Commit | Line | Data |
---|---|---|
64db4cff PM |
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 | |
87de1cfd PM |
15 | * along with this program; if not, you can access it online at |
16 | * http://www.gnu.org/licenses/gpl-2.0.html. | |
64db4cff PM |
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 - | |
a71fca58 | 28 | * Documentation/RCU |
64db4cff | 29 | */ |
a7538352 JP |
30 | |
31 | #define pr_fmt(fmt) "rcu: " fmt | |
32 | ||
64db4cff PM |
33 | #include <linux/types.h> |
34 | #include <linux/kernel.h> | |
35 | #include <linux/init.h> | |
36 | #include <linux/spinlock.h> | |
37 | #include <linux/smp.h> | |
f9411ebe | 38 | #include <linux/rcupdate_wait.h> |
64db4cff PM |
39 | #include <linux/interrupt.h> |
40 | #include <linux/sched.h> | |
b17b0153 | 41 | #include <linux/sched/debug.h> |
c1dc0b9c | 42 | #include <linux/nmi.h> |
8826f3b0 | 43 | #include <linux/atomic.h> |
64db4cff | 44 | #include <linux/bitops.h> |
9984de1a | 45 | #include <linux/export.h> |
64db4cff PM |
46 | #include <linux/completion.h> |
47 | #include <linux/moduleparam.h> | |
48 | #include <linux/percpu.h> | |
49 | #include <linux/notifier.h> | |
50 | #include <linux/cpu.h> | |
51 | #include <linux/mutex.h> | |
52 | #include <linux/time.h> | |
bbad9379 | 53 | #include <linux/kernel_stat.h> |
a26ac245 PM |
54 | #include <linux/wait.h> |
55 | #include <linux/kthread.h> | |
ae7e81c0 | 56 | #include <uapi/linux/sched/types.h> |
268bb0ce | 57 | #include <linux/prefetch.h> |
3d3b7db0 PM |
58 | #include <linux/delay.h> |
59 | #include <linux/stop_machine.h> | |
661a85dc | 60 | #include <linux/random.h> |
af658dca | 61 | #include <linux/trace_events.h> |
d1d74d14 | 62 | #include <linux/suspend.h> |
a278d471 | 63 | #include <linux/ftrace.h> |
d3052109 | 64 | #include <linux/tick.h> |
64db4cff | 65 | |
4102adab | 66 | #include "tree.h" |
29c00b4a | 67 | #include "rcu.h" |
9f77da9f | 68 | |
4102adab PM |
69 | #ifdef MODULE_PARAM_PREFIX |
70 | #undef MODULE_PARAM_PREFIX | |
71 | #endif | |
72 | #define MODULE_PARAM_PREFIX "rcutree." | |
73 | ||
64db4cff PM |
74 | /* Data structures. */ |
75 | ||
f7f7bac9 | 76 | /* |
dc5a4f29 PM |
77 | * Steal a bit from the bottom of ->dynticks for idle entry/exit |
78 | * control. Initially this is for TLB flushing. | |
f7f7bac9 | 79 | */ |
dc5a4f29 PM |
80 | #define RCU_DYNTICK_CTRL_MASK 0x1 |
81 | #define RCU_DYNTICK_CTRL_CTR (RCU_DYNTICK_CTRL_MASK + 1) | |
82 | #ifndef rcu_eqs_special_exit | |
83 | #define rcu_eqs_special_exit() do { } while (0) | |
a8a29b3b AB |
84 | #endif |
85 | ||
4c5273bf PM |
86 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, rcu_data) = { |
87 | .dynticks_nesting = 1, | |
88 | .dynticks_nmi_nesting = DYNTICK_IRQ_NONIDLE, | |
dc5a4f29 | 89 | .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR), |
4c5273bf | 90 | }; |
358be2d3 PM |
91 | struct rcu_state rcu_state = { |
92 | .level = { &rcu_state.node[0] }, | |
358be2d3 PM |
93 | .gp_state = RCU_GP_IDLE, |
94 | .gp_seq = (0UL - 300UL) << RCU_SEQ_CTR_SHIFT, | |
95 | .barrier_mutex = __MUTEX_INITIALIZER(rcu_state.barrier_mutex), | |
96 | .name = RCU_NAME, | |
97 | .abbr = RCU_ABBR, | |
98 | .exp_mutex = __MUTEX_INITIALIZER(rcu_state.exp_mutex), | |
99 | .exp_wake_mutex = __MUTEX_INITIALIZER(rcu_state.exp_wake_mutex), | |
894d45bb | 100 | .ofl_lock = __RAW_SPIN_LOCK_UNLOCKED(rcu_state.ofl_lock), |
358be2d3 | 101 | }; |
27f4d280 | 102 | |
a3dc2948 PM |
103 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
104 | static bool dump_tree; | |
105 | module_param(dump_tree, bool, 0444); | |
7fa27001 PM |
106 | /* Control rcu_node-tree auto-balancing at boot time. */ |
107 | static bool rcu_fanout_exact; | |
108 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
109 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
110 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 111 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 112 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 | 113 | /* Number of rcu_nodes at specified level. */ |
e95d68d2 | 114 | int num_rcu_lvl[] = NUM_RCU_LVL_INIT; |
f885b7f2 | 115 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
088e9d25 DBO |
116 | /* panic() on RCU Stall sysctl. */ |
117 | int sysctl_panic_on_rcu_stall __read_mostly; | |
f885b7f2 | 118 | |
b0d30417 | 119 | /* |
52d7e48b PM |
120 | * The rcu_scheduler_active variable is initialized to the value |
121 | * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the | |
122 | * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE, | |
123 | * RCU can assume that there is but one task, allowing RCU to (for example) | |
0d95092c | 124 | * optimize synchronize_rcu() to a simple barrier(). When this variable |
52d7e48b PM |
125 | * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required |
126 | * to detect real grace periods. This variable is also used to suppress | |
127 | * boot-time false positives from lockdep-RCU error checking. Finally, it | |
128 | * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU | |
129 | * is fully initialized, including all of its kthreads having been spawned. | |
b0d30417 | 130 | */ |
bbad9379 PM |
131 | int rcu_scheduler_active __read_mostly; |
132 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
133 | ||
b0d30417 PM |
134 | /* |
135 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
136 | * during the early_initcall() processing, which is after the scheduler | |
137 | * is capable of creating new tasks. So RCU processing (for example, | |
138 | * creating tasks for RCU priority boosting) must be delayed until after | |
139 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
140 | * currently delay invocation of any RCU callbacks until after this point. | |
141 | * | |
142 | * It might later prove better for people registering RCU callbacks during | |
143 | * early boot to take responsibility for these callbacks, but one step at | |
144 | * a time. | |
145 | */ | |
146 | static int rcu_scheduler_fully_active __read_mostly; | |
147 | ||
b50912d0 PM |
148 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
149 | unsigned long gps, unsigned long flags); | |
0aa04b05 PM |
150 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
151 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 152 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 | 153 | static void invoke_rcu_core(void); |
aff4e9ed | 154 | static void invoke_rcu_callbacks(struct rcu_data *rdp); |
63d4c8c9 | 155 | static void rcu_report_exp_rdp(struct rcu_data *rdp); |
3549c2bc | 156 | static void sync_sched_exp_online_cleanup(int cpu); |
a26ac245 | 157 | |
a94844b2 | 158 | /* rcuc/rcub kthread realtime priority */ |
26730f55 | 159 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; |
a94844b2 PM |
160 | module_param(kthread_prio, int, 0644); |
161 | ||
8d7dc928 | 162 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd | 163 | |
90040c9e PM |
164 | static int gp_preinit_delay; |
165 | module_param(gp_preinit_delay, int, 0444); | |
166 | static int gp_init_delay; | |
167 | module_param(gp_init_delay, int, 0444); | |
168 | static int gp_cleanup_delay; | |
169 | module_param(gp_cleanup_delay, int, 0444); | |
0f41c0dd | 170 | |
4cf439a2 | 171 | /* Retrieve RCU kthreads priority for rcutorture */ |
4babd855 JFG |
172 | int rcu_get_gp_kthreads_prio(void) |
173 | { | |
174 | return kthread_prio; | |
175 | } | |
176 | EXPORT_SYMBOL_GPL(rcu_get_gp_kthreads_prio); | |
177 | ||
eab128e8 PM |
178 | /* |
179 | * Number of grace periods between delays, normalized by the duration of | |
bfd090be | 180 | * the delay. The longer the delay, the more the grace periods between |
eab128e8 PM |
181 | * each delay. The reason for this normalization is that it means that, |
182 | * for non-zero delays, the overall slowdown of grace periods is constant | |
183 | * regardless of the duration of the delay. This arrangement balances | |
184 | * the need for long delays to increase some race probabilities with the | |
185 | * need for fast grace periods to increase other race probabilities. | |
186 | */ | |
187 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */ | |
37745d28 | 188 | |
0aa04b05 PM |
189 | /* |
190 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
191 | * This will not be stable unless the rcu_node structure's ->lock is | |
192 | * held, but the bit corresponding to the current CPU will be stable | |
193 | * in most contexts. | |
194 | */ | |
195 | unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) | |
196 | { | |
7d0ae808 | 197 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
198 | } |
199 | ||
fc2219d4 | 200 | /* |
7d0ae808 | 201 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
202 | * permit this function to be invoked without holding the root rcu_node |
203 | * structure's ->lock, but of course results can be subject to change. | |
204 | */ | |
de8e8730 | 205 | static int rcu_gp_in_progress(void) |
fc2219d4 | 206 | { |
de8e8730 | 207 | return rcu_seq_state(rcu_seq_current(&rcu_state.gp_seq)); |
b1f77b05 IM |
208 | } |
209 | ||
903ee83d PM |
210 | /* |
211 | * Return the number of callbacks queued on the specified CPU. | |
212 | * Handles both the nocbs and normal cases. | |
213 | */ | |
214 | static long rcu_get_n_cbs_cpu(int cpu) | |
215 | { | |
216 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); | |
217 | ||
218 | if (rcu_segcblist_is_enabled(&rdp->cblist)) /* Online normal CPU? */ | |
219 | return rcu_segcblist_n_cbs(&rdp->cblist); | |
220 | return rcu_get_n_cbs_nocb_cpu(rdp); /* Works for offline, too. */ | |
221 | } | |
222 | ||
d28139c4 | 223 | void rcu_softirq_qs(void) |
b1f77b05 | 224 | { |
45975c7d | 225 | rcu_qs(); |
d28139c4 | 226 | rcu_preempt_deferred_qs(current); |
b1f77b05 | 227 | } |
64db4cff | 228 | |
2625d469 PM |
229 | /* |
230 | * Record entry into an extended quiescent state. This is only to be | |
231 | * called when not already in an extended quiescent state. | |
232 | */ | |
233 | static void rcu_dynticks_eqs_enter(void) | |
234 | { | |
dc5a4f29 | 235 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
b8c17e66 | 236 | int seq; |
2625d469 PM |
237 | |
238 | /* | |
b8c17e66 | 239 | * CPUs seeing atomic_add_return() must see prior RCU read-side |
2625d469 PM |
240 | * critical sections, and we also must force ordering with the |
241 | * next idle sojourn. | |
242 | */ | |
dc5a4f29 | 243 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks); |
b8c17e66 PM |
244 | /* Better be in an extended quiescent state! */ |
245 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
246 | (seq & RCU_DYNTICK_CTRL_CTR)); | |
247 | /* Better not have special action (TLB flush) pending! */ | |
248 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
249 | (seq & RCU_DYNTICK_CTRL_MASK)); | |
2625d469 PM |
250 | } |
251 | ||
252 | /* | |
253 | * Record exit from an extended quiescent state. This is only to be | |
254 | * called from an extended quiescent state. | |
255 | */ | |
256 | static void rcu_dynticks_eqs_exit(void) | |
257 | { | |
dc5a4f29 | 258 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
b8c17e66 | 259 | int seq; |
2625d469 PM |
260 | |
261 | /* | |
b8c17e66 | 262 | * CPUs seeing atomic_add_return() must see prior idle sojourns, |
2625d469 PM |
263 | * and we also must force ordering with the next RCU read-side |
264 | * critical section. | |
265 | */ | |
dc5a4f29 | 266 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks); |
b8c17e66 PM |
267 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
268 | !(seq & RCU_DYNTICK_CTRL_CTR)); | |
269 | if (seq & RCU_DYNTICK_CTRL_MASK) { | |
dc5a4f29 | 270 | atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdp->dynticks); |
b8c17e66 PM |
271 | smp_mb__after_atomic(); /* _exit after clearing mask. */ |
272 | /* Prefer duplicate flushes to losing a flush. */ | |
273 | rcu_eqs_special_exit(); | |
274 | } | |
2625d469 PM |
275 | } |
276 | ||
277 | /* | |
278 | * Reset the current CPU's ->dynticks counter to indicate that the | |
279 | * newly onlined CPU is no longer in an extended quiescent state. | |
280 | * This will either leave the counter unchanged, or increment it | |
281 | * to the next non-quiescent value. | |
282 | * | |
283 | * The non-atomic test/increment sequence works because the upper bits | |
284 | * of the ->dynticks counter are manipulated only by the corresponding CPU, | |
285 | * or when the corresponding CPU is offline. | |
286 | */ | |
287 | static void rcu_dynticks_eqs_online(void) | |
288 | { | |
dc5a4f29 | 289 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
2625d469 | 290 | |
dc5a4f29 | 291 | if (atomic_read(&rdp->dynticks) & RCU_DYNTICK_CTRL_CTR) |
2625d469 | 292 | return; |
dc5a4f29 | 293 | atomic_add(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks); |
2625d469 PM |
294 | } |
295 | ||
02a5c550 PM |
296 | /* |
297 | * Is the current CPU in an extended quiescent state? | |
298 | * | |
299 | * No ordering, as we are sampling CPU-local information. | |
300 | */ | |
301 | bool rcu_dynticks_curr_cpu_in_eqs(void) | |
302 | { | |
dc5a4f29 | 303 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
02a5c550 | 304 | |
dc5a4f29 | 305 | return !(atomic_read(&rdp->dynticks) & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
306 | } |
307 | ||
8b2f63ab PM |
308 | /* |
309 | * Snapshot the ->dynticks counter with full ordering so as to allow | |
310 | * stable comparison of this counter with past and future snapshots. | |
311 | */ | |
dc5a4f29 | 312 | int rcu_dynticks_snap(struct rcu_data *rdp) |
8b2f63ab | 313 | { |
dc5a4f29 | 314 | int snap = atomic_add_return(0, &rdp->dynticks); |
8b2f63ab | 315 | |
b8c17e66 | 316 | return snap & ~RCU_DYNTICK_CTRL_MASK; |
8b2f63ab PM |
317 | } |
318 | ||
02a5c550 PM |
319 | /* |
320 | * Return true if the snapshot returned from rcu_dynticks_snap() | |
321 | * indicates that RCU is in an extended quiescent state. | |
322 | */ | |
323 | static bool rcu_dynticks_in_eqs(int snap) | |
324 | { | |
b8c17e66 | 325 | return !(snap & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
326 | } |
327 | ||
328 | /* | |
dc5a4f29 | 329 | * Return true if the CPU corresponding to the specified rcu_data |
02a5c550 PM |
330 | * structure has spent some time in an extended quiescent state since |
331 | * rcu_dynticks_snap() returned the specified snapshot. | |
332 | */ | |
dc5a4f29 | 333 | static bool rcu_dynticks_in_eqs_since(struct rcu_data *rdp, int snap) |
02a5c550 | 334 | { |
dc5a4f29 | 335 | return snap != rcu_dynticks_snap(rdp); |
02a5c550 PM |
336 | } |
337 | ||
b8c17e66 PM |
338 | /* |
339 | * Set the special (bottom) bit of the specified CPU so that it | |
340 | * will take special action (such as flushing its TLB) on the | |
341 | * next exit from an extended quiescent state. Returns true if | |
342 | * the bit was successfully set, or false if the CPU was not in | |
343 | * an extended quiescent state. | |
344 | */ | |
345 | bool rcu_eqs_special_set(int cpu) | |
346 | { | |
347 | int old; | |
348 | int new; | |
dc5a4f29 | 349 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
b8c17e66 PM |
350 | |
351 | do { | |
dc5a4f29 | 352 | old = atomic_read(&rdp->dynticks); |
b8c17e66 PM |
353 | if (old & RCU_DYNTICK_CTRL_CTR) |
354 | return false; | |
355 | new = old | RCU_DYNTICK_CTRL_MASK; | |
dc5a4f29 | 356 | } while (atomic_cmpxchg(&rdp->dynticks, old, new) != old); |
b8c17e66 | 357 | return true; |
6563de9d | 358 | } |
5cd37193 | 359 | |
4a81e832 PM |
360 | /* |
361 | * Let the RCU core know that this CPU has gone through the scheduler, | |
362 | * which is a quiescent state. This is called when the need for a | |
363 | * quiescent state is urgent, so we burn an atomic operation and full | |
364 | * memory barriers to let the RCU core know about it, regardless of what | |
365 | * this CPU might (or might not) do in the near future. | |
366 | * | |
0f9be8ca | 367 | * We inform the RCU core by emulating a zero-duration dyntick-idle period. |
46a5d164 | 368 | * |
3b57a399 | 369 | * The caller must have disabled interrupts and must not be idle. |
4a81e832 | 370 | */ |
395a2f09 | 371 | static void __maybe_unused rcu_momentary_dyntick_idle(void) |
4a81e832 | 372 | { |
3b57a399 PM |
373 | int special; |
374 | ||
2dba13f0 | 375 | raw_cpu_write(rcu_data.rcu_need_heavy_qs, false); |
dc5a4f29 PM |
376 | special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR, |
377 | &this_cpu_ptr(&rcu_data)->dynticks); | |
3b57a399 PM |
378 | /* It is illegal to call this from idle state. */ |
379 | WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR)); | |
3e310098 | 380 | rcu_preempt_deferred_qs(current); |
4a81e832 PM |
381 | } |
382 | ||
45975c7d PM |
383 | /** |
384 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle | |
bb73c52b | 385 | * |
45975c7d PM |
386 | * If the current CPU is idle or running at a first-level (not nested) |
387 | * interrupt from idle, return true. The caller must have at least | |
388 | * disabled preemption. | |
5cd37193 | 389 | */ |
45975c7d | 390 | static int rcu_is_cpu_rrupt_from_idle(void) |
5cd37193 | 391 | { |
4c5273bf PM |
392 | return __this_cpu_read(rcu_data.dynticks_nesting) <= 0 && |
393 | __this_cpu_read(rcu_data.dynticks_nmi_nesting) <= 1; | |
5cd37193 | 394 | } |
5cd37193 | 395 | |
17c7798b PM |
396 | #define DEFAULT_RCU_BLIMIT 10 /* Maximum callbacks per rcu_do_batch. */ |
397 | static long blimit = DEFAULT_RCU_BLIMIT; | |
398 | #define DEFAULT_RCU_QHIMARK 10000 /* If this many pending, ignore blimit. */ | |
399 | static long qhimark = DEFAULT_RCU_QHIMARK; | |
400 | #define DEFAULT_RCU_QLOMARK 100 /* Once only this many pending, use blimit. */ | |
401 | static long qlowmark = DEFAULT_RCU_QLOMARK; | |
64db4cff | 402 | |
878d7439 ED |
403 | module_param(blimit, long, 0444); |
404 | module_param(qhimark, long, 0444); | |
405 | module_param(qlowmark, long, 0444); | |
3d76c082 | 406 | |
026ad283 PM |
407 | static ulong jiffies_till_first_fqs = ULONG_MAX; |
408 | static ulong jiffies_till_next_fqs = ULONG_MAX; | |
8c7c4829 | 409 | static bool rcu_kick_kthreads; |
d40011f6 | 410 | |
c06aed0e PM |
411 | /* |
412 | * How long the grace period must be before we start recruiting | |
413 | * quiescent-state help from rcu_note_context_switch(). | |
414 | */ | |
415 | static ulong jiffies_till_sched_qs = ULONG_MAX; | |
416 | module_param(jiffies_till_sched_qs, ulong, 0444); | |
417 | static ulong jiffies_to_sched_qs; /* Adjusted version of above if not default */ | |
418 | module_param(jiffies_to_sched_qs, ulong, 0444); /* Display only! */ | |
419 | ||
420 | /* | |
421 | * Make sure that we give the grace-period kthread time to detect any | |
422 | * idle CPUs before taking active measures to force quiescent states. | |
423 | * However, don't go below 100 milliseconds, adjusted upwards for really | |
424 | * large systems. | |
425 | */ | |
426 | static void adjust_jiffies_till_sched_qs(void) | |
427 | { | |
428 | unsigned long j; | |
429 | ||
430 | /* If jiffies_till_sched_qs was specified, respect the request. */ | |
431 | if (jiffies_till_sched_qs != ULONG_MAX) { | |
432 | WRITE_ONCE(jiffies_to_sched_qs, jiffies_till_sched_qs); | |
433 | return; | |
434 | } | |
435 | j = READ_ONCE(jiffies_till_first_fqs) + | |
436 | 2 * READ_ONCE(jiffies_till_next_fqs); | |
437 | if (j < HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV) | |
438 | j = HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
439 | pr_info("RCU calculated value of scheduler-enlistment delay is %ld jiffies.\n", j); | |
440 | WRITE_ONCE(jiffies_to_sched_qs, j); | |
441 | } | |
442 | ||
67abb96c BP |
443 | static int param_set_first_fqs_jiffies(const char *val, const struct kernel_param *kp) |
444 | { | |
445 | ulong j; | |
446 | int ret = kstrtoul(val, 0, &j); | |
447 | ||
c06aed0e | 448 | if (!ret) { |
67abb96c | 449 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : j); |
c06aed0e PM |
450 | adjust_jiffies_till_sched_qs(); |
451 | } | |
67abb96c BP |
452 | return ret; |
453 | } | |
454 | ||
455 | static int param_set_next_fqs_jiffies(const char *val, const struct kernel_param *kp) | |
456 | { | |
457 | ulong j; | |
458 | int ret = kstrtoul(val, 0, &j); | |
459 | ||
c06aed0e | 460 | if (!ret) { |
67abb96c | 461 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : (j ?: 1)); |
c06aed0e PM |
462 | adjust_jiffies_till_sched_qs(); |
463 | } | |
67abb96c BP |
464 | return ret; |
465 | } | |
466 | ||
467 | static struct kernel_param_ops first_fqs_jiffies_ops = { | |
468 | .set = param_set_first_fqs_jiffies, | |
469 | .get = param_get_ulong, | |
470 | }; | |
471 | ||
472 | static struct kernel_param_ops next_fqs_jiffies_ops = { | |
473 | .set = param_set_next_fqs_jiffies, | |
474 | .get = param_get_ulong, | |
475 | }; | |
476 | ||
477 | module_param_cb(jiffies_till_first_fqs, &first_fqs_jiffies_ops, &jiffies_till_first_fqs, 0644); | |
478 | module_param_cb(jiffies_till_next_fqs, &next_fqs_jiffies_ops, &jiffies_till_next_fqs, 0644); | |
8c7c4829 | 479 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 480 | |
8ff0b907 | 481 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)); |
e9ecb780 | 482 | static void force_quiescent_state(void); |
e3950ecd | 483 | static int rcu_pending(void); |
64db4cff PM |
484 | |
485 | /* | |
17ef2fe9 | 486 | * Return the number of RCU GPs completed thus far for debug & stats. |
64db4cff | 487 | */ |
17ef2fe9 | 488 | unsigned long rcu_get_gp_seq(void) |
917963d0 | 489 | { |
16fc9c60 | 490 | return READ_ONCE(rcu_state.gp_seq); |
917963d0 | 491 | } |
17ef2fe9 | 492 | EXPORT_SYMBOL_GPL(rcu_get_gp_seq); |
917963d0 | 493 | |
291783b8 PM |
494 | /* |
495 | * Return the number of RCU expedited batches completed thus far for | |
496 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
