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, \ |
42c3533e PM |
98 | .gpnum = 0UL - 300UL, \ |
99 | .completed = 0UL - 300UL, \ | |
7be7f0be | 100 | .barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \ |
a8a29b3b | 101 | .name = RCU_STATE_NAME(sname), \ |
a4889858 | 102 | .abbr = sabbr, \ |
f6a12f34 | 103 | .exp_mutex = __MUTEX_INITIALIZER(sname##_state.exp_mutex), \ |
3b5f668e | 104 | .exp_wake_mutex = __MUTEX_INITIALIZER(sname##_state.exp_wake_mutex), \ |
2723249a | 105 | } |
64db4cff | 106 | |
a41bfeb2 SRRH |
107 | RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched); |
108 | RCU_STATE_INITIALIZER(rcu_bh, 'b', call_rcu_bh); | |
b1f77b05 | 109 | |
b28a7c01 | 110 | static struct rcu_state *const rcu_state_p; |
6ce75a23 | 111 | LIST_HEAD(rcu_struct_flavors); |
27f4d280 | 112 | |
a3dc2948 PM |
113 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
114 | static bool dump_tree; | |
115 | module_param(dump_tree, bool, 0444); | |
7fa27001 PM |
116 | /* Control rcu_node-tree auto-balancing at boot time. */ |
117 | static bool rcu_fanout_exact; | |
118 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
119 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
120 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 121 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 122 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 | 123 | /* Number of rcu_nodes at specified level. */ |
e95d68d2 | 124 | int num_rcu_lvl[] = NUM_RCU_LVL_INIT; |
f885b7f2 | 125 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
088e9d25 DBO |
126 | /* panic() on RCU Stall sysctl. */ |
127 | int sysctl_panic_on_rcu_stall __read_mostly; | |
f885b7f2 | 128 | |
b0d30417 | 129 | /* |
52d7e48b PM |
130 | * The rcu_scheduler_active variable is initialized to the value |
131 | * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the | |
132 | * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE, | |
133 | * RCU can assume that there is but one task, allowing RCU to (for example) | |
0d95092c | 134 | * optimize synchronize_rcu() to a simple barrier(). When this variable |
52d7e48b PM |
135 | * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required |
136 | * to detect real grace periods. This variable is also used to suppress | |
137 | * boot-time false positives from lockdep-RCU error checking. Finally, it | |
138 | * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU | |
139 | * is fully initialized, including all of its kthreads having been spawned. | |
b0d30417 | 140 | */ |
bbad9379 PM |
141 | int rcu_scheduler_active __read_mostly; |
142 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
143 | ||
b0d30417 PM |
144 | /* |
145 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
146 | * during the early_initcall() processing, which is after the scheduler | |
147 | * is capable of creating new tasks. So RCU processing (for example, | |
148 | * creating tasks for RCU priority boosting) must be delayed until after | |
149 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
150 | * currently delay invocation of any RCU callbacks until after this point. | |
151 | * | |
152 | * It might later prove better for people registering RCU callbacks during | |
153 | * early boot to take responsibility for these callbacks, but one step at | |
154 | * a time. | |
155 | */ | |
156 | static int rcu_scheduler_fully_active __read_mostly; | |
157 | ||
0aa04b05 PM |
158 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
159 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 160 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 PM |
161 | static void invoke_rcu_core(void); |
162 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); | |
6587a23b PM |
163 | static void rcu_report_exp_rdp(struct rcu_state *rsp, |
164 | struct rcu_data *rdp, bool wake); | |
3549c2bc | 165 | static void sync_sched_exp_online_cleanup(int cpu); |
a26ac245 | 166 | |
a94844b2 | 167 | /* rcuc/rcub kthread realtime priority */ |
26730f55 | 168 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; |
a94844b2 PM |
169 | module_param(kthread_prio, int, 0644); |
170 | ||
8d7dc928 | 171 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd | 172 | |
90040c9e PM |
173 | static int gp_preinit_delay; |
174 | module_param(gp_preinit_delay, int, 0444); | |
175 | static int gp_init_delay; | |
176 | module_param(gp_init_delay, int, 0444); | |
177 | static int gp_cleanup_delay; | |
178 | module_param(gp_cleanup_delay, int, 0444); | |
0f41c0dd | 179 | |
eab128e8 PM |
180 | /* |
181 | * Number of grace periods between delays, normalized by the duration of | |
bfd090be | 182 | * the delay. The longer the delay, the more the grace periods between |
eab128e8 PM |
183 | * each delay. The reason for this normalization is that it means that, |
184 | * for non-zero delays, the overall slowdown of grace periods is constant | |
185 | * regardless of the duration of the delay. This arrangement balances | |
186 | * the need for long delays to increase some race probabilities with the | |
187 | * need for fast grace periods to increase other race probabilities. | |
188 | */ | |
189 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */ | |
37745d28 | 190 | |
4a298656 PM |
191 | /* |
192 | * Track the rcutorture test sequence number and the update version | |
193 | * number within a given test. The rcutorture_testseq is incremented | |
194 | * on every rcutorture module load and unload, so has an odd value | |
195 | * when a test is running. The rcutorture_vernum is set to zero | |
196 | * when rcutorture starts and is incremented on each rcutorture update. | |
197 | * These variables enable correlating rcutorture output with the | |
198 | * RCU tracing information. | |
199 | */ | |
200 | unsigned long rcutorture_testseq; | |
201 | unsigned long rcutorture_vernum; | |
202 | ||
0aa04b05 PM |
203 | /* |
204 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
205 | * This will not be stable unless the rcu_node structure's ->lock is | |
206 | * held, but the bit corresponding to the current CPU will be stable | |
207 | * in most contexts. | |
208 | */ | |
209 | unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) | |
210 | { | |
7d0ae808 | 211 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
212 | } |
213 | ||
fc2219d4 | 214 | /* |
7d0ae808 | 215 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
216 | * permit this function to be invoked without holding the root rcu_node |
217 | * structure's ->lock, but of course results can be subject to change. | |
218 | */ | |
219 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
220 | { | |
7d0ae808 | 221 | return READ_ONCE(rsp->completed) != READ_ONCE(rsp->gpnum); |
fc2219d4 PM |
222 | } |
223 | ||
b1f77b05 | 224 | /* |
d6714c22 | 225 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 226 | * how many quiescent states passed, just if there was at least |
d6714c22 | 227 | * one since the start of the grace period, this just sets a flag. |
e4cc1f22 | 228 | * The caller must have disabled preemption. |
b1f77b05 | 229 | */ |
284a8c93 | 230 | void rcu_sched_qs(void) |
b1f77b05 | 231 | { |
f4687d26 | 232 | RCU_LOCKDEP_WARN(preemptible(), "rcu_sched_qs() invoked with preemption enabled!!!"); |
fecbf6f0 PM |
233 | if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.s)) |
234 | return; | |
235 | trace_rcu_grace_period(TPS("rcu_sched"), | |
236 | __this_cpu_read(rcu_sched_data.gpnum), | |
237 | TPS("cpuqs")); | |
238 | __this_cpu_write(rcu_sched_data.cpu_no_qs.b.norm, false); | |
239 | if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.b.exp)) | |
240 | return; | |
46a5d164 PM |
241 | __this_cpu_write(rcu_sched_data.cpu_no_qs.b.exp, false); |
242 | rcu_report_exp_rdp(&rcu_sched_state, | |
243 | this_cpu_ptr(&rcu_sched_data), true); | |
b1f77b05 IM |
244 | } |
245 | ||
284a8c93 | 246 | void rcu_bh_qs(void) |
b1f77b05 | 247 | { |
f4687d26 | 248 | RCU_LOCKDEP_WARN(preemptible(), "rcu_bh_qs() invoked with preemption enabled!!!"); |
5b74c458 | 249 | if (__this_cpu_read(rcu_bh_data.cpu_no_qs.s)) { |
284a8c93 PM |
250 | trace_rcu_grace_period(TPS("rcu_bh"), |
251 | __this_cpu_read(rcu_bh_data.gpnum), | |
252 | TPS("cpuqs")); | |
5b74c458 | 253 | __this_cpu_write(rcu_bh_data.cpu_no_qs.b.norm, false); |
284a8c93 | 254 | } |
b1f77b05 | 255 | } |
64db4cff | 256 | |
b8c17e66 PM |
257 | /* |
258 | * Steal a bit from the bottom of ->dynticks for idle entry/exit | |
259 | * control. Initially this is for TLB flushing. | |
260 | */ | |
261 | #define RCU_DYNTICK_CTRL_MASK 0x1 | |
262 | #define RCU_DYNTICK_CTRL_CTR (RCU_DYNTICK_CTRL_MASK + 1) | |
263 | #ifndef rcu_eqs_special_exit | |
264 | #define rcu_eqs_special_exit() do { } while (0) | |
265 | #endif | |
4a81e832 PM |
266 | |
267 | static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { | |
51a1fd30 | 268 | .dynticks_nesting = 1, |
58721f5d | 269 | .dynticks_nmi_nesting = DYNTICK_IRQ_NONIDLE, |
b8c17e66 | 270 | .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR), |
4a81e832 PM |
271 | }; |
272 | ||
2625d469 PM |
273 | /* |
274 | * Record entry into an extended quiescent state. This is only to be | |
275 | * called when not already in an extended quiescent state. | |
276 | */ | |
277 | static void rcu_dynticks_eqs_enter(void) | |
278 | { | |
279 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 280 | int seq; |
2625d469 PM |
281 | |
282 | /* | |
b8c17e66 | 283 | * CPUs seeing atomic_add_return() must see prior RCU read-side |
2625d469 PM |
284 | * critical sections, and we also must force ordering with the |
285 | * next idle sojourn. | |
286 | */ | |
b8c17e66 PM |
287 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
288 | /* Better be in an extended quiescent state! */ | |
289 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
290 | (seq & RCU_DYNTICK_CTRL_CTR)); | |
291 | /* Better not have special action (TLB flush) pending! */ | |
292 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
293 | (seq & RCU_DYNTICK_CTRL_MASK)); | |
2625d469 PM |
294 | } |
295 | ||
296 | /* | |
297 | * Record exit from an extended quiescent state. This is only to be | |
298 | * called from an extended quiescent state. | |
299 | */ | |
300 | static void rcu_dynticks_eqs_exit(void) | |
301 | { | |
302 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 303 | int seq; |
2625d469 PM |
304 | |
305 | /* | |
b8c17e66 | 306 | * CPUs seeing atomic_add_return() must see prior idle sojourns, |
2625d469 PM |
307 | * and we also must force ordering with the next RCU read-side |
308 | * critical section. | |
309 | */ | |
b8c17e66 PM |
310 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
311 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
312 | !(seq & RCU_DYNTICK_CTRL_CTR)); | |
313 | if (seq & RCU_DYNTICK_CTRL_MASK) { | |
314 | atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdtp->dynticks); | |
315 | smp_mb__after_atomic(); /* _exit after clearing mask. */ | |
316 | /* Prefer duplicate flushes to losing a flush. */ | |
317 | rcu_eqs_special_exit(); | |
318 | } | |
2625d469 PM |
319 | } |
320 | ||
321 | /* | |
322 | * Reset the current CPU's ->dynticks counter to indicate that the | |
323 | * newly onlined CPU is no longer in an extended quiescent state. | |
324 | * This will either leave the counter unchanged, or increment it | |
325 | * to the next non-quiescent value. | |
326 | * | |
327 | * The non-atomic test/increment sequence works because the upper bits | |
328 | * of the ->dynticks counter are manipulated only by the corresponding CPU, | |
329 | * or when the corresponding CPU is offline. | |
330 | */ | |
331 | static void rcu_dynticks_eqs_online(void) | |
332 | { | |
333 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
334 | ||
b8c17e66 | 335 | if (atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR) |
2625d469 | 336 | return; |
b8c17e66 | 337 | atomic_add(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
2625d469 PM |
338 | } |
339 | ||
02a5c550 PM |
340 | /* |
341 | * Is the current CPU in an extended quiescent state? | |
342 | * | |
343 | * No ordering, as we are sampling CPU-local information. | |
344 | */ | |
345 | bool rcu_dynticks_curr_cpu_in_eqs(void) | |
346 | { | |
347 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
348 | ||
b8c17e66 | 349 | return !(atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
350 | } |
351 | ||
8b2f63ab PM |
352 | /* |
353 | * Snapshot the ->dynticks counter with full ordering so as to allow | |
354 | * stable comparison of this counter with past and future snapshots. | |
355 | */ | |
02a5c550 | 356 | int rcu_dynticks_snap(struct rcu_dynticks *rdtp) |
8b2f63ab PM |
357 | { |
358 | int snap = atomic_add_return(0, &rdtp->dynticks); | |
359 | ||
b8c17e66 | 360 | return snap & ~RCU_DYNTICK_CTRL_MASK; |
8b2f63ab PM |
361 | } |
362 | ||
02a5c550 PM |
363 | /* |
364 | * Return true if the snapshot returned from rcu_dynticks_snap() | |
365 | * indicates that RCU is in an extended quiescent state. | |
366 | */ | |
367 | static bool rcu_dynticks_in_eqs(int snap) | |
368 | { | |
b8c17e66 | 369 | return !(snap & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
370 | } |
371 | ||
372 | /* | |
373 | * Return true if the CPU corresponding to the specified rcu_dynticks | |
374 | * structure has spent some time in an extended quiescent state since | |
375 | * rcu_dynticks_snap() returned the specified snapshot. | |
376 | */ | |
377 | static bool rcu_dynticks_in_eqs_since(struct rcu_dynticks *rdtp, int snap) | |
378 | { | |
379 | return snap != rcu_dynticks_snap(rdtp); | |
380 | } | |
381 | ||
6563de9d PM |
382 | /* |
383 | * Do a double-increment of the ->dynticks counter to emulate a | |
384 | * momentary idle-CPU quiescent state. | |
385 | */ | |
386 | static void rcu_dynticks_momentary_idle(void) | |
387 | { | |
388 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 PM |
389 | int special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR, |
390 | &rdtp->dynticks); | |
6563de9d PM |
391 | |
392 | /* It is illegal to call this from idle state. */ | |
b8c17e66 | 393 | WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR)); |
6563de9d PM |
394 | } |
395 | ||
b8c17e66 PM |
396 | /* |
397 | * Set the special (bottom) bit of the specified CPU so that it | |
398 | * will take special action (such as flushing its TLB) on the | |
399 | * next exit from an extended quiescent state. Returns true if | |
400 | * the bit was successfully set, or false if the CPU was not in | |
401 | * an extended quiescent state. | |
402 | */ | |
403 | bool rcu_eqs_special_set(int cpu) | |
404 | { | |
405 | int old; | |
406 | int new; | |
407 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); | |
408 | ||
409 | do { | |
410 | old = atomic_read(&rdtp->dynticks); | |
411 | if (old & RCU_DYNTICK_CTRL_CTR) | |
412 | return false; | |
413 | new = old | RCU_DYNTICK_CTRL_MASK; | |
414 | } while (atomic_cmpxchg(&rdtp->dynticks, old, new) != old); | |
415 | return true; | |
6563de9d | 416 | } |
5cd37193 | 417 | |
4a81e832 PM |
418 | /* |
419 | * Let the RCU core know that this CPU has gone through the scheduler, | |
420 | * which is a quiescent state. This is called when the need for a | |
421 | * quiescent state is urgent, so we burn an atomic operation and full | |
422 | * memory barriers to let the RCU core know about it, regardless of what | |
423 | * this CPU might (or might not) do in the near future. | |
424 | * | |
0f9be8ca | 425 | * We inform the RCU core by emulating a zero-duration dyntick-idle period. |
46a5d164 PM |
426 | * |
427 | * The caller must have disabled interrupts. | |
4a81e832 PM |
428 | */ |
429 | static void rcu_momentary_dyntick_idle(void) | |
430 | { | |
0f9be8ca PM |
431 | raw_cpu_write(rcu_dynticks.rcu_need_heavy_qs, false); |
432 | rcu_dynticks_momentary_idle(); | |
4a81e832 PM |
433 | } |
434 | ||
25502a6c PM |
435 | /* |
436 | * Note a context switch. This is a quiescent state for RCU-sched, | |
437 | * and requires special handling for preemptible RCU. | |
46a5d164 | 438 | * The caller must have disabled interrupts. |
25502a6c | 439 | */ |
bcbfdd01 | 440 | void rcu_note_context_switch(bool preempt) |
25502a6c | 441 | { |
bb73c52b | 442 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
f7f7bac9 | 443 | trace_rcu_utilization(TPS("Start context switch")); |
284a8c93 | 444 | rcu_sched_qs(); |
5b72f964 | 445 | rcu_preempt_note_context_switch(preempt); |
9226b10d PM |
446 | /* Load rcu_urgent_qs before other flags. */ |
447 | if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs))) | |
448 | goto out; | |
449 | this_cpu_write(rcu_dynticks.rcu_urgent_qs, false); | |
0f9be8ca | 450 | if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) |
4a81e832 | 451 | rcu_momentary_dyntick_idle(); |
9226b10d | 452 | this_cpu_inc(rcu_dynticks.rcu_qs_ctr); |
bcbfdd01 PM |
453 | if (!preempt) |
454 | rcu_note_voluntary_context_switch_lite(current); | |
9226b10d | 455 | out: |
f7f7bac9 | 456 | trace_rcu_utilization(TPS("End context switch")); |
bb73c52b | 457 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
25502a6c | 458 | } |
29ce8310 | 459 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
25502a6c | 460 | |
5cd37193 | 461 | /* |
1925d196 | 462 | * Register a quiescent state for all RCU flavors. If there is an |
5cd37193 PM |
463 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight |
464 | * dyntick-idle quiescent state visible to other CPUs (but only for those | |
1925d196 | 465 | * RCU flavors in desperate need of a quiescent state, which will normally |
5cd37193 PM |
466 | * be none of them). Either way, do a lightweight quiescent state for |
467 | * all RCU flavors. | |
bb73c52b BF |
468 | * |
469 | * The barrier() calls are redundant in the common case when this is | |
470 | * called externally, but just in case this is called from within this | |
471 | * file. | |
472 | * | |
5cd37193 PM |
473 | */ |
474 | void rcu_all_qs(void) | |
475 | { | |
46a5d164 PM |
476 | unsigned long flags; |
477 | ||
9226b10d PM |
478 | if (!raw_cpu_read(rcu_dynticks.rcu_urgent_qs)) |
479 | return; | |
480 | preempt_disable(); | |
481 | /* Load rcu_urgent_qs before other flags. */ | |
482 | if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs))) { | |
483 | preempt_enable(); | |
484 | return; | |
485 | } | |
486 | this_cpu_write(rcu_dynticks.rcu_urgent_qs, false); | |
bb73c52b | 487 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
0f9be8ca | 488 | if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) { |
46a5d164 | 489 | local_irq_save(flags); |
5cd37193 | 490 | rcu_momentary_dyntick_idle(); |
46a5d164 PM |
491 | local_irq_restore(flags); |
492 | } | |
9226b10d | 493 | if (unlikely(raw_cpu_read(rcu_sched_data.cpu_no_qs.b.exp))) |
a1e12248 | 494 | rcu_sched_qs(); |
9577df9a | 495 | this_cpu_inc(rcu_dynticks.rcu_qs_ctr); |
bb73c52b | 496 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
9226b10d | 497 | preempt_enable(); |
5cd37193 PM |
498 | } |
499 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
500 | ||
17c7798b PM |
501 | #define DEFAULT_RCU_BLIMIT 10 /* Maximum callbacks per rcu_do_batch. */ |
502 | static long blimit = DEFAULT_RCU_BLIMIT; | |
503 | #define DEFAULT_RCU_QHIMARK 10000 /* If this many pending, ignore blimit. */ | |
504 | static long qhimark = DEFAULT_RCU_QHIMARK; | |
505 | #define DEFAULT_RCU_QLOMARK 100 /* Once only this many pending, use blimit. */ | |
506 | static long qlowmark = DEFAULT_RCU_QLOMARK; | |
64db4cff | 507 | |
878d7439 ED |
508 | module_param(blimit, long, 0444); |
509 | module_param(qhimark, long, 0444); | |
510 | module_param(qlowmark, long, 0444); | |
3d76c082 | 511 | |
026ad283 PM |
512 | static ulong jiffies_till_first_fqs = ULONG_MAX; |
513 | static ulong jiffies_till_next_fqs = ULONG_MAX; | |
8c7c4829 | 514 | static bool rcu_kick_kthreads; |
d40011f6 PM |
515 | |
516 | module_param(jiffies_till_first_fqs, ulong, 0644); | |
517 | module_param(jiffies_till_next_fqs, ulong, 0644); | |
8c7c4829 | 518 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 519 | |
4a81e832 PM |
520 | /* |
521 | * How long the grace period must be before we start recruiting | |
522 | * quiescent-state help from rcu_note_context_switch(). | |
523 | */ | |
f79c3ad6 PM |
524 | static ulong jiffies_till_sched_qs = HZ / 10; |
525 | module_param(jiffies_till_sched_qs, ulong, 0444); | |
4a81e832 | 526 | |
48a7639c | 527 | static bool rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp, |
910ee45d | 528 | struct rcu_data *rdp); |
fe5ac724 | 529 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *rsp)); |
4cdfc175 | 530 | static void force_quiescent_state(struct rcu_state *rsp); |
e3950ecd | 531 | static int rcu_pending(void); |
64db4cff PM |
532 | |
533 | /* | |
917963d0 | 534 | * Return the number of RCU batches started thus far for debug & stats. |
64db4cff | 535 | */ |
917963d0 PM |
536 | unsigned long rcu_batches_started(void) |
537 | { | |
538 | return rcu_state_p->gpnum; | |
539 | } | |
540 | EXPORT_SYMBOL_GPL(rcu_batches_started); | |
541 | ||
542 | /* | |
543 | * Return the number of RCU-sched batches started thus far for debug & stats. | |
64db4cff | 544 | */ |
917963d0 PM |
545 | unsigned long rcu_batches_started_sched(void) |
546 | { | |
547 | return rcu_sched_state.gpnum; | |
548 | } | |
549 | EXPORT_SYMBOL_GPL(rcu_batches_started_sched); | |
550 | ||
551 | /* | |
552 | * Return the number of RCU BH batches started thus far for debug & stats. | |
553 | */ | |
554 | unsigned long rcu_batches_started_bh(void) | |
555 | { | |
556 | return rcu_bh_state.gpnum; | |
557 | } | |
558 | EXPORT_SYMBOL_GPL(rcu_batches_started_bh); | |
559 | ||
560 | /* | |
561 | * Return the number of RCU batches completed thus far for debug & stats. | |
562 | */ | |
563 | unsigned long rcu_batches_completed(void) | |
564 | { | |
565 | return rcu_state_p->completed; | |
566 | } | |
567 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
568 | ||
569 | /* | |
570 | * Return the number of RCU-sched batches completed thus far for debug & stats. | |
64db4cff | 571 | */ |
9733e4f0 | 572 | unsigned long rcu_batches_completed_sched(void) |
64db4cff | 573 | { |
d6714c22 | 574 | return rcu_sched_state.completed; |
64db4cff | 575 | } |
d6714c22 | 576 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
577 | |
578 | /* | |
917963d0 | 579 | * Return the number of RCU BH batches completed thus far for debug & stats. |
64db4cff | 580 | */ |
9733e4f0 | 581 | unsigned long rcu_batches_completed_bh(void) |
64db4cff PM |
582 | { |
583 | return rcu_bh_state.completed; | |
584 | } | |
585 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
586 | ||
291783b8 PM |
587 | /* |
588 | * Return the number of RCU expedited batches completed thus far for | |
589 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
590 | * numbers mean idle. The value returned will thus be roughly double | |
591 | * the cumulative batches since boot. | |
592 | */ | |
593 | unsigned long rcu_exp_batches_completed(void) | |
594 | { | |
595 | return rcu_state_p->expedited_sequence; | |
596 | } | |
597 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
598 | ||
599 | /* | |
600 | * Return the number of RCU-sched expedited batches completed thus far | |
601 | * for debug & stats. Similar to rcu_exp_batches_completed(). | |
602 | */ | |
