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 PM |
29 | */ |
30 | #include <linux/types.h> | |
31 | #include <linux/kernel.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/spinlock.h> | |
34 | #include <linux/smp.h> | |
f9411ebe | 35 | #include <linux/rcupdate_wait.h> |
64db4cff PM |
36 | #include <linux/interrupt.h> |
37 | #include <linux/sched.h> | |
b17b0153 | 38 | #include <linux/sched/debug.h> |
c1dc0b9c | 39 | #include <linux/nmi.h> |
8826f3b0 | 40 | #include <linux/atomic.h> |
64db4cff | 41 | #include <linux/bitops.h> |
9984de1a | 42 | #include <linux/export.h> |
64db4cff PM |
43 | #include <linux/completion.h> |
44 | #include <linux/moduleparam.h> | |
45 | #include <linux/percpu.h> | |
46 | #include <linux/notifier.h> | |
47 | #include <linux/cpu.h> | |
48 | #include <linux/mutex.h> | |
49 | #include <linux/time.h> | |
bbad9379 | 50 | #include <linux/kernel_stat.h> |
a26ac245 PM |
51 | #include <linux/wait.h> |
52 | #include <linux/kthread.h> | |
ae7e81c0 | 53 | #include <uapi/linux/sched/types.h> |
268bb0ce | 54 | #include <linux/prefetch.h> |
3d3b7db0 PM |
55 | #include <linux/delay.h> |
56 | #include <linux/stop_machine.h> | |
661a85dc | 57 | #include <linux/random.h> |
af658dca | 58 | #include <linux/trace_events.h> |
d1d74d14 | 59 | #include <linux/suspend.h> |
a278d471 | 60 | #include <linux/ftrace.h> |
64db4cff | 61 | |
4102adab | 62 | #include "tree.h" |
29c00b4a | 63 | #include "rcu.h" |
9f77da9f | 64 | |
4102adab PM |
65 | #ifdef MODULE_PARAM_PREFIX |
66 | #undef MODULE_PARAM_PREFIX | |
67 | #endif | |
68 | #define MODULE_PARAM_PREFIX "rcutree." | |
69 | ||
64db4cff PM |
70 | /* Data structures. */ |
71 | ||
f7f7bac9 SRRH |
72 | /* |
73 | * In order to export the rcu_state name to the tracing tools, it | |
74 | * needs to be added in the __tracepoint_string section. | |
75 | * This requires defining a separate variable tp_<sname>_varname | |
76 | * that points to the string being used, and this will allow | |
77 | * the tracing userspace tools to be able to decipher the string | |
78 | * address to the matching string. | |
79 | */ | |
a8a29b3b AB |
80 | #ifdef CONFIG_TRACING |
81 | # define DEFINE_RCU_TPS(sname) \ | |
f7f7bac9 | 82 | static char sname##_varname[] = #sname; \ |
a8a29b3b AB |
83 | static const char *tp_##sname##_varname __used __tracepoint_string = sname##_varname; |
84 | # define RCU_STATE_NAME(sname) sname##_varname | |
85 | #else | |
86 | # define DEFINE_RCU_TPS(sname) | |
87 | # define RCU_STATE_NAME(sname) __stringify(sname) | |
88 | #endif | |
89 | ||
90 | #define RCU_STATE_INITIALIZER(sname, sabbr, cr) \ | |
91 | DEFINE_RCU_TPS(sname) \ | |
c92fb057 | 92 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, sname##_data); \ |
a41bfeb2 | 93 | struct rcu_state sname##_state = { \ |
6c90cc7b | 94 | .level = { &sname##_state.node[0] }, \ |
2723249a | 95 | .rda = &sname##_data, \ |
037b64ed | 96 | .call = cr, \ |
77f81fe0 | 97 | .gp_state = RCU_GP_IDLE, \ |
de30ad51 | 98 | .gp_seq = (0UL - 300UL) << RCU_SEQ_CTR_SHIFT, \ |
7be7f0be | 99 | .barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \ |
a8a29b3b | 100 | .name = RCU_STATE_NAME(sname), \ |
a4889858 | 101 | .abbr = sabbr, \ |
f6a12f34 | 102 | .exp_mutex = __MUTEX_INITIALIZER(sname##_state.exp_mutex), \ |
3b5f668e | 103 | .exp_wake_mutex = __MUTEX_INITIALIZER(sname##_state.exp_wake_mutex), \ |
2723249a | 104 | } |
64db4cff | 105 | |
a41bfeb2 SRRH |
106 | RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched); |
107 | RCU_STATE_INITIALIZER(rcu_bh, 'b', call_rcu_bh); | |
b1f77b05 | 108 | |
b28a7c01 | 109 | static struct rcu_state *const rcu_state_p; |
6ce75a23 | 110 | LIST_HEAD(rcu_struct_flavors); |
27f4d280 | 111 | |
a3dc2948 PM |
112 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
113 | static bool dump_tree; | |
114 | module_param(dump_tree, bool, 0444); | |
7fa27001 PM |
115 | /* Control rcu_node-tree auto-balancing at boot time. */ |
116 | static bool rcu_fanout_exact; | |
117 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
118 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
119 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 120 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 121 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 | 122 | /* Number of rcu_nodes at specified level. */ |
e95d68d2 | 123 | int num_rcu_lvl[] = NUM_RCU_LVL_INIT; |
f885b7f2 | 124 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
088e9d25 DBO |
125 | /* panic() on RCU Stall sysctl. */ |
126 | int sysctl_panic_on_rcu_stall __read_mostly; | |
f885b7f2 | 127 | |
b0d30417 | 128 | /* |
52d7e48b PM |
129 | * The rcu_scheduler_active variable is initialized to the value |
130 | * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the | |
131 | * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE, | |
132 | * RCU can assume that there is but one task, allowing RCU to (for example) | |
0d95092c | 133 | * optimize synchronize_rcu() to a simple barrier(). When this variable |
52d7e48b PM |
134 | * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required |
135 | * to detect real grace periods. This variable is also used to suppress | |
136 | * boot-time false positives from lockdep-RCU error checking. Finally, it | |
137 | * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU | |
138 | * is fully initialized, including all of its kthreads having been spawned. | |
b0d30417 | 139 | */ |
bbad9379 PM |
140 | int rcu_scheduler_active __read_mostly; |
141 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
142 | ||
b0d30417 PM |
143 | /* |
144 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
145 | * during the early_initcall() processing, which is after the scheduler | |
146 | * is capable of creating new tasks. So RCU processing (for example, | |
147 | * creating tasks for RCU priority boosting) must be delayed until after | |
148 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
149 | * currently delay invocation of any RCU callbacks until after this point. | |
150 | * | |
151 | * It might later prove better for people registering RCU callbacks during | |
152 | * early boot to take responsibility for these callbacks, but one step at | |
153 | * a time. | |
154 | */ | |
155 | static int rcu_scheduler_fully_active __read_mostly; | |
156 | ||
0aa04b05 PM |
157 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
158 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 159 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 PM |
160 | static void invoke_rcu_core(void); |
161 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); | |
6587a23b PM |
162 | static void rcu_report_exp_rdp(struct rcu_state *rsp, |
163 | struct rcu_data *rdp, bool wake); | |
3549c2bc | 164 | static void sync_sched_exp_online_cleanup(int cpu); |
a26ac245 | 165 | |
a94844b2 | 166 | /* rcuc/rcub kthread realtime priority */ |
26730f55 | 167 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; |
a94844b2 PM |
168 | module_param(kthread_prio, int, 0644); |
169 | ||
8d7dc928 | 170 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd | 171 | |
90040c9e PM |
172 | static int gp_preinit_delay; |
173 | module_param(gp_preinit_delay, int, 0444); | |
174 | static int gp_init_delay; | |
175 | module_param(gp_init_delay, int, 0444); | |
176 | static int gp_cleanup_delay; | |
177 | module_param(gp_cleanup_delay, int, 0444); | |
0f41c0dd | 178 | |
eab128e8 PM |
179 | /* |
180 | * Number of grace periods between delays, normalized by the duration of | |
bfd090be | 181 | * the delay. The longer the delay, the more the grace periods between |
eab128e8 PM |
182 | * each delay. The reason for this normalization is that it means that, |
183 | * for non-zero delays, the overall slowdown of grace periods is constant | |
184 | * regardless of the duration of the delay. This arrangement balances | |
185 | * the need for long delays to increase some race probabilities with the | |
186 | * need for fast grace periods to increase other race probabilities. | |
187 | */ | |
188 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */ | |
37745d28 | 189 | |
4a298656 PM |
190 | /* |
191 | * Track the rcutorture test sequence number and the update version | |
192 | * number within a given test. The rcutorture_testseq is incremented | |
193 | * on every rcutorture module load and unload, so has an odd value | |
194 | * when a test is running. The rcutorture_vernum is set to zero | |
195 | * when rcutorture starts and is incremented on each rcutorture update. | |
196 | * These variables enable correlating rcutorture output with the | |
197 | * RCU tracing information. | |
198 | */ | |
199 | unsigned long rcutorture_testseq; | |
200 | unsigned long rcutorture_vernum; | |
201 | ||
0aa04b05 PM |
202 | /* |
203 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
204 | * This will not be stable unless the rcu_node structure's ->lock is | |
205 | * held, but the bit corresponding to the current CPU will be stable | |
206 | * in most contexts. | |
207 | */ | |
208 | unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) | |
209 | { | |
7d0ae808 | 210 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
211 | } |
212 | ||
fc2219d4 | 213 | /* |
7d0ae808 | 214 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
215 | * permit this function to be invoked without holding the root rcu_node |
216 | * structure's ->lock, but of course results can be subject to change. | |
217 | */ | |
218 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
219 | { | |
ba04107f | 220 | return rcu_seq_state(rcu_seq_current(&rsp->gp_seq)); |
fc2219d4 PM |
221 | } |
222 | ||
b1f77b05 | 223 | /* |
d6714c22 | 224 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 225 | * how many quiescent states passed, just if there was at least |
d6714c22 | 226 | * one since the start of the grace period, this just sets a flag. |
e4cc1f22 | 227 | * The caller must have disabled preemption. |
b1f77b05 | 228 | */ |
284a8c93 | 229 | void rcu_sched_qs(void) |
b1f77b05 | 230 | { |
f4687d26 | 231 | RCU_LOCKDEP_WARN(preemptible(), "rcu_sched_qs() invoked with preemption enabled!!!"); |
fecbf6f0 PM |
232 | if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.s)) |
233 | return; | |
234 | trace_rcu_grace_period(TPS("rcu_sched"), | |
477351f7 | 235 | __this_cpu_read(rcu_sched_data.gp_seq), |
fecbf6f0 PM |
236 | TPS("cpuqs")); |
237 | __this_cpu_write(rcu_sched_data.cpu_no_qs.b.norm, false); | |
238 | if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.b.exp)) | |
239 | return; | |
46a5d164 PM |
240 | __this_cpu_write(rcu_sched_data.cpu_no_qs.b.exp, false); |
241 | rcu_report_exp_rdp(&rcu_sched_state, | |
242 | this_cpu_ptr(&rcu_sched_data), true); | |
b1f77b05 IM |
243 | } |
244 | ||
284a8c93 | 245 | void rcu_bh_qs(void) |
b1f77b05 | 246 | { |
f4687d26 | 247 | RCU_LOCKDEP_WARN(preemptible(), "rcu_bh_qs() invoked with preemption enabled!!!"); |
5b74c458 | 248 | if (__this_cpu_read(rcu_bh_data.cpu_no_qs.s)) { |
284a8c93 | 249 | trace_rcu_grace_period(TPS("rcu_bh"), |
477351f7 | 250 | __this_cpu_read(rcu_bh_data.gp_seq), |
284a8c93 | 251 | TPS("cpuqs")); |
5b74c458 | 252 | __this_cpu_write(rcu_bh_data.cpu_no_qs.b.norm, false); |
284a8c93 | 253 | } |
b1f77b05 | 254 | } |
64db4cff | 255 | |
b8c17e66 PM |
256 | /* |
257 | * Steal a bit from the bottom of ->dynticks for idle entry/exit | |
258 | * control. Initially this is for TLB flushing. | |
259 | */ | |
260 | #define RCU_DYNTICK_CTRL_MASK 0x1 | |
261 | #define RCU_DYNTICK_CTRL_CTR (RCU_DYNTICK_CTRL_MASK + 1) | |
262 | #ifndef rcu_eqs_special_exit | |
263 | #define rcu_eqs_special_exit() do { } while (0) | |
264 | #endif | |
4a81e832 PM |
265 | |
266 | static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { | |
51a1fd30 | 267 | .dynticks_nesting = 1, |
58721f5d | 268 | .dynticks_nmi_nesting = DYNTICK_IRQ_NONIDLE, |
b8c17e66 | 269 | .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR), |
4a81e832 PM |
270 | }; |
271 | ||
2625d469 PM |
272 | /* |
273 | * Record entry into an extended quiescent state. This is only to be | |
274 | * called when not already in an extended quiescent state. | |
275 | */ | |
276 | static void rcu_dynticks_eqs_enter(void) | |
277 | { | |
278 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 279 | int seq; |
2625d469 PM |
280 | |
281 | /* | |
b8c17e66 | 282 | * CPUs seeing atomic_add_return() must see prior RCU read-side |
2625d469 PM |
283 | * critical sections, and we also must force ordering with the |
284 | * next idle sojourn. | |
285 | */ | |
b8c17e66 PM |
286 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
287 | /* Better be in an extended quiescent state! */ | |
288 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
289 | (seq & RCU_DYNTICK_CTRL_CTR)); | |
290 | /* Better not have special action (TLB flush) pending! */ | |
291 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
292 | (seq & RCU_DYNTICK_CTRL_MASK)); | |
2625d469 PM |
293 | } |
294 | ||
295 | /* | |
296 | * Record exit from an extended quiescent state. This is only to be | |
297 | * called from an extended quiescent state. | |
298 | */ | |
299 | static void rcu_dynticks_eqs_exit(void) | |
300 | { | |
301 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 302 | int seq; |
2625d469 PM |
303 | |
304 | /* | |
b8c17e66 | 305 | * CPUs seeing atomic_add_return() must see prior idle sojourns, |
2625d469 PM |
306 | * and we also must force ordering with the next RCU read-side |
307 | * critical section. | |
308 | */ | |
b8c17e66 PM |
309 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
310 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
311 | !(seq & RCU_DYNTICK_CTRL_CTR)); | |
312 | if (seq & RCU_DYNTICK_CTRL_MASK) { | |
313 | atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdtp->dynticks); | |
314 | smp_mb__after_atomic(); /* _exit after clearing mask. */ | |
315 | /* Prefer duplicate flushes to losing a flush. */ | |
316 | rcu_eqs_special_exit(); | |
317 | } | |
2625d469 PM |
318 | } |
319 | ||
320 | /* | |
321 | * Reset the current CPU's ->dynticks counter to indicate that the | |
322 | * newly onlined CPU is no longer in an extended quiescent state. | |
323 | * This will either leave the counter unchanged, or increment it | |
324 | * to the next non-quiescent value. | |
325 | * | |
326 | * The non-atomic test/increment sequence works because the upper bits | |
327 | * of the ->dynticks counter are manipulated only by the corresponding CPU, | |
328 | * or when the corresponding CPU is offline. | |
329 | */ | |
330 | static void rcu_dynticks_eqs_online(void) | |
331 | { | |
332 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
333 | ||
b8c17e66 | 334 | if (atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR) |
2625d469 | 335 | return; |
b8c17e66 | 336 | atomic_add(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
2625d469 PM |
337 | } |
338 | ||
02a5c550 PM |
339 | /* |
340 | * Is the current CPU in an extended quiescent state? | |
341 | * | |
342 | * No ordering, as we are sampling CPU-local information. | |
343 | */ | |
344 | bool rcu_dynticks_curr_cpu_in_eqs(void) | |
345 | { | |
346 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
347 | ||
b8c17e66 | 348 | return !(atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
349 | } |
350 | ||
8b2f63ab PM |
351 | /* |
352 | * Snapshot the ->dynticks counter with full ordering so as to allow | |
353 | * stable comparison of this counter with past and future snapshots. | |
354 | */ | |
02a5c550 | 355 | int rcu_dynticks_snap(struct rcu_dynticks *rdtp) |
8b2f63ab PM |
356 | { |
357 | int snap = atomic_add_return(0, &rdtp->dynticks); | |
358 | ||
b8c17e66 | 359 | return snap & ~RCU_DYNTICK_CTRL_MASK; |
8b2f63ab PM |
360 | } |
361 | ||
02a5c550 PM |
362 | /* |
363 | * Return true if the snapshot returned from rcu_dynticks_snap() | |
364 | * indicates that RCU is in an extended quiescent state. | |
365 | */ | |
366 | static bool rcu_dynticks_in_eqs(int snap) | |
367 | { | |
b8c17e66 | 368 | return !(snap & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
369 | } |
370 | ||
371 | /* | |
372 | * Return true if the CPU corresponding to the specified rcu_dynticks | |
373 | * structure has spent some time in an extended quiescent state since | |
374 | * rcu_dynticks_snap() returned the specified snapshot. | |
375 | */ | |
376 | static bool rcu_dynticks_in_eqs_since(struct rcu_dynticks *rdtp, int snap) | |
377 | { | |
378 | return snap != rcu_dynticks_snap(rdtp); | |
379 | } | |
380 | ||
6563de9d PM |
381 | /* |
382 | * Do a double-increment of the ->dynticks counter to emulate a | |
383 | * momentary idle-CPU quiescent state. | |
384 | */ | |
385 | static void rcu_dynticks_momentary_idle(void) | |
386 | { | |
387 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 PM |
388 | int special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR, |
389 | &rdtp->dynticks); | |
6563de9d PM |
390 | |
391 | /* It is illegal to call this from idle state. */ | |
b8c17e66 | 392 | WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR)); |
6563de9d PM |
393 | } |
394 | ||
b8c17e66 PM |
395 | /* |
396 | * Set the special (bottom) bit of the specified CPU so that it | |
397 | * will take special action (such as flushing its TLB) on the | |
398 | * next exit from an extended quiescent state. Returns true if | |
399 | * the bit was successfully set, or false if the CPU was not in | |
400 | * an extended quiescent state. | |
401 | */ | |
402 | bool rcu_eqs_special_set(int cpu) | |
403 | { | |
404 | int old; | |
405 | int new; | |
406 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); | |
407 | ||
408 | do { | |
409 | old = atomic_read(&rdtp->dynticks); | |
410 | if (old & RCU_DYNTICK_CTRL_CTR) | |
411 | return false; | |
412 | new = old | RCU_DYNTICK_CTRL_MASK; | |
413 | } while (atomic_cmpxchg(&rdtp->dynticks, old, new) != old); | |
414 | return true; | |
6563de9d | 415 | } |
5cd37193 | 416 | |
4a81e832 PM |
417 | /* |
418 | * Let the RCU core know that this CPU has gone through the scheduler, | |
419 | * which is a quiescent state. This is called when the need for a | |
420 | * quiescent state is urgent, so we burn an atomic operation and full | |
421 | * memory barriers to let the RCU core know about it, regardless of what | |
422 | * this CPU might (or might not) do in the near future. | |
423 | * | |
0f9be8ca | 424 | * We inform the RCU core by emulating a zero-duration dyntick-idle period. |
46a5d164 PM |
425 | * |
426 | * The caller must have disabled interrupts. | |
4a81e832 PM |
427 | */ |
428 | static void rcu_momentary_dyntick_idle(void) | |
429 | { | |
0f9be8ca PM |
430 | raw_cpu_write(rcu_dynticks.rcu_need_heavy_qs, false); |
431 | rcu_dynticks_momentary_idle(); | |
4a81e832 PM |
432 | } |
433 | ||
25502a6c PM |
434 | /* |
435 | * Note a context switch. This is a quiescent state for RCU-sched, | |
436 | * and requires special handling for preemptible RCU. | |
46a5d164 | 437 | * The caller must have disabled interrupts. |
25502a6c | 438 | */ |
bcbfdd01 | 439 | void rcu_note_context_switch(bool preempt) |
25502a6c | 440 | { |
bb73c52b | 441 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
f7f7bac9 | 442 | trace_rcu_utilization(TPS("Start context switch")); |
284a8c93 | 443 | rcu_sched_qs(); |
5b72f964 | 444 | rcu_preempt_note_context_switch(preempt); |
9226b10d PM |
445 | /* Load rcu_urgent_qs before other flags. */ |
446 | if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs))) | |
447 | goto out; | |
448 | this_cpu_write(rcu_dynticks.rcu_urgent_qs, false); | |
0f9be8ca | 449 | if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) |
4a81e832 | 450 | rcu_momentary_dyntick_idle(); |
9226b10d | 451 | this_cpu_inc(rcu_dynticks.rcu_qs_ctr); |
bcbfdd01 PM |
452 | if (!preempt) |
453 | rcu_note_voluntary_context_switch_lite(current); | |
9226b10d | 454 | out: |
f7f7bac9 | 455 | trace_rcu_utilization(TPS("End context switch")); |
bb73c52b | 456 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
25502a6c | 457 | } |
29ce8310 | 458 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
25502a6c | 459 | |
5cd37193 | 460 | /* |
1925d196 | 461 | * Register a quiescent state for all RCU flavors. If there is an |
5cd37193 PM |
462 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight |
463 | * dyntick-idle quiescent state visible to other CPUs (but only for those | |
1925d196 | 464 | * RCU flavors in desperate need of a quiescent state, which will normally |
5cd37193 PM |
465 | * be none of them). Either way, do a lightweight quiescent state for |
466 | * all RCU flavors. | |
bb73c52b BF |
467 | * |
468 | * The barrier() calls are redundant in the common case when this is | |
469 | * called externally, but just in case this is called from within this | |
470 | * file. | |
471 | * | |
5cd37193 PM |
472 | */ |
473 | void rcu_all_qs(void) | |
474 | { | |
46a5d164 PM |
475 | unsigned long flags; |
476 | ||
9226b10d PM |
477 | if (!raw_cpu_read(rcu_dynticks.rcu_urgent_qs)) |
478 | return; | |
479 | preempt_disable(); | |
480 | /* Load rcu_urgent_qs before other flags. */ | |
481 | if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs))) { | |
482 | preempt_enable(); | |
483 | return; | |
484 | } | |
485 | this_cpu_write(rcu_dynticks.rcu_urgent_qs, false); | |
bb73c52b | 486 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
0f9be8ca | 487 | if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) { |
46a5d164 | 488 | local_irq_save(flags); |
5cd37193 | 489 | rcu_momentary_dyntick_idle(); |
46a5d164 PM |
490 | local_irq_restore(flags); |
491 | } | |
9226b10d | 492 | if (unlikely(raw_cpu_read(rcu_sched_data.cpu_no_qs.b.exp))) |
a1e12248 | 493 | rcu_sched_qs(); |
9577df9a | 494 | this_cpu_inc(rcu_dynticks.rcu_qs_ctr); |
bb73c52b | 495 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
9226b10d | 496 | preempt_enable(); |
5cd37193 PM |
497 | } |
498 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
499 | ||
17c7798b PM |
500 | #define DEFAULT_RCU_BLIMIT 10 /* Maximum callbacks per rcu_do_batch. */ |
501 | static long blimit = DEFAULT_RCU_BLIMIT; | |
502 | #define DEFAULT_RCU_QHIMARK 10000 /* If this many pending, ignore blimit. */ | |
503 | static long qhimark = DEFAULT_RCU_QHIMARK; | |
504 | #define DEFAULT_RCU_QLOMARK 100 /* Once only this many pending, use blimit. */ | |
505 | static long qlowmark = DEFAULT_RCU_QLOMARK; | |
64db4cff | 506 | |
878d7439 ED |
507 | module_param(blimit, long, 0444); |
508 | module_param(qhimark, long, 0444); | |
509 | module_param(qlowmark, long, 0444); | |
3d76c082 | 510 | |
026ad283 PM |
511 | static ulong jiffies_till_first_fqs = ULONG_MAX; |
512 | static ulong jiffies_till_next_fqs = ULONG_MAX; | |
8c7c4829 | 513 | static bool rcu_kick_kthreads; |
d40011f6 PM |
514 | |
515 | module_param(jiffies_till_first_fqs, ulong, 0644); | |
516 | module_param(jiffies_till_next_fqs, ulong, 0644); | |
8c7c4829 | 517 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 518 | |
4a81e832 PM |
519 | /* |
520 | * How long the grace period must be before we start recruiting | |