497 | * numbers mean idle. The value returned will thus be roughly double | |
498 | * the cumulative batches since boot. | |
499 | */ | |
500 | unsigned long rcu_exp_batches_completed(void) | |
501 | { | |
16fc9c60 | 502 | return rcu_state.expedited_sequence; |
291783b8 PM |
503 | } |
504 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
505 | ||
a381d757 ACB |
506 | /* |
507 | * Force a quiescent state. | |
508 | */ | |
509 | void rcu_force_quiescent_state(void) | |
510 | { | |
e9ecb780 | 511 | force_quiescent_state(); |
a381d757 ACB |
512 | } |
513 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
514 | ||
69196019 PM |
515 | /* |
516 | * Convert a ->gp_state value to a character string. | |
517 | */ | |
518 | static const char *gp_state_getname(short gs) | |
519 | { | |
520 | if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names)) | |
521 | return "???"; | |
522 | return gp_state_names[gs]; | |
523 | } | |
524 | ||
afea227f PM |
525 | /* |
526 | * Show the state of the grace-period kthreads. | |
527 | */ | |
528 | void show_rcu_gp_kthreads(void) | |
529 | { | |
47199a08 | 530 | int cpu; |
c669c014 | 531 | unsigned long j; |
47199a08 PM |
532 | struct rcu_data *rdp; |
533 | struct rcu_node *rnp; | |
afea227f | 534 | |
c669c014 PM |
535 | j = jiffies - READ_ONCE(rcu_state.gp_activity); |
536 | pr_info("%s: wait state: %s(%d) ->state: %#lx delta ->gp_activity %ld\n", | |
537 | rcu_state.name, gp_state_getname(rcu_state.gp_state), | |
538 | rcu_state.gp_state, rcu_state.gp_kthread->state, j); | |
b97d23c5 PM |
539 | rcu_for_each_node_breadth_first(rnp) { |
540 | if (ULONG_CMP_GE(rcu_state.gp_seq, rnp->gp_seq_needed)) | |
541 | continue; | |
542 | pr_info("\trcu_node %d:%d ->gp_seq %lu ->gp_seq_needed %lu\n", | |
543 | rnp->grplo, rnp->grphi, rnp->gp_seq, | |
544 | rnp->gp_seq_needed); | |
545 | if (!rcu_is_leaf_node(rnp)) | |
546 | continue; | |
547 | for_each_leaf_node_possible_cpu(rnp, cpu) { | |
548 | rdp = per_cpu_ptr(&rcu_data, cpu); | |
549 | if (rdp->gpwrap || | |
550 | ULONG_CMP_GE(rcu_state.gp_seq, | |
551 | rdp->gp_seq_needed)) | |
47199a08 | 552 | continue; |
b97d23c5 PM |
553 | pr_info("\tcpu %d ->gp_seq_needed %lu\n", |
554 | cpu, rdp->gp_seq_needed); | |
47199a08 | 555 | } |
afea227f | 556 | } |
b97d23c5 | 557 | /* sched_show_task(rcu_state.gp_kthread); */ |
afea227f PM |
558 | } |
559 | EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads); | |
560 | ||
ad0dc7f9 PM |
561 | /* |
562 | * Send along grace-period-related data for rcutorture diagnostics. | |
563 | */ | |
564 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
aebc8264 | 565 | unsigned long *gp_seq) |
ad0dc7f9 | 566 | { |
ad0dc7f9 PM |
567 | switch (test_type) { |
568 | case RCU_FLAVOR: | |
ad0dc7f9 | 569 | case RCU_BH_FLAVOR: |
ad0dc7f9 | 570 | case RCU_SCHED_FLAVOR: |
f7dd7d44 PM |
571 | *flags = READ_ONCE(rcu_state.gp_flags); |
572 | *gp_seq = rcu_seq_current(&rcu_state.gp_seq); | |
ad0dc7f9 PM |
573 | break; |
574 | default: | |
575 | break; | |
576 | } | |
ad0dc7f9 PM |
577 | } |
578 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
579 | ||
365187fb | 580 | /* |
49918a54 | 581 | * Return the root node of the rcu_state structure. |
365187fb | 582 | */ |
336a4f6c | 583 | static struct rcu_node *rcu_get_root(void) |
365187fb | 584 | { |
336a4f6c | 585 | return &rcu_state.node[0]; |
365187fb PM |
586 | } |
587 | ||
9b2e4f18 | 588 | /* |
215bba9f PM |
589 | * Enter an RCU extended quiescent state, which can be either the |
590 | * idle loop or adaptive-tickless usermode execution. | |
9b2e4f18 | 591 | * |
215bba9f PM |
592 | * We crowbar the ->dynticks_nmi_nesting field to zero to allow for |
593 | * the possibility of usermode upcalls having messed up our count | |
594 | * of interrupt nesting level during the prior busy period. | |
9b2e4f18 | 595 | */ |
215bba9f | 596 | static void rcu_eqs_enter(bool user) |
9b2e4f18 | 597 | { |
4c5273bf | 598 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
96d3fd0d | 599 | |
4c5273bf PM |
600 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting != DYNTICK_IRQ_NONIDLE); |
601 | WRITE_ONCE(rdp->dynticks_nmi_nesting, 0); | |
215bba9f | 602 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
4c5273bf PM |
603 | rdp->dynticks_nesting == 0); |
604 | if (rdp->dynticks_nesting != 1) { | |
605 | rdp->dynticks_nesting--; | |
215bba9f | 606 | return; |
9b2e4f18 | 607 | } |
96d3fd0d | 608 | |
b04db8e1 | 609 | lockdep_assert_irqs_disabled(); |
dc5a4f29 | 610 | trace_rcu_dyntick(TPS("Start"), rdp->dynticks_nesting, 0, rdp->dynticks); |
e68bbb26 | 611 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
b97d23c5 PM |
612 | rdp = this_cpu_ptr(&rcu_data); |
613 | do_nocb_deferred_wakeup(rdp); | |
198bbf81 | 614 | rcu_prepare_for_idle(); |
3e310098 | 615 | rcu_preempt_deferred_qs(current); |
4c5273bf | 616 | WRITE_ONCE(rdp->dynticks_nesting, 0); /* Avoid irq-access tearing. */ |
844ccdd7 | 617 | rcu_dynticks_eqs_enter(); |
176f8f7a | 618 | rcu_dynticks_task_enter(); |
64db4cff | 619 | } |
adf5091e FW |
620 | |
621 | /** | |
622 | * rcu_idle_enter - inform RCU that current CPU is entering idle | |
623 | * | |
624 | * Enter idle mode, in other words, -leave- the mode in which RCU | |
625 | * read-side critical sections can occur. (Though RCU read-side | |
626 | * critical sections can occur in irq handlers in idle, a possibility | |
627 | * handled by irq_enter() and irq_exit().) | |
628 | * | |
c0da313e PM |
629 | * If you add or remove a call to rcu_idle_enter(), be sure to test with |
630 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
631 | */ |
632 | void rcu_idle_enter(void) | |
633 | { | |
b04db8e1 | 634 | lockdep_assert_irqs_disabled(); |
cb349ca9 | 635 | rcu_eqs_enter(false); |
adf5091e | 636 | } |
64db4cff | 637 | |
d1ec4c34 | 638 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
639 | /** |
640 | * rcu_user_enter - inform RCU that we are resuming userspace. | |
641 | * | |
642 | * Enter RCU idle mode right before resuming userspace. No use of RCU | |
643 | * is permitted between this call and rcu_user_exit(). This way the | |
644 | * CPU doesn't need to maintain the tick for RCU maintenance purposes | |
645 | * when the CPU runs in userspace. | |
c0da313e PM |
646 | * |
647 | * If you add or remove a call to rcu_user_enter(), be sure to test with | |
648 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
649 | */ |
650 | void rcu_user_enter(void) | |
651 | { | |
b04db8e1 | 652 | lockdep_assert_irqs_disabled(); |
d4db30af | 653 | rcu_eqs_enter(true); |
adf5091e | 654 | } |
d1ec4c34 | 655 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 656 | |
cf7614e1 | 657 | /* |
fd581a91 | 658 | * If we are returning from the outermost NMI handler that interrupted an |
dc5a4f29 | 659 | * RCU-idle period, update rdp->dynticks and rdp->dynticks_nmi_nesting |
fd581a91 PM |
660 | * to let the RCU grace-period handling know that the CPU is back to |
661 | * being RCU-idle. | |
662 | * | |
cf7614e1 | 663 | * If you add or remove a call to rcu_nmi_exit_common(), be sure to test |
fd581a91 PM |
664 | * with CONFIG_RCU_EQS_DEBUG=y. |
665 | */ | |
cf7614e1 | 666 | static __always_inline void rcu_nmi_exit_common(bool irq) |
fd581a91 | 667 | { |
4c5273bf | 668 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
fd581a91 PM |
669 | |
670 | /* | |
671 | * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks. | |
672 | * (We are exiting an NMI handler, so RCU better be paying attention | |
673 | * to us!) | |
674 | */ | |
4c5273bf | 675 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting <= 0); |
fd581a91 PM |
676 | WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs()); |
677 | ||
678 | /* | |
679 | * If the nesting level is not 1, the CPU wasn't RCU-idle, so | |
680 | * leave it in non-RCU-idle state. | |
681 | */ | |
4c5273bf | 682 | if (rdp->dynticks_nmi_nesting != 1) { |
dc5a4f29 | 683 | trace_rcu_dyntick(TPS("--="), rdp->dynticks_nmi_nesting, rdp->dynticks_nmi_nesting - 2, rdp->dynticks); |
4c5273bf PM |
684 | WRITE_ONCE(rdp->dynticks_nmi_nesting, /* No store tearing. */ |
685 | rdp->dynticks_nmi_nesting - 2); | |
fd581a91 PM |
686 | return; |
687 | } | |
688 | ||
689 | /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */ | |
dc5a4f29 | 690 | trace_rcu_dyntick(TPS("Startirq"), rdp->dynticks_nmi_nesting, 0, rdp->dynticks); |
4c5273bf | 691 | WRITE_ONCE(rdp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */ |
cf7614e1 BP |
692 | |
693 | if (irq) | |
694 | rcu_prepare_for_idle(); | |
695 | ||
fd581a91 | 696 | rcu_dynticks_eqs_enter(); |
cf7614e1 BP |
697 | |
698 | if (irq) | |
699 | rcu_dynticks_task_enter(); | |
700 | } | |
701 | ||
702 | /** | |
703 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
704 | * @irq: Is this call from rcu_irq_exit? | |
705 | * | |
706 | * If you add or remove a call to rcu_nmi_exit(), be sure to test | |
707 | * with CONFIG_RCU_EQS_DEBUG=y. | |
708 | */ | |
709 | void rcu_nmi_exit(void) | |
710 | { | |
711 | rcu_nmi_exit_common(false); | |
fd581a91 PM |
712 | } |
713 | ||
9b2e4f18 PM |
714 | /** |
715 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
716 | * | |
717 | * Exit from an interrupt handler, which might possibly result in entering | |
718 | * idle mode, in other words, leaving the mode in which read-side critical | |
7c9906ca | 719 | * sections can occur. The caller must have disabled interrupts. |
64db4cff | 720 | * |
9b2e4f18 PM |
721 | * This code assumes that the idle loop never does anything that might |
722 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
58721f5d PM |
723 | * architecture's idle loop violates this assumption, RCU will give you what |
724 | * you deserve, good and hard. But very infrequently and irreproducibly. | |
9b2e4f18 PM |
725 | * |
726 | * Use things like work queues to work around this limitation. | |
727 | * | |
728 | * You have been warned. | |
c0da313e PM |
729 | * |
730 | * If you add or remove a call to rcu_irq_exit(), be sure to test with | |
731 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 732 | */ |
9b2e4f18 | 733 | void rcu_irq_exit(void) |
64db4cff | 734 | { |
b04db8e1 | 735 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 736 | rcu_nmi_exit_common(true); |
7c9906ca PM |
737 | } |
738 | ||
739 | /* | |
740 | * Wrapper for rcu_irq_exit() where interrupts are enabled. | |
c0da313e PM |
741 | * |
742 | * If you add or remove a call to rcu_irq_exit_irqson(), be sure to test | |
743 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
744 | */ |
745 | void rcu_irq_exit_irqson(void) | |
746 | { | |
747 | unsigned long flags; | |
748 | ||
749 | local_irq_save(flags); | |
750 | rcu_irq_exit(); | |
9b2e4f18 PM |
751 | local_irq_restore(flags); |
752 | } | |
753 | ||
adf5091e FW |
754 | /* |
755 | * Exit an RCU extended quiescent state, which can be either the | |
756 | * idle loop or adaptive-tickless usermode execution. | |
51a1fd30 PM |
757 | * |
758 | * We crowbar the ->dynticks_nmi_nesting field to DYNTICK_IRQ_NONIDLE to | |
759 | * allow for the possibility of usermode upcalls messing up our count of | |
760 | * interrupt nesting level during the busy period that is just now starting. | |
9b2e4f18 | 761 | */ |
adf5091e | 762 | static void rcu_eqs_exit(bool user) |
9b2e4f18 | 763 | { |
4c5273bf | 764 | struct rcu_data *rdp; |
84585aa8 | 765 | long oldval; |
9b2e4f18 | 766 | |
b04db8e1 | 767 | lockdep_assert_irqs_disabled(); |
4c5273bf PM |
768 | rdp = this_cpu_ptr(&rcu_data); |
769 | oldval = rdp->dynticks_nesting; | |
1ce46ee5 | 770 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0); |
51a1fd30 | 771 | if (oldval) { |
4c5273bf | 772 | rdp->dynticks_nesting++; |
9dd238e2 | 773 | return; |
3a592405 | 774 | } |
9dd238e2 PM |
775 | rcu_dynticks_task_exit(); |
776 | rcu_dynticks_eqs_exit(); | |
777 | rcu_cleanup_after_idle(); | |
dc5a4f29 | 778 | trace_rcu_dyntick(TPS("End"), rdp->dynticks_nesting, 1, rdp->dynticks); |
e68bbb26 | 779 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
4c5273bf PM |
780 | WRITE_ONCE(rdp->dynticks_nesting, 1); |
781 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting); | |
782 | WRITE_ONCE(rdp->dynticks_nmi_nesting, DYNTICK_IRQ_NONIDLE); | |
9b2e4f18 | 783 | } |
adf5091e FW |
784 | |
785 | /** | |
786 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
787 | * | |
788 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
789 | * read-side critical sections can occur. | |
790 | * | |
c0da313e PM |
791 | * If you add or remove a call to rcu_idle_exit(), be sure to test with |
792 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
793 | */ |
794 | void rcu_idle_exit(void) | |
795 | { | |
c5d900bf FW |
796 | unsigned long flags; |
797 | ||
798 | local_irq_save(flags); | |
cb349ca9 | 799 | rcu_eqs_exit(false); |
c5d900bf | 800 | local_irq_restore(flags); |
adf5091e | 801 | } |
9b2e4f18 | 802 | |
d1ec4c34 | 803 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
804 | /** |
805 | * rcu_user_exit - inform RCU that we are exiting userspace. | |
806 | * | |
807 | * Exit RCU idle mode while entering the kernel because it can | |
808 | * run a RCU read side critical section anytime. | |
c0da313e PM |
809 | * |
810 | * If you add or remove a call to rcu_user_exit(), be sure to test with | |
811 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
812 | */ |
813 | void rcu_user_exit(void) | |
814 | { | |
91d1aa43 | 815 | rcu_eqs_exit(1); |
adf5091e | 816 | } |
d1ec4c34 | 817 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 818 | |
64db4cff | 819 | /** |
cf7614e1 BP |
820 | * rcu_nmi_enter_common - inform RCU of entry to NMI context |
821 | * @irq: Is this call from rcu_irq_enter? | |
64db4cff | 822 | * |
dc5a4f29 | 823 | * If the CPU was idle from RCU's viewpoint, update rdp->dynticks and |
4c5273bf | 824 | * rdp->dynticks_nmi_nesting to let the RCU grace-period handling know |
734d1680 PM |
825 | * that the CPU is active. This implementation permits nested NMIs, as |
826 | * long as the nesting level does not overflow an int. (You will probably | |
827 | * run out of stack space first.) | |
c0da313e | 828 | * |
cf7614e1 | 829 | * If you add or remove a call to rcu_nmi_enter_common(), be sure to test |
c0da313e | 830 | * with CONFIG_RCU_EQS_DEBUG=y. |
64db4cff | 831 | */ |
cf7614e1 | 832 | static __always_inline void rcu_nmi_enter_common(bool irq) |
64db4cff | 833 | { |
4c5273bf | 834 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
84585aa8 | 835 | long incby = 2; |
64db4cff | 836 | |
734d1680 | 837 | /* Complain about underflow. */ |
4c5273bf | 838 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting < 0); |
734d1680 PM |
839 | |
840 | /* | |
841 | * If idle from RCU viewpoint, atomically increment ->dynticks | |
842 | * to mark non-idle and increment ->dynticks_nmi_nesting by one. | |
843 | * Otherwise, increment ->dynticks_nmi_nesting by two. This means | |
844 | * if ->dynticks_nmi_nesting is equal to one, we are guaranteed | |
845 | * to be in the outermost NMI handler that interrupted an RCU-idle | |
846 | * period (observation due to Andy Lutomirski). | |
847 | */ | |
02a5c550 | 848 | if (rcu_dynticks_curr_cpu_in_eqs()) { |
cf7614e1 BP |
849 | |
850 | if (irq) | |
851 | rcu_dynticks_task_exit(); | |
852 | ||
2625d469 | 853 | rcu_dynticks_eqs_exit(); |
cf7614e1 BP |
854 | |
855 | if (irq) | |
856 | rcu_cleanup_after_idle(); | |
857 | ||
734d1680 PM |
858 | incby = 1; |
859 | } | |
bd2b879a | 860 | trace_rcu_dyntick(incby == 1 ? TPS("Endirq") : TPS("++="), |
4c5273bf | 861 | rdp->dynticks_nmi_nesting, |
dc5a4f29 | 862 | rdp->dynticks_nmi_nesting + incby, rdp->dynticks); |
4c5273bf PM |
863 | WRITE_ONCE(rdp->dynticks_nmi_nesting, /* Prevent store tearing. */ |
864 | rdp->dynticks_nmi_nesting + incby); | |
734d1680 | 865 | barrier(); |
64db4cff PM |
866 | } |
867 | ||
cf7614e1 BP |
868 | /** |
869 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
870 | */ | |
871 | void rcu_nmi_enter(void) | |
872 | { | |
873 | rcu_nmi_enter_common(false); | |
874 | } | |
875 | ||
64db4cff | 876 | /** |
9b2e4f18 | 877 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle |
64db4cff | 878 | * |
9b2e4f18 PM |
879 | * Enter an interrupt handler, which might possibly result in exiting |
880 | * idle mode, in other words, entering the mode in which read-side critical | |
7c9906ca | 881 | * sections can occur. The caller must have disabled interrupts. |
c0da313e | 882 | * |
9b2e4f18 | 883 | * Note that the Linux kernel is fully capable of entering an interrupt |
58721f5d PM |
884 | * handler that it never exits, for example when doing upcalls to user mode! |
885 | * This code assumes that the idle loop never does upcalls to user mode. | |
886 | * If your architecture's idle loop does do upcalls to user mode (or does | |
887 | * anything else that results in unbalanced calls to the irq_enter() and | |
888 | * irq_exit() functions), RCU will give you what you deserve, good and hard. | |
889 | * But very infrequently and irreproducibly. | |
9b2e4f18 PM |
890 | * |
891 | * Use things like work queues to work around this limitation. | |
892 | * | |
893 | * You have been warned. | |
c0da313e PM |
894 | * |
895 | * If you add or remove a call to rcu_irq_enter(), be sure to test with | |
896 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 897 | */ |
9b2e4f18 | 898 | void rcu_irq_enter(void) |
64db4cff | 899 | { |
b04db8e1 | 900 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 901 | rcu_nmi_enter_common(true); |
7c9906ca | 902 | } |
734d1680 | 903 | |
7c9906ca PM |
904 | /* |
905 | * Wrapper for rcu_irq_enter() where interrupts are enabled. | |
c0da313e PM |
906 | * |
907 | * If you add or remove a call to rcu_irq_enter_irqson(), be sure to test | |
908 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
909 | */ |
910 | void rcu_irq_enter_irqson(void) | |
911 | { | |
912 | unsigned long flags; | |
734d1680 | 913 | |
7c9906ca PM |
914 | local_irq_save(flags); |
915 | rcu_irq_enter(); | |
64db4cff | 916 | local_irq_restore(flags); |
64db4cff PM |
917 | } |
918 | ||
5c173eb8 | 919 | /** |
2320bda2 | 920 | * rcu_is_watching - see if RCU thinks that the current CPU is not idle |
64db4cff | 921 | * |
791875d1 PM |
922 | * Return true if RCU is watching the running CPU, which means that this |
923 | * CPU can safely enter RCU read-side critical sections. In other words, | |
2320bda2 ZZ |
924 | * if the current CPU is not in its idle loop or is in an interrupt or |
925 | * NMI handler, return true. | |
64db4cff | 926 | */ |
9418fb20 | 927 | bool notrace rcu_is_watching(void) |
64db4cff | 928 | { |
f534ed1f | 929 | bool ret; |
34240697 | 930 | |
46f00d18 | 931 | preempt_disable_notrace(); |
791875d1 | 932 | ret = !rcu_dynticks_curr_cpu_in_eqs(); |
46f00d18 | 933 | preempt_enable_notrace(); |
34240697 | 934 | return ret; |
64db4cff | 935 | } |
5c173eb8 | 936 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 937 | |
bcbfdd01 PM |
938 | /* |
939 | * If a holdout task is actually running, request an urgent quiescent | |
940 | * state from its CPU. This is unsynchronized, so migrations can cause | |
941 | * the request to go to the wrong CPU. Which is OK, all that will happen | |
942 | * is that the CPU's next context switch will be a bit slower and next | |
943 | * time around this task will generate another request. | |
944 | */ | |
945 | void rcu_request_urgent_qs_task(struct task_struct *t) | |
946 | { | |
947 | int cpu; | |
948 | ||
949 | barrier(); | |
950 | cpu = task_cpu(t); | |
951 | if (!task_curr(t)) | |
952 | return; /* This task is not running on that CPU. */ | |
2dba13f0 | 953 | smp_store_release(per_cpu_ptr(&rcu_data.rcu_urgent_qs, cpu), true); |
bcbfdd01 PM |
954 | } |
955 | ||
62fde6ed | 956 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
957 | |
958 | /* | |
5554788e | 959 | * Is the current CPU online as far as RCU is concerned? |
2036d94a | 960 | * |
5554788e PM |
961 | * Disable preemption to avoid false positives that could otherwise |
962 | * happen due to the current CPU number being sampled, this task being | |
963 | * preempted, its old CPU being taken offline, resuming on some other CPU, | |
49918a54 | 964 | * then determining that its old CPU is now offline. |
c0d6d01b | 965 | * |
5554788e PM |
966 | * Disable checking if in an NMI handler because we cannot safely |
967 | * report errors from NMI handlers anyway. In addition, it is OK to use | |
968 | * RCU on an offline processor during initial boot, hence the check for | |
969 | * rcu_scheduler_fully_active. | |
c0d6d01b PM |
970 | */ |
971 | bool rcu_lockdep_current_cpu_online(void) | |
972 | { | |
2036d94a PM |
973 | struct rcu_data *rdp; |
974 | struct rcu_node *rnp; | |
b97d23c5 | 975 | bool ret = false; |
c0d6d01b | 976 | |
5554788e | 977 | if (in_nmi() || !rcu_scheduler_fully_active) |
f6f7ee9a | 978 | return true; |
c0d6d01b | 979 | preempt_disable(); |
b97d23c5 PM |
980 | rdp = this_cpu_ptr(&rcu_data); |
981 | rnp = rdp->mynode; | |
982 | if (rdp->grpmask & rcu_rnp_online_cpus(rnp)) | |
983 | ret = true; | |
c0d6d01b | 984 | preempt_enable(); |
b97d23c5 | 985 | return ret; |
c0d6d01b PM |
986 | } |
987 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
988 | ||
62fde6ed | 989 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 990 | |
9b9500da PM |
991 | /* |
992 | * We are reporting a quiescent state on behalf of some other CPU, so | |
993 | * it is our responsibility to check for and handle potential overflow | |
a66ae8ae | 994 | * of the rcu_node ->gp_seq counter with respect to the rcu_data counters. |
9b9500da PM |
995 | * After all, the CPU might be in deep idle state, and thus executing no |
996 | * code whatsoever. | |
997 | */ | |
