~hc/RK356X_SDK_RELEASE.git

..	..	@@ -1,63 +1,55 @@
	1	+/* SPDX-License-Identifier: GPL-2.0+ */
1	2	/*
2	3	* RCU expedited grace periods
3	4	*
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	5	* Copyright IBM Corporation, 2016
19	6	*
20		- * Authors: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
	7	+ * Authors: Paul E. McKenney <paulmck@linux.ibm.com>
21	8	*/
22	9
23	10	#include <linux/lockdep.h>
24	11
	12	+static void rcu_exp_handler(void *unused);
	13	+static int rcu_print_task_exp_stall(struct rcu_node *rnp);
	14	+
25	15	/*
26	16	* Record the start of an expedited grace period.
27	17	*/
28		-static void rcu_exp_gp_seq_start(struct rcu_state *rsp)
	18	+static void rcu_exp_gp_seq_start(void)
29	19	{
30		- rcu_seq_start(&rsp->expedited_sequence);
	20	+ rcu_seq_start(&rcu_state.expedited_sequence);
31	21	}
32	22
33	23	/*
34		- * Return then value that expedited-grace-period counter will have
	24	+ * Return the value that the expedited-grace-period counter will have
35	25	* at the end of the current grace period.
36	26	*/
37		-static __maybe_unused unsigned long rcu_exp_gp_seq_endval(struct rcu_state *rsp)
	27	+static __maybe_unused unsigned long rcu_exp_gp_seq_endval(void)
38	28	{
39		- return rcu_seq_endval(&rsp->expedited_sequence);
	29	+ return rcu_seq_endval(&rcu_state.expedited_sequence);
40	30	}
41	31
42	32	/*
43	33	* Record the end of an expedited grace period.
44	34	*/
45		-static void rcu_exp_gp_seq_end(struct rcu_state *rsp)
	35	+static void rcu_exp_gp_seq_end(void)
46	36	{
47		- rcu_seq_end(&rsp->expedited_sequence);
	37	+ rcu_seq_end(&rcu_state.expedited_sequence);
48	38	smp_mb(); /* Ensure that consecutive grace periods serialize. */
49	39	}
50	40
51	41	/*
52		- * Take a snapshot of the expedited-grace-period counter.
	42	+ * Take a snapshot of the expedited-grace-period counter, which is the
	43	+ * earliest value that will indicate that a full grace period has
	44	+ * elapsed since the current time.
53	45	*/
54		-static unsigned long rcu_exp_gp_seq_snap(struct rcu_state *rsp)
	46	+static unsigned long rcu_exp_gp_seq_snap(void)
55	47	{
56	48	unsigned long s;
57	49
58	50	smp_mb(); /* Caller's modifications seen first by other CPUs. */
59		- s = rcu_seq_snap(&rsp->expedited_sequence);
60		- trace_rcu_exp_grace_period(rsp->name, s, TPS("snap"));
	51	+ s = rcu_seq_snap(&rcu_state.expedited_sequence);
	52	+ trace_rcu_exp_grace_period(rcu_state.name, s, TPS("snap"));
61	53	return s;
62	54	}
63	55
..	..	@@ -66,9 +58,9 @@
66	58	* if a full expedited grace period has elapsed since that snapshot
67	59	* was taken.
68	60	*/
69		-static bool rcu_exp_gp_seq_done(struct rcu_state *rsp, unsigned long s)
	61	+static bool rcu_exp_gp_seq_done(unsigned long s)
70	62	{
71		- return rcu_seq_done(&rsp->expedited_sequence, s);
	63	+ return rcu_seq_done(&rcu_state.expedited_sequence, s);
72	64	}
73	65
74	66	/*
..	..	@@ -78,26 +70,26 @@
78	70	* ever been online. This means that this function normally takes its
79	71	* no-work-to-do fastpath.
80	72	*/
81		-static void sync_exp_reset_tree_hotplug(struct rcu_state *rsp)
	73	+static void sync_exp_reset_tree_hotplug(void)
82	74	{
83	75	bool done;
84	76	unsigned long flags;
85	77	unsigned long mask;
86	78	unsigned long oldmask;
87		- int ncpus = smp_load_acquire(&rsp->ncpus); /* Order against locking. */
	79	+ int ncpus = smp_load_acquire(&rcu_state.ncpus); /* Order vs. locking. */
88	80	struct rcu_node *rnp;
89	81	struct rcu_node *rnp_up;
90	82
91	83	/* If no new CPUs onlined since last time, nothing to do. */
92		- if (likely(ncpus == rsp->ncpus_snap))
	84	+ if (likely(ncpus == rcu_state.ncpus_snap))
93	85	return;
94		- rsp->ncpus_snap = ncpus;
	86	+ rcu_state.ncpus_snap = ncpus;
95	87
96	88	/*
97	89	* Each pass through the following loop propagates newly onlined
98	90	* CPUs for the current rcu_node structure up the rcu_node tree.
99	91	*/
100		- rcu_for_each_leaf_node(rsp, rnp) {
	92	+ rcu_for_each_leaf_node(rnp) {
101	93	raw_spin_lock_irqsave_rcu_node(rnp, flags);
102	94	if (rnp->expmaskinit == rnp->expmaskinitnext) {
103	95	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
..	..	@@ -135,16 +127,16 @@
135	127	* Reset the ->expmask values in the rcu_node tree in preparation for
136	128	* a new expedited grace period.
137	129	*/
138		-static void __maybe_unused sync_exp_reset_tree(struct rcu_state *rsp)
	130	+static void __maybe_unused sync_exp_reset_tree(void)
139	131	{
140	132	unsigned long flags;
141	133	struct rcu_node *rnp;
142	134
143		- sync_exp_reset_tree_hotplug(rsp);
144		- rcu_for_each_node_breadth_first(rsp, rnp) {
	135	+ sync_exp_reset_tree_hotplug();
	136	+ rcu_for_each_node_breadth_first(rnp) {
145	137	raw_spin_lock_irqsave_rcu_node(rnp, flags);
146	138	WARN_ON_ONCE(rnp->expmask);
147		- rnp->expmask = rnp->expmaskinit;
	139	+ WRITE_ONCE(rnp->expmask, rnp->expmaskinit);
148	140	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
149	141	}
150	142	}
..	..	@@ -153,31 +145,26 @@
153	145	* Return non-zero if there is no RCU expedited grace period in progress
154	146	* for the specified rcu_node structure, in other words, if all CPUs and
155	147	* tasks covered by the specified rcu_node structure have done their bit
156		- * for the current expedited grace period. Works only for preemptible
157		- * RCU -- other RCU implementation use other means.
158		- *
159		- * Caller must hold the specificed rcu_node structure's ->lock
	148	+ * for the current expedited grace period.
