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> |
64db4cff | 60 | |
4102adab | 61 | #include "tree.h" |
29c00b4a | 62 | #include "rcu.h" |
9f77da9f | 63 | |
4102adab PM |
64 | #ifdef MODULE_PARAM_PREFIX |
65 | #undef MODULE_PARAM_PREFIX | |
66 | #endif | |
67 | #define MODULE_PARAM_PREFIX "rcutree." | |
68 | ||
64db4cff PM |
69 | /* Data structures. */ |
70 | ||
f7f7bac9 SRRH |
71 | /* |
72 | * In order to export the rcu_state name to the tracing tools, it | |
73 | * needs to be added in the __tracepoint_string section. | |
74 | * This requires defining a separate variable tp_<sname>_varname | |
75 | * that points to the string being used, and this will allow | |
76 | * the tracing userspace tools to be able to decipher the string | |
77 | * address to the matching string. | |
78 | */ | |
a8a29b3b AB |
79 | #ifdef CONFIG_TRACING |
80 | # define DEFINE_RCU_TPS(sname) \ | |
f7f7bac9 | 81 | static char sname##_varname[] = #sname; \ |
a8a29b3b AB |
82 | static const char *tp_##sname##_varname __used __tracepoint_string = sname##_varname; |
83 | # define RCU_STATE_NAME(sname) sname##_varname | |
84 | #else | |
85 | # define DEFINE_RCU_TPS(sname) | |
86 | # define RCU_STATE_NAME(sname) __stringify(sname) | |
87 | #endif | |
88 | ||
89 | #define RCU_STATE_INITIALIZER(sname, sabbr, cr) \ | |
90 | DEFINE_RCU_TPS(sname) \ | |
c92fb057 | 91 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, sname##_data); \ |
a41bfeb2 | 92 | struct rcu_state sname##_state = { \ |
6c90cc7b | 93 | .level = { &sname##_state.node[0] }, \ |
2723249a | 94 | .rda = &sname##_data, \ |
037b64ed | 95 | .call = cr, \ |
77f81fe0 | 96 | .gp_state = RCU_GP_IDLE, \ |
42c3533e PM |
97 | .gpnum = 0UL - 300UL, \ |
98 | .completed = 0UL - 300UL, \ | |
7b2e6011 | 99 | .orphan_lock = __RAW_SPIN_LOCK_UNLOCKED(&sname##_state.orphan_lock), \ |
6c90cc7b PM |
100 | .orphan_nxttail = &sname##_state.orphan_nxtlist, \ |
101 | .orphan_donetail = &sname##_state.orphan_donelist, \ | |
7be7f0be | 102 | .barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \ |
a8a29b3b | 103 | .name = RCU_STATE_NAME(sname), \ |
a4889858 | 104 | .abbr = sabbr, \ |
f6a12f34 | 105 | .exp_mutex = __MUTEX_INITIALIZER(sname##_state.exp_mutex), \ |
3b5f668e | 106 | .exp_wake_mutex = __MUTEX_INITIALIZER(sname##_state.exp_wake_mutex), \ |
2723249a | 107 | } |
64db4cff | 108 | |
a41bfeb2 SRRH |
109 | RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched); |
110 | RCU_STATE_INITIALIZER(rcu_bh, 'b', call_rcu_bh); | |
b1f77b05 | 111 | |
b28a7c01 | 112 | static struct rcu_state *const rcu_state_p; |
6ce75a23 | 113 | LIST_HEAD(rcu_struct_flavors); |
27f4d280 | 114 | |
a3dc2948 PM |
115 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
116 | static bool dump_tree; | |
117 | module_param(dump_tree, bool, 0444); | |
7fa27001 PM |
118 | /* Control rcu_node-tree auto-balancing at boot time. */ |
119 | static bool rcu_fanout_exact; | |
120 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
121 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
122 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 123 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 124 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 AG |
125 | /* Number of rcu_nodes at specified level. */ |
126 | static int num_rcu_lvl[] = NUM_RCU_LVL_INIT; | |
f885b7f2 | 127 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
088e9d25 DBO |
128 | /* panic() on RCU Stall sysctl. */ |
129 | int sysctl_panic_on_rcu_stall __read_mostly; | |
f885b7f2 | 130 | |
b0d30417 | 131 | /* |
52d7e48b PM |
132 | * The rcu_scheduler_active variable is initialized to the value |
133 | * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the | |
134 | * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE, | |
135 | * RCU can assume that there is but one task, allowing RCU to (for example) | |
0d95092c | 136 | * optimize synchronize_rcu() to a simple barrier(). When this variable |
52d7e48b PM |
137 | * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required |
138 | * to detect real grace periods. This variable is also used to suppress | |
139 | * boot-time false positives from lockdep-RCU error checking. Finally, it | |
140 | * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU | |
141 | * is fully initialized, including all of its kthreads having been spawned. | |
b0d30417 | 142 | */ |
bbad9379 PM |
143 | int rcu_scheduler_active __read_mostly; |
144 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
145 | ||
b0d30417 PM |
146 | /* |
147 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
148 | * during the early_initcall() processing, which is after the scheduler | |
149 | * is capable of creating new tasks. So RCU processing (for example, | |
150 | * creating tasks for RCU priority boosting) must be delayed until after | |
151 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
152 | * currently delay invocation of any RCU callbacks until after this point. | |
153 | * | |
154 | * It might later prove better for people registering RCU callbacks during | |
155 | * early boot to take responsibility for these callbacks, but one step at | |
156 | * a time. | |
157 | */ | |
158 | static int rcu_scheduler_fully_active __read_mostly; | |
159 | ||
0aa04b05 PM |
160 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
161 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 162 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 PM |
163 | static void invoke_rcu_core(void); |
164 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); | |
6587a23b PM |
165 | static void rcu_report_exp_rdp(struct rcu_state *rsp, |
166 | struct rcu_data *rdp, bool wake); | |
3549c2bc | 167 | static void sync_sched_exp_online_cleanup(int cpu); |
a26ac245 | 168 | |
a94844b2 | 169 | /* rcuc/rcub kthread realtime priority */ |
26730f55 | 170 | #ifdef CONFIG_RCU_KTHREAD_PRIO |
a94844b2 | 171 | static int kthread_prio = CONFIG_RCU_KTHREAD_PRIO; |
26730f55 PM |
172 | #else /* #ifdef CONFIG_RCU_KTHREAD_PRIO */ |
173 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; | |
174 | #endif /* #else #ifdef CONFIG_RCU_KTHREAD_PRIO */ | |
a94844b2 PM |
175 | module_param(kthread_prio, int, 0644); |
176 | ||
8d7dc928 | 177 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd PM |
178 | |
179 | #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT | |
180 | static int gp_preinit_delay = CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT_DELAY; | |
181 | module_param(gp_preinit_delay, int, 0644); | |
182 | #else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT */ | |
183 | static const int gp_preinit_delay; | |
184 | #endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT */ | |
185 | ||
8d7dc928 PM |
186 | #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT |
187 | static int gp_init_delay = CONFIG_RCU_TORTURE_TEST_SLOW_INIT_DELAY; | |
37745d28 | 188 | module_param(gp_init_delay, int, 0644); |
8d7dc928 PM |
189 | #else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT */ |
190 | static const int gp_init_delay; | |
191 | #endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT */ | |
eab128e8 | 192 | |
0f41c0dd PM |
193 | #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP |
194 | static int gp_cleanup_delay = CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP_DELAY; | |
195 | module_param(gp_cleanup_delay, int, 0644); | |
196 | #else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP */ | |
197 | static const int gp_cleanup_delay; | |
198 | #endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP */ | |
199 | ||
eab128e8 PM |
200 | /* |
201 | * Number of grace periods between delays, normalized by the duration of | |
202 | * the delay. The longer the the delay, the more the grace periods between | |
203 | * each delay. The reason for this normalization is that it means that, | |
204 | * for non-zero delays, the overall slowdown of grace periods is constant | |
205 | * regardless of the duration of the delay. This arrangement balances | |
206 | * the need for long delays to increase some race probabilities with the | |
207 | * need for fast grace periods to increase other race probabilities. | |
208 | */ | |
209 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */ | |
37745d28 | 210 | |
4a298656 PM |
211 | /* |
212 | * Track the rcutorture test sequence number and the update version | |
213 | * number within a given test. The rcutorture_testseq is incremented | |
214 | * on every rcutorture module load and unload, so has an odd value | |
215 | * when a test is running. The rcutorture_vernum is set to zero | |
216 | * when rcutorture starts and is incremented on each rcutorture update. | |
217 | * These variables enable correlating rcutorture output with the | |
218 | * RCU tracing information. | |
219 | */ | |
220 | unsigned long rcutorture_testseq; | |
221 | unsigned long rcutorture_vernum; | |
222 | ||
0aa04b05 PM |
223 | /* |
224 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
225 | * This will not be stable unless the rcu_node structure's ->lock is | |
226 | * held, but the bit corresponding to the current CPU will be stable | |
227 | * in most contexts. | |
228 | */ | |
229 | unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) | |
230 | { | |
7d0ae808 | 231 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
232 | } |
233 | ||
fc2219d4 | 234 | /* |
7d0ae808 | 235 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
236 | * permit this function to be invoked without holding the root rcu_node |
237 | * structure's ->lock, but of course results can be subject to change. | |
238 | */ | |
239 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
240 | { | |
7d0ae808 | 241 | return READ_ONCE(rsp->completed) != READ_ONCE(rsp->gpnum); |
fc2219d4 PM |
242 | } |
243 | ||
b1f77b05 | 244 | /* |
d6714c22 | 245 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 246 | * how many quiescent states passed, just if there was at least |
d6714c22 | 247 | * one since the start of the grace period, this just sets a flag. |
e4cc1f22 | 248 | * The caller must have disabled preemption. |
b1f77b05 | 249 | */ |
284a8c93 | 250 | void rcu_sched_qs(void) |
b1f77b05 | 251 | { |
fecbf6f0 PM |
252 | if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.s)) |
253 | return; | |
254 | trace_rcu_grace_period(TPS("rcu_sched"), | |
255 | __this_cpu_read(rcu_sched_data.gpnum), | |
256 | TPS("cpuqs")); | |
257 | __this_cpu_write(rcu_sched_data.cpu_no_qs.b.norm, false); | |
258 | if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.b.exp)) | |
259 | return; | |
46a5d164 PM |
260 | __this_cpu_write(rcu_sched_data.cpu_no_qs.b.exp, false); |
261 | rcu_report_exp_rdp(&rcu_sched_state, | |
262 | this_cpu_ptr(&rcu_sched_data), true); | |
b1f77b05 IM |
263 | } |
264 | ||
284a8c93 | 265 | void rcu_bh_qs(void) |
b1f77b05 | 266 | { |
5b74c458 | 267 | if (__this_cpu_read(rcu_bh_data.cpu_no_qs.s)) { |
284a8c93 PM |
268 | trace_rcu_grace_period(TPS("rcu_bh"), |
269 | __this_cpu_read(rcu_bh_data.gpnum), | |
270 | TPS("cpuqs")); | |
5b74c458 | 271 | __this_cpu_write(rcu_bh_data.cpu_no_qs.b.norm, false); |
284a8c93 | 272 | } |
b1f77b05 | 273 | } |
64db4cff | 274 | |
b8c17e66 PM |
275 | /* |
276 | * Steal a bit from the bottom of ->dynticks for idle entry/exit | |
277 | * control. Initially this is for TLB flushing. | |
278 | */ | |
279 | #define RCU_DYNTICK_CTRL_MASK 0x1 | |
280 | #define RCU_DYNTICK_CTRL_CTR (RCU_DYNTICK_CTRL_MASK + 1) | |
281 | #ifndef rcu_eqs_special_exit | |
282 | #define rcu_eqs_special_exit() do { } while (0) | |
283 | #endif | |
284 | ||
4a81e832 PM |
285 | static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
286 | .dynticks_nesting = DYNTICK_TASK_EXIT_IDLE, | |
b8c17e66 | 287 | .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR), |
4a81e832 PM |
288 | #ifdef CONFIG_NO_HZ_FULL_SYSIDLE |
289 | .dynticks_idle_nesting = DYNTICK_TASK_NEST_VALUE, | |
290 | .dynticks_idle = ATOMIC_INIT(1), | |
291 | #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ | |
292 | }; | |
293 | ||
2625d469 PM |
294 | /* |
295 | * Record entry into an extended quiescent state. This is only to be | |
296 | * called when not already in an extended quiescent state. | |
297 | */ | |
298 | static void rcu_dynticks_eqs_enter(void) | |
299 | { | |
300 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 301 | int seq; |
2625d469 PM |
302 | |
303 | /* | |
b8c17e66 | 304 | * CPUs seeing atomic_add_return() must see prior RCU read-side |
2625d469 PM |
305 | * critical sections, and we also must force ordering with the |
306 | * next idle sojourn. | |
307 | */ | |
b8c17e66 PM |
308 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
309 | /* Better be in an extended quiescent state! */ | |
310 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
311 | (seq & RCU_DYNTICK_CTRL_CTR)); | |
312 | /* Better not have special action (TLB flush) pending! */ | |
313 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
314 | (seq & RCU_DYNTICK_CTRL_MASK)); | |
2625d469 PM |
315 | } |
316 | ||
317 | /* | |
318 | * Record exit from an extended quiescent state. This is only to be | |
319 | * called from an extended quiescent state. | |
320 | */ | |
321 | static void rcu_dynticks_eqs_exit(void) | |
322 | { | |
323 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 324 | int seq; |
2625d469 PM |
325 | |
326 | /* | |
b8c17e66 | 327 | * CPUs seeing atomic_add_return() must see prior idle sojourns, |
2625d469 PM |
328 | * and we also must force ordering with the next RCU read-side |
329 | * critical section. | |
330 | */ | |
b8c17e66 PM |
331 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
332 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
333 | !(seq & RCU_DYNTICK_CTRL_CTR)); | |
334 | if (seq & RCU_DYNTICK_CTRL_MASK) { | |
335 | atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdtp->dynticks); | |
336 | smp_mb__after_atomic(); /* _exit after clearing mask. */ | |
337 | /* Prefer duplicate flushes to losing a flush. */ | |
338 | rcu_eqs_special_exit(); | |
339 | } | |
2625d469 PM |
340 | } |
341 | ||
342 | /* | |
343 | * Reset the current CPU's ->dynticks counter to indicate that the | |
344 | * newly onlined CPU is no longer in an extended quiescent state. | |
345 | * This will either leave the counter unchanged, or increment it | |
346 | * to the next non-quiescent value. | |
347 | * | |
348 | * The non-atomic test/increment sequence works because the upper bits | |
349 | * of the ->dynticks counter are manipulated only by the corresponding CPU, | |
350 | * or when the corresponding CPU is offline. | |
351 | */ | |
352 | static void rcu_dynticks_eqs_online(void) | |
353 | { | |
354 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
355 | ||
b8c17e66 | 356 | if (atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR) |
2625d469 | 357 | return; |
b8c17e66 | 358 | atomic_add(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
2625d469 PM |
359 | } |
360 | ||
02a5c550 PM |
361 | /* |
362 | * Is the current CPU in an extended quiescent state? | |
363 | * | |
364 | * No ordering, as we are sampling CPU-local information. | |
365 | */ | |
366 | bool rcu_dynticks_curr_cpu_in_eqs(void) | |
367 | { | |
368 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
369 | ||
b8c17e66 | 370 | return !(atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
371 | } |
372 | ||
8b2f63ab PM |
373 | /* |
374 | * Snapshot the ->dynticks counter with full ordering so as to allow | |
375 | * stable comparison of this counter with past and future snapshots. | |
376 | */ | |
02a5c550 | 377 | int rcu_dynticks_snap(struct rcu_dynticks *rdtp) |
8b2f63ab PM |
378 | { |
379 | int snap = atomic_add_return(0, &rdtp->dynticks); | |
380 | ||
b8c17e66 | 381 | return snap & ~RCU_DYNTICK_CTRL_MASK; |
8b2f63ab PM |
382 | } |
383 | ||
02a5c550 PM |
384 | /* |
385 | * Return true if the snapshot returned from rcu_dynticks_snap() | |
386 | * indicates that RCU is in an extended quiescent state. | |
387 | */ | |
388 | static bool rcu_dynticks_in_eqs(int snap) | |
389 | { | |
b8c17e66 | 390 | return !(snap & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
391 | } |
392 | ||
393 | /* | |
394 | * Return true if the CPU corresponding to the specified rcu_dynticks | |
395 | * structure has spent some time in an extended quiescent state since | |
396 | * rcu_dynticks_snap() returned the specified snapshot. | |
397 | */ | |
398 | static bool rcu_dynticks_in_eqs_since(struct rcu_dynticks *rdtp, int snap) | |
399 | { | |
400 | return snap != rcu_dynticks_snap(rdtp); | |
401 | } | |
402 | ||
6563de9d PM |
403 | /* |
404 | * Do a double-increment of the ->dynticks counter to emulate a | |
405 | * momentary idle-CPU quiescent state. | |
406 | */ | |
407 | static void rcu_dynticks_momentary_idle(void) | |
408 | { | |
409 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 PM |
410 | int special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR, |
411 | &rdtp->dynticks); | |
6563de9d PM |
412 | |
413 | /* It is illegal to call this from idle state. */ | |
b8c17e66 PM |
414 | WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR)); |
415 | } | |
416 | ||
417 | /* | |
418 | * Set the special (bottom) bit of the specified CPU so that it | |
419 | * will take special action (such as flushing its TLB) on the | |
420 | * next exit from an extended quiescent state. Returns true if | |
421 | * the bit was successfully set, or false if the CPU was not in | |
422 | * an extended quiescent state. | |
423 | */ | |
424 | bool rcu_eqs_special_set(int cpu) | |
425 | { | |
426 | int old; | |
427 | int new; | |
428 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); | |
429 | ||
430 | do { | |
431 | old = atomic_read(&rdtp->dynticks); | |
432 | if (old & RCU_DYNTICK_CTRL_CTR) | |
433 | return false; | |
434 | new = old | RCU_DYNTICK_CTRL_MASK; | |
435 | } while (atomic_cmpxchg(&rdtp->dynticks, old, new) != old); | |
436 | return true; | |
6563de9d PM |
437 | } |
438 | ||
4a81e832 PM |
439 | /* |
440 | * Let the RCU core know that this CPU has gone through the scheduler, | |
441 | * which is a quiescent state. This is called when the need for a | |
442 | * quiescent state is urgent, so we burn an atomic operation and full | |
443 | * memory barriers to let the RCU core know about it, regardless of what | |
444 | * this CPU might (or might not) do in the near future. | |
445 | * | |
0f9be8ca | 446 | * We inform the RCU core by emulating a zero-duration dyntick-idle period. |
46a5d164 PM |
447 | * |
448 | * The caller must have disabled interrupts. | |
4a81e832 PM |
449 | */ |
450 | static void rcu_momentary_dyntick_idle(void) | |
451 | { | |
0f9be8ca PM |
452 | raw_cpu_write(rcu_dynticks.rcu_need_heavy_qs, false); |
453 | rcu_dynticks_momentary_idle(); | |
4a81e832 PM |
454 | } |
455 | ||
25502a6c PM |
456 | /* |
457 | * Note a context switch. This is a quiescent state for RCU-sched, | |
458 | * and requires special handling for preemptible RCU. | |
46a5d164 | 459 | * The caller must have disabled interrupts. |
25502a6c | 460 | */ |
38200cf2 | 461 | void rcu_note_context_switch(void) |
25502a6c | 462 | { |
bb73c52b | 463 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
f7f7bac9 | 464 | trace_rcu_utilization(TPS("Start context switch")); |
284a8c93 | 465 | rcu_sched_qs(); |
38200cf2 | 466 | rcu_preempt_note_context_switch(); |
0f9be8ca | 467 | if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) |
4a81e832 | 468 | rcu_momentary_dyntick_idle(); |
f7f7bac9 | 469 | trace_rcu_utilization(TPS("End context switch")); |
bb73c52b | 470 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
25502a6c | 471 | } |
29ce8310 | 472 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
25502a6c | 473 | |
5cd37193 | 474 | /* |
1925d196 | 475 | * Register a quiescent state for all RCU flavors. If there is an |
5cd37193 PM |
476 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight |
477 | * dyntick-idle quiescent state visible to other CPUs (but only for those | |
1925d196 | 478 | * RCU flavors in desperate need of a quiescent state, which will normally |
5cd37193 PM |
479 | * be none of them). Either way, do a lightweight quiescent state for |
480 | * all RCU flavors. | |
bb73c52b BF |
481 | * |
482 | * The barrier() calls are redundant in the common case when this is | |
483 | * called externally, but just in case this is called from within this | |
484 | * file. | |
485 | * | |
5cd37193 PM |
486 | */ |
487 | void rcu_all_qs(void) | |
488 | { | |
46a5d164 PM |
489 | unsigned long flags; |
490 | ||
bb73c52b | 491 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
0f9be8ca | 492 | if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) { |
46a5d164 | 493 | local_irq_save(flags); |
5cd37193 | 494 | rcu_momentary_dyntick_idle(); |
46a5d164 PM |
495 | local_irq_restore(flags); |
496 | } | |
a1e12248 PM |
497 | if (unlikely(raw_cpu_read(rcu_sched_data.cpu_no_qs.b.exp))) { |
498 | /* | |
499 | * Yes, we just checked a per-CPU variable with preemption | |
500 | * enabled, so we might be migrated to some other CPU at | |
501 | * this point. That is OK because in that case, the | |
502 | * migration will supply the needed quiescent state. | |
503 | * We might end up needlessly disabling preemption and | |
504 | * invoking rcu_sched_qs() on the destination CPU, but | |
505 | * the probability and cost are both quite low, so this | |
506 | * should not be a problem in practice. | |
507 | */ | |
508 | preempt_disable(); | |
509 | rcu_sched_qs(); | |
510 | preempt_enable(); | |
511 | } | |
9577df9a | 512 | this_cpu_inc(rcu_dynticks.rcu_qs_ctr); |
bb73c52b | 513 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
5cd37193 PM |
514 | } |
515 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
516 | ||
878d7439 ED |
517 | static long blimit = 10; /* Maximum callbacks per rcu_do_batch. */ |
518 | static long qhimark = 10000; /* If this many pending, ignore blimit. */ | |
519 | static long qlowmark = 100; /* Once only this many pending, use blimit. */ | |
64db4cff | 520 | |
878d7439 ED |
521 | module_param(blimit, long, 0444); |
522 | module_param(qhimark, long, 0444); | |
523 | module_param(qlowmark, long, 0444); | |
3d76c082 | 524 | |
026ad283 PM |
525 | static ulong jiffies_till_first_fqs = ULONG_MAX; |
526 | static ulong jiffies_till_next_fqs = ULONG_MAX; | |
8c7c4829 | 527 | static bool rcu_kick_kthreads; |
d40011f6 PM |
528 | |
529 | module_param(jiffies_till_first_fqs, ulong, 0644); | |
530 | module_param(jiffies_till_next_fqs, ulong, 0644); | |
8c7c4829 | 531 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 532 | |
4a81e832 PM |
533 | /* |
534 | * How long the grace period must be before we start recruiting | |
535 | * quiescent-state help from rcu_note_context_switch(). | |
536 | */ | |
537 | static ulong jiffies_till_sched_qs = HZ / 20; | |
538 | module_param(jiffies_till_sched_qs, ulong, 0644); | |
539 | ||
48a7639c | 540 | static bool rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp, |
910ee45d | 541 | struct rcu_data *rdp); |
217af2a2 PM |
542 | static void force_qs_rnp(struct rcu_state *rsp, |
543 | int (*f)(struct rcu_data *rsp, bool *isidle, | |
544 | unsigned long *maxj), | |
545 | bool *isidle, unsigned long *maxj); | |
4cdfc175 | 546 | static void force_quiescent_state(struct rcu_state *rsp); |
e3950ecd | 547 | static int rcu_pending(void); |
64db4cff PM |
548 | |
549 | /* | |
917963d0 | 550 | * Return the number of RCU batches started thus far for debug & stats. |
64db4cff | 551 | */ |
917963d0 PM |
552 | unsigned long rcu_batches_started(void) |
553 | { | |
554 | return rcu_state_p->gpnum; | |
555 | } | |
556 | EXPORT_SYMBOL_GPL(rcu_batches_started); | |
557 | ||
558 | /* | |
559 | * Return the number of RCU-sched batches started thus far for debug & stats. | |
64db4cff | 560 | */ |
917963d0 PM |
561 | unsigned long rcu_batches_started_sched(void) |
562 | { | |
563 | return rcu_sched_state.gpnum; | |
564 | } | |
565 | EXPORT_SYMBOL_GPL(rcu_batches_started_sched); | |
566 | ||
567 | /* | |
568 | * Return the number of RCU BH batches started thus far for debug & stats. | |
569 | */ | |
570 | unsigned long rcu_batches_started_bh(void) | |
571 | { | |
572 | return rcu_bh_state.gpnum; | |
573 | } | |
574 | EXPORT_SYMBOL_GPL(rcu_batches_started_bh); | |
575 | ||
576 | /* | |
577 | * Return the number of RCU batches completed thus far for debug & stats. | |
578 | */ | |
579 | unsigned long rcu_batches_completed(void) | |
580 | { | |
581 | return rcu_state_p->completed; | |
582 | } | |
583 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
584 | ||
585 | /* | |
586 | * Return the number of RCU-sched batches completed thus far for debug & stats. | |
64db4cff | 587 | */ |
9733e4f0 | 588 | unsigned long rcu_batches_completed_sched(void) |
64db4cff | 589 | { |
d6714c22 | 590 | return rcu_sched_state.completed; |
64db4cff | 591 | } |
d6714c22 | 592 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
593 | |
594 | /* | |
917963d0 | 595 | * Return the number of RCU BH batches completed thus far for debug & stats. |
64db4cff | 596 | */ |
9733e4f0 | 597 | unsigned long rcu_batches_completed_bh(void) |
64db4cff PM |
598 | { |
599 | return rcu_bh_state.completed; | |
600 | } | |
601 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
602 | ||
291783b8 PM |
603 | /* |
604 | * Return the number of RCU expedited batches completed thus far for | |
605 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
606 | * numbers mean idle. The value returned will thus be roughly double | |
607 | * the cumulative batches since boot. | |
608 | */ | |
609 | unsigned long rcu_exp_batches_completed(void) | |
610 | { | |
611 | return rcu_state_p->expedited_sequence; | |
612 | } | |
613 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
614 | ||
615 | /* | |
616 | * Return the number of RCU-sched expedited batches completed thus far | |
617 | * for debug & stats. Similar to rcu_exp_batches_completed(). | |
618 | */ | |
619 | unsigned long rcu_exp_batches_completed_sched(void) | |
620 | { | |
621 | return rcu_sched_state.expedited_sequence; | |
622 | } | |
623 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed_sched); | |
624 | ||
a381d757 ACB |
625 | /* |
626 | * Force a quiescent state. | |
627 | */ | |
628 | void rcu_force_quiescent_state(void) | |
629 | { | |
e534165b | 630 | force_quiescent_state(rcu_state_p); |
a381d757 ACB |
631 | } |
632 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
633 | ||
bf66f18e PM |
634 | /* |
635 | * Force a quiescent state for RCU BH. | |
636 | */ | |
637 | void rcu_bh_force_quiescent_state(void) | |
638 | { | |
4cdfc175 | 639 | force_quiescent_state(&rcu_bh_state); |
bf66f18e PM |
640 | } |
641 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
642 | ||
e7580f33 PM |
643 | /* |
644 | * Force a quiescent state for RCU-sched. | |
645 | */ | |
646 | void rcu_sched_force_quiescent_state(void) | |
647 | { | |
648 | force_quiescent_state(&rcu_sched_state); | |
649 | } | |
650 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
651 | ||
afea227f PM |
652 | /* |
653 | * Show the state of the grace-period kthreads. | |
654 | */ | |
655 | void show_rcu_gp_kthreads(void) | |
656 | { | |
657 | struct rcu_state *rsp; | |
658 | ||
659 | for_each_rcu_flavor(rsp) { | |
660 | pr_info("%s: wait state: %d ->state: %#lx\n", | |
661 | rsp->name, rsp->gp_state, rsp->gp_kthread->state); | |
662 | /* sched_show_task(rsp->gp_kthread); */ | |
663 | } | |
664 | } | |
665 | EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads); | |
666 | ||
4a298656 PM |
667 | /* |
668 | * Record the number of times rcutorture tests have been initiated and | |
669 | * terminated. This information allows the debugfs tracing stats to be | |
670 | * correlated to the rcutorture messages, even when the rcutorture module | |
671 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
672 | * store this state in rcutorture itself. | |
673 | */ | |
674 | void rcutorture_record_test_transition(void) | |
675 | { | |
676 | rcutorture_testseq++; | |
677 | rcutorture_vernum = 0; | |
678 | } | |
679 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
680 | ||
ad0dc7f9 PM |
681 | /* |
682 | * Send along grace-period-related data for rcutorture diagnostics. | |
683 | */ | |
684 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
685 | unsigned long *gpnum, unsigned long *completed) | |
686 | { | |
687 | struct rcu_state *rsp = NULL; | |
688 | ||
689 | switch (test_type) { | |
690 | case RCU_FLAVOR: | |
e534165b | 691 | rsp = rcu_state_p; |
ad0dc7f9 PM |
692 | break; |
693 | case RCU_BH_FLAVOR: | |
694 | rsp = &rcu_bh_state; | |
695 | break; | |
696 | case RCU_SCHED_FLAVOR: | |
697 | rsp = &rcu_sched_state; | |
698 | break; | |
699 | default: | |
700 | break; | |
701 | } | |
702 | if (rsp != NULL) { | |
7d0ae808 PM |
703 | *flags = READ_ONCE(rsp->gp_flags); |
704 | *gpnum = READ_ONCE(rsp->gpnum); | |
705 | *completed = READ_ONCE(rsp->completed); | |
ad0dc7f9 PM |
706 | return; |
707 | } | |
708 | *flags = 0; | |
709 | *gpnum = 0; | |
710 | *completed = 0; | |
711 | } | |
712 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
713 | ||
4a298656 PM |
714 | /* |
715 | * Record the number of writer passes through the current rcutorture test. | |
716 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
717 | * messages. | |
718 | */ | |
719 | void rcutorture_record_progress(unsigned long vernum) | |
720 | { | |
721 | rcutorture_vernum++; | |
722 | } | |
723 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
724 | ||
64db4cff PM |
725 | /* |
726 | * Does the CPU have callbacks ready to be invoked? | |
727 | */ | |
728 | static int | |
729 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
730 | { | |
3fbfbf7a | 731 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL] && |
630c7ed9 | 732 | rdp->nxttail[RCU_NEXT_TAIL] != NULL; |
64db4cff PM |
733 | } |
734 | ||
365187fb PM |
735 | /* |
736 | * Return the root node of the specified rcu_state structure. | |
737 | */ | |
738 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
739 | { | |
740 | return &rsp->node[0]; | |
741 | } | |
742 | ||
743 | /* | |
744 | * Is there any need for future grace periods? | |
745 | * Interrupts must be disabled. If the caller does not hold the root | |
746 | * rnp_node structure's ->lock, the results are advisory only. | |
747 | */ | |
748 | static int rcu_future_needs_gp(struct rcu_state *rsp) | |
749 | { | |
750 | struct rcu_node *rnp = rcu_get_root(rsp); | |
7d0ae808 | 751 | int idx = (READ_ONCE(rnp->completed) + 1) & 0x1; |
365187fb PM |
752 | int *fp = &rnp->need_future_gp[idx]; |
753 | ||
7d0ae808 | 754 | return READ_ONCE(*fp); |
365187fb PM |
755 | } |
756 | ||
64db4cff | 757 | /* |
dc35c893 PM |
758 | * Does the current CPU require a not-yet-started grace period? |
759 | * The caller must have disabled interrupts to prevent races with | |
760 | * normal callback registry. | |
64db4cff | 761 | */ |
d117c8aa | 762 | static bool |
64db4cff PM |
763 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) |
764 | { | |
dc35c893 | 765 | int i; |
3fbfbf7a | 766 | |
dc35c893 | 767 | if (rcu_gp_in_progress(rsp)) |
d117c8aa | 768 | return false; /* No, a grace period is already in progress. */ |
365187fb | 769 | if (rcu_future_needs_gp(rsp)) |
d117c8aa | 770 | return true; /* Yes, a no-CBs CPU needs one. */ |
dc35c893 | 771 | if (!rdp->nxttail[RCU_NEXT_TAIL]) |
d117c8aa | 772 | return false; /* No, this is a no-CBs (or offline) CPU. */ |
dc35c893 | 773 | if (*rdp->nxttail[RCU_NEXT_READY_TAIL]) |
d117c8aa | 774 | return true; /* Yes, CPU has newly registered callbacks. */ |
dc35c893 PM |
775 | for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) |
776 | if (rdp->nxttail[i - 1] != rdp->nxttail[i] && | |
7d0ae808 | 777 | ULONG_CMP_LT(READ_ONCE(rsp->completed), |
dc35c893 | 778 | rdp->nxtcompleted[i])) |
d117c8aa PM |
779 | return true; /* Yes, CBs for future grace period. */ |
780 | return false; /* No grace period needed. */ | |
64db4cff PM |
781 | } |
782 | ||
9b2e4f18 | 783 | /* |
adf5091e | 784 | * rcu_eqs_enter_common - current CPU is moving towards extended quiescent state |
9b2e4f18 PM |
785 | * |
786 | * If the new value of the ->dynticks_nesting counter now is zero, | |
787 | * we really have entered idle, and must do the appropriate accounting. | |
788 | * The caller must have disabled interrupts. | |
789 | */ | |
28ced795 | 790 | static void rcu_eqs_enter_common(long long oldval, bool user) |
9b2e4f18 | 791 | { |
96d3fd0d PM |
792 | struct rcu_state *rsp; |
793 | struct rcu_data *rdp; | |
2625d469 | 794 | RCU_TRACE(struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);) |
96d3fd0d | 795 | |
f7f7bac9 | 796 | trace_rcu_dyntick(TPS("Start"), oldval, rdtp->dynticks_nesting); |
1ce46ee5 PM |
797 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
798 | !user && !is_idle_task(current)) { | |
289828e6 PM |
799 | struct task_struct *idle __maybe_unused = |
800 | idle_task(smp_processor_id()); | |
0989cb46 | 801 | |
f7f7bac9 | 802 | trace_rcu_dyntick(TPS("Error on entry: not idle task"), oldval, 0); |
274529ba | 803 | rcu_ftrace_dump(DUMP_ORIG); |
0989cb46 PM |
804 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
805 | current->pid, current->comm, | |
806 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 | 807 | } |
96d3fd0d PM |
808 | for_each_rcu_flavor(rsp) { |
809 | rdp = this_cpu_ptr(rsp->rda); | |
810 | do_nocb_deferred_wakeup(rdp); | |
811 | } | |
198bbf81 | 812 | rcu_prepare_for_idle(); |
2625d469 | 813 | rcu_dynticks_eqs_enter(); |
176f8f7a | 814 | rcu_dynticks_task_enter(); |
c44e2cdd PM |
815 | |
816 | /* | |
adf5091e | 817 | * It is illegal to enter an extended quiescent state while |
c44e2cdd PM |
818 | * in an RCU read-side critical section. |
819 | */ | |
f78f5b90 PM |
820 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), |
821 | "Illegal idle entry in RCU read-side critical section."); | |
822 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), | |
823 | "Illegal idle entry in RCU-bh read-side critical section."); | |
824 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), | |
825 | "Illegal idle entry in RCU-sched read-side critical section."); | |
9b2e4f18 | 826 | } |
64db4cff | 827 | |
adf5091e FW |
828 | /* |
829 | * Enter an RCU extended quiescent state, which can be either the | |
830 | * idle loop or adaptive-tickless usermode execution. | |
64db4cff | 831 | */ |
adf5091e | 832 | static void rcu_eqs_enter(bool user) |
64db4cff | 833 | { |
4145fa7f | 834 | long long oldval; |
64db4cff PM |
835 | struct rcu_dynticks *rdtp; |
836 | ||
c9d4b0af | 837 | rdtp = this_cpu_ptr(&rcu_dynticks); |
4145fa7f | 838 | oldval = rdtp->dynticks_nesting; |
1ce46ee5 PM |
839 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
840 | (oldval & DYNTICK_TASK_NEST_MASK) == 0); | |
3a592405 | 841 | if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) { |
29e37d81 | 842 | rdtp->dynticks_nesting = 0; |
28ced795 | 843 | rcu_eqs_enter_common(oldval, user); |
3a592405 | 844 | } else { |
29e37d81 | 845 | rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE; |
3a592405 | 846 | } |
64db4cff | 847 | } |
adf5091e FW |
848 | |
849 | /** | |
850 | * rcu_idle_enter - inform RCU that current CPU is entering idle | |
851 | * | |
852 | * Enter idle mode, in other words, -leave- the mode in which RCU | |
853 | * read-side critical sections can occur. (Though RCU read-side | |
854 | * critical sections can occur in irq handlers in idle, a possibility | |
855 | * handled by irq_enter() and irq_exit().) | |
856 | * | |
857 | * We crowbar the ->dynticks_nesting field to zero to allow for | |
858 | * the possibility of usermode upcalls having messed up our count | |
859 | * of interrupt nesting level during the prior busy period. | |
860 | */ | |
861 | void rcu_idle_enter(void) | |
862 | { | |
c5d900bf FW |
863 | unsigned long flags; |
864 | ||
865 | local_irq_save(flags); | |
cb349ca9 | 866 | rcu_eqs_enter(false); |
28ced795 | 867 | rcu_sysidle_enter(0); |
c5d900bf | 868 | local_irq_restore(flags); |
adf5091e | 869 | } |
8a2ecf47 | 870 | EXPORT_SYMBOL_GPL(rcu_idle_enter); |
64db4cff | 871 | |
d1ec4c34 | 872 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
873 | /** |
874 | * rcu_user_enter - inform RCU that we are resuming userspace. | |
875 | * | |
876 | * Enter RCU idle mode right before resuming userspace. No use of RCU | |
877 | * is permitted between this call and rcu_user_exit(). This way the | |
878 | * CPU doesn't need to maintain the tick for RCU maintenance purposes | |
879 | * when the CPU runs in userspace. | |
880 | */ | |
881 | void rcu_user_enter(void) | |
882 | { | |
91d1aa43 | 883 | rcu_eqs_enter(1); |
adf5091e | 884 | } |
d1ec4c34 | 885 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 886 | |
9b2e4f18 PM |
887 | /** |
888 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
889 | * | |
890 | * Exit from an interrupt handler, which might possibly result in entering | |
891 | * idle mode, in other words, leaving the mode in which read-side critical | |
7c9906ca | 892 | * sections can occur. The caller must have disabled interrupts. |
64db4cff | 893 | * |
9b2e4f18 PM |
894 | * This code assumes that the idle loop never does anything that might |
895 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
896 | * architecture violates this assumption, RCU will give you what you | |
897 | * deserve, good and hard. But very infrequently and irreproducibly. | |
898 | * | |
899 | * Use things like work queues to work around this limitation. | |
900 | * | |
901 | * You have been warned. | |
64db4cff | 902 | */ |
9b2e4f18 | 903 | void rcu_irq_exit(void) |
64db4cff | 904 | { |
4145fa7f | 905 | long long oldval; |
64db4cff PM |
906 | struct rcu_dynticks *rdtp; |
907 | ||
7c9906ca | 908 | RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_irq_exit() invoked with irqs enabled!!!"); |
c9d4b0af | 909 | rdtp = this_cpu_ptr(&rcu_dynticks); |
4145fa7f | 910 | oldval = rdtp->dynticks_nesting; |
9b2e4f18 | 911 | rdtp->dynticks_nesting--; |
1ce46ee5 PM |
912 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
913 | rdtp->dynticks_nesting < 0); | |
b6fc6020 | 914 | if (rdtp->dynticks_nesting) |
f7f7bac9 | 915 | trace_rcu_dyntick(TPS("--="), oldval, rdtp->dynticks_nesting); |
b6fc6020 | 916 | else |
28ced795 CL |
917 | rcu_eqs_enter_common(oldval, true); |
918 | rcu_sysidle_enter(1); | |
7c9906ca PM |
919 | } |
920 | ||
921 | /* | |
922 | * Wrapper for rcu_irq_exit() where interrupts are enabled. | |
923 | */ | |
924 | void rcu_irq_exit_irqson(void) | |
925 | { | |
926 | unsigned long flags; | |
927 | ||
928 | local_irq_save(flags); | |
929 | rcu_irq_exit(); | |
9b2e4f18 PM |
930 | local_irq_restore(flags); |
931 | } | |
932 | ||
933 | /* | |
adf5091e | 934 | * rcu_eqs_exit_common - current CPU moving away from extended quiescent state |
9b2e4f18 PM |
935 | * |
936 | * If the new value of the ->dynticks_nesting counter was previously zero, | |
937 | * we really have exited idle, and must do the appropriate accounting. | |
938 | * The caller must have disabled interrupts. | |
939 | */ | |
28ced795 | 940 | static void rcu_eqs_exit_common(long long oldval, int user) |
9b2e4f18 | 941 | { |
2625d469 | 942 | RCU_TRACE(struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);) |
28ced795 | 943 | |
176f8f7a | 944 | rcu_dynticks_task_exit(); |
2625d469 | 945 | rcu_dynticks_eqs_exit(); |
8fa7845d | 946 | rcu_cleanup_after_idle(); |
f7f7bac9 | 947 | trace_rcu_dyntick(TPS("End"), oldval, rdtp->dynticks_nesting); |
1ce46ee5 PM |
948 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
949 | !user && !is_idle_task(current)) { | |
289828e6 PM |
950 | struct task_struct *idle __maybe_unused = |
951 | idle_task(smp_processor_id()); | |
0989cb46 | 952 | |
f7f7bac9 | 953 | trace_rcu_dyntick(TPS("Error on exit: not idle task"), |
4145fa7f | 954 | oldval, rdtp->dynticks_nesting); |
274529ba | 955 | rcu_ftrace_dump(DUMP_ORIG); |
0989cb46 PM |
956 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
957 | current->pid, current->comm, | |
958 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 PM |
959 | } |
960 | } | |
961 | ||
adf5091e FW |
962 | /* |
963 | * Exit an RCU extended quiescent state, which can be either the | |
964 | * idle loop or adaptive-tickless usermode execution. | |
9b2e4f18 | 965 | */ |
adf5091e | 966 | static void rcu_eqs_exit(bool user) |
9b2e4f18 | 967 | { |
9b2e4f18 PM |
968 | struct rcu_dynticks *rdtp; |
969 | long long oldval; | |
970 | ||
c9d4b0af | 971 | rdtp = this_cpu_ptr(&rcu_dynticks); |
9b2e4f18 | 972 | oldval = rdtp->dynticks_nesting; |
1ce46ee5 | 973 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0); |
3a592405 | 974 | if (oldval & DYNTICK_TASK_NEST_MASK) { |
29e37d81 | 975 | rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE; |
3a592405 | 976 | } else { |
29e37d81 | 977 | rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE; |
28ced795 | 978 | rcu_eqs_exit_common(oldval, user); |
3a592405 | 979 | } |
9b2e4f18 | 980 | } |
adf5091e FW |
981 | |
982 | /** | |
983 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
984 | * | |
985 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
986 | * read-side critical sections can occur. | |
987 | * | |
988 | * We crowbar the ->dynticks_nesting field to DYNTICK_TASK_NEST to | |
989 | * allow for the possibility of usermode upcalls messing up our count | |
990 | * of interrupt nesting level during the busy period that is just | |
991 | * now starting. | |
992 | */ | |
993 | void rcu_idle_exit(void) | |
994 | { | |
c5d900bf FW |
995 | unsigned long flags; |
996 | ||
997 | local_irq_save(flags); | |
cb349ca9 | 998 | rcu_eqs_exit(false); |
28ced795 | 999 | rcu_sysidle_exit(0); |
c5d900bf | 1000 | local_irq_restore(flags); |
adf5091e | 1001 | } |
8a2ecf47 | 1002 | EXPORT_SYMBOL_GPL(rcu_idle_exit); |
9b2e4f18 | 1003 | |
d1ec4c34 | 1004 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
1005 | /** |
1006 | * rcu_user_exit - inform RCU that we are exiting userspace. | |
1007 | * | |
1008 | * Exit RCU idle mode while entering the kernel because it can | |
1009 | * run a RCU read side critical section anytime. | |
1010 | */ | |
1011 | void rcu_user_exit(void) | |
1012 | { | |
91d1aa43 | 1013 | rcu_eqs_exit(1); |
adf5091e | 1014 | } |
d1ec4c34 | 1015 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 1016 | |
9b2e4f18 PM |
1017 | /** |
1018 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle | |
1019 | * | |
1020 | * Enter an interrupt handler, which might possibly result in exiting | |
1021 | * idle mode, in other words, entering the mode in which read-side critical | |
7c9906ca | 1022 | * sections can occur. The caller must have disabled interrupts. |
9b2e4f18 PM |
1023 | * |
1024 | * Note that the Linux kernel is fully capable of entering an interrupt | |
1025 | * handler that it never exits, for example when doing upcalls to | |
1026 | * user mode! This code assumes that the idle loop never does upcalls to | |
1027 | * user mode. If your architecture does do upcalls from the idle loop (or | |
1028 | * does anything else that results in unbalanced calls to the irq_enter() | |
1029 | * and irq_exit() functions), RCU will give you what you deserve, good | |
1030 | * and hard. But very infrequently and irreproducibly. | |
1031 | * | |
1032 | * Use things like work queues to work around this limitation. | |
1033 | * | |
1034 | * You have been warned. | |
1035 | */ | |
1036 | void rcu_irq_enter(void) | |
1037 | { | |
9b2e4f18 PM |
1038 | struct rcu_dynticks *rdtp; |
1039 | long long oldval; | |
1040 | ||
7c9906ca | 1041 | RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_irq_enter() invoked with irqs enabled!!!"); |
c9d4b0af | 1042 | rdtp = this_cpu_ptr(&rcu_dynticks); |
9b2e4f18 PM |
1043 | oldval = rdtp->dynticks_nesting; |
1044 | rdtp->dynticks_nesting++; | |
1ce46ee5 PM |
1045 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
1046 | rdtp->dynticks_nesting == 0); | |
b6fc6020 | 1047 | if (oldval) |
f7f7bac9 | 1048 | trace_rcu_dyntick(TPS("++="), oldval, rdtp->dynticks_nesting); |
b6fc6020 | 1049 | else |
28ced795 CL |
1050 | rcu_eqs_exit_common(oldval, true); |
1051 | rcu_sysidle_exit(1); | |
7c9906ca PM |
1052 | } |
1053 | ||
1054 | /* | |
1055 | * Wrapper for rcu_irq_enter() where interrupts are enabled. | |
1056 | */ | |
1057 | void rcu_irq_enter_irqson(void) | |
1058 | { | |
1059 | unsigned long flags; | |
1060 | ||
1061 | local_irq_save(flags); | |
1062 | rcu_irq_enter(); | |
64db4cff | 1063 | local_irq_restore(flags); |
64db4cff PM |
1064 | } |
1065 | ||
1066 | /** | |
1067 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
1068 | * | |
734d1680 PM |
1069 | * If the CPU was idle from RCU's viewpoint, update rdtp->dynticks and |
1070 | * rdtp->dynticks_nmi_nesting to let the RCU grace-period handling know | |
1071 | * that the CPU is active. This implementation permits nested NMIs, as | |
1072 | * long as the nesting level does not overflow an int. (You will probably | |
1073 | * run out of stack space first.) | |
64db4cff PM |
1074 | */ |
1075 | void rcu_nmi_enter(void) | |
1076 | { | |
c9d4b0af | 1077 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
734d1680 | 1078 | int incby = 2; |
64db4cff | 1079 | |
734d1680 PM |
1080 | /* Complain about underflow. */ |
1081 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting < 0); | |
1082 | ||
1083 | /* | |
1084 | * If idle from RCU viewpoint, atomically increment ->dynticks | |
1085 | * to mark non-idle and increment ->dynticks_nmi_nesting by one. | |
1086 | * Otherwise, increment ->dynticks_nmi_nesting by two. This means | |
1087 | * if ->dynticks_nmi_nesting is equal to one, we are guaranteed | |
1088 | * to be in the outermost NMI handler that interrupted an RCU-idle | |
1089 | * period (observation due to Andy Lutomirski). | |
1090 | */ | |
02a5c550 | 1091 | if (rcu_dynticks_curr_cpu_in_eqs()) { |
2625d469 | 1092 | rcu_dynticks_eqs_exit(); |
734d1680 PM |
1093 | incby = 1; |
1094 | } | |
1095 | rdtp->dynticks_nmi_nesting += incby; | |
1096 | barrier(); | |
64db4cff PM |
1097 | } |
1098 | ||
1099 | /** | |
1100 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
1101 | * | |
734d1680 PM |
1102 | * If we are returning from the outermost NMI handler that interrupted an |
1103 | * RCU-idle period, update rdtp->dynticks and rdtp->dynticks_nmi_nesting | |
1104 | * to let the RCU grace-period handling know that the CPU is back to | |
1105 | * being RCU-idle. | |
64db4cff PM |
1106 | */ |
1107 | void rcu_nmi_exit(void) | |
1108 | { | |
c9d4b0af | 1109 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
64db4cff | 1110 | |
734d1680 PM |
1111 | /* |
1112 | * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks. | |
1113 | * (We are exiting an NMI handler, so RCU better be paying attention | |
1114 | * to us!) | |
1115 | */ | |
1116 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting <= 0); | |
02a5c550 | 1117 | WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs()); |
734d1680 PM |
1118 | |
1119 | /* | |
1120 | * If the nesting level is not 1, the CPU wasn't RCU-idle, so | |
1121 | * leave it in non-RCU-idle state. | |
1122 | */ | |
1123 | if (rdtp->dynticks_nmi_nesting != 1) { | |
1124 | rdtp->dynticks_nmi_nesting -= 2; | |
64db4cff | 1125 | return; |
734d1680 PM |
1126 | } |
1127 | ||
1128 | /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */ | |
1129 | rdtp->dynticks_nmi_nesting = 0; | |
2625d469 | 1130 | rcu_dynticks_eqs_enter(); |
64db4cff PM |
1131 | } |
1132 | ||
1133 | /** | |
5c173eb8 PM |
1134 | * __rcu_is_watching - are RCU read-side critical sections safe? |
1135 | * | |
1136 | * Return true if RCU is watching the running CPU, which means that | |
1137 | * this CPU can safely enter RCU read-side critical sections. Unlike | |
1138 | * rcu_is_watching(), the caller of __rcu_is_watching() must have at | |
1139 | * least disabled preemption. | |
1140 | */ | |
9418fb20 | 1141 | bool notrace __rcu_is_watching(void) |
5c173eb8 | 1142 | { |
02a5c550 | 1143 | return !rcu_dynticks_curr_cpu_in_eqs(); |
5c173eb8 PM |
1144 | } |
1145 | ||
1146 | /** | |
1147 | * rcu_is_watching - see if RCU thinks that the current CPU is idle | |
64db4cff | 1148 | * |
9b2e4f18 | 1149 | * If the current CPU is in its idle loop and is neither in an interrupt |
34240697 | 1150 | * or NMI handler, return true. |
64db4cff | 1151 | */ |
9418fb20 | 1152 | bool notrace rcu_is_watching(void) |
64db4cff | 1153 | { |
f534ed1f | 1154 | bool ret; |
34240697 | 1155 | |
46f00d18 | 1156 | preempt_disable_notrace(); |
5c173eb8 | 1157 | ret = __rcu_is_watching(); |
46f00d18 | 1158 | preempt_enable_notrace(); |
34240697 | 1159 | return ret; |
64db4cff | 1160 | } |
5c173eb8 | 1161 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 1162 | |
62fde6ed | 1163 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
1164 | |
1165 | /* | |
1166 | * Is the current CPU online? Disable preemption to avoid false positives | |
1167 | * that could otherwise happen due to the current CPU number being sampled, | |
1168 | * this task being preempted, its old CPU being taken offline, resuming | |
1169 | * on some other CPU, then determining that its old CPU is now offline. | |
1170 | * It is OK to use RCU on an offline processor during initial boot, hence | |
2036d94a PM |
1171 | * the check for rcu_scheduler_fully_active. Note also that it is OK |
1172 | * for a CPU coming online to use RCU for one jiffy prior to marking itself | |
1173 | * online in the cpu_online_mask. Similarly, it is OK for a CPU going | |
1174 | * offline to continue to use RCU for one jiffy after marking itself | |
1175 | * offline in the cpu_online_mask. This leniency is necessary given the | |
1176 | * non-atomic nature of the online and offline processing, for example, | |
4df83742 TG |
1177 | * the fact that a CPU enters the scheduler after completing the teardown |
1178 | * of the CPU. | |
2036d94a | 1179 | * |
4df83742 TG |
1180 | * This is also why RCU internally marks CPUs online during in the |
1181 | * preparation phase and offline after the CPU has been taken down. | |
c0d6d01b PM |
1182 | * |
1183 | * Disable checking if in an NMI handler because we cannot safely report | |
1184 | * errors from NMI handlers anyway. | |
1185 | */ | |
1186 | bool rcu_lockdep_current_cpu_online(void) | |
1187 | { | |
2036d94a PM |
1188 | struct rcu_data *rdp; |
1189 | struct rcu_node *rnp; | |
c0d6d01b PM |
1190 | bool ret; |
1191 | ||
1192 | if (in_nmi()) | |
f6f7ee9a | 1193 | return true; |
c0d6d01b | 1194 | preempt_disable(); |
c9d4b0af | 1195 | rdp = this_cpu_ptr(&rcu_sched_data); |
2036d94a | 1196 | rnp = rdp->mynode; |
0aa04b05 | 1197 | ret = (rdp->grpmask & rcu_rnp_online_cpus(rnp)) || |
c0d6d01b PM |
1198 | !rcu_scheduler_fully_active; |
1199 | preempt_enable(); | |
1200 | return ret; | |
1201 | } | |
1202 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
1203 | ||
62fde6ed | 1204 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 1205 | |
64db4cff | 1206 | /** |
9b2e4f18 | 1207 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle |
64db4cff | 1208 | * |
9b2e4f18 PM |
1209 | * If the current CPU is idle or running at a first-level (not nested) |
1210 | * interrupt from idle, return true. The caller must have at least | |
1211 | * disabled preemption. | |
64db4cff | 1212 | */ |
62e3cb14 | 1213 | static int rcu_is_cpu_rrupt_from_idle(void) |
64db4cff | 1214 | { |
c9d4b0af | 1215 | return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 1; |
64db4cff PM |
1216 | } |
1217 | ||
64db4cff PM |
1218 | /* |
1219 | * Snapshot the specified CPU's dynticks counter so that we can later | |
1220 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 1221 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 1222 | */ |
217af2a2 PM |
1223 | static int dyntick_save_progress_counter(struct rcu_data *rdp, |
1224 | bool *isidle, unsigned long *maxj) | |
64db4cff | 1225 | { |
8b2f63ab | 1226 | rdp->dynticks_snap = rcu_dynticks_snap(rdp->dynticks); |
0edd1b17 | 1227 | rcu_sysidle_check_cpu(rdp, isidle, maxj); |
02a5c550 | 1228 | if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) { |
7941dbde | 1229 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti")); |
7d0ae808 | 1230 | if (ULONG_CMP_LT(READ_ONCE(rdp->gpnum) + ULONG_MAX / 4, |
e3663b10 | 1231 | rdp->mynode->gpnum)) |
7d0ae808 | 1232 | WRITE_ONCE(rdp->gpwrap, true); |
23a9bacd | 1233 | return 1; |
7941dbde | 1234 | } |
23a9bacd | 1235 | return 0; |
64db4cff PM |
1236 | } |
1237 | ||
1238 | /* | |
1239 | * Return true if the specified CPU has passed through a quiescent | |
1240 | * state by virtue of being in or having passed through an dynticks | |
1241 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 1242 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 1243 | */ |
217af2a2 PM |
1244 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp, |
1245 | bool *isidle, unsigned long *maxj) | |
64db4cff | 1246 | { |
3a19b46a | 1247 | unsigned long jtsq; |
0f9be8ca | 1248 | bool *rnhqp; |
3a19b46a PM |
1249 | unsigned long rjtsc; |
1250 | struct rcu_node *rnp; | |
64db4cff PM |
1251 | |
1252 | /* | |
1253 | * If the CPU passed through or entered a dynticks idle phase with | |
1254 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
1255 | * already acknowledged the request to pass through a quiescent | |
1256 | * state. Either way, that CPU cannot possibly be in an RCU | |
1257 | * read-side critical section that started before the beginning | |
1258 | * of the current RCU grace period. | |
1259 | */ | |
02a5c550 | 1260 | if (rcu_dynticks_in_eqs_since(rdp->dynticks, rdp->dynticks_snap)) { |
f7f7bac9 | 1261 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti")); |
64db4cff PM |
1262 | rdp->dynticks_fqs++; |
1263 | return 1; | |
1264 | } | |
1265 | ||
3a19b46a PM |
1266 | /* Compute and saturate jiffies_till_sched_qs. */ |
1267 | jtsq = jiffies_till_sched_qs; | |
1268 | rjtsc = rcu_jiffies_till_stall_check(); | |
1269 | if (jtsq > rjtsc / 2) { | |
1270 | WRITE_ONCE(jiffies_till_sched_qs, rjtsc); | |
1271 | jtsq = rjtsc / 2; | |
1272 | } else if (jtsq < 1) { | |
1273 | WRITE_ONCE(jiffies_till_sched_qs, 1); | |
1274 | jtsq = 1; | |
1275 | } | |
1276 | ||
a82dcc76 | 1277 | /* |
3a19b46a PM |
1278 | * Has this CPU encountered a cond_resched_rcu_qs() since the |
1279 | * beginning of the grace period? For this to be the case, | |
1280 | * the CPU has to have noticed the current grace period. This | |
1281 | * might not be the case for nohz_full CPUs looping in the kernel. | |
a82dcc76 | 1282 | */ |
3a19b46a PM |
1283 | rnp = rdp->mynode; |
1284 | if (time_after(jiffies, rdp->rsp->gp_start + jtsq) && | |
9577df9a | 1285 | READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) && |
3a19b46a PM |
1286 | READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) { |
1287 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc")); | |
1288 | return 1; | |
1289 | } | |
1290 | ||
38d30b33 PM |
1291 | /* Check for the CPU being offline. */ |
1292 | if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp))) { | |
f7f7bac9 | 1293 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl")); |
a82dcc76 PM |
1294 | rdp->offline_fqs++; |
1295 | return 1; | |
1296 | } | |
65d798f0 PM |
1297 | |
1298 | /* | |
4a81e832 PM |
1299 | * A CPU running for an extended time within the kernel can |
1300 | * delay RCU grace periods. When the CPU is in NO_HZ_FULL mode, | |
1301 | * even context-switching back and forth between a pair of | |
1302 | * in-kernel CPU-bound tasks cannot advance grace periods. | |
1303 | * So if the grace period is old enough, make the CPU pay attention. | |
1304 | * Note that the unsynchronized assignments to the per-CPU | |
0f9be8ca | 1305 | * rcu_need_heavy_qs variable are safe. Yes, setting of |
4a81e832 PM |
1306 | * bits can be lost, but they will be set again on the next |
1307 | * force-quiescent-state pass. So lost bit sets do not result | |
1308 | * in incorrect behavior, merely in a grace period lasting | |
1309 | * a few jiffies longer than it might otherwise. Because | |
1310 | * there are at most four threads involved, and because the | |
1311 | * updates are only once every few jiffies, the probability of | |
1312 | * lossage (and thus of slight grace-period extension) is | |
1313 | * quite low. | |
1314 | * | |
1315 | * Note that if the jiffies_till_sched_qs boot/sysfs parameter | |
1316 | * is set too high, we override with half of the RCU CPU stall | |
1317 | * warning delay. | |
6193c76a | 1318 | */ |
0f9be8ca PM |
1319 | rnhqp = &per_cpu(rcu_dynticks.rcu_need_heavy_qs, rdp->cpu); |
1320 | if (!READ_ONCE(*rnhqp) && | |
1321 | (time_after(jiffies, rdp->rsp->gp_start + jtsq) || | |
1322 | time_after(jiffies, rdp->rsp->jiffies_resched))) { | |
1323 | WRITE_ONCE(*rnhqp, true); | |
4914950a | 1324 | rdp->rsp->jiffies_resched += 5; /* Re-enable beating. */ |
6193c76a PM |
1325 | } |
1326 | ||
28053bc7 PM |
1327 | /* |
1328 | * If more than halfway to RCU CPU stall-warning time, do | |
1329 | * a resched_cpu() to try to loosen things up a bit. | |
1330 | */ | |
1331 | if (jiffies - rdp->rsp->gp_start > rcu_jiffies_till_stall_check() / 2) | |
1332 | resched_cpu(rdp->cpu); | |
4914950a | 1333 | |
a82dcc76 | 1334 | return 0; |
64db4cff PM |
1335 | } |
1336 | ||
64db4cff PM |
1337 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
1338 | { | |
cb1e78cf | 1339 | unsigned long j = jiffies; |
6193c76a | 1340 | unsigned long j1; |
26cdfedf PM |
1341 | |
1342 | rsp->gp_start = j; | |
1343 | smp_wmb(); /* Record start time before stall time. */ | |
6193c76a | 1344 | j1 = rcu_jiffies_till_stall_check(); |
7d0ae808 | 1345 | WRITE_ONCE(rsp->jiffies_stall, j + j1); |
6193c76a | 1346 | rsp->jiffies_resched = j + j1 / 2; |
7d0ae808 | 1347 | rsp->n_force_qs_gpstart = READ_ONCE(rsp->n_force_qs); |
64db4cff PM |
1348 | } |
1349 | ||
6b50e119 PM |
1350 | /* |
1351 | * Convert a ->gp_state value to a character string. | |
1352 | */ | |
1353 | static const char *gp_state_getname(short gs) | |
1354 | { | |
1355 | if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names)) | |
1356 | return "???"; | |
1357 | return gp_state_names[gs]; | |
1358 | } | |
1359 | ||
fb81a44b PM |
1360 | /* |
1361 | * Complain about starvation of grace-period kthread. | |
1362 | */ | |
1363 | static void rcu_check_gp_kthread_starvation(struct rcu_state *rsp) | |
1364 | { | |
1365 | unsigned long gpa; | |
1366 | unsigned long j; | |
1367 | ||
1368 | j = jiffies; | |
7d0ae808 | 1369 | gpa = READ_ONCE(rsp->gp_activity); |
b1adb3e2 | 1370 | if (j - gpa > 2 * HZ) { |
6b50e119 | 1371 | pr_err("%s kthread starved for %ld jiffies! g%lu c%lu f%#x %s(%d) ->state=%#lx\n", |
81e701e4 | 1372 | rsp->name, j - gpa, |
319362c9 | 1373 | rsp->gpnum, rsp->completed, |
6b50e119 PM |
1374 | rsp->gp_flags, |
1375 | gp_state_getname(rsp->gp_state), rsp->gp_state, | |
a0e3a3aa | 1376 | rsp->gp_kthread ? rsp->gp_kthread->state : ~0); |
86057b80 | 1377 | if (rsp->gp_kthread) { |
b1adb3e2 | 1378 | sched_show_task(rsp->gp_kthread); |
86057b80 PM |
1379 | wake_up_process(rsp->gp_kthread); |
1380 | } | |
b1adb3e2 | 1381 | } |
64db4cff PM |
1382 | } |
1383 | ||
b637a328 | 1384 | /* |
7aa92230 PM |
1385 | * Dump stacks of all tasks running on stalled CPUs. First try using |
1386 | * NMIs, but fall back to manual remote stack tracing on architectures | |
1387 | * that don't support NMI-based stack dumps. The NMI-triggered stack | |
1388 | * traces are more accurate because they are printed by the target CPU. | |
b637a328 PM |
1389 | */ |
1390 | static void rcu_dump_cpu_stacks(struct rcu_state *rsp) | |
1391 | { | |
1392 | int cpu; | |
1393 | unsigned long flags; | |
1394 | struct rcu_node *rnp; | |
1395 | ||
1396 | rcu_for_each_leaf_node(rsp, rnp) { | |
6cf10081 | 1397 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7aa92230 PM |
1398 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1399 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) | |
1400 | if (!trigger_single_cpu_backtrace(cpu)) | |
bc75e999 | 1401 | dump_cpu_task(cpu); |
67c583a7 | 1402 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b637a328 PM |
1403 | } |
1404 | } | |
1405 | ||
8c7c4829 PM |
1406 | /* |
1407 | * If too much time has passed in the current grace period, and if | |
1408 | * so configured, go kick the relevant kthreads. | |
1409 | */ | |
1410 | static void rcu_stall_kick_kthreads(struct rcu_state *rsp) | |
1411 | { | |
1412 | unsigned long j; | |
1413 | ||
1414 | if (!rcu_kick_kthreads) | |
1415 | return; | |
1416 | j = READ_ONCE(rsp->jiffies_kick_kthreads); | |
aa3e0bf1 PM |
1417 | if (time_after(jiffies, j) && rsp->gp_kthread && |
1418 | (rcu_gp_in_progress(rsp) || READ_ONCE(rsp->gp_flags))) { | |
8c7c4829 | 1419 | WARN_ONCE(1, "Kicking %s grace-period kthread\n", rsp->name); |
5dffed1e | 1420 | rcu_ftrace_dump(DUMP_ALL); |
8c7c4829 PM |
1421 | wake_up_process(rsp->gp_kthread); |
1422 | WRITE_ONCE(rsp->jiffies_kick_kthreads, j + HZ); | |
1423 | } | |
1424 | } | |
1425 | ||
088e9d25 DBO |
1426 | static inline void panic_on_rcu_stall(void) |
1427 | { | |
1428 | if (sysctl_panic_on_rcu_stall) | |
1429 | panic("RCU Stall\n"); | |
1430 | } | |
1431 | ||
6ccd2ecd | 1432 | static void print_other_cpu_stall(struct rcu_state *rsp, unsigned long gpnum) |
64db4cff PM |
1433 | { |
1434 | int cpu; | |
1435 | long delta; | |
1436 | unsigned long flags; | |
6ccd2ecd PM |
1437 | unsigned long gpa; |
1438 | unsigned long j; | |
285fe294 | 1439 | int ndetected = 0; |
64db4cff | 1440 | struct rcu_node *rnp = rcu_get_root(rsp); |
53bb857c | 1441 | long totqlen = 0; |
64db4cff | 1442 | |
8c7c4829 PM |
1443 | /* Kick and suppress, if so configured. */ |
1444 | rcu_stall_kick_kthreads(rsp); | |
1445 | if (rcu_cpu_stall_suppress) | |
1446 | return; | |
1447 | ||
64db4cff PM |
1448 | /* Only let one CPU complain about others per time interval. */ |
1449 | ||
6cf10081 | 1450 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7d0ae808 | 1451 | delta = jiffies - READ_ONCE(rsp->jiffies_stall); |
fc2219d4 | 1452 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
67c583a7 | 1453 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1454 | return; |
1455 | } | |
7d0ae808 PM |
1456 | WRITE_ONCE(rsp->jiffies_stall, |
1457 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
67c583a7 | 1458 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1459 | |
8cdd32a9 PM |
1460 | /* |
1461 | * OK, time to rat on our buddy... | |
1462 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1463 | * RCU CPU stall warnings. | |
1464 | */ | |
d7f3e207 | 1465 | pr_err("INFO: %s detected stalls on CPUs/tasks:", |
4300aa64 | 1466 | rsp->name); |
a858af28 | 1467 | print_cpu_stall_info_begin(); |
a0b6c9a7 | 1468 | rcu_for_each_leaf_node(rsp, rnp) { |
6cf10081 | 1469 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
9bc8b558 | 1470 | ndetected += rcu_print_task_stall(rnp); |
c8020a67 | 1471 | if (rnp->qsmask != 0) { |
bc75e999 MR |
1472 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1473 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) { | |
1474 | print_cpu_stall_info(rsp, cpu); | |
c8020a67 PM |
1475 | ndetected++; |
1476 | } | |
1477 | } | |
67c583a7 | 1478 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1479 | } |
a858af28 | 1480 | |
a858af28 | 1481 | print_cpu_stall_info_end(); |
53bb857c PM |
1482 | for_each_possible_cpu(cpu) |
1483 | totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen; | |
83ebe63e | 1484 | pr_cont("(detected by %d, t=%ld jiffies, g=%ld, c=%ld, q=%lu)\n", |
eee05882 | 1485 | smp_processor_id(), (long)(jiffies - rsp->gp_start), |
83ebe63e | 1486 | (long)rsp->gpnum, (long)rsp->completed, totqlen); |
6ccd2ecd | 1487 | if (ndetected) { |
b637a328 | 1488 | rcu_dump_cpu_stacks(rsp); |
c4402b27 BP |
1489 | |
1490 | /* Complain about tasks blocking the grace period. */ | |
1491 | rcu_print_detail_task_stall(rsp); | |
6ccd2ecd | 1492 | } else { |
7d0ae808 PM |
1493 | if (READ_ONCE(rsp->gpnum) != gpnum || |
1494 | READ_ONCE(rsp->completed) == gpnum) { | |
6ccd2ecd PM |
1495 | pr_err("INFO: Stall ended before state dump start\n"); |
1496 | } else { | |
1497 | j = jiffies; | |
7d0ae808 | 1498 | gpa = READ_ONCE(rsp->gp_activity); |
237a0f21 | 1499 | pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n", |
6ccd2ecd | 1500 | rsp->name, j - gpa, j, gpa, |
237a0f21 PM |
1501 | jiffies_till_next_fqs, |
1502 | rcu_get_root(rsp)->qsmask); | |
6ccd2ecd PM |
1503 | /* In this case, the current CPU might be at fault. */ |
1504 | sched_show_task(current); | |
1505 | } | |
1506 | } | |
c1dc0b9c | 1507 | |
fb81a44b PM |
1508 | rcu_check_gp_kthread_starvation(rsp); |
1509 | ||
088e9d25 DBO |
1510 | panic_on_rcu_stall(); |
1511 | ||
4cdfc175 | 1512 | force_quiescent_state(rsp); /* Kick them all. */ |
64db4cff PM |
1513 | } |
1514 | ||
1515 | static void print_cpu_stall(struct rcu_state *rsp) | |
1516 | { | |
53bb857c | 1517 | int cpu; |
64db4cff PM |
1518 | unsigned long flags; |
1519 | struct rcu_node *rnp = rcu_get_root(rsp); | |
53bb857c | 1520 | long totqlen = 0; |
64db4cff | 1521 | |
8c7c4829 PM |
1522 | /* Kick and suppress, if so configured. */ |
1523 | rcu_stall_kick_kthreads(rsp); | |
1524 | if (rcu_cpu_stall_suppress) | |
1525 | return; | |
1526 | ||
8cdd32a9 PM |
1527 | /* |
1528 | * OK, time to rat on ourselves... | |
1529 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1530 | * RCU CPU stall warnings. | |
1531 | */ | |
d7f3e207 | 1532 | pr_err("INFO: %s self-detected stall on CPU", rsp->name); |
a858af28 PM |
1533 | print_cpu_stall_info_begin(); |
1534 | print_cpu_stall_info(rsp, smp_processor_id()); | |
1535 | print_cpu_stall_info_end(); | |
53bb857c PM |
1536 | for_each_possible_cpu(cpu) |
1537 | totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen; | |
83ebe63e PM |
1538 | pr_cont(" (t=%lu jiffies g=%ld c=%ld q=%lu)\n", |
1539 | jiffies - rsp->gp_start, | |
1540 | (long)rsp->gpnum, (long)rsp->completed, totqlen); | |
fb81a44b PM |
1541 | |
1542 | rcu_check_gp_kthread_starvation(rsp); | |
1543 | ||
bc1dce51 | 1544 | rcu_dump_cpu_stacks(rsp); |
c1dc0b9c | 1545 | |
6cf10081 | 1546 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7d0ae808 PM |
1547 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall))) |
1548 | WRITE_ONCE(rsp->jiffies_stall, | |
1549 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
67c583a7 | 1550 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c1dc0b9c | 1551 | |
088e9d25 DBO |
1552 | panic_on_rcu_stall(); |
1553 | ||
b021fe3e PZ |
1554 | /* |
1555 | * Attempt to revive the RCU machinery by forcing a context switch. | |
1556 | * | |
1557 | * A context switch would normally allow the RCU state machine to make | |
1558 | * progress and it could be we're stuck in kernel space without context | |
1559 | * switches for an entirely unreasonable amount of time. | |
1560 | */ | |
1561 | resched_cpu(smp_processor_id()); | |
64db4cff PM |
1562 | } |
1563 | ||
1564 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
1565 | { | |
26cdfedf PM |
1566 | unsigned long completed; |
1567 | unsigned long gpnum; | |
1568 | unsigned long gps; | |
bad6e139 PM |
1569 | unsigned long j; |
1570 | unsigned long js; | |
64db4cff PM |
1571 | struct rcu_node *rnp; |
1572 | ||
8c7c4829 PM |
1573 | if ((rcu_cpu_stall_suppress && !rcu_kick_kthreads) || |
1574 | !rcu_gp_in_progress(rsp)) | |
c68de209 | 1575 | return; |
8c7c4829 | 1576 | rcu_stall_kick_kthreads(rsp); |
cb1e78cf | 1577 | j = jiffies; |
26cdfedf PM |
1578 | |
1579 | /* | |
1580 | * Lots of memory barriers to reject false positives. | |
1581 | * | |
1582 | * The idea is to pick up rsp->gpnum, then rsp->jiffies_stall, | |
1583 | * then rsp->gp_start, and finally rsp->completed. These values | |
1584 | * are updated in the opposite order with memory barriers (or | |
1585 | * equivalent) during grace-period initialization and cleanup. | |
1586 | * Now, a false positive can occur if we get an new value of | |
1587 | * rsp->gp_start and a old value of rsp->jiffies_stall. But given | |
1588 | * the memory barriers, the only way that this can happen is if one | |
1589 | * grace period ends and another starts between these two fetches. | |
1590 | * Detect this by comparing rsp->completed with the previous fetch | |
1591 | * from rsp->gpnum. | |
1592 | * | |
1593 | * Given this check, comparisons of jiffies, rsp->jiffies_stall, | |
1594 | * and rsp->gp_start suffice to forestall false positives. | |
1595 | */ | |
7d0ae808 | 1596 | gpnum = READ_ONCE(rsp->gpnum); |
26cdfedf | 1597 | smp_rmb(); /* Pick up ->gpnum first... */ |
7d0ae808 | 1598 | js = READ_ONCE(rsp->jiffies_stall); |
26cdfedf | 1599 | smp_rmb(); /* ...then ->jiffies_stall before the rest... */ |
7d0ae808 | 1600 | gps = READ_ONCE(rsp->gp_start); |
26cdfedf | 1601 | smp_rmb(); /* ...and finally ->gp_start before ->completed. */ |
7d0ae808 | 1602 | completed = READ_ONCE(rsp->completed); |
26cdfedf PM |
1603 | if (ULONG_CMP_GE(completed, gpnum) || |
1604 | ULONG_CMP_LT(j, js) || | |
1605 | ULONG_CMP_GE(gps, js)) | |
1606 | return; /* No stall or GP completed since entering function. */ | |
64db4cff | 1607 | rnp = rdp->mynode; |
c96ea7cf | 1608 | if (rcu_gp_in_progress(rsp) && |
7d0ae808 | 1609 | (READ_ONCE(rnp->qsmask) & rdp->grpmask)) { |
64db4cff PM |
1610 | |
1611 | /* We haven't checked in, so go dump stack. */ | |
1612 | print_cpu_stall(rsp); | |
1613 | ||
bad6e139 PM |
1614 | } else if (rcu_gp_in_progress(rsp) && |
1615 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 1616 | |
bad6e139 | 1617 | /* They had a few time units to dump stack, so complain. */ |
6ccd2ecd | 1618 | print_other_cpu_stall(rsp, gpnum); |
64db4cff PM |
1619 | } |
1620 | } | |
1621 | ||
53d84e00 PM |
1622 | /** |
1623 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
1624 | * | |
1625 | * Set the stall-warning timeout way off into the future, thus preventing | |
1626 | * any RCU CPU stall-warning messages from appearing in the current set of | |
1627 | * RCU grace periods. | |
1628 | * | |
1629 | * The caller must disable hard irqs. | |
1630 | */ | |
1631 | void rcu_cpu_stall_reset(void) | |
1632 | { | |
6ce75a23 PM |
1633 | struct rcu_state *rsp; |
1634 | ||
1635 | for_each_rcu_flavor(rsp) | |
7d0ae808 | 1636 | WRITE_ONCE(rsp->jiffies_stall, jiffies + ULONG_MAX / 2); |
53d84e00 PM |
1637 | } |
1638 | ||
3f5d3ea6 | 1639 | /* |
d3f3f3f2 PM |
1640 | * Initialize the specified rcu_data structure's default callback list |
1641 | * to empty. The default callback list is the one that is not used by | |
1642 | * no-callbacks CPUs. | |
3f5d3ea6 | 1643 | */ |
d3f3f3f2 | 1644 | static void init_default_callback_list(struct rcu_data *rdp) |
3f5d3ea6 PM |
1645 | { |
1646 | int i; | |
1647 | ||
1648 | rdp->nxtlist = NULL; | |
1649 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1650 | rdp->nxttail[i] = &rdp->nxtlist; | |
1651 | } | |
1652 | ||
d3f3f3f2 PM |
1653 | /* |
1654 | * Initialize the specified rcu_data structure's callback list to empty. | |
1655 | */ | |
1656 | static void init_callback_list(struct rcu_data *rdp) | |
1657 | { | |
1658 | if (init_nocb_callback_list(rdp)) | |
1659 | return; | |
1660 | init_default_callback_list(rdp); | |
1661 | } | |
1662 | ||
dc35c893 PM |
1663 | /* |
1664 | * Determine the value that ->completed will have at the end of the | |
1665 | * next subsequent grace period. This is used to tag callbacks so that | |
1666 | * a CPU can invoke callbacks in a timely fashion even if that CPU has | |
1667 | * been dyntick-idle for an extended period with callbacks under the | |
1668 | * influence of RCU_FAST_NO_HZ. | |
1669 | * | |
1670 | * The caller must hold rnp->lock with interrupts disabled. | |
1671 | */ | |
1672 | static unsigned long rcu_cbs_completed(struct rcu_state *rsp, | |
1673 | struct rcu_node *rnp) | |
1674 | { | |
1675 | /* | |
1676 | * If RCU is idle, we just wait for the next grace period. | |
1677 | * But we can only be sure that RCU is idle if we are looking | |
1678 | * at the root rcu_node structure -- otherwise, a new grace | |
1679 | * period might have started, but just not yet gotten around | |
1680 | * to initializing the current non-root rcu_node structure. | |
1681 | */ | |
1682 | if (rcu_get_root(rsp) == rnp && rnp->gpnum == rnp->completed) | |
1683 | return rnp->completed + 1; | |
1684 | ||
1685 | /* | |
1686 | * Otherwise, wait for a possible partial grace period and | |
1687 | * then the subsequent full grace period. | |
1688 | */ | |
1689 | return rnp->completed + 2; | |
1690 | } | |
1691 | ||
0446be48 PM |
1692 | /* |
1693 | * Trace-event helper function for rcu_start_future_gp() and | |
1694 | * rcu_nocb_wait_gp(). | |
1695 | */ | |
1696 | static void trace_rcu_future_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
e66c33d5 | 1697 | unsigned long c, const char *s) |
0446be48 PM |
1698 | { |
1699 | trace_rcu_future_grace_period(rdp->rsp->name, rnp->gpnum, | |
1700 | rnp->completed, c, rnp->level, | |
1701 | rnp->grplo, rnp->grphi, s); | |
1702 | } | |
1703 | ||
1704 | /* | |
1705 | * Start some future grace period, as needed to handle newly arrived | |
1706 | * callbacks. The required future grace periods are recorded in each | |
48a7639c PM |
1707 | * rcu_node structure's ->need_future_gp field. Returns true if there |
1708 | * is reason to awaken the grace-period kthread. | |
0446be48 PM |
1709 | * |
1710 | * The caller must hold the specified rcu_node structure's ->lock. | |
1711 | */ | |
48a7639c PM |
1712 | static bool __maybe_unused |
1713 | rcu_start_future_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
1714 | unsigned long *c_out) | |
0446be48 PM |
1715 | { |
1716 | unsigned long c; | |
1717 | int i; | |
48a7639c | 1718 | bool ret = false; |
0446be48 PM |
1719 | struct rcu_node *rnp_root = rcu_get_root(rdp->rsp); |
1720 | ||
1721 | /* | |
1722 | * Pick up grace-period number for new callbacks. If this | |
1723 | * grace period is already marked as needed, return to the caller. | |
1724 | */ | |
1725 | c = rcu_cbs_completed(rdp->rsp, rnp); | |
f7f7bac9 | 1726 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startleaf")); |
0446be48 | 1727 | if (rnp->need_future_gp[c & 0x1]) { |
f7f7bac9 | 1728 | trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartleaf")); |
48a7639c | 1729 | goto out; |
0446be48 PM |
1730 | } |
1731 | ||
1732 | /* | |
1733 | * If either this rcu_node structure or the root rcu_node structure | |
1734 | * believe that a grace period is in progress, then we must wait | |
1735 | * for the one following, which is in "c". Because our request | |
1736 | * will be noticed at the end of the current grace period, we don't | |
48bd8e9b PK |
1737 | * need to explicitly start one. We only do the lockless check |
1738 | * of rnp_root's fields if the current rcu_node structure thinks | |
1739 | * there is no grace period in flight, and because we hold rnp->lock, | |
1740 | * the only possible change is when rnp_root's two fields are | |
1741 | * equal, in which case rnp_root->gpnum might be concurrently | |
1742 | * incremented. But that is OK, as it will just result in our | |
1743 | * doing some extra useless work. | |
0446be48 PM |
1744 | */ |
1745 | if (rnp->gpnum != rnp->completed || | |
7d0ae808 | 1746 | READ_ONCE(rnp_root->gpnum) != READ_ONCE(rnp_root->completed)) { |
0446be48 | 1747 | rnp->need_future_gp[c & 0x1]++; |
f7f7bac9 | 1748 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleaf")); |
48a7639c | 1749 | goto out; |
0446be48 PM |
1750 | } |
1751 | ||
1752 | /* | |
1753 | * There might be no grace period in progress. If we don't already | |
1754 | * hold it, acquire the root rcu_node structure's lock in order to | |
1755 | * start one (if needed). | |
1756 | */ | |
2a67e741 PZ |
1757 | if (rnp != rnp_root) |
1758 | raw_spin_lock_rcu_node(rnp_root); | |
0446be48 PM |
1759 | |
1760 | /* | |
1761 | * Get a new grace-period number. If there really is no grace | |
1762 | * period in progress, it will be smaller than the one we obtained | |
1763 | * earlier. Adjust callbacks as needed. Note that even no-CBs | |
1764 | * CPUs have a ->nxtcompleted[] array, so no no-CBs checks needed. | |
1765 | */ | |
1766 | c = rcu_cbs_completed(rdp->rsp, rnp_root); | |
1767 | for (i = RCU_DONE_TAIL; i < RCU_NEXT_TAIL; i++) | |
1768 | if (ULONG_CMP_LT(c, rdp->nxtcompleted[i])) | |
1769 | rdp->nxtcompleted[i] = c; | |
1770 | ||
1771 | /* | |
1772 | * If the needed for the required grace period is already | |
1773 | * recorded, trace and leave. | |
1774 | */ | |
1775 | if (rnp_root->need_future_gp[c & 0x1]) { | |
f7f7bac9 | 1776 | trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartedroot")); |
0446be48 PM |
1777 | goto unlock_out; |
1778 | } | |
1779 | ||
1780 | /* Record the need for the future grace period. */ | |
1781 | rnp_root->need_future_gp[c & 0x1]++; | |
1782 | ||
1783 | /* If a grace period is not already in progress, start one. */ | |
1784 | if (rnp_root->gpnum != rnp_root->completed) { | |
f7f7bac9 | 1785 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleafroot")); |
0446be48 | 1786 | } else { |
f7f7bac9 | 1787 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startedroot")); |
48a7639c | 1788 | ret = rcu_start_gp_advanced(rdp->rsp, rnp_root, rdp); |
0446be48 PM |
1789 | } |
1790 | unlock_out: | |
1791 | if (rnp != rnp_root) | |
67c583a7 | 1792 | raw_spin_unlock_rcu_node(rnp_root); |
48a7639c PM |
1793 | out: |
1794 | if (c_out != NULL) | |
1795 | *c_out = c; | |
1796 | return ret; | |
0446be48 PM |
1797 | } |
1798 | ||
1799 | /* | |
1800 | * Clean up any old requests for the just-ended grace period. Also return | |
1801 | * whether any additional grace periods have been requested. Also invoke | |
1802 | * rcu_nocb_gp_cleanup() in order to wake up any no-callbacks kthreads | |
1803 | * waiting for this grace period to complete. | |
1804 | */ | |
1805 | static int rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp) | |
1806 | { | |
1807 | int c = rnp->completed; | |
1808 | int needmore; | |
1809 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
1810 | ||
0446be48 PM |
1811 | rnp->need_future_gp[c & 0x1] = 0; |
1812 | needmore = rnp->need_future_gp[(c + 1) & 0x1]; | |
f7f7bac9 SRRH |
1813 | trace_rcu_future_gp(rnp, rdp, c, |
1814 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); | |
0446be48 PM |
1815 | return needmore; |
1816 | } | |
1817 | ||
48a7639c PM |
1818 | /* |
1819 | * Awaken the grace-period kthread for the specified flavor of RCU. | |
1820 | * Don't do a self-awaken, and don't bother awakening when there is | |
1821 | * nothing for the grace-period kthread to do (as in several CPUs | |
1822 | * raced to awaken, and we lost), and finally don't try to awaken | |
1823 | * a kthread that has not yet been created. | |
1824 | */ | |
1825 | static void rcu_gp_kthread_wake(struct rcu_state *rsp) | |
1826 | { | |
1827 | if (current == rsp->gp_kthread || | |
7d0ae808 | 1828 | !READ_ONCE(rsp->gp_flags) || |
48a7639c PM |
1829 | !rsp->gp_kthread) |
1830 | return; | |
abedf8e2 | 1831 | swake_up(&rsp->gp_wq); |
48a7639c PM |
1832 | } |
1833 | ||
dc35c893 PM |
1834 | /* |
1835 | * If there is room, assign a ->completed number to any callbacks on | |
1836 | * this CPU that have not already been assigned. Also accelerate any | |
1837 | * callbacks that were previously assigned a ->completed number that has | |
1838 | * since proven to be too conservative, which can happen if callbacks get | |
1839 | * assigned a ->completed number while RCU is idle, but with reference to | |
1840 | * a non-root rcu_node structure. This function is idempotent, so it does | |
48a7639c PM |
1841 | * not hurt to call it repeatedly. Returns an flag saying that we should |
1842 | * awaken the RCU grace-period kthread. | |
dc35c893 PM |
1843 | * |
1844 | * The caller must hold rnp->lock with interrupts disabled. | |
1845 | */ | |
48a7639c | 1846 | static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1847 | struct rcu_data *rdp) |
1848 | { | |
1849 | unsigned long c; | |
1850 | int i; | |
48a7639c | 1851 | bool ret; |
dc35c893 PM |
1852 | |
1853 | /* If the CPU has no callbacks, nothing to do. */ | |
1854 | if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL]) | |
48a7639c | 1855 | return false; |
dc35c893 PM |
1856 | |
1857 | /* | |
1858 | * Starting from the sublist containing the callbacks most | |
1859 | * recently assigned a ->completed number and working down, find the | |
1860 | * first sublist that is not assignable to an upcoming grace period. | |
1861 | * Such a sublist has something in it (first two tests) and has | |
1862 | * a ->completed number assigned that will complete sooner than | |
1863 | * the ->completed number for newly arrived callbacks (last test). | |
1864 | * | |
1865 | * The key point is that any later sublist can be assigned the | |
1866 | * same ->completed number as the newly arrived callbacks, which | |
1867 | * means that the callbacks in any of these later sublist can be | |
1868 | * grouped into a single sublist, whether or not they have already | |
1869 | * been assigned a ->completed number. | |
1870 | */ | |
1871 | c = rcu_cbs_completed(rsp, rnp); | |
1872 | for (i = RCU_NEXT_TAIL - 1; i > RCU_DONE_TAIL; i--) | |
1873 | if (rdp->nxttail[i] != rdp->nxttail[i - 1] && | |
1874 | !ULONG_CMP_GE(rdp->nxtcompleted[i], c)) | |
1875 | break; | |
1876 | ||
1877 | /* | |
1878 | * If there are no sublist for unassigned callbacks, leave. | |
1879 | * At the same time, advance "i" one sublist, so that "i" will | |
1880 | * index into the sublist where all the remaining callbacks should | |
1881 | * be grouped into. | |
1882 | */ | |
1883 | if (++i >= RCU_NEXT_TAIL) | |
48a7639c | 1884 | return false; |
dc35c893 PM |
1885 | |
1886 | /* | |
1887 | * Assign all subsequent callbacks' ->completed number to the next | |
1888 | * full grace period and group them all in the sublist initially | |
1889 | * indexed by "i". | |
1890 | */ | |
1891 | for (; i <= RCU_NEXT_TAIL; i++) { | |
1892 | rdp->nxttail[i] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1893 | rdp->nxtcompleted[i] = c; | |
1894 | } | |
910ee45d | 1895 | /* Record any needed additional grace periods. */ |
48a7639c | 1896 | ret = rcu_start_future_gp(rnp, rdp, NULL); |
6d4b418c PM |
1897 | |
1898 | /* Trace depending on how much we were able to accelerate. */ | |
1899 | if (!*rdp->nxttail[RCU_WAIT_TAIL]) | |
f7f7bac9 | 1900 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccWaitCB")); |
6d4b418c | 1901 | else |
f7f7bac9 | 1902 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccReadyCB")); |
48a7639c | 1903 | return ret; |
dc35c893 PM |
1904 | } |
1905 | ||
1906 | /* | |
1907 | * Move any callbacks whose grace period has completed to the | |
1908 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
1909 | * assign ->completed numbers to any callbacks in the RCU_NEXT_TAIL | |
1910 | * sublist. This function is idempotent, so it does not hurt to | |
1911 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1912 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1913 | * |
1914 | * The caller must hold rnp->lock with interrupts disabled. | |
1915 | */ | |
48a7639c | 1916 | static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1917 | struct rcu_data *rdp) |
1918 | { | |
1919 | int i, j; | |
1920 | ||
1921 | /* If the CPU has no callbacks, nothing to do. */ | |
1922 | if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL]) | |
48a7639c | 1923 | return false; |
dc35c893 PM |
1924 | |
1925 | /* | |
1926 | * Find all callbacks whose ->completed numbers indicate that they | |
1927 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. | |
1928 | */ | |
1929 | for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) { | |
1930 | if (ULONG_CMP_LT(rnp->completed, rdp->nxtcompleted[i])) | |
1931 | break; | |
1932 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[i]; | |
1933 | } | |
1934 | /* Clean up any sublist tail pointers that were misordered above. */ | |
1935 | for (j = RCU_WAIT_TAIL; j < i; j++) | |
1936 | rdp->nxttail[j] = rdp->nxttail[RCU_DONE_TAIL]; | |
1937 | ||
1938 | /* Copy down callbacks to fill in empty sublists. */ | |
1939 | for (j = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++, j++) { | |
1940 | if (rdp->nxttail[j] == rdp->nxttail[RCU_NEXT_TAIL]) | |
1941 | break; | |
1942 | rdp->nxttail[j] = rdp->nxttail[i]; | |
1943 | rdp->nxtcompleted[j] = rdp->nxtcompleted[i]; | |
1944 | } | |
1945 | ||
1946 | /* Classify any remaining callbacks. */ | |
48a7639c | 1947 | return rcu_accelerate_cbs(rsp, rnp, rdp); |
dc35c893 PM |
1948 | } |
1949 | ||
d09b62df | 1950 | /* |
ba9fbe95 PM |
1951 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1952 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1953 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1954 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1955 | */ |
48a7639c PM |
1956 | static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, |
1957 | struct rcu_data *rdp) | |
d09b62df | 1958 | { |
48a7639c | 1959 | bool ret; |
3563a438 | 1960 | bool need_gp; |
48a7639c | 1961 | |
ba9fbe95 | 1962 | /* Handle the ends of any preceding grace periods first. */ |
e3663b10 | 1963 | if (rdp->completed == rnp->completed && |
7d0ae808 | 1964 | !unlikely(READ_ONCE(rdp->gpwrap))) { |
d09b62df | 1965 | |
ba9fbe95 | 1966 | /* No grace period end, so just accelerate recent callbacks. */ |
48a7639c | 1967 | ret = rcu_accelerate_cbs(rsp, rnp, rdp); |
d09b62df | 1968 | |
dc35c893 PM |
1969 | } else { |
1970 | ||
1971 | /* Advance callbacks. */ | |
48a7639c | 1972 | ret = rcu_advance_cbs(rsp, rnp, rdp); |
d09b62df PM |
1973 | |
1974 | /* Remember that we saw this grace-period completion. */ | |
1975 | rdp->completed = rnp->completed; | |
f7f7bac9 | 1976 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuend")); |
d09b62df | 1977 | } |
398ebe60 | 1978 | |
7d0ae808 | 1979 | if (rdp->gpnum != rnp->gpnum || unlikely(READ_ONCE(rdp->gpwrap))) { |
6eaef633 PM |
1980 | /* |
1981 | * If the current grace period is waiting for this CPU, | |
1982 | * set up to detect a quiescent state, otherwise don't | |
1983 | * go looking for one. | |
1984 | */ | |
1985 | rdp->gpnum = rnp->gpnum; | |
f7f7bac9 | 1986 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart")); |
3563a438 PM |
1987 | need_gp = !!(rnp->qsmask & rdp->grpmask); |
1988 | rdp->cpu_no_qs.b.norm = need_gp; | |
9577df9a | 1989 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
3563a438 | 1990 | rdp->core_needs_qs = need_gp; |
6eaef633 | 1991 | zero_cpu_stall_ticks(rdp); |
7d0ae808 | 1992 | WRITE_ONCE(rdp->gpwrap, false); |
6eaef633 | 1993 | } |
48a7639c | 1994 | return ret; |
6eaef633 PM |
1995 | } |
1996 | ||
d34ea322 | 1997 | static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp) |
6eaef633 PM |
1998 | { |
1999 | unsigned long flags; | |
48a7639c | 2000 | bool needwake; |
6eaef633 PM |
2001 | struct rcu_node *rnp; |
2002 | ||
2003 | local_irq_save(flags); | |
2004 | rnp = rdp->mynode; | |
7d0ae808 PM |
2005 | if ((rdp->gpnum == READ_ONCE(rnp->gpnum) && |
2006 | rdp->completed == READ_ONCE(rnp->completed) && | |
2007 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ | |
2a67e741 | 2008 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
2009 | local_irq_restore(flags); |
2010 | return; | |
2011 | } | |
48a7639c | 2012 | needwake = __note_gp_changes(rsp, rnp, rdp); |
67c583a7 | 2013 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c PM |
2014 | if (needwake) |
2015 | rcu_gp_kthread_wake(rsp); | |
6eaef633 PM |
2016 | } |
2017 | ||
0f41c0dd PM |
2018 | static void rcu_gp_slow(struct rcu_state *rsp, int delay) |
2019 | { | |
2020 | if (delay > 0 && | |
2021 | !(rsp->gpnum % (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) | |
2022 | schedule_timeout_uninterruptible(delay); | |
2023 | } | |
2024 | ||
b3dbec76 | 2025 | /* |
45fed3e7 | 2026 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 2027 | */ |
45fed3e7 | 2028 | static bool rcu_gp_init(struct rcu_state *rsp) |
b3dbec76 | 2029 | { |
0aa04b05 | 2030 | unsigned long oldmask; |
b3dbec76 | 2031 | struct rcu_data *rdp; |
7fdefc10 | 2032 | struct rcu_node *rnp = rcu_get_root(rsp); |
b3dbec76 | 2033 | |
7d0ae808 | 2034 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 2035 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 | 2036 | if (!READ_ONCE(rsp->gp_flags)) { |
f7be8209 | 2037 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 2038 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 2039 | return false; |
f7be8209 | 2040 | } |
7d0ae808 | 2041 | WRITE_ONCE(rsp->gp_flags, 0); /* Clear all flags: New grace period. */ |
b3dbec76 | 2042 | |
f7be8209 PM |
2043 | if (WARN_ON_ONCE(rcu_gp_in_progress(rsp))) { |
2044 | /* | |
2045 | * Grace period already in progress, don't start another. | |
2046 | * Not supposed to be able to happen. | |
2047 | */ | |
67c583a7 | 2048 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 2049 | return false; |
7fdefc10 PM |
2050 | } |
2051 | ||
7fdefc10 | 2052 | /* Advance to a new grace period and initialize state. */ |
26cdfedf | 2053 | record_gp_stall_check_time(rsp); |
765a3f4f PM |
2054 | /* Record GP times before starting GP, hence smp_store_release(). */ |
2055 | smp_store_release(&rsp->gpnum, rsp->gpnum + 1); | |
f7f7bac9 | 2056 | trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start")); |
67c583a7 | 2057 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 2058 | |
0aa04b05 PM |
2059 | /* |
2060 | * Apply per-leaf buffered online and offline operations to the | |
2061 | * rcu_node tree. Note that this new grace period need not wait | |
2062 | * for subsequent online CPUs, and that quiescent-state forcing | |
2063 | * will handle subsequent offline CPUs. | |
2064 | */ | |
2065 | rcu_for_each_leaf_node(rsp, rnp) { | |
0f41c0dd | 2066 | rcu_gp_slow(rsp, gp_preinit_delay); |
2a67e741 | 2067 | raw_spin_lock_irq_rcu_node(rnp); |
0aa04b05 PM |
2068 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
2069 | !rnp->wait_blkd_tasks) { | |
2070 | /* Nothing to do on this leaf rcu_node structure. */ | |
67c583a7 | 2071 | raw_spin_unlock_irq_rcu_node(rnp); |
0aa04b05 PM |
2072 | continue; |
2073 | } | |
2074 | ||
2075 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
2076 | oldmask = rnp->qsmaskinit; | |
2077 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
2078 | ||
2079 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
2080 | if (!oldmask != !rnp->qsmaskinit) { | |
2081 | if (!oldmask) /* First online CPU for this rcu_node. */ | |
2082 | rcu_init_new_rnp(rnp); | |
2083 | else if (rcu_preempt_has_tasks(rnp)) /* blocked tasks */ | |
2084 | rnp->wait_blkd_tasks = true; | |
2085 | else /* Last offline CPU and can propagate. */ | |
2086 | rcu_cleanup_dead_rnp(rnp); | |
2087 | } | |
2088 | ||
2089 | /* | |
2090 | * If all waited-on tasks from prior grace period are | |
2091 | * done, and if all this rcu_node structure's CPUs are | |
2092 | * still offline, propagate up the rcu_node tree and | |
2093 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
2094 | * rcu_node structure's CPUs has since come back online, | |
2095 | * simply clear ->wait_blkd_tasks (but rcu_cleanup_dead_rnp() | |
2096 | * checks for this, so just call it unconditionally). | |
2097 | */ | |
2098 | if (rnp->wait_blkd_tasks && | |
2099 | (!rcu_preempt_has_tasks(rnp) || | |
2100 | rnp->qsmaskinit)) { | |
2101 | rnp->wait_blkd_tasks = false; | |
2102 | rcu_cleanup_dead_rnp(rnp); | |
2103 | } | |
2104 | ||
67c583a7 | 2105 | raw_spin_unlock_irq_rcu_node(rnp); |
0aa04b05 | 2106 | } |
7fdefc10 PM |
2107 | |
2108 | /* | |
2109 | * Set the quiescent-state-needed bits in all the rcu_node | |
2110 | * structures for all currently online CPUs in breadth-first order, | |
2111 | * starting from the root rcu_node structure, relying on the layout | |
2112 | * of the tree within the rsp->node[] array. Note that other CPUs | |
2113 | * will access only the leaves of the hierarchy, thus seeing that no | |
2114 | * grace period is in progress, at least until the corresponding | |
590d1757 | 2115 | * leaf node has been initialized. |
7fdefc10 PM |
2116 | * |
2117 | * The grace period cannot complete until the initialization | |
2118 | * process finishes, because this kthread handles both. | |
2119 | */ | |
2120 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
0f41c0dd | 2121 | rcu_gp_slow(rsp, gp_init_delay); |
2a67e741 | 2122 | raw_spin_lock_irq_rcu_node(rnp); |
b3dbec76 | 2123 | rdp = this_cpu_ptr(rsp->rda); |
7fdefc10 PM |
2124 | rcu_preempt_check_blocked_tasks(rnp); |
2125 | rnp->qsmask = rnp->qsmaskinit; | |
7d0ae808 | 2126 | WRITE_ONCE(rnp->gpnum, rsp->gpnum); |
3f47da0f | 2127 | if (WARN_ON_ONCE(rnp->completed != rsp->completed)) |
7d0ae808 | 2128 | WRITE_ONCE(rnp->completed, rsp->completed); |
7fdefc10 | 2129 | if (rnp == rdp->mynode) |
48a7639c | 2130 | (void)__note_gp_changes(rsp, rnp, rdp); |
7fdefc10 PM |
2131 | rcu_preempt_boost_start_gp(rnp); |
2132 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, | |
2133 | rnp->level, rnp->grplo, | |
2134 | rnp->grphi, rnp->qsmask); | |
67c583a7 | 2135 | raw_spin_unlock_irq_rcu_node(rnp); |
bde6c3aa | 2136 | cond_resched_rcu_qs(); |
7d0ae808 | 2137 | WRITE_ONCE(rsp->gp_activity, jiffies); |
7fdefc10 | 2138 | } |
b3dbec76 | 2139 | |
45fed3e7 | 2140 | return true; |
7fdefc10 | 2141 | } |
b3dbec76 | 2142 | |
b9a425cf PM |
2143 | /* |
2144 | * Helper function for wait_event_interruptible_timeout() wakeup | |
2145 | * at force-quiescent-state time. | |
2146 | */ | |
2147 | static bool rcu_gp_fqs_check_wake(struct rcu_state *rsp, int *gfp) | |
2148 | { | |
2149 | struct rcu_node *rnp = rcu_get_root(rsp); | |
2150 | ||
2151 | /* Someone like call_rcu() requested a force-quiescent-state scan. */ | |
2152 | *gfp = READ_ONCE(rsp->gp_flags); | |
2153 | if (*gfp & RCU_GP_FLAG_FQS) | |
2154 | return true; | |
2155 | ||
2156 | /* The current grace period has completed. */ | |
2157 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) | |
2158 | return true; | |
2159 | ||
2160 | return false; | |
2161 | } | |
2162 | ||
4cdfc175 PM |
2163 | /* |
2164 | * Do one round of quiescent-state forcing. | |
2165 | */ | |
77f81fe0 | 2166 | static void rcu_gp_fqs(struct rcu_state *rsp, bool first_time) |
4cdfc175 | 2167 | { |
217af2a2 PM |
2168 | bool isidle = false; |
2169 | unsigned long maxj; | |
4cdfc175 PM |
2170 | struct rcu_node *rnp = rcu_get_root(rsp); |
2171 | ||
7d0ae808 | 2172 | WRITE_ONCE(rsp->gp_activity, jiffies); |
4cdfc175 | 2173 | rsp->n_force_qs++; |
77f81fe0 | 2174 | if (first_time) { |
4cdfc175 | 2175 | /* Collect dyntick-idle snapshots. */ |
0edd1b17 | 2176 | if (is_sysidle_rcu_state(rsp)) { |
e02b2edf | 2177 | isidle = true; |
0edd1b17 PM |
2178 | maxj = jiffies - ULONG_MAX / 4; |
2179 | } | |
217af2a2 PM |
2180 | force_qs_rnp(rsp, dyntick_save_progress_counter, |
2181 | &isidle, &maxj); | |
0edd1b17 | 2182 | rcu_sysidle_report_gp(rsp, isidle, maxj); |
4cdfc175 PM |
2183 | } else { |
2184 | /* Handle dyntick-idle and offline CPUs. */ | |
675da67f | 2185 | isidle = true; |
217af2a2 | 2186 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs, &isidle, &maxj); |
4cdfc175 PM |
2187 | } |
2188 | /* Clear flag to prevent immediate re-entry. */ | |
7d0ae808 | 2189 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 2190 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 PM |
2191 | WRITE_ONCE(rsp->gp_flags, |
2192 | READ_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 2193 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 2194 | } |
4cdfc175 PM |
2195 | } |
2196 | ||
7fdefc10 PM |
2197 | /* |
2198 | * Clean up after the old grace period. | |
2199 | */ | |
4cdfc175 | 2200 | static void rcu_gp_cleanup(struct rcu_state *rsp) |
7fdefc10 PM |
2201 | { |
2202 | unsigned long gp_duration; | |
48a7639c | 2203 | bool needgp = false; |
dae6e64d | 2204 | int nocb = 0; |
7fdefc10 PM |
2205 | struct rcu_data *rdp; |
2206 | struct rcu_node *rnp = rcu_get_root(rsp); | |
abedf8e2 | 2207 | struct swait_queue_head *sq; |
b3dbec76 | 2208 | |
7d0ae808 | 2209 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 2210 | raw_spin_lock_irq_rcu_node(rnp); |
7fdefc10 PM |
2211 | gp_duration = jiffies - rsp->gp_start; |
2212 | if (gp_duration > rsp->gp_max) | |
2213 | rsp->gp_max = gp_duration; | |
b3dbec76 | 2214 | |
7fdefc10 PM |
2215 | /* |
2216 | * We know the grace period is complete, but to everyone else | |
2217 | * it appears to still be ongoing. But it is also the case | |
2218 | * that to everyone else it looks like there is nothing that | |
2219 | * they can do to advance the grace period. It is therefore | |
2220 | * safe for us to drop the lock in order to mark the grace | |
2221 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 2222 | */ |
67c583a7 | 2223 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 2224 | |
5d4b8659 PM |
2225 | /* |
2226 | * Propagate new ->completed value to rcu_node structures so | |
2227 | * that other CPUs don't have to wait until the start of the next | |
2228 | * grace period to process their callbacks. This also avoids | |
2229 | * some nasty RCU grace-period initialization races by forcing | |
2230 | * the end of the current grace period to be completely recorded in | |
2231 | * all of the rcu_node structures before the beginning of the next | |
2232 | * grace period is recorded in any of the rcu_node structures. | |
2233 | */ | |
2234 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
2a67e741 | 2235 | raw_spin_lock_irq_rcu_node(rnp); |
5c60d25f PM |
2236 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)); |
2237 | WARN_ON_ONCE(rnp->qsmask); | |
7d0ae808 | 2238 | WRITE_ONCE(rnp->completed, rsp->gpnum); |
b11cc576 PM |
2239 | rdp = this_cpu_ptr(rsp->rda); |
2240 | if (rnp == rdp->mynode) | |
48a7639c | 2241 | needgp = __note_gp_changes(rsp, rnp, rdp) || needgp; |
78e4bc34 | 2242 | /* smp_mb() provided by prior unlock-lock pair. */ |
0446be48 | 2243 | nocb += rcu_future_gp_cleanup(rsp, rnp); |
065bb78c | 2244 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 2245 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 2246 | rcu_nocb_gp_cleanup(sq); |
bde6c3aa | 2247 | cond_resched_rcu_qs(); |
7d0ae808 | 2248 | WRITE_ONCE(rsp->gp_activity, jiffies); |
0f41c0dd | 2249 | rcu_gp_slow(rsp, gp_cleanup_delay); |
7fdefc10 | 2250 | } |
5d4b8659 | 2251 | rnp = rcu_get_root(rsp); |
2a67e741 | 2252 | raw_spin_lock_irq_rcu_node(rnp); /* Order GP before ->completed update. */ |
dae6e64d | 2253 | rcu_nocb_gp_set(rnp, nocb); |
7fdefc10 | 2254 | |
765a3f4f | 2255 | /* Declare grace period done. */ |
7d0ae808 | 2256 | WRITE_ONCE(rsp->completed, rsp->gpnum); |
f7f7bac9 | 2257 | trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end")); |
77f81fe0 | 2258 | rsp->gp_state = RCU_GP_IDLE; |
5d4b8659 | 2259 | rdp = this_cpu_ptr(rsp->rda); |
48a7639c PM |
2260 | /* Advance CBs to reduce false positives below. */ |
2261 | needgp = rcu_advance_cbs(rsp, rnp, rdp) || needgp; | |
2262 | if (needgp || cpu_needs_another_gp(rsp, rdp)) { | |
7d0ae808 | 2263 | WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT); |
bb311ecc | 2264 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2265 | READ_ONCE(rsp->gpnum), |
bb311ecc PM |
2266 | TPS("newreq")); |
2267 | } | |
67c583a7 | 2268 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 PM |
2269 | } |
2270 | ||
2271 | /* | |
2272 | * Body of kthread that handles grace periods. | |
2273 | */ | |
2274 | static int __noreturn rcu_gp_kthread(void *arg) | |
2275 | { | |
77f81fe0 | 2276 | bool first_gp_fqs; |
88d6df61 | 2277 | int gf; |
d40011f6 | 2278 | unsigned long j; |
4cdfc175 | 2279 | int ret; |
7fdefc10 PM |
2280 | struct rcu_state *rsp = arg; |
2281 | struct rcu_node *rnp = rcu_get_root(rsp); | |
2282 | ||
5871968d | 2283 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
2284 | for (;;) { |
2285 | ||
2286 | /* Handle grace-period start. */ | |
2287 | for (;;) { | |
63c4db78 | 2288 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2289 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2290 | TPS("reqwait")); |
afea227f | 2291 | rsp->gp_state = RCU_GP_WAIT_GPS; |
abedf8e2 | 2292 | swait_event_interruptible(rsp->gp_wq, |
7d0ae808 | 2293 | READ_ONCE(rsp->gp_flags) & |
4cdfc175 | 2294 | RCU_GP_FLAG_INIT); |
319362c9 | 2295 | rsp->gp_state = RCU_GP_DONE_GPS; |
78e4bc34 | 2296 | /* Locking provides needed memory barrier. */ |
f7be8209 | 2297 | if (rcu_gp_init(rsp)) |
7fdefc10 | 2298 | break; |
bde6c3aa | 2299 | cond_resched_rcu_qs(); |
7d0ae808 | 2300 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2301 | WARN_ON(signal_pending(current)); |
63c4db78 | 2302 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2303 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2304 | TPS("reqwaitsig")); |
7fdefc10 | 2305 | } |
cabc49c1 | 2306 | |
4cdfc175 | 2307 | /* Handle quiescent-state forcing. */ |
77f81fe0 | 2308 | first_gp_fqs = true; |
d40011f6 PM |
2309 | j = jiffies_till_first_fqs; |
2310 | if (j > HZ) { | |
2311 | j = HZ; | |
2312 | jiffies_till_first_fqs = HZ; | |
2313 | } | |
88d6df61 | 2314 | ret = 0; |
cabc49c1 | 2315 | for (;;) { |
8c7c4829 | 2316 | if (!ret) { |
88d6df61 | 2317 | rsp->jiffies_force_qs = jiffies + j; |
8c7c4829 PM |
2318 | WRITE_ONCE(rsp->jiffies_kick_kthreads, |
2319 | jiffies + 3 * j); | |
2320 | } | |
63c4db78 | 2321 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2322 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2323 | TPS("fqswait")); |
afea227f | 2324 | rsp->gp_state = RCU_GP_WAIT_FQS; |
abedf8e2 | 2325 | ret = swait_event_interruptible_timeout(rsp->gp_wq, |
b9a425cf | 2326 | rcu_gp_fqs_check_wake(rsp, &gf), j); |
32bb1c79 | 2327 | rsp->gp_state = RCU_GP_DOING_FQS; |
78e4bc34 | 2328 | /* Locking provides needed memory barriers. */ |
4cdfc175 | 2329 | /* If grace period done, leave loop. */ |
7d0ae808 | 2330 | if (!READ_ONCE(rnp->qsmask) && |
4cdfc175 | 2331 | !rcu_preempt_blocked_readers_cgp(rnp)) |
cabc49c1 | 2332 | break; |
4cdfc175 | 2333 | /* If time for quiescent-state forcing, do it. */ |
88d6df61 PM |
2334 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) || |
2335 | (gf & RCU_GP_FLAG_FQS)) { | |
63c4db78 | 2336 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2337 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2338 | TPS("fqsstart")); |
77f81fe0 PM |
2339 | rcu_gp_fqs(rsp, first_gp_fqs); |
2340 | first_gp_fqs = false; | |
63c4db78 | 2341 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2342 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2343 | TPS("fqsend")); |
bde6c3aa | 2344 | cond_resched_rcu_qs(); |
7d0ae808 | 2345 | WRITE_ONCE(rsp->gp_activity, jiffies); |
fcfd0a23 PM |
2346 | ret = 0; /* Force full wait till next FQS. */ |
2347 | j = jiffies_till_next_fqs; | |
2348 | if (j > HZ) { | |
2349 | j = HZ; | |
2350 | jiffies_till_next_fqs = HZ; | |
2351 | } else if (j < 1) { | |
2352 | j = 1; | |
2353 | jiffies_till_next_fqs = 1; | |
2354 | } | |
4cdfc175 PM |
2355 | } else { |
2356 | /* Deal with stray signal. */ | |
bde6c3aa | 2357 | cond_resched_rcu_qs(); |
7d0ae808 | 2358 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2359 | WARN_ON(signal_pending(current)); |
63c4db78 | 2360 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2361 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2362 | TPS("fqswaitsig")); |
fcfd0a23 PM |
2363 | ret = 1; /* Keep old FQS timing. */ |
2364 | j = jiffies; | |
2365 | if (time_after(jiffies, rsp->jiffies_force_qs)) | |
2366 | j = 1; | |
2367 | else | |
2368 | j = rsp->jiffies_force_qs - j; | |
d40011f6 | 2369 | } |
cabc49c1 | 2370 | } |
4cdfc175 PM |
2371 | |
2372 | /* Handle grace-period end. */ | |
319362c9 | 2373 | rsp->gp_state = RCU_GP_CLEANUP; |
4cdfc175 | 2374 | rcu_gp_cleanup(rsp); |
319362c9 | 2375 | rsp->gp_state = RCU_GP_CLEANED; |
b3dbec76 | 2376 | } |
b3dbec76 PM |
2377 | } |
2378 | ||
64db4cff PM |
2379 | /* |
2380 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
2381 | * in preparation for detecting the next grace period. The caller must hold | |
b8462084 | 2382 | * the root node's ->lock and hard irqs must be disabled. |
e5601400 PM |
2383 | * |
2384 | * Note that it is legal for a dying CPU (which is marked as offline) to | |
2385 | * invoke this function. This can happen when the dying CPU reports its | |
2386 | * quiescent state. | |
48a7639c PM |
2387 | * |
2388 | * Returns true if the grace-period kthread must be awakened. | |
64db4cff | 2389 | */ |
48a7639c | 2390 | static bool |
910ee45d PM |
2391 | rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp, |
2392 | struct rcu_data *rdp) | |
64db4cff | 2393 | { |
b8462084 | 2394 | if (!rsp->gp_kthread || !cpu_needs_another_gp(rsp, rdp)) { |
afe24b12 | 2395 | /* |
b3dbec76 | 2396 | * Either we have not yet spawned the grace-period |
62da1921 PM |
2397 | * task, this CPU does not need another grace period, |
2398 | * or a grace period is already in progress. | |
b3dbec76 | 2399 | * Either way, don't start a new grace period. |
afe24b12 | 2400 | */ |
48a7639c | 2401 | return false; |
afe24b12 | 2402 | } |
7d0ae808 PM |
2403 | WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT); |
2404 | trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gpnum), | |
bb311ecc | 2405 | TPS("newreq")); |
62da1921 | 2406 | |
016a8d5b SR |
2407 | /* |
2408 | * We can't do wakeups while holding the rnp->lock, as that | |
1eafd31c | 2409 | * could cause possible deadlocks with the rq->lock. Defer |
48a7639c | 2410 | * the wakeup to our caller. |
016a8d5b | 2411 | */ |
48a7639c | 2412 | return true; |
64db4cff PM |
2413 | } |
2414 | ||
910ee45d PM |
2415 | /* |
2416 | * Similar to rcu_start_gp_advanced(), but also advance the calling CPU's | |
2417 | * callbacks. Note that rcu_start_gp_advanced() cannot do this because it | |
2418 | * is invoked indirectly from rcu_advance_cbs(), which would result in | |
2419 | * endless recursion -- or would do so if it wasn't for the self-deadlock | |
2420 | * that is encountered beforehand. | |
48a7639c PM |
2421 | * |
2422 | * Returns true if the grace-period kthread needs to be awakened. | |
910ee45d | 2423 | */ |
48a7639c | 2424 | static bool rcu_start_gp(struct rcu_state *rsp) |
910ee45d PM |
2425 | { |
2426 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
2427 | struct rcu_node *rnp = rcu_get_root(rsp); | |
48a7639c | 2428 | bool ret = false; |
910ee45d PM |
2429 | |
2430 | /* | |
2431 | * If there is no grace period in progress right now, any | |
2432 | * callbacks we have up to this point will be satisfied by the | |
2433 | * next grace period. Also, advancing the callbacks reduces the | |
2434 | * probability of false positives from cpu_needs_another_gp() | |
2435 | * resulting in pointless grace periods. So, advance callbacks | |
2436 | * then start the grace period! | |
2437 | */ | |
48a7639c PM |
2438 | ret = rcu_advance_cbs(rsp, rnp, rdp) || ret; |
2439 | ret = rcu_start_gp_advanced(rsp, rnp, rdp) || ret; | |
2440 | return ret; | |
910ee45d PM |
2441 | } |
2442 | ||
f41d911f | 2443 | /* |
8994515c PM |
2444 | * Report a full set of quiescent states to the specified rcu_state data |
2445 | * structure. Invoke rcu_gp_kthread_wake() to awaken the grace-period | |
2446 | * kthread if another grace period is required. Whether we wake | |
2447 | * the grace-period kthread or it awakens itself for the next round | |
2448 | * of quiescent-state forcing, that kthread will clean up after the | |
2449 | * just-completed grace period. Note that the caller must hold rnp->lock, | |
2450 | * which is released before return. | |
f41d911f | 2451 | */ |
d3f6bad3 | 2452 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 2453 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 2454 | { |
fc2219d4 | 2455 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
cd73ca21 | 2456 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 2457 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags); |
94d44776 | 2458 | rcu_gp_kthread_wake(rsp); |
f41d911f PM |
2459 | } |
2460 | ||
64db4cff | 2461 | /* |
d3f6bad3 PM |
2462 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
2463 | * Allows quiescent states for a group of CPUs to be reported at one go | |
2464 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
2465 | * must be represented by the same rcu_node structure (which need not be a |
2466 | * leaf rcu_node structure, though it often will be). The gps parameter | |
2467 | * is the grace-period snapshot, which means that the quiescent states | |
2468 | * are valid only if rnp->gpnum is equal to gps. That structure's lock | |
2469 | * must be held upon entry, and it is released before return. | |
64db4cff PM |
2470 | */ |
2471 | static void | |
d3f6bad3 | 2472 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
654e9533 | 2473 | struct rcu_node *rnp, unsigned long gps, unsigned long flags) |
64db4cff PM |
2474 | __releases(rnp->lock) |
2475 | { | |
654e9533 | 2476 | unsigned long oldmask = 0; |
28ecd580 PM |
2477 | struct rcu_node *rnp_c; |
2478 | ||
64db4cff PM |
2479 | /* Walk up the rcu_node hierarchy. */ |
2480 | for (;;) { | |
654e9533 | 2481 | if (!(rnp->qsmask & mask) || rnp->gpnum != gps) { |
64db4cff | 2482 | |
654e9533 PM |
2483 | /* |
2484 | * Our bit has already been cleared, or the | |
2485 | * relevant grace period is already over, so done. | |
2486 | */ | |
67c583a7 | 2487 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2488 | return; |
2489 | } | |
654e9533 | 2490 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
64db4cff | 2491 | rnp->qsmask &= ~mask; |
d4c08f2a PM |
2492 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
2493 | mask, rnp->qsmask, rnp->level, | |
2494 | rnp->grplo, rnp->grphi, | |
2495 | !!rnp->gp_tasks); | |
27f4d280 | 2496 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
2497 | |
2498 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 2499 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2500 | return; |
2501 | } | |
2502 | mask = rnp->grpmask; | |
2503 | if (rnp->parent == NULL) { | |
2504 | ||
2505 | /* No more levels. Exit loop holding root lock. */ | |
2506 | ||
2507 | break; | |
2508 | } | |
67c583a7 | 2509 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 2510 | rnp_c = rnp; |
64db4cff | 2511 | rnp = rnp->parent; |
2a67e741 | 2512 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
654e9533 | 2513 | oldmask = rnp_c->qsmask; |
64db4cff PM |
2514 | } |
2515 | ||
2516 | /* | |
2517 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 2518 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 2519 | * to clean up and start the next grace period if one is needed. |
64db4cff | 2520 | */ |
d3f6bad3 | 2521 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
2522 | } |
2523 | ||
cc99a310 PM |
2524 | /* |
2525 | * Record a quiescent state for all tasks that were previously queued | |
2526 | * on the specified rcu_node structure and that were blocking the current | |
2527 | * RCU grace period. The caller must hold the specified rnp->lock with | |
2528 | * irqs disabled, and this lock is released upon return, but irqs remain | |
2529 | * disabled. | |
2530 | */ | |
0aa04b05 | 2531 | static void rcu_report_unblock_qs_rnp(struct rcu_state *rsp, |
cc99a310 PM |
2532 | struct rcu_node *rnp, unsigned long flags) |
2533 | __releases(rnp->lock) | |
2534 | { | |
654e9533 | 2535 | unsigned long gps; |
cc99a310 PM |
2536 | unsigned long mask; |
2537 | struct rcu_node *rnp_p; | |
2538 | ||
a77da14c PM |
2539 | if (rcu_state_p == &rcu_sched_state || rsp != rcu_state_p || |
2540 | rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { | |
67c583a7 | 2541 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
2542 | return; /* Still need more quiescent states! */ |
2543 | } | |
2544 | ||
2545 | rnp_p = rnp->parent; | |
2546 | if (rnp_p == NULL) { | |
2547 | /* | |
a77da14c PM |
2548 | * Only one rcu_node structure in the tree, so don't |
2549 | * try to report up to its nonexistent parent! | |
cc99a310 PM |
2550 | */ |
2551 | rcu_report_qs_rsp(rsp, flags); | |
2552 | return; | |
2553 | } | |
2554 | ||
654e9533 PM |
2555 | /* Report up the rest of the hierarchy, tracking current ->gpnum. */ |
2556 | gps = rnp->gpnum; | |
cc99a310 | 2557 | mask = rnp->grpmask; |
67c583a7 | 2558 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 2559 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
654e9533 | 2560 | rcu_report_qs_rnp(mask, rsp, rnp_p, gps, flags); |
cc99a310 PM |
2561 | } |
2562 | ||
64db4cff | 2563 | /* |
d3f6bad3 | 2564 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 2565 | * structure. This must be called from the specified CPU. |
64db4cff PM |
2566 | */ |
2567 | static void | |
d7d6a11e | 2568 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2569 | { |
2570 | unsigned long flags; | |
2571 | unsigned long mask; | |
48a7639c | 2572 | bool needwake; |
64db4cff PM |
2573 | struct rcu_node *rnp; |
2574 | ||
2575 | rnp = rdp->mynode; | |
2a67e741 | 2576 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
3a19b46a PM |
2577 | if (rdp->cpu_no_qs.b.norm || rdp->gpnum != rnp->gpnum || |
2578 | rnp->completed == rnp->gpnum || rdp->gpwrap) { | |
64db4cff PM |
2579 | |
2580 | /* | |
e4cc1f22 PM |
2581 | * The grace period in which this quiescent state was |
2582 | * recorded has ended, so don't report it upwards. | |
2583 | * We will instead need a new quiescent state that lies | |
2584 | * within the current grace period. | |
64db4cff | 2585 | */ |
5b74c458 | 2586 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
9577df9a | 2587 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
67c583a7 | 2588 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2589 | return; |
2590 | } | |
2591 | mask = rdp->grpmask; | |
2592 | if ((rnp->qsmask & mask) == 0) { | |
67c583a7 | 2593 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2594 | } else { |
bb53e416 | 2595 | rdp->core_needs_qs = false; |
64db4cff PM |
2596 | |
2597 | /* | |
2598 | * This GP can't end until cpu checks in, so all of our | |
2599 | * callbacks can be processed during the next GP. | |
2600 | */ | |
48a7639c | 2601 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
64db4cff | 2602 | |
654e9533 PM |
2603 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags); |
2604 | /* ^^^ Released rnp->lock */ | |
48a7639c PM |
2605 | if (needwake) |
2606 | rcu_gp_kthread_wake(rsp); | |
64db4cff PM |
2607 | } |
2608 | } | |
2609 | ||
2610 | /* | |
2611 | * Check to see if there is a new grace period of which this CPU | |
2612 | * is not yet aware, and if so, set up local rcu_data state for it. | |
2613 | * Otherwise, see if this CPU has just passed through its first | |
2614 | * quiescent state for this grace period, and record that fact if so. | |
2615 | */ | |
2616 | static void | |
2617 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
2618 | { | |
05eb552b PM |
2619 | /* Check for grace-period ends and beginnings. */ |
2620 | note_gp_changes(rsp, rdp); | |
64db4cff PM |
2621 | |
2622 | /* | |
2623 | * Does this CPU still need to do its part for current grace period? | |
2624 | * If no, return and let the other CPUs do their part as well. | |
2625 | */ | |
97c668b8 | 2626 | if (!rdp->core_needs_qs) |
64db4cff PM |
2627 | return; |
2628 | ||
2629 | /* | |
2630 | * Was there a quiescent state since the beginning of the grace | |
2631 | * period? If no, then exit and wait for the next call. | |
2632 | */ | |
3a19b46a | 2633 | if (rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2634 | return; |
2635 | ||
d3f6bad3 PM |
2636 | /* |
2637 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
2638 | * judge of that). | |
2639 | */ | |
d7d6a11e | 2640 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp); |
64db4cff PM |
2641 | } |
2642 | ||
e74f4c45 | 2643 | /* |
b1420f1c PM |
2644 | * Send the specified CPU's RCU callbacks to the orphanage. The |
2645 | * specified CPU must be offline, and the caller must hold the | |
7b2e6011 | 2646 | * ->orphan_lock. |
e74f4c45 | 2647 | */ |
b1420f1c PM |
2648 | static void |
2649 | rcu_send_cbs_to_orphanage(int cpu, struct rcu_state *rsp, | |
2650 | struct rcu_node *rnp, struct rcu_data *rdp) | |
e74f4c45 | 2651 | { |
3fbfbf7a | 2652 | /* No-CBs CPUs do not have orphanable callbacks. */ |
ea46351c | 2653 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || rcu_is_nocb_cpu(rdp->cpu)) |
3fbfbf7a PM |
2654 | return; |
2655 | ||
b1420f1c PM |
2656 | /* |
2657 | * Orphan the callbacks. First adjust the counts. This is safe | |
abfd6e58 PM |
2658 | * because _rcu_barrier() excludes CPU-hotplug operations, so it |
2659 | * cannot be running now. Thus no memory barrier is required. | |
b1420f1c | 2660 | */ |
a50c3af9 | 2661 | if (rdp->nxtlist != NULL) { |
b1420f1c PM |
2662 | rsp->qlen_lazy += rdp->qlen_lazy; |
2663 | rsp->qlen += rdp->qlen; | |
2664 | rdp->n_cbs_orphaned += rdp->qlen; | |
a50c3af9 | 2665 | rdp->qlen_lazy = 0; |
7d0ae808 | 2666 | WRITE_ONCE(rdp->qlen, 0); |
a50c3af9 PM |
2667 | } |
2668 | ||
2669 | /* | |
b1420f1c PM |
2670 | * Next, move those callbacks still needing a grace period to |
2671 | * the orphanage, where some other CPU will pick them up. | |
2672 | * Some of the callbacks might have gone partway through a grace | |
2673 | * period, but that is too bad. They get to start over because we | |
2674 | * cannot assume that grace periods are synchronized across CPUs. | |
2675 | * We don't bother updating the ->nxttail[] array yet, instead | |
2676 | * we just reset the whole thing later on. | |
a50c3af9 | 2677 | */ |
b1420f1c PM |
2678 | if (*rdp->nxttail[RCU_DONE_TAIL] != NULL) { |
2679 | *rsp->orphan_nxttail = *rdp->nxttail[RCU_DONE_TAIL]; | |
2680 | rsp->orphan_nxttail = rdp->nxttail[RCU_NEXT_TAIL]; | |
2681 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
a50c3af9 PM |
2682 | } |
2683 | ||
2684 | /* | |
b1420f1c PM |
2685 | * Then move the ready-to-invoke callbacks to the orphanage, |
2686 | * where some other CPU will pick them up. These will not be | |
2687 | * required to pass though another grace period: They are done. | |
a50c3af9 | 2688 | */ |
e5601400 | 2689 | if (rdp->nxtlist != NULL) { |
b1420f1c PM |
2690 | *rsp->orphan_donetail = rdp->nxtlist; |
2691 | rsp->orphan_donetail = rdp->nxttail[RCU_DONE_TAIL]; | |
e5601400 | 2692 | } |
e74f4c45 | 2693 | |
b33078b6 PM |
2694 | /* |
2695 | * Finally, initialize the rcu_data structure's list to empty and | |
2696 | * disallow further callbacks on this CPU. | |
2697 | */ | |
3f5d3ea6 | 2698 | init_callback_list(rdp); |
b33078b6 | 2699 | rdp->nxttail[RCU_NEXT_TAIL] = NULL; |
b1420f1c PM |
2700 | } |
2701 | ||
2702 | /* | |
2703 | * Adopt the RCU callbacks from the specified rcu_state structure's | |
7b2e6011 | 2704 | * orphanage. The caller must hold the ->orphan_lock. |
b1420f1c | 2705 | */ |
96d3fd0d | 2706 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags) |
b1420f1c PM |
2707 | { |
2708 | int i; | |
fa07a58f | 2709 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
b1420f1c | 2710 | |
3fbfbf7a | 2711 | /* No-CBs CPUs are handled specially. */ |
ea46351c PM |
2712 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
2713 | rcu_nocb_adopt_orphan_cbs(rsp, rdp, flags)) | |
3fbfbf7a PM |
2714 | return; |
2715 | ||
b1420f1c PM |
2716 | /* Do the accounting first. */ |
2717 | rdp->qlen_lazy += rsp->qlen_lazy; | |
2718 | rdp->qlen += rsp->qlen; | |
2719 | rdp->n_cbs_adopted += rsp->qlen; | |
8f5af6f1 PM |
2720 | if (rsp->qlen_lazy != rsp->qlen) |
2721 | rcu_idle_count_callbacks_posted(); | |
b1420f1c PM |
2722 | rsp->qlen_lazy = 0; |
2723 | rsp->qlen = 0; | |
2724 | ||
2725 | /* | |
2726 | * We do not need a memory barrier here because the only way we | |
2727 | * can get here if there is an rcu_barrier() in flight is if | |
2728 | * we are the task doing the rcu_barrier(). | |
2729 | */ | |
2730 | ||
2731 | /* First adopt the ready-to-invoke callbacks. */ | |
2732 | if (rsp->orphan_donelist != NULL) { | |
2733 | *rsp->orphan_donetail = *rdp->nxttail[RCU_DONE_TAIL]; | |
2734 | *rdp->nxttail[RCU_DONE_TAIL] = rsp->orphan_donelist; | |
2735 | for (i = RCU_NEXT_SIZE - 1; i >= RCU_DONE_TAIL; i--) | |
2736 | if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL]) | |
2737 | rdp->nxttail[i] = rsp->orphan_donetail; | |
2738 | rsp->orphan_donelist = NULL; | |
2739 | rsp->orphan_donetail = &rsp->orphan_donelist; | |
2740 | } | |
2741 | ||
2742 | /* And then adopt the callbacks that still need a grace period. */ | |
2743 | if (rsp->orphan_nxtlist != NULL) { | |
2744 | *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxtlist; | |
2745 | rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxttail; | |
2746 | rsp->orphan_nxtlist = NULL; | |
2747 | rsp->orphan_nxttail = &rsp->orphan_nxtlist; | |
2748 | } | |
2749 | } | |
2750 | ||
2751 | /* | |
2752 | * Trace the fact that this CPU is going offline. | |
2753 | */ | |
2754 | static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) | |
2755 | { | |
88a4976d PM |
2756 | RCU_TRACE(unsigned long mask;) |
2757 | RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda);) | |
2758 | RCU_TRACE(struct rcu_node *rnp = rdp->mynode;) | |
b1420f1c | 2759 | |
ea46351c PM |
2760 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2761 | return; | |
2762 | ||
88a4976d | 2763 | RCU_TRACE(mask = rdp->grpmask;) |
e5601400 PM |
2764 | trace_rcu_grace_period(rsp->name, |
2765 | rnp->gpnum + 1 - !!(rnp->qsmask & mask), | |
f7f7bac9 | 2766 | TPS("cpuofl")); |
64db4cff PM |
2767 | } |
2768 | ||
8af3a5e7 PM |
2769 | /* |
2770 | * All CPUs for the specified rcu_node structure have gone offline, | |
2771 | * and all tasks that were preempted within an RCU read-side critical | |
2772 | * section while running on one of those CPUs have since exited their RCU | |
2773 | * read-side critical section. Some other CPU is reporting this fact with | |
2774 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2775 | * This function therefore goes up the tree of rcu_node structures, | |
2776 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2777 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
2778 | * updated | |
2779 | * | |
2780 | * This function does check that the specified rcu_node structure has | |
2781 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2782 | * prematurely. That said, invoking it after the fact will cost you | |
2783 | * a needless lock acquisition. So once it has done its work, don't | |
2784 | * invoke it again. | |
2785 | */ | |
2786 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2787 | { | |
2788 | long mask; | |
2789 | struct rcu_node *rnp = rnp_leaf; | |
2790 | ||
ea46351c PM |
2791 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
2792 | rnp->qsmaskinit || rcu_preempt_has_tasks(rnp)) | |
8af3a5e7 PM |
2793 | return; |
2794 | for (;;) { | |
2795 | mask = rnp->grpmask; | |
2796 | rnp = rnp->parent; | |
2797 | if (!rnp) | |
2798 | break; | |
2a67e741 | 2799 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2800 | rnp->qsmaskinit &= ~mask; |
0aa04b05 | 2801 | rnp->qsmask &= ~mask; |
8af3a5e7 | 2802 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2803 | raw_spin_unlock_rcu_node(rnp); |
2804 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2805 | return; |
2806 | } | |
67c583a7 | 2807 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2808 | } |
2809 | } | |
2810 | ||
64db4cff | 2811 | /* |
e5601400 | 2812 | * The CPU has been completely removed, and some other CPU is reporting |
b1420f1c PM |
2813 | * this fact from process context. Do the remainder of the cleanup, |
2814 | * including orphaning the outgoing CPU's RCU callbacks, and also | |
1331e7a1 PM |
2815 | * adopting them. There can only be one CPU hotplug operation at a time, |
2816 | * so no other CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff | 2817 | */ |
e5601400 | 2818 | static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) |
64db4cff | 2819 | { |
2036d94a | 2820 | unsigned long flags; |
e5601400 | 2821 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
b1420f1c | 2822 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2823 | |
ea46351c PM |
2824 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2825 | return; | |
2826 | ||
2036d94a | 2827 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2828 | rcu_boost_kthread_setaffinity(rnp, -1); |
2036d94a | 2829 | |
b1420f1c | 2830 | /* Orphan the dead CPU's callbacks, and adopt them if appropriate. */ |
78043c46 | 2831 | raw_spin_lock_irqsave(&rsp->orphan_lock, flags); |
b1420f1c | 2832 | rcu_send_cbs_to_orphanage(cpu, rsp, rnp, rdp); |
96d3fd0d | 2833 | rcu_adopt_orphan_cbs(rsp, flags); |
a8f4cbad | 2834 | raw_spin_unlock_irqrestore(&rsp->orphan_lock, flags); |
b1420f1c | 2835 | |
cf01537e PM |
2836 | WARN_ONCE(rdp->qlen != 0 || rdp->nxtlist != NULL, |
2837 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, nxtlist=%p\n", | |
2838 | cpu, rdp->qlen, rdp->nxtlist); | |
64db4cff PM |
2839 | } |
2840 | ||
64db4cff PM |
2841 | /* |
2842 | * Invoke any RCU callbacks that have made it to the end of their grace | |
2843 | * period. Thottle as specified by rdp->blimit. | |
2844 | */ | |
37c72e56 | 2845 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2846 | { |
2847 | unsigned long flags; | |
2848 | struct rcu_head *next, *list, **tail; | |
878d7439 ED |
2849 | long bl, count, count_lazy; |
2850 | int i; | |
64db4cff | 2851 | |
dc35c893 | 2852 | /* If no callbacks are ready, just return. */ |
29c00b4a | 2853 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) { |
486e2593 | 2854 | trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0); |
7d0ae808 | 2855 | trace_rcu_batch_end(rsp->name, 0, !!READ_ONCE(rdp->nxtlist), |
4968c300 PM |
2856 | need_resched(), is_idle_task(current), |
2857 | rcu_is_callbacks_kthread()); | |
64db4cff | 2858 | return; |
29c00b4a | 2859 | } |
64db4cff PM |
2860 | |
2861 | /* | |
2862 | * Extract the list of ready callbacks, disabling to prevent | |
2863 | * races with call_rcu() from interrupt handlers. | |
2864 | */ | |
2865 | local_irq_save(flags); | |
8146c4e2 | 2866 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 2867 | bl = rdp->blimit; |
486e2593 | 2868 | trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl); |
64db4cff PM |
2869 | list = rdp->nxtlist; |
2870 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
2871 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
2872 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
b41772ab PM |
2873 | for (i = RCU_NEXT_SIZE - 1; i >= 0; i--) |
2874 | if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL]) | |
2875 | rdp->nxttail[i] = &rdp->nxtlist; | |
64db4cff PM |
2876 | local_irq_restore(flags); |
2877 | ||
2878 | /* Invoke callbacks. */ | |
486e2593 | 2879 | count = count_lazy = 0; |
64db4cff PM |
2880 | while (list) { |
2881 | next = list->next; | |
2882 | prefetch(next); | |
551d55a9 | 2883 | debug_rcu_head_unqueue(list); |
486e2593 PM |
2884 | if (__rcu_reclaim(rsp->name, list)) |
2885 | count_lazy++; | |
64db4cff | 2886 | list = next; |
dff1672d PM |
2887 | /* Stop only if limit reached and CPU has something to do. */ |
2888 | if (++count >= bl && | |
2889 | (need_resched() || | |
2890 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
2891 | break; |
2892 | } | |
2893 | ||
2894 | local_irq_save(flags); | |
4968c300 PM |
2895 | trace_rcu_batch_end(rsp->name, count, !!list, need_resched(), |
2896 | is_idle_task(current), | |
2897 | rcu_is_callbacks_kthread()); | |
64db4cff PM |
2898 | |
2899 | /* Update count, and requeue any remaining callbacks. */ | |
64db4cff PM |
2900 | if (list != NULL) { |
2901 | *tail = rdp->nxtlist; | |
2902 | rdp->nxtlist = list; | |
b41772ab PM |
2903 | for (i = 0; i < RCU_NEXT_SIZE; i++) |
2904 | if (&rdp->nxtlist == rdp->nxttail[i]) | |
2905 | rdp->nxttail[i] = tail; | |
64db4cff PM |
2906 | else |
2907 | break; | |
2908 | } | |
b1420f1c PM |
2909 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
2910 | rdp->qlen_lazy -= count_lazy; | |
7d0ae808 | 2911 | WRITE_ONCE(rdp->qlen, rdp->qlen - count); |
b1420f1c | 2912 | rdp->n_cbs_invoked += count; |
64db4cff PM |
2913 | |
2914 | /* Reinstate batch limit if we have worked down the excess. */ | |
2915 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
2916 | rdp->blimit = blimit; | |
2917 | ||
37c72e56 PM |
2918 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
2919 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
2920 | rdp->qlen_last_fqs_check = 0; | |
2921 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
2922 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
2923 | rdp->qlen_last_fqs_check = rdp->qlen; | |
cfca9279 | 2924 | WARN_ON_ONCE((rdp->nxtlist == NULL) != (rdp->qlen == 0)); |
37c72e56 | 2925 | |
64db4cff PM |
2926 | local_irq_restore(flags); |
2927 | ||
e0f23060 | 2928 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
64db4cff | 2929 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 2930 | invoke_rcu_core(); |
64db4cff PM |
2931 | } |
2932 | ||
2933 | /* | |
2934 | * Check to see if this CPU is in a non-context-switch quiescent state | |
2935 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 2936 | * Also schedule RCU core processing. |
64db4cff | 2937 | * |
9b2e4f18 | 2938 | * This function must be called from hardirq context. It is normally |
5403d367 | 2939 | * invoked from the scheduling-clock interrupt. |
64db4cff | 2940 | */ |
c3377c2d | 2941 | void rcu_check_callbacks(int user) |
64db4cff | 2942 | { |
f7f7bac9 | 2943 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
a858af28 | 2944 | increment_cpu_stall_ticks(); |
9b2e4f18 | 2945 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
64db4cff PM |
2946 | |
2947 | /* | |
2948 | * Get here if this CPU took its interrupt from user | |
2949 | * mode or from the idle loop, and if this is not a | |
2950 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 2951 | * a quiescent state, so note it. |
64db4cff PM |
2952 | * |
2953 | * No memory barrier is required here because both | |
d6714c22 PM |
2954 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
2955 | * variables that other CPUs neither access nor modify, | |
2956 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
2957 | */ |
2958 | ||
284a8c93 PM |
2959 | rcu_sched_qs(); |
2960 | rcu_bh_qs(); | |
64db4cff PM |
2961 | |
2962 | } else if (!in_softirq()) { | |
2963 | ||
2964 | /* | |
2965 | * Get here if this CPU did not take its interrupt from | |
2966 | * softirq, in other words, if it is not interrupting | |
2967 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 2968 | * critical section, so note it. |
64db4cff PM |
2969 | */ |
2970 | ||
284a8c93 | 2971 | rcu_bh_qs(); |
64db4cff | 2972 | } |
86aea0e6 | 2973 | rcu_preempt_check_callbacks(); |
e3950ecd | 2974 | if (rcu_pending()) |
a46e0899 | 2975 | invoke_rcu_core(); |
8315f422 PM |
2976 | if (user) |
2977 | rcu_note_voluntary_context_switch(current); | |
f7f7bac9 | 2978 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2979 | } |
2980 | ||
64db4cff PM |
2981 | /* |
2982 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
2983 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
2984 | * Also initiate boosting for any threads blocked on the root rcu_node. |
2985 | * | |
ee47eb9f | 2986 | * The caller must have suppressed start of new grace periods. |
64db4cff | 2987 | */ |
217af2a2 PM |
2988 | static void force_qs_rnp(struct rcu_state *rsp, |
2989 | int (*f)(struct rcu_data *rsp, bool *isidle, | |
2990 | unsigned long *maxj), | |
2991 | bool *isidle, unsigned long *maxj) | |
64db4cff | 2992 | { |
64db4cff PM |
2993 | int cpu; |
2994 | unsigned long flags; | |
2995 | unsigned long mask; | |
a0b6c9a7 | 2996 | struct rcu_node *rnp; |
64db4cff | 2997 | |
a0b6c9a7 | 2998 | rcu_for_each_leaf_node(rsp, rnp) { |
bde6c3aa | 2999 | cond_resched_rcu_qs(); |
64db4cff | 3000 | mask = 0; |
2a67e741 | 3001 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
a0b6c9a7 | 3002 | if (rnp->qsmask == 0) { |
a77da14c PM |
3003 | if (rcu_state_p == &rcu_sched_state || |
3004 | rsp != rcu_state_p || | |
3005 | rcu_preempt_blocked_readers_cgp(rnp)) { | |
3006 | /* | |
3007 | * No point in scanning bits because they | |
3008 | * are all zero. But we might need to | |
3009 | * priority-boost blocked readers. | |
3010 | */ | |
3011 | rcu_initiate_boost(rnp, flags); | |
3012 | /* rcu_initiate_boost() releases rnp->lock */ | |
3013 | continue; | |
3014 | } | |
3015 | if (rnp->parent && | |
3016 | (rnp->parent->qsmask & rnp->grpmask)) { | |
3017 | /* | |
3018 | * Race between grace-period | |
3019 | * initialization and task exiting RCU | |
3020 | * read-side critical section: Report. | |
3021 | */ | |
3022 | rcu_report_unblock_qs_rnp(rsp, rnp, flags); | |
3023 | /* rcu_report_unblock_qs_rnp() rlses ->lock */ | |
3024 | continue; | |
3025 | } | |
64db4cff | 3026 | } |
bc75e999 MR |
3027 | for_each_leaf_node_possible_cpu(rnp, cpu) { |
3028 | unsigned long bit = leaf_node_cpu_bit(rnp, cpu); | |
0edd1b17 | 3029 | if ((rnp->qsmask & bit) != 0) { |
0edd1b17 PM |
3030 | if (f(per_cpu_ptr(rsp->rda, cpu), isidle, maxj)) |
3031 | mask |= bit; | |
3032 | } | |
64db4cff | 3033 | } |
45f014c5 | 3034 | if (mask != 0) { |
654e9533 PM |
3035 | /* Idle/offline CPUs, report (releases rnp->lock. */ |
3036 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags); | |
0aa04b05 PM |
3037 | } else { |
3038 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 3039 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 3040 | } |
64db4cff | 3041 | } |
64db4cff PM |
3042 | } |
3043 | ||
3044 | /* | |
3045 | * Force quiescent states on reluctant CPUs, and also detect which | |
3046 | * CPUs are in dyntick-idle mode. | |
3047 | */ | |
4cdfc175 | 3048 | static void force_quiescent_state(struct rcu_state *rsp) |
64db4cff PM |
3049 | { |
3050 | unsigned long flags; | |
394f2769 PM |
3051 | bool ret; |
3052 | struct rcu_node *rnp; | |
3053 | struct rcu_node *rnp_old = NULL; | |
3054 | ||
3055 | /* Funnel through hierarchy to reduce memory contention. */ | |
d860d403 | 3056 | rnp = __this_cpu_read(rsp->rda->mynode); |
394f2769 | 3057 | for (; rnp != NULL; rnp = rnp->parent) { |
7d0ae808 | 3058 | ret = (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) || |
394f2769 PM |
3059 | !raw_spin_trylock(&rnp->fqslock); |
3060 | if (rnp_old != NULL) | |
3061 | raw_spin_unlock(&rnp_old->fqslock); | |
3062 | if (ret) { | |
a792563b | 3063 | rsp->n_force_qs_lh++; |
394f2769 PM |
3064 | return; |
3065 | } | |
3066 | rnp_old = rnp; | |
3067 | } | |
3068 | /* rnp_old == rcu_get_root(rsp), rnp == NULL. */ | |
64db4cff | 3069 | |
394f2769 | 3070 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 3071 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 3072 | raw_spin_unlock(&rnp_old->fqslock); |
7d0ae808 | 3073 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
a792563b | 3074 | rsp->n_force_qs_lh++; |
67c583a7 | 3075 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 3076 | return; /* Someone beat us to it. */ |
46a1e34e | 3077 | } |
7d0ae808 | 3078 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 3079 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
94d44776 | 3080 | rcu_gp_kthread_wake(rsp); |
64db4cff PM |
3081 | } |
3082 | ||
64db4cff | 3083 | /* |
e0f23060 PM |
3084 | * This does the RCU core processing work for the specified rcu_state |
3085 | * and rcu_data structures. This may be called only from the CPU to | |
3086 | * whom the rdp belongs. | |
64db4cff PM |
3087 | */ |
3088 | static void | |
1bca8cf1 | 3089 | __rcu_process_callbacks(struct rcu_state *rsp) |
64db4cff PM |
3090 | { |
3091 | unsigned long flags; | |
48a7639c | 3092 | bool needwake; |
fa07a58f | 3093 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
64db4cff | 3094 | |
2e597558 PM |
3095 | WARN_ON_ONCE(rdp->beenonline == 0); |
3096 | ||
64db4cff PM |
3097 | /* Update RCU state based on any recent quiescent states. */ |
3098 | rcu_check_quiescent_state(rsp, rdp); | |
3099 | ||
3100 | /* Does this CPU require a not-yet-started grace period? */ | |
dc35c893 | 3101 | local_irq_save(flags); |
64db4cff | 3102 | if (cpu_needs_another_gp(rsp, rdp)) { |
6cf10081 | 3103 | raw_spin_lock_rcu_node(rcu_get_root(rsp)); /* irqs disabled. */ |
48a7639c | 3104 | needwake = rcu_start_gp(rsp); |
67c583a7 | 3105 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags); |
48a7639c PM |
3106 | if (needwake) |
3107 | rcu_gp_kthread_wake(rsp); | |
dc35c893 PM |
3108 | } else { |
3109 | local_irq_restore(flags); | |
64db4cff PM |
3110 | } |
3111 | ||
3112 | /* If there are callbacks ready, invoke them. */ | |
09223371 | 3113 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 3114 | invoke_rcu_callbacks(rsp, rdp); |
96d3fd0d PM |
3115 | |
3116 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
3117 | do_nocb_deferred_wakeup(rdp); | |
09223371 SL |
3118 | } |
3119 | ||
64db4cff | 3120 | /* |
e0f23060 | 3121 | * Do RCU core processing for the current CPU. |
64db4cff | 3122 | */ |
0766f788 | 3123 | static __latent_entropy void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 3124 | { |
6ce75a23 PM |
3125 | struct rcu_state *rsp; |
3126 | ||
bfa00b4c PM |
3127 | if (cpu_is_offline(smp_processor_id())) |
3128 | return; | |
f7f7bac9 | 3129 | trace_rcu_utilization(TPS("Start RCU core")); |
6ce75a23 PM |
3130 | for_each_rcu_flavor(rsp) |
3131 | __rcu_process_callbacks(rsp); | |
f7f7bac9 | 3132 | trace_rcu_utilization(TPS("End RCU core")); |
64db4cff PM |
3133 | } |
3134 | ||
a26ac245 | 3135 | /* |
e0f23060 PM |
3136 | * Schedule RCU callback invocation. If the specified type of RCU |
3137 | * does not support RCU priority boosting, just do a direct call, | |
3138 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
924df8a0 | 3139 | * are running on the current CPU with softirqs disabled, the |
e0f23060 | 3140 | * rcu_cpu_kthread_task cannot disappear out from under us. |
a26ac245 | 3141 | */ |
a46e0899 | 3142 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 3143 | { |
7d0ae808 | 3144 | if (unlikely(!READ_ONCE(rcu_scheduler_fully_active))) |
b0d30417 | 3145 | return; |
a46e0899 PM |
3146 | if (likely(!rsp->boost)) { |
3147 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
3148 | return; |
3149 | } | |
a46e0899 | 3150 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
3151 | } |
3152 | ||
a46e0899 | 3153 | static void invoke_rcu_core(void) |
09223371 | 3154 | { |
b0f74036 PM |
3155 | if (cpu_online(smp_processor_id())) |
3156 | raise_softirq(RCU_SOFTIRQ); | |
09223371 SL |
3157 | } |
3158 | ||
29154c57 PM |
3159 | /* |
3160 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
3161 | */ | |
3162 | static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp, | |
3163 | struct rcu_head *head, unsigned long flags) | |
64db4cff | 3164 | { |
48a7639c PM |
3165 | bool needwake; |
3166 | ||
62fde6ed PM |
3167 | /* |
3168 | * If called from an extended quiescent state, invoke the RCU | |
3169 | * core in order to force a re-evaluation of RCU's idleness. | |
3170 | */ | |
9910affa | 3171 | if (!rcu_is_watching()) |
62fde6ed PM |
3172 | invoke_rcu_core(); |
3173 | ||
a16b7a69 | 3174 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 3175 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 3176 | return; |
64db4cff | 3177 | |
37c72e56 PM |
3178 | /* |
3179 | * Force the grace period if too many callbacks or too long waiting. | |
3180 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
3181 | * if some other CPU has recently done so. Also, don't bother | |
3182 | * invoking force_quiescent_state() if the newly enqueued callback | |
3183 | * is the only one waiting for a grace period to complete. | |
3184 | */ | |
2655d57e | 3185 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
b52573d2 PM |
3186 | |
3187 | /* Are we ignoring a completed grace period? */ | |
470716fc | 3188 | note_gp_changes(rsp, rdp); |
b52573d2 PM |
3189 | |
3190 | /* Start a new grace period if one not already started. */ | |
3191 | if (!rcu_gp_in_progress(rsp)) { | |
b52573d2 PM |
3192 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
3193 | ||
2a67e741 | 3194 | raw_spin_lock_rcu_node(rnp_root); |
48a7639c | 3195 | needwake = rcu_start_gp(rsp); |
67c583a7 | 3196 | raw_spin_unlock_rcu_node(rnp_root); |
48a7639c PM |
3197 | if (needwake) |
3198 | rcu_gp_kthread_wake(rsp); | |
b52573d2 PM |
3199 | } else { |
3200 | /* Give the grace period a kick. */ | |
3201 | rdp->blimit = LONG_MAX; | |
3202 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
3203 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
4cdfc175 | 3204 | force_quiescent_state(rsp); |
b52573d2 PM |
3205 | rdp->n_force_qs_snap = rsp->n_force_qs; |
3206 | rdp->qlen_last_fqs_check = rdp->qlen; | |
3207 | } | |
4cdfc175 | 3208 | } |
29154c57 PM |
3209 | } |
3210 | ||
ae150184 PM |
3211 | /* |
3212 | * RCU callback function to leak a callback. | |
3213 | */ | |
3214 | static void rcu_leak_callback(struct rcu_head *rhp) | |
3215 | { | |
3216 | } | |
3217 | ||
3fbfbf7a PM |
3218 | /* |
3219 | * Helper function for call_rcu() and friends. The cpu argument will | |
3220 | * normally be -1, indicating "currently running CPU". It may specify | |
3221 | * a CPU only if that CPU is a no-CBs CPU. Currently, only _rcu_barrier() | |
3222 | * is expected to specify a CPU. | |
3223 | */ | |
64db4cff | 3224 | static void |
b6a4ae76 | 3225 | __call_rcu(struct rcu_head *head, rcu_callback_t func, |
3fbfbf7a | 3226 | struct rcu_state *rsp, int cpu, bool lazy) |
64db4cff PM |
3227 | { |
3228 | unsigned long flags; | |
3229 | struct rcu_data *rdp; | |
3230 | ||
b8f2ed53 PM |
3231 | /* Misaligned rcu_head! */ |
3232 | WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1)); | |
3233 | ||
ae150184 PM |
3234 | if (debug_rcu_head_queue(head)) { |
3235 | /* Probable double call_rcu(), so leak the callback. */ | |
7d0ae808 | 3236 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
3237 | WARN_ONCE(1, "__call_rcu(): Leaked duplicate callback\n"); |
3238 | return; | |
3239 | } | |
64db4cff PM |
3240 | head->func = func; |
3241 | head->next = NULL; | |
64db4cff | 3242 | local_irq_save(flags); |
394f99a9 | 3243 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
3244 | |
3245 | /* Add the callback to our list. */ | |
3fbfbf7a PM |
3246 | if (unlikely(rdp->nxttail[RCU_NEXT_TAIL] == NULL) || cpu != -1) { |
3247 | int offline; | |
3248 | ||
3249 | if (cpu != -1) | |
3250 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
143da9c2 PM |
3251 | if (likely(rdp->mynode)) { |
3252 | /* Post-boot, so this should be for a no-CBs CPU. */ | |
3253 | offline = !__call_rcu_nocb(rdp, head, lazy, flags); | |
3254 | WARN_ON_ONCE(offline); | |
3255 | /* Offline CPU, _call_rcu() illegal, leak callback. */ | |
3256 | local_irq_restore(flags); | |
3257 | return; | |
3258 | } | |
3259 | /* | |
3260 | * Very early boot, before rcu_init(). Initialize if needed | |
3261 | * and then drop through to queue the callback. | |
3262 | */ | |
3263 | BUG_ON(cpu != -1); | |
34404ca8 | 3264 | WARN_ON_ONCE(!rcu_is_watching()); |
143da9c2 PM |
3265 | if (!likely(rdp->nxtlist)) |
3266 | init_default_callback_list(rdp); | |
0d8ee37e | 3267 | } |
7d0ae808 | 3268 | WRITE_ONCE(rdp->qlen, rdp->qlen + 1); |
486e2593 PM |
3269 | if (lazy) |
3270 | rdp->qlen_lazy++; | |
c57afe80 PM |
3271 | else |
3272 | rcu_idle_count_callbacks_posted(); | |
b1420f1c PM |
3273 | smp_mb(); /* Count before adding callback for rcu_barrier(). */ |
3274 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
3275 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
2655d57e | 3276 | |
d4c08f2a PM |
3277 | if (__is_kfree_rcu_offset((unsigned long)func)) |
3278 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
486e2593 | 3279 | rdp->qlen_lazy, rdp->qlen); |
d4c08f2a | 3280 | else |
486e2593 | 3281 | trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen); |
d4c08f2a | 3282 | |
29154c57 PM |
3283 | /* Go handle any RCU core processing required. */ |
3284 | __call_rcu_core(rsp, rdp, head, flags); | |
64db4cff PM |
3285 | local_irq_restore(flags); |
3286 | } | |
3287 | ||
3288 | /* | |
d6714c22 | 3289 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 3290 | */ |
b6a4ae76 | 3291 | void call_rcu_sched(struct rcu_head *head, rcu_callback_t func) |
64db4cff | 3292 | { |
3fbfbf7a | 3293 | __call_rcu(head, func, &rcu_sched_state, -1, 0); |
64db4cff | 3294 | } |
d6714c22 | 3295 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
3296 | |
3297 | /* | |
486e2593 | 3298 | * Queue an RCU callback for invocation after a quicker grace period. |
64db4cff | 3299 | */ |
b6a4ae76 | 3300 | void call_rcu_bh(struct rcu_head *head, rcu_callback_t func) |
64db4cff | 3301 | { |
3fbfbf7a | 3302 | __call_rcu(head, func, &rcu_bh_state, -1, 0); |
64db4cff PM |
3303 | } |
3304 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
3305 | ||
495aa969 ACB |
3306 | /* |
3307 | * Queue an RCU callback for lazy invocation after a grace period. | |
3308 | * This will likely be later named something like "call_rcu_lazy()", | |
3309 | * but this change will require some way of tagging the lazy RCU | |
3310 | * callbacks in the list of pending callbacks. Until then, this | |
3311 | * function may only be called from __kfree_rcu(). | |
3312 | */ | |
3313 | void kfree_call_rcu(struct rcu_head *head, | |
b6a4ae76 | 3314 | rcu_callback_t func) |
495aa969 | 3315 | { |
e534165b | 3316 | __call_rcu(head, func, rcu_state_p, -1, 1); |
495aa969 ACB |
3317 | } |
3318 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
3319 | ||
6d813391 PM |
3320 | /* |
3321 | * Because a context switch is a grace period for RCU-sched and RCU-bh, | |
3322 | * any blocking grace-period wait automatically implies a grace period | |
3323 | * if there is only one CPU online at any point time during execution | |
3324 | * of either synchronize_sched() or synchronize_rcu_bh(). It is OK to | |
3325 | * occasionally incorrectly indicate that there are multiple CPUs online | |
3326 | * when there was in fact only one the whole time, as this just adds | |
3327 | * some overhead: RCU still operates correctly. | |
6d813391 PM |
3328 | */ |
3329 | static inline int rcu_blocking_is_gp(void) | |
3330 | { | |
95f0c1de PM |
3331 | int ret; |
3332 | ||
6d813391 | 3333 | might_sleep(); /* Check for RCU read-side critical section. */ |
95f0c1de PM |
3334 | preempt_disable(); |
3335 | ret = num_online_cpus() <= 1; | |
3336 | preempt_enable(); | |
3337 | return ret; | |
6d813391 PM |
3338 | } |
3339 | ||
6ebb237b PM |
3340 | /** |
3341 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
3342 | * | |
3343 | * Control will return to the caller some time after a full rcu-sched | |
3344 | * grace period has elapsed, in other words after all currently executing | |
3345 | * rcu-sched read-side critical sections have completed. These read-side | |
3346 | * critical sections are delimited by rcu_read_lock_sched() and | |
3347 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
3348 | * local_irq_disable(), and so on may be used in place of | |
3349 | * rcu_read_lock_sched(). | |
3350 | * | |
3351 | * This means that all preempt_disable code sequences, including NMI and | |
f0a0e6f2 PM |
3352 | * non-threaded hardware-interrupt handlers, in progress on entry will |
3353 | * have completed before this primitive returns. However, this does not | |
3354 | * guarantee that softirq handlers will have completed, since in some | |
3355 | * kernels, these handlers can run in process context, and can block. | |
3356 | * | |
3357 | * Note that this guarantee implies further memory-ordering guarantees. | |
3358 | * On systems with more than one CPU, when synchronize_sched() returns, | |
3359 | * each CPU is guaranteed to have executed a full memory barrier since the | |
3360 | * end of its last RCU-sched read-side critical section whose beginning | |
3361 | * preceded the call to synchronize_sched(). In addition, each CPU having | |
3362 | * an RCU read-side critical section that extends beyond the return from | |
3363 | * synchronize_sched() is guaranteed to have executed a full memory barrier | |
3364 | * after the beginning of synchronize_sched() and before the beginning of | |
3365 | * that RCU read-side critical section. Note that these guarantees include | |
3366 | * CPUs that are offline, idle, or executing in user mode, as well as CPUs | |
3367 | * that are executing in the kernel. | |
3368 | * | |
3369 | * Furthermore, if CPU A invoked synchronize_sched(), which returned | |
3370 | * to its caller on CPU B, then both CPU A and CPU B are guaranteed | |
3371 | * to have executed a full memory barrier during the execution of | |
3372 | * synchronize_sched() -- even if CPU A and CPU B are the same CPU (but | |
3373 | * again only if the system has more than one CPU). | |
6ebb237b PM |
3374 | * |
3375 | * This primitive provides the guarantees made by the (now removed) | |
3376 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
3377 | * guarantees that rcu_read_lock() sections will have completed. | |
3378 | * In "classic RCU", these two guarantees happen to be one and | |
3379 | * the same, but can differ in realtime RCU implementations. | |
3380 | */ | |
3381 | void synchronize_sched(void) | |
3382 | { | |
f78f5b90 PM |
3383 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
3384 | lock_is_held(&rcu_lock_map) || | |
3385 | lock_is_held(&rcu_sched_lock_map), | |
3386 | "Illegal synchronize_sched() in RCU-sched read-side critical section"); | |
6ebb237b PM |
3387 | if (rcu_blocking_is_gp()) |
3388 | return; | |
5afff48b | 3389 | if (rcu_gp_is_expedited()) |
3705b88d AM |
3390 | synchronize_sched_expedited(); |
3391 | else | |
3392 | wait_rcu_gp(call_rcu_sched); | |
6ebb237b PM |
3393 | } |
3394 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
3395 | ||
3396 | /** | |
3397 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
3398 | * | |
3399 | * Control will return to the caller some time after a full rcu_bh grace | |
3400 | * period has elapsed, in other words after all currently executing rcu_bh | |
3401 | * read-side critical sections have completed. RCU read-side critical | |
3402 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
3403 | * and may be nested. | |
f0a0e6f2 PM |
3404 | * |
3405 | * See the description of synchronize_sched() for more detailed information | |
3406 | * on memory ordering guarantees. | |
6ebb237b PM |
3407 | */ |
3408 | void synchronize_rcu_bh(void) | |
3409 | { | |
f78f5b90 PM |
3410 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
3411 | lock_is_held(&rcu_lock_map) || | |
3412 | lock_is_held(&rcu_sched_lock_map), | |
3413 | "Illegal synchronize_rcu_bh() in RCU-bh read-side critical section"); | |
6ebb237b PM |
3414 | if (rcu_blocking_is_gp()) |
3415 | return; | |
5afff48b | 3416 | if (rcu_gp_is_expedited()) |
3705b88d AM |
3417 | synchronize_rcu_bh_expedited(); |
3418 | else | |
3419 | wait_rcu_gp(call_rcu_bh); | |
6ebb237b PM |
3420 | } |
3421 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
3422 | ||
765a3f4f PM |
3423 | /** |
3424 | * get_state_synchronize_rcu - Snapshot current RCU state | |
3425 | * | |
3426 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
3427 | * to determine whether or not a full grace period has elapsed in the | |
3428 | * meantime. | |
3429 | */ | |
3430 | unsigned long get_state_synchronize_rcu(void) | |
3431 | { | |
3432 | /* | |
3433 | * Any prior manipulation of RCU-protected data must happen | |
3434 | * before the load from ->gpnum. | |
3435 | */ | |
3436 | smp_mb(); /* ^^^ */ | |
3437 | ||
3438 | /* | |
3439 | * Make sure this load happens before the purportedly | |
3440 | * time-consuming work between get_state_synchronize_rcu() | |
3441 | * and cond_synchronize_rcu(). | |
3442 | */ | |
e534165b | 3443 | return smp_load_acquire(&rcu_state_p->gpnum); |
765a3f4f PM |
3444 | } |
3445 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
3446 | ||
3447 | /** | |
3448 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
3449 | * | |
3450 | * @oldstate: return value from earlier call to get_state_synchronize_rcu() | |
3451 | * | |
3452 | * If a full RCU grace period has elapsed since the earlier call to | |
3453 | * get_state_synchronize_rcu(), just return. Otherwise, invoke | |
3454 | * synchronize_rcu() to wait for a full grace period. | |
3455 | * | |
3456 | * Yes, this function does not take counter wrap into account. But | |
3457 | * counter wrap is harmless. If the counter wraps, we have waited for | |
3458 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3459 | * so waiting for one additional grace period should be just fine. | |
3460 | */ | |
3461 | void cond_synchronize_rcu(unsigned long oldstate) | |
3462 | { | |
3463 | unsigned long newstate; | |
3464 | ||
3465 | /* | |
3466 | * Ensure that this load happens before any RCU-destructive | |
3467 | * actions the caller might carry out after we return. | |
3468 | */ | |
e534165b | 3469 | newstate = smp_load_acquire(&rcu_state_p->completed); |
765a3f4f PM |
3470 | if (ULONG_CMP_GE(oldstate, newstate)) |
3471 | synchronize_rcu(); | |
3472 | } | |
3473 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
3474 | ||
24560056 PM |
3475 | /** |
3476 | * get_state_synchronize_sched - Snapshot current RCU-sched state | |
3477 | * | |
3478 | * Returns a cookie that is used by a later call to cond_synchronize_sched() | |
3479 | * to determine whether or not a full grace period has elapsed in the | |
3480 | * meantime. | |
3481 | */ | |
3482 | unsigned long get_state_synchronize_sched(void) | |
3483 | { | |
3484 | /* | |
3485 | * Any prior manipulation of RCU-protected data must happen | |
3486 | * before the load from ->gpnum. | |
3487 | */ | |
3488 | smp_mb(); /* ^^^ */ | |
3489 | ||
3490 | /* | |
3491 | * Make sure this load happens before the purportedly | |
3492 | * time-consuming work between get_state_synchronize_sched() | |
3493 | * and cond_synchronize_sched(). | |
3494 | */ | |
3495 | return smp_load_acquire(&rcu_sched_state.gpnum); | |
3496 | } | |
3497 | EXPORT_SYMBOL_GPL(get_state_synchronize_sched); | |
3498 | ||
3499 | /** | |
3500 | * cond_synchronize_sched - Conditionally wait for an RCU-sched grace period | |
3501 | * | |
3502 | * @oldstate: return value from earlier call to get_state_synchronize_sched() | |
3503 | * | |
3504 | * If a full RCU-sched grace period has elapsed since the earlier call to | |
3505 | * get_state_synchronize_sched(), just return. Otherwise, invoke | |
3506 | * synchronize_sched() to wait for a full grace period. | |
3507 | * | |
3508 | * Yes, this function does not take counter wrap into account. But | |
3509 | * counter wrap is harmless. If the counter wraps, we have waited for | |
3510 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3511 | * so waiting for one additional grace period should be just fine. | |
3512 | */ | |
3513 | void cond_synchronize_sched(unsigned long oldstate) | |
3514 | { | |
3515 | unsigned long newstate; | |
3516 | ||
3517 | /* | |
3518 | * Ensure that this load happens before any RCU-destructive | |
3519 | * actions the caller might carry out after we return. | |
3520 | */ | |
3521 | newstate = smp_load_acquire(&rcu_sched_state.completed); | |
3522 | if (ULONG_CMP_GE(oldstate, newstate)) | |
3523 | synchronize_sched(); | |
3524 | } | |
3525 | EXPORT_SYMBOL_GPL(cond_synchronize_sched); | |
3526 | ||
28f00767 PM |
3527 | /* Adjust sequence number for start of update-side operation. */ |
3528 | static void rcu_seq_start(unsigned long *sp) | |
3529 | { | |
3530 | WRITE_ONCE(*sp, *sp + 1); | |
3531 | smp_mb(); /* Ensure update-side operation after counter increment. */ | |
3532 | WARN_ON_ONCE(!(*sp & 0x1)); | |
3533 | } | |
3534 | ||
3535 | /* Adjust sequence number for end of update-side operation. */ | |
3536 | static void rcu_seq_end(unsigned long *sp) | |
3537 | { | |
3538 | smp_mb(); /* Ensure update-side operation before counter increment. */ | |
3539 | WRITE_ONCE(*sp, *sp + 1); | |
3540 | WARN_ON_ONCE(*sp & 0x1); | |
3541 | } | |
3542 | ||
3543 | /* Take a snapshot of the update side's sequence number. */ | |
3544 | static unsigned long rcu_seq_snap(unsigned long *sp) | |
3545 | { | |
3546 | unsigned long s; | |
3547 | ||
28f00767 PM |
3548 | s = (READ_ONCE(*sp) + 3) & ~0x1; |
3549 | smp_mb(); /* Above access must not bleed into critical section. */ | |
3550 | return s; | |
3551 | } | |
3552 | ||
3553 | /* | |
3554 | * Given a snapshot from rcu_seq_snap(), determine whether or not a | |
3555 | * full update-side operation has occurred. | |
3556 | */ | |
3557 | static bool rcu_seq_done(unsigned long *sp, unsigned long s) | |
3558 | { | |
3559 | return ULONG_CMP_GE(READ_ONCE(*sp), s); | |
3560 | } | |
3561 | ||
64db4cff PM |
3562 | /* |
3563 | * Check to see if there is any immediate RCU-related work to be done | |
3564 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
3565 | * The checks are in order of increasing expense: checks that can be | |
3566 | * carried out against CPU-local state are performed first. However, | |
3567 | * we must check for CPU stalls first, else we might not get a chance. | |
3568 | */ | |
3569 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
3570 | { | |
2f51f988 PM |
3571 | struct rcu_node *rnp = rdp->mynode; |
3572 | ||
64db4cff PM |
3573 | rdp->n_rcu_pending++; |
3574 | ||
3575 | /* Check for CPU stalls, if enabled. */ | |
3576 | check_cpu_stall(rsp, rdp); | |
3577 | ||
a096932f PM |
3578 | /* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */ |
3579 | if (rcu_nohz_full_cpu(rsp)) | |
3580 | return 0; | |
3581 | ||
64db4cff | 3582 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
5c51dd73 | 3583 | if (rcu_scheduler_fully_active && |
5b74c458 | 3584 | rdp->core_needs_qs && rdp->cpu_no_qs.b.norm && |
9577df9a | 3585 | rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_dynticks.rcu_qs_ctr)) { |
97c668b8 | 3586 | rdp->n_rp_core_needs_qs++; |
3a19b46a | 3587 | } else if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) { |
d21670ac | 3588 | rdp->n_rp_report_qs++; |
64db4cff | 3589 | return 1; |
7ba5c840 | 3590 | } |
64db4cff PM |
3591 | |
3592 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
3593 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
3594 | rdp->n_rp_cb_ready++; | |
64db4cff | 3595 | return 1; |
7ba5c840 | 3596 | } |
64db4cff PM |
3597 | |
3598 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
3599 | if (cpu_needs_another_gp(rsp, rdp)) { |
3600 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 3601 | return 1; |
7ba5c840 | 3602 | } |
64db4cff PM |
3603 | |
3604 | /* Has another RCU grace period completed? */ | |
7d0ae808 | 3605 | if (READ_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 3606 | rdp->n_rp_gp_completed++; |
64db4cff | 3607 | return 1; |
7ba5c840 | 3608 | } |
64db4cff PM |
3609 | |
3610 | /* Has a new RCU grace period started? */ | |
7d0ae808 PM |
3611 | if (READ_ONCE(rnp->gpnum) != rdp->gpnum || |
3612 | unlikely(READ_ONCE(rdp->gpwrap))) { /* outside lock */ | |
7ba5c840 | 3613 | rdp->n_rp_gp_started++; |
64db4cff | 3614 | return 1; |
7ba5c840 | 3615 | } |
64db4cff | 3616 | |
96d3fd0d PM |
3617 | /* Does this CPU need a deferred NOCB wakeup? */ |
3618 | if (rcu_nocb_need_deferred_wakeup(rdp)) { | |
3619 | rdp->n_rp_nocb_defer_wakeup++; | |
3620 | return 1; | |
3621 | } | |
3622 | ||
64db4cff | 3623 | /* nothing to do */ |
7ba5c840 | 3624 | rdp->n_rp_need_nothing++; |
64db4cff PM |
3625 | return 0; |
3626 | } | |
3627 | ||
3628 | /* | |
3629 | * Check to see if there is any immediate RCU-related work to be done | |
3630 | * by the current CPU, returning 1 if so. This function is part of the | |
3631 | * RCU implementation; it is -not- an exported member of the RCU API. | |
3632 | */ | |
e3950ecd | 3633 | static int rcu_pending(void) |
64db4cff | 3634 | { |
6ce75a23 PM |
3635 | struct rcu_state *rsp; |
3636 | ||
3637 | for_each_rcu_flavor(rsp) | |
e3950ecd | 3638 | if (__rcu_pending(rsp, this_cpu_ptr(rsp->rda))) |
6ce75a23 PM |
3639 | return 1; |
3640 | return 0; | |
64db4cff PM |
3641 | } |
3642 | ||
3643 | /* | |
c0f4dfd4 PM |
3644 | * Return true if the specified CPU has any callback. If all_lazy is |
3645 | * non-NULL, store an indication of whether all callbacks are lazy. | |
3646 | * (If there are no callbacks, all of them are deemed to be lazy.) | |
64db4cff | 3647 | */ |
82072c4f | 3648 | static bool __maybe_unused rcu_cpu_has_callbacks(bool *all_lazy) |
64db4cff | 3649 | { |
c0f4dfd4 PM |
3650 | bool al = true; |
3651 | bool hc = false; | |
3652 | struct rcu_data *rdp; | |
6ce75a23 PM |
3653 | struct rcu_state *rsp; |
3654 | ||
c0f4dfd4 | 3655 | for_each_rcu_flavor(rsp) { |
aa6da514 | 3656 | rdp = this_cpu_ptr(rsp->rda); |
69c8d28c PM |
3657 | if (!rdp->nxtlist) |
3658 | continue; | |
3659 | hc = true; | |
3660 | if (rdp->qlen != rdp->qlen_lazy || !all_lazy) { | |
c0f4dfd4 | 3661 | al = false; |
69c8d28c PM |
3662 | break; |
3663 | } | |
c0f4dfd4 PM |
3664 | } |
3665 | if (all_lazy) | |
3666 | *all_lazy = al; | |
3667 | return hc; | |
64db4cff PM |
3668 | } |
3669 | ||
a83eff0a PM |
3670 | /* |
3671 | * Helper function for _rcu_barrier() tracing. If tracing is disabled, | |
3672 | * the compiler is expected to optimize this away. | |
3673 | */ | |
e66c33d5 | 3674 | static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s, |
a83eff0a PM |
3675 | int cpu, unsigned long done) |
3676 | { | |
3677 | trace_rcu_barrier(rsp->name, s, cpu, | |
3678 | atomic_read(&rsp->barrier_cpu_count), done); | |
3679 | } | |
3680 | ||
b1420f1c PM |
3681 | /* |
3682 | * RCU callback function for _rcu_barrier(). If we are last, wake | |
3683 | * up the task executing _rcu_barrier(). | |
3684 | */ | |
24ebbca8 | 3685 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 3686 | { |
24ebbca8 PM |
3687 | struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head); |
3688 | struct rcu_state *rsp = rdp->rsp; | |
3689 | ||
a83eff0a | 3690 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) { |
4f525a52 | 3691 | _rcu_barrier_trace(rsp, "LastCB", -1, rsp->barrier_sequence); |
7db74df8 | 3692 | complete(&rsp->barrier_completion); |
a83eff0a | 3693 | } else { |
4f525a52 | 3694 | _rcu_barrier_trace(rsp, "CB", -1, rsp->barrier_sequence); |
a83eff0a | 3695 | } |
d0ec774c PM |
3696 | } |
3697 | ||
3698 | /* | |
3699 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
3700 | */ | |
3701 | static void rcu_barrier_func(void *type) | |
3702 | { | |
037b64ed | 3703 | struct rcu_state *rsp = type; |
fa07a58f | 3704 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
d0ec774c | 3705 | |
4f525a52 | 3706 | _rcu_barrier_trace(rsp, "IRQ", -1, rsp->barrier_sequence); |
24ebbca8 | 3707 | atomic_inc(&rsp->barrier_cpu_count); |
06668efa | 3708 | rsp->call(&rdp->barrier_head, rcu_barrier_callback); |
d0ec774c PM |
3709 | } |
3710 | ||
d0ec774c PM |
3711 | /* |
3712 | * Orchestrate the specified type of RCU barrier, waiting for all | |
3713 | * RCU callbacks of the specified type to complete. | |
3714 | */ | |
037b64ed | 3715 | static void _rcu_barrier(struct rcu_state *rsp) |
d0ec774c | 3716 | { |
b1420f1c | 3717 | int cpu; |
b1420f1c | 3718 | struct rcu_data *rdp; |
4f525a52 | 3719 | unsigned long s = rcu_seq_snap(&rsp->barrier_sequence); |
b1420f1c | 3720 | |
4f525a52 | 3721 | _rcu_barrier_trace(rsp, "Begin", -1, s); |
b1420f1c | 3722 | |
e74f4c45 | 3723 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
7be7f0be | 3724 | mutex_lock(&rsp->barrier_mutex); |
b1420f1c | 3725 | |
4f525a52 PM |
3726 | /* Did someone else do our work for us? */ |
3727 | if (rcu_seq_done(&rsp->barrier_sequence, s)) { | |
3728 | _rcu_barrier_trace(rsp, "EarlyExit", -1, rsp->barrier_sequence); | |
cf3a9c48 PM |
3729 | smp_mb(); /* caller's subsequent code after above check. */ |
3730 | mutex_unlock(&rsp->barrier_mutex); | |
3731 | return; | |
3732 | } | |
3733 | ||
4f525a52 PM |
3734 | /* Mark the start of the barrier operation. */ |
3735 | rcu_seq_start(&rsp->barrier_sequence); | |
3736 | _rcu_barrier_trace(rsp, "Inc1", -1, rsp->barrier_sequence); | |
b1420f1c | 3737 | |
d0ec774c | 3738 | /* |
b1420f1c PM |
3739 | * Initialize the count to one rather than to zero in order to |
3740 | * avoid a too-soon return to zero in case of a short grace period | |
1331e7a1 PM |
3741 | * (or preemption of this task). Exclude CPU-hotplug operations |
3742 | * to ensure that no offline CPU has callbacks queued. | |
d0ec774c | 3743 | */ |
7db74df8 | 3744 | init_completion(&rsp->barrier_completion); |
24ebbca8 | 3745 | atomic_set(&rsp->barrier_cpu_count, 1); |
1331e7a1 | 3746 | get_online_cpus(); |
b1420f1c PM |
3747 | |
3748 | /* | |
1331e7a1 PM |
3749 | * Force each CPU with callbacks to register a new callback. |
3750 | * When that callback is invoked, we will know that all of the | |
3751 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 3752 | */ |
3fbfbf7a | 3753 | for_each_possible_cpu(cpu) { |
d1e43fa5 | 3754 | if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu)) |
3fbfbf7a | 3755 | continue; |
b1420f1c | 3756 | rdp = per_cpu_ptr(rsp->rda, cpu); |
d1e43fa5 | 3757 | if (rcu_is_nocb_cpu(cpu)) { |
d7e29933 PM |
3758 | if (!rcu_nocb_cpu_needs_barrier(rsp, cpu)) { |
3759 | _rcu_barrier_trace(rsp, "OfflineNoCB", cpu, | |
4f525a52 | 3760 | rsp->barrier_sequence); |
d7e29933 PM |
3761 | } else { |
3762 | _rcu_barrier_trace(rsp, "OnlineNoCB", cpu, | |
4f525a52 | 3763 | rsp->barrier_sequence); |
41050a00 | 3764 | smp_mb__before_atomic(); |
d7e29933 PM |
3765 | atomic_inc(&rsp->barrier_cpu_count); |
3766 | __call_rcu(&rdp->barrier_head, | |
3767 | rcu_barrier_callback, rsp, cpu, 0); | |
3768 | } | |
7d0ae808 | 3769 | } else if (READ_ONCE(rdp->qlen)) { |
a83eff0a | 3770 | _rcu_barrier_trace(rsp, "OnlineQ", cpu, |
4f525a52 | 3771 | rsp->barrier_sequence); |
037b64ed | 3772 | smp_call_function_single(cpu, rcu_barrier_func, rsp, 1); |
b1420f1c | 3773 | } else { |
a83eff0a | 3774 | _rcu_barrier_trace(rsp, "OnlineNQ", cpu, |
4f525a52 | 3775 | rsp->barrier_sequence); |
b1420f1c PM |
3776 | } |
3777 | } | |
1331e7a1 | 3778 | put_online_cpus(); |
b1420f1c PM |
3779 | |
3780 | /* | |
3781 | * Now that we have an rcu_barrier_callback() callback on each | |
3782 | * CPU, and thus each counted, remove the initial count. | |
3783 | */ | |
24ebbca8 | 3784 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) |
7db74df8 | 3785 | complete(&rsp->barrier_completion); |
b1420f1c PM |
3786 | |
3787 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
7db74df8 | 3788 | wait_for_completion(&rsp->barrier_completion); |
b1420f1c | 3789 | |
4f525a52 PM |
3790 | /* Mark the end of the barrier operation. */ |
3791 | _rcu_barrier_trace(rsp, "Inc2", -1, rsp->barrier_sequence); | |
3792 | rcu_seq_end(&rsp->barrier_sequence); | |
3793 | ||
b1420f1c | 3794 | /* Other rcu_barrier() invocations can now safely proceed. */ |
7be7f0be | 3795 | mutex_unlock(&rsp->barrier_mutex); |
d0ec774c | 3796 | } |
d0ec774c PM |
3797 | |
3798 | /** | |
3799 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
3800 | */ | |
3801 | void rcu_barrier_bh(void) | |
3802 | { | |
037b64ed | 3803 | _rcu_barrier(&rcu_bh_state); |
d0ec774c PM |
3804 | } |
3805 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
3806 | ||
3807 | /** | |
3808 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
3809 | */ | |
3810 | void rcu_barrier_sched(void) | |
3811 | { | |
037b64ed | 3812 | _rcu_barrier(&rcu_sched_state); |
d0ec774c PM |
3813 | } |
3814 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
3815 | ||
0aa04b05 PM |
3816 | /* |
3817 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
3818 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
3819 | * must hold the corresponding leaf rcu_node ->lock with interrrupts | |
3820 | * disabled. | |
3821 | */ | |
3822 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
3823 | { | |
3824 | long mask; | |
3825 | struct rcu_node *rnp = rnp_leaf; | |
3826 | ||
3827 | for (;;) { | |
3828 | mask = rnp->grpmask; | |
3829 | rnp = rnp->parent; | |
3830 | if (rnp == NULL) | |
3831 | return; | |
6cf10081 | 3832 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
0aa04b05 | 3833 | rnp->qsmaskinit |= mask; |
67c583a7 | 3834 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
0aa04b05 PM |
3835 | } |
3836 | } | |
3837 | ||
64db4cff | 3838 | /* |
27569620 | 3839 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 3840 | */ |
27569620 PM |
3841 | static void __init |
3842 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
3843 | { |
3844 | unsigned long flags; | |
394f99a9 | 3845 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
3846 | struct rcu_node *rnp = rcu_get_root(rsp); |
3847 | ||
3848 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 3849 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
bc75e999 | 3850 | rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu); |
27569620 | 3851 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
29e37d81 | 3852 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE); |
02a5c550 | 3853 | WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp->dynticks))); |
27569620 | 3854 | rdp->cpu = cpu; |
d4c08f2a | 3855 | rdp->rsp = rsp; |
3fbfbf7a | 3856 | rcu_boot_init_nocb_percpu_data(rdp); |
67c583a7 | 3857 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27569620 PM |
3858 | } |
3859 | ||
3860 | /* | |
3861 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
3862 | * offline event can be happening at a given time. Note also that we | |
3863 | * can accept some slop in the rsp->completed access due to the fact | |
3864 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 3865 | */ |
49fb4c62 | 3866 | static void |
9b67122a | 3867 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp) |
64db4cff PM |
3868 | { |
3869 | unsigned long flags; | |
394f99a9 | 3870 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
3871 | struct rcu_node *rnp = rcu_get_root(rsp); |
3872 | ||
3873 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 3874 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 PM |
3875 | rdp->qlen_last_fqs_check = 0; |
3876 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 3877 | rdp->blimit = blimit; |
39c8d313 PM |
3878 | if (!rdp->nxtlist) |
3879 | init_callback_list(rdp); /* Re-enable callbacks on this CPU. */ | |
29e37d81 | 3880 | rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE; |
2333210b | 3881 | rcu_sysidle_init_percpu_data(rdp->dynticks); |
2625d469 | 3882 | rcu_dynticks_eqs_online(); |
67c583a7 | 3883 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
64db4cff | 3884 | |
0aa04b05 PM |
3885 | /* |
3886 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
3887 | * propagation up the rcu_node tree will happen at the beginning | |
3888 | * of the next grace period. | |
3889 | */ | |
64db4cff | 3890 | rnp = rdp->mynode; |
2a67e741 | 3891 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
b9585e94 PM |
3892 | if (!rdp->beenonline) |
3893 | WRITE_ONCE(rsp->ncpus, READ_ONCE(rsp->ncpus) + 1); | |
3894 | rdp->beenonline = true; /* We have now been online. */ | |
0aa04b05 PM |
3895 | rdp->gpnum = rnp->completed; /* Make CPU later note any new GP. */ |
3896 | rdp->completed = rnp->completed; | |
5b74c458 | 3897 | rdp->cpu_no_qs.b.norm = true; |
9577df9a | 3898 | rdp->rcu_qs_ctr_snap = per_cpu(rcu_dynticks.rcu_qs_ctr, cpu); |
97c668b8 | 3899 | rdp->core_needs_qs = false; |
0aa04b05 | 3900 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl")); |
67c583a7 | 3901 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
3902 | } |
3903 | ||
4df83742 | 3904 | int rcutree_prepare_cpu(unsigned int cpu) |
64db4cff | 3905 | { |
6ce75a23 PM |
3906 | struct rcu_state *rsp; |
3907 | ||
3908 | for_each_rcu_flavor(rsp) | |
9b67122a | 3909 | rcu_init_percpu_data(cpu, rsp); |
4df83742 TG |
3910 | |
3911 | rcu_prepare_kthreads(cpu); | |
3912 | rcu_spawn_all_nocb_kthreads(cpu); | |
3913 | ||
3914 | return 0; | |
3915 | } | |
3916 | ||
3917 | static void rcutree_affinity_setting(unsigned int cpu, int outgoing) | |
3918 | { | |
3919 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); | |
3920 | ||
3921 | rcu_boost_kthread_setaffinity(rdp->mynode, outgoing); | |
3922 | } | |
3923 | ||
3924 | int rcutree_online_cpu(unsigned int cpu) | |
3925 | { | |
3926 | sync_sched_exp_online_cleanup(cpu); | |
3927 | rcutree_affinity_setting(cpu, -1); | |
3928 | return 0; | |
3929 | } | |
3930 | ||
3931 | int rcutree_offline_cpu(unsigned int cpu) | |
3932 | { | |
3933 | rcutree_affinity_setting(cpu, cpu); | |
3934 | return 0; | |
3935 | } | |
3936 | ||
3937 | ||
3938 | int rcutree_dying_cpu(unsigned int cpu) | |
3939 | { | |
3940 | struct rcu_state *rsp; | |
3941 | ||
3942 | for_each_rcu_flavor(rsp) | |
3943 | rcu_cleanup_dying_cpu(rsp); | |
3944 | return 0; | |
3945 | } | |
3946 | ||
3947 | int rcutree_dead_cpu(unsigned int cpu) | |
3948 | { | |
3949 | struct rcu_state *rsp; | |
3950 | ||
3951 | for_each_rcu_flavor(rsp) { | |
3952 | rcu_cleanup_dead_cpu(cpu, rsp); | |
3953 | do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu)); | |
3954 | } | |
3955 | return 0; | |
64db4cff PM |
3956 | } |
3957 | ||
7ec99de3 PM |
3958 | /* |
3959 | * Mark the specified CPU as being online so that subsequent grace periods | |
3960 | * (both expedited and normal) will wait on it. Note that this means that | |
3961 | * incoming CPUs are not allowed to use RCU read-side critical sections | |
3962 | * until this function is called. Failing to observe this restriction | |
3963 | * will result in lockdep splats. | |
3964 | */ | |
3965 | void rcu_cpu_starting(unsigned int cpu) | |
3966 | { | |
3967 | unsigned long flags; | |
3968 | unsigned long mask; | |
3969 | struct rcu_data *rdp; | |
3970 | struct rcu_node *rnp; | |
3971 | struct rcu_state *rsp; | |
3972 | ||
3973 | for_each_rcu_flavor(rsp) { | |
fdbb9b31 | 3974 | rdp = per_cpu_ptr(rsp->rda, cpu); |
7ec99de3 PM |
3975 | rnp = rdp->mynode; |
3976 | mask = rdp->grpmask; | |
3977 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3978 | rnp->qsmaskinitnext |= mask; | |
3979 | rnp->expmaskinitnext |= mask; | |
3980 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
3981 | } | |
3982 | } | |
3983 | ||
27d50c7e TG |
3984 | #ifdef CONFIG_HOTPLUG_CPU |
3985 | /* | |
710d60cb LT |
3986 | * The CPU is exiting the idle loop into the arch_cpu_idle_dead() |
3987 | * function. We now remove it from the rcu_node tree's ->qsmaskinit | |
3988 | * bit masks. | |
27d50c7e TG |
3989 | * The CPU is exiting the idle loop into the arch_cpu_idle_dead() |
3990 | * function. We now remove it from the rcu_node tree's ->qsmaskinit | |
3991 | * bit masks. | |
3992 | */ | |
3993 | static void rcu_cleanup_dying_idle_cpu(int cpu, struct rcu_state *rsp) | |
3994 | { | |
3995 | unsigned long flags; | |
3996 | unsigned long mask; | |
3997 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
3998 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ | |
3999 | ||
27d50c7e TG |
4000 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ |
4001 | mask = rdp->grpmask; | |
4002 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ | |
4003 | rnp->qsmaskinitnext &= ~mask; | |
710d60cb | 4004 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27d50c7e TG |
4005 | } |
4006 | ||
4007 | void rcu_report_dead(unsigned int cpu) | |
4008 | { | |
4009 | struct rcu_state *rsp; | |
4010 | ||
4011 | /* QS for any half-done expedited RCU-sched GP. */ | |
4012 | preempt_disable(); | |
4013 | rcu_report_exp_rdp(&rcu_sched_state, | |
4014 | this_cpu_ptr(rcu_sched_state.rda), true); | |
4015 | preempt_enable(); | |
4016 | for_each_rcu_flavor(rsp) | |
4017 | rcu_cleanup_dying_idle_cpu(cpu, rsp); | |
4018 | } | |
4019 | #endif | |
4020 | ||
d1d74d14 BP |
4021 | static int rcu_pm_notify(struct notifier_block *self, |
4022 | unsigned long action, void *hcpu) | |
4023 | { | |
4024 | switch (action) { | |
4025 | case PM_HIBERNATION_PREPARE: | |
4026 | case PM_SUSPEND_PREPARE: | |
4027 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ | |
5afff48b | 4028 | rcu_expedite_gp(); |
d1d74d14 BP |
4029 | break; |
4030 | case PM_POST_HIBERNATION: | |
4031 | case PM_POST_SUSPEND: | |
5afff48b PM |
4032 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ |
4033 | rcu_unexpedite_gp(); | |
d1d74d14 BP |
4034 | break; |
4035 | default: | |
4036 | break; | |
4037 | } | |
4038 | return NOTIFY_OK; | |
4039 | } | |
4040 | ||
b3dbec76 | 4041 | /* |
9386c0b7 | 4042 | * Spawn the kthreads that handle each RCU flavor's grace periods. |
b3dbec76 PM |
4043 | */ |
4044 | static int __init rcu_spawn_gp_kthread(void) | |
4045 | { | |
4046 | unsigned long flags; | |
a94844b2 | 4047 | int kthread_prio_in = kthread_prio; |
b3dbec76 PM |
4048 | struct rcu_node *rnp; |
4049 | struct rcu_state *rsp; | |
a94844b2 | 4050 | struct sched_param sp; |
b3dbec76 PM |
4051 | struct task_struct *t; |
4052 | ||
a94844b2 PM |
4053 | /* Force priority into range. */ |
4054 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
4055 | kthread_prio = 1; | |
4056 | else if (kthread_prio < 0) | |
4057 | kthread_prio = 0; | |
4058 | else if (kthread_prio > 99) | |
4059 | kthread_prio = 99; | |
4060 | if (kthread_prio != kthread_prio_in) | |
4061 | pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n", | |
4062 | kthread_prio, kthread_prio_in); | |
4063 | ||
9386c0b7 | 4064 | rcu_scheduler_fully_active = 1; |
b3dbec76 | 4065 | for_each_rcu_flavor(rsp) { |
a94844b2 | 4066 | t = kthread_create(rcu_gp_kthread, rsp, "%s", rsp->name); |
b3dbec76 PM |
4067 | BUG_ON(IS_ERR(t)); |
4068 | rnp = rcu_get_root(rsp); | |
6cf10081 | 4069 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b3dbec76 | 4070 | rsp->gp_kthread = t; |
a94844b2 PM |
4071 | if (kthread_prio) { |
4072 | sp.sched_priority = kthread_prio; | |
4073 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
4074 | } | |
67c583a7 | 4075 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
e11f1335 | 4076 | wake_up_process(t); |
b3dbec76 | 4077 | } |
35ce7f29 | 4078 | rcu_spawn_nocb_kthreads(); |
9386c0b7 | 4079 | rcu_spawn_boost_kthreads(); |
b3dbec76 PM |
4080 | return 0; |
4081 | } | |
4082 | early_initcall(rcu_spawn_gp_kthread); | |
4083 | ||
bbad9379 | 4084 | /* |
52d7e48b PM |
4085 | * This function is invoked towards the end of the scheduler's |
4086 | * initialization process. Before this is called, the idle task might | |
4087 | * contain synchronous grace-period primitives (during which time, this idle | |
4088 | * task is booting the system, and such primitives are no-ops). After this | |
4089 | * function is called, any synchronous grace-period primitives are run as | |
4090 | * expedited, with the requesting task driving the grace period forward. | |
4091 | * A later core_initcall() rcu_exp_runtime_mode() will switch to full | |
4092 | * runtime RCU functionality. | |
bbad9379 PM |
4093 | */ |
4094 | void rcu_scheduler_starting(void) | |
4095 | { | |
4096 | WARN_ON(num_online_cpus() != 1); | |
4097 | WARN_ON(nr_context_switches() > 0); | |
52d7e48b PM |
4098 | rcu_test_sync_prims(); |
4099 | rcu_scheduler_active = RCU_SCHEDULER_INIT; | |
4100 | rcu_test_sync_prims(); | |
bbad9379 PM |
4101 | } |
4102 | ||
64db4cff PM |
4103 | /* |
4104 | * Compute the per-level fanout, either using the exact fanout specified | |
7fa27001 | 4105 | * or balancing the tree, depending on the rcu_fanout_exact boot parameter. |
64db4cff | 4106 | */ |
199977bf | 4107 | static void __init rcu_init_levelspread(int *levelspread, const int *levelcnt) |
64db4cff | 4108 | { |
64db4cff PM |
4109 | int i; |
4110 | ||
7fa27001 | 4111 | if (rcu_fanout_exact) { |
199977bf | 4112 | levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf; |
66292405 | 4113 | for (i = rcu_num_lvls - 2; i >= 0; i--) |
199977bf | 4114 | levelspread[i] = RCU_FANOUT; |
66292405 PM |
4115 | } else { |
4116 | int ccur; | |
4117 | int cprv; | |
4118 | ||
4119 | cprv = nr_cpu_ids; | |
4120 | for (i = rcu_num_lvls - 1; i >= 0; i--) { | |
199977bf AG |
4121 | ccur = levelcnt[i]; |
4122 | levelspread[i] = (cprv + ccur - 1) / ccur; | |
66292405 PM |
4123 | cprv = ccur; |
4124 | } | |
64db4cff PM |
4125 | } |
4126 | } | |
64db4cff PM |
4127 | |
4128 | /* | |
4129 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
4130 | */ | |
a87f203e | 4131 | static void __init rcu_init_one(struct rcu_state *rsp) |
64db4cff | 4132 | { |
cb007102 AG |
4133 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
4134 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
4135 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
4136 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
199977bf AG |
4137 | |
4138 | int levelcnt[RCU_NUM_LVLS]; /* # nodes in each level. */ | |
4139 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ | |
64db4cff PM |
4140 | int cpustride = 1; |
4141 | int i; | |
4142 | int j; | |
4143 | struct rcu_node *rnp; | |
4144 | ||
05b84aec | 4145 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 4146 | |
3eaaaf6c PM |
4147 | /* Silence gcc 4.8 false positive about array index out of range. */ |
4148 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
4149 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 4150 | |
64db4cff PM |
4151 | /* Initialize the level-tracking arrays. */ |
4152 | ||
f885b7f2 | 4153 | for (i = 0; i < rcu_num_lvls; i++) |
199977bf | 4154 | levelcnt[i] = num_rcu_lvl[i]; |
f885b7f2 | 4155 | for (i = 1; i < rcu_num_lvls; i++) |
199977bf AG |
4156 | rsp->level[i] = rsp->level[i - 1] + levelcnt[i - 1]; |
4157 | rcu_init_levelspread(levelspread, levelcnt); | |
64db4cff PM |
4158 | |
4159 | /* Initialize the elements themselves, starting from the leaves. */ | |
4160 | ||
f885b7f2 | 4161 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 4162 | cpustride *= levelspread[i]; |
64db4cff | 4163 | rnp = rsp->level[i]; |
199977bf | 4164 | for (j = 0; j < levelcnt[i]; j++, rnp++) { |
67c583a7 BF |
4165 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
4166 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 4167 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
4168 | raw_spin_lock_init(&rnp->fqslock); |
4169 | lockdep_set_class_and_name(&rnp->fqslock, | |
4170 | &rcu_fqs_class[i], fqs[i]); | |
25d30cf4 PM |
4171 | rnp->gpnum = rsp->gpnum; |
4172 | rnp->completed = rsp->completed; | |
64db4cff PM |
4173 | rnp->qsmask = 0; |
4174 | rnp->qsmaskinit = 0; | |
4175 | rnp->grplo = j * cpustride; | |
4176 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
4177 | if (rnp->grphi >= nr_cpu_ids) |
4178 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
4179 | if (i == 0) { |
4180 | rnp->grpnum = 0; | |
4181 | rnp->grpmask = 0; | |
4182 | rnp->parent = NULL; | |
4183 | } else { | |
199977bf | 4184 | rnp->grpnum = j % levelspread[i - 1]; |
64db4cff PM |
4185 | rnp->grpmask = 1UL << rnp->grpnum; |
4186 | rnp->parent = rsp->level[i - 1] + | |
199977bf | 4187 | j / levelspread[i - 1]; |
64db4cff PM |
4188 | } |
4189 | rnp->level = i; | |
12f5f524 | 4190 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 4191 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
4192 | init_waitqueue_head(&rnp->exp_wq[0]); |
4193 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
4194 | init_waitqueue_head(&rnp->exp_wq[2]); |
4195 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 4196 | spin_lock_init(&rnp->exp_lock); |
64db4cff PM |
4197 | } |
4198 | } | |
0c34029a | 4199 | |
abedf8e2 PG |
4200 | init_swait_queue_head(&rsp->gp_wq); |
4201 | init_swait_queue_head(&rsp->expedited_wq); | |
f885b7f2 | 4202 | rnp = rsp->level[rcu_num_lvls - 1]; |
0c34029a | 4203 | for_each_possible_cpu(i) { |
4a90a068 | 4204 | while (i > rnp->grphi) |
0c34029a | 4205 | rnp++; |
394f99a9 | 4206 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
4207 | rcu_boot_init_percpu_data(i, rsp); |
4208 | } | |
6ce75a23 | 4209 | list_add(&rsp->flavors, &rcu_struct_flavors); |
64db4cff PM |
4210 | } |
4211 | ||
f885b7f2 PM |
4212 | /* |
4213 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 4214 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
4215 | * the ->node array in the rcu_state structure. |
4216 | */ | |
4217 | static void __init rcu_init_geometry(void) | |
4218 | { | |
026ad283 | 4219 | ulong d; |
f885b7f2 | 4220 | int i; |
05b84aec | 4221 | int rcu_capacity[RCU_NUM_LVLS]; |
f885b7f2 | 4222 | |
026ad283 PM |
4223 | /* |
4224 | * Initialize any unspecified boot parameters. | |
4225 | * The default values of jiffies_till_first_fqs and | |
4226 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
4227 | * value, which is a function of HZ, then adding one for each | |
4228 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
4229 | */ | |
4230 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
4231 | if (jiffies_till_first_fqs == ULONG_MAX) | |
4232 | jiffies_till_first_fqs = d; | |
4233 | if (jiffies_till_next_fqs == ULONG_MAX) | |
4234 | jiffies_till_next_fqs = d; | |
4235 | ||
f885b7f2 | 4236 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 4237 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 4238 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 4239 | return; |
39479098 PM |
4240 | pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%d\n", |
4241 | rcu_fanout_leaf, nr_cpu_ids); | |
f885b7f2 | 4242 | |
f885b7f2 | 4243 | /* |
ee968ac6 PM |
4244 | * The boot-time rcu_fanout_leaf parameter must be at least two |
4245 | * and cannot exceed the number of bits in the rcu_node masks. | |
4246 | * Complain and fall back to the compile-time values if this | |
4247 | * limit is exceeded. | |
f885b7f2 | 4248 | */ |
ee968ac6 | 4249 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 4250 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 4251 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
4252 | WARN_ON(1); |
4253 | return; | |
4254 | } | |
4255 | ||
f885b7f2 PM |
4256 | /* |
4257 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 4258 | * with the given number of levels. |
f885b7f2 | 4259 | */ |
9618138b | 4260 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 4261 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 4262 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
4263 | |
4264 | /* | |
75cf15a4 | 4265 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 4266 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 4267 | */ |
ee968ac6 PM |
4268 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
4269 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
4270 | WARN_ON(1); | |
4271 | return; | |
4272 | } | |
f885b7f2 | 4273 | |
679f9858 | 4274 | /* Calculate the number of levels in the tree. */ |
9618138b | 4275 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 4276 | } |
9618138b | 4277 | rcu_num_lvls = i + 1; |
679f9858 | 4278 | |
f885b7f2 | 4279 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 4280 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 4281 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
4282 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
4283 | } | |
f885b7f2 PM |
4284 | |
4285 | /* Calculate the total number of rcu_node structures. */ | |
4286 | rcu_num_nodes = 0; | |
679f9858 | 4287 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 4288 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
4289 | } |
4290 | ||
a3dc2948 PM |
4291 | /* |
4292 | * Dump out the structure of the rcu_node combining tree associated | |
4293 | * with the rcu_state structure referenced by rsp. | |
4294 | */ | |
4295 | static void __init rcu_dump_rcu_node_tree(struct rcu_state *rsp) | |
4296 | { | |
4297 | int level = 0; | |
4298 | struct rcu_node *rnp; | |
4299 | ||
4300 | pr_info("rcu_node tree layout dump\n"); | |
4301 | pr_info(" "); | |
4302 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
4303 | if (rnp->level != level) { | |
4304 | pr_cont("\n"); | |
4305 | pr_info(" "); | |
4306 | level = rnp->level; | |
4307 | } | |
4308 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
4309 | } | |
4310 | pr_cont("\n"); | |
4311 | } | |
4312 | ||
9f680ab4 | 4313 | void __init rcu_init(void) |
64db4cff | 4314 | { |
017c4261 | 4315 | int cpu; |
9f680ab4 | 4316 | |
47627678 PM |
4317 | rcu_early_boot_tests(); |
4318 | ||
f41d911f | 4319 | rcu_bootup_announce(); |
f885b7f2 | 4320 | rcu_init_geometry(); |
a87f203e PM |
4321 | rcu_init_one(&rcu_bh_state); |
4322 | rcu_init_one(&rcu_sched_state); | |
a3dc2948 PM |
4323 | if (dump_tree) |
4324 | rcu_dump_rcu_node_tree(&rcu_sched_state); | |
f41d911f | 4325 | __rcu_init_preempt(); |
b5b39360 | 4326 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
4327 | |
4328 | /* | |
4329 | * We don't need protection against CPU-hotplug here because | |
4330 | * this is called early in boot, before either interrupts | |
4331 | * or the scheduler are operational. | |
4332 | */ | |
d1d74d14 | 4333 | pm_notifier(rcu_pm_notify, 0); |
7ec99de3 | 4334 | for_each_online_cpu(cpu) { |
4df83742 | 4335 | rcutree_prepare_cpu(cpu); |
7ec99de3 PM |
4336 | rcu_cpu_starting(cpu); |
4337 | } | |
64db4cff PM |
4338 | } |
4339 | ||
3549c2bc | 4340 | #include "tree_exp.h" |
4102adab | 4341 | #include "tree_plugin.h" |