998 | static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp) | |
999 | { | |
a32e01ee | 1000 | raw_lockdep_assert_held_rcu_node(rnp); |
a66ae8ae PM |
1001 | if (ULONG_CMP_LT(rcu_seq_current(&rdp->gp_seq) + ULONG_MAX / 4, |
1002 | rnp->gp_seq)) | |
9b9500da | 1003 | WRITE_ONCE(rdp->gpwrap, true); |
8aa670cd PM |
1004 | if (ULONG_CMP_LT(rdp->rcu_iw_gp_seq + ULONG_MAX / 4, rnp->gp_seq)) |
1005 | rdp->rcu_iw_gp_seq = rnp->gp_seq + ULONG_MAX / 4; | |
9b9500da PM |
1006 | } |
1007 | ||
64db4cff PM |
1008 | /* |
1009 | * Snapshot the specified CPU's dynticks counter so that we can later | |
1010 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 1011 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 1012 | */ |
fe5ac724 | 1013 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
64db4cff | 1014 | { |
dc5a4f29 | 1015 | rdp->dynticks_snap = rcu_dynticks_snap(rdp); |
02a5c550 | 1016 | if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) { |
88d1bead | 1017 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 1018 | rcu_gpnum_ovf(rdp->mynode, rdp); |
23a9bacd | 1019 | return 1; |
7941dbde | 1020 | } |
23a9bacd | 1021 | return 0; |
64db4cff PM |
1022 | } |
1023 | ||
9b9500da PM |
1024 | /* |
1025 | * Handler for the irq_work request posted when a grace period has | |
1026 | * gone on for too long, but not yet long enough for an RCU CPU | |
1027 | * stall warning. Set state appropriately, but just complain if | |
1028 | * there is unexpected state on entry. | |
1029 | */ | |
1030 | static void rcu_iw_handler(struct irq_work *iwp) | |
1031 | { | |
1032 | struct rcu_data *rdp; | |
1033 | struct rcu_node *rnp; | |
1034 | ||
1035 | rdp = container_of(iwp, struct rcu_data, rcu_iw); | |
1036 | rnp = rdp->mynode; | |
1037 | raw_spin_lock_rcu_node(rnp); | |
1038 | if (!WARN_ON_ONCE(!rdp->rcu_iw_pending)) { | |
8aa670cd | 1039 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1040 | rdp->rcu_iw_pending = false; |
1041 | } | |
1042 | raw_spin_unlock_rcu_node(rnp); | |
1043 | } | |
1044 | ||
64db4cff PM |
1045 | /* |
1046 | * Return true if the specified CPU has passed through a quiescent | |
1047 | * state by virtue of being in or having passed through an dynticks | |
1048 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 1049 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 1050 | */ |
fe5ac724 | 1051 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
64db4cff | 1052 | { |
3a19b46a | 1053 | unsigned long jtsq; |
0f9be8ca | 1054 | bool *rnhqp; |
9226b10d | 1055 | bool *ruqp; |
9b9500da | 1056 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
1057 | |
1058 | /* | |
1059 | * If the CPU passed through or entered a dynticks idle phase with | |
1060 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
1061 | * already acknowledged the request to pass through a quiescent | |
1062 | * state. Either way, that CPU cannot possibly be in an RCU | |
1063 | * read-side critical section that started before the beginning | |
1064 | * of the current RCU grace period. | |
1065 | */ | |
dc5a4f29 | 1066 | if (rcu_dynticks_in_eqs_since(rdp, rdp->dynticks_snap)) { |
88d1bead | 1067 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 1068 | rcu_gpnum_ovf(rnp, rdp); |
3a19b46a PM |
1069 | return 1; |
1070 | } | |
1071 | ||
f2e2df59 PM |
1072 | /* If waiting too long on an offline CPU, complain. */ |
1073 | if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp)) && | |
88d1bead | 1074 | time_after(jiffies, rcu_state.gp_start + HZ)) { |
f2e2df59 PM |
1075 | bool onl; |
1076 | struct rcu_node *rnp1; | |
1077 | ||
1078 | WARN_ON(1); /* Offline CPUs are supposed to report QS! */ | |
1079 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", | |
1080 | __func__, rnp->grplo, rnp->grphi, rnp->level, | |
1081 | (long)rnp->gp_seq, (long)rnp->completedqs); | |
1082 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
1083 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx ->rcu_gp_init_mask %#lx\n", | |
1084 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext, rnp1->rcu_gp_init_mask); | |
1085 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); | |
1086 | pr_info("%s %d: %c online: %ld(%d) offline: %ld(%d)\n", | |
1087 | __func__, rdp->cpu, ".o"[onl], | |
1088 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
1089 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
1090 | return 1; /* Break things loose after complaining. */ | |
1091 | } | |
1092 | ||
65d798f0 | 1093 | /* |
4a81e832 | 1094 | * A CPU running for an extended time within the kernel can |
c06aed0e PM |
1095 | * delay RCU grace periods: (1) At age jiffies_to_sched_qs, |
1096 | * set .rcu_urgent_qs, (2) At age 2*jiffies_to_sched_qs, set | |
7e28c5af PM |
1097 | * both .rcu_need_heavy_qs and .rcu_urgent_qs. Note that the |
1098 | * unsynchronized assignments to the per-CPU rcu_need_heavy_qs | |
1099 | * variable are safe because the assignments are repeated if this | |
1100 | * CPU failed to pass through a quiescent state. This code | |
c06aed0e | 1101 | * also checks .jiffies_resched in case jiffies_to_sched_qs |
7e28c5af | 1102 | * is set way high. |
6193c76a | 1103 | */ |
c06aed0e | 1104 | jtsq = READ_ONCE(jiffies_to_sched_qs); |
2dba13f0 PM |
1105 | ruqp = per_cpu_ptr(&rcu_data.rcu_urgent_qs, rdp->cpu); |
1106 | rnhqp = &per_cpu(rcu_data.rcu_need_heavy_qs, rdp->cpu); | |
0f9be8ca | 1107 | if (!READ_ONCE(*rnhqp) && |
7e28c5af | 1108 | (time_after(jiffies, rcu_state.gp_start + jtsq * 2) || |
88d1bead | 1109 | time_after(jiffies, rcu_state.jiffies_resched))) { |
0f9be8ca | 1110 | WRITE_ONCE(*rnhqp, true); |
9226b10d PM |
1111 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ |
1112 | smp_store_release(ruqp, true); | |
7e28c5af PM |
1113 | } else if (time_after(jiffies, rcu_state.gp_start + jtsq)) { |
1114 | WRITE_ONCE(*ruqp, true); | |
6193c76a PM |
1115 | } |
1116 | ||
28053bc7 | 1117 | /* |
d3052109 PM |
1118 | * NO_HZ_FULL CPUs can run in-kernel without rcu_check_callbacks! |
1119 | * The above code handles this, but only for straight cond_resched(). | |
1120 | * And some in-kernel loops check need_resched() before calling | |
1121 | * cond_resched(), which defeats the above code for CPUs that are | |
1122 | * running in-kernel with scheduling-clock interrupts disabled. | |
1123 | * So hit them over the head with the resched_cpu() hammer! | |
28053bc7 | 1124 | */ |
d3052109 PM |
1125 | if (tick_nohz_full_cpu(rdp->cpu) && |
1126 | time_after(jiffies, | |
1127 | READ_ONCE(rdp->last_fqs_resched) + jtsq * 3)) { | |
28053bc7 | 1128 | resched_cpu(rdp->cpu); |
d3052109 PM |
1129 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * If more than halfway to RCU CPU stall-warning time, invoke | |
1134 | * resched_cpu() more frequently to try to loosen things up a bit. | |
1135 | * Also check to see if the CPU is getting hammered with interrupts, | |
1136 | * but only once per grace period, just to keep the IPIs down to | |
1137 | * a dull roar. | |
1138 | */ | |
1139 | if (time_after(jiffies, rcu_state.jiffies_resched)) { | |
1140 | if (time_after(jiffies, | |
1141 | READ_ONCE(rdp->last_fqs_resched) + jtsq)) { | |
1142 | resched_cpu(rdp->cpu); | |
1143 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); | |
1144 | } | |
9b9500da | 1145 | if (IS_ENABLED(CONFIG_IRQ_WORK) && |
8aa670cd | 1146 | !rdp->rcu_iw_pending && rdp->rcu_iw_gp_seq != rnp->gp_seq && |
9b9500da PM |
1147 | (rnp->ffmask & rdp->grpmask)) { |
1148 | init_irq_work(&rdp->rcu_iw, rcu_iw_handler); | |
1149 | rdp->rcu_iw_pending = true; | |
8aa670cd | 1150 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1151 | irq_work_queue_on(&rdp->rcu_iw, rdp->cpu); |
1152 | } | |
1153 | } | |
4914950a | 1154 | |
a82dcc76 | 1155 | return 0; |
64db4cff PM |
1156 | } |
1157 | ||
ad3832e9 | 1158 | static void record_gp_stall_check_time(void) |
64db4cff | 1159 | { |
cb1e78cf | 1160 | unsigned long j = jiffies; |
6193c76a | 1161 | unsigned long j1; |
26cdfedf | 1162 | |
ad3832e9 | 1163 | rcu_state.gp_start = j; |
6193c76a | 1164 | j1 = rcu_jiffies_till_stall_check(); |
91f63ced | 1165 | /* Record ->gp_start before ->jiffies_stall. */ |
ad3832e9 PM |
1166 | smp_store_release(&rcu_state.jiffies_stall, j + j1); /* ^^^ */ |
1167 | rcu_state.jiffies_resched = j + j1 / 2; | |
1168 | rcu_state.n_force_qs_gpstart = READ_ONCE(rcu_state.n_force_qs); | |
64db4cff PM |
1169 | } |
1170 | ||
fb81a44b PM |
1171 | /* |
1172 | * Complain about starvation of grace-period kthread. | |
1173 | */ | |
8fd119b6 | 1174 | static void rcu_check_gp_kthread_starvation(void) |
fb81a44b | 1175 | { |
7cba4775 | 1176 | struct task_struct *gpk = rcu_state.gp_kthread; |
fb81a44b PM |
1177 | unsigned long j; |
1178 | ||
7cba4775 PM |
1179 | j = jiffies - READ_ONCE(rcu_state.gp_activity); |
1180 | if (j > 2 * HZ) { | |
78c5a67f | 1181 | pr_err("%s kthread starved for %ld jiffies! g%ld f%#x %s(%d) ->state=%#lx ->cpu=%d\n", |
7cba4775 PM |
1182 | rcu_state.name, j, |
1183 | (long)rcu_seq_current(&rcu_state.gp_seq), | |
1184 | rcu_state.gp_flags, | |
1185 | gp_state_getname(rcu_state.gp_state), rcu_state.gp_state, | |
1186 | gpk ? gpk->state : ~0, gpk ? task_cpu(gpk) : -1); | |
1187 | if (gpk) { | |
d07aee2c | 1188 | pr_err("RCU grace-period kthread stack dump:\n"); |
7cba4775 PM |
1189 | sched_show_task(gpk); |
1190 | wake_up_process(gpk); | |
86057b80 | 1191 | } |
b1adb3e2 | 1192 | } |
64db4cff PM |
1193 | } |
1194 | ||
b637a328 | 1195 | /* |
7aa92230 PM |
1196 | * Dump stacks of all tasks running on stalled CPUs. First try using |
1197 | * NMIs, but fall back to manual remote stack tracing on architectures | |
1198 | * that don't support NMI-based stack dumps. The NMI-triggered stack | |
1199 | * traces are more accurate because they are printed by the target CPU. | |
b637a328 | 1200 | */ |
33dbdbf0 | 1201 | static void rcu_dump_cpu_stacks(void) |
b637a328 PM |
1202 | { |
1203 | int cpu; | |
1204 | unsigned long flags; | |
1205 | struct rcu_node *rnp; | |
1206 | ||
aedf4ba9 | 1207 | rcu_for_each_leaf_node(rnp) { |
6cf10081 | 1208 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7aa92230 PM |
1209 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1210 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) | |
1211 | if (!trigger_single_cpu_backtrace(cpu)) | |
bc75e999 | 1212 | dump_cpu_task(cpu); |
67c583a7 | 1213 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b637a328 PM |
1214 | } |
1215 | } | |
1216 | ||
8c7c4829 PM |
1217 | /* |
1218 | * If too much time has passed in the current grace period, and if | |
1219 | * so configured, go kick the relevant kthreads. | |
1220 | */ | |
e1741c69 | 1221 | static void rcu_stall_kick_kthreads(void) |
8c7c4829 PM |
1222 | { |
1223 | unsigned long j; | |
1224 | ||
1225 | if (!rcu_kick_kthreads) | |
1226 | return; | |
4c6ed437 PM |
1227 | j = READ_ONCE(rcu_state.jiffies_kick_kthreads); |
1228 | if (time_after(jiffies, j) && rcu_state.gp_kthread && | |
1229 | (rcu_gp_in_progress() || READ_ONCE(rcu_state.gp_flags))) { | |
1230 | WARN_ONCE(1, "Kicking %s grace-period kthread\n", | |
1231 | rcu_state.name); | |
5dffed1e | 1232 | rcu_ftrace_dump(DUMP_ALL); |
4c6ed437 PM |
1233 | wake_up_process(rcu_state.gp_kthread); |
1234 | WRITE_ONCE(rcu_state.jiffies_kick_kthreads, j + HZ); | |
8c7c4829 PM |
1235 | } |
1236 | } | |
1237 | ||
95394e69 | 1238 | static void panic_on_rcu_stall(void) |
088e9d25 DBO |
1239 | { |
1240 | if (sysctl_panic_on_rcu_stall) | |
1241 | panic("RCU Stall\n"); | |
1242 | } | |
1243 | ||
a91e7e58 | 1244 | static void print_other_cpu_stall(unsigned long gp_seq) |
64db4cff PM |
1245 | { |
1246 | int cpu; | |
64db4cff | 1247 | unsigned long flags; |
6ccd2ecd PM |
1248 | unsigned long gpa; |
1249 | unsigned long j; | |
285fe294 | 1250 | int ndetected = 0; |
336a4f6c | 1251 | struct rcu_node *rnp = rcu_get_root(); |
53bb857c | 1252 | long totqlen = 0; |
64db4cff | 1253 | |
8c7c4829 | 1254 | /* Kick and suppress, if so configured. */ |
e1741c69 | 1255 | rcu_stall_kick_kthreads(); |
8c7c4829 PM |
1256 | if (rcu_cpu_stall_suppress) |
1257 | return; | |
1258 | ||
8cdd32a9 PM |
1259 | /* |
1260 | * OK, time to rat on our buddy... | |
1261 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1262 | * RCU CPU stall warnings. | |
1263 | */ | |
4c6ed437 | 1264 | pr_err("INFO: %s detected stalls on CPUs/tasks:", rcu_state.name); |
a858af28 | 1265 | print_cpu_stall_info_begin(); |
aedf4ba9 | 1266 | rcu_for_each_leaf_node(rnp) { |
6cf10081 | 1267 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
9bc8b558 | 1268 | ndetected += rcu_print_task_stall(rnp); |
c8020a67 | 1269 | if (rnp->qsmask != 0) { |
bc75e999 MR |
1270 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1271 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) { | |
b21ebed9 | 1272 | print_cpu_stall_info(cpu); |
c8020a67 PM |
1273 | ndetected++; |
1274 | } | |
1275 | } | |
67c583a7 | 1276 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1277 | } |
a858af28 | 1278 | |
a858af28 | 1279 | print_cpu_stall_info_end(); |
53bb857c | 1280 | for_each_possible_cpu(cpu) |
903ee83d | 1281 | totqlen += rcu_get_n_cbs_cpu(cpu); |
471f87c3 | 1282 | pr_cont("(detected by %d, t=%ld jiffies, g=%ld, q=%lu)\n", |
4c6ed437 PM |
1283 | smp_processor_id(), (long)(jiffies - rcu_state.gp_start), |
1284 | (long)rcu_seq_current(&rcu_state.gp_seq), totqlen); | |
6ccd2ecd | 1285 | if (ndetected) { |
33dbdbf0 | 1286 | rcu_dump_cpu_stacks(); |
c4402b27 BP |
1287 | |
1288 | /* Complain about tasks blocking the grace period. */ | |
a2887cd8 | 1289 | rcu_print_detail_task_stall(); |
6ccd2ecd | 1290 | } else { |
4c6ed437 | 1291 | if (rcu_seq_current(&rcu_state.gp_seq) != gp_seq) { |
6ccd2ecd PM |
1292 | pr_err("INFO: Stall ended before state dump start\n"); |
1293 | } else { | |
1294 | j = jiffies; | |
4c6ed437 | 1295 | gpa = READ_ONCE(rcu_state.gp_activity); |
237a0f21 | 1296 | pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n", |
4c6ed437 | 1297 | rcu_state.name, j - gpa, j, gpa, |
c06aed0e | 1298 | READ_ONCE(jiffies_till_next_fqs), |
336a4f6c | 1299 | rcu_get_root()->qsmask); |
6ccd2ecd PM |
1300 | /* In this case, the current CPU might be at fault. */ |
1301 | sched_show_task(current); | |
1302 | } | |
1303 | } | |
8c42b1f3 | 1304 | /* Rewrite if needed in case of slow consoles. */ |
4c6ed437 PM |
1305 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall))) |
1306 | WRITE_ONCE(rcu_state.jiffies_stall, | |
8c42b1f3 | 1307 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); |
c1dc0b9c | 1308 | |
8fd119b6 | 1309 | rcu_check_gp_kthread_starvation(); |
fb81a44b | 1310 | |
088e9d25 DBO |
1311 | panic_on_rcu_stall(); |
1312 | ||
e9ecb780 | 1313 | force_quiescent_state(); /* Kick them all. */ |
64db4cff PM |
1314 | } |
1315 | ||
4e8b8e08 | 1316 | static void print_cpu_stall(void) |
64db4cff | 1317 | { |
53bb857c | 1318 | int cpu; |
64db4cff | 1319 | unsigned long flags; |
da1df50d | 1320 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
336a4f6c | 1321 | struct rcu_node *rnp = rcu_get_root(); |
53bb857c | 1322 | long totqlen = 0; |
64db4cff | 1323 | |
8c7c4829 | 1324 | /* Kick and suppress, if so configured. */ |
e1741c69 | 1325 | rcu_stall_kick_kthreads(); |
8c7c4829 PM |
1326 | if (rcu_cpu_stall_suppress) |
1327 | return; | |
1328 | ||
8cdd32a9 PM |
1329 | /* |
1330 | * OK, time to rat on ourselves... | |
1331 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1332 | * RCU CPU stall warnings. | |
1333 | */ | |
4c6ed437 | 1334 | pr_err("INFO: %s self-detected stall on CPU", rcu_state.name); |
a858af28 | 1335 | print_cpu_stall_info_begin(); |
9b9500da | 1336 | raw_spin_lock_irqsave_rcu_node(rdp->mynode, flags); |
b21ebed9 | 1337 | print_cpu_stall_info(smp_processor_id()); |
9b9500da | 1338 | raw_spin_unlock_irqrestore_rcu_node(rdp->mynode, flags); |
a858af28 | 1339 | print_cpu_stall_info_end(); |
53bb857c | 1340 | for_each_possible_cpu(cpu) |
903ee83d | 1341 | totqlen += rcu_get_n_cbs_cpu(cpu); |
471f87c3 | 1342 | pr_cont(" (t=%lu jiffies g=%ld q=%lu)\n", |
4c6ed437 PM |
1343 | jiffies - rcu_state.gp_start, |
1344 | (long)rcu_seq_current(&rcu_state.gp_seq), totqlen); | |
fb81a44b | 1345 | |
8fd119b6 | 1346 | rcu_check_gp_kthread_starvation(); |
fb81a44b | 1347 | |
33dbdbf0 | 1348 | rcu_dump_cpu_stacks(); |
c1dc0b9c | 1349 | |
6cf10081 | 1350 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
8c42b1f3 | 1351 | /* Rewrite if needed in case of slow consoles. */ |
4c6ed437 PM |
1352 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall))) |
1353 | WRITE_ONCE(rcu_state.jiffies_stall, | |
7d0ae808 | 1354 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); |
67c583a7 | 1355 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c1dc0b9c | 1356 | |
088e9d25 DBO |
1357 | panic_on_rcu_stall(); |
1358 | ||
b021fe3e PZ |
1359 | /* |
1360 | * Attempt to revive the RCU machinery by forcing a context switch. | |
1361 | * | |
1362 | * A context switch would normally allow the RCU state machine to make | |
1363 | * progress and it could be we're stuck in kernel space without context | |
1364 | * switches for an entirely unreasonable amount of time. | |
1365 | */ | |
fced9c8c PM |
1366 | set_tsk_need_resched(current); |
1367 | set_preempt_need_resched(); | |
64db4cff PM |
1368 | } |
1369 | ||
ea12ff2b | 1370 | static void check_cpu_stall(struct rcu_data *rdp) |
64db4cff | 1371 | { |
471f87c3 PM |
1372 | unsigned long gs1; |
1373 | unsigned long gs2; | |
26cdfedf | 1374 | unsigned long gps; |
bad6e139 | 1375 | unsigned long j; |
8c42b1f3 | 1376 | unsigned long jn; |
bad6e139 | 1377 | unsigned long js; |
64db4cff PM |
1378 | struct rcu_node *rnp; |
1379 | ||
8c7c4829 | 1380 | if ((rcu_cpu_stall_suppress && !rcu_kick_kthreads) || |
de8e8730 | 1381 | !rcu_gp_in_progress()) |
c68de209 | 1382 | return; |
e1741c69 | 1383 | rcu_stall_kick_kthreads(); |
cb1e78cf | 1384 | j = jiffies; |
26cdfedf PM |
1385 | |
1386 | /* | |
1387 | * Lots of memory barriers to reject false positives. | |
1388 | * | |
4c6ed437 PM |
1389 | * The idea is to pick up rcu_state.gp_seq, then |
1390 | * rcu_state.jiffies_stall, then rcu_state.gp_start, and finally | |
1391 | * another copy of rcu_state.gp_seq. These values are updated in | |
1392 | * the opposite order with memory barriers (or equivalent) during | |
1393 | * grace-period initialization and cleanup. Now, a false positive | |
1394 | * can occur if we get an new value of rcu_state.gp_start and a old | |
1395 | * value of rcu_state.jiffies_stall. But given the memory barriers, | |
1396 | * the only way that this can happen is if one grace period ends | |
1397 | * and another starts between these two fetches. This is detected | |
1398 | * by comparing the second fetch of rcu_state.gp_seq with the | |
1399 | * previous fetch from rcu_state.gp_seq. | |
26cdfedf | 1400 | * |
4c6ed437 PM |
1401 | * Given this check, comparisons of jiffies, rcu_state.jiffies_stall, |
1402 | * and rcu_state.gp_start suffice to forestall false positives. | |
26cdfedf | 1403 | */ |
4c6ed437 | 1404 | gs1 = READ_ONCE(rcu_state.gp_seq); |
471f87c3 | 1405 | smp_rmb(); /* Pick up ->gp_seq first... */ |
4c6ed437 | 1406 | js = READ_ONCE(rcu_state.jiffies_stall); |
26cdfedf | 1407 | smp_rmb(); /* ...then ->jiffies_stall before the rest... */ |
4c6ed437 | 1408 | gps = READ_ONCE(rcu_state.gp_start); |
471f87c3 | 1409 | smp_rmb(); /* ...and finally ->gp_start before ->gp_seq again. */ |
4c6ed437 | 1410 | gs2 = READ_ONCE(rcu_state.gp_seq); |
471f87c3 | 1411 | if (gs1 != gs2 || |
26cdfedf PM |
1412 | ULONG_CMP_LT(j, js) || |
1413 | ULONG_CMP_GE(gps, js)) | |
1414 | return; /* No stall or GP completed since entering function. */ | |
64db4cff | 1415 | rnp = rdp->mynode; |
8c42b1f3 | 1416 | jn = jiffies + 3 * rcu_jiffies_till_stall_check() + 3; |
de8e8730 | 1417 | if (rcu_gp_in_progress() && |
8c42b1f3 | 1418 | (READ_ONCE(rnp->qsmask) & rdp->grpmask) && |
4c6ed437 | 1419 | cmpxchg(&rcu_state.jiffies_stall, js, jn) == js) { |
64db4cff PM |
1420 | |
1421 | /* We haven't checked in, so go dump stack. */ | |
4e8b8e08 | 1422 | print_cpu_stall(); |
64db4cff | 1423 | |
de8e8730 | 1424 | } else if (rcu_gp_in_progress() && |
8c42b1f3 | 1425 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY) && |
4c6ed437 | 1426 | cmpxchg(&rcu_state.jiffies_stall, js, jn) == js) { |
64db4cff | 1427 | |
bad6e139 | 1428 | /* They had a few time units to dump stack, so complain. */ |
a91e7e58 | 1429 | print_other_cpu_stall(gs2); |
64db4cff PM |
1430 | } |
1431 | } | |
1432 | ||
53d84e00 PM |
1433 | /** |
1434 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
1435 | * | |
1436 | * Set the stall-warning timeout way off into the future, thus preventing | |
1437 | * any RCU CPU stall-warning messages from appearing in the current set of | |
1438 | * RCU grace periods. | |
1439 | * | |
1440 | * The caller must disable hard irqs. | |
1441 | */ | |
1442 | void rcu_cpu_stall_reset(void) | |
1443 | { | |
b97d23c5 | 1444 | WRITE_ONCE(rcu_state.jiffies_stall, jiffies + ULONG_MAX / 2); |
53d84e00 PM |
1445 | } |
1446 | ||
41e80595 PM |
1447 | /* Trace-event wrapper function for trace_rcu_future_grace_period. */ |
1448 | static void trace_rcu_this_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
b73de91d | 1449 | unsigned long gp_seq_req, const char *s) |
0446be48 | 1450 | { |
88d1bead | 1451 | trace_rcu_future_grace_period(rcu_state.name, rnp->gp_seq, gp_seq_req, |
abd13fdd | 1452 | rnp->level, rnp->grplo, rnp->grphi, s); |
0446be48 PM |
1453 | } |
1454 | ||
1455 | /* | |
b73de91d | 1456 | * rcu_start_this_gp - Request the start of a particular grace period |
df2bf8f7 | 1457 | * @rnp_start: The leaf node of the CPU from which to start. |
b73de91d JF |
1458 | * @rdp: The rcu_data corresponding to the CPU from which to start. |