160	149	*/
161		-static bool sync_rcu_preempt_exp_done(struct rcu_node *rnp)
	150	+static bool sync_rcu_exp_done(struct rcu_node *rnp)
162	151	{
163	152	raw_lockdep_assert_held_rcu_node(rnp);
164		-
165		- return rnp->exp_tasks == NULL &&
	153	+ return READ_ONCE(rnp->exp_tasks) == NULL &&
166	154	READ_ONCE(rnp->expmask) == 0;
167	155	}
168	156
169	157	/*
170		- * Like sync_rcu_preempt_exp_done(), but this function assumes the caller
171		- * doesn't hold the rcu_node's ->lock, and will acquire and release the lock
172		- * itself
	158	+ * Like sync_rcu_exp_done(), but where the caller does not hold the
	159	+ * rcu_node's ->lock.
173	160	*/
174		-static bool sync_rcu_preempt_exp_done_unlocked(struct rcu_node *rnp)
	161	+static bool sync_rcu_exp_done_unlocked(struct rcu_node *rnp)
175	162	{
176	163	unsigned long flags;
177	164	bool ret;
178	165
179	166	raw_spin_lock_irqsave_rcu_node(rnp, flags);
180		- ret = sync_rcu_preempt_exp_done(rnp);
	167	+ ret = sync_rcu_exp_done(rnp);
181	168	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
182	169
183	170	return ret;
..	..	@@ -191,17 +178,16 @@
191	178	* which the task was queued or to one of that rcu_node structure's ancestors,
192	179	* recursively up the tree. (Calm down, calm down, we do the recursion
193	180	* iteratively!)
194		- *
195		- * Caller must hold the specified rcu_node structure's ->lock.
196	181	*/
197		-static void __rcu_report_exp_rnp(struct rcu_state rsp, struct rcu_node rnp,
	182	+static void __rcu_report_exp_rnp(struct rcu_node *rnp,
198	183	bool wake, unsigned long flags)
199	184	__releases(rnp->lock)
200	185	{
201	186	unsigned long mask;
202	187
	188	+ raw_lockdep_assert_held_rcu_node(rnp);
203	189	for (;;) {
204		- if (!sync_rcu_preempt_exp_done(rnp)) {
	190	+ if (!sync_rcu_exp_done(rnp)) {
205	191	if (!rnp->expmask)
206	192	rcu_initiate_boost(rnp, flags);
207	193	else
..	..	@@ -212,7 +198,7 @@
212	198	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
213	199	if (wake) {
214	200	smp_mb(); /* EGP done before wake_up(). */
215		- swake_up_one(&rsp->expedited_wq);
	201	+ swake_up_one(&rcu_state.expedited_wq);
216	202	}
217	203	break;
218	204	}
..	..	@@ -221,7 +207,7 @@
221	207	rnp = rnp->parent;
222	208	raw_spin_lock_rcu_node(rnp); /* irqs already disabled */
223	209	WARN_ON_ONCE(!(rnp->expmask & mask));
224		- rnp->expmask &= ~mask;
	210	+ WRITE_ONCE(rnp->expmask, rnp->expmask & ~mask);
225	211	}
226	212	}
227	213
..	..	@@ -229,49 +215,56 @@
229	215	* Report expedited quiescent state for specified node. This is a
230	216	* lock-acquisition wrapper function for __rcu_report_exp_rnp().
231	217	*/
232		-static void __maybe_unused rcu_report_exp_rnp(struct rcu_state *rsp,
233		- struct rcu_node *rnp, bool wake)
	218	+static void __maybe_unused rcu_report_exp_rnp(struct rcu_node *rnp, bool wake)
234	219	{
235	220	unsigned long flags;
236	221
237	222	raw_spin_lock_irqsave_rcu_node(rnp, flags);
238		- __rcu_report_exp_rnp(rsp, rnp, wake, flags);
	223	+ __rcu_report_exp_rnp(rnp, wake, flags);
239	224	}
240	225
241	226	/*
242	227	* Report expedited quiescent state for multiple CPUs, all covered by the
243	228	* specified leaf rcu_node structure.
244	229	*/
245		-static void rcu_report_exp_cpu_mult(struct rcu_state rsp, struct rcu_node rnp,
	230	+static void rcu_report_exp_cpu_mult(struct rcu_node *rnp,
246	231	unsigned long mask, bool wake)
247	232	{
	233	+ int cpu;
248	234	unsigned long flags;
	235	+ struct rcu_data *rdp;
249	236
250	237	raw_spin_lock_irqsave_rcu_node(rnp, flags);
251	238	if (!(rnp->expmask & mask)) {
252	239	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
253	240	return;
254	241	}
255		- rnp->expmask &= ~mask;
256		- __rcu_report_exp_rnp(rsp, rnp, wake, flags); /* Releases rnp->lock. */
	242	+ WRITE_ONCE(rnp->expmask, rnp->expmask & ~mask);
	243	+ for_each_leaf_node_cpu_mask(rnp, cpu, mask) {
	244	+ rdp = per_cpu_ptr(&rcu_data, cpu);
	245	+ if (!IS_ENABLED(CONFIG_NO_HZ_FULL) \|\| !rdp->rcu_forced_tick_exp)
	246	+ continue;
	247	+ rdp->rcu_forced_tick_exp = false;
	248	+ tick_dep_clear_cpu(cpu, TICK_DEP_BIT_RCU_EXP);
	249	+ }
	250	+ __rcu_report_exp_rnp(rnp, wake, flags); /* Releases rnp->lock. */
257	251	}
258	252
259	253	/*
260	254	* Report expedited quiescent state for specified rcu_data (CPU).
261	255	*/
262		-static void rcu_report_exp_rdp(struct rcu_state rsp, struct rcu_data rdp,
263		- bool wake)
	256	+static void rcu_report_exp_rdp(struct rcu_data *rdp)
264	257	{
265		- rcu_report_exp_cpu_mult(rsp, rdp->mynode, rdp->grpmask, wake);
	258	+ WRITE_ONCE(rdp->exp_deferred_qs, false);
	259	+ rcu_report_exp_cpu_mult(rdp->mynode, rdp->grpmask, true);
266	260	}
267	261
268		-/* Common code for synchronize_{rcu,sched}_expedited() work-done checking. */
269		-static bool sync_exp_work_done(struct rcu_state *rsp, unsigned long s)
	262	+/* Common code for work-done checking. */
	263	+static bool sync_exp_work_done(unsigned long s)
270	264	{
271		- if (rcu_exp_gp_seq_done(rsp, s)) {
272		- trace_rcu_exp_grace_period(rsp->name, s, TPS("done"));
273		- /* Ensure test happens before caller kfree(). */
274		- smp_mb__before_atomic(); /* ^^^ */
	265	+ if (rcu_exp_gp_seq_done(s)) {
	266	+ trace_rcu_exp_grace_period(rcu_state.name, s, TPS("done"));
	267	+ smp_mb(); /* Ensure test happens before caller kfree(). */
275	268	return true;
276	269	}
277	270	return false;
..	..	@@ -284,28 +277,28 @@
284	277	* with the mutex held, indicating that the caller must actually do the
285	278	* expedited grace period.