603 | unsigned long rcu_exp_batches_completed_sched(void) | |
604 | { | |
605 | return rcu_sched_state.expedited_sequence; | |
606 | } | |
607 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed_sched); | |
608 | ||
a381d757 ACB |
609 | /* |
610 | * Force a quiescent state. | |
611 | */ | |
612 | void rcu_force_quiescent_state(void) | |
613 | { | |
e534165b | 614 | force_quiescent_state(rcu_state_p); |
a381d757 ACB |
615 | } |
616 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
617 | ||
bf66f18e PM |
618 | /* |
619 | * Force a quiescent state for RCU BH. | |
620 | */ | |
621 | void rcu_bh_force_quiescent_state(void) | |
622 | { | |
4cdfc175 | 623 | force_quiescent_state(&rcu_bh_state); |
bf66f18e PM |
624 | } |
625 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
626 | ||
e7580f33 PM |
627 | /* |
628 | * Force a quiescent state for RCU-sched. | |
629 | */ | |
630 | void rcu_sched_force_quiescent_state(void) | |
631 | { | |
632 | force_quiescent_state(&rcu_sched_state); | |
633 | } | |
634 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
635 | ||
afea227f PM |
636 | /* |
637 | * Show the state of the grace-period kthreads. | |
638 | */ | |
639 | void show_rcu_gp_kthreads(void) | |
640 | { | |
641 | struct rcu_state *rsp; | |
642 | ||
643 | for_each_rcu_flavor(rsp) { | |
644 | pr_info("%s: wait state: %d ->state: %#lx\n", | |
645 | rsp->name, rsp->gp_state, rsp->gp_kthread->state); | |
646 | /* sched_show_task(rsp->gp_kthread); */ | |
647 | } | |
648 | } | |
649 | EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads); | |
650 | ||
4a298656 PM |
651 | /* |
652 | * Record the number of times rcutorture tests have been initiated and | |
653 | * terminated. This information allows the debugfs tracing stats to be | |
654 | * correlated to the rcutorture messages, even when the rcutorture module | |
655 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
656 | * store this state in rcutorture itself. | |
657 | */ | |
658 | void rcutorture_record_test_transition(void) | |
659 | { | |
660 | rcutorture_testseq++; | |
661 | rcutorture_vernum = 0; | |
662 | } | |
663 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
664 | ||
ad0dc7f9 PM |
665 | /* |
666 | * Send along grace-period-related data for rcutorture diagnostics. | |
667 | */ | |
668 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
669 | unsigned long *gpnum, unsigned long *completed) | |
670 | { | |
671 | struct rcu_state *rsp = NULL; | |
672 | ||
673 | switch (test_type) { | |
674 | case RCU_FLAVOR: | |
e534165b | 675 | rsp = rcu_state_p; |
ad0dc7f9 PM |
676 | break; |
677 | case RCU_BH_FLAVOR: | |
678 | rsp = &rcu_bh_state; | |
679 | break; | |
680 | case RCU_SCHED_FLAVOR: | |
681 | rsp = &rcu_sched_state; | |
682 | break; | |
683 | default: | |
684 | break; | |
685 | } | |
7f6733c3 | 686 | if (rsp == NULL) |
ad0dc7f9 | 687 | return; |
7f6733c3 PM |
688 | *flags = READ_ONCE(rsp->gp_flags); |
689 | *gpnum = READ_ONCE(rsp->gpnum); | |
690 | *completed = READ_ONCE(rsp->completed); | |
ad0dc7f9 PM |
691 | } |
692 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
693 | ||
4a298656 PM |
694 | /* |
695 | * Record the number of writer passes through the current rcutorture test. | |
696 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
697 | * messages. | |
698 | */ | |
699 | void rcutorture_record_progress(unsigned long vernum) | |
700 | { | |
701 | rcutorture_vernum++; | |
702 | } | |
703 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
704 | ||
365187fb PM |
705 | /* |
706 | * Return the root node of the specified rcu_state structure. | |
707 | */ | |
708 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
709 | { | |
710 | return &rsp->node[0]; | |
711 | } | |
712 | ||
713 | /* | |
714 | * Is there any need for future grace periods? | |
715 | * Interrupts must be disabled. If the caller does not hold the root | |
716 | * rnp_node structure's ->lock, the results are advisory only. | |
717 | */ | |
718 | static int rcu_future_needs_gp(struct rcu_state *rsp) | |
719 | { | |
720 | struct rcu_node *rnp = rcu_get_root(rsp); | |
7d0ae808 | 721 | int idx = (READ_ONCE(rnp->completed) + 1) & 0x1; |
365187fb PM |
722 | int *fp = &rnp->need_future_gp[idx]; |
723 | ||
b04db8e1 | 724 | lockdep_assert_irqs_disabled(); |
7d0ae808 | 725 | return READ_ONCE(*fp); |
365187fb PM |
726 | } |
727 | ||
64db4cff | 728 | /* |
dc35c893 PM |
729 | * Does the current CPU require a not-yet-started grace period? |
730 | * The caller must have disabled interrupts to prevent races with | |
731 | * normal callback registry. | |
64db4cff | 732 | */ |
d117c8aa | 733 | static bool |
64db4cff PM |
734 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) |
735 | { | |
b04db8e1 | 736 | lockdep_assert_irqs_disabled(); |
dc35c893 | 737 | if (rcu_gp_in_progress(rsp)) |
d117c8aa | 738 | return false; /* No, a grace period is already in progress. */ |
365187fb | 739 | if (rcu_future_needs_gp(rsp)) |
d117c8aa | 740 | return true; /* Yes, a no-CBs CPU needs one. */ |
15fecf89 | 741 | if (!rcu_segcblist_is_enabled(&rdp->cblist)) |
d117c8aa | 742 | return false; /* No, this is a no-CBs (or offline) CPU. */ |
15fecf89 | 743 | if (!rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
d117c8aa | 744 | return true; /* Yes, CPU has newly registered callbacks. */ |
15fecf89 PM |
745 | if (rcu_segcblist_future_gp_needed(&rdp->cblist, |
746 | READ_ONCE(rsp->completed))) | |
747 | return true; /* Yes, CBs for future grace period. */ | |
d117c8aa | 748 | return false; /* No grace period needed. */ |
64db4cff PM |
749 | } |
750 | ||
9b2e4f18 | 751 | /* |
a278d471 | 752 | * rcu_eqs_enter_common - current CPU is entering an extended quiescent state |
9b2e4f18 | 753 | * |
a278d471 PM |
754 | * Enter idle, doing appropriate accounting. The caller must have |
755 | * disabled interrupts. | |
9b2e4f18 | 756 | */ |
a278d471 | 757 | static void rcu_eqs_enter_common(bool user) |
9b2e4f18 | 758 | { |
96d3fd0d PM |
759 | struct rcu_state *rsp; |
760 | struct rcu_data *rdp; | |
a278d471 | 761 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
96d3fd0d | 762 | |
b04db8e1 | 763 | lockdep_assert_irqs_disabled(); |
a278d471 | 764 | trace_rcu_dyntick(TPS("Start"), rdtp->dynticks_nesting, 0); |
1ce46ee5 PM |
765 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
766 | !user && !is_idle_task(current)) { | |
289828e6 PM |
767 | struct task_struct *idle __maybe_unused = |
768 | idle_task(smp_processor_id()); | |
0989cb46 | 769 | |
a278d471 | 770 | trace_rcu_dyntick(TPS("Error on entry: not idle task"), rdtp->dynticks_nesting, 0); |
274529ba | 771 | rcu_ftrace_dump(DUMP_ORIG); |
0989cb46 PM |
772 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
773 | current->pid, current->comm, | |
774 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 | 775 | } |
96d3fd0d PM |
776 | for_each_rcu_flavor(rsp) { |
777 | rdp = this_cpu_ptr(rsp->rda); | |
778 | do_nocb_deferred_wakeup(rdp); | |
779 | } | |
198bbf81 | 780 | rcu_prepare_for_idle(); |
844ccdd7 PM |
781 | rdtp->dynticks_nesting = 0; |
782 | rcu_dynticks_eqs_enter(); | |
176f8f7a | 783 | rcu_dynticks_task_enter(); |
c44e2cdd PM |
784 | |
785 | /* | |
adf5091e | 786 | * It is illegal to enter an extended quiescent state while |
c44e2cdd PM |
787 | * in an RCU read-side critical section. |
788 | */ | |
f78f5b90 PM |
789 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), |
790 | "Illegal idle entry in RCU read-side critical section."); | |
791 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), | |
792 | "Illegal idle entry in RCU-bh read-side critical section."); | |
793 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), | |
794 | "Illegal idle entry in RCU-sched read-side critical section."); | |
9b2e4f18 | 795 | } |
64db4cff | 796 | |
adf5091e FW |
797 | /* |
798 | * Enter an RCU extended quiescent state, which can be either the | |
799 | * idle loop or adaptive-tickless usermode execution. | |
51a1fd30 PM |
800 | * |
801 | * We crowbar the ->dynticks_nmi_nesting field to zero to allow for | |
802 | * the possibility of usermode upcalls having messed up our count | |
803 | * of interrupt nesting level during the prior busy period. | |
64db4cff | 804 | */ |
adf5091e | 805 | static void rcu_eqs_enter(bool user) |
64db4cff | 806 | { |
64db4cff PM |
807 | struct rcu_dynticks *rdtp; |
808 | ||
c9d4b0af | 809 | rdtp = this_cpu_ptr(&rcu_dynticks); |
6136d6e4 | 810 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0); |
1ce46ee5 | 811 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
51a1fd30 PM |
812 | rdtp->dynticks_nesting == 0); |
813 | if (rdtp->dynticks_nesting == 1) | |
a278d471 PM |
814 | rcu_eqs_enter_common(user); |
815 | else | |
51a1fd30 | 816 | rdtp->dynticks_nesting--; |
64db4cff | 817 | } |
adf5091e FW |
818 | |
819 | /** | |
820 | * rcu_idle_enter - inform RCU that current CPU is entering idle | |
821 | * | |
822 | * Enter idle mode, in other words, -leave- the mode in which RCU | |
823 | * read-side critical sections can occur. (Though RCU read-side | |
824 | * critical sections can occur in irq handlers in idle, a possibility | |
825 | * handled by irq_enter() and irq_exit().) | |
826 | * | |
c0da313e PM |
827 | * If you add or remove a call to rcu_idle_enter(), be sure to test with |
828 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
829 | */ |
830 | void rcu_idle_enter(void) | |
831 | { | |
b04db8e1 | 832 | lockdep_assert_irqs_disabled(); |
cb349ca9 | 833 | rcu_eqs_enter(false); |
adf5091e | 834 | } |
64db4cff | 835 | |
d1ec4c34 | 836 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
837 | /** |
838 | * rcu_user_enter - inform RCU that we are resuming userspace. | |
839 | * | |
840 | * Enter RCU idle mode right before resuming userspace. No use of RCU | |
841 | * is permitted between this call and rcu_user_exit(). This way the | |
842 | * CPU doesn't need to maintain the tick for RCU maintenance purposes | |
843 | * when the CPU runs in userspace. | |
c0da313e PM |
844 | * |
845 | * If you add or remove a call to rcu_user_enter(), be sure to test with | |
846 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
847 | */ |
848 | void rcu_user_enter(void) | |
849 | { | |
b04db8e1 | 850 | lockdep_assert_irqs_disabled(); |
d4db30af | 851 | rcu_eqs_enter(true); |
adf5091e | 852 | } |
d1ec4c34 | 853 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 854 | |
fd581a91 PM |
855 | /** |
856 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
857 | * | |
858 | * If we are returning from the outermost NMI handler that interrupted an | |
859 | * RCU-idle period, update rdtp->dynticks and rdtp->dynticks_nmi_nesting | |
860 | * to let the RCU grace-period handling know that the CPU is back to | |
861 | * being RCU-idle. | |
862 | * | |
863 | * If you add or remove a call to rcu_nmi_exit(), be sure to test | |
864 | * with CONFIG_RCU_EQS_DEBUG=y. | |
865 | */ | |
866 | void rcu_nmi_exit(void) | |
867 | { | |
868 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
869 | ||
870 | /* | |
871 | * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks. | |
872 | * (We are exiting an NMI handler, so RCU better be paying attention | |
873 | * to us!) | |
874 | */ | |
875 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting <= 0); | |
876 | WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs()); | |
877 | ||
878 | /* | |
879 | * If the nesting level is not 1, the CPU wasn't RCU-idle, so | |
880 | * leave it in non-RCU-idle state. | |
881 | */ | |
882 | if (rdtp->dynticks_nmi_nesting != 1) { | |
883 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* No store tearing. */ | |
884 | rdtp->dynticks_nmi_nesting - 2); | |
885 | return; | |
886 | } | |
887 | ||
888 | /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */ | |
889 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */ | |
890 | rcu_dynticks_eqs_enter(); | |
891 | } | |
892 | ||
9b2e4f18 PM |
893 | /** |
894 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
895 | * | |
896 | * Exit from an interrupt handler, which might possibly result in entering | |
897 | * idle mode, in other words, leaving the mode in which read-side critical | |
7c9906ca | 898 | * sections can occur. The caller must have disabled interrupts. |
64db4cff | 899 | * |
9b2e4f18 PM |
900 | * This code assumes that the idle loop never does anything that might |
901 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
58721f5d PM |
902 | * architecture's idle loop violates this assumption, RCU will give you what |
903 | * you deserve, good and hard. But very infrequently and irreproducibly. | |
9b2e4f18 PM |
904 | * |
905 | * Use things like work queues to work around this limitation. | |
906 | * | |
907 | * You have been warned. | |
c0da313e PM |
908 | * |
909 | * If you add or remove a call to rcu_irq_exit(), be sure to test with | |
910 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 911 | */ |
9b2e4f18 | 912 | void rcu_irq_exit(void) |
64db4cff | 913 | { |
58721f5d | 914 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
64db4cff | 915 | |
b04db8e1 | 916 | lockdep_assert_irqs_disabled(); |
58721f5d PM |
917 | if (rdtp->dynticks_nmi_nesting == 1) |
918 | rcu_prepare_for_idle(); | |
919 | rcu_nmi_exit(); | |
920 | if (rdtp->dynticks_nmi_nesting == 0) | |
921 | rcu_dynticks_task_enter(); | |
7c9906ca PM |
922 | } |
923 | ||
924 | /* | |
925 | * Wrapper for rcu_irq_exit() where interrupts are enabled. | |
c0da313e PM |
926 | * |
927 | * If you add or remove a call to rcu_irq_exit_irqson(), be sure to test | |
928 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
929 | */ |
930 | void rcu_irq_exit_irqson(void) | |
931 | { | |
932 | unsigned long flags; | |
933 | ||
934 | local_irq_save(flags); | |
935 | rcu_irq_exit(); | |
9b2e4f18 PM |
936 | local_irq_restore(flags); |
937 | } | |
938 | ||
939 | /* | |
adf5091e | 940 | * rcu_eqs_exit_common - current CPU moving away from extended quiescent state |
9b2e4f18 PM |
941 | * |
942 | * If the new value of the ->dynticks_nesting counter was previously zero, | |
943 | * we really have exited idle, and must do the appropriate accounting. | |
944 | * The caller must have disabled interrupts. | |
945 | */ | |
a0eb22bf | 946 | static void rcu_eqs_exit_common(long long newval, int user) |
9b2e4f18 | 947 | { |
2625d469 | 948 | RCU_TRACE(struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);) |
28ced795 | 949 | |
176f8f7a | 950 | rcu_dynticks_task_exit(); |
2625d469 | 951 | rcu_dynticks_eqs_exit(); |
8fa7845d | 952 | rcu_cleanup_after_idle(); |
a0eb22bf | 953 | trace_rcu_dyntick(TPS("End"), rdtp->dynticks_nesting, newval); |
1ce46ee5 PM |
954 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
955 | !user && !is_idle_task(current)) { | |
289828e6 PM |
956 | struct task_struct *idle __maybe_unused = |
957 | idle_task(smp_processor_id()); | |
0989cb46 | 958 | |
f7f7bac9 | 959 | trace_rcu_dyntick(TPS("Error on exit: not idle task"), |
a0eb22bf | 960 | rdtp->dynticks_nesting, newval); |
274529ba | 961 | rcu_ftrace_dump(DUMP_ORIG); |
0989cb46 PM |
962 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
963 | current->pid, current->comm, | |
964 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 PM |
965 | } |
966 | } | |
967 | ||
adf5091e FW |
968 | /* |
969 | * Exit an RCU extended quiescent state, which can be either the | |
970 | * idle loop or adaptive-tickless usermode execution. | |
51a1fd30 PM |
971 | * |
972 | * We crowbar the ->dynticks_nmi_nesting field to DYNTICK_IRQ_NONIDLE to | |
973 | * allow for the possibility of usermode upcalls messing up our count of | |
974 | * interrupt nesting level during the busy period that is just now starting. | |
9b2e4f18 | 975 | */ |
adf5091e | 976 | static void rcu_eqs_exit(bool user) |
9b2e4f18 | 977 | { |
9b2e4f18 PM |
978 | struct rcu_dynticks *rdtp; |
979 | long long oldval; | |
980 | ||
b04db8e1 | 981 | lockdep_assert_irqs_disabled(); |
c9d4b0af | 982 | rdtp = this_cpu_ptr(&rcu_dynticks); |
9b2e4f18 | 983 | oldval = rdtp->dynticks_nesting; |
1ce46ee5 | 984 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0); |
51a1fd30 PM |
985 | if (oldval) { |
986 | rdtp->dynticks_nesting++; | |
3a592405 | 987 | } else { |
51a1fd30 PM |
988 | rcu_eqs_exit_common(1, user); |
989 | rdtp->dynticks_nesting = 1; | |
6136d6e4 | 990 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, DYNTICK_IRQ_NONIDLE); |
3a592405 | 991 | } |
9b2e4f18 | 992 | } |
adf5091e FW |
993 | |
994 | /** | |
995 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
996 | * | |
997 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
998 | * read-side critical sections can occur. | |
999 | * | |
c0da313e PM |
1000 | * If you add or remove a call to rcu_idle_exit(), be sure to test with |
1001 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
1002 | */ |
1003 | void rcu_idle_exit(void) | |
1004 | { | |
c5d900bf FW |
1005 | unsigned long flags; |
1006 | ||
1007 | local_irq_save(flags); | |
cb349ca9 | 1008 | rcu_eqs_exit(false); |
c5d900bf | 1009 | local_irq_restore(flags); |
adf5091e | 1010 | } |
9b2e4f18 | 1011 | |
d1ec4c34 | 1012 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
1013 | /** |
1014 | * rcu_user_exit - inform RCU that we are exiting userspace. | |
1015 | * | |
1016 | * Exit RCU idle mode while entering the kernel because it can | |
1017 | * run a RCU read side critical section anytime. | |
c0da313e PM |
1018 | * |
1019 | * If you add or remove a call to rcu_user_exit(), be sure to test with | |
1020 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
1021 | */ |
1022 | void rcu_user_exit(void) | |
1023 | { | |
91d1aa43 | 1024 | rcu_eqs_exit(1); |
adf5091e | 1025 | } |
d1ec4c34 | 1026 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 1027 | |
fd581a91 PM |
1028 | /** |
1029 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
1030 | * | |
1031 | * If the CPU was idle from RCU's viewpoint, update rdtp->dynticks and | |
1032 | * rdtp->dynticks_nmi_nesting to let the RCU grace-period handling know | |
1033 | * that the CPU is active. This implementation permits nested NMIs, as | |
1034 | * long as the nesting level does not overflow an int. (You will probably | |
1035 | * run out of stack space first.) | |
1036 | * | |
1037 | * If you add or remove a call to rcu_nmi_enter(), be sure to test | |
1038 | * with CONFIG_RCU_EQS_DEBUG=y. | |
1039 | */ | |
1040 | void rcu_nmi_enter(void) | |
1041 | { | |
1042 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
1043 | int incby = 2; | |
1044 | ||
1045 | /* Complain about underflow. */ | |
1046 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting < 0); | |
1047 | ||
1048 | /* | |
1049 | * If idle from RCU viewpoint, atomically increment ->dynticks | |
1050 | * to mark non-idle and increment ->dynticks_nmi_nesting by one. | |
1051 | * Otherwise, increment ->dynticks_nmi_nesting by two. This means | |
1052 | * if ->dynticks_nmi_nesting is equal to one, we are guaranteed | |
1053 | * to be in the outermost NMI handler that interrupted an RCU-idle | |
1054 | * period (observation due to Andy Lutomirski). | |
1055 | */ | |
1056 | if (rcu_dynticks_curr_cpu_in_eqs()) { | |
1057 | rcu_dynticks_eqs_exit(); | |
1058 | incby = 1; | |
1059 | } | |
1060 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* Prevent store tearing. */ | |
1061 | rdtp->dynticks_nmi_nesting + incby); | |
1062 | barrier(); | |
1063 | } | |
1064 | ||
9b2e4f18 PM |
1065 | /** |
1066 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle | |
1067 | * | |
1068 | * Enter an interrupt handler, which might possibly result in exiting | |
1069 | * idle mode, in other words, entering the mode in which read-side critical | |
7c9906ca | 1070 | * sections can occur. The caller must have disabled interrupts. |
9b2e4f18 PM |
1071 | * |
1072 | * Note that the Linux kernel is fully capable of entering an interrupt | |
58721f5d PM |
1073 | * handler that it never exits, for example when doing upcalls to user mode! |
1074 | * This code assumes that the idle loop never does upcalls to user mode. | |
1075 | * If your architecture's idle loop does do upcalls to user mode (or does | |
1076 | * anything else that results in unbalanced calls to the irq_enter() and | |
1077 | * irq_exit() functions), RCU will give you what you deserve, good and hard. | |
1078 | * But very infrequently and irreproducibly. | |
9b2e4f18 PM |
1079 | * |
1080 | * Use things like work queues to work around this limitation. | |
1081 | * | |
1082 | * You have been warned. | |
c0da313e PM |
1083 | * |
1084 | * If you add or remove a call to rcu_irq_enter(), be sure to test with | |
1085 | * CONFIG_RCU_EQS_DEBUG=y. | |
9b2e4f18 PM |
1086 | */ |
1087 | void rcu_irq_enter(void) | |
1088 | { | |
58721f5d | 1089 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
9b2e4f18 | 1090 | |
b04db8e1 | 1091 | lockdep_assert_irqs_disabled(); |
58721f5d PM |
1092 | if (rdtp->dynticks_nmi_nesting == 0) |
1093 | rcu_dynticks_task_exit(); | |
1094 | rcu_nmi_enter(); | |
1095 | if (rdtp->dynticks_nmi_nesting == 1) | |
1096 | rcu_cleanup_after_idle(); | |
7c9906ca PM |
1097 | } |
1098 | ||
1099 | /* | |
1100 | * Wrapper for rcu_irq_enter() where interrupts are enabled. | |
c0da313e PM |
1101 | * |
1102 | * If you add or remove a call to rcu_irq_enter_irqson(), be sure to test | |
1103 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
1104 | */ |
1105 | void rcu_irq_enter_irqson(void) | |
1106 | { | |
1107 | unsigned long flags; | |
1108 | ||
1109 | local_irq_save(flags); | |
1110 | rcu_irq_enter(); | |
64db4cff | 1111 | local_irq_restore(flags); |
64db4cff PM |
1112 | } |
1113 | ||
5c173eb8 PM |
1114 | /** |
1115 | * rcu_is_watching - see if RCU thinks that the current CPU is idle | |
64db4cff | 1116 | * |
791875d1 PM |
1117 | * Return true if RCU is watching the running CPU, which means that this |
1118 | * CPU can safely enter RCU read-side critical sections. In other words, | |
1119 | * if the current CPU is in its idle loop and is neither in an interrupt | |
34240697 | 1120 | * or NMI handler, return true. |
64db4cff | 1121 | */ |
9418fb20 | 1122 | bool notrace rcu_is_watching(void) |
64db4cff | 1123 | { |
f534ed1f | 1124 | bool ret; |
34240697 | 1125 | |
46f00d18 | 1126 | preempt_disable_notrace(); |
791875d1 | 1127 | ret = !rcu_dynticks_curr_cpu_in_eqs(); |
46f00d18 | 1128 | preempt_enable_notrace(); |
34240697 | 1129 | return ret; |
64db4cff | 1130 | } |
5c173eb8 | 1131 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 1132 | |
bcbfdd01 PM |
1133 | /* |
1134 | * If a holdout task is actually running, request an urgent quiescent | |
1135 | * state from its CPU. This is unsynchronized, so migrations can cause | |
1136 | * the request to go to the wrong CPU. Which is OK, all that will happen | |
1137 | * is that the CPU's next context switch will be a bit slower and next | |
1138 | * time around this task will generate another request. | |
1139 | */ | |
1140 | void rcu_request_urgent_qs_task(struct task_struct *t) | |
1141 | { | |
1142 | int cpu; | |
1143 | ||
1144 | barrier(); | |
1145 | cpu = task_cpu(t); | |
1146 | if (!task_curr(t)) | |
1147 | return; /* This task is not running on that CPU. */ | |
1148 | smp_store_release(per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, cpu), true); | |
1149 | } | |
1150 | ||
