Commit | Line | Data |
---|---|---|
64db4cff PM |
1 | /* |
2 | * Read-Copy Update mechanism for mutual exclusion | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
87de1cfd PM |
15 | * along with this program; if not, you can access it online at |
16 | * http://www.gnu.org/licenses/gpl-2.0.html. | |
64db4cff PM |
17 | * |
18 | * Copyright IBM Corporation, 2008 | |
19 | * | |
20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | |
21 | * Manfred Spraul <manfred@colorfullife.com> | |
22 | * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version | |
23 | * | |
24 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> | |
25 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. | |
26 | * | |
27 | * For detailed explanation of Read-Copy Update mechanism see - | |
a71fca58 | 28 | * Documentation/RCU |
64db4cff | 29 | */ |
a7538352 JP |
30 | |
31 | #define pr_fmt(fmt) "rcu: " fmt | |
32 | ||
64db4cff PM |
33 | #include <linux/types.h> |
34 | #include <linux/kernel.h> | |
35 | #include <linux/init.h> | |
36 | #include <linux/spinlock.h> | |
37 | #include <linux/smp.h> | |
f9411ebe | 38 | #include <linux/rcupdate_wait.h> |
64db4cff PM |
39 | #include <linux/interrupt.h> |
40 | #include <linux/sched.h> | |
b17b0153 | 41 | #include <linux/sched/debug.h> |
c1dc0b9c | 42 | #include <linux/nmi.h> |
8826f3b0 | 43 | #include <linux/atomic.h> |
64db4cff | 44 | #include <linux/bitops.h> |
9984de1a | 45 | #include <linux/export.h> |
64db4cff PM |
46 | #include <linux/completion.h> |
47 | #include <linux/moduleparam.h> | |
48 | #include <linux/percpu.h> | |
49 | #include <linux/notifier.h> | |
50 | #include <linux/cpu.h> | |
51 | #include <linux/mutex.h> | |
52 | #include <linux/time.h> | |
bbad9379 | 53 | #include <linux/kernel_stat.h> |
a26ac245 PM |
54 | #include <linux/wait.h> |
55 | #include <linux/kthread.h> | |
ae7e81c0 | 56 | #include <uapi/linux/sched/types.h> |
268bb0ce | 57 | #include <linux/prefetch.h> |
3d3b7db0 PM |
58 | #include <linux/delay.h> |
59 | #include <linux/stop_machine.h> | |
661a85dc | 60 | #include <linux/random.h> |
af658dca | 61 | #include <linux/trace_events.h> |
d1d74d14 | 62 | #include <linux/suspend.h> |
a278d471 | 63 | #include <linux/ftrace.h> |
64db4cff | 64 | |
4102adab | 65 | #include "tree.h" |
29c00b4a | 66 | #include "rcu.h" |
9f77da9f | 67 | |
4102adab PM |
68 | #ifdef MODULE_PARAM_PREFIX |
69 | #undef MODULE_PARAM_PREFIX | |
70 | #endif | |
71 | #define MODULE_PARAM_PREFIX "rcutree." | |
72 | ||
64db4cff PM |
73 | /* Data structures. */ |
74 | ||
358be2d3 PM |
75 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, rcu_data); |
76 | struct rcu_state rcu_state = { | |
77 | .level = { &rcu_state.node[0] }, | |
78 | .rda = &rcu_data, | |
358be2d3 PM |
79 | .gp_state = RCU_GP_IDLE, |
80 | .gp_seq = (0UL - 300UL) << RCU_SEQ_CTR_SHIFT, | |
81 | .barrier_mutex = __MUTEX_INITIALIZER(rcu_state.barrier_mutex), | |
82 | .name = RCU_NAME, | |
83 | .abbr = RCU_ABBR, | |
84 | .exp_mutex = __MUTEX_INITIALIZER(rcu_state.exp_mutex), | |
85 | .exp_wake_mutex = __MUTEX_INITIALIZER(rcu_state.exp_wake_mutex), | |
86 | .ofl_lock = __SPIN_LOCK_UNLOCKED(rcu_state.ofl_lock), | |
87 | }; | |
b1f77b05 | 88 | |
45975c7d PM |
89 | static struct rcu_state *const rcu_state_p = &rcu_state; |
90 | static struct rcu_data __percpu *const rcu_data_p = &rcu_data; | |
6ce75a23 | 91 | LIST_HEAD(rcu_struct_flavors); |
27f4d280 | 92 | |
a3dc2948 PM |
93 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
94 | static bool dump_tree; | |
95 | module_param(dump_tree, bool, 0444); | |
7fa27001 PM |
96 | /* Control rcu_node-tree auto-balancing at boot time. */ |
97 | static bool rcu_fanout_exact; | |
98 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
99 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
100 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 101 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 102 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 | 103 | /* Number of rcu_nodes at specified level. */ |
e95d68d2 | 104 | int num_rcu_lvl[] = NUM_RCU_LVL_INIT; |
f885b7f2 | 105 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
088e9d25 DBO |
106 | /* panic() on RCU Stall sysctl. */ |
107 | int sysctl_panic_on_rcu_stall __read_mostly; | |
f885b7f2 | 108 | |
b0d30417 | 109 | /* |
52d7e48b PM |
110 | * The rcu_scheduler_active variable is initialized to the value |
111 | * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the | |
112 | * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE, | |
113 | * RCU can assume that there is but one task, allowing RCU to (for example) | |
0d95092c | 114 | * optimize synchronize_rcu() to a simple barrier(). When this variable |
52d7e48b PM |
115 | * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required |
116 | * to detect real grace periods. This variable is also used to suppress | |
117 | * boot-time false positives from lockdep-RCU error checking. Finally, it | |
118 | * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU | |
119 | * is fully initialized, including all of its kthreads having been spawned. | |
b0d30417 | 120 | */ |
bbad9379 PM |
121 | int rcu_scheduler_active __read_mostly; |
122 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
123 | ||
b0d30417 PM |
124 | /* |
125 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
126 | * during the early_initcall() processing, which is after the scheduler | |
127 | * is capable of creating new tasks. So RCU processing (for example, | |
128 | * creating tasks for RCU priority boosting) must be delayed until after | |
129 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
130 | * currently delay invocation of any RCU callbacks until after this point. | |
131 | * | |
132 | * It might later prove better for people registering RCU callbacks during | |
133 | * early boot to take responsibility for these callbacks, but one step at | |
134 | * a time. | |
135 | */ | |
136 | static int rcu_scheduler_fully_active __read_mostly; | |
137 | ||
ec2c2976 PM |
138 | static void |
139 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, | |
140 | struct rcu_node *rnp, unsigned long gps, unsigned long flags); | |
0aa04b05 PM |
141 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
142 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 143 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 PM |
144 | static void invoke_rcu_core(void); |
145 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); | |
2bbfc25b | 146 | static void rcu_report_exp_rdp(struct rcu_state *rsp, struct rcu_data *rdp); |
3549c2bc | 147 | static void sync_sched_exp_online_cleanup(int cpu); |
a26ac245 | 148 | |
a94844b2 | 149 | /* rcuc/rcub kthread realtime priority */ |
26730f55 | 150 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; |
a94844b2 PM |
151 | module_param(kthread_prio, int, 0644); |
152 | ||
8d7dc928 | 153 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd | 154 | |
90040c9e PM |
155 | static int gp_preinit_delay; |
156 | module_param(gp_preinit_delay, int, 0444); | |
157 | static int gp_init_delay; | |
158 | module_param(gp_init_delay, int, 0444); | |
159 | static int gp_cleanup_delay; | |
160 | module_param(gp_cleanup_delay, int, 0444); | |
0f41c0dd | 161 | |
4cf439a2 | 162 | /* Retrieve RCU kthreads priority for rcutorture */ |
4babd855 JFG |
163 | int rcu_get_gp_kthreads_prio(void) |
164 | { | |
165 | return kthread_prio; | |
166 | } | |
167 | EXPORT_SYMBOL_GPL(rcu_get_gp_kthreads_prio); | |
168 | ||
eab128e8 PM |
169 | /* |
170 | * Number of grace periods between delays, normalized by the duration of | |
bfd090be | 171 | * the delay. The longer the delay, the more the grace periods between |
eab128e8 PM |
172 | * each delay. The reason for this normalization is that it means that, |
173 | * for non-zero delays, the overall slowdown of grace periods is constant | |
174 | * regardless of the duration of the delay. This arrangement balances | |
175 | * the need for long delays to increase some race probabilities with the | |
176 | * need for fast grace periods to increase other race probabilities. | |
177 | */ | |
178 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */ | |
37745d28 | 179 | |
0aa04b05 PM |
180 | /* |
181 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
182 | * This will not be stable unless the rcu_node structure's ->lock is | |
183 | * held, but the bit corresponding to the current CPU will be stable | |
184 | * in most contexts. | |
185 | */ | |
186 | unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) | |
187 | { | |
7d0ae808 | 188 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
189 | } |
190 | ||
fc2219d4 | 191 | /* |
7d0ae808 | 192 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
193 | * permit this function to be invoked without holding the root rcu_node |
194 | * structure's ->lock, but of course results can be subject to change. | |
195 | */ | |
196 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
197 | { | |
ba04107f | 198 | return rcu_seq_state(rcu_seq_current(&rsp->gp_seq)); |
fc2219d4 PM |
199 | } |
200 | ||
d28139c4 PM |
201 | void rcu_softirq_qs(void) |
202 | { | |
45975c7d | 203 | rcu_qs(); |
d28139c4 PM |
204 | rcu_preempt_deferred_qs(current); |
205 | } | |
206 | ||
b8c17e66 PM |
207 | /* |