286	279	*/
287		-static bool exp_funnel_lock(struct rcu_state *rsp, unsigned long s)
	280	+static bool exp_funnel_lock(unsigned long s)
288	281	{
289		- struct rcu_data *rdp = per_cpu_ptr(rsp->rda, raw_smp_processor_id());
	282	+ struct rcu_data *rdp = per_cpu_ptr(&rcu_data, raw_smp_processor_id());
290	283	struct rcu_node *rnp = rdp->mynode;
291		- struct rcu_node *rnp_root = rcu_get_root(rsp);
	284	+ struct rcu_node *rnp_root = rcu_get_root();
292	285
293	286	/* Low-contention fastpath. */
294	287	if (ULONG_CMP_LT(READ_ONCE(rnp->exp_seq_rq), s) &&
295	288	(rnp == rnp_root \|\|
296	289	ULONG_CMP_LT(READ_ONCE(rnp_root->exp_seq_rq), s)) &&
297		- mutex_trylock(&rsp->exp_mutex))
	290	+ mutex_trylock(&rcu_state.exp_mutex))
298	291	goto fastpath;
299	292
300	293	/*
301	294	* Each pass through the following loop works its way up
302	295	* the rcu_node tree, returning if others have done the work or
303		- * otherwise falls through to acquire rsp->exp_mutex. The mapping
	296	+ * otherwise falls through to acquire ->exp_mutex. The mapping
304	297	* from CPU to rcu_node structure can be inexact, as it is just
305	298	* promoting locality and is not strictly needed for correctness.
306	299	*/
307	300	for (; rnp != NULL; rnp = rnp->parent) {
308		- if (sync_exp_work_done(rsp, s))
	301	+ if (sync_exp_work_done(s))
309	302	return true;
310	303
311	304	/* Work not done, either wait here or go up. */
..	..	@@ -314,66 +307,27 @@
314	307
315	308	/* Someone else doing GP, so wait for them. */
316	309	spin_unlock(&rnp->exp_lock);
317		- trace_rcu_exp_funnel_lock(rsp->name, rnp->level,
	310	+ trace_rcu_exp_funnel_lock(rcu_state.name, rnp->level,
318	311	rnp->grplo, rnp->grphi,
319	312	TPS("wait"));
320	313	wait_event(rnp->exp_wq[rcu_seq_ctr(s) & 0x3],
321		- sync_exp_work_done(rsp, s));
	314	+ sync_exp_work_done(s));
322	315	return true;
323	316	}
324		- rnp->exp_seq_rq = s; /* Followers can wait on us. */
	317	+ WRITE_ONCE(rnp->exp_seq_rq, s); /* Followers can wait on us. */
325	318	spin_unlock(&rnp->exp_lock);
326		- trace_rcu_exp_funnel_lock(rsp->name, rnp->level, rnp->grplo,
327		- rnp->grphi, TPS("nxtlvl"));
	319	+ trace_rcu_exp_funnel_lock(rcu_state.name, rnp->level,
	320	+ rnp->grplo, rnp->grphi, TPS("nxtlvl"));
328	321	}
329		- mutex_lock(&rsp->exp_mutex);
	322	+ mutex_lock(&rcu_state.exp_mutex);
330	323	fastpath:
331		- if (sync_exp_work_done(rsp, s)) {
332		- mutex_unlock(&rsp->exp_mutex);
	324	+ if (sync_exp_work_done(s)) {
	325	+ mutex_unlock(&rcu_state.exp_mutex);
333	326	return true;
334	327	}
335		- rcu_exp_gp_seq_start(rsp);
336		- trace_rcu_exp_grace_period(rsp->name, s, TPS("start"));
	328	+ rcu_exp_gp_seq_start();
	329	+ trace_rcu_exp_grace_period(rcu_state.name, s, TPS("start"));
337	330	return false;
338		-}
339		-
340		-/* Invoked on each online non-idle CPU for expedited quiescent state. */
341		-static void sync_sched_exp_handler(void *data)
342		-{
343		- struct rcu_data *rdp;
344		- struct rcu_node *rnp;
345		- struct rcu_state *rsp = data;
346		-
347		- rdp = this_cpu_ptr(rsp->rda);
348		- rnp = rdp->mynode;
349		- if (!(READ_ONCE(rnp->expmask) & rdp->grpmask) \|\|
350		- __this_cpu_read(rcu_sched_data.cpu_no_qs.b.exp))
351		- return;
352		- if (rcu_is_cpu_rrupt_from_idle()) {
353		- rcu_report_exp_rdp(&rcu_sched_state,
354		- this_cpu_ptr(&rcu_sched_data), true);
355		- return;
356		- }
357		- __this_cpu_write(rcu_sched_data.cpu_no_qs.b.exp, true);
358		- /* Store .exp before .rcu_urgent_qs. */
359		- smp_store_release(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs), true);
360		- resched_cpu(smp_processor_id());
361		-}
362		-
363		-/* Send IPI for expedited cleanup if needed at end of CPU-hotplug operation. */
364		-static void sync_sched_exp_online_cleanup(int cpu)
365		-{
366		- struct rcu_data *rdp;
367		- int ret;
368		- struct rcu_node *rnp;
369		- struct rcu_state *rsp = &rcu_sched_state;
370		-
371		- rdp = per_cpu_ptr(rsp->rda, cpu);
372		- rnp = rdp->mynode;
373		- if (!(READ_ONCE(rnp->expmask) & rdp->grpmask))
374		- return;
375		- ret = smp_call_function_single(cpu, sync_sched_exp_handler, rsp, 0);
376		- WARN_ON_ONCE(ret);
377	331	}
378	332
379	333	/*
..	..	@@ -384,31 +338,27 @@
384	338	{
385	339	int cpu;
386	340	unsigned long flags;
387		- smp_call_func_t func;
388	341	unsigned long mask_ofl_test;
389	342	unsigned long mask_ofl_ipi;
390	343	int ret;
391	344	struct rcu_exp_work *rewp =
392	345	container_of(wp, struct rcu_exp_work, rew_work);
393	346	struct rcu_node *rnp = container_of(rewp, struct rcu_node, rew);
394		- struct rcu_state *rsp = rewp->rew_rsp;
395	347
396		- func = rewp->rew_func;
397	348	raw_spin_lock_irqsave_rcu_node(rnp, flags);
398	349
399	350	/* Each pass checks a CPU for identity, offline, and idle. */
400	351	mask_ofl_test = 0;
401	352	for_each_leaf_node_cpu_mask(rnp, cpu, rnp->expmask) {
402		- unsigned long mask = leaf_node_cpu_bit(rnp, cpu);
403		- struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
404		- struct rcu_dynticks *rdtp = per_cpu_ptr(&rcu_dynticks, cpu);
	353	+ struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
	354	+ unsigned long mask = rdp->grpmask;
405	355	int snap;
406	356
407	357	if (raw_smp_processor_id() == cpu \|\|
408	358	!(rnp->qsmaskinitnext & mask)) {
409	359	mask_ofl_test \|= mask;
410	360	} else {
411		- snap = rcu_dynticks_snap(rdtp);
	361	+ snap = rcu_dynticks_snap(rdp);
412	362	if (rcu_dynticks_in_eqs(snap))
413	363	mask_ofl_test \|= mask;
414	364	else
..	..	@@ -423,125 +373,172 @@
423	373	* until such time as the ->expmask bits are cleared.