521 | * quiescent-state help from rcu_note_context_switch(). | |
522 | */ | |
f79c3ad6 PM |
523 | static ulong jiffies_till_sched_qs = HZ / 10; |
524 | module_param(jiffies_till_sched_qs, ulong, 0444); | |
4a81e832 | 525 | |
fe5ac724 | 526 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *rsp)); |
4cdfc175 | 527 | static void force_quiescent_state(struct rcu_state *rsp); |
e3950ecd | 528 | static int rcu_pending(void); |
64db4cff PM |
529 | |
530 | /* | |
17ef2fe9 | 531 | * Return the number of RCU GPs completed thus far for debug & stats. |
64db4cff | 532 | */ |
17ef2fe9 | 533 | unsigned long rcu_get_gp_seq(void) |
917963d0 | 534 | { |
d7219312 | 535 | return READ_ONCE(rcu_state_p->gp_seq); |
917963d0 | 536 | } |
17ef2fe9 | 537 | EXPORT_SYMBOL_GPL(rcu_get_gp_seq); |
917963d0 PM |
538 | |
539 | /* | |
17ef2fe9 | 540 | * Return the number of RCU-sched GPs completed thus far for debug & stats. |
64db4cff | 541 | */ |
17ef2fe9 | 542 | unsigned long rcu_sched_get_gp_seq(void) |
917963d0 | 543 | { |
d7219312 | 544 | return READ_ONCE(rcu_sched_state.gp_seq); |
917963d0 | 545 | } |
17ef2fe9 | 546 | EXPORT_SYMBOL_GPL(rcu_sched_get_gp_seq); |
917963d0 PM |
547 | |
548 | /* | |
17ef2fe9 | 549 | * Return the number of RCU-bh GPs completed thus far for debug & stats. |
917963d0 | 550 | */ |
17ef2fe9 | 551 | unsigned long rcu_bh_get_gp_seq(void) |
917963d0 | 552 | { |
d7219312 | 553 | return READ_ONCE(rcu_bh_state.gp_seq); |
917963d0 | 554 | } |
17ef2fe9 | 555 | EXPORT_SYMBOL_GPL(rcu_bh_get_gp_seq); |
64db4cff | 556 | |
291783b8 PM |
557 | /* |
558 | * Return the number of RCU expedited batches completed thus far for | |
559 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
560 | * numbers mean idle. The value returned will thus be roughly double | |
561 | * the cumulative batches since boot. | |
562 | */ | |
563 | unsigned long rcu_exp_batches_completed(void) | |
564 | { | |
565 | return rcu_state_p->expedited_sequence; | |
566 | } | |
567 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
568 | ||
569 | /* | |
570 | * Return the number of RCU-sched expedited batches completed thus far | |
571 | * for debug & stats. Similar to rcu_exp_batches_completed(). | |
572 | */ | |
573 | unsigned long rcu_exp_batches_completed_sched(void) | |
574 | { | |
575 | return rcu_sched_state.expedited_sequence; | |
576 | } | |
577 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed_sched); | |
578 | ||
a381d757 ACB |
579 | /* |
580 | * Force a quiescent state. | |
581 | */ | |
582 | void rcu_force_quiescent_state(void) | |
583 | { | |
e534165b | 584 | force_quiescent_state(rcu_state_p); |
a381d757 ACB |
585 | } |
586 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
587 | ||
bf66f18e PM |
588 | /* |
589 | * Force a quiescent state for RCU BH. | |
590 | */ | |
591 | void rcu_bh_force_quiescent_state(void) | |
592 | { | |
4cdfc175 | 593 | force_quiescent_state(&rcu_bh_state); |
bf66f18e PM |
594 | } |
595 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
596 | ||
e7580f33 PM |
597 | /* |
598 | * Force a quiescent state for RCU-sched. | |
599 | */ | |
600 | void rcu_sched_force_quiescent_state(void) | |
601 | { | |
602 | force_quiescent_state(&rcu_sched_state); | |
603 | } | |
604 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
605 | ||
afea227f PM |
606 | /* |
607 | * Show the state of the grace-period kthreads. | |
608 | */ | |
609 | void show_rcu_gp_kthreads(void) | |
610 | { | |
611 | struct rcu_state *rsp; | |
612 | ||
613 | for_each_rcu_flavor(rsp) { | |
614 | pr_info("%s: wait state: %d ->state: %#lx\n", | |
615 | rsp->name, rsp->gp_state, rsp->gp_kthread->state); | |
616 | /* sched_show_task(rsp->gp_kthread); */ | |
617 | } | |
618 | } | |
619 | EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads); | |
620 | ||
4a298656 PM |
621 | /* |
622 | * Record the number of times rcutorture tests have been initiated and | |
623 | * terminated. This information allows the debugfs tracing stats to be | |
624 | * correlated to the rcutorture messages, even when the rcutorture module | |
625 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
626 | * store this state in rcutorture itself. | |
627 | */ | |
628 | void rcutorture_record_test_transition(void) | |
629 | { | |
630 | rcutorture_testseq++; | |
631 | rcutorture_vernum = 0; | |
632 | } | |
633 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
634 | ||
ad0dc7f9 PM |
635 | /* |
636 | * Send along grace-period-related data for rcutorture diagnostics. | |
637 | */ | |
638 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
aebc8264 | 639 | unsigned long *gp_seq) |
ad0dc7f9 PM |
640 | { |
641 | struct rcu_state *rsp = NULL; | |
642 | ||
643 | switch (test_type) { | |
644 | case RCU_FLAVOR: | |
e534165b | 645 | rsp = rcu_state_p; |
ad0dc7f9 PM |
646 | break; |
647 | case RCU_BH_FLAVOR: | |
648 | rsp = &rcu_bh_state; | |
649 | break; | |
650 | case RCU_SCHED_FLAVOR: | |
651 | rsp = &rcu_sched_state; | |
652 | break; | |
653 | default: | |
654 | break; | |
655 | } | |
7f6733c3 | 656 | if (rsp == NULL) |
ad0dc7f9 | 657 | return; |
7f6733c3 | 658 | *flags = READ_ONCE(rsp->gp_flags); |
aebc8264 | 659 | *gp_seq = rcu_seq_current(&rsp->gp_seq); |
ad0dc7f9 PM |
660 | } |
661 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
662 | ||
4a298656 PM |
663 | /* |
664 | * Record the number of writer passes through the current rcutorture test. | |
665 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
666 | * messages. | |
667 | */ | |
668 | void rcutorture_record_progress(unsigned long vernum) | |
669 | { | |
670 | rcutorture_vernum++; | |
671 | } | |
672 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
673 | ||
365187fb PM |
674 | /* |
675 | * Return the root node of the specified rcu_state structure. | |
676 | */ | |
677 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
678 | { | |
679 | return &rsp->node[0]; | |
680 | } | |
681 | ||
9b2e4f18 | 682 | /* |
215bba9f PM |
683 | * Enter an RCU extended quiescent state, which can be either the |
684 | * idle loop or adaptive-tickless usermode execution. | |
9b2e4f18 | 685 | * |
215bba9f PM |
686 | * We crowbar the ->dynticks_nmi_nesting field to zero to allow for |
687 | * the possibility of usermode upcalls having messed up our count | |
688 | * of interrupt nesting level during the prior busy period. | |
9b2e4f18 | 689 | */ |
215bba9f | 690 | static void rcu_eqs_enter(bool user) |
9b2e4f18 | 691 | { |
96d3fd0d PM |
692 | struct rcu_state *rsp; |
693 | struct rcu_data *rdp; | |
215bba9f | 694 | struct rcu_dynticks *rdtp; |
96d3fd0d | 695 | |
215bba9f PM |
696 | rdtp = this_cpu_ptr(&rcu_dynticks); |
697 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0); | |
698 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
699 | rdtp->dynticks_nesting == 0); | |
700 | if (rdtp->dynticks_nesting != 1) { | |
701 | rdtp->dynticks_nesting--; | |
702 | return; | |
9b2e4f18 | 703 | } |
96d3fd0d | 704 | |
b04db8e1 | 705 | lockdep_assert_irqs_disabled(); |
dec98900 | 706 | trace_rcu_dyntick(TPS("Start"), rdtp->dynticks_nesting, 0, rdtp->dynticks); |
e68bbb26 | 707 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
96d3fd0d PM |
708 | for_each_rcu_flavor(rsp) { |
709 | rdp = this_cpu_ptr(rsp->rda); | |
710 | do_nocb_deferred_wakeup(rdp); | |
711 | } | |
198bbf81 | 712 | rcu_prepare_for_idle(); |
2342172f | 713 | WRITE_ONCE(rdtp->dynticks_nesting, 0); /* Avoid irq-access tearing. */ |
844ccdd7 | 714 | rcu_dynticks_eqs_enter(); |
176f8f7a | 715 | rcu_dynticks_task_enter(); |
64db4cff | 716 | } |
adf5091e FW |
717 | |
718 | /** | |
719 | * rcu_idle_enter - inform RCU that current CPU is entering idle | |
720 | * | |
721 | * Enter idle mode, in other words, -leave- the mode in which RCU | |
722 | * read-side critical sections can occur. (Though RCU read-side | |
723 | * critical sections can occur in irq handlers in idle, a possibility | |
724 | * handled by irq_enter() and irq_exit().) | |
725 | * | |
c0da313e PM |
726 | * If you add or remove a call to rcu_idle_enter(), be sure to test with |
727 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
728 | */ |
729 | void rcu_idle_enter(void) | |
730 | { | |
b04db8e1 | 731 | lockdep_assert_irqs_disabled(); |
cb349ca9 | 732 | rcu_eqs_enter(false); |
adf5091e | 733 | } |
64db4cff | 734 | |
d1ec4c34 | 735 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
736 | /** |
737 | * rcu_user_enter - inform RCU that we are resuming userspace. | |
738 | * | |
739 | * Enter RCU idle mode right before resuming userspace. No use of RCU | |
740 | * is permitted between this call and rcu_user_exit(). This way the | |
741 | * CPU doesn't need to maintain the tick for RCU maintenance purposes | |
742 | * when the CPU runs in userspace. | |
c0da313e PM |
743 | * |
744 | * If you add or remove a call to rcu_user_enter(), be sure to test with | |
745 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
746 | */ |
747 | void rcu_user_enter(void) | |
748 | { | |
b04db8e1 | 749 | lockdep_assert_irqs_disabled(); |
d4db30af | 750 | rcu_eqs_enter(true); |
adf5091e | 751 | } |
d1ec4c34 | 752 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 753 | |
fd581a91 PM |
754 | /** |
755 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
756 | * | |
757 | * If we are returning from the outermost NMI handler that interrupted an | |
758 | * RCU-idle period, update rdtp->dynticks and rdtp->dynticks_nmi_nesting | |
759 | * to let the RCU grace-period handling know that the CPU is back to | |
760 | * being RCU-idle. | |
761 | * | |
762 | * If you add or remove a call to rcu_nmi_exit(), be sure to test | |
763 | * with CONFIG_RCU_EQS_DEBUG=y. | |
764 | */ | |
765 | void rcu_nmi_exit(void) | |
766 | { | |
767 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
768 | ||
769 | /* | |
770 | * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks. | |
771 | * (We are exiting an NMI handler, so RCU better be paying attention | |
772 | * to us!) | |
773 | */ | |
774 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting <= 0); | |
775 | WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs()); | |
776 | ||
777 | /* | |
778 | * If the nesting level is not 1, the CPU wasn't RCU-idle, so | |
779 | * leave it in non-RCU-idle state. | |
780 | */ | |
781 | if (rdtp->dynticks_nmi_nesting != 1) { | |
dec98900 | 782 | trace_rcu_dyntick(TPS("--="), rdtp->dynticks_nmi_nesting, rdtp->dynticks_nmi_nesting - 2, rdtp->dynticks); |
fd581a91 PM |
783 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* No store tearing. */ |
784 | rdtp->dynticks_nmi_nesting - 2); | |
785 | return; | |
786 | } | |
787 | ||
788 | /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */ | |
dec98900 | 789 | trace_rcu_dyntick(TPS("Startirq"), rdtp->dynticks_nmi_nesting, 0, rdtp->dynticks); |
fd581a91 PM |
790 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */ |
791 | rcu_dynticks_eqs_enter(); | |
792 | } | |
793 | ||
9b2e4f18 PM |
794 | /** |
795 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
796 | * | |
797 | * Exit from an interrupt handler, which might possibly result in entering | |
798 | * idle mode, in other words, leaving the mode in which read-side critical | |
7c9906ca | 799 | * sections can occur. The caller must have disabled interrupts. |
64db4cff | 800 | * |
9b2e4f18 PM |
801 | * This code assumes that the idle loop never does anything that might |
802 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
58721f5d PM |
803 | * architecture's idle loop violates this assumption, RCU will give you what |
804 | * you deserve, good and hard. But very infrequently and irreproducibly. | |
9b2e4f18 PM |
805 | * |
806 | * Use things like work queues to work around this limitation. | |
807 | * | |
808 | * You have been warned. | |
c0da313e PM |
809 | * |
810 | * If you add or remove a call to rcu_irq_exit(), be sure to test with | |
811 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 812 | */ |
9b2e4f18 | 813 | void rcu_irq_exit(void) |
64db4cff | 814 | { |
58721f5d | 815 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
64db4cff | 816 | |
b04db8e1 | 817 | lockdep_assert_irqs_disabled(); |
58721f5d PM |
818 | if (rdtp->dynticks_nmi_nesting == 1) |
819 | rcu_prepare_for_idle(); | |
820 | rcu_nmi_exit(); | |
821 | if (rdtp->dynticks_nmi_nesting == 0) | |
822 | rcu_dynticks_task_enter(); | |
7c9906ca PM |
823 | } |
824 | ||
825 | /* | |
826 | * Wrapper for rcu_irq_exit() where interrupts are enabled. | |
c0da313e PM |
827 | * |
828 | * If you add or remove a call to rcu_irq_exit_irqson(), be sure to test | |
829 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
830 | */ |
831 | void rcu_irq_exit_irqson(void) | |
832 | { | |
833 | unsigned long flags; | |
834 | ||
835 | local_irq_save(flags); | |
836 | rcu_irq_exit(); | |
9b2e4f18 PM |
837 | local_irq_restore(flags); |
838 | } | |
839 | ||
adf5091e FW |
840 | /* |
841 | * Exit an RCU extended quiescent state, which can be either the | |
842 | * idle loop or adaptive-tickless usermode execution. | |
51a1fd30 PM |
843 | * |
844 | * We crowbar the ->dynticks_nmi_nesting field to DYNTICK_IRQ_NONIDLE to | |
845 | * allow for the possibility of usermode upcalls messing up our count of | |
846 | * interrupt nesting level during the busy period that is just now starting. | |
9b2e4f18 | 847 | */ |
adf5091e | 848 | static void rcu_eqs_exit(bool user) |
9b2e4f18 | 849 | { |
9b2e4f18 | 850 | struct rcu_dynticks *rdtp; |
84585aa8 | 851 | long oldval; |
9b2e4f18 | 852 | |
b04db8e1 | 853 | lockdep_assert_irqs_disabled(); |
c9d4b0af | 854 | rdtp = this_cpu_ptr(&rcu_dynticks); |
9b2e4f18 | 855 | oldval = rdtp->dynticks_nesting; |
1ce46ee5 | 856 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0); |
51a1fd30 PM |
857 | if (oldval) { |
858 | rdtp->dynticks_nesting++; | |
9dd238e2 | 859 | return; |
3a592405 | 860 | } |
9dd238e2 PM |
861 | rcu_dynticks_task_exit(); |
862 | rcu_dynticks_eqs_exit(); | |
863 | rcu_cleanup_after_idle(); | |
864 | trace_rcu_dyntick(TPS("End"), rdtp->dynticks_nesting, 1, rdtp->dynticks); | |
e68bbb26 | 865 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
9dd238e2 PM |
866 | WRITE_ONCE(rdtp->dynticks_nesting, 1); |
867 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, DYNTICK_IRQ_NONIDLE); | |
9b2e4f18 | 868 | } |
adf5091e FW |
869 | |
870 | /** | |
871 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
872 | * | |
873 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
874 | * read-side critical sections can occur. | |
875 | * | |
c0da313e PM |
876 | * If you add or remove a call to rcu_idle_exit(), be sure to test with |
877 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
878 | */ |
879 | void rcu_idle_exit(void) | |
880 | { | |
c5d900bf FW |
881 | unsigned long flags; |
882 | ||
883 | local_irq_save(flags); | |
cb349ca9 | 884 | rcu_eqs_exit(false); |
c5d900bf | 885 | local_irq_restore(flags); |
adf5091e | 886 | } |
9b2e4f18 | 887 | |
d1ec4c34 | 888 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
889 | /** |
890 | * rcu_user_exit - inform RCU that we are exiting userspace. | |
891 | * | |
892 | * Exit RCU idle mode while entering the kernel because it can | |
893 | * run a RCU read side critical section anytime. | |
c0da313e PM |
894 | * |
895 | * If you add or remove a call to rcu_user_exit(), be sure to test with | |
896 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
897 | */ |
898 | void rcu_user_exit(void) | |
899 | { | |
91d1aa43 | 900 | rcu_eqs_exit(1); |
adf5091e | 901 | } |
d1ec4c34 | 902 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 903 | |
64db4cff PM |
904 | /** |
905 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
906 | * | |
734d1680 PM |
907 | * If the CPU was idle from RCU's viewpoint, update rdtp->dynticks and |
908 | * rdtp->dynticks_nmi_nesting to let the RCU grace-period handling know | |
909 | * that the CPU is active. This implementation permits nested NMIs, as | |
910 | * long as the nesting level does not overflow an int. (You will probably | |
911 | * run out of stack space first.) | |
c0da313e PM |
912 | * |
913 | * If you add or remove a call to rcu_nmi_enter(), be sure to test | |
914 | * with CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff PM |
915 | */ |
916 | void rcu_nmi_enter(void) | |
917 | { | |
c9d4b0af | 918 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
84585aa8 | 919 | long incby = 2; |
64db4cff | 920 | |
734d1680 PM |
921 | /* Complain about underflow. */ |
922 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting < 0); | |
923 | ||
924 | /* | |
925 | * If idle from RCU viewpoint, atomically increment ->dynticks | |
926 | * to mark non-idle and increment ->dynticks_nmi_nesting by one. | |
927 | * Otherwise, increment ->dynticks_nmi_nesting by two. This means | |
928 | * if ->dynticks_nmi_nesting is equal to one, we are guaranteed | |
929 | * to be in the outermost NMI handler that interrupted an RCU-idle | |
930 | * period (observation due to Andy Lutomirski). | |
931 | */ | |
02a5c550 | 932 | if (rcu_dynticks_curr_cpu_in_eqs()) { |
2625d469 | 933 | rcu_dynticks_eqs_exit(); |
734d1680 PM |
934 | incby = 1; |
935 | } | |
bd2b879a PM |
936 | trace_rcu_dyntick(incby == 1 ? TPS("Endirq") : TPS("++="), |
937 | rdtp->dynticks_nmi_nesting, | |
dec98900 | 938 | rdtp->dynticks_nmi_nesting + incby, rdtp->dynticks); |
fd581a91 PM |
939 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* Prevent store tearing. */ |
940 | rdtp->dynticks_nmi_nesting + incby); | |
734d1680 | 941 | barrier(); |
64db4cff PM |
942 | } |
943 | ||
944 | /** | |
9b2e4f18 | 945 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle |
64db4cff | 946 | * |
9b2e4f18 PM |
947 | * Enter an interrupt handler, which might possibly result in exiting |
948 | * idle mode, in other words, entering the mode in which read-side critical | |
7c9906ca | 949 | * sections can occur. The caller must have disabled interrupts. |
c0da313e | 950 | * |
9b2e4f18 | 951 | * Note that the Linux kernel is fully capable of entering an interrupt |
58721f5d PM |
952 | * handler that it never exits, for example when doing upcalls to user mode! |
953 | * This code assumes that the idle loop never does upcalls to user mode. | |
954 | * If your architecture's idle loop does do upcalls to user mode (or does | |
955 | * anything else that results in unbalanced calls to the irq_enter() and | |
956 | * irq_exit() functions), RCU will give you what you deserve, good and hard. | |
957 | * But very infrequently and irreproducibly. | |
9b2e4f18 PM |
958 | * |
959 | * Use things like work queues to work around this limitation. | |
960 | * | |
961 | * You have been warned. | |
c0da313e PM |
962 | * |
963 | * If you add or remove a call to rcu_irq_enter(), be sure to test with | |
964 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 965 | */ |
9b2e4f18 | 966 | void rcu_irq_enter(void) |
64db4cff | 967 | { |
c9d4b0af | 968 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
64db4cff | 969 | |
b04db8e1 | 970 | lockdep_assert_irqs_disabled(); |
58721f5d PM |
971 | if (rdtp->dynticks_nmi_nesting == 0) |
972 | rcu_dynticks_task_exit(); | |
973 | rcu_nmi_enter(); | |
974 | if (rdtp->dynticks_nmi_nesting == 1) | |
975 | rcu_cleanup_after_idle(); | |
7c9906ca | 976 | } |
734d1680 | 977 | |
7c9906ca PM |
978 | /* |
979 | * Wrapper for rcu_irq_enter() where interrupts are enabled. | |
c0da313e PM |
980 | * |
981 | * If you add or remove a call to rcu_irq_enter_irqson(), be sure to test | |
982 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
983 | */ |
984 | void rcu_irq_enter_irqson(void) | |
985 | { | |
986 | unsigned long flags; | |
734d1680 | 987 | |
7c9906ca PM |
988 | local_irq_save(flags); |
989 | rcu_irq_enter(); | |
64db4cff | 990 | local_irq_restore(flags); |
64db4cff PM |
991 | } |
992 | ||
5c173eb8 PM |
993 | /** |
994 | * rcu_is_watching - see if RCU thinks that the current CPU is idle | |
64db4cff | 995 | * |
791875d1 PM |
996 | * Return true if RCU is watching the running CPU, which means that this |
997 | * CPU can safely enter RCU read-side critical sections. In other words, | |
998 | * if the current CPU is in its idle loop and is neither in an interrupt | |
34240697 | 999 | * or NMI handler, return true. |
64db4cff | 1000 | */ |
9418fb20 | 1001 | bool notrace rcu_is_watching(void) |
64db4cff | 1002 | { |
f534ed1f | 1003 | bool ret; |
34240697 | 1004 | |
46f00d18 | 1005 | preempt_disable_notrace(); |
791875d1 | 1006 | ret = !rcu_dynticks_curr_cpu_in_eqs(); |
46f00d18 | 1007 | preempt_enable_notrace(); |
34240697 | 1008 | return ret; |
64db4cff | 1009 | } |
5c173eb8 | 1010 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 1011 | |
bcbfdd01 PM |
1012 | /* |
1013 | * If a holdout task is actually running, request an urgent quiescent | |
1014 | * state from its CPU. This is unsynchronized, so migrations can cause | |
1015 | * the request to go to the wrong CPU. Which is OK, all that will happen | |
1016 | * is that the CPU's next context switch will be a bit slower and next | |
1017 | * time around this task will generate another request. | |
1018 | */ | |
1019 | void rcu_request_urgent_qs_task(struct task_struct *t) | |
1020 | { | |
1021 | int cpu; | |
1022 | ||
1023 | barrier(); | |
1024 | cpu = task_cpu(t); | |
1025 | if (!task_curr(t)) | |
1026 | return; /* This task is not running on that CPU. */ | |
1027 | smp_store_release(per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, cpu), true); | |
1028 | } | |
1029 | ||
62fde6ed | 1030 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
1031 | |
1032 | /* | |
1033 | * Is the current CPU online? Disable preemption to avoid false positives | |
1034 | * that could otherwise happen due to the current CPU number being sampled, | |
1035 | * this task being preempted, its old CPU being taken offline, resuming | |
1036 | * on some other CPU, then determining that its old CPU is now offline. | |
1037 | * It is OK to use RCU on an offline processor during initial boot, hence | |
2036d94a PM |
1038 | * the check for rcu_scheduler_fully_active. Note also that it is OK |
1039 | * for a CPU coming online to use RCU for one jiffy prior to marking itself | |
1040 | * online in the cpu_online_mask. Similarly, it is OK for a CPU going | |
1041 | * offline to continue to use RCU for one jiffy after marking itself | |
1042 | * offline in the cpu_online_mask. This leniency is necessary given the | |
1043 | * non-atomic nature of the online and offline processing, for example, | |
4df83742 TG |
1044 | * the fact that a CPU enters the scheduler after completing the teardown |
1045 | * of the CPU. | |
2036d94a | 1046 | * |
4df83742 TG |
1047 | * This is also why RCU internally marks CPUs online during in the |
1048 | * preparation phase and offline after the CPU has been taken down. | |
c0d6d01b PM |
1049 | * |
1050 | * Disable checking if in an NMI handler because we cannot safely report | |
1051 | * errors from NMI handlers anyway. | |
1052 | */ | |
