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> |
64db4cff | 64 | |
4102adab | 65 | #include "tree.h" |
29c00b4a | 66 | #include "rcu.h" |
9f77da9f | 67 | |
4102adab PM |
68 | #ifdef MODULE_PARAM_PREFIX |
69 | #undef MODULE_PARAM_PREFIX | |
70 | #endif | |
71 | #define MODULE_PARAM_PREFIX "rcutree." | |
72 | ||
64db4cff PM |
73 | /* Data structures. */ |
74 | ||
358be2d3 PM |
75 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, rcu_data); |
76 | struct rcu_state rcu_state = { | |
77 | .level = { &rcu_state.node[0] }, | |
358be2d3 PM |
78 | .gp_state = RCU_GP_IDLE, |
79 | .gp_seq = (0UL - 300UL) << RCU_SEQ_CTR_SHIFT, | |
80 | .barrier_mutex = __MUTEX_INITIALIZER(rcu_state.barrier_mutex), | |
81 | .name = RCU_NAME, | |
82 | .abbr = RCU_ABBR, | |
83 | .exp_mutex = __MUTEX_INITIALIZER(rcu_state.exp_mutex), | |
84 | .exp_wake_mutex = __MUTEX_INITIALIZER(rcu_state.exp_wake_mutex), | |
85 | .ofl_lock = __SPIN_LOCK_UNLOCKED(rcu_state.ofl_lock), | |
86 | }; | |
b1f77b05 | 87 | |
a3dc2948 PM |
88 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
89 | static bool dump_tree; | |
90 | module_param(dump_tree, bool, 0444); | |
7fa27001 PM |
91 | /* Control rcu_node-tree auto-balancing at boot time. */ |
92 | static bool rcu_fanout_exact; | |
93 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
94 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
95 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 96 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 97 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 | 98 | /* Number of rcu_nodes at specified level. */ |
e95d68d2 | 99 | int num_rcu_lvl[] = NUM_RCU_LVL_INIT; |
f885b7f2 | 100 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
088e9d25 DBO |
101 | /* panic() on RCU Stall sysctl. */ |
102 | int sysctl_panic_on_rcu_stall __read_mostly; | |
f885b7f2 | 103 | |
b0d30417 | 104 | /* |
52d7e48b PM |
105 | * The rcu_scheduler_active variable is initialized to the value |
106 | * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the | |
107 | * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE, | |
108 | * RCU can assume that there is but one task, allowing RCU to (for example) | |
0d95092c | 109 | * optimize synchronize_rcu() to a simple barrier(). When this variable |
52d7e48b PM |
110 | * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required |
111 | * to detect real grace periods. This variable is also used to suppress | |
112 | * boot-time false positives from lockdep-RCU error checking. Finally, it | |
113 | * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU | |
114 | * is fully initialized, including all of its kthreads having been spawned. | |
b0d30417 | 115 | */ |
bbad9379 PM |
116 | int rcu_scheduler_active __read_mostly; |
117 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
118 | ||
b0d30417 PM |
119 | /* |
120 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
121 | * during the early_initcall() processing, which is after the scheduler | |
122 | * is capable of creating new tasks. So RCU processing (for example, | |
123 | * creating tasks for RCU priority boosting) must be delayed until after | |
124 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
125 | * currently delay invocation of any RCU callbacks until after this point. | |
126 | * | |
127 | * It might later prove better for people registering RCU callbacks during | |
128 | * early boot to take responsibility for these callbacks, but one step at | |
129 | * a time. | |
130 | */ | |
131 | static int rcu_scheduler_fully_active __read_mostly; | |
132 | ||
b50912d0 PM |
133 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
134 | unsigned long gps, unsigned long flags); | |
0aa04b05 PM |
135 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
136 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 137 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 | 138 | static void invoke_rcu_core(void); |
aff4e9ed | 139 | static void invoke_rcu_callbacks(struct rcu_data *rdp); |
63d4c8c9 | 140 | static void rcu_report_exp_rdp(struct rcu_data *rdp); |
3549c2bc | 141 | static void sync_sched_exp_online_cleanup(int cpu); |
a26ac245 | 142 | |
a94844b2 | 143 | /* rcuc/rcub kthread realtime priority */ |
26730f55 | 144 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; |
a94844b2 PM |
145 | module_param(kthread_prio, int, 0644); |
146 | ||
8d7dc928 | 147 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd | 148 | |
90040c9e PM |
149 | static int gp_preinit_delay; |
150 | module_param(gp_preinit_delay, int, 0444); | |
151 | static int gp_init_delay; | |
152 | module_param(gp_init_delay, int, 0444); | |
153 | static int gp_cleanup_delay; | |
154 | module_param(gp_cleanup_delay, int, 0444); | |
0f41c0dd | 155 | |
4cf439a2 | 156 | /* Retrieve RCU kthreads priority for rcutorture */ |
4babd855 JFG |
157 | int rcu_get_gp_kthreads_prio(void) |
158 | { | |
159 | return kthread_prio; | |
160 | } | |
161 | EXPORT_SYMBOL_GPL(rcu_get_gp_kthreads_prio); | |
162 | ||
eab128e8 PM |
163 | /* |
164 | * Number of grace periods between delays, normalized by the duration of | |
bfd090be | 165 | * the delay. The longer the delay, the more the grace periods between |
eab128e8 PM |
166 | * each delay. The reason for this normalization is that it means that, |
167 | * for non-zero delays, the overall slowdown of grace periods is constant | |
168 | * regardless of the duration of the delay. This arrangement balances | |
169 | * the need for long delays to increase some race probabilities with the | |
170 | * need for fast grace periods to increase other race probabilities. | |
171 | */ | |
172 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */ | |
37745d28 | 173 | |
0aa04b05 PM |
174 | /* |
175 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
176 | * This will not be stable unless the rcu_node structure's ->lock is | |
177 | * held, but the bit corresponding to the current CPU will be stable | |
178 | * in most contexts. | |
179 | */ | |
180 | unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) | |
181 | { | |
7d0ae808 | 182 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
183 | } |
184 | ||
fc2219d4 | 185 | /* |
7d0ae808 | 186 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
187 | * permit this function to be invoked without holding the root rcu_node |
188 | * structure's ->lock, but of course results can be subject to change. | |
189 | */ | |
de8e8730 | 190 | static int rcu_gp_in_progress(void) |
fc2219d4 | 191 | { |
de8e8730 | 192 | return rcu_seq_state(rcu_seq_current(&rcu_state.gp_seq)); |
fc2219d4 PM |
193 | } |
194 | ||
d28139c4 PM |
195 | void rcu_softirq_qs(void) |
196 | { | |
45975c7d | 197 | rcu_qs(); |
d28139c4 PM |
198 | rcu_preempt_deferred_qs(current); |
199 | } | |
200 | ||
b8c17e66 PM |
201 | /* |
202 | * Steal a bit from the bottom of ->dynticks for idle entry/exit | |
203 | * control. Initially this is for TLB flushing. | |
204 | */ | |
205 | #define RCU_DYNTICK_CTRL_MASK 0x1 | |
206 | #define RCU_DYNTICK_CTRL_CTR (RCU_DYNTICK_CTRL_MASK + 1) | |
207 | #ifndef rcu_eqs_special_exit | |
208 | #define rcu_eqs_special_exit() do { } while (0) | |
209 | #endif | |
4a81e832 PM |
210 | |
211 | static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { | |
51a1fd30 | 212 | .dynticks_nesting = 1, |
58721f5d | 213 | .dynticks_nmi_nesting = DYNTICK_IRQ_NONIDLE, |
b8c17e66 | 214 | .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR), |
4a81e832 PM |
215 | }; |
216 | ||
2625d469 PM |
217 | /* |
218 | * Record entry into an extended quiescent state. This is only to be | |
219 | * called when not already in an extended quiescent state. | |
220 | */ | |
221 | static void rcu_dynticks_eqs_enter(void) | |
222 | { | |
223 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 224 | int seq; |
2625d469 PM |
225 | |
226 | /* | |
b8c17e66 | 227 | * CPUs seeing atomic_add_return() must see prior RCU read-side |
2625d469 PM |
228 | * critical sections, and we also must force ordering with the |
229 | * next idle sojourn. | |
230 | */ | |
b8c17e66 PM |
231 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
232 | /* Better be in an extended quiescent state! */ | |
233 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
234 | (seq & RCU_DYNTICK_CTRL_CTR)); | |
235 | /* Better not have special action (TLB flush) pending! */ | |
236 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
237 | (seq & RCU_DYNTICK_CTRL_MASK)); | |
2625d469 PM |
238 | } |
239 | ||
240 | /* | |
241 | * Record exit from an extended quiescent state. This is only to be | |
242 | * called from an extended quiescent state. | |
243 | */ | |
244 | static void rcu_dynticks_eqs_exit(void) | |
245 | { | |
246 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 247 | int seq; |
2625d469 PM |
248 | |
249 | /* | |
b8c17e66 | 250 | * CPUs seeing atomic_add_return() must see prior idle sojourns, |
2625d469 PM |
251 | * and we also must force ordering with the next RCU read-side |
252 | * critical section. | |
253 | */ | |
b8c17e66 PM |
254 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
255 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
256 | !(seq & RCU_DYNTICK_CTRL_CTR)); | |
257 | if (seq & RCU_DYNTICK_CTRL_MASK) { | |
258 | atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdtp->dynticks); | |
259 | smp_mb__after_atomic(); /* _exit after clearing mask. */ | |
260 | /* Prefer duplicate flushes to losing a flush. */ | |
261 | rcu_eqs_special_exit(); | |
262 | } | |
2625d469 PM |
263 | } |
264 | ||
265 | /* | |
266 | * Reset the current CPU's ->dynticks counter to indicate that the | |
267 | * newly onlined CPU is no longer in an extended quiescent state. | |
268 | * This will either leave the counter unchanged, or increment it | |
269 | * to the next non-quiescent value. | |
270 | * | |
271 | * The non-atomic test/increment sequence works because the upper bits | |
272 | * of the ->dynticks counter are manipulated only by the corresponding CPU, | |
273 | * or when the corresponding CPU is offline. | |
274 | */ | |
275 | static void rcu_dynticks_eqs_online(void) | |
276 | { | |
277 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
278 | ||
b8c17e66 | 279 | if (atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR) |
2625d469 | 280 | return; |
b8c17e66 | 281 | atomic_add(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
2625d469 PM |
282 | } |
283 | ||
02a5c550 PM |
284 | /* |
285 | * Is the current CPU in an extended quiescent state? | |
286 | * | |
287 | * No ordering, as we are sampling CPU-local information. | |
288 | */ | |
289 | bool rcu_dynticks_curr_cpu_in_eqs(void) | |
290 | { | |
291 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
292 | ||
b8c17e66 | 293 | return !(atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
294 | } |
295 | ||
8b2f63ab PM |
296 | /* |
297 | * Snapshot the ->dynticks counter with full ordering so as to allow | |
298 | * stable comparison of this counter with past and future snapshots. | |
299 | */ | |
02a5c550 | 300 | int rcu_dynticks_snap(struct rcu_dynticks *rdtp) |
8b2f63ab PM |
301 | { |
302 | int snap = atomic_add_return(0, &rdtp->dynticks); | |
303 | ||
b8c17e66 | 304 | return snap & ~RCU_DYNTICK_CTRL_MASK; |
8b2f63ab PM |
305 | } |
306 | ||
02a5c550 PM |
307 | /* |
308 | * Return true if the snapshot returned from rcu_dynticks_snap() | |
309 | * indicates that RCU is in an extended quiescent state. | |
310 | */ | |
311 | static bool rcu_dynticks_in_eqs(int snap) | |
312 | { | |
b8c17e66 | 313 | return !(snap & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
314 | } |
315 | ||
316 | /* | |
317 | * Return true if the CPU corresponding to the specified rcu_dynticks | |
318 | * structure has spent some time in an extended quiescent state since | |
319 | * rcu_dynticks_snap() returned the specified snapshot. | |
320 | */ | |
321 | static bool rcu_dynticks_in_eqs_since(struct rcu_dynticks *rdtp, int snap) | |
322 | { | |
323 | return snap != rcu_dynticks_snap(rdtp); | |
324 | } | |
325 | ||
b8c17e66 PM |
326 | /* |
327 | * Set the special (bottom) bit of the specified CPU so that it | |
328 | * will take special action (such as flushing its TLB) on the | |
329 | * next exit from an extended quiescent state. Returns true if | |
330 | * the bit was successfully set, or false if the CPU was not in | |
331 | * an extended quiescent state. | |
332 | */ | |
333 | bool rcu_eqs_special_set(int cpu) | |
334 | { | |
335 | int old; | |
336 | int new; | |
337 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); | |
338 | ||
339 | do { | |
340 | old = atomic_read(&rdtp->dynticks); | |
341 | if (old & RCU_DYNTICK_CTRL_CTR) | |
342 | return false; | |
343 | new = old | RCU_DYNTICK_CTRL_MASK; | |
344 | } while (atomic_cmpxchg(&rdtp->dynticks, old, new) != old); | |
345 | return true; | |
6563de9d | 346 | } |
5cd37193 | 347 | |
4a81e832 PM |
348 | /* |
349 | * Let the RCU core know that this CPU has gone through the scheduler, | |
350 | * which is a quiescent state. This is called when the need for a | |
351 | * quiescent state is urgent, so we burn an atomic operation and full | |
352 | * memory barriers to let the RCU core know about it, regardless of what | |
353 | * this CPU might (or might not) do in the near future. | |
354 | * | |
0f9be8ca | 355 | * We inform the RCU core by emulating a zero-duration dyntick-idle period. |
46a5d164 | 356 | * |
3b57a399 | 357 | * The caller must have disabled interrupts and must not be idle. |
4a81e832 PM |
358 | */ |
359 | static void rcu_momentary_dyntick_idle(void) | |
360 | { | |
3b57a399 PM |
361 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
362 | int special; | |
363 | ||
0f9be8ca | 364 | raw_cpu_write(rcu_dynticks.rcu_need_heavy_qs, false); |
3b57a399 PM |
365 | special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
366 | /* It is illegal to call this from idle state. */ | |
367 | WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR)); | |
3e310098 | 368 | rcu_preempt_deferred_qs(current); |
4a81e832 PM |
369 | } |
370 | ||
45975c7d PM |
371 | /** |
372 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle | |
373 | * | |
374 | * If the current CPU is idle or running at a first-level (not nested) | |
375 | * interrupt from idle, return true. The caller must have at least | |
376 | * disabled preemption. | |
25502a6c | 377 | */ |
45975c7d | 378 | static int rcu_is_cpu_rrupt_from_idle(void) |
25502a6c | 379 | { |
45975c7d PM |
380 | return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 0 && |
381 | __this_cpu_read(rcu_dynticks.dynticks_nmi_nesting) <= 1; | |
25502a6c PM |
382 | } |
383 | ||
5cd37193 | 384 | /* |
1925d196 | 385 | * Register a quiescent state for all RCU flavors. If there is an |
5cd37193 PM |
386 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight |
387 | * dyntick-idle quiescent state visible to other CPUs (but only for those | |
1925d196 | 388 | * RCU flavors in desperate need of a quiescent state, which will normally |
5cd37193 PM |
389 | * be none of them). Either way, do a lightweight quiescent state for |
390 | * all RCU flavors. | |
bb73c52b BF |
391 | * |
392 | * The barrier() calls are redundant in the common case when this is | |
393 | * called externally, but just in case this is called from within this | |
394 | * file. | |
395 | * | |
5cd37193 PM |
396 | */ |
397 | void rcu_all_qs(void) | |
398 | { | |
46a5d164 PM |
399 | unsigned long flags; |
400 | ||
9226b10d PM |
401 | if (!raw_cpu_read(rcu_dynticks.rcu_urgent_qs)) |
402 | return; | |
403 | preempt_disable(); | |
404 | /* Load rcu_urgent_qs before other flags. */ | |
405 | if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs))) { | |
406 | preempt_enable(); | |
407 | return; | |
408 | } | |
409 | this_cpu_write(rcu_dynticks.rcu_urgent_qs, false); | |
bb73c52b | 410 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
0f9be8ca | 411 | if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) { |
46a5d164 | 412 | local_irq_save(flags); |
5cd37193 | 413 | rcu_momentary_dyntick_idle(); |
46a5d164 PM |
414 | local_irq_restore(flags); |
415 | } | |
45975c7d PM |
416 | if (unlikely(raw_cpu_read(rcu_data.cpu_no_qs.b.exp))) |
417 | rcu_qs(); | |
9577df9a | 418 | this_cpu_inc(rcu_dynticks.rcu_qs_ctr); |
bb73c52b | 419 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
9226b10d | 420 | preempt_enable(); |
5cd37193 PM |
421 | } |
422 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
423 | ||
17c7798b PM |
424 | #define DEFAULT_RCU_BLIMIT 10 /* Maximum callbacks per rcu_do_batch. */ |
425 | static long blimit = DEFAULT_RCU_BLIMIT; | |
426 | #define DEFAULT_RCU_QHIMARK 10000 /* If this many pending, ignore blimit. */ | |
427 | static long qhimark = DEFAULT_RCU_QHIMARK; | |
428 | #define DEFAULT_RCU_QLOMARK 100 /* Once only this many pending, use blimit. */ | |
429 | static long qlowmark = DEFAULT_RCU_QLOMARK; | |
64db4cff | 430 | |
878d7439 ED |
431 | module_param(blimit, long, 0444); |
432 | module_param(qhimark, long, 0444); | |
433 | module_param(qlowmark, long, 0444); | |
3d76c082 | 434 | |
026ad283 PM |
435 | static ulong jiffies_till_first_fqs = ULONG_MAX; |
436 | static ulong jiffies_till_next_fqs = ULONG_MAX; | |
8c7c4829 | 437 | static bool rcu_kick_kthreads; |
d40011f6 | 438 | |
67abb96c BP |
439 | static int param_set_first_fqs_jiffies(const char *val, const struct kernel_param *kp) |
440 | { | |
441 | ulong j; | |
442 | int ret = kstrtoul(val, 0, &j); | |
443 | ||
444 | if (!ret) | |
445 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : j); | |
446 | return ret; | |
447 | } | |
448 | ||
449 | static int param_set_next_fqs_jiffies(const char *val, const struct kernel_param *kp) | |
450 | { | |
451 | ulong j; | |
452 | int ret = kstrtoul(val, 0, &j); | |
453 | ||
454 | if (!ret) | |
455 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : (j ?: 1)); | |
456 | return ret; | |
457 | } | |
458 | ||
459 | static struct kernel_param_ops first_fqs_jiffies_ops = { | |
460 | .set = param_set_first_fqs_jiffies, | |
461 | .get = param_get_ulong, | |
462 | }; | |
463 | ||
464 | static struct kernel_param_ops next_fqs_jiffies_ops = { | |
465 | .set = param_set_next_fqs_jiffies, | |
466 | .get = param_get_ulong, | |
467 | }; | |
468 | ||
469 | module_param_cb(jiffies_till_first_fqs, &first_fqs_jiffies_ops, &jiffies_till_first_fqs, 0644); | |
470 | module_param_cb(jiffies_till_next_fqs, &next_fqs_jiffies_ops, &jiffies_till_next_fqs, 0644); | |
8c7c4829 | 471 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 472 | |
4a81e832 PM |
473 | /* |
474 | * How long the grace period must be before we start recruiting | |
475 | * quiescent-state help from rcu_note_context_switch(). | |
476 | */ | |
f79c3ad6 PM |
477 | static ulong jiffies_till_sched_qs = HZ / 10; |
478 | module_param(jiffies_till_sched_qs, ulong, 0444); | |
4a81e832 | 479 | |
8ff0b907 | 480 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)); |
e9ecb780 | 481 | static void force_quiescent_state(void); |
e3950ecd | 482 | static int rcu_pending(void); |
64db4cff PM |
483 | |
484 | /* | |
17ef2fe9 | 485 | * Return the number of RCU GPs completed thus far for debug & stats. |
64db4cff | 486 | */ |
17ef2fe9 | 487 | unsigned long rcu_get_gp_seq(void) |
917963d0 | 488 | { |
16fc9c60 | 489 | return READ_ONCE(rcu_state.gp_seq); |
917963d0 | 490 | } |
17ef2fe9 | 491 | EXPORT_SYMBOL_GPL(rcu_get_gp_seq); |
917963d0 | 492 | |
291783b8 PM |
493 | /* |
494 | * Return the number of RCU expedited batches completed thus far for | |
495 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
496 | * numbers mean idle. The value returned will thus be roughly double | |
497 | * the cumulative batches since boot. | |
498 | */ | |
499 | unsigned long rcu_exp_batches_completed(void) | |
500 | { | |
16fc9c60 | 501 | return rcu_state.expedited_sequence; |
291783b8 PM |
502 | } |
503 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
504 | ||
a381d757 ACB |
505 | /* |
506 | * Force a quiescent state. | |
507 | */ | |
508 | void rcu_force_quiescent_state(void) | |
509 | { | |
e9ecb780 | 510 | force_quiescent_state(); |
a381d757 ACB |
511 | } |
512 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
513 | ||
afea227f PM |
514 | /* |
515 | * Show the state of the grace-period kthreads. | |
516 | */ | |
517 | void show_rcu_gp_kthreads(void) | |
518 | { | |
47199a08 PM |
519 | int cpu; |
520 | struct rcu_data *rdp; | |
521 | struct rcu_node *rnp; | |
afea227f | 522 | |
b97d23c5 PM |
523 | pr_info("%s: wait state: %d ->state: %#lx\n", rcu_state.name, |
524 | rcu_state.gp_state, rcu_state.gp_kthread->state); | |
525 | rcu_for_each_node_breadth_first(rnp) { | |
526 | if (ULONG_CMP_GE(rcu_state.gp_seq, rnp->gp_seq_needed)) | |
527 | continue; | |
528 | pr_info("\trcu_node %d:%d ->gp_seq %lu ->gp_seq_needed %lu\n", | |
529 | rnp->grplo, rnp->grphi, rnp->gp_seq, | |
530 | rnp->gp_seq_needed); | |
531 | if (!rcu_is_leaf_node(rnp)) | |
532 | continue; | |
533 | for_each_leaf_node_possible_cpu(rnp, cpu) { | |
534 | rdp = per_cpu_ptr(&rcu_data, cpu); | |
535 | if (rdp->gpwrap || | |
536 | ULONG_CMP_GE(rcu_state.gp_seq, | |
537 | rdp->gp_seq_needed)) | |
47199a08 | 538 | continue; |
b97d23c5 PM |
539 | pr_info("\tcpu %d ->gp_seq_needed %lu\n", |