424	374	*/
425	375	if (rcu_preempt_has_tasks(rnp))
426		- rnp->exp_tasks = rnp->blkd_tasks.next;
	376	+ WRITE_ONCE(rnp->exp_tasks, rnp->blkd_tasks.next);
427	377	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
428	378
429	379	/* IPI the remaining CPUs for expedited quiescent state. */
430		- for_each_leaf_node_cpu_mask(rnp, cpu, rnp->expmask) {
431		- unsigned long mask = leaf_node_cpu_bit(rnp, cpu);
432		- struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
	380	+ for_each_leaf_node_cpu_mask(rnp, cpu, mask_ofl_ipi) {
	381	+ struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
	382	+ unsigned long mask = rdp->grpmask;
433	383
434		- if (!(mask_ofl_ipi & mask))
435		- continue;
436	384	retry_ipi:
437		- if (rcu_dynticks_in_eqs_since(rdp->dynticks,
438		- rdp->exp_dynticks_snap)) {
	385	+ if (rcu_dynticks_in_eqs_since(rdp, rdp->exp_dynticks_snap)) {
439	386	mask_ofl_test \|= mask;
440	387	continue;
441	388	}
442		- ret = smp_call_function_single(cpu, func, rsp, 0);
443		- if (!ret) {
444		- mask_ofl_ipi &= ~mask;
	389	+ if (get_cpu() == cpu) {
	390	+ mask_ofl_test \|= mask;
	391	+ put_cpu();
445	392	continue;
446	393	}
	394	+ ret = smp_call_function_single(cpu, rcu_exp_handler, NULL, 0);
	395	+ put_cpu();
	396	+ /* The CPU will report the QS in response to the IPI. */
	397	+ if (!ret)
	398	+ continue;
	399	+
447	400	/* Failed, raced with CPU hotplug operation. */
448	401	raw_spin_lock_irqsave_rcu_node(rnp, flags);
449	402	if ((rnp->qsmaskinitnext & mask) &&
450	403	(rnp->expmask & mask)) {
451	404	/* Online, so delay for a bit and try again. */
452	405	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
453		- trace_rcu_exp_grace_period(rsp->name, rcu_exp_gp_seq_endval(rsp), TPS("selectofl"));
454		- schedule_timeout_uninterruptible(1);
	406	+ trace_rcu_exp_grace_period(rcu_state.name, rcu_exp_gp_seq_endval(), TPS("selectofl"));
	407	+ schedule_timeout_idle(1);
455	408	goto retry_ipi;
456	409	}
457		- /* CPU really is offline, so we can ignore it. */
458		- if (!(rnp->expmask & mask))
459		- mask_ofl_ipi &= ~mask;
	410	+ /* CPU really is offline, so we must report its QS. */
	411	+ if (rnp->expmask & mask)
	412	+ mask_ofl_test \|= mask;
460	413	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
461	414	}
462	415	/* Report quiescent states for those that went offline. */
463		- mask_ofl_test \|= mask_ofl_ipi;
464	416	if (mask_ofl_test)
465		- rcu_report_exp_cpu_mult(rsp, rnp, mask_ofl_test, false);
	417	+ rcu_report_exp_cpu_mult(rnp, mask_ofl_test, false);
466	418	}
467	419
468	420	/*
469	421	* Select the nodes that the upcoming expedited grace period needs
470	422	* to wait for.
471	423	*/
472		-static void sync_rcu_exp_select_cpus(struct rcu_state *rsp,
473		- smp_call_func_t func)
	424	+static void sync_rcu_exp_select_cpus(void)
474	425	{
	426	+ int cpu;
475	427	struct rcu_node *rnp;
476	428
477		- trace_rcu_exp_grace_period(rsp->name, rcu_exp_gp_seq_endval(rsp), TPS("reset"));
478		- sync_exp_reset_tree(rsp);
479		- trace_rcu_exp_grace_period(rsp->name, rcu_exp_gp_seq_endval(rsp), TPS("select"));
	429	+ trace_rcu_exp_grace_period(rcu_state.name, rcu_exp_gp_seq_endval(), TPS("reset"));
	430	+ sync_exp_reset_tree();
	431	+ trace_rcu_exp_grace_period(rcu_state.name, rcu_exp_gp_seq_endval(), TPS("select"));
480	432
481	433	/* Schedule work for each leaf rcu_node structure. */
482		- rcu_for_each_leaf_node(rsp, rnp) {
	434	+ rcu_for_each_leaf_node(rnp) {
483	435	rnp->exp_need_flush = false;
484	436	if (!READ_ONCE(rnp->expmask))
485	437	continue; /* Avoid early boot non-existent wq. */
486		- rnp->rew.rew_func = func;
487		- rnp->rew.rew_rsp = rsp;
488	438	if (!READ_ONCE(rcu_par_gp_wq) \|\|
489	439	rcu_scheduler_active != RCU_SCHEDULER_RUNNING \|\|
490		- rcu_is_last_leaf_node(rsp, rnp)) {
	440	+ rcu_is_last_leaf_node(rnp)) {
491	441	/* No workqueues yet or last leaf, do direct call. */
492	442	sync_rcu_exp_select_node_cpus(&rnp->rew.rew_work);
493	443	continue;
494	444	}
495	445	INIT_WORK(&rnp->rew.rew_work, sync_rcu_exp_select_node_cpus);
496		- queue_work_on(rnp->grplo, rcu_par_gp_wq, &rnp->rew.rew_work);
	446	+ cpu = find_next_bit(&rnp->ffmask, BITS_PER_LONG, -1);
	447	+ /* If all offline, queue the work on an unbound CPU. */
	448	+ if (unlikely(cpu > rnp->grphi - rnp->grplo))
	449	+ cpu = WORK_CPU_UNBOUND;
	450	+ else
	451	+ cpu += rnp->grplo;
	452	+ queue_work_on(cpu, rcu_par_gp_wq, &rnp->rew.rew_work);
497	453	rnp->exp_need_flush = true;
498	454	}
499	455
500	456	/* Wait for workqueue jobs (if any) to complete. */
501		- rcu_for_each_leaf_node(rsp, rnp)
	457	+ rcu_for_each_leaf_node(rnp)
502	458	if (rnp->exp_need_flush)
503	459	flush_work(&rnp->rew.rew_work);
504	460	}
505	461
506		-static void synchronize_sched_expedited_wait(struct rcu_state *rsp)
	462	+/*
	463	+ * Wait for the expedited grace period to elapse, within time limit.