208 | * Steal a bit from the bottom of ->dynticks for idle entry/exit | |
209 | * control. Initially this is for TLB flushing. | |
210 | */ | |
211 | #define RCU_DYNTICK_CTRL_MASK 0x1 | |
212 | #define RCU_DYNTICK_CTRL_CTR (RCU_DYNTICK_CTRL_MASK + 1) | |
213 | #ifndef rcu_eqs_special_exit | |
214 | #define rcu_eqs_special_exit() do { } while (0) | |
215 | #endif | |
4a81e832 PM |
216 | |
217 | static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { | |
51a1fd30 | 218 | .dynticks_nesting = 1, |
58721f5d | 219 | .dynticks_nmi_nesting = DYNTICK_IRQ_NONIDLE, |
b8c17e66 | 220 | .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR), |
4a81e832 PM |
221 | }; |
222 | ||
2625d469 PM |
223 | /* |
224 | * Record entry into an extended quiescent state. This is only to be | |
225 | * called when not already in an extended quiescent state. | |
226 | */ | |
227 | static void rcu_dynticks_eqs_enter(void) | |
228 | { | |
229 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 230 | int seq; |
2625d469 PM |
231 | |
232 | /* | |
b8c17e66 | 233 | * CPUs seeing atomic_add_return() must see prior RCU read-side |
2625d469 PM |
234 | * critical sections, and we also must force ordering with the |
235 | * next idle sojourn. | |
236 | */ | |
b8c17e66 PM |
237 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
238 | /* Better be in an extended quiescent state! */ | |
239 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
240 | (seq & RCU_DYNTICK_CTRL_CTR)); | |
241 | /* Better not have special action (TLB flush) pending! */ | |
242 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
243 | (seq & RCU_DYNTICK_CTRL_MASK)); | |
2625d469 PM |
244 | } |
245 | ||
246 | /* | |
247 | * Record exit from an extended quiescent state. This is only to be | |
248 | * called from an extended quiescent state. | |
249 | */ | |
250 | static void rcu_dynticks_eqs_exit(void) | |
251 | { | |
252 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
b8c17e66 | 253 | int seq; |
2625d469 PM |
254 | |
255 | /* | |
b8c17e66 | 256 | * CPUs seeing atomic_add_return() must see prior idle sojourns, |
2625d469 PM |
257 | * and we also must force ordering with the next RCU read-side |
258 | * critical section. | |
259 | */ | |
b8c17e66 PM |
260 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
261 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
262 | !(seq & RCU_DYNTICK_CTRL_CTR)); | |
263 | if (seq & RCU_DYNTICK_CTRL_MASK) { | |
264 | atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdtp->dynticks); | |
265 | smp_mb__after_atomic(); /* _exit after clearing mask. */ | |
266 | /* Prefer duplicate flushes to losing a flush. */ | |
267 | rcu_eqs_special_exit(); | |
268 | } | |
2625d469 PM |
269 | } |
270 | ||
271 | /* | |
272 | * Reset the current CPU's ->dynticks counter to indicate that the | |
273 | * newly onlined CPU is no longer in an extended quiescent state. | |
274 | * This will either leave the counter unchanged, or increment it | |
275 | * to the next non-quiescent value. | |
276 | * | |
277 | * The non-atomic test/increment sequence works because the upper bits | |
278 | * of the ->dynticks counter are manipulated only by the corresponding CPU, | |
279 | * or when the corresponding CPU is offline. | |
280 | */ | |
281 | static void rcu_dynticks_eqs_online(void) | |
282 | { | |
283 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
284 | ||
b8c17e66 | 285 | if (atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR) |
2625d469 | 286 | return; |
b8c17e66 | 287 | atomic_add(RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
2625d469 PM |
288 | } |
289 | ||
02a5c550 PM |
290 | /* |
291 | * Is the current CPU in an extended quiescent state? | |
292 | * | |
293 | * No ordering, as we are sampling CPU-local information. | |
294 | */ | |
295 | bool rcu_dynticks_curr_cpu_in_eqs(void) | |
296 | { | |
297 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
298 | ||
b8c17e66 | 299 | return !(atomic_read(&rdtp->dynticks) & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
300 | } |
301 | ||
8b2f63ab PM |
302 | /* |
303 | * Snapshot the ->dynticks counter with full ordering so as to allow | |
304 | * stable comparison of this counter with past and future snapshots. | |
305 | */ | |
02a5c550 | 306 | int rcu_dynticks_snap(struct rcu_dynticks *rdtp) |
8b2f63ab PM |
307 | { |
308 | int snap = atomic_add_return(0, &rdtp->dynticks); | |
309 | ||
b8c17e66 | 310 | return snap & ~RCU_DYNTICK_CTRL_MASK; |
8b2f63ab PM |
311 | } |
312 | ||
02a5c550 PM |
313 | /* |
314 | * Return true if the snapshot returned from rcu_dynticks_snap() | |
315 | * indicates that RCU is in an extended quiescent state. | |
316 | */ | |
317 | static bool rcu_dynticks_in_eqs(int snap) | |
318 | { | |
b8c17e66 | 319 | return !(snap & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
320 | } |
321 | ||
322 | /* | |
323 | * Return true if the CPU corresponding to the specified rcu_dynticks | |
324 | * structure has spent some time in an extended quiescent state since | |
325 | * rcu_dynticks_snap() returned the specified snapshot. | |
326 | */ | |
327 | static bool rcu_dynticks_in_eqs_since(struct rcu_dynticks *rdtp, int snap) | |
328 | { | |
329 | return snap != rcu_dynticks_snap(rdtp); | |
330 | } | |
331 | ||
b8c17e66 PM |
332 | /* |
333 | * Set the special (bottom) bit of the specified CPU so that it | |
334 | * will take special action (such as flushing its TLB) on the | |
335 | * next exit from an extended quiescent state. Returns true if | |
336 | * the bit was successfully set, or false if the CPU was not in | |
337 | * an extended quiescent state. | |
338 | */ | |
339 | bool rcu_eqs_special_set(int cpu) | |
340 | { | |
341 | int old; | |
342 | int new; | |
343 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); | |
344 | ||
345 | do { | |
346 | old = atomic_read(&rdtp->dynticks); | |
347 | if (old & RCU_DYNTICK_CTRL_CTR) | |
348 | return false; | |
349 | new = old | RCU_DYNTICK_CTRL_MASK; | |
350 | } while (atomic_cmpxchg(&rdtp->dynticks, old, new) != old); | |
351 | return true; | |
6563de9d | 352 | } |
5cd37193 | 353 | |
4a81e832 PM |
354 | /* |
355 | * Let the RCU core know that this CPU has gone through the scheduler, | |
356 | * which is a quiescent state. This is called when the need for a | |
357 | * quiescent state is urgent, so we burn an atomic operation and full | |
358 | * memory barriers to let the RCU core know about it, regardless of what | |
359 | * this CPU might (or might not) do in the near future. | |
360 | * | |
0f9be8ca | 361 | * We inform the RCU core by emulating a zero-duration dyntick-idle period. |
46a5d164 | 362 | * |
3b57a399 | 363 | * The caller must have disabled interrupts and must not be idle. |
4a81e832 PM |
364 | */ |
365 | static void rcu_momentary_dyntick_idle(void) | |
366 | { | |
3b57a399 PM |
367 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
368 | int special; | |
369 | ||
0f9be8ca | 370 | raw_cpu_write(rcu_dynticks.rcu_need_heavy_qs, false); |
3b57a399 PM |
371 | special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR, &rdtp->dynticks); |
372 | /* It is illegal to call this from idle state. */ | |
373 | WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR)); | |
3e310098 | 374 | rcu_preempt_deferred_qs(current); |
4a81e832 PM |
375 | } |
376 | ||
45975c7d PM |
377 | /** |
378 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle | |
379 | * | |
380 | * If the current CPU is idle or running at a first-level (not nested) | |
381 | * interrupt from idle, return true. The caller must have at least | |
382 | * disabled preemption. | |
25502a6c | 383 | */ |
45975c7d | 384 | static int rcu_is_cpu_rrupt_from_idle(void) |
25502a6c | 385 | { |
45975c7d PM |
386 | return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 0 && |
387 | __this_cpu_read(rcu_dynticks.dynticks_nmi_nesting) <= 1; | |
25502a6c PM |
388 | } |
389 | ||
5cd37193 | 390 | /* |
1925d196 | 391 | * Register a quiescent state for all RCU flavors. If there is an |
5cd37193 PM |
392 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight |
393 | * dyntick-idle quiescent state visible to other CPUs (but only for those | |
1925d196 | 394 | * RCU flavors in desperate need of a quiescent state, which will normally |
5cd37193 PM |
395 | * be none of them). Either way, do a lightweight quiescent state for |
396 | * all RCU flavors. | |
bb73c52b BF |
397 | * |
398 | * The barrier() calls are redundant in the common case when this is | |
399 | * called externally, but just in case this is called from within this | |
400 | * file. | |
401 | * | |
5cd37193 PM |
402 | */ |
403 | void rcu_all_qs(void) | |
404 | { | |
46a5d164 PM |
405 | unsigned long flags; |
406 | ||
9226b10d PM |
407 | if (!raw_cpu_read(rcu_dynticks.rcu_urgent_qs)) |
408 | return; | |
409 | preempt_disable(); | |
410 | /* Load rcu_urgent_qs before other flags. */ | |
411 | if (!smp_load_acquire(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs))) { | |
412 | preempt_enable(); | |
413 | return; | |
414 | } | |
415 | this_cpu_write(rcu_dynticks.rcu_urgent_qs, false); | |
bb73c52b | 416 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
0f9be8ca | 417 | if (unlikely(raw_cpu_read(rcu_dynticks.rcu_need_heavy_qs))) { |