1053 | bool rcu_lockdep_current_cpu_online(void) | |
1054 | { | |
2036d94a PM |
1055 | struct rcu_data *rdp; |
1056 | struct rcu_node *rnp; | |
c0d6d01b PM |
1057 | bool ret; |
1058 | ||
1059 | if (in_nmi()) | |
f6f7ee9a | 1060 | return true; |
c0d6d01b | 1061 | preempt_disable(); |
c9d4b0af | 1062 | rdp = this_cpu_ptr(&rcu_sched_data); |
2036d94a | 1063 | rnp = rdp->mynode; |
0aa04b05 | 1064 | ret = (rdp->grpmask & rcu_rnp_online_cpus(rnp)) || |
c0d6d01b PM |
1065 | !rcu_scheduler_fully_active; |
1066 | preempt_enable(); | |
1067 | return ret; | |
1068 | } | |
1069 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
1070 | ||
62fde6ed | 1071 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 1072 | |
64db4cff | 1073 | /** |
9b2e4f18 | 1074 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle |
64db4cff | 1075 | * |
9b2e4f18 PM |
1076 | * If the current CPU is idle or running at a first-level (not nested) |
1077 | * interrupt from idle, return true. The caller must have at least | |
1078 | * disabled preemption. | |
64db4cff | 1079 | */ |
62e3cb14 | 1080 | static int rcu_is_cpu_rrupt_from_idle(void) |
64db4cff | 1081 | { |
51a1fd30 PM |
1082 | return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 0 && |
1083 | __this_cpu_read(rcu_dynticks.dynticks_nmi_nesting) <= 1; | |
64db4cff PM |
1084 | } |
1085 | ||
9b9500da PM |
1086 | /* |
1087 | * We are reporting a quiescent state on behalf of some other CPU, so | |
1088 | * it is our responsibility to check for and handle potential overflow | |
a66ae8ae | 1089 | * of the rcu_node ->gp_seq counter with respect to the rcu_data counters. |
9b9500da PM |
1090 | * After all, the CPU might be in deep idle state, and thus executing no |
1091 | * code whatsoever. | |
1092 | */ | |
1093 | static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp) | |
1094 | { | |
a32e01ee | 1095 | raw_lockdep_assert_held_rcu_node(rnp); |
a66ae8ae PM |
1096 | if (ULONG_CMP_LT(rcu_seq_current(&rdp->gp_seq) + ULONG_MAX / 4, |
1097 | rnp->gp_seq)) | |
9b9500da | 1098 | WRITE_ONCE(rdp->gpwrap, true); |
8aa670cd PM |
1099 | if (ULONG_CMP_LT(rdp->rcu_iw_gp_seq + ULONG_MAX / 4, rnp->gp_seq)) |
1100 | rdp->rcu_iw_gp_seq = rnp->gp_seq + ULONG_MAX / 4; | |
9b9500da PM |
1101 | } |
1102 | ||
64db4cff PM |
1103 | /* |
1104 | * Snapshot the specified CPU's dynticks counter so that we can later | |
1105 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 1106 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 1107 | */ |
fe5ac724 | 1108 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
64db4cff | 1109 | { |
8b2f63ab | 1110 | rdp->dynticks_snap = rcu_dynticks_snap(rdp->dynticks); |
02a5c550 | 1111 | if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) { |
fee5997c | 1112 | trace_rcu_fqs(rdp->rsp->name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 1113 | rcu_gpnum_ovf(rdp->mynode, rdp); |
23a9bacd | 1114 | return 1; |
7941dbde | 1115 | } |
23a9bacd | 1116 | return 0; |
64db4cff PM |
1117 | } |
1118 | ||
9b9500da PM |
1119 | /* |
1120 | * Handler for the irq_work request posted when a grace period has | |
1121 | * gone on for too long, but not yet long enough for an RCU CPU | |
1122 | * stall warning. Set state appropriately, but just complain if | |
1123 | * there is unexpected state on entry. | |
1124 | */ | |
1125 | static void rcu_iw_handler(struct irq_work *iwp) | |
1126 | { | |
1127 | struct rcu_data *rdp; | |
1128 | struct rcu_node *rnp; | |
1129 | ||
1130 | rdp = container_of(iwp, struct rcu_data, rcu_iw); | |
1131 | rnp = rdp->mynode; | |
1132 | raw_spin_lock_rcu_node(rnp); | |
1133 | if (!WARN_ON_ONCE(!rdp->rcu_iw_pending)) { | |
8aa670cd | 1134 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1135 | rdp->rcu_iw_pending = false; |
1136 | } | |
1137 | raw_spin_unlock_rcu_node(rnp); | |
1138 | } | |
1139 | ||
64db4cff PM |
1140 | /* |
1141 | * Return true if the specified CPU has passed through a quiescent | |
1142 | * state by virtue of being in or having passed through an dynticks | |
1143 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 1144 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 1145 | */ |
fe5ac724 | 1146 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
64db4cff | 1147 | { |
3a19b46a | 1148 | unsigned long jtsq; |
0f9be8ca | 1149 | bool *rnhqp; |
9226b10d | 1150 | bool *ruqp; |
9b9500da | 1151 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
1152 | |
1153 | /* | |
1154 | * If the CPU passed through or entered a dynticks idle phase with | |
1155 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
1156 | * already acknowledged the request to pass through a quiescent | |
1157 | * state. Either way, that CPU cannot possibly be in an RCU | |
1158 | * read-side critical section that started before the beginning | |
1159 | * of the current RCU grace period. | |
1160 | */ | |
02a5c550 | 1161 | if (rcu_dynticks_in_eqs_since(rdp->dynticks, rdp->dynticks_snap)) { |
fee5997c | 1162 | trace_rcu_fqs(rdp->rsp->name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
64db4cff | 1163 | rdp->dynticks_fqs++; |
9b9500da | 1164 | rcu_gpnum_ovf(rnp, rdp); |
64db4cff PM |
1165 | return 1; |
1166 | } | |
1167 | ||
a82dcc76 | 1168 | /* |
cee43939 PM |
1169 | * Has this CPU encountered a cond_resched() since the beginning |
1170 | * of the grace period? For this to be the case, the CPU has to | |
1171 | * have noticed the current grace period. This might not be the | |
1172 | * case for nohz_full CPUs looping in the kernel. | |
a82dcc76 | 1173 | */ |
f79c3ad6 | 1174 | jtsq = jiffies_till_sched_qs; |
9226b10d | 1175 | ruqp = per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, rdp->cpu); |
3a19b46a | 1176 | if (time_after(jiffies, rdp->rsp->gp_start + jtsq) && |
9577df9a | 1177 | READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) && |
e05720b0 | 1178 | rcu_seq_current(&rdp->gp_seq) == rnp->gp_seq && !rdp->gpwrap) { |
fee5997c | 1179 | trace_rcu_fqs(rdp->rsp->name, rdp->gp_seq, rdp->cpu, TPS("rqc")); |
9b9500da | 1180 | rcu_gpnum_ovf(rnp, rdp); |
3a19b46a | 1181 | return 1; |
f79c3ad6 | 1182 | } else if (time_after(jiffies, rdp->rsp->gp_start + jtsq)) { |
9226b10d PM |
1183 | /* Load rcu_qs_ctr before store to rcu_urgent_qs. */ |
1184 | smp_store_release(ruqp, true); | |
3a19b46a PM |
1185 | } |
1186 | ||
38d30b33 PM |
1187 | /* Check for the CPU being offline. */ |
1188 | if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp))) { | |
fee5997c | 1189 | trace_rcu_fqs(rdp->rsp->name, rdp->gp_seq, rdp->cpu, TPS("ofl")); |
a82dcc76 | 1190 | rdp->offline_fqs++; |
9b9500da | 1191 | rcu_gpnum_ovf(rnp, rdp); |
a82dcc76 PM |
1192 | return 1; |
1193 | } | |
65d798f0 PM |
1194 | |
1195 | /* | |
4a81e832 PM |
1196 | * A CPU running for an extended time within the kernel can |
1197 | * delay RCU grace periods. When the CPU is in NO_HZ_FULL mode, | |
1198 | * even context-switching back and forth between a pair of | |
1199 | * in-kernel CPU-bound tasks cannot advance grace periods. | |
1200 | * So if the grace period is old enough, make the CPU pay attention. | |
1201 | * Note that the unsynchronized assignments to the per-CPU | |
0f9be8ca | 1202 | * rcu_need_heavy_qs variable are safe. Yes, setting of |
4a81e832 PM |
1203 | * bits can be lost, but they will be set again on the next |
1204 | * force-quiescent-state pass. So lost bit sets do not result | |
1205 | * in incorrect behavior, merely in a grace period lasting | |
1206 | * a few jiffies longer than it might otherwise. Because | |
1207 | * there are at most four threads involved, and because the | |
1208 | * updates are only once every few jiffies, the probability of | |
1209 | * lossage (and thus of slight grace-period extension) is | |
1210 | * quite low. | |
6193c76a | 1211 | */ |
0f9be8ca PM |
1212 | rnhqp = &per_cpu(rcu_dynticks.rcu_need_heavy_qs, rdp->cpu); |
1213 | if (!READ_ONCE(*rnhqp) && | |
1214 | (time_after(jiffies, rdp->rsp->gp_start + jtsq) || | |
1215 | time_after(jiffies, rdp->rsp->jiffies_resched))) { | |
1216 | WRITE_ONCE(*rnhqp, true); | |
9226b10d PM |
1217 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ |
1218 | smp_store_release(ruqp, true); | |
f79c3ad6 | 1219 | rdp->rsp->jiffies_resched += jtsq; /* Re-enable beating. */ |
6193c76a PM |
1220 | } |
1221 | ||
28053bc7 | 1222 | /* |
9b9500da PM |
1223 | * If more than halfway to RCU CPU stall-warning time, do a |
1224 | * resched_cpu() to try to loosen things up a bit. Also check to | |
1225 | * see if the CPU is getting hammered with interrupts, but only | |
1226 | * once per grace period, just to keep the IPIs down to a dull roar. | |
28053bc7 | 1227 | */ |
9b9500da | 1228 | if (jiffies - rdp->rsp->gp_start > rcu_jiffies_till_stall_check() / 2) { |
28053bc7 | 1229 | resched_cpu(rdp->cpu); |
9b9500da | 1230 | if (IS_ENABLED(CONFIG_IRQ_WORK) && |
8aa670cd | 1231 | !rdp->rcu_iw_pending && rdp->rcu_iw_gp_seq != rnp->gp_seq && |
9b9500da PM |
1232 | (rnp->ffmask & rdp->grpmask)) { |
1233 | init_irq_work(&rdp->rcu_iw, rcu_iw_handler); | |
1234 | rdp->rcu_iw_pending = true; | |
8aa670cd | 1235 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1236 | irq_work_queue_on(&rdp->rcu_iw, rdp->cpu); |
1237 | } | |
1238 | } | |
4914950a | 1239 | |
a82dcc76 | 1240 | return 0; |
64db4cff PM |
1241 | } |
1242 | ||
64db4cff PM |
1243 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
1244 | { | |
cb1e78cf | 1245 | unsigned long j = jiffies; |
6193c76a | 1246 | unsigned long j1; |
26cdfedf PM |
1247 | |
1248 | rsp->gp_start = j; | |
1249 | smp_wmb(); /* Record start time before stall time. */ | |
6193c76a | 1250 | j1 = rcu_jiffies_till_stall_check(); |
7d0ae808 | 1251 | WRITE_ONCE(rsp->jiffies_stall, j + j1); |
6193c76a | 1252 | rsp->jiffies_resched = j + j1 / 2; |
7d0ae808 | 1253 | rsp->n_force_qs_gpstart = READ_ONCE(rsp->n_force_qs); |
64db4cff PM |
1254 | } |
1255 | ||
6b50e119 PM |
1256 | /* |
1257 | * Convert a ->gp_state value to a character string. | |
1258 | */ | |
1259 | static const char *gp_state_getname(short gs) | |
1260 | { | |
1261 | if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names)) | |
1262 | return "???"; | |
1263 | return gp_state_names[gs]; | |
1264 | } | |
1265 | ||
fb81a44b PM |
1266 | /* |
1267 | * Complain about starvation of grace-period kthread. | |
1268 | */ | |
1269 | static void rcu_check_gp_kthread_starvation(struct rcu_state *rsp) | |
1270 | { | |
1271 | unsigned long gpa; | |
1272 | unsigned long j; | |
1273 | ||
1274 | j = jiffies; | |
7d0ae808 | 1275 | gpa = READ_ONCE(rsp->gp_activity); |
b1adb3e2 | 1276 | if (j - gpa > 2 * HZ) { |
78c5a67f | 1277 | pr_err("%s kthread starved for %ld jiffies! g%ld f%#x %s(%d) ->state=%#lx ->cpu=%d\n", |
81e701e4 | 1278 | rsp->name, j - gpa, |
78c5a67f | 1279 | (long)rcu_seq_current(&rsp->gp_seq), |
6b50e119 PM |
1280 | rsp->gp_flags, |
1281 | gp_state_getname(rsp->gp_state), rsp->gp_state, | |
96036c43 PM |
1282 | rsp->gp_kthread ? rsp->gp_kthread->state : ~0, |
1283 | rsp->gp_kthread ? task_cpu(rsp->gp_kthread) : -1); | |
86057b80 | 1284 | if (rsp->gp_kthread) { |
d07aee2c | 1285 | pr_err("RCU grace-period kthread stack dump:\n"); |
b1adb3e2 | 1286 | sched_show_task(rsp->gp_kthread); |
86057b80 PM |
1287 | wake_up_process(rsp->gp_kthread); |
1288 | } | |
b1adb3e2 | 1289 | } |
64db4cff PM |
1290 | } |
1291 | ||
b637a328 | 1292 | /* |
7aa92230 PM |
1293 | * Dump stacks of all tasks running on stalled CPUs. First try using |
1294 | * NMIs, but fall back to manual remote stack tracing on architectures | |
1295 | * that don't support NMI-based stack dumps. The NMI-triggered stack | |
1296 | * traces are more accurate because they are printed by the target CPU. | |
b637a328 PM |
1297 | */ |
1298 | static void rcu_dump_cpu_stacks(struct rcu_state *rsp) | |
1299 | { | |
1300 | int cpu; | |
1301 | unsigned long flags; | |
1302 | struct rcu_node *rnp; | |
1303 | ||
1304 | rcu_for_each_leaf_node(rsp, rnp) { | |
6cf10081 | 1305 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7aa92230 PM |
1306 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1307 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) | |
1308 | if (!trigger_single_cpu_backtrace(cpu)) | |
bc75e999 | 1309 | dump_cpu_task(cpu); |
67c583a7 | 1310 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b637a328 PM |
1311 | } |
1312 | } | |
1313 | ||
8c7c4829 PM |
1314 | /* |
1315 | * If too much time has passed in the current grace period, and if | |
1316 | * so configured, go kick the relevant kthreads. | |
1317 | */ | |
1318 | static void rcu_stall_kick_kthreads(struct rcu_state *rsp) | |
1319 | { | |
1320 | unsigned long j; | |
1321 | ||
1322 | if (!rcu_kick_kthreads) | |
1323 | return; | |
1324 | j = READ_ONCE(rsp->jiffies_kick_kthreads); | |
aa3e0bf1 PM |
1325 | if (time_after(jiffies, j) && rsp->gp_kthread && |
1326 | (rcu_gp_in_progress(rsp) || READ_ONCE(rsp->gp_flags))) { | |
8c7c4829 | 1327 | WARN_ONCE(1, "Kicking %s grace-period kthread\n", rsp->name); |
5dffed1e | 1328 | rcu_ftrace_dump(DUMP_ALL); |
8c7c4829 PM |
1329 | wake_up_process(rsp->gp_kthread); |
1330 | WRITE_ONCE(rsp->jiffies_kick_kthreads, j + HZ); | |
1331 | } | |
1332 | } | |
1333 | ||
088e9d25 DBO |
1334 | static inline void panic_on_rcu_stall(void) |
1335 | { | |
1336 | if (sysctl_panic_on_rcu_stall) | |
1337 | panic("RCU Stall\n"); | |
1338 | } | |
1339 | ||
471f87c3 | 1340 | static void print_other_cpu_stall(struct rcu_state *rsp, unsigned long gp_seq) |
64db4cff PM |
1341 | { |
1342 | int cpu; | |
64db4cff | 1343 | unsigned long flags; |
6ccd2ecd PM |
1344 | unsigned long gpa; |
1345 | unsigned long j; | |
285fe294 | 1346 | int ndetected = 0; |
64db4cff | 1347 | struct rcu_node *rnp = rcu_get_root(rsp); |
53bb857c | 1348 | long totqlen = 0; |
64db4cff | 1349 | |
8c7c4829 PM |
1350 | /* Kick and suppress, if so configured. */ |
1351 | rcu_stall_kick_kthreads(rsp); | |
1352 | if (rcu_cpu_stall_suppress) | |
1353 | return; | |
1354 | ||
8cdd32a9 PM |
1355 | /* |
1356 | * OK, time to rat on our buddy... | |
1357 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1358 | * RCU CPU stall warnings. | |
1359 | */ | |
d7f3e207 | 1360 | pr_err("INFO: %s detected stalls on CPUs/tasks:", |
4300aa64 | 1361 | rsp->name); |
a858af28 | 1362 | print_cpu_stall_info_begin(); |
a0b6c9a7 | 1363 | rcu_for_each_leaf_node(rsp, rnp) { |
6cf10081 | 1364 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
9bc8b558 | 1365 | ndetected += rcu_print_task_stall(rnp); |
c8020a67 | 1366 | if (rnp->qsmask != 0) { |
bc75e999 MR |
1367 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1368 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) { | |
1369 | print_cpu_stall_info(rsp, cpu); | |
c8020a67 PM |
1370 | ndetected++; |
1371 | } | |
1372 | } | |
67c583a7 | 1373 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1374 | } |
a858af28 | 1375 | |
a858af28 | 1376 | print_cpu_stall_info_end(); |
53bb857c | 1377 | for_each_possible_cpu(cpu) |
15fecf89 PM |
1378 | totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(rsp->rda, |
1379 | cpu)->cblist); | |
471f87c3 | 1380 | pr_cont("(detected by %d, t=%ld jiffies, g=%ld, q=%lu)\n", |
eee05882 | 1381 | smp_processor_id(), (long)(jiffies - rsp->gp_start), |
471f87c3 | 1382 | (long)rcu_seq_current(&rsp->gp_seq), totqlen); |
6ccd2ecd | 1383 | if (ndetected) { |
b637a328 | 1384 | rcu_dump_cpu_stacks(rsp); |
c4402b27 BP |
1385 | |
1386 | /* Complain about tasks blocking the grace period. */ | |
1387 | rcu_print_detail_task_stall(rsp); | |
6ccd2ecd | 1388 | } else { |
471f87c3 | 1389 | if (rcu_seq_current(&rsp->gp_seq) != gp_seq) { |
6ccd2ecd PM |
1390 | pr_err("INFO: Stall ended before state dump start\n"); |
1391 | } else { | |
1392 | j = jiffies; | |
7d0ae808 | 1393 | gpa = READ_ONCE(rsp->gp_activity); |
237a0f21 | 1394 | pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n", |
6ccd2ecd | 1395 | rsp->name, j - gpa, j, gpa, |
237a0f21 PM |
1396 | jiffies_till_next_fqs, |
1397 | rcu_get_root(rsp)->qsmask); | |
6ccd2ecd PM |
1398 | /* In this case, the current CPU might be at fault. */ |
1399 | sched_show_task(current); | |
1400 | } | |
1401 | } | |
8c42b1f3 PM |
1402 | /* Rewrite if needed in case of slow consoles. */ |
1403 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall))) | |
1404 | WRITE_ONCE(rsp->jiffies_stall, | |
1405 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
c1dc0b9c | 1406 | |
fb81a44b PM |
1407 | rcu_check_gp_kthread_starvation(rsp); |
1408 | ||
088e9d25 DBO |
1409 | panic_on_rcu_stall(); |
1410 | ||
4cdfc175 | 1411 | force_quiescent_state(rsp); /* Kick them all. */ |
64db4cff PM |
1412 | } |
1413 | ||
1414 | static void print_cpu_stall(struct rcu_state *rsp) | |
1415 | { | |
53bb857c | 1416 | int cpu; |
64db4cff | 1417 | unsigned long flags; |
9b9500da | 1418 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 1419 | struct rcu_node *rnp = rcu_get_root(rsp); |
53bb857c | 1420 | long totqlen = 0; |
64db4cff | 1421 | |
8c7c4829 PM |
1422 | /* Kick and suppress, if so configured. */ |
1423 | rcu_stall_kick_kthreads(rsp); | |
1424 | if (rcu_cpu_stall_suppress) | |
1425 | return; | |
1426 | ||
8cdd32a9 PM |
1427 | /* |
1428 | * OK, time to rat on ourselves... | |
1429 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1430 | * RCU CPU stall warnings. | |
1431 | */ | |
d7f3e207 | 1432 | pr_err("INFO: %s self-detected stall on CPU", rsp->name); |
a858af28 | 1433 | print_cpu_stall_info_begin(); |
9b9500da | 1434 | raw_spin_lock_irqsave_rcu_node(rdp->mynode, flags); |
a858af28 | 1435 | print_cpu_stall_info(rsp, smp_processor_id()); |
9b9500da | 1436 | raw_spin_unlock_irqrestore_rcu_node(rdp->mynode, flags); |
a858af28 | 1437 | print_cpu_stall_info_end(); |
53bb857c | 1438 | for_each_possible_cpu(cpu) |
15fecf89 PM |
1439 | totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(rsp->rda, |
1440 | cpu)->cblist); | |
471f87c3 | 1441 | pr_cont(" (t=%lu jiffies g=%ld q=%lu)\n", |
83ebe63e | 1442 | jiffies - rsp->gp_start, |
471f87c3 | 1443 | (long)rcu_seq_current(&rsp->gp_seq), totqlen); |
fb81a44b PM |
1444 | |
1445 | rcu_check_gp_kthread_starvation(rsp); | |
1446 | ||
bc1dce51 | 1447 | rcu_dump_cpu_stacks(rsp); |
c1dc0b9c | 1448 | |
6cf10081 | 1449 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
8c42b1f3 | 1450 | /* Rewrite if needed in case of slow consoles. */ |
7d0ae808 PM |
1451 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall))) |
1452 | WRITE_ONCE(rsp->jiffies_stall, | |
1453 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
67c583a7 | 1454 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c1dc0b9c | 1455 | |
088e9d25 DBO |
1456 | panic_on_rcu_stall(); |
1457 | ||
b021fe3e PZ |
1458 | /* |
1459 | * Attempt to revive the RCU machinery by forcing a context switch. | |
1460 | * | |
1461 | * A context switch would normally allow the RCU state machine to make | |
1462 | * progress and it could be we're stuck in kernel space without context | |
1463 | * switches for an entirely unreasonable amount of time. | |
1464 | */ | |
1465 | resched_cpu(smp_processor_id()); | |
64db4cff PM |
1466 | } |
1467 | ||
1468 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
1469 | { | |
471f87c3 PM |
1470 | unsigned long gs1; |
1471 | unsigned long gs2; | |
26cdfedf | 1472 | unsigned long gps; |
bad6e139 | 1473 | unsigned long j; |
8c42b1f3 | 1474 | unsigned long jn; |
bad6e139 | 1475 | unsigned long js; |
64db4cff PM |
1476 | struct rcu_node *rnp; |
1477 | ||
8c7c4829 PM |
1478 | if ((rcu_cpu_stall_suppress && !rcu_kick_kthreads) || |
1479 | !rcu_gp_in_progress(rsp)) | |
c68de209 | 1480 | return; |
8c7c4829 | 1481 | rcu_stall_kick_kthreads(rsp); |
cb1e78cf | 1482 | j = jiffies; |
26cdfedf PM |
1483 | |
1484 | /* | |
1485 | * Lots of memory barriers to reject false positives. | |
1486 | * | |
471f87c3 PM |
1487 | * The idea is to pick up rsp->gp_seq, then rsp->jiffies_stall, |
1488 | * then rsp->gp_start, and finally another copy of rsp->gp_seq. | |
1489 | * These values are updated in the opposite order with memory | |
1490 | * barriers (or equivalent) during grace-period initialization | |
1491 | * and cleanup. Now, a false positive can occur if we get an new | |
1492 | * value of rsp->gp_start and a old value of rsp->jiffies_stall. | |
1493 | * But given the memory barriers, the only way that this can happen | |
1494 | * is if one grace period ends and another starts between these | |
1495 | * two fetches. This is detected by comparing the second fetch | |
1496 | * of rsp->gp_seq with the previous fetch from rsp->gp_seq. | |
26cdfedf PM |
1497 | * |
1498 | * Given this check, comparisons of jiffies, rsp->jiffies_stall, | |
1499 | * and rsp->gp_start suffice to forestall false positives. | |
1500 | */ | |
471f87c3 PM |
1501 | gs1 = READ_ONCE(rsp->gp_seq); |
1502 | smp_rmb(); /* Pick up ->gp_seq first... */ | |
7d0ae808 | 1503 | js = READ_ONCE(rsp->jiffies_stall); |
26cdfedf | 1504 | smp_rmb(); /* ...then ->jiffies_stall before the rest... */ |
7d0ae808 | 1505 | gps = READ_ONCE(rsp->gp_start); |
471f87c3 PM |
1506 | smp_rmb(); /* ...and finally ->gp_start before ->gp_seq again. */ |
1507 | gs2 = READ_ONCE(rsp->gp_seq); | |
1508 | if (gs1 != gs2 || | |
26cdfedf PM |
1509 | ULONG_CMP_LT(j, js) || |
1510 | ULONG_CMP_GE(gps, js)) | |
1511 | return; /* No stall or GP completed since entering function. */ | |
64db4cff | 1512 | rnp = rdp->mynode; |
8c42b1f3 | 1513 | jn = jiffies + 3 * rcu_jiffies_till_stall_check() + 3; |
c96ea7cf | 1514 | if (rcu_gp_in_progress(rsp) && |
8c42b1f3 PM |
1515 | (READ_ONCE(rnp->qsmask) & rdp->grpmask) && |
1516 | cmpxchg(&rsp->jiffies_stall, js, jn) == js) { | |
64db4cff PM |
1517 | |
1518 | /* We haven't checked in, so go dump stack. */ | |
1519 | print_cpu_stall(rsp); | |
1520 | ||
bad6e139 | 1521 | } else if (rcu_gp_in_progress(rsp) && |
8c42b1f3 PM |
1522 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY) && |
1523 | cmpxchg(&rsp->jiffies_stall, js, jn) == js) { | |
64db4cff | 1524 | |
bad6e139 | 1525 | /* They had a few time units to dump stack, so complain. */ |
471f87c3 | 1526 | print_other_cpu_stall(rsp, gs2); |
64db4cff PM |
1527 | } |
1528 | } | |
1529 | ||
53d84e00 PM |
1530 | /** |
1531 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
1532 | * | |
1533 | * Set the stall-warning timeout way off into the future, thus preventing | |
1534 | * any RCU CPU stall-warning messages from appearing in the current set of | |
1535 | * RCU grace periods. | |
1536 | * | |
1537 | * The caller must disable hard irqs. | |
1538 | */ | |
1539 | void rcu_cpu_stall_reset(void) | |
1540 | { | |
6ce75a23 PM |
1541 | struct rcu_state *rsp; |
1542 | ||
1543 | for_each_rcu_flavor(rsp) | |
7d0ae808 | 1544 | WRITE_ONCE(rsp->jiffies_stall, jiffies + ULONG_MAX / 2); |
53d84e00 PM |
1545 | } |
1546 | ||
41e80595 PM |
1547 | /* Trace-event wrapper function for trace_rcu_future_grace_period. */ |
1548 | static void trace_rcu_this_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
b73de91d | 1549 | unsigned long gp_seq_req, const char *s) |
0446be48 | 1550 | { |