1459 | * @gp_seq_req: The gp_seq of the grace period to start. | |
1460 | * | |
41e80595 | 1461 | * Start the specified grace period, as needed to handle newly arrived |
0446be48 | 1462 | * callbacks. The required future grace periods are recorded in each |
7a1d0f23 | 1463 | * rcu_node structure's ->gp_seq_needed field. Returns true if there |
48a7639c | 1464 | * is reason to awaken the grace-period kthread. |
0446be48 | 1465 | * |
d5cd9685 PM |
1466 | * The caller must hold the specified rcu_node structure's ->lock, which |
1467 | * is why the caller is responsible for waking the grace-period kthread. | |
b73de91d JF |
1468 | * |
1469 | * Returns true if the GP thread needs to be awakened else false. | |
0446be48 | 1470 | */ |
df2bf8f7 | 1471 | static bool rcu_start_this_gp(struct rcu_node *rnp_start, struct rcu_data *rdp, |
b73de91d | 1472 | unsigned long gp_seq_req) |
0446be48 | 1473 | { |
48a7639c | 1474 | bool ret = false; |
df2bf8f7 | 1475 | struct rcu_node *rnp; |
0446be48 PM |
1476 | |
1477 | /* | |
360e0da6 PM |
1478 | * Use funnel locking to either acquire the root rcu_node |
1479 | * structure's lock or bail out if the need for this grace period | |
df2bf8f7 JFG |
1480 | * has already been recorded -- or if that grace period has in |
1481 | * fact already started. If there is already a grace period in | |
1482 | * progress in a non-leaf node, no recording is needed because the | |
1483 | * end of the grace period will scan the leaf rcu_node structures. | |
1484 | * Note that rnp_start->lock must not be released. | |
0446be48 | 1485 | */ |
df2bf8f7 JFG |
1486 | raw_lockdep_assert_held_rcu_node(rnp_start); |
1487 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, TPS("Startleaf")); | |
1488 | for (rnp = rnp_start; 1; rnp = rnp->parent) { | |
1489 | if (rnp != rnp_start) | |
1490 | raw_spin_lock_rcu_node(rnp); | |
1491 | if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req) || | |
1492 | rcu_seq_started(&rnp->gp_seq, gp_seq_req) || | |
1493 | (rnp != rnp_start && | |
1494 | rcu_seq_state(rcu_seq_current(&rnp->gp_seq)))) { | |
1495 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, | |
b73de91d | 1496 | TPS("Prestarted")); |
360e0da6 PM |
1497 | goto unlock_out; |
1498 | } | |
df2bf8f7 | 1499 | rnp->gp_seq_needed = gp_seq_req; |
226ca5e7 | 1500 | if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) { |
a2165e41 | 1501 | /* |
226ca5e7 JFG |
1502 | * We just marked the leaf or internal node, and a |
1503 | * grace period is in progress, which means that | |
1504 | * rcu_gp_cleanup() will see the marking. Bail to | |
1505 | * reduce contention. | |
a2165e41 | 1506 | */ |
df2bf8f7 | 1507 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, |
b73de91d | 1508 | TPS("Startedleaf")); |
a2165e41 PM |
1509 | goto unlock_out; |
1510 | } | |
df2bf8f7 JFG |
1511 | if (rnp != rnp_start && rnp->parent != NULL) |
1512 | raw_spin_unlock_rcu_node(rnp); | |
1513 | if (!rnp->parent) | |
360e0da6 | 1514 | break; /* At root, and perhaps also leaf. */ |
0446be48 PM |
1515 | } |
1516 | ||
360e0da6 | 1517 | /* If GP already in progress, just leave, otherwise start one. */ |
de8e8730 | 1518 | if (rcu_gp_in_progress()) { |
df2bf8f7 | 1519 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedleafroot")); |
0446be48 PM |
1520 | goto unlock_out; |
1521 | } | |
df2bf8f7 | 1522 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedroot")); |
9cbc5b97 PM |
1523 | WRITE_ONCE(rcu_state.gp_flags, rcu_state.gp_flags | RCU_GP_FLAG_INIT); |
1524 | rcu_state.gp_req_activity = jiffies; | |
1525 | if (!rcu_state.gp_kthread) { | |
df2bf8f7 | 1526 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("NoGPkthread")); |
360e0da6 | 1527 | goto unlock_out; |
0446be48 | 1528 | } |
9cbc5b97 | 1529 | trace_rcu_grace_period(rcu_state.name, READ_ONCE(rcu_state.gp_seq), TPS("newreq")); |
360e0da6 | 1530 | ret = true; /* Caller must wake GP kthread. */ |
0446be48 | 1531 | unlock_out: |
ab5e869c | 1532 | /* Push furthest requested GP to leaf node and rcu_data structure. */ |
df2bf8f7 JFG |
1533 | if (ULONG_CMP_LT(gp_seq_req, rnp->gp_seq_needed)) { |
1534 | rnp_start->gp_seq_needed = rnp->gp_seq_needed; | |
1535 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
ab5e869c | 1536 | } |
df2bf8f7 JFG |
1537 | if (rnp != rnp_start) |
1538 | raw_spin_unlock_rcu_node(rnp); | |
48a7639c | 1539 | return ret; |
0446be48 PM |
1540 | } |
1541 | ||
1542 | /* | |
1543 | * Clean up any old requests for the just-ended grace period. Also return | |
d1e4f01d | 1544 | * whether any additional grace periods have been requested. |
0446be48 | 1545 | */ |
3481f2ea | 1546 | static bool rcu_future_gp_cleanup(struct rcu_node *rnp) |
0446be48 | 1547 | { |
fb31340f | 1548 | bool needmore; |
da1df50d | 1549 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
0446be48 | 1550 | |
7a1d0f23 PM |
1551 | needmore = ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed); |
1552 | if (!needmore) | |
1553 | rnp->gp_seq_needed = rnp->gp_seq; /* Avoid counter wrap. */ | |
b73de91d | 1554 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq, |
41e80595 | 1555 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); |
0446be48 PM |
1556 | return needmore; |
1557 | } | |
1558 | ||
48a7639c | 1559 | /* |
49918a54 PM |
1560 | * Awaken the grace-period kthread. Don't do a self-awaken, and don't |
1561 | * bother awakening when there is nothing for the grace-period kthread | |
1562 | * to do (as in several CPUs raced to awaken, and we lost), and finally | |
1563 | * don't try to awaken a kthread that has not yet been created. | |
48a7639c | 1564 | */ |
532c00c9 | 1565 | static void rcu_gp_kthread_wake(void) |
48a7639c | 1566 | { |
532c00c9 PM |
1567 | if (current == rcu_state.gp_kthread || |
1568 | !READ_ONCE(rcu_state.gp_flags) || | |
1569 | !rcu_state.gp_kthread) | |
48a7639c | 1570 | return; |
532c00c9 | 1571 | swake_up_one(&rcu_state.gp_wq); |
48a7639c PM |
1572 | } |
1573 | ||
dc35c893 | 1574 | /* |
29365e56 PM |
1575 | * If there is room, assign a ->gp_seq number to any callbacks on this |
1576 | * CPU that have not already been assigned. Also accelerate any callbacks | |
1577 | * that were previously assigned a ->gp_seq number that has since proven | |
1578 | * to be too conservative, which can happen if callbacks get assigned a | |
1579 | * ->gp_seq number while RCU is idle, but with reference to a non-root | |
1580 | * rcu_node structure. This function is idempotent, so it does not hurt | |
1581 | * to call it repeatedly. Returns an flag saying that we should awaken | |
1582 | * the RCU grace-period kthread. | |
dc35c893 PM |
1583 | * |
1584 | * The caller must hold rnp->lock with interrupts disabled. | |
1585 | */ | |
02f50142 | 1586 | static bool rcu_accelerate_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1587 | { |
b73de91d | 1588 | unsigned long gp_seq_req; |
15fecf89 | 1589 | bool ret = false; |
dc35c893 | 1590 | |
a32e01ee | 1591 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1592 | |
15fecf89 PM |
1593 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1594 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1595 | return false; |
dc35c893 PM |
1596 | |
1597 | /* | |
15fecf89 PM |
1598 | * Callbacks are often registered with incomplete grace-period |
1599 | * information. Something about the fact that getting exact | |
1600 | * information requires acquiring a global lock... RCU therefore | |
1601 | * makes a conservative estimate of the grace period number at which | |
1602 | * a given callback will become ready to invoke. The following | |
1603 | * code checks this estimate and improves it when possible, thus | |
1604 | * accelerating callback invocation to an earlier grace-period | |
1605 | * number. | |
dc35c893 | 1606 | */ |
9cbc5b97 | 1607 | gp_seq_req = rcu_seq_snap(&rcu_state.gp_seq); |
b73de91d JF |
1608 | if (rcu_segcblist_accelerate(&rdp->cblist, gp_seq_req)) |
1609 | ret = rcu_start_this_gp(rnp, rdp, gp_seq_req); | |
6d4b418c PM |
1610 | |
1611 | /* Trace depending on how much we were able to accelerate. */ | |
15fecf89 | 1612 | if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL)) |
9cbc5b97 | 1613 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccWaitCB")); |
6d4b418c | 1614 | else |
9cbc5b97 | 1615 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccReadyCB")); |
48a7639c | 1616 | return ret; |
dc35c893 PM |
1617 | } |
1618 | ||
e44e73ca PM |
1619 | /* |
1620 | * Similar to rcu_accelerate_cbs(), but does not require that the leaf | |
1621 | * rcu_node structure's ->lock be held. It consults the cached value | |
1622 | * of ->gp_seq_needed in the rcu_data structure, and if that indicates | |
1623 | * that a new grace-period request be made, invokes rcu_accelerate_cbs() | |
1624 | * while holding the leaf rcu_node structure's ->lock. | |
1625 | */ | |
c6e09b97 | 1626 | static void rcu_accelerate_cbs_unlocked(struct rcu_node *rnp, |
e44e73ca PM |
1627 | struct rcu_data *rdp) |
1628 | { | |
1629 | unsigned long c; | |
1630 | bool needwake; | |
1631 | ||
1632 | lockdep_assert_irqs_disabled(); | |
c6e09b97 | 1633 | c = rcu_seq_snap(&rcu_state.gp_seq); |
e44e73ca PM |
1634 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { |
1635 | /* Old request still live, so mark recent callbacks. */ | |
1636 | (void)rcu_segcblist_accelerate(&rdp->cblist, c); | |
1637 | return; | |
1638 | } | |
1639 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
02f50142 | 1640 | needwake = rcu_accelerate_cbs(rnp, rdp); |
e44e73ca PM |
1641 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
1642 | if (needwake) | |
532c00c9 | 1643 | rcu_gp_kthread_wake(); |
e44e73ca PM |
1644 | } |
1645 | ||
dc35c893 PM |
1646 | /* |
1647 | * Move any callbacks whose grace period has completed to the | |
1648 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
29365e56 | 1649 | * assign ->gp_seq numbers to any callbacks in the RCU_NEXT_TAIL |
dc35c893 PM |
1650 | * sublist. This function is idempotent, so it does not hurt to |
1651 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1652 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1653 | * |
1654 | * The caller must hold rnp->lock with interrupts disabled. | |
1655 | */ | |
834f56bf | 1656 | static bool rcu_advance_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1657 | { |
a32e01ee | 1658 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1659 | |
15fecf89 PM |
1660 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1661 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1662 | return false; |
dc35c893 PM |
1663 | |
1664 | /* | |
29365e56 | 1665 | * Find all callbacks whose ->gp_seq numbers indicate that they |
dc35c893 PM |
1666 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. |
1667 | */ | |
29365e56 | 1668 | rcu_segcblist_advance(&rdp->cblist, rnp->gp_seq); |
dc35c893 PM |
1669 | |
1670 | /* Classify any remaining callbacks. */ | |
02f50142 | 1671 | return rcu_accelerate_cbs(rnp, rdp); |
dc35c893 PM |
1672 | } |
1673 | ||
d09b62df | 1674 | /* |
ba9fbe95 PM |
1675 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1676 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1677 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1678 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1679 | */ |
c7e48f7b | 1680 | static bool __note_gp_changes(struct rcu_node *rnp, struct rcu_data *rdp) |
d09b62df | 1681 | { |
48a7639c | 1682 | bool ret; |
3563a438 | 1683 | bool need_gp; |
48a7639c | 1684 | |
a32e01ee | 1685 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1686 | |
67e14c1e PM |
1687 | if (rdp->gp_seq == rnp->gp_seq) |
1688 | return false; /* Nothing to do. */ | |
d09b62df | 1689 | |
67e14c1e PM |
1690 | /* Handle the ends of any preceding grace periods first. */ |
1691 | if (rcu_seq_completed_gp(rdp->gp_seq, rnp->gp_seq) || | |
1692 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
834f56bf | 1693 | ret = rcu_advance_cbs(rnp, rdp); /* Advance callbacks. */ |
9cbc5b97 | 1694 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuend")); |
67e14c1e | 1695 | } else { |
02f50142 | 1696 | ret = rcu_accelerate_cbs(rnp, rdp); /* Recent callbacks. */ |
d09b62df | 1697 | } |
398ebe60 | 1698 | |
67e14c1e PM |
1699 | /* Now handle the beginnings of any new-to-this-CPU grace periods. */ |
1700 | if (rcu_seq_new_gp(rdp->gp_seq, rnp->gp_seq) || | |
1701 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
6eaef633 PM |
1702 | /* |
1703 | * If the current grace period is waiting for this CPU, | |
1704 | * set up to detect a quiescent state, otherwise don't | |
1705 | * go looking for one. | |
1706 | */ | |
9cbc5b97 | 1707 | trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, TPS("cpustart")); |
3563a438 PM |
1708 | need_gp = !!(rnp->qsmask & rdp->grpmask); |
1709 | rdp->cpu_no_qs.b.norm = need_gp; | |
3563a438 | 1710 | rdp->core_needs_qs = need_gp; |
6eaef633 PM |
1711 | zero_cpu_stall_ticks(rdp); |
1712 | } | |
67e14c1e | 1713 | rdp->gp_seq = rnp->gp_seq; /* Remember new grace-period state. */ |
3d18469a PM |
1714 | if (ULONG_CMP_GE(rnp->gp_seq_needed, rdp->gp_seq_needed) || rdp->gpwrap) |
1715 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
1716 | WRITE_ONCE(rdp->gpwrap, false); | |
1717 | rcu_gpnum_ovf(rnp, rdp); | |
48a7639c | 1718 | return ret; |
6eaef633 PM |
1719 | } |
1720 | ||
15cabdff | 1721 | static void note_gp_changes(struct rcu_data *rdp) |
6eaef633 PM |
1722 | { |
1723 | unsigned long flags; | |
48a7639c | 1724 | bool needwake; |
6eaef633 PM |
1725 | struct rcu_node *rnp; |
1726 | ||
1727 | local_irq_save(flags); | |
1728 | rnp = rdp->mynode; | |
67e14c1e | 1729 | if ((rdp->gp_seq == rcu_seq_current(&rnp->gp_seq) && |
7d0ae808 | 1730 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ |
2a67e741 | 1731 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
1732 | local_irq_restore(flags); |
1733 | return; | |
1734 | } | |
c7e48f7b | 1735 | needwake = __note_gp_changes(rnp, rdp); |
67c583a7 | 1736 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c | 1737 | if (needwake) |
532c00c9 | 1738 | rcu_gp_kthread_wake(); |
6eaef633 PM |
1739 | } |
1740 | ||
22212332 | 1741 | static void rcu_gp_slow(int delay) |
0f41c0dd PM |
1742 | { |
1743 | if (delay > 0 && | |
22212332 | 1744 | !(rcu_seq_ctr(rcu_state.gp_seq) % |
dee4f422 | 1745 | (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) |
0f41c0dd PM |
1746 | schedule_timeout_uninterruptible(delay); |
1747 | } | |
1748 | ||
b3dbec76 | 1749 | /* |
45fed3e7 | 1750 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 1751 | */ |
0854a05c | 1752 | static bool rcu_gp_init(void) |
b3dbec76 | 1753 | { |
ec2c2976 | 1754 | unsigned long flags; |
0aa04b05 | 1755 | unsigned long oldmask; |
ec2c2976 | 1756 | unsigned long mask; |
b3dbec76 | 1757 | struct rcu_data *rdp; |
336a4f6c | 1758 | struct rcu_node *rnp = rcu_get_root(); |
b3dbec76 | 1759 | |
9cbc5b97 | 1760 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 1761 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 | 1762 | if (!READ_ONCE(rcu_state.gp_flags)) { |
f7be8209 | 1763 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 1764 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1765 | return false; |
f7be8209 | 1766 | } |
9cbc5b97 | 1767 | WRITE_ONCE(rcu_state.gp_flags, 0); /* Clear all flags: New GP. */ |
b3dbec76 | 1768 | |
de8e8730 | 1769 | if (WARN_ON_ONCE(rcu_gp_in_progress())) { |
f7be8209 PM |
1770 | /* |
1771 | * Grace period already in progress, don't start another. | |
1772 | * Not supposed to be able to happen. | |
1773 | */ | |
67c583a7 | 1774 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1775 | return false; |
7fdefc10 PM |
1776 | } |
1777 | ||
7fdefc10 | 1778 | /* Advance to a new grace period and initialize state. */ |
ad3832e9 | 1779 | record_gp_stall_check_time(); |
ff3bb6f4 | 1780 | /* Record GP times before starting GP, hence rcu_seq_start(). */ |
9cbc5b97 PM |
1781 | rcu_seq_start(&rcu_state.gp_seq); |
1782 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("start")); | |
67c583a7 | 1783 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 1784 | |
0aa04b05 PM |
1785 | /* |
1786 | * Apply per-leaf buffered online and offline operations to the | |
1787 | * rcu_node tree. Note that this new grace period need not wait | |
1788 | * for subsequent online CPUs, and that quiescent-state forcing | |
1789 | * will handle subsequent offline CPUs. | |
1790 | */ | |
9cbc5b97 | 1791 | rcu_state.gp_state = RCU_GP_ONOFF; |
aedf4ba9 | 1792 | rcu_for_each_leaf_node(rnp) { |
894d45bb | 1793 | raw_spin_lock(&rcu_state.ofl_lock); |
2a67e741 | 1794 | raw_spin_lock_irq_rcu_node(rnp); |
0aa04b05 PM |
1795 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
1796 | !rnp->wait_blkd_tasks) { | |
1797 | /* Nothing to do on this leaf rcu_node structure. */ | |
67c583a7 | 1798 | raw_spin_unlock_irq_rcu_node(rnp); |
894d45bb | 1799 | raw_spin_unlock(&rcu_state.ofl_lock); |
0aa04b05 PM |
1800 | continue; |
1801 | } | |
1802 | ||
1803 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
1804 | oldmask = rnp->qsmaskinit; | |
1805 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
1806 | ||
1807 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
1808 | if (!oldmask != !rnp->qsmaskinit) { | |
962aff03 PM |
1809 | if (!oldmask) { /* First online CPU for rcu_node. */ |
1810 | if (!rnp->wait_blkd_tasks) /* Ever offline? */ | |
1811 | rcu_init_new_rnp(rnp); | |
1812 | } else if (rcu_preempt_has_tasks(rnp)) { | |
1813 | rnp->wait_blkd_tasks = true; /* blocked tasks */ | |
1814 | } else { /* Last offline CPU and can propagate. */ | |
0aa04b05 | 1815 | rcu_cleanup_dead_rnp(rnp); |
962aff03 | 1816 | } |
0aa04b05 PM |
1817 | } |
1818 | ||
1819 | /* | |
1820 | * If all waited-on tasks from prior grace period are | |
1821 | * done, and if all this rcu_node structure's CPUs are | |
1822 | * still offline, propagate up the rcu_node tree and | |
1823 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
1824 | * rcu_node structure's CPUs has since come back online, | |
962aff03 | 1825 | * simply clear ->wait_blkd_tasks. |
0aa04b05 PM |
1826 | */ |
1827 | if (rnp->wait_blkd_tasks && | |
962aff03 | 1828 | (!rcu_preempt_has_tasks(rnp) || rnp->qsmaskinit)) { |
0aa04b05 | 1829 | rnp->wait_blkd_tasks = false; |
962aff03 PM |
1830 | if (!rnp->qsmaskinit) |
1831 | rcu_cleanup_dead_rnp(rnp); | |
0aa04b05 PM |
1832 | } |
1833 | ||
67c583a7 | 1834 | raw_spin_unlock_irq_rcu_node(rnp); |
894d45bb | 1835 | raw_spin_unlock(&rcu_state.ofl_lock); |
0aa04b05 | 1836 | } |
22212332 | 1837 | rcu_gp_slow(gp_preinit_delay); /* Races with CPU hotplug. */ |
7fdefc10 PM |
1838 | |
1839 | /* | |
1840 | * Set the quiescent-state-needed bits in all the rcu_node | |
9cbc5b97 PM |
1841 | * structures for all currently online CPUs in breadth-first |
1842 | * order, starting from the root rcu_node structure, relying on the | |
1843 | * layout of the tree within the rcu_state.node[] array. Note that | |
1844 | * other CPUs will access only the leaves of the hierarchy, thus | |
1845 | * seeing that no grace period is in progress, at least until the | |
1846 | * corresponding leaf node has been initialized. | |
7fdefc10 PM |
1847 | * |
1848 | * The grace period cannot complete until the initialization | |
1849 | * process finishes, because this kthread handles both. | |
1850 | */ | |
9cbc5b97 | 1851 | rcu_state.gp_state = RCU_GP_INIT; |
aedf4ba9 | 1852 | rcu_for_each_node_breadth_first(rnp) { |
22212332 | 1853 | rcu_gp_slow(gp_init_delay); |
ec2c2976 | 1854 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
da1df50d | 1855 | rdp = this_cpu_ptr(&rcu_data); |
81ab59a3 | 1856 | rcu_preempt_check_blocked_tasks(rnp); |
7fdefc10 | 1857 | rnp->qsmask = rnp->qsmaskinit; |
9cbc5b97 | 1858 | WRITE_ONCE(rnp->gp_seq, rcu_state.gp_seq); |
7fdefc10 | 1859 | if (rnp == rdp->mynode) |
c7e48f7b | 1860 | (void)__note_gp_changes(rnp, rdp); |
7fdefc10 | 1861 | rcu_preempt_boost_start_gp(rnp); |
9cbc5b97 | 1862 | trace_rcu_grace_period_init(rcu_state.name, rnp->gp_seq, |
7fdefc10 PM |
1863 | rnp->level, rnp->grplo, |
1864 | rnp->grphi, rnp->qsmask); | |
ec2c2976 PM |
1865 | /* Quiescent states for tasks on any now-offline CPUs. */ |
1866 | mask = rnp->qsmask & ~rnp->qsmaskinitnext; | |
f2e2df59 | 1867 | rnp->rcu_gp_init_mask = mask; |
ec2c2976 | 1868 | if ((mask || rnp->wait_blkd_tasks) && rcu_is_leaf_node(rnp)) |
b50912d0 | 1869 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
ec2c2976 PM |
1870 | else |
1871 | raw_spin_unlock_irq_rcu_node(rnp); | |
cee43939 | 1872 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1873 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
7fdefc10 | 1874 | } |
b3dbec76 | 1875 | |
45fed3e7 | 1876 | return true; |
7fdefc10 | 1877 | } |
b3dbec76 | 1878 | |
b9a425cf | 1879 | /* |
b3dae109 | 1880 | * Helper function for swait_event_idle_exclusive() wakeup at force-quiescent-state |
d5374226 | 1881 | * time. |
b9a425cf | 1882 | */ |
0854a05c | 1883 | static bool rcu_gp_fqs_check_wake(int *gfp) |
b9a425cf | 1884 | { |
336a4f6c | 1885 | struct rcu_node *rnp = rcu_get_root(); |
b9a425cf PM |
1886 | |
1887 | /* Someone like call_rcu() requested a force-quiescent-state scan. */ | |
0854a05c | 1888 | *gfp = READ_ONCE(rcu_state.gp_flags); |
b9a425cf PM |
1889 | if (*gfp & RCU_GP_FLAG_FQS) |
1890 | return true; | |
1891 | ||
1892 | /* The current grace period has completed. */ | |
1893 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) | |
1894 | return true; | |
1895 | ||
1896 | return false; | |
1897 | } | |
1898 | ||
4cdfc175 PM |
1899 | /* |
1900 | * Do one round of quiescent-state forcing. | |
1901 | */ | |
0854a05c | 1902 | static void rcu_gp_fqs(bool first_time) |
4cdfc175 | 1903 | { |