46a5d164 | 418 | local_irq_save(flags); |
5cd37193 | 419 | rcu_momentary_dyntick_idle(); |
46a5d164 PM |
420 | local_irq_restore(flags); |
421 | } | |
45975c7d PM |
422 | if (unlikely(raw_cpu_read(rcu_data.cpu_no_qs.b.exp))) |
423 | rcu_qs(); | |
9577df9a | 424 | this_cpu_inc(rcu_dynticks.rcu_qs_ctr); |
bb73c52b | 425 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
9226b10d | 426 | preempt_enable(); |
5cd37193 PM |
427 | } |
428 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
429 | ||
17c7798b PM |
430 | #define DEFAULT_RCU_BLIMIT 10 /* Maximum callbacks per rcu_do_batch. */ |
431 | static long blimit = DEFAULT_RCU_BLIMIT; | |
432 | #define DEFAULT_RCU_QHIMARK 10000 /* If this many pending, ignore blimit. */ | |
433 | static long qhimark = DEFAULT_RCU_QHIMARK; | |
434 | #define DEFAULT_RCU_QLOMARK 100 /* Once only this many pending, use blimit. */ | |
435 | static long qlowmark = DEFAULT_RCU_QLOMARK; | |
64db4cff | 436 | |
878d7439 ED |
437 | module_param(blimit, long, 0444); |
438 | module_param(qhimark, long, 0444); | |
439 | module_param(qlowmark, long, 0444); | |
3d76c082 | 440 | |
026ad283 PM |
441 | static ulong jiffies_till_first_fqs = ULONG_MAX; |
442 | static ulong jiffies_till_next_fqs = ULONG_MAX; | |
8c7c4829 | 443 | static bool rcu_kick_kthreads; |
d40011f6 | 444 | |
67abb96c BP |
445 | static int param_set_first_fqs_jiffies(const char *val, const struct kernel_param *kp) |
446 | { | |
447 | ulong j; | |
448 | int ret = kstrtoul(val, 0, &j); | |
449 | ||
450 | if (!ret) | |
451 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : j); | |
452 | return ret; | |
453 | } | |
454 | ||
455 | static int param_set_next_fqs_jiffies(const char *val, const struct kernel_param *kp) | |
456 | { | |
457 | ulong j; | |
458 | int ret = kstrtoul(val, 0, &j); | |
459 | ||
460 | if (!ret) | |
461 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : (j ?: 1)); | |
462 | return ret; | |
463 | } | |
464 | ||
465 | static struct kernel_param_ops first_fqs_jiffies_ops = { | |
466 | .set = param_set_first_fqs_jiffies, | |
467 | .get = param_get_ulong, | |
468 | }; | |
469 | ||
470 | static struct kernel_param_ops next_fqs_jiffies_ops = { | |
471 | .set = param_set_next_fqs_jiffies, | |
472 | .get = param_get_ulong, | |
473 | }; | |
474 | ||
475 | module_param_cb(jiffies_till_first_fqs, &first_fqs_jiffies_ops, &jiffies_till_first_fqs, 0644); | |
476 | module_param_cb(jiffies_till_next_fqs, &next_fqs_jiffies_ops, &jiffies_till_next_fqs, 0644); | |
8c7c4829 | 477 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 478 | |
4a81e832 PM |
479 | /* |
480 | * How long the grace period must be before we start recruiting | |
481 | * quiescent-state help from rcu_note_context_switch(). | |
482 | */ | |
f79c3ad6 PM |
483 | static ulong jiffies_till_sched_qs = HZ / 10; |
484 | module_param(jiffies_till_sched_qs, ulong, 0444); | |
4a81e832 | 485 | |
fe5ac724 | 486 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *rsp)); |
4cdfc175 | 487 | static void force_quiescent_state(struct rcu_state *rsp); |
e3950ecd | 488 | static int rcu_pending(void); |
64db4cff PM |
489 | |
490 | /* | |
17ef2fe9 | 491 | * Return the number of RCU GPs completed thus far for debug & stats. |
64db4cff | 492 | */ |
17ef2fe9 | 493 | unsigned long rcu_get_gp_seq(void) |
917963d0 | 494 | { |
d7219312 | 495 | return READ_ONCE(rcu_state_p->gp_seq); |
917963d0 | 496 | } |
17ef2fe9 | 497 | EXPORT_SYMBOL_GPL(rcu_get_gp_seq); |
917963d0 PM |
498 | |
499 | /* | |
17ef2fe9 | 500 | * Return the number of RCU-sched GPs completed thus far for debug & stats. |
64db4cff | 501 | */ |
17ef2fe9 | 502 | unsigned long rcu_sched_get_gp_seq(void) |
917963d0 | 503 | { |
45975c7d | 504 | return rcu_get_gp_seq(); |
917963d0 | 505 | } |
17ef2fe9 | 506 | EXPORT_SYMBOL_GPL(rcu_sched_get_gp_seq); |
917963d0 PM |
507 | |
508 | /* | |
82fcecfa PM |
509 | * Return the number of RCU GPs completed thus far for debug & stats. |
510 | * This is a transitional API and will soon be removed. | |
917963d0 | 511 | */ |
17ef2fe9 | 512 | unsigned long rcu_bh_get_gp_seq(void) |
917963d0 | 513 | { |
65cfe358 | 514 | return READ_ONCE(rcu_state_p->gp_seq); |
917963d0 | 515 | } |
17ef2fe9 | 516 | EXPORT_SYMBOL_GPL(rcu_bh_get_gp_seq); |
64db4cff | 517 | |
291783b8 PM |
518 | /* |
519 | * Return the number of RCU expedited batches completed thus far for | |
520 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
521 | * numbers mean idle. The value returned will thus be roughly double | |
522 | * the cumulative batches since boot. | |
523 | */ | |
524 | unsigned long rcu_exp_batches_completed(void) | |
525 | { | |
526 | return rcu_state_p->expedited_sequence; | |
527 | } | |
528 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
529 | ||
530 | /* | |
531 | * Return the number of RCU-sched expedited batches completed thus far | |
532 | * for debug & stats. Similar to rcu_exp_batches_completed(). | |
533 | */ | |
534 | unsigned long rcu_exp_batches_completed_sched(void) | |
535 | { | |
45975c7d | 536 | return rcu_state.expedited_sequence; |
291783b8 PM |
537 | } |
538 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed_sched); | |
539 | ||
a381d757 ACB |
540 | /* |
541 | * Force a quiescent state. | |
542 | */ | |
543 | void rcu_force_quiescent_state(void) | |
544 | { | |
e534165b | 545 | force_quiescent_state(rcu_state_p); |
a381d757 ACB |
546 | } |
547 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
548 | ||
bf66f18e PM |
549 | /* |
550 | * Force a quiescent state for RCU BH. | |
551 | */ | |
552 | void rcu_bh_force_quiescent_state(void) | |
553 | { | |
65cfe358 | 554 | force_quiescent_state(rcu_state_p); |
bf66f18e PM |
555 | } |
556 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
557 | ||
e7580f33 PM |
558 | /* |
559 | * Force a quiescent state for RCU-sched. | |
560 | */ | |
561 | void rcu_sched_force_quiescent_state(void) | |
562 | { | |
45975c7d | 563 | rcu_force_quiescent_state(); |
e7580f33 PM |
564 | } |
565 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
566 | ||
afea227f PM |
567 | /* |
568 | * Show the state of the grace-period kthreads. | |
569 | */ | |
570 | void show_rcu_gp_kthreads(void) | |
571 | { | |
47199a08 PM |
572 | int cpu; |
573 | struct rcu_data *rdp; | |
574 | struct rcu_node *rnp; | |
afea227f PM |
575 | struct rcu_state *rsp; |
576 | ||
577 | for_each_rcu_flavor(rsp) { | |
578 | pr_info("%s: wait state: %d ->state: %#lx\n", | |
579 | rsp->name, rsp->gp_state, rsp->gp_kthread->state); | |
47199a08 PM |
580 | rcu_for_each_node_breadth_first(rsp, rnp) { |
581 | if (ULONG_CMP_GE(rsp->gp_seq, rnp->gp_seq_needed)) | |
582 | continue; | |
583 | pr_info("\trcu_node %d:%d ->gp_seq %lu ->gp_seq_needed %lu\n", | |
584 | rnp->grplo, rnp->grphi, rnp->gp_seq, | |
585 | rnp->gp_seq_needed); | |
586 | if (!rcu_is_leaf_node(rnp)) | |
587 | continue; | |
588 | for_each_leaf_node_possible_cpu(rnp, cpu) { | |
589 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
590 | if (rdp->gpwrap || | |
591 | ULONG_CMP_GE(rsp->gp_seq, | |
592 | rdp->gp_seq_needed)) | |
593 | continue; | |
594 | pr_info("\tcpu %d ->gp_seq_needed %lu\n", | |
595 | cpu, rdp->gp_seq_needed); | |
596 | } | |
597 | } | |
afea227f PM |
598 | /* sched_show_task(rsp->gp_kthread); */ |
599 | } | |
600 | } | |
601 | EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads); | |
602 | ||
ad0dc7f9 PM |
603 | /* |
604 | * Send along grace-period-related data for rcutorture diagnostics. | |
605 | */ | |
606 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
aebc8264 | 607 | unsigned long *gp_seq) |
ad0dc7f9 PM |
608 | { |
609 | struct rcu_state *rsp = NULL; | |
610 | ||
611 | switch (test_type) { | |
612 | case RCU_FLAVOR: | |
ad0dc7f9 | 613 | case RCU_BH_FLAVOR: |
ad0dc7f9 | 614 | case RCU_SCHED_FLAVOR: |
45975c7d | 615 | rsp = rcu_state_p; |
ad0dc7f9 PM |
616 | break; |
617 | default: | |
618 | break; | |
619 | } | |
7f6733c3 | 620 | if (rsp == NULL) |
ad0dc7f9 | 621 | return; |
7f6733c3 | 622 | *flags = READ_ONCE(rsp->gp_flags); |
aebc8264 | 623 | *gp_seq = rcu_seq_current(&rsp->gp_seq); |
ad0dc7f9 PM |
624 | } |
625 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
626 | ||
365187fb PM |
627 | /* |
628 | * Return the root node of the specified rcu_state structure. | |
629 | */ | |
630 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
631 | { | |
632 | return &rsp->node[0]; | |
633 | } | |
634 | ||
9b2e4f18 | 635 | /* |
215bba9f PM |
636 | * Enter an RCU extended quiescent state, which can be either the |
637 | * idle loop or adaptive-tickless usermode execution. | |
9b2e4f18 | 638 | * |
215bba9f PM |
639 | * We crowbar the ->dynticks_nmi_nesting field to zero to allow for |
640 | * the possibility of usermode upcalls having messed up our count | |
641 | * of interrupt nesting level during the prior busy period. | |
9b2e4f18 | 642 | */ |
215bba9f | 643 | static void rcu_eqs_enter(bool user) |
9b2e4f18 | 644 | { |
96d3fd0d PM |
645 | struct rcu_state *rsp; |
646 | struct rcu_data *rdp; | |
215bba9f | 647 | struct rcu_dynticks *rdtp; |
96d3fd0d | 648 | |
215bba9f | 649 | rdtp = this_cpu_ptr(&rcu_dynticks); |
e11ec65c | 650 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting != DYNTICK_IRQ_NONIDLE); |
215bba9f PM |
651 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0); |