62fde6ed | 1151 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
1152 | |
1153 | /* | |
1154 | * Is the current CPU online? Disable preemption to avoid false positives | |
1155 | * that could otherwise happen due to the current CPU number being sampled, | |
1156 | * this task being preempted, its old CPU being taken offline, resuming | |
1157 | * on some other CPU, then determining that its old CPU is now offline. | |
1158 | * It is OK to use RCU on an offline processor during initial boot, hence | |
2036d94a PM |
1159 | * the check for rcu_scheduler_fully_active. Note also that it is OK |
1160 | * for a CPU coming online to use RCU for one jiffy prior to marking itself | |
1161 | * online in the cpu_online_mask. Similarly, it is OK for a CPU going | |
1162 | * offline to continue to use RCU for one jiffy after marking itself | |
1163 | * offline in the cpu_online_mask. This leniency is necessary given the | |
1164 | * non-atomic nature of the online and offline processing, for example, | |
4df83742 TG |
1165 | * the fact that a CPU enters the scheduler after completing the teardown |
1166 | * of the CPU. | |
2036d94a | 1167 | * |
4df83742 TG |
1168 | * This is also why RCU internally marks CPUs online during in the |
1169 | * preparation phase and offline after the CPU has been taken down. | |
c0d6d01b PM |
1170 | * |
1171 | * Disable checking if in an NMI handler because we cannot safely report | |
1172 | * errors from NMI handlers anyway. | |
1173 | */ | |
1174 | bool rcu_lockdep_current_cpu_online(void) | |
1175 | { | |
2036d94a PM |
1176 | struct rcu_data *rdp; |
1177 | struct rcu_node *rnp; | |
c0d6d01b PM |
1178 | bool ret; |
1179 | ||
1180 | if (in_nmi()) | |
f6f7ee9a | 1181 | return true; |
c0d6d01b | 1182 | preempt_disable(); |
c9d4b0af | 1183 | rdp = this_cpu_ptr(&rcu_sched_data); |
2036d94a | 1184 | rnp = rdp->mynode; |
0aa04b05 | 1185 | ret = (rdp->grpmask & rcu_rnp_online_cpus(rnp)) || |
c0d6d01b PM |
1186 | !rcu_scheduler_fully_active; |
1187 | preempt_enable(); | |
1188 | return ret; | |
1189 | } | |
1190 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
1191 | ||
62fde6ed | 1192 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 1193 | |
64db4cff | 1194 | /** |
9b2e4f18 | 1195 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle |
64db4cff | 1196 | * |
9b2e4f18 PM |
1197 | * If the current CPU is idle or running at a first-level (not nested) |
1198 | * interrupt from idle, return true. The caller must have at least | |
1199 | * disabled preemption. | |
64db4cff | 1200 | */ |
62e3cb14 | 1201 | static int rcu_is_cpu_rrupt_from_idle(void) |
64db4cff | 1202 | { |
51a1fd30 PM |
1203 | return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 0 && |
1204 | __this_cpu_read(rcu_dynticks.dynticks_nmi_nesting) <= 1; | |
64db4cff PM |
1205 | } |
1206 | ||
9b9500da PM |
1207 | /* |
1208 | * We are reporting a quiescent state on behalf of some other CPU, so | |
1209 | * it is our responsibility to check for and handle potential overflow | |
1210 | * of the rcu_node ->gpnum counter with respect to the rcu_data counters. | |
1211 | * After all, the CPU might be in deep idle state, and thus executing no | |
1212 | * code whatsoever. | |
1213 | */ | |
1214 | static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp) | |
1215 | { | |
1216 | lockdep_assert_held(&rnp->lock); | |
1217 | if (ULONG_CMP_LT(READ_ONCE(rdp->gpnum) + ULONG_MAX / 4, rnp->gpnum)) | |
1218 | WRITE_ONCE(rdp->gpwrap, true); | |
1219 | if (ULONG_CMP_LT(rdp->rcu_iw_gpnum + ULONG_MAX / 4, rnp->gpnum)) | |
1220 | rdp->rcu_iw_gpnum = rnp->gpnum + ULONG_MAX / 4; | |
1221 | } | |
1222 | ||
64db4cff PM |
1223 | /* |
1224 | * Snapshot the specified CPU's dynticks counter so that we can later | |
1225 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 1226 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 1227 | */ |
fe5ac724 | 1228 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
64db4cff | 1229 | { |
8b2f63ab | 1230 | rdp->dynticks_snap = rcu_dynticks_snap(rdp->dynticks); |
02a5c550 | 1231 | if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) { |
7941dbde | 1232 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti")); |
9b9500da | 1233 | rcu_gpnum_ovf(rdp->mynode, rdp); |
23a9bacd | 1234 | return 1; |
7941dbde | 1235 | } |
23a9bacd | 1236 | return 0; |
64db4cff PM |
1237 | } |
1238 | ||
9b9500da PM |
1239 | /* |
1240 | * Handler for the irq_work request posted when a grace period has | |
1241 | * gone on for too long, but not yet long enough for an RCU CPU | |
1242 | * stall warning. Set state appropriately, but just complain if | |
1243 | * there is unexpected state on entry. | |
1244 | */ | |
1245 | static void rcu_iw_handler(struct irq_work *iwp) | |
1246 | { | |
1247 | struct rcu_data *rdp; | |
1248 | struct rcu_node *rnp; | |
1249 | ||
1250 | rdp = container_of(iwp, struct rcu_data, rcu_iw); | |
1251 | rnp = rdp->mynode; | |
1252 | raw_spin_lock_rcu_node(rnp); | |
1253 | if (!WARN_ON_ONCE(!rdp->rcu_iw_pending)) { | |
1254 | rdp->rcu_iw_gpnum = rnp->gpnum; | |
1255 | rdp->rcu_iw_pending = false; | |
1256 | } | |
1257 | raw_spin_unlock_rcu_node(rnp); | |
1258 | } | |
1259 | ||
64db4cff PM |
1260 | /* |
1261 | * Return true if the specified CPU has passed through a quiescent | |
1262 | * state by virtue of being in or having passed through an dynticks | |
1263 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 1264 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 1265 | */ |
fe5ac724 | 1266 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
64db4cff | 1267 | { |
3a19b46a | 1268 | unsigned long jtsq; |
0f9be8ca | 1269 | bool *rnhqp; |
9226b10d | 1270 | bool *ruqp; |
9b9500da | 1271 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
1272 | |
1273 | /* | |
1274 | * If the CPU passed through or entered a dynticks idle phase with | |
1275 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
1276 | * already acknowledged the request to pass through a quiescent | |
1277 | * state. Either way, that CPU cannot possibly be in an RCU | |
1278 | * read-side critical section that started before the beginning | |
1279 | * of the current RCU grace period. | |
1280 | */ | |
02a5c550 | 1281 | if (rcu_dynticks_in_eqs_since(rdp->dynticks, rdp->dynticks_snap)) { |
f7f7bac9 | 1282 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti")); |
64db4cff | 1283 | rdp->dynticks_fqs++; |
9b9500da | 1284 | rcu_gpnum_ovf(rnp, rdp); |
64db4cff PM |
1285 | return 1; |
1286 | } | |
1287 | ||
a82dcc76 | 1288 | /* |
3a19b46a PM |
1289 | * Has this CPU encountered a cond_resched_rcu_qs() since the |
1290 | * beginning of the grace period? For this to be the case, | |
1291 | * the CPU has to have noticed the current grace period. This | |
1292 | * might not be the case for nohz_full CPUs looping in the kernel. | |
a82dcc76 | 1293 | */ |
f79c3ad6 | 1294 | jtsq = jiffies_till_sched_qs; |
9226b10d | 1295 | ruqp = per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, rdp->cpu); |
3a19b46a | 1296 | if (time_after(jiffies, rdp->rsp->gp_start + jtsq) && |
9577df9a | 1297 | READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) && |
3a19b46a PM |
1298 | READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) { |
1299 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc")); | |
9b9500da | 1300 | rcu_gpnum_ovf(rnp, rdp); |
3a19b46a | 1301 | return 1; |
f79c3ad6 | 1302 | } else if (time_after(jiffies, rdp->rsp->gp_start + jtsq)) { |
9226b10d PM |
1303 | /* Load rcu_qs_ctr before store to rcu_urgent_qs. */ |
1304 | smp_store_release(ruqp, true); | |
3a19b46a PM |
1305 | } |
1306 | ||
38d30b33 PM |
1307 | /* Check for the CPU being offline. */ |
1308 | if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp))) { | |
f7f7bac9 | 1309 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl")); |
a82dcc76 | 1310 | rdp->offline_fqs++; |
9b9500da | 1311 | rcu_gpnum_ovf(rnp, rdp); |
a82dcc76 PM |
1312 | return 1; |
1313 | } | |
65d798f0 PM |
1314 | |
1315 | /* | |
4a81e832 PM |
1316 | * A CPU running for an extended time within the kernel can |
1317 | * delay RCU grace periods. When the CPU is in NO_HZ_FULL mode, | |
1318 | * even context-switching back and forth between a pair of | |
1319 | * in-kernel CPU-bound tasks cannot advance grace periods. | |
1320 | * So if the grace period is old enough, make the CPU pay attention. | |
1321 | * Note that the unsynchronized assignments to the per-CPU | |
0f9be8ca | 1322 | * rcu_need_heavy_qs variable are safe. Yes, setting of |
4a81e832 PM |
1323 | * bits can be lost, but they will be set again on the next |
1324 | * force-quiescent-state pass. So lost bit sets do not result | |
1325 | * in incorrect behavior, merely in a grace period lasting | |
1326 | * a few jiffies longer than it might otherwise. Because | |
1327 | * there are at most four threads involved, and because the | |
1328 | * updates are only once every few jiffies, the probability of | |
1329 | * lossage (and thus of slight grace-period extension) is | |
1330 | * quite low. | |
6193c76a | 1331 | */ |
0f9be8ca PM |
1332 | rnhqp = &per_cpu(rcu_dynticks.rcu_need_heavy_qs, rdp->cpu); |
1333 | if (!READ_ONCE(*rnhqp) && | |
1334 | (time_after(jiffies, rdp->rsp->gp_start + jtsq) || | |
1335 | time_after(jiffies, rdp->rsp->jiffies_resched))) { | |
1336 | WRITE_ONCE(*rnhqp, true); | |
9226b10d PM |
1337 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ |
1338 | smp_store_release(ruqp, true); | |
f79c3ad6 | 1339 | rdp->rsp->jiffies_resched += jtsq; /* Re-enable beating. */ |
6193c76a PM |
1340 | } |
1341 | ||
28053bc7 | 1342 | /* |
9b9500da PM |
1343 | * If more than halfway to RCU CPU stall-warning time, do a |
1344 | * resched_cpu() to try to loosen things up a bit. Also check to | |
1345 | * see if the CPU is getting hammered with interrupts, but only | |
1346 | * once per grace period, just to keep the IPIs down to a dull roar. | |
28053bc7 | 1347 | */ |
9b9500da | 1348 | if (jiffies - rdp->rsp->gp_start > rcu_jiffies_till_stall_check() / 2) { |
28053bc7 | 1349 | resched_cpu(rdp->cpu); |
9b9500da PM |
1350 | if (IS_ENABLED(CONFIG_IRQ_WORK) && |
1351 | !rdp->rcu_iw_pending && rdp->rcu_iw_gpnum != rnp->gpnum && | |
1352 | (rnp->ffmask & rdp->grpmask)) { | |
1353 | init_irq_work(&rdp->rcu_iw, rcu_iw_handler); | |
1354 | rdp->rcu_iw_pending = true; | |
1355 | rdp->rcu_iw_gpnum = rnp->gpnum; | |
1356 | irq_work_queue_on(&rdp->rcu_iw, rdp->cpu); | |
1357 | } | |
1358 | } | |
4914950a | 1359 | |
a82dcc76 | 1360 | return 0; |
64db4cff PM |
1361 | } |
1362 | ||
64db4cff PM |
1363 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
1364 | { | |
cb1e78cf | 1365 | unsigned long j = jiffies; |
6193c76a | 1366 | unsigned long j1; |
26cdfedf PM |
1367 | |
1368 | rsp->gp_start = j; | |
1369 | smp_wmb(); /* Record start time before stall time. */ | |
6193c76a | 1370 | j1 = rcu_jiffies_till_stall_check(); |
7d0ae808 | 1371 | WRITE_ONCE(rsp->jiffies_stall, j + j1); |
6193c76a | 1372 | rsp->jiffies_resched = j + j1 / 2; |
7d0ae808 | 1373 | rsp->n_force_qs_gpstart = READ_ONCE(rsp->n_force_qs); |
64db4cff PM |
1374 | } |
1375 | ||
6b50e119 PM |
1376 | /* |
1377 | * Convert a ->gp_state value to a character string. | |
1378 | */ | |
1379 | static const char *gp_state_getname(short gs) | |
1380 | { | |
1381 | if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names)) | |
1382 | return "???"; | |
1383 | return gp_state_names[gs]; | |
1384 | } | |
1385 | ||
fb81a44b PM |
1386 | /* |
1387 | * Complain about starvation of grace-period kthread. | |
1388 | */ | |
1389 | static void rcu_check_gp_kthread_starvation(struct rcu_state *rsp) | |
1390 | { | |
1391 | unsigned long gpa; | |
1392 | unsigned long j; | |
1393 | ||
1394 | j = jiffies; | |
7d0ae808 | 1395 | gpa = READ_ONCE(rsp->gp_activity); |
b1adb3e2 | 1396 | if (j - gpa > 2 * HZ) { |
96036c43 | 1397 | pr_err("%s kthread starved for %ld jiffies! g%lu c%lu f%#x %s(%d) ->state=%#lx ->cpu=%d\n", |
81e701e4 | 1398 | rsp->name, j - gpa, |
319362c9 | 1399 | rsp->gpnum, rsp->completed, |
6b50e119 PM |
1400 | rsp->gp_flags, |
1401 | gp_state_getname(rsp->gp_state), rsp->gp_state, | |
96036c43 PM |
1402 | rsp->gp_kthread ? rsp->gp_kthread->state : ~0, |
1403 | rsp->gp_kthread ? task_cpu(rsp->gp_kthread) : -1); | |
86057b80 | 1404 | if (rsp->gp_kthread) { |
b1adb3e2 | 1405 | sched_show_task(rsp->gp_kthread); |
86057b80 PM |
1406 | wake_up_process(rsp->gp_kthread); |
1407 | } | |
b1adb3e2 | 1408 | } |
64db4cff PM |
1409 | } |
1410 | ||
b637a328 | 1411 | /* |
7aa92230 PM |
1412 | * Dump stacks of all tasks running on stalled CPUs. First try using |
1413 | * NMIs, but fall back to manual remote stack tracing on architectures | |
1414 | * that don't support NMI-based stack dumps. The NMI-triggered stack | |
1415 | * traces are more accurate because they are printed by the target CPU. | |
b637a328 PM |
1416 | */ |
1417 | static void rcu_dump_cpu_stacks(struct rcu_state *rsp) | |
1418 | { | |
1419 | int cpu; | |
1420 | unsigned long flags; | |
1421 | struct rcu_node *rnp; | |
1422 | ||
1423 | rcu_for_each_leaf_node(rsp, rnp) { | |
6cf10081 | 1424 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7aa92230 PM |
1425 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1426 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) | |
1427 | if (!trigger_single_cpu_backtrace(cpu)) | |
bc75e999 | 1428 | dump_cpu_task(cpu); |
67c583a7 | 1429 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b637a328 PM |
1430 | } |
1431 | } | |
1432 | ||
8c7c4829 PM |
1433 | /* |
1434 | * If too much time has passed in the current grace period, and if | |
1435 | * so configured, go kick the relevant kthreads. | |
1436 | */ | |
1437 | static void rcu_stall_kick_kthreads(struct rcu_state *rsp) | |
1438 | { | |
1439 | unsigned long j; | |
1440 | ||
1441 | if (!rcu_kick_kthreads) | |
1442 | return; | |
1443 | j = READ_ONCE(rsp->jiffies_kick_kthreads); | |
aa3e0bf1 PM |
1444 | if (time_after(jiffies, j) && rsp->gp_kthread && |
1445 | (rcu_gp_in_progress(rsp) || READ_ONCE(rsp->gp_flags))) { | |
8c7c4829 | 1446 | WARN_ONCE(1, "Kicking %s grace-period kthread\n", rsp->name); |
5dffed1e | 1447 | rcu_ftrace_dump(DUMP_ALL); |
8c7c4829 PM |
1448 | wake_up_process(rsp->gp_kthread); |
1449 | WRITE_ONCE(rsp->jiffies_kick_kthreads, j + HZ); | |
1450 | } | |
1451 | } | |
1452 | ||
088e9d25 DBO |
1453 | static inline void panic_on_rcu_stall(void) |
1454 | { | |
1455 | if (sysctl_panic_on_rcu_stall) | |
1456 | panic("RCU Stall\n"); | |
1457 | } | |
1458 | ||
6ccd2ecd | 1459 | static void print_other_cpu_stall(struct rcu_state *rsp, unsigned long gpnum) |
64db4cff PM |
1460 | { |
1461 | int cpu; | |
1462 | long delta; | |
1463 | unsigned long flags; | |
6ccd2ecd PM |
1464 | unsigned long gpa; |
1465 | unsigned long j; | |
285fe294 | 1466 | int ndetected = 0; |
64db4cff | 1467 | struct rcu_node *rnp = rcu_get_root(rsp); |
53bb857c | 1468 | long totqlen = 0; |
64db4cff | 1469 | |
8c7c4829 PM |
1470 | /* Kick and suppress, if so configured. */ |
1471 | rcu_stall_kick_kthreads(rsp); | |
1472 | if (rcu_cpu_stall_suppress) | |
1473 | return; | |
1474 | ||
64db4cff PM |
1475 | /* Only let one CPU complain about others per time interval. */ |
1476 | ||
6cf10081 | 1477 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7d0ae808 | 1478 | delta = jiffies - READ_ONCE(rsp->jiffies_stall); |
fc2219d4 | 1479 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
67c583a7 | 1480 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1481 | return; |
1482 | } | |
7d0ae808 PM |
1483 | WRITE_ONCE(rsp->jiffies_stall, |
1484 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
67c583a7 | 1485 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1486 | |
8cdd32a9 PM |
1487 | /* |
1488 | * OK, time to rat on our buddy... | |
1489 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1490 | * RCU CPU stall warnings. | |
1491 | */ | |
d7f3e207 | 1492 | pr_err("INFO: %s detected stalls on CPUs/tasks:", |
4300aa64 | 1493 | rsp->name); |
a858af28 | 1494 | print_cpu_stall_info_begin(); |
a0b6c9a7 | 1495 | rcu_for_each_leaf_node(rsp, rnp) { |
6cf10081 | 1496 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
9bc8b558 | 1497 | ndetected += rcu_print_task_stall(rnp); |
c8020a67 | 1498 | if (rnp->qsmask != 0) { |
bc75e999 MR |
1499 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1500 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) { | |
1501 | print_cpu_stall_info(rsp, cpu); | |
c8020a67 PM |
1502 | ndetected++; |
1503 | } | |
1504 | } | |
67c583a7 | 1505 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1506 | } |
a858af28 | 1507 | |
a858af28 | 1508 | print_cpu_stall_info_end(); |
53bb857c | 1509 | for_each_possible_cpu(cpu) |
15fecf89 PM |
1510 | totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(rsp->rda, |
1511 | cpu)->cblist); | |
83ebe63e | 1512 | pr_cont("(detected by %d, t=%ld jiffies, g=%ld, c=%ld, q=%lu)\n", |
eee05882 | 1513 | smp_processor_id(), (long)(jiffies - rsp->gp_start), |
83ebe63e | 1514 | (long)rsp->gpnum, (long)rsp->completed, totqlen); |
6ccd2ecd | 1515 | if (ndetected) { |
b637a328 | 1516 | rcu_dump_cpu_stacks(rsp); |
c4402b27 BP |
1517 | |
1518 | /* Complain about tasks blocking the grace period. */ | |
1519 | rcu_print_detail_task_stall(rsp); | |
6ccd2ecd | 1520 | } else { |
7d0ae808 PM |
1521 | if (READ_ONCE(rsp->gpnum) != gpnum || |
1522 | READ_ONCE(rsp->completed) == gpnum) { | |
6ccd2ecd PM |
1523 | pr_err("INFO: Stall ended before state dump start\n"); |
1524 | } else { | |
1525 | j = jiffies; | |
7d0ae808 | 1526 | gpa = READ_ONCE(rsp->gp_activity); |
237a0f21 | 1527 | pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n", |
6ccd2ecd | 1528 | rsp->name, j - gpa, j, gpa, |
237a0f21 PM |
1529 | jiffies_till_next_fqs, |
1530 | rcu_get_root(rsp)->qsmask); | |
6ccd2ecd PM |
1531 | /* In this case, the current CPU might be at fault. */ |
1532 | sched_show_task(current); | |
1533 | } | |
1534 | } | |
c1dc0b9c | 1535 | |
fb81a44b PM |
1536 | rcu_check_gp_kthread_starvation(rsp); |
1537 | ||
088e9d25 DBO |
1538 | panic_on_rcu_stall(); |
1539 | ||
4cdfc175 | 1540 | force_quiescent_state(rsp); /* Kick them all. */ |
64db4cff PM |
1541 | } |
1542 | ||
1543 | static void print_cpu_stall(struct rcu_state *rsp) | |
1544 | { | |
53bb857c | 1545 | int cpu; |
64db4cff | 1546 | unsigned long flags; |
9b9500da | 1547 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 1548 | struct rcu_node *rnp = rcu_get_root(rsp); |
53bb857c | 1549 | long totqlen = 0; |
64db4cff | 1550 | |
8c7c4829 PM |
1551 | /* Kick and suppress, if so configured. */ |
1552 | rcu_stall_kick_kthreads(rsp); | |
1553 | if (rcu_cpu_stall_suppress) | |
1554 | return; | |
1555 | ||
8cdd32a9 PM |
1556 | /* |
1557 | * OK, time to rat on ourselves... | |
1558 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1559 | * RCU CPU stall warnings. | |
1560 | */ | |
d7f3e207 | 1561 | pr_err("INFO: %s self-detected stall on CPU", rsp->name); |
a858af28 | 1562 | print_cpu_stall_info_begin(); |
9b9500da | 1563 | raw_spin_lock_irqsave_rcu_node(rdp->mynode, flags); |
a858af28 | 1564 | print_cpu_stall_info(rsp, smp_processor_id()); |
9b9500da | 1565 | raw_spin_unlock_irqrestore_rcu_node(rdp->mynode, flags); |
a858af28 | 1566 | print_cpu_stall_info_end(); |
53bb857c | 1567 | for_each_possible_cpu(cpu) |
15fecf89 PM |
1568 | totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(rsp->rda, |
1569 | cpu)->cblist); | |
83ebe63e PM |
1570 | pr_cont(" (t=%lu jiffies g=%ld c=%ld q=%lu)\n", |
1571 | jiffies - rsp->gp_start, | |
1572 | (long)rsp->gpnum, (long)rsp->completed, totqlen); | |
fb81a44b PM |
1573 | |
1574 | rcu_check_gp_kthread_starvation(rsp); | |
1575 | ||
bc1dce51 | 1576 | rcu_dump_cpu_stacks(rsp); |
c1dc0b9c | 1577 | |
6cf10081 | 1578 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7d0ae808 PM |
1579 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall))) |
1580 | WRITE_ONCE(rsp->jiffies_stall, | |
1581 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
67c583a7 | 1582 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c1dc0b9c | 1583 | |
088e9d25 DBO |
1584 | panic_on_rcu_stall(); |
1585 | ||
b021fe3e PZ |
1586 | /* |
1587 | * Attempt to revive the RCU machinery by forcing a context switch. | |
1588 | * | |
1589 | * A context switch would normally allow the RCU state machine to make | |
1590 | * progress and it could be we're stuck in kernel space without context | |
1591 | * switches for an entirely unreasonable amount of time. | |
1592 | */ | |
1593 | resched_cpu(smp_processor_id()); | |
64db4cff PM |
1594 | } |
1595 | ||
1596 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
1597 | { | |
26cdfedf PM |
1598 | unsigned long completed; |
1599 | unsigned long gpnum; | |
1600 | unsigned long gps; | |
bad6e139 PM |
1601 | unsigned long j; |
1602 | unsigned long js; | |
64db4cff PM |
1603 | struct rcu_node *rnp; |
1604 | ||
8c7c4829 PM |
1605 | if ((rcu_cpu_stall_suppress && !rcu_kick_kthreads) || |
1606 | !rcu_gp_in_progress(rsp)) | |
c68de209 | 1607 | return; |
8c7c4829 | 1608 | rcu_stall_kick_kthreads(rsp); |
cb1e78cf | 1609 | j = jiffies; |
26cdfedf PM |
1610 | |
1611 | /* | |
1612 | * Lots of memory barriers to reject false positives. | |
1613 | * | |
1614 | * The idea is to pick up rsp->gpnum, then rsp->jiffies_stall, | |
1615 | * then rsp->gp_start, and finally rsp->completed. These values | |
1616 | * are updated in the opposite order with memory barriers (or | |
1617 | * equivalent) during grace-period initialization and cleanup. | |
1618 | * Now, a false positive can occur if we get an new value of | |
1619 | * rsp->gp_start and a old value of rsp->jiffies_stall. But given | |
1620 | * the memory barriers, the only way that this can happen is if one | |
1621 | * grace period ends and another starts between these two fetches. | |
1622 | * Detect this by comparing rsp->completed with the previous fetch | |
1623 | * from rsp->gpnum. | |
1624 | * | |
1625 | * Given this check, comparisons of jiffies, rsp->jiffies_stall, | |
1626 | * and rsp->gp_start suffice to forestall false positives. | |
1627 | */ | |
7d0ae808 | 1628 | gpnum = READ_ONCE(rsp->gpnum); |
26cdfedf | 1629 | smp_rmb(); /* Pick up ->gpnum first... */ |
7d0ae808 | 1630 | js = READ_ONCE(rsp->jiffies_stall); |
26cdfedf | 1631 | smp_rmb(); /* ...then ->jiffies_stall before the rest... */ |
7d0ae808 | 1632 | gps = READ_ONCE(rsp->gp_start); |
26cdfedf | 1633 | smp_rmb(); /* ...and finally ->gp_start before ->completed. */ |
7d0ae808 | 1634 | completed = READ_ONCE(rsp->completed); |
26cdfedf PM |
1635 | if (ULONG_CMP_GE(completed, gpnum) || |
1636 | ULONG_CMP_LT(j, js) || | |
1637 | ULONG_CMP_GE(gps, js)) | |
1638 | return; /* No stall or GP completed since entering function. */ | |
64db4cff | 1639 | rnp = rdp->mynode; |
c96ea7cf | 1640 | if (rcu_gp_in_progress(rsp) && |
7d0ae808 | 1641 | (READ_ONCE(rnp->qsmask) & rdp->grpmask)) { |
64db4cff PM |
1642 | |
1643 | /* We haven't checked in, so go dump stack. */ | |
1644 | print_cpu_stall(rsp); | |
1645 | ||
bad6e139 PM |
1646 | } else if (rcu_gp_in_progress(rsp) && |
1647 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 1648 | |
bad6e139 | 1649 | /* They had a few time units to dump stack, so complain. */ |
6ccd2ecd | 1650 | print_other_cpu_stall(rsp, gpnum); |