336a4f6c | 1904 | struct rcu_node *rnp = rcu_get_root(); |
4cdfc175 | 1905 | |
9cbc5b97 PM |
1906 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
1907 | rcu_state.n_force_qs++; | |
77f81fe0 | 1908 | if (first_time) { |
4cdfc175 | 1909 | /* Collect dyntick-idle snapshots. */ |
e9ecb780 | 1910 | force_qs_rnp(dyntick_save_progress_counter); |
4cdfc175 PM |
1911 | } else { |
1912 | /* Handle dyntick-idle and offline CPUs. */ | |
e9ecb780 | 1913 | force_qs_rnp(rcu_implicit_dynticks_qs); |
4cdfc175 PM |
1914 | } |
1915 | /* Clear flag to prevent immediate re-entry. */ | |
9cbc5b97 | 1916 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 1917 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 PM |
1918 | WRITE_ONCE(rcu_state.gp_flags, |
1919 | READ_ONCE(rcu_state.gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 1920 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 1921 | } |
4cdfc175 PM |
1922 | } |
1923 | ||
c3854a05 PM |
1924 | /* |
1925 | * Loop doing repeated quiescent-state forcing until the grace period ends. | |
1926 | */ | |
1927 | static void rcu_gp_fqs_loop(void) | |
1928 | { | |
1929 | bool first_gp_fqs; | |
1930 | int gf; | |
1931 | unsigned long j; | |
1932 | int ret; | |
1933 | struct rcu_node *rnp = rcu_get_root(); | |
1934 | ||
1935 | first_gp_fqs = true; | |
c06aed0e | 1936 | j = READ_ONCE(jiffies_till_first_fqs); |
c3854a05 PM |
1937 | ret = 0; |
1938 | for (;;) { | |
1939 | if (!ret) { | |
1940 | rcu_state.jiffies_force_qs = jiffies + j; | |
1941 | WRITE_ONCE(rcu_state.jiffies_kick_kthreads, | |
1942 | jiffies + 3 * j); | |
1943 | } | |
1944 | trace_rcu_grace_period(rcu_state.name, | |
1945 | READ_ONCE(rcu_state.gp_seq), | |
1946 | TPS("fqswait")); | |
1947 | rcu_state.gp_state = RCU_GP_WAIT_FQS; | |
1948 | ret = swait_event_idle_timeout_exclusive( | |
1949 | rcu_state.gp_wq, rcu_gp_fqs_check_wake(&gf), j); | |
1950 | rcu_state.gp_state = RCU_GP_DOING_FQS; | |
1951 | /* Locking provides needed memory barriers. */ | |
1952 | /* If grace period done, leave loop. */ | |
1953 | if (!READ_ONCE(rnp->qsmask) && | |
1954 | !rcu_preempt_blocked_readers_cgp(rnp)) | |
1955 | break; | |
1956 | /* If time for quiescent-state forcing, do it. */ | |
1957 | if (ULONG_CMP_GE(jiffies, rcu_state.jiffies_force_qs) || | |
1958 | (gf & RCU_GP_FLAG_FQS)) { | |
1959 | trace_rcu_grace_period(rcu_state.name, | |
1960 | READ_ONCE(rcu_state.gp_seq), | |
1961 | TPS("fqsstart")); | |
1962 | rcu_gp_fqs(first_gp_fqs); | |
1963 | first_gp_fqs = false; | |
1964 | trace_rcu_grace_period(rcu_state.name, | |
1965 | READ_ONCE(rcu_state.gp_seq), | |
1966 | TPS("fqsend")); | |
1967 | cond_resched_tasks_rcu_qs(); | |
1968 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1969 | ret = 0; /* Force full wait till next FQS. */ | |
c06aed0e | 1970 | j = READ_ONCE(jiffies_till_next_fqs); |
c3854a05 PM |
1971 | } else { |
1972 | /* Deal with stray signal. */ | |
1973 | cond_resched_tasks_rcu_qs(); | |
1974 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1975 | WARN_ON(signal_pending(current)); | |
1976 | trace_rcu_grace_period(rcu_state.name, | |
1977 | READ_ONCE(rcu_state.gp_seq), | |
1978 | TPS("fqswaitsig")); | |
1979 | ret = 1; /* Keep old FQS timing. */ | |
1980 | j = jiffies; | |
1981 | if (time_after(jiffies, rcu_state.jiffies_force_qs)) | |
1982 | j = 1; | |
1983 | else | |
1984 | j = rcu_state.jiffies_force_qs - j; | |
1985 | } | |
1986 | } | |
1987 | } | |
1988 | ||
7fdefc10 PM |
1989 | /* |
1990 | * Clean up after the old grace period. | |
1991 | */ | |
0854a05c | 1992 | static void rcu_gp_cleanup(void) |
7fdefc10 PM |
1993 | { |
1994 | unsigned long gp_duration; | |
48a7639c | 1995 | bool needgp = false; |
de30ad51 | 1996 | unsigned long new_gp_seq; |
7fdefc10 | 1997 | struct rcu_data *rdp; |
336a4f6c | 1998 | struct rcu_node *rnp = rcu_get_root(); |
abedf8e2 | 1999 | struct swait_queue_head *sq; |
b3dbec76 | 2000 | |
9cbc5b97 | 2001 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 2002 | raw_spin_lock_irq_rcu_node(rnp); |
c51d7b5e PM |
2003 | rcu_state.gp_end = jiffies; |
2004 | gp_duration = rcu_state.gp_end - rcu_state.gp_start; | |
9cbc5b97 PM |
2005 | if (gp_duration > rcu_state.gp_max) |
2006 | rcu_state.gp_max = gp_duration; | |
b3dbec76 | 2007 | |
7fdefc10 PM |
2008 | /* |
2009 | * We know the grace period is complete, but to everyone else | |
2010 | * it appears to still be ongoing. But it is also the case | |
2011 | * that to everyone else it looks like there is nothing that | |
2012 | * they can do to advance the grace period. It is therefore | |
2013 | * safe for us to drop the lock in order to mark the grace | |
2014 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 2015 | */ |
67c583a7 | 2016 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 2017 | |
5d4b8659 | 2018 | /* |
ff3bb6f4 PM |
2019 | * Propagate new ->gp_seq value to rcu_node structures so that |
2020 | * other CPUs don't have to wait until the start of the next grace | |
2021 | * period to process their callbacks. This also avoids some nasty | |
2022 | * RCU grace-period initialization races by forcing the end of | |
2023 | * the current grace period to be completely recorded in all of | |
2024 | * the rcu_node structures before the beginning of the next grace | |
2025 | * period is recorded in any of the rcu_node structures. | |
5d4b8659 | 2026 | */ |
9cbc5b97 | 2027 | new_gp_seq = rcu_state.gp_seq; |
de30ad51 | 2028 | rcu_seq_end(&new_gp_seq); |
aedf4ba9 | 2029 | rcu_for_each_node_breadth_first(rnp) { |
2a67e741 | 2030 | raw_spin_lock_irq_rcu_node(rnp); |
4bc8d555 | 2031 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 2032 | dump_blkd_tasks(rnp, 10); |
5c60d25f | 2033 | WARN_ON_ONCE(rnp->qsmask); |
de30ad51 | 2034 | WRITE_ONCE(rnp->gp_seq, new_gp_seq); |
da1df50d | 2035 | rdp = this_cpu_ptr(&rcu_data); |
b11cc576 | 2036 | if (rnp == rdp->mynode) |
c7e48f7b | 2037 | needgp = __note_gp_changes(rnp, rdp) || needgp; |
78e4bc34 | 2038 | /* smp_mb() provided by prior unlock-lock pair. */ |
3481f2ea | 2039 | needgp = rcu_future_gp_cleanup(rnp) || needgp; |
065bb78c | 2040 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 2041 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 2042 | rcu_nocb_gp_cleanup(sq); |
cee43939 | 2043 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 2044 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
22212332 | 2045 | rcu_gp_slow(gp_cleanup_delay); |
7fdefc10 | 2046 | } |
336a4f6c | 2047 | rnp = rcu_get_root(); |
9cbc5b97 | 2048 | raw_spin_lock_irq_rcu_node(rnp); /* GP before ->gp_seq update. */ |
7fdefc10 | 2049 | |
0a89e5a4 | 2050 | /* Declare grace period done, trace first to use old GP number. */ |
9cbc5b97 | 2051 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("end")); |
0a89e5a4 | 2052 | rcu_seq_end(&rcu_state.gp_seq); |
9cbc5b97 | 2053 | rcu_state.gp_state = RCU_GP_IDLE; |
fb31340f | 2054 | /* Check for GP requests since above loop. */ |
da1df50d | 2055 | rdp = this_cpu_ptr(&rcu_data); |
5b55072f | 2056 | if (!needgp && ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed)) { |
abd13fdd | 2057 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq_needed, |
41e80595 | 2058 | TPS("CleanupMore")); |
fb31340f PM |
2059 | needgp = true; |
2060 | } | |
48a7639c | 2061 | /* Advance CBs to reduce false positives below. */ |
02f50142 | 2062 | if (!rcu_accelerate_cbs(rnp, rdp) && needgp) { |
9cbc5b97 PM |
2063 | WRITE_ONCE(rcu_state.gp_flags, RCU_GP_FLAG_INIT); |
2064 | rcu_state.gp_req_activity = jiffies; | |
2065 | trace_rcu_grace_period(rcu_state.name, | |
2066 | READ_ONCE(rcu_state.gp_seq), | |
bb311ecc | 2067 | TPS("newreq")); |
18390aea | 2068 | } else { |
9cbc5b97 PM |
2069 | WRITE_ONCE(rcu_state.gp_flags, |
2070 | rcu_state.gp_flags & RCU_GP_FLAG_INIT); | |
bb311ecc | 2071 | } |
67c583a7 | 2072 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 PM |
2073 | } |
2074 | ||
2075 | /* | |
2076 | * Body of kthread that handles grace periods. | |
2077 | */ | |
0854a05c | 2078 | static int __noreturn rcu_gp_kthread(void *unused) |
7fdefc10 | 2079 | { |
5871968d | 2080 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
2081 | for (;;) { |
2082 | ||
2083 | /* Handle grace-period start. */ | |
2084 | for (;;) { | |
9cbc5b97 PM |
2085 | trace_rcu_grace_period(rcu_state.name, |
2086 | READ_ONCE(rcu_state.gp_seq), | |
63c4db78 | 2087 | TPS("reqwait")); |
9cbc5b97 PM |
2088 | rcu_state.gp_state = RCU_GP_WAIT_GPS; |
2089 | swait_event_idle_exclusive(rcu_state.gp_wq, | |
2090 | READ_ONCE(rcu_state.gp_flags) & | |
2091 | RCU_GP_FLAG_INIT); | |
2092 | rcu_state.gp_state = RCU_GP_DONE_GPS; | |
78e4bc34 | 2093 | /* Locking provides needed memory barrier. */ |
0854a05c | 2094 | if (rcu_gp_init()) |
7fdefc10 | 2095 | break; |
cee43939 | 2096 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 2097 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
73a860cd | 2098 | WARN_ON(signal_pending(current)); |
9cbc5b97 PM |
2099 | trace_rcu_grace_period(rcu_state.name, |
2100 | READ_ONCE(rcu_state.gp_seq), | |
63c4db78 | 2101 | TPS("reqwaitsig")); |
7fdefc10 | 2102 | } |
cabc49c1 | 2103 | |
4cdfc175 | 2104 | /* Handle quiescent-state forcing. */ |
c3854a05 | 2105 | rcu_gp_fqs_loop(); |
4cdfc175 PM |
2106 | |
2107 | /* Handle grace-period end. */ | |
9cbc5b97 | 2108 | rcu_state.gp_state = RCU_GP_CLEANUP; |
0854a05c | 2109 | rcu_gp_cleanup(); |
9cbc5b97 | 2110 | rcu_state.gp_state = RCU_GP_CLEANED; |
b3dbec76 | 2111 | } |
b3dbec76 PM |
2112 | } |
2113 | ||
f41d911f | 2114 | /* |
49918a54 PM |
2115 | * Report a full set of quiescent states to the rcu_state data structure. |
2116 | * Invoke rcu_gp_kthread_wake() to awaken the grace-period kthread if | |
2117 | * another grace period is required. Whether we wake the grace-period | |
2118 | * kthread or it awakens itself for the next round of quiescent-state | |
2119 | * forcing, that kthread will clean up after the just-completed grace | |
2120 | * period. Note that the caller must hold rnp->lock, which is released | |
2121 | * before return. | |
f41d911f | 2122 | */ |
aff4e9ed | 2123 | static void rcu_report_qs_rsp(unsigned long flags) |
336a4f6c | 2124 | __releases(rcu_get_root()->lock) |
f41d911f | 2125 | { |
336a4f6c | 2126 | raw_lockdep_assert_held_rcu_node(rcu_get_root()); |
de8e8730 | 2127 | WARN_ON_ONCE(!rcu_gp_in_progress()); |
9cbc5b97 PM |
2128 | WRITE_ONCE(rcu_state.gp_flags, |
2129 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
336a4f6c | 2130 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(), flags); |
532c00c9 | 2131 | rcu_gp_kthread_wake(); |
f41d911f PM |
2132 | } |
2133 | ||
64db4cff | 2134 | /* |
d3f6bad3 PM |
2135 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
2136 | * Allows quiescent states for a group of CPUs to be reported at one go | |
2137 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
2138 | * must be represented by the same rcu_node structure (which need not be a |
2139 | * leaf rcu_node structure, though it often will be). The gps parameter | |
2140 | * is the grace-period snapshot, which means that the quiescent states | |
c9a24e2d | 2141 | * are valid only if rnp->gp_seq is equal to gps. That structure's lock |
654e9533 | 2142 | * must be held upon entry, and it is released before return. |
ec2c2976 PM |
2143 | * |
2144 | * As a special case, if mask is zero, the bit-already-cleared check is | |
2145 | * disabled. This allows propagating quiescent state due to resumed tasks | |
2146 | * during grace-period initialization. | |
64db4cff | 2147 | */ |
b50912d0 PM |
2148 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
2149 | unsigned long gps, unsigned long flags) | |
64db4cff PM |
2150 | __releases(rnp->lock) |
2151 | { | |
654e9533 | 2152 | unsigned long oldmask = 0; |
28ecd580 PM |
2153 | struct rcu_node *rnp_c; |
2154 | ||
a32e01ee | 2155 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 2156 | |
64db4cff PM |
2157 | /* Walk up the rcu_node hierarchy. */ |
2158 | for (;;) { | |
ec2c2976 | 2159 | if ((!(rnp->qsmask & mask) && mask) || rnp->gp_seq != gps) { |
64db4cff | 2160 | |
654e9533 PM |
2161 | /* |
2162 | * Our bit has already been cleared, or the | |
2163 | * relevant grace period is already over, so done. | |
2164 | */ | |
67c583a7 | 2165 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2166 | return; |
2167 | } | |
654e9533 | 2168 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
5b4c11d5 | 2169 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp) && |
2dee9404 | 2170 | rcu_preempt_blocked_readers_cgp(rnp)); |
64db4cff | 2171 | rnp->qsmask &= ~mask; |
67a0edbf | 2172 | trace_rcu_quiescent_state_report(rcu_state.name, rnp->gp_seq, |
d4c08f2a PM |
2173 | mask, rnp->qsmask, rnp->level, |
2174 | rnp->grplo, rnp->grphi, | |
2175 | !!rnp->gp_tasks); | |
27f4d280 | 2176 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
2177 | |
2178 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 2179 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2180 | return; |
2181 | } | |
d43a5d32 | 2182 | rnp->completedqs = rnp->gp_seq; |
64db4cff PM |
2183 | mask = rnp->grpmask; |
2184 | if (rnp->parent == NULL) { | |
2185 | ||
2186 | /* No more levels. Exit loop holding root lock. */ | |
2187 | ||
2188 | break; | |
2189 | } | |
67c583a7 | 2190 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 2191 | rnp_c = rnp; |
64db4cff | 2192 | rnp = rnp->parent; |
2a67e741 | 2193 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
654e9533 | 2194 | oldmask = rnp_c->qsmask; |
64db4cff PM |
2195 | } |
2196 | ||
2197 | /* | |
2198 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 2199 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 2200 | * to clean up and start the next grace period if one is needed. |
64db4cff | 2201 | */ |
aff4e9ed | 2202 | rcu_report_qs_rsp(flags); /* releases rnp->lock. */ |
64db4cff PM |
2203 | } |
2204 | ||
cc99a310 PM |
2205 | /* |
2206 | * Record a quiescent state for all tasks that were previously queued | |
2207 | * on the specified rcu_node structure and that were blocking the current | |
49918a54 | 2208 | * RCU grace period. The caller must hold the corresponding rnp->lock with |
cc99a310 PM |
2209 | * irqs disabled, and this lock is released upon return, but irqs remain |
2210 | * disabled. | |
2211 | */ | |
17a8212b | 2212 | static void __maybe_unused |
139ad4da | 2213 | rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
cc99a310 PM |
2214 | __releases(rnp->lock) |
2215 | { | |
654e9533 | 2216 | unsigned long gps; |
cc99a310 PM |
2217 | unsigned long mask; |
2218 | struct rcu_node *rnp_p; | |
2219 | ||
a32e01ee | 2220 | raw_lockdep_assert_held_rcu_node(rnp); |
45975c7d | 2221 | if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPT)) || |
c74859d1 PM |
2222 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)) || |
2223 | rnp->qsmask != 0) { | |
67c583a7 | 2224 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
2225 | return; /* Still need more quiescent states! */ |
2226 | } | |
2227 | ||
77cfc7bf | 2228 | rnp->completedqs = rnp->gp_seq; |
cc99a310 PM |
2229 | rnp_p = rnp->parent; |
2230 | if (rnp_p == NULL) { | |
2231 | /* | |
a77da14c PM |
2232 | * Only one rcu_node structure in the tree, so don't |
2233 | * try to report up to its nonexistent parent! | |
cc99a310 | 2234 | */ |
aff4e9ed | 2235 | rcu_report_qs_rsp(flags); |
cc99a310 PM |
2236 | return; |
2237 | } | |
2238 | ||
c9a24e2d PM |
2239 | /* Report up the rest of the hierarchy, tracking current ->gp_seq. */ |
2240 | gps = rnp->gp_seq; | |
cc99a310 | 2241 | mask = rnp->grpmask; |
67c583a7 | 2242 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 2243 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
b50912d0 | 2244 | rcu_report_qs_rnp(mask, rnp_p, gps, flags); |
cc99a310 PM |
2245 | } |
2246 | ||
64db4cff | 2247 | /* |
d3f6bad3 | 2248 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 2249 | * structure. This must be called from the specified CPU. |
64db4cff PM |
2250 | */ |
2251 | static void | |
33085c46 | 2252 | rcu_report_qs_rdp(int cpu, struct rcu_data *rdp) |
64db4cff PM |
2253 | { |
2254 | unsigned long flags; | |
2255 | unsigned long mask; | |
48a7639c | 2256 | bool needwake; |
64db4cff PM |
2257 | struct rcu_node *rnp; |
2258 | ||
2259 | rnp = rdp->mynode; | |
2a67e741 | 2260 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
c9a24e2d PM |
2261 | if (rdp->cpu_no_qs.b.norm || rdp->gp_seq != rnp->gp_seq || |
2262 | rdp->gpwrap) { | |
64db4cff PM |
2263 | |
2264 | /* | |
e4cc1f22 PM |
2265 | * The grace period in which this quiescent state was |
2266 | * recorded has ended, so don't report it upwards. | |
2267 | * We will instead need a new quiescent state that lies | |
2268 | * within the current grace period. | |
64db4cff | 2269 | */ |
5b74c458 | 2270 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
67c583a7 | 2271 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2272 | return; |
2273 | } | |
2274 | mask = rdp->grpmask; | |
2275 | if ((rnp->qsmask & mask) == 0) { | |
67c583a7 | 2276 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2277 | } else { |
bb53e416 | 2278 | rdp->core_needs_qs = false; |
64db4cff PM |
2279 | |
2280 | /* | |
2281 | * This GP can't end until cpu checks in, so all of our | |
2282 | * callbacks can be processed during the next GP. | |
2283 | */ | |
02f50142 | 2284 | needwake = rcu_accelerate_cbs(rnp, rdp); |
64db4cff | 2285 | |
b50912d0 | 2286 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
654e9533 | 2287 | /* ^^^ Released rnp->lock */ |
48a7639c | 2288 | if (needwake) |
532c00c9 | 2289 | rcu_gp_kthread_wake(); |
64db4cff PM |
2290 | } |
2291 | } | |
2292 | ||
2293 | /* | |
2294 | * Check to see if there is a new grace period of which this CPU | |
2295 | * is not yet aware, and if so, set up local rcu_data state for it. | |
2296 | * Otherwise, see if this CPU has just passed through its first | |
2297 | * quiescent state for this grace period, and record that fact if so. | |
2298 | */ | |
2299 | static void | |
8087d3e3 | 2300 | rcu_check_quiescent_state(struct rcu_data *rdp) |
64db4cff | 2301 | { |
05eb552b | 2302 | /* Check for grace-period ends and beginnings. */ |
15cabdff | 2303 | note_gp_changes(rdp); |
64db4cff PM |
2304 | |
2305 | /* | |
2306 | * Does this CPU still need to do its part for current grace period? | |
2307 | * If no, return and let the other CPUs do their part as well. | |
2308 | */ | |
97c668b8 | 2309 | if (!rdp->core_needs_qs) |
64db4cff PM |
2310 | return; |
2311 | ||
2312 | /* | |
2313 | * Was there a quiescent state since the beginning of the grace | |
2314 | * period? If no, then exit and wait for the next call. | |
2315 | */ | |
3a19b46a | 2316 | if (rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2317 | return; |
2318 | ||
d3f6bad3 PM |
2319 | /* |
2320 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
2321 | * judge of that). | |
2322 | */ | |
33085c46 | 2323 | rcu_report_qs_rdp(rdp->cpu, rdp); |
64db4cff PM |
2324 | } |
2325 | ||
b1420f1c | 2326 | /* |
780cd590 PM |
2327 | * Near the end of the offline process. Trace the fact that this CPU |
2328 | * is going offline. | |
b1420f1c | 2329 | */ |
780cd590 | 2330 | int rcutree_dying_cpu(unsigned int cpu) |
b1420f1c | 2331 | { |
477351f7 | 2332 | RCU_TRACE(bool blkd;) |
da1df50d | 2333 | RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(&rcu_data);) |
88a4976d | 2334 | RCU_TRACE(struct rcu_node *rnp = rdp->mynode;) |
b1420f1c | 2335 | |
ea46351c | 2336 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 2337 | return 0; |
ea46351c | 2338 | |
477351f7 | 2339 | RCU_TRACE(blkd = !!(rnp->qsmask & rdp->grpmask);) |
780cd590 | 2340 | trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, |
477351f7 | 2341 | blkd ? TPS("cpuofl") : TPS("cpuofl-bgp")); |
780cd590 | 2342 | return 0; |
64db4cff PM |
2343 | } |
2344 | ||
8af3a5e7 PM |
2345 | /* |
2346 | * All CPUs for the specified rcu_node structure have gone offline, | |
2347 | * and all tasks that were preempted within an RCU read-side critical | |
2348 | * section while running on one of those CPUs have since exited their RCU | |
2349 | * read-side critical section. Some other CPU is reporting this fact with | |
2350 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2351 | * This function therefore goes up the tree of rcu_node structures, | |
2352 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2353 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
c50cbe53 | 2354 | * updated. |
8af3a5e7 PM |
2355 | * |
2356 | * This function does check that the specified rcu_node structure has | |
2357 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2358 | * prematurely. That said, invoking it after the fact will cost you | |
2359 | * a needless lock acquisition. So once it has done its work, don't | |
2360 | * invoke it again. | |
2361 | */ | |
2362 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2363 | { | |
2364 | long mask; | |
2365 | struct rcu_node *rnp = rnp_leaf; | |
2366 | ||
962aff03 | 2367 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
ea46351c | 2368 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
962aff03 PM |
2369 | WARN_ON_ONCE(rnp_leaf->qsmaskinit) || |
2370 | WARN_ON_ONCE(rcu_preempt_has_tasks(rnp_leaf))) | |
8af3a5e7 PM |
2371 | return; |
2372 | for (;;) { | |
2373 | mask = rnp->grpmask; | |
2374 | rnp = rnp->parent; | |
2375 | if (!rnp) | |
2376 | break; | |
2a67e741 | 2377 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2378 | rnp->qsmaskinit &= ~mask; |
962aff03 PM |
2379 | /* Between grace periods, so better already be zero! */ |
2380 | WARN_ON_ONCE(rnp->qsmask); | |
8af3a5e7 | 2381 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2382 | raw_spin_unlock_rcu_node(rnp); |