b73de91d | 1551 | trace_rcu_future_grace_period(rdp->rsp->name, rnp->gp_seq, gp_seq_req, |
abd13fdd | 1552 | rnp->level, rnp->grplo, rnp->grphi, s); |
0446be48 PM |
1553 | } |
1554 | ||
1555 | /* | |
b73de91d | 1556 | * rcu_start_this_gp - Request the start of a particular grace period |
df2bf8f7 | 1557 | * @rnp_start: The leaf node of the CPU from which to start. |
b73de91d JF |
1558 | * @rdp: The rcu_data corresponding to the CPU from which to start. |
1559 | * @gp_seq_req: The gp_seq of the grace period to start. | |
1560 | * | |
41e80595 | 1561 | * Start the specified grace period, as needed to handle newly arrived |
0446be48 | 1562 | * callbacks. The required future grace periods are recorded in each |
7a1d0f23 | 1563 | * rcu_node structure's ->gp_seq_needed field. Returns true if there |
48a7639c | 1564 | * is reason to awaken the grace-period kthread. |
0446be48 | 1565 | * |
d5cd9685 PM |
1566 | * The caller must hold the specified rcu_node structure's ->lock, which |
1567 | * is why the caller is responsible for waking the grace-period kthread. | |
b73de91d JF |
1568 | * |
1569 | * Returns true if the GP thread needs to be awakened else false. | |
0446be48 | 1570 | */ |
df2bf8f7 | 1571 | static bool rcu_start_this_gp(struct rcu_node *rnp_start, struct rcu_data *rdp, |
b73de91d | 1572 | unsigned long gp_seq_req) |
0446be48 | 1573 | { |
48a7639c | 1574 | bool ret = false; |
d5cd9685 | 1575 | struct rcu_state *rsp = rdp->rsp; |
df2bf8f7 | 1576 | struct rcu_node *rnp; |
0446be48 PM |
1577 | |
1578 | /* | |
360e0da6 PM |
1579 | * Use funnel locking to either acquire the root rcu_node |
1580 | * structure's lock or bail out if the need for this grace period | |
df2bf8f7 JFG |
1581 | * has already been recorded -- or if that grace period has in |
1582 | * fact already started. If there is already a grace period in | |
1583 | * progress in a non-leaf node, no recording is needed because the | |
1584 | * end of the grace period will scan the leaf rcu_node structures. | |
1585 | * Note that rnp_start->lock must not be released. | |
0446be48 | 1586 | */ |
df2bf8f7 JFG |
1587 | raw_lockdep_assert_held_rcu_node(rnp_start); |
1588 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, TPS("Startleaf")); | |
1589 | for (rnp = rnp_start; 1; rnp = rnp->parent) { | |
1590 | if (rnp != rnp_start) | |
1591 | raw_spin_lock_rcu_node(rnp); | |
1592 | if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req) || | |
1593 | rcu_seq_started(&rnp->gp_seq, gp_seq_req) || | |
1594 | (rnp != rnp_start && | |
1595 | rcu_seq_state(rcu_seq_current(&rnp->gp_seq)))) { | |
1596 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, | |
b73de91d | 1597 | TPS("Prestarted")); |
360e0da6 PM |
1598 | goto unlock_out; |
1599 | } | |
df2bf8f7 | 1600 | rnp->gp_seq_needed = gp_seq_req; |
226ca5e7 | 1601 | if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) { |
a2165e41 | 1602 | /* |
226ca5e7 JFG |
1603 | * We just marked the leaf or internal node, and a |
1604 | * grace period is in progress, which means that | |
1605 | * rcu_gp_cleanup() will see the marking. Bail to | |
1606 | * reduce contention. | |
a2165e41 | 1607 | */ |
df2bf8f7 | 1608 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, |
b73de91d | 1609 | TPS("Startedleaf")); |
a2165e41 PM |
1610 | goto unlock_out; |
1611 | } | |
df2bf8f7 JFG |
1612 | if (rnp != rnp_start && rnp->parent != NULL) |
1613 | raw_spin_unlock_rcu_node(rnp); | |
1614 | if (!rnp->parent) | |
360e0da6 | 1615 | break; /* At root, and perhaps also leaf. */ |
0446be48 PM |
1616 | } |
1617 | ||
360e0da6 | 1618 | /* If GP already in progress, just leave, otherwise start one. */ |
29365e56 | 1619 | if (rcu_gp_in_progress(rsp)) { |
df2bf8f7 | 1620 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedleafroot")); |
0446be48 PM |
1621 | goto unlock_out; |
1622 | } | |
df2bf8f7 | 1623 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedroot")); |
360e0da6 | 1624 | WRITE_ONCE(rsp->gp_flags, rsp->gp_flags | RCU_GP_FLAG_INIT); |
26d950a9 | 1625 | rsp->gp_req_activity = jiffies; |
360e0da6 | 1626 | if (!rsp->gp_kthread) { |
df2bf8f7 | 1627 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("NoGPkthread")); |
360e0da6 | 1628 | goto unlock_out; |
0446be48 | 1629 | } |
477351f7 | 1630 | trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gp_seq), TPS("newreq")); |
360e0da6 | 1631 | ret = true; /* Caller must wake GP kthread. */ |
0446be48 | 1632 | unlock_out: |
ab5e869c | 1633 | /* Push furthest requested GP to leaf node and rcu_data structure. */ |
df2bf8f7 JFG |
1634 | if (ULONG_CMP_LT(gp_seq_req, rnp->gp_seq_needed)) { |
1635 | rnp_start->gp_seq_needed = rnp->gp_seq_needed; | |
1636 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
ab5e869c | 1637 | } |
df2bf8f7 JFG |
1638 | if (rnp != rnp_start) |
1639 | raw_spin_unlock_rcu_node(rnp); | |
48a7639c | 1640 | return ret; |
0446be48 PM |
1641 | } |
1642 | ||
1643 | /* | |
1644 | * Clean up any old requests for the just-ended grace period. Also return | |
d1e4f01d | 1645 | * whether any additional grace periods have been requested. |
0446be48 | 1646 | */ |
fb31340f | 1647 | static bool rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp) |
0446be48 | 1648 | { |
fb31340f | 1649 | bool needmore; |
0446be48 PM |
1650 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
1651 | ||
7a1d0f23 PM |
1652 | needmore = ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed); |
1653 | if (!needmore) | |
1654 | rnp->gp_seq_needed = rnp->gp_seq; /* Avoid counter wrap. */ | |
b73de91d | 1655 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq, |
41e80595 | 1656 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); |
0446be48 PM |
1657 | return needmore; |
1658 | } | |
1659 | ||
48a7639c PM |
1660 | /* |
1661 | * Awaken the grace-period kthread for the specified flavor of RCU. | |
1662 | * Don't do a self-awaken, and don't bother awakening when there is | |
1663 | * nothing for the grace-period kthread to do (as in several CPUs | |
1664 | * raced to awaken, and we lost), and finally don't try to awaken | |
1665 | * a kthread that has not yet been created. | |
1666 | */ | |
1667 | static void rcu_gp_kthread_wake(struct rcu_state *rsp) | |
1668 | { | |
1669 | if (current == rsp->gp_kthread || | |
7d0ae808 | 1670 | !READ_ONCE(rsp->gp_flags) || |
48a7639c PM |
1671 | !rsp->gp_kthread) |
1672 | return; | |
abedf8e2 | 1673 | swake_up(&rsp->gp_wq); |
48a7639c PM |
1674 | } |
1675 | ||
dc35c893 | 1676 | /* |
29365e56 PM |
1677 | * If there is room, assign a ->gp_seq number to any callbacks on this |
1678 | * CPU that have not already been assigned. Also accelerate any callbacks | |
1679 | * that were previously assigned a ->gp_seq number that has since proven | |
1680 | * to be too conservative, which can happen if callbacks get assigned a | |
1681 | * ->gp_seq number while RCU is idle, but with reference to a non-root | |
1682 | * rcu_node structure. This function is idempotent, so it does not hurt | |
1683 | * to call it repeatedly. Returns an flag saying that we should awaken | |
1684 | * the RCU grace-period kthread. | |
dc35c893 PM |
1685 | * |
1686 | * The caller must hold rnp->lock with interrupts disabled. | |
1687 | */ | |
48a7639c | 1688 | static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1689 | struct rcu_data *rdp) |
1690 | { | |
b73de91d | 1691 | unsigned long gp_seq_req; |
15fecf89 | 1692 | bool ret = false; |
dc35c893 | 1693 | |
a32e01ee | 1694 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1695 | |
15fecf89 PM |
1696 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1697 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1698 | return false; |
dc35c893 PM |
1699 | |
1700 | /* | |
15fecf89 PM |
1701 | * Callbacks are often registered with incomplete grace-period |
1702 | * information. Something about the fact that getting exact | |
1703 | * information requires acquiring a global lock... RCU therefore | |
1704 | * makes a conservative estimate of the grace period number at which | |
1705 | * a given callback will become ready to invoke. The following | |
1706 | * code checks this estimate and improves it when possible, thus | |
1707 | * accelerating callback invocation to an earlier grace-period | |
1708 | * number. | |
dc35c893 | 1709 | */ |
b73de91d JF |
1710 | gp_seq_req = rcu_seq_snap(&rsp->gp_seq); |
1711 | if (rcu_segcblist_accelerate(&rdp->cblist, gp_seq_req)) | |
1712 | ret = rcu_start_this_gp(rnp, rdp, gp_seq_req); | |
6d4b418c PM |
1713 | |
1714 | /* Trace depending on how much we were able to accelerate. */ | |
15fecf89 | 1715 | if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL)) |
477351f7 | 1716 | trace_rcu_grace_period(rsp->name, rdp->gp_seq, TPS("AccWaitCB")); |
6d4b418c | 1717 | else |
477351f7 | 1718 | trace_rcu_grace_period(rsp->name, rdp->gp_seq, TPS("AccReadyCB")); |
48a7639c | 1719 | return ret; |
dc35c893 PM |
1720 | } |
1721 | ||
e44e73ca PM |
1722 | /* |
1723 | * Similar to rcu_accelerate_cbs(), but does not require that the leaf | |
1724 | * rcu_node structure's ->lock be held. It consults the cached value | |
1725 | * of ->gp_seq_needed in the rcu_data structure, and if that indicates | |
1726 | * that a new grace-period request be made, invokes rcu_accelerate_cbs() | |
1727 | * while holding the leaf rcu_node structure's ->lock. | |
1728 | */ | |
1729 | static void rcu_accelerate_cbs_unlocked(struct rcu_state *rsp, | |
1730 | struct rcu_node *rnp, | |
1731 | struct rcu_data *rdp) | |
1732 | { | |
1733 | unsigned long c; | |
1734 | bool needwake; | |
1735 | ||
1736 | lockdep_assert_irqs_disabled(); | |
1737 | c = rcu_seq_snap(&rsp->gp_seq); | |
1738 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { | |
1739 | /* Old request still live, so mark recent callbacks. */ | |
1740 | (void)rcu_segcblist_accelerate(&rdp->cblist, c); | |
1741 | return; | |
1742 | } | |
1743 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
1744 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); | |
1745 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ | |
1746 | if (needwake) | |
1747 | rcu_gp_kthread_wake(rsp); | |
1748 | } | |
1749 | ||
dc35c893 PM |
1750 | /* |
1751 | * Move any callbacks whose grace period has completed to the | |
1752 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
29365e56 | 1753 | * assign ->gp_seq numbers to any callbacks in the RCU_NEXT_TAIL |
dc35c893 PM |
1754 | * sublist. This function is idempotent, so it does not hurt to |
1755 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1756 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1757 | * |
1758 | * The caller must hold rnp->lock with interrupts disabled. | |
1759 | */ | |
48a7639c | 1760 | static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1761 | struct rcu_data *rdp) |
1762 | { | |
a32e01ee | 1763 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1764 | |
15fecf89 PM |
1765 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1766 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1767 | return false; |
dc35c893 PM |
1768 | |
1769 | /* | |
29365e56 | 1770 | * Find all callbacks whose ->gp_seq numbers indicate that they |
dc35c893 PM |
1771 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. |
1772 | */ | |
29365e56 | 1773 | rcu_segcblist_advance(&rdp->cblist, rnp->gp_seq); |
dc35c893 PM |
1774 | |
1775 | /* Classify any remaining callbacks. */ | |
48a7639c | 1776 | return rcu_accelerate_cbs(rsp, rnp, rdp); |
dc35c893 PM |
1777 | } |
1778 | ||
d09b62df | 1779 | /* |
ba9fbe95 PM |
1780 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1781 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1782 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1783 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1784 | */ |
48a7639c PM |
1785 | static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, |
1786 | struct rcu_data *rdp) | |
d09b62df | 1787 | { |
48a7639c | 1788 | bool ret; |
3563a438 | 1789 | bool need_gp; |
48a7639c | 1790 | |
a32e01ee | 1791 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1792 | |
67e14c1e PM |
1793 | if (rdp->gp_seq == rnp->gp_seq) |
1794 | return false; /* Nothing to do. */ | |
d09b62df | 1795 | |
67e14c1e PM |
1796 | /* Handle the ends of any preceding grace periods first. */ |
1797 | if (rcu_seq_completed_gp(rdp->gp_seq, rnp->gp_seq) || | |
1798 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
1799 | ret = rcu_advance_cbs(rsp, rnp, rdp); /* Advance callbacks. */ | |
67e14c1e PM |
1800 | trace_rcu_grace_period(rsp->name, rdp->gp_seq, TPS("cpuend")); |
1801 | } else { | |
1802 | ret = rcu_accelerate_cbs(rsp, rnp, rdp); /* Recent callbacks. */ | |
d09b62df | 1803 | } |
398ebe60 | 1804 | |
67e14c1e PM |
1805 | /* Now handle the beginnings of any new-to-this-CPU grace periods. */ |
1806 | if (rcu_seq_new_gp(rdp->gp_seq, rnp->gp_seq) || | |
1807 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
6eaef633 PM |
1808 | /* |
1809 | * If the current grace period is waiting for this CPU, | |
1810 | * set up to detect a quiescent state, otherwise don't | |
1811 | * go looking for one. | |
1812 | */ | |
5ca0905f | 1813 | trace_rcu_grace_period(rsp->name, rnp->gp_seq, TPS("cpustart")); |
3563a438 PM |
1814 | need_gp = !!(rnp->qsmask & rdp->grpmask); |
1815 | rdp->cpu_no_qs.b.norm = need_gp; | |
9577df9a | 1816 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
3563a438 | 1817 | rdp->core_needs_qs = need_gp; |
6eaef633 PM |
1818 | zero_cpu_stall_ticks(rdp); |
1819 | } | |
67e14c1e | 1820 | rdp->gp_seq = rnp->gp_seq; /* Remember new grace-period state. */ |
3d18469a PM |
1821 | if (ULONG_CMP_GE(rnp->gp_seq_needed, rdp->gp_seq_needed) || rdp->gpwrap) |
1822 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
1823 | WRITE_ONCE(rdp->gpwrap, false); | |
1824 | rcu_gpnum_ovf(rnp, rdp); | |
48a7639c | 1825 | return ret; |
6eaef633 PM |
1826 | } |
1827 | ||
d34ea322 | 1828 | static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp) |
6eaef633 PM |
1829 | { |
1830 | unsigned long flags; | |
48a7639c | 1831 | bool needwake; |
6eaef633 PM |
1832 | struct rcu_node *rnp; |
1833 | ||
1834 | local_irq_save(flags); | |
1835 | rnp = rdp->mynode; | |
67e14c1e | 1836 | if ((rdp->gp_seq == rcu_seq_current(&rnp->gp_seq) && |
7d0ae808 | 1837 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ |
2a67e741 | 1838 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
1839 | local_irq_restore(flags); |
1840 | return; | |
1841 | } | |
48a7639c | 1842 | needwake = __note_gp_changes(rsp, rnp, rdp); |
67c583a7 | 1843 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c PM |
1844 | if (needwake) |
1845 | rcu_gp_kthread_wake(rsp); | |
6eaef633 PM |
1846 | } |
1847 | ||
0f41c0dd PM |
1848 | static void rcu_gp_slow(struct rcu_state *rsp, int delay) |
1849 | { | |
1850 | if (delay > 0 && | |
dee4f422 PM |
1851 | !(rcu_seq_ctr(rsp->gp_seq) % |
1852 | (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) | |
0f41c0dd PM |
1853 | schedule_timeout_uninterruptible(delay); |
1854 | } | |
1855 | ||
b3dbec76 | 1856 | /* |
45fed3e7 | 1857 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 1858 | */ |
45fed3e7 | 1859 | static bool rcu_gp_init(struct rcu_state *rsp) |
b3dbec76 | 1860 | { |
0aa04b05 | 1861 | unsigned long oldmask; |
b3dbec76 | 1862 | struct rcu_data *rdp; |
7fdefc10 | 1863 | struct rcu_node *rnp = rcu_get_root(rsp); |
b3dbec76 | 1864 | |
7d0ae808 | 1865 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 1866 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 | 1867 | if (!READ_ONCE(rsp->gp_flags)) { |
f7be8209 | 1868 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 1869 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1870 | return false; |
f7be8209 | 1871 | } |
7d0ae808 | 1872 | WRITE_ONCE(rsp->gp_flags, 0); /* Clear all flags: New grace period. */ |
b3dbec76 | 1873 | |
f7be8209 PM |
1874 | if (WARN_ON_ONCE(rcu_gp_in_progress(rsp))) { |
1875 | /* | |
1876 | * Grace period already in progress, don't start another. | |
1877 | * Not supposed to be able to happen. | |
1878 | */ | |
67c583a7 | 1879 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1880 | return false; |
7fdefc10 PM |
1881 | } |
1882 | ||
7fdefc10 | 1883 | /* Advance to a new grace period and initialize state. */ |
26cdfedf | 1884 | record_gp_stall_check_time(rsp); |
ff3bb6f4 | 1885 | /* Record GP times before starting GP, hence rcu_seq_start(). */ |
de30ad51 | 1886 | rcu_seq_start(&rsp->gp_seq); |
477351f7 | 1887 | trace_rcu_grace_period(rsp->name, rsp->gp_seq, TPS("start")); |
67c583a7 | 1888 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 1889 | |
0aa04b05 PM |
1890 | /* |
1891 | * Apply per-leaf buffered online and offline operations to the | |
1892 | * rcu_node tree. Note that this new grace period need not wait | |
1893 | * for subsequent online CPUs, and that quiescent-state forcing | |
1894 | * will handle subsequent offline CPUs. | |
1895 | */ | |
1896 | rcu_for_each_leaf_node(rsp, rnp) { | |
0f41c0dd | 1897 | rcu_gp_slow(rsp, gp_preinit_delay); |
2a67e741 | 1898 | raw_spin_lock_irq_rcu_node(rnp); |
0aa04b05 PM |
1899 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
1900 | !rnp->wait_blkd_tasks) { | |
1901 | /* Nothing to do on this leaf rcu_node structure. */ | |
67c583a7 | 1902 | raw_spin_unlock_irq_rcu_node(rnp); |
0aa04b05 PM |
1903 | continue; |
1904 | } | |
1905 | ||
1906 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
1907 | oldmask = rnp->qsmaskinit; | |
1908 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
1909 | ||
1910 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
1911 | if (!oldmask != !rnp->qsmaskinit) { | |
1912 | if (!oldmask) /* First online CPU for this rcu_node. */ | |
1913 | rcu_init_new_rnp(rnp); | |
1914 | else if (rcu_preempt_has_tasks(rnp)) /* blocked tasks */ | |
1915 | rnp->wait_blkd_tasks = true; | |
1916 | else /* Last offline CPU and can propagate. */ | |
1917 | rcu_cleanup_dead_rnp(rnp); | |
1918 | } | |
1919 | ||
1920 | /* | |
1921 | * If all waited-on tasks from prior grace period are | |
1922 | * done, and if all this rcu_node structure's CPUs are | |
1923 | * still offline, propagate up the rcu_node tree and | |
1924 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
1925 | * rcu_node structure's CPUs has since come back online, | |
1926 | * simply clear ->wait_blkd_tasks (but rcu_cleanup_dead_rnp() | |
1927 | * checks for this, so just call it unconditionally). | |
1928 | */ | |
1929 | if (rnp->wait_blkd_tasks && | |
1930 | (!rcu_preempt_has_tasks(rnp) || | |
1931 | rnp->qsmaskinit)) { | |
1932 | rnp->wait_blkd_tasks = false; | |
1933 | rcu_cleanup_dead_rnp(rnp); | |
1934 | } | |
1935 | ||
67c583a7 | 1936 | raw_spin_unlock_irq_rcu_node(rnp); |
0aa04b05 | 1937 | } |
7fdefc10 PM |
1938 | |
1939 | /* | |
1940 | * Set the quiescent-state-needed bits in all the rcu_node | |
1941 | * structures for all currently online CPUs in breadth-first order, | |
1942 | * starting from the root rcu_node structure, relying on the layout | |
1943 | * of the tree within the rsp->node[] array. Note that other CPUs | |
1944 | * will access only the leaves of the hierarchy, thus seeing that no | |
1945 | * grace period is in progress, at least until the corresponding | |
590d1757 | 1946 | * leaf node has been initialized. |
7fdefc10 PM |
1947 | * |
1948 | * The grace period cannot complete until the initialization | |
1949 | * process finishes, because this kthread handles both. | |
1950 | */ | |
1951 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
0f41c0dd | 1952 | rcu_gp_slow(rsp, gp_init_delay); |
2a67e741 | 1953 | raw_spin_lock_irq_rcu_node(rnp); |
b3dbec76 | 1954 | rdp = this_cpu_ptr(rsp->rda); |
7fdefc10 PM |
1955 | rcu_preempt_check_blocked_tasks(rnp); |
1956 | rnp->qsmask = rnp->qsmaskinit; | |
de30ad51 | 1957 | WRITE_ONCE(rnp->gp_seq, rsp->gp_seq); |
7fdefc10 | 1958 | if (rnp == rdp->mynode) |
48a7639c | 1959 | (void)__note_gp_changes(rsp, rnp, rdp); |
7fdefc10 | 1960 | rcu_preempt_boost_start_gp(rnp); |
477351f7 | 1961 | trace_rcu_grace_period_init(rsp->name, rnp->gp_seq, |
7fdefc10 PM |
1962 | rnp->level, rnp->grplo, |
1963 | rnp->grphi, rnp->qsmask); | |
67c583a7 | 1964 | raw_spin_unlock_irq_rcu_node(rnp); |
cee43939 | 1965 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 1966 | WRITE_ONCE(rsp->gp_activity, jiffies); |
7fdefc10 | 1967 | } |
b3dbec76 | 1968 | |
45fed3e7 | 1969 | return true; |
7fdefc10 | 1970 | } |
b3dbec76 | 1971 | |
b9a425cf | 1972 | /* |
d5374226 LR |
1973 | * Helper function for swait_event_idle() wakeup at force-quiescent-state |
1974 | * time. | |
b9a425cf PM |
1975 | */ |
1976 | static bool rcu_gp_fqs_check_wake(struct rcu_state *rsp, int *gfp) | |
1977 | { | |
1978 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1979 | ||
1980 | /* Someone like call_rcu() requested a force-quiescent-state scan. */ | |
1981 | *gfp = READ_ONCE(rsp->gp_flags); | |
1982 | if (*gfp & RCU_GP_FLAG_FQS) | |
1983 | return true; | |
1984 | ||
1985 | /* The current grace period has completed. */ | |
1986 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) | |
1987 | return true; | |
1988 | ||
1989 | return false; | |
1990 | } | |
1991 | ||
4cdfc175 PM |
1992 | /* |
1993 | * Do one round of quiescent-state forcing. | |
1994 | */ | |
77f81fe0 | 1995 | static void rcu_gp_fqs(struct rcu_state *rsp, bool first_time) |
4cdfc175 | 1996 | { |
4cdfc175 PM |
1997 | struct rcu_node *rnp = rcu_get_root(rsp); |
1998 | ||
7d0ae808 | 1999 | WRITE_ONCE(rsp->gp_activity, jiffies); |
4cdfc175 | 2000 | rsp->n_force_qs++; |
77f81fe0 | 2001 | if (first_time) { |
4cdfc175 | 2002 | /* Collect dyntick-idle snapshots. */ |
fe5ac724 | 2003 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
4cdfc175 PM |
2004 | } else { |
2005 | /* Handle dyntick-idle and offline CPUs. */ | |
fe5ac724 | 2006 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
4cdfc175 PM |
2007 | } |
2008 | /* Clear flag to prevent immediate re-entry. */ | |
7d0ae808 | 2009 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 2010 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 PM |
2011 | WRITE_ONCE(rsp->gp_flags, |
2012 | READ_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 2013 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 2014 | } |
4cdfc175 PM |
2015 | } |
2016 | ||
7fdefc10 PM |
2017 | /* |
2018 | * Clean up after the old grace period. | |
2019 | */ | |
4cdfc175 | 2020 | static void rcu_gp_cleanup(struct rcu_state *rsp) |
7fdefc10 PM |
2021 | { |
2022 | unsigned long gp_duration; | |
48a7639c | 2023 | bool needgp = false; |
de30ad51 | 2024 | unsigned long new_gp_seq; |
7fdefc10 PM |
2025 | struct rcu_data *rdp; |
2026 | struct rcu_node *rnp = rcu_get_root(rsp); | |
abedf8e2 | 2027 | struct swait_queue_head *sq; |
b3dbec76 | 2028 | |
7d0ae808 | 2029 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 2030 | raw_spin_lock_irq_rcu_node(rnp); |
7fdefc10 PM |
2031 | gp_duration = jiffies - rsp->gp_start; |
2032 | if (gp_duration > rsp->gp_max) | |
2033 | rsp->gp_max = gp_duration; | |
b3dbec76 | 2034 | |
7fdefc10 PM |
2035 | /* |