540 | cpu, rdp->gp_seq_needed); | |
47199a08 | 541 | } |
afea227f | 542 | } |
b97d23c5 | 543 | /* sched_show_task(rcu_state.gp_kthread); */ |
afea227f PM |
544 | } |
545 | EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads); | |
546 | ||
ad0dc7f9 PM |
547 | /* |
548 | * Send along grace-period-related data for rcutorture diagnostics. | |
549 | */ | |
550 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
aebc8264 | 551 | unsigned long *gp_seq) |
ad0dc7f9 | 552 | { |
ad0dc7f9 PM |
553 | switch (test_type) { |
554 | case RCU_FLAVOR: | |
ad0dc7f9 | 555 | case RCU_BH_FLAVOR: |
ad0dc7f9 | 556 | case RCU_SCHED_FLAVOR: |
f7dd7d44 PM |
557 | *flags = READ_ONCE(rcu_state.gp_flags); |
558 | *gp_seq = rcu_seq_current(&rcu_state.gp_seq); | |
ad0dc7f9 PM |
559 | break; |
560 | default: | |
561 | break; | |
562 | } | |
ad0dc7f9 PM |
563 | } |
564 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
565 | ||
365187fb PM |
566 | /* |
567 | * Return the root node of the specified rcu_state structure. | |
568 | */ | |
336a4f6c | 569 | static struct rcu_node *rcu_get_root(void) |
365187fb | 570 | { |
336a4f6c | 571 | return &rcu_state.node[0]; |
365187fb PM |
572 | } |
573 | ||
9b2e4f18 | 574 | /* |
215bba9f PM |
575 | * Enter an RCU extended quiescent state, which can be either the |
576 | * idle loop or adaptive-tickless usermode execution. | |
9b2e4f18 | 577 | * |
215bba9f PM |
578 | * We crowbar the ->dynticks_nmi_nesting field to zero to allow for |
579 | * the possibility of usermode upcalls having messed up our count | |
580 | * of interrupt nesting level during the prior busy period. | |
9b2e4f18 | 581 | */ |
215bba9f | 582 | static void rcu_eqs_enter(bool user) |
9b2e4f18 | 583 | { |
96d3fd0d | 584 | struct rcu_data *rdp; |
215bba9f | 585 | struct rcu_dynticks *rdtp; |
96d3fd0d | 586 | |
215bba9f | 587 | rdtp = this_cpu_ptr(&rcu_dynticks); |
e11ec65c | 588 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting != DYNTICK_IRQ_NONIDLE); |
215bba9f PM |
589 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0); |
590 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
591 | rdtp->dynticks_nesting == 0); | |
592 | if (rdtp->dynticks_nesting != 1) { | |
593 | rdtp->dynticks_nesting--; | |
594 | return; | |
9b2e4f18 | 595 | } |
96d3fd0d | 596 | |
b04db8e1 | 597 | lockdep_assert_irqs_disabled(); |
dec98900 | 598 | trace_rcu_dyntick(TPS("Start"), rdtp->dynticks_nesting, 0, rdtp->dynticks); |
e68bbb26 | 599 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
b97d23c5 PM |
600 | rdp = this_cpu_ptr(&rcu_data); |
601 | do_nocb_deferred_wakeup(rdp); | |
198bbf81 | 602 | rcu_prepare_for_idle(); |
3e310098 | 603 | rcu_preempt_deferred_qs(current); |
2342172f | 604 | WRITE_ONCE(rdtp->dynticks_nesting, 0); /* Avoid irq-access tearing. */ |
844ccdd7 | 605 | rcu_dynticks_eqs_enter(); |
176f8f7a | 606 | rcu_dynticks_task_enter(); |
64db4cff | 607 | } |
adf5091e FW |
608 | |
609 | /** | |
610 | * rcu_idle_enter - inform RCU that current CPU is entering idle | |
611 | * | |
612 | * Enter idle mode, in other words, -leave- the mode in which RCU | |
613 | * read-side critical sections can occur. (Though RCU read-side | |
614 | * critical sections can occur in irq handlers in idle, a possibility | |
615 | * handled by irq_enter() and irq_exit().) | |
616 | * | |
c0da313e PM |
617 | * If you add or remove a call to rcu_idle_enter(), be sure to test with |
618 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
619 | */ |
620 | void rcu_idle_enter(void) | |
621 | { | |
b04db8e1 | 622 | lockdep_assert_irqs_disabled(); |
cb349ca9 | 623 | rcu_eqs_enter(false); |
adf5091e | 624 | } |
64db4cff | 625 | |
d1ec4c34 | 626 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
627 | /** |
628 | * rcu_user_enter - inform RCU that we are resuming userspace. | |
629 | * | |
630 | * Enter RCU idle mode right before resuming userspace. No use of RCU | |
631 | * is permitted between this call and rcu_user_exit(). This way the | |
632 | * CPU doesn't need to maintain the tick for RCU maintenance purposes | |
633 | * when the CPU runs in userspace. | |
c0da313e PM |
634 | * |
635 | * If you add or remove a call to rcu_user_enter(), be sure to test with | |
636 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
637 | */ |
638 | void rcu_user_enter(void) | |
639 | { | |
b04db8e1 | 640 | lockdep_assert_irqs_disabled(); |
d4db30af | 641 | rcu_eqs_enter(true); |
adf5091e | 642 | } |
d1ec4c34 | 643 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 644 | |
cf7614e1 | 645 | /* |
fd581a91 PM |
646 | * If we are returning from the outermost NMI handler that interrupted an |
647 | * RCU-idle period, update rdtp->dynticks and rdtp->dynticks_nmi_nesting | |
648 | * to let the RCU grace-period handling know that the CPU is back to | |
649 | * being RCU-idle. | |
650 | * | |
cf7614e1 | 651 | * If you add or remove a call to rcu_nmi_exit_common(), be sure to test |
fd581a91 PM |
652 | * with CONFIG_RCU_EQS_DEBUG=y. |
653 | */ | |
cf7614e1 | 654 | static __always_inline void rcu_nmi_exit_common(bool irq) |
fd581a91 PM |
655 | { |
656 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
657 | ||
658 | /* | |
659 | * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks. | |
660 | * (We are exiting an NMI handler, so RCU better be paying attention | |
661 | * to us!) | |
662 | */ | |
663 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting <= 0); | |
664 | WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs()); | |
665 | ||
666 | /* | |
667 | * If the nesting level is not 1, the CPU wasn't RCU-idle, so | |
668 | * leave it in non-RCU-idle state. | |
669 | */ | |
670 | if (rdtp->dynticks_nmi_nesting != 1) { | |
dec98900 | 671 | trace_rcu_dyntick(TPS("--="), rdtp->dynticks_nmi_nesting, rdtp->dynticks_nmi_nesting - 2, rdtp->dynticks); |
fd581a91 PM |
672 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* No store tearing. */ |
673 | rdtp->dynticks_nmi_nesting - 2); | |
674 | return; | |
675 | } | |
676 | ||
677 | /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */ | |
dec98900 | 678 | trace_rcu_dyntick(TPS("Startirq"), rdtp->dynticks_nmi_nesting, 0, rdtp->dynticks); |
fd581a91 | 679 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */ |
cf7614e1 BP |
680 | |
681 | if (irq) | |
682 | rcu_prepare_for_idle(); | |
683 | ||
fd581a91 | 684 | rcu_dynticks_eqs_enter(); |
cf7614e1 BP |
685 | |
686 | if (irq) | |
687 | rcu_dynticks_task_enter(); | |
688 | } | |
689 | ||
690 | /** | |
691 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
692 | * @irq: Is this call from rcu_irq_exit? | |
693 | * | |
694 | * If you add or remove a call to rcu_nmi_exit(), be sure to test | |
695 | * with CONFIG_RCU_EQS_DEBUG=y. | |
696 | */ | |
697 | void rcu_nmi_exit(void) | |
698 | { | |
699 | rcu_nmi_exit_common(false); | |
fd581a91 PM |
700 | } |
701 | ||
9b2e4f18 PM |
702 | /** |
703 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
704 | * | |
705 | * Exit from an interrupt handler, which might possibly result in entering | |
706 | * idle mode, in other words, leaving the mode in which read-side critical | |
7c9906ca | 707 | * sections can occur. The caller must have disabled interrupts. |
64db4cff | 708 | * |
9b2e4f18 PM |
709 | * This code assumes that the idle loop never does anything that might |
710 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
58721f5d PM |
711 | * architecture's idle loop violates this assumption, RCU will give you what |
712 | * you deserve, good and hard. But very infrequently and irreproducibly. | |
9b2e4f18 PM |
713 | * |
714 | * Use things like work queues to work around this limitation. | |
715 | * | |
716 | * You have been warned. | |
c0da313e PM |
717 | * |
718 | * If you add or remove a call to rcu_irq_exit(), be sure to test with | |
719 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 720 | */ |
9b2e4f18 | 721 | void rcu_irq_exit(void) |
64db4cff | 722 | { |
b04db8e1 | 723 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 724 | rcu_nmi_exit_common(true); |
7c9906ca PM |
725 | } |
726 | ||
727 | /* | |
728 | * Wrapper for rcu_irq_exit() where interrupts are enabled. | |
c0da313e PM |
729 | * |
730 | * If you add or remove a call to rcu_irq_exit_irqson(), be sure to test | |
731 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
732 | */ |
733 | void rcu_irq_exit_irqson(void) | |
734 | { | |
735 | unsigned long flags; | |
736 | ||
737 | local_irq_save(flags); | |
738 | rcu_irq_exit(); | |
9b2e4f18 PM |
739 | local_irq_restore(flags); |
740 | } | |
741 | ||
adf5091e FW |
742 | /* |
743 | * Exit an RCU extended quiescent state, which can be either the | |
744 | * idle loop or adaptive-tickless usermode execution. | |
51a1fd30 PM |
745 | * |
746 | * We crowbar the ->dynticks_nmi_nesting field to DYNTICK_IRQ_NONIDLE to | |
747 | * allow for the possibility of usermode upcalls messing up our count of | |
748 | * interrupt nesting level during the busy period that is just now starting. | |
9b2e4f18 | 749 | */ |
adf5091e | 750 | static void rcu_eqs_exit(bool user) |
9b2e4f18 | 751 | { |
9b2e4f18 | 752 | struct rcu_dynticks *rdtp; |
84585aa8 | 753 | long oldval; |
9b2e4f18 | 754 | |
b04db8e1 | 755 | lockdep_assert_irqs_disabled(); |
c9d4b0af | 756 | rdtp = this_cpu_ptr(&rcu_dynticks); |
9b2e4f18 | 757 | oldval = rdtp->dynticks_nesting; |
1ce46ee5 | 758 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0); |
51a1fd30 PM |
759 | if (oldval) { |
760 | rdtp->dynticks_nesting++; | |
9dd238e2 | 761 | return; |
3a592405 | 762 | } |
9dd238e2 PM |
763 | rcu_dynticks_task_exit(); |
764 | rcu_dynticks_eqs_exit(); | |
765 | rcu_cleanup_after_idle(); | |
766 | trace_rcu_dyntick(TPS("End"), rdtp->dynticks_nesting, 1, rdtp->dynticks); | |
e68bbb26 | 767 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
9dd238e2 | 768 | WRITE_ONCE(rdtp->dynticks_nesting, 1); |
e11ec65c | 769 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting); |
9dd238e2 | 770 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, DYNTICK_IRQ_NONIDLE); |
9b2e4f18 | 771 | } |
adf5091e FW |
772 | |
773 | /** | |
774 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
775 | * | |
776 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
777 | * read-side critical sections can occur. | |
778 | * | |
c0da313e PM |
779 | * If you add or remove a call to rcu_idle_exit(), be sure to test with |
780 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
781 | */ |
782 | void rcu_idle_exit(void) | |
783 | { | |
c5d900bf FW |
784 | unsigned long flags; |
785 | ||
786 | local_irq_save(flags); | |
cb349ca9 | 787 | rcu_eqs_exit(false); |
c5d900bf | 788 | local_irq_restore(flags); |
adf5091e | 789 | } |
9b2e4f18 | 790 | |
d1ec4c34 | 791 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
792 | /** |
793 | * rcu_user_exit - inform RCU that we are exiting userspace. | |
794 | * | |
795 | * Exit RCU idle mode while entering the kernel because it can | |
796 | * run a RCU read side critical section anytime. | |
c0da313e PM |
797 | * |
798 | * If you add or remove a call to rcu_user_exit(), be sure to test with | |
799 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
800 | */ |
801 | void rcu_user_exit(void) | |
802 | { | |
91d1aa43 | 803 | rcu_eqs_exit(1); |
adf5091e | 804 | } |
d1ec4c34 | 805 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 806 | |
64db4cff | 807 | /** |
cf7614e1 BP |
808 | * rcu_nmi_enter_common - inform RCU of entry to NMI context |
809 | * @irq: Is this call from rcu_irq_enter? | |
64db4cff | 810 | * |
734d1680 PM |
811 | * If the CPU was idle from RCU's viewpoint, update rdtp->dynticks and |
812 | * rdtp->dynticks_nmi_nesting to let the RCU grace-period handling know | |
813 | * that the CPU is active. This implementation permits nested NMIs, as | |
814 | * long as the nesting level does not overflow an int. (You will probably | |
815 | * run out of stack space first.) | |
c0da313e | 816 | * |
cf7614e1 | 817 | * If you add or remove a call to rcu_nmi_enter_common(), be sure to test |
c0da313e | 818 | * with CONFIG_RCU_EQS_DEBUG=y. |
64db4cff | 819 | */ |
cf7614e1 | 820 | static __always_inline void rcu_nmi_enter_common(bool irq) |
64db4cff | 821 | { |
c9d4b0af | 822 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
84585aa8 | 823 | long incby = 2; |
64db4cff | 824 | |
734d1680 PM |
825 | /* Complain about underflow. */ |
826 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting < 0); | |
827 | ||
828 | /* | |
829 | * If idle from RCU viewpoint, atomically increment ->dynticks | |
830 | * to mark non-idle and increment ->dynticks_nmi_nesting by one. | |
831 | * Otherwise, increment ->dynticks_nmi_nesting by two. This means | |
832 | * if ->dynticks_nmi_nesting is equal to one, we are guaranteed | |
833 | * to be in the outermost NMI handler that interrupted an RCU-idle | |
834 | * period (observation due to Andy Lutomirski). | |
835 | */ | |
02a5c550 | 836 | if (rcu_dynticks_curr_cpu_in_eqs()) { |
cf7614e1 BP |
837 | |
838 | if (irq) | |
839 | rcu_dynticks_task_exit(); | |
840 | ||
2625d469 | 841 | rcu_dynticks_eqs_exit(); |
cf7614e1 BP |
842 | |
843 | if (irq) | |
844 | rcu_cleanup_after_idle(); | |
845 | ||
734d1680 PM |
846 | incby = 1; |
847 | } | |
bd2b879a PM |
848 | trace_rcu_dyntick(incby == 1 ? TPS("Endirq") : TPS("++="), |
849 | rdtp->dynticks_nmi_nesting, | |
dec98900 | 850 | rdtp->dynticks_nmi_nesting + incby, rdtp->dynticks); |
fd581a91 PM |
851 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* Prevent store tearing. */ |
852 | rdtp->dynticks_nmi_nesting + incby); | |
734d1680 | 853 | barrier(); |
64db4cff PM |
854 | } |
855 | ||
cf7614e1 BP |
856 | /** |
857 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
858 | */ | |
859 | void rcu_nmi_enter(void) | |
860 | { | |
861 | rcu_nmi_enter_common(false); | |
862 | } | |
863 | ||
64db4cff | 864 | /** |
9b2e4f18 | 865 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle |
64db4cff | 866 | * |
9b2e4f18 PM |
867 | * Enter an interrupt handler, which might possibly result in exiting |
868 | * idle mode, in other words, entering the mode in which read-side critical | |
7c9906ca | 869 | * sections can occur. The caller must have disabled interrupts. |
c0da313e | 870 | * |
9b2e4f18 | 871 | * Note that the Linux kernel is fully capable of entering an interrupt |
58721f5d PM |
872 | * handler that it never exits, for example when doing upcalls to user mode! |
873 | * This code assumes that the idle loop never does upcalls to user mode. | |
874 | * If your architecture's idle loop does do upcalls to user mode (or does | |
875 | * anything else that results in unbalanced calls to the irq_enter() and | |
876 | * irq_exit() functions), RCU will give you what you deserve, good and hard. | |
877 | * But very infrequently and irreproducibly. | |
9b2e4f18 PM |
878 | * |
879 | * Use things like work queues to work around this limitation. | |
880 | * | |
881 | * You have been warned. | |
c0da313e PM |
882 | * |
883 | * If you add or remove a call to rcu_irq_enter(), be sure to test with | |
884 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 885 | */ |
9b2e4f18 | 886 | void rcu_irq_enter(void) |
64db4cff | 887 | { |
b04db8e1 | 888 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 889 | rcu_nmi_enter_common(true); |
7c9906ca | 890 | } |
734d1680 | 891 | |
7c9906ca PM |
892 | /* |
893 | * Wrapper for rcu_irq_enter() where interrupts are enabled. | |
c0da313e PM |
894 | * |
895 | * If you add or remove a call to rcu_irq_enter_irqson(), be sure to test | |
896 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
897 | */ |
898 | void rcu_irq_enter_irqson(void) | |
899 | { | |
900 | unsigned long flags; | |
734d1680 | 901 | |
7c9906ca PM |
902 | local_irq_save(flags); |
903 | rcu_irq_enter(); | |
64db4cff | 904 | local_irq_restore(flags); |
64db4cff PM |
905 | } |
906 | ||
5c173eb8 PM |
907 | /** |
908 | * rcu_is_watching - see if RCU thinks that the current CPU is idle | |
64db4cff | 909 | * |
791875d1 PM |
910 | * Return true if RCU is watching the running CPU, which means that this |
911 | * CPU can safely enter RCU read-side critical sections. In other words, | |
912 | * if the current CPU is in its idle loop and is neither in an interrupt | |
34240697 | 913 | * or NMI handler, return true. |
64db4cff | 914 | */ |
9418fb20 | 915 | bool notrace rcu_is_watching(void) |
64db4cff | 916 | { |
f534ed1f | 917 | bool ret; |
34240697 | 918 | |
46f00d18 | 919 | preempt_disable_notrace(); |
791875d1 | 920 | ret = !rcu_dynticks_curr_cpu_in_eqs(); |
46f00d18 | 921 | preempt_enable_notrace(); |
34240697 | 922 | return ret; |
64db4cff | 923 | } |
5c173eb8 | 924 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 925 | |
bcbfdd01 PM |
926 | /* |
927 | * If a holdout task is actually running, request an urgent quiescent | |
928 | * state from its CPU. This is unsynchronized, so migrations can cause | |
929 | * the request to go to the wrong CPU. Which is OK, all that will happen | |
930 | * is that the CPU's next context switch will be a bit slower and next | |
931 | * time around this task will generate another request. | |
932 | */ | |
933 | void rcu_request_urgent_qs_task(struct task_struct *t) | |
934 | { | |
935 | int cpu; | |
936 | ||
937 | barrier(); | |
938 | cpu = task_cpu(t); | |
939 | if (!task_curr(t)) | |
940 | return; /* This task is not running on that CPU. */ | |
941 | smp_store_release(per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, cpu), true); | |
942 | } | |
943 | ||
62fde6ed | 944 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
945 | |
946 | /* | |
5554788e | 947 | * Is the current CPU online as far as RCU is concerned? |
2036d94a | 948 | * |
5554788e PM |
949 | * Disable preemption to avoid false positives that could otherwise |
950 | * happen due to the current CPU number being sampled, this task being | |
951 | * preempted, its old CPU being taken offline, resuming on some other CPU, | |
952 | * then determining that its old CPU is now offline. Because there are | |
953 | * multiple flavors of RCU, and because this function can be called in the | |
954 | * midst of updating the flavors while a given CPU coming online or going | |
955 | * offline, it is necessary to check all flavors. If any of the flavors | |
956 | * believe that given CPU is online, it is considered to be online. | |
c0d6d01b | 957 | * |
5554788e PM |
958 | * Disable checking if in an NMI handler because we cannot safely |
959 | * report errors from NMI handlers anyway. In addition, it is OK to use | |
960 | * RCU on an offline processor during initial boot, hence the check for | |
961 | * rcu_scheduler_fully_active. | |
c0d6d01b PM |
962 | */ |
963 | bool rcu_lockdep_current_cpu_online(void) | |
964 | { | |
2036d94a PM |
965 | struct rcu_data *rdp; |
966 | struct rcu_node *rnp; | |
b97d23c5 | 967 | bool ret = false; |
c0d6d01b | 968 | |
5554788e | 969 | if (in_nmi() || !rcu_scheduler_fully_active) |
f6f7ee9a | 970 | return true; |
c0d6d01b | 971 | preempt_disable(); |
b97d23c5 PM |
972 | rdp = this_cpu_ptr(&rcu_data); |
973 | rnp = rdp->mynode; | |
974 | if (rdp->grpmask & rcu_rnp_online_cpus(rnp)) | |
975 | ret = true; | |
c0d6d01b | 976 | preempt_enable(); |
b97d23c5 | 977 | return ret; |
c0d6d01b PM |
978 | } |
979 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
980 | ||
62fde6ed | 981 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 982 | |
9b9500da PM |
983 | /* |
984 | * We are reporting a quiescent state on behalf of some other CPU, so | |
985 | * it is our responsibility to check for and handle potential overflow | |
a66ae8ae | 986 | * of the rcu_node ->gp_seq counter with respect to the rcu_data counters. |
9b9500da PM |
987 | * After all, the CPU might be in deep idle state, and thus executing no |
988 | * code whatsoever. | |
989 | */ | |
990 | static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp) | |
991 | { | |
a32e01ee | 992 | raw_lockdep_assert_held_rcu_node(rnp); |
a66ae8ae PM |
993 | if (ULONG_CMP_LT(rcu_seq_current(&rdp->gp_seq) + ULONG_MAX / 4, |
994 | rnp->gp_seq)) | |
9b9500da | 995 | WRITE_ONCE(rdp->gpwrap, true); |
8aa670cd PM |
996 | if (ULONG_CMP_LT(rdp->rcu_iw_gp_seq + ULONG_MAX / 4, rnp->gp_seq)) |
997 | rdp->rcu_iw_gp_seq = rnp->gp_seq + ULONG_MAX / 4; | |
9b9500da PM |
998 | } |
999 | ||
64db4cff PM |
1000 | /* |
1001 | * Snapshot the specified CPU's dynticks counter so that we can later | |
1002 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 1003 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 1004 | */ |
fe5ac724 | 1005 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
64db4cff | 1006 | { |
8b2f63ab | 1007 | rdp->dynticks_snap = rcu_dynticks_snap(rdp->dynticks); |
02a5c550 | 1008 | if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) { |
88d1bead | 1009 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 1010 | rcu_gpnum_ovf(rdp->mynode, rdp); |
23a9bacd | 1011 | return 1; |
7941dbde | 1012 | } |
23a9bacd | 1013 | return 0; |
64db4cff PM |
1014 | } |
1015 | ||
9b9500da PM |
1016 | /* |
1017 | * Handler for the irq_work request posted when a grace period has | |
1018 | * gone on for too long, but not yet long enough for an RCU CPU | |
1019 | * stall warning. Set state appropriately, but just complain if | |