64db4cff PM |
1651 | } |
1652 | } | |
1653 | ||
53d84e00 PM |
1654 | /** |
1655 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
1656 | * | |
1657 | * Set the stall-warning timeout way off into the future, thus preventing | |
1658 | * any RCU CPU stall-warning messages from appearing in the current set of | |
1659 | * RCU grace periods. | |
1660 | * | |
1661 | * The caller must disable hard irqs. | |
1662 | */ | |
1663 | void rcu_cpu_stall_reset(void) | |
1664 | { | |
6ce75a23 PM |
1665 | struct rcu_state *rsp; |
1666 | ||
1667 | for_each_rcu_flavor(rsp) | |
7d0ae808 | 1668 | WRITE_ONCE(rsp->jiffies_stall, jiffies + ULONG_MAX / 2); |
53d84e00 PM |
1669 | } |
1670 | ||
dc35c893 PM |
1671 | /* |
1672 | * Determine the value that ->completed will have at the end of the | |
1673 | * next subsequent grace period. This is used to tag callbacks so that | |
1674 | * a CPU can invoke callbacks in a timely fashion even if that CPU has | |
1675 | * been dyntick-idle for an extended period with callbacks under the | |
1676 | * influence of RCU_FAST_NO_HZ. | |
1677 | * | |
1678 | * The caller must hold rnp->lock with interrupts disabled. | |
1679 | */ | |
1680 | static unsigned long rcu_cbs_completed(struct rcu_state *rsp, | |
1681 | struct rcu_node *rnp) | |
1682 | { | |
c0b334c5 PM |
1683 | lockdep_assert_held(&rnp->lock); |
1684 | ||
dc35c893 PM |
1685 | /* |
1686 | * If RCU is idle, we just wait for the next grace period. | |
1687 | * But we can only be sure that RCU is idle if we are looking | |
1688 | * at the root rcu_node structure -- otherwise, a new grace | |
1689 | * period might have started, but just not yet gotten around | |
1690 | * to initializing the current non-root rcu_node structure. | |
1691 | */ | |
1692 | if (rcu_get_root(rsp) == rnp && rnp->gpnum == rnp->completed) | |
1693 | return rnp->completed + 1; | |
1694 | ||
1695 | /* | |
1696 | * Otherwise, wait for a possible partial grace period and | |
1697 | * then the subsequent full grace period. | |
1698 | */ | |
1699 | return rnp->completed + 2; | |
1700 | } | |
1701 | ||
0446be48 PM |
1702 | /* |
1703 | * Trace-event helper function for rcu_start_future_gp() and | |
1704 | * rcu_nocb_wait_gp(). | |
1705 | */ | |
1706 | static void trace_rcu_future_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
e66c33d5 | 1707 | unsigned long c, const char *s) |
0446be48 PM |
1708 | { |
1709 | trace_rcu_future_grace_period(rdp->rsp->name, rnp->gpnum, | |
1710 | rnp->completed, c, rnp->level, | |
1711 | rnp->grplo, rnp->grphi, s); | |
1712 | } | |
1713 | ||
1714 | /* | |
1715 | * Start some future grace period, as needed to handle newly arrived | |
1716 | * callbacks. The required future grace periods are recorded in each | |
48a7639c PM |
1717 | * rcu_node structure's ->need_future_gp field. Returns true if there |
1718 | * is reason to awaken the grace-period kthread. | |
0446be48 PM |
1719 | * |
1720 | * The caller must hold the specified rcu_node structure's ->lock. | |
1721 | */ | |
48a7639c PM |
1722 | static bool __maybe_unused |
1723 | rcu_start_future_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
1724 | unsigned long *c_out) | |
0446be48 PM |
1725 | { |
1726 | unsigned long c; | |
48a7639c | 1727 | bool ret = false; |
0446be48 PM |
1728 | struct rcu_node *rnp_root = rcu_get_root(rdp->rsp); |
1729 | ||
c0b334c5 PM |
1730 | lockdep_assert_held(&rnp->lock); |
1731 | ||
0446be48 PM |
1732 | /* |
1733 | * Pick up grace-period number for new callbacks. If this | |
1734 | * grace period is already marked as needed, return to the caller. | |
1735 | */ | |
1736 | c = rcu_cbs_completed(rdp->rsp, rnp); | |
f7f7bac9 | 1737 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startleaf")); |
0446be48 | 1738 | if (rnp->need_future_gp[c & 0x1]) { |
f7f7bac9 | 1739 | trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartleaf")); |
48a7639c | 1740 | goto out; |
0446be48 PM |
1741 | } |
1742 | ||
1743 | /* | |
1744 | * If either this rcu_node structure or the root rcu_node structure | |
1745 | * believe that a grace period is in progress, then we must wait | |
1746 | * for the one following, which is in "c". Because our request | |
1747 | * will be noticed at the end of the current grace period, we don't | |
48bd8e9b PK |
1748 | * need to explicitly start one. We only do the lockless check |
1749 | * of rnp_root's fields if the current rcu_node structure thinks | |
1750 | * there is no grace period in flight, and because we hold rnp->lock, | |
1751 | * the only possible change is when rnp_root's two fields are | |
1752 | * equal, in which case rnp_root->gpnum might be concurrently | |
1753 | * incremented. But that is OK, as it will just result in our | |
1754 | * doing some extra useless work. | |
0446be48 PM |
1755 | */ |
1756 | if (rnp->gpnum != rnp->completed || | |
7d0ae808 | 1757 | READ_ONCE(rnp_root->gpnum) != READ_ONCE(rnp_root->completed)) { |
0446be48 | 1758 | rnp->need_future_gp[c & 0x1]++; |
f7f7bac9 | 1759 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleaf")); |
48a7639c | 1760 | goto out; |
0446be48 PM |
1761 | } |
1762 | ||
1763 | /* | |
1764 | * There might be no grace period in progress. If we don't already | |
1765 | * hold it, acquire the root rcu_node structure's lock in order to | |
1766 | * start one (if needed). | |
1767 | */ | |
2a67e741 PZ |
1768 | if (rnp != rnp_root) |
1769 | raw_spin_lock_rcu_node(rnp_root); | |
0446be48 PM |
1770 | |
1771 | /* | |
1772 | * Get a new grace-period number. If there really is no grace | |
1773 | * period in progress, it will be smaller than the one we obtained | |
15fecf89 | 1774 | * earlier. Adjust callbacks as needed. |
0446be48 PM |
1775 | */ |
1776 | c = rcu_cbs_completed(rdp->rsp, rnp_root); | |
15fecf89 PM |
1777 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
1778 | (void)rcu_segcblist_accelerate(&rdp->cblist, c); | |
0446be48 PM |
1779 | |
1780 | /* | |
1781 | * If the needed for the required grace period is already | |
1782 | * recorded, trace and leave. | |
1783 | */ | |
1784 | if (rnp_root->need_future_gp[c & 0x1]) { | |
f7f7bac9 | 1785 | trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartedroot")); |
0446be48 PM |
1786 | goto unlock_out; |
1787 | } | |
1788 | ||
1789 | /* Record the need for the future grace period. */ | |
1790 | rnp_root->need_future_gp[c & 0x1]++; | |
1791 | ||
1792 | /* If a grace period is not already in progress, start one. */ | |
1793 | if (rnp_root->gpnum != rnp_root->completed) { | |
f7f7bac9 | 1794 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleafroot")); |
0446be48 | 1795 | } else { |
f7f7bac9 | 1796 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startedroot")); |
48a7639c | 1797 | ret = rcu_start_gp_advanced(rdp->rsp, rnp_root, rdp); |
0446be48 PM |
1798 | } |
1799 | unlock_out: | |
1800 | if (rnp != rnp_root) | |
67c583a7 | 1801 | raw_spin_unlock_rcu_node(rnp_root); |
48a7639c PM |
1802 | out: |
1803 | if (c_out != NULL) | |
1804 | *c_out = c; | |
1805 | return ret; | |
0446be48 PM |
1806 | } |
1807 | ||
1808 | /* | |
1809 | * Clean up any old requests for the just-ended grace period. Also return | |
d1e4f01d | 1810 | * whether any additional grace periods have been requested. |
0446be48 PM |
1811 | */ |
1812 | static int rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp) | |
1813 | { | |
1814 | int c = rnp->completed; | |
1815 | int needmore; | |
1816 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
1817 | ||
0446be48 PM |
1818 | rnp->need_future_gp[c & 0x1] = 0; |
1819 | needmore = rnp->need_future_gp[(c + 1) & 0x1]; | |
f7f7bac9 SRRH |
1820 | trace_rcu_future_gp(rnp, rdp, c, |
1821 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); | |
0446be48 PM |
1822 | return needmore; |
1823 | } | |
1824 | ||
48a7639c PM |
1825 | /* |
1826 | * Awaken the grace-period kthread for the specified flavor of RCU. | |
1827 | * Don't do a self-awaken, and don't bother awakening when there is | |
1828 | * nothing for the grace-period kthread to do (as in several CPUs | |
1829 | * raced to awaken, and we lost), and finally don't try to awaken | |
1830 | * a kthread that has not yet been created. | |
1831 | */ | |
1832 | static void rcu_gp_kthread_wake(struct rcu_state *rsp) | |
1833 | { | |
1834 | if (current == rsp->gp_kthread || | |
7d0ae808 | 1835 | !READ_ONCE(rsp->gp_flags) || |
48a7639c PM |
1836 | !rsp->gp_kthread) |
1837 | return; | |
abedf8e2 | 1838 | swake_up(&rsp->gp_wq); |
48a7639c PM |
1839 | } |
1840 | ||
dc35c893 PM |
1841 | /* |
1842 | * If there is room, assign a ->completed number to any callbacks on | |
1843 | * this CPU that have not already been assigned. Also accelerate any | |
1844 | * callbacks that were previously assigned a ->completed number that has | |
1845 | * since proven to be too conservative, which can happen if callbacks get | |
1846 | * assigned a ->completed number while RCU is idle, but with reference to | |
1847 | * a non-root rcu_node structure. This function is idempotent, so it does | |
48a7639c PM |
1848 | * not hurt to call it repeatedly. Returns an flag saying that we should |
1849 | * awaken the RCU grace-period kthread. | |
dc35c893 PM |
1850 | * |
1851 | * The caller must hold rnp->lock with interrupts disabled. | |
1852 | */ | |
48a7639c | 1853 | static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1854 | struct rcu_data *rdp) |
1855 | { | |
15fecf89 | 1856 | bool ret = false; |
dc35c893 | 1857 | |
c0b334c5 PM |
1858 | lockdep_assert_held(&rnp->lock); |
1859 | ||
15fecf89 PM |
1860 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1861 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1862 | return false; |
dc35c893 PM |
1863 | |
1864 | /* | |
15fecf89 PM |
1865 | * Callbacks are often registered with incomplete grace-period |
1866 | * information. Something about the fact that getting exact | |
1867 | * information requires acquiring a global lock... RCU therefore | |
1868 | * makes a conservative estimate of the grace period number at which | |
1869 | * a given callback will become ready to invoke. The following | |
1870 | * code checks this estimate and improves it when possible, thus | |
1871 | * accelerating callback invocation to an earlier grace-period | |
1872 | * number. | |
dc35c893 | 1873 | */ |
15fecf89 PM |
1874 | if (rcu_segcblist_accelerate(&rdp->cblist, rcu_cbs_completed(rsp, rnp))) |
1875 | ret = rcu_start_future_gp(rnp, rdp, NULL); | |
6d4b418c PM |
1876 | |
1877 | /* Trace depending on how much we were able to accelerate. */ | |
15fecf89 | 1878 | if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL)) |
f7f7bac9 | 1879 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccWaitCB")); |
6d4b418c | 1880 | else |
f7f7bac9 | 1881 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccReadyCB")); |
48a7639c | 1882 | return ret; |
dc35c893 PM |
1883 | } |
1884 | ||
1885 | /* | |
1886 | * Move any callbacks whose grace period has completed to the | |
1887 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
1888 | * assign ->completed numbers to any callbacks in the RCU_NEXT_TAIL | |
1889 | * sublist. This function is idempotent, so it does not hurt to | |
1890 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1891 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1892 | * |
1893 | * The caller must hold rnp->lock with interrupts disabled. | |
1894 | */ | |
48a7639c | 1895 | static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1896 | struct rcu_data *rdp) |
1897 | { | |
c0b334c5 PM |
1898 | lockdep_assert_held(&rnp->lock); |
1899 | ||
15fecf89 PM |
1900 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1901 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1902 | return false; |
dc35c893 PM |
1903 | |
1904 | /* | |
1905 | * Find all callbacks whose ->completed numbers indicate that they | |
1906 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. | |
1907 | */ | |
15fecf89 | 1908 | rcu_segcblist_advance(&rdp->cblist, rnp->completed); |
dc35c893 PM |
1909 | |
1910 | /* Classify any remaining callbacks. */ | |
48a7639c | 1911 | return rcu_accelerate_cbs(rsp, rnp, rdp); |
dc35c893 PM |
1912 | } |
1913 | ||
d09b62df | 1914 | /* |
ba9fbe95 PM |
1915 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1916 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1917 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1918 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1919 | */ |
48a7639c PM |
1920 | static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, |
1921 | struct rcu_data *rdp) | |
d09b62df | 1922 | { |
48a7639c | 1923 | bool ret; |
3563a438 | 1924 | bool need_gp; |
48a7639c | 1925 | |
c0b334c5 PM |
1926 | lockdep_assert_held(&rnp->lock); |
1927 | ||
ba9fbe95 | 1928 | /* Handle the ends of any preceding grace periods first. */ |
e3663b10 | 1929 | if (rdp->completed == rnp->completed && |
7d0ae808 | 1930 | !unlikely(READ_ONCE(rdp->gpwrap))) { |
d09b62df | 1931 | |
ba9fbe95 | 1932 | /* No grace period end, so just accelerate recent callbacks. */ |
48a7639c | 1933 | ret = rcu_accelerate_cbs(rsp, rnp, rdp); |
d09b62df | 1934 | |
dc35c893 PM |
1935 | } else { |
1936 | ||
1937 | /* Advance callbacks. */ | |
48a7639c | 1938 | ret = rcu_advance_cbs(rsp, rnp, rdp); |
d09b62df PM |
1939 | |
1940 | /* Remember that we saw this grace-period completion. */ | |
1941 | rdp->completed = rnp->completed; | |
f7f7bac9 | 1942 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuend")); |
d09b62df | 1943 | } |
398ebe60 | 1944 | |
7d0ae808 | 1945 | if (rdp->gpnum != rnp->gpnum || unlikely(READ_ONCE(rdp->gpwrap))) { |
6eaef633 PM |
1946 | /* |
1947 | * If the current grace period is waiting for this CPU, | |
1948 | * set up to detect a quiescent state, otherwise don't | |
1949 | * go looking for one. | |
1950 | */ | |
1951 | rdp->gpnum = rnp->gpnum; | |
f7f7bac9 | 1952 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart")); |
3563a438 PM |
1953 | need_gp = !!(rnp->qsmask & rdp->grpmask); |
1954 | rdp->cpu_no_qs.b.norm = need_gp; | |
9577df9a | 1955 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
3563a438 | 1956 | rdp->core_needs_qs = need_gp; |
6eaef633 | 1957 | zero_cpu_stall_ticks(rdp); |
7d0ae808 | 1958 | WRITE_ONCE(rdp->gpwrap, false); |
9b9500da | 1959 | rcu_gpnum_ovf(rnp, rdp); |
6eaef633 | 1960 | } |
48a7639c | 1961 | return ret; |
6eaef633 PM |
1962 | } |
1963 | ||
d34ea322 | 1964 | static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp) |
6eaef633 PM |
1965 | { |
1966 | unsigned long flags; | |
48a7639c | 1967 | bool needwake; |
6eaef633 PM |
1968 | struct rcu_node *rnp; |
1969 | ||
1970 | local_irq_save(flags); | |
1971 | rnp = rdp->mynode; | |
7d0ae808 PM |
1972 | if ((rdp->gpnum == READ_ONCE(rnp->gpnum) && |
1973 | rdp->completed == READ_ONCE(rnp->completed) && | |
1974 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ | |
2a67e741 | 1975 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
1976 | local_irq_restore(flags); |
1977 | return; | |
1978 | } | |
48a7639c | 1979 | needwake = __note_gp_changes(rsp, rnp, rdp); |
67c583a7 | 1980 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c PM |
1981 | if (needwake) |
1982 | rcu_gp_kthread_wake(rsp); | |
6eaef633 PM |
1983 | } |
1984 | ||
0f41c0dd PM |
1985 | static void rcu_gp_slow(struct rcu_state *rsp, int delay) |
1986 | { | |
1987 | if (delay > 0 && | |
1988 | !(rsp->gpnum % (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) | |
1989 | schedule_timeout_uninterruptible(delay); | |
1990 | } | |
1991 | ||
b3dbec76 | 1992 | /* |
45fed3e7 | 1993 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 1994 | */ |
45fed3e7 | 1995 | static bool rcu_gp_init(struct rcu_state *rsp) |
b3dbec76 | 1996 | { |
0aa04b05 | 1997 | unsigned long oldmask; |
b3dbec76 | 1998 | struct rcu_data *rdp; |
7fdefc10 | 1999 | struct rcu_node *rnp = rcu_get_root(rsp); |
b3dbec76 | 2000 | |
7d0ae808 | 2001 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 2002 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 | 2003 | if (!READ_ONCE(rsp->gp_flags)) { |
f7be8209 | 2004 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 2005 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 2006 | return false; |
f7be8209 | 2007 | } |
7d0ae808 | 2008 | WRITE_ONCE(rsp->gp_flags, 0); /* Clear all flags: New grace period. */ |
b3dbec76 | 2009 | |
f7be8209 PM |
2010 | if (WARN_ON_ONCE(rcu_gp_in_progress(rsp))) { |
2011 | /* | |
2012 | * Grace period already in progress, don't start another. | |
2013 | * Not supposed to be able to happen. | |
2014 | */ | |
67c583a7 | 2015 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 2016 | return false; |
7fdefc10 PM |
2017 | } |
2018 | ||
7fdefc10 | 2019 | /* Advance to a new grace period and initialize state. */ |
26cdfedf | 2020 | record_gp_stall_check_time(rsp); |
765a3f4f PM |
2021 | /* Record GP times before starting GP, hence smp_store_release(). */ |
2022 | smp_store_release(&rsp->gpnum, rsp->gpnum + 1); | |
f7f7bac9 | 2023 | trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start")); |
67c583a7 | 2024 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 2025 | |
0aa04b05 PM |
2026 | /* |
2027 | * Apply per-leaf buffered online and offline operations to the | |
2028 | * rcu_node tree. Note that this new grace period need not wait | |
2029 | * for subsequent online CPUs, and that quiescent-state forcing | |
2030 | * will handle subsequent offline CPUs. | |
2031 | */ | |
2032 | rcu_for_each_leaf_node(rsp, rnp) { | |
0f41c0dd | 2033 | rcu_gp_slow(rsp, gp_preinit_delay); |
2a67e741 | 2034 | raw_spin_lock_irq_rcu_node(rnp); |
0aa04b05 PM |
2035 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
2036 | !rnp->wait_blkd_tasks) { | |
2037 | /* Nothing to do on this leaf rcu_node structure. */ | |
67c583a7 | 2038 | raw_spin_unlock_irq_rcu_node(rnp); |
0aa04b05 PM |
2039 | continue; |
2040 | } | |
2041 | ||
2042 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
2043 | oldmask = rnp->qsmaskinit; | |
2044 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
2045 | ||
2046 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
2047 | if (!oldmask != !rnp->qsmaskinit) { | |
2048 | if (!oldmask) /* First online CPU for this rcu_node. */ | |
2049 | rcu_init_new_rnp(rnp); | |
2050 | else if (rcu_preempt_has_tasks(rnp)) /* blocked tasks */ | |
2051 | rnp->wait_blkd_tasks = true; | |
2052 | else /* Last offline CPU and can propagate. */ | |
2053 | rcu_cleanup_dead_rnp(rnp); | |
2054 | } | |
2055 | ||
2056 | /* | |
2057 | * If all waited-on tasks from prior grace period are | |
2058 | * done, and if all this rcu_node structure's CPUs are | |
2059 | * still offline, propagate up the rcu_node tree and | |
2060 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
2061 | * rcu_node structure's CPUs has since come back online, | |
2062 | * simply clear ->wait_blkd_tasks (but rcu_cleanup_dead_rnp() | |
2063 | * checks for this, so just call it unconditionally). | |
2064 | */ | |
2065 | if (rnp->wait_blkd_tasks && | |
2066 | (!rcu_preempt_has_tasks(rnp) || | |
2067 | rnp->qsmaskinit)) { | |
2068 | rnp->wait_blkd_tasks = false; | |
2069 | rcu_cleanup_dead_rnp(rnp); | |
2070 | } | |
2071 | ||
67c583a7 | 2072 | raw_spin_unlock_irq_rcu_node(rnp); |
0aa04b05 | 2073 | } |
7fdefc10 PM |
2074 | |
2075 | /* | |
2076 | * Set the quiescent-state-needed bits in all the rcu_node | |
2077 | * structures for all currently online CPUs in breadth-first order, | |
2078 | * starting from the root rcu_node structure, relying on the layout | |
2079 | * of the tree within the rsp->node[] array. Note that other CPUs | |
2080 | * will access only the leaves of the hierarchy, thus seeing that no | |
2081 | * grace period is in progress, at least until the corresponding | |
590d1757 | 2082 | * leaf node has been initialized. |
7fdefc10 PM |
2083 | * |
2084 | * The grace period cannot complete until the initialization | |
2085 | * process finishes, because this kthread handles both. | |
2086 | */ | |
2087 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
0f41c0dd | 2088 | rcu_gp_slow(rsp, gp_init_delay); |
2a67e741 | 2089 | raw_spin_lock_irq_rcu_node(rnp); |
b3dbec76 | 2090 | rdp = this_cpu_ptr(rsp->rda); |
7fdefc10 PM |
2091 | rcu_preempt_check_blocked_tasks(rnp); |
2092 | rnp->qsmask = rnp->qsmaskinit; | |
7d0ae808 | 2093 | WRITE_ONCE(rnp->gpnum, rsp->gpnum); |
3f47da0f | 2094 | if (WARN_ON_ONCE(rnp->completed != rsp->completed)) |
7d0ae808 | 2095 | WRITE_ONCE(rnp->completed, rsp->completed); |
7fdefc10 | 2096 | if (rnp == rdp->mynode) |
48a7639c | 2097 | (void)__note_gp_changes(rsp, rnp, rdp); |
7fdefc10 PM |
2098 | rcu_preempt_boost_start_gp(rnp); |
2099 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, | |
2100 | rnp->level, rnp->grplo, | |
2101 | rnp->grphi, rnp->qsmask); | |
67c583a7 | 2102 | raw_spin_unlock_irq_rcu_node(rnp); |
bde6c3aa | 2103 | cond_resched_rcu_qs(); |
7d0ae808 | 2104 | WRITE_ONCE(rsp->gp_activity, jiffies); |
7fdefc10 | 2105 | } |
b3dbec76 | 2106 | |
45fed3e7 | 2107 | return true; |
7fdefc10 | 2108 | } |
b3dbec76 | 2109 | |
b9a425cf | 2110 | /* |
d5374226 LR |
2111 | * Helper function for swait_event_idle() wakeup at force-quiescent-state |
2112 | * time. | |
b9a425cf PM |
2113 | */ |
2114 | static bool rcu_gp_fqs_check_wake(struct rcu_state *rsp, int *gfp) | |
2115 | { | |
2116 | struct rcu_node *rnp = rcu_get_root(rsp); | |
2117 | ||
2118 | /* Someone like call_rcu() requested a force-quiescent-state scan. */ | |
2119 | *gfp = READ_ONCE(rsp->gp_flags); | |
2120 | if (*gfp & RCU_GP_FLAG_FQS) | |
2121 | return true; | |
2122 | ||
2123 | /* The current grace period has completed. */ | |
2124 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) | |
2125 | return true; | |
2126 | ||
2127 | return false; | |
2128 | } | |
2129 | ||
4cdfc175 PM |
2130 | /* |
2131 | * Do one round of quiescent-state forcing. | |
2132 | */ | |
77f81fe0 | 2133 | static void rcu_gp_fqs(struct rcu_state *rsp, bool first_time) |
4cdfc175 | 2134 | { |
4cdfc175 PM |
2135 | struct rcu_node *rnp = rcu_get_root(rsp); |
2136 | ||
7d0ae808 | 2137 | WRITE_ONCE(rsp->gp_activity, jiffies); |
4cdfc175 | 2138 | rsp->n_force_qs++; |
77f81fe0 | 2139 | if (first_time) { |
4cdfc175 | 2140 | /* Collect dyntick-idle snapshots. */ |
fe5ac724 | 2141 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
4cdfc175 PM |
2142 | } else { |
2143 | /* Handle dyntick-idle and offline CPUs. */ | |
fe5ac724 | 2144 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
4cdfc175 PM |
2145 | } |
2146 | /* Clear flag to prevent immediate re-entry. */ | |
7d0ae808 | 2147 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 2148 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 PM |
2149 | WRITE_ONCE(rsp->gp_flags, |
2150 | READ_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 2151 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 2152 | } |
4cdfc175 PM |
2153 | } |
2154 | ||
7fdefc10 PM |
2155 | /* |
2156 | * Clean up after the old grace period. | |
2157 | */ | |
4cdfc175 | 2158 | static void rcu_gp_cleanup(struct rcu_state *rsp) |
7fdefc10 PM |
2159 | { |
2160 | unsigned long gp_duration; | |
48a7639c | 2161 | bool needgp = false; |
dae6e64d | 2162 | int nocb = 0; |
7fdefc10 PM |
2163 | struct rcu_data *rdp; |
2164 | struct rcu_node *rnp = rcu_get_root(rsp); | |
abedf8e2 | 2165 | struct swait_queue_head *sq; |