2036 | * We know the grace period is complete, but to everyone else | |
2037 | * it appears to still be ongoing. But it is also the case | |
2038 | * that to everyone else it looks like there is nothing that | |
2039 | * they can do to advance the grace period. It is therefore | |
2040 | * safe for us to drop the lock in order to mark the grace | |
2041 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 2042 | */ |
67c583a7 | 2043 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 2044 | |
5d4b8659 | 2045 | /* |
ff3bb6f4 PM |
2046 | * Propagate new ->gp_seq value to rcu_node structures so that |
2047 | * other CPUs don't have to wait until the start of the next grace | |
2048 | * period to process their callbacks. This also avoids some nasty | |
2049 | * RCU grace-period initialization races by forcing the end of | |
2050 | * the current grace period to be completely recorded in all of | |
2051 | * the rcu_node structures before the beginning of the next grace | |
2052 | * period is recorded in any of the rcu_node structures. | |
5d4b8659 | 2053 | */ |
de30ad51 PM |
2054 | new_gp_seq = rsp->gp_seq; |
2055 | rcu_seq_end(&new_gp_seq); | |
5d4b8659 | 2056 | rcu_for_each_node_breadth_first(rsp, rnp) { |
2a67e741 | 2057 | raw_spin_lock_irq_rcu_node(rnp); |
4bc8d555 PM |
2058 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
2059 | dump_blkd_tasks(rnp, 10); | |
5c60d25f | 2060 | WARN_ON_ONCE(rnp->qsmask); |
de30ad51 | 2061 | WRITE_ONCE(rnp->gp_seq, new_gp_seq); |
b11cc576 PM |
2062 | rdp = this_cpu_ptr(rsp->rda); |
2063 | if (rnp == rdp->mynode) | |
48a7639c | 2064 | needgp = __note_gp_changes(rsp, rnp, rdp) || needgp; |
78e4bc34 | 2065 | /* smp_mb() provided by prior unlock-lock pair. */ |
fb31340f | 2066 | needgp = rcu_future_gp_cleanup(rsp, rnp) || needgp; |
065bb78c | 2067 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 2068 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 2069 | rcu_nocb_gp_cleanup(sq); |
cee43939 | 2070 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 2071 | WRITE_ONCE(rsp->gp_activity, jiffies); |
0f41c0dd | 2072 | rcu_gp_slow(rsp, gp_cleanup_delay); |
7fdefc10 | 2073 | } |
5d4b8659 | 2074 | rnp = rcu_get_root(rsp); |
de30ad51 | 2075 | raw_spin_lock_irq_rcu_node(rnp); /* GP before rsp->gp_seq update. */ |
7fdefc10 | 2076 | |
765a3f4f | 2077 | /* Declare grace period done. */ |
de30ad51 | 2078 | rcu_seq_end(&rsp->gp_seq); |
477351f7 | 2079 | trace_rcu_grace_period(rsp->name, rsp->gp_seq, TPS("end")); |
77f81fe0 | 2080 | rsp->gp_state = RCU_GP_IDLE; |
fb31340f | 2081 | /* Check for GP requests since above loop. */ |
5d4b8659 | 2082 | rdp = this_cpu_ptr(rsp->rda); |
5b55072f | 2083 | if (!needgp && ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed)) { |
abd13fdd | 2084 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq_needed, |
41e80595 | 2085 | TPS("CleanupMore")); |
fb31340f PM |
2086 | needgp = true; |
2087 | } | |
48a7639c | 2088 | /* Advance CBs to reduce false positives below. */ |
384f77f4 | 2089 | if (!rcu_accelerate_cbs(rsp, rnp, rdp) && needgp) { |
7d0ae808 | 2090 | WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT); |
26d950a9 | 2091 | rsp->gp_req_activity = jiffies; |
477351f7 | 2092 | trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gp_seq), |
bb311ecc | 2093 | TPS("newreq")); |
18390aea PM |
2094 | } else { |
2095 | WRITE_ONCE(rsp->gp_flags, rsp->gp_flags & RCU_GP_FLAG_INIT); | |
bb311ecc | 2096 | } |
67c583a7 | 2097 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 PM |
2098 | } |
2099 | ||
2100 | /* | |
2101 | * Body of kthread that handles grace periods. | |
2102 | */ | |
2103 | static int __noreturn rcu_gp_kthread(void *arg) | |
2104 | { | |
77f81fe0 | 2105 | bool first_gp_fqs; |
88d6df61 | 2106 | int gf; |
d40011f6 | 2107 | unsigned long j; |
4cdfc175 | 2108 | int ret; |
7fdefc10 PM |
2109 | struct rcu_state *rsp = arg; |
2110 | struct rcu_node *rnp = rcu_get_root(rsp); | |
2111 | ||
5871968d | 2112 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
2113 | for (;;) { |
2114 | ||
2115 | /* Handle grace-period start. */ | |
2116 | for (;;) { | |
63c4db78 | 2117 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2118 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2119 | TPS("reqwait")); |
afea227f | 2120 | rsp->gp_state = RCU_GP_WAIT_GPS; |
d5374226 LR |
2121 | swait_event_idle(rsp->gp_wq, READ_ONCE(rsp->gp_flags) & |
2122 | RCU_GP_FLAG_INIT); | |
319362c9 | 2123 | rsp->gp_state = RCU_GP_DONE_GPS; |
78e4bc34 | 2124 | /* Locking provides needed memory barrier. */ |
f7be8209 | 2125 | if (rcu_gp_init(rsp)) |
7fdefc10 | 2126 | break; |
cee43939 | 2127 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 2128 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2129 | WARN_ON(signal_pending(current)); |
63c4db78 | 2130 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2131 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2132 | TPS("reqwaitsig")); |
7fdefc10 | 2133 | } |
cabc49c1 | 2134 | |
4cdfc175 | 2135 | /* Handle quiescent-state forcing. */ |
77f81fe0 | 2136 | first_gp_fqs = true; |
d40011f6 PM |
2137 | j = jiffies_till_first_fqs; |
2138 | if (j > HZ) { | |
2139 | j = HZ; | |
2140 | jiffies_till_first_fqs = HZ; | |
2141 | } | |
88d6df61 | 2142 | ret = 0; |
cabc49c1 | 2143 | for (;;) { |
8c7c4829 | 2144 | if (!ret) { |
88d6df61 | 2145 | rsp->jiffies_force_qs = jiffies + j; |
8c7c4829 PM |
2146 | WRITE_ONCE(rsp->jiffies_kick_kthreads, |
2147 | jiffies + 3 * j); | |
2148 | } | |
63c4db78 | 2149 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2150 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2151 | TPS("fqswait")); |
afea227f | 2152 | rsp->gp_state = RCU_GP_WAIT_FQS; |
d5374226 | 2153 | ret = swait_event_idle_timeout(rsp->gp_wq, |
b9a425cf | 2154 | rcu_gp_fqs_check_wake(rsp, &gf), j); |
32bb1c79 | 2155 | rsp->gp_state = RCU_GP_DOING_FQS; |
78e4bc34 | 2156 | /* Locking provides needed memory barriers. */ |
4cdfc175 | 2157 | /* If grace period done, leave loop. */ |
7d0ae808 | 2158 | if (!READ_ONCE(rnp->qsmask) && |
4cdfc175 | 2159 | !rcu_preempt_blocked_readers_cgp(rnp)) |
cabc49c1 | 2160 | break; |
4cdfc175 | 2161 | /* If time for quiescent-state forcing, do it. */ |
88d6df61 PM |
2162 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) || |
2163 | (gf & RCU_GP_FLAG_FQS)) { | |
63c4db78 | 2164 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2165 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2166 | TPS("fqsstart")); |
77f81fe0 PM |
2167 | rcu_gp_fqs(rsp, first_gp_fqs); |
2168 | first_gp_fqs = false; | |
63c4db78 | 2169 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2170 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2171 | TPS("fqsend")); |
cee43939 | 2172 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 2173 | WRITE_ONCE(rsp->gp_activity, jiffies); |
fcfd0a23 PM |
2174 | ret = 0; /* Force full wait till next FQS. */ |
2175 | j = jiffies_till_next_fqs; | |
2176 | if (j > HZ) { | |
2177 | j = HZ; | |
2178 | jiffies_till_next_fqs = HZ; | |
2179 | } else if (j < 1) { | |
2180 | j = 1; | |
2181 | jiffies_till_next_fqs = 1; | |
2182 | } | |
4cdfc175 PM |
2183 | } else { |
2184 | /* Deal with stray signal. */ | |
cee43939 | 2185 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 2186 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2187 | WARN_ON(signal_pending(current)); |
63c4db78 | 2188 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2189 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2190 | TPS("fqswaitsig")); |
fcfd0a23 PM |
2191 | ret = 1; /* Keep old FQS timing. */ |
2192 | j = jiffies; | |
2193 | if (time_after(jiffies, rsp->jiffies_force_qs)) | |
2194 | j = 1; | |
2195 | else | |
2196 | j = rsp->jiffies_force_qs - j; | |
d40011f6 | 2197 | } |
cabc49c1 | 2198 | } |
4cdfc175 PM |
2199 | |
2200 | /* Handle grace-period end. */ | |
319362c9 | 2201 | rsp->gp_state = RCU_GP_CLEANUP; |
4cdfc175 | 2202 | rcu_gp_cleanup(rsp); |
319362c9 | 2203 | rsp->gp_state = RCU_GP_CLEANED; |
b3dbec76 | 2204 | } |
b3dbec76 PM |
2205 | } |
2206 | ||
f41d911f | 2207 | /* |
8994515c PM |
2208 | * Report a full set of quiescent states to the specified rcu_state data |
2209 | * structure. Invoke rcu_gp_kthread_wake() to awaken the grace-period | |
2210 | * kthread if another grace period is required. Whether we wake | |
2211 | * the grace-period kthread or it awakens itself for the next round | |
2212 | * of quiescent-state forcing, that kthread will clean up after the | |
2213 | * just-completed grace period. Note that the caller must hold rnp->lock, | |
2214 | * which is released before return. | |
f41d911f | 2215 | */ |
d3f6bad3 | 2216 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 2217 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 2218 | { |
a32e01ee | 2219 | raw_lockdep_assert_held_rcu_node(rcu_get_root(rsp)); |
fc2219d4 | 2220 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
cd73ca21 | 2221 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 2222 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags); |
94d44776 | 2223 | rcu_gp_kthread_wake(rsp); |
f41d911f PM |
2224 | } |
2225 | ||
64db4cff | 2226 | /* |
d3f6bad3 PM |
2227 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
2228 | * Allows quiescent states for a group of CPUs to be reported at one go | |
2229 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
2230 | * must be represented by the same rcu_node structure (which need not be a |
2231 | * leaf rcu_node structure, though it often will be). The gps parameter | |
2232 | * is the grace-period snapshot, which means that the quiescent states | |
c9a24e2d | 2233 | * are valid only if rnp->gp_seq is equal to gps. That structure's lock |
654e9533 | 2234 | * must be held upon entry, and it is released before return. |
64db4cff PM |
2235 | */ |
2236 | static void | |
d3f6bad3 | 2237 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
654e9533 | 2238 | struct rcu_node *rnp, unsigned long gps, unsigned long flags) |
64db4cff PM |
2239 | __releases(rnp->lock) |
2240 | { | |
654e9533 | 2241 | unsigned long oldmask = 0; |
28ecd580 PM |
2242 | struct rcu_node *rnp_c; |
2243 | ||
a32e01ee | 2244 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 2245 | |
64db4cff PM |
2246 | /* Walk up the rcu_node hierarchy. */ |
2247 | for (;;) { | |
c9a24e2d | 2248 | if (!(rnp->qsmask & mask) || rnp->gp_seq != gps) { |
64db4cff | 2249 | |
654e9533 PM |
2250 | /* |
2251 | * Our bit has already been cleared, or the | |
2252 | * relevant grace period is already over, so done. | |
2253 | */ | |
67c583a7 | 2254 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2255 | return; |
2256 | } | |
654e9533 | 2257 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
5b4c11d5 | 2258 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp) && |
2dee9404 | 2259 | rcu_preempt_blocked_readers_cgp(rnp)); |
64db4cff | 2260 | rnp->qsmask &= ~mask; |
db023296 | 2261 | trace_rcu_quiescent_state_report(rsp->name, rnp->gp_seq, |
d4c08f2a PM |
2262 | mask, rnp->qsmask, rnp->level, |
2263 | rnp->grplo, rnp->grphi, | |
2264 | !!rnp->gp_tasks); | |
27f4d280 | 2265 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
2266 | |
2267 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 2268 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2269 | return; |
2270 | } | |
d43a5d32 | 2271 | rnp->completedqs = rnp->gp_seq; |
64db4cff PM |
2272 | mask = rnp->grpmask; |
2273 | if (rnp->parent == NULL) { | |
2274 | ||
2275 | /* No more levels. Exit loop holding root lock. */ | |
2276 | ||
2277 | break; | |
2278 | } | |
67c583a7 | 2279 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 2280 | rnp_c = rnp; |
64db4cff | 2281 | rnp = rnp->parent; |
2a67e741 | 2282 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
654e9533 | 2283 | oldmask = rnp_c->qsmask; |
64db4cff PM |
2284 | } |
2285 | ||
2286 | /* | |
2287 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 2288 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 2289 | * to clean up and start the next grace period if one is needed. |
64db4cff | 2290 | */ |
d3f6bad3 | 2291 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
2292 | } |
2293 | ||
cc99a310 PM |
2294 | /* |
2295 | * Record a quiescent state for all tasks that were previously queued | |
2296 | * on the specified rcu_node structure and that were blocking the current | |
2297 | * RCU grace period. The caller must hold the specified rnp->lock with | |
2298 | * irqs disabled, and this lock is released upon return, but irqs remain | |
2299 | * disabled. | |
2300 | */ | |
0aa04b05 | 2301 | static void rcu_report_unblock_qs_rnp(struct rcu_state *rsp, |
cc99a310 PM |
2302 | struct rcu_node *rnp, unsigned long flags) |
2303 | __releases(rnp->lock) | |
2304 | { | |
654e9533 | 2305 | unsigned long gps; |
cc99a310 PM |
2306 | unsigned long mask; |
2307 | struct rcu_node *rnp_p; | |
2308 | ||
a32e01ee | 2309 | raw_lockdep_assert_held_rcu_node(rnp); |
a77da14c PM |
2310 | if (rcu_state_p == &rcu_sched_state || rsp != rcu_state_p || |
2311 | rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { | |
67c583a7 | 2312 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
2313 | return; /* Still need more quiescent states! */ |
2314 | } | |
2315 | ||
2316 | rnp_p = rnp->parent; | |
2317 | if (rnp_p == NULL) { | |
2318 | /* | |
a77da14c PM |
2319 | * Only one rcu_node structure in the tree, so don't |
2320 | * try to report up to its nonexistent parent! | |
cc99a310 PM |
2321 | */ |
2322 | rcu_report_qs_rsp(rsp, flags); | |
2323 | return; | |
2324 | } | |
2325 | ||
c9a24e2d PM |
2326 | /* Report up the rest of the hierarchy, tracking current ->gp_seq. */ |
2327 | gps = rnp->gp_seq; | |
cc99a310 | 2328 | mask = rnp->grpmask; |
67c583a7 | 2329 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 2330 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
654e9533 | 2331 | rcu_report_qs_rnp(mask, rsp, rnp_p, gps, flags); |
cc99a310 PM |
2332 | } |
2333 | ||
64db4cff | 2334 | /* |
d3f6bad3 | 2335 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 2336 | * structure. This must be called from the specified CPU. |
64db4cff PM |
2337 | */ |
2338 | static void | |
d7d6a11e | 2339 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2340 | { |
2341 | unsigned long flags; | |
2342 | unsigned long mask; | |
48a7639c | 2343 | bool needwake; |
64db4cff PM |
2344 | struct rcu_node *rnp; |
2345 | ||
2346 | rnp = rdp->mynode; | |
2a67e741 | 2347 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
c9a24e2d PM |
2348 | if (rdp->cpu_no_qs.b.norm || rdp->gp_seq != rnp->gp_seq || |
2349 | rdp->gpwrap) { | |
64db4cff PM |
2350 | |
2351 | /* | |
e4cc1f22 PM |
2352 | * The grace period in which this quiescent state was |
2353 | * recorded has ended, so don't report it upwards. | |
2354 | * We will instead need a new quiescent state that lies | |
2355 | * within the current grace period. | |
64db4cff | 2356 | */ |
5b74c458 | 2357 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
9577df9a | 2358 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
67c583a7 | 2359 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2360 | return; |
2361 | } | |
2362 | mask = rdp->grpmask; | |
2363 | if ((rnp->qsmask & mask) == 0) { | |
67c583a7 | 2364 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2365 | } else { |
bb53e416 | 2366 | rdp->core_needs_qs = false; |
64db4cff PM |
2367 | |
2368 | /* | |
2369 | * This GP can't end until cpu checks in, so all of our | |
2370 | * callbacks can be processed during the next GP. | |
2371 | */ | |
48a7639c | 2372 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
64db4cff | 2373 | |
c9a24e2d | 2374 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gp_seq, flags); |
654e9533 | 2375 | /* ^^^ Released rnp->lock */ |
48a7639c PM |
2376 | if (needwake) |
2377 | rcu_gp_kthread_wake(rsp); | |
64db4cff PM |
2378 | } |
2379 | } | |
2380 | ||
2381 | /* | |
2382 | * Check to see if there is a new grace period of which this CPU | |
2383 | * is not yet aware, and if so, set up local rcu_data state for it. | |
2384 | * Otherwise, see if this CPU has just passed through its first | |
2385 | * quiescent state for this grace period, and record that fact if so. | |
2386 | */ | |
2387 | static void | |
2388 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
2389 | { | |
05eb552b PM |
2390 | /* Check for grace-period ends and beginnings. */ |
2391 | note_gp_changes(rsp, rdp); | |
64db4cff PM |
2392 | |
2393 | /* | |
2394 | * Does this CPU still need to do its part for current grace period? | |
2395 | * If no, return and let the other CPUs do their part as well. | |
2396 | */ | |
97c668b8 | 2397 | if (!rdp->core_needs_qs) |
64db4cff PM |
2398 | return; |
2399 | ||
2400 | /* | |
2401 | * Was there a quiescent state since the beginning of the grace | |
2402 | * period? If no, then exit and wait for the next call. | |
2403 | */ | |
3a19b46a | 2404 | if (rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2405 | return; |
2406 | ||
d3f6bad3 PM |
2407 | /* |
2408 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
2409 | * judge of that). | |
2410 | */ | |
d7d6a11e | 2411 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp); |
64db4cff PM |
2412 | } |
2413 | ||
b1420f1c PM |
2414 | /* |
2415 | * Trace the fact that this CPU is going offline. | |
2416 | */ | |
2417 | static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) | |
2418 | { | |
477351f7 | 2419 | RCU_TRACE(bool blkd;) |
88a4976d PM |
2420 | RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda);) |
2421 | RCU_TRACE(struct rcu_node *rnp = rdp->mynode;) | |
b1420f1c | 2422 | |
ea46351c PM |
2423 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2424 | return; | |
2425 | ||
477351f7 PM |
2426 | RCU_TRACE(blkd = !!(rnp->qsmask & rdp->grpmask);) |
2427 | trace_rcu_grace_period(rsp->name, rnp->gp_seq, | |
2428 | blkd ? TPS("cpuofl") : TPS("cpuofl-bgp")); | |
64db4cff PM |
2429 | } |
2430 | ||
8af3a5e7 PM |
2431 | /* |
2432 | * All CPUs for the specified rcu_node structure have gone offline, | |
2433 | * and all tasks that were preempted within an RCU read-side critical | |
2434 | * section while running on one of those CPUs have since exited their RCU | |
2435 | * read-side critical section. Some other CPU is reporting this fact with | |
2436 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2437 | * This function therefore goes up the tree of rcu_node structures, | |
2438 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2439 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
2440 | * updated | |
2441 | * | |
2442 | * This function does check that the specified rcu_node structure has | |
2443 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2444 | * prematurely. That said, invoking it after the fact will cost you | |
2445 | * a needless lock acquisition. So once it has done its work, don't | |
2446 | * invoke it again. | |
2447 | */ | |
2448 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2449 | { | |
2450 | long mask; | |
2451 | struct rcu_node *rnp = rnp_leaf; | |
2452 | ||
a32e01ee | 2453 | raw_lockdep_assert_held_rcu_node(rnp); |
ea46351c PM |
2454 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
2455 | rnp->qsmaskinit || rcu_preempt_has_tasks(rnp)) | |
8af3a5e7 PM |
2456 | return; |
2457 | for (;;) { | |
2458 | mask = rnp->grpmask; | |
2459 | rnp = rnp->parent; | |
2460 | if (!rnp) | |
2461 | break; | |
2a67e741 | 2462 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2463 | rnp->qsmaskinit &= ~mask; |
0aa04b05 | 2464 | rnp->qsmask &= ~mask; |
8af3a5e7 | 2465 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2466 | raw_spin_unlock_rcu_node(rnp); |
2467 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2468 | return; |
2469 | } | |
67c583a7 | 2470 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2471 | } |
2472 | } | |
2473 | ||
64db4cff | 2474 | /* |
e5601400 | 2475 | * The CPU has been completely removed, and some other CPU is reporting |
a58163d8 PM |
2476 | * this fact from process context. Do the remainder of the cleanup. |
2477 | * There can only be one CPU hotplug operation at a time, so no need for | |
2478 | * explicit locking. | |
64db4cff | 2479 | */ |
e5601400 | 2480 | static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) |
64db4cff | 2481 | { |
e5601400 | 2482 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
b1420f1c | 2483 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2484 | |
ea46351c PM |
2485 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2486 | return; | |
2487 | ||
2036d94a | 2488 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2489 | rcu_boost_kthread_setaffinity(rnp, -1); |
64db4cff PM |
2490 | } |
2491 | ||
64db4cff PM |
2492 | /* |
2493 | * Invoke any RCU callbacks that have made it to the end of their grace | |
2494 | * period. Thottle as specified by rdp->blimit. | |
2495 | */ | |
37c72e56 | 2496 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2497 | { |
2498 | unsigned long flags; | |
15fecf89 PM |
2499 | struct rcu_head *rhp; |
2500 | struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl); | |
2501 | long bl, count; | |
64db4cff | 2502 | |
dc35c893 | 2503 | /* If no callbacks are ready, just return. */ |
15fecf89 PM |
2504 | if (!rcu_segcblist_ready_cbs(&rdp->cblist)) { |
2505 | trace_rcu_batch_start(rsp->name, | |
2506 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
2507 | rcu_segcblist_n_cbs(&rdp->cblist), 0); | |
2508 | trace_rcu_batch_end(rsp->name, 0, | |
2509 | !rcu_segcblist_empty(&rdp->cblist), | |
4968c300 PM |
2510 | need_resched(), is_idle_task(current), |
2511 | rcu_is_callbacks_kthread()); | |
64db4cff | 2512 | return; |
29c00b4a | 2513 | } |
64db4cff PM |
2514 | |
2515 | /* | |
2516 | * Extract the list of ready callbacks, disabling to prevent | |
15fecf89 PM |
2517 | * races with call_rcu() from interrupt handlers. Leave the |
2518 | * callback counts, as rcu_barrier() needs to be conservative. | |
64db4cff PM |
2519 | */ |
2520 | local_irq_save(flags); | |
8146c4e2 | 2521 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 2522 | bl = rdp->blimit; |
15fecf89 PM |
2523 | trace_rcu_batch_start(rsp->name, rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2524 | rcu_segcblist_n_cbs(&rdp->cblist), bl); | |
2525 | rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl); | |
64db4cff PM |
2526 | local_irq_restore(flags); |
2527 | ||
2528 | /* Invoke callbacks. */ | |
15fecf89 PM |
2529 | rhp = rcu_cblist_dequeue(&rcl); |
2530 | for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) { | |
2531 | debug_rcu_head_unqueue(rhp); | |
2532 | if (__rcu_reclaim(rsp->name, rhp)) | |
2533 | rcu_cblist_dequeued_lazy(&rcl); | |
2534 | /* | |
2535 | * Stop only if limit reached and CPU has something to do. | |
2536 | * Note: The rcl structure counts down from zero. | |
2537 | */ | |
4b27f20b | 2538 | if (-rcl.len >= bl && |
dff1672d PM |
2539 | (need_resched() || |
2540 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
2541 | break; |