1020 | * there is unexpected state on entry. | |
1021 | */ | |
1022 | static void rcu_iw_handler(struct irq_work *iwp) | |
1023 | { | |
1024 | struct rcu_data *rdp; | |
1025 | struct rcu_node *rnp; | |
1026 | ||
1027 | rdp = container_of(iwp, struct rcu_data, rcu_iw); | |
1028 | rnp = rdp->mynode; | |
1029 | raw_spin_lock_rcu_node(rnp); | |
1030 | if (!WARN_ON_ONCE(!rdp->rcu_iw_pending)) { | |
8aa670cd | 1031 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1032 | rdp->rcu_iw_pending = false; |
1033 | } | |
1034 | raw_spin_unlock_rcu_node(rnp); | |
1035 | } | |
1036 | ||
64db4cff PM |
1037 | /* |
1038 | * Return true if the specified CPU has passed through a quiescent | |
1039 | * state by virtue of being in or having passed through an dynticks | |
1040 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 1041 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 1042 | */ |
fe5ac724 | 1043 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
64db4cff | 1044 | { |
3a19b46a | 1045 | unsigned long jtsq; |
0f9be8ca | 1046 | bool *rnhqp; |
9226b10d | 1047 | bool *ruqp; |
9b9500da | 1048 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
1049 | |
1050 | /* | |
1051 | * If the CPU passed through or entered a dynticks idle phase with | |
1052 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
1053 | * already acknowledged the request to pass through a quiescent | |
1054 | * state. Either way, that CPU cannot possibly be in an RCU | |
1055 | * read-side critical section that started before the beginning | |
1056 | * of the current RCU grace period. | |
1057 | */ | |
02a5c550 | 1058 | if (rcu_dynticks_in_eqs_since(rdp->dynticks, rdp->dynticks_snap)) { |
88d1bead | 1059 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
64db4cff | 1060 | rdp->dynticks_fqs++; |
9b9500da | 1061 | rcu_gpnum_ovf(rnp, rdp); |
64db4cff PM |
1062 | return 1; |
1063 | } | |
1064 | ||
a82dcc76 | 1065 | /* |
cee43939 PM |
1066 | * Has this CPU encountered a cond_resched() since the beginning |
1067 | * of the grace period? For this to be the case, the CPU has to | |
1068 | * have noticed the current grace period. This might not be the | |
1069 | * case for nohz_full CPUs looping in the kernel. | |
a82dcc76 | 1070 | */ |
f79c3ad6 | 1071 | jtsq = jiffies_till_sched_qs; |
9226b10d | 1072 | ruqp = per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, rdp->cpu); |
88d1bead | 1073 | if (time_after(jiffies, rcu_state.gp_start + jtsq) && |
9577df9a | 1074 | READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) && |
e05720b0 | 1075 | rcu_seq_current(&rdp->gp_seq) == rnp->gp_seq && !rdp->gpwrap) { |
88d1bead | 1076 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("rqc")); |
9b9500da | 1077 | rcu_gpnum_ovf(rnp, rdp); |
3a19b46a | 1078 | return 1; |
88d1bead | 1079 | } else if (time_after(jiffies, rcu_state.gp_start + jtsq)) { |
9226b10d PM |
1080 | /* Load rcu_qs_ctr before store to rcu_urgent_qs. */ |
1081 | smp_store_release(ruqp, true); | |
3a19b46a PM |
1082 | } |
1083 | ||
f2e2df59 PM |
1084 | /* If waiting too long on an offline CPU, complain. */ |
1085 | if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp)) && | |
88d1bead | 1086 | time_after(jiffies, rcu_state.gp_start + HZ)) { |
f2e2df59 PM |
1087 | bool onl; |
1088 | struct rcu_node *rnp1; | |
1089 | ||
1090 | WARN_ON(1); /* Offline CPUs are supposed to report QS! */ | |
1091 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", | |
1092 | __func__, rnp->grplo, rnp->grphi, rnp->level, | |
1093 | (long)rnp->gp_seq, (long)rnp->completedqs); | |
1094 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
1095 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx ->rcu_gp_init_mask %#lx\n", | |
1096 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext, rnp1->rcu_gp_init_mask); | |
1097 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); | |
1098 | pr_info("%s %d: %c online: %ld(%d) offline: %ld(%d)\n", | |
1099 | __func__, rdp->cpu, ".o"[onl], | |
1100 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
1101 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
1102 | return 1; /* Break things loose after complaining. */ | |
1103 | } | |
1104 | ||
65d798f0 | 1105 | /* |
4a81e832 PM |
1106 | * A CPU running for an extended time within the kernel can |
1107 | * delay RCU grace periods. When the CPU is in NO_HZ_FULL mode, | |
1108 | * even context-switching back and forth between a pair of | |
1109 | * in-kernel CPU-bound tasks cannot advance grace periods. | |
1110 | * So if the grace period is old enough, make the CPU pay attention. | |
1111 | * Note that the unsynchronized assignments to the per-CPU | |
0f9be8ca | 1112 | * rcu_need_heavy_qs variable are safe. Yes, setting of |
4a81e832 PM |
1113 | * bits can be lost, but they will be set again on the next |
1114 | * force-quiescent-state pass. So lost bit sets do not result | |
1115 | * in incorrect behavior, merely in a grace period lasting | |
1116 | * a few jiffies longer than it might otherwise. Because | |
1117 | * there are at most four threads involved, and because the | |
1118 | * updates are only once every few jiffies, the probability of | |
1119 | * lossage (and thus of slight grace-period extension) is | |
1120 | * quite low. | |
6193c76a | 1121 | */ |
0f9be8ca PM |
1122 | rnhqp = &per_cpu(rcu_dynticks.rcu_need_heavy_qs, rdp->cpu); |
1123 | if (!READ_ONCE(*rnhqp) && | |
88d1bead PM |
1124 | (time_after(jiffies, rcu_state.gp_start + jtsq) || |
1125 | time_after(jiffies, rcu_state.jiffies_resched))) { | |
0f9be8ca | 1126 | WRITE_ONCE(*rnhqp, true); |
9226b10d PM |
1127 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ |
1128 | smp_store_release(ruqp, true); | |
88d1bead | 1129 | rcu_state.jiffies_resched += jtsq; /* Re-enable beating. */ |
6193c76a PM |
1130 | } |
1131 | ||
28053bc7 | 1132 | /* |
9b9500da PM |
1133 | * If more than halfway to RCU CPU stall-warning time, do a |
1134 | * resched_cpu() to try to loosen things up a bit. Also check to | |
1135 | * see if the CPU is getting hammered with interrupts, but only | |
1136 | * once per grace period, just to keep the IPIs down to a dull roar. | |
28053bc7 | 1137 | */ |
88d1bead | 1138 | if (jiffies - rcu_state.gp_start > rcu_jiffies_till_stall_check() / 2) { |
28053bc7 | 1139 | resched_cpu(rdp->cpu); |
9b9500da | 1140 | if (IS_ENABLED(CONFIG_IRQ_WORK) && |
8aa670cd | 1141 | !rdp->rcu_iw_pending && rdp->rcu_iw_gp_seq != rnp->gp_seq && |
9b9500da PM |
1142 | (rnp->ffmask & rdp->grpmask)) { |
1143 | init_irq_work(&rdp->rcu_iw, rcu_iw_handler); | |
1144 | rdp->rcu_iw_pending = true; | |
8aa670cd | 1145 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1146 | irq_work_queue_on(&rdp->rcu_iw, rdp->cpu); |
1147 | } | |
1148 | } | |
4914950a | 1149 | |
a82dcc76 | 1150 | return 0; |
64db4cff PM |
1151 | } |
1152 | ||
ad3832e9 | 1153 | static void record_gp_stall_check_time(void) |
64db4cff | 1154 | { |
cb1e78cf | 1155 | unsigned long j = jiffies; |
6193c76a | 1156 | unsigned long j1; |
26cdfedf | 1157 | |
ad3832e9 | 1158 | rcu_state.gp_start = j; |
6193c76a | 1159 | j1 = rcu_jiffies_till_stall_check(); |
91f63ced | 1160 | /* Record ->gp_start before ->jiffies_stall. */ |
ad3832e9 PM |
1161 | smp_store_release(&rcu_state.jiffies_stall, j + j1); /* ^^^ */ |
1162 | rcu_state.jiffies_resched = j + j1 / 2; | |
1163 | rcu_state.n_force_qs_gpstart = READ_ONCE(rcu_state.n_force_qs); | |
64db4cff PM |
1164 | } |
1165 | ||
6b50e119 PM |
1166 | /* |
1167 | * Convert a ->gp_state value to a character string. | |
1168 | */ | |
1169 | static const char *gp_state_getname(short gs) | |
1170 | { | |
1171 | if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names)) | |
1172 | return "???"; | |
1173 | return gp_state_names[gs]; | |
1174 | } | |
1175 | ||
fb81a44b PM |
1176 | /* |
1177 | * Complain about starvation of grace-period kthread. | |
1178 | */ | |
8fd119b6 | 1179 | static void rcu_check_gp_kthread_starvation(void) |
fb81a44b | 1180 | { |
7cba4775 | 1181 | struct task_struct *gpk = rcu_state.gp_kthread; |
fb81a44b PM |
1182 | unsigned long j; |
1183 | ||
7cba4775 PM |
1184 | j = jiffies - READ_ONCE(rcu_state.gp_activity); |
1185 | if (j > 2 * HZ) { | |
78c5a67f | 1186 | pr_err("%s kthread starved for %ld jiffies! g%ld f%#x %s(%d) ->state=%#lx ->cpu=%d\n", |
7cba4775 PM |
1187 | rcu_state.name, j, |
1188 | (long)rcu_seq_current(&rcu_state.gp_seq), | |
1189 | rcu_state.gp_flags, | |
1190 | gp_state_getname(rcu_state.gp_state), rcu_state.gp_state, | |
1191 | gpk ? gpk->state : ~0, gpk ? task_cpu(gpk) : -1); | |
1192 | if (gpk) { | |
d07aee2c | 1193 | pr_err("RCU grace-period kthread stack dump:\n"); |
7cba4775 PM |
1194 | sched_show_task(gpk); |
1195 | wake_up_process(gpk); | |
86057b80 | 1196 | } |
b1adb3e2 | 1197 | } |
64db4cff PM |
1198 | } |
1199 | ||
b637a328 | 1200 | /* |
7aa92230 PM |
1201 | * Dump stacks of all tasks running on stalled CPUs. First try using |
1202 | * NMIs, but fall back to manual remote stack tracing on architectures | |
1203 | * that don't support NMI-based stack dumps. The NMI-triggered stack | |
1204 | * traces are more accurate because they are printed by the target CPU. | |
b637a328 | 1205 | */ |
33dbdbf0 | 1206 | static void rcu_dump_cpu_stacks(void) |
b637a328 PM |
1207 | { |
1208 | int cpu; | |
1209 | unsigned long flags; | |
1210 | struct rcu_node *rnp; | |
1211 | ||
aedf4ba9 | 1212 | rcu_for_each_leaf_node(rnp) { |
6cf10081 | 1213 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7aa92230 PM |
1214 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1215 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) | |
1216 | if (!trigger_single_cpu_backtrace(cpu)) | |
bc75e999 | 1217 | dump_cpu_task(cpu); |
67c583a7 | 1218 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b637a328 PM |
1219 | } |
1220 | } | |
1221 | ||
8c7c4829 PM |
1222 | /* |
1223 | * If too much time has passed in the current grace period, and if | |
1224 | * so configured, go kick the relevant kthreads. | |
1225 | */ | |
e1741c69 | 1226 | static void rcu_stall_kick_kthreads(void) |
8c7c4829 PM |
1227 | { |
1228 | unsigned long j; | |
1229 | ||
1230 | if (!rcu_kick_kthreads) | |
1231 | return; | |
4c6ed437 PM |
1232 | j = READ_ONCE(rcu_state.jiffies_kick_kthreads); |
1233 | if (time_after(jiffies, j) && rcu_state.gp_kthread && | |
1234 | (rcu_gp_in_progress() || READ_ONCE(rcu_state.gp_flags))) { | |
1235 | WARN_ONCE(1, "Kicking %s grace-period kthread\n", | |
1236 | rcu_state.name); | |
5dffed1e | 1237 | rcu_ftrace_dump(DUMP_ALL); |
4c6ed437 PM |
1238 | wake_up_process(rcu_state.gp_kthread); |
1239 | WRITE_ONCE(rcu_state.jiffies_kick_kthreads, j + HZ); | |
8c7c4829 PM |
1240 | } |
1241 | } | |
1242 | ||
95394e69 | 1243 | static void panic_on_rcu_stall(void) |
088e9d25 DBO |
1244 | { |
1245 | if (sysctl_panic_on_rcu_stall) | |
1246 | panic("RCU Stall\n"); | |
1247 | } | |
1248 | ||
a91e7e58 | 1249 | static void print_other_cpu_stall(unsigned long gp_seq) |
64db4cff PM |
1250 | { |
1251 | int cpu; | |
64db4cff | 1252 | unsigned long flags; |
6ccd2ecd PM |
1253 | unsigned long gpa; |
1254 | unsigned long j; | |
285fe294 | 1255 | int ndetected = 0; |
336a4f6c | 1256 | struct rcu_node *rnp = rcu_get_root(); |
53bb857c | 1257 | long totqlen = 0; |
64db4cff | 1258 | |
8c7c4829 | 1259 | /* Kick and suppress, if so configured. */ |
e1741c69 | 1260 | rcu_stall_kick_kthreads(); |
8c7c4829 PM |
1261 | if (rcu_cpu_stall_suppress) |
1262 | return; | |
1263 | ||
8cdd32a9 PM |
1264 | /* |
1265 | * OK, time to rat on our buddy... | |
1266 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1267 | * RCU CPU stall warnings. | |
1268 | */ | |
4c6ed437 | 1269 | pr_err("INFO: %s detected stalls on CPUs/tasks:", rcu_state.name); |
a858af28 | 1270 | print_cpu_stall_info_begin(); |
aedf4ba9 | 1271 | rcu_for_each_leaf_node(rnp) { |
6cf10081 | 1272 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
9bc8b558 | 1273 | ndetected += rcu_print_task_stall(rnp); |
c8020a67 | 1274 | if (rnp->qsmask != 0) { |
bc75e999 MR |
1275 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1276 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) { | |
b21ebed9 | 1277 | print_cpu_stall_info(cpu); |
c8020a67 PM |
1278 | ndetected++; |
1279 | } | |
1280 | } | |
67c583a7 | 1281 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1282 | } |
a858af28 | 1283 | |
a858af28 | 1284 | print_cpu_stall_info_end(); |
53bb857c | 1285 | for_each_possible_cpu(cpu) |
da1df50d | 1286 | totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(&rcu_data, |
15fecf89 | 1287 | cpu)->cblist); |
471f87c3 | 1288 | pr_cont("(detected by %d, t=%ld jiffies, g=%ld, q=%lu)\n", |
4c6ed437 PM |
1289 | smp_processor_id(), (long)(jiffies - rcu_state.gp_start), |
1290 | (long)rcu_seq_current(&rcu_state.gp_seq), totqlen); | |
6ccd2ecd | 1291 | if (ndetected) { |
33dbdbf0 | 1292 | rcu_dump_cpu_stacks(); |
c4402b27 BP |
1293 | |
1294 | /* Complain about tasks blocking the grace period. */ | |
a2887cd8 | 1295 | rcu_print_detail_task_stall(); |
6ccd2ecd | 1296 | } else { |
4c6ed437 | 1297 | if (rcu_seq_current(&rcu_state.gp_seq) != gp_seq) { |
6ccd2ecd PM |
1298 | pr_err("INFO: Stall ended before state dump start\n"); |
1299 | } else { | |
1300 | j = jiffies; | |
4c6ed437 | 1301 | gpa = READ_ONCE(rcu_state.gp_activity); |
237a0f21 | 1302 | pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n", |
4c6ed437 | 1303 | rcu_state.name, j - gpa, j, gpa, |
237a0f21 | 1304 | jiffies_till_next_fqs, |
336a4f6c | 1305 | rcu_get_root()->qsmask); |
6ccd2ecd PM |
1306 | /* In this case, the current CPU might be at fault. */ |
1307 | sched_show_task(current); | |
1308 | } | |
1309 | } | |
8c42b1f3 | 1310 | /* Rewrite if needed in case of slow consoles. */ |
4c6ed437 PM |
1311 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall))) |
1312 | WRITE_ONCE(rcu_state.jiffies_stall, | |
8c42b1f3 | 1313 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); |
c1dc0b9c | 1314 | |
8fd119b6 | 1315 | rcu_check_gp_kthread_starvation(); |
fb81a44b | 1316 | |
088e9d25 DBO |
1317 | panic_on_rcu_stall(); |
1318 | ||
e9ecb780 | 1319 | force_quiescent_state(); /* Kick them all. */ |
64db4cff PM |
1320 | } |
1321 | ||
4e8b8e08 | 1322 | static void print_cpu_stall(void) |
64db4cff | 1323 | { |
53bb857c | 1324 | int cpu; |
64db4cff | 1325 | unsigned long flags; |
da1df50d | 1326 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
336a4f6c | 1327 | struct rcu_node *rnp = rcu_get_root(); |
53bb857c | 1328 | long totqlen = 0; |
64db4cff | 1329 | |
8c7c4829 | 1330 | /* Kick and suppress, if so configured. */ |
e1741c69 | 1331 | rcu_stall_kick_kthreads(); |
8c7c4829 PM |
1332 | if (rcu_cpu_stall_suppress) |
1333 | return; | |
1334 | ||
8cdd32a9 PM |
1335 | /* |
1336 | * OK, time to rat on ourselves... | |
1337 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1338 | * RCU CPU stall warnings. | |
1339 | */ | |
4c6ed437 | 1340 | pr_err("INFO: %s self-detected stall on CPU", rcu_state.name); |
a858af28 | 1341 | print_cpu_stall_info_begin(); |
9b9500da | 1342 | raw_spin_lock_irqsave_rcu_node(rdp->mynode, flags); |
b21ebed9 | 1343 | print_cpu_stall_info(smp_processor_id()); |
9b9500da | 1344 | raw_spin_unlock_irqrestore_rcu_node(rdp->mynode, flags); |
a858af28 | 1345 | print_cpu_stall_info_end(); |
53bb857c | 1346 | for_each_possible_cpu(cpu) |
da1df50d | 1347 | totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(&rcu_data, |
15fecf89 | 1348 | cpu)->cblist); |
471f87c3 | 1349 | pr_cont(" (t=%lu jiffies g=%ld q=%lu)\n", |
4c6ed437 PM |
1350 | jiffies - rcu_state.gp_start, |
1351 | (long)rcu_seq_current(&rcu_state.gp_seq), totqlen); | |
fb81a44b | 1352 | |
8fd119b6 | 1353 | rcu_check_gp_kthread_starvation(); |
fb81a44b | 1354 | |
33dbdbf0 | 1355 | rcu_dump_cpu_stacks(); |
c1dc0b9c | 1356 | |
6cf10081 | 1357 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
8c42b1f3 | 1358 | /* Rewrite if needed in case of slow consoles. */ |
4c6ed437 PM |
1359 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rcu_state.jiffies_stall))) |
1360 | WRITE_ONCE(rcu_state.jiffies_stall, | |
7d0ae808 | 1361 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); |
67c583a7 | 1362 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c1dc0b9c | 1363 | |
088e9d25 DBO |
1364 | panic_on_rcu_stall(); |
1365 | ||
b021fe3e PZ |
1366 | /* |
1367 | * Attempt to revive the RCU machinery by forcing a context switch. | |
1368 | * | |
1369 | * A context switch would normally allow the RCU state machine to make | |
1370 | * progress and it could be we're stuck in kernel space without context | |
1371 | * switches for an entirely unreasonable amount of time. | |
1372 | */ | |
1373 | resched_cpu(smp_processor_id()); | |
64db4cff PM |
1374 | } |
1375 | ||
ea12ff2b | 1376 | static void check_cpu_stall(struct rcu_data *rdp) |
64db4cff | 1377 | { |
471f87c3 PM |
1378 | unsigned long gs1; |
1379 | unsigned long gs2; | |
26cdfedf | 1380 | unsigned long gps; |
bad6e139 | 1381 | unsigned long j; |
8c42b1f3 | 1382 | unsigned long jn; |
bad6e139 | 1383 | unsigned long js; |
64db4cff PM |
1384 | struct rcu_node *rnp; |
1385 | ||
8c7c4829 | 1386 | if ((rcu_cpu_stall_suppress && !rcu_kick_kthreads) || |
de8e8730 | 1387 | !rcu_gp_in_progress()) |
c68de209 | 1388 | return; |
e1741c69 | 1389 | rcu_stall_kick_kthreads(); |
cb1e78cf | 1390 | j = jiffies; |
26cdfedf PM |
1391 | |
1392 | /* | |
1393 | * Lots of memory barriers to reject false positives. | |
1394 | * | |
4c6ed437 PM |
1395 | * The idea is to pick up rcu_state.gp_seq, then |
1396 | * rcu_state.jiffies_stall, then rcu_state.gp_start, and finally | |
1397 | * another copy of rcu_state.gp_seq. These values are updated in | |
1398 | * the opposite order with memory barriers (or equivalent) during | |
1399 | * grace-period initialization and cleanup. Now, a false positive | |
1400 | * can occur if we get an new value of rcu_state.gp_start and a old | |
1401 | * value of rcu_state.jiffies_stall. But given the memory barriers, | |
1402 | * the only way that this can happen is if one grace period ends | |
1403 | * and another starts between these two fetches. This is detected | |
1404 | * by comparing the second fetch of rcu_state.gp_seq with the | |
1405 | * previous fetch from rcu_state.gp_seq. | |
26cdfedf | 1406 | * |
4c6ed437 PM |
1407 | * Given this check, comparisons of jiffies, rcu_state.jiffies_stall, |
1408 | * and rcu_state.gp_start suffice to forestall false positives. | |
26cdfedf | 1409 | */ |
4c6ed437 | 1410 | gs1 = READ_ONCE(rcu_state.gp_seq); |
471f87c3 | 1411 | smp_rmb(); /* Pick up ->gp_seq first... */ |
4c6ed437 | 1412 | js = READ_ONCE(rcu_state.jiffies_stall); |
26cdfedf | 1413 | smp_rmb(); /* ...then ->jiffies_stall before the rest... */ |
4c6ed437 | 1414 | gps = READ_ONCE(rcu_state.gp_start); |
471f87c3 | 1415 | smp_rmb(); /* ...and finally ->gp_start before ->gp_seq again. */ |
4c6ed437 | 1416 | gs2 = READ_ONCE(rcu_state.gp_seq); |
471f87c3 | 1417 | if (gs1 != gs2 || |
26cdfedf PM |
1418 | ULONG_CMP_LT(j, js) || |
1419 | ULONG_CMP_GE(gps, js)) | |
1420 | return; /* No stall or GP completed since entering function. */ | |
64db4cff | 1421 | rnp = rdp->mynode; |
8c42b1f3 | 1422 | jn = jiffies + 3 * rcu_jiffies_till_stall_check() + 3; |
de8e8730 | 1423 | if (rcu_gp_in_progress() && |
8c42b1f3 | 1424 | (READ_ONCE(rnp->qsmask) & rdp->grpmask) && |
4c6ed437 | 1425 | cmpxchg(&rcu_state.jiffies_stall, js, jn) == js) { |
64db4cff PM |
1426 | |
1427 | /* We haven't checked in, so go dump stack. */ | |
4e8b8e08 | 1428 | print_cpu_stall(); |
64db4cff | 1429 | |
de8e8730 | 1430 | } else if (rcu_gp_in_progress() && |
8c42b1f3 | 1431 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY) && |
4c6ed437 | 1432 | cmpxchg(&rcu_state.jiffies_stall, js, jn) == js) { |
64db4cff | 1433 | |
bad6e139 | 1434 | /* They had a few time units to dump stack, so complain. */ |
a91e7e58 | 1435 | print_other_cpu_stall(gs2); |
64db4cff PM |
1436 | } |
1437 | } | |
1438 | ||
53d84e00 PM |
1439 | /** |
1440 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
1441 | * | |
1442 | * Set the stall-warning timeout way off into the future, thus preventing | |
1443 | * any RCU CPU stall-warning messages from appearing in the current set of | |
1444 | * RCU grace periods. | |
1445 | * | |
1446 | * The caller must disable hard irqs. | |
1447 | */ | |
1448 | void rcu_cpu_stall_reset(void) | |
1449 | { | |
b97d23c5 | 1450 | WRITE_ONCE(rcu_state.jiffies_stall, jiffies + ULONG_MAX / 2); |
53d84e00 PM |
1451 | } |
1452 | ||
41e80595 PM |
1453 | /* Trace-event wrapper function for trace_rcu_future_grace_period. */ |
1454 | static void trace_rcu_this_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
b73de91d | 1455 | unsigned long gp_seq_req, const char *s) |
0446be48 | 1456 | { |
88d1bead | 1457 | trace_rcu_future_grace_period(rcu_state.name, rnp->gp_seq, gp_seq_req, |
abd13fdd | 1458 | rnp->level, rnp->grplo, rnp->grphi, s); |
0446be48 PM |
1459 | } |
1460 | ||
1461 | /* | |
b73de91d | 1462 | * rcu_start_this_gp - Request the start of a particular grace period |
df2bf8f7 | 1463 | * @rnp_start: The leaf node of the CPU from which to start. |
b73de91d JF |
1464 | * @rdp: The rcu_data corresponding to the CPU from which to start. |
1465 | * @gp_seq_req: The gp_seq of the grace period to start. | |