	464	+ * If the time limit is exceeded without the grace period elapsing,
	465	+ * return false, otherwise return true.
	466	+ */
	467	+static bool synchronize_rcu_expedited_wait_once(long tlimit)
	468	+{
	469	+ int t;
	470	+ struct rcu_node *rnp_root = rcu_get_root();
	471	+
	472	+ t = swait_event_timeout_exclusive(rcu_state.expedited_wq,
	473	+ sync_rcu_exp_done_unlocked(rnp_root),
	474	+ tlimit);
	475	+ // Workqueues should not be signaled.
	476	+ if (t > 0 \|\| sync_rcu_exp_done_unlocked(rnp_root))
	477	+ return true;
	478	+ WARN_ON(t < 0); /* workqueues should not be signaled. */
	479	+ return false;
	480	+}
	481	+
	482	+/*
	483	+ * Wait for the expedited grace period to elapse, issuing any needed
	484	+ * RCU CPU stall warnings along the way.
	485	+ */
	486	+static void synchronize_rcu_expedited_wait(void)
507	487	{
508	488	int cpu;
	489	+ unsigned long j;
509	490	unsigned long jiffies_stall;
510	491	unsigned long jiffies_start;
511	492	unsigned long mask;
512	493	int ndetected;
	494	+ struct rcu_data *rdp;
513	495	struct rcu_node *rnp;
514		- struct rcu_node *rnp_root = rcu_get_root(rsp);
515		- int ret;
	496	+ struct rcu_node *rnp_root = rcu_get_root();
516	497
517		- trace_rcu_exp_grace_period(rsp->name, rcu_exp_gp_seq_endval(rsp), TPS("startwait"));
	498	+ trace_rcu_exp_grace_period(rcu_state.name, rcu_exp_gp_seq_endval(), TPS("startwait"));
518	499	jiffies_stall = rcu_jiffies_till_stall_check();
519	500	jiffies_start = jiffies;
	501	+ if (tick_nohz_full_enabled() && rcu_inkernel_boot_has_ended()) {
	502	+ if (synchronize_rcu_expedited_wait_once(1))
	503	+ return;
	504	+ rcu_for_each_leaf_node(rnp) {
	505	+ for_each_leaf_node_cpu_mask(rnp, cpu, rnp->expmask) {
	506	+ rdp = per_cpu_ptr(&rcu_data, cpu);
	507	+ if (rdp->rcu_forced_tick_exp)
	508	+ continue;
	509	+ rdp->rcu_forced_tick_exp = true;
	510	+ preempt_disable();
	511	+ if (cpu_online(cpu))
	512	+ tick_dep_set_cpu(cpu, TICK_DEP_BIT_RCU_EXP);
	513	+ preempt_enable();
	514	+ }
	515	+ }
	516	+ j = READ_ONCE(jiffies_till_first_fqs);
	517	+ if (synchronize_rcu_expedited_wait_once(j + HZ))
	518	+ return;
	519	+ WARN_ON_ONCE(IS_ENABLED(CONFIG_PREEMPT_RT));
	520	+ }
520	521
521	522	for (;;) {
522		- ret = swait_event_timeout_exclusive(
523		- rsp->expedited_wq,
524		- sync_rcu_preempt_exp_done_unlocked(rnp_root),
525		- jiffies_stall);
526		- if (ret > 0 \|\| sync_rcu_preempt_exp_done_unlocked(rnp_root))
	523	+ if (synchronize_rcu_expedited_wait_once(jiffies_stall))
527	524	return;
528		- WARN_ON(ret < 0); /* workqueues should not be signaled. */
529		- if (rcu_cpu_stall_suppress)
	525	+ if (rcu_stall_is_suppressed())
530	526	continue;
531	527	panic_on_rcu_stall();
	528	+ trace_rcu_stall_warning(rcu_state.name, TPS("ExpeditedStall"));
532	529	pr_err("INFO: %s detected expedited stalls on CPUs/tasks: {",
533		- rsp->name);
	530	+ rcu_state.name);
534	531	ndetected = 0;
535		- rcu_for_each_leaf_node(rsp, rnp) {
	532	+ rcu_for_each_leaf_node(rnp) {
536	533	ndetected += rcu_print_task_exp_stall(rnp);
537	534	for_each_leaf_node_possible_cpu(rnp, cpu) {
538	535	struct rcu_data *rdp;
539	536
540	537	mask = leaf_node_cpu_bit(rnp, cpu);
541		- if (!(rnp->expmask & mask))
	538	+ if (!(READ_ONCE(rnp->expmask) & mask))
542	539	continue;
543	540	ndetected++;
544		- rdp = per_cpu_ptr(rsp->rda, cpu);
	541	+ rdp = per_cpu_ptr(&rcu_data, cpu);
545	542	pr_cont(" %d-%c%c%c", cpu,
546	543	"O."[!!cpu_online(cpu)],
547	544	"o."[!!(rdp->grpmask & rnp->expmaskinit)],
..	..	@@ -549,28 +546,31 @@
549	546	}
550	547	}
551	548	pr_cont(" } %lu jiffies s: %lu root: %#lx/%c\n",
552		- jiffies - jiffies_start, rsp->expedited_sequence,
553		- rnp_root->expmask, ".T"[!!rnp_root->exp_tasks]);
	549	+ jiffies - jiffies_start, rcu_state.expedited_sequence,
	550	+ data_race(rnp_root->expmask),
	551	+ ".T"[!!data_race(rnp_root->exp_tasks)]);
554	552	if (ndetected) {
555	553	pr_err("blocking rcu_node structures:");
556		- rcu_for_each_node_breadth_first(rsp, rnp) {
	554	+ rcu_for_each_node_breadth_first(rnp) {
557	555	if (rnp == rnp_root)
558	556	continue; /* printed unconditionally */
559		- if (sync_rcu_preempt_exp_done_unlocked(rnp))
	557	+ if (sync_rcu_exp_done_unlocked(rnp))
560	558	continue;
561	559	pr_cont(" l=%u:%d-%d:%#lx/%c",
562	560	rnp->level, rnp->grplo, rnp->grphi,
563		- rnp->expmask,
564		- ".T"[!!rnp->exp_tasks]);
	561	+ data_race(rnp->expmask),
	562	+ ".T"[!!data_race(rnp->exp_tasks)]);
565	563	}
566	564	pr_cont("\n");
567	565	}
568		- rcu_for_each_leaf_node(rsp, rnp) {
	566	+ rcu_for_each_leaf_node(rnp) {
569	567	for_each_leaf_node_possible_cpu(rnp, cpu) {
570	568	mask = leaf_node_cpu_bit(rnp, cpu);
571		- if (!(rnp->expmask & mask))
	569	+ if (!(READ_ONCE(rnp->expmask) & mask))
572	570	continue;
	571	+ preempt_disable(); // For smp_processor_id() in dump_cpu_task().