2542 | } | |
2543 | ||
2544 | local_irq_save(flags); | |
4b27f20b | 2545 | count = -rcl.len; |
8ef0f37e PM |
2546 | trace_rcu_batch_end(rsp->name, count, !!rcl.head, need_resched(), |
2547 | is_idle_task(current), rcu_is_callbacks_kthread()); | |
64db4cff | 2548 | |
15fecf89 PM |
2549 | /* Update counts and requeue any remaining callbacks. */ |
2550 | rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl); | |
b1420f1c | 2551 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
15fecf89 | 2552 | rcu_segcblist_insert_count(&rdp->cblist, &rcl); |
64db4cff PM |
2553 | |
2554 | /* Reinstate batch limit if we have worked down the excess. */ | |
15fecf89 PM |
2555 | count = rcu_segcblist_n_cbs(&rdp->cblist); |
2556 | if (rdp->blimit == LONG_MAX && count <= qlowmark) | |
64db4cff PM |
2557 | rdp->blimit = blimit; |
2558 | ||
37c72e56 | 2559 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
15fecf89 | 2560 | if (count == 0 && rdp->qlen_last_fqs_check != 0) { |
37c72e56 PM |
2561 | rdp->qlen_last_fqs_check = 0; |
2562 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
15fecf89 PM |
2563 | } else if (count < rdp->qlen_last_fqs_check - qhimark) |
2564 | rdp->qlen_last_fqs_check = count; | |
efd88b02 PM |
2565 | |
2566 | /* | |
2567 | * The following usually indicates a double call_rcu(). To track | |
2568 | * this down, try building with CONFIG_DEBUG_OBJECTS_RCU_HEAD=y. | |
2569 | */ | |
15fecf89 | 2570 | WARN_ON_ONCE(rcu_segcblist_empty(&rdp->cblist) != (count == 0)); |
37c72e56 | 2571 | |
64db4cff PM |
2572 | local_irq_restore(flags); |
2573 | ||
e0f23060 | 2574 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
15fecf89 | 2575 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2576 | invoke_rcu_core(); |
64db4cff PM |
2577 | } |
2578 | ||
2579 | /* | |
2580 | * Check to see if this CPU is in a non-context-switch quiescent state | |
2581 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 2582 | * Also schedule RCU core processing. |
64db4cff | 2583 | * |
9b2e4f18 | 2584 | * This function must be called from hardirq context. It is normally |
5403d367 | 2585 | * invoked from the scheduling-clock interrupt. |
64db4cff | 2586 | */ |
c3377c2d | 2587 | void rcu_check_callbacks(int user) |
64db4cff | 2588 | { |
f7f7bac9 | 2589 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
a858af28 | 2590 | increment_cpu_stall_ticks(); |
9b2e4f18 | 2591 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
64db4cff PM |
2592 | |
2593 | /* | |
2594 | * Get here if this CPU took its interrupt from user | |
2595 | * mode or from the idle loop, and if this is not a | |
2596 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 2597 | * a quiescent state, so note it. |
64db4cff PM |
2598 | * |
2599 | * No memory barrier is required here because both | |
d6714c22 PM |
2600 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
2601 | * variables that other CPUs neither access nor modify, | |
2602 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
2603 | */ |
2604 | ||
284a8c93 PM |
2605 | rcu_sched_qs(); |
2606 | rcu_bh_qs(); | |
64db4cff PM |
2607 | |
2608 | } else if (!in_softirq()) { | |
2609 | ||
2610 | /* | |
2611 | * Get here if this CPU did not take its interrupt from | |
2612 | * softirq, in other words, if it is not interrupting | |
2613 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 2614 | * critical section, so note it. |
64db4cff PM |
2615 | */ |
2616 | ||
284a8c93 | 2617 | rcu_bh_qs(); |
64db4cff | 2618 | } |
86aea0e6 | 2619 | rcu_preempt_check_callbacks(); |
e3950ecd | 2620 | if (rcu_pending()) |
a46e0899 | 2621 | invoke_rcu_core(); |
8315f422 PM |
2622 | if (user) |
2623 | rcu_note_voluntary_context_switch(current); | |
f7f7bac9 | 2624 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2625 | } |
2626 | ||
64db4cff PM |
2627 | /* |
2628 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
2629 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
2630 | * Also initiate boosting for any threads blocked on the root rcu_node. |
2631 | * | |
ee47eb9f | 2632 | * The caller must have suppressed start of new grace periods. |
64db4cff | 2633 | */ |
fe5ac724 | 2634 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *rsp)) |
64db4cff | 2635 | { |
64db4cff PM |
2636 | int cpu; |
2637 | unsigned long flags; | |
2638 | unsigned long mask; | |
a0b6c9a7 | 2639 | struct rcu_node *rnp; |
64db4cff | 2640 | |
a0b6c9a7 | 2641 | rcu_for_each_leaf_node(rsp, rnp) { |
cee43939 | 2642 | cond_resched_tasks_rcu_qs(); |
64db4cff | 2643 | mask = 0; |
2a67e741 | 2644 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
a0b6c9a7 | 2645 | if (rnp->qsmask == 0) { |
a77da14c PM |
2646 | if (rcu_state_p == &rcu_sched_state || |
2647 | rsp != rcu_state_p || | |
2648 | rcu_preempt_blocked_readers_cgp(rnp)) { | |
2649 | /* | |
2650 | * No point in scanning bits because they | |
2651 | * are all zero. But we might need to | |
2652 | * priority-boost blocked readers. | |
2653 | */ | |
2654 | rcu_initiate_boost(rnp, flags); | |
2655 | /* rcu_initiate_boost() releases rnp->lock */ | |
2656 | continue; | |
2657 | } | |
2658 | if (rnp->parent && | |
2659 | (rnp->parent->qsmask & rnp->grpmask)) { | |
2660 | /* | |
2661 | * Race between grace-period | |
2662 | * initialization and task exiting RCU | |
2663 | * read-side critical section: Report. | |
2664 | */ | |
2665 | rcu_report_unblock_qs_rnp(rsp, rnp, flags); | |
2666 | /* rcu_report_unblock_qs_rnp() rlses ->lock */ | |
2667 | continue; | |
2668 | } | |
64db4cff | 2669 | } |
bc75e999 MR |
2670 | for_each_leaf_node_possible_cpu(rnp, cpu) { |
2671 | unsigned long bit = leaf_node_cpu_bit(rnp, cpu); | |
0edd1b17 | 2672 | if ((rnp->qsmask & bit) != 0) { |
fe5ac724 | 2673 | if (f(per_cpu_ptr(rsp->rda, cpu))) |
0edd1b17 PM |
2674 | mask |= bit; |
2675 | } | |
64db4cff | 2676 | } |
45f014c5 | 2677 | if (mask != 0) { |
c9a24e2d PM |
2678 | /* Idle/offline CPUs, report (releases rnp->lock). */ |
2679 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gp_seq, flags); | |
0aa04b05 PM |
2680 | } else { |
2681 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 2682 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2683 | } |
64db4cff | 2684 | } |
64db4cff PM |
2685 | } |
2686 | ||
2687 | /* | |
2688 | * Force quiescent states on reluctant CPUs, and also detect which | |
2689 | * CPUs are in dyntick-idle mode. | |
2690 | */ | |
4cdfc175 | 2691 | static void force_quiescent_state(struct rcu_state *rsp) |
64db4cff PM |
2692 | { |
2693 | unsigned long flags; | |
394f2769 PM |
2694 | bool ret; |
2695 | struct rcu_node *rnp; | |
2696 | struct rcu_node *rnp_old = NULL; | |
2697 | ||
2698 | /* Funnel through hierarchy to reduce memory contention. */ | |
d860d403 | 2699 | rnp = __this_cpu_read(rsp->rda->mynode); |
394f2769 | 2700 | for (; rnp != NULL; rnp = rnp->parent) { |
7d0ae808 | 2701 | ret = (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) || |
394f2769 PM |
2702 | !raw_spin_trylock(&rnp->fqslock); |
2703 | if (rnp_old != NULL) | |
2704 | raw_spin_unlock(&rnp_old->fqslock); | |
d62df573 | 2705 | if (ret) |
394f2769 | 2706 | return; |
394f2769 PM |
2707 | rnp_old = rnp; |
2708 | } | |
2709 | /* rnp_old == rcu_get_root(rsp), rnp == NULL. */ | |
64db4cff | 2710 | |
394f2769 | 2711 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 2712 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 2713 | raw_spin_unlock(&rnp_old->fqslock); |
7d0ae808 | 2714 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
67c583a7 | 2715 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 2716 | return; /* Someone beat us to it. */ |
46a1e34e | 2717 | } |
7d0ae808 | 2718 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 2719 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
94d44776 | 2720 | rcu_gp_kthread_wake(rsp); |
64db4cff PM |
2721 | } |
2722 | ||
26d950a9 PM |
2723 | /* |
2724 | * This function checks for grace-period requests that fail to motivate | |
2725 | * RCU to come out of its idle mode. | |
2726 | */ | |
2727 | static void | |
2728 | rcu_check_gp_start_stall(struct rcu_state *rsp, struct rcu_node *rnp, | |
2729 | struct rcu_data *rdp) | |
2730 | { | |
2731 | unsigned long flags; | |
2732 | unsigned long j; | |
2733 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
2734 | static atomic_t warned = ATOMIC_INIT(0); | |
2735 | ||
7a1d0f23 PM |
2736 | if (!IS_ENABLED(CONFIG_PROVE_RCU) || rcu_gp_in_progress(rsp) || |
2737 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed)) | |
26d950a9 PM |
2738 | return; |
2739 | j = jiffies; /* Expensive access, and in common case don't get here. */ | |
2740 | if (time_before(j, READ_ONCE(rsp->gp_req_activity) + HZ) || | |
2741 | time_before(j, READ_ONCE(rsp->gp_activity) + HZ) || | |
2742 | atomic_read(&warned)) | |
2743 | return; | |
2744 | ||
2745 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
2746 | j = jiffies; | |
7a1d0f23 PM |
2747 | if (rcu_gp_in_progress(rsp) || |
2748 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) || | |
26d950a9 PM |
2749 | time_before(j, READ_ONCE(rsp->gp_req_activity) + HZ) || |
2750 | time_before(j, READ_ONCE(rsp->gp_activity) + HZ) || | |
2751 | atomic_read(&warned)) { | |
2752 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2753 | return; | |
2754 | } | |
2755 | /* Hold onto the leaf lock to make others see warned==1. */ | |
2756 | ||
2757 | if (rnp_root != rnp) | |
2758 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ | |
2759 | j = jiffies; | |
7a1d0f23 PM |
2760 | if (rcu_gp_in_progress(rsp) || |
2761 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) || | |
26d950a9 PM |
2762 | time_before(j, rsp->gp_req_activity + HZ) || |
2763 | time_before(j, rsp->gp_activity + HZ) || | |
2764 | atomic_xchg(&warned, 1)) { | |
2765 | raw_spin_unlock_rcu_node(rnp_root); /* irqs remain disabled. */ | |
2766 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2767 | return; | |
2768 | } | |
7a1d0f23 PM |
2769 | pr_alert("%s: g%ld->%ld gar:%lu ga:%lu f%#x %s->state:%#lx\n", |
2770 | __func__, (long)READ_ONCE(rsp->gp_seq), | |
2771 | (long)READ_ONCE(rnp_root->gp_seq_needed), | |
26d950a9 PM |
2772 | j - rsp->gp_req_activity, j - rsp->gp_activity, |
2773 | rsp->gp_flags, rsp->name, | |
2774 | rsp->gp_kthread ? rsp->gp_kthread->state : 0x1ffffL); | |
2775 | WARN_ON(1); | |
2776 | if (rnp_root != rnp) | |
2777 | raw_spin_unlock_rcu_node(rnp_root); | |
2778 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2779 | } | |
2780 | ||
64db4cff | 2781 | /* |
e0f23060 PM |
2782 | * This does the RCU core processing work for the specified rcu_state |
2783 | * and rcu_data structures. This may be called only from the CPU to | |
2784 | * whom the rdp belongs. | |
64db4cff PM |
2785 | */ |
2786 | static void | |
1bca8cf1 | 2787 | __rcu_process_callbacks(struct rcu_state *rsp) |
64db4cff PM |
2788 | { |
2789 | unsigned long flags; | |
fa07a58f | 2790 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
26d950a9 | 2791 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 2792 | |
50dc7def | 2793 | WARN_ON_ONCE(!rdp->beenonline); |
2e597558 | 2794 | |
64db4cff PM |
2795 | /* Update RCU state based on any recent quiescent states. */ |
2796 | rcu_check_quiescent_state(rsp, rdp); | |
2797 | ||
bd7af846 PM |
2798 | /* No grace period and unregistered callbacks? */ |
2799 | if (!rcu_gp_in_progress(rsp) && | |
2800 | rcu_segcblist_is_enabled(&rdp->cblist)) { | |
2801 | local_irq_save(flags); | |
e44e73ca PM |
2802 | if (!rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
2803 | rcu_accelerate_cbs_unlocked(rsp, rnp, rdp); | |
2804 | local_irq_restore(flags); | |
64db4cff PM |
2805 | } |
2806 | ||
26d950a9 PM |
2807 | rcu_check_gp_start_stall(rsp, rnp, rdp); |
2808 | ||
64db4cff | 2809 | /* If there are callbacks ready, invoke them. */ |
15fecf89 | 2810 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2811 | invoke_rcu_callbacks(rsp, rdp); |
96d3fd0d PM |
2812 | |
2813 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
2814 | do_nocb_deferred_wakeup(rdp); | |
09223371 SL |
2815 | } |
2816 | ||
64db4cff | 2817 | /* |
e0f23060 | 2818 | * Do RCU core processing for the current CPU. |
64db4cff | 2819 | */ |
0766f788 | 2820 | static __latent_entropy void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 2821 | { |
6ce75a23 PM |
2822 | struct rcu_state *rsp; |
2823 | ||
bfa00b4c PM |
2824 | if (cpu_is_offline(smp_processor_id())) |
2825 | return; | |
f7f7bac9 | 2826 | trace_rcu_utilization(TPS("Start RCU core")); |
6ce75a23 PM |
2827 | for_each_rcu_flavor(rsp) |
2828 | __rcu_process_callbacks(rsp); | |
f7f7bac9 | 2829 | trace_rcu_utilization(TPS("End RCU core")); |
64db4cff PM |
2830 | } |
2831 | ||
a26ac245 | 2832 | /* |
e0f23060 PM |
2833 | * Schedule RCU callback invocation. If the specified type of RCU |
2834 | * does not support RCU priority boosting, just do a direct call, | |
2835 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
924df8a0 | 2836 | * are running on the current CPU with softirqs disabled, the |
e0f23060 | 2837 | * rcu_cpu_kthread_task cannot disappear out from under us. |
a26ac245 | 2838 | */ |
a46e0899 | 2839 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 2840 | { |
7d0ae808 | 2841 | if (unlikely(!READ_ONCE(rcu_scheduler_fully_active))) |
b0d30417 | 2842 | return; |
a46e0899 PM |
2843 | if (likely(!rsp->boost)) { |
2844 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
2845 | return; |
2846 | } | |
a46e0899 | 2847 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
2848 | } |
2849 | ||
a46e0899 | 2850 | static void invoke_rcu_core(void) |
09223371 | 2851 | { |
b0f74036 PM |
2852 | if (cpu_online(smp_processor_id())) |
2853 | raise_softirq(RCU_SOFTIRQ); | |
09223371 SL |
2854 | } |
2855 | ||
29154c57 PM |
2856 | /* |
2857 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
2858 | */ | |
2859 | static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp, | |
2860 | struct rcu_head *head, unsigned long flags) | |
64db4cff | 2861 | { |
62fde6ed PM |
2862 | /* |
2863 | * If called from an extended quiescent state, invoke the RCU | |
2864 | * core in order to force a re-evaluation of RCU's idleness. | |
2865 | */ | |
9910affa | 2866 | if (!rcu_is_watching()) |
62fde6ed PM |
2867 | invoke_rcu_core(); |
2868 | ||
a16b7a69 | 2869 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 2870 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 2871 | return; |
64db4cff | 2872 | |
37c72e56 PM |
2873 | /* |
2874 | * Force the grace period if too many callbacks or too long waiting. | |
2875 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
2876 | * if some other CPU has recently done so. Also, don't bother | |
2877 | * invoking force_quiescent_state() if the newly enqueued callback | |
2878 | * is the only one waiting for a grace period to complete. | |
2879 | */ | |
15fecf89 PM |
2880 | if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) > |
2881 | rdp->qlen_last_fqs_check + qhimark)) { | |
b52573d2 PM |
2882 | |
2883 | /* Are we ignoring a completed grace period? */ | |
470716fc | 2884 | note_gp_changes(rsp, rdp); |
b52573d2 PM |
2885 | |
2886 | /* Start a new grace period if one not already started. */ | |
2887 | if (!rcu_gp_in_progress(rsp)) { | |
e44e73ca | 2888 | rcu_accelerate_cbs_unlocked(rsp, rdp->mynode, rdp); |
b52573d2 PM |
2889 | } else { |
2890 | /* Give the grace period a kick. */ | |
2891 | rdp->blimit = LONG_MAX; | |
2892 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
15fecf89 | 2893 | rcu_segcblist_first_pend_cb(&rdp->cblist) != head) |
4cdfc175 | 2894 | force_quiescent_state(rsp); |
b52573d2 | 2895 | rdp->n_force_qs_snap = rsp->n_force_qs; |
15fecf89 | 2896 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); |
b52573d2 | 2897 | } |
4cdfc175 | 2898 | } |
29154c57 PM |
2899 | } |
2900 | ||
ae150184 PM |
2901 | /* |
2902 | * RCU callback function to leak a callback. | |
2903 | */ | |
2904 | static void rcu_leak_callback(struct rcu_head *rhp) | |
2905 | { | |
2906 | } | |
2907 | ||
3fbfbf7a PM |
2908 | /* |
2909 | * Helper function for call_rcu() and friends. The cpu argument will | |
2910 | * normally be -1, indicating "currently running CPU". It may specify | |
2911 | * a CPU only if that CPU is a no-CBs CPU. Currently, only _rcu_barrier() | |
2912 | * is expected to specify a CPU. | |
2913 | */ | |
64db4cff | 2914 | static void |
b6a4ae76 | 2915 | __call_rcu(struct rcu_head *head, rcu_callback_t func, |
3fbfbf7a | 2916 | struct rcu_state *rsp, int cpu, bool lazy) |
64db4cff PM |
2917 | { |
2918 | unsigned long flags; | |
2919 | struct rcu_data *rdp; | |
2920 | ||
b8f2ed53 PM |
2921 | /* Misaligned rcu_head! */ |
2922 | WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1)); | |
2923 | ||
ae150184 | 2924 | if (debug_rcu_head_queue(head)) { |
fa3c6647 PM |
2925 | /* |
2926 | * Probable double call_rcu(), so leak the callback. | |
2927 | * Use rcu:rcu_callback trace event to find the previous | |
2928 | * time callback was passed to __call_rcu(). | |
2929 | */ | |
2930 | WARN_ONCE(1, "__call_rcu(): Double-freed CB %p->%pF()!!!\n", | |
2931 | head, head->func); | |
7d0ae808 | 2932 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
2933 | return; |
2934 | } | |
64db4cff PM |
2935 | head->func = func; |
2936 | head->next = NULL; | |
64db4cff | 2937 | local_irq_save(flags); |
394f99a9 | 2938 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
2939 | |
2940 | /* Add the callback to our list. */ | |
15fecf89 | 2941 | if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist)) || cpu != -1) { |
3fbfbf7a PM |
2942 | int offline; |
2943 | ||
2944 | if (cpu != -1) | |
2945 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
143da9c2 PM |
2946 | if (likely(rdp->mynode)) { |
2947 | /* Post-boot, so this should be for a no-CBs CPU. */ | |
2948 | offline = !__call_rcu_nocb(rdp, head, lazy, flags); | |
2949 | WARN_ON_ONCE(offline); | |
2950 | /* Offline CPU, _call_rcu() illegal, leak callback. */ | |
2951 | local_irq_restore(flags); | |
2952 | return; | |
2953 | } | |
2954 | /* | |
2955 | * Very early boot, before rcu_init(). Initialize if needed | |
2956 | * and then drop through to queue the callback. | |
2957 | */ | |
2958 | BUG_ON(cpu != -1); | |
34404ca8 | 2959 | WARN_ON_ONCE(!rcu_is_watching()); |
15fecf89 PM |
2960 | if (rcu_segcblist_empty(&rdp->cblist)) |
2961 | rcu_segcblist_init(&rdp->cblist); | |
0d8ee37e | 2962 | } |
15fecf89 PM |
2963 | rcu_segcblist_enqueue(&rdp->cblist, head, lazy); |
2964 | if (!lazy) | |
c57afe80 | 2965 | rcu_idle_count_callbacks_posted(); |
2655d57e | 2966 | |
d4c08f2a PM |
2967 | if (__is_kfree_rcu_offset((unsigned long)func)) |
2968 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
15fecf89 PM |
2969 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2970 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2971 | else |
15fecf89 PM |
2972 | trace_rcu_callback(rsp->name, head, |
2973 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
2974 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2975 | |
29154c57 PM |
2976 | /* Go handle any RCU core processing required. */ |
2977 | __call_rcu_core(rsp, rdp, head, flags); | |
64db4cff PM |
2978 | local_irq_restore(flags); |
2979 | } | |
2980 | ||
a68a2bb2 PM |
2981 | /** |
2982 | * call_rcu_sched() - Queue an RCU for invocation after sched grace period. | |
2983 | * @head: structure to be used for queueing the RCU updates. | |
2984 | * @func: actual callback function to be invoked after the grace period | |
2985 | * | |
2986 | * The callback function will be invoked some time after a full grace | |
2987 | * period elapses, in other words after all currently executing RCU | |
2988 | * read-side critical sections have completed. call_rcu_sched() assumes | |
2989 | * that the read-side critical sections end on enabling of preemption | |
2990 | * or on voluntary preemption. | |
27fdb35f PM |
2991 | * RCU read-side critical sections are delimited by: |
2992 | * | |
2993 | * - rcu_read_lock_sched() and rcu_read_unlock_sched(), OR | |
2994 | * - anything that disables preemption. | |
a68a2bb2 PM |
2995 | * |
2996 | * These may be nested. | |
2997 | * | |
2998 | * See the description of call_rcu() for more detailed information on | |
2999 | * memory ordering guarantees. | |
64db4cff | 3000 | */ |
b6a4ae76 | 3001 | void call_rcu_sched(struct rcu_head *head, rcu_callback_t func) |
64db4cff | 3002 | { |
3fbfbf7a | 3003 | __call_rcu(head, func, &rcu_sched_state, -1, 0); |
64db4cff | 3004 | } |
d6714c22 | 3005 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff | 3006 | |
a68a2bb2 PM |
3007 | /** |
3008 | * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period. | |
3009 | * @head: structure to be used for queueing the RCU updates. | |
3010 | * @func: actual callback function to be invoked after the grace period | |
3011 | * | |
3012 | * The callback function will be invoked some time after a full grace | |
3013 | * period elapses, in other words after all currently executing RCU | |
3014 | * read-side critical sections have completed. call_rcu_bh() assumes | |
3015 | * that the read-side critical sections end on completion of a softirq | |
3016 | * handler. This means that read-side critical sections in process | |
3017 | * context must not be interrupted by softirqs. This interface is to be | |
3018 | * used when most of the read-side critical sections are in softirq context. | |
27fdb35f PM |
3019 | * RCU read-side critical sections are delimited by: |
3020 | * | |
3021 | * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context, OR | |
3022 | * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. | |
3023 | * | |
3024 | * These may be nested. | |
a68a2bb2 PM |
3025 | * |
3026 | * See the description of call_rcu() for more detailed information on | |
3027 | * memory ordering guarantees. | |
64db4cff | 3028 | */ |
b6a4ae76 | 3029 | void call_rcu_bh(struct rcu_head *head, rcu_callback_t func) |
64db4cff | 3030 | { |
3fbfbf7a | 3031 | __call_rcu(head, func, &rcu_bh_state, -1, 0); |
64db4cff PM |
3032 | } |
3033 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
3034 | ||
495aa969 ACB |
3035 | /* |
3036 | * Queue an RCU callback for lazy invocation after a grace period. | |
3037 | * This will likely be later named something like "call_rcu_lazy()", | |
3038 | * but this change will require some way of tagging the lazy RCU | |
3039 | * callbacks in the list of pending callbacks. Until then, this | |
3040 | * function may only be called from __kfree_rcu(). | |
3041 | */ | |
3042 | void kfree_call_rcu(struct rcu_head *head, | |
b6a4ae76 | 3043 | rcu_callback_t func) |
495aa969 | 3044 | { |
e534165b | 3045 | __call_rcu(head, func, rcu_state_p, -1, 1); |
495aa969 ACB |
3046 | } |
3047 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
3048 | ||
6d813391 PM |
3049 | /* |
3050 | * Because a context switch is a grace period for RCU-sched and RCU-bh, | |
3051 | * any blocking grace-period wait automatically implies a grace period | |
3052 | * if there is only one CPU online at any point time during execution | |