b3dbec76 | 2166 | |
7d0ae808 | 2167 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 2168 | raw_spin_lock_irq_rcu_node(rnp); |
7fdefc10 PM |
2169 | gp_duration = jiffies - rsp->gp_start; |
2170 | if (gp_duration > rsp->gp_max) | |
2171 | rsp->gp_max = gp_duration; | |
b3dbec76 | 2172 | |
7fdefc10 PM |
2173 | /* |
2174 | * We know the grace period is complete, but to everyone else | |
2175 | * it appears to still be ongoing. But it is also the case | |
2176 | * that to everyone else it looks like there is nothing that | |
2177 | * they can do to advance the grace period. It is therefore | |
2178 | * safe for us to drop the lock in order to mark the grace | |
2179 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 2180 | */ |
67c583a7 | 2181 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 2182 | |
5d4b8659 PM |
2183 | /* |
2184 | * Propagate new ->completed value to rcu_node structures so | |
2185 | * that other CPUs don't have to wait until the start of the next | |
2186 | * grace period to process their callbacks. This also avoids | |
2187 | * some nasty RCU grace-period initialization races by forcing | |
2188 | * the end of the current grace period to be completely recorded in | |
2189 | * all of the rcu_node structures before the beginning of the next | |
2190 | * grace period is recorded in any of the rcu_node structures. | |
2191 | */ | |
2192 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
2a67e741 | 2193 | raw_spin_lock_irq_rcu_node(rnp); |
5c60d25f PM |
2194 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)); |
2195 | WARN_ON_ONCE(rnp->qsmask); | |
7d0ae808 | 2196 | WRITE_ONCE(rnp->completed, rsp->gpnum); |
b11cc576 PM |
2197 | rdp = this_cpu_ptr(rsp->rda); |
2198 | if (rnp == rdp->mynode) | |
48a7639c | 2199 | needgp = __note_gp_changes(rsp, rnp, rdp) || needgp; |
78e4bc34 | 2200 | /* smp_mb() provided by prior unlock-lock pair. */ |
0446be48 | 2201 | nocb += rcu_future_gp_cleanup(rsp, rnp); |
065bb78c | 2202 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 2203 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 2204 | rcu_nocb_gp_cleanup(sq); |
bde6c3aa | 2205 | cond_resched_rcu_qs(); |
7d0ae808 | 2206 | WRITE_ONCE(rsp->gp_activity, jiffies); |
0f41c0dd | 2207 | rcu_gp_slow(rsp, gp_cleanup_delay); |
7fdefc10 | 2208 | } |
5d4b8659 | 2209 | rnp = rcu_get_root(rsp); |
2a67e741 | 2210 | raw_spin_lock_irq_rcu_node(rnp); /* Order GP before ->completed update. */ |
dae6e64d | 2211 | rcu_nocb_gp_set(rnp, nocb); |
7fdefc10 | 2212 | |
765a3f4f | 2213 | /* Declare grace period done. */ |
7d0ae808 | 2214 | WRITE_ONCE(rsp->completed, rsp->gpnum); |
f7f7bac9 | 2215 | trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end")); |
77f81fe0 | 2216 | rsp->gp_state = RCU_GP_IDLE; |
5d4b8659 | 2217 | rdp = this_cpu_ptr(rsp->rda); |
48a7639c PM |
2218 | /* Advance CBs to reduce false positives below. */ |
2219 | needgp = rcu_advance_cbs(rsp, rnp, rdp) || needgp; | |
2220 | if (needgp || cpu_needs_another_gp(rsp, rdp)) { | |
7d0ae808 | 2221 | WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT); |
bb311ecc | 2222 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2223 | READ_ONCE(rsp->gpnum), |
bb311ecc PM |
2224 | TPS("newreq")); |
2225 | } | |
67c583a7 | 2226 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 PM |
2227 | } |
2228 | ||
2229 | /* | |
2230 | * Body of kthread that handles grace periods. | |
2231 | */ | |
2232 | static int __noreturn rcu_gp_kthread(void *arg) | |
2233 | { | |
77f81fe0 | 2234 | bool first_gp_fqs; |
88d6df61 | 2235 | int gf; |
d40011f6 | 2236 | unsigned long j; |
4cdfc175 | 2237 | int ret; |
7fdefc10 PM |
2238 | struct rcu_state *rsp = arg; |
2239 | struct rcu_node *rnp = rcu_get_root(rsp); | |
2240 | ||
5871968d | 2241 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
2242 | for (;;) { |
2243 | ||
2244 | /* Handle grace-period start. */ | |
2245 | for (;;) { | |
63c4db78 | 2246 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2247 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2248 | TPS("reqwait")); |
afea227f | 2249 | rsp->gp_state = RCU_GP_WAIT_GPS; |
d5374226 LR |
2250 | swait_event_idle(rsp->gp_wq, READ_ONCE(rsp->gp_flags) & |
2251 | RCU_GP_FLAG_INIT); | |
319362c9 | 2252 | rsp->gp_state = RCU_GP_DONE_GPS; |
78e4bc34 | 2253 | /* Locking provides needed memory barrier. */ |
f7be8209 | 2254 | if (rcu_gp_init(rsp)) |
7fdefc10 | 2255 | break; |
bde6c3aa | 2256 | cond_resched_rcu_qs(); |
7d0ae808 | 2257 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2258 | WARN_ON(signal_pending(current)); |
63c4db78 | 2259 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2260 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2261 | TPS("reqwaitsig")); |
7fdefc10 | 2262 | } |
cabc49c1 | 2263 | |
4cdfc175 | 2264 | /* Handle quiescent-state forcing. */ |
77f81fe0 | 2265 | first_gp_fqs = true; |
d40011f6 PM |
2266 | j = jiffies_till_first_fqs; |
2267 | if (j > HZ) { | |
2268 | j = HZ; | |
2269 | jiffies_till_first_fqs = HZ; | |
2270 | } | |
88d6df61 | 2271 | ret = 0; |
cabc49c1 | 2272 | for (;;) { |
8c7c4829 | 2273 | if (!ret) { |
88d6df61 | 2274 | rsp->jiffies_force_qs = jiffies + j; |
8c7c4829 PM |
2275 | WRITE_ONCE(rsp->jiffies_kick_kthreads, |
2276 | jiffies + 3 * j); | |
2277 | } | |
63c4db78 | 2278 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2279 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2280 | TPS("fqswait")); |
afea227f | 2281 | rsp->gp_state = RCU_GP_WAIT_FQS; |
d5374226 | 2282 | ret = swait_event_idle_timeout(rsp->gp_wq, |
b9a425cf | 2283 | rcu_gp_fqs_check_wake(rsp, &gf), j); |
32bb1c79 | 2284 | rsp->gp_state = RCU_GP_DOING_FQS; |
78e4bc34 | 2285 | /* Locking provides needed memory barriers. */ |
4cdfc175 | 2286 | /* If grace period done, leave loop. */ |
7d0ae808 | 2287 | if (!READ_ONCE(rnp->qsmask) && |
4cdfc175 | 2288 | !rcu_preempt_blocked_readers_cgp(rnp)) |
cabc49c1 | 2289 | break; |
4cdfc175 | 2290 | /* If time for quiescent-state forcing, do it. */ |
88d6df61 PM |
2291 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) || |
2292 | (gf & RCU_GP_FLAG_FQS)) { | |
63c4db78 | 2293 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2294 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2295 | TPS("fqsstart")); |
77f81fe0 PM |
2296 | rcu_gp_fqs(rsp, first_gp_fqs); |
2297 | first_gp_fqs = false; | |
63c4db78 | 2298 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2299 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2300 | TPS("fqsend")); |
bde6c3aa | 2301 | cond_resched_rcu_qs(); |
7d0ae808 | 2302 | WRITE_ONCE(rsp->gp_activity, jiffies); |
fcfd0a23 PM |
2303 | ret = 0; /* Force full wait till next FQS. */ |
2304 | j = jiffies_till_next_fqs; | |
2305 | if (j > HZ) { | |
2306 | j = HZ; | |
2307 | jiffies_till_next_fqs = HZ; | |
2308 | } else if (j < 1) { | |
2309 | j = 1; | |
2310 | jiffies_till_next_fqs = 1; | |
2311 | } | |
4cdfc175 PM |
2312 | } else { |
2313 | /* Deal with stray signal. */ | |
bde6c3aa | 2314 | cond_resched_rcu_qs(); |
7d0ae808 | 2315 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2316 | WARN_ON(signal_pending(current)); |
63c4db78 | 2317 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2318 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2319 | TPS("fqswaitsig")); |
fcfd0a23 PM |
2320 | ret = 1; /* Keep old FQS timing. */ |
2321 | j = jiffies; | |
2322 | if (time_after(jiffies, rsp->jiffies_force_qs)) | |
2323 | j = 1; | |
2324 | else | |
2325 | j = rsp->jiffies_force_qs - j; | |
d40011f6 | 2326 | } |
cabc49c1 | 2327 | } |
4cdfc175 PM |
2328 | |
2329 | /* Handle grace-period end. */ | |
319362c9 | 2330 | rsp->gp_state = RCU_GP_CLEANUP; |
4cdfc175 | 2331 | rcu_gp_cleanup(rsp); |
319362c9 | 2332 | rsp->gp_state = RCU_GP_CLEANED; |
b3dbec76 | 2333 | } |
b3dbec76 PM |
2334 | } |
2335 | ||
64db4cff PM |
2336 | /* |
2337 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
2338 | * in preparation for detecting the next grace period. The caller must hold | |
b8462084 | 2339 | * the root node's ->lock and hard irqs must be disabled. |
e5601400 PM |
2340 | * |
2341 | * Note that it is legal for a dying CPU (which is marked as offline) to | |
2342 | * invoke this function. This can happen when the dying CPU reports its | |
2343 | * quiescent state. | |
48a7639c PM |
2344 | * |
2345 | * Returns true if the grace-period kthread must be awakened. | |
64db4cff | 2346 | */ |
48a7639c | 2347 | static bool |
910ee45d PM |
2348 | rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp, |
2349 | struct rcu_data *rdp) | |
64db4cff | 2350 | { |
c0b334c5 | 2351 | lockdep_assert_held(&rnp->lock); |
b8462084 | 2352 | if (!rsp->gp_kthread || !cpu_needs_another_gp(rsp, rdp)) { |
afe24b12 | 2353 | /* |
b3dbec76 | 2354 | * Either we have not yet spawned the grace-period |
62da1921 PM |
2355 | * task, this CPU does not need another grace period, |
2356 | * or a grace period is already in progress. | |
b3dbec76 | 2357 | * Either way, don't start a new grace period. |
afe24b12 | 2358 | */ |
48a7639c | 2359 | return false; |
afe24b12 | 2360 | } |
7d0ae808 PM |
2361 | WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT); |
2362 | trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gpnum), | |
bb311ecc | 2363 | TPS("newreq")); |
62da1921 | 2364 | |
016a8d5b SR |
2365 | /* |
2366 | * We can't do wakeups while holding the rnp->lock, as that | |
1eafd31c | 2367 | * could cause possible deadlocks with the rq->lock. Defer |
48a7639c | 2368 | * the wakeup to our caller. |
016a8d5b | 2369 | */ |
48a7639c | 2370 | return true; |
64db4cff PM |
2371 | } |
2372 | ||
910ee45d PM |
2373 | /* |
2374 | * Similar to rcu_start_gp_advanced(), but also advance the calling CPU's | |
2375 | * callbacks. Note that rcu_start_gp_advanced() cannot do this because it | |
2376 | * is invoked indirectly from rcu_advance_cbs(), which would result in | |
2377 | * endless recursion -- or would do so if it wasn't for the self-deadlock | |
2378 | * that is encountered beforehand. | |
48a7639c PM |
2379 | * |
2380 | * Returns true if the grace-period kthread needs to be awakened. | |
910ee45d | 2381 | */ |
48a7639c | 2382 | static bool rcu_start_gp(struct rcu_state *rsp) |
910ee45d PM |
2383 | { |
2384 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
2385 | struct rcu_node *rnp = rcu_get_root(rsp); | |
48a7639c | 2386 | bool ret = false; |
910ee45d PM |
2387 | |
2388 | /* | |
2389 | * If there is no grace period in progress right now, any | |
2390 | * callbacks we have up to this point will be satisfied by the | |
2391 | * next grace period. Also, advancing the callbacks reduces the | |
2392 | * probability of false positives from cpu_needs_another_gp() | |
2393 | * resulting in pointless grace periods. So, advance callbacks | |
2394 | * then start the grace period! | |
2395 | */ | |
48a7639c PM |
2396 | ret = rcu_advance_cbs(rsp, rnp, rdp) || ret; |
2397 | ret = rcu_start_gp_advanced(rsp, rnp, rdp) || ret; | |
2398 | return ret; | |
910ee45d PM |
2399 | } |
2400 | ||
f41d911f | 2401 | /* |
8994515c PM |
2402 | * Report a full set of quiescent states to the specified rcu_state data |
2403 | * structure. Invoke rcu_gp_kthread_wake() to awaken the grace-period | |
2404 | * kthread if another grace period is required. Whether we wake | |
2405 | * the grace-period kthread or it awakens itself for the next round | |
2406 | * of quiescent-state forcing, that kthread will clean up after the | |
2407 | * just-completed grace period. Note that the caller must hold rnp->lock, | |
2408 | * which is released before return. | |
f41d911f | 2409 | */ |
d3f6bad3 | 2410 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 2411 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 2412 | { |
c0b334c5 | 2413 | lockdep_assert_held(&rcu_get_root(rsp)->lock); |
fc2219d4 | 2414 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
cd73ca21 | 2415 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 2416 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags); |
94d44776 | 2417 | rcu_gp_kthread_wake(rsp); |
f41d911f PM |
2418 | } |
2419 | ||
64db4cff | 2420 | /* |
d3f6bad3 PM |
2421 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
2422 | * Allows quiescent states for a group of CPUs to be reported at one go | |
2423 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
2424 | * must be represented by the same rcu_node structure (which need not be a |
2425 | * leaf rcu_node structure, though it often will be). The gps parameter | |
2426 | * is the grace-period snapshot, which means that the quiescent states | |
2427 | * are valid only if rnp->gpnum is equal to gps. That structure's lock | |
2428 | * must be held upon entry, and it is released before return. | |
64db4cff PM |
2429 | */ |
2430 | static void | |
d3f6bad3 | 2431 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
654e9533 | 2432 | struct rcu_node *rnp, unsigned long gps, unsigned long flags) |
64db4cff PM |
2433 | __releases(rnp->lock) |
2434 | { | |
654e9533 | 2435 | unsigned long oldmask = 0; |
28ecd580 PM |
2436 | struct rcu_node *rnp_c; |
2437 | ||
c0b334c5 PM |
2438 | lockdep_assert_held(&rnp->lock); |
2439 | ||
64db4cff PM |
2440 | /* Walk up the rcu_node hierarchy. */ |
2441 | for (;;) { | |
654e9533 | 2442 | if (!(rnp->qsmask & mask) || rnp->gpnum != gps) { |
64db4cff | 2443 | |
654e9533 PM |
2444 | /* |
2445 | * Our bit has already been cleared, or the | |
2446 | * relevant grace period is already over, so done. | |
2447 | */ | |
67c583a7 | 2448 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2449 | return; |
2450 | } | |
654e9533 | 2451 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
2dee9404 PM |
2452 | WARN_ON_ONCE(rnp->level != rcu_num_lvls - 1 && |
2453 | rcu_preempt_blocked_readers_cgp(rnp)); | |
64db4cff | 2454 | rnp->qsmask &= ~mask; |
d4c08f2a PM |
2455 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
2456 | mask, rnp->qsmask, rnp->level, | |
2457 | rnp->grplo, rnp->grphi, | |
2458 | !!rnp->gp_tasks); | |
27f4d280 | 2459 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
2460 | |
2461 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 2462 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2463 | return; |
2464 | } | |
2465 | mask = rnp->grpmask; | |
2466 | if (rnp->parent == NULL) { | |
2467 | ||
2468 | /* No more levels. Exit loop holding root lock. */ | |
2469 | ||
2470 | break; | |
2471 | } | |
67c583a7 | 2472 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 2473 | rnp_c = rnp; |
64db4cff | 2474 | rnp = rnp->parent; |
2a67e741 | 2475 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
654e9533 | 2476 | oldmask = rnp_c->qsmask; |
64db4cff PM |
2477 | } |
2478 | ||
2479 | /* | |
2480 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 2481 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 2482 | * to clean up and start the next grace period if one is needed. |
64db4cff | 2483 | */ |
d3f6bad3 | 2484 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
2485 | } |
2486 | ||
cc99a310 PM |
2487 | /* |
2488 | * Record a quiescent state for all tasks that were previously queued | |
2489 | * on the specified rcu_node structure and that were blocking the current | |
2490 | * RCU grace period. The caller must hold the specified rnp->lock with | |
2491 | * irqs disabled, and this lock is released upon return, but irqs remain | |
2492 | * disabled. | |
2493 | */ | |
0aa04b05 | 2494 | static void rcu_report_unblock_qs_rnp(struct rcu_state *rsp, |
cc99a310 PM |
2495 | struct rcu_node *rnp, unsigned long flags) |
2496 | __releases(rnp->lock) | |
2497 | { | |
654e9533 | 2498 | unsigned long gps; |
cc99a310 PM |
2499 | unsigned long mask; |
2500 | struct rcu_node *rnp_p; | |
2501 | ||
c0b334c5 | 2502 | lockdep_assert_held(&rnp->lock); |
a77da14c PM |
2503 | if (rcu_state_p == &rcu_sched_state || rsp != rcu_state_p || |
2504 | rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { | |
67c583a7 | 2505 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
2506 | return; /* Still need more quiescent states! */ |
2507 | } | |
2508 | ||
2509 | rnp_p = rnp->parent; | |
2510 | if (rnp_p == NULL) { | |
2511 | /* | |
a77da14c PM |
2512 | * Only one rcu_node structure in the tree, so don't |
2513 | * try to report up to its nonexistent parent! | |
cc99a310 PM |
2514 | */ |
2515 | rcu_report_qs_rsp(rsp, flags); | |
2516 | return; | |
2517 | } | |
2518 | ||
654e9533 PM |
2519 | /* Report up the rest of the hierarchy, tracking current ->gpnum. */ |
2520 | gps = rnp->gpnum; | |
cc99a310 | 2521 | mask = rnp->grpmask; |
67c583a7 | 2522 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 2523 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
654e9533 | 2524 | rcu_report_qs_rnp(mask, rsp, rnp_p, gps, flags); |
cc99a310 PM |
2525 | } |
2526 | ||
64db4cff | 2527 | /* |
d3f6bad3 | 2528 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 2529 | * structure. This must be called from the specified CPU. |
64db4cff PM |
2530 | */ |
2531 | static void | |
d7d6a11e | 2532 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2533 | { |
2534 | unsigned long flags; | |
2535 | unsigned long mask; | |
48a7639c | 2536 | bool needwake; |
64db4cff PM |
2537 | struct rcu_node *rnp; |
2538 | ||
2539 | rnp = rdp->mynode; | |
2a67e741 | 2540 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
3a19b46a PM |
2541 | if (rdp->cpu_no_qs.b.norm || rdp->gpnum != rnp->gpnum || |
2542 | rnp->completed == rnp->gpnum || rdp->gpwrap) { | |
64db4cff PM |
2543 | |
2544 | /* | |
e4cc1f22 PM |
2545 | * The grace period in which this quiescent state was |
2546 | * recorded has ended, so don't report it upwards. | |
2547 | * We will instead need a new quiescent state that lies | |
2548 | * within the current grace period. | |
64db4cff | 2549 | */ |
5b74c458 | 2550 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
9577df9a | 2551 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
67c583a7 | 2552 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2553 | return; |
2554 | } | |
2555 | mask = rdp->grpmask; | |
2556 | if ((rnp->qsmask & mask) == 0) { | |
67c583a7 | 2557 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2558 | } else { |
bb53e416 | 2559 | rdp->core_needs_qs = false; |
64db4cff PM |
2560 | |
2561 | /* | |
2562 | * This GP can't end until cpu checks in, so all of our | |
2563 | * callbacks can be processed during the next GP. | |
2564 | */ | |
48a7639c | 2565 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
64db4cff | 2566 | |
654e9533 PM |
2567 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags); |
2568 | /* ^^^ Released rnp->lock */ | |
48a7639c PM |
2569 | if (needwake) |
2570 | rcu_gp_kthread_wake(rsp); | |
64db4cff PM |
2571 | } |
2572 | } | |
2573 | ||
2574 | /* | |
2575 | * Check to see if there is a new grace period of which this CPU | |
2576 | * is not yet aware, and if so, set up local rcu_data state for it. | |
2577 | * Otherwise, see if this CPU has just passed through its first | |
2578 | * quiescent state for this grace period, and record that fact if so. | |
2579 | */ | |
2580 | static void | |
2581 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
2582 | { | |
05eb552b PM |
2583 | /* Check for grace-period ends and beginnings. */ |
2584 | note_gp_changes(rsp, rdp); | |
64db4cff PM |
2585 | |
2586 | /* | |
2587 | * Does this CPU still need to do its part for current grace period? | |
2588 | * If no, return and let the other CPUs do their part as well. | |
2589 | */ | |
97c668b8 | 2590 | if (!rdp->core_needs_qs) |
64db4cff PM |
2591 | return; |
2592 | ||
2593 | /* | |
2594 | * Was there a quiescent state since the beginning of the grace | |
2595 | * period? If no, then exit and wait for the next call. | |
2596 | */ | |
3a19b46a | 2597 | if (rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2598 | return; |
2599 | ||
d3f6bad3 PM |
2600 | /* |
2601 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
2602 | * judge of that). | |
2603 | */ | |
d7d6a11e | 2604 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp); |
64db4cff PM |
2605 | } |
2606 | ||
b1420f1c PM |
2607 | /* |
2608 | * Trace the fact that this CPU is going offline. | |
2609 | */ | |
2610 | static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) | |
2611 | { | |
88a4976d PM |
2612 | RCU_TRACE(unsigned long mask;) |
2613 | RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda);) | |
2614 | RCU_TRACE(struct rcu_node *rnp = rdp->mynode;) | |
b1420f1c | 2615 | |
ea46351c PM |
2616 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2617 | return; | |
2618 | ||
88a4976d | 2619 | RCU_TRACE(mask = rdp->grpmask;) |
e5601400 PM |
2620 | trace_rcu_grace_period(rsp->name, |
2621 | rnp->gpnum + 1 - !!(rnp->qsmask & mask), | |
f7f7bac9 | 2622 | TPS("cpuofl")); |
64db4cff PM |
2623 | } |
2624 | ||
8af3a5e7 PM |
2625 | /* |
2626 | * All CPUs for the specified rcu_node structure have gone offline, | |
2627 | * and all tasks that were preempted within an RCU read-side critical | |
2628 | * section while running on one of those CPUs have since exited their RCU | |
2629 | * read-side critical section. Some other CPU is reporting this fact with | |
2630 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2631 | * This function therefore goes up the tree of rcu_node structures, | |
2632 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2633 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
2634 | * updated | |
2635 | * | |
2636 | * This function does check that the specified rcu_node structure has | |
2637 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2638 | * prematurely. That said, invoking it after the fact will cost you | |
2639 | * a needless lock acquisition. So once it has done its work, don't | |
2640 | * invoke it again. | |
2641 | */ | |
2642 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2643 | { | |
2644 | long mask; | |
2645 | struct rcu_node *rnp = rnp_leaf; | |
2646 | ||
c0b334c5 | 2647 | lockdep_assert_held(&rnp->lock); |
ea46351c PM |
2648 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
2649 | rnp->qsmaskinit || rcu_preempt_has_tasks(rnp)) | |
8af3a5e7 PM |
2650 | return; |
2651 | for (;;) { | |
2652 | mask = rnp->grpmask; | |
2653 | rnp = rnp->parent; | |
2654 | if (!rnp) | |
2655 | break; | |
2a67e741 | 2656 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2657 | rnp->qsmaskinit &= ~mask; |
0aa04b05 | 2658 | rnp->qsmask &= ~mask; |
8af3a5e7 | 2659 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2660 | raw_spin_unlock_rcu_node(rnp); |
2661 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2662 | return; |
2663 | } | |
67c583a7 | 2664 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2665 | } |
2666 | } | |
2667 | ||
64db4cff | 2668 | /* |
e5601400 | 2669 | * The CPU has been completely removed, and some other CPU is reporting |
a58163d8 PM |
2670 | * this fact from process context. Do the remainder of the cleanup. |
2671 | * There can only be one CPU hotplug operation at a time, so no need for | |
2672 | * explicit locking. | |
64db4cff | 2673 | */ |
e5601400 | 2674 | static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) |
64db4cff | 2675 | { |
e5601400 | 2676 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
b1420f1c | 2677 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2678 | |
ea46351c PM |
2679 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2680 | return; | |
2681 | ||
2036d94a | 2682 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2683 | rcu_boost_kthread_setaffinity(rnp, -1); |
64db4cff PM |
2684 | } |
2685 | ||
64db4cff PM |
2686 | /* |
2687 | * Invoke any RCU callbacks that have made it to the end of their grace | |
2688 | * period. Thottle as specified by rdp->blimit. | |
2689 | */ | |
37c72e56 | 2690 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2691 | { |
2692 | unsigned long flags; | |
15fecf89 PM |
2693 | struct rcu_head *rhp; |
2694 | struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl); | |
2695 | long bl, count; | |
64db4cff | 2696 | |
dc35c893 | 2697 | /* If no callbacks are ready, just return. */ |
15fecf89 PM |
2698 | if (!rcu_segcblist_ready_cbs(&rdp->cblist)) { |
2699 | trace_rcu_batch_start(rsp->name, | |
2700 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
2701 | rcu_segcblist_n_cbs(&rdp->cblist), 0); | |
2702 | trace_rcu_batch_end(rsp->name, 0, | |
2703 | !rcu_segcblist_empty(&rdp->cblist), | |
4968c300 PM |
2704 | need_resched(), is_idle_task(current), |
2705 | rcu_is_callbacks_kthread()); | |
64db4cff | 2706 | return; |
29c00b4a | 2707 | } |
64db4cff PM |
2708 | |
2709 | /* | |
2710 | * Extract the list of ready callbacks, disabling to prevent | |
15fecf89 PM |
2711 | * races with call_rcu() from interrupt handlers. Leave the |
2712 | * callback counts, as rcu_barrier() needs to be conservative. | |
64db4cff PM |
2713 | */ |
2714 | local_irq_save(flags); | |
8146c4e2 | 2715 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 2716 | bl = rdp->blimit; |
15fecf89 PM |
2717 | trace_rcu_batch_start(rsp->name, rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2718 | rcu_segcblist_n_cbs(&rdp->cblist), bl); | |
2719 | rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl); | |
64db4cff PM |
2720 | local_irq_restore(flags); |
2721 | ||
2722 | /* Invoke callbacks. */ | |
15fecf89 PM |
2723 | rhp = rcu_cblist_dequeue(&rcl); |
2724 | for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) { | |
2725 | debug_rcu_head_unqueue(rhp); | |
2726 | if (__rcu_reclaim(rsp->name, rhp)) | |
2727 | rcu_cblist_dequeued_lazy(&rcl); | |
2728 | /* | |
2729 | * Stop only if limit reached and CPU has something to do. | |
2730 | * Note: The rcl structure counts down from zero. | |
2731 | */ | |
4b27f20b | 2732 | if (-rcl.len >= bl && |
dff1672d PM |
2733 | (need_resched() || |
2734 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
2735 | break; |
2736 | } | |
2737 | ||
2738 | local_irq_save(flags); | |
4b27f20b | 2739 | count = -rcl.len; |
8ef0f37e PM |
2740 | trace_rcu_batch_end(rsp->name, count, !!rcl.head, need_resched(), |
2741 | is_idle_task(current), rcu_is_callbacks_kthread()); | |
64db4cff | 2742 | |
15fecf89 PM |
2743 | /* Update counts and requeue any remaining callbacks. */ |
2744 | rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl); | |
b1420f1c | 2745 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
b1420f1c | 2746 | rdp->n_cbs_invoked += count; |
15fecf89 | 2747 | rcu_segcblist_insert_count(&rdp->cblist, &rcl); |
64db4cff PM |
2748 | |
2749 | /* Reinstate batch limit if we have worked down the excess. */ | |
15fecf89 PM |
2750 | count = rcu_segcblist_n_cbs(&rdp->cblist); |
2751 | if (rdp->blimit == LONG_MAX && count <= qlowmark) | |
64db4cff PM |
2752 | rdp->blimit = blimit; |
2753 | ||
37c72e56 | 2754 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
15fecf89 | 2755 | if (count == 0 && rdp->qlen_last_fqs_check != 0) { |
37c72e56 PM |
2756 | rdp->qlen_last_fqs_check = 0; |
2757 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
15fecf89 PM |
2758 | } else if (count < rdp->qlen_last_fqs_check - qhimark) |
2759 | rdp->qlen_last_fqs_check = count; | |
2760 | WARN_ON_ONCE(rcu_segcblist_empty(&rdp->cblist) != (count == 0)); | |
37c72e56 | 2761 | |
64db4cff PM |
2762 | local_irq_restore(flags); |
2763 | ||
e0f23060 | 2764 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
15fecf89 | 2765 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2766 | invoke_rcu_core(); |
64db4cff PM |
2767 | } |
2768 | ||
2769 | /* | |
2770 | * Check to see if this CPU is in a non-context-switch quiescent state | |
2771 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 2772 | * Also schedule RCU core processing. |
64db4cff | 2773 | * |
9b2e4f18 | 2774 | * This function must be called from hardirq context. It is normally |
5403d367 | 2775 | * invoked from the scheduling-clock interrupt. |
64db4cff | 2776 | */ |
c3377c2d | 2777 | void rcu_check_callbacks(int user) |
64db4cff | 2778 | { |
f7f7bac9 | 2779 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
a858af28 | 2780 | increment_cpu_stall_ticks(); |
9b2e4f18 | 2781 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
64db4cff PM |
2782 | |
2783 | /* | |
2784 | * Get here if this CPU took its interrupt from user | |
2785 | * mode or from the idle loop, and if this is not a | |
2786 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 2787 | * a quiescent state, so note it. |
64db4cff PM |
2788 | * |
2789 | * No memory barrier is required here because both | |
d6714c22 PM |
2790 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
2791 | * variables that other CPUs neither access nor modify, | |
2792 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
2793 | */ |
2794 | ||
284a8c93 PM |
2795 | rcu_sched_qs(); |
2796 | rcu_bh_qs(); | |
64db4cff PM |
2797 | |
2798 | } else if (!in_softirq()) { | |
2799 | ||
2800 | /* | |
2801 | * Get here if this CPU did not take its interrupt from | |
2802 | * softirq, in other words, if it is not interrupting | |
2803 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 2804 | * critical section, so note it. |
64db4cff PM |
2805 | */ |
2806 | ||
284a8c93 | 2807 | rcu_bh_qs(); |
64db4cff | 2808 | } |
86aea0e6 | 2809 | rcu_preempt_check_callbacks(); |
e3950ecd | 2810 | if (rcu_pending()) |
a46e0899 | 2811 | invoke_rcu_core(); |
8315f422 PM |
2812 | if (user) |
2813 | rcu_note_voluntary_context_switch(current); | |
f7f7bac9 | 2814 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2815 | } |
2816 | ||
64db4cff PM |
2817 | /* |
2818 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
2819 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
2820 | * Also initiate boosting for any threads blocked on the root rcu_node. |
2821 | * | |
ee47eb9f | 2822 | * The caller must have suppressed start of new grace periods. |
64db4cff | 2823 | */ |
fe5ac724 | 2824 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *rsp)) |
64db4cff | 2825 | { |
64db4cff PM |
2826 | int cpu; |
2827 | unsigned long flags; | |
2828 | unsigned long mask; | |
a0b6c9a7 | 2829 | struct rcu_node *rnp; |
64db4cff | 2830 | |
a0b6c9a7 | 2831 | rcu_for_each_leaf_node(rsp, rnp) { |
bde6c3aa | 2832 | cond_resched_rcu_qs(); |
64db4cff | 2833 | mask = 0; |
2a67e741 | 2834 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
a0b6c9a7 | 2835 | if (rnp->qsmask == 0) { |
a77da14c PM |
2836 | if (rcu_state_p == &rcu_sched_state || |
2837 | rsp != rcu_state_p || | |
2838 | rcu_preempt_blocked_readers_cgp(rnp)) { | |
2839 | /* | |
2840 | * No point in scanning bits because they | |
2841 | * are all zero. But we might need to | |
2842 | * priority-boost blocked readers. | |
2843 | */ | |
2844 | rcu_initiate_boost(rnp, flags); | |
2845 | /* rcu_initiate_boost() releases rnp->lock */ | |
2846 | continue; | |
2847 | } | |
2848 | if (rnp->parent && | |
2849 | (rnp->parent->qsmask & rnp->grpmask)) { | |
2850 | /* | |
2851 | * Race between grace-period | |
2852 | * initialization and task exiting RCU | |
2853 | * read-side critical section: Report. | |
2854 | */ | |
2855 | rcu_report_unblock_qs_rnp(rsp, rnp, flags); | |
2856 | /* rcu_report_unblock_qs_rnp() rlses ->lock */ | |
2857 | continue; | |
2858 | } | |
64db4cff | 2859 | } |
bc75e999 MR |
2860 | for_each_leaf_node_possible_cpu(rnp, cpu) { |
2861 | unsigned long bit = leaf_node_cpu_bit(rnp, cpu); | |
0edd1b17 | 2862 | if ((rnp->qsmask & bit) != 0) { |
fe5ac724 | 2863 | if (f(per_cpu_ptr(rsp->rda, cpu))) |
0edd1b17 PM |
2864 | mask |= bit; |
2865 | } | |
64db4cff | 2866 | } |
45f014c5 | 2867 | if (mask != 0) { |
654e9533 PM |
2868 | /* Idle/offline CPUs, report (releases rnp->lock. */ |
2869 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags); | |
0aa04b05 PM |
2870 | } else { |
2871 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 2872 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2873 | } |
64db4cff | 2874 | } |
64db4cff PM |
2875 | } |
2876 | ||
2877 | /* | |
2878 | * Force quiescent states on reluctant CPUs, and also detect which | |
2879 | * CPUs are in dyntick-idle mode. | |
2880 | */ | |
4cdfc175 | 2881 | static void force_quiescent_state(struct rcu_state *rsp) |
64db4cff PM |
2882 | { |
2883 | unsigned long flags; | |
394f2769 PM |
2884 | bool ret; |
2885 | struct rcu_node *rnp; | |
2886 | struct rcu_node *rnp_old = NULL; | |
2887 | ||
2888 | /* Funnel through hierarchy to reduce memory contention. */ | |
d860d403 | 2889 | rnp = __this_cpu_read(rsp->rda->mynode); |
394f2769 | 2890 | for (; rnp != NULL; rnp = rnp->parent) { |
7d0ae808 | 2891 | ret = (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) || |
394f2769 PM |
2892 | !raw_spin_trylock(&rnp->fqslock); |
2893 | if (rnp_old != NULL) | |
2894 | raw_spin_unlock(&rnp_old->fqslock); | |
2895 | if (ret) { | |
a792563b | 2896 | rsp->n_force_qs_lh++; |
394f2769 PM |
2897 | return; |
2898 | } | |
2899 | rnp_old = rnp; | |
2900 | } | |
2901 | /* rnp_old == rcu_get_root(rsp), rnp == NULL. */ | |
64db4cff | 2902 | |
394f2769 | 2903 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 2904 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 2905 | raw_spin_unlock(&rnp_old->fqslock); |
7d0ae808 | 2906 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
a792563b | 2907 | rsp->n_force_qs_lh++; |
67c583a7 | 2908 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 2909 | return; /* Someone beat us to it. */ |
46a1e34e | 2910 | } |
7d0ae808 | 2911 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 2912 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
94d44776 | 2913 | rcu_gp_kthread_wake(rsp); |
64db4cff PM |
2914 | } |
2915 | ||
64db4cff | 2916 | /* |
e0f23060 PM |
2917 | * This does the RCU core processing work for the specified rcu_state |
2918 | * and rcu_data structures. This may be called only from the CPU to | |
2919 | * whom the rdp belongs. | |
64db4cff PM |
2920 | */ |
2921 | static void | |
1bca8cf1 | 2922 | __rcu_process_callbacks(struct rcu_state *rsp) |
64db4cff PM |
2923 | { |
2924 | unsigned long flags; | |
48a7639c | 2925 | bool needwake; |
fa07a58f | 2926 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
64db4cff | 2927 | |
50dc7def | 2928 | WARN_ON_ONCE(!rdp->beenonline); |
2e597558 | 2929 | |
64db4cff PM |
2930 | /* Update RCU state based on any recent quiescent states. */ |
2931 | rcu_check_quiescent_state(rsp, rdp); | |
2932 | ||
2933 | /* Does this CPU require a not-yet-started grace period? */ | |
dc35c893 | 2934 | local_irq_save(flags); |
64db4cff | 2935 | if (cpu_needs_another_gp(rsp, rdp)) { |
6cf10081 | 2936 | raw_spin_lock_rcu_node(rcu_get_root(rsp)); /* irqs disabled. */ |
48a7639c | 2937 | needwake = rcu_start_gp(rsp); |
67c583a7 | 2938 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags); |
48a7639c PM |
2939 | if (needwake) |
2940 | rcu_gp_kthread_wake(rsp); | |
dc35c893 PM |
2941 | } else { |
2942 | local_irq_restore(flags); | |
64db4cff PM |
2943 | } |
2944 | ||
2945 | /* If there are callbacks ready, invoke them. */ | |
15fecf89 | 2946 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2947 | invoke_rcu_callbacks(rsp, rdp); |
96d3fd0d PM |
2948 | |
2949 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
2950 | do_nocb_deferred_wakeup(rdp); | |
09223371 SL |
2951 | } |
2952 | ||
64db4cff | 2953 | /* |
e0f23060 | 2954 | * Do RCU core processing for the current CPU. |
64db4cff | 2955 | */ |
0766f788 | 2956 | static __latent_entropy void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 2957 | { |
6ce75a23 PM |
2958 | struct rcu_state *rsp; |
2959 | ||
bfa00b4c PM |
2960 | if (cpu_is_offline(smp_processor_id())) |
2961 | return; | |
f7f7bac9 | 2962 | trace_rcu_utilization(TPS("Start RCU core")); |
6ce75a23 PM |
2963 | for_each_rcu_flavor(rsp) |
2964 | __rcu_process_callbacks(rsp); | |
f7f7bac9 | 2965 | trace_rcu_utilization(TPS("End RCU core")); |
64db4cff PM |
2966 | } |
2967 | ||
a26ac245 | 2968 | /* |
e0f23060 PM |
2969 | * Schedule RCU callback invocation. If the specified type of RCU |
2970 | * does not support RCU priority boosting, just do a direct call, | |
2971 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
924df8a0 | 2972 | * are running on the current CPU with softirqs disabled, the |
e0f23060 | 2973 | * rcu_cpu_kthread_task cannot disappear out from under us. |
a26ac245 | 2974 | */ |
a46e0899 | 2975 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 2976 | { |
7d0ae808 | 2977 | if (unlikely(!READ_ONCE(rcu_scheduler_fully_active))) |
b0d30417 | 2978 | return; |
a46e0899 PM |
2979 | if (likely(!rsp->boost)) { |
2980 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
2981 | return; |
2982 | } | |
a46e0899 | 2983 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
2984 | } |
2985 | ||
a46e0899 | 2986 | static void invoke_rcu_core(void) |
09223371 | 2987 | { |
b0f74036 PM |
2988 | if (cpu_online(smp_processor_id())) |
2989 | raise_softirq(RCU_SOFTIRQ); | |
09223371 SL |
2990 | } |
2991 | ||
29154c57 PM |
2992 | /* |
2993 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
2994 | */ | |
2995 | static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp, | |
2996 | struct rcu_head *head, unsigned long flags) | |
64db4cff | 2997 | { |
48a7639c PM |
2998 | bool needwake; |
2999 | ||
62fde6ed PM |
3000 | /* |
3001 | * If called from an extended quiescent state, invoke the RCU | |
3002 | * core in order to force a re-evaluation of RCU's idleness. | |
3003 | */ | |
9910affa | 3004 | if (!rcu_is_watching()) |
62fde6ed PM |
3005 | invoke_rcu_core(); |
3006 | ||
a16b7a69 | 3007 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 3008 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 3009 | return; |
64db4cff | 3010 | |
37c72e56 PM |
3011 | /* |
3012 | * Force the grace period if too many callbacks or too long waiting. | |
3013 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
3014 | * if some other CPU has recently done so. Also, don't bother | |
3015 | * invoking force_quiescent_state() if the newly enqueued callback | |
3016 | * is the only one waiting for a grace period to complete. | |
3017 | */ | |
15fecf89 PM |
3018 | if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) > |
3019 | rdp->qlen_last_fqs_check + qhimark)) { | |
b52573d2 PM |
3020 | |
3021 | /* Are we ignoring a completed grace period? */ | |
470716fc | 3022 | note_gp_changes(rsp, rdp); |
b52573d2 PM |
3023 | |
3024 | /* Start a new grace period if one not already started. */ | |
3025 | if (!rcu_gp_in_progress(rsp)) { | |
b52573d2 PM |
3026 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
3027 | ||
2a67e741 | 3028 | raw_spin_lock_rcu_node(rnp_root); |
48a7639c | 3029 | needwake = rcu_start_gp(rsp); |
67c583a7 | 3030 | raw_spin_unlock_rcu_node(rnp_root); |
48a7639c PM |
3031 | if (needwake) |
3032 | rcu_gp_kthread_wake(rsp); | |
b52573d2 PM |
3033 | } else { |
3034 | /* Give the grace period a kick. */ | |
3035 | rdp->blimit = LONG_MAX; | |
3036 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
15fecf89 | 3037 | rcu_segcblist_first_pend_cb(&rdp->cblist) != head) |
4cdfc175 | 3038 | force_quiescent_state(rsp); |
b52573d2 | 3039 | rdp->n_force_qs_snap = rsp->n_force_qs; |
15fecf89 | 3040 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); |
b52573d2 | 3041 | } |
4cdfc175 | 3042 | } |
29154c57 PM |
3043 | } |
3044 | ||
ae150184 PM |
3045 | /* |
3046 | * RCU callback function to leak a callback. | |
3047 | */ | |
3048 | static void rcu_leak_callback(struct rcu_head *rhp) | |
3049 | { | |
3050 | } | |
3051 | ||
3fbfbf7a PM |
3052 | /* |
3053 | * Helper function for call_rcu() and friends. The cpu argument will | |
3054 | * normally be -1, indicating "currently running CPU". It may specify | |
3055 | * a CPU only if that CPU is a no-CBs CPU. Currently, only _rcu_barrier() | |
3056 | * is expected to specify a CPU. | |
3057 | */ | |
64db4cff | 3058 | static void |
b6a4ae76 | 3059 | __call_rcu(struct rcu_head *head, rcu_callback_t func, |
3fbfbf7a | 3060 | struct rcu_state *rsp, int cpu, bool lazy) |
64db4cff PM |
3061 | { |
3062 | unsigned long flags; | |
3063 | struct rcu_data *rdp; | |
3064 | ||
b8f2ed53 PM |
3065 | /* Misaligned rcu_head! */ |
3066 | WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1)); | |
3067 | ||
ae150184 | 3068 | if (debug_rcu_head_queue(head)) { |
fa3c6647 PM |
3069 | /* |
3070 | * Probable double call_rcu(), so leak the callback. | |
3071 | * Use rcu:rcu_callback trace event to find the previous | |
3072 | * time callback was passed to __call_rcu(). | |
3073 | */ | |
3074 | WARN_ONCE(1, "__call_rcu(): Double-freed CB %p->%pF()!!!\n", | |
3075 | head, head->func); | |
7d0ae808 | 3076 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
3077 | return; |
3078 | } | |
64db4cff PM |
3079 | head->func = func; |
3080 | head->next = NULL; | |
64db4cff | 3081 | local_irq_save(flags); |
394f99a9 | 3082 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
3083 | |
3084 | /* Add the callback to our list. */ | |
15fecf89 | 3085 | if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist)) || cpu != -1) { |
3fbfbf7a PM |
3086 | int offline; |
3087 | ||
3088 | if (cpu != -1) | |
3089 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
143da9c2 PM |
3090 | if (likely(rdp->mynode)) { |
3091 | /* Post-boot, so this should be for a no-CBs CPU. */ | |
3092 | offline = !__call_rcu_nocb(rdp, head, lazy, flags); | |
3093 | WARN_ON_ONCE(offline); | |
3094 | /* Offline CPU, _call_rcu() illegal, leak callback. */ | |
3095 | local_irq_restore(flags); | |
3096 | return; | |
3097 | } | |
3098 | /* | |
3099 | * Very early boot, before rcu_init(). Initialize if needed | |
3100 | * and then drop through to queue the callback. | |
3101 | */ | |
3102 | BUG_ON(cpu != -1); | |
34404ca8 | 3103 | WARN_ON_ONCE(!rcu_is_watching()); |
15fecf89 PM |
3104 | if (rcu_segcblist_empty(&rdp->cblist)) |
3105 | rcu_segcblist_init(&rdp->cblist); | |
0d8ee37e | 3106 | } |
15fecf89 PM |
3107 | rcu_segcblist_enqueue(&rdp->cblist, head, lazy); |
3108 | if (!lazy) | |
c57afe80 | 3109 | rcu_idle_count_callbacks_posted(); |
2655d57e | 3110 | |
d4c08f2a PM |
3111 | if (__is_kfree_rcu_offset((unsigned long)func)) |
3112 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
15fecf89 PM |
3113 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
3114 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 3115 | else |
15fecf89 PM |
3116 | trace_rcu_callback(rsp->name, head, |
3117 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
3118 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 3119 | |
29154c57 PM |
3120 | /* Go handle any RCU core processing required. */ |
3121 | __call_rcu_core(rsp, rdp, head, flags); | |
64db4cff PM |
3122 | local_irq_restore(flags); |
3123 | } | |
3124 | ||
a68a2bb2 PM |
3125 | /** |
3126 | * call_rcu_sched() - Queue an RCU for invocation after sched grace period. | |
3127 | * @head: structure to be used for queueing the RCU updates. | |
3128 | * @func: actual callback function to be invoked after the grace period | |
3129 | * | |
3130 | * The callback function will be invoked some time after a full grace | |
3131 | * period elapses, in other words after all currently executing RCU | |
3132 | * read-side critical sections have completed. call_rcu_sched() assumes | |
3133 | * that the read-side critical sections end on enabling of preemption | |
3134 | * or on voluntary preemption. | |
27fdb35f PM |
3135 | * RCU read-side critical sections are delimited by: |
3136 | * | |
3137 | * - rcu_read_lock_sched() and rcu_read_unlock_sched(), OR | |
3138 | * - anything that disables preemption. | |
a68a2bb2 PM |
3139 | * |
3140 | * These may be nested. | |
3141 | * | |
3142 | * See the description of call_rcu() for more detailed information on | |
3143 | * memory ordering guarantees. | |
64db4cff | 3144 | */ |
b6a4ae76 | 3145 | void call_rcu_sched(struct rcu_head *head, rcu_callback_t func) |
64db4cff | 3146 | { |
3fbfbf7a | 3147 | __call_rcu(head, func, &rcu_sched_state, -1, 0); |
64db4cff | 3148 | } |
d6714c22 | 3149 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff | 3150 | |
a68a2bb2 PM |
3151 | /** |
3152 | * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period. | |
3153 | * @head: structure to be used for queueing the RCU updates. | |
3154 | * @func: actual callback function to be invoked after the grace period | |
3155 | * | |
3156 | * The callback function will be invoked some time after a full grace | |
3157 | * period elapses, in other words after all currently executing RCU | |
3158 | * read-side critical sections have completed. call_rcu_bh() assumes | |
3159 | * that the read-side critical sections end on completion of a softirq | |
3160 | * handler. This means that read-side critical sections in process | |
3161 | * context must not be interrupted by softirqs. This interface is to be | |
3162 | * used when most of the read-side critical sections are in softirq context. | |
27fdb35f PM |
3163 | * RCU read-side critical sections are delimited by: |
3164 | * | |
3165 | * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context, OR | |
3166 | * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. | |
3167 | * | |
3168 | * These may be nested. | |
a68a2bb2 PM |
3169 | * |
3170 | * See the description of call_rcu() for more detailed information on | |
3171 | * memory ordering guarantees. | |
64db4cff | 3172 | */ |
b6a4ae76 | 3173 | void call_rcu_bh(struct rcu_head *head, rcu_callback_t func) |
64db4cff | 3174 | { |
3fbfbf7a | 3175 | __call_rcu(head, func, &rcu_bh_state, -1, 0); |
64db4cff PM |
3176 | } |
3177 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
3178 | ||
495aa969 ACB |
3179 | /* |
3180 | * Queue an RCU callback for lazy invocation after a grace period. | |
3181 | * This will likely be later named something like "call_rcu_lazy()", | |
3182 | * but this change will require some way of tagging the lazy RCU | |
3183 | * callbacks in the list of pending callbacks. Until then, this | |
3184 | * function may only be called from __kfree_rcu(). | |
3185 | */ | |
3186 | void kfree_call_rcu(struct rcu_head *head, | |
b6a4ae76 | 3187 | rcu_callback_t func) |
495aa969 | 3188 | { |
e534165b | 3189 | __call_rcu(head, func, rcu_state_p, -1, 1); |
495aa969 ACB |
3190 | } |
3191 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
3192 | ||
6d813391 PM |
3193 | /* |
3194 | * Because a context switch is a grace period for RCU-sched and RCU-bh, | |
3195 | * any blocking grace-period wait automatically implies a grace period | |
3196 | * if there is only one CPU online at any point time during execution | |
3197 | * of either synchronize_sched() or synchronize_rcu_bh(). It is OK to | |
3198 | * occasionally incorrectly indicate that there are multiple CPUs online | |
3199 | * when there was in fact only one the whole time, as this just adds | |
3200 | * some overhead: RCU still operates correctly. | |
6d813391 PM |
3201 | */ |
3202 | static inline int rcu_blocking_is_gp(void) | |
3203 | { | |
95f0c1de PM |
3204 | int ret; |
3205 | ||
6d813391 | 3206 | might_sleep(); /* Check for RCU read-side critical section. */ |
95f0c1de PM |
3207 | preempt_disable(); |
3208 | ret = num_online_cpus() <= 1; | |
3209 | preempt_enable(); | |
3210 | return ret; | |
6d813391 PM |
3211 | } |
3212 | ||
6ebb237b PM |
3213 | /** |
3214 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
3215 | * | |
3216 | * Control will return to the caller some time after a full rcu-sched | |
3217 | * grace period has elapsed, in other words after all currently executing | |
3218 | * rcu-sched read-side critical sections have completed. These read-side | |
3219 | * critical sections are delimited by rcu_read_lock_sched() and | |
3220 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
3221 | * local_irq_disable(), and so on may be used in place of | |
3222 | * rcu_read_lock_sched(). | |
3223 | * | |
3224 | * This means that all preempt_disable code sequences, including NMI and | |
f0a0e6f2 PM |
3225 | * non-threaded hardware-interrupt handlers, in progress on entry will |
3226 | * have completed before this primitive returns. However, this does not | |
3227 | * guarantee that softirq handlers will have completed, since in some | |
3228 | * kernels, these handlers can run in process context, and can block. | |
3229 | * | |
3230 | * Note that this guarantee implies further memory-ordering guarantees. | |
3231 | * On systems with more than one CPU, when synchronize_sched() returns, | |
3232 | * each CPU is guaranteed to have executed a full memory barrier since the | |
3233 | * end of its last RCU-sched read-side critical section whose beginning | |
3234 | * preceded the call to synchronize_sched(). In addition, each CPU having | |
3235 | * an RCU read-side critical section that extends beyond the return from | |
3236 | * synchronize_sched() is guaranteed to have executed a full memory barrier | |
3237 | * after the beginning of synchronize_sched() and before the beginning of | |
3238 | * that RCU read-side critical section. Note that these guarantees include | |
3239 | * CPUs that are offline, idle, or executing in user mode, as well as CPUs | |
3240 | * that are executing in the kernel. | |
3241 | * | |
3242 | * Furthermore, if CPU A invoked synchronize_sched(), which returned | |
3243 | * to its caller on CPU B, then both CPU A and CPU B are guaranteed | |
3244 | * to have executed a full memory barrier during the execution of | |
3245 | * synchronize_sched() -- even if CPU A and CPU B are the same CPU (but | |
3246 | * again only if the system has more than one CPU). | |
6ebb237b PM |
3247 | */ |
3248 | void synchronize_sched(void) | |
3249 | { | |
f78f5b90 PM |
3250 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
3251 | lock_is_held(&rcu_lock_map) || | |
3252 | lock_is_held(&rcu_sched_lock_map), | |
3253 | "Illegal synchronize_sched() in RCU-sched read-side critical section"); | |
6ebb237b PM |
3254 | if (rcu_blocking_is_gp()) |
3255 | return; | |
5afff48b | 3256 | if (rcu_gp_is_expedited()) |
3705b88d AM |
3257 | synchronize_sched_expedited(); |
3258 | else | |
3259 | wait_rcu_gp(call_rcu_sched); | |
6ebb237b PM |
3260 | } |
3261 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
3262 | ||
3263 | /** | |
3264 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
3265 | * | |
3266 | * Control will return to the caller some time after a full rcu_bh grace | |
3267 | * period has elapsed, in other words after all currently executing rcu_bh | |
3268 | * read-side critical sections have completed. RCU read-side critical | |
3269 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
3270 | * and may be nested. | |
f0a0e6f2 PM |
3271 | * |
3272 | * See the description of synchronize_sched() for more detailed information | |
3273 | * on memory ordering guarantees. | |
6ebb237b PM |
3274 | */ |
3275 | void synchronize_rcu_bh(void) | |
3276 | { | |
f78f5b90 PM |
3277 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
3278 | lock_is_held(&rcu_lock_map) || | |
3279 | lock_is_held(&rcu_sched_lock_map), | |
3280 | "Illegal synchronize_rcu_bh() in RCU-bh read-side critical section"); | |
6ebb237b PM |
3281 | if (rcu_blocking_is_gp()) |
3282 | return; | |
5afff48b | 3283 | if (rcu_gp_is_expedited()) |
3705b88d AM |
3284 | synchronize_rcu_bh_expedited(); |
3285 | else | |
3286 | wait_rcu_gp(call_rcu_bh); | |
6ebb237b PM |
3287 | } |
3288 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
3289 | ||
765a3f4f PM |
3290 | /** |
3291 | * get_state_synchronize_rcu - Snapshot current RCU state | |
3292 | * | |
3293 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
3294 | * to determine whether or not a full grace period has elapsed in the | |
3295 | * meantime. | |
3296 | */ | |
3297 | unsigned long get_state_synchronize_rcu(void) | |
3298 | { | |
3299 | /* | |
3300 | * Any prior manipulation of RCU-protected data must happen | |
3301 | * before the load from ->gpnum. | |
3302 | */ | |
3303 | smp_mb(); /* ^^^ */ | |
3304 | ||
3305 | /* | |
3306 | * Make sure this load happens before the purportedly | |
3307 | * time-consuming work between get_state_synchronize_rcu() | |
3308 | * and cond_synchronize_rcu(). | |
3309 | */ | |
e534165b | 3310 | return smp_load_acquire(&rcu_state_p->gpnum); |
765a3f4f PM |
3311 | } |
3312 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
3313 | ||
3314 | /** | |
3315 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
3316 | * | |
3317 | * @oldstate: return value from earlier call to get_state_synchronize_rcu() | |
3318 | * | |
3319 | * If a full RCU grace period has elapsed since the earlier call to | |
3320 | * get_state_synchronize_rcu(), just return. Otherwise, invoke | |
3321 | * synchronize_rcu() to wait for a full grace period. | |
3322 | * | |
3323 | * Yes, this function does not take counter wrap into account. But | |
3324 | * counter wrap is harmless. If the counter wraps, we have waited for | |
3325 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3326 | * so waiting for one additional grace period should be just fine. | |
3327 | */ | |
3328 | void cond_synchronize_rcu(unsigned long oldstate) | |
3329 | { | |
3330 | unsigned long newstate; | |
3331 | ||
3332 | /* | |
3333 | * Ensure that this load happens before any RCU-destructive | |
3334 | * actions the caller might carry out after we return. | |
3335 | */ | |
e534165b | 3336 | newstate = smp_load_acquire(&rcu_state_p->completed); |
765a3f4f PM |
3337 | if (ULONG_CMP_GE(oldstate, newstate)) |
3338 | synchronize_rcu(); | |
3339 | } | |
3340 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
3341 | ||
24560056 PM |
3342 | /** |
3343 | * get_state_synchronize_sched - Snapshot current RCU-sched state | |
3344 | * | |
3345 | * Returns a cookie that is used by a later call to cond_synchronize_sched() | |
3346 | * to determine whether or not a full grace period has elapsed in the | |
3347 | * meantime. | |
3348 | */ | |
3349 | unsigned long get_state_synchronize_sched(void) | |
3350 | { | |
3351 | /* | |
3352 | * Any prior manipulation of RCU-protected data must happen | |
3353 | * before the load from ->gpnum. | |
3354 | */ | |
3355 | smp_mb(); /* ^^^ */ | |
3356 | ||
3357 | /* | |
3358 | * Make sure this load happens before the purportedly | |
3359 | * time-consuming work between get_state_synchronize_sched() | |
3360 | * and cond_synchronize_sched(). | |
3361 | */ | |
3362 | return smp_load_acquire(&rcu_sched_state.gpnum); | |
3363 | } | |
3364 | EXPORT_SYMBOL_GPL(get_state_synchronize_sched); | |
3365 | ||
3366 | /** | |
3367 | * cond_synchronize_sched - Conditionally wait for an RCU-sched grace period | |
3368 | * | |
3369 | * @oldstate: return value from earlier call to get_state_synchronize_sched() | |
3370 | * | |
3371 | * If a full RCU-sched grace period has elapsed since the earlier call to | |
3372 | * get_state_synchronize_sched(), just return. Otherwise, invoke | |
3373 | * synchronize_sched() to wait for a full grace period. | |
3374 | * | |
3375 | * Yes, this function does not take counter wrap into account. But | |
3376 | * counter wrap is harmless. If the counter wraps, we have waited for | |
3377 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3378 | * so waiting for one additional grace period should be just fine. | |
3379 | */ | |
3380 | void cond_synchronize_sched(unsigned long oldstate) | |
3381 | { | |
3382 | unsigned long newstate; | |
3383 | ||
3384 | /* | |
3385 | * Ensure that this load happens before any RCU-destructive | |
3386 | * actions the caller might carry out after we return. | |
3387 | */ | |
3388 | newstate = smp_load_acquire(&rcu_sched_state.completed); | |
3389 | if (ULONG_CMP_GE(oldstate, newstate)) | |
3390 | synchronize_sched(); | |
3391 | } | |
3392 | EXPORT_SYMBOL_GPL(cond_synchronize_sched); | |
3393 | ||
64db4cff PM |
3394 | /* |
3395 | * Check to see if there is any immediate RCU-related work to be done | |
3396 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
3397 | * The checks are in order of increasing expense: checks that can be | |
3398 | * carried out against CPU-local state are performed first. However, | |
3399 | * we must check for CPU stalls first, else we might not get a chance. | |
3400 | */ | |
3401 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
3402 | { | |
2f51f988 PM |
3403 | struct rcu_node *rnp = rdp->mynode; |
3404 | ||
64db4cff PM |
3405 | rdp->n_rcu_pending++; |
3406 | ||
3407 | /* Check for CPU stalls, if enabled. */ | |
3408 | check_cpu_stall(rsp, rdp); | |
3409 | ||
a096932f PM |
3410 | /* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */ |
3411 | if (rcu_nohz_full_cpu(rsp)) | |
3412 | return 0; | |
3413 | ||
64db4cff | 3414 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
5c51dd73 | 3415 | if (rcu_scheduler_fully_active && |
5b74c458 | 3416 | rdp->core_needs_qs && rdp->cpu_no_qs.b.norm && |
9577df9a | 3417 | rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_dynticks.rcu_qs_ctr)) { |
97c668b8 | 3418 | rdp->n_rp_core_needs_qs++; |
3a19b46a | 3419 | } else if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) { |
d21670ac | 3420 | rdp->n_rp_report_qs++; |
64db4cff | 3421 | return 1; |
7ba5c840 | 3422 | } |
64db4cff PM |
3423 | |
3424 | /* Does this CPU have callbacks ready to invoke? */ | |
15fecf89 | 3425 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) { |
7ba5c840 | 3426 | rdp->n_rp_cb_ready++; |
64db4cff | 3427 | return 1; |
7ba5c840 | 3428 | } |
64db4cff PM |
3429 | |
3430 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
3431 | if (cpu_needs_another_gp(rsp, rdp)) { |
3432 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 3433 | return 1; |
7ba5c840 | 3434 | } |
64db4cff PM |
3435 | |
3436 | /* Has another RCU grace period completed? */ | |
7d0ae808 | 3437 | if (READ_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 3438 | rdp->n_rp_gp_completed++; |
64db4cff | 3439 | return 1; |
7ba5c840 | 3440 | } |
64db4cff PM |
3441 | |
3442 | /* Has a new RCU grace period started? */ | |
7d0ae808 PM |
3443 | if (READ_ONCE(rnp->gpnum) != rdp->gpnum || |
3444 | unlikely(READ_ONCE(rdp->gpwrap))) { /* outside lock */ | |
7ba5c840 | 3445 | rdp->n_rp_gp_started++; |
64db4cff | 3446 | return 1; |
7ba5c840 | 3447 | } |
64db4cff | 3448 | |
96d3fd0d PM |
3449 | /* Does this CPU need a deferred NOCB wakeup? */ |
3450 | if (rcu_nocb_need_deferred_wakeup(rdp)) { | |
3451 | rdp->n_rp_nocb_defer_wakeup++; | |
3452 | return 1; | |
3453 | } | |
3454 | ||
64db4cff | 3455 | /* nothing to do */ |
7ba5c840 | 3456 | rdp->n_rp_need_nothing++; |
64db4cff PM |
3457 | return 0; |
3458 | } | |
3459 | ||
3460 | /* | |
3461 | * Check to see if there is any immediate RCU-related work to be done | |
3462 | * by the current CPU, returning 1 if so. This function is part of the | |
3463 | * RCU implementation; it is -not- an exported member of the RCU API. | |
3464 | */ | |
e3950ecd | 3465 | static int rcu_pending(void) |
64db4cff | 3466 | { |
6ce75a23 PM |
3467 | struct rcu_state *rsp; |
3468 | ||
3469 | for_each_rcu_flavor(rsp) | |
e3950ecd | 3470 | if (__rcu_pending(rsp, this_cpu_ptr(rsp->rda))) |
6ce75a23 PM |
3471 | return 1; |
3472 | return 0; | |
64db4cff PM |
3473 | } |
3474 | ||
3475 | /* | |
c0f4dfd4 PM |
3476 | * Return true if the specified CPU has any callback. If all_lazy is |
3477 | * non-NULL, store an indication of whether all callbacks are lazy. | |
3478 | * (If there are no callbacks, all of them are deemed to be lazy.) | |
64db4cff | 3479 | */ |
82072c4f | 3480 | static bool __maybe_unused rcu_cpu_has_callbacks(bool *all_lazy) |
64db4cff | 3481 | { |
c0f4dfd4 PM |
3482 | bool al = true; |
3483 | bool hc = false; | |
3484 | struct rcu_data *rdp; | |
6ce75a23 PM |
3485 | struct rcu_state *rsp; |
3486 | ||
c0f4dfd4 | 3487 | for_each_rcu_flavor(rsp) { |
aa6da514 | 3488 | rdp = this_cpu_ptr(rsp->rda); |
15fecf89 | 3489 | if (rcu_segcblist_empty(&rdp->cblist)) |
69c8d28c PM |
3490 | continue; |
3491 | hc = true; | |
15fecf89 | 3492 | if (rcu_segcblist_n_nonlazy_cbs(&rdp->cblist) || !all_lazy) { |
c0f4dfd4 | 3493 | al = false; |
69c8d28c PM |
3494 | break; |
3495 | } | |
c0f4dfd4 PM |
3496 | } |
3497 | if (all_lazy) | |
3498 | *all_lazy = al; | |
3499 | return hc; | |
64db4cff PM |
3500 | } |
3501 | ||
a83eff0a PM |
3502 | /* |
3503 | * Helper function for _rcu_barrier() tracing. If tracing is disabled, | |
3504 | * the compiler is expected to optimize this away. | |
3505 | */ | |
e66c33d5 | 3506 | static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s, |
a83eff0a PM |
3507 | int cpu, unsigned long done) |
3508 | { | |
3509 | trace_rcu_barrier(rsp->name, s, cpu, | |
3510 | atomic_read(&rsp->barrier_cpu_count), done); | |
3511 | } | |
3512 | ||
b1420f1c PM |
3513 | /* |
3514 | * RCU callback function for _rcu_barrier(). If we are last, wake | |
3515 | * up the task executing _rcu_barrier(). | |
3516 | */ | |
24ebbca8 | 3517 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 3518 | { |
24ebbca8 PM |
3519 | struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head); |
3520 | struct rcu_state *rsp = rdp->rsp; | |
3521 | ||
a83eff0a | 3522 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) { |
d8db2e86 PM |
3523 | _rcu_barrier_trace(rsp, TPS("LastCB"), -1, |
3524 | rsp->barrier_sequence); | |
7db74df8 | 3525 | complete(&rsp->barrier_completion); |
a83eff0a | 3526 | } else { |
d8db2e86 | 3527 | _rcu_barrier_trace(rsp, TPS("CB"), -1, rsp->barrier_sequence); |
a83eff0a | 3528 | } |
d0ec774c PM |
3529 | } |
3530 | ||
3531 | /* | |
3532 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
3533 | */ | |
3534 | static void rcu_barrier_func(void *type) | |
3535 | { | |
037b64ed | 3536 | struct rcu_state *rsp = type; |
fa07a58f | 3537 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
d0ec774c | 3538 | |
d8db2e86 | 3539 | _rcu_barrier_trace(rsp, TPS("IRQ"), -1, rsp->barrier_sequence); |
f92c734f PM |
3540 | rdp->barrier_head.func = rcu_barrier_callback; |
3541 | debug_rcu_head_queue(&rdp->barrier_head); | |
3542 | if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head, 0)) { | |
3543 | atomic_inc(&rsp->barrier_cpu_count); | |
3544 | } else { | |
3545 | debug_rcu_head_unqueue(&rdp->barrier_head); | |
d8db2e86 PM |
3546 | _rcu_barrier_trace(rsp, TPS("IRQNQ"), -1, |
3547 | rsp->barrier_sequence); | |
f92c734f | 3548 | } |
d0ec774c PM |
3549 | } |
3550 | ||
d0ec774c PM |
3551 | /* |
3552 | * Orchestrate the specified type of RCU barrier, waiting for all | |
3553 | * RCU callbacks of the specified type to complete. | |
3554 | */ | |
037b64ed | 3555 | static void _rcu_barrier(struct rcu_state *rsp) |
d0ec774c | 3556 | { |
b1420f1c | 3557 | int cpu; |
b1420f1c | 3558 | struct rcu_data *rdp; |
4f525a52 | 3559 | unsigned long s = rcu_seq_snap(&rsp->barrier_sequence); |
b1420f1c | 3560 | |
d8db2e86 | 3561 | _rcu_barrier_trace(rsp, TPS("Begin"), -1, s); |
b1420f1c | 3562 | |
e74f4c45 | 3563 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
7be7f0be | 3564 | mutex_lock(&rsp->barrier_mutex); |
b1420f1c | 3565 | |
4f525a52 PM |
3566 | /* Did someone else do our work for us? */ |
3567 | if (rcu_seq_done(&rsp->barrier_sequence, s)) { | |
d8db2e86 PM |
3568 | _rcu_barrier_trace(rsp, TPS("EarlyExit"), -1, |
3569 | rsp->barrier_sequence); | |
cf3a9c48 PM |
3570 | smp_mb(); /* caller's subsequent code after above check. */ |
3571 | mutex_unlock(&rsp->barrier_mutex); | |
3572 | return; | |
3573 | } | |
3574 | ||
4f525a52 PM |
3575 | /* Mark the start of the barrier operation. */ |
3576 | rcu_seq_start(&rsp->barrier_sequence); | |
d8db2e86 | 3577 | _rcu_barrier_trace(rsp, TPS("Inc1"), -1, rsp->barrier_sequence); |
b1420f1c | 3578 | |
d0ec774c | 3579 | /* |
b1420f1c PM |
3580 | * Initialize the count to one rather than to zero in order to |
3581 | * avoid a too-soon return to zero in case of a short grace period | |
1331e7a1 PM |
3582 | * (or preemption of this task). Exclude CPU-hotplug operations |
3583 | * to ensure that no offline CPU has callbacks queued. | |
d0ec774c | 3584 | */ |
7db74df8 | 3585 | init_completion(&rsp->barrier_completion); |
24ebbca8 | 3586 | atomic_set(&rsp->barrier_cpu_count, 1); |
1331e7a1 | 3587 | get_online_cpus(); |
b1420f1c PM |
3588 | |
3589 | /* | |
1331e7a1 PM |
3590 | * Force each CPU with callbacks to register a new callback. |
3591 | * When that callback is invoked, we will know that all of the | |
3592 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 3593 | */ |
3fbfbf7a | 3594 | for_each_possible_cpu(cpu) { |
d1e43fa5 | 3595 | if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu)) |
3fbfbf7a | 3596 | continue; |
b1420f1c | 3597 | rdp = per_cpu_ptr(rsp->rda, cpu); |
d1e43fa5 | 3598 | if (rcu_is_nocb_cpu(cpu)) { |
d7e29933 | 3599 | if (!rcu_nocb_cpu_needs_barrier(rsp, cpu)) { |
d8db2e86 | 3600 | _rcu_barrier_trace(rsp, TPS("OfflineNoCB"), cpu, |
4f525a52 | 3601 | rsp->barrier_sequence); |
d7e29933 | 3602 | } else { |
d8db2e86 | 3603 | _rcu_barrier_trace(rsp, TPS("OnlineNoCB"), cpu, |
4f525a52 | 3604 | rsp->barrier_sequence); |
41050a00 | 3605 | smp_mb__before_atomic(); |
d7e29933 PM |
3606 | atomic_inc(&rsp->barrier_cpu_count); |
3607 | __call_rcu(&rdp->barrier_head, | |
3608 | rcu_barrier_callback, rsp, cpu, 0); | |
3609 | } | |
15fecf89 | 3610 | } else if (rcu_segcblist_n_cbs(&rdp->cblist)) { |
d8db2e86 | 3611 | _rcu_barrier_trace(rsp, TPS("OnlineQ"), cpu, |
4f525a52 | 3612 | rsp->barrier_sequence); |
037b64ed | 3613 | smp_call_function_single(cpu, rcu_barrier_func, rsp, 1); |
b1420f1c | 3614 | } else { |
d8db2e86 | 3615 | _rcu_barrier_trace(rsp, TPS("OnlineNQ"), cpu, |
4f525a52 | 3616 | rsp->barrier_sequence); |
b1420f1c PM |
3617 | } |
3618 | } | |
1331e7a1 | 3619 | put_online_cpus(); |
b1420f1c PM |
3620 | |
3621 | /* | |
3622 | * Now that we have an rcu_barrier_callback() callback on each | |
3623 | * CPU, and thus each counted, remove the initial count. | |
3624 | */ | |
24ebbca8 | 3625 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) |
7db74df8 | 3626 | complete(&rsp->barrier_completion); |
b1420f1c PM |
3627 | |
3628 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
7db74df8 | 3629 | wait_for_completion(&rsp->barrier_completion); |
b1420f1c | 3630 | |
4f525a52 | 3631 | /* Mark the end of the barrier operation. */ |
d8db2e86 | 3632 | _rcu_barrier_trace(rsp, TPS("Inc2"), -1, rsp->barrier_sequence); |
4f525a52 PM |
3633 | rcu_seq_end(&rsp->barrier_sequence); |
3634 | ||
b1420f1c | 3635 | /* Other rcu_barrier() invocations can now safely proceed. */ |
7be7f0be | 3636 | mutex_unlock(&rsp->barrier_mutex); |
d0ec774c | 3637 | } |
d0ec774c PM |
3638 | |
3639 | /** | |
3640 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
3641 | */ | |
3642 | void rcu_barrier_bh(void) | |
3643 | { | |
037b64ed | 3644 | _rcu_barrier(&rcu_bh_state); |
d0ec774c PM |
3645 | } |
3646 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
3647 | ||
3648 | /** | |
3649 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
3650 | */ | |
3651 | void rcu_barrier_sched(void) | |
3652 | { | |
037b64ed | 3653 | _rcu_barrier(&rcu_sched_state); |
d0ec774c PM |
3654 | } |
3655 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
3656 | ||
0aa04b05 PM |
3657 | /* |
3658 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
3659 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
3660 | * must hold the corresponding leaf rcu_node ->lock with interrrupts | |
3661 | * disabled. | |
3662 | */ | |
3663 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
3664 | { | |
3665 | long mask; | |
3666 | struct rcu_node *rnp = rnp_leaf; | |
3667 | ||
c0b334c5 | 3668 | lockdep_assert_held(&rnp->lock); |
0aa04b05 PM |
3669 | for (;;) { |
3670 | mask = rnp->grpmask; | |
3671 | rnp = rnp->parent; | |
3672 | if (rnp == NULL) | |
3673 | return; | |
6cf10081 | 3674 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
0aa04b05 | 3675 | rnp->qsmaskinit |= mask; |
67c583a7 | 3676 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
0aa04b05 PM |
3677 | } |
3678 | } | |
3679 | ||
64db4cff | 3680 | /* |
27569620 | 3681 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 3682 | */ |
27569620 PM |
3683 | static void __init |
3684 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
3685 | { |
3686 | unsigned long flags; | |
394f99a9 | 3687 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
3688 | struct rcu_node *rnp = rcu_get_root(rsp); |
3689 | ||
3690 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 3691 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
bc75e999 | 3692 | rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu); |
27569620 | 3693 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
51a1fd30 | 3694 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != 1); |
02a5c550 | 3695 | WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp->dynticks))); |
27569620 | 3696 | rdp->cpu = cpu; |
d4c08f2a | 3697 | rdp->rsp = rsp; |
3fbfbf7a | 3698 | rcu_boot_init_nocb_percpu_data(rdp); |
67c583a7 | 3699 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27569620 PM |
3700 | } |
3701 | ||
3702 | /* | |
3703 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
3704 | * offline event can be happening at a given time. Note also that we | |
3705 | * can accept some slop in the rsp->completed access due to the fact | |
3706 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 3707 | */ |
49fb4c62 | 3708 | static void |
9b67122a | 3709 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp) |
64db4cff PM |
3710 | { |
3711 | unsigned long flags; | |
394f99a9 | 3712 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
3713 | struct rcu_node *rnp = rcu_get_root(rsp); |
3714 | ||
3715 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 3716 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 PM |
3717 | rdp->qlen_last_fqs_check = 0; |
3718 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 3719 | rdp->blimit = blimit; |
15fecf89 PM |
3720 | if (rcu_segcblist_empty(&rdp->cblist) && /* No early-boot CBs? */ |
3721 | !init_nocb_callback_list(rdp)) | |
3722 | rcu_segcblist_init(&rdp->cblist); /* Re-enable callbacks. */ | |
51a1fd30 | 3723 | rdp->dynticks->dynticks_nesting = 1; |
2625d469 | 3724 | rcu_dynticks_eqs_online(); |
67c583a7 | 3725 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
64db4cff | 3726 | |
0aa04b05 PM |
3727 | /* |
3728 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
3729 | * propagation up the rcu_node tree will happen at the beginning | |
3730 | * of the next grace period. | |
3731 | */ | |
64db4cff | 3732 | rnp = rdp->mynode; |
2a67e741 | 3733 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
b9585e94 | 3734 | rdp->beenonline = true; /* We have now been online. */ |
0aa04b05 PM |
3735 | rdp->gpnum = rnp->completed; /* Make CPU later note any new GP. */ |
3736 | rdp->completed = rnp->completed; | |
5b74c458 | 3737 | rdp->cpu_no_qs.b.norm = true; |
9577df9a | 3738 | rdp->rcu_qs_ctr_snap = per_cpu(rcu_dynticks.rcu_qs_ctr, cpu); |
97c668b8 | 3739 | rdp->core_needs_qs = false; |
9b9500da PM |
3740 | rdp->rcu_iw_pending = false; |
3741 | rdp->rcu_iw_gpnum = rnp->gpnum - 1; | |
0aa04b05 | 3742 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl")); |
67c583a7 | 3743 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
3744 | } |
3745 | ||
deb34f36 PM |
3746 | /* |
3747 | * Invoked early in the CPU-online process, when pretty much all | |
3748 | * services are available. The incoming CPU is not present. | |
3749 | */ | |
4df83742 | 3750 | int rcutree_prepare_cpu(unsigned int cpu) |
64db4cff | 3751 | { |
6ce75a23 PM |
3752 | struct rcu_state *rsp; |
3753 | ||
3754 | for_each_rcu_flavor(rsp) | |
9b67122a | 3755 | rcu_init_percpu_data(cpu, rsp); |
4df83742 TG |
3756 | |
3757 | rcu_prepare_kthreads(cpu); | |
3758 | rcu_spawn_all_nocb_kthreads(cpu); | |
3759 | ||
3760 | return 0; | |
3761 | } | |
3762 | ||
deb34f36 PM |
3763 | /* |
3764 | * Update RCU priority boot kthread affinity for CPU-hotplug changes. | |
3765 | */ | |
4df83742 TG |
3766 | static void rcutree_affinity_setting(unsigned int cpu, int outgoing) |
3767 | { | |
3768 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); | |
3769 | ||
3770 | rcu_boost_kthread_setaffinity(rdp->mynode, outgoing); | |
3771 | } | |
3772 | ||
deb34f36 PM |
3773 | /* |
3774 | * Near the end of the CPU-online process. Pretty much all services | |
3775 | * enabled, and the CPU is now very much alive. | |
3776 | */ | |
4df83742 TG |
3777 | int rcutree_online_cpu(unsigned int cpu) |
3778 | { | |
9b9500da PM |
3779 | unsigned long flags; |
3780 | struct rcu_data *rdp; | |
3781 | struct rcu_node *rnp; | |
3782 | struct rcu_state *rsp; | |
3783 | ||
3784 | for_each_rcu_flavor(rsp) { | |
3785 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
3786 | rnp = rdp->mynode; | |
3787 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3788 | rnp->ffmask |= rdp->grpmask; | |
3789 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
3790 | } | |
da915ad5 PM |
3791 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3792 | srcu_online_cpu(cpu); | |
9b9500da PM |
3793 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
3794 | return 0; /* Too early in boot for scheduler work. */ | |
3795 | sync_sched_exp_online_cleanup(cpu); | |
3796 | rcutree_affinity_setting(cpu, -1); | |
4df83742 TG |
3797 | return 0; |
3798 | } | |
3799 | ||
deb34f36 PM |
3800 | /* |
3801 | * Near the beginning of the process. The CPU is still very much alive | |
3802 | * with pretty much all services enabled. | |
3803 | */ | |
4df83742 TG |
3804 | int rcutree_offline_cpu(unsigned int cpu) |
3805 | { | |
9b9500da PM |
3806 | unsigned long flags; |
3807 | struct rcu_data *rdp; | |
3808 | struct rcu_node *rnp; | |
3809 | struct rcu_state *rsp; | |
3810 | ||
3811 | for_each_rcu_flavor(rsp) { | |
3812 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
3813 | rnp = rdp->mynode; | |
3814 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3815 | rnp->ffmask &= ~rdp->grpmask; | |
3816 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
3817 | } | |
3818 | ||
4df83742 | 3819 | rcutree_affinity_setting(cpu, cpu); |
da915ad5 PM |
3820 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3821 | srcu_offline_cpu(cpu); | |
4df83742 TG |
3822 | return 0; |
3823 | } | |
3824 | ||
deb34f36 PM |
3825 | /* |
3826 | * Near the end of the offline process. We do only tracing here. | |
3827 | */ | |
4df83742 TG |
3828 | int rcutree_dying_cpu(unsigned int cpu) |
3829 | { | |
3830 | struct rcu_state *rsp; | |
3831 | ||
3832 | for_each_rcu_flavor(rsp) | |
3833 | rcu_cleanup_dying_cpu(rsp); | |
3834 | return 0; | |
3835 | } | |
3836 | ||
deb34f36 PM |
3837 | /* |
3838 | * The outgoing CPU is gone and we are running elsewhere. | |
3839 | */ | |
4df83742 TG |
3840 | int rcutree_dead_cpu(unsigned int cpu) |
3841 | { | |
3842 | struct rcu_state *rsp; | |
3843 | ||
3844 | for_each_rcu_flavor(rsp) { | |
3845 | rcu_cleanup_dead_cpu(cpu, rsp); | |
3846 | do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu)); | |
3847 | } | |
3848 | return 0; | |
64db4cff PM |
3849 | } |
3850 | ||
7ec99de3 PM |
3851 | /* |
3852 | * Mark the specified CPU as being online so that subsequent grace periods | |
3853 | * (both expedited and normal) will wait on it. Note that this means that | |
3854 | * incoming CPUs are not allowed to use RCU read-side critical sections | |
3855 | * until this function is called. Failing to observe this restriction | |
3856 | * will result in lockdep splats. | |
deb34f36 PM |
3857 | * |
3858 | * Note that this function is special in that it is invoked directly | |
3859 | * from the incoming CPU rather than from the cpuhp_step mechanism. | |
3860 | * This is because this function must be invoked at a precise location. | |
7ec99de3 PM |
3861 | */ |
3862 | void rcu_cpu_starting(unsigned int cpu) | |
3863 | { | |
3864 | unsigned long flags; | |
3865 | unsigned long mask; | |
313517fc PM |
3866 | int nbits; |
3867 | unsigned long oldmask; | |
7ec99de3 PM |
3868 | struct rcu_data *rdp; |
3869 | struct rcu_node *rnp; | |
3870 | struct rcu_state *rsp; | |
3871 | ||
3872 | for_each_rcu_flavor(rsp) { | |
fdbb9b31 | 3873 | rdp = per_cpu_ptr(rsp->rda, cpu); |
7ec99de3 PM |
3874 | rnp = rdp->mynode; |
3875 | mask = rdp->grpmask; | |
3876 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3877 | rnp->qsmaskinitnext |= mask; | |
313517fc | 3878 | oldmask = rnp->expmaskinitnext; |
7ec99de3 | 3879 | rnp->expmaskinitnext |= mask; |
313517fc PM |
3880 | oldmask ^= rnp->expmaskinitnext; |
3881 | nbits = bitmap_weight(&oldmask, BITS_PER_LONG); | |
3882 | /* Allow lockless access for expedited grace periods. */ | |
3883 | smp_store_release(&rsp->ncpus, rsp->ncpus + nbits); /* ^^^ */ | |
7ec99de3 PM |
3884 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
3885 | } | |
313517fc | 3886 | smp_mb(); /* Ensure RCU read-side usage follows above initialization. */ |
7ec99de3 PM |
3887 | } |
3888 | ||
27d50c7e TG |
3889 | #ifdef CONFIG_HOTPLUG_CPU |
3890 | /* | |
3891 | * The CPU is exiting the idle loop into the arch_cpu_idle_dead() | |
3892 | * function. We now remove it from the rcu_node tree's ->qsmaskinit | |
3893 | * bit masks. | |
3894 | */ | |
3895 | static void rcu_cleanup_dying_idle_cpu(int cpu, struct rcu_state *rsp) | |
3896 | { | |
3897 | unsigned long flags; | |
3898 | unsigned long mask; | |
3899 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
3900 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ | |
3901 | ||
27d50c7e TG |
3902 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ |
3903 | mask = rdp->grpmask; | |
3904 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ | |
3905 | rnp->qsmaskinitnext &= ~mask; | |
710d60cb | 3906 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27d50c7e TG |
3907 | } |
3908 | ||
deb34f36 PM |
3909 | /* |
3910 | * The outgoing function has no further need of RCU, so remove it from | |
3911 | * the list of CPUs that RCU must track. | |
3912 | * | |
3913 | * Note that this function is special in that it is invoked directly | |
3914 | * from the outgoing CPU rather than from the cpuhp_step mechanism. | |
3915 | * This is because this function must be invoked at a precise location. | |
3916 | */ | |
27d50c7e TG |
3917 | void rcu_report_dead(unsigned int cpu) |
3918 | { | |
3919 | struct rcu_state *rsp; | |
3920 | ||
3921 | /* QS for any half-done expedited RCU-sched GP. */ | |
3922 | preempt_disable(); | |
3923 | rcu_report_exp_rdp(&rcu_sched_state, | |
3924 | this_cpu_ptr(rcu_sched_state.rda), true); | |
3925 | preempt_enable(); | |
3926 | for_each_rcu_flavor(rsp) | |
3927 | rcu_cleanup_dying_idle_cpu(cpu, rsp); | |
3928 | } | |
a58163d8 | 3929 | |
f2dbe4a5 | 3930 | /* Migrate the dead CPU's callbacks to the current CPU. */ |
a58163d8 PM |
3931 | static void rcu_migrate_callbacks(int cpu, struct rcu_state *rsp) |
3932 | { | |
3933 | unsigned long flags; | |
b1a2d79f | 3934 | struct rcu_data *my_rdp; |
a58163d8 | 3935 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
9fa46fb8 | 3936 | struct rcu_node *rnp_root = rcu_get_root(rdp->rsp); |
a58163d8 | 3937 | |
95335c03 PM |
3938 | if (rcu_is_nocb_cpu(cpu) || rcu_segcblist_empty(&rdp->cblist)) |
3939 | return; /* No callbacks to migrate. */ | |
3940 | ||
b1a2d79f PM |
3941 | local_irq_save(flags); |
3942 | my_rdp = this_cpu_ptr(rsp->rda); | |
3943 | if (rcu_nocb_adopt_orphan_cbs(my_rdp, rdp, flags)) { | |
3944 | local_irq_restore(flags); | |
3945 | return; | |
3946 | } | |
9fa46fb8 PM |
3947 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ |
3948 | rcu_advance_cbs(rsp, rnp_root, rdp); /* Leverage recent GPs. */ | |
21cc2483 | 3949 | rcu_advance_cbs(rsp, rnp_root, my_rdp); /* Assign GP to pending CBs. */ |
f2dbe4a5 | 3950 | rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist); |
09efeeee PM |
3951 | WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) != |
3952 | !rcu_segcblist_n_cbs(&my_rdp->cblist)); | |
537b85c8 | 3953 | raw_spin_unlock_irqrestore_rcu_node(rnp_root, flags); |
a58163d8 PM |
3954 | WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 || |
3955 | !rcu_segcblist_empty(&rdp->cblist), | |
3956 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n", | |
3957 | cpu, rcu_segcblist_n_cbs(&rdp->cblist), | |
3958 | rcu_segcblist_first_cb(&rdp->cblist)); | |
3959 | } | |
3960 | ||
3961 | /* | |
3962 | * The outgoing CPU has just passed through the dying-idle state, | |
3963 | * and we are being invoked from the CPU that was IPIed to continue the | |
3964 | * offline operation. We need to migrate the outgoing CPU's callbacks. | |
3965 | */ | |
3966 | void rcutree_migrate_callbacks(int cpu) | |
3967 | { | |
3968 | struct rcu_state *rsp; | |
3969 | ||
3970 | for_each_rcu_flavor(rsp) | |
3971 | rcu_migrate_callbacks(cpu, rsp); | |
3972 | } | |
27d50c7e TG |
3973 | #endif |
3974 | ||
deb34f36 PM |
3975 | /* |
3976 | * On non-huge systems, use expedited RCU grace periods to make suspend | |
3977 | * and hibernation run faster. | |
3978 | */ | |
d1d74d14 BP |
3979 | static int rcu_pm_notify(struct notifier_block *self, |
3980 | unsigned long action, void *hcpu) | |
3981 | { | |
3982 | switch (action) { | |
3983 | case PM_HIBERNATION_PREPARE: | |
3984 | case PM_SUSPEND_PREPARE: | |
3985 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ | |
5afff48b | 3986 | rcu_expedite_gp(); |
d1d74d14 BP |
3987 | break; |
3988 | case PM_POST_HIBERNATION: | |
3989 | case PM_POST_SUSPEND: | |
5afff48b PM |
3990 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ |
3991 | rcu_unexpedite_gp(); | |
d1d74d14 BP |
3992 | break; |
3993 | default: | |
3994 | break; | |
3995 | } | |
3996 | return NOTIFY_OK; | |
3997 | } | |
3998 | ||
b3dbec76 | 3999 | /* |
9386c0b7 | 4000 | * Spawn the kthreads that handle each RCU flavor's grace periods. |
b3dbec76 PM |
4001 | */ |
4002 | static int __init rcu_spawn_gp_kthread(void) | |
4003 | { | |
4004 | unsigned long flags; | |
a94844b2 | 4005 | int kthread_prio_in = kthread_prio; |
b3dbec76 PM |
4006 | struct rcu_node *rnp; |
4007 | struct rcu_state *rsp; | |
a94844b2 | 4008 | struct sched_param sp; |
b3dbec76 PM |
4009 | struct task_struct *t; |
4010 | ||
a94844b2 PM |
4011 | /* Force priority into range. */ |
4012 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
4013 | kthread_prio = 1; | |
4014 | else if (kthread_prio < 0) | |
4015 | kthread_prio = 0; | |
4016 | else if (kthread_prio > 99) | |
4017 | kthread_prio = 99; | |
4018 | if (kthread_prio != kthread_prio_in) | |
4019 | pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n", | |
4020 | kthread_prio, kthread_prio_in); | |
4021 | ||
9386c0b7 | 4022 | rcu_scheduler_fully_active = 1; |
b3dbec76 | 4023 | for_each_rcu_flavor(rsp) { |
a94844b2 | 4024 | t = kthread_create(rcu_gp_kthread, rsp, "%s", rsp->name); |
b3dbec76 PM |
4025 | BUG_ON(IS_ERR(t)); |
4026 | rnp = rcu_get_root(rsp); | |
6cf10081 | 4027 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b3dbec76 | 4028 | rsp->gp_kthread = t; |
a94844b2 PM |
4029 | if (kthread_prio) { |
4030 | sp.sched_priority = kthread_prio; | |
4031 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
4032 | } | |
67c583a7 | 4033 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
e11f1335 | 4034 | wake_up_process(t); |
b3dbec76 | 4035 | } |
35ce7f29 | 4036 | rcu_spawn_nocb_kthreads(); |
9386c0b7 | 4037 | rcu_spawn_boost_kthreads(); |
b3dbec76 PM |
4038 | return 0; |
4039 | } | |
4040 | early_initcall(rcu_spawn_gp_kthread); | |
4041 | ||
bbad9379 | 4042 | /* |
52d7e48b PM |
4043 | * This function is invoked towards the end of the scheduler's |
4044 | * initialization process. Before this is called, the idle task might | |
4045 | * contain synchronous grace-period primitives (during which time, this idle | |
4046 | * task is booting the system, and such primitives are no-ops). After this | |
4047 | * function is called, any synchronous grace-period primitives are run as | |
4048 | * expedited, with the requesting task driving the grace period forward. | |
900b1028 | 4049 | * A later core_initcall() rcu_set_runtime_mode() will switch to full |
52d7e48b | 4050 | * runtime RCU functionality. |
bbad9379 PM |
4051 | */ |
4052 | void rcu_scheduler_starting(void) | |
4053 | { | |
4054 | WARN_ON(num_online_cpus() != 1); | |
4055 | WARN_ON(nr_context_switches() > 0); | |
52d7e48b PM |
4056 | rcu_test_sync_prims(); |
4057 | rcu_scheduler_active = RCU_SCHEDULER_INIT; | |
4058 | rcu_test_sync_prims(); | |
bbad9379 PM |
4059 | } |
4060 | ||
64db4cff PM |
4061 | /* |
4062 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
4063 | */ | |
a87f203e | 4064 | static void __init rcu_init_one(struct rcu_state *rsp) |
64db4cff | 4065 | { |
cb007102 AG |
4066 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
4067 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
4068 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
4069 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
199977bf | 4070 | |
199977bf | 4071 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ |
64db4cff PM |
4072 | int cpustride = 1; |
4073 | int i; | |
4074 | int j; | |
4075 | struct rcu_node *rnp; | |
4076 | ||
05b84aec | 4077 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 4078 | |
3eaaaf6c PM |
4079 | /* Silence gcc 4.8 false positive about array index out of range. */ |
4080 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
4081 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 4082 | |
64db4cff PM |
4083 | /* Initialize the level-tracking arrays. */ |
4084 | ||
f885b7f2 | 4085 | for (i = 1; i < rcu_num_lvls; i++) |
41f5c631 PM |
4086 | rsp->level[i] = rsp->level[i - 1] + num_rcu_lvl[i - 1]; |
4087 | rcu_init_levelspread(levelspread, num_rcu_lvl); | |
64db4cff PM |
4088 | |
4089 | /* Initialize the elements themselves, starting from the leaves. */ | |
4090 | ||
f885b7f2 | 4091 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 4092 | cpustride *= levelspread[i]; |
64db4cff | 4093 | rnp = rsp->level[i]; |
41f5c631 | 4094 | for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) { |
67c583a7 BF |
4095 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
4096 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 4097 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
4098 | raw_spin_lock_init(&rnp->fqslock); |
4099 | lockdep_set_class_and_name(&rnp->fqslock, | |
4100 | &rcu_fqs_class[i], fqs[i]); | |
25d30cf4 PM |
4101 | rnp->gpnum = rsp->gpnum; |
4102 | rnp->completed = rsp->completed; | |
64db4cff PM |
4103 | rnp->qsmask = 0; |
4104 | rnp->qsmaskinit = 0; | |
4105 | rnp->grplo = j * cpustride; | |
4106 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
4107 | if (rnp->grphi >= nr_cpu_ids) |
4108 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
4109 | if (i == 0) { |
4110 | rnp->grpnum = 0; | |
4111 | rnp->grpmask = 0; | |
4112 | rnp->parent = NULL; | |
4113 | } else { | |
199977bf | 4114 | rnp->grpnum = j % levelspread[i - 1]; |
64db4cff PM |
4115 | rnp->grpmask = 1UL << rnp->grpnum; |
4116 | rnp->parent = rsp->level[i - 1] + | |
199977bf | 4117 | j / levelspread[i - 1]; |
64db4cff PM |
4118 | } |
4119 | rnp->level = i; | |
12f5f524 | 4120 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 4121 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
4122 | init_waitqueue_head(&rnp->exp_wq[0]); |
4123 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
4124 | init_waitqueue_head(&rnp->exp_wq[2]); |
4125 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 4126 | spin_lock_init(&rnp->exp_lock); |
64db4cff PM |
4127 | } |
4128 | } | |
0c34029a | 4129 | |
abedf8e2 PG |
4130 | init_swait_queue_head(&rsp->gp_wq); |
4131 | init_swait_queue_head(&rsp->expedited_wq); | |
f885b7f2 | 4132 | rnp = rsp->level[rcu_num_lvls - 1]; |
0c34029a | 4133 | for_each_possible_cpu(i) { |
4a90a068 | 4134 | while (i > rnp->grphi) |
0c34029a | 4135 | rnp++; |
394f99a9 | 4136 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
4137 | rcu_boot_init_percpu_data(i, rsp); |
4138 | } | |
6ce75a23 | 4139 | list_add(&rsp->flavors, &rcu_struct_flavors); |
64db4cff PM |
4140 | } |
4141 | ||
f885b7f2 PM |
4142 | /* |
4143 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 4144 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
4145 | * the ->node array in the rcu_state structure. |
4146 | */ | |
4147 | static void __init rcu_init_geometry(void) | |
4148 | { | |
026ad283 | 4149 | ulong d; |
f885b7f2 | 4150 | int i; |
05b84aec | 4151 | int rcu_capacity[RCU_NUM_LVLS]; |
f885b7f2 | 4152 | |
026ad283 PM |
4153 | /* |
4154 | * Initialize any unspecified boot parameters. | |
4155 | * The default values of jiffies_till_first_fqs and | |
4156 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
4157 | * value, which is a function of HZ, then adding one for each | |
4158 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
4159 | */ | |
4160 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
4161 | if (jiffies_till_first_fqs == ULONG_MAX) | |
4162 | jiffies_till_first_fqs = d; | |
4163 | if (jiffies_till_next_fqs == ULONG_MAX) | |
4164 | jiffies_till_next_fqs = d; | |
4165 | ||
f885b7f2 | 4166 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 4167 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 4168 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 4169 | return; |
9b130ad5 | 4170 | pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n", |
39479098 | 4171 | rcu_fanout_leaf, nr_cpu_ids); |
f885b7f2 | 4172 | |
f885b7f2 | 4173 | /* |
ee968ac6 PM |
4174 | * The boot-time rcu_fanout_leaf parameter must be at least two |
4175 | * and cannot exceed the number of bits in the rcu_node masks. | |
4176 | * Complain and fall back to the compile-time values if this | |
4177 | * limit is exceeded. | |
f885b7f2 | 4178 | */ |
ee968ac6 | 4179 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 4180 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 4181 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
4182 | WARN_ON(1); |
4183 | return; | |
4184 | } | |
4185 | ||
f885b7f2 PM |
4186 | /* |
4187 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 4188 | * with the given number of levels. |
f885b7f2 | 4189 | */ |
9618138b | 4190 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 4191 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 4192 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
4193 | |
4194 | /* | |
75cf15a4 | 4195 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 4196 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 4197 | */ |
ee968ac6 PM |
4198 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
4199 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
4200 | WARN_ON(1); | |
4201 | return; | |
4202 | } | |
f885b7f2 | 4203 | |
679f9858 | 4204 | /* Calculate the number of levels in the tree. */ |
9618138b | 4205 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 4206 | } |
9618138b | 4207 | rcu_num_lvls = i + 1; |
679f9858 | 4208 | |
f885b7f2 | 4209 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 4210 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 4211 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
4212 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
4213 | } | |
f885b7f2 PM |
4214 | |
4215 | /* Calculate the total number of rcu_node structures. */ | |
4216 | rcu_num_nodes = 0; | |
679f9858 | 4217 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 4218 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
4219 | } |
4220 | ||
a3dc2948 PM |
4221 | /* |
4222 | * Dump out the structure of the rcu_node combining tree associated | |
4223 | * with the rcu_state structure referenced by rsp. | |
4224 | */ | |
4225 | static void __init rcu_dump_rcu_node_tree(struct rcu_state *rsp) | |
4226 | { | |
4227 | int level = 0; | |
4228 | struct rcu_node *rnp; | |
4229 | ||
4230 | pr_info("rcu_node tree layout dump\n"); | |
4231 | pr_info(" "); | |
4232 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
4233 | if (rnp->level != level) { | |
4234 | pr_cont("\n"); | |
4235 | pr_info(" "); | |
4236 | level = rnp->level; | |
4237 | } | |
4238 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
4239 | } | |
4240 | pr_cont("\n"); | |
4241 | } | |
4242 | ||
9f680ab4 | 4243 | void __init rcu_init(void) |
64db4cff | 4244 | { |
017c4261 | 4245 | int cpu; |
9f680ab4 | 4246 | |
47627678 PM |
4247 | rcu_early_boot_tests(); |
4248 | ||
f41d911f | 4249 | rcu_bootup_announce(); |
f885b7f2 | 4250 | rcu_init_geometry(); |
a87f203e PM |
4251 | rcu_init_one(&rcu_bh_state); |
4252 | rcu_init_one(&rcu_sched_state); | |
a3dc2948 PM |
4253 | if (dump_tree) |
4254 | rcu_dump_rcu_node_tree(&rcu_sched_state); | |
f41d911f | 4255 | __rcu_init_preempt(); |
b5b39360 | 4256 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
4257 | |
4258 | /* | |
4259 | * We don't need protection against CPU-hotplug here because | |
4260 | * this is called early in boot, before either interrupts | |
4261 | * or the scheduler are operational. | |
4262 | */ | |
d1d74d14 | 4263 | pm_notifier(rcu_pm_notify, 0); |
7ec99de3 | 4264 | for_each_online_cpu(cpu) { |
4df83742 | 4265 | rcutree_prepare_cpu(cpu); |
7ec99de3 | 4266 | rcu_cpu_starting(cpu); |
9b9500da | 4267 | rcutree_online_cpu(cpu); |
7ec99de3 | 4268 | } |
64db4cff PM |
4269 | } |
4270 | ||
3549c2bc | 4271 | #include "tree_exp.h" |
4102adab | 4272 | #include "tree_plugin.h" |