652 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
653 | rdtp->dynticks_nesting == 0); | |
654 | if (rdtp->dynticks_nesting != 1) { | |
655 | rdtp->dynticks_nesting--; | |
656 | return; | |
9b2e4f18 | 657 | } |
96d3fd0d | 658 | |
b04db8e1 | 659 | lockdep_assert_irqs_disabled(); |
dec98900 | 660 | trace_rcu_dyntick(TPS("Start"), rdtp->dynticks_nesting, 0, rdtp->dynticks); |
e68bbb26 | 661 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
96d3fd0d PM |
662 | for_each_rcu_flavor(rsp) { |
663 | rdp = this_cpu_ptr(rsp->rda); | |
664 | do_nocb_deferred_wakeup(rdp); | |
665 | } | |
198bbf81 | 666 | rcu_prepare_for_idle(); |
3e310098 | 667 | rcu_preempt_deferred_qs(current); |
2342172f | 668 | WRITE_ONCE(rdtp->dynticks_nesting, 0); /* Avoid irq-access tearing. */ |
844ccdd7 | 669 | rcu_dynticks_eqs_enter(); |
176f8f7a | 670 | rcu_dynticks_task_enter(); |
64db4cff | 671 | } |
adf5091e FW |
672 | |
673 | /** | |
674 | * rcu_idle_enter - inform RCU that current CPU is entering idle | |
675 | * | |
676 | * Enter idle mode, in other words, -leave- the mode in which RCU | |
677 | * read-side critical sections can occur. (Though RCU read-side | |
678 | * critical sections can occur in irq handlers in idle, a possibility | |
679 | * handled by irq_enter() and irq_exit().) | |
680 | * | |
c0da313e PM |
681 | * If you add or remove a call to rcu_idle_enter(), be sure to test with |
682 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
683 | */ |
684 | void rcu_idle_enter(void) | |
685 | { | |
b04db8e1 | 686 | lockdep_assert_irqs_disabled(); |
cb349ca9 | 687 | rcu_eqs_enter(false); |
adf5091e | 688 | } |
64db4cff | 689 | |
d1ec4c34 | 690 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
691 | /** |
692 | * rcu_user_enter - inform RCU that we are resuming userspace. | |
693 | * | |
694 | * Enter RCU idle mode right before resuming userspace. No use of RCU | |
695 | * is permitted between this call and rcu_user_exit(). This way the | |
696 | * CPU doesn't need to maintain the tick for RCU maintenance purposes | |
697 | * when the CPU runs in userspace. | |
c0da313e PM |
698 | * |
699 | * If you add or remove a call to rcu_user_enter(), be sure to test with | |
700 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
701 | */ |
702 | void rcu_user_enter(void) | |
703 | { | |
b04db8e1 | 704 | lockdep_assert_irqs_disabled(); |
d4db30af | 705 | rcu_eqs_enter(true); |
adf5091e | 706 | } |
d1ec4c34 | 707 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 708 | |
cf7614e1 | 709 | /* |
fd581a91 PM |
710 | * If we are returning from the outermost NMI handler that interrupted an |
711 | * RCU-idle period, update rdtp->dynticks and rdtp->dynticks_nmi_nesting | |
712 | * to let the RCU grace-period handling know that the CPU is back to | |
713 | * being RCU-idle. | |
714 | * | |
cf7614e1 | 715 | * If you add or remove a call to rcu_nmi_exit_common(), be sure to test |
fd581a91 PM |
716 | * with CONFIG_RCU_EQS_DEBUG=y. |
717 | */ | |
cf7614e1 | 718 | static __always_inline void rcu_nmi_exit_common(bool irq) |
fd581a91 PM |
719 | { |
720 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); | |
721 | ||
722 | /* | |
723 | * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks. | |
724 | * (We are exiting an NMI handler, so RCU better be paying attention | |
725 | * to us!) | |
726 | */ | |
727 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting <= 0); | |
728 | WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs()); | |
729 | ||
730 | /* | |
731 | * If the nesting level is not 1, the CPU wasn't RCU-idle, so | |
732 | * leave it in non-RCU-idle state. | |
733 | */ | |
734 | if (rdtp->dynticks_nmi_nesting != 1) { | |
dec98900 | 735 | trace_rcu_dyntick(TPS("--="), rdtp->dynticks_nmi_nesting, rdtp->dynticks_nmi_nesting - 2, rdtp->dynticks); |
fd581a91 PM |
736 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* No store tearing. */ |
737 | rdtp->dynticks_nmi_nesting - 2); | |
738 | return; | |
739 | } | |
740 | ||
741 | /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */ | |
dec98900 | 742 | trace_rcu_dyntick(TPS("Startirq"), rdtp->dynticks_nmi_nesting, 0, rdtp->dynticks); |
fd581a91 | 743 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */ |
cf7614e1 BP |
744 | |
745 | if (irq) | |
746 | rcu_prepare_for_idle(); | |
747 | ||
fd581a91 | 748 | rcu_dynticks_eqs_enter(); |
cf7614e1 BP |
749 | |
750 | if (irq) | |
751 | rcu_dynticks_task_enter(); | |
752 | } | |
753 | ||
754 | /** | |
755 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
756 | * @irq: Is this call from rcu_irq_exit? | |
757 | * | |
758 | * If you add or remove a call to rcu_nmi_exit(), be sure to test | |
759 | * with CONFIG_RCU_EQS_DEBUG=y. | |
760 | */ | |
761 | void rcu_nmi_exit(void) | |
762 | { | |
763 | rcu_nmi_exit_common(false); | |
fd581a91 PM |
764 | } |
765 | ||
9b2e4f18 PM |
766 | /** |
767 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
768 | * | |
769 | * Exit from an interrupt handler, which might possibly result in entering | |
770 | * idle mode, in other words, leaving the mode in which read-side critical | |
7c9906ca | 771 | * sections can occur. The caller must have disabled interrupts. |
64db4cff | 772 | * |
9b2e4f18 PM |
773 | * This code assumes that the idle loop never does anything that might |
774 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
58721f5d PM |
775 | * architecture's idle loop violates this assumption, RCU will give you what |
776 | * you deserve, good and hard. But very infrequently and irreproducibly. | |
9b2e4f18 PM |
777 | * |
778 | * Use things like work queues to work around this limitation. | |
779 | * | |
780 | * You have been warned. | |
c0da313e PM |
781 | * |
782 | * If you add or remove a call to rcu_irq_exit(), be sure to test with | |
783 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 784 | */ |
9b2e4f18 | 785 | void rcu_irq_exit(void) |
64db4cff | 786 | { |
b04db8e1 | 787 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 788 | rcu_nmi_exit_common(true); |
7c9906ca PM |
789 | } |
790 | ||
791 | /* | |
792 | * Wrapper for rcu_irq_exit() where interrupts are enabled. | |
c0da313e PM |
793 | * |
794 | * If you add or remove a call to rcu_irq_exit_irqson(), be sure to test | |
795 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
796 | */ |
797 | void rcu_irq_exit_irqson(void) | |
798 | { | |
799 | unsigned long flags; | |
800 | ||
801 | local_irq_save(flags); | |
802 | rcu_irq_exit(); | |
9b2e4f18 PM |
803 | local_irq_restore(flags); |
804 | } | |
805 | ||
adf5091e FW |
806 | /* |
807 | * Exit an RCU extended quiescent state, which can be either the | |
808 | * idle loop or adaptive-tickless usermode execution. | |
51a1fd30 PM |
809 | * |
810 | * We crowbar the ->dynticks_nmi_nesting field to DYNTICK_IRQ_NONIDLE to | |
811 | * allow for the possibility of usermode upcalls messing up our count of | |
812 | * interrupt nesting level during the busy period that is just now starting. | |
9b2e4f18 | 813 | */ |
adf5091e | 814 | static void rcu_eqs_exit(bool user) |
9b2e4f18 | 815 | { |
9b2e4f18 | 816 | struct rcu_dynticks *rdtp; |
84585aa8 | 817 | long oldval; |
9b2e4f18 | 818 | |
b04db8e1 | 819 | lockdep_assert_irqs_disabled(); |
c9d4b0af | 820 | rdtp = this_cpu_ptr(&rcu_dynticks); |
9b2e4f18 | 821 | oldval = rdtp->dynticks_nesting; |
1ce46ee5 | 822 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0); |
51a1fd30 PM |
823 | if (oldval) { |
824 | rdtp->dynticks_nesting++; | |
9dd238e2 | 825 | return; |
3a592405 | 826 | } |
9dd238e2 PM |
827 | rcu_dynticks_task_exit(); |
828 | rcu_dynticks_eqs_exit(); | |
829 | rcu_cleanup_after_idle(); | |
830 | trace_rcu_dyntick(TPS("End"), rdtp->dynticks_nesting, 1, rdtp->dynticks); | |
e68bbb26 | 831 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
9dd238e2 | 832 | WRITE_ONCE(rdtp->dynticks_nesting, 1); |
e11ec65c | 833 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting); |
9dd238e2 | 834 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, DYNTICK_IRQ_NONIDLE); |
9b2e4f18 | 835 | } |
adf5091e FW |
836 | |
837 | /** | |
838 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
839 | * | |
840 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
841 | * read-side critical sections can occur. | |
842 | * | |
c0da313e PM |
843 | * If you add or remove a call to rcu_idle_exit(), be sure to test with |
844 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
845 | */ |
846 | void rcu_idle_exit(void) | |
847 | { | |
c5d900bf FW |
848 | unsigned long flags; |
849 | ||
850 | local_irq_save(flags); | |
cb349ca9 | 851 | rcu_eqs_exit(false); |
c5d900bf | 852 | local_irq_restore(flags); |
adf5091e | 853 | } |
9b2e4f18 | 854 | |
d1ec4c34 | 855 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
856 | /** |
857 | * rcu_user_exit - inform RCU that we are exiting userspace. | |
858 | * | |
859 | * Exit RCU idle mode while entering the kernel because it can | |
860 | * run a RCU read side critical section anytime. | |
c0da313e PM |
861 | * |
862 | * If you add or remove a call to rcu_user_exit(), be sure to test with | |
863 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
864 | */ |
865 | void rcu_user_exit(void) | |
866 | { | |
91d1aa43 | 867 | rcu_eqs_exit(1); |
adf5091e | 868 | } |