1466 | * | |
41e80595 | 1467 | * Start the specified grace period, as needed to handle newly arrived |
0446be48 | 1468 | * callbacks. The required future grace periods are recorded in each |
7a1d0f23 | 1469 | * rcu_node structure's ->gp_seq_needed field. Returns true if there |
48a7639c | 1470 | * is reason to awaken the grace-period kthread. |
0446be48 | 1471 | * |
d5cd9685 PM |
1472 | * The caller must hold the specified rcu_node structure's ->lock, which |
1473 | * is why the caller is responsible for waking the grace-period kthread. | |
b73de91d JF |
1474 | * |
1475 | * Returns true if the GP thread needs to be awakened else false. | |
0446be48 | 1476 | */ |
df2bf8f7 | 1477 | static bool rcu_start_this_gp(struct rcu_node *rnp_start, struct rcu_data *rdp, |
b73de91d | 1478 | unsigned long gp_seq_req) |
0446be48 | 1479 | { |
48a7639c | 1480 | bool ret = false; |
df2bf8f7 | 1481 | struct rcu_node *rnp; |
0446be48 PM |
1482 | |
1483 | /* | |
360e0da6 PM |
1484 | * Use funnel locking to either acquire the root rcu_node |
1485 | * structure's lock or bail out if the need for this grace period | |
df2bf8f7 JFG |
1486 | * has already been recorded -- or if that grace period has in |
1487 | * fact already started. If there is already a grace period in | |
1488 | * progress in a non-leaf node, no recording is needed because the | |
1489 | * end of the grace period will scan the leaf rcu_node structures. | |
1490 | * Note that rnp_start->lock must not be released. | |
0446be48 | 1491 | */ |
df2bf8f7 JFG |
1492 | raw_lockdep_assert_held_rcu_node(rnp_start); |
1493 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, TPS("Startleaf")); | |
1494 | for (rnp = rnp_start; 1; rnp = rnp->parent) { | |
1495 | if (rnp != rnp_start) | |
1496 | raw_spin_lock_rcu_node(rnp); | |
1497 | if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req) || | |
1498 | rcu_seq_started(&rnp->gp_seq, gp_seq_req) || | |
1499 | (rnp != rnp_start && | |
1500 | rcu_seq_state(rcu_seq_current(&rnp->gp_seq)))) { | |
1501 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, | |
b73de91d | 1502 | TPS("Prestarted")); |
360e0da6 PM |
1503 | goto unlock_out; |
1504 | } | |
df2bf8f7 | 1505 | rnp->gp_seq_needed = gp_seq_req; |
226ca5e7 | 1506 | if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) { |
a2165e41 | 1507 | /* |
226ca5e7 JFG |
1508 | * We just marked the leaf or internal node, and a |
1509 | * grace period is in progress, which means that | |
1510 | * rcu_gp_cleanup() will see the marking. Bail to | |
1511 | * reduce contention. | |
a2165e41 | 1512 | */ |
df2bf8f7 | 1513 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, |
b73de91d | 1514 | TPS("Startedleaf")); |
a2165e41 PM |
1515 | goto unlock_out; |
1516 | } | |
df2bf8f7 JFG |
1517 | if (rnp != rnp_start && rnp->parent != NULL) |
1518 | raw_spin_unlock_rcu_node(rnp); | |
1519 | if (!rnp->parent) | |
360e0da6 | 1520 | break; /* At root, and perhaps also leaf. */ |
0446be48 PM |
1521 | } |
1522 | ||
360e0da6 | 1523 | /* If GP already in progress, just leave, otherwise start one. */ |
de8e8730 | 1524 | if (rcu_gp_in_progress()) { |
df2bf8f7 | 1525 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedleafroot")); |
0446be48 PM |
1526 | goto unlock_out; |
1527 | } | |
df2bf8f7 | 1528 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedroot")); |
9cbc5b97 PM |
1529 | WRITE_ONCE(rcu_state.gp_flags, rcu_state.gp_flags | RCU_GP_FLAG_INIT); |
1530 | rcu_state.gp_req_activity = jiffies; | |
1531 | if (!rcu_state.gp_kthread) { | |
df2bf8f7 | 1532 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("NoGPkthread")); |
360e0da6 | 1533 | goto unlock_out; |
0446be48 | 1534 | } |
9cbc5b97 | 1535 | trace_rcu_grace_period(rcu_state.name, READ_ONCE(rcu_state.gp_seq), TPS("newreq")); |
360e0da6 | 1536 | ret = true; /* Caller must wake GP kthread. */ |
0446be48 | 1537 | unlock_out: |
ab5e869c | 1538 | /* Push furthest requested GP to leaf node and rcu_data structure. */ |
df2bf8f7 JFG |
1539 | if (ULONG_CMP_LT(gp_seq_req, rnp->gp_seq_needed)) { |
1540 | rnp_start->gp_seq_needed = rnp->gp_seq_needed; | |
1541 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
ab5e869c | 1542 | } |
df2bf8f7 JFG |
1543 | if (rnp != rnp_start) |
1544 | raw_spin_unlock_rcu_node(rnp); | |
48a7639c | 1545 | return ret; |
0446be48 PM |
1546 | } |
1547 | ||
1548 | /* | |
1549 | * Clean up any old requests for the just-ended grace period. Also return | |
d1e4f01d | 1550 | * whether any additional grace periods have been requested. |
0446be48 | 1551 | */ |
3481f2ea | 1552 | static bool rcu_future_gp_cleanup(struct rcu_node *rnp) |
0446be48 | 1553 | { |
fb31340f | 1554 | bool needmore; |
da1df50d | 1555 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
0446be48 | 1556 | |
7a1d0f23 PM |
1557 | needmore = ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed); |
1558 | if (!needmore) | |
1559 | rnp->gp_seq_needed = rnp->gp_seq; /* Avoid counter wrap. */ | |
b73de91d | 1560 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq, |
41e80595 | 1561 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); |
0446be48 PM |
1562 | return needmore; |
1563 | } | |
1564 | ||
48a7639c PM |
1565 | /* |
1566 | * Awaken the grace-period kthread for the specified flavor of RCU. | |
1567 | * Don't do a self-awaken, and don't bother awakening when there is | |
1568 | * nothing for the grace-period kthread to do (as in several CPUs | |
1569 | * raced to awaken, and we lost), and finally don't try to awaken | |
1570 | * a kthread that has not yet been created. | |
1571 | */ | |
532c00c9 | 1572 | static void rcu_gp_kthread_wake(void) |
48a7639c | 1573 | { |
532c00c9 PM |
1574 | if (current == rcu_state.gp_kthread || |
1575 | !READ_ONCE(rcu_state.gp_flags) || | |
1576 | !rcu_state.gp_kthread) | |
48a7639c | 1577 | return; |
532c00c9 | 1578 | swake_up_one(&rcu_state.gp_wq); |
48a7639c PM |
1579 | } |
1580 | ||
dc35c893 | 1581 | /* |
29365e56 PM |
1582 | * If there is room, assign a ->gp_seq number to any callbacks on this |
1583 | * CPU that have not already been assigned. Also accelerate any callbacks | |
1584 | * that were previously assigned a ->gp_seq number that has since proven | |
1585 | * to be too conservative, which can happen if callbacks get assigned a | |
1586 | * ->gp_seq number while RCU is idle, but with reference to a non-root | |
1587 | * rcu_node structure. This function is idempotent, so it does not hurt | |
1588 | * to call it repeatedly. Returns an flag saying that we should awaken | |
1589 | * the RCU grace-period kthread. | |
dc35c893 PM |
1590 | * |
1591 | * The caller must hold rnp->lock with interrupts disabled. | |
1592 | */ | |
02f50142 | 1593 | static bool rcu_accelerate_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1594 | { |
b73de91d | 1595 | unsigned long gp_seq_req; |
15fecf89 | 1596 | bool ret = false; |
dc35c893 | 1597 | |
a32e01ee | 1598 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1599 | |
15fecf89 PM |
1600 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1601 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1602 | return false; |
dc35c893 PM |
1603 | |
1604 | /* | |
15fecf89 PM |
1605 | * Callbacks are often registered with incomplete grace-period |
1606 | * information. Something about the fact that getting exact | |
1607 | * information requires acquiring a global lock... RCU therefore | |
1608 | * makes a conservative estimate of the grace period number at which | |
1609 | * a given callback will become ready to invoke. The following | |
1610 | * code checks this estimate and improves it when possible, thus | |
1611 | * accelerating callback invocation to an earlier grace-period | |
1612 | * number. | |
dc35c893 | 1613 | */ |
9cbc5b97 | 1614 | gp_seq_req = rcu_seq_snap(&rcu_state.gp_seq); |
b73de91d JF |
1615 | if (rcu_segcblist_accelerate(&rdp->cblist, gp_seq_req)) |
1616 | ret = rcu_start_this_gp(rnp, rdp, gp_seq_req); | |
6d4b418c PM |
1617 | |
1618 | /* Trace depending on how much we were able to accelerate. */ | |
15fecf89 | 1619 | if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL)) |
9cbc5b97 | 1620 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccWaitCB")); |
6d4b418c | 1621 | else |
9cbc5b97 | 1622 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccReadyCB")); |
48a7639c | 1623 | return ret; |
dc35c893 PM |
1624 | } |
1625 | ||
e44e73ca PM |
1626 | /* |
1627 | * Similar to rcu_accelerate_cbs(), but does not require that the leaf | |
1628 | * rcu_node structure's ->lock be held. It consults the cached value | |
1629 | * of ->gp_seq_needed in the rcu_data structure, and if that indicates | |
1630 | * that a new grace-period request be made, invokes rcu_accelerate_cbs() | |
1631 | * while holding the leaf rcu_node structure's ->lock. | |
1632 | */ | |
c6e09b97 | 1633 | static void rcu_accelerate_cbs_unlocked(struct rcu_node *rnp, |
e44e73ca PM |
1634 | struct rcu_data *rdp) |
1635 | { | |
1636 | unsigned long c; | |
1637 | bool needwake; | |
1638 | ||
1639 | lockdep_assert_irqs_disabled(); | |
c6e09b97 | 1640 | c = rcu_seq_snap(&rcu_state.gp_seq); |
e44e73ca PM |
1641 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { |
1642 | /* Old request still live, so mark recent callbacks. */ | |
1643 | (void)rcu_segcblist_accelerate(&rdp->cblist, c); | |
1644 | return; | |
1645 | } | |
1646 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
02f50142 | 1647 | needwake = rcu_accelerate_cbs(rnp, rdp); |
e44e73ca PM |
1648 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
1649 | if (needwake) | |
532c00c9 | 1650 | rcu_gp_kthread_wake(); |
e44e73ca PM |
1651 | } |
1652 | ||
dc35c893 PM |
1653 | /* |
1654 | * Move any callbacks whose grace period has completed to the | |
1655 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
29365e56 | 1656 | * assign ->gp_seq numbers to any callbacks in the RCU_NEXT_TAIL |
dc35c893 PM |
1657 | * sublist. This function is idempotent, so it does not hurt to |
1658 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1659 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1660 | * |
1661 | * The caller must hold rnp->lock with interrupts disabled. | |
1662 | */ | |
834f56bf | 1663 | static bool rcu_advance_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1664 | { |
a32e01ee | 1665 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1666 | |
15fecf89 PM |
1667 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1668 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1669 | return false; |
dc35c893 PM |
1670 | |
1671 | /* | |
29365e56 | 1672 | * Find all callbacks whose ->gp_seq numbers indicate that they |
dc35c893 PM |
1673 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. |
1674 | */ | |
29365e56 | 1675 | rcu_segcblist_advance(&rdp->cblist, rnp->gp_seq); |
dc35c893 PM |
1676 | |
1677 | /* Classify any remaining callbacks. */ | |
02f50142 | 1678 | return rcu_accelerate_cbs(rnp, rdp); |
dc35c893 PM |
1679 | } |
1680 | ||
d09b62df | 1681 | /* |
ba9fbe95 PM |
1682 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1683 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1684 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1685 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1686 | */ |
c7e48f7b | 1687 | static bool __note_gp_changes(struct rcu_node *rnp, struct rcu_data *rdp) |
d09b62df | 1688 | { |
48a7639c | 1689 | bool ret; |
3563a438 | 1690 | bool need_gp; |
48a7639c | 1691 | |
a32e01ee | 1692 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1693 | |
67e14c1e PM |
1694 | if (rdp->gp_seq == rnp->gp_seq) |
1695 | return false; /* Nothing to do. */ | |
d09b62df | 1696 | |
67e14c1e PM |
1697 | /* Handle the ends of any preceding grace periods first. */ |
1698 | if (rcu_seq_completed_gp(rdp->gp_seq, rnp->gp_seq) || | |
1699 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
834f56bf | 1700 | ret = rcu_advance_cbs(rnp, rdp); /* Advance callbacks. */ |
9cbc5b97 | 1701 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuend")); |
67e14c1e | 1702 | } else { |
02f50142 | 1703 | ret = rcu_accelerate_cbs(rnp, rdp); /* Recent callbacks. */ |
d09b62df | 1704 | } |
398ebe60 | 1705 | |
67e14c1e PM |
1706 | /* Now handle the beginnings of any new-to-this-CPU grace periods. */ |
1707 | if (rcu_seq_new_gp(rdp->gp_seq, rnp->gp_seq) || | |
1708 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
6eaef633 PM |
1709 | /* |
1710 | * If the current grace period is waiting for this CPU, | |
1711 | * set up to detect a quiescent state, otherwise don't | |
1712 | * go looking for one. | |
1713 | */ | |
9cbc5b97 | 1714 | trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, TPS("cpustart")); |
3563a438 PM |
1715 | need_gp = !!(rnp->qsmask & rdp->grpmask); |
1716 | rdp->cpu_no_qs.b.norm = need_gp; | |
9577df9a | 1717 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
3563a438 | 1718 | rdp->core_needs_qs = need_gp; |
6eaef633 PM |
1719 | zero_cpu_stall_ticks(rdp); |
1720 | } | |
67e14c1e | 1721 | rdp->gp_seq = rnp->gp_seq; /* Remember new grace-period state. */ |
3d18469a PM |
1722 | if (ULONG_CMP_GE(rnp->gp_seq_needed, rdp->gp_seq_needed) || rdp->gpwrap) |
1723 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
1724 | WRITE_ONCE(rdp->gpwrap, false); | |
1725 | rcu_gpnum_ovf(rnp, rdp); | |
48a7639c | 1726 | return ret; |
6eaef633 PM |
1727 | } |
1728 | ||
15cabdff | 1729 | static void note_gp_changes(struct rcu_data *rdp) |
6eaef633 PM |
1730 | { |
1731 | unsigned long flags; | |
48a7639c | 1732 | bool needwake; |
6eaef633 PM |
1733 | struct rcu_node *rnp; |
1734 | ||
1735 | local_irq_save(flags); | |
1736 | rnp = rdp->mynode; | |
67e14c1e | 1737 | if ((rdp->gp_seq == rcu_seq_current(&rnp->gp_seq) && |
7d0ae808 | 1738 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ |
2a67e741 | 1739 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
1740 | local_irq_restore(flags); |
1741 | return; | |
1742 | } | |
c7e48f7b | 1743 | needwake = __note_gp_changes(rnp, rdp); |
67c583a7 | 1744 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c | 1745 | if (needwake) |
532c00c9 | 1746 | rcu_gp_kthread_wake(); |
6eaef633 PM |
1747 | } |
1748 | ||
22212332 | 1749 | static void rcu_gp_slow(int delay) |
0f41c0dd PM |
1750 | { |
1751 | if (delay > 0 && | |
22212332 | 1752 | !(rcu_seq_ctr(rcu_state.gp_seq) % |
dee4f422 | 1753 | (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) |
0f41c0dd PM |
1754 | schedule_timeout_uninterruptible(delay); |
1755 | } | |
1756 | ||
b3dbec76 | 1757 | /* |
45fed3e7 | 1758 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 1759 | */ |
0854a05c | 1760 | static bool rcu_gp_init(void) |
b3dbec76 | 1761 | { |
ec2c2976 | 1762 | unsigned long flags; |
0aa04b05 | 1763 | unsigned long oldmask; |
ec2c2976 | 1764 | unsigned long mask; |
b3dbec76 | 1765 | struct rcu_data *rdp; |
336a4f6c | 1766 | struct rcu_node *rnp = rcu_get_root(); |
b3dbec76 | 1767 | |
9cbc5b97 | 1768 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 1769 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 | 1770 | if (!READ_ONCE(rcu_state.gp_flags)) { |
f7be8209 | 1771 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 1772 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1773 | return false; |
f7be8209 | 1774 | } |
9cbc5b97 | 1775 | WRITE_ONCE(rcu_state.gp_flags, 0); /* Clear all flags: New GP. */ |
b3dbec76 | 1776 | |
de8e8730 | 1777 | if (WARN_ON_ONCE(rcu_gp_in_progress())) { |
f7be8209 PM |
1778 | /* |
1779 | * Grace period already in progress, don't start another. | |
1780 | * Not supposed to be able to happen. | |
1781 | */ | |
67c583a7 | 1782 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1783 | return false; |
7fdefc10 PM |
1784 | } |
1785 | ||
7fdefc10 | 1786 | /* Advance to a new grace period and initialize state. */ |
ad3832e9 | 1787 | record_gp_stall_check_time(); |
ff3bb6f4 | 1788 | /* Record GP times before starting GP, hence rcu_seq_start(). */ |
9cbc5b97 PM |
1789 | rcu_seq_start(&rcu_state.gp_seq); |
1790 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("start")); | |
67c583a7 | 1791 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 1792 | |
0aa04b05 PM |
1793 | /* |
1794 | * Apply per-leaf buffered online and offline operations to the | |
1795 | * rcu_node tree. Note that this new grace period need not wait | |
1796 | * for subsequent online CPUs, and that quiescent-state forcing | |
1797 | * will handle subsequent offline CPUs. | |
1798 | */ | |
9cbc5b97 | 1799 | rcu_state.gp_state = RCU_GP_ONOFF; |
aedf4ba9 | 1800 | rcu_for_each_leaf_node(rnp) { |
9cbc5b97 | 1801 | spin_lock(&rcu_state.ofl_lock); |
2a67e741 | 1802 | raw_spin_lock_irq_rcu_node(rnp); |
0aa04b05 PM |
1803 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
1804 | !rnp->wait_blkd_tasks) { | |
1805 | /* Nothing to do on this leaf rcu_node structure. */ | |
67c583a7 | 1806 | raw_spin_unlock_irq_rcu_node(rnp); |
9cbc5b97 | 1807 | spin_unlock(&rcu_state.ofl_lock); |
0aa04b05 PM |
1808 | continue; |
1809 | } | |
1810 | ||
1811 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
1812 | oldmask = rnp->qsmaskinit; | |
1813 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
1814 | ||
1815 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
1816 | if (!oldmask != !rnp->qsmaskinit) { | |
962aff03 PM |
1817 | if (!oldmask) { /* First online CPU for rcu_node. */ |
1818 | if (!rnp->wait_blkd_tasks) /* Ever offline? */ | |
1819 | rcu_init_new_rnp(rnp); | |
1820 | } else if (rcu_preempt_has_tasks(rnp)) { | |
1821 | rnp->wait_blkd_tasks = true; /* blocked tasks */ | |
1822 | } else { /* Last offline CPU and can propagate. */ | |
0aa04b05 | 1823 | rcu_cleanup_dead_rnp(rnp); |
962aff03 | 1824 | } |
0aa04b05 PM |
1825 | } |
1826 | ||
1827 | /* | |
1828 | * If all waited-on tasks from prior grace period are | |
1829 | * done, and if all this rcu_node structure's CPUs are | |
1830 | * still offline, propagate up the rcu_node tree and | |
1831 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
1832 | * rcu_node structure's CPUs has since come back online, | |
962aff03 | 1833 | * simply clear ->wait_blkd_tasks. |
0aa04b05 PM |
1834 | */ |
1835 | if (rnp->wait_blkd_tasks && | |
962aff03 | 1836 | (!rcu_preempt_has_tasks(rnp) || rnp->qsmaskinit)) { |
0aa04b05 | 1837 | rnp->wait_blkd_tasks = false; |
962aff03 PM |
1838 | if (!rnp->qsmaskinit) |
1839 | rcu_cleanup_dead_rnp(rnp); | |
0aa04b05 PM |
1840 | } |
1841 | ||
67c583a7 | 1842 | raw_spin_unlock_irq_rcu_node(rnp); |
9cbc5b97 | 1843 | spin_unlock(&rcu_state.ofl_lock); |
0aa04b05 | 1844 | } |
22212332 | 1845 | rcu_gp_slow(gp_preinit_delay); /* Races with CPU hotplug. */ |
7fdefc10 PM |
1846 | |
1847 | /* | |
1848 | * Set the quiescent-state-needed bits in all the rcu_node | |
9cbc5b97 PM |
1849 | * structures for all currently online CPUs in breadth-first |
1850 | * order, starting from the root rcu_node structure, relying on the | |
1851 | * layout of the tree within the rcu_state.node[] array. Note that | |
1852 | * other CPUs will access only the leaves of the hierarchy, thus | |
1853 | * seeing that no grace period is in progress, at least until the | |
1854 | * corresponding leaf node has been initialized. | |
7fdefc10 PM |
1855 | * |
1856 | * The grace period cannot complete until the initialization | |
1857 | * process finishes, because this kthread handles both. | |
1858 | */ | |
9cbc5b97 | 1859 | rcu_state.gp_state = RCU_GP_INIT; |
aedf4ba9 | 1860 | rcu_for_each_node_breadth_first(rnp) { |
22212332 | 1861 | rcu_gp_slow(gp_init_delay); |
ec2c2976 | 1862 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
da1df50d | 1863 | rdp = this_cpu_ptr(&rcu_data); |
81ab59a3 | 1864 | rcu_preempt_check_blocked_tasks(rnp); |
7fdefc10 | 1865 | rnp->qsmask = rnp->qsmaskinit; |
9cbc5b97 | 1866 | WRITE_ONCE(rnp->gp_seq, rcu_state.gp_seq); |
7fdefc10 | 1867 | if (rnp == rdp->mynode) |
c7e48f7b | 1868 | (void)__note_gp_changes(rnp, rdp); |
7fdefc10 | 1869 | rcu_preempt_boost_start_gp(rnp); |
9cbc5b97 | 1870 | trace_rcu_grace_period_init(rcu_state.name, rnp->gp_seq, |
7fdefc10 PM |
1871 | rnp->level, rnp->grplo, |
1872 | rnp->grphi, rnp->qsmask); | |
ec2c2976 PM |
1873 | /* Quiescent states for tasks on any now-offline CPUs. */ |
1874 | mask = rnp->qsmask & ~rnp->qsmaskinitnext; | |
f2e2df59 | 1875 | rnp->rcu_gp_init_mask = mask; |
ec2c2976 | 1876 | if ((mask || rnp->wait_blkd_tasks) && rcu_is_leaf_node(rnp)) |
b50912d0 | 1877 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
ec2c2976 PM |
1878 | else |
1879 | raw_spin_unlock_irq_rcu_node(rnp); | |
cee43939 | 1880 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1881 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
7fdefc10 | 1882 | } |
b3dbec76 | 1883 | |
45fed3e7 | 1884 | return true; |
7fdefc10 | 1885 | } |
b3dbec76 | 1886 | |
b9a425cf | 1887 | /* |
b3dae109 | 1888 | * Helper function for swait_event_idle_exclusive() wakeup at force-quiescent-state |
d5374226 | 1889 | * time. |
b9a425cf | 1890 | */ |
0854a05c | 1891 | static bool rcu_gp_fqs_check_wake(int *gfp) |
b9a425cf | 1892 | { |
336a4f6c | 1893 | struct rcu_node *rnp = rcu_get_root(); |
b9a425cf PM |
1894 | |
1895 | /* Someone like call_rcu() requested a force-quiescent-state scan. */ | |
0854a05c | 1896 | *gfp = READ_ONCE(rcu_state.gp_flags); |
b9a425cf PM |