573	572	dump_cpu_task(cpu);
	573	+ preempt_enable();
574	574	}
575	575	}
576	576	jiffies_stall = 3 * rcu_jiffies_till_stall_check() + 3;
..	..	@@ -583,47 +583,45 @@
583	583	* grace period. Also update all the ->exp_seq_rq counters as needed
584	584	* in order to avoid counter-wrap problems.
585	585	*/
586		-static void rcu_exp_wait_wake(struct rcu_state *rsp, unsigned long s)
	586	+static void rcu_exp_wait_wake(unsigned long s)
587	587	{
588	588	struct rcu_node *rnp;
589	589
590		- synchronize_sched_expedited_wait(rsp);
591		- rcu_exp_gp_seq_end(rsp);
592		- trace_rcu_exp_grace_period(rsp->name, s, TPS("end"));
	590	+ synchronize_rcu_expedited_wait();
593	591
594		- /*
595		- * Switch over to wakeup mode, allowing the next GP, but -only- the
596		- * next GP, to proceed.
597		- */
598		- mutex_lock(&rsp->exp_wake_mutex);
	592	+ // Switch over to wakeup mode, allowing the next GP to proceed.
	593	+ // End the previous grace period only after acquiring the mutex
	594	+ // to ensure that only one GP runs concurrently with wakeups.
	595	+ mutex_lock(&rcu_state.exp_wake_mutex);
	596	+ rcu_exp_gp_seq_end();
	597	+ trace_rcu_exp_grace_period(rcu_state.name, s, TPS("end"));
599	598
600		- rcu_for_each_node_breadth_first(rsp, rnp) {
	599	+ rcu_for_each_node_breadth_first(rnp) {
601	600	if (ULONG_CMP_LT(READ_ONCE(rnp->exp_seq_rq), s)) {
602	601	spin_lock(&rnp->exp_lock);
603	602	/* Recheck, avoid hang in case someone just arrived. */
604	603	if (ULONG_CMP_LT(rnp->exp_seq_rq, s))
605		- rnp->exp_seq_rq = s;
	604	+ WRITE_ONCE(rnp->exp_seq_rq, s);
606	605	spin_unlock(&rnp->exp_lock);
607	606	}
608	607	smp_mb(); /* All above changes before wakeup. */
609	608	wake_up_all(&rnp->exp_wq[rcu_seq_ctr(s) & 0x3]);
610	609	}
611		- trace_rcu_exp_grace_period(rsp->name, s, TPS("endwake"));
612		- mutex_unlock(&rsp->exp_wake_mutex);
	610	+ trace_rcu_exp_grace_period(rcu_state.name, s, TPS("endwake"));
	611	+ mutex_unlock(&rcu_state.exp_wake_mutex);
613	612	}
614	613
615	614	/*
616	615	* Common code to drive an expedited grace period forward, used by
617	616	* workqueues and mid-boot-time tasks.
618	617	*/
619		-static void rcu_exp_sel_wait_wake(struct rcu_state *rsp,
620		- smp_call_func_t func, unsigned long s)
	618	+static void rcu_exp_sel_wait_wake(unsigned long s)
621	619	{
622	620	/* Initialize the rcu_node tree in preparation for the wait. */
623		- sync_rcu_exp_select_cpus(rsp, func);
	621	+ sync_rcu_exp_select_cpus();
624	622
625	623	/* Wait and clean up, including waking everyone. */
626		- rcu_exp_wait_wake(rsp, s);
	624	+ rcu_exp_wait_wake(s);
627	625	}
628	626
629	627	/*
..	..	@@ -634,88 +632,8 @@
634	632	struct rcu_exp_work *rewp;
635	633
636	634	rewp = container_of(wp, struct rcu_exp_work, rew_work);
637		- rcu_exp_sel_wait_wake(rewp->rew_rsp, rewp->rew_func, rewp->rew_s);
	635	+ rcu_exp_sel_wait_wake(rewp->rew_s);
638	636	}
639		-
640		-/*
641		- * Given an rcu_state pointer and a smp_call_function() handler, kick
642		- * off the specified flavor of expedited grace period.
643		- */
644		-static void _synchronize_rcu_expedited(struct rcu_state *rsp,
645		- smp_call_func_t func)
646		-{
647		- struct rcu_data *rdp;
648		- struct rcu_exp_work rew;
649		- struct rcu_node *rnp;
650		- unsigned long s;
651		-
652		- /* If expedited grace periods are prohibited, fall back to normal. */
653		- if (rcu_gp_is_normal()) {
654		- wait_rcu_gp(rsp->call);
655		- return;
656		- }
657		-
658		- /* Take a snapshot of the sequence number. */
659		- s = rcu_exp_gp_seq_snap(rsp);
660		- if (exp_funnel_lock(rsp, s))
661		- return; /* Someone else did our work for us. */
662		-
663		- /* Ensure that load happens before action based on it. */
664		- if (unlikely(rcu_scheduler_active == RCU_SCHEDULER_INIT)) {
665		- /* Direct call during scheduler init and early_initcalls(). */
666		- rcu_exp_sel_wait_wake(rsp, func, s);
667		- } else {
668		- /* Marshall arguments & schedule the expedited grace period. */
669		- rew.rew_func = func;
670		- rew.rew_rsp = rsp;
671		- rew.rew_s = s;
672		- INIT_WORK_ONSTACK(&rew.rew_work, wait_rcu_exp_gp);
673		- queue_work(rcu_gp_wq, &rew.rew_work);
674		- }
675		-
676		- /* Wait for expedited grace period to complete. */
677		- rdp = per_cpu_ptr(rsp->rda, raw_smp_processor_id());
678		- rnp = rcu_get_root(rsp);
679		- wait_event(rnp->exp_wq[rcu_seq_ctr(s) & 0x3],
680		- sync_exp_work_done(rsp, s));
681		- smp_mb(); /* Workqueue actions happen before return. */
682		-
683		- /* Let the next expedited grace period start. */
684		- mutex_unlock(&rsp->exp_mutex);
685		-}
686		-
687		-/**
688		- * synchronize_sched_expedited - Brute-force RCU-sched grace period
689		- *
690		- * Wait for an RCU-sched grace period to elapse, but use a "big hammer"
691		- * approach to force the grace period to end quickly. This consumes
692		- * significant time on all CPUs and is unfriendly to real-time workloads,
693		- * so is thus not recommended for any sort of common-case code. In fact,
694		- * if you are using synchronize_sched_expedited() in a loop, please
695		- * restructure your code to batch your updates, and then use a single
696		- * synchronize_sched() instead.