d1ec4c34 | 869 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 870 | |
64db4cff | 871 | /** |
cf7614e1 BP |
872 | * rcu_nmi_enter_common - inform RCU of entry to NMI context |
873 | * @irq: Is this call from rcu_irq_enter? | |
64db4cff | 874 | * |
734d1680 PM |
875 | * If the CPU was idle from RCU's viewpoint, update rdtp->dynticks and |
876 | * rdtp->dynticks_nmi_nesting to let the RCU grace-period handling know | |
877 | * that the CPU is active. This implementation permits nested NMIs, as | |
878 | * long as the nesting level does not overflow an int. (You will probably | |
879 | * run out of stack space first.) | |
c0da313e | 880 | * |
cf7614e1 | 881 | * If you add or remove a call to rcu_nmi_enter_common(), be sure to test |
c0da313e | 882 | * with CONFIG_RCU_EQS_DEBUG=y. |
64db4cff | 883 | */ |
cf7614e1 | 884 | static __always_inline void rcu_nmi_enter_common(bool irq) |
64db4cff | 885 | { |
c9d4b0af | 886 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
84585aa8 | 887 | long incby = 2; |
64db4cff | 888 | |
734d1680 PM |
889 | /* Complain about underflow. */ |
890 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting < 0); | |
891 | ||
892 | /* | |
893 | * If idle from RCU viewpoint, atomically increment ->dynticks | |
894 | * to mark non-idle and increment ->dynticks_nmi_nesting by one. | |
895 | * Otherwise, increment ->dynticks_nmi_nesting by two. This means | |
896 | * if ->dynticks_nmi_nesting is equal to one, we are guaranteed | |
897 | * to be in the outermost NMI handler that interrupted an RCU-idle | |
898 | * period (observation due to Andy Lutomirski). | |
899 | */ | |
02a5c550 | 900 | if (rcu_dynticks_curr_cpu_in_eqs()) { |
cf7614e1 BP |
901 | |
902 | if (irq) | |
903 | rcu_dynticks_task_exit(); | |
904 | ||
2625d469 | 905 | rcu_dynticks_eqs_exit(); |
cf7614e1 BP |
906 | |
907 | if (irq) | |
908 | rcu_cleanup_after_idle(); | |
909 | ||
734d1680 PM |
910 | incby = 1; |
911 | } | |
bd2b879a PM |
912 | trace_rcu_dyntick(incby == 1 ? TPS("Endirq") : TPS("++="), |
913 | rdtp->dynticks_nmi_nesting, | |
dec98900 | 914 | rdtp->dynticks_nmi_nesting + incby, rdtp->dynticks); |
fd581a91 PM |
915 | WRITE_ONCE(rdtp->dynticks_nmi_nesting, /* Prevent store tearing. */ |
916 | rdtp->dynticks_nmi_nesting + incby); | |
734d1680 | 917 | barrier(); |
64db4cff PM |
918 | } |
919 | ||
cf7614e1 BP |
920 | /** |
921 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
922 | */ | |
923 | void rcu_nmi_enter(void) | |
924 | { | |
925 | rcu_nmi_enter_common(false); | |
926 | } | |
927 | ||
64db4cff | 928 | /** |
9b2e4f18 | 929 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle |
64db4cff | 930 | * |
9b2e4f18 PM |
931 | * Enter an interrupt handler, which might possibly result in exiting |
932 | * idle mode, in other words, entering the mode in which read-side critical | |
7c9906ca | 933 | * sections can occur. The caller must have disabled interrupts. |
c0da313e | 934 | * |
9b2e4f18 | 935 | * Note that the Linux kernel is fully capable of entering an interrupt |
58721f5d PM |
936 | * handler that it never exits, for example when doing upcalls to user mode! |
937 | * This code assumes that the idle loop never does upcalls to user mode. | |
938 | * If your architecture's idle loop does do upcalls to user mode (or does | |
939 | * anything else that results in unbalanced calls to the irq_enter() and | |
940 | * irq_exit() functions), RCU will give you what you deserve, good and hard. | |
941 | * But very infrequently and irreproducibly. | |
9b2e4f18 PM |
942 | * |
943 | * Use things like work queues to work around this limitation. | |
944 | * | |
945 | * You have been warned. | |
c0da313e PM |
946 | * |
947 | * If you add or remove a call to rcu_irq_enter(), be sure to test with | |
948 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 949 | */ |
9b2e4f18 | 950 | void rcu_irq_enter(void) |
64db4cff | 951 | { |
b04db8e1 | 952 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 953 | rcu_nmi_enter_common(true); |
7c9906ca | 954 | } |
734d1680 | 955 | |
7c9906ca PM |
956 | /* |
957 | * Wrapper for rcu_irq_enter() where interrupts are enabled. | |
c0da313e PM |
958 | * |
959 | * If you add or remove a call to rcu_irq_enter_irqson(), be sure to test | |
960 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
961 | */ |
962 | void rcu_irq_enter_irqson(void) | |
963 | { | |
964 | unsigned long flags; | |
734d1680 | 965 | |
7c9906ca PM |
966 | local_irq_save(flags); |
967 | rcu_irq_enter(); | |
64db4cff | 968 | local_irq_restore(flags); |
64db4cff PM |
969 | } |
970 | ||
5c173eb8 PM |
971 | /** |
972 | * rcu_is_watching - see if RCU thinks that the current CPU is idle | |
64db4cff | 973 | * |
791875d1 PM |
974 | * Return true if RCU is watching the running CPU, which means that this |
975 | * CPU can safely enter RCU read-side critical sections. In other words, | |
976 | * if the current CPU is in its idle loop and is neither in an interrupt | |
34240697 | 977 | * or NMI handler, return true. |
64db4cff | 978 | */ |
9418fb20 | 979 | bool notrace rcu_is_watching(void) |
64db4cff | 980 | { |
f534ed1f | 981 | bool ret; |
34240697 | 982 | |
46f00d18 | 983 | preempt_disable_notrace(); |
791875d1 | 984 | ret = !rcu_dynticks_curr_cpu_in_eqs(); |
46f00d18 | 985 | preempt_enable_notrace(); |
34240697 | 986 | return ret; |
64db4cff | 987 | } |
5c173eb8 | 988 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 989 | |
bcbfdd01 PM |
990 | /* |
991 | * If a holdout task is actually running, request an urgent quiescent | |
992 | * state from its CPU. This is unsynchronized, so migrations can cause | |
993 | * the request to go to the wrong CPU. Which is OK, all that will happen | |
994 | * is that the CPU's next context switch will be a bit slower and next | |
995 | * time around this task will generate another request. | |
996 | */ | |
997 | void rcu_request_urgent_qs_task(struct task_struct *t) | |
998 | { | |
999 | int cpu; | |
1000 | ||
1001 | barrier(); | |
1002 | cpu = task_cpu(t); | |
1003 | if (!task_curr(t)) | |
1004 | return; /* This task is not running on that CPU. */ | |
1005 | smp_store_release(per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, cpu), true); | |
1006 | } | |
1007 | ||
62fde6ed | 1008 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
1009 | |
1010 | /* | |
5554788e | 1011 | * Is the current CPU online as far as RCU is concerned? |
2036d94a | 1012 | * |
5554788e PM |
1013 | * Disable preemption to avoid false positives that could otherwise |
1014 | * happen due to the current CPU number being sampled, this task being | |
1015 | * preempted, its old CPU being taken offline, resuming on some other CPU, | |
1016 | * then determining that its old CPU is now offline. Because there are | |
1017 | * multiple flavors of RCU, and because this function can be called in the | |
1018 | * midst of updating the flavors while a given CPU coming online or going | |
1019 | * offline, it is necessary to check all flavors. If any of the flavors | |
1020 | * believe that given CPU is online, it is considered to be online. | |
c0d6d01b | 1021 | * |
5554788e PM |
1022 | * Disable checking if in an NMI handler because we cannot safely |
1023 | * report errors from NMI handlers anyway. In addition, it is OK to use | |
1024 | * RCU on an offline processor during initial boot, hence the check for | |
1025 | * rcu_scheduler_fully_active. | |
c0d6d01b PM |
1026 | */ |
1027 | bool rcu_lockdep_current_cpu_online(void) | |
1028 | { | |
2036d94a PM |
1029 | struct rcu_data *rdp; |
1030 | struct rcu_node *rnp; | |
5554788e | 1031 | struct rcu_state *rsp; |
c0d6d01b | 1032 | |
5554788e | 1033 | if (in_nmi() || !rcu_scheduler_fully_active) |
f6f7ee9a | 1034 | return true; |
c0d6d01b | 1035 | preempt_disable(); |
5554788e PM |
1036 | for_each_rcu_flavor(rsp) { |
1037 | rdp = this_cpu_ptr(rsp->rda); | |
1038 | rnp = rdp->mynode; | |
1039 | if (rdp->grpmask & rcu_rnp_online_cpus(rnp)) { | |
1040 | preempt_enable(); | |
1041 | return true; | |
1042 | } | |
1043 | } | |
c0d6d01b | 1044 | preempt_enable(); |
5554788e | 1045 | return false; |
c0d6d01b PM |
1046 | } |
1047 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
1048 | ||
62fde6ed | 1049 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 1050 | |
9b9500da PM |
1051 | /* |
1052 | * We are reporting a quiescent state on behalf of some other CPU, so | |
1053 | * it is our responsibility to check for and handle potential overflow | |
a66ae8ae | 1054 | * of the rcu_node ->gp_seq counter with respect to the rcu_data counters. |
9b9500da PM |
1055 | * After all, the CPU might be in deep idle state, and thus executing no |
1056 | * code whatsoever. | |
1057 | */ | |
1058 | static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp) | |
1059 | { | |
a32e01ee | 1060 | raw_lockdep_assert_held_rcu_node(rnp); |
a66ae8ae PM |
1061 | if (ULONG_CMP_LT(rcu_seq_current(&rdp->gp_seq) + ULONG_MAX / 4, |
1062 | rnp->gp_seq)) | |
9b9500da | 1063 | WRITE_ONCE(rdp->gpwrap, true); |
8aa670cd PM |
1064 | if (ULONG_CMP_LT(rdp->rcu_iw_gp_seq + ULONG_MAX / 4, rnp->gp_seq)) |
1065 | rdp->rcu_iw_gp_seq = rnp->gp_seq + ULONG_MAX / 4; | |
9b9500da PM |
1066 | } |
1067 | ||
64db4cff PM |
1068 | /* |
1069 | * Snapshot the specified CPU's dynticks counter so that we can later | |