2383 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2384 | return; |
2385 | } | |
67c583a7 | 2386 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2387 | } |
2388 | } | |
2389 | ||
64db4cff | 2390 | /* |
e5601400 | 2391 | * The CPU has been completely removed, and some other CPU is reporting |
a58163d8 PM |
2392 | * this fact from process context. Do the remainder of the cleanup. |
2393 | * There can only be one CPU hotplug operation at a time, so no need for | |
2394 | * explicit locking. | |
64db4cff | 2395 | */ |
780cd590 | 2396 | int rcutree_dead_cpu(unsigned int cpu) |
64db4cff | 2397 | { |
da1df50d | 2398 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
b1420f1c | 2399 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2400 | |
ea46351c | 2401 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 2402 | return 0; |
ea46351c | 2403 | |
2036d94a | 2404 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2405 | rcu_boost_kthread_setaffinity(rnp, -1); |
780cd590 PM |
2406 | /* Do any needed no-CB deferred wakeups from this CPU. */ |
2407 | do_nocb_deferred_wakeup(per_cpu_ptr(&rcu_data, cpu)); | |
2408 | return 0; | |
64db4cff PM |
2409 | } |
2410 | ||
64db4cff PM |
2411 | /* |
2412 | * Invoke any RCU callbacks that have made it to the end of their grace | |
2413 | * period. Thottle as specified by rdp->blimit. | |
2414 | */ | |
5bb5d09c | 2415 | static void rcu_do_batch(struct rcu_data *rdp) |
64db4cff PM |
2416 | { |
2417 | unsigned long flags; | |
15fecf89 PM |
2418 | struct rcu_head *rhp; |
2419 | struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl); | |
2420 | long bl, count; | |
64db4cff | 2421 | |
dc35c893 | 2422 | /* If no callbacks are ready, just return. */ |
15fecf89 | 2423 | if (!rcu_segcblist_ready_cbs(&rdp->cblist)) { |
3c779dfe | 2424 | trace_rcu_batch_start(rcu_state.name, |
15fecf89 PM |
2425 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2426 | rcu_segcblist_n_cbs(&rdp->cblist), 0); | |
3c779dfe | 2427 | trace_rcu_batch_end(rcu_state.name, 0, |
15fecf89 | 2428 | !rcu_segcblist_empty(&rdp->cblist), |
4968c300 PM |
2429 | need_resched(), is_idle_task(current), |
2430 | rcu_is_callbacks_kthread()); | |
64db4cff | 2431 | return; |
29c00b4a | 2432 | } |
64db4cff PM |
2433 | |
2434 | /* | |
2435 | * Extract the list of ready callbacks, disabling to prevent | |
15fecf89 PM |
2436 | * races with call_rcu() from interrupt handlers. Leave the |
2437 | * callback counts, as rcu_barrier() needs to be conservative. | |
64db4cff PM |
2438 | */ |
2439 | local_irq_save(flags); | |
8146c4e2 | 2440 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 2441 | bl = rdp->blimit; |
3c779dfe PM |
2442 | trace_rcu_batch_start(rcu_state.name, |
2443 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
15fecf89 PM |
2444 | rcu_segcblist_n_cbs(&rdp->cblist), bl); |
2445 | rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl); | |
64db4cff PM |
2446 | local_irq_restore(flags); |
2447 | ||
2448 | /* Invoke callbacks. */ | |
15fecf89 PM |
2449 | rhp = rcu_cblist_dequeue(&rcl); |
2450 | for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) { | |
2451 | debug_rcu_head_unqueue(rhp); | |
3c779dfe | 2452 | if (__rcu_reclaim(rcu_state.name, rhp)) |
15fecf89 PM |
2453 | rcu_cblist_dequeued_lazy(&rcl); |
2454 | /* | |
2455 | * Stop only if limit reached and CPU has something to do. | |
2456 | * Note: The rcl structure counts down from zero. | |
2457 | */ | |
4b27f20b | 2458 | if (-rcl.len >= bl && |
dff1672d PM |
2459 | (need_resched() || |
2460 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
2461 | break; |
2462 | } | |
2463 | ||
2464 | local_irq_save(flags); | |
4b27f20b | 2465 | count = -rcl.len; |
3c779dfe | 2466 | trace_rcu_batch_end(rcu_state.name, count, !!rcl.head, need_resched(), |
8ef0f37e | 2467 | is_idle_task(current), rcu_is_callbacks_kthread()); |
64db4cff | 2468 | |
15fecf89 PM |
2469 | /* Update counts and requeue any remaining callbacks. */ |
2470 | rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl); | |
b1420f1c | 2471 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
15fecf89 | 2472 | rcu_segcblist_insert_count(&rdp->cblist, &rcl); |
64db4cff PM |
2473 | |
2474 | /* Reinstate batch limit if we have worked down the excess. */ | |
15fecf89 PM |
2475 | count = rcu_segcblist_n_cbs(&rdp->cblist); |
2476 | if (rdp->blimit == LONG_MAX && count <= qlowmark) | |
64db4cff PM |
2477 | rdp->blimit = blimit; |
2478 | ||
37c72e56 | 2479 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
15fecf89 | 2480 | if (count == 0 && rdp->qlen_last_fqs_check != 0) { |
37c72e56 | 2481 | rdp->qlen_last_fqs_check = 0; |
3c779dfe | 2482 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
15fecf89 PM |
2483 | } else if (count < rdp->qlen_last_fqs_check - qhimark) |
2484 | rdp->qlen_last_fqs_check = count; | |
efd88b02 PM |
2485 | |
2486 | /* | |
2487 | * The following usually indicates a double call_rcu(). To track | |
2488 | * this down, try building with CONFIG_DEBUG_OBJECTS_RCU_HEAD=y. | |
2489 | */ | |
15fecf89 | 2490 | WARN_ON_ONCE(rcu_segcblist_empty(&rdp->cblist) != (count == 0)); |
37c72e56 | 2491 | |
64db4cff PM |
2492 | local_irq_restore(flags); |
2493 | ||
e0f23060 | 2494 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
15fecf89 | 2495 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2496 | invoke_rcu_core(); |
64db4cff PM |
2497 | } |
2498 | ||
2499 | /* | |
2500 | * Check to see if this CPU is in a non-context-switch quiescent state | |
2501 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 2502 | * Also schedule RCU core processing. |
64db4cff | 2503 | * |
9b2e4f18 | 2504 | * This function must be called from hardirq context. It is normally |
5403d367 | 2505 | * invoked from the scheduling-clock interrupt. |
64db4cff | 2506 | */ |
c3377c2d | 2507 | void rcu_check_callbacks(int user) |
64db4cff | 2508 | { |
f7f7bac9 | 2509 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
4e95020c | 2510 | raw_cpu_inc(rcu_data.ticks_this_gp); |
92aa39e9 | 2511 | /* The load-acquire pairs with the store-release setting to true. */ |
2dba13f0 | 2512 | if (smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
92aa39e9 | 2513 | /* Idle and userspace execution already are quiescent states. */ |
a0ef9ec2 | 2514 | if (!rcu_is_cpu_rrupt_from_idle() && !user) { |
92aa39e9 PM |
2515 | set_tsk_need_resched(current); |
2516 | set_preempt_need_resched(); | |
2517 | } | |
2dba13f0 | 2518 | __this_cpu_write(rcu_data.rcu_urgent_qs, false); |
64db4cff | 2519 | } |
45975c7d | 2520 | rcu_flavor_check_callbacks(user); |
e3950ecd | 2521 | if (rcu_pending()) |
a46e0899 | 2522 | invoke_rcu_core(); |
07f27570 | 2523 | |
f7f7bac9 | 2524 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2525 | } |
2526 | ||
64db4cff PM |
2527 | /* |
2528 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
2529 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
2530 | * Also initiate boosting for any threads blocked on the root rcu_node. |
2531 | * | |
ee47eb9f | 2532 | * The caller must have suppressed start of new grace periods. |
64db4cff | 2533 | */ |
8ff0b907 | 2534 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)) |
64db4cff | 2535 | { |
64db4cff PM |
2536 | int cpu; |
2537 | unsigned long flags; | |
2538 | unsigned long mask; | |
a0b6c9a7 | 2539 | struct rcu_node *rnp; |
64db4cff | 2540 | |
aedf4ba9 | 2541 | rcu_for_each_leaf_node(rnp) { |
cee43939 | 2542 | cond_resched_tasks_rcu_qs(); |
64db4cff | 2543 | mask = 0; |
2a67e741 | 2544 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
a0b6c9a7 | 2545 | if (rnp->qsmask == 0) { |
45975c7d | 2546 | if (!IS_ENABLED(CONFIG_PREEMPT) || |
a77da14c PM |
2547 | rcu_preempt_blocked_readers_cgp(rnp)) { |
2548 | /* | |
2549 | * No point in scanning bits because they | |
2550 | * are all zero. But we might need to | |
2551 | * priority-boost blocked readers. | |
2552 | */ | |
2553 | rcu_initiate_boost(rnp, flags); | |
2554 | /* rcu_initiate_boost() releases rnp->lock */ | |
2555 | continue; | |
2556 | } | |
92816435 PM |
2557 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
2558 | continue; | |
64db4cff | 2559 | } |
bc75e999 MR |
2560 | for_each_leaf_node_possible_cpu(rnp, cpu) { |
2561 | unsigned long bit = leaf_node_cpu_bit(rnp, cpu); | |
0edd1b17 | 2562 | if ((rnp->qsmask & bit) != 0) { |
da1df50d | 2563 | if (f(per_cpu_ptr(&rcu_data, cpu))) |
0edd1b17 PM |
2564 | mask |= bit; |
2565 | } | |
64db4cff | 2566 | } |
45f014c5 | 2567 | if (mask != 0) { |
c9a24e2d | 2568 | /* Idle/offline CPUs, report (releases rnp->lock). */ |
b50912d0 | 2569 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
0aa04b05 PM |
2570 | } else { |
2571 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 2572 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2573 | } |
64db4cff | 2574 | } |
64db4cff PM |
2575 | } |
2576 | ||
2577 | /* | |
2578 | * Force quiescent states on reluctant CPUs, and also detect which | |
2579 | * CPUs are in dyntick-idle mode. | |
2580 | */ | |
e9ecb780 | 2581 | static void force_quiescent_state(void) |
64db4cff PM |
2582 | { |
2583 | unsigned long flags; | |
394f2769 PM |
2584 | bool ret; |
2585 | struct rcu_node *rnp; | |
2586 | struct rcu_node *rnp_old = NULL; | |
2587 | ||
2588 | /* Funnel through hierarchy to reduce memory contention. */ | |
da1df50d | 2589 | rnp = __this_cpu_read(rcu_data.mynode); |
394f2769 | 2590 | for (; rnp != NULL; rnp = rnp->parent) { |
67a0edbf | 2591 | ret = (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) || |
394f2769 PM |
2592 | !raw_spin_trylock(&rnp->fqslock); |
2593 | if (rnp_old != NULL) | |
2594 | raw_spin_unlock(&rnp_old->fqslock); | |
d62df573 | 2595 | if (ret) |
394f2769 | 2596 | return; |
394f2769 PM |
2597 | rnp_old = rnp; |
2598 | } | |
336a4f6c | 2599 | /* rnp_old == rcu_get_root(), rnp == NULL. */ |
64db4cff | 2600 | |
394f2769 | 2601 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 2602 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 2603 | raw_spin_unlock(&rnp_old->fqslock); |
67a0edbf | 2604 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
67c583a7 | 2605 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 2606 | return; /* Someone beat us to it. */ |
46a1e34e | 2607 | } |
67a0edbf PM |
2608 | WRITE_ONCE(rcu_state.gp_flags, |
2609 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
67c583a7 | 2610 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
532c00c9 | 2611 | rcu_gp_kthread_wake(); |
64db4cff PM |
2612 | } |
2613 | ||
26d950a9 PM |
2614 | /* |
2615 | * This function checks for grace-period requests that fail to motivate | |
2616 | * RCU to come out of its idle mode. | |
2617 | */ | |
791416c4 PM |
2618 | void |
2619 | rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp, | |
2620 | const unsigned long gpssdelay) | |
26d950a9 PM |
2621 | { |
2622 | unsigned long flags; | |
2623 | unsigned long j; | |
336a4f6c | 2624 | struct rcu_node *rnp_root = rcu_get_root(); |
26d950a9 PM |
2625 | static atomic_t warned = ATOMIC_INIT(0); |
2626 | ||
de8e8730 | 2627 | if (!IS_ENABLED(CONFIG_PROVE_RCU) || rcu_gp_in_progress() || |
7a1d0f23 | 2628 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed)) |
26d950a9 PM |
2629 | return; |
2630 | j = jiffies; /* Expensive access, and in common case don't get here. */ | |
67a0edbf PM |
2631 | if (time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) || |
2632 | time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) || | |
26d950a9 PM |
2633 | atomic_read(&warned)) |
2634 | return; | |
2635 | ||
2636 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
2637 | j = jiffies; | |
de8e8730 | 2638 | if (rcu_gp_in_progress() || |
7a1d0f23 | 2639 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) || |
67a0edbf PM |
2640 | time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) || |
2641 | time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) || | |
26d950a9 PM |
2642 | atomic_read(&warned)) { |
2643 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2644 | return; | |
2645 | } | |
2646 | /* Hold onto the leaf lock to make others see warned==1. */ | |
2647 | ||
2648 | if (rnp_root != rnp) | |
2649 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ | |
2650 | j = jiffies; | |
de8e8730 | 2651 | if (rcu_gp_in_progress() || |
7a1d0f23 | 2652 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) || |
67a0edbf PM |
2653 | time_before(j, rcu_state.gp_req_activity + gpssdelay) || |
2654 | time_before(j, rcu_state.gp_activity + gpssdelay) || | |
26d950a9 PM |
2655 | atomic_xchg(&warned, 1)) { |
2656 | raw_spin_unlock_rcu_node(rnp_root); /* irqs remain disabled. */ | |
2657 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2658 | return; | |
2659 | } | |
b06ae25a | 2660 | pr_alert("%s: g%ld->%ld gar:%lu ga:%lu f%#x gs:%d %s->state:%#lx\n", |
67a0edbf | 2661 | __func__, (long)READ_ONCE(rcu_state.gp_seq), |
7a1d0f23 | 2662 | (long)READ_ONCE(rnp_root->gp_seq_needed), |
67a0edbf PM |
2663 | j - rcu_state.gp_req_activity, j - rcu_state.gp_activity, |
2664 | rcu_state.gp_flags, rcu_state.gp_state, rcu_state.name, | |
2665 | rcu_state.gp_kthread ? rcu_state.gp_kthread->state : 0x1ffffL); | |
26d950a9 PM |
2666 | WARN_ON(1); |
2667 | if (rnp_root != rnp) | |
2668 | raw_spin_unlock_rcu_node(rnp_root); | |
2669 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2670 | } | |
2671 | ||
e0aff973 PM |
2672 | /* |
2673 | * Do a forward-progress check for rcutorture. This is normally invoked | |
2674 | * due to an OOM event. The argument "j" gives the time period during | |
2675 | * which rcutorture would like progress to have been made. | |
2676 | */ | |
2677 | void rcu_fwd_progress_check(unsigned long j) | |
2678 | { | |
bfcfcffc PM |
2679 | unsigned long cbs; |
2680 | int cpu; | |
2681 | unsigned long max_cbs = 0; | |
2682 | int max_cpu = -1; | |
e0aff973 PM |
2683 | struct rcu_data *rdp; |
2684 | ||
2685 | if (rcu_gp_in_progress()) { | |
8dd3b546 PM |
2686 | pr_info("%s: GP age %lu jiffies\n", |
2687 | __func__, jiffies - rcu_state.gp_start); | |
e0aff973 PM |
2688 | show_rcu_gp_kthreads(); |
2689 | } else { | |
c51d7b5e PM |
2690 | pr_info("%s: Last GP end %lu jiffies ago\n", |
2691 | __func__, jiffies - rcu_state.gp_end); | |
e0aff973 PM |
2692 | preempt_disable(); |
2693 | rdp = this_cpu_ptr(&rcu_data); | |
2694 | rcu_check_gp_start_stall(rdp->mynode, rdp, j); | |
2695 | preempt_enable(); | |
2696 | } | |
bfcfcffc PM |
2697 | for_each_possible_cpu(cpu) { |
2698 | cbs = rcu_get_n_cbs_cpu(cpu); | |
2699 | if (!cbs) | |
2700 | continue; | |
2701 | if (max_cpu < 0) | |
2702 | pr_info("%s: callbacks", __func__); | |
2703 | pr_cont(" %d: %lu", cpu, cbs); | |
2704 | if (cbs <= max_cbs) | |
2705 | continue; | |
2706 | max_cbs = cbs; | |
2707 | max_cpu = cpu; | |
2708 | } | |
2709 | if (max_cpu >= 0) | |
2710 | pr_cont("\n"); | |
e0aff973 PM |
2711 | } |
2712 | EXPORT_SYMBOL_GPL(rcu_fwd_progress_check); | |
2713 | ||
64db4cff | 2714 | /* |
b049fdf8 PM |
2715 | * This does the RCU core processing work for the specified rcu_data |
2716 | * structures. This may be called only from the CPU to whom the rdp | |
2717 | * belongs. | |
64db4cff | 2718 | */ |
b049fdf8 | 2719 | static __latent_entropy void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff PM |
2720 | { |
2721 | unsigned long flags; | |
da1df50d | 2722 | struct rcu_data *rdp = raw_cpu_ptr(&rcu_data); |
26d950a9 | 2723 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 2724 | |
b049fdf8 PM |
2725 | if (cpu_is_offline(smp_processor_id())) |
2726 | return; | |
2727 | trace_rcu_utilization(TPS("Start RCU core")); | |
50dc7def | 2728 | WARN_ON_ONCE(!rdp->beenonline); |
2e597558 | 2729 | |
3e310098 | 2730 | /* Report any deferred quiescent states if preemption enabled. */ |
fced9c8c | 2731 | if (!(preempt_count() & PREEMPT_MASK)) { |
3e310098 | 2732 | rcu_preempt_deferred_qs(current); |
fced9c8c PM |
2733 | } else if (rcu_preempt_need_deferred_qs(current)) { |
2734 | set_tsk_need_resched(current); | |
2735 | set_preempt_need_resched(); | |
2736 | } | |
3e310098 | 2737 | |
64db4cff | 2738 | /* Update RCU state based on any recent quiescent states. */ |
8087d3e3 | 2739 | rcu_check_quiescent_state(rdp); |
64db4cff | 2740 | |
bd7af846 | 2741 | /* No grace period and unregistered callbacks? */ |
de8e8730 | 2742 | if (!rcu_gp_in_progress() && |
bd7af846 PM |
2743 | rcu_segcblist_is_enabled(&rdp->cblist)) { |
2744 | local_irq_save(flags); | |
e44e73ca | 2745 | if (!rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
c6e09b97 | 2746 | rcu_accelerate_cbs_unlocked(rnp, rdp); |
e44e73ca | 2747 | local_irq_restore(flags); |
64db4cff PM |
2748 | } |
2749 | ||
791416c4 | 2750 | rcu_check_gp_start_stall(rnp, rdp, rcu_jiffies_till_stall_check()); |
26d950a9 | 2751 | |
64db4cff | 2752 | /* If there are callbacks ready, invoke them. */ |
15fecf89 | 2753 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
aff4e9ed | 2754 | invoke_rcu_callbacks(rdp); |
96d3fd0d PM |
2755 | |
2756 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
2757 | do_nocb_deferred_wakeup(rdp); | |
f7f7bac9 | 2758 | trace_rcu_utilization(TPS("End RCU core")); |
64db4cff PM |
2759 | } |
2760 | ||
a26ac245 | 2761 | /* |
49918a54 PM |
2762 | * Schedule RCU callback invocation. If the running implementation of RCU |
2763 | * does not support RCU priority boosting, just do a direct call, otherwise | |
2764 | * wake up the per-CPU kernel kthread. Note that because we are running | |
2765 | * on the current CPU with softirqs disabled, the rcu_cpu_kthread_task | |
2766 | * cannot disappear out from under us. | |
a26ac245 | 2767 | */ |
aff4e9ed | 2768 | static void invoke_rcu_callbacks(struct rcu_data *rdp) |
a26ac245 | 2769 | { |
7d0ae808 | 2770 | if (unlikely(!READ_ONCE(rcu_scheduler_fully_active))) |
b0d30417 | 2771 | return; |
3c779dfe | 2772 | if (likely(!rcu_state.boost)) { |
5bb5d09c | 2773 | rcu_do_batch(rdp); |
a26ac245 PM |
2774 | return; |
2775 | } | |
a46e0899 | 2776 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
2777 | } |
2778 | ||
a46e0899 | 2779 | static void invoke_rcu_core(void) |
09223371 | 2780 | { |
b0f74036 PM |
2781 | if (cpu_online(smp_processor_id())) |
2782 | raise_softirq(RCU_SOFTIRQ); | |
09223371 SL |
2783 | } |
2784 | ||
29154c57 PM |
2785 | /* |
2786 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
2787 | */ | |
5c7d8967 PM |
2788 | static void __call_rcu_core(struct rcu_data *rdp, struct rcu_head *head, |
2789 | unsigned long flags) | |
64db4cff | 2790 | { |
62fde6ed PM |
2791 | /* |
2792 | * If called from an extended quiescent state, invoke the RCU | |
2793 | * core in order to force a re-evaluation of RCU's idleness. | |
2794 | */ | |
9910affa | 2795 | if (!rcu_is_watching()) |
62fde6ed PM |
2796 | invoke_rcu_core(); |
2797 | ||
a16b7a69 | 2798 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 2799 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 2800 | return; |
64db4cff | 2801 | |
37c72e56 PM |
2802 | /* |
2803 | * Force the grace period if too many callbacks or too long waiting. | |
2804 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
2805 | * if some other CPU has recently done so. Also, don't bother | |
2806 | * invoking force_quiescent_state() if the newly enqueued callback | |
2807 | * is the only one waiting for a grace period to complete. | |
2808 | */ | |
15fecf89 PM |
2809 | if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) > |
2810 | rdp->qlen_last_fqs_check + qhimark)) { | |
b52573d2 PM |
2811 | |
2812 | /* Are we ignoring a completed grace period? */ | |
15cabdff | 2813 | note_gp_changes(rdp); |
b52573d2 PM |
2814 | |
2815 | /* Start a new grace period if one not already started. */ | |
de8e8730 | 2816 | if (!rcu_gp_in_progress()) { |
c6e09b97 | 2817 | rcu_accelerate_cbs_unlocked(rdp->mynode, rdp); |
b52573d2 PM |
2818 | } else { |
2819 | /* Give the grace period a kick. */ | |
2820 | rdp->blimit = LONG_MAX; | |
5c7d8967 | 2821 | if (rcu_state.n_force_qs == rdp->n_force_qs_snap && |
15fecf89 | 2822 | rcu_segcblist_first_pend_cb(&rdp->cblist) != head) |
e9ecb780 | 2823 | force_quiescent_state(); |
5c7d8967 | 2824 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
15fecf89 | 2825 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); |
b52573d2 | 2826 | } |
4cdfc175 | 2827 | } |
29154c57 PM |
2828 | } |
2829 | ||
ae150184 PM |
2830 | /* |
2831 | * RCU callback function to leak a callback. | |
2832 | */ | |
2833 | static void rcu_leak_callback(struct rcu_head *rhp) | |
2834 | { | |
2835 | } | |
2836 | ||
3fbfbf7a PM |
2837 | /* |
2838 | * Helper function for call_rcu() and friends. The cpu argument will | |
2839 | * normally be -1, indicating "currently running CPU". It may specify | |
dd46a788 | 2840 | * a CPU only if that CPU is a no-CBs CPU. Currently, only rcu_barrier() |
3fbfbf7a PM |
2841 | * is expected to specify a CPU. |
2842 | */ | |
64db4cff | 2843 | static void |
5c7d8967 | 2844 | __call_rcu(struct rcu_head *head, rcu_callback_t func, int cpu, bool lazy) |
64db4cff PM |
2845 | { |
2846 | unsigned long flags; | |
2847 | struct rcu_data *rdp; | |
2848 | ||
b8f2ed53 PM |
2849 | /* Misaligned rcu_head! */ |
2850 | WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1)); | |
2851 | ||
ae150184 | 2852 | if (debug_rcu_head_queue(head)) { |
fa3c6647 PM |
2853 | /* |
2854 | * Probable double call_rcu(), so leak the callback. | |
2855 | * Use rcu:rcu_callback trace event to find the previous | |
2856 | * time callback was passed to __call_rcu(). | |
2857 | */ | |
2858 | WARN_ONCE(1, "__call_rcu(): Double-freed CB %p->%pF()!!!\n", | |
2859 | head, head->func); | |
7d0ae808 | 2860 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
2861 | return; |