697		- *
698		- * This implementation can be thought of as an application of sequence
699		- * locking to expedited grace periods, but using the sequence counter to
700		- * determine when someone else has already done the work instead of for
701		- * retrying readers.
702		- */
703		-void synchronize_sched_expedited(void)
704		-{
705		- struct rcu_state *rsp = &rcu_sched_state;
706		-
707		- RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) \|\|
708		- lock_is_held(&rcu_lock_map) \|\|
709		- lock_is_held(&rcu_sched_lock_map),
710		- "Illegal synchronize_sched_expedited() in RCU read-side critical section");
711		-
712		- /* If only one CPU, this is automatically a grace period. */
713		- if (rcu_blocking_is_gp())
714		- return;
715		-
716		- _synchronize_rcu_expedited(rsp, sync_sched_exp_handler);
717		-}
718		-EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
719	637
720	638	#ifdef CONFIG_PREEMPT_RCU
721	639
..	..	@@ -726,79 +644,234 @@
726	644	* ->expmask fields in the rcu_node tree. Otherwise, immediately
727	645	* report the quiescent state.
728	646	*/
729		-static void sync_rcu_exp_handler(void *info)
	647	+static void rcu_exp_handler(void *unused)
730	648	{
731		- struct rcu_data *rdp;
732		- struct rcu_state *rsp = info;
	649	+ int depth = rcu_preempt_depth();
	650	+ unsigned long flags;
	651	+ struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
	652	+ struct rcu_node *rnp = rdp->mynode;
733	653	struct task_struct *t = current;
734	654
735	655	/*
736		- * Within an RCU read-side critical section, request that the next
737		- * rcu_read_unlock() report. Unless this RCU read-side critical
738		- * section has already blocked, in which case it is already set
739		- * up for the expedited grace period to wait on it.
	656	+ * First, the common case of not being in an RCU read-side
	657	+ * critical section. If also enabled or idle, immediately
	658	+ * report the quiescent state, otherwise defer.
740	659	*/
741		- if (t->rcu_read_lock_nesting > 0 &&
742		- !t->rcu_read_unlock_special.b.blocked) {
743		- t->rcu_read_unlock_special.b.exp_need_qs = true;
	660	+ if (!depth) {
	661	+ if (!(preempt_count() & (PREEMPT_MASK \| SOFTIRQ_MASK)) \|\|
	662	+ rcu_dynticks_curr_cpu_in_eqs()) {
	663	+ rcu_report_exp_rdp(rdp);
	664	+ } else {
	665	+ rdp->exp_deferred_qs = true;
	666	+ set_tsk_need_resched(t);
	667	+ set_preempt_need_resched();
	668	+ }
744	669	return;
745	670	}
746	671
747	672	/*
748		- * We are either exiting an RCU read-side critical section (negative
749		- * values of t->rcu_read_lock_nesting) or are not in one at all
750		- * (zero value of t->rcu_read_lock_nesting). Or we are in an RCU
751		- * read-side critical section that blocked before this expedited
752		- * grace period started. Either way, we can immediately report
753		- * the quiescent state.
	673	+ * Second, the less-common case of being in an RCU read-side
	674	+ * critical section. In this case we can count on a future
	675	+ * rcu_read_unlock(). However, this rcu_read_unlock() might
	676	+ * execute on some other CPU, but in that case there will be
	677	+ * a future context switch. Either way, if the expedited
	678	+ * grace period is still waiting on this CPU, set ->deferred_qs
	679	+ * so that the eventual quiescent state will be reported.
	680	+ * Note that there is a large group of race conditions that
	681	+ * can have caused this quiescent state to already have been
	682	+ * reported, so we really do need to check ->expmask.
754	683	*/
755		- rdp = this_cpu_ptr(rsp->rda);
756		- rcu_report_exp_rdp(rsp, rdp, true);
	684	+ if (depth > 0) {
	685	+ raw_spin_lock_irqsave_rcu_node(rnp, flags);
	686	+ if (rnp->expmask & rdp->grpmask) {
	687	+ rdp->exp_deferred_qs = true;
	688	+ t->rcu_read_unlock_special.b.exp_hint = true;
	689	+ }
	690	+ raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
	691	+ return;
	692	+ }
	693	+
	694	+ // Finally, negative nesting depth should not happen.
	695	+ WARN_ON_ONCE(1);
757	696	}
	697	+
	698	+/* PREEMPTION=y, so no PREEMPTION=n expedited grace period to clean up after. */
	699	+static void sync_sched_exp_online_cleanup(int cpu)
	700	+{
	701	+}
	702	+
	703	+/*
	704	+ * Scan the current list of tasks blocked within RCU read-side critical
	705	+ * sections, printing out the tid of each that is blocking the current
	706	+ * expedited grace period.
	707	+ */
	708	+static int rcu_print_task_exp_stall(struct rcu_node *rnp)
	709	+{
	710	+ unsigned long flags;
	711	+ int ndetected = 0;
	712	+ struct task_struct *t;
	713	+
	714	+ raw_spin_lock_irqsave_rcu_node(rnp, flags);
	715	+ if (!rnp->exp_tasks) {
	716	+ raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
	717	+ return 0;
	718	+ }
	719	+ t = list_entry(rnp->exp_tasks->prev,
	720	+ struct task_struct, rcu_node_entry);
	721	+ list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
	722	+ pr_cont(" P%d", t->pid);
	723	+ ndetected++;
	724	+ }
	725	+ raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
	726	+ return ndetected;
	727	+}
	728	+
	729	+#else /* #ifdef CONFIG_PREEMPT_RCU */
	730	+
	731	+/* Request an expedited quiescent state. */
	732	+static void rcu_exp_need_qs(void)
	733	+{
	734	+ __this_cpu_write(rcu_data.cpu_no_qs.b.exp, true);
	735	+ /* Store .exp before .rcu_urgent_qs. */
	736	+ smp_store_release(this_cpu_ptr(&rcu_data.rcu_urgent_qs), true);
	737	+ set_tsk_need_resched(current);
	738	+ set_preempt_need_resched();
	739	+}
	740	+
	741	+/* Invoked on each online non-idle CPU for expedited quiescent state. */
	742	+static void rcu_exp_handler(void *unused)
	743	+{
	744	+ struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
	745	+ struct rcu_node *rnp = rdp->mynode;
	746	+
	747	+ if (!(READ_ONCE(rnp->expmask) & rdp->grpmask) \|\|
	748	+ __this_cpu_read(rcu_data.cpu_no_qs.b.exp))
	749	+ return;
	750	+ if (rcu_is_cpu_rrupt_from_idle()) {
	751	+ rcu_report_exp_rdp(this_cpu_ptr(&rcu_data));
	752	+ return;
	753	+ }
	754	+ rcu_exp_need_qs();
	755	+}
	756	+
	757	+/* Send IPI for expedited cleanup if needed at end of CPU-hotplug operation. */
	758	+static void sync_sched_exp_online_cleanup(int cpu)
	759	+{
	760	+ unsigned long flags;
	761	+ int my_cpu;
	762	+ struct rcu_data *rdp;
	763	+ int ret;
	764	+ struct rcu_node *rnp;
	765	+
	766	+ rdp = per_cpu_ptr(&rcu_data, cpu);
	767	+ rnp = rdp->mynode;
	768	+ my_cpu = get_cpu();
	769	+ /* Quiescent state either not needed or already requested, leave. */
	770	+ if (!(READ_ONCE(rnp->expmask) & rdp->grpmask) \|\|
	771	+ rdp->cpu_no_qs.b.exp) {
	772	+ put_cpu();
	773	+ return;
	774	+ }
	775	+ /* Quiescent state needed on current CPU, so set it up locally. */
	776	+ if (my_cpu == cpu) {
	777	+ local_irq_save(flags);
	778	+ rcu_exp_need_qs();
	779	+ local_irq_restore(flags);
	780	+ put_cpu();
	781	+ return;
	782	+ }
	783	+ /* Quiescent state needed on some other CPU, send IPI. */
	784	+ ret = smp_call_function_single(cpu, rcu_exp_handler, NULL, 0);
	785	+ put_cpu();
	786	+ WARN_ON_ONCE(ret);
	787	+}
	788	+
	789	+/*
	790	+ * Because preemptible RCU does not exist, we never have to check for
	791	+ * tasks blocked within RCU read-side critical sections that are
	792	+ * blocking the current expedited grace period.