1070 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 1071 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 1072 | */ |
fe5ac724 | 1073 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
64db4cff | 1074 | { |
8b2f63ab | 1075 | rdp->dynticks_snap = rcu_dynticks_snap(rdp->dynticks); |
02a5c550 | 1076 | if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) { |
fee5997c | 1077 | trace_rcu_fqs(rdp->rsp->name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 1078 | rcu_gpnum_ovf(rdp->mynode, rdp); |
23a9bacd | 1079 | return 1; |
7941dbde | 1080 | } |
23a9bacd | 1081 | return 0; |
64db4cff PM |
1082 | } |
1083 | ||
9b9500da PM |
1084 | /* |
1085 | * Handler for the irq_work request posted when a grace period has | |
1086 | * gone on for too long, but not yet long enough for an RCU CPU | |
1087 | * stall warning. Set state appropriately, but just complain if | |
1088 | * there is unexpected state on entry. | |
1089 | */ | |
1090 | static void rcu_iw_handler(struct irq_work *iwp) | |
1091 | { | |
1092 | struct rcu_data *rdp; | |
1093 | struct rcu_node *rnp; | |
1094 | ||
1095 | rdp = container_of(iwp, struct rcu_data, rcu_iw); | |
1096 | rnp = rdp->mynode; | |
1097 | raw_spin_lock_rcu_node(rnp); | |
1098 | if (!WARN_ON_ONCE(!rdp->rcu_iw_pending)) { | |
8aa670cd | 1099 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1100 | rdp->rcu_iw_pending = false; |
1101 | } | |
1102 | raw_spin_unlock_rcu_node(rnp); | |
1103 | } | |
1104 | ||
64db4cff PM |
1105 | /* |
1106 | * Return true if the specified CPU has passed through a quiescent | |
1107 | * state by virtue of being in or having passed through an dynticks | |
1108 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 1109 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 1110 | */ |
fe5ac724 | 1111 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
64db4cff | 1112 | { |
3a19b46a | 1113 | unsigned long jtsq; |
0f9be8ca | 1114 | bool *rnhqp; |
9226b10d | 1115 | bool *ruqp; |
9b9500da | 1116 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
1117 | |
1118 | /* | |
1119 | * If the CPU passed through or entered a dynticks idle phase with | |
1120 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
1121 | * already acknowledged the request to pass through a quiescent | |
1122 | * state. Either way, that CPU cannot possibly be in an RCU | |
1123 | * read-side critical section that started before the beginning | |
1124 | * of the current RCU grace period. | |
1125 | */ | |
02a5c550 | 1126 | if (rcu_dynticks_in_eqs_since(rdp->dynticks, rdp->dynticks_snap)) { |
fee5997c | 1127 | trace_rcu_fqs(rdp->rsp->name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
64db4cff | 1128 | rdp->dynticks_fqs++; |
9b9500da | 1129 | rcu_gpnum_ovf(rnp, rdp); |
64db4cff PM |
1130 | return 1; |
1131 | } | |
1132 | ||
a82dcc76 | 1133 | /* |
cee43939 PM |
1134 | * Has this CPU encountered a cond_resched() since the beginning |
1135 | * of the grace period? For this to be the case, the CPU has to | |
1136 | * have noticed the current grace period. This might not be the | |
1137 | * case for nohz_full CPUs looping in the kernel. | |
a82dcc76 | 1138 | */ |
f79c3ad6 | 1139 | jtsq = jiffies_till_sched_qs; |
9226b10d | 1140 | ruqp = per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, rdp->cpu); |
3a19b46a | 1141 | if (time_after(jiffies, rdp->rsp->gp_start + jtsq) && |
9577df9a | 1142 | READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_dynticks.rcu_qs_ctr, rdp->cpu) && |
e05720b0 | 1143 | rcu_seq_current(&rdp->gp_seq) == rnp->gp_seq && !rdp->gpwrap) { |
fee5997c | 1144 | trace_rcu_fqs(rdp->rsp->name, rdp->gp_seq, rdp->cpu, TPS("rqc")); |
9b9500da | 1145 | rcu_gpnum_ovf(rnp, rdp); |
3a19b46a | 1146 | return 1; |
f79c3ad6 | 1147 | } else if (time_after(jiffies, rdp->rsp->gp_start + jtsq)) { |
9226b10d PM |
1148 | /* Load rcu_qs_ctr before store to rcu_urgent_qs. */ |
1149 | smp_store_release(ruqp, true); | |
3a19b46a PM |
1150 | } |
1151 | ||
f2e2df59 PM |
1152 | /* If waiting too long on an offline CPU, complain. */ |
1153 | if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp)) && | |
1154 | time_after(jiffies, rdp->rsp->gp_start + HZ)) { | |
1155 | bool onl; | |
1156 | struct rcu_node *rnp1; | |
1157 | ||
1158 | WARN_ON(1); /* Offline CPUs are supposed to report QS! */ | |
1159 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", | |
1160 | __func__, rnp->grplo, rnp->grphi, rnp->level, | |
1161 | (long)rnp->gp_seq, (long)rnp->completedqs); | |
1162 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
1163 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx ->rcu_gp_init_mask %#lx\n", | |
1164 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext, rnp1->rcu_gp_init_mask); | |
1165 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); | |
1166 | pr_info("%s %d: %c online: %ld(%d) offline: %ld(%d)\n", | |
1167 | __func__, rdp->cpu, ".o"[onl], | |
1168 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
1169 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
1170 | return 1; /* Break things loose after complaining. */ | |
1171 | } | |
1172 | ||
65d798f0 | 1173 | /* |
4a81e832 PM |
1174 | * A CPU running for an extended time within the kernel can |
1175 | * delay RCU grace periods. When the CPU is in NO_HZ_FULL mode, | |
1176 | * even context-switching back and forth between a pair of | |
1177 | * in-kernel CPU-bound tasks cannot advance grace periods. | |
1178 | * So if the grace period is old enough, make the CPU pay attention. | |
1179 | * Note that the unsynchronized assignments to the per-CPU | |
0f9be8ca | 1180 | * rcu_need_heavy_qs variable are safe. Yes, setting of |
4a81e832 PM |
1181 | * bits can be lost, but they will be set again on the next |
1182 | * force-quiescent-state pass. So lost bit sets do not result | |
1183 | * in incorrect behavior, merely in a grace period lasting | |
1184 | * a few jiffies longer than it might otherwise. Because | |
1185 | * there are at most four threads involved, and because the | |
1186 | * updates are only once every few jiffies, the probability of | |
1187 | * lossage (and thus of slight grace-period extension) is | |
1188 | * quite low. | |
6193c76a | 1189 | */ |
0f9be8ca PM |
1190 | rnhqp = &per_cpu(rcu_dynticks.rcu_need_heavy_qs, rdp->cpu); |
1191 | if (!READ_ONCE(*rnhqp) && | |
1192 | (time_after(jiffies, rdp->rsp->gp_start + jtsq) || | |
1193 | time_after(jiffies, rdp->rsp->jiffies_resched))) { | |
1194 | WRITE_ONCE(*rnhqp, true); | |
9226b10d PM |
1195 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ |
1196 | smp_store_release(ruqp, true); | |
f79c3ad6 | 1197 | rdp->rsp->jiffies_resched += jtsq; /* Re-enable beating. */ |
6193c76a PM |
1198 | } |
1199 | ||
28053bc7 | 1200 | /* |
9b9500da PM |
1201 | * If more than halfway to RCU CPU stall-warning time, do a |
1202 | * resched_cpu() to try to loosen things up a bit. Also check to | |
1203 | * see if the CPU is getting hammered with interrupts, but only | |
1204 | * once per grace period, just to keep the IPIs down to a dull roar. | |
28053bc7 | 1205 | */ |
9b9500da | 1206 | if (jiffies - rdp->rsp->gp_start > rcu_jiffies_till_stall_check() / 2) { |
28053bc7 | 1207 | resched_cpu(rdp->cpu); |
9b9500da | 1208 | if (IS_ENABLED(CONFIG_IRQ_WORK) && |
8aa670cd | 1209 | !rdp->rcu_iw_pending && rdp->rcu_iw_gp_seq != rnp->gp_seq && |
9b9500da PM |
1210 | (rnp->ffmask & rdp->grpmask)) { |
1211 | init_irq_work(&rdp->rcu_iw, rcu_iw_handler); | |
1212 | rdp->rcu_iw_pending = true; | |
8aa670cd | 1213 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1214 | irq_work_queue_on(&rdp->rcu_iw, rdp->cpu); |
1215 | } | |
1216 | } | |
4914950a | 1217 | |
a82dcc76 | 1218 | return 0; |
64db4cff PM |
1219 | } |
1220 | ||
64db4cff PM |
1221 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
1222 | { | |
cb1e78cf | 1223 | unsigned long j = jiffies; |
6193c76a | 1224 | unsigned long j1; |
26cdfedf PM |
1225 | |
1226 | rsp->gp_start = j; | |
6193c76a | 1227 | j1 = rcu_jiffies_till_stall_check(); |
91f63ced PM |
1228 | /* Record ->gp_start before ->jiffies_stall. */ |
1229 | smp_store_release(&rsp->jiffies_stall, j + j1); /* ^^^ */ | |
6193c76a | 1230 | rsp->jiffies_resched = j + j1 / 2; |
7d0ae808 | 1231 | rsp->n_force_qs_gpstart = READ_ONCE(rsp->n_force_qs); |
64db4cff PM |
1232 | } |
1233 | ||
6b50e119 PM |
1234 | /* |
1235 | * Convert a ->gp_state value to a character string. | |
1236 | */ | |
1237 | static const char *gp_state_getname(short gs) | |
1238 | { | |
1239 | if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names)) | |
1240 | return "???"; | |
1241 | return gp_state_names[gs]; | |
1242 | } | |
1243 | ||
fb81a44b PM |
1244 | /* |
1245 | * Complain about starvation of grace-period kthread. | |
1246 | */ | |
1247 | static void rcu_check_gp_kthread_starvation(struct rcu_state *rsp) | |
1248 | { | |
1249 | unsigned long gpa; | |
1250 | unsigned long j; | |
1251 | ||
1252 | j = jiffies; | |
7d0ae808 | 1253 | gpa = READ_ONCE(rsp->gp_activity); |
b1adb3e2 | 1254 | if (j - gpa > 2 * HZ) { |
78c5a67f | 1255 | pr_err("%s kthread starved for %ld jiffies! g%ld f%#x %s(%d) ->state=%#lx ->cpu=%d\n", |
81e701e4 | 1256 | rsp->name, j - gpa, |
78c5a67f | 1257 | (long)rcu_seq_current(&rsp->gp_seq), |