2862 | } | |
64db4cff PM |
2863 | head->func = func; |
2864 | head->next = NULL; | |
64db4cff | 2865 | local_irq_save(flags); |
da1df50d | 2866 | rdp = this_cpu_ptr(&rcu_data); |
64db4cff PM |
2867 | |
2868 | /* Add the callback to our list. */ | |
15fecf89 | 2869 | if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist)) || cpu != -1) { |
3fbfbf7a PM |
2870 | int offline; |
2871 | ||
2872 | if (cpu != -1) | |
da1df50d | 2873 | rdp = per_cpu_ptr(&rcu_data, cpu); |
143da9c2 PM |
2874 | if (likely(rdp->mynode)) { |
2875 | /* Post-boot, so this should be for a no-CBs CPU. */ | |
2876 | offline = !__call_rcu_nocb(rdp, head, lazy, flags); | |
2877 | WARN_ON_ONCE(offline); | |
2878 | /* Offline CPU, _call_rcu() illegal, leak callback. */ | |
2879 | local_irq_restore(flags); | |
2880 | return; | |
2881 | } | |
2882 | /* | |
2883 | * Very early boot, before rcu_init(). Initialize if needed | |
2884 | * and then drop through to queue the callback. | |
2885 | */ | |
08543bda | 2886 | WARN_ON_ONCE(cpu != -1); |
34404ca8 | 2887 | WARN_ON_ONCE(!rcu_is_watching()); |
15fecf89 PM |
2888 | if (rcu_segcblist_empty(&rdp->cblist)) |
2889 | rcu_segcblist_init(&rdp->cblist); | |
0d8ee37e | 2890 | } |
15fecf89 PM |
2891 | rcu_segcblist_enqueue(&rdp->cblist, head, lazy); |
2892 | if (!lazy) | |
c57afe80 | 2893 | rcu_idle_count_callbacks_posted(); |
2655d57e | 2894 | |
d4c08f2a | 2895 | if (__is_kfree_rcu_offset((unsigned long)func)) |
3c779dfe PM |
2896 | trace_rcu_kfree_callback(rcu_state.name, head, |
2897 | (unsigned long)func, | |
15fecf89 PM |
2898 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2899 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2900 | else |
3c779dfe | 2901 | trace_rcu_callback(rcu_state.name, head, |
15fecf89 PM |
2902 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2903 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2904 | |
29154c57 | 2905 | /* Go handle any RCU core processing required. */ |
5c7d8967 | 2906 | __call_rcu_core(rdp, head, flags); |
64db4cff PM |
2907 | local_irq_restore(flags); |
2908 | } | |
2909 | ||
a68a2bb2 | 2910 | /** |
45975c7d | 2911 | * call_rcu() - Queue an RCU callback for invocation after a grace period. |
a68a2bb2 PM |
2912 | * @head: structure to be used for queueing the RCU updates. |
2913 | * @func: actual callback function to be invoked after the grace period | |
2914 | * | |
2915 | * The callback function will be invoked some time after a full grace | |
45975c7d PM |
2916 | * period elapses, in other words after all pre-existing RCU read-side |
2917 | * critical sections have completed. However, the callback function | |
2918 | * might well execute concurrently with RCU read-side critical sections | |
2919 | * that started after call_rcu() was invoked. RCU read-side critical | |
2920 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), and | |
2921 | * may be nested. In addition, regions of code across which interrupts, | |
2922 | * preemption, or softirqs have been disabled also serve as RCU read-side | |
2923 | * critical sections. This includes hardware interrupt handlers, softirq | |
2924 | * handlers, and NMI handlers. | |
2925 | * | |
2926 | * Note that all CPUs must agree that the grace period extended beyond | |
2927 | * all pre-existing RCU read-side critical section. On systems with more | |
2928 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
2929 | * guaranteed to have executed a full memory barrier since the end of its | |
2930 | * last RCU read-side critical section whose beginning preceded the call | |
2931 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
2932 | * critical section that continues beyond the start of "func()" must have | |
2933 | * executed a memory barrier after the call_rcu() but before the beginning | |
2934 | * of that RCU read-side critical section. Note that these guarantees | |
2935 | * include CPUs that are offline, idle, or executing in user mode, as | |
2936 | * well as CPUs that are executing in the kernel. | |
2937 | * | |
2938 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
2939 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
2940 | * guaranteed to execute a full memory barrier during the time interval | |
2941 | * between the call to call_rcu() and the invocation of "func()" -- even | |
2942 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
2943 | * more than one CPU). | |
2944 | */ | |
2945 | void call_rcu(struct rcu_head *head, rcu_callback_t func) | |
2946 | { | |
5c7d8967 | 2947 | __call_rcu(head, func, -1, 0); |
45975c7d PM |
2948 | } |
2949 | EXPORT_SYMBOL_GPL(call_rcu); | |
64db4cff | 2950 | |
495aa969 ACB |
2951 | /* |
2952 | * Queue an RCU callback for lazy invocation after a grace period. | |
2953 | * This will likely be later named something like "call_rcu_lazy()", | |
2954 | * but this change will require some way of tagging the lazy RCU | |
2955 | * callbacks in the list of pending callbacks. Until then, this | |
2956 | * function may only be called from __kfree_rcu(). | |
2957 | */ | |
98ece508 | 2958 | void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func) |
495aa969 | 2959 | { |
5c7d8967 | 2960 | __call_rcu(head, func, -1, 1); |
495aa969 ACB |
2961 | } |
2962 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
2963 | ||
765a3f4f PM |
2964 | /** |
2965 | * get_state_synchronize_rcu - Snapshot current RCU state | |
2966 | * | |
2967 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
2968 | * to determine whether or not a full grace period has elapsed in the | |
2969 | * meantime. | |
2970 | */ | |
2971 | unsigned long get_state_synchronize_rcu(void) | |
2972 | { | |
2973 | /* | |
2974 | * Any prior manipulation of RCU-protected data must happen | |
e4be81a2 | 2975 | * before the load from ->gp_seq. |
765a3f4f PM |
2976 | */ |
2977 | smp_mb(); /* ^^^ */ | |
16fc9c60 | 2978 | return rcu_seq_snap(&rcu_state.gp_seq); |
765a3f4f PM |
2979 | } |
2980 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
2981 | ||
2982 | /** | |
2983 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
2984 | * | |
2985 | * @oldstate: return value from earlier call to get_state_synchronize_rcu() | |
2986 | * | |
2987 | * If a full RCU grace period has elapsed since the earlier call to | |
2988 | * get_state_synchronize_rcu(), just return. Otherwise, invoke | |
2989 | * synchronize_rcu() to wait for a full grace period. | |
2990 | * | |
2991 | * Yes, this function does not take counter wrap into account. But | |
2992 | * counter wrap is harmless. If the counter wraps, we have waited for | |
2993 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
2994 | * so waiting for one additional grace period should be just fine. | |
2995 | */ | |
2996 | void cond_synchronize_rcu(unsigned long oldstate) | |
2997 | { | |
16fc9c60 | 2998 | if (!rcu_seq_done(&rcu_state.gp_seq, oldstate)) |
765a3f4f | 2999 | synchronize_rcu(); |
e4be81a2 PM |
3000 | else |
3001 | smp_mb(); /* Ensure GP ends before subsequent accesses. */ | |
765a3f4f PM |
3002 | } |
3003 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
3004 | ||
64db4cff | 3005 | /* |
98ece508 | 3006 | * Check to see if there is any immediate RCU-related work to be done by |
49918a54 PM |
3007 | * the current CPU, returning 1 if so and zero otherwise. The checks are |
3008 | * in order of increasing expense: checks that can be carried out against | |
3009 | * CPU-local state are performed first. However, we must check for CPU | |
3010 | * stalls first, else we might not get a chance. | |
64db4cff | 3011 | */ |
98ece508 | 3012 | static int rcu_pending(void) |
64db4cff | 3013 | { |
98ece508 | 3014 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
2f51f988 PM |
3015 | struct rcu_node *rnp = rdp->mynode; |
3016 | ||
64db4cff | 3017 | /* Check for CPU stalls, if enabled. */ |
ea12ff2b | 3018 | check_cpu_stall(rdp); |
64db4cff | 3019 | |
a096932f | 3020 | /* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */ |
4580b054 | 3021 | if (rcu_nohz_full_cpu()) |
a096932f PM |
3022 | return 0; |
3023 | ||
64db4cff | 3024 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
01c495f7 | 3025 | if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) |
64db4cff PM |
3026 | return 1; |
3027 | ||
3028 | /* Does this CPU have callbacks ready to invoke? */ | |
01c495f7 | 3029 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
64db4cff PM |
3030 | return 1; |
3031 | ||
3032 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
de8e8730 | 3033 | if (!rcu_gp_in_progress() && |
c1935209 PM |
3034 | rcu_segcblist_is_enabled(&rdp->cblist) && |
3035 | !rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) | |
64db4cff PM |
3036 | return 1; |
3037 | ||
67e14c1e PM |
3038 | /* Have RCU grace period completed or started? */ |
3039 | if (rcu_seq_current(&rnp->gp_seq) != rdp->gp_seq || | |
01c495f7 | 3040 | unlikely(READ_ONCE(rdp->gpwrap))) /* outside lock */ |
64db4cff PM |
3041 | return 1; |
3042 | ||
96d3fd0d | 3043 | /* Does this CPU need a deferred NOCB wakeup? */ |
01c495f7 | 3044 | if (rcu_nocb_need_deferred_wakeup(rdp)) |
96d3fd0d | 3045 | return 1; |
96d3fd0d | 3046 | |
64db4cff PM |
3047 | /* nothing to do */ |
3048 | return 0; | |
3049 | } | |
3050 | ||
64db4cff | 3051 | /* |
c0f4dfd4 PM |
3052 | * Return true if the specified CPU has any callback. If all_lazy is |
3053 | * non-NULL, store an indication of whether all callbacks are lazy. | |
3054 | * (If there are no callbacks, all of them are deemed to be lazy.) | |
64db4cff | 3055 | */ |
51fbb910 | 3056 | static bool rcu_cpu_has_callbacks(bool *all_lazy) |
64db4cff | 3057 | { |
c0f4dfd4 PM |
3058 | bool al = true; |
3059 | bool hc = false; | |
3060 | struct rcu_data *rdp; | |
6ce75a23 | 3061 | |
b97d23c5 PM |
3062 | rdp = this_cpu_ptr(&rcu_data); |
3063 | if (!rcu_segcblist_empty(&rdp->cblist)) { | |
69c8d28c | 3064 | hc = true; |
b97d23c5 | 3065 | if (rcu_segcblist_n_nonlazy_cbs(&rdp->cblist)) |
c0f4dfd4 | 3066 | al = false; |
c0f4dfd4 PM |
3067 | } |
3068 | if (all_lazy) | |
3069 | *all_lazy = al; | |
3070 | return hc; | |
64db4cff PM |
3071 | } |
3072 | ||
a83eff0a | 3073 | /* |
dd46a788 | 3074 | * Helper function for rcu_barrier() tracing. If tracing is disabled, |
a83eff0a PM |
3075 | * the compiler is expected to optimize this away. |
3076 | */ | |
dd46a788 | 3077 | static void rcu_barrier_trace(const char *s, int cpu, unsigned long done) |
a83eff0a | 3078 | { |
8344b871 PM |
3079 | trace_rcu_barrier(rcu_state.name, s, cpu, |
3080 | atomic_read(&rcu_state.barrier_cpu_count), done); | |
a83eff0a PM |
3081 | } |
3082 | ||
b1420f1c | 3083 | /* |
dd46a788 PM |
3084 | * RCU callback function for rcu_barrier(). If we are last, wake |
3085 | * up the task executing rcu_barrier(). | |
b1420f1c | 3086 | */ |
24ebbca8 | 3087 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 3088 | { |
ec9f5835 | 3089 | if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) { |
dd46a788 | 3090 | rcu_barrier_trace(TPS("LastCB"), -1, |
ec9f5835 PM |
3091 | rcu_state.barrier_sequence); |
3092 | complete(&rcu_state.barrier_completion); | |
a83eff0a | 3093 | } else { |
dd46a788 | 3094 | rcu_barrier_trace(TPS("CB"), -1, rcu_state.barrier_sequence); |
a83eff0a | 3095 | } |
d0ec774c PM |
3096 | } |
3097 | ||
3098 | /* | |
3099 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
3100 | */ | |
ec9f5835 | 3101 | static void rcu_barrier_func(void *unused) |
d0ec774c | 3102 | { |
da1df50d | 3103 | struct rcu_data *rdp = raw_cpu_ptr(&rcu_data); |
d0ec774c | 3104 | |
dd46a788 | 3105 | rcu_barrier_trace(TPS("IRQ"), -1, rcu_state.barrier_sequence); |
f92c734f PM |
3106 | rdp->barrier_head.func = rcu_barrier_callback; |
3107 | debug_rcu_head_queue(&rdp->barrier_head); | |
3108 | if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head, 0)) { | |
ec9f5835 | 3109 | atomic_inc(&rcu_state.barrier_cpu_count); |
f92c734f PM |
3110 | } else { |
3111 | debug_rcu_head_unqueue(&rdp->barrier_head); | |
dd46a788 | 3112 | rcu_barrier_trace(TPS("IRQNQ"), -1, |
ec9f5835 | 3113 | rcu_state.barrier_sequence); |
f92c734f | 3114 | } |
d0ec774c PM |
3115 | } |
3116 | ||
dd46a788 PM |
3117 | /** |
3118 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
3119 | * | |
3120 | * Note that this primitive does not necessarily wait for an RCU grace period | |
3121 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
3122 | * in the system, then rcu_barrier() is within its rights to return | |
3123 | * immediately, without waiting for anything, much less an RCU grace period. | |
d0ec774c | 3124 | */ |
dd46a788 | 3125 | void rcu_barrier(void) |
d0ec774c | 3126 | { |
b1420f1c | 3127 | int cpu; |
b1420f1c | 3128 | struct rcu_data *rdp; |
ec9f5835 | 3129 | unsigned long s = rcu_seq_snap(&rcu_state.barrier_sequence); |
b1420f1c | 3130 | |
dd46a788 | 3131 | rcu_barrier_trace(TPS("Begin"), -1, s); |
b1420f1c | 3132 | |
e74f4c45 | 3133 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
ec9f5835 | 3134 | mutex_lock(&rcu_state.barrier_mutex); |
b1420f1c | 3135 | |
4f525a52 | 3136 | /* Did someone else do our work for us? */ |
ec9f5835 | 3137 | if (rcu_seq_done(&rcu_state.barrier_sequence, s)) { |
dd46a788 | 3138 | rcu_barrier_trace(TPS("EarlyExit"), -1, |
ec9f5835 | 3139 | rcu_state.barrier_sequence); |
cf3a9c48 | 3140 | smp_mb(); /* caller's subsequent code after above check. */ |
ec9f5835 | 3141 | mutex_unlock(&rcu_state.barrier_mutex); |
cf3a9c48 PM |
3142 | return; |
3143 | } | |
3144 | ||
4f525a52 | 3145 | /* Mark the start of the barrier operation. */ |
ec9f5835 | 3146 | rcu_seq_start(&rcu_state.barrier_sequence); |
dd46a788 | 3147 | rcu_barrier_trace(TPS("Inc1"), -1, rcu_state.barrier_sequence); |
b1420f1c | 3148 | |
d0ec774c | 3149 | /* |
b1420f1c PM |
3150 | * Initialize the count to one rather than to zero in order to |
3151 | * avoid a too-soon return to zero in case of a short grace period | |
1331e7a1 PM |
3152 | * (or preemption of this task). Exclude CPU-hotplug operations |
3153 | * to ensure that no offline CPU has callbacks queued. | |
d0ec774c | 3154 | */ |
ec9f5835 PM |
3155 | init_completion(&rcu_state.barrier_completion); |
3156 | atomic_set(&rcu_state.barrier_cpu_count, 1); | |
1331e7a1 | 3157 | get_online_cpus(); |
b1420f1c PM |
3158 | |
3159 | /* | |
1331e7a1 PM |
3160 | * Force each CPU with callbacks to register a new callback. |
3161 | * When that callback is invoked, we will know that all of the | |
3162 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 3163 | */ |
3fbfbf7a | 3164 | for_each_possible_cpu(cpu) { |
d1e43fa5 | 3165 | if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu)) |
3fbfbf7a | 3166 | continue; |
da1df50d | 3167 | rdp = per_cpu_ptr(&rcu_data, cpu); |
d1e43fa5 | 3168 | if (rcu_is_nocb_cpu(cpu)) { |
4580b054 | 3169 | if (!rcu_nocb_cpu_needs_barrier(cpu)) { |
dd46a788 | 3170 | rcu_barrier_trace(TPS("OfflineNoCB"), cpu, |
ec9f5835 | 3171 | rcu_state.barrier_sequence); |
d7e29933 | 3172 | } else { |
dd46a788 | 3173 | rcu_barrier_trace(TPS("OnlineNoCB"), cpu, |
ec9f5835 | 3174 | rcu_state.barrier_sequence); |
41050a00 | 3175 | smp_mb__before_atomic(); |
ec9f5835 | 3176 | atomic_inc(&rcu_state.barrier_cpu_count); |
d7e29933 | 3177 | __call_rcu(&rdp->barrier_head, |
5c7d8967 | 3178 | rcu_barrier_callback, cpu, 0); |
d7e29933 | 3179 | } |
15fecf89 | 3180 | } else if (rcu_segcblist_n_cbs(&rdp->cblist)) { |
dd46a788 | 3181 | rcu_barrier_trace(TPS("OnlineQ"), cpu, |
ec9f5835 PM |
3182 | rcu_state.barrier_sequence); |
3183 | smp_call_function_single(cpu, rcu_barrier_func, NULL, 1); | |
b1420f1c | 3184 | } else { |
dd46a788 | 3185 | rcu_barrier_trace(TPS("OnlineNQ"), cpu, |
ec9f5835 | 3186 | rcu_state.barrier_sequence); |
b1420f1c PM |
3187 | } |
3188 | } | |
1331e7a1 | 3189 | put_online_cpus(); |
b1420f1c PM |
3190 | |
3191 | /* | |
3192 | * Now that we have an rcu_barrier_callback() callback on each | |
3193 | * CPU, and thus each counted, remove the initial count. | |
3194 | */ | |
ec9f5835 PM |
3195 | if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) |
3196 | complete(&rcu_state.barrier_completion); | |
b1420f1c PM |
3197 | |
3198 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
ec9f5835 | 3199 | wait_for_completion(&rcu_state.barrier_completion); |
b1420f1c | 3200 | |
4f525a52 | 3201 | /* Mark the end of the barrier operation. */ |
dd46a788 | 3202 | rcu_barrier_trace(TPS("Inc2"), -1, rcu_state.barrier_sequence); |
ec9f5835 | 3203 | rcu_seq_end(&rcu_state.barrier_sequence); |
4f525a52 | 3204 | |
b1420f1c | 3205 | /* Other rcu_barrier() invocations can now safely proceed. */ |
ec9f5835 | 3206 | mutex_unlock(&rcu_state.barrier_mutex); |
d0ec774c | 3207 | } |
45975c7d | 3208 | EXPORT_SYMBOL_GPL(rcu_barrier); |
d0ec774c | 3209 | |
0aa04b05 PM |
3210 | /* |
3211 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
3212 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
3213 | * must hold the corresponding leaf rcu_node ->lock with interrrupts | |
3214 | * disabled. | |
3215 | */ | |
3216 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
3217 | { | |
3218 | long mask; | |
8d672fa6 | 3219 | long oldmask; |
0aa04b05 PM |
3220 | struct rcu_node *rnp = rnp_leaf; |
3221 | ||
8d672fa6 | 3222 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
962aff03 | 3223 | WARN_ON_ONCE(rnp->wait_blkd_tasks); |
0aa04b05 PM |
3224 | for (;;) { |
3225 | mask = rnp->grpmask; | |
3226 | rnp = rnp->parent; | |
3227 | if (rnp == NULL) | |
3228 | return; | |
6cf10081 | 3229 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
8d672fa6 | 3230 | oldmask = rnp->qsmaskinit; |
0aa04b05 | 3231 | rnp->qsmaskinit |= mask; |
67c583a7 | 3232 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
8d672fa6 PM |
3233 | if (oldmask) |
3234 | return; | |
0aa04b05 PM |
3235 | } |
3236 | } | |
3237 | ||
64db4cff | 3238 | /* |
27569620 | 3239 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 3240 | */ |
27569620 | 3241 | static void __init |
53b46303 | 3242 | rcu_boot_init_percpu_data(int cpu) |
64db4cff | 3243 | { |
da1df50d | 3244 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27569620 PM |
3245 | |
3246 | /* Set up local state, ensuring consistent view of global state. */ | |
bc75e999 | 3247 | rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu); |
4c5273bf | 3248 | WARN_ON_ONCE(rdp->dynticks_nesting != 1); |
dc5a4f29 | 3249 | WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp))); |
53b46303 | 3250 | rdp->rcu_ofl_gp_seq = rcu_state.gp_seq; |
57738942 | 3251 | rdp->rcu_ofl_gp_flags = RCU_GP_CLEANED; |
53b46303 | 3252 | rdp->rcu_onl_gp_seq = rcu_state.gp_seq; |
57738942 | 3253 | rdp->rcu_onl_gp_flags = RCU_GP_CLEANED; |
27569620 | 3254 | rdp->cpu = cpu; |
3fbfbf7a | 3255 | rcu_boot_init_nocb_percpu_data(rdp); |
27569620 PM |
3256 | } |
3257 | ||
3258 | /* | |
53b46303 PM |
3259 | * Invoked early in the CPU-online process, when pretty much all services |
3260 | * are available. The incoming CPU is not present. | |
3261 | * | |
3262 | * Initializes a CPU's per-CPU RCU data. Note that only one online or | |
ff3bb6f4 PM |
3263 | * offline event can be happening at a given time. Note also that we can |
3264 | * accept some slop in the rsp->gp_seq access due to the fact that this | |
3265 | * CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 3266 | */ |
53b46303 | 3267 | int rcutree_prepare_cpu(unsigned int cpu) |
64db4cff PM |
3268 | { |
3269 | unsigned long flags; | |
da1df50d | 3270 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
336a4f6c | 3271 | struct rcu_node *rnp = rcu_get_root(); |
64db4cff PM |
3272 | |
3273 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 3274 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 | 3275 | rdp->qlen_last_fqs_check = 0; |
53b46303 | 3276 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
64db4cff | 3277 | rdp->blimit = blimit; |
15fecf89 PM |
3278 | if (rcu_segcblist_empty(&rdp->cblist) && /* No early-boot CBs? */ |
3279 | !init_nocb_callback_list(rdp)) | |
3280 | rcu_segcblist_init(&rdp->cblist); /* Re-enable callbacks. */ | |
4c5273bf | 3281 | rdp->dynticks_nesting = 1; /* CPU not up, no tearing. */ |
2625d469 | 3282 | rcu_dynticks_eqs_online(); |
67c583a7 | 3283 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
64db4cff | 3284 | |
0aa04b05 PM |
3285 | /* |
3286 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
3287 | * propagation up the rcu_node tree will happen at the beginning | |
3288 | * of the next grace period. | |
3289 | */ | |
64db4cff | 3290 | rnp = rdp->mynode; |
2a67e741 | 3291 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
b9585e94 | 3292 | rdp->beenonline = true; /* We have now been online. */ |
de30ad51 | 3293 | rdp->gp_seq = rnp->gp_seq; |
7a1d0f23 | 3294 | rdp->gp_seq_needed = rnp->gp_seq; |
5b74c458 | 3295 | rdp->cpu_no_qs.b.norm = true; |
97c668b8 | 3296 | rdp->core_needs_qs = false; |
9b9500da | 3297 | rdp->rcu_iw_pending = false; |
8aa670cd | 3298 | rdp->rcu_iw_gp_seq = rnp->gp_seq - 1; |
53b46303 | 3299 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuonl")); |
67c583a7 | 3300 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
4df83742 TG |
3301 | rcu_prepare_kthreads(cpu); |
3302 | rcu_spawn_all_nocb_kthreads(cpu); | |
3303 | ||
3304 | return 0; | |
3305 | } | |
3306 | ||
deb34f36 PM |
3307 | /* |
3308 | * Update RCU priority boot kthread affinity for CPU-hotplug changes. | |
3309 | */ | |
4df83742 TG |
3310 | static void rcutree_affinity_setting(unsigned int cpu, int outgoing) |
3311 | { | |
da1df50d | 3312 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
4df83742 TG |
3313 | |
3314 | rcu_boost_kthread_setaffinity(rdp->mynode, outgoing); | |
3315 | } | |
3316 | ||
deb34f36 PM |
3317 | /* |
3318 | * Near the end of the CPU-online process. Pretty much all services | |