1897 | if (*gfp & RCU_GP_FLAG_FQS) |
1898 | return true; | |
1899 | ||
1900 | /* The current grace period has completed. */ | |
1901 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) | |
1902 | return true; | |
1903 | ||
1904 | return false; | |
1905 | } | |
1906 | ||
4cdfc175 PM |
1907 | /* |
1908 | * Do one round of quiescent-state forcing. | |
1909 | */ | |
0854a05c | 1910 | static void rcu_gp_fqs(bool first_time) |
4cdfc175 | 1911 | { |
336a4f6c | 1912 | struct rcu_node *rnp = rcu_get_root(); |
4cdfc175 | 1913 | |
9cbc5b97 PM |
1914 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
1915 | rcu_state.n_force_qs++; | |
77f81fe0 | 1916 | if (first_time) { |
4cdfc175 | 1917 | /* Collect dyntick-idle snapshots. */ |
e9ecb780 | 1918 | force_qs_rnp(dyntick_save_progress_counter); |
4cdfc175 PM |
1919 | } else { |
1920 | /* Handle dyntick-idle and offline CPUs. */ | |
e9ecb780 | 1921 | force_qs_rnp(rcu_implicit_dynticks_qs); |
4cdfc175 PM |
1922 | } |
1923 | /* Clear flag to prevent immediate re-entry. */ | |
9cbc5b97 | 1924 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 1925 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 PM |
1926 | WRITE_ONCE(rcu_state.gp_flags, |
1927 | READ_ONCE(rcu_state.gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 1928 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 1929 | } |
4cdfc175 PM |
1930 | } |
1931 | ||
c3854a05 PM |
1932 | /* |
1933 | * Loop doing repeated quiescent-state forcing until the grace period ends. | |
1934 | */ | |
1935 | static void rcu_gp_fqs_loop(void) | |
1936 | { | |
1937 | bool first_gp_fqs; | |
1938 | int gf; | |
1939 | unsigned long j; | |
1940 | int ret; | |
1941 | struct rcu_node *rnp = rcu_get_root(); | |
1942 | ||
1943 | first_gp_fqs = true; | |
1944 | j = jiffies_till_first_fqs; | |
1945 | ret = 0; | |
1946 | for (;;) { | |
1947 | if (!ret) { | |
1948 | rcu_state.jiffies_force_qs = jiffies + j; | |
1949 | WRITE_ONCE(rcu_state.jiffies_kick_kthreads, | |
1950 | jiffies + 3 * j); | |
1951 | } | |
1952 | trace_rcu_grace_period(rcu_state.name, | |
1953 | READ_ONCE(rcu_state.gp_seq), | |
1954 | TPS("fqswait")); | |
1955 | rcu_state.gp_state = RCU_GP_WAIT_FQS; | |
1956 | ret = swait_event_idle_timeout_exclusive( | |
1957 | rcu_state.gp_wq, rcu_gp_fqs_check_wake(&gf), j); | |
1958 | rcu_state.gp_state = RCU_GP_DOING_FQS; | |
1959 | /* Locking provides needed memory barriers. */ | |
1960 | /* If grace period done, leave loop. */ | |
1961 | if (!READ_ONCE(rnp->qsmask) && | |
1962 | !rcu_preempt_blocked_readers_cgp(rnp)) | |
1963 | break; | |
1964 | /* If time for quiescent-state forcing, do it. */ | |
1965 | if (ULONG_CMP_GE(jiffies, rcu_state.jiffies_force_qs) || | |
1966 | (gf & RCU_GP_FLAG_FQS)) { | |
1967 | trace_rcu_grace_period(rcu_state.name, | |
1968 | READ_ONCE(rcu_state.gp_seq), | |
1969 | TPS("fqsstart")); | |
1970 | rcu_gp_fqs(first_gp_fqs); | |
1971 | first_gp_fqs = false; | |
1972 | trace_rcu_grace_period(rcu_state.name, | |
1973 | READ_ONCE(rcu_state.gp_seq), | |
1974 | TPS("fqsend")); | |
1975 | cond_resched_tasks_rcu_qs(); | |
1976 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1977 | ret = 0; /* Force full wait till next FQS. */ | |
1978 | j = jiffies_till_next_fqs; | |
1979 | } else { | |
1980 | /* Deal with stray signal. */ | |
1981 | cond_resched_tasks_rcu_qs(); | |
1982 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1983 | WARN_ON(signal_pending(current)); | |
1984 | trace_rcu_grace_period(rcu_state.name, | |
1985 | READ_ONCE(rcu_state.gp_seq), | |
1986 | TPS("fqswaitsig")); | |
1987 | ret = 1; /* Keep old FQS timing. */ | |
1988 | j = jiffies; | |
1989 | if (time_after(jiffies, rcu_state.jiffies_force_qs)) | |
1990 | j = 1; | |
1991 | else | |
1992 | j = rcu_state.jiffies_force_qs - j; | |
1993 | } | |
1994 | } | |
1995 | } | |
1996 | ||
7fdefc10 PM |
1997 | /* |
1998 | * Clean up after the old grace period. | |
1999 | */ | |
0854a05c | 2000 | static void rcu_gp_cleanup(void) |
7fdefc10 PM |
2001 | { |
2002 | unsigned long gp_duration; | |
48a7639c | 2003 | bool needgp = false; |
de30ad51 | 2004 | unsigned long new_gp_seq; |
7fdefc10 | 2005 | struct rcu_data *rdp; |
336a4f6c | 2006 | struct rcu_node *rnp = rcu_get_root(); |
abedf8e2 | 2007 | struct swait_queue_head *sq; |
b3dbec76 | 2008 | |
9cbc5b97 | 2009 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 2010 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 PM |
2011 | gp_duration = jiffies - rcu_state.gp_start; |
2012 | if (gp_duration > rcu_state.gp_max) | |
2013 | rcu_state.gp_max = gp_duration; | |
b3dbec76 | 2014 | |
7fdefc10 PM |
2015 | /* |
2016 | * We know the grace period is complete, but to everyone else | |
2017 | * it appears to still be ongoing. But it is also the case | |
2018 | * that to everyone else it looks like there is nothing that | |
2019 | * they can do to advance the grace period. It is therefore | |
2020 | * safe for us to drop the lock in order to mark the grace | |
2021 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 2022 | */ |
67c583a7 | 2023 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 2024 | |
5d4b8659 | 2025 | /* |
ff3bb6f4 PM |
2026 | * Propagate new ->gp_seq value to rcu_node structures so that |
2027 | * other CPUs don't have to wait until the start of the next grace | |
2028 | * period to process their callbacks. This also avoids some nasty | |
2029 | * RCU grace-period initialization races by forcing the end of | |
2030 | * the current grace period to be completely recorded in all of | |
2031 | * the rcu_node structures before the beginning of the next grace | |
2032 | * period is recorded in any of the rcu_node structures. | |
5d4b8659 | 2033 | */ |
9cbc5b97 | 2034 | new_gp_seq = rcu_state.gp_seq; |
de30ad51 | 2035 | rcu_seq_end(&new_gp_seq); |
aedf4ba9 | 2036 | rcu_for_each_node_breadth_first(rnp) { |
2a67e741 | 2037 | raw_spin_lock_irq_rcu_node(rnp); |
4bc8d555 | 2038 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 2039 | dump_blkd_tasks(rnp, 10); |
5c60d25f | 2040 | WARN_ON_ONCE(rnp->qsmask); |
de30ad51 | 2041 | WRITE_ONCE(rnp->gp_seq, new_gp_seq); |
da1df50d | 2042 | rdp = this_cpu_ptr(&rcu_data); |
b11cc576 | 2043 | if (rnp == rdp->mynode) |
c7e48f7b | 2044 | needgp = __note_gp_changes(rnp, rdp) || needgp; |
78e4bc34 | 2045 | /* smp_mb() provided by prior unlock-lock pair. */ |
3481f2ea | 2046 | needgp = rcu_future_gp_cleanup(rnp) || needgp; |
065bb78c | 2047 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 2048 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 2049 | rcu_nocb_gp_cleanup(sq); |
cee43939 | 2050 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 2051 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
22212332 | 2052 | rcu_gp_slow(gp_cleanup_delay); |
7fdefc10 | 2053 | } |
336a4f6c | 2054 | rnp = rcu_get_root(); |
9cbc5b97 | 2055 | raw_spin_lock_irq_rcu_node(rnp); /* GP before ->gp_seq update. */ |
7fdefc10 | 2056 | |
765a3f4f | 2057 | /* Declare grace period done. */ |
9cbc5b97 PM |
2058 | rcu_seq_end(&rcu_state.gp_seq); |
2059 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("end")); | |
2060 | rcu_state.gp_state = RCU_GP_IDLE; | |
fb31340f | 2061 | /* Check for GP requests since above loop. */ |
da1df50d | 2062 | rdp = this_cpu_ptr(&rcu_data); |
5b55072f | 2063 | if (!needgp && ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed)) { |
abd13fdd | 2064 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq_needed, |
41e80595 | 2065 | TPS("CleanupMore")); |
fb31340f PM |
2066 | needgp = true; |
2067 | } | |
48a7639c | 2068 | /* Advance CBs to reduce false positives below. */ |
02f50142 | 2069 | if (!rcu_accelerate_cbs(rnp, rdp) && needgp) { |
9cbc5b97 PM |
2070 | WRITE_ONCE(rcu_state.gp_flags, RCU_GP_FLAG_INIT); |
2071 | rcu_state.gp_req_activity = jiffies; | |
2072 | trace_rcu_grace_period(rcu_state.name, | |
2073 | READ_ONCE(rcu_state.gp_seq), | |
bb311ecc | 2074 | TPS("newreq")); |
18390aea | 2075 | } else { |
9cbc5b97 PM |
2076 | WRITE_ONCE(rcu_state.gp_flags, |
2077 | rcu_state.gp_flags & RCU_GP_FLAG_INIT); | |
bb311ecc | 2078 | } |
67c583a7 | 2079 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 PM |
2080 | } |
2081 | ||
2082 | /* | |
2083 | * Body of kthread that handles grace periods. | |
2084 | */ | |
0854a05c | 2085 | static int __noreturn rcu_gp_kthread(void *unused) |
7fdefc10 | 2086 | { |
5871968d | 2087 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
2088 | for (;;) { |
2089 | ||
2090 | /* Handle grace-period start. */ | |
2091 | for (;;) { | |
9cbc5b97 PM |
2092 | trace_rcu_grace_period(rcu_state.name, |
2093 | READ_ONCE(rcu_state.gp_seq), | |
63c4db78 | 2094 | TPS("reqwait")); |
9cbc5b97 PM |
2095 | rcu_state.gp_state = RCU_GP_WAIT_GPS; |
2096 | swait_event_idle_exclusive(rcu_state.gp_wq, | |
2097 | READ_ONCE(rcu_state.gp_flags) & | |
2098 | RCU_GP_FLAG_INIT); | |
2099 | rcu_state.gp_state = RCU_GP_DONE_GPS; | |
78e4bc34 | 2100 | /* Locking provides needed memory barrier. */ |
0854a05c | 2101 | if (rcu_gp_init()) |
7fdefc10 | 2102 | break; |
cee43939 | 2103 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 2104 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
73a860cd | 2105 | WARN_ON(signal_pending(current)); |
9cbc5b97 PM |
2106 | trace_rcu_grace_period(rcu_state.name, |
2107 | READ_ONCE(rcu_state.gp_seq), | |
63c4db78 | 2108 | TPS("reqwaitsig")); |
7fdefc10 | 2109 | } |
cabc49c1 | 2110 | |
4cdfc175 | 2111 | /* Handle quiescent-state forcing. */ |
c3854a05 | 2112 | rcu_gp_fqs_loop(); |
4cdfc175 PM |
2113 | |
2114 | /* Handle grace-period end. */ | |
9cbc5b97 | 2115 | rcu_state.gp_state = RCU_GP_CLEANUP; |
0854a05c | 2116 | rcu_gp_cleanup(); |
9cbc5b97 | 2117 | rcu_state.gp_state = RCU_GP_CLEANED; |
b3dbec76 | 2118 | } |
b3dbec76 PM |
2119 | } |
2120 | ||
f41d911f | 2121 | /* |
8994515c PM |
2122 | * Report a full set of quiescent states to the specified rcu_state data |
2123 | * structure. Invoke rcu_gp_kthread_wake() to awaken the grace-period | |
2124 | * kthread if another grace period is required. Whether we wake | |
2125 | * the grace-period kthread or it awakens itself for the next round | |
2126 | * of quiescent-state forcing, that kthread will clean up after the | |
2127 | * just-completed grace period. Note that the caller must hold rnp->lock, | |
2128 | * which is released before return. | |
f41d911f | 2129 | */ |
aff4e9ed | 2130 | static void rcu_report_qs_rsp(unsigned long flags) |
336a4f6c | 2131 | __releases(rcu_get_root()->lock) |
f41d911f | 2132 | { |
336a4f6c | 2133 | raw_lockdep_assert_held_rcu_node(rcu_get_root()); |
de8e8730 | 2134 | WARN_ON_ONCE(!rcu_gp_in_progress()); |
9cbc5b97 PM |
2135 | WRITE_ONCE(rcu_state.gp_flags, |
2136 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
336a4f6c | 2137 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(), flags); |
532c00c9 | 2138 | rcu_gp_kthread_wake(); |
f41d911f PM |
2139 | } |
2140 | ||
64db4cff | 2141 | /* |
d3f6bad3 PM |
2142 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
2143 | * Allows quiescent states for a group of CPUs to be reported at one go | |
2144 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
2145 | * must be represented by the same rcu_node structure (which need not be a |
2146 | * leaf rcu_node structure, though it often will be). The gps parameter | |
2147 | * is the grace-period snapshot, which means that the quiescent states | |
c9a24e2d | 2148 | * are valid only if rnp->gp_seq is equal to gps. That structure's lock |
654e9533 | 2149 | * must be held upon entry, and it is released before return. |
ec2c2976 PM |
2150 | * |
2151 | * As a special case, if mask is zero, the bit-already-cleared check is | |
2152 | * disabled. This allows propagating quiescent state due to resumed tasks | |
2153 | * during grace-period initialization. | |
64db4cff | 2154 | */ |
b50912d0 PM |
2155 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
2156 | unsigned long gps, unsigned long flags) | |
64db4cff PM |
2157 | __releases(rnp->lock) |
2158 | { | |
654e9533 | 2159 | unsigned long oldmask = 0; |
28ecd580 PM |
2160 | struct rcu_node *rnp_c; |
2161 | ||
a32e01ee | 2162 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 2163 | |
64db4cff PM |
2164 | /* Walk up the rcu_node hierarchy. */ |
2165 | for (;;) { | |
ec2c2976 | 2166 | if ((!(rnp->qsmask & mask) && mask) || rnp->gp_seq != gps) { |
64db4cff | 2167 | |
654e9533 PM |
2168 | /* |
2169 | * Our bit has already been cleared, or the | |
2170 | * relevant grace period is already over, so done. | |
2171 | */ | |
67c583a7 | 2172 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2173 | return; |
2174 | } | |
654e9533 | 2175 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
5b4c11d5 | 2176 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp) && |
2dee9404 | 2177 | rcu_preempt_blocked_readers_cgp(rnp)); |
64db4cff | 2178 | rnp->qsmask &= ~mask; |
67a0edbf | 2179 | trace_rcu_quiescent_state_report(rcu_state.name, rnp->gp_seq, |
d4c08f2a PM |
2180 | mask, rnp->qsmask, rnp->level, |
2181 | rnp->grplo, rnp->grphi, | |
2182 | !!rnp->gp_tasks); | |
27f4d280 | 2183 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
2184 | |
2185 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 2186 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2187 | return; |
2188 | } | |
d43a5d32 | 2189 | rnp->completedqs = rnp->gp_seq; |
64db4cff PM |
2190 | mask = rnp->grpmask; |
2191 | if (rnp->parent == NULL) { | |
2192 | ||
2193 | /* No more levels. Exit loop holding root lock. */ | |
2194 | ||
2195 | break; | |
2196 | } | |
67c583a7 | 2197 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 2198 | rnp_c = rnp; |
64db4cff | 2199 | rnp = rnp->parent; |
2a67e741 | 2200 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
654e9533 | 2201 | oldmask = rnp_c->qsmask; |
64db4cff PM |
2202 | } |
2203 | ||
2204 | /* | |
2205 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 2206 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 2207 | * to clean up and start the next grace period if one is needed. |
64db4cff | 2208 | */ |
aff4e9ed | 2209 | rcu_report_qs_rsp(flags); /* releases rnp->lock. */ |
64db4cff PM |
2210 | } |
2211 | ||
cc99a310 PM |
2212 | /* |
2213 | * Record a quiescent state for all tasks that were previously queued | |
2214 | * on the specified rcu_node structure and that were blocking the current | |
2215 | * RCU grace period. The caller must hold the specified rnp->lock with | |
2216 | * irqs disabled, and this lock is released upon return, but irqs remain | |
2217 | * disabled. | |
2218 | */ | |
17a8212b | 2219 | static void __maybe_unused |
139ad4da | 2220 | rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
cc99a310 PM |
2221 | __releases(rnp->lock) |
2222 | { | |
654e9533 | 2223 | unsigned long gps; |
cc99a310 PM |
2224 | unsigned long mask; |
2225 | struct rcu_node *rnp_p; | |
2226 | ||
a32e01ee | 2227 | raw_lockdep_assert_held_rcu_node(rnp); |
45975c7d | 2228 | if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPT)) || |
c74859d1 PM |
2229 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)) || |
2230 | rnp->qsmask != 0) { | |
67c583a7 | 2231 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
2232 | return; /* Still need more quiescent states! */ |
2233 | } | |
2234 | ||
77cfc7bf | 2235 | rnp->completedqs = rnp->gp_seq; |
cc99a310 PM |
2236 | rnp_p = rnp->parent; |
2237 | if (rnp_p == NULL) { | |
2238 | /* | |
a77da14c PM |
2239 | * Only one rcu_node structure in the tree, so don't |
2240 | * try to report up to its nonexistent parent! | |
cc99a310 | 2241 | */ |
aff4e9ed | 2242 | rcu_report_qs_rsp(flags); |
cc99a310 PM |
2243 | return; |
2244 | } | |
2245 | ||
c9a24e2d PM |
2246 | /* Report up the rest of the hierarchy, tracking current ->gp_seq. */ |
2247 | gps = rnp->gp_seq; | |
cc99a310 | 2248 | mask = rnp->grpmask; |
67c583a7 | 2249 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 2250 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
b50912d0 | 2251 | rcu_report_qs_rnp(mask, rnp_p, gps, flags); |
cc99a310 PM |
2252 | } |
2253 | ||
64db4cff | 2254 | /* |
d3f6bad3 | 2255 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 2256 | * structure. This must be called from the specified CPU. |
64db4cff PM |
2257 | */ |
2258 | static void | |
33085c46 | 2259 | rcu_report_qs_rdp(int cpu, struct rcu_data *rdp) |
64db4cff PM |
2260 | { |
2261 | unsigned long flags; | |
2262 | unsigned long mask; | |
48a7639c | 2263 | bool needwake; |
64db4cff PM |
2264 | struct rcu_node *rnp; |
2265 | ||
2266 | rnp = rdp->mynode; | |
2a67e741 | 2267 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
c9a24e2d PM |
2268 | if (rdp->cpu_no_qs.b.norm || rdp->gp_seq != rnp->gp_seq || |
2269 | rdp->gpwrap) { | |
64db4cff PM |
2270 | |
2271 | /* | |
e4cc1f22 PM |
2272 | * The grace period in which this quiescent state was |
2273 | * recorded has ended, so don't report it upwards. | |
2274 | * We will instead need a new quiescent state that lies | |
2275 | * within the current grace period. | |
64db4cff | 2276 | */ |
5b74c458 | 2277 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
9577df9a | 2278 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
67c583a7 | 2279 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2280 | return; |
2281 | } | |
2282 | mask = rdp->grpmask; | |
2283 | if ((rnp->qsmask & mask) == 0) { | |
67c583a7 | 2284 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2285 | } else { |
bb53e416 | 2286 | rdp->core_needs_qs = false; |
64db4cff PM |
2287 | |
2288 | /* | |
2289 | * This GP can't end until cpu checks in, so all of our | |
2290 | * callbacks can be processed during the next GP. | |
2291 | */ | |
02f50142 | 2292 | needwake = rcu_accelerate_cbs(rnp, rdp); |
64db4cff | 2293 | |
b50912d0 | 2294 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
654e9533 | 2295 | /* ^^^ Released rnp->lock */ |
48a7639c | 2296 | if (needwake) |
532c00c9 | 2297 | rcu_gp_kthread_wake(); |
64db4cff PM |
2298 | } |
2299 | } | |
2300 | ||
2301 | /* | |
2302 | * Check to see if there is a new grace period of which this CPU | |
2303 | * is not yet aware, and if so, set up local rcu_data state for it. | |
2304 | * Otherwise, see if this CPU has just passed through its first | |
2305 | * quiescent state for this grace period, and record that fact if so. | |
2306 | */ | |
2307 | static void | |
8087d3e3 | 2308 | rcu_check_quiescent_state(struct rcu_data *rdp) |
64db4cff | 2309 | { |
05eb552b | 2310 | /* Check for grace-period ends and beginnings. */ |
15cabdff | 2311 | note_gp_changes(rdp); |
64db4cff PM |
2312 | |
2313 | /* | |
2314 | * Does this CPU still need to do its part for current grace period? | |
2315 | * If no, return and let the other CPUs do their part as well. | |
2316 | */ | |
97c668b8 | 2317 | if (!rdp->core_needs_qs) |
64db4cff PM |
2318 | return; |
2319 | ||
2320 | /* | |
2321 | * Was there a quiescent state since the beginning of the grace | |
2322 | * period? If no, then exit and wait for the next call. | |
2323 | */ | |
3a19b46a | 2324 | if (rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2325 | return; |
2326 | ||
d3f6bad3 PM |
2327 | /* |
2328 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
2329 | * judge of that). | |
2330 | */ | |
33085c46 | 2331 | rcu_report_qs_rdp(rdp->cpu, rdp); |
64db4cff PM |
2332 | } |
2333 | ||
b1420f1c | 2334 | /* |
780cd590 PM |
2335 | * Near the end of the offline process. Trace the fact that this CPU |
2336 | * is going offline. | |
b1420f1c | 2337 | */ |
780cd590 | 2338 | int rcutree_dying_cpu(unsigned int cpu) |
b1420f1c | 2339 | { |
477351f7 | 2340 | RCU_TRACE(bool blkd;) |
da1df50d | 2341 | RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(&rcu_data);) |
88a4976d | 2342 | RCU_TRACE(struct rcu_node *rnp = rdp->mynode;) |
b1420f1c | 2343 | |
ea46351c | 2344 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 2345 | return 0; |
ea46351c | 2346 | |
477351f7 | 2347 | RCU_TRACE(blkd = !!(rnp->qsmask & rdp->grpmask);) |
780cd590 | 2348 | trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, |
477351f7 | 2349 | blkd ? TPS("cpuofl") : TPS("cpuofl-bgp")); |
780cd590 | 2350 | return 0; |
64db4cff PM |
2351 | } |
2352 | ||
8af3a5e7 PM |
2353 | /* |
2354 | * All CPUs for the specified rcu_node structure have gone offline, | |
2355 | * and all tasks that were preempted within an RCU read-side critical | |
2356 | * section while running on one of those CPUs have since exited their RCU | |
2357 | * read-side critical section. Some other CPU is reporting this fact with | |
2358 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2359 | * This function therefore goes up the tree of rcu_node structures, | |
2360 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2361 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
c50cbe53 | 2362 | * updated. |
8af3a5e7 PM |
2363 | * |
2364 | * This function does check that the specified rcu_node structure has | |
2365 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2366 | * prematurely. That said, invoking it after the fact will cost you | |
2367 | * a needless lock acquisition. So once it has done its work, don't | |
2368 | * invoke it again. | |
2369 | */ | |
2370 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2371 | { | |
2372 | long mask; | |