6b50e119 PM |
1258 | rsp->gp_flags, |
1259 | gp_state_getname(rsp->gp_state), rsp->gp_state, | |
96036c43 PM |
1260 | rsp->gp_kthread ? rsp->gp_kthread->state : ~0, |
1261 | rsp->gp_kthread ? task_cpu(rsp->gp_kthread) : -1); | |
86057b80 | 1262 | if (rsp->gp_kthread) { |
d07aee2c | 1263 | pr_err("RCU grace-period kthread stack dump:\n"); |
b1adb3e2 | 1264 | sched_show_task(rsp->gp_kthread); |
86057b80 PM |
1265 | wake_up_process(rsp->gp_kthread); |
1266 | } | |
b1adb3e2 | 1267 | } |
64db4cff PM |
1268 | } |
1269 | ||
b637a328 | 1270 | /* |
7aa92230 PM |
1271 | * Dump stacks of all tasks running on stalled CPUs. First try using |
1272 | * NMIs, but fall back to manual remote stack tracing on architectures | |
1273 | * that don't support NMI-based stack dumps. The NMI-triggered stack | |
1274 | * traces are more accurate because they are printed by the target CPU. | |
b637a328 PM |
1275 | */ |
1276 | static void rcu_dump_cpu_stacks(struct rcu_state *rsp) | |
1277 | { | |
1278 | int cpu; | |
1279 | unsigned long flags; | |
1280 | struct rcu_node *rnp; | |
1281 | ||
1282 | rcu_for_each_leaf_node(rsp, rnp) { | |
6cf10081 | 1283 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7aa92230 PM |
1284 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1285 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) | |
1286 | if (!trigger_single_cpu_backtrace(cpu)) | |
bc75e999 | 1287 | dump_cpu_task(cpu); |
67c583a7 | 1288 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b637a328 PM |
1289 | } |
1290 | } | |
1291 | ||
8c7c4829 PM |
1292 | /* |
1293 | * If too much time has passed in the current grace period, and if | |
1294 | * so configured, go kick the relevant kthreads. | |
1295 | */ | |
1296 | static void rcu_stall_kick_kthreads(struct rcu_state *rsp) | |
1297 | { | |
1298 | unsigned long j; | |
1299 | ||
1300 | if (!rcu_kick_kthreads) | |
1301 | return; | |
1302 | j = READ_ONCE(rsp->jiffies_kick_kthreads); | |
aa3e0bf1 PM |
1303 | if (time_after(jiffies, j) && rsp->gp_kthread && |
1304 | (rcu_gp_in_progress(rsp) || READ_ONCE(rsp->gp_flags))) { | |
8c7c4829 | 1305 | WARN_ONCE(1, "Kicking %s grace-period kthread\n", rsp->name); |
5dffed1e | 1306 | rcu_ftrace_dump(DUMP_ALL); |
8c7c4829 PM |
1307 | wake_up_process(rsp->gp_kthread); |
1308 | WRITE_ONCE(rsp->jiffies_kick_kthreads, j + HZ); | |
1309 | } | |
1310 | } | |
1311 | ||
95394e69 | 1312 | static void panic_on_rcu_stall(void) |
088e9d25 DBO |
1313 | { |
1314 | if (sysctl_panic_on_rcu_stall) | |
1315 | panic("RCU Stall\n"); | |
1316 | } | |
1317 | ||
471f87c3 | 1318 | static void print_other_cpu_stall(struct rcu_state *rsp, unsigned long gp_seq) |
64db4cff PM |
1319 | { |
1320 | int cpu; | |
64db4cff | 1321 | unsigned long flags; |
6ccd2ecd PM |
1322 | unsigned long gpa; |
1323 | unsigned long j; | |
285fe294 | 1324 | int ndetected = 0; |
64db4cff | 1325 | struct rcu_node *rnp = rcu_get_root(rsp); |
53bb857c | 1326 | long totqlen = 0; |
64db4cff | 1327 | |
8c7c4829 PM |
1328 | /* Kick and suppress, if so configured. */ |
1329 | rcu_stall_kick_kthreads(rsp); | |
1330 | if (rcu_cpu_stall_suppress) | |
1331 | return; | |
1332 | ||
8cdd32a9 PM |
1333 | /* |
1334 | * OK, time to rat on our buddy... | |
1335 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1336 | * RCU CPU stall warnings. | |
1337 | */ | |
a7538352 | 1338 | pr_err("INFO: %s detected stalls on CPUs/tasks:", rsp->name); |
a858af28 | 1339 | print_cpu_stall_info_begin(); |
a0b6c9a7 | 1340 | rcu_for_each_leaf_node(rsp, rnp) { |
6cf10081 | 1341 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
9bc8b558 | 1342 | ndetected += rcu_print_task_stall(rnp); |
c8020a67 | 1343 | if (rnp->qsmask != 0) { |
bc75e999 MR |
1344 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1345 | if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) { | |
1346 | print_cpu_stall_info(rsp, cpu); | |
c8020a67 PM |
1347 | ndetected++; |
1348 | } | |
1349 | } | |
67c583a7 | 1350 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1351 | } |
a858af28 | 1352 | |
a858af28 | 1353 | print_cpu_stall_info_end(); |
53bb857c | 1354 | for_each_possible_cpu(cpu) |
15fecf89 PM |
1355 | totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(rsp->rda, |
1356 | cpu)->cblist); | |
471f87c3 | 1357 | pr_cont("(detected by %d, t=%ld jiffies, g=%ld, q=%lu)\n", |
eee05882 | 1358 | smp_processor_id(), (long)(jiffies - rsp->gp_start), |
471f87c3 | 1359 | (long)rcu_seq_current(&rsp->gp_seq), totqlen); |
6ccd2ecd | 1360 | if (ndetected) { |
b637a328 | 1361 | rcu_dump_cpu_stacks(rsp); |
c4402b27 BP |
1362 | |
1363 | /* Complain about tasks blocking the grace period. */ | |
1364 | rcu_print_detail_task_stall(rsp); | |
6ccd2ecd | 1365 | } else { |
471f87c3 | 1366 | if (rcu_seq_current(&rsp->gp_seq) != gp_seq) { |
6ccd2ecd PM |
1367 | pr_err("INFO: Stall ended before state dump start\n"); |
1368 | } else { | |
1369 | j = jiffies; | |
7d0ae808 | 1370 | gpa = READ_ONCE(rsp->gp_activity); |
237a0f21 | 1371 | pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n", |
6ccd2ecd | 1372 | rsp->name, j - gpa, j, gpa, |
237a0f21 PM |
1373 | jiffies_till_next_fqs, |
1374 | rcu_get_root(rsp)->qsmask); | |
6ccd2ecd PM |
1375 | /* In this case, the current CPU might be at fault. */ |
1376 | sched_show_task(current); | |
1377 | } | |
1378 | } | |
8c42b1f3 PM |
1379 | /* Rewrite if needed in case of slow consoles. */ |
1380 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall))) | |
1381 | WRITE_ONCE(rsp->jiffies_stall, | |
1382 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
c1dc0b9c | 1383 | |
fb81a44b PM |
1384 | rcu_check_gp_kthread_starvation(rsp); |
1385 | ||
088e9d25 DBO |
1386 | panic_on_rcu_stall(); |
1387 | ||
4cdfc175 | 1388 | force_quiescent_state(rsp); /* Kick them all. */ |
64db4cff PM |
1389 | } |
1390 | ||
1391 | static void print_cpu_stall(struct rcu_state *rsp) | |
1392 | { | |
53bb857c | 1393 | int cpu; |
64db4cff | 1394 | unsigned long flags; |
9b9500da | 1395 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 1396 | struct rcu_node *rnp = rcu_get_root(rsp); |
53bb857c | 1397 | long totqlen = 0; |
64db4cff | 1398 | |
8c7c4829 PM |
1399 | /* Kick and suppress, if so configured. */ |
1400 | rcu_stall_kick_kthreads(rsp); | |
1401 | if (rcu_cpu_stall_suppress) | |
1402 | return; | |
1403 | ||
8cdd32a9 PM |
1404 | /* |
1405 | * OK, time to rat on ourselves... | |
1406 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1407 | * RCU CPU stall warnings. | |
1408 | */ | |
d7f3e207 | 1409 | pr_err("INFO: %s self-detected stall on CPU", rsp->name); |
a858af28 | 1410 | print_cpu_stall_info_begin(); |
9b9500da | 1411 | raw_spin_lock_irqsave_rcu_node(rdp->mynode, flags); |
a858af28 | 1412 | print_cpu_stall_info(rsp, smp_processor_id()); |
9b9500da | 1413 | raw_spin_unlock_irqrestore_rcu_node(rdp->mynode, flags); |
a858af28 | 1414 | print_cpu_stall_info_end(); |
53bb857c | 1415 | for_each_possible_cpu(cpu) |
15fecf89 PM |
1416 | totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(rsp->rda, |
1417 | cpu)->cblist); | |
471f87c3 | 1418 | pr_cont(" (t=%lu jiffies g=%ld q=%lu)\n", |
83ebe63e | 1419 | jiffies - rsp->gp_start, |
471f87c3 | 1420 | (long)rcu_seq_current(&rsp->gp_seq), totqlen); |
fb81a44b PM |
1421 | |
1422 | rcu_check_gp_kthread_starvation(rsp); | |
1423 | ||
bc1dce51 | 1424 | rcu_dump_cpu_stacks(rsp); |
c1dc0b9c | 1425 | |
6cf10081 | 1426 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
8c42b1f3 | 1427 | /* Rewrite if needed in case of slow consoles. */ |
7d0ae808 PM |
1428 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall))) |
1429 | WRITE_ONCE(rsp->jiffies_stall, | |
1430 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
67c583a7 | 1431 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c1dc0b9c | 1432 | |
088e9d25 DBO |
1433 | panic_on_rcu_stall(); |
1434 | ||
b021fe3e PZ |
1435 | /* |
1436 | * Attempt to revive the RCU machinery by forcing a context switch. | |
1437 | * | |
1438 | * A context switch would normally allow the RCU state machine to make | |
1439 | * progress and it could be we're stuck in kernel space without context | |
1440 | * switches for an entirely unreasonable amount of time. | |
1441 | */ | |
1442 | resched_cpu(smp_processor_id()); | |
64db4cff PM |
1443 | } |
1444 | ||
1445 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
1446 | { | |
471f87c3 PM |
1447 | unsigned long gs1; |
1448 | unsigned long gs2; | |
26cdfedf | 1449 | unsigned long gps; |
bad6e139 | 1450 | unsigned long j; |
8c42b1f3 | 1451 | unsigned long jn; |
bad6e139 | 1452 | unsigned long js; |
64db4cff PM |
1453 | struct rcu_node *rnp; |
1454 | ||
8c7c4829 PM |
1455 | if ((rcu_cpu_stall_suppress && !rcu_kick_kthreads) || |
1456 | !rcu_gp_in_progress(rsp)) | |
c68de209 | 1457 | return; |
8c7c4829 | 1458 | rcu_stall_kick_kthreads(rsp); |
cb1e78cf | 1459 | j = jiffies; |
26cdfedf PM |
1460 | |
1461 | /* | |
1462 | * Lots of memory barriers to reject false positives. | |
1463 | * | |
471f87c3 PM |
1464 | * The idea is to pick up rsp->gp_seq, then rsp->jiffies_stall, |
1465 | * then rsp->gp_start, and finally another copy of rsp->gp_seq. | |
1466 | * These values are updated in the opposite order with memory | |
1467 | * barriers (or equivalent) during grace-period initialization | |