3319 | * enabled, and the CPU is now very much alive. | |
3320 | */ | |
4df83742 TG |
3321 | int rcutree_online_cpu(unsigned int cpu) |
3322 | { | |
9b9500da PM |
3323 | unsigned long flags; |
3324 | struct rcu_data *rdp; | |
3325 | struct rcu_node *rnp; | |
9b9500da | 3326 | |
b97d23c5 PM |
3327 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3328 | rnp = rdp->mynode; | |
3329 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3330 | rnp->ffmask |= rdp->grpmask; | |
3331 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
da915ad5 PM |
3332 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3333 | srcu_online_cpu(cpu); | |
9b9500da PM |
3334 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
3335 | return 0; /* Too early in boot for scheduler work. */ | |
3336 | sync_sched_exp_online_cleanup(cpu); | |
3337 | rcutree_affinity_setting(cpu, -1); | |
4df83742 TG |
3338 | return 0; |
3339 | } | |
3340 | ||
deb34f36 PM |
3341 | /* |
3342 | * Near the beginning of the process. The CPU is still very much alive | |
3343 | * with pretty much all services enabled. | |
3344 | */ | |
4df83742 TG |
3345 | int rcutree_offline_cpu(unsigned int cpu) |
3346 | { | |
9b9500da PM |
3347 | unsigned long flags; |
3348 | struct rcu_data *rdp; | |
3349 | struct rcu_node *rnp; | |
9b9500da | 3350 | |
b97d23c5 PM |
3351 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3352 | rnp = rdp->mynode; | |
3353 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3354 | rnp->ffmask &= ~rdp->grpmask; | |
3355 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
9b9500da | 3356 | |
4df83742 | 3357 | rcutree_affinity_setting(cpu, cpu); |
da915ad5 PM |
3358 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3359 | srcu_offline_cpu(cpu); | |
4df83742 TG |
3360 | return 0; |
3361 | } | |
3362 | ||
f64c6013 PZ |
3363 | static DEFINE_PER_CPU(int, rcu_cpu_started); |
3364 | ||
7ec99de3 PM |
3365 | /* |
3366 | * Mark the specified CPU as being online so that subsequent grace periods | |
3367 | * (both expedited and normal) will wait on it. Note that this means that | |
3368 | * incoming CPUs are not allowed to use RCU read-side critical sections | |
3369 | * until this function is called. Failing to observe this restriction | |
3370 | * will result in lockdep splats. | |
deb34f36 PM |
3371 | * |
3372 | * Note that this function is special in that it is invoked directly | |
3373 | * from the incoming CPU rather than from the cpuhp_step mechanism. | |
3374 | * This is because this function must be invoked at a precise location. | |
7ec99de3 PM |
3375 | */ |
3376 | void rcu_cpu_starting(unsigned int cpu) | |
3377 | { | |
3378 | unsigned long flags; | |
3379 | unsigned long mask; | |
313517fc PM |
3380 | int nbits; |
3381 | unsigned long oldmask; | |
7ec99de3 PM |
3382 | struct rcu_data *rdp; |
3383 | struct rcu_node *rnp; | |
7ec99de3 | 3384 | |
f64c6013 PZ |
3385 | if (per_cpu(rcu_cpu_started, cpu)) |
3386 | return; | |
3387 | ||
3388 | per_cpu(rcu_cpu_started, cpu) = 1; | |
3389 | ||
b97d23c5 PM |
3390 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3391 | rnp = rdp->mynode; | |
3392 | mask = rdp->grpmask; | |
3393 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3394 | rnp->qsmaskinitnext |= mask; | |
3395 | oldmask = rnp->expmaskinitnext; | |
3396 | rnp->expmaskinitnext |= mask; | |
3397 | oldmask ^= rnp->expmaskinitnext; | |
3398 | nbits = bitmap_weight(&oldmask, BITS_PER_LONG); | |
3399 | /* Allow lockless access for expedited grace periods. */ | |
eb7a6653 | 3400 | smp_store_release(&rcu_state.ncpus, rcu_state.ncpus + nbits); /* ^^^ */ |
b97d23c5 | 3401 | rcu_gpnum_ovf(rnp, rdp); /* Offline-induced counter wrap? */ |
eb7a6653 PM |
3402 | rdp->rcu_onl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
3403 | rdp->rcu_onl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
b97d23c5 PM |
3404 | if (rnp->qsmask & mask) { /* RCU waiting on incoming CPU? */ |
3405 | /* Report QS -after- changing ->qsmaskinitnext! */ | |
3406 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); | |
3407 | } else { | |
3408 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
7ec99de3 | 3409 | } |
313517fc | 3410 | smp_mb(); /* Ensure RCU read-side usage follows above initialization. */ |
7ec99de3 PM |
3411 | } |
3412 | ||
27d50c7e TG |
3413 | #ifdef CONFIG_HOTPLUG_CPU |
3414 | /* | |
53b46303 PM |
3415 | * The outgoing function has no further need of RCU, so remove it from |
3416 | * the rcu_node tree's ->qsmaskinitnext bit masks. | |
3417 | * | |
3418 | * Note that this function is special in that it is invoked directly | |
3419 | * from the outgoing CPU rather than from the cpuhp_step mechanism. | |
3420 | * This is because this function must be invoked at a precise location. | |
27d50c7e | 3421 | */ |
53b46303 | 3422 | void rcu_report_dead(unsigned int cpu) |
27d50c7e TG |
3423 | { |
3424 | unsigned long flags; | |
3425 | unsigned long mask; | |
da1df50d | 3426 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27d50c7e TG |
3427 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
3428 | ||
49918a54 | 3429 | /* QS for any half-done expedited grace period. */ |
53b46303 | 3430 | preempt_disable(); |
63d4c8c9 | 3431 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); |
53b46303 PM |
3432 | preempt_enable(); |
3433 | rcu_preempt_deferred_qs(current); | |
3434 | ||
27d50c7e TG |
3435 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ |
3436 | mask = rdp->grpmask; | |
894d45bb | 3437 | raw_spin_lock(&rcu_state.ofl_lock); |
27d50c7e | 3438 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ |
53b46303 PM |
3439 | rdp->rcu_ofl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
3440 | rdp->rcu_ofl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
fece2776 PM |
3441 | if (rnp->qsmask & mask) { /* RCU waiting on outgoing CPU? */ |
3442 | /* Report quiescent state -before- changing ->qsmaskinitnext! */ | |
b50912d0 | 3443 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
fece2776 PM |
3444 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
3445 | } | |
27d50c7e | 3446 | rnp->qsmaskinitnext &= ~mask; |
710d60cb | 3447 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
894d45bb | 3448 | raw_spin_unlock(&rcu_state.ofl_lock); |
f64c6013 PZ |
3449 | |
3450 | per_cpu(rcu_cpu_started, cpu) = 0; | |
27d50c7e | 3451 | } |
a58163d8 | 3452 | |
53b46303 PM |
3453 | /* |
3454 | * The outgoing CPU has just passed through the dying-idle state, and we | |
3455 | * are being invoked from the CPU that was IPIed to continue the offline | |
3456 | * operation. Migrate the outgoing CPU's callbacks to the current CPU. | |
3457 | */ | |
3458 | void rcutree_migrate_callbacks(int cpu) | |
a58163d8 PM |
3459 | { |
3460 | unsigned long flags; | |
b1a2d79f | 3461 | struct rcu_data *my_rdp; |
da1df50d | 3462 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
336a4f6c | 3463 | struct rcu_node *rnp_root = rcu_get_root(); |
ec4eacce | 3464 | bool needwake; |
a58163d8 | 3465 | |
95335c03 PM |
3466 | if (rcu_is_nocb_cpu(cpu) || rcu_segcblist_empty(&rdp->cblist)) |
3467 | return; /* No callbacks to migrate. */ | |
3468 | ||
b1a2d79f | 3469 | local_irq_save(flags); |
da1df50d | 3470 | my_rdp = this_cpu_ptr(&rcu_data); |
b1a2d79f PM |
3471 | if (rcu_nocb_adopt_orphan_cbs(my_rdp, rdp, flags)) { |
3472 | local_irq_restore(flags); | |
3473 | return; | |
3474 | } | |
9fa46fb8 | 3475 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ |
ec4eacce | 3476 | /* Leverage recent GPs and set GP for new callbacks. */ |
834f56bf PM |
3477 | needwake = rcu_advance_cbs(rnp_root, rdp) || |
3478 | rcu_advance_cbs(rnp_root, my_rdp); | |
f2dbe4a5 | 3479 | rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist); |
09efeeee PM |
3480 | WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) != |
3481 | !rcu_segcblist_n_cbs(&my_rdp->cblist)); | |
537b85c8 | 3482 | raw_spin_unlock_irqrestore_rcu_node(rnp_root, flags); |
ec4eacce | 3483 | if (needwake) |
532c00c9 | 3484 | rcu_gp_kthread_wake(); |
a58163d8 PM |
3485 | WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 || |
3486 | !rcu_segcblist_empty(&rdp->cblist), | |
3487 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n", | |
3488 | cpu, rcu_segcblist_n_cbs(&rdp->cblist), | |
3489 | rcu_segcblist_first_cb(&rdp->cblist)); | |
3490 | } | |
27d50c7e TG |
3491 | #endif |
3492 | ||
deb34f36 PM |
3493 | /* |
3494 | * On non-huge systems, use expedited RCU grace periods to make suspend | |
3495 | * and hibernation run faster. | |
3496 | */ | |
d1d74d14 BP |
3497 | static int rcu_pm_notify(struct notifier_block *self, |
3498 | unsigned long action, void *hcpu) | |
3499 | { | |
3500 | switch (action) { | |
3501 | case PM_HIBERNATION_PREPARE: | |
3502 | case PM_SUSPEND_PREPARE: | |
3503 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ | |
5afff48b | 3504 | rcu_expedite_gp(); |
d1d74d14 BP |
3505 | break; |
3506 | case PM_POST_HIBERNATION: | |
3507 | case PM_POST_SUSPEND: | |
5afff48b PM |
3508 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ |
3509 | rcu_unexpedite_gp(); | |
d1d74d14 BP |
3510 | break; |
3511 | default: | |
3512 | break; | |
3513 | } | |
3514 | return NOTIFY_OK; | |
3515 | } | |
3516 | ||
b3dbec76 | 3517 | /* |
49918a54 | 3518 | * Spawn the kthreads that handle RCU's grace periods. |
b3dbec76 PM |
3519 | */ |
3520 | static int __init rcu_spawn_gp_kthread(void) | |
3521 | { | |
3522 | unsigned long flags; | |
a94844b2 | 3523 | int kthread_prio_in = kthread_prio; |
b3dbec76 | 3524 | struct rcu_node *rnp; |
a94844b2 | 3525 | struct sched_param sp; |
b3dbec76 PM |
3526 | struct task_struct *t; |
3527 | ||
a94844b2 | 3528 | /* Force priority into range. */ |
c7cd161e JFG |
3529 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 2 |
3530 | && IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)) | |
3531 | kthread_prio = 2; | |
3532 | else if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
a94844b2 PM |
3533 | kthread_prio = 1; |
3534 | else if (kthread_prio < 0) | |
3535 | kthread_prio = 0; | |
3536 | else if (kthread_prio > 99) | |
3537 | kthread_prio = 99; | |
c7cd161e | 3538 | |
a94844b2 PM |
3539 | if (kthread_prio != kthread_prio_in) |
3540 | pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n", | |
3541 | kthread_prio, kthread_prio_in); | |
3542 | ||
9386c0b7 | 3543 | rcu_scheduler_fully_active = 1; |
b97d23c5 | 3544 | t = kthread_create(rcu_gp_kthread, NULL, "%s", rcu_state.name); |
08543bda PM |
3545 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start grace-period kthread, OOM is now expected behavior\n", __func__)) |
3546 | return 0; | |
b97d23c5 PM |
3547 | rnp = rcu_get_root(); |
3548 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3549 | rcu_state.gp_kthread = t; | |
3550 | if (kthread_prio) { | |
3551 | sp.sched_priority = kthread_prio; | |
3552 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
b3dbec76 | 3553 | } |
b97d23c5 PM |
3554 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
3555 | wake_up_process(t); | |
35ce7f29 | 3556 | rcu_spawn_nocb_kthreads(); |
9386c0b7 | 3557 | rcu_spawn_boost_kthreads(); |
b3dbec76 PM |
3558 | return 0; |
3559 | } | |
3560 | early_initcall(rcu_spawn_gp_kthread); | |
3561 | ||
bbad9379 | 3562 | /* |
52d7e48b PM |
3563 | * This function is invoked towards the end of the scheduler's |
3564 | * initialization process. Before this is called, the idle task might | |
3565 | * contain synchronous grace-period primitives (during which time, this idle | |
3566 | * task is booting the system, and such primitives are no-ops). After this | |
3567 | * function is called, any synchronous grace-period primitives are run as | |
3568 | * expedited, with the requesting task driving the grace period forward. | |
900b1028 | 3569 | * A later core_initcall() rcu_set_runtime_mode() will switch to full |
52d7e48b | 3570 | * runtime RCU functionality. |
bbad9379 PM |
3571 | */ |
3572 | void rcu_scheduler_starting(void) | |
3573 | { | |
3574 | WARN_ON(num_online_cpus() != 1); | |
3575 | WARN_ON(nr_context_switches() > 0); | |
52d7e48b PM |
3576 | rcu_test_sync_prims(); |
3577 | rcu_scheduler_active = RCU_SCHEDULER_INIT; | |
3578 | rcu_test_sync_prims(); | |
bbad9379 PM |
3579 | } |
3580 | ||
64db4cff | 3581 | /* |
49918a54 | 3582 | * Helper function for rcu_init() that initializes the rcu_state structure. |
64db4cff | 3583 | */ |
b8bb1f63 | 3584 | static void __init rcu_init_one(void) |
64db4cff | 3585 | { |
cb007102 AG |
3586 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
3587 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
3588 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
3589 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
199977bf | 3590 | |
199977bf | 3591 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ |
64db4cff PM |
3592 | int cpustride = 1; |
3593 | int i; | |
3594 | int j; | |
3595 | struct rcu_node *rnp; | |
3596 | ||
05b84aec | 3597 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 3598 | |
3eaaaf6c PM |
3599 | /* Silence gcc 4.8 false positive about array index out of range. */ |
3600 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
3601 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 3602 | |
64db4cff PM |
3603 | /* Initialize the level-tracking arrays. */ |
3604 | ||
f885b7f2 | 3605 | for (i = 1; i < rcu_num_lvls; i++) |
eb7a6653 PM |
3606 | rcu_state.level[i] = |
3607 | rcu_state.level[i - 1] + num_rcu_lvl[i - 1]; | |
41f5c631 | 3608 | rcu_init_levelspread(levelspread, num_rcu_lvl); |
64db4cff PM |
3609 | |
3610 | /* Initialize the elements themselves, starting from the leaves. */ | |
3611 | ||
f885b7f2 | 3612 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 3613 | cpustride *= levelspread[i]; |
eb7a6653 | 3614 | rnp = rcu_state.level[i]; |
41f5c631 | 3615 | for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) { |
67c583a7 BF |
3616 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
3617 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 3618 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
3619 | raw_spin_lock_init(&rnp->fqslock); |
3620 | lockdep_set_class_and_name(&rnp->fqslock, | |
3621 | &rcu_fqs_class[i], fqs[i]); | |
eb7a6653 PM |
3622 | rnp->gp_seq = rcu_state.gp_seq; |
3623 | rnp->gp_seq_needed = rcu_state.gp_seq; | |
3624 | rnp->completedqs = rcu_state.gp_seq; | |
64db4cff PM |
3625 | rnp->qsmask = 0; |
3626 | rnp->qsmaskinit = 0; | |
3627 | rnp->grplo = j * cpustride; | |
3628 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
3629 | if (rnp->grphi >= nr_cpu_ids) |
3630 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
3631 | if (i == 0) { |
3632 | rnp->grpnum = 0; | |
3633 | rnp->grpmask = 0; | |
3634 | rnp->parent = NULL; | |
3635 | } else { | |
199977bf | 3636 | rnp->grpnum = j % levelspread[i - 1]; |
df63fa5b | 3637 | rnp->grpmask = BIT(rnp->grpnum); |
eb7a6653 | 3638 | rnp->parent = rcu_state.level[i - 1] + |
199977bf | 3639 | j / levelspread[i - 1]; |
64db4cff PM |
3640 | } |
3641 | rnp->level = i; | |
12f5f524 | 3642 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 3643 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
3644 | init_waitqueue_head(&rnp->exp_wq[0]); |
3645 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
3646 | init_waitqueue_head(&rnp->exp_wq[2]); |
3647 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 3648 | spin_lock_init(&rnp->exp_lock); |
64db4cff PM |
3649 | } |
3650 | } | |
0c34029a | 3651 | |
eb7a6653 PM |
3652 | init_swait_queue_head(&rcu_state.gp_wq); |
3653 | init_swait_queue_head(&rcu_state.expedited_wq); | |
aedf4ba9 | 3654 | rnp = rcu_first_leaf_node(); |
0c34029a | 3655 | for_each_possible_cpu(i) { |
4a90a068 | 3656 | while (i > rnp->grphi) |
0c34029a | 3657 | rnp++; |
da1df50d | 3658 | per_cpu_ptr(&rcu_data, i)->mynode = rnp; |
53b46303 | 3659 | rcu_boot_init_percpu_data(i); |
0c34029a | 3660 | } |
64db4cff PM |
3661 | } |
3662 | ||
f885b7f2 PM |
3663 | /* |
3664 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 3665 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
3666 | * the ->node array in the rcu_state structure. |
3667 | */ | |
3668 | static void __init rcu_init_geometry(void) | |
3669 | { | |
026ad283 | 3670 | ulong d; |
f885b7f2 | 3671 | int i; |
05b84aec | 3672 | int rcu_capacity[RCU_NUM_LVLS]; |
f885b7f2 | 3673 | |
026ad283 PM |
3674 | /* |
3675 | * Initialize any unspecified boot parameters. | |
3676 | * The default values of jiffies_till_first_fqs and | |
3677 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
3678 | * value, which is a function of HZ, then adding one for each | |
3679 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
3680 | */ | |
3681 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
3682 | if (jiffies_till_first_fqs == ULONG_MAX) | |
3683 | jiffies_till_first_fqs = d; | |
3684 | if (jiffies_till_next_fqs == ULONG_MAX) | |
3685 | jiffies_till_next_fqs = d; | |
c06aed0e PM |
3686 | if (jiffies_till_sched_qs == ULONG_MAX) |
3687 | adjust_jiffies_till_sched_qs(); | |
026ad283 | 3688 | |
f885b7f2 | 3689 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 3690 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 3691 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 3692 | return; |
a7538352 | 3693 | pr_info("Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n", |
39479098 | 3694 | rcu_fanout_leaf, nr_cpu_ids); |
f885b7f2 | 3695 | |
f885b7f2 | 3696 | /* |
ee968ac6 PM |
3697 | * The boot-time rcu_fanout_leaf parameter must be at least two |
3698 | * and cannot exceed the number of bits in the rcu_node masks. | |
3699 | * Complain and fall back to the compile-time values if this | |
3700 | * limit is exceeded. | |
f885b7f2 | 3701 | */ |
ee968ac6 | 3702 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 3703 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 3704 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
3705 | WARN_ON(1); |
3706 | return; | |
3707 | } | |
3708 | ||
f885b7f2 PM |
3709 | /* |
3710 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 3711 | * with the given number of levels. |
f885b7f2 | 3712 | */ |
9618138b | 3713 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 3714 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 3715 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
3716 | |
3717 | /* | |
75cf15a4 | 3718 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 3719 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 3720 | */ |
ee968ac6 PM |
3721 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
3722 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
3723 | WARN_ON(1); | |
3724 | return; | |
3725 | } | |
f885b7f2 | 3726 | |
679f9858 | 3727 | /* Calculate the number of levels in the tree. */ |
9618138b | 3728 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 3729 | } |
9618138b | 3730 | rcu_num_lvls = i + 1; |
679f9858 | 3731 | |
f885b7f2 | 3732 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 3733 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 3734 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
3735 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
3736 | } | |
f885b7f2 PM |
3737 | |
3738 | /* Calculate the total number of rcu_node structures. */ | |
3739 | rcu_num_nodes = 0; | |
679f9858 | 3740 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 3741 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
3742 | } |
3743 | ||
a3dc2948 PM |
3744 | /* |
3745 | * Dump out the structure of the rcu_node combining tree associated | |
49918a54 | 3746 | * with the rcu_state structure. |
a3dc2948 | 3747 | */ |
b8bb1f63 | 3748 | static void __init rcu_dump_rcu_node_tree(void) |
a3dc2948 PM |
3749 | { |
3750 | int level = 0; | |
3751 | struct rcu_node *rnp; | |
3752 | ||
3753 | pr_info("rcu_node tree layout dump\n"); | |
3754 | pr_info(" "); | |
aedf4ba9 | 3755 | rcu_for_each_node_breadth_first(rnp) { |
a3dc2948 PM |
3756 | if (rnp->level != level) { |
3757 | pr_cont("\n"); | |
3758 | pr_info(" "); | |
3759 | level = rnp->level; | |
3760 | } | |
3761 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
3762 | } | |
3763 | pr_cont("\n"); | |
3764 | } | |
3765 | ||
ad7c946b | 3766 | struct workqueue_struct *rcu_gp_wq; |
25f3d7ef | 3767 | struct workqueue_struct *rcu_par_gp_wq; |
ad7c946b | 3768 | |
9f680ab4 | 3769 | void __init rcu_init(void) |
64db4cff | 3770 | { |
017c4261 | 3771 | int cpu; |
9f680ab4 | 3772 | |
47627678 PM |
3773 | rcu_early_boot_tests(); |
3774 | ||
f41d911f | 3775 | rcu_bootup_announce(); |
f885b7f2 | 3776 | rcu_init_geometry(); |
b8bb1f63 | 3777 | rcu_init_one(); |
a3dc2948 | 3778 | if (dump_tree) |
b8bb1f63 | 3779 | rcu_dump_rcu_node_tree(); |
b5b39360 | 3780 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
3781 | |
3782 | /* | |
3783 | * We don't need protection against CPU-hotplug here because | |
3784 | * this is called early in boot, before either interrupts | |
3785 | * or the scheduler are operational. | |
3786 | */ | |
d1d74d14 | 3787 | pm_notifier(rcu_pm_notify, 0); |
7ec99de3 | 3788 | for_each_online_cpu(cpu) { |
4df83742 | 3789 | rcutree_prepare_cpu(cpu); |
7ec99de3 | 3790 | rcu_cpu_starting(cpu); |
9b9500da | 3791 | rcutree_online_cpu(cpu); |
7ec99de3 | 3792 | } |
ad7c946b PM |
3793 | |
3794 | /* Create workqueue for expedited GPs and for Tree SRCU. */ | |
3795 | rcu_gp_wq = alloc_workqueue("rcu_gp", WQ_MEM_RECLAIM, 0); | |
3796 | WARN_ON(!rcu_gp_wq); | |
25f3d7ef PM |
3797 | rcu_par_gp_wq = alloc_workqueue("rcu_par_gp", WQ_MEM_RECLAIM, 0); |
3798 | WARN_ON(!rcu_par_gp_wq); | |
e0fcba9a | 3799 | srcu_init(); |
64db4cff PM |
3800 | } |
3801 | ||
3549c2bc | 3802 | #include "tree_exp.h" |
4102adab | 3803 | #include "tree_plugin.h" |