	793	+ */
	794	+static int rcu_print_task_exp_stall(struct rcu_node *rnp)
	795	+{
	796	+ return 0;
	797	+}
	798	+
	799	+#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
758	800
759	801	/**
760	802	* synchronize_rcu_expedited - Brute-force RCU grace period
761	803	*
762		- * Wait for an RCU-preempt grace period, but expedite it. The basic
763		- * idea is to IPI all non-idle non-nohz online CPUs. The IPI handler
764		- * checks whether the CPU is in an RCU-preempt critical section, and
765		- * if so, it sets a flag that causes the outermost rcu_read_unlock()
766		- * to report the quiescent state. On the other hand, if the CPU is
767		- * not in an RCU read-side critical section, the IPI handler reports
768		- * the quiescent state immediately.
	804	+ * Wait for an RCU grace period, but expedite it. The basic idea is to
	805	+ * IPI all non-idle non-nohz online CPUs. The IPI handler checks whether
	806	+ * the CPU is in an RCU critical section, and if so, it sets a flag that
	807	+ * causes the outermost rcu_read_unlock() to report the quiescent state
	808	+ * for RCU-preempt or asks the scheduler for help for RCU-sched. On the
	809	+ * other hand, if the CPU is not in an RCU read-side critical section,
	810	+ * the IPI handler reports the quiescent state immediately.
769	811	*
770		- * Although this is a greate improvement over previous expedited
	812	+ * Although this is a great improvement over previous expedited
771	813	* implementations, it is still unfriendly to real-time workloads, so is
772	814	* thus not recommended for any sort of common-case code. In fact, if
773	815	* you are using synchronize_rcu_expedited() in a loop, please restructure
774		- * your code to batch your updates, and then Use a single synchronize_rcu()
	816	+ * your code to batch your updates, and then use a single synchronize_rcu()
775	817	* instead.
	818	+ *
	819	+ * This has the same semantics as (but is more brutal than) synchronize_rcu().
776	820	*/
777	821	void synchronize_rcu_expedited(void)
778	822	{
779		- struct rcu_state *rsp = rcu_state_p;
	823	+ bool no_wq;
	824	+ struct rcu_exp_work rew;
	825	+ struct rcu_node *rnp;
	826	+ unsigned long s;
780	827
781	828	RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) \|\|
782	829	lock_is_held(&rcu_lock_map) \|\|
783	830	lock_is_held(&rcu_sched_lock_map),
784	831	"Illegal synchronize_rcu_expedited() in RCU read-side critical section");
785	832
786		- if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE)
	833	+ /* Is the state is such that the call is a grace period? */
	834	+ if (rcu_blocking_is_gp())
787	835	return;
788		- _synchronize_rcu_expedited(rsp, sync_rcu_exp_handler);
	836	+
	837	+ /* If expedited grace periods are prohibited, fall back to normal. */
	838	+ if (rcu_gp_is_normal()) {
	839	+ wait_rcu_gp(call_rcu);
	840	+ return;
	841	+ }
	842	+
	843	+ /* Take a snapshot of the sequence number. */
	844	+ s = rcu_exp_gp_seq_snap();
	845	+ if (exp_funnel_lock(s))
	846	+ return; /* Someone else did our work for us. */
	847	+
	848	+ /* Don't use workqueue during boot or from an incoming CPU. */
	849	+ preempt_disable();
	850	+ no_wq = rcu_scheduler_active == RCU_SCHEDULER_INIT \|\|
	851	+ !cpumask_test_cpu(smp_processor_id(), cpu_active_mask);
	852	+ preempt_enable();
	853	+
	854	+ /* Ensure that load happens before action based on it. */
	855	+ if (unlikely(no_wq)) {
	856	+ /* Direct call for scheduler init, early_initcall()s, and incoming CPUs. */
	857	+ rcu_exp_sel_wait_wake(s);
	858	+ } else {
	859	+ /* Marshall arguments & schedule the expedited grace period. */
	860	+ rew.rew_s = s;
	861	+ INIT_WORK_ONSTACK(&rew.rew_work, wait_rcu_exp_gp);
	862	+ queue_work(rcu_gp_wq, &rew.rew_work);
	863	+ }
	864	+
	865	+ /* Wait for expedited grace period to complete. */
	866	+ rnp = rcu_get_root();
	867	+ wait_event(rnp->exp_wq[rcu_seq_ctr(s) & 0x3],
	868	+ sync_exp_work_done(s));
	869	+ smp_mb(); /* Workqueue actions happen before return. */
	870	+
	871	+ /* Let the next expedited grace period start. */
	872	+ mutex_unlock(&rcu_state.exp_mutex);
	873	+
	874	+ if (likely(!no_wq))
	875	+ destroy_work_on_stack(&rew.rew_work);
789	876	}
790	877	EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
791		-
792		-#else /* #ifdef CONFIG_PREEMPT_RCU */
793		-
794		-/*
795		- * Wait for an rcu-preempt grace period, but make it happen quickly.
796		- * But because preemptible RCU does not exist, map to rcu-sched.
797		- */
798		-void synchronize_rcu_expedited(void)
799		-{
800		- synchronize_sched_expedited();
801		-}
802		-EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
803		-
804		-#endif /* #else #ifdef CONFIG_PREEMPT_RCU */