3053 | * of either synchronize_sched() or synchronize_rcu_bh(). It is OK to | |
3054 | * occasionally incorrectly indicate that there are multiple CPUs online | |
3055 | * when there was in fact only one the whole time, as this just adds | |
3056 | * some overhead: RCU still operates correctly. | |
6d813391 PM |
3057 | */ |
3058 | static inline int rcu_blocking_is_gp(void) | |
3059 | { | |
95f0c1de PM |
3060 | int ret; |
3061 | ||
6d813391 | 3062 | might_sleep(); /* Check for RCU read-side critical section. */ |
95f0c1de PM |
3063 | preempt_disable(); |
3064 | ret = num_online_cpus() <= 1; | |
3065 | preempt_enable(); | |
3066 | return ret; | |
6d813391 PM |
3067 | } |
3068 | ||
6ebb237b PM |
3069 | /** |
3070 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
3071 | * | |
3072 | * Control will return to the caller some time after a full rcu-sched | |
3073 | * grace period has elapsed, in other words after all currently executing | |
3074 | * rcu-sched read-side critical sections have completed. These read-side | |
3075 | * critical sections are delimited by rcu_read_lock_sched() and | |
3076 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
3077 | * local_irq_disable(), and so on may be used in place of | |
3078 | * rcu_read_lock_sched(). | |
3079 | * | |
3080 | * This means that all preempt_disable code sequences, including NMI and | |
f0a0e6f2 PM |
3081 | * non-threaded hardware-interrupt handlers, in progress on entry will |
3082 | * have completed before this primitive returns. However, this does not | |
3083 | * guarantee that softirq handlers will have completed, since in some | |
3084 | * kernels, these handlers can run in process context, and can block. | |
3085 | * | |
3086 | * Note that this guarantee implies further memory-ordering guarantees. | |
3087 | * On systems with more than one CPU, when synchronize_sched() returns, | |
3088 | * each CPU is guaranteed to have executed a full memory barrier since the | |
3089 | * end of its last RCU-sched read-side critical section whose beginning | |
3090 | * preceded the call to synchronize_sched(). In addition, each CPU having | |
3091 | * an RCU read-side critical section that extends beyond the return from | |
3092 | * synchronize_sched() is guaranteed to have executed a full memory barrier | |
3093 | * after the beginning of synchronize_sched() and before the beginning of | |
3094 | * that RCU read-side critical section. Note that these guarantees include | |
3095 | * CPUs that are offline, idle, or executing in user mode, as well as CPUs | |
3096 | * that are executing in the kernel. | |
3097 | * | |
3098 | * Furthermore, if CPU A invoked synchronize_sched(), which returned | |
3099 | * to its caller on CPU B, then both CPU A and CPU B are guaranteed | |
3100 | * to have executed a full memory barrier during the execution of | |
3101 | * synchronize_sched() -- even if CPU A and CPU B are the same CPU (but | |
3102 | * again only if the system has more than one CPU). | |
6ebb237b PM |
3103 | */ |
3104 | void synchronize_sched(void) | |
3105 | { | |
f78f5b90 PM |
3106 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
3107 | lock_is_held(&rcu_lock_map) || | |
3108 | lock_is_held(&rcu_sched_lock_map), | |
3109 | "Illegal synchronize_sched() in RCU-sched read-side critical section"); | |
6ebb237b PM |
3110 | if (rcu_blocking_is_gp()) |
3111 | return; | |
5afff48b | 3112 | if (rcu_gp_is_expedited()) |
3705b88d AM |
3113 | synchronize_sched_expedited(); |
3114 | else | |
3115 | wait_rcu_gp(call_rcu_sched); | |
6ebb237b PM |
3116 | } |
3117 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
3118 | ||
3119 | /** | |
3120 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
3121 | * | |
3122 | * Control will return to the caller some time after a full rcu_bh grace | |
3123 | * period has elapsed, in other words after all currently executing rcu_bh | |
3124 | * read-side critical sections have completed. RCU read-side critical | |
3125 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
3126 | * and may be nested. | |
f0a0e6f2 PM |
3127 | * |
3128 | * See the description of synchronize_sched() for more detailed information | |
3129 | * on memory ordering guarantees. | |
6ebb237b PM |
3130 | */ |
3131 | void synchronize_rcu_bh(void) | |
3132 | { | |
f78f5b90 PM |
3133 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
3134 | lock_is_held(&rcu_lock_map) || | |
3135 | lock_is_held(&rcu_sched_lock_map), | |
3136 | "Illegal synchronize_rcu_bh() in RCU-bh read-side critical section"); | |
6ebb237b PM |
3137 | if (rcu_blocking_is_gp()) |
3138 | return; | |
5afff48b | 3139 | if (rcu_gp_is_expedited()) |
3705b88d AM |
3140 | synchronize_rcu_bh_expedited(); |
3141 | else | |
3142 | wait_rcu_gp(call_rcu_bh); | |
6ebb237b PM |
3143 | } |
3144 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
3145 | ||
765a3f4f PM |
3146 | /** |
3147 | * get_state_synchronize_rcu - Snapshot current RCU state | |
3148 | * | |
3149 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
3150 | * to determine whether or not a full grace period has elapsed in the | |
3151 | * meantime. | |
3152 | */ | |
3153 | unsigned long get_state_synchronize_rcu(void) | |
3154 | { | |
3155 | /* | |
3156 | * Any prior manipulation of RCU-protected data must happen | |
e4be81a2 | 3157 | * before the load from ->gp_seq. |
765a3f4f PM |
3158 | */ |
3159 | smp_mb(); /* ^^^ */ | |
e4be81a2 | 3160 | return rcu_seq_snap(&rcu_state_p->gp_seq); |
765a3f4f PM |
3161 | } |
3162 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
3163 | ||
3164 | /** | |
3165 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
3166 | * | |
3167 | * @oldstate: return value from earlier call to get_state_synchronize_rcu() | |
3168 | * | |
3169 | * If a full RCU grace period has elapsed since the earlier call to | |
3170 | * get_state_synchronize_rcu(), just return. Otherwise, invoke | |
3171 | * synchronize_rcu() to wait for a full grace period. | |
3172 | * | |
3173 | * Yes, this function does not take counter wrap into account. But | |
3174 | * counter wrap is harmless. If the counter wraps, we have waited for | |
3175 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3176 | * so waiting for one additional grace period should be just fine. | |
3177 | */ | |
3178 | void cond_synchronize_rcu(unsigned long oldstate) | |
3179 | { | |
e4be81a2 | 3180 | if (!rcu_seq_done(&rcu_state_p->gp_seq, oldstate)) |
765a3f4f | 3181 | synchronize_rcu(); |
e4be81a2 PM |
3182 | else |
3183 | smp_mb(); /* Ensure GP ends before subsequent accesses. */ | |
765a3f4f PM |
3184 | } |
3185 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
3186 | ||
24560056 PM |
3187 | /** |
3188 | * get_state_synchronize_sched - Snapshot current RCU-sched state | |
3189 | * | |
3190 | * Returns a cookie that is used by a later call to cond_synchronize_sched() | |
3191 | * to determine whether or not a full grace period has elapsed in the | |
3192 | * meantime. | |
3193 | */ | |
3194 | unsigned long get_state_synchronize_sched(void) | |
3195 | { | |
3196 | /* | |
3197 | * Any prior manipulation of RCU-protected data must happen | |
e4be81a2 | 3198 | * before the load from ->gp_seq. |
24560056 PM |
3199 | */ |
3200 | smp_mb(); /* ^^^ */ | |
e4be81a2 | 3201 | return rcu_seq_snap(&rcu_sched_state.gp_seq); |
24560056 PM |
3202 | } |
3203 | EXPORT_SYMBOL_GPL(get_state_synchronize_sched); | |
3204 | ||
3205 | /** | |
3206 | * cond_synchronize_sched - Conditionally wait for an RCU-sched grace period | |
3207 | * | |
3208 | * @oldstate: return value from earlier call to get_state_synchronize_sched() | |
3209 | * | |
3210 | * If a full RCU-sched grace period has elapsed since the earlier call to | |
3211 | * get_state_synchronize_sched(), just return. Otherwise, invoke | |
3212 | * synchronize_sched() to wait for a full grace period. | |
3213 | * | |
3214 | * Yes, this function does not take counter wrap into account. But | |
3215 | * counter wrap is harmless. If the counter wraps, we have waited for | |
3216 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3217 | * so waiting for one additional grace period should be just fine. | |
3218 | */ | |
3219 | void cond_synchronize_sched(unsigned long oldstate) | |
3220 | { | |
e4be81a2 | 3221 | if (!rcu_seq_done(&rcu_sched_state.gp_seq, oldstate)) |
24560056 | 3222 | synchronize_sched(); |
e4be81a2 PM |
3223 | else |
3224 | smp_mb(); /* Ensure GP ends before subsequent accesses. */ | |
24560056 PM |
3225 | } |
3226 | EXPORT_SYMBOL_GPL(cond_synchronize_sched); | |
3227 | ||
64db4cff PM |
3228 | /* |
3229 | * Check to see if there is any immediate RCU-related work to be done | |
3230 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
3231 | * The checks are in order of increasing expense: checks that can be | |
3232 | * carried out against CPU-local state are performed first. However, | |
3233 | * we must check for CPU stalls first, else we might not get a chance. | |
3234 | */ | |
3235 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
3236 | { | |
2f51f988 PM |
3237 | struct rcu_node *rnp = rdp->mynode; |
3238 | ||
64db4cff PM |
3239 | /* Check for CPU stalls, if enabled. */ |
3240 | check_cpu_stall(rsp, rdp); | |
3241 | ||
a096932f PM |
3242 | /* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */ |
3243 | if (rcu_nohz_full_cpu(rsp)) | |
3244 | return 0; | |
3245 | ||
64db4cff | 3246 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
01c495f7 | 3247 | if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) |
64db4cff PM |
3248 | return 1; |
3249 | ||
3250 | /* Does this CPU have callbacks ready to invoke? */ | |
01c495f7 | 3251 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
64db4cff PM |
3252 | return 1; |
3253 | ||
3254 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
c1935209 PM |
3255 | if (!rcu_gp_in_progress(rsp) && |
3256 | rcu_segcblist_is_enabled(&rdp->cblist) && | |
3257 | !rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) | |
64db4cff PM |
3258 | return 1; |
3259 | ||
67e14c1e PM |
3260 | /* Have RCU grace period completed or started? */ |
3261 | if (rcu_seq_current(&rnp->gp_seq) != rdp->gp_seq || | |
01c495f7 | 3262 | unlikely(READ_ONCE(rdp->gpwrap))) /* outside lock */ |
64db4cff PM |
3263 | return 1; |
3264 | ||
96d3fd0d | 3265 | /* Does this CPU need a deferred NOCB wakeup? */ |
01c495f7 | 3266 | if (rcu_nocb_need_deferred_wakeup(rdp)) |
96d3fd0d | 3267 | return 1; |
96d3fd0d | 3268 | |
64db4cff PM |
3269 | /* nothing to do */ |
3270 | return 0; | |
3271 | } | |
3272 | ||
3273 | /* | |
3274 | * Check to see if there is any immediate RCU-related work to be done | |
3275 | * by the current CPU, returning 1 if so. This function is part of the | |
3276 | * RCU implementation; it is -not- an exported member of the RCU API. | |
3277 | */ | |
e3950ecd | 3278 | static int rcu_pending(void) |
64db4cff | 3279 | { |
6ce75a23 PM |
3280 | struct rcu_state *rsp; |
3281 | ||
3282 | for_each_rcu_flavor(rsp) | |
e3950ecd | 3283 | if (__rcu_pending(rsp, this_cpu_ptr(rsp->rda))) |
6ce75a23 PM |
3284 | return 1; |
3285 | return 0; | |
64db4cff PM |
3286 | } |
3287 | ||
3288 | /* | |
c0f4dfd4 PM |
3289 | * Return true if the specified CPU has any callback. If all_lazy is |
3290 | * non-NULL, store an indication of whether all callbacks are lazy. | |
3291 | * (If there are no callbacks, all of them are deemed to be lazy.) | |
64db4cff | 3292 | */ |
82072c4f | 3293 | static bool __maybe_unused rcu_cpu_has_callbacks(bool *all_lazy) |
64db4cff | 3294 | { |
c0f4dfd4 PM |
3295 | bool al = true; |
3296 | bool hc = false; | |
3297 | struct rcu_data *rdp; | |
6ce75a23 PM |
3298 | struct rcu_state *rsp; |
3299 | ||
c0f4dfd4 | 3300 | for_each_rcu_flavor(rsp) { |
aa6da514 | 3301 | rdp = this_cpu_ptr(rsp->rda); |
15fecf89 | 3302 | if (rcu_segcblist_empty(&rdp->cblist)) |
69c8d28c PM |
3303 | continue; |
3304 | hc = true; | |
15fecf89 | 3305 | if (rcu_segcblist_n_nonlazy_cbs(&rdp->cblist) || !all_lazy) { |
c0f4dfd4 | 3306 | al = false; |
69c8d28c PM |
3307 | break; |
3308 | } | |
c0f4dfd4 PM |
3309 | } |
3310 | if (all_lazy) | |
3311 | *all_lazy = al; | |
3312 | return hc; | |
64db4cff PM |
3313 | } |
3314 | ||
a83eff0a PM |
3315 | /* |
3316 | * Helper function for _rcu_barrier() tracing. If tracing is disabled, | |
3317 | * the compiler is expected to optimize this away. | |
3318 | */ | |
e66c33d5 | 3319 | static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s, |
a83eff0a PM |
3320 | int cpu, unsigned long done) |
3321 | { | |
3322 | trace_rcu_barrier(rsp->name, s, cpu, | |
3323 | atomic_read(&rsp->barrier_cpu_count), done); | |
3324 | } | |
3325 | ||
b1420f1c PM |
3326 | /* |
3327 | * RCU callback function for _rcu_barrier(). If we are last, wake | |
3328 | * up the task executing _rcu_barrier(). | |
3329 | */ | |
24ebbca8 | 3330 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 3331 | { |
24ebbca8 PM |
3332 | struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head); |
3333 | struct rcu_state *rsp = rdp->rsp; | |
3334 | ||
a83eff0a | 3335 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) { |
d8db2e86 PM |
3336 | _rcu_barrier_trace(rsp, TPS("LastCB"), -1, |
3337 | rsp->barrier_sequence); | |
7db74df8 | 3338 | complete(&rsp->barrier_completion); |
a83eff0a | 3339 | } else { |
d8db2e86 | 3340 | _rcu_barrier_trace(rsp, TPS("CB"), -1, rsp->barrier_sequence); |
a83eff0a | 3341 | } |
d0ec774c PM |
3342 | } |
3343 | ||
3344 | /* | |
3345 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
3346 | */ | |
3347 | static void rcu_barrier_func(void *type) | |
3348 | { | |
037b64ed | 3349 | struct rcu_state *rsp = type; |
fa07a58f | 3350 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
d0ec774c | 3351 | |
d8db2e86 | 3352 | _rcu_barrier_trace(rsp, TPS("IRQ"), -1, rsp->barrier_sequence); |
f92c734f PM |
3353 | rdp->barrier_head.func = rcu_barrier_callback; |
3354 | debug_rcu_head_queue(&rdp->barrier_head); | |
3355 | if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head, 0)) { | |
3356 | atomic_inc(&rsp->barrier_cpu_count); | |
3357 | } else { | |
3358 | debug_rcu_head_unqueue(&rdp->barrier_head); | |
d8db2e86 PM |
3359 | _rcu_barrier_trace(rsp, TPS("IRQNQ"), -1, |
3360 | rsp->barrier_sequence); | |
f92c734f | 3361 | } |
d0ec774c PM |
3362 | } |
3363 | ||
d0ec774c PM |
3364 | /* |
3365 | * Orchestrate the specified type of RCU barrier, waiting for all | |
3366 | * RCU callbacks of the specified type to complete. | |
3367 | */ | |
037b64ed | 3368 | static void _rcu_barrier(struct rcu_state *rsp) |
d0ec774c | 3369 | { |
b1420f1c | 3370 | int cpu; |
b1420f1c | 3371 | struct rcu_data *rdp; |
4f525a52 | 3372 | unsigned long s = rcu_seq_snap(&rsp->barrier_sequence); |
b1420f1c | 3373 | |
d8db2e86 | 3374 | _rcu_barrier_trace(rsp, TPS("Begin"), -1, s); |
b1420f1c | 3375 | |
e74f4c45 | 3376 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
7be7f0be | 3377 | mutex_lock(&rsp->barrier_mutex); |
b1420f1c | 3378 | |
4f525a52 PM |
3379 | /* Did someone else do our work for us? */ |
3380 | if (rcu_seq_done(&rsp->barrier_sequence, s)) { | |
d8db2e86 PM |
3381 | _rcu_barrier_trace(rsp, TPS("EarlyExit"), -1, |
3382 | rsp->barrier_sequence); | |
cf3a9c48 PM |
3383 | smp_mb(); /* caller's subsequent code after above check. */ |
3384 | mutex_unlock(&rsp->barrier_mutex); | |
3385 | return; | |
3386 | } | |
3387 | ||
4f525a52 PM |
3388 | /* Mark the start of the barrier operation. */ |
3389 | rcu_seq_start(&rsp->barrier_sequence); | |
d8db2e86 | 3390 | _rcu_barrier_trace(rsp, TPS("Inc1"), -1, rsp->barrier_sequence); |
b1420f1c | 3391 | |
d0ec774c | 3392 | /* |
b1420f1c PM |
3393 | * Initialize the count to one rather than to zero in order to |
3394 | * avoid a too-soon return to zero in case of a short grace period | |
1331e7a1 PM |
3395 | * (or preemption of this task). Exclude CPU-hotplug operations |
3396 | * to ensure that no offline CPU has callbacks queued. | |
d0ec774c | 3397 | */ |
7db74df8 | 3398 | init_completion(&rsp->barrier_completion); |
24ebbca8 | 3399 | atomic_set(&rsp->barrier_cpu_count, 1); |
1331e7a1 | 3400 | get_online_cpus(); |
b1420f1c PM |
3401 | |
3402 | /* | |
1331e7a1 PM |
3403 | * Force each CPU with callbacks to register a new callback. |
3404 | * When that callback is invoked, we will know that all of the | |
3405 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 3406 | */ |
3fbfbf7a | 3407 | for_each_possible_cpu(cpu) { |
d1e43fa5 | 3408 | if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu)) |
3fbfbf7a | 3409 | continue; |
b1420f1c | 3410 | rdp = per_cpu_ptr(rsp->rda, cpu); |
d1e43fa5 | 3411 | if (rcu_is_nocb_cpu(cpu)) { |
d7e29933 | 3412 | if (!rcu_nocb_cpu_needs_barrier(rsp, cpu)) { |
d8db2e86 | 3413 | _rcu_barrier_trace(rsp, TPS("OfflineNoCB"), cpu, |
4f525a52 | 3414 | rsp->barrier_sequence); |
d7e29933 | 3415 | } else { |
d8db2e86 | 3416 | _rcu_barrier_trace(rsp, TPS("OnlineNoCB"), cpu, |
4f525a52 | 3417 | rsp->barrier_sequence); |
41050a00 | 3418 | smp_mb__before_atomic(); |
d7e29933 PM |
3419 | atomic_inc(&rsp->barrier_cpu_count); |
3420 | __call_rcu(&rdp->barrier_head, | |
3421 | rcu_barrier_callback, rsp, cpu, 0); | |
3422 | } | |
15fecf89 | 3423 | } else if (rcu_segcblist_n_cbs(&rdp->cblist)) { |
d8db2e86 | 3424 | _rcu_barrier_trace(rsp, TPS("OnlineQ"), cpu, |
4f525a52 | 3425 | rsp->barrier_sequence); |
037b64ed | 3426 | smp_call_function_single(cpu, rcu_barrier_func, rsp, 1); |
b1420f1c | 3427 | } else { |
d8db2e86 | 3428 | _rcu_barrier_trace(rsp, TPS("OnlineNQ"), cpu, |
4f525a52 | 3429 | rsp->barrier_sequence); |
b1420f1c PM |
3430 | } |
3431 | } | |
1331e7a1 | 3432 | put_online_cpus(); |
b1420f1c PM |
3433 | |
3434 | /* | |
3435 | * Now that we have an rcu_barrier_callback() callback on each | |
3436 | * CPU, and thus each counted, remove the initial count. | |
3437 | */ | |
24ebbca8 | 3438 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) |
7db74df8 | 3439 | complete(&rsp->barrier_completion); |
b1420f1c PM |
3440 | |
3441 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
7db74df8 | 3442 | wait_for_completion(&rsp->barrier_completion); |
b1420f1c | 3443 | |
4f525a52 | 3444 | /* Mark the end of the barrier operation. */ |
d8db2e86 | 3445 | _rcu_barrier_trace(rsp, TPS("Inc2"), -1, rsp->barrier_sequence); |
4f525a52 PM |
3446 | rcu_seq_end(&rsp->barrier_sequence); |
3447 | ||
b1420f1c | 3448 | /* Other rcu_barrier() invocations can now safely proceed. */ |
7be7f0be | 3449 | mutex_unlock(&rsp->barrier_mutex); |
d0ec774c | 3450 | } |
d0ec774c PM |
3451 | |
3452 | /** | |
3453 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
3454 | */ | |
3455 | void rcu_barrier_bh(void) | |
3456 | { | |
037b64ed | 3457 | _rcu_barrier(&rcu_bh_state); |
d0ec774c PM |
3458 | } |
3459 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
3460 | ||
3461 | /** | |
3462 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
3463 | */ | |
3464 | void rcu_barrier_sched(void) | |
3465 | { | |
037b64ed | 3466 | _rcu_barrier(&rcu_sched_state); |
d0ec774c PM |
3467 | } |
3468 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
3469 | ||
0aa04b05 PM |
3470 | /* |
3471 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
3472 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
3473 | * must hold the corresponding leaf rcu_node ->lock with interrrupts | |
3474 | * disabled. | |
3475 | */ | |
3476 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
3477 | { | |
3478 | long mask; | |
3479 | struct rcu_node *rnp = rnp_leaf; | |
3480 | ||
a32e01ee | 3481 | raw_lockdep_assert_held_rcu_node(rnp); |
0aa04b05 PM |
3482 | for (;;) { |
3483 | mask = rnp->grpmask; | |
3484 | rnp = rnp->parent; | |
3485 | if (rnp == NULL) | |
3486 | return; | |
6cf10081 | 3487 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
0aa04b05 | 3488 | rnp->qsmaskinit |= mask; |
67c583a7 | 3489 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
0aa04b05 PM |
3490 | } |
3491 | } | |
3492 | ||
64db4cff | 3493 | /* |
27569620 | 3494 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 3495 | */ |
27569620 PM |
3496 | static void __init |
3497 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff | 3498 | { |
394f99a9 | 3499 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
3500 | |
3501 | /* Set up local state, ensuring consistent view of global state. */ | |
bc75e999 | 3502 | rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu); |
27569620 | 3503 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
51a1fd30 | 3504 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != 1); |
02a5c550 | 3505 | WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp->dynticks))); |
27569620 | 3506 | rdp->cpu = cpu; |
d4c08f2a | 3507 | rdp->rsp = rsp; |
3fbfbf7a | 3508 | rcu_boot_init_nocb_percpu_data(rdp); |
27569620 PM |
3509 | } |
3510 | ||
3511 | /* | |
3512 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
ff3bb6f4 PM |
3513 | * offline event can be happening at a given time. Note also that we can |
3514 | * accept some slop in the rsp->gp_seq access due to the fact that this | |
3515 | * CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 3516 | */ |
49fb4c62 | 3517 | static void |
9b67122a | 3518 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp) |
64db4cff PM |
3519 | { |
3520 | unsigned long flags; | |
394f99a9 | 3521 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
3522 | struct rcu_node *rnp = rcu_get_root(rsp); |
3523 | ||
3524 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 3525 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 PM |
3526 | rdp->qlen_last_fqs_check = 0; |
3527 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 3528 | rdp->blimit = blimit; |
15fecf89 PM |
3529 | if (rcu_segcblist_empty(&rdp->cblist) && /* No early-boot CBs? */ |
3530 | !init_nocb_callback_list(rdp)) | |
3531 | rcu_segcblist_init(&rdp->cblist); /* Re-enable callbacks. */ | |
2342172f | 3532 | rdp->dynticks->dynticks_nesting = 1; /* CPU not up, no tearing. */ |
2625d469 | 3533 | rcu_dynticks_eqs_online(); |
67c583a7 | 3534 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
64db4cff | 3535 | |
0aa04b05 PM |
3536 | /* |
3537 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
3538 | * propagation up the rcu_node tree will happen at the beginning | |
3539 | * of the next grace period. | |
3540 | */ | |
64db4cff | 3541 | rnp = rdp->mynode; |
2a67e741 | 3542 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
b9585e94 | 3543 | rdp->beenonline = true; /* We have now been online. */ |
de30ad51 | 3544 | rdp->gp_seq = rnp->gp_seq; |
7a1d0f23 | 3545 | rdp->gp_seq_needed = rnp->gp_seq; |
5b74c458 | 3546 | rdp->cpu_no_qs.b.norm = true; |
9577df9a | 3547 | rdp->rcu_qs_ctr_snap = per_cpu(rcu_dynticks.rcu_qs_ctr, cpu); |
97c668b8 | 3548 | rdp->core_needs_qs = false; |
9b9500da | 3549 | rdp->rcu_iw_pending = false; |
8aa670cd | 3550 | rdp->rcu_iw_gp_seq = rnp->gp_seq - 1; |
477351f7 | 3551 | trace_rcu_grace_period(rsp->name, rdp->gp_seq, TPS("cpuonl")); |
67c583a7 | 3552 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