2373 | struct rcu_node *rnp = rnp_leaf; | |
2374 | ||
962aff03 | 2375 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
ea46351c | 2376 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
962aff03 PM |
2377 | WARN_ON_ONCE(rnp_leaf->qsmaskinit) || |
2378 | WARN_ON_ONCE(rcu_preempt_has_tasks(rnp_leaf))) | |
8af3a5e7 PM |
2379 | return; |
2380 | for (;;) { | |
2381 | mask = rnp->grpmask; | |
2382 | rnp = rnp->parent; | |
2383 | if (!rnp) | |
2384 | break; | |
2a67e741 | 2385 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2386 | rnp->qsmaskinit &= ~mask; |
962aff03 PM |
2387 | /* Between grace periods, so better already be zero! */ |
2388 | WARN_ON_ONCE(rnp->qsmask); | |
8af3a5e7 | 2389 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2390 | raw_spin_unlock_rcu_node(rnp); |
2391 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2392 | return; |
2393 | } | |
67c583a7 | 2394 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2395 | } |
2396 | } | |
2397 | ||
64db4cff | 2398 | /* |
e5601400 | 2399 | * The CPU has been completely removed, and some other CPU is reporting |
a58163d8 PM |
2400 | * this fact from process context. Do the remainder of the cleanup. |
2401 | * There can only be one CPU hotplug operation at a time, so no need for | |
2402 | * explicit locking. | |
64db4cff | 2403 | */ |
780cd590 | 2404 | int rcutree_dead_cpu(unsigned int cpu) |
64db4cff | 2405 | { |
da1df50d | 2406 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
b1420f1c | 2407 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2408 | |
ea46351c | 2409 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 2410 | return 0; |
ea46351c | 2411 | |
2036d94a | 2412 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2413 | rcu_boost_kthread_setaffinity(rnp, -1); |
780cd590 PM |
2414 | /* Do any needed no-CB deferred wakeups from this CPU. */ |
2415 | do_nocb_deferred_wakeup(per_cpu_ptr(&rcu_data, cpu)); | |
2416 | return 0; | |
64db4cff PM |
2417 | } |
2418 | ||
64db4cff PM |
2419 | /* |
2420 | * Invoke any RCU callbacks that have made it to the end of their grace | |
2421 | * period. Thottle as specified by rdp->blimit. | |
2422 | */ | |
5bb5d09c | 2423 | static void rcu_do_batch(struct rcu_data *rdp) |
64db4cff PM |
2424 | { |
2425 | unsigned long flags; | |
15fecf89 PM |
2426 | struct rcu_head *rhp; |
2427 | struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl); | |
2428 | long bl, count; | |
64db4cff | 2429 | |
dc35c893 | 2430 | /* If no callbacks are ready, just return. */ |
15fecf89 | 2431 | if (!rcu_segcblist_ready_cbs(&rdp->cblist)) { |
3c779dfe | 2432 | trace_rcu_batch_start(rcu_state.name, |
15fecf89 PM |
2433 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2434 | rcu_segcblist_n_cbs(&rdp->cblist), 0); | |
3c779dfe | 2435 | trace_rcu_batch_end(rcu_state.name, 0, |
15fecf89 | 2436 | !rcu_segcblist_empty(&rdp->cblist), |
4968c300 PM |
2437 | need_resched(), is_idle_task(current), |
2438 | rcu_is_callbacks_kthread()); | |
64db4cff | 2439 | return; |
29c00b4a | 2440 | } |
64db4cff PM |
2441 | |
2442 | /* | |
2443 | * Extract the list of ready callbacks, disabling to prevent | |
15fecf89 PM |
2444 | * races with call_rcu() from interrupt handlers. Leave the |
2445 | * callback counts, as rcu_barrier() needs to be conservative. | |
64db4cff PM |
2446 | */ |
2447 | local_irq_save(flags); | |
8146c4e2 | 2448 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 2449 | bl = rdp->blimit; |
3c779dfe PM |
2450 | trace_rcu_batch_start(rcu_state.name, |
2451 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
15fecf89 PM |
2452 | rcu_segcblist_n_cbs(&rdp->cblist), bl); |
2453 | rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl); | |
64db4cff PM |
2454 | local_irq_restore(flags); |
2455 | ||
2456 | /* Invoke callbacks. */ | |
15fecf89 PM |
2457 | rhp = rcu_cblist_dequeue(&rcl); |
2458 | for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) { | |
2459 | debug_rcu_head_unqueue(rhp); | |
3c779dfe | 2460 | if (__rcu_reclaim(rcu_state.name, rhp)) |
15fecf89 PM |
2461 | rcu_cblist_dequeued_lazy(&rcl); |
2462 | /* | |
2463 | * Stop only if limit reached and CPU has something to do. | |
2464 | * Note: The rcl structure counts down from zero. | |
2465 | */ | |
4b27f20b | 2466 | if (-rcl.len >= bl && |
dff1672d PM |
2467 | (need_resched() || |
2468 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
2469 | break; |
2470 | } | |
2471 | ||
2472 | local_irq_save(flags); | |
4b27f20b | 2473 | count = -rcl.len; |
3c779dfe | 2474 | trace_rcu_batch_end(rcu_state.name, count, !!rcl.head, need_resched(), |
8ef0f37e | 2475 | is_idle_task(current), rcu_is_callbacks_kthread()); |
64db4cff | 2476 | |
15fecf89 PM |
2477 | /* Update counts and requeue any remaining callbacks. */ |
2478 | rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl); | |
b1420f1c | 2479 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
15fecf89 | 2480 | rcu_segcblist_insert_count(&rdp->cblist, &rcl); |
64db4cff PM |
2481 | |
2482 | /* Reinstate batch limit if we have worked down the excess. */ | |
15fecf89 PM |
2483 | count = rcu_segcblist_n_cbs(&rdp->cblist); |
2484 | if (rdp->blimit == LONG_MAX && count <= qlowmark) | |
64db4cff PM |
2485 | rdp->blimit = blimit; |
2486 | ||
37c72e56 | 2487 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
15fecf89 | 2488 | if (count == 0 && rdp->qlen_last_fqs_check != 0) { |
37c72e56 | 2489 | rdp->qlen_last_fqs_check = 0; |
3c779dfe | 2490 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
15fecf89 PM |
2491 | } else if (count < rdp->qlen_last_fqs_check - qhimark) |
2492 | rdp->qlen_last_fqs_check = count; | |
efd88b02 PM |
2493 | |
2494 | /* | |
2495 | * The following usually indicates a double call_rcu(). To track | |
2496 | * this down, try building with CONFIG_DEBUG_OBJECTS_RCU_HEAD=y. | |
2497 | */ | |
15fecf89 | 2498 | WARN_ON_ONCE(rcu_segcblist_empty(&rdp->cblist) != (count == 0)); |
37c72e56 | 2499 | |
64db4cff PM |
2500 | local_irq_restore(flags); |
2501 | ||
e0f23060 | 2502 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
15fecf89 | 2503 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2504 | invoke_rcu_core(); |
64db4cff PM |
2505 | } |
2506 | ||
2507 | /* | |
2508 | * Check to see if this CPU is in a non-context-switch quiescent state | |
2509 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 2510 | * Also schedule RCU core processing. |
64db4cff | 2511 | * |
9b2e4f18 | 2512 | * This function must be called from hardirq context. It is normally |
5403d367 | 2513 | * invoked from the scheduling-clock interrupt. |
64db4cff | 2514 | */ |
c3377c2d | 2515 | void rcu_check_callbacks(int user) |
64db4cff | 2516 | { |
f7f7bac9 | 2517 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
4e95020c | 2518 | raw_cpu_inc(rcu_data.ticks_this_gp); |
45975c7d | 2519 | rcu_flavor_check_callbacks(user); |
e3950ecd | 2520 | if (rcu_pending()) |
a46e0899 | 2521 | invoke_rcu_core(); |
07f27570 | 2522 | |
f7f7bac9 | 2523 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2524 | } |
2525 | ||
64db4cff PM |
2526 | /* |
2527 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
2528 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
2529 | * Also initiate boosting for any threads blocked on the root rcu_node. |
2530 | * | |
ee47eb9f | 2531 | * The caller must have suppressed start of new grace periods. |
64db4cff | 2532 | */ |
8ff0b907 | 2533 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)) |
64db4cff | 2534 | { |
64db4cff PM |
2535 | int cpu; |
2536 | unsigned long flags; | |
2537 | unsigned long mask; | |
a0b6c9a7 | 2538 | struct rcu_node *rnp; |
64db4cff | 2539 | |
aedf4ba9 | 2540 | rcu_for_each_leaf_node(rnp) { |
cee43939 | 2541 | cond_resched_tasks_rcu_qs(); |
64db4cff | 2542 | mask = 0; |
2a67e741 | 2543 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
a0b6c9a7 | 2544 | if (rnp->qsmask == 0) { |
45975c7d | 2545 | if (!IS_ENABLED(CONFIG_PREEMPT) || |
a77da14c PM |
2546 | rcu_preempt_blocked_readers_cgp(rnp)) { |
2547 | /* | |
2548 | * No point in scanning bits because they | |
2549 | * are all zero. But we might need to | |
2550 | * priority-boost blocked readers. | |
2551 | */ | |
2552 | rcu_initiate_boost(rnp, flags); | |
2553 | /* rcu_initiate_boost() releases rnp->lock */ | |
2554 | continue; | |
2555 | } | |
92816435 PM |
2556 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
2557 | continue; | |
64db4cff | 2558 | } |
bc75e999 MR |
2559 | for_each_leaf_node_possible_cpu(rnp, cpu) { |
2560 | unsigned long bit = leaf_node_cpu_bit(rnp, cpu); | |
0edd1b17 | 2561 | if ((rnp->qsmask & bit) != 0) { |
da1df50d | 2562 | if (f(per_cpu_ptr(&rcu_data, cpu))) |
0edd1b17 PM |
2563 | mask |= bit; |
2564 | } | |
64db4cff | 2565 | } |
45f014c5 | 2566 | if (mask != 0) { |
c9a24e2d | 2567 | /* Idle/offline CPUs, report (releases rnp->lock). */ |
b50912d0 | 2568 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
0aa04b05 PM |
2569 | } else { |
2570 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 2571 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2572 | } |
64db4cff | 2573 | } |
64db4cff PM |
2574 | } |
2575 | ||
2576 | /* | |
2577 | * Force quiescent states on reluctant CPUs, and also detect which | |
2578 | * CPUs are in dyntick-idle mode. | |
2579 | */ | |
e9ecb780 | 2580 | static void force_quiescent_state(void) |
64db4cff PM |
2581 | { |
2582 | unsigned long flags; | |
394f2769 PM |
2583 | bool ret; |
2584 | struct rcu_node *rnp; | |
2585 | struct rcu_node *rnp_old = NULL; | |
2586 | ||
2587 | /* Funnel through hierarchy to reduce memory contention. */ | |
da1df50d | 2588 | rnp = __this_cpu_read(rcu_data.mynode); |
394f2769 | 2589 | for (; rnp != NULL; rnp = rnp->parent) { |
67a0edbf | 2590 | ret = (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) || |
394f2769 PM |
2591 | !raw_spin_trylock(&rnp->fqslock); |
2592 | if (rnp_old != NULL) | |
2593 | raw_spin_unlock(&rnp_old->fqslock); | |
d62df573 | 2594 | if (ret) |
394f2769 | 2595 | return; |
394f2769 PM |
2596 | rnp_old = rnp; |
2597 | } | |
336a4f6c | 2598 | /* rnp_old == rcu_get_root(), rnp == NULL. */ |
64db4cff | 2599 | |
394f2769 | 2600 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 2601 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 2602 | raw_spin_unlock(&rnp_old->fqslock); |
67a0edbf | 2603 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
67c583a7 | 2604 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 2605 | return; /* Someone beat us to it. */ |
46a1e34e | 2606 | } |
67a0edbf PM |
2607 | WRITE_ONCE(rcu_state.gp_flags, |
2608 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
67c583a7 | 2609 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
532c00c9 | 2610 | rcu_gp_kthread_wake(); |
64db4cff PM |
2611 | } |
2612 | ||
26d950a9 PM |
2613 | /* |
2614 | * This function checks for grace-period requests that fail to motivate | |
2615 | * RCU to come out of its idle mode. | |
2616 | */ | |
2617 | static void | |
b96f9dc4 | 2618 | rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp) |
26d950a9 | 2619 | { |
b06ae25a | 2620 | const unsigned long gpssdelay = rcu_jiffies_till_stall_check() * HZ; |
26d950a9 PM |
2621 | unsigned long flags; |
2622 | unsigned long j; | |
336a4f6c | 2623 | struct rcu_node *rnp_root = rcu_get_root(); |
26d950a9 PM |
2624 | static atomic_t warned = ATOMIC_INIT(0); |
2625 | ||
de8e8730 | 2626 | if (!IS_ENABLED(CONFIG_PROVE_RCU) || rcu_gp_in_progress() || |
7a1d0f23 | 2627 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed)) |
26d950a9 PM |
2628 | return; |
2629 | j = jiffies; /* Expensive access, and in common case don't get here. */ | |
67a0edbf PM |
2630 | if (time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) || |
2631 | time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) || | |
26d950a9 PM |
2632 | atomic_read(&warned)) |
2633 | return; | |
2634 | ||
2635 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
2636 | j = jiffies; | |
de8e8730 | 2637 | if (rcu_gp_in_progress() || |
7a1d0f23 | 2638 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) || |
67a0edbf PM |
2639 | time_before(j, READ_ONCE(rcu_state.gp_req_activity) + gpssdelay) || |
2640 | time_before(j, READ_ONCE(rcu_state.gp_activity) + gpssdelay) || | |
26d950a9 PM |
2641 | atomic_read(&warned)) { |
2642 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2643 | return; | |
2644 | } | |
2645 | /* Hold onto the leaf lock to make others see warned==1. */ | |
2646 | ||
2647 | if (rnp_root != rnp) | |
2648 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ | |
2649 | j = jiffies; | |
de8e8730 | 2650 | if (rcu_gp_in_progress() || |
7a1d0f23 | 2651 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) || |
67a0edbf PM |
2652 | time_before(j, rcu_state.gp_req_activity + gpssdelay) || |
2653 | time_before(j, rcu_state.gp_activity + gpssdelay) || | |
26d950a9 PM |
2654 | atomic_xchg(&warned, 1)) { |
2655 | raw_spin_unlock_rcu_node(rnp_root); /* irqs remain disabled. */ | |
2656 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2657 | return; | |
2658 | } | |
b06ae25a | 2659 | pr_alert("%s: g%ld->%ld gar:%lu ga:%lu f%#x gs:%d %s->state:%#lx\n", |
67a0edbf | 2660 | __func__, (long)READ_ONCE(rcu_state.gp_seq), |
7a1d0f23 | 2661 | (long)READ_ONCE(rnp_root->gp_seq_needed), |
67a0edbf PM |
2662 | j - rcu_state.gp_req_activity, j - rcu_state.gp_activity, |
2663 | rcu_state.gp_flags, rcu_state.gp_state, rcu_state.name, | |
2664 | rcu_state.gp_kthread ? rcu_state.gp_kthread->state : 0x1ffffL); | |
26d950a9 PM |
2665 | WARN_ON(1); |
2666 | if (rnp_root != rnp) | |
2667 | raw_spin_unlock_rcu_node(rnp_root); | |
2668 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2669 | } | |
2670 | ||
64db4cff | 2671 | /* |
b049fdf8 PM |
2672 | * This does the RCU core processing work for the specified rcu_data |
2673 | * structures. This may be called only from the CPU to whom the rdp | |
2674 | * belongs. | |
64db4cff | 2675 | */ |
b049fdf8 | 2676 | static __latent_entropy void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff PM |
2677 | { |
2678 | unsigned long flags; | |
da1df50d | 2679 | struct rcu_data *rdp = raw_cpu_ptr(&rcu_data); |
26d950a9 | 2680 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 2681 | |
b049fdf8 PM |
2682 | if (cpu_is_offline(smp_processor_id())) |
2683 | return; | |
2684 | trace_rcu_utilization(TPS("Start RCU core")); | |
50dc7def | 2685 | WARN_ON_ONCE(!rdp->beenonline); |
2e597558 | 2686 | |
3e310098 PM |
2687 | /* Report any deferred quiescent states if preemption enabled. */ |
2688 | if (!(preempt_count() & PREEMPT_MASK)) | |
2689 | rcu_preempt_deferred_qs(current); | |
2690 | else if (rcu_preempt_need_deferred_qs(current)) | |
2691 | resched_cpu(rdp->cpu); /* Provoke future context switch. */ | |
2692 | ||
64db4cff | 2693 | /* Update RCU state based on any recent quiescent states. */ |
8087d3e3 | 2694 | rcu_check_quiescent_state(rdp); |
64db4cff | 2695 | |
bd7af846 | 2696 | /* No grace period and unregistered callbacks? */ |
de8e8730 | 2697 | if (!rcu_gp_in_progress() && |
bd7af846 PM |
2698 | rcu_segcblist_is_enabled(&rdp->cblist)) { |
2699 | local_irq_save(flags); | |
e44e73ca | 2700 | if (!rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
c6e09b97 | 2701 | rcu_accelerate_cbs_unlocked(rnp, rdp); |
e44e73ca | 2702 | local_irq_restore(flags); |
64db4cff PM |
2703 | } |
2704 | ||
b96f9dc4 | 2705 | rcu_check_gp_start_stall(rnp, rdp); |
26d950a9 | 2706 | |
64db4cff | 2707 | /* If there are callbacks ready, invoke them. */ |
15fecf89 | 2708 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
aff4e9ed | 2709 | invoke_rcu_callbacks(rdp); |
96d3fd0d PM |
2710 | |
2711 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
2712 | do_nocb_deferred_wakeup(rdp); | |
f7f7bac9 | 2713 | trace_rcu_utilization(TPS("End RCU core")); |
64db4cff PM |
2714 | } |
2715 | ||
a26ac245 | 2716 | /* |
e0f23060 PM |
2717 | * Schedule RCU callback invocation. If the specified type of RCU |
2718 | * does not support RCU priority boosting, just do a direct call, | |
2719 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
924df8a0 | 2720 | * are running on the current CPU with softirqs disabled, the |
e0f23060 | 2721 | * rcu_cpu_kthread_task cannot disappear out from under us. |
a26ac245 | 2722 | */ |
aff4e9ed | 2723 | static void invoke_rcu_callbacks(struct rcu_data *rdp) |
a26ac245 | 2724 | { |
7d0ae808 | 2725 | if (unlikely(!READ_ONCE(rcu_scheduler_fully_active))) |
b0d30417 | 2726 | return; |
3c779dfe | 2727 | if (likely(!rcu_state.boost)) { |
5bb5d09c | 2728 | rcu_do_batch(rdp); |
a26ac245 PM |
2729 | return; |
2730 | } | |
a46e0899 | 2731 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
2732 | } |
2733 | ||
a46e0899 | 2734 | static void invoke_rcu_core(void) |
09223371 | 2735 | { |
b0f74036 PM |
2736 | if (cpu_online(smp_processor_id())) |
2737 | raise_softirq(RCU_SOFTIRQ); | |
09223371 SL |
2738 | } |
2739 | ||
29154c57 PM |
2740 | /* |
2741 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
2742 | */ | |
5c7d8967 PM |
2743 | static void __call_rcu_core(struct rcu_data *rdp, struct rcu_head *head, |
2744 | unsigned long flags) | |
64db4cff | 2745 | { |
62fde6ed PM |
2746 | /* |
2747 | * If called from an extended quiescent state, invoke the RCU | |
2748 | * core in order to force a re-evaluation of RCU's idleness. | |
2749 | */ | |
9910affa | 2750 | if (!rcu_is_watching()) |
62fde6ed PM |
2751 | invoke_rcu_core(); |
2752 | ||
a16b7a69 | 2753 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 2754 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 2755 | return; |
64db4cff | 2756 | |
37c72e56 PM |
2757 | /* |
2758 | * Force the grace period if too many callbacks or too long waiting. | |
2759 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
2760 | * if some other CPU has recently done so. Also, don't bother | |
2761 | * invoking force_quiescent_state() if the newly enqueued callback | |
2762 | * is the only one waiting for a grace period to complete. | |
2763 | */ | |
15fecf89 PM |
2764 | if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) > |
2765 | rdp->qlen_last_fqs_check + qhimark)) { | |
b52573d2 PM |
2766 | |
2767 | /* Are we ignoring a completed grace period? */ | |
15cabdff | 2768 | note_gp_changes(rdp); |
b52573d2 PM |
2769 | |
2770 | /* Start a new grace period if one not already started. */ | |
de8e8730 | 2771 | if (!rcu_gp_in_progress()) { |
c6e09b97 | 2772 | rcu_accelerate_cbs_unlocked(rdp->mynode, rdp); |
b52573d2 PM |
2773 | } else { |
2774 | /* Give the grace period a kick. */ | |
2775 | rdp->blimit = LONG_MAX; | |
5c7d8967 | 2776 | if (rcu_state.n_force_qs == rdp->n_force_qs_snap && |
15fecf89 | 2777 | rcu_segcblist_first_pend_cb(&rdp->cblist) != head) |
e9ecb780 | 2778 | force_quiescent_state(); |
5c7d8967 | 2779 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
15fecf89 | 2780 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); |
b52573d2 | 2781 | } |
4cdfc175 | 2782 | } |
29154c57 PM |
2783 | } |
2784 | ||
ae150184 PM |
2785 | /* |
2786 | * RCU callback function to leak a callback. | |
2787 | */ | |
2788 | static void rcu_leak_callback(struct rcu_head *rhp) | |
2789 | { | |
2790 | } | |
2791 | ||
3fbfbf7a PM |
2792 | /* |
2793 | * Helper function for call_rcu() and friends. The cpu argument will | |
2794 | * normally be -1, indicating "currently running CPU". It may specify | |
2795 | * a CPU only if that CPU is a no-CBs CPU. Currently, only _rcu_barrier() | |
2796 | * is expected to specify a CPU. | |
2797 | */ | |
64db4cff | 2798 | static void |
5c7d8967 | 2799 | __call_rcu(struct rcu_head *head, rcu_callback_t func, int cpu, bool lazy) |
64db4cff PM |
2800 | { |
2801 | unsigned long flags; | |
2802 | struct rcu_data *rdp; | |
2803 | ||
b8f2ed53 PM |
2804 | /* Misaligned rcu_head! */ |
2805 | WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1)); | |
2806 | ||
ae150184 | 2807 | if (debug_rcu_head_queue(head)) { |
fa3c6647 PM |
2808 | /* |
2809 | * Probable double call_rcu(), so leak the callback. | |
2810 | * Use rcu:rcu_callback trace event to find the previous | |
2811 | * time callback was passed to __call_rcu(). | |
2812 | */ | |
2813 | WARN_ONCE(1, "__call_rcu(): Double-freed CB %p->%pF()!!!\n", | |
2814 | head, head->func); | |
7d0ae808 | 2815 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
2816 | return; |
2817 | } | |
64db4cff PM |
2818 | head->func = func; |
2819 | head->next = NULL; | |
64db4cff | 2820 | local_irq_save(flags); |
da1df50d | 2821 | rdp = this_cpu_ptr(&rcu_data); |
64db4cff PM |
2822 | |
2823 | /* Add the callback to our list. */ | |
15fecf89 | 2824 | if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist)) || cpu != -1) { |
3fbfbf7a PM |
2825 | int offline; |
2826 | ||
2827 | if (cpu != -1) | |
da1df50d | 2828 | rdp = per_cpu_ptr(&rcu_data, cpu); |
143da9c2 PM |
2829 | if (likely(rdp->mynode)) { |
2830 | /* Post-boot, so this should be for a no-CBs CPU. */ | |
2831 | offline = !__call_rcu_nocb(rdp, head, lazy, flags); | |
2832 | WARN_ON_ONCE(offline); | |
2833 | /* Offline CPU, _call_rcu() illegal, leak callback. */ | |
2834 | local_irq_restore(flags); | |
2835 | return; | |
2836 | } | |
2837 | /* | |
2838 | * Very early boot, before rcu_init(). Initialize if needed | |
2839 | * and then drop through to queue the callback. | |
2840 | */ | |
2841 | BUG_ON(cpu != -1); | |
34404ca8 | 2842 | WARN_ON_ONCE(!rcu_is_watching()); |
15fecf89 PM |
2843 | if (rcu_segcblist_empty(&rdp->cblist)) |
2844 | rcu_segcblist_init(&rdp->cblist); | |
0d8ee37e | 2845 | } |
15fecf89 PM |
2846 | rcu_segcblist_enqueue(&rdp->cblist, head, lazy); |
2847 | if (!lazy) | |
c57afe80 | 2848 | rcu_idle_count_callbacks_posted(); |
2655d57e | 2849 | |
d4c08f2a | 2850 | if (__is_kfree_rcu_offset((unsigned long)func)) |
3c779dfe PM |
2851 | trace_rcu_kfree_callback(rcu_state.name, head, |
2852 | (unsigned long)func, | |
15fecf89 PM |
2853 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2854 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2855 | else |
3c779dfe | 2856 | trace_rcu_callback(rcu_state.name, head, |
15fecf89 PM |
2857 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2858 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2859 | |
29154c57 | 2860 | /* Go handle any RCU core processing required. */ |
5c7d8967 | 2861 | __call_rcu_core(rdp, head, flags); |
64db4cff PM |
2862 | local_irq_restore(flags); |
2863 | } | |
2864 | ||
a68a2bb2 | 2865 | /** |
45975c7d | 2866 | * call_rcu() - Queue an RCU callback for invocation after a grace period. |
a68a2bb2 PM |
2867 | * @head: structure to be used for queueing the RCU updates. |
2868 | * @func: actual callback function to be invoked after the grace period | |
2869 | * | |
2870 | * The callback function will be invoked some time after a full grace | |
45975c7d PM |
2871 | * period elapses, in other words after all pre-existing RCU read-side |
2872 | * critical sections have completed. However, the callback function | |
2873 | * might well execute concurrently with RCU read-side critical sections | |
2874 | * that started after call_rcu() was invoked. RCU read-side critical | |
2875 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), and | |
2876 | * may be nested. In addition, regions of code across which interrupts, | |
2877 | * preemption, or softirqs have been disabled also serve as RCU read-side | |
2878 | * critical sections. This includes hardware interrupt handlers, softirq | |
2879 | * handlers, and NMI handlers. | |
2880 | * | |
2881 | * Note that all CPUs must agree that the grace period extended beyond | |
2882 | * all pre-existing RCU read-side critical section. On systems with more | |
2883 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
2884 | * guaranteed to have executed a full memory barrier since the end of its | |
2885 | * last RCU read-side critical section whose beginning preceded the call | |
2886 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
2887 | * critical section that continues beyond the start of "func()" must have | |
2888 | * executed a memory barrier after the call_rcu() but before the beginning | |
2889 | * of that RCU read-side critical section. Note that these guarantees | |
2890 | * include CPUs that are offline, idle, or executing in user mode, as | |
2891 | * well as CPUs that are executing in the kernel. | |
2892 | * | |
2893 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
2894 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
2895 | * guaranteed to execute a full memory barrier during the time interval | |
2896 | * between the call to call_rcu() and the invocation of "func()" -- even | |
2897 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
2898 | * more than one CPU). | |
2899 | */ | |
2900 | void call_rcu(struct rcu_head *head, rcu_callback_t func) | |
2901 | { | |
5c7d8967 | 2902 | __call_rcu(head, func, -1, 0); |
45975c7d PM |
2903 | } |
2904 | EXPORT_SYMBOL_GPL(call_rcu); | |
2905 | ||
495aa969 ACB |
2906 | /* |
2907 | * Queue an RCU callback for lazy invocation after a grace period. | |
2908 | * This will likely be later named something like "call_rcu_lazy()", | |
2909 | * but this change will require some way of tagging the lazy RCU | |
2910 | * callbacks in the list of pending callbacks. Until then, this | |
2911 | * function may only be called from __kfree_rcu(). | |
2912 | */ | |
98ece508 | 2913 | void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func) |
495aa969 | 2914 | { |
5c7d8967 | 2915 | __call_rcu(head, func, -1, 1); |
495aa969 ACB |
2916 | } |
2917 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
2918 | ||
765a3f4f PM |
2919 | /** |
2920 | * get_state_synchronize_rcu - Snapshot current RCU state | |
2921 | * | |
2922 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
2923 | * to determine whether or not a full grace period has elapsed in the | |
2924 | * meantime. | |
2925 | */ | |
2926 | unsigned long get_state_synchronize_rcu(void) | |
2927 | { | |
2928 | /* | |
2929 | * Any prior manipulation of RCU-protected data must happen | |
e4be81a2 | 2930 | * before the load from ->gp_seq. |
765a3f4f PM |
2931 | */ |
2932 | smp_mb(); /* ^^^ */ | |
16fc9c60 | 2933 | return rcu_seq_snap(&rcu_state.gp_seq); |
765a3f4f PM |
2934 | } |
2935 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
2936 | ||
2937 | /** | |
2938 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
2939 | * | |
2940 | * @oldstate: return value from earlier call to get_state_synchronize_rcu() | |
2941 | * | |
2942 | * If a full RCU grace period has elapsed since the earlier call to | |
2943 | * get_state_synchronize_rcu(), just return. Otherwise, invoke | |
2944 | * synchronize_rcu() to wait for a full grace period. | |
2945 | * | |
2946 | * Yes, this function does not take counter wrap into account. But | |
2947 | * counter wrap is harmless. If the counter wraps, we have waited for | |
2948 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
2949 | * so waiting for one additional grace period should be just fine. | |
2950 | */ | |
2951 | void cond_synchronize_rcu(unsigned long oldstate) | |
2952 | { | |
16fc9c60 | 2953 | if (!rcu_seq_done(&rcu_state.gp_seq, oldstate)) |
765a3f4f | 2954 | synchronize_rcu(); |
e4be81a2 PM |
2955 | else |
2956 | smp_mb(); /* Ensure GP ends before subsequent accesses. */ | |
765a3f4f PM |
2957 | } |
2958 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
2959 | ||
64db4cff | 2960 | /* |
98ece508 PM |
2961 | * Check to see if there is any immediate RCU-related work to be done by |
2962 | * the current CPU, for the specified type of RCU, returning 1 if so and | |
2963 | * zero otherwise. The checks are in order of increasing expense: checks | |
2964 | * that can be carried out against CPU-local state are performed first. | |
2965 | * However, we must check for CPU stalls first, else we might not get | |
2966 | * a chance. | |
64db4cff | 2967 | */ |
98ece508 | 2968 | static int rcu_pending(void) |
64db4cff | 2969 | { |
98ece508 | 2970 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
2f51f988 PM |
2971 | struct rcu_node *rnp = rdp->mynode; |
2972 | ||
64db4cff | 2973 | /* Check for CPU stalls, if enabled. */ |
ea12ff2b | 2974 | check_cpu_stall(rdp); |
64db4cff | 2975 | |
a096932f | 2976 | /* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */ |
4580b054 | 2977 | if (rcu_nohz_full_cpu()) |
a096932f PM |
2978 | return 0; |
2979 | ||
64db4cff | 2980 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
01c495f7 | 2981 | if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2982 | return 1; |
2983 | ||
2984 | /* Does this CPU have callbacks ready to invoke? */ | |
01c495f7 | 2985 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
64db4cff PM |
2986 | return 1; |
2987 | ||
2988 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
de8e8730 | 2989 | if (!rcu_gp_in_progress() && |
c1935209 PM |
2990 | rcu_segcblist_is_enabled(&rdp->cblist) && |
2991 | !rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) | |
64db4cff PM |
2992 | return 1; |
2993 | ||
67e14c1e PM |
2994 | /* Have RCU grace period completed or started? */ |
2995 | if (rcu_seq_current(&rnp->gp_seq) != rdp->gp_seq || | |
01c495f7 | 2996 | unlikely(READ_ONCE(rdp->gpwrap))) /* outside lock */ |
64db4cff PM |
2997 | return 1; |
2998 | ||
96d3fd0d | 2999 | /* Does this CPU need a deferred NOCB wakeup? */ |
01c495f7 | 3000 | if (rcu_nocb_need_deferred_wakeup(rdp)) |
96d3fd0d | 3001 | return 1; |
96d3fd0d | 3002 | |
64db4cff PM |
3003 | /* nothing to do */ |
3004 | return 0; | |
3005 | } | |
3006 | ||
64db4cff | 3007 | /* |
c0f4dfd4 PM |
3008 | * Return true if the specified CPU has any callback. If all_lazy is |
3009 | * non-NULL, store an indication of whether all callbacks are lazy. | |
3010 | * (If there are no callbacks, all of them are deemed to be lazy.) | |
64db4cff | 3011 | */ |
51fbb910 | 3012 | static bool rcu_cpu_has_callbacks(bool *all_lazy) |
64db4cff | 3013 | { |
c0f4dfd4 PM |
3014 | bool al = true; |
3015 | bool hc = false; | |
3016 | struct rcu_data *rdp; | |
6ce75a23 | 3017 | |
b97d23c5 PM |
3018 | rdp = this_cpu_ptr(&rcu_data); |
3019 | if (!rcu_segcblist_empty(&rdp->cblist)) { | |
69c8d28c | 3020 | hc = true; |
b97d23c5 | 3021 | if (rcu_segcblist_n_nonlazy_cbs(&rdp->cblist)) |
c0f4dfd4 | 3022 | al = false; |
c0f4dfd4 PM |
3023 | } |
3024 | if (all_lazy) | |
3025 | *all_lazy = al; | |
3026 | return hc; | |
64db4cff PM |
3027 | } |
3028 | ||
a83eff0a PM |
3029 | /* |
3030 | * Helper function for _rcu_barrier() tracing. If tracing is disabled, | |
3031 | * the compiler is expected to optimize this away. | |
3032 | */ | |
8344b871 | 3033 | static void _rcu_barrier_trace(const char *s, int cpu, unsigned long done) |
a83eff0a | 3034 | { |
8344b871 PM |
3035 | trace_rcu_barrier(rcu_state.name, s, cpu, |
3036 | atomic_read(&rcu_state.barrier_cpu_count), done); | |
a83eff0a PM |
3037 | } |
3038 | ||
b1420f1c PM |
3039 | /* |
3040 | * RCU callback function for _rcu_barrier(). If we are last, wake | |
3041 | * up the task executing _rcu_barrier(). | |
3042 | */ | |
24ebbca8 | 3043 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 3044 | { |
ec9f5835 PM |
3045 | if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) { |
3046 | _rcu_barrier_trace(TPS("LastCB"), -1, | |
3047 | rcu_state.barrier_sequence); | |
3048 | complete(&rcu_state.barrier_completion); | |
a83eff0a | 3049 | } else { |
ec9f5835 | 3050 | _rcu_barrier_trace(TPS("CB"), -1, rcu_state.barrier_sequence); |
a83eff0a | 3051 | } |
d0ec774c PM |
3052 | } |
3053 | ||
3054 | /* | |
3055 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
3056 | */ | |
ec9f5835 | 3057 | static void rcu_barrier_func(void *unused) |
d0ec774c | 3058 | { |
da1df50d | 3059 | struct rcu_data *rdp = raw_cpu_ptr(&rcu_data); |
d0ec774c | 3060 | |
ec9f5835 | 3061 | _rcu_barrier_trace(TPS("IRQ"), -1, rcu_state.barrier_sequence); |
f92c734f PM |
3062 | rdp->barrier_head.func = rcu_barrier_callback; |
3063 | debug_rcu_head_queue(&rdp->barrier_head); | |
3064 | if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head, 0)) { | |
ec9f5835 | 3065 | atomic_inc(&rcu_state.barrier_cpu_count); |
f92c734f PM |
3066 | } else { |
3067 | debug_rcu_head_unqueue(&rdp->barrier_head); | |
ec9f5835 PM |
3068 | _rcu_barrier_trace(TPS("IRQNQ"), -1, |
3069 | rcu_state.barrier_sequence); | |
f92c734f | 3070 | } |
d0ec774c PM |
3071 | } |
3072 | ||
d0ec774c PM |
3073 | /* |
3074 | * Orchestrate the specified type of RCU barrier, waiting for all | |
3075 | * RCU callbacks of the specified type to complete. | |
3076 | */ | |
8344b871 | 3077 | static void _rcu_barrier(void) |
d0ec774c | 3078 | { |
b1420f1c | 3079 | int cpu; |
b1420f1c | 3080 | struct rcu_data *rdp; |
ec9f5835 | 3081 | unsigned long s = rcu_seq_snap(&rcu_state.barrier_sequence); |
b1420f1c | 3082 | |
8344b871 | 3083 | _rcu_barrier_trace(TPS("Begin"), -1, s); |
b1420f1c | 3084 | |
e74f4c45 | 3085 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
ec9f5835 | 3086 | mutex_lock(&rcu_state.barrier_mutex); |
b1420f1c | 3087 | |
4f525a52 | 3088 | /* Did someone else do our work for us? */ |
ec9f5835 PM |
3089 | if (rcu_seq_done(&rcu_state.barrier_sequence, s)) { |
3090 | _rcu_barrier_trace(TPS("EarlyExit"), -1, | |
3091 | rcu_state.barrier_sequence); | |
cf3a9c48 | 3092 | smp_mb(); /* caller's subsequent code after above check. */ |
ec9f5835 | 3093 | mutex_unlock(&rcu_state.barrier_mutex); |
cf3a9c48 PM |
3094 | return; |
3095 | } | |
3096 | ||
4f525a52 | 3097 | /* Mark the start of the barrier operation. */ |
ec9f5835 PM |
3098 | rcu_seq_start(&rcu_state.barrier_sequence); |
3099 | _rcu_barrier_trace(TPS("Inc1"), -1, rcu_state.barrier_sequence); | |
b1420f1c | 3100 | |
d0ec774c | 3101 | /* |
b1420f1c PM |
3102 | * Initialize the count to one rather than to zero in order to |
3103 | * avoid a too-soon return to zero in case of a short grace period | |
1331e7a1 PM |
3104 | * (or preemption of this task). Exclude CPU-hotplug operations |
3105 | * to ensure that no offline CPU has callbacks queued. | |
d0ec774c | 3106 | */ |
ec9f5835 PM |
3107 | init_completion(&rcu_state.barrier_completion); |
3108 | atomic_set(&rcu_state.barrier_cpu_count, 1); | |
1331e7a1 | 3109 | get_online_cpus(); |
b1420f1c PM |
3110 | |
3111 | /* | |
1331e7a1 PM |
3112 | * Force each CPU with callbacks to register a new callback. |
3113 | * When that callback is invoked, we will know that all of the | |
3114 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 3115 | */ |
3fbfbf7a | 3116 | for_each_possible_cpu(cpu) { |
d1e43fa5 | 3117 | if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu)) |
3fbfbf7a | 3118 | continue; |
da1df50d | 3119 | rdp = per_cpu_ptr(&rcu_data, cpu); |
d1e43fa5 | 3120 | if (rcu_is_nocb_cpu(cpu)) { |
4580b054 | 3121 | if (!rcu_nocb_cpu_needs_barrier(cpu)) { |
8344b871 | 3122 | _rcu_barrier_trace(TPS("OfflineNoCB"), cpu, |
ec9f5835 | 3123 | rcu_state.barrier_sequence); |
d7e29933 | 3124 | } else { |
8344b871 | 3125 | _rcu_barrier_trace(TPS("OnlineNoCB"), cpu, |
ec9f5835 | 3126 | rcu_state.barrier_sequence); |
41050a00 | 3127 | smp_mb__before_atomic(); |
ec9f5835 | 3128 | atomic_inc(&rcu_state.barrier_cpu_count); |
d7e29933 | 3129 | __call_rcu(&rdp->barrier_head, |
5c7d8967 | 3130 | rcu_barrier_callback, cpu, 0); |
d7e29933 | 3131 | } |
15fecf89 | 3132 | } else if (rcu_segcblist_n_cbs(&rdp->cblist)) { |
8344b871 | 3133 | _rcu_barrier_trace(TPS("OnlineQ"), cpu, |
ec9f5835 PM |
3134 | rcu_state.barrier_sequence); |
3135 | smp_call_function_single(cpu, rcu_barrier_func, NULL, 1); | |
b1420f1c | 3136 | } else { |
8344b871 | 3137 | _rcu_barrier_trace(TPS("OnlineNQ"), cpu, |
ec9f5835 | 3138 | rcu_state.barrier_sequence); |
b1420f1c PM |
3139 | } |
3140 | } | |
1331e7a1 | 3141 | put_online_cpus(); |
b1420f1c PM |
3142 | |
3143 | /* | |
3144 | * Now that we have an rcu_barrier_callback() callback on each | |
3145 | * CPU, and thus each counted, remove the initial count. | |
3146 | */ | |
ec9f5835 PM |
3147 | if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) |
3148 | complete(&rcu_state.barrier_completion); | |
b1420f1c PM |
3149 | |
3150 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
ec9f5835 | 3151 | wait_for_completion(&rcu_state.barrier_completion); |
b1420f1c | 3152 | |
4f525a52 | 3153 | /* Mark the end of the barrier operation. */ |
ec9f5835 PM |
3154 | _rcu_barrier_trace(TPS("Inc2"), -1, rcu_state.barrier_sequence); |
3155 | rcu_seq_end(&rcu_state.barrier_sequence); | |
4f525a52 | 3156 | |
b1420f1c | 3157 | /* Other rcu_barrier() invocations can now safely proceed. */ |
ec9f5835 | 3158 | mutex_unlock(&rcu_state.barrier_mutex); |
d0ec774c | 3159 | } |
d0ec774c | 3160 | |
45975c7d PM |
3161 | /** |
3162 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
3163 | * | |
3164 | * Note that this primitive does not necessarily wait for an RCU grace period | |
3165 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
3166 | * in the system, then rcu_barrier() is within its rights to return | |
3167 | * immediately, without waiting for anything, much less an RCU grace period. | |
3168 | */ | |
3169 | void rcu_barrier(void) | |
3170 | { | |
8344b871 | 3171 | _rcu_barrier(); |
45975c7d PM |
3172 | } |
3173 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
3174 | ||
0aa04b05 PM |
3175 | /* |
3176 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
3177 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
3178 | * must hold the corresponding leaf rcu_node ->lock with interrrupts | |
3179 | * disabled. | |
3180 | */ | |
3181 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
3182 | { | |
3183 | long mask; | |
8d672fa6 | 3184 | long oldmask; |
0aa04b05 PM |
3185 | struct rcu_node *rnp = rnp_leaf; |
3186 | ||
8d672fa6 | 3187 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
962aff03 | 3188 | WARN_ON_ONCE(rnp->wait_blkd_tasks); |
0aa04b05 PM |
3189 | for (;;) { |
3190 | mask = rnp->grpmask; | |
3191 | rnp = rnp->parent; | |
3192 | if (rnp == NULL) | |
3193 | return; | |
6cf10081 | 3194 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
8d672fa6 | 3195 | oldmask = rnp->qsmaskinit; |
0aa04b05 | 3196 | rnp->qsmaskinit |= mask; |
67c583a7 | 3197 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
8d672fa6 PM |
3198 | if (oldmask) |
3199 | return; | |
0aa04b05 PM |
3200 | } |
3201 | } | |
3202 | ||
64db4cff | 3203 | /* |
27569620 | 3204 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 3205 | */ |
27569620 | 3206 | static void __init |
53b46303 | 3207 | rcu_boot_init_percpu_data(int cpu) |
64db4cff | 3208 | { |
da1df50d | 3209 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27569620 PM |
3210 | |
3211 | /* Set up local state, ensuring consistent view of global state. */ | |
bc75e999 | 3212 | rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu); |
27569620 | 3213 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
51a1fd30 | 3214 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != 1); |
02a5c550 | 3215 | WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp->dynticks))); |
53b46303 | 3216 | rdp->rcu_ofl_gp_seq = rcu_state.gp_seq; |
57738942 | 3217 | rdp->rcu_ofl_gp_flags = RCU_GP_CLEANED; |
53b46303 | 3218 | rdp->rcu_onl_gp_seq = rcu_state.gp_seq; |
57738942 | 3219 | rdp->rcu_onl_gp_flags = RCU_GP_CLEANED; |
27569620 | 3220 | rdp->cpu = cpu; |
3fbfbf7a | 3221 | rcu_boot_init_nocb_percpu_data(rdp); |
27569620 PM |
3222 | } |
3223 | ||
3224 | /* | |
53b46303 PM |
3225 | * Invoked early in the CPU-online process, when pretty much all services |
3226 | * are available. The incoming CPU is not present. | |
3227 | * | |
3228 | * Initializes a CPU's per-CPU RCU data. Note that only one online or | |
ff3bb6f4 PM |
3229 | * offline event can be happening at a given time. Note also that we can |
3230 | * accept some slop in the rsp->gp_seq access due to the fact that this | |
3231 | * CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 3232 | */ |
53b46303 | 3233 | int rcutree_prepare_cpu(unsigned int cpu) |
64db4cff PM |
3234 | { |
3235 | unsigned long flags; | |
da1df50d | 3236 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
336a4f6c | 3237 | struct rcu_node *rnp = rcu_get_root(); |
64db4cff PM |
3238 | |
3239 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 3240 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 | 3241 | rdp->qlen_last_fqs_check = 0; |
53b46303 | 3242 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
64db4cff | 3243 | rdp->blimit = blimit; |
15fecf89 PM |
3244 | if (rcu_segcblist_empty(&rdp->cblist) && /* No early-boot CBs? */ |
3245 | !init_nocb_callback_list(rdp)) | |
3246 | rcu_segcblist_init(&rdp->cblist); /* Re-enable callbacks. */ | |
2342172f | 3247 | rdp->dynticks->dynticks_nesting = 1; /* CPU not up, no tearing. */ |
2625d469 | 3248 | rcu_dynticks_eqs_online(); |
67c583a7 | 3249 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
64db4cff | 3250 | |
0aa04b05 PM |
3251 | /* |
3252 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
3253 | * propagation up the rcu_node tree will happen at the beginning | |
3254 | * of the next grace period. | |
3255 | */ | |
64db4cff | 3256 | rnp = rdp->mynode; |
2a67e741 | 3257 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
b9585e94 | 3258 | rdp->beenonline = true; /* We have now been online. */ |
de30ad51 | 3259 | rdp->gp_seq = rnp->gp_seq; |
7a1d0f23 | 3260 | rdp->gp_seq_needed = rnp->gp_seq; |
5b74c458 | 3261 | rdp->cpu_no_qs.b.norm = true; |
9577df9a | 3262 | rdp->rcu_qs_ctr_snap = per_cpu(rcu_dynticks.rcu_qs_ctr, cpu); |
97c668b8 | 3263 | rdp->core_needs_qs = false; |
9b9500da | 3264 | rdp->rcu_iw_pending = false; |
8aa670cd | 3265 | rdp->rcu_iw_gp_seq = rnp->gp_seq - 1; |
53b46303 | 3266 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuonl")); |
67c583a7 | 3267 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
4df83742 TG |
3268 | rcu_prepare_kthreads(cpu); |
3269 | rcu_spawn_all_nocb_kthreads(cpu); | |
3270 | ||
3271 | return 0; | |
3272 | } | |
3273 | ||
deb34f36 PM |
3274 | /* |
3275 | * Update RCU priority boot kthread affinity for CPU-hotplug changes. | |
3276 | */ | |
4df83742 TG |
3277 | static void rcutree_affinity_setting(unsigned int cpu, int outgoing) |
3278 | { | |
da1df50d | 3279 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
4df83742 TG |
3280 | |
3281 | rcu_boost_kthread_setaffinity(rdp->mynode, outgoing); | |
3282 | } | |
3283 | ||
deb34f36 PM |
3284 | /* |
3285 | * Near the end of the CPU-online process. Pretty much all services | |
3286 | * enabled, and the CPU is now very much alive. | |
3287 | */ | |
4df83742 TG |
3288 | int rcutree_online_cpu(unsigned int cpu) |
3289 | { | |
9b9500da PM |
3290 | unsigned long flags; |
3291 | struct rcu_data *rdp; | |
3292 | struct rcu_node *rnp; | |
9b9500da | 3293 | |
b97d23c5 PM |
3294 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3295 | rnp = rdp->mynode; | |
3296 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3297 | rnp->ffmask |= rdp->grpmask; | |
3298 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
da915ad5 PM |
3299 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3300 | srcu_online_cpu(cpu); | |
9b9500da PM |
3301 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
3302 | return 0; /* Too early in boot for scheduler work. */ | |
3303 | sync_sched_exp_online_cleanup(cpu); | |
3304 | rcutree_affinity_setting(cpu, -1); | |
4df83742 TG |
3305 | return 0; |
3306 | } | |
3307 | ||
deb34f36 PM |
3308 | /* |
3309 | * Near the beginning of the process. The CPU is still very much alive | |
3310 | * with pretty much all services enabled. | |
3311 | */ | |
4df83742 TG |
3312 | int rcutree_offline_cpu(unsigned int cpu) |
3313 | { | |
9b9500da PM |
3314 | unsigned long flags; |
3315 | struct rcu_data *rdp; | |
3316 | struct rcu_node *rnp; | |
9b9500da | 3317 | |
b97d23c5 PM |
3318 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3319 | rnp = rdp->mynode; | |
3320 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3321 | rnp->ffmask &= ~rdp->grpmask; | |
3322 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
9b9500da | 3323 | |
4df83742 | 3324 | rcutree_affinity_setting(cpu, cpu); |
da915ad5 PM |
3325 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3326 | srcu_offline_cpu(cpu); | |
4df83742 TG |
3327 | return 0; |
3328 | } | |
3329 | ||
f64c6013 PZ |
3330 | static DEFINE_PER_CPU(int, rcu_cpu_started); |
3331 | ||
7ec99de3 PM |
3332 | /* |
3333 | * Mark the specified CPU as being online so that subsequent grace periods | |
3334 | * (both expedited and normal) will wait on it. Note that this means that | |
3335 | * incoming CPUs are not allowed to use RCU read-side critical sections | |
3336 | * until this function is called. Failing to observe this restriction | |
3337 | * will result in lockdep splats. | |
deb34f36 PM |
3338 | * |
3339 | * Note that this function is special in that it is invoked directly | |
3340 | * from the incoming CPU rather than from the cpuhp_step mechanism. | |
3341 | * This is because this function must be invoked at a precise location. | |
7ec99de3 PM |
3342 | */ |
3343 | void rcu_cpu_starting(unsigned int cpu) | |
3344 | { | |
3345 | unsigned long flags; | |
3346 | unsigned long mask; | |
313517fc PM |
3347 | int nbits; |
3348 | unsigned long oldmask; | |
7ec99de3 PM |
3349 | struct rcu_data *rdp; |
3350 | struct rcu_node *rnp; | |
7ec99de3 | 3351 | |
f64c6013 PZ |
3352 | if (per_cpu(rcu_cpu_started, cpu)) |
3353 | return; | |
3354 | ||
3355 | per_cpu(rcu_cpu_started, cpu) = 1; | |
3356 | ||
b97d23c5 PM |
3357 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3358 | rnp = rdp->mynode; | |
3359 | mask = rdp->grpmask; | |
3360 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3361 | rnp->qsmaskinitnext |= mask; | |
3362 | oldmask = rnp->expmaskinitnext; | |
3363 | rnp->expmaskinitnext |= mask; | |
3364 | oldmask ^= rnp->expmaskinitnext; | |
3365 | nbits = bitmap_weight(&oldmask, BITS_PER_LONG); | |
3366 | /* Allow lockless access for expedited grace periods. */ | |
eb7a6653 | 3367 | smp_store_release(&rcu_state.ncpus, rcu_state.ncpus + nbits); /* ^^^ */ |
b97d23c5 | 3368 | rcu_gpnum_ovf(rnp, rdp); /* Offline-induced counter wrap? */ |
eb7a6653 PM |
3369 | rdp->rcu_onl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
3370 | rdp->rcu_onl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
b97d23c5 PM |
3371 | if (rnp->qsmask & mask) { /* RCU waiting on incoming CPU? */ |
3372 | /* Report QS -after- changing ->qsmaskinitnext! */ | |
3373 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); | |
3374 | } else { | |
3375 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
7ec99de3 | 3376 | } |
313517fc | 3377 | smp_mb(); /* Ensure RCU read-side usage follows above initialization. */ |
7ec99de3 PM |
3378 | } |
3379 | ||
27d50c7e TG |
3380 | #ifdef CONFIG_HOTPLUG_CPU |
3381 | /* | |
53b46303 PM |
3382 | * The outgoing function has no further need of RCU, so remove it from |
3383 | * the rcu_node tree's ->qsmaskinitnext bit masks. | |
3384 | * | |
3385 | * Note that this function is special in that it is invoked directly | |
3386 | * from the outgoing CPU rather than from the cpuhp_step mechanism. | |
3387 | * This is because this function must be invoked at a precise location. | |
27d50c7e | 3388 | */ |
53b46303 | 3389 | void rcu_report_dead(unsigned int cpu) |
27d50c7e TG |
3390 | { |
3391 | unsigned long flags; | |
3392 | unsigned long mask; | |
da1df50d | 3393 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27d50c7e TG |
3394 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
3395 | ||
53b46303 PM |
3396 | /* QS for any half-done expedited RCU-sched GP. */ |
3397 | preempt_disable(); | |
63d4c8c9 | 3398 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); |
53b46303 PM |
3399 | preempt_enable(); |
3400 | rcu_preempt_deferred_qs(current); | |
3401 | ||
27d50c7e TG |
3402 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ |
3403 | mask = rdp->grpmask; | |
53b46303 | 3404 | spin_lock(&rcu_state.ofl_lock); |
27d50c7e | 3405 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ |
53b46303 PM |
3406 | rdp->rcu_ofl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
3407 | rdp->rcu_ofl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
fece2776 PM |
3408 | if (rnp->qsmask & mask) { /* RCU waiting on outgoing CPU? */ |
3409 | /* Report quiescent state -before- changing ->qsmaskinitnext! */ | |
b50912d0 | 3410 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
fece2776 PM |
3411 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
3412 | } | |
27d50c7e | 3413 | rnp->qsmaskinitnext &= ~mask; |
710d60cb | 3414 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
53b46303 | 3415 | spin_unlock(&rcu_state.ofl_lock); |
f64c6013 PZ |
3416 | |
3417 | per_cpu(rcu_cpu_started, cpu) = 0; | |
27d50c7e | 3418 | } |
a58163d8 | 3419 | |
53b46303 PM |
3420 | /* |
3421 | * The outgoing CPU has just passed through the dying-idle state, and we | |
3422 | * are being invoked from the CPU that was IPIed to continue the offline | |
3423 | * operation. Migrate the outgoing CPU's callbacks to the current CPU. | |
3424 | */ | |
3425 | void rcutree_migrate_callbacks(int cpu) | |
a58163d8 PM |
3426 | { |
3427 | unsigned long flags; | |
b1a2d79f | 3428 | struct rcu_data *my_rdp; |
da1df50d | 3429 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
336a4f6c | 3430 | struct rcu_node *rnp_root = rcu_get_root(); |
ec4eacce | 3431 | bool needwake; |
a58163d8 | 3432 | |
95335c03 PM |
3433 | if (rcu_is_nocb_cpu(cpu) || rcu_segcblist_empty(&rdp->cblist)) |
3434 | return; /* No callbacks to migrate. */ | |
3435 | ||
b1a2d79f | 3436 | local_irq_save(flags); |
da1df50d | 3437 | my_rdp = this_cpu_ptr(&rcu_data); |
b1a2d79f PM |
3438 | if (rcu_nocb_adopt_orphan_cbs(my_rdp, rdp, flags)) { |
3439 | local_irq_restore(flags); | |
3440 | return; | |
3441 | } | |
9fa46fb8 | 3442 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ |
ec4eacce | 3443 | /* Leverage recent GPs and set GP for new callbacks. */ |
834f56bf PM |
3444 | needwake = rcu_advance_cbs(rnp_root, rdp) || |
3445 | rcu_advance_cbs(rnp_root, my_rdp); | |
f2dbe4a5 | 3446 | rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist); |
09efeeee PM |
3447 | WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) != |
3448 | !rcu_segcblist_n_cbs(&my_rdp->cblist)); | |
537b85c8 | 3449 | raw_spin_unlock_irqrestore_rcu_node(rnp_root, flags); |
ec4eacce | 3450 | if (needwake) |
532c00c9 | 3451 | rcu_gp_kthread_wake(); |
a58163d8 PM |
3452 | WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 || |
3453 | !rcu_segcblist_empty(&rdp->cblist), | |
3454 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n", | |
3455 | cpu, rcu_segcblist_n_cbs(&rdp->cblist), | |
3456 | rcu_segcblist_first_cb(&rdp->cblist)); | |
3457 | } | |
27d50c7e TG |
3458 | #endif |
3459 | ||
deb34f36 PM |
3460 | /* |
3461 | * On non-huge systems, use expedited RCU grace periods to make suspend | |
3462 | * and hibernation run faster. | |
3463 | */ | |
d1d74d14 BP |
3464 | static int rcu_pm_notify(struct notifier_block *self, |
3465 | unsigned long action, void *hcpu) | |
3466 | { | |
3467 | switch (action) { | |
3468 | case PM_HIBERNATION_PREPARE: | |
3469 | case PM_SUSPEND_PREPARE: | |
3470 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ | |
5afff48b | 3471 | rcu_expedite_gp(); |
d1d74d14 BP |
3472 | break; |
3473 | case PM_POST_HIBERNATION: | |
3474 | case PM_POST_SUSPEND: | |
5afff48b PM |
3475 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ |
3476 | rcu_unexpedite_gp(); | |
d1d74d14 BP |
3477 | break; |
3478 | default: | |
3479 | break; | |
3480 | } | |
3481 | return NOTIFY_OK; | |
3482 | } | |
3483 | ||
b3dbec76 | 3484 | /* |
9386c0b7 | 3485 | * Spawn the kthreads that handle each RCU flavor's grace periods. |
b3dbec76 PM |
3486 | */ |
3487 | static int __init rcu_spawn_gp_kthread(void) | |
3488 | { | |
3489 | unsigned long flags; | |
a94844b2 | 3490 | int kthread_prio_in = kthread_prio; |
b3dbec76 | 3491 | struct rcu_node *rnp; |
a94844b2 | 3492 | struct sched_param sp; |
b3dbec76 PM |
3493 | struct task_struct *t; |
3494 | ||
a94844b2 | 3495 | /* Force priority into range. */ |
c7cd161e JFG |
3496 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 2 |
3497 | && IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)) | |
3498 | kthread_prio = 2; | |
3499 | else if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
a94844b2 PM |
3500 | kthread_prio = 1; |
3501 | else if (kthread_prio < 0) | |
3502 | kthread_prio = 0; | |
3503 | else if (kthread_prio > 99) | |
3504 | kthread_prio = 99; | |
c7cd161e | 3505 | |
a94844b2 PM |
3506 | if (kthread_prio != kthread_prio_in) |
3507 | pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n", | |
3508 | kthread_prio, kthread_prio_in); | |
3509 | ||
9386c0b7 | 3510 | rcu_scheduler_fully_active = 1; |
b97d23c5 PM |
3511 | t = kthread_create(rcu_gp_kthread, NULL, "%s", rcu_state.name); |
3512 | BUG_ON(IS_ERR(t)); | |
3513 | rnp = rcu_get_root(); | |
3514 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3515 | rcu_state.gp_kthread = t; | |
3516 | if (kthread_prio) { | |
3517 | sp.sched_priority = kthread_prio; | |
3518 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
b3dbec76 | 3519 | } |
b97d23c5 PM |
3520 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
3521 | wake_up_process(t); | |
35ce7f29 | 3522 | rcu_spawn_nocb_kthreads(); |
9386c0b7 | 3523 | rcu_spawn_boost_kthreads(); |
b3dbec76 PM |
3524 | return 0; |
3525 | } | |
3526 | early_initcall(rcu_spawn_gp_kthread); | |
3527 | ||
bbad9379 | 3528 | /* |
52d7e48b PM |
3529 | * This function is invoked towards the end of the scheduler's |
3530 | * initialization process. Before this is called, the idle task might | |
3531 | * contain synchronous grace-period primitives (during which time, this idle | |
3532 | * task is booting the system, and such primitives are no-ops). After this | |
3533 | * function is called, any synchronous grace-period primitives are run as | |
3534 | * expedited, with the requesting task driving the grace period forward. | |
900b1028 | 3535 | * A later core_initcall() rcu_set_runtime_mode() will switch to full |
52d7e48b | 3536 | * runtime RCU functionality. |
bbad9379 PM |
3537 | */ |
3538 | void rcu_scheduler_starting(void) | |
3539 | { | |
3540 | WARN_ON(num_online_cpus() != 1); | |
3541 | WARN_ON(nr_context_switches() > 0); | |
52d7e48b PM |
3542 | rcu_test_sync_prims(); |
3543 | rcu_scheduler_active = RCU_SCHEDULER_INIT; | |
3544 | rcu_test_sync_prims(); | |
bbad9379 PM |
3545 | } |
3546 | ||
64db4cff PM |
3547 | /* |
3548 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
3549 | */ | |
b8bb1f63 | 3550 | static void __init rcu_init_one(void) |
64db4cff | 3551 | { |
cb007102 AG |
3552 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
3553 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
3554 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
3555 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
199977bf | 3556 | |
199977bf | 3557 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ |
64db4cff PM |
3558 | int cpustride = 1; |
3559 | int i; | |
3560 | int j; | |
3561 | struct rcu_node *rnp; | |
3562 | ||
05b84aec | 3563 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 3564 | |
3eaaaf6c PM |
3565 | /* Silence gcc 4.8 false positive about array index out of range. */ |
3566 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
3567 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 3568 | |
64db4cff PM |
3569 | /* Initialize the level-tracking arrays. */ |
3570 | ||
f885b7f2 | 3571 | for (i = 1; i < rcu_num_lvls; i++) |
eb7a6653 PM |
3572 | rcu_state.level[i] = |
3573 | rcu_state.level[i - 1] + num_rcu_lvl[i - 1]; | |
41f5c631 | 3574 | rcu_init_levelspread(levelspread, num_rcu_lvl); |
64db4cff PM |
3575 | |
3576 | /* Initialize the elements themselves, starting from the leaves. */ | |
3577 | ||
f885b7f2 | 3578 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 3579 | cpustride *= levelspread[i]; |
eb7a6653 | 3580 | rnp = rcu_state.level[i]; |
41f5c631 | 3581 | for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) { |
67c583a7 BF |
3582 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
3583 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 3584 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
3585 | raw_spin_lock_init(&rnp->fqslock); |
3586 | lockdep_set_class_and_name(&rnp->fqslock, | |
3587 | &rcu_fqs_class[i], fqs[i]); | |
eb7a6653 PM |
3588 | rnp->gp_seq = rcu_state.gp_seq; |
3589 | rnp->gp_seq_needed = rcu_state.gp_seq; | |
3590 | rnp->completedqs = rcu_state.gp_seq; | |
64db4cff PM |
3591 | rnp->qsmask = 0; |
3592 | rnp->qsmaskinit = 0; | |
3593 | rnp->grplo = j * cpustride; | |
3594 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
3595 | if (rnp->grphi >= nr_cpu_ids) |
3596 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
3597 | if (i == 0) { |
3598 | rnp->grpnum = 0; | |
3599 | rnp->grpmask = 0; | |
3600 | rnp->parent = NULL; | |
3601 | } else { | |
199977bf | 3602 | rnp->grpnum = j % levelspread[i - 1]; |
64db4cff | 3603 | rnp->grpmask = 1UL << rnp->grpnum; |
eb7a6653 | 3604 | rnp->parent = rcu_state.level[i - 1] + |
199977bf | 3605 | j / levelspread[i - 1]; |
64db4cff PM |
3606 | } |
3607 | rnp->level = i; | |
12f5f524 | 3608 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 3609 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
3610 | init_waitqueue_head(&rnp->exp_wq[0]); |
3611 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
3612 | init_waitqueue_head(&rnp->exp_wq[2]); |
3613 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 3614 | spin_lock_init(&rnp->exp_lock); |
64db4cff PM |
3615 | } |
3616 | } | |
0c34029a | 3617 | |
eb7a6653 PM |
3618 | init_swait_queue_head(&rcu_state.gp_wq); |
3619 | init_swait_queue_head(&rcu_state.expedited_wq); | |
aedf4ba9 | 3620 | rnp = rcu_first_leaf_node(); |
0c34029a | 3621 | for_each_possible_cpu(i) { |
4a90a068 | 3622 | while (i > rnp->grphi) |
0c34029a | 3623 | rnp++; |
da1df50d | 3624 | per_cpu_ptr(&rcu_data, i)->mynode = rnp; |
53b46303 | 3625 | rcu_boot_init_percpu_data(i); |
0c34029a | 3626 | } |
64db4cff PM |
3627 | } |
3628 | ||
f885b7f2 PM |
3629 | /* |
3630 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 3631 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
3632 | * the ->node array in the rcu_state structure. |
3633 | */ | |
3634 | static void __init rcu_init_geometry(void) | |
3635 | { | |
026ad283 | 3636 | ulong d; |
f885b7f2 | 3637 | int i; |
05b84aec | 3638 | int rcu_capacity[RCU_NUM_LVLS]; |
f885b7f2 | 3639 | |
026ad283 PM |
3640 | /* |
3641 | * Initialize any unspecified boot parameters. | |
3642 | * The default values of jiffies_till_first_fqs and | |
3643 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
3644 | * value, which is a function of HZ, then adding one for each | |
3645 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
3646 | */ | |
3647 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
3648 | if (jiffies_till_first_fqs == ULONG_MAX) | |
3649 | jiffies_till_first_fqs = d; | |
3650 | if (jiffies_till_next_fqs == ULONG_MAX) | |
3651 | jiffies_till_next_fqs = d; | |
3652 | ||
f885b7f2 | 3653 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 3654 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 3655 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 3656 | return; |
a7538352 | 3657 | pr_info("Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n", |
39479098 | 3658 | rcu_fanout_leaf, nr_cpu_ids); |
f885b7f2 | 3659 | |
f885b7f2 | 3660 | /* |
ee968ac6 PM |
3661 | * The boot-time rcu_fanout_leaf parameter must be at least two |
3662 | * and cannot exceed the number of bits in the rcu_node masks. | |
3663 | * Complain and fall back to the compile-time values if this | |
3664 | * limit is exceeded. | |
f885b7f2 | 3665 | */ |
ee968ac6 | 3666 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 3667 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 3668 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
3669 | WARN_ON(1); |
3670 | return; | |
3671 | } | |
3672 | ||
f885b7f2 PM |
3673 | /* |
3674 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 3675 | * with the given number of levels. |
f885b7f2 | 3676 | */ |
9618138b | 3677 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 3678 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 3679 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
3680 | |
3681 | /* | |
75cf15a4 | 3682 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 3683 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 3684 | */ |
ee968ac6 PM |
3685 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
3686 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
3687 | WARN_ON(1); | |
3688 | return; | |
3689 | } | |
f885b7f2 | 3690 | |
679f9858 | 3691 | /* Calculate the number of levels in the tree. */ |
9618138b | 3692 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 3693 | } |
9618138b | 3694 | rcu_num_lvls = i + 1; |
679f9858 | 3695 | |
f885b7f2 | 3696 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 3697 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 3698 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
3699 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
3700 | } | |
f885b7f2 PM |
3701 | |
3702 | /* Calculate the total number of rcu_node structures. */ | |
3703 | rcu_num_nodes = 0; | |
679f9858 | 3704 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 3705 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
3706 | } |
3707 | ||
a3dc2948 PM |
3708 | /* |
3709 | * Dump out the structure of the rcu_node combining tree associated | |
3710 | * with the rcu_state structure referenced by rsp. | |
3711 | */ | |
b8bb1f63 | 3712 | static void __init rcu_dump_rcu_node_tree(void) |
a3dc2948 PM |
3713 | { |
3714 | int level = 0; | |
3715 | struct rcu_node *rnp; | |
3716 | ||
3717 | pr_info("rcu_node tree layout dump\n"); | |
3718 | pr_info(" "); | |
aedf4ba9 | 3719 | rcu_for_each_node_breadth_first(rnp) { |
a3dc2948 PM |
3720 | if (rnp->level != level) { |
3721 | pr_cont("\n"); | |
3722 | pr_info(" "); | |
3723 | level = rnp->level; | |
3724 | } | |
3725 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
3726 | } | |
3727 | pr_cont("\n"); | |
3728 | } | |
3729 | ||
ad7c946b | 3730 | struct workqueue_struct *rcu_gp_wq; |
25f3d7ef | 3731 | struct workqueue_struct *rcu_par_gp_wq; |
ad7c946b | 3732 | |
9f680ab4 | 3733 | void __init rcu_init(void) |
64db4cff | 3734 | { |
017c4261 | 3735 | int cpu; |
9f680ab4 | 3736 | |
47627678 PM |
3737 | rcu_early_boot_tests(); |
3738 | ||
f41d911f | 3739 | rcu_bootup_announce(); |
f885b7f2 | 3740 | rcu_init_geometry(); |
b8bb1f63 | 3741 | rcu_init_one(); |
a3dc2948 | 3742 | if (dump_tree) |
b8bb1f63 | 3743 | rcu_dump_rcu_node_tree(); |
b5b39360 | 3744 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
3745 | |
3746 | /* | |
3747 | * We don't need protection against CPU-hotplug here because | |
3748 | * this is called early in boot, before either interrupts | |
3749 | * or the scheduler are operational. | |
3750 | */ | |
d1d74d14 | 3751 | pm_notifier(rcu_pm_notify, 0); |
7ec99de3 | 3752 | for_each_online_cpu(cpu) { |
4df83742 | 3753 | rcutree_prepare_cpu(cpu); |
7ec99de3 | 3754 | rcu_cpu_starting(cpu); |
9b9500da | 3755 | rcutree_online_cpu(cpu); |
7ec99de3 | 3756 | } |
ad7c946b PM |
3757 | |
3758 | /* Create workqueue for expedited GPs and for Tree SRCU. */ | |
3759 | rcu_gp_wq = alloc_workqueue("rcu_gp", WQ_MEM_RECLAIM, 0); | |
3760 | WARN_ON(!rcu_gp_wq); | |
25f3d7ef PM |
3761 | rcu_par_gp_wq = alloc_workqueue("rcu_par_gp", WQ_MEM_RECLAIM, 0); |
3762 | WARN_ON(!rcu_par_gp_wq); | |
64db4cff PM |
3763 | } |
3764 | ||
3549c2bc | 3765 | #include "tree_exp.h" |
4102adab | 3766 | #include "tree_plugin.h" |