp->level;
4067                 }
4068                 pr_cont("%d:%d ^%d  ", rnp->grplo, rnp->grphi, rnp->grpnum);
4069         }
4070         pr_cont("\n");
4071 }
4072 
4073 struct workqueue_struct *rcu_gp_wq;
4074 struct workqueue_struct *rcu_par_gp_wq;
4075 
4076 void __init rcu_init(void)
4077 {
4078         int cpu;
4079 
4080         rcu_early_boot_tests();
4081 
4082         rcu_bootup_announce();
4083         rcu_init_geometry();
4084         rcu_init_one(&rcu_bh_state);
4085         rcu_init_one(&rcu_sched_state);
4086         if (dump_tree)
4087                 rcu_dump_rcu_node_tree(&rcu_sched_state);
4088         __rcu_init_preempt();
4089         open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
4090 
4091         /*
4092          * We don't need protection against CPU-hotplug here because
4093          * this is called early in boot, before either interrupts
4094          * or the scheduler are operational.
4095          */
4096         pm_notifier(rcu_pm_notify, 0);
4097         for_each_online_cpu(cpu) {
4098                 rcutree_prepare_cpu(cpu);
4099                 rcu_cpu_starting(cpu);
4100                 rcutree_online_cpu(cpu);
4101         }
4102 
4103         /* Create workqueue for expedited GPs and for Tree SRCU. */
4104         rcu_gp_wq = alloc_workqueue("rcu_gp", WQ_MEM_RECLAIM, 0);
4105         WARN_ON(!rcu_gp_wq);
4106         rcu_par_gp_wq = alloc_workqueue("rcu_par_gp", WQ_MEM_RECLAIM, 0);
4107         WARN_ON(!rcu_par_gp_wq);
4108 }
4109 
4110 #include "tree_exp.h"
4111 #include "tree_plugin.h"
4112 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp