]>
Commit | Line | Data |
---|---|---|
22e40925 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
64db4cff PM |
2 | /* |
3 | * Read-Copy Update mechanism for mutual exclusion | |
4 | * | |
64db4cff PM |
5 | * Copyright IBM Corporation, 2008 |
6 | * | |
7 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | |
8 | * Manfred Spraul <manfred@colorfullife.com> | |
22e40925 | 9 | * Paul E. McKenney <paulmck@linux.ibm.com> Hierarchical version |
64db4cff | 10 | * |
22e40925 | 11 | * Based on the original work by Paul McKenney <paulmck@linux.ibm.com> |
64db4cff PM |
12 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. |
13 | * | |
14 | * For detailed explanation of Read-Copy Update mechanism see - | |
a71fca58 | 15 | * Documentation/RCU |
64db4cff | 16 | */ |
a7538352 JP |
17 | |
18 | #define pr_fmt(fmt) "rcu: " fmt | |
19 | ||
64db4cff PM |
20 | #include <linux/types.h> |
21 | #include <linux/kernel.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/spinlock.h> | |
24 | #include <linux/smp.h> | |
f9411ebe | 25 | #include <linux/rcupdate_wait.h> |
64db4cff PM |
26 | #include <linux/interrupt.h> |
27 | #include <linux/sched.h> | |
b17b0153 | 28 | #include <linux/sched/debug.h> |
c1dc0b9c | 29 | #include <linux/nmi.h> |
8826f3b0 | 30 | #include <linux/atomic.h> |
64db4cff | 31 | #include <linux/bitops.h> |
9984de1a | 32 | #include <linux/export.h> |
64db4cff PM |
33 | #include <linux/completion.h> |
34 | #include <linux/moduleparam.h> | |
35 | #include <linux/percpu.h> | |
36 | #include <linux/notifier.h> | |
37 | #include <linux/cpu.h> | |
38 | #include <linux/mutex.h> | |
39 | #include <linux/time.h> | |
bbad9379 | 40 | #include <linux/kernel_stat.h> |
a26ac245 PM |
41 | #include <linux/wait.h> |
42 | #include <linux/kthread.h> | |
ae7e81c0 | 43 | #include <uapi/linux/sched/types.h> |
268bb0ce | 44 | #include <linux/prefetch.h> |
3d3b7db0 PM |
45 | #include <linux/delay.h> |
46 | #include <linux/stop_machine.h> | |
661a85dc | 47 | #include <linux/random.h> |
af658dca | 48 | #include <linux/trace_events.h> |
d1d74d14 | 49 | #include <linux/suspend.h> |
a278d471 | 50 | #include <linux/ftrace.h> |
d3052109 | 51 | #include <linux/tick.h> |
2ccaff10 | 52 | #include <linux/sysrq.h> |
c13324a5 | 53 | #include <linux/kprobes.h> |
48d07c04 SAS |
54 | #include <linux/gfp.h> |
55 | #include <linux/oom.h> | |
56 | #include <linux/smpboot.h> | |
57 | #include <linux/jiffies.h> | |
58 | #include <linux/sched/isolation.h> | |
cfcdef5e | 59 | #include <linux/sched/clock.h> |
48d07c04 | 60 | #include "../time/tick-internal.h" |
64db4cff | 61 | |
4102adab | 62 | #include "tree.h" |
29c00b4a | 63 | #include "rcu.h" |
9f77da9f | 64 | |
4102adab PM |
65 | #ifdef MODULE_PARAM_PREFIX |
66 | #undef MODULE_PARAM_PREFIX | |
67 | #endif | |
68 | #define MODULE_PARAM_PREFIX "rcutree." | |
69 | ||
64db4cff PM |
70 | /* Data structures. */ |
71 | ||
f7f7bac9 | 72 | /* |
dc5a4f29 PM |
73 | * Steal a bit from the bottom of ->dynticks for idle entry/exit |
74 | * control. Initially this is for TLB flushing. | |
f7f7bac9 | 75 | */ |
dc5a4f29 PM |
76 | #define RCU_DYNTICK_CTRL_MASK 0x1 |
77 | #define RCU_DYNTICK_CTRL_CTR (RCU_DYNTICK_CTRL_MASK + 1) | |
78 | #ifndef rcu_eqs_special_exit | |
79 | #define rcu_eqs_special_exit() do { } while (0) | |
a8a29b3b AB |
80 | #endif |
81 | ||
4c5273bf PM |
82 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, rcu_data) = { |
83 | .dynticks_nesting = 1, | |
84 | .dynticks_nmi_nesting = DYNTICK_IRQ_NONIDLE, | |
dc5a4f29 | 85 | .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR), |
4c5273bf | 86 | }; |
358be2d3 PM |
87 | struct rcu_state rcu_state = { |
88 | .level = { &rcu_state.node[0] }, | |
358be2d3 PM |
89 | .gp_state = RCU_GP_IDLE, |
90 | .gp_seq = (0UL - 300UL) << RCU_SEQ_CTR_SHIFT, | |
91 | .barrier_mutex = __MUTEX_INITIALIZER(rcu_state.barrier_mutex), | |
92 | .name = RCU_NAME, | |
93 | .abbr = RCU_ABBR, | |
94 | .exp_mutex = __MUTEX_INITIALIZER(rcu_state.exp_mutex), | |
95 | .exp_wake_mutex = __MUTEX_INITIALIZER(rcu_state.exp_wake_mutex), | |
894d45bb | 96 | .ofl_lock = __RAW_SPIN_LOCK_UNLOCKED(rcu_state.ofl_lock), |
358be2d3 | 97 | }; |
27f4d280 | 98 | |
a3dc2948 PM |
99 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
100 | static bool dump_tree; | |
101 | module_param(dump_tree, bool, 0444); | |
48d07c04 SAS |
102 | /* By default, use RCU_SOFTIRQ instead of rcuc kthreads. */ |
103 | static bool use_softirq = 1; | |
104 | module_param(use_softirq, bool, 0444); | |
7fa27001 PM |
105 | /* Control rcu_node-tree auto-balancing at boot time. */ |
106 | static bool rcu_fanout_exact; | |
107 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
108 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
109 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 110 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 111 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 | 112 | /* Number of rcu_nodes at specified level. */ |
e95d68d2 | 113 | int num_rcu_lvl[] = NUM_RCU_LVL_INIT; |
f885b7f2 PM |
114 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
115 | ||
b0d30417 | 116 | /* |
52d7e48b PM |
117 | * The rcu_scheduler_active variable is initialized to the value |
118 | * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the | |
119 | * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE, | |
120 | * RCU can assume that there is but one task, allowing RCU to (for example) | |
0d95092c | 121 | * optimize synchronize_rcu() to a simple barrier(). When this variable |
52d7e48b PM |
122 | * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required |
123 | * to detect real grace periods. This variable is also used to suppress | |
124 | * boot-time false positives from lockdep-RCU error checking. Finally, it | |
125 | * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU | |
126 | * is fully initialized, including all of its kthreads having been spawned. | |
b0d30417 | 127 | */ |
bbad9379 PM |
128 | int rcu_scheduler_active __read_mostly; |
129 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
130 | ||
b0d30417 PM |
131 | /* |
132 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
133 | * during the early_initcall() processing, which is after the scheduler | |
134 | * is capable of creating new tasks. So RCU processing (for example, | |
135 | * creating tasks for RCU priority boosting) must be delayed until after | |
136 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
137 | * currently delay invocation of any RCU callbacks until after this point. | |
138 | * | |
139 | * It might later prove better for people registering RCU callbacks during | |
140 | * early boot to take responsibility for these callbacks, but one step at | |
141 | * a time. | |
142 | */ | |
143 | static int rcu_scheduler_fully_active __read_mostly; | |
144 | ||
b50912d0 PM |
145 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
146 | unsigned long gps, unsigned long flags); | |
0aa04b05 PM |
147 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
148 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 149 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 | 150 | static void invoke_rcu_core(void); |
63d4c8c9 | 151 | static void rcu_report_exp_rdp(struct rcu_data *rdp); |
3549c2bc | 152 | static void sync_sched_exp_online_cleanup(int cpu); |
a26ac245 | 153 | |
a94844b2 | 154 | /* rcuc/rcub kthread realtime priority */ |
26730f55 | 155 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; |
3ffe3d1a | 156 | module_param(kthread_prio, int, 0444); |
a94844b2 | 157 | |
8d7dc928 | 158 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd | 159 | |
90040c9e PM |
160 | static int gp_preinit_delay; |
161 | module_param(gp_preinit_delay, int, 0444); | |
162 | static int gp_init_delay; | |
163 | module_param(gp_init_delay, int, 0444); | |
164 | static int gp_cleanup_delay; | |
165 | module_param(gp_cleanup_delay, int, 0444); | |
0f41c0dd | 166 | |
4cf439a2 | 167 | /* Retrieve RCU kthreads priority for rcutorture */ |
4babd855 JFG |
168 | int rcu_get_gp_kthreads_prio(void) |
169 | { | |
170 | return kthread_prio; | |
171 | } | |
172 | EXPORT_SYMBOL_GPL(rcu_get_gp_kthreads_prio); | |
173 | ||
eab128e8 PM |
174 | /* |
175 | * Number of grace periods between delays, normalized by the duration of | |
bfd090be | 176 | * the delay. The longer the delay, the more the grace periods between |
eab128e8 PM |
177 | * each delay. The reason for this normalization is that it means that, |
178 | * for non-zero delays, the overall slowdown of grace periods is constant | |
179 | * regardless of the duration of the delay. This arrangement balances | |
180 | * the need for long delays to increase some race probabilities with the | |
181 | * need for fast grace periods to increase other race probabilities. | |
182 | */ | |
183 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */ | |
37745d28 | 184 | |
0aa04b05 PM |
185 | /* |
186 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
187 | * This will not be stable unless the rcu_node structure's ->lock is | |
188 | * held, but the bit corresponding to the current CPU will be stable | |
189 | * in most contexts. | |
190 | */ | |
191 | unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) | |
192 | { | |
7d0ae808 | 193 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
194 | } |
195 | ||
fc2219d4 | 196 | /* |
7d0ae808 | 197 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
198 | * permit this function to be invoked without holding the root rcu_node |
199 | * structure's ->lock, but of course results can be subject to change. | |
200 | */ | |
de8e8730 | 201 | static int rcu_gp_in_progress(void) |
fc2219d4 | 202 | { |
de8e8730 | 203 | return rcu_seq_state(rcu_seq_current(&rcu_state.gp_seq)); |
b1f77b05 IM |
204 | } |
205 | ||
903ee83d PM |
206 | /* |
207 | * Return the number of callbacks queued on the specified CPU. | |
208 | * Handles both the nocbs and normal cases. | |
209 | */ | |
210 | static long rcu_get_n_cbs_cpu(int cpu) | |
211 | { | |
212 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); | |
213 | ||
c035280f | 214 | if (rcu_segcblist_is_enabled(&rdp->cblist)) |
903ee83d | 215 | return rcu_segcblist_n_cbs(&rdp->cblist); |
c035280f | 216 | return 0; |
903ee83d PM |
217 | } |
218 | ||
d28139c4 | 219 | void rcu_softirq_qs(void) |
b1f77b05 | 220 | { |
45975c7d | 221 | rcu_qs(); |
d28139c4 | 222 | rcu_preempt_deferred_qs(current); |
b1f77b05 | 223 | } |
64db4cff | 224 | |
2625d469 PM |
225 | /* |
226 | * Record entry into an extended quiescent state. This is only to be | |
227 | * called when not already in an extended quiescent state. | |
228 | */ | |
229 | static void rcu_dynticks_eqs_enter(void) | |
230 | { | |
dc5a4f29 | 231 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
b8c17e66 | 232 | int seq; |
2625d469 PM |
233 | |
234 | /* | |
b8c17e66 | 235 | * CPUs seeing atomic_add_return() must see prior RCU read-side |
2625d469 PM |
236 | * critical sections, and we also must force ordering with the |
237 | * next idle sojourn. | |
238 | */ | |
dc5a4f29 | 239 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks); |
b8c17e66 PM |
240 | /* Better be in an extended quiescent state! */ |
241 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
242 | (seq & RCU_DYNTICK_CTRL_CTR)); | |
243 | /* Better not have special action (TLB flush) pending! */ | |
244 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
245 | (seq & RCU_DYNTICK_CTRL_MASK)); | |
2625d469 PM |
246 | } |
247 | ||
248 | /* | |
249 | * Record exit from an extended quiescent state. This is only to be | |
250 | * called from an extended quiescent state. | |
251 | */ | |
252 | static void rcu_dynticks_eqs_exit(void) | |
253 | { | |
dc5a4f29 | 254 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
b8c17e66 | 255 | int seq; |
2625d469 PM |
256 | |
257 | /* | |
b8c17e66 | 258 | * CPUs seeing atomic_add_return() must see prior idle sojourns, |
2625d469 PM |
259 | * and we also must force ordering with the next RCU read-side |
260 | * critical section. | |
261 | */ | |
dc5a4f29 | 262 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks); |
b8c17e66 PM |
263 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
264 | !(seq & RCU_DYNTICK_CTRL_CTR)); | |
265 | if (seq & RCU_DYNTICK_CTRL_MASK) { | |
dc5a4f29 | 266 | atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdp->dynticks); |
b8c17e66 PM |
267 | smp_mb__after_atomic(); /* _exit after clearing mask. */ |
268 | /* Prefer duplicate flushes to losing a flush. */ | |
269 | rcu_eqs_special_exit(); | |
270 | } | |
2625d469 PM |
271 | } |
272 | ||
273 | /* | |
274 | * Reset the current CPU's ->dynticks counter to indicate that the | |
275 | * newly onlined CPU is no longer in an extended quiescent state. | |
276 | * This will either leave the counter unchanged, or increment it | |
277 | * to the next non-quiescent value. | |
278 | * | |
279 | * The non-atomic test/increment sequence works because the upper bits | |
280 | * of the ->dynticks counter are manipulated only by the corresponding CPU, | |
281 | * or when the corresponding CPU is offline. | |
282 | */ | |
283 | static void rcu_dynticks_eqs_online(void) | |
284 | { | |
dc5a4f29 | 285 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
2625d469 | 286 | |
dc5a4f29 | 287 | if (atomic_read(&rdp->dynticks) & RCU_DYNTICK_CTRL_CTR) |
2625d469 | 288 | return; |
dc5a4f29 | 289 | atomic_add(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks); |
2625d469 PM |
290 | } |
291 | ||
02a5c550 PM |
292 | /* |
293 | * Is the current CPU in an extended quiescent state? | |
294 | * | |
295 | * No ordering, as we are sampling CPU-local information. | |
296 | */ | |
297 | bool rcu_dynticks_curr_cpu_in_eqs(void) | |
298 | { | |
dc5a4f29 | 299 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
02a5c550 | 300 | |
dc5a4f29 | 301 | return !(atomic_read(&rdp->dynticks) & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
302 | } |
303 | ||
8b2f63ab PM |
304 | /* |
305 | * Snapshot the ->dynticks counter with full ordering so as to allow | |
306 | * stable comparison of this counter with past and future snapshots. | |
307 | */ | |
dc5a4f29 | 308 | int rcu_dynticks_snap(struct rcu_data *rdp) |
8b2f63ab | 309 | { |
dc5a4f29 | 310 | int snap = atomic_add_return(0, &rdp->dynticks); |
8b2f63ab | 311 | |
b8c17e66 | 312 | return snap & ~RCU_DYNTICK_CTRL_MASK; |
8b2f63ab PM |
313 | } |
314 | ||
02a5c550 PM |
315 | /* |
316 | * Return true if the snapshot returned from rcu_dynticks_snap() | |
317 | * indicates that RCU is in an extended quiescent state. | |
318 | */ | |
319 | static bool rcu_dynticks_in_eqs(int snap) | |
320 | { | |
b8c17e66 | 321 | return !(snap & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
322 | } |
323 | ||
324 | /* | |
dc5a4f29 | 325 | * Return true if the CPU corresponding to the specified rcu_data |
02a5c550 PM |
326 | * structure has spent some time in an extended quiescent state since |
327 | * rcu_dynticks_snap() returned the specified snapshot. | |
328 | */ | |
dc5a4f29 | 329 | static bool rcu_dynticks_in_eqs_since(struct rcu_data *rdp, int snap) |
02a5c550 | 330 | { |
dc5a4f29 | 331 | return snap != rcu_dynticks_snap(rdp); |
02a5c550 PM |
332 | } |
333 | ||
b8c17e66 PM |
334 | /* |
335 | * Set the special (bottom) bit of the specified CPU so that it | |
336 | * will take special action (such as flushing its TLB) on the | |
337 | * next exit from an extended quiescent state. Returns true if | |
338 | * the bit was successfully set, or false if the CPU was not in | |
339 | * an extended quiescent state. | |
340 | */ | |
341 | bool rcu_eqs_special_set(int cpu) | |
342 | { | |
343 | int old; | |
344 | int new; | |
dc5a4f29 | 345 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
b8c17e66 PM |
346 | |
347 | do { | |
dc5a4f29 | 348 | old = atomic_read(&rdp->dynticks); |
b8c17e66 PM |
349 | if (old & RCU_DYNTICK_CTRL_CTR) |
350 | return false; | |
351 | new = old | RCU_DYNTICK_CTRL_MASK; | |
dc5a4f29 | 352 | } while (atomic_cmpxchg(&rdp->dynticks, old, new) != old); |
b8c17e66 | 353 | return true; |
6563de9d | 354 | } |
5cd37193 | 355 | |
4a81e832 PM |
356 | /* |
357 | * Let the RCU core know that this CPU has gone through the scheduler, | |
358 | * which is a quiescent state. This is called when the need for a | |
359 | * quiescent state is urgent, so we burn an atomic operation and full | |
360 | * memory barriers to let the RCU core know about it, regardless of what | |
361 | * this CPU might (or might not) do in the near future. | |
362 | * | |
0f9be8ca | 363 | * We inform the RCU core by emulating a zero-duration dyntick-idle period. |
46a5d164 | 364 | * |
3b57a399 | 365 | * The caller must have disabled interrupts and must not be idle. |
4a81e832 | 366 | */ |
366237e7 | 367 | void rcu_momentary_dyntick_idle(void) |
4a81e832 | 368 | { |
3b57a399 PM |
369 | int special; |
370 | ||
2dba13f0 | 371 | raw_cpu_write(rcu_data.rcu_need_heavy_qs, false); |
dc5a4f29 PM |
372 | special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR, |
373 | &this_cpu_ptr(&rcu_data)->dynticks); | |
3b57a399 PM |
374 | /* It is illegal to call this from idle state. */ |
375 | WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR)); | |
3e310098 | 376 | rcu_preempt_deferred_qs(current); |
4a81e832 | 377 | } |
79ba7ff5 | 378 | EXPORT_SYMBOL_GPL(rcu_momentary_dyntick_idle); |
4a81e832 | 379 | |
45975c7d | 380 | /** |
eddded80 | 381 | * rcu_is_cpu_rrupt_from_idle - see if interrupted from idle |
bb73c52b | 382 | * |
eddded80 | 383 | * If the current CPU is idle and running at a first-level (not nested) |
45975c7d PM |
384 | * interrupt from idle, return true. The caller must have at least |
385 | * disabled preemption. | |
5cd37193 | 386 | */ |
45975c7d | 387 | static int rcu_is_cpu_rrupt_from_idle(void) |
5cd37193 | 388 | { |
eddded80 JFG |
389 | /* Called only from within the scheduling-clock interrupt */ |
390 | lockdep_assert_in_irq(); | |
391 | ||
392 | /* Check for counter underflows */ | |
393 | RCU_LOCKDEP_WARN(__this_cpu_read(rcu_data.dynticks_nesting) < 0, | |
394 | "RCU dynticks_nesting counter underflow!"); | |
395 | RCU_LOCKDEP_WARN(__this_cpu_read(rcu_data.dynticks_nmi_nesting) <= 0, | |
396 | "RCU dynticks_nmi_nesting counter underflow/zero!"); | |
397 | ||
398 | /* Are we at first interrupt nesting level? */ | |
399 | if (__this_cpu_read(rcu_data.dynticks_nmi_nesting) != 1) | |
400 | return false; | |
401 | ||
402 | /* Does CPU appear to be idle from an RCU standpoint? */ | |
403 | return __this_cpu_read(rcu_data.dynticks_nesting) == 0; | |
5cd37193 | 404 | } |
5cd37193 | 405 | |
d5a9a8c3 PM |
406 | #define DEFAULT_RCU_BLIMIT 10 /* Maximum callbacks per rcu_do_batch ... */ |
407 | #define DEFAULT_MAX_RCU_BLIMIT 10000 /* ... even during callback flood. */ | |
17c7798b PM |
408 | static long blimit = DEFAULT_RCU_BLIMIT; |
409 | #define DEFAULT_RCU_QHIMARK 10000 /* If this many pending, ignore blimit. */ | |
410 | static long qhimark = DEFAULT_RCU_QHIMARK; | |
411 | #define DEFAULT_RCU_QLOMARK 100 /* Once only this many pending, use blimit. */ | |
412 | static long qlowmark = DEFAULT_RCU_QLOMARK; | |
64db4cff | 413 | |
878d7439 ED |
414 | module_param(blimit, long, 0444); |
415 | module_param(qhimark, long, 0444); | |
416 | module_param(qlowmark, long, 0444); | |
3d76c082 | 417 | |
026ad283 PM |
418 | static ulong jiffies_till_first_fqs = ULONG_MAX; |
419 | static ulong jiffies_till_next_fqs = ULONG_MAX; | |
8c7c4829 | 420 | static bool rcu_kick_kthreads; |
cfcdef5e ED |
421 | static int rcu_divisor = 7; |
422 | module_param(rcu_divisor, int, 0644); | |
423 | ||
424 | /* Force an exit from rcu_do_batch() after 3 milliseconds. */ | |
425 | static long rcu_resched_ns = 3 * NSEC_PER_MSEC; | |
426 | module_param(rcu_resched_ns, long, 0644); | |
d40011f6 | 427 | |
c06aed0e PM |
428 | /* |
429 | * How long the grace period must be before we start recruiting | |
430 | * quiescent-state help from rcu_note_context_switch(). | |
431 | */ | |
432 | static ulong jiffies_till_sched_qs = ULONG_MAX; | |
433 | module_param(jiffies_till_sched_qs, ulong, 0444); | |
85f2b60c | 434 | static ulong jiffies_to_sched_qs; /* See adjust_jiffies_till_sched_qs(). */ |
c06aed0e PM |
435 | module_param(jiffies_to_sched_qs, ulong, 0444); /* Display only! */ |
436 | ||
437 | /* | |
438 | * Make sure that we give the grace-period kthread time to detect any | |
439 | * idle CPUs before taking active measures to force quiescent states. | |
440 | * However, don't go below 100 milliseconds, adjusted upwards for really | |
441 | * large systems. | |
442 | */ | |
443 | static void adjust_jiffies_till_sched_qs(void) | |
444 | { | |
445 | unsigned long j; | |
446 | ||
447 | /* If jiffies_till_sched_qs was specified, respect the request. */ | |
448 | if (jiffies_till_sched_qs != ULONG_MAX) { | |
449 | WRITE_ONCE(jiffies_to_sched_qs, jiffies_till_sched_qs); | |
450 | return; | |
451 | } | |
85f2b60c | 452 | /* Otherwise, set to third fqs scan, but bound below on large system. */ |
c06aed0e PM |
453 | j = READ_ONCE(jiffies_till_first_fqs) + |
454 | 2 * READ_ONCE(jiffies_till_next_fqs); | |
455 | if (j < HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV) | |
456 | j = HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
457 | pr_info("RCU calculated value of scheduler-enlistment delay is %ld jiffies.\n", j); | |
458 | WRITE_ONCE(jiffies_to_sched_qs, j); | |
459 | } | |
460 | ||
67abb96c BP |
461 | static int param_set_first_fqs_jiffies(const char *val, const struct kernel_param *kp) |
462 | { | |
463 | ulong j; | |
464 | int ret = kstrtoul(val, 0, &j); | |
465 | ||
c06aed0e | 466 | if (!ret) { |
67abb96c | 467 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : j); |
c06aed0e PM |
468 | adjust_jiffies_till_sched_qs(); |
469 | } | |
67abb96c BP |
470 | return ret; |
471 | } | |
472 | ||
473 | static int param_set_next_fqs_jiffies(const char *val, const struct kernel_param *kp) | |
474 | { | |
475 | ulong j; | |
476 | int ret = kstrtoul(val, 0, &j); | |
477 | ||
c06aed0e | 478 | if (!ret) { |
67abb96c | 479 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : (j ?: 1)); |
c06aed0e PM |
480 | adjust_jiffies_till_sched_qs(); |
481 | } | |
67abb96c BP |
482 | return ret; |
483 | } | |
484 | ||
485 | static struct kernel_param_ops first_fqs_jiffies_ops = { | |
486 | .set = param_set_first_fqs_jiffies, | |
487 | .get = param_get_ulong, | |
488 | }; | |
489 | ||
490 | static struct kernel_param_ops next_fqs_jiffies_ops = { | |
491 | .set = param_set_next_fqs_jiffies, | |
492 | .get = param_get_ulong, | |
493 | }; | |
494 | ||
495 | module_param_cb(jiffies_till_first_fqs, &first_fqs_jiffies_ops, &jiffies_till_first_fqs, 0644); | |
496 | module_param_cb(jiffies_till_next_fqs, &next_fqs_jiffies_ops, &jiffies_till_next_fqs, 0644); | |
8c7c4829 | 497 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 498 | |
8ff0b907 | 499 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)); |
dd7dafd1 | 500 | static int rcu_pending(int user); |
64db4cff PM |
501 | |
502 | /* | |
17ef2fe9 | 503 | * Return the number of RCU GPs completed thus far for debug & stats. |
64db4cff | 504 | */ |
17ef2fe9 | 505 | unsigned long rcu_get_gp_seq(void) |
917963d0 | 506 | { |
16fc9c60 | 507 | return READ_ONCE(rcu_state.gp_seq); |
917963d0 | 508 | } |
17ef2fe9 | 509 | EXPORT_SYMBOL_GPL(rcu_get_gp_seq); |
917963d0 | 510 | |
291783b8 PM |
511 | /* |
512 | * Return the number of RCU expedited batches completed thus far for | |
513 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
514 | * numbers mean idle. The value returned will thus be roughly double | |
515 | * the cumulative batches since boot. | |
516 | */ | |
517 | unsigned long rcu_exp_batches_completed(void) | |
518 | { | |
16fc9c60 | 519 | return rcu_state.expedited_sequence; |
291783b8 PM |
520 | } |
521 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
522 | ||
fd897573 PM |
523 | /* |
524 | * Return the root node of the rcu_state structure. | |
525 | */ | |
526 | static struct rcu_node *rcu_get_root(void) | |
527 | { | |
528 | return &rcu_state.node[0]; | |
529 | } | |
530 | ||
69196019 PM |
531 | /* |
532 | * Convert a ->gp_state value to a character string. | |
533 | */ | |
534 | static const char *gp_state_getname(short gs) | |
535 | { | |
536 | if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names)) | |
537 | return "???"; | |
538 | return gp_state_names[gs]; | |
539 | } | |
540 | ||
ad0dc7f9 PM |
541 | /* |
542 | * Send along grace-period-related data for rcutorture diagnostics. | |
543 | */ | |
544 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
aebc8264 | 545 | unsigned long *gp_seq) |
ad0dc7f9 | 546 | { |
ad0dc7f9 PM |
547 | switch (test_type) { |
548 | case RCU_FLAVOR: | |
f7dd7d44 PM |
549 | *flags = READ_ONCE(rcu_state.gp_flags); |
550 | *gp_seq = rcu_seq_current(&rcu_state.gp_seq); | |
ad0dc7f9 PM |
551 | break; |
552 | default: | |
553 | break; | |
554 | } | |
ad0dc7f9 PM |
555 | } |
556 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
557 | ||
9b2e4f18 | 558 | /* |
215bba9f PM |
559 | * Enter an RCU extended quiescent state, which can be either the |
560 | * idle loop or adaptive-tickless usermode execution. | |
9b2e4f18 | 561 | * |
215bba9f PM |
562 | * We crowbar the ->dynticks_nmi_nesting field to zero to allow for |
563 | * the possibility of usermode upcalls having messed up our count | |
564 | * of interrupt nesting level during the prior busy period. | |
9b2e4f18 | 565 | */ |
215bba9f | 566 | static void rcu_eqs_enter(bool user) |
9b2e4f18 | 567 | { |
4c5273bf | 568 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
96d3fd0d | 569 | |
4c5273bf PM |
570 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting != DYNTICK_IRQ_NONIDLE); |
571 | WRITE_ONCE(rdp->dynticks_nmi_nesting, 0); | |
215bba9f | 572 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
4c5273bf PM |
573 | rdp->dynticks_nesting == 0); |
574 | if (rdp->dynticks_nesting != 1) { | |
575 | rdp->dynticks_nesting--; | |
215bba9f | 576 | return; |
9b2e4f18 | 577 | } |
96d3fd0d | 578 | |
b04db8e1 | 579 | lockdep_assert_irqs_disabled(); |
dc5a4f29 | 580 | trace_rcu_dyntick(TPS("Start"), rdp->dynticks_nesting, 0, rdp->dynticks); |
e68bbb26 | 581 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
b97d23c5 PM |
582 | rdp = this_cpu_ptr(&rcu_data); |
583 | do_nocb_deferred_wakeup(rdp); | |
198bbf81 | 584 | rcu_prepare_for_idle(); |
3e310098 | 585 | rcu_preempt_deferred_qs(current); |
4c5273bf | 586 | WRITE_ONCE(rdp->dynticks_nesting, 0); /* Avoid irq-access tearing. */ |
844ccdd7 | 587 | rcu_dynticks_eqs_enter(); |
176f8f7a | 588 | rcu_dynticks_task_enter(); |
64db4cff | 589 | } |
adf5091e FW |
590 | |
591 | /** | |
592 | * rcu_idle_enter - inform RCU that current CPU is entering idle | |
593 | * | |
594 | * Enter idle mode, in other words, -leave- the mode in which RCU | |
595 | * read-side critical sections can occur. (Though RCU read-side | |
596 | * critical sections can occur in irq handlers in idle, a possibility | |
597 | * handled by irq_enter() and irq_exit().) | |
598 | * | |
c0da313e PM |
599 | * If you add or remove a call to rcu_idle_enter(), be sure to test with |
600 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
601 | */ |
602 | void rcu_idle_enter(void) | |
603 | { | |
b04db8e1 | 604 | lockdep_assert_irqs_disabled(); |
cb349ca9 | 605 | rcu_eqs_enter(false); |
adf5091e | 606 | } |
64db4cff | 607 | |
d1ec4c34 | 608 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
609 | /** |
610 | * rcu_user_enter - inform RCU that we are resuming userspace. | |
611 | * | |
612 | * Enter RCU idle mode right before resuming userspace. No use of RCU | |
613 | * is permitted between this call and rcu_user_exit(). This way the | |
614 | * CPU doesn't need to maintain the tick for RCU maintenance purposes | |
615 | * when the CPU runs in userspace. | |
c0da313e PM |
616 | * |
617 | * If you add or remove a call to rcu_user_enter(), be sure to test with | |
618 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
619 | */ |
620 | void rcu_user_enter(void) | |
621 | { | |
b04db8e1 | 622 | lockdep_assert_irqs_disabled(); |
d4db30af | 623 | rcu_eqs_enter(true); |
adf5091e | 624 | } |
d1ec4c34 | 625 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 626 | |
cf7614e1 | 627 | /* |
fd581a91 | 628 | * If we are returning from the outermost NMI handler that interrupted an |
dc5a4f29 | 629 | * RCU-idle period, update rdp->dynticks and rdp->dynticks_nmi_nesting |
fd581a91 PM |
630 | * to let the RCU grace-period handling know that the CPU is back to |
631 | * being RCU-idle. | |
632 | * | |
cf7614e1 | 633 | * If you add or remove a call to rcu_nmi_exit_common(), be sure to test |
fd581a91 PM |
634 | * with CONFIG_RCU_EQS_DEBUG=y. |
635 | */ | |
cf7614e1 | 636 | static __always_inline void rcu_nmi_exit_common(bool irq) |
fd581a91 | 637 | { |
4c5273bf | 638 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
fd581a91 PM |
639 | |
640 | /* | |
641 | * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks. | |
642 | * (We are exiting an NMI handler, so RCU better be paying attention | |
643 | * to us!) | |
644 | */ | |
4c5273bf | 645 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting <= 0); |
fd581a91 PM |
646 | WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs()); |
647 | ||
648 | /* | |
649 | * If the nesting level is not 1, the CPU wasn't RCU-idle, so | |
650 | * leave it in non-RCU-idle state. | |
651 | */ | |
4c5273bf | 652 | if (rdp->dynticks_nmi_nesting != 1) { |
dc5a4f29 | 653 | trace_rcu_dyntick(TPS("--="), rdp->dynticks_nmi_nesting, rdp->dynticks_nmi_nesting - 2, rdp->dynticks); |
4c5273bf PM |
654 | WRITE_ONCE(rdp->dynticks_nmi_nesting, /* No store tearing. */ |
655 | rdp->dynticks_nmi_nesting - 2); | |
fd581a91 PM |
656 | return; |
657 | } | |
658 | ||
659 | /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */ | |
dc5a4f29 | 660 | trace_rcu_dyntick(TPS("Startirq"), rdp->dynticks_nmi_nesting, 0, rdp->dynticks); |
4c5273bf | 661 | WRITE_ONCE(rdp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */ |
cf7614e1 BP |
662 | |
663 | if (irq) | |
664 | rcu_prepare_for_idle(); | |
665 | ||
fd581a91 | 666 | rcu_dynticks_eqs_enter(); |
cf7614e1 BP |
667 | |
668 | if (irq) | |
669 | rcu_dynticks_task_enter(); | |
670 | } | |
671 | ||
672 | /** | |
673 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
cf7614e1 BP |
674 | * |
675 | * If you add or remove a call to rcu_nmi_exit(), be sure to test | |
676 | * with CONFIG_RCU_EQS_DEBUG=y. | |
677 | */ | |
678 | void rcu_nmi_exit(void) | |
679 | { | |
680 | rcu_nmi_exit_common(false); | |
fd581a91 PM |
681 | } |
682 | ||
9b2e4f18 PM |
683 | /** |
684 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
685 | * | |
686 | * Exit from an interrupt handler, which might possibly result in entering | |
687 | * idle mode, in other words, leaving the mode in which read-side critical | |
7c9906ca | 688 | * sections can occur. The caller must have disabled interrupts. |
64db4cff | 689 | * |
9b2e4f18 PM |
690 | * This code assumes that the idle loop never does anything that might |
691 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
58721f5d PM |
692 | * architecture's idle loop violates this assumption, RCU will give you what |
693 | * you deserve, good and hard. But very infrequently and irreproducibly. | |
9b2e4f18 PM |
694 | * |
695 | * Use things like work queues to work around this limitation. | |
696 | * | |
697 | * You have been warned. | |
c0da313e PM |
698 | * |
699 | * If you add or remove a call to rcu_irq_exit(), be sure to test with | |
700 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 701 | */ |
9b2e4f18 | 702 | void rcu_irq_exit(void) |
64db4cff | 703 | { |
b04db8e1 | 704 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 705 | rcu_nmi_exit_common(true); |
7c9906ca PM |
706 | } |
707 | ||
708 | /* | |
709 | * Wrapper for rcu_irq_exit() where interrupts are enabled. | |
c0da313e PM |
710 | * |
711 | * If you add or remove a call to rcu_irq_exit_irqson(), be sure to test | |
712 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
713 | */ |
714 | void rcu_irq_exit_irqson(void) | |
715 | { | |
716 | unsigned long flags; | |
717 | ||
718 | local_irq_save(flags); | |
719 | rcu_irq_exit(); | |
9b2e4f18 PM |
720 | local_irq_restore(flags); |
721 | } | |
722 | ||
adf5091e FW |
723 | /* |
724 | * Exit an RCU extended quiescent state, which can be either the | |
725 | * idle loop or adaptive-tickless usermode execution. | |
51a1fd30 PM |
726 | * |
727 | * We crowbar the ->dynticks_nmi_nesting field to DYNTICK_IRQ_NONIDLE to | |
728 | * allow for the possibility of usermode upcalls messing up our count of | |
729 | * interrupt nesting level during the busy period that is just now starting. | |
9b2e4f18 | 730 | */ |
adf5091e | 731 | static void rcu_eqs_exit(bool user) |
9b2e4f18 | 732 | { |
4c5273bf | 733 | struct rcu_data *rdp; |
84585aa8 | 734 | long oldval; |
9b2e4f18 | 735 | |
b04db8e1 | 736 | lockdep_assert_irqs_disabled(); |
4c5273bf PM |
737 | rdp = this_cpu_ptr(&rcu_data); |
738 | oldval = rdp->dynticks_nesting; | |
1ce46ee5 | 739 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0); |
51a1fd30 | 740 | if (oldval) { |
4c5273bf | 741 | rdp->dynticks_nesting++; |
9dd238e2 | 742 | return; |
3a592405 | 743 | } |
9dd238e2 PM |
744 | rcu_dynticks_task_exit(); |
745 | rcu_dynticks_eqs_exit(); | |
746 | rcu_cleanup_after_idle(); | |
dc5a4f29 | 747 | trace_rcu_dyntick(TPS("End"), rdp->dynticks_nesting, 1, rdp->dynticks); |
e68bbb26 | 748 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
4c5273bf PM |
749 | WRITE_ONCE(rdp->dynticks_nesting, 1); |
750 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting); | |
751 | WRITE_ONCE(rdp->dynticks_nmi_nesting, DYNTICK_IRQ_NONIDLE); | |
9b2e4f18 | 752 | } |
adf5091e FW |
753 | |
754 | /** | |
755 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
756 | * | |
757 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
758 | * read-side critical sections can occur. | |
759 | * | |
c0da313e PM |
760 | * If you add or remove a call to rcu_idle_exit(), be sure to test with |
761 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
762 | */ |
763 | void rcu_idle_exit(void) | |
764 | { | |
c5d900bf FW |
765 | unsigned long flags; |
766 | ||
767 | local_irq_save(flags); | |
cb349ca9 | 768 | rcu_eqs_exit(false); |
c5d900bf | 769 | local_irq_restore(flags); |
adf5091e | 770 | } |
9b2e4f18 | 771 | |
d1ec4c34 | 772 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
773 | /** |
774 | * rcu_user_exit - inform RCU that we are exiting userspace. | |
775 | * | |
776 | * Exit RCU idle mode while entering the kernel because it can | |
777 | * run a RCU read side critical section anytime. | |
c0da313e PM |
778 | * |
779 | * If you add or remove a call to rcu_user_exit(), be sure to test with | |
780 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
781 | */ |
782 | void rcu_user_exit(void) | |
783 | { | |
91d1aa43 | 784 | rcu_eqs_exit(1); |
adf5091e | 785 | } |
d1ec4c34 | 786 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 787 | |
64db4cff | 788 | /** |
cf7614e1 BP |
789 | * rcu_nmi_enter_common - inform RCU of entry to NMI context |
790 | * @irq: Is this call from rcu_irq_enter? | |
64db4cff | 791 | * |
dc5a4f29 | 792 | * If the CPU was idle from RCU's viewpoint, update rdp->dynticks and |
4c5273bf | 793 | * rdp->dynticks_nmi_nesting to let the RCU grace-period handling know |
734d1680 PM |
794 | * that the CPU is active. This implementation permits nested NMIs, as |
795 | * long as the nesting level does not overflow an int. (You will probably | |
796 | * run out of stack space first.) | |
c0da313e | 797 | * |
cf7614e1 | 798 | * If you add or remove a call to rcu_nmi_enter_common(), be sure to test |
c0da313e | 799 | * with CONFIG_RCU_EQS_DEBUG=y. |
64db4cff | 800 | */ |
cf7614e1 | 801 | static __always_inline void rcu_nmi_enter_common(bool irq) |
64db4cff | 802 | { |
4c5273bf | 803 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
84585aa8 | 804 | long incby = 2; |
64db4cff | 805 | |
734d1680 | 806 | /* Complain about underflow. */ |
4c5273bf | 807 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting < 0); |
734d1680 PM |
808 | |
809 | /* | |
810 | * If idle from RCU viewpoint, atomically increment ->dynticks | |
811 | * to mark non-idle and increment ->dynticks_nmi_nesting by one. | |
812 | * Otherwise, increment ->dynticks_nmi_nesting by two. This means | |
813 | * if ->dynticks_nmi_nesting is equal to one, we are guaranteed | |
814 | * to be in the outermost NMI handler that interrupted an RCU-idle | |
815 | * period (observation due to Andy Lutomirski). | |
816 | */ | |
02a5c550 | 817 | if (rcu_dynticks_curr_cpu_in_eqs()) { |
cf7614e1 BP |
818 | |
819 | if (irq) | |
820 | rcu_dynticks_task_exit(); | |
821 | ||
2625d469 | 822 | rcu_dynticks_eqs_exit(); |
cf7614e1 BP |
823 | |
824 | if (irq) | |
825 | rcu_cleanup_after_idle(); | |
826 | ||
734d1680 | 827 | incby = 1; |
b200a048 PM |
828 | } else if (tick_nohz_full_cpu(rdp->cpu) && |
829 | rdp->dynticks_nmi_nesting == DYNTICK_IRQ_NONIDLE && | |
516e5ae0 | 830 | READ_ONCE(rdp->rcu_urgent_qs) && !rdp->rcu_forced_tick) { |
b200a048 PM |
831 | rdp->rcu_forced_tick = true; |
832 | tick_dep_set_cpu(rdp->cpu, TICK_DEP_BIT_RCU); | |
734d1680 | 833 | } |
bd2b879a | 834 | trace_rcu_dyntick(incby == 1 ? TPS("Endirq") : TPS("++="), |
4c5273bf | 835 | rdp->dynticks_nmi_nesting, |
dc5a4f29 | 836 | rdp->dynticks_nmi_nesting + incby, rdp->dynticks); |
4c5273bf PM |
837 | WRITE_ONCE(rdp->dynticks_nmi_nesting, /* Prevent store tearing. */ |
838 | rdp->dynticks_nmi_nesting + incby); | |
734d1680 | 839 | barrier(); |
64db4cff PM |
840 | } |
841 | ||
cf7614e1 BP |
842 | /** |
843 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
844 | */ | |
845 | void rcu_nmi_enter(void) | |
846 | { | |
847 | rcu_nmi_enter_common(false); | |
848 | } | |
c13324a5 | 849 | NOKPROBE_SYMBOL(rcu_nmi_enter); |
cf7614e1 | 850 | |
64db4cff | 851 | /** |
9b2e4f18 | 852 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle |
64db4cff | 853 | * |
9b2e4f18 PM |
854 | * Enter an interrupt handler, which might possibly result in exiting |
855 | * idle mode, in other words, entering the mode in which read-side critical | |
7c9906ca | 856 | * sections can occur. The caller must have disabled interrupts. |
c0da313e | 857 | * |
9b2e4f18 | 858 | * Note that the Linux kernel is fully capable of entering an interrupt |
58721f5d PM |
859 | * handler that it never exits, for example when doing upcalls to user mode! |
860 | * This code assumes that the idle loop never does upcalls to user mode. | |
861 | * If your architecture's idle loop does do upcalls to user mode (or does | |
862 | * anything else that results in unbalanced calls to the irq_enter() and | |
863 | * irq_exit() functions), RCU will give you what you deserve, good and hard. | |
864 | * But very infrequently and irreproducibly. | |
9b2e4f18 PM |
865 | * |
866 | * Use things like work queues to work around this limitation. | |
867 | * | |
868 | * You have been warned. | |
c0da313e PM |
869 | * |
870 | * If you add or remove a call to rcu_irq_enter(), be sure to test with | |
871 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 872 | */ |
9b2e4f18 | 873 | void rcu_irq_enter(void) |
64db4cff | 874 | { |
b04db8e1 | 875 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 876 | rcu_nmi_enter_common(true); |
7c9906ca | 877 | } |
734d1680 | 878 | |
7c9906ca PM |
879 | /* |
880 | * Wrapper for rcu_irq_enter() where interrupts are enabled. | |
c0da313e PM |
881 | * |
882 | * If you add or remove a call to rcu_irq_enter_irqson(), be sure to test | |
883 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
884 | */ |
885 | void rcu_irq_enter_irqson(void) | |
886 | { | |
887 | unsigned long flags; | |
734d1680 | 888 | |
7c9906ca PM |
889 | local_irq_save(flags); |
890 | rcu_irq_enter(); | |
64db4cff | 891 | local_irq_restore(flags); |
64db4cff PM |
892 | } |
893 | ||
66e4c33b | 894 | /* |
516e5ae0 JFG |
895 | * If any sort of urgency was applied to the current CPU (for example, |
896 | * the scheduler-clock interrupt was enabled on a nohz_full CPU) in order | |
897 | * to get to a quiescent state, disable it. | |
66e4c33b | 898 | */ |
516e5ae0 | 899 | static void rcu_disable_urgency_upon_qs(struct rcu_data *rdp) |
66e4c33b | 900 | { |
516e5ae0 JFG |
901 | WRITE_ONCE(rdp->rcu_urgent_qs, false); |
902 | WRITE_ONCE(rdp->rcu_need_heavy_qs, false); | |
66e4c33b PM |
903 | if (tick_nohz_full_cpu(rdp->cpu) && rdp->rcu_forced_tick) { |
904 | tick_dep_clear_cpu(rdp->cpu, TICK_DEP_BIT_RCU); | |
905 | rdp->rcu_forced_tick = false; | |
906 | } | |
907 | } | |
908 | ||
5c173eb8 | 909 | /** |
2320bda2 | 910 | * rcu_is_watching - see if RCU thinks that the current CPU is not idle |
64db4cff | 911 | * |
791875d1 PM |
912 | * Return true if RCU is watching the running CPU, which means that this |
913 | * CPU can safely enter RCU read-side critical sections. In other words, | |
2320bda2 ZZ |
914 | * if the current CPU is not in its idle loop or is in an interrupt or |
915 | * NMI handler, return true. | |
64db4cff | 916 | */ |
9418fb20 | 917 | bool notrace rcu_is_watching(void) |
64db4cff | 918 | { |
f534ed1f | 919 | bool ret; |
34240697 | 920 | |
46f00d18 | 921 | preempt_disable_notrace(); |
791875d1 | 922 | ret = !rcu_dynticks_curr_cpu_in_eqs(); |
46f00d18 | 923 | preempt_enable_notrace(); |
34240697 | 924 | return ret; |
64db4cff | 925 | } |
5c173eb8 | 926 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 927 | |
bcbfdd01 PM |
928 | /* |
929 | * If a holdout task is actually running, request an urgent quiescent | |
930 | * state from its CPU. This is unsynchronized, so migrations can cause | |
931 | * the request to go to the wrong CPU. Which is OK, all that will happen | |
932 | * is that the CPU's next context switch will be a bit slower and next | |
933 | * time around this task will generate another request. | |
934 | */ | |
935 | void rcu_request_urgent_qs_task(struct task_struct *t) | |
936 | { | |
937 | int cpu; | |
938 | ||
939 | barrier(); | |
940 | cpu = task_cpu(t); | |
941 | if (!task_curr(t)) | |
942 | return; /* This task is not running on that CPU. */ | |
2dba13f0 | 943 | smp_store_release(per_cpu_ptr(&rcu_data.rcu_urgent_qs, cpu), true); |
bcbfdd01 PM |
944 | } |
945 | ||
62fde6ed | 946 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
947 | |
948 | /* | |
5554788e | 949 | * Is the current CPU online as far as RCU is concerned? |
2036d94a | 950 | * |
5554788e PM |
951 | * Disable preemption to avoid false positives that could otherwise |
952 | * happen due to the current CPU number being sampled, this task being | |
953 | * preempted, its old CPU being taken offline, resuming on some other CPU, | |
49918a54 | 954 | * then determining that its old CPU is now offline. |
c0d6d01b | 955 | * |
5554788e PM |
956 | * Disable checking if in an NMI handler because we cannot safely |
957 | * report errors from NMI handlers anyway. In addition, it is OK to use | |
958 | * RCU on an offline processor during initial boot, hence the check for | |
959 | * rcu_scheduler_fully_active. | |
c0d6d01b PM |
960 | */ |
961 | bool rcu_lockdep_current_cpu_online(void) | |
962 | { | |
2036d94a PM |
963 | struct rcu_data *rdp; |
964 | struct rcu_node *rnp; | |
b97d23c5 | 965 | bool ret = false; |
c0d6d01b | 966 | |
5554788e | 967 | if (in_nmi() || !rcu_scheduler_fully_active) |
f6f7ee9a | 968 | return true; |
c0d6d01b | 969 | preempt_disable(); |
b97d23c5 PM |
970 | rdp = this_cpu_ptr(&rcu_data); |
971 | rnp = rdp->mynode; | |
972 | if (rdp->grpmask & rcu_rnp_online_cpus(rnp)) | |
973 | ret = true; | |
c0d6d01b | 974 | preempt_enable(); |
b97d23c5 | 975 | return ret; |
c0d6d01b PM |
976 | } |
977 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
978 | ||
62fde6ed | 979 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 980 | |
9b9500da PM |
981 | /* |
982 | * We are reporting a quiescent state on behalf of some other CPU, so | |
983 | * it is our responsibility to check for and handle potential overflow | |
a66ae8ae | 984 | * of the rcu_node ->gp_seq counter with respect to the rcu_data counters. |
9b9500da PM |
985 | * After all, the CPU might be in deep idle state, and thus executing no |
986 | * code whatsoever. | |
987 | */ | |
988 | static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp) | |
989 | { | |
a32e01ee | 990 | raw_lockdep_assert_held_rcu_node(rnp); |
a66ae8ae PM |
991 | if (ULONG_CMP_LT(rcu_seq_current(&rdp->gp_seq) + ULONG_MAX / 4, |
992 | rnp->gp_seq)) | |
9b9500da | 993 | WRITE_ONCE(rdp->gpwrap, true); |
8aa670cd PM |
994 | if (ULONG_CMP_LT(rdp->rcu_iw_gp_seq + ULONG_MAX / 4, rnp->gp_seq)) |
995 | rdp->rcu_iw_gp_seq = rnp->gp_seq + ULONG_MAX / 4; | |
9b9500da PM |
996 | } |
997 | ||
64db4cff PM |
998 | /* |
999 | * Snapshot the specified CPU's dynticks counter so that we can later | |
1000 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 1001 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 1002 | */ |
fe5ac724 | 1003 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
64db4cff | 1004 | { |
dc5a4f29 | 1005 | rdp->dynticks_snap = rcu_dynticks_snap(rdp); |
02a5c550 | 1006 | if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) { |
88d1bead | 1007 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 1008 | rcu_gpnum_ovf(rdp->mynode, rdp); |
23a9bacd | 1009 | return 1; |
7941dbde | 1010 | } |
23a9bacd | 1011 | return 0; |
64db4cff PM |
1012 | } |
1013 | ||
1014 | /* | |
1015 | * Return true if the specified CPU has passed through a quiescent | |
1016 | * state by virtue of being in or having passed through an dynticks | |
1017 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 1018 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 1019 | */ |
fe5ac724 | 1020 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
64db4cff | 1021 | { |
3a19b46a | 1022 | unsigned long jtsq; |
0f9be8ca | 1023 | bool *rnhqp; |
9226b10d | 1024 | bool *ruqp; |
9b9500da | 1025 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
1026 | |
1027 | /* | |
1028 | * If the CPU passed through or entered a dynticks idle phase with | |
1029 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
1030 | * already acknowledged the request to pass through a quiescent | |
1031 | * state. Either way, that CPU cannot possibly be in an RCU | |
1032 | * read-side critical section that started before the beginning | |
1033 | * of the current RCU grace period. | |
1034 | */ | |
dc5a4f29 | 1035 | if (rcu_dynticks_in_eqs_since(rdp, rdp->dynticks_snap)) { |
88d1bead | 1036 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 1037 | rcu_gpnum_ovf(rnp, rdp); |
3a19b46a PM |
1038 | return 1; |
1039 | } | |
1040 | ||
f2e2df59 PM |
1041 | /* If waiting too long on an offline CPU, complain. */ |
1042 | if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp)) && | |
88d1bead | 1043 | time_after(jiffies, rcu_state.gp_start + HZ)) { |
f2e2df59 PM |
1044 | bool onl; |
1045 | struct rcu_node *rnp1; | |
1046 | ||
1047 | WARN_ON(1); /* Offline CPUs are supposed to report QS! */ | |
1048 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", | |
1049 | __func__, rnp->grplo, rnp->grphi, rnp->level, | |
1050 | (long)rnp->gp_seq, (long)rnp->completedqs); | |
1051 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
1052 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx ->rcu_gp_init_mask %#lx\n", | |
1053 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext, rnp1->rcu_gp_init_mask); | |
1054 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); | |
1055 | pr_info("%s %d: %c online: %ld(%d) offline: %ld(%d)\n", | |
1056 | __func__, rdp->cpu, ".o"[onl], | |
1057 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
1058 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
1059 | return 1; /* Break things loose after complaining. */ | |
1060 | } | |
1061 | ||
65d798f0 | 1062 | /* |
4a81e832 | 1063 | * A CPU running for an extended time within the kernel can |
c06aed0e PM |
1064 | * delay RCU grace periods: (1) At age jiffies_to_sched_qs, |
1065 | * set .rcu_urgent_qs, (2) At age 2*jiffies_to_sched_qs, set | |
7e28c5af PM |
1066 | * both .rcu_need_heavy_qs and .rcu_urgent_qs. Note that the |
1067 | * unsynchronized assignments to the per-CPU rcu_need_heavy_qs | |
1068 | * variable are safe because the assignments are repeated if this | |
1069 | * CPU failed to pass through a quiescent state. This code | |
c06aed0e | 1070 | * also checks .jiffies_resched in case jiffies_to_sched_qs |
7e28c5af | 1071 | * is set way high. |
6193c76a | 1072 | */ |
c06aed0e | 1073 | jtsq = READ_ONCE(jiffies_to_sched_qs); |
2dba13f0 PM |
1074 | ruqp = per_cpu_ptr(&rcu_data.rcu_urgent_qs, rdp->cpu); |
1075 | rnhqp = &per_cpu(rcu_data.rcu_need_heavy_qs, rdp->cpu); | |
0f9be8ca | 1076 | if (!READ_ONCE(*rnhqp) && |
7e28c5af | 1077 | (time_after(jiffies, rcu_state.gp_start + jtsq * 2) || |
88d1bead | 1078 | time_after(jiffies, rcu_state.jiffies_resched))) { |
0f9be8ca | 1079 | WRITE_ONCE(*rnhqp, true); |
9226b10d PM |
1080 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ |
1081 | smp_store_release(ruqp, true); | |
7e28c5af PM |
1082 | } else if (time_after(jiffies, rcu_state.gp_start + jtsq)) { |
1083 | WRITE_ONCE(*ruqp, true); | |
6193c76a PM |
1084 | } |
1085 | ||
28053bc7 | 1086 | /* |
c98cac60 | 1087 | * NO_HZ_FULL CPUs can run in-kernel without rcu_sched_clock_irq! |
d3052109 PM |
1088 | * The above code handles this, but only for straight cond_resched(). |
1089 | * And some in-kernel loops check need_resched() before calling | |
1090 | * cond_resched(), which defeats the above code for CPUs that are | |
1091 | * running in-kernel with scheduling-clock interrupts disabled. | |
1092 | * So hit them over the head with the resched_cpu() hammer! | |
28053bc7 | 1093 | */ |
d3052109 PM |
1094 | if (tick_nohz_full_cpu(rdp->cpu) && |
1095 | time_after(jiffies, | |
1096 | READ_ONCE(rdp->last_fqs_resched) + jtsq * 3)) { | |
05ef9e9e | 1097 | WRITE_ONCE(*ruqp, true); |
28053bc7 | 1098 | resched_cpu(rdp->cpu); |
d3052109 PM |
1099 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); |
1100 | } | |
1101 | ||
1102 | /* | |
1103 | * If more than halfway to RCU CPU stall-warning time, invoke | |
1104 | * resched_cpu() more frequently to try to loosen things up a bit. | |
1105 | * Also check to see if the CPU is getting hammered with interrupts, | |
1106 | * but only once per grace period, just to keep the IPIs down to | |
1107 | * a dull roar. | |
1108 | */ | |
1109 | if (time_after(jiffies, rcu_state.jiffies_resched)) { | |
1110 | if (time_after(jiffies, | |
1111 | READ_ONCE(rdp->last_fqs_resched) + jtsq)) { | |
1112 | resched_cpu(rdp->cpu); | |
1113 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); | |
1114 | } | |
9b9500da | 1115 | if (IS_ENABLED(CONFIG_IRQ_WORK) && |
8aa670cd | 1116 | !rdp->rcu_iw_pending && rdp->rcu_iw_gp_seq != rnp->gp_seq && |
9b9500da PM |
1117 | (rnp->ffmask & rdp->grpmask)) { |
1118 | init_irq_work(&rdp->rcu_iw, rcu_iw_handler); | |
1119 | rdp->rcu_iw_pending = true; | |
8aa670cd | 1120 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1121 | irq_work_queue_on(&rdp->rcu_iw, rdp->cpu); |
1122 | } | |
1123 | } | |
4914950a | 1124 | |
a82dcc76 | 1125 | return 0; |
64db4cff PM |
1126 | } |
1127 | ||
41e80595 PM |
1128 | /* Trace-event wrapper function for trace_rcu_future_grace_period. */ |
1129 | static void trace_rcu_this_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
b73de91d | 1130 | unsigned long gp_seq_req, const char *s) |
0446be48 | 1131 | { |
88d1bead | 1132 | trace_rcu_future_grace_period(rcu_state.name, rnp->gp_seq, gp_seq_req, |
abd13fdd | 1133 | rnp->level, rnp->grplo, rnp->grphi, s); |
0446be48 PM |
1134 | } |
1135 | ||
1136 | /* | |
b73de91d | 1137 | * rcu_start_this_gp - Request the start of a particular grace period |
df2bf8f7 | 1138 | * @rnp_start: The leaf node of the CPU from which to start. |
b73de91d JF |
1139 | * @rdp: The rcu_data corresponding to the CPU from which to start. |
1140 | * @gp_seq_req: The gp_seq of the grace period to start. | |
1141 | * | |
41e80595 | 1142 | * Start the specified grace period, as needed to handle newly arrived |
0446be48 | 1143 | * callbacks. The required future grace periods are recorded in each |
7a1d0f23 | 1144 | * rcu_node structure's ->gp_seq_needed field. Returns true if there |
48a7639c | 1145 | * is reason to awaken the grace-period kthread. |
0446be48 | 1146 | * |
d5cd9685 PM |
1147 | * The caller must hold the specified rcu_node structure's ->lock, which |
1148 | * is why the caller is responsible for waking the grace-period kthread. | |
b73de91d JF |
1149 | * |
1150 | * Returns true if the GP thread needs to be awakened else false. | |
0446be48 | 1151 | */ |
df2bf8f7 | 1152 | static bool rcu_start_this_gp(struct rcu_node *rnp_start, struct rcu_data *rdp, |
b73de91d | 1153 | unsigned long gp_seq_req) |
0446be48 | 1154 | { |
48a7639c | 1155 | bool ret = false; |
df2bf8f7 | 1156 | struct rcu_node *rnp; |
0446be48 PM |
1157 | |
1158 | /* | |
360e0da6 PM |
1159 | * Use funnel locking to either acquire the root rcu_node |
1160 | * structure's lock or bail out if the need for this grace period | |
df2bf8f7 JFG |
1161 | * has already been recorded -- or if that grace period has in |
1162 | * fact already started. If there is already a grace period in | |
1163 | * progress in a non-leaf node, no recording is needed because the | |
1164 | * end of the grace period will scan the leaf rcu_node structures. | |
1165 | * Note that rnp_start->lock must not be released. | |
0446be48 | 1166 | */ |
df2bf8f7 JFG |
1167 | raw_lockdep_assert_held_rcu_node(rnp_start); |
1168 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, TPS("Startleaf")); | |
1169 | for (rnp = rnp_start; 1; rnp = rnp->parent) { | |
1170 | if (rnp != rnp_start) | |
1171 | raw_spin_lock_rcu_node(rnp); | |
1172 | if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req) || | |
1173 | rcu_seq_started(&rnp->gp_seq, gp_seq_req) || | |
1174 | (rnp != rnp_start && | |
1175 | rcu_seq_state(rcu_seq_current(&rnp->gp_seq)))) { | |
1176 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, | |
b73de91d | 1177 | TPS("Prestarted")); |
360e0da6 PM |
1178 | goto unlock_out; |
1179 | } | |
df2bf8f7 | 1180 | rnp->gp_seq_needed = gp_seq_req; |
226ca5e7 | 1181 | if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) { |
a2165e41 | 1182 | /* |
226ca5e7 JFG |
1183 | * We just marked the leaf or internal node, and a |
1184 | * grace period is in progress, which means that | |
1185 | * rcu_gp_cleanup() will see the marking. Bail to | |
1186 | * reduce contention. | |
a2165e41 | 1187 | */ |
df2bf8f7 | 1188 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, |
b73de91d | 1189 | TPS("Startedleaf")); |
a2165e41 PM |
1190 | goto unlock_out; |
1191 | } | |
df2bf8f7 JFG |
1192 | if (rnp != rnp_start && rnp->parent != NULL) |
1193 | raw_spin_unlock_rcu_node(rnp); | |
1194 | if (!rnp->parent) | |
360e0da6 | 1195 | break; /* At root, and perhaps also leaf. */ |
0446be48 PM |
1196 | } |
1197 | ||
360e0da6 | 1198 | /* If GP already in progress, just leave, otherwise start one. */ |
de8e8730 | 1199 | if (rcu_gp_in_progress()) { |
df2bf8f7 | 1200 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedleafroot")); |
0446be48 PM |
1201 | goto unlock_out; |
1202 | } | |
df2bf8f7 | 1203 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedroot")); |
9cbc5b97 PM |
1204 | WRITE_ONCE(rcu_state.gp_flags, rcu_state.gp_flags | RCU_GP_FLAG_INIT); |
1205 | rcu_state.gp_req_activity = jiffies; | |
1206 | if (!rcu_state.gp_kthread) { | |
df2bf8f7 | 1207 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("NoGPkthread")); |
360e0da6 | 1208 | goto unlock_out; |
0446be48 | 1209 | } |
9cbc5b97 | 1210 | trace_rcu_grace_period(rcu_state.name, READ_ONCE(rcu_state.gp_seq), TPS("newreq")); |
360e0da6 | 1211 | ret = true; /* Caller must wake GP kthread. */ |
0446be48 | 1212 | unlock_out: |
ab5e869c | 1213 | /* Push furthest requested GP to leaf node and rcu_data structure. */ |
df2bf8f7 JFG |
1214 | if (ULONG_CMP_LT(gp_seq_req, rnp->gp_seq_needed)) { |
1215 | rnp_start->gp_seq_needed = rnp->gp_seq_needed; | |
1216 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
ab5e869c | 1217 | } |
df2bf8f7 JFG |
1218 | if (rnp != rnp_start) |
1219 | raw_spin_unlock_rcu_node(rnp); | |
48a7639c | 1220 | return ret; |
0446be48 PM |
1221 | } |
1222 | ||
1223 | /* | |
1224 | * Clean up any old requests for the just-ended grace period. Also return | |
d1e4f01d | 1225 | * whether any additional grace periods have been requested. |
0446be48 | 1226 | */ |
3481f2ea | 1227 | static bool rcu_future_gp_cleanup(struct rcu_node *rnp) |
0446be48 | 1228 | { |
fb31340f | 1229 | bool needmore; |
da1df50d | 1230 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
0446be48 | 1231 | |
7a1d0f23 PM |
1232 | needmore = ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed); |
1233 | if (!needmore) | |
1234 | rnp->gp_seq_needed = rnp->gp_seq; /* Avoid counter wrap. */ | |
b73de91d | 1235 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq, |
41e80595 | 1236 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); |
0446be48 PM |
1237 | return needmore; |
1238 | } | |
1239 | ||
48a7639c | 1240 | /* |
1d1f898d ZJ |
1241 | * Awaken the grace-period kthread. Don't do a self-awaken (unless in |
1242 | * an interrupt or softirq handler), and don't bother awakening when there | |
1243 | * is nothing for the grace-period kthread to do (as in several CPUs raced | |
1244 | * to awaken, and we lost), and finally don't try to awaken a kthread that | |
1245 | * has not yet been created. If all those checks are passed, track some | |
1246 | * debug information and awaken. | |
1247 | * | |
1248 | * So why do the self-wakeup when in an interrupt or softirq handler | |
1249 | * in the grace-period kthread's context? Because the kthread might have | |
1250 | * been interrupted just as it was going to sleep, and just after the final | |
1251 | * pre-sleep check of the awaken condition. In this case, a wakeup really | |
1252 | * is required, and is therefore supplied. | |
48a7639c | 1253 | */ |
532c00c9 | 1254 | static void rcu_gp_kthread_wake(void) |
48a7639c | 1255 | { |
1d1f898d | 1256 | if ((current == rcu_state.gp_kthread && |
0f58d2ac | 1257 | !in_irq() && !in_serving_softirq()) || |
532c00c9 PM |
1258 | !READ_ONCE(rcu_state.gp_flags) || |
1259 | !rcu_state.gp_kthread) | |
48a7639c | 1260 | return; |
fd897573 PM |
1261 | WRITE_ONCE(rcu_state.gp_wake_time, jiffies); |
1262 | WRITE_ONCE(rcu_state.gp_wake_seq, READ_ONCE(rcu_state.gp_seq)); | |
532c00c9 | 1263 | swake_up_one(&rcu_state.gp_wq); |
48a7639c PM |
1264 | } |
1265 | ||
dc35c893 | 1266 | /* |
29365e56 PM |
1267 | * If there is room, assign a ->gp_seq number to any callbacks on this |
1268 | * CPU that have not already been assigned. Also accelerate any callbacks | |
1269 | * that were previously assigned a ->gp_seq number that has since proven | |
1270 | * to be too conservative, which can happen if callbacks get assigned a | |
1271 | * ->gp_seq number while RCU is idle, but with reference to a non-root | |
1272 | * rcu_node structure. This function is idempotent, so it does not hurt | |
1273 | * to call it repeatedly. Returns an flag saying that we should awaken | |
1274 | * the RCU grace-period kthread. | |
dc35c893 PM |
1275 | * |
1276 | * The caller must hold rnp->lock with interrupts disabled. | |
1277 | */ | |
02f50142 | 1278 | static bool rcu_accelerate_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1279 | { |
b73de91d | 1280 | unsigned long gp_seq_req; |
15fecf89 | 1281 | bool ret = false; |
dc35c893 | 1282 | |
d1b222c6 | 1283 | rcu_lockdep_assert_cblist_protected(rdp); |
a32e01ee | 1284 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1285 | |
15fecf89 PM |
1286 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1287 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1288 | return false; |
dc35c893 PM |
1289 | |
1290 | /* | |
15fecf89 PM |
1291 | * Callbacks are often registered with incomplete grace-period |
1292 | * information. Something about the fact that getting exact | |
1293 | * information requires acquiring a global lock... RCU therefore | |
1294 | * makes a conservative estimate of the grace period number at which | |
1295 | * a given callback will become ready to invoke. The following | |
1296 | * code checks this estimate and improves it when possible, thus | |
1297 | * accelerating callback invocation to an earlier grace-period | |
1298 | * number. | |
dc35c893 | 1299 | */ |
9cbc5b97 | 1300 | gp_seq_req = rcu_seq_snap(&rcu_state.gp_seq); |
b73de91d JF |
1301 | if (rcu_segcblist_accelerate(&rdp->cblist, gp_seq_req)) |
1302 | ret = rcu_start_this_gp(rnp, rdp, gp_seq_req); | |
6d4b418c PM |
1303 | |
1304 | /* Trace depending on how much we were able to accelerate. */ | |
15fecf89 | 1305 | if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL)) |
9cbc5b97 | 1306 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccWaitCB")); |
6d4b418c | 1307 | else |
9cbc5b97 | 1308 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccReadyCB")); |
48a7639c | 1309 | return ret; |
dc35c893 PM |
1310 | } |
1311 | ||
e44e73ca PM |
1312 | /* |
1313 | * Similar to rcu_accelerate_cbs(), but does not require that the leaf | |
1314 | * rcu_node structure's ->lock be held. It consults the cached value | |
1315 | * of ->gp_seq_needed in the rcu_data structure, and if that indicates | |
1316 | * that a new grace-period request be made, invokes rcu_accelerate_cbs() | |
1317 | * while holding the leaf rcu_node structure's ->lock. | |
1318 | */ | |
c6e09b97 | 1319 | static void rcu_accelerate_cbs_unlocked(struct rcu_node *rnp, |
e44e73ca PM |
1320 | struct rcu_data *rdp) |
1321 | { | |
1322 | unsigned long c; | |
1323 | bool needwake; | |
1324 | ||
d1b222c6 | 1325 | rcu_lockdep_assert_cblist_protected(rdp); |
c6e09b97 | 1326 | c = rcu_seq_snap(&rcu_state.gp_seq); |
e44e73ca PM |
1327 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { |
1328 | /* Old request still live, so mark recent callbacks. */ | |
1329 | (void)rcu_segcblist_accelerate(&rdp->cblist, c); | |
1330 | return; | |
1331 | } | |
1332 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
02f50142 | 1333 | needwake = rcu_accelerate_cbs(rnp, rdp); |
e44e73ca PM |
1334 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
1335 | if (needwake) | |
532c00c9 | 1336 | rcu_gp_kthread_wake(); |
e44e73ca PM |
1337 | } |
1338 | ||
dc35c893 PM |
1339 | /* |
1340 | * Move any callbacks whose grace period has completed to the | |
1341 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
29365e56 | 1342 | * assign ->gp_seq numbers to any callbacks in the RCU_NEXT_TAIL |
dc35c893 PM |
1343 | * sublist. This function is idempotent, so it does not hurt to |
1344 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1345 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1346 | * |
1347 | * The caller must hold rnp->lock with interrupts disabled. | |
1348 | */ | |
834f56bf | 1349 | static bool rcu_advance_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1350 | { |
d1b222c6 | 1351 | rcu_lockdep_assert_cblist_protected(rdp); |
a32e01ee | 1352 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1353 | |
15fecf89 PM |
1354 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1355 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1356 | return false; |
dc35c893 PM |
1357 | |
1358 | /* | |
29365e56 | 1359 | * Find all callbacks whose ->gp_seq numbers indicate that they |
dc35c893 PM |
1360 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. |
1361 | */ | |
29365e56 | 1362 | rcu_segcblist_advance(&rdp->cblist, rnp->gp_seq); |
dc35c893 PM |
1363 | |
1364 | /* Classify any remaining callbacks. */ | |
02f50142 | 1365 | return rcu_accelerate_cbs(rnp, rdp); |
dc35c893 PM |
1366 | } |
1367 | ||
7f36ef82 PM |
1368 | /* |
1369 | * Move and classify callbacks, but only if doing so won't require | |
1370 | * that the RCU grace-period kthread be awakened. | |
1371 | */ | |
1372 | static void __maybe_unused rcu_advance_cbs_nowake(struct rcu_node *rnp, | |
1373 | struct rcu_data *rdp) | |
1374 | { | |
d1b222c6 | 1375 | rcu_lockdep_assert_cblist_protected(rdp); |
6608c3a0 PM |
1376 | if (!rcu_seq_state(rcu_seq_current(&rnp->gp_seq)) || |
1377 | !raw_spin_trylock_rcu_node(rnp)) | |
7f36ef82 PM |
1378 | return; |
1379 | WARN_ON_ONCE(rcu_advance_cbs(rnp, rdp)); | |
6608c3a0 | 1380 | raw_spin_unlock_rcu_node(rnp); |
7f36ef82 PM |
1381 | } |
1382 | ||
d09b62df | 1383 | /* |
ba9fbe95 PM |
1384 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1385 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1386 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1387 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1388 | */ |
c7e48f7b | 1389 | static bool __note_gp_changes(struct rcu_node *rnp, struct rcu_data *rdp) |
d09b62df | 1390 | { |
5d6742b3 | 1391 | bool ret = false; |
3563a438 | 1392 | bool need_gp; |
5d6742b3 PM |
1393 | const bool offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) && |
1394 | rcu_segcblist_is_offloaded(&rdp->cblist); | |
48a7639c | 1395 | |
a32e01ee | 1396 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1397 | |
67e14c1e PM |
1398 | if (rdp->gp_seq == rnp->gp_seq) |
1399 | return false; /* Nothing to do. */ | |
d09b62df | 1400 | |
67e14c1e PM |
1401 | /* Handle the ends of any preceding grace periods first. */ |
1402 | if (rcu_seq_completed_gp(rdp->gp_seq, rnp->gp_seq) || | |
1403 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
5d6742b3 PM |
1404 | if (!offloaded) |
1405 | ret = rcu_advance_cbs(rnp, rdp); /* Advance CBs. */ | |
9cbc5b97 | 1406 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuend")); |
67e14c1e | 1407 | } else { |
5d6742b3 PM |
1408 | if (!offloaded) |
1409 | ret = rcu_accelerate_cbs(rnp, rdp); /* Recent CBs. */ | |
d09b62df | 1410 | } |
398ebe60 | 1411 | |
67e14c1e PM |
1412 | /* Now handle the beginnings of any new-to-this-CPU grace periods. */ |
1413 | if (rcu_seq_new_gp(rdp->gp_seq, rnp->gp_seq) || | |
1414 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
6eaef633 PM |
1415 | /* |
1416 | * If the current grace period is waiting for this CPU, | |
1417 | * set up to detect a quiescent state, otherwise don't | |
1418 | * go looking for one. | |
1419 | */ | |
9cbc5b97 | 1420 | trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, TPS("cpustart")); |
3563a438 PM |
1421 | need_gp = !!(rnp->qsmask & rdp->grpmask); |
1422 | rdp->cpu_no_qs.b.norm = need_gp; | |
3563a438 | 1423 | rdp->core_needs_qs = need_gp; |
6eaef633 PM |
1424 | zero_cpu_stall_ticks(rdp); |
1425 | } | |
67e14c1e | 1426 | rdp->gp_seq = rnp->gp_seq; /* Remember new grace-period state. */ |
13dc7d0c | 1427 | if (ULONG_CMP_LT(rdp->gp_seq_needed, rnp->gp_seq_needed) || rdp->gpwrap) |
3d18469a PM |
1428 | rdp->gp_seq_needed = rnp->gp_seq_needed; |
1429 | WRITE_ONCE(rdp->gpwrap, false); | |
1430 | rcu_gpnum_ovf(rnp, rdp); | |
48a7639c | 1431 | return ret; |
6eaef633 PM |
1432 | } |
1433 | ||
15cabdff | 1434 | static void note_gp_changes(struct rcu_data *rdp) |
6eaef633 PM |
1435 | { |
1436 | unsigned long flags; | |
48a7639c | 1437 | bool needwake; |
6eaef633 PM |
1438 | struct rcu_node *rnp; |
1439 | ||
1440 | local_irq_save(flags); | |
1441 | rnp = rdp->mynode; | |
67e14c1e | 1442 | if ((rdp->gp_seq == rcu_seq_current(&rnp->gp_seq) && |
7d0ae808 | 1443 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ |
2a67e741 | 1444 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
1445 | local_irq_restore(flags); |
1446 | return; | |
1447 | } | |
c7e48f7b | 1448 | needwake = __note_gp_changes(rnp, rdp); |
67c583a7 | 1449 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c | 1450 | if (needwake) |
532c00c9 | 1451 | rcu_gp_kthread_wake(); |
6eaef633 PM |
1452 | } |
1453 | ||
22212332 | 1454 | static void rcu_gp_slow(int delay) |
0f41c0dd PM |
1455 | { |
1456 | if (delay > 0 && | |
22212332 | 1457 | !(rcu_seq_ctr(rcu_state.gp_seq) % |
dee4f422 | 1458 | (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) |
0f41c0dd PM |
1459 | schedule_timeout_uninterruptible(delay); |
1460 | } | |
1461 | ||
b3dbec76 | 1462 | /* |
45fed3e7 | 1463 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 1464 | */ |
0854a05c | 1465 | static bool rcu_gp_init(void) |
b3dbec76 | 1466 | { |
ec2c2976 | 1467 | unsigned long flags; |
0aa04b05 | 1468 | unsigned long oldmask; |
ec2c2976 | 1469 | unsigned long mask; |
b3dbec76 | 1470 | struct rcu_data *rdp; |
336a4f6c | 1471 | struct rcu_node *rnp = rcu_get_root(); |
b3dbec76 | 1472 | |
9cbc5b97 | 1473 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 1474 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 | 1475 | if (!READ_ONCE(rcu_state.gp_flags)) { |
f7be8209 | 1476 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 1477 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1478 | return false; |
f7be8209 | 1479 | } |
9cbc5b97 | 1480 | WRITE_ONCE(rcu_state.gp_flags, 0); /* Clear all flags: New GP. */ |
b3dbec76 | 1481 | |
de8e8730 | 1482 | if (WARN_ON_ONCE(rcu_gp_in_progress())) { |
f7be8209 PM |
1483 | /* |
1484 | * Grace period already in progress, don't start another. | |
1485 | * Not supposed to be able to happen. | |
1486 | */ | |
67c583a7 | 1487 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1488 | return false; |
7fdefc10 PM |
1489 | } |
1490 | ||
7fdefc10 | 1491 | /* Advance to a new grace period and initialize state. */ |
ad3832e9 | 1492 | record_gp_stall_check_time(); |
ff3bb6f4 | 1493 | /* Record GP times before starting GP, hence rcu_seq_start(). */ |
9cbc5b97 PM |
1494 | rcu_seq_start(&rcu_state.gp_seq); |
1495 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("start")); | |
67c583a7 | 1496 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 1497 | |
0aa04b05 PM |
1498 | /* |
1499 | * Apply per-leaf buffered online and offline operations to the | |
1500 | * rcu_node tree. Note that this new grace period need not wait | |
1501 | * for subsequent online CPUs, and that quiescent-state forcing | |
1502 | * will handle subsequent offline CPUs. | |
1503 | */ | |
9cbc5b97 | 1504 | rcu_state.gp_state = RCU_GP_ONOFF; |
aedf4ba9 | 1505 | rcu_for_each_leaf_node(rnp) { |
894d45bb | 1506 | raw_spin_lock(&rcu_state.ofl_lock); |
2a67e741 | 1507 | raw_spin_lock_irq_rcu_node(rnp); |
0aa04b05 PM |
1508 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
1509 | !rnp->wait_blkd_tasks) { | |
1510 | /* Nothing to do on this leaf rcu_node structure. */ | |
67c583a7 | 1511 | raw_spin_unlock_irq_rcu_node(rnp); |
894d45bb | 1512 | raw_spin_unlock(&rcu_state.ofl_lock); |
0aa04b05 PM |
1513 | continue; |
1514 | } | |
1515 | ||
1516 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
1517 | oldmask = rnp->qsmaskinit; | |
1518 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
1519 | ||
1520 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
1521 | if (!oldmask != !rnp->qsmaskinit) { | |
962aff03 PM |
1522 | if (!oldmask) { /* First online CPU for rcu_node. */ |
1523 | if (!rnp->wait_blkd_tasks) /* Ever offline? */ | |
1524 | rcu_init_new_rnp(rnp); | |
1525 | } else if (rcu_preempt_has_tasks(rnp)) { | |
1526 | rnp->wait_blkd_tasks = true; /* blocked tasks */ | |
1527 | } else { /* Last offline CPU and can propagate. */ | |
0aa04b05 | 1528 | rcu_cleanup_dead_rnp(rnp); |
962aff03 | 1529 | } |
0aa04b05 PM |
1530 | } |
1531 | ||
1532 | /* | |
1533 | * If all waited-on tasks from prior grace period are | |
1534 | * done, and if all this rcu_node structure's CPUs are | |
1535 | * still offline, propagate up the rcu_node tree and | |
1536 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
1537 | * rcu_node structure's CPUs has since come back online, | |
962aff03 | 1538 | * simply clear ->wait_blkd_tasks. |
0aa04b05 PM |
1539 | */ |
1540 | if (rnp->wait_blkd_tasks && | |
962aff03 | 1541 | (!rcu_preempt_has_tasks(rnp) || rnp->qsmaskinit)) { |
0aa04b05 | 1542 | rnp->wait_blkd_tasks = false; |
962aff03 PM |
1543 | if (!rnp->qsmaskinit) |
1544 | rcu_cleanup_dead_rnp(rnp); | |
0aa04b05 PM |
1545 | } |
1546 | ||
67c583a7 | 1547 | raw_spin_unlock_irq_rcu_node(rnp); |
894d45bb | 1548 | raw_spin_unlock(&rcu_state.ofl_lock); |
0aa04b05 | 1549 | } |
22212332 | 1550 | rcu_gp_slow(gp_preinit_delay); /* Races with CPU hotplug. */ |
7fdefc10 PM |
1551 | |
1552 | /* | |
1553 | * Set the quiescent-state-needed bits in all the rcu_node | |
9cbc5b97 PM |
1554 | * structures for all currently online CPUs in breadth-first |
1555 | * order, starting from the root rcu_node structure, relying on the | |
1556 | * layout of the tree within the rcu_state.node[] array. Note that | |
1557 | * other CPUs will access only the leaves of the hierarchy, thus | |
1558 | * seeing that no grace period is in progress, at least until the | |
1559 | * corresponding leaf node has been initialized. | |
7fdefc10 PM |
1560 | * |
1561 | * The grace period cannot complete until the initialization | |
1562 | * process finishes, because this kthread handles both. | |
1563 | */ | |
9cbc5b97 | 1564 | rcu_state.gp_state = RCU_GP_INIT; |
aedf4ba9 | 1565 | rcu_for_each_node_breadth_first(rnp) { |
22212332 | 1566 | rcu_gp_slow(gp_init_delay); |
ec2c2976 | 1567 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
da1df50d | 1568 | rdp = this_cpu_ptr(&rcu_data); |
81ab59a3 | 1569 | rcu_preempt_check_blocked_tasks(rnp); |
7fdefc10 | 1570 | rnp->qsmask = rnp->qsmaskinit; |
9cbc5b97 | 1571 | WRITE_ONCE(rnp->gp_seq, rcu_state.gp_seq); |
7fdefc10 | 1572 | if (rnp == rdp->mynode) |
c7e48f7b | 1573 | (void)__note_gp_changes(rnp, rdp); |
7fdefc10 | 1574 | rcu_preempt_boost_start_gp(rnp); |
9cbc5b97 | 1575 | trace_rcu_grace_period_init(rcu_state.name, rnp->gp_seq, |
7fdefc10 PM |
1576 | rnp->level, rnp->grplo, |
1577 | rnp->grphi, rnp->qsmask); | |
ec2c2976 PM |
1578 | /* Quiescent states for tasks on any now-offline CPUs. */ |
1579 | mask = rnp->qsmask & ~rnp->qsmaskinitnext; | |
f2e2df59 | 1580 | rnp->rcu_gp_init_mask = mask; |
ec2c2976 | 1581 | if ((mask || rnp->wait_blkd_tasks) && rcu_is_leaf_node(rnp)) |
b50912d0 | 1582 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
ec2c2976 PM |
1583 | else |
1584 | raw_spin_unlock_irq_rcu_node(rnp); | |
cee43939 | 1585 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1586 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
7fdefc10 | 1587 | } |
b3dbec76 | 1588 | |
45fed3e7 | 1589 | return true; |
7fdefc10 | 1590 | } |
b3dbec76 | 1591 | |
b9a425cf | 1592 | /* |
b3dae109 | 1593 | * Helper function for swait_event_idle_exclusive() wakeup at force-quiescent-state |
d5374226 | 1594 | * time. |
b9a425cf | 1595 | */ |
0854a05c | 1596 | static bool rcu_gp_fqs_check_wake(int *gfp) |
b9a425cf | 1597 | { |
336a4f6c | 1598 | struct rcu_node *rnp = rcu_get_root(); |
b9a425cf PM |
1599 | |
1600 | /* Someone like call_rcu() requested a force-quiescent-state scan. */ | |
0854a05c | 1601 | *gfp = READ_ONCE(rcu_state.gp_flags); |
b9a425cf PM |
1602 | if (*gfp & RCU_GP_FLAG_FQS) |
1603 | return true; | |
1604 | ||
1605 | /* The current grace period has completed. */ | |
1606 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) | |
1607 | return true; | |
1608 | ||
1609 | return false; | |
1610 | } | |
1611 | ||
4cdfc175 PM |
1612 | /* |
1613 | * Do one round of quiescent-state forcing. | |
1614 | */ | |
0854a05c | 1615 | static void rcu_gp_fqs(bool first_time) |
4cdfc175 | 1616 | { |
336a4f6c | 1617 | struct rcu_node *rnp = rcu_get_root(); |
4cdfc175 | 1618 | |
9cbc5b97 PM |
1619 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
1620 | rcu_state.n_force_qs++; | |
77f81fe0 | 1621 | if (first_time) { |
4cdfc175 | 1622 | /* Collect dyntick-idle snapshots. */ |
e9ecb780 | 1623 | force_qs_rnp(dyntick_save_progress_counter); |
4cdfc175 PM |
1624 | } else { |
1625 | /* Handle dyntick-idle and offline CPUs. */ | |
e9ecb780 | 1626 | force_qs_rnp(rcu_implicit_dynticks_qs); |
4cdfc175 PM |
1627 | } |
1628 | /* Clear flag to prevent immediate re-entry. */ | |
9cbc5b97 | 1629 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 1630 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 PM |
1631 | WRITE_ONCE(rcu_state.gp_flags, |
1632 | READ_ONCE(rcu_state.gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 1633 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 1634 | } |
4cdfc175 PM |
1635 | } |
1636 | ||
c3854a05 PM |
1637 | /* |
1638 | * Loop doing repeated quiescent-state forcing until the grace period ends. | |
1639 | */ | |
1640 | static void rcu_gp_fqs_loop(void) | |
1641 | { | |
1642 | bool first_gp_fqs; | |
1643 | int gf; | |
1644 | unsigned long j; | |
1645 | int ret; | |
1646 | struct rcu_node *rnp = rcu_get_root(); | |
1647 | ||
1648 | first_gp_fqs = true; | |
c06aed0e | 1649 | j = READ_ONCE(jiffies_till_first_fqs); |
c3854a05 PM |
1650 | ret = 0; |
1651 | for (;;) { | |
1652 | if (!ret) { | |
1653 | rcu_state.jiffies_force_qs = jiffies + j; | |
1654 | WRITE_ONCE(rcu_state.jiffies_kick_kthreads, | |
9cf422a8 | 1655 | jiffies + (j ? 3 * j : 2)); |
c3854a05 PM |
1656 | } |
1657 | trace_rcu_grace_period(rcu_state.name, | |
1658 | READ_ONCE(rcu_state.gp_seq), | |
1659 | TPS("fqswait")); | |
1660 | rcu_state.gp_state = RCU_GP_WAIT_FQS; | |
1661 | ret = swait_event_idle_timeout_exclusive( | |
1662 | rcu_state.gp_wq, rcu_gp_fqs_check_wake(&gf), j); | |
1663 | rcu_state.gp_state = RCU_GP_DOING_FQS; | |
1664 | /* Locking provides needed memory barriers. */ | |
1665 | /* If grace period done, leave loop. */ | |
1666 | if (!READ_ONCE(rnp->qsmask) && | |
1667 | !rcu_preempt_blocked_readers_cgp(rnp)) | |
1668 | break; | |
1669 | /* If time for quiescent-state forcing, do it. */ | |
1670 | if (ULONG_CMP_GE(jiffies, rcu_state.jiffies_force_qs) || | |
1671 | (gf & RCU_GP_FLAG_FQS)) { | |
1672 | trace_rcu_grace_period(rcu_state.name, | |
1673 | READ_ONCE(rcu_state.gp_seq), | |
1674 | TPS("fqsstart")); | |
1675 | rcu_gp_fqs(first_gp_fqs); | |
1676 | first_gp_fqs = false; | |
1677 | trace_rcu_grace_period(rcu_state.name, | |
1678 | READ_ONCE(rcu_state.gp_seq), | |
1679 | TPS("fqsend")); | |
1680 | cond_resched_tasks_rcu_qs(); | |
1681 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1682 | ret = 0; /* Force full wait till next FQS. */ | |
c06aed0e | 1683 | j = READ_ONCE(jiffies_till_next_fqs); |
c3854a05 PM |
1684 | } else { |
1685 | /* Deal with stray signal. */ | |
1686 | cond_resched_tasks_rcu_qs(); | |
1687 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1688 | WARN_ON(signal_pending(current)); | |
1689 | trace_rcu_grace_period(rcu_state.name, | |
1690 | READ_ONCE(rcu_state.gp_seq), | |
1691 | TPS("fqswaitsig")); | |
1692 | ret = 1; /* Keep old FQS timing. */ | |
1693 | j = jiffies; | |
1694 | if (time_after(jiffies, rcu_state.jiffies_force_qs)) | |
1695 | j = 1; | |
1696 | else | |
1697 | j = rcu_state.jiffies_force_qs - j; | |
1698 | } | |
1699 | } | |
1700 | } | |
1701 | ||
7fdefc10 PM |
1702 | /* |
1703 | * Clean up after the old grace period. | |
1704 | */ | |
0854a05c | 1705 | static void rcu_gp_cleanup(void) |
7fdefc10 PM |
1706 | { |
1707 | unsigned long gp_duration; | |
48a7639c | 1708 | bool needgp = false; |
de30ad51 | 1709 | unsigned long new_gp_seq; |
5d6742b3 | 1710 | bool offloaded; |
7fdefc10 | 1711 | struct rcu_data *rdp; |
336a4f6c | 1712 | struct rcu_node *rnp = rcu_get_root(); |
abedf8e2 | 1713 | struct swait_queue_head *sq; |
b3dbec76 | 1714 | |
9cbc5b97 | 1715 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 1716 | raw_spin_lock_irq_rcu_node(rnp); |
c51d7b5e PM |
1717 | rcu_state.gp_end = jiffies; |
1718 | gp_duration = rcu_state.gp_end - rcu_state.gp_start; | |
9cbc5b97 PM |
1719 | if (gp_duration > rcu_state.gp_max) |
1720 | rcu_state.gp_max = gp_duration; | |
b3dbec76 | 1721 | |
7fdefc10 PM |
1722 | /* |
1723 | * We know the grace period is complete, but to everyone else | |
1724 | * it appears to still be ongoing. But it is also the case | |
1725 | * that to everyone else it looks like there is nothing that | |
1726 | * they can do to advance the grace period. It is therefore | |
1727 | * safe for us to drop the lock in order to mark the grace | |
1728 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 1729 | */ |
67c583a7 | 1730 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 1731 | |
5d4b8659 | 1732 | /* |
ff3bb6f4 PM |
1733 | * Propagate new ->gp_seq value to rcu_node structures so that |
1734 | * other CPUs don't have to wait until the start of the next grace | |
1735 | * period to process their callbacks. This also avoids some nasty | |
1736 | * RCU grace-period initialization races by forcing the end of | |
1737 | * the current grace period to be completely recorded in all of | |
1738 | * the rcu_node structures before the beginning of the next grace | |
1739 | * period is recorded in any of the rcu_node structures. | |
5d4b8659 | 1740 | */ |
9cbc5b97 | 1741 | new_gp_seq = rcu_state.gp_seq; |
de30ad51 | 1742 | rcu_seq_end(&new_gp_seq); |
aedf4ba9 | 1743 | rcu_for_each_node_breadth_first(rnp) { |
2a67e741 | 1744 | raw_spin_lock_irq_rcu_node(rnp); |
4bc8d555 | 1745 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 1746 | dump_blkd_tasks(rnp, 10); |
5c60d25f | 1747 | WARN_ON_ONCE(rnp->qsmask); |
de30ad51 | 1748 | WRITE_ONCE(rnp->gp_seq, new_gp_seq); |
da1df50d | 1749 | rdp = this_cpu_ptr(&rcu_data); |
b11cc576 | 1750 | if (rnp == rdp->mynode) |
c7e48f7b | 1751 | needgp = __note_gp_changes(rnp, rdp) || needgp; |
78e4bc34 | 1752 | /* smp_mb() provided by prior unlock-lock pair. */ |
3481f2ea | 1753 | needgp = rcu_future_gp_cleanup(rnp) || needgp; |
065bb78c | 1754 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 1755 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 1756 | rcu_nocb_gp_cleanup(sq); |
cee43939 | 1757 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1758 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
22212332 | 1759 | rcu_gp_slow(gp_cleanup_delay); |
7fdefc10 | 1760 | } |
336a4f6c | 1761 | rnp = rcu_get_root(); |
9cbc5b97 | 1762 | raw_spin_lock_irq_rcu_node(rnp); /* GP before ->gp_seq update. */ |
7fdefc10 | 1763 | |
0a89e5a4 | 1764 | /* Declare grace period done, trace first to use old GP number. */ |
9cbc5b97 | 1765 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("end")); |
0a89e5a4 | 1766 | rcu_seq_end(&rcu_state.gp_seq); |
9cbc5b97 | 1767 | rcu_state.gp_state = RCU_GP_IDLE; |
fb31340f | 1768 | /* Check for GP requests since above loop. */ |
da1df50d | 1769 | rdp = this_cpu_ptr(&rcu_data); |
5b55072f | 1770 | if (!needgp && ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed)) { |
abd13fdd | 1771 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq_needed, |
41e80595 | 1772 | TPS("CleanupMore")); |
fb31340f PM |
1773 | needgp = true; |
1774 | } | |
48a7639c | 1775 | /* Advance CBs to reduce false positives below. */ |
5d6742b3 PM |
1776 | offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) && |
1777 | rcu_segcblist_is_offloaded(&rdp->cblist); | |
1778 | if ((offloaded || !rcu_accelerate_cbs(rnp, rdp)) && needgp) { | |
9cbc5b97 PM |
1779 | WRITE_ONCE(rcu_state.gp_flags, RCU_GP_FLAG_INIT); |
1780 | rcu_state.gp_req_activity = jiffies; | |
1781 | trace_rcu_grace_period(rcu_state.name, | |
1782 | READ_ONCE(rcu_state.gp_seq), | |
bb311ecc | 1783 | TPS("newreq")); |
18390aea | 1784 | } else { |
9cbc5b97 PM |
1785 | WRITE_ONCE(rcu_state.gp_flags, |
1786 | rcu_state.gp_flags & RCU_GP_FLAG_INIT); | |
bb311ecc | 1787 | } |
67c583a7 | 1788 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 PM |
1789 | } |
1790 | ||
1791 | /* | |
1792 | * Body of kthread that handles grace periods. | |
1793 | */ | |
0854a05c | 1794 | static int __noreturn rcu_gp_kthread(void *unused) |
7fdefc10 | 1795 | { |
5871968d | 1796 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
1797 | for (;;) { |
1798 | ||
1799 | /* Handle grace-period start. */ | |
1800 | for (;;) { | |
9cbc5b97 PM |
1801 | trace_rcu_grace_period(rcu_state.name, |
1802 | READ_ONCE(rcu_state.gp_seq), | |
63c4db78 | 1803 | TPS("reqwait")); |
9cbc5b97 PM |
1804 | rcu_state.gp_state = RCU_GP_WAIT_GPS; |
1805 | swait_event_idle_exclusive(rcu_state.gp_wq, | |
1806 | READ_ONCE(rcu_state.gp_flags) & | |
1807 | RCU_GP_FLAG_INIT); | |
1808 | rcu_state.gp_state = RCU_GP_DONE_GPS; | |
78e4bc34 | 1809 | /* Locking provides needed memory barrier. */ |
0854a05c | 1810 | if (rcu_gp_init()) |
7fdefc10 | 1811 | break; |
cee43939 | 1812 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1813 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
73a860cd | 1814 | WARN_ON(signal_pending(current)); |
9cbc5b97 PM |
1815 | trace_rcu_grace_period(rcu_state.name, |
1816 | READ_ONCE(rcu_state.gp_seq), | |
63c4db78 | 1817 | TPS("reqwaitsig")); |
7fdefc10 | 1818 | } |
cabc49c1 | 1819 | |
4cdfc175 | 1820 | /* Handle quiescent-state forcing. */ |
c3854a05 | 1821 | rcu_gp_fqs_loop(); |
4cdfc175 PM |
1822 | |
1823 | /* Handle grace-period end. */ | |
9cbc5b97 | 1824 | rcu_state.gp_state = RCU_GP_CLEANUP; |
0854a05c | 1825 | rcu_gp_cleanup(); |
9cbc5b97 | 1826 | rcu_state.gp_state = RCU_GP_CLEANED; |
b3dbec76 | 1827 | } |
b3dbec76 PM |
1828 | } |
1829 | ||
f41d911f | 1830 | /* |
49918a54 PM |
1831 | * Report a full set of quiescent states to the rcu_state data structure. |
1832 | * Invoke rcu_gp_kthread_wake() to awaken the grace-period kthread if | |
1833 | * another grace period is required. Whether we wake the grace-period | |
1834 | * kthread or it awakens itself for the next round of quiescent-state | |
1835 | * forcing, that kthread will clean up after the just-completed grace | |
1836 | * period. Note that the caller must hold rnp->lock, which is released | |
1837 | * before return. | |
f41d911f | 1838 | */ |
aff4e9ed | 1839 | static void rcu_report_qs_rsp(unsigned long flags) |
336a4f6c | 1840 | __releases(rcu_get_root()->lock) |
f41d911f | 1841 | { |
336a4f6c | 1842 | raw_lockdep_assert_held_rcu_node(rcu_get_root()); |
de8e8730 | 1843 | WARN_ON_ONCE(!rcu_gp_in_progress()); |
9cbc5b97 PM |
1844 | WRITE_ONCE(rcu_state.gp_flags, |
1845 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
336a4f6c | 1846 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(), flags); |
532c00c9 | 1847 | rcu_gp_kthread_wake(); |
f41d911f PM |
1848 | } |
1849 | ||
64db4cff | 1850 | /* |
d3f6bad3 PM |
1851 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
1852 | * Allows quiescent states for a group of CPUs to be reported at one go | |
1853 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
1854 | * must be represented by the same rcu_node structure (which need not be a |
1855 | * leaf rcu_node structure, though it often will be). The gps parameter | |
1856 | * is the grace-period snapshot, which means that the quiescent states | |
c9a24e2d | 1857 | * are valid only if rnp->gp_seq is equal to gps. That structure's lock |
654e9533 | 1858 | * must be held upon entry, and it is released before return. |
ec2c2976 PM |
1859 | * |
1860 | * As a special case, if mask is zero, the bit-already-cleared check is | |
1861 | * disabled. This allows propagating quiescent state due to resumed tasks | |
1862 | * during grace-period initialization. | |
64db4cff | 1863 | */ |
b50912d0 PM |
1864 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
1865 | unsigned long gps, unsigned long flags) | |
64db4cff PM |
1866 | __releases(rnp->lock) |
1867 | { | |
654e9533 | 1868 | unsigned long oldmask = 0; |
28ecd580 PM |
1869 | struct rcu_node *rnp_c; |
1870 | ||
a32e01ee | 1871 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1872 | |
64db4cff PM |
1873 | /* Walk up the rcu_node hierarchy. */ |
1874 | for (;;) { | |
ec2c2976 | 1875 | if ((!(rnp->qsmask & mask) && mask) || rnp->gp_seq != gps) { |
64db4cff | 1876 | |
654e9533 PM |
1877 | /* |
1878 | * Our bit has already been cleared, or the | |
1879 | * relevant grace period is already over, so done. | |
1880 | */ | |
67c583a7 | 1881 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1882 | return; |
1883 | } | |
654e9533 | 1884 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
5b4c11d5 | 1885 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp) && |
2dee9404 | 1886 | rcu_preempt_blocked_readers_cgp(rnp)); |
64db4cff | 1887 | rnp->qsmask &= ~mask; |
67a0edbf | 1888 | trace_rcu_quiescent_state_report(rcu_state.name, rnp->gp_seq, |
d4c08f2a PM |
1889 | mask, rnp->qsmask, rnp->level, |
1890 | rnp->grplo, rnp->grphi, | |
1891 | !!rnp->gp_tasks); | |
27f4d280 | 1892 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
1893 | |
1894 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 1895 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1896 | return; |
1897 | } | |
d43a5d32 | 1898 | rnp->completedqs = rnp->gp_seq; |
64db4cff PM |
1899 | mask = rnp->grpmask; |
1900 | if (rnp->parent == NULL) { | |
1901 | ||
1902 | /* No more levels. Exit loop holding root lock. */ | |
1903 | ||
1904 | break; | |
1905 | } | |
67c583a7 | 1906 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 1907 | rnp_c = rnp; |
64db4cff | 1908 | rnp = rnp->parent; |
2a67e741 | 1909 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
654e9533 | 1910 | oldmask = rnp_c->qsmask; |
64db4cff PM |
1911 | } |
1912 | ||
1913 | /* | |
1914 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 1915 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 1916 | * to clean up and start the next grace period if one is needed. |
64db4cff | 1917 | */ |
aff4e9ed | 1918 | rcu_report_qs_rsp(flags); /* releases rnp->lock. */ |
64db4cff PM |
1919 | } |
1920 | ||
cc99a310 PM |
1921 | /* |
1922 | * Record a quiescent state for all tasks that were previously queued | |
1923 | * on the specified rcu_node structure and that were blocking the current | |
49918a54 | 1924 | * RCU grace period. The caller must hold the corresponding rnp->lock with |
cc99a310 PM |
1925 | * irqs disabled, and this lock is released upon return, but irqs remain |
1926 | * disabled. | |
1927 | */ | |
17a8212b | 1928 | static void __maybe_unused |
139ad4da | 1929 | rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
cc99a310 PM |
1930 | __releases(rnp->lock) |
1931 | { | |
654e9533 | 1932 | unsigned long gps; |
cc99a310 PM |
1933 | unsigned long mask; |
1934 | struct rcu_node *rnp_p; | |
1935 | ||
a32e01ee | 1936 | raw_lockdep_assert_held_rcu_node(rnp); |
01b1d88b | 1937 | if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPTION)) || |
c74859d1 PM |
1938 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)) || |
1939 | rnp->qsmask != 0) { | |
67c583a7 | 1940 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
1941 | return; /* Still need more quiescent states! */ |
1942 | } | |
1943 | ||
77cfc7bf | 1944 | rnp->completedqs = rnp->gp_seq; |
cc99a310 PM |
1945 | rnp_p = rnp->parent; |
1946 | if (rnp_p == NULL) { | |
1947 | /* | |
a77da14c PM |
1948 | * Only one rcu_node structure in the tree, so don't |
1949 | * try to report up to its nonexistent parent! | |
cc99a310 | 1950 | */ |
aff4e9ed | 1951 | rcu_report_qs_rsp(flags); |
cc99a310 PM |
1952 | return; |
1953 | } | |
1954 | ||
c9a24e2d PM |
1955 | /* Report up the rest of the hierarchy, tracking current ->gp_seq. */ |
1956 | gps = rnp->gp_seq; | |
cc99a310 | 1957 | mask = rnp->grpmask; |
67c583a7 | 1958 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 1959 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
b50912d0 | 1960 | rcu_report_qs_rnp(mask, rnp_p, gps, flags); |
cc99a310 PM |
1961 | } |
1962 | ||
64db4cff | 1963 | /* |
d3f6bad3 | 1964 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 1965 | * structure. This must be called from the specified CPU. |
64db4cff PM |
1966 | */ |
1967 | static void | |
33085c46 | 1968 | rcu_report_qs_rdp(int cpu, struct rcu_data *rdp) |
64db4cff PM |
1969 | { |
1970 | unsigned long flags; | |
1971 | unsigned long mask; | |
5d6742b3 PM |
1972 | bool needwake = false; |
1973 | const bool offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) && | |
1974 | rcu_segcblist_is_offloaded(&rdp->cblist); | |
64db4cff PM |
1975 | struct rcu_node *rnp; |
1976 | ||
1977 | rnp = rdp->mynode; | |
2a67e741 | 1978 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
c9a24e2d PM |
1979 | if (rdp->cpu_no_qs.b.norm || rdp->gp_seq != rnp->gp_seq || |
1980 | rdp->gpwrap) { | |
64db4cff PM |
1981 | |
1982 | /* | |
e4cc1f22 PM |
1983 | * The grace period in which this quiescent state was |
1984 | * recorded has ended, so don't report it upwards. | |
1985 | * We will instead need a new quiescent state that lies | |
1986 | * within the current grace period. | |
64db4cff | 1987 | */ |
5b74c458 | 1988 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
67c583a7 | 1989 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1990 | return; |
1991 | } | |
1992 | mask = rdp->grpmask; | |
1993 | if ((rnp->qsmask & mask) == 0) { | |
67c583a7 | 1994 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1995 | } else { |
64db4cff PM |
1996 | /* |
1997 | * This GP can't end until cpu checks in, so all of our | |
1998 | * callbacks can be processed during the next GP. | |
1999 | */ | |
5d6742b3 PM |
2000 | if (!offloaded) |
2001 | needwake = rcu_accelerate_cbs(rnp, rdp); | |
64db4cff | 2002 | |
516e5ae0 | 2003 | rcu_disable_urgency_upon_qs(rdp); |
b50912d0 | 2004 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
654e9533 | 2005 | /* ^^^ Released rnp->lock */ |
48a7639c | 2006 | if (needwake) |
532c00c9 | 2007 | rcu_gp_kthread_wake(); |
64db4cff PM |
2008 | } |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * Check to see if there is a new grace period of which this CPU | |
2013 | * is not yet aware, and if so, set up local rcu_data state for it. | |
2014 | * Otherwise, see if this CPU has just passed through its first | |
2015 | * quiescent state for this grace period, and record that fact if so. | |
2016 | */ | |
2017 | static void | |
8087d3e3 | 2018 | rcu_check_quiescent_state(struct rcu_data *rdp) |
64db4cff | 2019 | { |
05eb552b | 2020 | /* Check for grace-period ends and beginnings. */ |
15cabdff | 2021 | note_gp_changes(rdp); |
64db4cff PM |
2022 | |
2023 | /* | |
2024 | * Does this CPU still need to do its part for current grace period? | |
2025 | * If no, return and let the other CPUs do their part as well. | |
2026 | */ | |
97c668b8 | 2027 | if (!rdp->core_needs_qs) |
64db4cff PM |
2028 | return; |
2029 | ||
2030 | /* | |
2031 | * Was there a quiescent state since the beginning of the grace | |
2032 | * period? If no, then exit and wait for the next call. | |
2033 | */ | |
3a19b46a | 2034 | if (rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2035 | return; |
2036 | ||
d3f6bad3 PM |
2037 | /* |
2038 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
2039 | * judge of that). | |
2040 | */ | |
33085c46 | 2041 | rcu_report_qs_rdp(rdp->cpu, rdp); |
64db4cff PM |
2042 | } |
2043 | ||
b1420f1c | 2044 | /* |
780cd590 PM |
2045 | * Near the end of the offline process. Trace the fact that this CPU |
2046 | * is going offline. | |
b1420f1c | 2047 | */ |
780cd590 | 2048 | int rcutree_dying_cpu(unsigned int cpu) |
b1420f1c | 2049 | { |
4f5fbd78 YS |
2050 | bool blkd; |
2051 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); | |
2052 | struct rcu_node *rnp = rdp->mynode; | |
b1420f1c | 2053 | |
ea46351c | 2054 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 2055 | return 0; |
ea46351c | 2056 | |
4f5fbd78 | 2057 | blkd = !!(rnp->qsmask & rdp->grpmask); |
780cd590 | 2058 | trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, |
477351f7 | 2059 | blkd ? TPS("cpuofl") : TPS("cpuofl-bgp")); |
780cd590 | 2060 | return 0; |
64db4cff PM |
2061 | } |
2062 | ||
8af3a5e7 PM |
2063 | /* |
2064 | * All CPUs for the specified rcu_node structure have gone offline, | |
2065 | * and all tasks that were preempted within an RCU read-side critical | |
2066 | * section while running on one of those CPUs have since exited their RCU | |
2067 | * read-side critical section. Some other CPU is reporting this fact with | |
2068 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2069 | * This function therefore goes up the tree of rcu_node structures, | |
2070 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2071 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
c50cbe53 | 2072 | * updated. |
8af3a5e7 PM |
2073 | * |
2074 | * This function does check that the specified rcu_node structure has | |
2075 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2076 | * prematurely. That said, invoking it after the fact will cost you | |
2077 | * a needless lock acquisition. So once it has done its work, don't | |
2078 | * invoke it again. | |
2079 | */ | |
2080 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2081 | { | |
2082 | long mask; | |
2083 | struct rcu_node *rnp = rnp_leaf; | |
2084 | ||
962aff03 | 2085 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
ea46351c | 2086 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
962aff03 PM |
2087 | WARN_ON_ONCE(rnp_leaf->qsmaskinit) || |
2088 | WARN_ON_ONCE(rcu_preempt_has_tasks(rnp_leaf))) | |
8af3a5e7 PM |
2089 | return; |
2090 | for (;;) { | |
2091 | mask = rnp->grpmask; | |
2092 | rnp = rnp->parent; | |
2093 | if (!rnp) | |
2094 | break; | |
2a67e741 | 2095 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2096 | rnp->qsmaskinit &= ~mask; |
962aff03 PM |
2097 | /* Between grace periods, so better already be zero! */ |
2098 | WARN_ON_ONCE(rnp->qsmask); | |
8af3a5e7 | 2099 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2100 | raw_spin_unlock_rcu_node(rnp); |
2101 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2102 | return; |
2103 | } | |
67c583a7 | 2104 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2105 | } |
2106 | } | |
2107 | ||
64db4cff | 2108 | /* |
e5601400 | 2109 | * The CPU has been completely removed, and some other CPU is reporting |
a58163d8 PM |
2110 | * this fact from process context. Do the remainder of the cleanup. |
2111 | * There can only be one CPU hotplug operation at a time, so no need for | |
2112 | * explicit locking. | |
64db4cff | 2113 | */ |
780cd590 | 2114 | int rcutree_dead_cpu(unsigned int cpu) |
64db4cff | 2115 | { |
da1df50d | 2116 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
b1420f1c | 2117 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2118 | |
ea46351c | 2119 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 2120 | return 0; |
ea46351c | 2121 | |
2036d94a | 2122 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2123 | rcu_boost_kthread_setaffinity(rnp, -1); |
780cd590 PM |
2124 | /* Do any needed no-CB deferred wakeups from this CPU. */ |
2125 | do_nocb_deferred_wakeup(per_cpu_ptr(&rcu_data, cpu)); | |
96926686 PM |
2126 | |
2127 | // Stop-machine done, so allow nohz_full to disable tick. | |
2128 | tick_dep_clear(TICK_DEP_BIT_RCU); | |
780cd590 | 2129 | return 0; |
64db4cff PM |
2130 | } |
2131 | ||
64db4cff PM |
2132 | /* |
2133 | * Invoke any RCU callbacks that have made it to the end of their grace | |
2134 | * period. Thottle as specified by rdp->blimit. | |
2135 | */ | |
5bb5d09c | 2136 | static void rcu_do_batch(struct rcu_data *rdp) |
64db4cff PM |
2137 | { |
2138 | unsigned long flags; | |
ec5ef87b PM |
2139 | const bool offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) && |
2140 | rcu_segcblist_is_offloaded(&rdp->cblist); | |
15fecf89 PM |
2141 | struct rcu_head *rhp; |
2142 | struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl); | |
2143 | long bl, count; | |
cfcdef5e | 2144 | long pending, tlimit = 0; |
64db4cff | 2145 | |
dc35c893 | 2146 | /* If no callbacks are ready, just return. */ |
15fecf89 | 2147 | if (!rcu_segcblist_ready_cbs(&rdp->cblist)) { |
3c779dfe | 2148 | trace_rcu_batch_start(rcu_state.name, |
15fecf89 PM |
2149 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2150 | rcu_segcblist_n_cbs(&rdp->cblist), 0); | |
3c779dfe | 2151 | trace_rcu_batch_end(rcu_state.name, 0, |
15fecf89 | 2152 | !rcu_segcblist_empty(&rdp->cblist), |
4968c300 PM |
2153 | need_resched(), is_idle_task(current), |
2154 | rcu_is_callbacks_kthread()); | |
64db4cff | 2155 | return; |
29c00b4a | 2156 | } |
64db4cff PM |
2157 | |
2158 | /* | |
2159 | * Extract the list of ready callbacks, disabling to prevent | |
15fecf89 PM |
2160 | * races with call_rcu() from interrupt handlers. Leave the |
2161 | * callback counts, as rcu_barrier() needs to be conservative. | |
64db4cff PM |
2162 | */ |
2163 | local_irq_save(flags); | |
5d6742b3 | 2164 | rcu_nocb_lock(rdp); |
8146c4e2 | 2165 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
cfcdef5e ED |
2166 | pending = rcu_segcblist_n_cbs(&rdp->cblist); |
2167 | bl = max(rdp->blimit, pending >> rcu_divisor); | |
2168 | if (unlikely(bl > 100)) | |
2169 | tlimit = local_clock() + rcu_resched_ns; | |
3c779dfe PM |
2170 | trace_rcu_batch_start(rcu_state.name, |
2171 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
15fecf89 PM |
2172 | rcu_segcblist_n_cbs(&rdp->cblist), bl); |
2173 | rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl); | |
7f36ef82 PM |
2174 | if (offloaded) |
2175 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); | |
5d6742b3 | 2176 | rcu_nocb_unlock_irqrestore(rdp, flags); |
64db4cff PM |
2177 | |
2178 | /* Invoke callbacks. */ | |
6a949b7a | 2179 | tick_dep_set_task(current, TICK_DEP_BIT_RCU); |
15fecf89 PM |
2180 | rhp = rcu_cblist_dequeue(&rcl); |
2181 | for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) { | |
2182 | debug_rcu_head_unqueue(rhp); | |
3c779dfe | 2183 | if (__rcu_reclaim(rcu_state.name, rhp)) |
15fecf89 PM |
2184 | rcu_cblist_dequeued_lazy(&rcl); |
2185 | /* | |
2186 | * Stop only if limit reached and CPU has something to do. | |
2187 | * Note: The rcl structure counts down from zero. | |
2188 | */ | |
ec5ef87b | 2189 | if (-rcl.len >= bl && !offloaded && |
dff1672d PM |
2190 | (need_resched() || |
2191 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff | 2192 | break; |
cfcdef5e ED |
2193 | if (unlikely(tlimit)) { |
2194 | /* only call local_clock() every 32 callbacks */ | |
2195 | if (likely((-rcl.len & 31) || local_clock() < tlimit)) | |
2196 | continue; | |
2197 | /* Exceeded the time limit, so leave. */ | |
2198 | break; | |
2199 | } | |
ec5ef87b | 2200 | if (offloaded) { |
5d6742b3 PM |
2201 | WARN_ON_ONCE(in_serving_softirq()); |
2202 | local_bh_enable(); | |
2203 | lockdep_assert_irqs_enabled(); | |
2204 | cond_resched_tasks_rcu_qs(); | |
2205 | lockdep_assert_irqs_enabled(); | |
2206 | local_bh_disable(); | |
2207 | } | |
64db4cff PM |
2208 | } |
2209 | ||
2210 | local_irq_save(flags); | |
5d6742b3 | 2211 | rcu_nocb_lock(rdp); |
4b27f20b | 2212 | count = -rcl.len; |
3c779dfe | 2213 | trace_rcu_batch_end(rcu_state.name, count, !!rcl.head, need_resched(), |
8ef0f37e | 2214 | is_idle_task(current), rcu_is_callbacks_kthread()); |
64db4cff | 2215 | |
15fecf89 PM |
2216 | /* Update counts and requeue any remaining callbacks. */ |
2217 | rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl); | |
b1420f1c | 2218 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
15fecf89 | 2219 | rcu_segcblist_insert_count(&rdp->cblist, &rcl); |
64db4cff PM |
2220 | |
2221 | /* Reinstate batch limit if we have worked down the excess. */ | |
15fecf89 | 2222 | count = rcu_segcblist_n_cbs(&rdp->cblist); |
d5a9a8c3 | 2223 | if (rdp->blimit >= DEFAULT_MAX_RCU_BLIMIT && count <= qlowmark) |
64db4cff PM |
2224 | rdp->blimit = blimit; |
2225 | ||
37c72e56 | 2226 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
15fecf89 | 2227 | if (count == 0 && rdp->qlen_last_fqs_check != 0) { |
37c72e56 | 2228 | rdp->qlen_last_fqs_check = 0; |
3c779dfe | 2229 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
15fecf89 PM |
2230 | } else if (count < rdp->qlen_last_fqs_check - qhimark) |
2231 | rdp->qlen_last_fqs_check = count; | |
efd88b02 PM |
2232 | |
2233 | /* | |
2234 | * The following usually indicates a double call_rcu(). To track | |
2235 | * this down, try building with CONFIG_DEBUG_OBJECTS_RCU_HEAD=y. | |
2236 | */ | |
d1b222c6 PM |
2237 | WARN_ON_ONCE(count == 0 && !rcu_segcblist_empty(&rdp->cblist)); |
2238 | WARN_ON_ONCE(!IS_ENABLED(CONFIG_RCU_NOCB_CPU) && | |
2239 | count != 0 && rcu_segcblist_empty(&rdp->cblist)); | |
37c72e56 | 2240 | |
5d6742b3 | 2241 | rcu_nocb_unlock_irqrestore(rdp, flags); |
64db4cff | 2242 | |
e0f23060 | 2243 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
ec5ef87b | 2244 | if (!offloaded && rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2245 | invoke_rcu_core(); |
6a949b7a | 2246 | tick_dep_clear_task(current, TICK_DEP_BIT_RCU); |
64db4cff PM |
2247 | } |
2248 | ||
2249 | /* | |
c98cac60 PM |
2250 | * This function is invoked from each scheduling-clock interrupt, |
2251 | * and checks to see if this CPU is in a non-context-switch quiescent | |
2252 | * state, for example, user mode or idle loop. It also schedules RCU | |
2253 | * core processing. If the current grace period has gone on too long, | |
2254 | * it will ask the scheduler to manufacture a context switch for the sole | |
2255 | * purpose of providing a providing the needed quiescent state. | |
64db4cff | 2256 | */ |
c98cac60 | 2257 | void rcu_sched_clock_irq(int user) |
64db4cff | 2258 | { |
f7f7bac9 | 2259 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
4e95020c | 2260 | raw_cpu_inc(rcu_data.ticks_this_gp); |
92aa39e9 | 2261 | /* The load-acquire pairs with the store-release setting to true. */ |
2dba13f0 | 2262 | if (smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
92aa39e9 | 2263 | /* Idle and userspace execution already are quiescent states. */ |
a0ef9ec2 | 2264 | if (!rcu_is_cpu_rrupt_from_idle() && !user) { |
92aa39e9 PM |
2265 | set_tsk_need_resched(current); |
2266 | set_preempt_need_resched(); | |
2267 | } | |
2dba13f0 | 2268 | __this_cpu_write(rcu_data.rcu_urgent_qs, false); |
64db4cff | 2269 | } |
c98cac60 | 2270 | rcu_flavor_sched_clock_irq(user); |
dd7dafd1 | 2271 | if (rcu_pending(user)) |
a46e0899 | 2272 | invoke_rcu_core(); |
07f27570 | 2273 | |
f7f7bac9 | 2274 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2275 | } |
2276 | ||
64db4cff | 2277 | /* |
5d8a752e ZZ |
2278 | * Scan the leaf rcu_node structures. For each structure on which all |
2279 | * CPUs have reported a quiescent state and on which there are tasks | |
2280 | * blocking the current grace period, initiate RCU priority boosting. | |
2281 | * Otherwise, invoke the specified function to check dyntick state for | |
2282 | * each CPU that has not yet reported a quiescent state. | |
64db4cff | 2283 | */ |
8ff0b907 | 2284 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)) |
64db4cff | 2285 | { |
64db4cff PM |
2286 | int cpu; |
2287 | unsigned long flags; | |
2288 | unsigned long mask; | |
66e4c33b | 2289 | struct rcu_data *rdp; |
a0b6c9a7 | 2290 | struct rcu_node *rnp; |
64db4cff | 2291 | |
aedf4ba9 | 2292 | rcu_for_each_leaf_node(rnp) { |
cee43939 | 2293 | cond_resched_tasks_rcu_qs(); |
64db4cff | 2294 | mask = 0; |
2a67e741 | 2295 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
a0b6c9a7 | 2296 | if (rnp->qsmask == 0) { |
01b1d88b | 2297 | if (!IS_ENABLED(CONFIG_PREEMPTION) || |
a77da14c PM |
2298 | rcu_preempt_blocked_readers_cgp(rnp)) { |
2299 | /* | |
2300 | * No point in scanning bits because they | |
2301 | * are all zero. But we might need to | |
2302 | * priority-boost blocked readers. | |
2303 | */ | |
2304 | rcu_initiate_boost(rnp, flags); | |
2305 | /* rcu_initiate_boost() releases rnp->lock */ | |
2306 | continue; | |
2307 | } | |
92816435 PM |
2308 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
2309 | continue; | |
64db4cff | 2310 | } |
bc75e999 MR |
2311 | for_each_leaf_node_possible_cpu(rnp, cpu) { |
2312 | unsigned long bit = leaf_node_cpu_bit(rnp, cpu); | |
0edd1b17 | 2313 | if ((rnp->qsmask & bit) != 0) { |
66e4c33b PM |
2314 | rdp = per_cpu_ptr(&rcu_data, cpu); |
2315 | if (f(rdp)) { | |
0edd1b17 | 2316 | mask |= bit; |
516e5ae0 | 2317 | rcu_disable_urgency_upon_qs(rdp); |
66e4c33b | 2318 | } |
0edd1b17 | 2319 | } |
64db4cff | 2320 | } |
45f014c5 | 2321 | if (mask != 0) { |
c9a24e2d | 2322 | /* Idle/offline CPUs, report (releases rnp->lock). */ |
b50912d0 | 2323 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
0aa04b05 PM |
2324 | } else { |
2325 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 2326 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2327 | } |
64db4cff | 2328 | } |
64db4cff PM |
2329 | } |
2330 | ||
2331 | /* | |
2332 | * Force quiescent states on reluctant CPUs, and also detect which | |
2333 | * CPUs are in dyntick-idle mode. | |
2334 | */ | |
cd920e5a | 2335 | void rcu_force_quiescent_state(void) |
64db4cff PM |
2336 | { |
2337 | unsigned long flags; | |
394f2769 PM |
2338 | bool ret; |
2339 | struct rcu_node *rnp; | |
2340 | struct rcu_node *rnp_old = NULL; | |
2341 | ||
2342 | /* Funnel through hierarchy to reduce memory contention. */ | |
da1df50d | 2343 | rnp = __this_cpu_read(rcu_data.mynode); |
394f2769 | 2344 | for (; rnp != NULL; rnp = rnp->parent) { |
67a0edbf | 2345 | ret = (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) || |
66e4c33b | 2346 | !raw_spin_trylock(&rnp->fqslock); |
394f2769 PM |
2347 | if (rnp_old != NULL) |
2348 | raw_spin_unlock(&rnp_old->fqslock); | |
d62df573 | 2349 | if (ret) |
394f2769 | 2350 | return; |
394f2769 PM |
2351 | rnp_old = rnp; |
2352 | } | |
336a4f6c | 2353 | /* rnp_old == rcu_get_root(), rnp == NULL. */ |
64db4cff | 2354 | |
394f2769 | 2355 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 2356 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 2357 | raw_spin_unlock(&rnp_old->fqslock); |
67a0edbf | 2358 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
67c583a7 | 2359 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 2360 | return; /* Someone beat us to it. */ |
46a1e34e | 2361 | } |
67a0edbf PM |
2362 | WRITE_ONCE(rcu_state.gp_flags, |
2363 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
67c583a7 | 2364 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
532c00c9 | 2365 | rcu_gp_kthread_wake(); |
64db4cff | 2366 | } |
cd920e5a | 2367 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); |
64db4cff | 2368 | |
fb60e533 | 2369 | /* Perform RCU core processing work for the current CPU. */ |
48d07c04 | 2370 | static __latent_entropy void rcu_core(void) |
64db4cff PM |
2371 | { |
2372 | unsigned long flags; | |
da1df50d | 2373 | struct rcu_data *rdp = raw_cpu_ptr(&rcu_data); |
26d950a9 | 2374 | struct rcu_node *rnp = rdp->mynode; |
c1ab99d6 PM |
2375 | const bool offloaded = IS_ENABLED(CONFIG_RCU_NOCB_CPU) && |
2376 | rcu_segcblist_is_offloaded(&rdp->cblist); | |
64db4cff | 2377 | |
b049fdf8 PM |
2378 | if (cpu_is_offline(smp_processor_id())) |
2379 | return; | |
2380 | trace_rcu_utilization(TPS("Start RCU core")); | |
50dc7def | 2381 | WARN_ON_ONCE(!rdp->beenonline); |
2e597558 | 2382 | |
3e310098 | 2383 | /* Report any deferred quiescent states if preemption enabled. */ |
fced9c8c | 2384 | if (!(preempt_count() & PREEMPT_MASK)) { |
3e310098 | 2385 | rcu_preempt_deferred_qs(current); |
fced9c8c PM |
2386 | } else if (rcu_preempt_need_deferred_qs(current)) { |
2387 | set_tsk_need_resched(current); | |
2388 | set_preempt_need_resched(); | |
2389 | } | |
3e310098 | 2390 | |
64db4cff | 2391 | /* Update RCU state based on any recent quiescent states. */ |
8087d3e3 | 2392 | rcu_check_quiescent_state(rdp); |
64db4cff | 2393 | |
bd7af846 | 2394 | /* No grace period and unregistered callbacks? */ |
de8e8730 | 2395 | if (!rcu_gp_in_progress() && |
c1ab99d6 | 2396 | rcu_segcblist_is_enabled(&rdp->cblist) && !offloaded) { |
bd7af846 | 2397 | local_irq_save(flags); |
e44e73ca | 2398 | if (!rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
c6e09b97 | 2399 | rcu_accelerate_cbs_unlocked(rnp, rdp); |
e44e73ca | 2400 | local_irq_restore(flags); |
64db4cff PM |
2401 | } |
2402 | ||
791416c4 | 2403 | rcu_check_gp_start_stall(rnp, rdp, rcu_jiffies_till_stall_check()); |
26d950a9 | 2404 | |
64db4cff | 2405 | /* If there are callbacks ready, invoke them. */ |
c1ab99d6 | 2406 | if (!offloaded && rcu_segcblist_ready_cbs(&rdp->cblist) && |
43e903ad PM |
2407 | likely(READ_ONCE(rcu_scheduler_fully_active))) |
2408 | rcu_do_batch(rdp); | |
96d3fd0d PM |
2409 | |
2410 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
2411 | do_nocb_deferred_wakeup(rdp); | |
f7f7bac9 | 2412 | trace_rcu_utilization(TPS("End RCU core")); |
64db4cff PM |
2413 | } |
2414 | ||
48d07c04 SAS |
2415 | static void rcu_core_si(struct softirq_action *h) |
2416 | { | |
2417 | rcu_core(); | |
2418 | } | |
2419 | ||
2420 | static void rcu_wake_cond(struct task_struct *t, int status) | |
2421 | { | |
2422 | /* | |
2423 | * If the thread is yielding, only wake it when this | |
2424 | * is invoked from idle | |
2425 | */ | |
2426 | if (t && (status != RCU_KTHREAD_YIELDING || is_idle_task(current))) | |
2427 | wake_up_process(t); | |
2428 | } | |
2429 | ||
2430 | static void invoke_rcu_core_kthread(void) | |
2431 | { | |
2432 | struct task_struct *t; | |
2433 | unsigned long flags; | |
2434 | ||
2435 | local_irq_save(flags); | |
2436 | __this_cpu_write(rcu_data.rcu_cpu_has_work, 1); | |
2437 | t = __this_cpu_read(rcu_data.rcu_cpu_kthread_task); | |
2438 | if (t != NULL && t != current) | |
2439 | rcu_wake_cond(t, __this_cpu_read(rcu_data.rcu_cpu_kthread_status)); | |
2440 | local_irq_restore(flags); | |
2441 | } | |
2442 | ||
48d07c04 SAS |
2443 | /* |
2444 | * Wake up this CPU's rcuc kthread to do RCU core processing. | |
2445 | */ | |
a46e0899 | 2446 | static void invoke_rcu_core(void) |
09223371 | 2447 | { |
48d07c04 SAS |
2448 | if (!cpu_online(smp_processor_id())) |
2449 | return; | |
2450 | if (use_softirq) | |
b0f74036 | 2451 | raise_softirq(RCU_SOFTIRQ); |
48d07c04 SAS |
2452 | else |
2453 | invoke_rcu_core_kthread(); | |
2454 | } | |
2455 | ||
2456 | static void rcu_cpu_kthread_park(unsigned int cpu) | |
2457 | { | |
2458 | per_cpu(rcu_data.rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; | |
2459 | } | |
2460 | ||
2461 | static int rcu_cpu_kthread_should_run(unsigned int cpu) | |
2462 | { | |
2463 | return __this_cpu_read(rcu_data.rcu_cpu_has_work); | |
2464 | } | |
2465 | ||
2466 | /* | |
2467 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces | |
2468 | * the RCU softirq used in configurations of RCU that do not support RCU | |
2469 | * priority boosting. | |
2470 | */ | |
2471 | static void rcu_cpu_kthread(unsigned int cpu) | |
2472 | { | |
2473 | unsigned int *statusp = this_cpu_ptr(&rcu_data.rcu_cpu_kthread_status); | |
2474 | char work, *workp = this_cpu_ptr(&rcu_data.rcu_cpu_has_work); | |
2475 | int spincnt; | |
2476 | ||
2477 | for (spincnt = 0; spincnt < 10; spincnt++) { | |
2478 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); | |
2479 | local_bh_disable(); | |
2480 | *statusp = RCU_KTHREAD_RUNNING; | |
2481 | local_irq_disable(); | |
2482 | work = *workp; | |
2483 | *workp = 0; | |
2484 | local_irq_enable(); | |
2485 | if (work) | |
2486 | rcu_core(); | |
2487 | local_bh_enable(); | |
2488 | if (*workp == 0) { | |
2489 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); | |
2490 | *statusp = RCU_KTHREAD_WAITING; | |
2491 | return; | |
2492 | } | |
2493 | } | |
2494 | *statusp = RCU_KTHREAD_YIELDING; | |
2495 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); | |
2496 | schedule_timeout_interruptible(2); | |
2497 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); | |
2498 | *statusp = RCU_KTHREAD_WAITING; | |
2499 | } | |
2500 | ||
2501 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { | |
2502 | .store = &rcu_data.rcu_cpu_kthread_task, | |
2503 | .thread_should_run = rcu_cpu_kthread_should_run, | |
2504 | .thread_fn = rcu_cpu_kthread, | |
2505 | .thread_comm = "rcuc/%u", | |
2506 | .setup = rcu_cpu_kthread_setup, | |
2507 | .park = rcu_cpu_kthread_park, | |
2508 | }; | |
2509 | ||
2510 | /* | |
2511 | * Spawn per-CPU RCU core processing kthreads. | |
2512 | */ | |
2513 | static int __init rcu_spawn_core_kthreads(void) | |
2514 | { | |
2515 | int cpu; | |
2516 | ||
2517 | for_each_possible_cpu(cpu) | |
2518 | per_cpu(rcu_data.rcu_cpu_has_work, cpu) = 0; | |
2519 | if (!IS_ENABLED(CONFIG_RCU_BOOST) && use_softirq) | |
2520 | return 0; | |
2521 | WARN_ONCE(smpboot_register_percpu_thread(&rcu_cpu_thread_spec), | |
2522 | "%s: Could not start rcuc kthread, OOM is now expected behavior\n", __func__); | |
2523 | return 0; | |
09223371 | 2524 | } |
48d07c04 | 2525 | early_initcall(rcu_spawn_core_kthreads); |
09223371 | 2526 | |
29154c57 PM |
2527 | /* |
2528 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
2529 | */ | |
5c7d8967 PM |
2530 | static void __call_rcu_core(struct rcu_data *rdp, struct rcu_head *head, |
2531 | unsigned long flags) | |
64db4cff | 2532 | { |
62fde6ed PM |
2533 | /* |
2534 | * If called from an extended quiescent state, invoke the RCU | |
2535 | * core in order to force a re-evaluation of RCU's idleness. | |
2536 | */ | |
9910affa | 2537 | if (!rcu_is_watching()) |
62fde6ed PM |
2538 | invoke_rcu_core(); |
2539 | ||
a16b7a69 | 2540 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 2541 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 2542 | return; |
64db4cff | 2543 | |
37c72e56 PM |
2544 | /* |
2545 | * Force the grace period if too many callbacks or too long waiting. | |
cd920e5a | 2546 | * Enforce hysteresis, and don't invoke rcu_force_quiescent_state() |
37c72e56 | 2547 | * if some other CPU has recently done so. Also, don't bother |
cd920e5a | 2548 | * invoking rcu_force_quiescent_state() if the newly enqueued callback |
37c72e56 PM |
2549 | * is the only one waiting for a grace period to complete. |
2550 | */ | |
15fecf89 PM |
2551 | if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) > |
2552 | rdp->qlen_last_fqs_check + qhimark)) { | |
b52573d2 PM |
2553 | |
2554 | /* Are we ignoring a completed grace period? */ | |
15cabdff | 2555 | note_gp_changes(rdp); |
b52573d2 PM |
2556 | |
2557 | /* Start a new grace period if one not already started. */ | |
de8e8730 | 2558 | if (!rcu_gp_in_progress()) { |
c6e09b97 | 2559 | rcu_accelerate_cbs_unlocked(rdp->mynode, rdp); |
b52573d2 PM |
2560 | } else { |
2561 | /* Give the grace period a kick. */ | |
d5a9a8c3 | 2562 | rdp->blimit = DEFAULT_MAX_RCU_BLIMIT; |
5c7d8967 | 2563 | if (rcu_state.n_force_qs == rdp->n_force_qs_snap && |
15fecf89 | 2564 | rcu_segcblist_first_pend_cb(&rdp->cblist) != head) |
cd920e5a | 2565 | rcu_force_quiescent_state(); |
5c7d8967 | 2566 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
15fecf89 | 2567 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); |
b52573d2 | 2568 | } |
4cdfc175 | 2569 | } |
29154c57 PM |
2570 | } |
2571 | ||
ae150184 PM |
2572 | /* |
2573 | * RCU callback function to leak a callback. | |
2574 | */ | |
2575 | static void rcu_leak_callback(struct rcu_head *rhp) | |
2576 | { | |
2577 | } | |
2578 | ||
3fbfbf7a PM |
2579 | /* |
2580 | * Helper function for call_rcu() and friends. The cpu argument will | |
2581 | * normally be -1, indicating "currently running CPU". It may specify | |
dd46a788 | 2582 | * a CPU only if that CPU is a no-CBs CPU. Currently, only rcu_barrier() |
3fbfbf7a PM |
2583 | * is expected to specify a CPU. |
2584 | */ | |
64db4cff | 2585 | static void |
5d6742b3 | 2586 | __call_rcu(struct rcu_head *head, rcu_callback_t func, bool lazy) |
64db4cff PM |
2587 | { |
2588 | unsigned long flags; | |
2589 | struct rcu_data *rdp; | |
5d6742b3 | 2590 | bool was_alldone; |
64db4cff | 2591 | |
b8f2ed53 PM |
2592 | /* Misaligned rcu_head! */ |
2593 | WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1)); | |
2594 | ||
ae150184 | 2595 | if (debug_rcu_head_queue(head)) { |
fa3c6647 PM |
2596 | /* |
2597 | * Probable double call_rcu(), so leak the callback. | |
2598 | * Use rcu:rcu_callback trace event to find the previous | |
2599 | * time callback was passed to __call_rcu(). | |
2600 | */ | |
d75f773c | 2601 | WARN_ONCE(1, "__call_rcu(): Double-freed CB %p->%pS()!!!\n", |
fa3c6647 | 2602 | head, head->func); |
7d0ae808 | 2603 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
2604 | return; |
2605 | } | |
64db4cff PM |
2606 | head->func = func; |
2607 | head->next = NULL; | |
64db4cff | 2608 | local_irq_save(flags); |
da1df50d | 2609 | rdp = this_cpu_ptr(&rcu_data); |
64db4cff PM |
2610 | |
2611 | /* Add the callback to our list. */ | |
5d6742b3 PM |
2612 | if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist))) { |
2613 | // This can trigger due to call_rcu() from offline CPU: | |
2614 | WARN_ON_ONCE(rcu_scheduler_active != RCU_SCHEDULER_INACTIVE); | |
34404ca8 | 2615 | WARN_ON_ONCE(!rcu_is_watching()); |
5d6742b3 PM |
2616 | // Very early boot, before rcu_init(). Initialize if needed |
2617 | // and then drop through to queue the callback. | |
15fecf89 PM |
2618 | if (rcu_segcblist_empty(&rdp->cblist)) |
2619 | rcu_segcblist_init(&rdp->cblist); | |
0d8ee37e | 2620 | } |
d1b222c6 PM |
2621 | if (rcu_nocb_try_bypass(rdp, head, &was_alldone, flags)) |
2622 | return; // Enqueued onto ->nocb_bypass, so just leave. | |
2623 | /* If we get here, rcu_nocb_try_bypass() acquired ->nocb_lock. */ | |
15fecf89 | 2624 | rcu_segcblist_enqueue(&rdp->cblist, head, lazy); |
d4c08f2a | 2625 | if (__is_kfree_rcu_offset((unsigned long)func)) |
3c779dfe PM |
2626 | trace_rcu_kfree_callback(rcu_state.name, head, |
2627 | (unsigned long)func, | |
15fecf89 PM |
2628 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2629 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2630 | else |
3c779dfe | 2631 | trace_rcu_callback(rcu_state.name, head, |
15fecf89 PM |
2632 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2633 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2634 | |
29154c57 | 2635 | /* Go handle any RCU core processing required. */ |
5d6742b3 PM |
2636 | if (IS_ENABLED(CONFIG_RCU_NOCB_CPU) && |
2637 | unlikely(rcu_segcblist_is_offloaded(&rdp->cblist))) { | |
2638 | __call_rcu_nocb_wake(rdp, was_alldone, flags); /* unlocks */ | |
2639 | } else { | |
2640 | __call_rcu_core(rdp, head, flags); | |
2641 | local_irq_restore(flags); | |
2642 | } | |
64db4cff PM |
2643 | } |
2644 | ||
a68a2bb2 | 2645 | /** |
45975c7d | 2646 | * call_rcu() - Queue an RCU callback for invocation after a grace period. |
a68a2bb2 PM |
2647 | * @head: structure to be used for queueing the RCU updates. |
2648 | * @func: actual callback function to be invoked after the grace period | |
2649 | * | |
2650 | * The callback function will be invoked some time after a full grace | |
45975c7d PM |
2651 | * period elapses, in other words after all pre-existing RCU read-side |
2652 | * critical sections have completed. However, the callback function | |
2653 | * might well execute concurrently with RCU read-side critical sections | |
2654 | * that started after call_rcu() was invoked. RCU read-side critical | |
2655 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), and | |
2656 | * may be nested. In addition, regions of code across which interrupts, | |
2657 | * preemption, or softirqs have been disabled also serve as RCU read-side | |
2658 | * critical sections. This includes hardware interrupt handlers, softirq | |
2659 | * handlers, and NMI handlers. | |
2660 | * | |
2661 | * Note that all CPUs must agree that the grace period extended beyond | |
2662 | * all pre-existing RCU read-side critical section. On systems with more | |
2663 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
2664 | * guaranteed to have executed a full memory barrier since the end of its | |
2665 | * last RCU read-side critical section whose beginning preceded the call | |
2666 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
2667 | * critical section that continues beyond the start of "func()" must have | |
2668 | * executed a memory barrier after the call_rcu() but before the beginning | |
2669 | * of that RCU read-side critical section. Note that these guarantees | |
2670 | * include CPUs that are offline, idle, or executing in user mode, as | |
2671 | * well as CPUs that are executing in the kernel. | |
2672 | * | |
2673 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
2674 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
2675 | * guaranteed to execute a full memory barrier during the time interval | |
2676 | * between the call to call_rcu() and the invocation of "func()" -- even | |
2677 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
2678 | * more than one CPU). | |
2679 | */ | |
2680 | void call_rcu(struct rcu_head *head, rcu_callback_t func) | |
2681 | { | |
5d6742b3 | 2682 | __call_rcu(head, func, 0); |
45975c7d PM |
2683 | } |
2684 | EXPORT_SYMBOL_GPL(call_rcu); | |
64db4cff | 2685 | |
495aa969 ACB |
2686 | /* |
2687 | * Queue an RCU callback for lazy invocation after a grace period. | |
2688 | * This will likely be later named something like "call_rcu_lazy()", | |
2689 | * but this change will require some way of tagging the lazy RCU | |
2690 | * callbacks in the list of pending callbacks. Until then, this | |
2691 | * function may only be called from __kfree_rcu(). | |
2692 | */ | |
98ece508 | 2693 | void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func) |
495aa969 | 2694 | { |
5d6742b3 | 2695 | __call_rcu(head, func, 1); |
495aa969 ACB |
2696 | } |
2697 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
2698 | ||
e5bc3af7 PM |
2699 | /* |
2700 | * During early boot, any blocking grace-period wait automatically | |
2701 | * implies a grace period. Later on, this is never the case for PREEMPT. | |
2702 | * | |
2703 | * Howevr, because a context switch is a grace period for !PREEMPT, any | |
2704 | * blocking grace-period wait automatically implies a grace period if | |
2705 | * there is only one CPU online at any point time during execution of | |
2706 | * either synchronize_rcu() or synchronize_rcu_expedited(). It is OK to | |
2707 | * occasionally incorrectly indicate that there are multiple CPUs online | |
2708 | * when there was in fact only one the whole time, as this just adds some | |
2709 | * overhead: RCU still operates correctly. | |
2710 | */ | |
2711 | static int rcu_blocking_is_gp(void) | |
2712 | { | |
2713 | int ret; | |
2714 | ||
01b1d88b | 2715 | if (IS_ENABLED(CONFIG_PREEMPTION)) |
e5bc3af7 PM |
2716 | return rcu_scheduler_active == RCU_SCHEDULER_INACTIVE; |
2717 | might_sleep(); /* Check for RCU read-side critical section. */ | |
2718 | preempt_disable(); | |
2719 | ret = num_online_cpus() <= 1; | |
2720 | preempt_enable(); | |
2721 | return ret; | |
2722 | } | |
2723 | ||
2724 | /** | |
2725 | * synchronize_rcu - wait until a grace period has elapsed. | |
2726 | * | |
2727 | * Control will return to the caller some time after a full grace | |
2728 | * period has elapsed, in other words after all currently executing RCU | |
2729 | * read-side critical sections have completed. Note, however, that | |
2730 | * upon return from synchronize_rcu(), the caller might well be executing | |
2731 | * concurrently with new RCU read-side critical sections that began while | |
2732 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
2733 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
2734 | * In addition, regions of code across which interrupts, preemption, or | |
2735 | * softirqs have been disabled also serve as RCU read-side critical | |
2736 | * sections. This includes hardware interrupt handlers, softirq handlers, | |
2737 | * and NMI handlers. | |
2738 | * | |
2739 | * Note that this guarantee implies further memory-ordering guarantees. | |
2740 | * On systems with more than one CPU, when synchronize_rcu() returns, | |
2741 | * each CPU is guaranteed to have executed a full memory barrier since | |
2742 | * the end of its last RCU read-side critical section whose beginning | |
2743 | * preceded the call to synchronize_rcu(). In addition, each CPU having | |
2744 | * an RCU read-side critical section that extends beyond the return from | |
2745 | * synchronize_rcu() is guaranteed to have executed a full memory barrier | |
2746 | * after the beginning of synchronize_rcu() and before the beginning of | |
2747 | * that RCU read-side critical section. Note that these guarantees include | |
2748 | * CPUs that are offline, idle, or executing in user mode, as well as CPUs | |
2749 | * that are executing in the kernel. | |
2750 | * | |
2751 | * Furthermore, if CPU A invoked synchronize_rcu(), which returned | |
2752 | * to its caller on CPU B, then both CPU A and CPU B are guaranteed | |
2753 | * to have executed a full memory barrier during the execution of | |
2754 | * synchronize_rcu() -- even if CPU A and CPU B are the same CPU (but | |
2755 | * again only if the system has more than one CPU). | |
2756 | */ | |
2757 | void synchronize_rcu(void) | |
2758 | { | |
2759 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || | |
2760 | lock_is_held(&rcu_lock_map) || | |
2761 | lock_is_held(&rcu_sched_lock_map), | |
2762 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
2763 | if (rcu_blocking_is_gp()) | |
2764 | return; | |
2765 | if (rcu_gp_is_expedited()) | |
2766 | synchronize_rcu_expedited(); | |
2767 | else | |
2768 | wait_rcu_gp(call_rcu); | |
2769 | } | |
2770 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
2771 | ||
765a3f4f PM |
2772 | /** |
2773 | * get_state_synchronize_rcu - Snapshot current RCU state | |
2774 | * | |
2775 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
2776 | * to determine whether or not a full grace period has elapsed in the | |
2777 | * meantime. | |
2778 | */ | |
2779 | unsigned long get_state_synchronize_rcu(void) | |
2780 | { | |
2781 | /* | |
2782 | * Any prior manipulation of RCU-protected data must happen | |
e4be81a2 | 2783 | * before the load from ->gp_seq. |
765a3f4f PM |
2784 | */ |
2785 | smp_mb(); /* ^^^ */ | |
16fc9c60 | 2786 | return rcu_seq_snap(&rcu_state.gp_seq); |
765a3f4f PM |
2787 | } |
2788 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
2789 | ||
2790 | /** | |
2791 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
2792 | * | |
2793 | * @oldstate: return value from earlier call to get_state_synchronize_rcu() | |
2794 | * | |
2795 | * If a full RCU grace period has elapsed since the earlier call to | |
2796 | * get_state_synchronize_rcu(), just return. Otherwise, invoke | |
2797 | * synchronize_rcu() to wait for a full grace period. | |
2798 | * | |
2799 | * Yes, this function does not take counter wrap into account. But | |
2800 | * counter wrap is harmless. If the counter wraps, we have waited for | |
2801 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
2802 | * so waiting for one additional grace period should be just fine. | |
2803 | */ | |
2804 | void cond_synchronize_rcu(unsigned long oldstate) | |
2805 | { | |
16fc9c60 | 2806 | if (!rcu_seq_done(&rcu_state.gp_seq, oldstate)) |
765a3f4f | 2807 | synchronize_rcu(); |
e4be81a2 PM |
2808 | else |
2809 | smp_mb(); /* Ensure GP ends before subsequent accesses. */ | |
765a3f4f PM |
2810 | } |
2811 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
2812 | ||
64db4cff | 2813 | /* |
98ece508 | 2814 | * Check to see if there is any immediate RCU-related work to be done by |
49918a54 PM |
2815 | * the current CPU, returning 1 if so and zero otherwise. The checks are |
2816 | * in order of increasing expense: checks that can be carried out against | |
2817 | * CPU-local state are performed first. However, we must check for CPU | |
2818 | * stalls first, else we might not get a chance. | |
64db4cff | 2819 | */ |
dd7dafd1 | 2820 | static int rcu_pending(int user) |
64db4cff | 2821 | { |
ed93dfc6 | 2822 | bool gp_in_progress; |
98ece508 | 2823 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
2f51f988 PM |
2824 | struct rcu_node *rnp = rdp->mynode; |
2825 | ||
64db4cff | 2826 | /* Check for CPU stalls, if enabled. */ |
ea12ff2b | 2827 | check_cpu_stall(rdp); |
64db4cff | 2828 | |
85f69b32 PM |
2829 | /* Does this CPU need a deferred NOCB wakeup? */ |
2830 | if (rcu_nocb_need_deferred_wakeup(rdp)) | |
2831 | return 1; | |
2832 | ||
dd7dafd1 PM |
2833 | /* Is this a nohz_full CPU in userspace or idle? (Ignore RCU if so.) */ |
2834 | if ((user || rcu_is_cpu_rrupt_from_idle()) && rcu_nohz_full_cpu()) | |
a096932f PM |
2835 | return 0; |
2836 | ||
64db4cff | 2837 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
ed93dfc6 PM |
2838 | gp_in_progress = rcu_gp_in_progress(); |
2839 | if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm && gp_in_progress) | |
64db4cff PM |
2840 | return 1; |
2841 | ||
2842 | /* Does this CPU have callbacks ready to invoke? */ | |
01c495f7 | 2843 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
64db4cff PM |
2844 | return 1; |
2845 | ||
2846 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
ed93dfc6 | 2847 | if (!gp_in_progress && rcu_segcblist_is_enabled(&rdp->cblist) && |
921bb5fa PM |
2848 | (!IS_ENABLED(CONFIG_RCU_NOCB_CPU) || |
2849 | !rcu_segcblist_is_offloaded(&rdp->cblist)) && | |
c1935209 | 2850 | !rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
64db4cff PM |
2851 | return 1; |
2852 | ||
67e14c1e PM |
2853 | /* Have RCU grace period completed or started? */ |
2854 | if (rcu_seq_current(&rnp->gp_seq) != rdp->gp_seq || | |
01c495f7 | 2855 | unlikely(READ_ONCE(rdp->gpwrap))) /* outside lock */ |
64db4cff PM |
2856 | return 1; |
2857 | ||
64db4cff PM |
2858 | /* nothing to do */ |
2859 | return 0; | |
2860 | } | |
2861 | ||
a83eff0a | 2862 | /* |
dd46a788 | 2863 | * Helper function for rcu_barrier() tracing. If tracing is disabled, |
a83eff0a PM |
2864 | * the compiler is expected to optimize this away. |
2865 | */ | |
dd46a788 | 2866 | static void rcu_barrier_trace(const char *s, int cpu, unsigned long done) |
a83eff0a | 2867 | { |
8344b871 PM |
2868 | trace_rcu_barrier(rcu_state.name, s, cpu, |
2869 | atomic_read(&rcu_state.barrier_cpu_count), done); | |
a83eff0a PM |
2870 | } |
2871 | ||
b1420f1c | 2872 | /* |
dd46a788 PM |
2873 | * RCU callback function for rcu_barrier(). If we are last, wake |
2874 | * up the task executing rcu_barrier(). | |
b1420f1c | 2875 | */ |
24ebbca8 | 2876 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 2877 | { |
ec9f5835 | 2878 | if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) { |
dd46a788 | 2879 | rcu_barrier_trace(TPS("LastCB"), -1, |
66e4c33b | 2880 | rcu_state.barrier_sequence); |
ec9f5835 | 2881 | complete(&rcu_state.barrier_completion); |
a83eff0a | 2882 | } else { |
dd46a788 | 2883 | rcu_barrier_trace(TPS("CB"), -1, rcu_state.barrier_sequence); |
a83eff0a | 2884 | } |
d0ec774c PM |
2885 | } |
2886 | ||
2887 | /* | |
2888 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
2889 | */ | |
ec9f5835 | 2890 | static void rcu_barrier_func(void *unused) |
d0ec774c | 2891 | { |
da1df50d | 2892 | struct rcu_data *rdp = raw_cpu_ptr(&rcu_data); |
d0ec774c | 2893 | |
dd46a788 | 2894 | rcu_barrier_trace(TPS("IRQ"), -1, rcu_state.barrier_sequence); |
f92c734f PM |
2895 | rdp->barrier_head.func = rcu_barrier_callback; |
2896 | debug_rcu_head_queue(&rdp->barrier_head); | |
5d6742b3 | 2897 | rcu_nocb_lock(rdp); |
d1b222c6 | 2898 | WARN_ON_ONCE(!rcu_nocb_flush_bypass(rdp, NULL, jiffies)); |
f92c734f | 2899 | if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head, 0)) { |
ec9f5835 | 2900 | atomic_inc(&rcu_state.barrier_cpu_count); |
f92c734f PM |
2901 | } else { |
2902 | debug_rcu_head_unqueue(&rdp->barrier_head); | |
dd46a788 | 2903 | rcu_barrier_trace(TPS("IRQNQ"), -1, |
66e4c33b | 2904 | rcu_state.barrier_sequence); |
f92c734f | 2905 | } |
5d6742b3 | 2906 | rcu_nocb_unlock(rdp); |
d0ec774c PM |
2907 | } |
2908 | ||
dd46a788 PM |
2909 | /** |
2910 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
2911 | * | |
2912 | * Note that this primitive does not necessarily wait for an RCU grace period | |
2913 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
2914 | * in the system, then rcu_barrier() is within its rights to return | |
2915 | * immediately, without waiting for anything, much less an RCU grace period. | |
d0ec774c | 2916 | */ |
dd46a788 | 2917 | void rcu_barrier(void) |
d0ec774c | 2918 | { |
b1420f1c | 2919 | int cpu; |
b1420f1c | 2920 | struct rcu_data *rdp; |
ec9f5835 | 2921 | unsigned long s = rcu_seq_snap(&rcu_state.barrier_sequence); |
b1420f1c | 2922 | |
dd46a788 | 2923 | rcu_barrier_trace(TPS("Begin"), -1, s); |
b1420f1c | 2924 | |
e74f4c45 | 2925 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
ec9f5835 | 2926 | mutex_lock(&rcu_state.barrier_mutex); |
b1420f1c | 2927 | |
4f525a52 | 2928 | /* Did someone else do our work for us? */ |
ec9f5835 | 2929 | if (rcu_seq_done(&rcu_state.barrier_sequence, s)) { |
dd46a788 | 2930 | rcu_barrier_trace(TPS("EarlyExit"), -1, |
66e4c33b | 2931 | rcu_state.barrier_sequence); |
cf3a9c48 | 2932 | smp_mb(); /* caller's subsequent code after above check. */ |
ec9f5835 | 2933 | mutex_unlock(&rcu_state.barrier_mutex); |
cf3a9c48 PM |
2934 | return; |
2935 | } | |
2936 | ||
4f525a52 | 2937 | /* Mark the start of the barrier operation. */ |
ec9f5835 | 2938 | rcu_seq_start(&rcu_state.barrier_sequence); |
dd46a788 | 2939 | rcu_barrier_trace(TPS("Inc1"), -1, rcu_state.barrier_sequence); |
b1420f1c | 2940 | |
d0ec774c | 2941 | /* |
b1420f1c PM |
2942 | * Initialize the count to one rather than to zero in order to |
2943 | * avoid a too-soon return to zero in case of a short grace period | |
1331e7a1 PM |
2944 | * (or preemption of this task). Exclude CPU-hotplug operations |
2945 | * to ensure that no offline CPU has callbacks queued. | |
d0ec774c | 2946 | */ |
ec9f5835 PM |
2947 | init_completion(&rcu_state.barrier_completion); |
2948 | atomic_set(&rcu_state.barrier_cpu_count, 1); | |
1331e7a1 | 2949 | get_online_cpus(); |
b1420f1c PM |
2950 | |
2951 | /* | |
1331e7a1 PM |
2952 | * Force each CPU with callbacks to register a new callback. |
2953 | * When that callback is invoked, we will know that all of the | |
2954 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 2955 | */ |
3fbfbf7a | 2956 | for_each_possible_cpu(cpu) { |
da1df50d | 2957 | rdp = per_cpu_ptr(&rcu_data, cpu); |
ce5215c1 PM |
2958 | if (!cpu_online(cpu) && |
2959 | !rcu_segcblist_is_offloaded(&rdp->cblist)) | |
2960 | continue; | |
5d6742b3 | 2961 | if (rcu_segcblist_n_cbs(&rdp->cblist)) { |
dd46a788 | 2962 | rcu_barrier_trace(TPS("OnlineQ"), cpu, |
66e4c33b | 2963 | rcu_state.barrier_sequence); |
ec9f5835 | 2964 | smp_call_function_single(cpu, rcu_barrier_func, NULL, 1); |
b1420f1c | 2965 | } else { |
dd46a788 | 2966 | rcu_barrier_trace(TPS("OnlineNQ"), cpu, |
66e4c33b | 2967 | rcu_state.barrier_sequence); |
b1420f1c PM |
2968 | } |
2969 | } | |
1331e7a1 | 2970 | put_online_cpus(); |
b1420f1c PM |
2971 | |
2972 | /* | |
2973 | * Now that we have an rcu_barrier_callback() callback on each | |
2974 | * CPU, and thus each counted, remove the initial count. | |
2975 | */ | |
ec9f5835 PM |
2976 | if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) |
2977 | complete(&rcu_state.barrier_completion); | |
b1420f1c PM |
2978 | |
2979 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
ec9f5835 | 2980 | wait_for_completion(&rcu_state.barrier_completion); |
b1420f1c | 2981 | |
4f525a52 | 2982 | /* Mark the end of the barrier operation. */ |
dd46a788 | 2983 | rcu_barrier_trace(TPS("Inc2"), -1, rcu_state.barrier_sequence); |
ec9f5835 | 2984 | rcu_seq_end(&rcu_state.barrier_sequence); |
4f525a52 | 2985 | |
b1420f1c | 2986 | /* Other rcu_barrier() invocations can now safely proceed. */ |
ec9f5835 | 2987 | mutex_unlock(&rcu_state.barrier_mutex); |
d0ec774c | 2988 | } |
45975c7d | 2989 | EXPORT_SYMBOL_GPL(rcu_barrier); |
d0ec774c | 2990 | |
0aa04b05 PM |
2991 | /* |
2992 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
2993 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
2994 | * must hold the corresponding leaf rcu_node ->lock with interrrupts | |
2995 | * disabled. | |
2996 | */ | |
2997 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
2998 | { | |
2999 | long mask; | |
8d672fa6 | 3000 | long oldmask; |
0aa04b05 PM |
3001 | struct rcu_node *rnp = rnp_leaf; |
3002 | ||
8d672fa6 | 3003 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
962aff03 | 3004 | WARN_ON_ONCE(rnp->wait_blkd_tasks); |
0aa04b05 PM |
3005 | for (;;) { |
3006 | mask = rnp->grpmask; | |
3007 | rnp = rnp->parent; | |
3008 | if (rnp == NULL) | |
3009 | return; | |
6cf10081 | 3010 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
8d672fa6 | 3011 | oldmask = rnp->qsmaskinit; |
0aa04b05 | 3012 | rnp->qsmaskinit |= mask; |
67c583a7 | 3013 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
8d672fa6 PM |
3014 | if (oldmask) |
3015 | return; | |
0aa04b05 PM |
3016 | } |
3017 | } | |
3018 | ||
64db4cff | 3019 | /* |
27569620 | 3020 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 3021 | */ |
27569620 | 3022 | static void __init |
53b46303 | 3023 | rcu_boot_init_percpu_data(int cpu) |
64db4cff | 3024 | { |
da1df50d | 3025 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27569620 PM |
3026 | |
3027 | /* Set up local state, ensuring consistent view of global state. */ | |
bc75e999 | 3028 | rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu); |
4c5273bf | 3029 | WARN_ON_ONCE(rdp->dynticks_nesting != 1); |
dc5a4f29 | 3030 | WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp))); |
53b46303 | 3031 | rdp->rcu_ofl_gp_seq = rcu_state.gp_seq; |
57738942 | 3032 | rdp->rcu_ofl_gp_flags = RCU_GP_CLEANED; |
53b46303 | 3033 | rdp->rcu_onl_gp_seq = rcu_state.gp_seq; |
57738942 | 3034 | rdp->rcu_onl_gp_flags = RCU_GP_CLEANED; |
27569620 | 3035 | rdp->cpu = cpu; |
3fbfbf7a | 3036 | rcu_boot_init_nocb_percpu_data(rdp); |
27569620 PM |
3037 | } |
3038 | ||
3039 | /* | |
53b46303 PM |
3040 | * Invoked early in the CPU-online process, when pretty much all services |
3041 | * are available. The incoming CPU is not present. | |
3042 | * | |
3043 | * Initializes a CPU's per-CPU RCU data. Note that only one online or | |
ff3bb6f4 PM |
3044 | * offline event can be happening at a given time. Note also that we can |
3045 | * accept some slop in the rsp->gp_seq access due to the fact that this | |
e83e73f5 PM |
3046 | * CPU cannot possibly have any non-offloaded RCU callbacks in flight yet. |
3047 | * And any offloaded callbacks are being numbered elsewhere. | |
64db4cff | 3048 | */ |
53b46303 | 3049 | int rcutree_prepare_cpu(unsigned int cpu) |
64db4cff PM |
3050 | { |
3051 | unsigned long flags; | |
da1df50d | 3052 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
336a4f6c | 3053 | struct rcu_node *rnp = rcu_get_root(); |
64db4cff PM |
3054 | |
3055 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 3056 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 | 3057 | rdp->qlen_last_fqs_check = 0; |
53b46303 | 3058 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
64db4cff | 3059 | rdp->blimit = blimit; |
15fecf89 | 3060 | if (rcu_segcblist_empty(&rdp->cblist) && /* No early-boot CBs? */ |
e83e73f5 | 3061 | !rcu_segcblist_is_offloaded(&rdp->cblist)) |
15fecf89 | 3062 | rcu_segcblist_init(&rdp->cblist); /* Re-enable callbacks. */ |
4c5273bf | 3063 | rdp->dynticks_nesting = 1; /* CPU not up, no tearing. */ |
2625d469 | 3064 | rcu_dynticks_eqs_online(); |
67c583a7 | 3065 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
64db4cff | 3066 | |
0aa04b05 PM |
3067 | /* |
3068 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
3069 | * propagation up the rcu_node tree will happen at the beginning | |
3070 | * of the next grace period. | |
3071 | */ | |
64db4cff | 3072 | rnp = rdp->mynode; |
2a67e741 | 3073 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
b9585e94 | 3074 | rdp->beenonline = true; /* We have now been online. */ |
de30ad51 | 3075 | rdp->gp_seq = rnp->gp_seq; |
7a1d0f23 | 3076 | rdp->gp_seq_needed = rnp->gp_seq; |
5b74c458 | 3077 | rdp->cpu_no_qs.b.norm = true; |
97c668b8 | 3078 | rdp->core_needs_qs = false; |
9b9500da | 3079 | rdp->rcu_iw_pending = false; |
8aa670cd | 3080 | rdp->rcu_iw_gp_seq = rnp->gp_seq - 1; |
53b46303 | 3081 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuonl")); |
67c583a7 | 3082 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
4df83742 | 3083 | rcu_prepare_kthreads(cpu); |
ad368d15 | 3084 | rcu_spawn_cpu_nocb_kthread(cpu); |
4df83742 TG |
3085 | |
3086 | return 0; | |
3087 | } | |
3088 | ||
deb34f36 PM |
3089 | /* |
3090 | * Update RCU priority boot kthread affinity for CPU-hotplug changes. | |
3091 | */ | |
4df83742 TG |
3092 | static void rcutree_affinity_setting(unsigned int cpu, int outgoing) |
3093 | { | |
da1df50d | 3094 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
4df83742 TG |
3095 | |
3096 | rcu_boost_kthread_setaffinity(rdp->mynode, outgoing); | |
3097 | } | |
3098 | ||
deb34f36 PM |
3099 | /* |
3100 | * Near the end of the CPU-online process. Pretty much all services | |
3101 | * enabled, and the CPU is now very much alive. | |
3102 | */ | |
4df83742 TG |
3103 | int rcutree_online_cpu(unsigned int cpu) |
3104 | { | |
9b9500da PM |
3105 | unsigned long flags; |
3106 | struct rcu_data *rdp; | |
3107 | struct rcu_node *rnp; | |
9b9500da | 3108 | |
b97d23c5 PM |
3109 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3110 | rnp = rdp->mynode; | |
3111 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3112 | rnp->ffmask |= rdp->grpmask; | |
3113 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
9b9500da PM |
3114 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
3115 | return 0; /* Too early in boot for scheduler work. */ | |
3116 | sync_sched_exp_online_cleanup(cpu); | |
3117 | rcutree_affinity_setting(cpu, -1); | |
96926686 PM |
3118 | |
3119 | // Stop-machine done, so allow nohz_full to disable tick. | |
3120 | tick_dep_clear(TICK_DEP_BIT_RCU); | |
4df83742 TG |
3121 | return 0; |
3122 | } | |
3123 | ||
deb34f36 PM |
3124 | /* |
3125 | * Near the beginning of the process. The CPU is still very much alive | |
3126 | * with pretty much all services enabled. | |
3127 | */ | |
4df83742 TG |
3128 | int rcutree_offline_cpu(unsigned int cpu) |
3129 | { | |
9b9500da PM |
3130 | unsigned long flags; |
3131 | struct rcu_data *rdp; | |
3132 | struct rcu_node *rnp; | |
9b9500da | 3133 | |
b97d23c5 PM |
3134 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3135 | rnp = rdp->mynode; | |
3136 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3137 | rnp->ffmask &= ~rdp->grpmask; | |
3138 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
9b9500da | 3139 | |
4df83742 | 3140 | rcutree_affinity_setting(cpu, cpu); |
96926686 PM |
3141 | |
3142 | // nohz_full CPUs need the tick for stop-machine to work quickly | |
3143 | tick_dep_set(TICK_DEP_BIT_RCU); | |
4df83742 TG |
3144 | return 0; |
3145 | } | |
3146 | ||
f64c6013 PZ |
3147 | static DEFINE_PER_CPU(int, rcu_cpu_started); |
3148 | ||
7ec99de3 PM |
3149 | /* |
3150 | * Mark the specified CPU as being online so that subsequent grace periods | |
3151 | * (both expedited and normal) will wait on it. Note that this means that | |
3152 | * incoming CPUs are not allowed to use RCU read-side critical sections | |
3153 | * until this function is called. Failing to observe this restriction | |
3154 | * will result in lockdep splats. | |
deb34f36 PM |
3155 | * |
3156 | * Note that this function is special in that it is invoked directly | |
3157 | * from the incoming CPU rather than from the cpuhp_step mechanism. | |
3158 | * This is because this function must be invoked at a precise location. | |
7ec99de3 PM |
3159 | */ |
3160 | void rcu_cpu_starting(unsigned int cpu) | |
3161 | { | |
3162 | unsigned long flags; | |
3163 | unsigned long mask; | |
313517fc PM |
3164 | int nbits; |
3165 | unsigned long oldmask; | |
7ec99de3 PM |
3166 | struct rcu_data *rdp; |
3167 | struct rcu_node *rnp; | |
7ec99de3 | 3168 | |
f64c6013 PZ |
3169 | if (per_cpu(rcu_cpu_started, cpu)) |
3170 | return; | |
3171 | ||
3172 | per_cpu(rcu_cpu_started, cpu) = 1; | |
3173 | ||
b97d23c5 PM |
3174 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3175 | rnp = rdp->mynode; | |
3176 | mask = rdp->grpmask; | |
3177 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3178 | rnp->qsmaskinitnext |= mask; | |
3179 | oldmask = rnp->expmaskinitnext; | |
3180 | rnp->expmaskinitnext |= mask; | |
3181 | oldmask ^= rnp->expmaskinitnext; | |
3182 | nbits = bitmap_weight(&oldmask, BITS_PER_LONG); | |
3183 | /* Allow lockless access for expedited grace periods. */ | |
eb7a6653 | 3184 | smp_store_release(&rcu_state.ncpus, rcu_state.ncpus + nbits); /* ^^^ */ |
b97d23c5 | 3185 | rcu_gpnum_ovf(rnp, rdp); /* Offline-induced counter wrap? */ |
eb7a6653 PM |
3186 | rdp->rcu_onl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
3187 | rdp->rcu_onl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
b97d23c5 | 3188 | if (rnp->qsmask & mask) { /* RCU waiting on incoming CPU? */ |
516e5ae0 | 3189 | rcu_disable_urgency_upon_qs(rdp); |
b97d23c5 PM |
3190 | /* Report QS -after- changing ->qsmaskinitnext! */ |
3191 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); | |
3192 | } else { | |
3193 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
7ec99de3 | 3194 | } |
313517fc | 3195 | smp_mb(); /* Ensure RCU read-side usage follows above initialization. */ |
7ec99de3 PM |
3196 | } |
3197 | ||
27d50c7e TG |
3198 | #ifdef CONFIG_HOTPLUG_CPU |
3199 | /* | |
53b46303 PM |
3200 | * The outgoing function has no further need of RCU, so remove it from |
3201 | * the rcu_node tree's ->qsmaskinitnext bit masks. | |
3202 | * | |
3203 | * Note that this function is special in that it is invoked directly | |
3204 | * from the outgoing CPU rather than from the cpuhp_step mechanism. | |
3205 | * This is because this function must be invoked at a precise location. | |
27d50c7e | 3206 | */ |
53b46303 | 3207 | void rcu_report_dead(unsigned int cpu) |
27d50c7e TG |
3208 | { |
3209 | unsigned long flags; | |
3210 | unsigned long mask; | |
da1df50d | 3211 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27d50c7e TG |
3212 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
3213 | ||
49918a54 | 3214 | /* QS for any half-done expedited grace period. */ |
53b46303 | 3215 | preempt_disable(); |
63d4c8c9 | 3216 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); |
53b46303 PM |
3217 | preempt_enable(); |
3218 | rcu_preempt_deferred_qs(current); | |
3219 | ||
27d50c7e TG |
3220 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ |
3221 | mask = rdp->grpmask; | |
894d45bb | 3222 | raw_spin_lock(&rcu_state.ofl_lock); |
27d50c7e | 3223 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ |
53b46303 PM |
3224 | rdp->rcu_ofl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
3225 | rdp->rcu_ofl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
fece2776 PM |
3226 | if (rnp->qsmask & mask) { /* RCU waiting on outgoing CPU? */ |
3227 | /* Report quiescent state -before- changing ->qsmaskinitnext! */ | |
b50912d0 | 3228 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
fece2776 PM |
3229 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
3230 | } | |
27d50c7e | 3231 | rnp->qsmaskinitnext &= ~mask; |
710d60cb | 3232 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
894d45bb | 3233 | raw_spin_unlock(&rcu_state.ofl_lock); |
f64c6013 PZ |
3234 | |
3235 | per_cpu(rcu_cpu_started, cpu) = 0; | |
27d50c7e | 3236 | } |
a58163d8 | 3237 | |
53b46303 PM |
3238 | /* |
3239 | * The outgoing CPU has just passed through the dying-idle state, and we | |
3240 | * are being invoked from the CPU that was IPIed to continue the offline | |
3241 | * operation. Migrate the outgoing CPU's callbacks to the current CPU. | |
3242 | */ | |
3243 | void rcutree_migrate_callbacks(int cpu) | |
a58163d8 PM |
3244 | { |
3245 | unsigned long flags; | |
b1a2d79f | 3246 | struct rcu_data *my_rdp; |
c00045be | 3247 | struct rcu_node *my_rnp; |
da1df50d | 3248 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
ec4eacce | 3249 | bool needwake; |
a58163d8 | 3250 | |
ce5215c1 PM |
3251 | if (rcu_segcblist_is_offloaded(&rdp->cblist) || |
3252 | rcu_segcblist_empty(&rdp->cblist)) | |
95335c03 PM |
3253 | return; /* No callbacks to migrate. */ |
3254 | ||
b1a2d79f | 3255 | local_irq_save(flags); |
da1df50d | 3256 | my_rdp = this_cpu_ptr(&rcu_data); |
c00045be | 3257 | my_rnp = my_rdp->mynode; |
5d6742b3 | 3258 | rcu_nocb_lock(my_rdp); /* irqs already disabled. */ |
d1b222c6 | 3259 | WARN_ON_ONCE(!rcu_nocb_flush_bypass(my_rdp, NULL, jiffies)); |
c00045be | 3260 | raw_spin_lock_rcu_node(my_rnp); /* irqs already disabled. */ |
ec4eacce | 3261 | /* Leverage recent GPs and set GP for new callbacks. */ |
c00045be PM |
3262 | needwake = rcu_advance_cbs(my_rnp, rdp) || |
3263 | rcu_advance_cbs(my_rnp, my_rdp); | |
f2dbe4a5 | 3264 | rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist); |
23651d9b | 3265 | needwake = needwake || rcu_advance_cbs(my_rnp, my_rdp); |
c035280f | 3266 | rcu_segcblist_disable(&rdp->cblist); |
09efeeee PM |
3267 | WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) != |
3268 | !rcu_segcblist_n_cbs(&my_rdp->cblist)); | |
5d6742b3 PM |
3269 | if (rcu_segcblist_is_offloaded(&my_rdp->cblist)) { |
3270 | raw_spin_unlock_rcu_node(my_rnp); /* irqs remain disabled. */ | |
3271 | __call_rcu_nocb_wake(my_rdp, true, flags); | |
3272 | } else { | |
3273 | rcu_nocb_unlock(my_rdp); /* irqs remain disabled. */ | |
3274 | raw_spin_unlock_irqrestore_rcu_node(my_rnp, flags); | |
3275 | } | |
ec4eacce | 3276 | if (needwake) |
532c00c9 | 3277 | rcu_gp_kthread_wake(); |
5d6742b3 | 3278 | lockdep_assert_irqs_enabled(); |
a58163d8 PM |
3279 | WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 || |
3280 | !rcu_segcblist_empty(&rdp->cblist), | |
3281 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n", | |
3282 | cpu, rcu_segcblist_n_cbs(&rdp->cblist), | |
3283 | rcu_segcblist_first_cb(&rdp->cblist)); | |
3284 | } | |
27d50c7e TG |
3285 | #endif |
3286 | ||
deb34f36 PM |
3287 | /* |
3288 | * On non-huge systems, use expedited RCU grace periods to make suspend | |
3289 | * and hibernation run faster. | |
3290 | */ | |
d1d74d14 BP |
3291 | static int rcu_pm_notify(struct notifier_block *self, |
3292 | unsigned long action, void *hcpu) | |
3293 | { | |
3294 | switch (action) { | |
3295 | case PM_HIBERNATION_PREPARE: | |
3296 | case PM_SUSPEND_PREPARE: | |
e85e6a21 | 3297 | rcu_expedite_gp(); |
d1d74d14 BP |
3298 | break; |
3299 | case PM_POST_HIBERNATION: | |
3300 | case PM_POST_SUSPEND: | |
e85e6a21 | 3301 | rcu_unexpedite_gp(); |
d1d74d14 BP |
3302 | break; |
3303 | default: | |
3304 | break; | |
3305 | } | |
3306 | return NOTIFY_OK; | |
3307 | } | |
3308 | ||
b3dbec76 | 3309 | /* |
49918a54 | 3310 | * Spawn the kthreads that handle RCU's grace periods. |
b3dbec76 PM |
3311 | */ |
3312 | static int __init rcu_spawn_gp_kthread(void) | |
3313 | { | |
3314 | unsigned long flags; | |
a94844b2 | 3315 | int kthread_prio_in = kthread_prio; |
b3dbec76 | 3316 | struct rcu_node *rnp; |
a94844b2 | 3317 | struct sched_param sp; |
b3dbec76 PM |
3318 | struct task_struct *t; |
3319 | ||
a94844b2 | 3320 | /* Force priority into range. */ |
c7cd161e JFG |
3321 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 2 |
3322 | && IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)) | |
3323 | kthread_prio = 2; | |
3324 | else if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
a94844b2 PM |
3325 | kthread_prio = 1; |
3326 | else if (kthread_prio < 0) | |
3327 | kthread_prio = 0; | |
3328 | else if (kthread_prio > 99) | |
3329 | kthread_prio = 99; | |
c7cd161e | 3330 | |
a94844b2 PM |
3331 | if (kthread_prio != kthread_prio_in) |
3332 | pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n", | |
3333 | kthread_prio, kthread_prio_in); | |
3334 | ||
9386c0b7 | 3335 | rcu_scheduler_fully_active = 1; |
b97d23c5 | 3336 | t = kthread_create(rcu_gp_kthread, NULL, "%s", rcu_state.name); |
08543bda PM |
3337 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start grace-period kthread, OOM is now expected behavior\n", __func__)) |
3338 | return 0; | |
b97d23c5 PM |
3339 | if (kthread_prio) { |
3340 | sp.sched_priority = kthread_prio; | |
3341 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
b3dbec76 | 3342 | } |
b97d23c5 PM |
3343 | rnp = rcu_get_root(); |
3344 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3345 | rcu_state.gp_kthread = t; | |
b97d23c5 PM |
3346 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
3347 | wake_up_process(t); | |
35ce7f29 | 3348 | rcu_spawn_nocb_kthreads(); |
9386c0b7 | 3349 | rcu_spawn_boost_kthreads(); |
b3dbec76 PM |
3350 | return 0; |
3351 | } | |
3352 | early_initcall(rcu_spawn_gp_kthread); | |
3353 | ||
bbad9379 | 3354 | /* |
52d7e48b PM |
3355 | * This function is invoked towards the end of the scheduler's |
3356 | * initialization process. Before this is called, the idle task might | |
3357 | * contain synchronous grace-period primitives (during which time, this idle | |
3358 | * task is booting the system, and such primitives are no-ops). After this | |
3359 | * function is called, any synchronous grace-period primitives are run as | |
3360 | * expedited, with the requesting task driving the grace period forward. | |
900b1028 | 3361 | * A later core_initcall() rcu_set_runtime_mode() will switch to full |
52d7e48b | 3362 | * runtime RCU functionality. |
bbad9379 PM |
3363 | */ |
3364 | void rcu_scheduler_starting(void) | |
3365 | { | |
3366 | WARN_ON(num_online_cpus() != 1); | |
3367 | WARN_ON(nr_context_switches() > 0); | |
52d7e48b PM |
3368 | rcu_test_sync_prims(); |
3369 | rcu_scheduler_active = RCU_SCHEDULER_INIT; | |
3370 | rcu_test_sync_prims(); | |
bbad9379 PM |
3371 | } |
3372 | ||
64db4cff | 3373 | /* |
49918a54 | 3374 | * Helper function for rcu_init() that initializes the rcu_state structure. |
64db4cff | 3375 | */ |
b8bb1f63 | 3376 | static void __init rcu_init_one(void) |
64db4cff | 3377 | { |
cb007102 AG |
3378 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
3379 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
3380 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
3381 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
199977bf | 3382 | |
199977bf | 3383 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ |
64db4cff PM |
3384 | int cpustride = 1; |
3385 | int i; | |
3386 | int j; | |
3387 | struct rcu_node *rnp; | |
3388 | ||
05b84aec | 3389 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 3390 | |
3eaaaf6c PM |
3391 | /* Silence gcc 4.8 false positive about array index out of range. */ |
3392 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
3393 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 3394 | |
64db4cff PM |
3395 | /* Initialize the level-tracking arrays. */ |
3396 | ||
f885b7f2 | 3397 | for (i = 1; i < rcu_num_lvls; i++) |
eb7a6653 PM |
3398 | rcu_state.level[i] = |
3399 | rcu_state.level[i - 1] + num_rcu_lvl[i - 1]; | |
41f5c631 | 3400 | rcu_init_levelspread(levelspread, num_rcu_lvl); |
64db4cff PM |
3401 | |
3402 | /* Initialize the elements themselves, starting from the leaves. */ | |
3403 | ||
f885b7f2 | 3404 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 3405 | cpustride *= levelspread[i]; |
eb7a6653 | 3406 | rnp = rcu_state.level[i]; |
41f5c631 | 3407 | for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) { |
67c583a7 BF |
3408 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
3409 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 3410 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
3411 | raw_spin_lock_init(&rnp->fqslock); |
3412 | lockdep_set_class_and_name(&rnp->fqslock, | |
3413 | &rcu_fqs_class[i], fqs[i]); | |
eb7a6653 PM |
3414 | rnp->gp_seq = rcu_state.gp_seq; |
3415 | rnp->gp_seq_needed = rcu_state.gp_seq; | |
3416 | rnp->completedqs = rcu_state.gp_seq; | |
64db4cff PM |
3417 | rnp->qsmask = 0; |
3418 | rnp->qsmaskinit = 0; | |
3419 | rnp->grplo = j * cpustride; | |
3420 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
3421 | if (rnp->grphi >= nr_cpu_ids) |
3422 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
3423 | if (i == 0) { |
3424 | rnp->grpnum = 0; | |
3425 | rnp->grpmask = 0; | |
3426 | rnp->parent = NULL; | |
3427 | } else { | |
199977bf | 3428 | rnp->grpnum = j % levelspread[i - 1]; |
df63fa5b | 3429 | rnp->grpmask = BIT(rnp->grpnum); |
eb7a6653 | 3430 | rnp->parent = rcu_state.level[i - 1] + |
199977bf | 3431 | j / levelspread[i - 1]; |
64db4cff PM |
3432 | } |
3433 | rnp->level = i; | |
12f5f524 | 3434 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 3435 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
3436 | init_waitqueue_head(&rnp->exp_wq[0]); |
3437 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
3438 | init_waitqueue_head(&rnp->exp_wq[2]); |
3439 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 3440 | spin_lock_init(&rnp->exp_lock); |
64db4cff PM |
3441 | } |
3442 | } | |
0c34029a | 3443 | |
eb7a6653 PM |
3444 | init_swait_queue_head(&rcu_state.gp_wq); |
3445 | init_swait_queue_head(&rcu_state.expedited_wq); | |
aedf4ba9 | 3446 | rnp = rcu_first_leaf_node(); |
0c34029a | 3447 | for_each_possible_cpu(i) { |
4a90a068 | 3448 | while (i > rnp->grphi) |
0c34029a | 3449 | rnp++; |
da1df50d | 3450 | per_cpu_ptr(&rcu_data, i)->mynode = rnp; |
53b46303 | 3451 | rcu_boot_init_percpu_data(i); |
0c34029a | 3452 | } |
64db4cff PM |
3453 | } |
3454 | ||
f885b7f2 PM |
3455 | /* |
3456 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 3457 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
3458 | * the ->node array in the rcu_state structure. |
3459 | */ | |
3460 | static void __init rcu_init_geometry(void) | |
3461 | { | |
026ad283 | 3462 | ulong d; |
f885b7f2 | 3463 | int i; |
05b84aec | 3464 | int rcu_capacity[RCU_NUM_LVLS]; |
f885b7f2 | 3465 | |
026ad283 PM |
3466 | /* |
3467 | * Initialize any unspecified boot parameters. | |
3468 | * The default values of jiffies_till_first_fqs and | |
3469 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
3470 | * value, which is a function of HZ, then adding one for each | |
3471 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
3472 | */ | |
3473 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
3474 | if (jiffies_till_first_fqs == ULONG_MAX) | |
3475 | jiffies_till_first_fqs = d; | |
3476 | if (jiffies_till_next_fqs == ULONG_MAX) | |
3477 | jiffies_till_next_fqs = d; | |
6973032a | 3478 | adjust_jiffies_till_sched_qs(); |
026ad283 | 3479 | |
f885b7f2 | 3480 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 3481 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 3482 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 3483 | return; |
a7538352 | 3484 | pr_info("Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n", |
39479098 | 3485 | rcu_fanout_leaf, nr_cpu_ids); |
f885b7f2 | 3486 | |
f885b7f2 | 3487 | /* |
ee968ac6 PM |
3488 | * The boot-time rcu_fanout_leaf parameter must be at least two |
3489 | * and cannot exceed the number of bits in the rcu_node masks. | |
3490 | * Complain and fall back to the compile-time values if this | |
3491 | * limit is exceeded. | |
f885b7f2 | 3492 | */ |
ee968ac6 | 3493 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 3494 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 3495 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
3496 | WARN_ON(1); |
3497 | return; | |
3498 | } | |
3499 | ||
f885b7f2 PM |
3500 | /* |
3501 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 3502 | * with the given number of levels. |
f885b7f2 | 3503 | */ |
9618138b | 3504 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 3505 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 3506 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
3507 | |
3508 | /* | |
75cf15a4 | 3509 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 3510 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 3511 | */ |
ee968ac6 PM |
3512 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
3513 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
3514 | WARN_ON(1); | |
3515 | return; | |
3516 | } | |
f885b7f2 | 3517 | |
679f9858 | 3518 | /* Calculate the number of levels in the tree. */ |
9618138b | 3519 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 3520 | } |
9618138b | 3521 | rcu_num_lvls = i + 1; |
679f9858 | 3522 | |
f885b7f2 | 3523 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 3524 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 3525 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
3526 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
3527 | } | |
f885b7f2 PM |
3528 | |
3529 | /* Calculate the total number of rcu_node structures. */ | |
3530 | rcu_num_nodes = 0; | |
679f9858 | 3531 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 3532 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
3533 | } |
3534 | ||
a3dc2948 PM |
3535 | /* |
3536 | * Dump out the structure of the rcu_node combining tree associated | |
49918a54 | 3537 | * with the rcu_state structure. |
a3dc2948 | 3538 | */ |
b8bb1f63 | 3539 | static void __init rcu_dump_rcu_node_tree(void) |
a3dc2948 PM |
3540 | { |
3541 | int level = 0; | |
3542 | struct rcu_node *rnp; | |
3543 | ||
3544 | pr_info("rcu_node tree layout dump\n"); | |
3545 | pr_info(" "); | |
aedf4ba9 | 3546 | rcu_for_each_node_breadth_first(rnp) { |
a3dc2948 PM |
3547 | if (rnp->level != level) { |
3548 | pr_cont("\n"); | |
3549 | pr_info(" "); | |
3550 | level = rnp->level; | |
3551 | } | |
3552 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
3553 | } | |
3554 | pr_cont("\n"); | |
3555 | } | |
3556 | ||
ad7c946b | 3557 | struct workqueue_struct *rcu_gp_wq; |
25f3d7ef | 3558 | struct workqueue_struct *rcu_par_gp_wq; |
ad7c946b | 3559 | |
9f680ab4 | 3560 | void __init rcu_init(void) |
64db4cff | 3561 | { |
017c4261 | 3562 | int cpu; |
9f680ab4 | 3563 | |
47627678 PM |
3564 | rcu_early_boot_tests(); |
3565 | ||
f41d911f | 3566 | rcu_bootup_announce(); |
f885b7f2 | 3567 | rcu_init_geometry(); |
b8bb1f63 | 3568 | rcu_init_one(); |
a3dc2948 | 3569 | if (dump_tree) |
b8bb1f63 | 3570 | rcu_dump_rcu_node_tree(); |
48d07c04 SAS |
3571 | if (use_softirq) |
3572 | open_softirq(RCU_SOFTIRQ, rcu_core_si); | |
9f680ab4 PM |
3573 | |
3574 | /* | |
3575 | * We don't need protection against CPU-hotplug here because | |
3576 | * this is called early in boot, before either interrupts | |
3577 | * or the scheduler are operational. | |
3578 | */ | |
d1d74d14 | 3579 | pm_notifier(rcu_pm_notify, 0); |
7ec99de3 | 3580 | for_each_online_cpu(cpu) { |
4df83742 | 3581 | rcutree_prepare_cpu(cpu); |
7ec99de3 | 3582 | rcu_cpu_starting(cpu); |
9b9500da | 3583 | rcutree_online_cpu(cpu); |
7ec99de3 | 3584 | } |
ad7c946b PM |
3585 | |
3586 | /* Create workqueue for expedited GPs and for Tree SRCU. */ | |
3587 | rcu_gp_wq = alloc_workqueue("rcu_gp", WQ_MEM_RECLAIM, 0); | |
3588 | WARN_ON(!rcu_gp_wq); | |
25f3d7ef PM |
3589 | rcu_par_gp_wq = alloc_workqueue("rcu_par_gp", WQ_MEM_RECLAIM, 0); |
3590 | WARN_ON(!rcu_par_gp_wq); | |
e0fcba9a | 3591 | srcu_init(); |
64db4cff PM |
3592 | } |
3593 | ||
10462d6f | 3594 | #include "tree_stall.h" |
3549c2bc | 3595 | #include "tree_exp.h" |
4102adab | 3596 | #include "tree_plugin.h" |