3553 | } |
3554 | ||
deb34f36 PM |
3555 | /* |
3556 | * Invoked early in the CPU-online process, when pretty much all | |
3557 | * services are available. The incoming CPU is not present. | |
3558 | */ | |
4df83742 | 3559 | int rcutree_prepare_cpu(unsigned int cpu) |
64db4cff | 3560 | { |
6ce75a23 PM |
3561 | struct rcu_state *rsp; |
3562 | ||
3563 | for_each_rcu_flavor(rsp) | |
9b67122a | 3564 | rcu_init_percpu_data(cpu, rsp); |
4df83742 TG |
3565 | |
3566 | rcu_prepare_kthreads(cpu); | |
3567 | rcu_spawn_all_nocb_kthreads(cpu); | |
3568 | ||
3569 | return 0; | |
3570 | } | |
3571 | ||
deb34f36 PM |
3572 | /* |
3573 | * Update RCU priority boot kthread affinity for CPU-hotplug changes. | |
3574 | */ | |
4df83742 TG |
3575 | static void rcutree_affinity_setting(unsigned int cpu, int outgoing) |
3576 | { | |
3577 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); | |
3578 | ||
3579 | rcu_boost_kthread_setaffinity(rdp->mynode, outgoing); | |
3580 | } | |
3581 | ||
deb34f36 PM |
3582 | /* |
3583 | * Near the end of the CPU-online process. Pretty much all services | |
3584 | * enabled, and the CPU is now very much alive. | |
3585 | */ | |
4df83742 TG |
3586 | int rcutree_online_cpu(unsigned int cpu) |
3587 | { | |
9b9500da PM |
3588 | unsigned long flags; |
3589 | struct rcu_data *rdp; | |
3590 | struct rcu_node *rnp; | |
3591 | struct rcu_state *rsp; | |
3592 | ||
3593 | for_each_rcu_flavor(rsp) { | |
3594 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
3595 | rnp = rdp->mynode; | |
3596 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3597 | rnp->ffmask |= rdp->grpmask; | |
3598 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
3599 | } | |
da915ad5 PM |
3600 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3601 | srcu_online_cpu(cpu); | |
9b9500da PM |
3602 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
3603 | return 0; /* Too early in boot for scheduler work. */ | |
3604 | sync_sched_exp_online_cleanup(cpu); | |
3605 | rcutree_affinity_setting(cpu, -1); | |
4df83742 TG |
3606 | return 0; |
3607 | } | |
3608 | ||
deb34f36 PM |
3609 | /* |
3610 | * Near the beginning of the process. The CPU is still very much alive | |
3611 | * with pretty much all services enabled. | |
3612 | */ | |
4df83742 TG |
3613 | int rcutree_offline_cpu(unsigned int cpu) |
3614 | { | |
9b9500da PM |
3615 | unsigned long flags; |
3616 | struct rcu_data *rdp; | |
3617 | struct rcu_node *rnp; | |
3618 | struct rcu_state *rsp; | |
3619 | ||
3620 | for_each_rcu_flavor(rsp) { | |
3621 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
3622 | rnp = rdp->mynode; | |
3623 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3624 | rnp->ffmask &= ~rdp->grpmask; | |
3625 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
3626 | } | |
3627 | ||
4df83742 | 3628 | rcutree_affinity_setting(cpu, cpu); |
da915ad5 PM |
3629 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3630 | srcu_offline_cpu(cpu); | |
4df83742 TG |
3631 | return 0; |
3632 | } | |
3633 | ||
deb34f36 PM |
3634 | /* |
3635 | * Near the end of the offline process. We do only tracing here. | |
3636 | */ | |
4df83742 TG |
3637 | int rcutree_dying_cpu(unsigned int cpu) |
3638 | { | |
3639 | struct rcu_state *rsp; | |
3640 | ||
3641 | for_each_rcu_flavor(rsp) | |
3642 | rcu_cleanup_dying_cpu(rsp); | |
3643 | return 0; | |
3644 | } | |
3645 | ||
deb34f36 PM |
3646 | /* |
3647 | * The outgoing CPU is gone and we are running elsewhere. | |
3648 | */ | |
4df83742 TG |
3649 | int rcutree_dead_cpu(unsigned int cpu) |
3650 | { | |
3651 | struct rcu_state *rsp; | |
3652 | ||
3653 | for_each_rcu_flavor(rsp) { | |
3654 | rcu_cleanup_dead_cpu(cpu, rsp); | |
3655 | do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu)); | |
3656 | } | |
3657 | return 0; | |
64db4cff PM |
3658 | } |
3659 | ||
f64c6013 PZ |
3660 | static DEFINE_PER_CPU(int, rcu_cpu_started); |
3661 | ||
7ec99de3 PM |
3662 | /* |
3663 | * Mark the specified CPU as being online so that subsequent grace periods | |
3664 | * (both expedited and normal) will wait on it. Note that this means that | |
3665 | * incoming CPUs are not allowed to use RCU read-side critical sections | |
3666 | * until this function is called. Failing to observe this restriction | |
3667 | * will result in lockdep splats. | |
deb34f36 PM |
3668 | * |
3669 | * Note that this function is special in that it is invoked directly | |
3670 | * from the incoming CPU rather than from the cpuhp_step mechanism. | |
3671 | * This is because this function must be invoked at a precise location. | |
7ec99de3 PM |
3672 | */ |
3673 | void rcu_cpu_starting(unsigned int cpu) | |
3674 | { | |
3675 | unsigned long flags; | |
3676 | unsigned long mask; | |
313517fc PM |
3677 | int nbits; |
3678 | unsigned long oldmask; | |
7ec99de3 PM |
3679 | struct rcu_data *rdp; |
3680 | struct rcu_node *rnp; | |
3681 | struct rcu_state *rsp; | |
3682 | ||
f64c6013 PZ |
3683 | if (per_cpu(rcu_cpu_started, cpu)) |
3684 | return; | |
3685 | ||
3686 | per_cpu(rcu_cpu_started, cpu) = 1; | |
3687 | ||
7ec99de3 | 3688 | for_each_rcu_flavor(rsp) { |
fdbb9b31 | 3689 | rdp = per_cpu_ptr(rsp->rda, cpu); |
7ec99de3 PM |
3690 | rnp = rdp->mynode; |
3691 | mask = rdp->grpmask; | |
3692 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3693 | rnp->qsmaskinitnext |= mask; | |
313517fc | 3694 | oldmask = rnp->expmaskinitnext; |
7ec99de3 | 3695 | rnp->expmaskinitnext |= mask; |
313517fc PM |
3696 | oldmask ^= rnp->expmaskinitnext; |
3697 | nbits = bitmap_weight(&oldmask, BITS_PER_LONG); | |
3698 | /* Allow lockless access for expedited grace periods. */ | |
3699 | smp_store_release(&rsp->ncpus, rsp->ncpus + nbits); /* ^^^ */ | |
7ec99de3 PM |
3700 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
3701 | } | |
313517fc | 3702 | smp_mb(); /* Ensure RCU read-side usage follows above initialization. */ |
7ec99de3 PM |
3703 | } |
3704 | ||
27d50c7e TG |
3705 | #ifdef CONFIG_HOTPLUG_CPU |
3706 | /* | |
3707 | * The CPU is exiting the idle loop into the arch_cpu_idle_dead() | |
3708 | * function. We now remove it from the rcu_node tree's ->qsmaskinit | |
3709 | * bit masks. | |
3710 | */ | |
3711 | static void rcu_cleanup_dying_idle_cpu(int cpu, struct rcu_state *rsp) | |
3712 | { | |
3713 | unsigned long flags; | |
3714 | unsigned long mask; | |
3715 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
3716 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ | |
3717 | ||
27d50c7e TG |
3718 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ |
3719 | mask = rdp->grpmask; | |
3720 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ | |
3721 | rnp->qsmaskinitnext &= ~mask; | |
710d60cb | 3722 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27d50c7e TG |
3723 | } |
3724 | ||
deb34f36 PM |
3725 | /* |
3726 | * The outgoing function has no further need of RCU, so remove it from | |
3727 | * the list of CPUs that RCU must track. | |
3728 | * | |
3729 | * Note that this function is special in that it is invoked directly | |
3730 | * from the outgoing CPU rather than from the cpuhp_step mechanism. | |
3731 | * This is because this function must be invoked at a precise location. | |
3732 | */ | |
27d50c7e TG |
3733 | void rcu_report_dead(unsigned int cpu) |
3734 | { | |
3735 | struct rcu_state *rsp; | |
3736 | ||
3737 | /* QS for any half-done expedited RCU-sched GP. */ | |
3738 | preempt_disable(); | |
3739 | rcu_report_exp_rdp(&rcu_sched_state, | |
3740 | this_cpu_ptr(rcu_sched_state.rda), true); | |
3741 | preempt_enable(); | |
3742 | for_each_rcu_flavor(rsp) | |
3743 | rcu_cleanup_dying_idle_cpu(cpu, rsp); | |
f64c6013 PZ |
3744 | |
3745 | per_cpu(rcu_cpu_started, cpu) = 0; | |
27d50c7e | 3746 | } |
a58163d8 | 3747 | |
f2dbe4a5 | 3748 | /* Migrate the dead CPU's callbacks to the current CPU. */ |
a58163d8 PM |
3749 | static void rcu_migrate_callbacks(int cpu, struct rcu_state *rsp) |
3750 | { | |
3751 | unsigned long flags; | |
b1a2d79f | 3752 | struct rcu_data *my_rdp; |
a58163d8 | 3753 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
9fa46fb8 | 3754 | struct rcu_node *rnp_root = rcu_get_root(rdp->rsp); |
ec4eacce | 3755 | bool needwake; |
a58163d8 | 3756 | |
95335c03 PM |
3757 | if (rcu_is_nocb_cpu(cpu) || rcu_segcblist_empty(&rdp->cblist)) |
3758 | return; /* No callbacks to migrate. */ | |
3759 | ||
b1a2d79f PM |
3760 | local_irq_save(flags); |
3761 | my_rdp = this_cpu_ptr(rsp->rda); | |
3762 | if (rcu_nocb_adopt_orphan_cbs(my_rdp, rdp, flags)) { | |
3763 | local_irq_restore(flags); | |
3764 | return; | |
3765 | } | |
9fa46fb8 | 3766 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ |
ec4eacce PM |
3767 | /* Leverage recent GPs and set GP for new callbacks. */ |
3768 | needwake = rcu_advance_cbs(rsp, rnp_root, rdp) || | |
3769 | rcu_advance_cbs(rsp, rnp_root, my_rdp); | |
f2dbe4a5 | 3770 | rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist); |
09efeeee PM |
3771 | WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) != |
3772 | !rcu_segcblist_n_cbs(&my_rdp->cblist)); | |
537b85c8 | 3773 | raw_spin_unlock_irqrestore_rcu_node(rnp_root, flags); |
ec4eacce PM |
3774 | if (needwake) |
3775 | rcu_gp_kthread_wake(rsp); | |
a58163d8 PM |
3776 | WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 || |
3777 | !rcu_segcblist_empty(&rdp->cblist), | |
3778 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n", | |
3779 | cpu, rcu_segcblist_n_cbs(&rdp->cblist), | |
3780 | rcu_segcblist_first_cb(&rdp->cblist)); | |
3781 | } | |
3782 | ||
3783 | /* | |
3784 | * The outgoing CPU has just passed through the dying-idle state, | |
3785 | * and we are being invoked from the CPU that was IPIed to continue the | |
3786 | * offline operation. We need to migrate the outgoing CPU's callbacks. | |
3787 | */ | |
3788 | void rcutree_migrate_callbacks(int cpu) | |
3789 | { | |
3790 | struct rcu_state *rsp; | |
3791 | ||
3792 | for_each_rcu_flavor(rsp) | |
3793 | rcu_migrate_callbacks(cpu, rsp); | |
3794 | } | |
27d50c7e TG |
3795 | #endif |
3796 | ||
deb34f36 PM |
3797 | /* |
3798 | * On non-huge systems, use expedited RCU grace periods to make suspend | |
3799 | * and hibernation run faster. | |
3800 | */ | |
d1d74d14 BP |
3801 | static int rcu_pm_notify(struct notifier_block *self, |
3802 | unsigned long action, void *hcpu) | |
3803 | { | |
3804 | switch (action) { | |
3805 | case PM_HIBERNATION_PREPARE: | |
3806 | case PM_SUSPEND_PREPARE: | |
3807 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ | |
5afff48b | 3808 | rcu_expedite_gp(); |
d1d74d14 BP |
3809 | break; |
3810 | case PM_POST_HIBERNATION: | |
3811 | case PM_POST_SUSPEND: | |
5afff48b PM |
3812 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ |
3813 | rcu_unexpedite_gp(); | |
d1d74d14 BP |
3814 | break; |
3815 | default: | |
3816 | break; | |
3817 | } | |
3818 | return NOTIFY_OK; | |
3819 | } | |
3820 | ||
b3dbec76 | 3821 | /* |
9386c0b7 | 3822 | * Spawn the kthreads that handle each RCU flavor's grace periods. |
b3dbec76 PM |
3823 | */ |
3824 | static int __init rcu_spawn_gp_kthread(void) | |
3825 | { | |
3826 | unsigned long flags; | |
a94844b2 | 3827 | int kthread_prio_in = kthread_prio; |
b3dbec76 PM |
3828 | struct rcu_node *rnp; |
3829 | struct rcu_state *rsp; | |
a94844b2 | 3830 | struct sched_param sp; |
b3dbec76 PM |
3831 | struct task_struct *t; |
3832 | ||
a94844b2 PM |
3833 | /* Force priority into range. */ |
3834 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
3835 | kthread_prio = 1; | |
3836 | else if (kthread_prio < 0) | |
3837 | kthread_prio = 0; | |
3838 | else if (kthread_prio > 99) | |
3839 | kthread_prio = 99; | |
3840 | if (kthread_prio != kthread_prio_in) | |
3841 | pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n", | |
3842 | kthread_prio, kthread_prio_in); | |
3843 | ||
9386c0b7 | 3844 | rcu_scheduler_fully_active = 1; |
b3dbec76 | 3845 | for_each_rcu_flavor(rsp) { |
a94844b2 | 3846 | t = kthread_create(rcu_gp_kthread, rsp, "%s", rsp->name); |
b3dbec76 PM |
3847 | BUG_ON(IS_ERR(t)); |
3848 | rnp = rcu_get_root(rsp); | |
6cf10081 | 3849 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b3dbec76 | 3850 | rsp->gp_kthread = t; |
a94844b2 PM |
3851 | if (kthread_prio) { |
3852 | sp.sched_priority = kthread_prio; | |
3853 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
3854 | } | |
67c583a7 | 3855 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
e11f1335 | 3856 | wake_up_process(t); |
b3dbec76 | 3857 | } |
35ce7f29 | 3858 | rcu_spawn_nocb_kthreads(); |
9386c0b7 | 3859 | rcu_spawn_boost_kthreads(); |
b3dbec76 PM |
3860 | return 0; |
3861 | } | |
3862 | early_initcall(rcu_spawn_gp_kthread); | |
3863 | ||
bbad9379 | 3864 | /* |
52d7e48b PM |
3865 | * This function is invoked towards the end of the scheduler's |
3866 | * initialization process. Before this is called, the idle task might | |
3867 | * contain synchronous grace-period primitives (during which time, this idle | |
3868 | * task is booting the system, and such primitives are no-ops). After this | |
3869 | * function is called, any synchronous grace-period primitives are run as | |
3870 | * expedited, with the requesting task driving the grace period forward. | |
900b1028 | 3871 | * A later core_initcall() rcu_set_runtime_mode() will switch to full |
52d7e48b | 3872 | * runtime RCU functionality. |
bbad9379 PM |
3873 | */ |
3874 | void rcu_scheduler_starting(void) | |
3875 | { | |
3876 | WARN_ON(num_online_cpus() != 1); | |
3877 | WARN_ON(nr_context_switches() > 0); | |
52d7e48b PM |
3878 | rcu_test_sync_prims(); |
3879 | rcu_scheduler_active = RCU_SCHEDULER_INIT; | |
3880 | rcu_test_sync_prims(); | |
bbad9379 PM |
3881 | } |
3882 | ||
64db4cff PM |
3883 | /* |
3884 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
3885 | */ | |
a87f203e | 3886 | static void __init rcu_init_one(struct rcu_state *rsp) |
64db4cff | 3887 | { |
cb007102 AG |
3888 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
3889 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
3890 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
3891 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
199977bf | 3892 | |
199977bf | 3893 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ |
64db4cff PM |
3894 | int cpustride = 1; |
3895 | int i; | |
3896 | int j; | |
3897 | struct rcu_node *rnp; | |
3898 | ||
05b84aec | 3899 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 3900 | |
3eaaaf6c PM |
3901 | /* Silence gcc 4.8 false positive about array index out of range. */ |
3902 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
3903 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 3904 | |
64db4cff PM |
3905 | /* Initialize the level-tracking arrays. */ |
3906 | ||
f885b7f2 | 3907 | for (i = 1; i < rcu_num_lvls; i++) |
41f5c631 PM |
3908 | rsp->level[i] = rsp->level[i - 1] + num_rcu_lvl[i - 1]; |
3909 | rcu_init_levelspread(levelspread, num_rcu_lvl); | |
64db4cff PM |
3910 | |
3911 | /* Initialize the elements themselves, starting from the leaves. */ | |
3912 | ||
f885b7f2 | 3913 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 3914 | cpustride *= levelspread[i]; |
64db4cff | 3915 | rnp = rsp->level[i]; |
41f5c631 | 3916 | for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) { |
67c583a7 BF |
3917 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
3918 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 3919 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
3920 | raw_spin_lock_init(&rnp->fqslock); |
3921 | lockdep_set_class_and_name(&rnp->fqslock, | |
3922 | &rcu_fqs_class[i], fqs[i]); | |
de30ad51 | 3923 | rnp->gp_seq = rsp->gp_seq; |
7a1d0f23 | 3924 | rnp->gp_seq_needed = rsp->gp_seq; |
d43a5d32 | 3925 | rnp->completedqs = rsp->gp_seq; |
64db4cff PM |
3926 | rnp->qsmask = 0; |
3927 | rnp->qsmaskinit = 0; | |
3928 | rnp->grplo = j * cpustride; | |
3929 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
3930 | if (rnp->grphi >= nr_cpu_ids) |
3931 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
3932 | if (i == 0) { |
3933 | rnp->grpnum = 0; | |
3934 | rnp->grpmask = 0; | |
3935 | rnp->parent = NULL; | |
3936 | } else { | |
199977bf | 3937 | rnp->grpnum = j % levelspread[i - 1]; |
64db4cff PM |
3938 | rnp->grpmask = 1UL << rnp->grpnum; |
3939 | rnp->parent = rsp->level[i - 1] + | |
199977bf | 3940 | j / levelspread[i - 1]; |
64db4cff PM |
3941 | } |
3942 | rnp->level = i; | |
12f5f524 | 3943 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 3944 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
3945 | init_waitqueue_head(&rnp->exp_wq[0]); |
3946 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
3947 | init_waitqueue_head(&rnp->exp_wq[2]); |
3948 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 3949 | spin_lock_init(&rnp->exp_lock); |
64db4cff PM |
3950 | } |
3951 | } | |
0c34029a | 3952 | |
abedf8e2 PG |
3953 | init_swait_queue_head(&rsp->gp_wq); |
3954 | init_swait_queue_head(&rsp->expedited_wq); | |
5b4c11d5 | 3955 | rnp = rcu_first_leaf_node(rsp); |
0c34029a | 3956 | for_each_possible_cpu(i) { |
4a90a068 | 3957 | while (i > rnp->grphi) |
0c34029a | 3958 | rnp++; |
394f99a9 | 3959 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
3960 | rcu_boot_init_percpu_data(i, rsp); |
3961 | } | |
6ce75a23 | 3962 | list_add(&rsp->flavors, &rcu_struct_flavors); |
64db4cff PM |
3963 | } |
3964 | ||
f885b7f2 PM |
3965 | /* |
3966 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 3967 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
3968 | * the ->node array in the rcu_state structure. |
3969 | */ | |
3970 | static void __init rcu_init_geometry(void) | |
3971 | { | |
026ad283 | 3972 | ulong d; |
f885b7f2 | 3973 | int i; |
05b84aec | 3974 | int rcu_capacity[RCU_NUM_LVLS]; |
f885b7f2 | 3975 | |
026ad283 PM |
3976 | /* |
3977 | * Initialize any unspecified boot parameters. | |
3978 | * The default values of jiffies_till_first_fqs and | |
3979 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
3980 | * value, which is a function of HZ, then adding one for each | |
3981 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
3982 | */ | |
3983 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
3984 | if (jiffies_till_first_fqs == ULONG_MAX) | |
3985 | jiffies_till_first_fqs = d; | |
3986 | if (jiffies_till_next_fqs == ULONG_MAX) | |
3987 | jiffies_till_next_fqs = d; | |
3988 | ||
f885b7f2 | 3989 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 3990 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 3991 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 3992 | return; |
9b130ad5 | 3993 | pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n", |
39479098 | 3994 | rcu_fanout_leaf, nr_cpu_ids); |
f885b7f2 | 3995 | |
f885b7f2 | 3996 | /* |
ee968ac6 PM |
3997 | * The boot-time rcu_fanout_leaf parameter must be at least two |
3998 | * and cannot exceed the number of bits in the rcu_node masks. | |
3999 | * Complain and fall back to the compile-time values if this | |
4000 | * limit is exceeded. | |
f885b7f2 | 4001 | */ |
ee968ac6 | 4002 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 4003 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 4004 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
4005 | WARN_ON(1); |
4006 | return; | |
4007 | } | |
4008 | ||
f885b7f2 PM |
4009 | /* |
4010 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 4011 | * with the given number of levels. |
f885b7f2 | 4012 | */ |
9618138b | 4013 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 4014 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 4015 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
4016 | |
4017 | /* | |
75cf15a4 | 4018 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 4019 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 4020 | */ |
ee968ac6 PM |
4021 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
4022 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
4023 | WARN_ON(1); | |
4024 | return; | |
4025 | } | |
f885b7f2 | 4026 | |
679f9858 | 4027 | /* Calculate the number of levels in the tree. */ |
9618138b | 4028 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 4029 | } |
9618138b | 4030 | rcu_num_lvls = i + 1; |
679f9858 | 4031 | |
f885b7f2 | 4032 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 4033 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 4034 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
4035 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
4036 | } | |
f885b7f2 PM |
4037 | |
4038 | /* Calculate the total number of rcu_node structures. */ | |
4039 | rcu_num_nodes = 0; | |
679f9858 | 4040 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 4041 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
4042 | } |
4043 | ||
a3dc2948 PM |
4044 | /* |
4045 | * Dump out the structure of the rcu_node combining tree associated | |
4046 | * with the rcu_state structure referenced by rsp. | |
4047 | */ | |
4048 | static void __init rcu_dump_rcu_node_tree(struct rcu_state *rsp) | |
4049 | { | |
4050 | int level = 0; | |
4051 | struct rcu_node *rnp; | |
4052 | ||
4053 | pr_info("rcu_node tree layout dump\n"); | |
4054 | pr_info(" "); | |
4055 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
4056 | if (rnp->level != level) { | |
4057 | pr_cont("\n"); | |
4058 | pr_info(" "); | |
4059 | level = rnp->level; | |
4060 | } | |
4061 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
4062 | } | |
4063 | pr_cont("\n"); | |
4064 | } | |
4065 | ||
ad7c946b | 4066 | struct workqueue_struct *rcu_gp_wq; |
25f3d7ef | 4067 | struct workqueue_struct *rcu_par_gp_wq; |
ad7c946b | 4068 | |
9f680ab4 | 4069 | void __init rcu_init(void) |
64db4cff | 4070 | { |
017c4261 | 4071 | int cpu; |
9f680ab4 | 4072 | |
47627678 PM |
4073 | rcu_early_boot_tests(); |
4074 | ||
f41d911f | 4075 | rcu_bootup_announce(); |
f885b7f2 | 4076 | rcu_init_geometry(); |
a87f203e PM |
4077 | rcu_init_one(&rcu_bh_state); |
4078 | rcu_init_one(&rcu_sched_state); | |
a3dc2948 PM |
4079 | if (dump_tree) |
4080 | rcu_dump_rcu_node_tree(&rcu_sched_state); | |
f41d911f | 4081 | __rcu_init_preempt(); |
b5b39360 | 4082 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
4083 | |
4084 | /* | |
4085 | * We don't need protection against CPU-hotplug here because | |
4086 | * this is called early in boot, before either interrupts | |
4087 | * or the scheduler are operational. | |
4088 | */ | |
d1d74d14 | 4089 | pm_notifier(rcu_pm_notify, 0); |
7ec99de3 | 4090 | for_each_online_cpu(cpu) { |
4df83742 | 4091 | rcutree_prepare_cpu(cpu); |
7ec99de3 | 4092 | rcu_cpu_starting(cpu); |
9b9500da | 4093 | rcutree_online_cpu(cpu); |
7ec99de3 | 4094 | } |
ad7c946b PM |
4095 | |
4096 | /* Create workqueue for expedited GPs and for Tree SRCU. */ | |
4097 | rcu_gp_wq = alloc_workqueue("rcu_gp", WQ_MEM_RECLAIM, 0); | |
4098 | WARN_ON(!rcu_gp_wq); | |
25f3d7ef PM |
4099 | rcu_par_gp_wq = alloc_workqueue("rcu_par_gp", WQ_MEM_RECLAIM, 0); |
4100 | WARN_ON(!rcu_par_gp_wq); | |
64db4cff PM |
4101 | } |
4102 | ||
3549c2bc | 4103 | #include "tree_exp.h" |
4102adab | 4104 | #include "tree_plugin.h" |