1468 | * and cleanup. Now, a false positive can occur if we get an new | |
1469 | * value of rsp->gp_start and a old value of rsp->jiffies_stall. | |
1470 | * But given the memory barriers, the only way that this can happen | |
1471 | * is if one grace period ends and another starts between these | |
1472 | * two fetches. This is detected by comparing the second fetch | |
1473 | * of rsp->gp_seq with the previous fetch from rsp->gp_seq. | |
26cdfedf PM |
1474 | * |
1475 | * Given this check, comparisons of jiffies, rsp->jiffies_stall, | |
1476 | * and rsp->gp_start suffice to forestall false positives. | |
1477 | */ | |
471f87c3 PM |
1478 | gs1 = READ_ONCE(rsp->gp_seq); |
1479 | smp_rmb(); /* Pick up ->gp_seq first... */ | |
7d0ae808 | 1480 | js = READ_ONCE(rsp->jiffies_stall); |
26cdfedf | 1481 | smp_rmb(); /* ...then ->jiffies_stall before the rest... */ |
7d0ae808 | 1482 | gps = READ_ONCE(rsp->gp_start); |
471f87c3 PM |
1483 | smp_rmb(); /* ...and finally ->gp_start before ->gp_seq again. */ |
1484 | gs2 = READ_ONCE(rsp->gp_seq); | |
1485 | if (gs1 != gs2 || | |
26cdfedf PM |
1486 | ULONG_CMP_LT(j, js) || |
1487 | ULONG_CMP_GE(gps, js)) | |
1488 | return; /* No stall or GP completed since entering function. */ | |
64db4cff | 1489 | rnp = rdp->mynode; |
8c42b1f3 | 1490 | jn = jiffies + 3 * rcu_jiffies_till_stall_check() + 3; |
c96ea7cf | 1491 | if (rcu_gp_in_progress(rsp) && |
8c42b1f3 PM |
1492 | (READ_ONCE(rnp->qsmask) & rdp->grpmask) && |
1493 | cmpxchg(&rsp->jiffies_stall, js, jn) == js) { | |
64db4cff PM |
1494 | |
1495 | /* We haven't checked in, so go dump stack. */ | |
1496 | print_cpu_stall(rsp); | |
1497 | ||
bad6e139 | 1498 | } else if (rcu_gp_in_progress(rsp) && |
8c42b1f3 PM |
1499 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY) && |
1500 | cmpxchg(&rsp->jiffies_stall, js, jn) == js) { | |
64db4cff | 1501 | |
bad6e139 | 1502 | /* They had a few time units to dump stack, so complain. */ |
471f87c3 | 1503 | print_other_cpu_stall(rsp, gs2); |
64db4cff PM |
1504 | } |
1505 | } | |
1506 | ||
53d84e00 PM |
1507 | /** |
1508 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
1509 | * | |
1510 | * Set the stall-warning timeout way off into the future, thus preventing | |
1511 | * any RCU CPU stall-warning messages from appearing in the current set of | |
1512 | * RCU grace periods. | |
1513 | * | |
1514 | * The caller must disable hard irqs. | |
1515 | */ | |
1516 | void rcu_cpu_stall_reset(void) | |
1517 | { | |
6ce75a23 PM |
1518 | struct rcu_state *rsp; |
1519 | ||
1520 | for_each_rcu_flavor(rsp) | |
7d0ae808 | 1521 | WRITE_ONCE(rsp->jiffies_stall, jiffies + ULONG_MAX / 2); |
53d84e00 PM |
1522 | } |
1523 | ||
41e80595 PM |
1524 | /* Trace-event wrapper function for trace_rcu_future_grace_period. */ |
1525 | static void trace_rcu_this_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
b73de91d | 1526 | unsigned long gp_seq_req, const char *s) |
0446be48 | 1527 | { |
b73de91d | 1528 | trace_rcu_future_grace_period(rdp->rsp->name, rnp->gp_seq, gp_seq_req, |
abd13fdd | 1529 | rnp->level, rnp->grplo, rnp->grphi, s); |
0446be48 PM |
1530 | } |
1531 | ||
1532 | /* | |
b73de91d | 1533 | * rcu_start_this_gp - Request the start of a particular grace period |
df2bf8f7 | 1534 | * @rnp_start: The leaf node of the CPU from which to start. |
b73de91d JF |
1535 | * @rdp: The rcu_data corresponding to the CPU from which to start. |
1536 | * @gp_seq_req: The gp_seq of the grace period to start. | |
1537 | * | |
41e80595 | 1538 | * Start the specified grace period, as needed to handle newly arrived |
0446be48 | 1539 | * callbacks. The required future grace periods are recorded in each |
7a1d0f23 | 1540 | * rcu_node structure's ->gp_seq_needed field. Returns true if there |
48a7639c | 1541 | * is reason to awaken the grace-period kthread. |
0446be48 | 1542 | * |
d5cd9685 PM |
1543 | * The caller must hold the specified rcu_node structure's ->lock, which |
1544 | * is why the caller is responsible for waking the grace-period kthread. | |
b73de91d JF |
1545 | * |
1546 | * Returns true if the GP thread needs to be awakened else false. | |
0446be48 | 1547 | */ |
df2bf8f7 | 1548 | static bool rcu_start_this_gp(struct rcu_node *rnp_start, struct rcu_data *rdp, |
b73de91d | 1549 | unsigned long gp_seq_req) |
0446be48 | 1550 | { |
48a7639c | 1551 | bool ret = false; |
d5cd9685 | 1552 | struct rcu_state *rsp = rdp->rsp; |
df2bf8f7 | 1553 | struct rcu_node *rnp; |
0446be48 PM |
1554 | |
1555 | /* | |
360e0da6 PM |
1556 | * Use funnel locking to either acquire the root rcu_node |
1557 | * structure's lock or bail out if the need for this grace period | |
df2bf8f7 JFG |
1558 | * has already been recorded -- or if that grace period has in |
1559 | * fact already started. If there is already a grace period in | |
1560 | * progress in a non-leaf node, no recording is needed because the | |
1561 | * end of the grace period will scan the leaf rcu_node structures. | |
1562 | * Note that rnp_start->lock must not be released. | |
0446be48 | 1563 | */ |
df2bf8f7 JFG |
1564 | raw_lockdep_assert_held_rcu_node(rnp_start); |
1565 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, TPS("Startleaf")); | |
1566 | for (rnp = rnp_start; 1; rnp = rnp->parent) { | |
1567 | if (rnp != rnp_start) | |
1568 | raw_spin_lock_rcu_node(rnp); | |
1569 | if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req) || | |
1570 | rcu_seq_started(&rnp->gp_seq, gp_seq_req) || | |
1571 | (rnp != rnp_start && | |
1572 | rcu_seq_state(rcu_seq_current(&rnp->gp_seq)))) { | |
1573 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, | |
b73de91d | 1574 | TPS("Prestarted")); |
360e0da6 PM |
1575 | goto unlock_out; |
1576 | } | |
df2bf8f7 | 1577 | rnp->gp_seq_needed = gp_seq_req; |
226ca5e7 | 1578 | if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) { |
a2165e41 | 1579 | /* |
226ca5e7 JFG |
1580 | * We just marked the leaf or internal node, and a |
1581 | * grace period is in progress, which means that | |
1582 | * rcu_gp_cleanup() will see the marking. Bail to | |
1583 | * reduce contention. | |
a2165e41 | 1584 | */ |
df2bf8f7 | 1585 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, |
b73de91d | 1586 | TPS("Startedleaf")); |
a2165e41 PM |
1587 | goto unlock_out; |
1588 | } | |
df2bf8f7 JFG |
1589 | if (rnp != rnp_start && rnp->parent != NULL) |
1590 | raw_spin_unlock_rcu_node(rnp); | |
1591 | if (!rnp->parent) | |
360e0da6 | 1592 | break; /* At root, and perhaps also leaf. */ |
0446be48 PM |
1593 | } |
1594 | ||
360e0da6 | 1595 | /* If GP already in progress, just leave, otherwise start one. */ |
29365e56 | 1596 | if (rcu_gp_in_progress(rsp)) { |
df2bf8f7 | 1597 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedleafroot")); |
0446be48 PM |
1598 | goto unlock_out; |
1599 | } | |
df2bf8f7 | 1600 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedroot")); |
360e0da6 | 1601 | WRITE_ONCE(rsp->gp_flags, rsp->gp_flags | RCU_GP_FLAG_INIT); |
26d950a9 | 1602 | rsp->gp_req_activity = jiffies; |
360e0da6 | 1603 | if (!rsp->gp_kthread) { |
df2bf8f7 | 1604 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("NoGPkthread")); |
360e0da6 | 1605 | goto unlock_out; |
0446be48 | 1606 | } |
477351f7 | 1607 | trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gp_seq), TPS("newreq")); |
360e0da6 | 1608 | ret = true; /* Caller must wake GP kthread. */ |
0446be48 | 1609 | unlock_out: |
ab5e869c | 1610 | /* Push furthest requested GP to leaf node and rcu_data structure. */ |
df2bf8f7 JFG |
1611 | if (ULONG_CMP_LT(gp_seq_req, rnp->gp_seq_needed)) { |
1612 | rnp_start->gp_seq_needed = rnp->gp_seq_needed; | |
1613 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
ab5e869c | 1614 | } |
df2bf8f7 JFG |
1615 | if (rnp != rnp_start) |
1616 | raw_spin_unlock_rcu_node(rnp); | |
48a7639c | 1617 | return ret; |
0446be48 PM |
1618 | } |
1619 | ||
1620 | /* | |
1621 | * Clean up any old requests for the just-ended grace period. Also return | |
d1e4f01d | 1622 | * whether any additional grace periods have been requested. |
0446be48 | 1623 | */ |
fb31340f | 1624 | static bool rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp) |
0446be48 | 1625 | { |
fb31340f | 1626 | bool needmore; |
0446be48 PM |
1627 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
1628 | ||
7a1d0f23 PM |
1629 | needmore = ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed); |
1630 | if (!needmore) | |
1631 | rnp->gp_seq_needed = rnp->gp_seq; /* Avoid counter wrap. */ | |
b73de91d | 1632 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq, |
41e80595 | 1633 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); |
0446be48 PM |
1634 | return needmore; |
1635 | } | |
1636 | ||
48a7639c PM |
1637 | /* |
1638 | * Awaken the grace-period kthread for the specified flavor of RCU. | |
1639 | * Don't do a self-awaken, and don't bother awakening when there is | |
1640 | * nothing for the grace-period kthread to do (as in several CPUs | |
1641 | * raced to awaken, and we lost), and finally don't try to awaken | |
1642 | * a kthread that has not yet been created. | |
1643 | */ | |
1644 | static void rcu_gp_kthread_wake(struct rcu_state *rsp) | |
1645 | { | |
1646 | if (current == rsp->gp_kthread || | |
7d0ae808 | 1647 | !READ_ONCE(rsp->gp_flags) || |
48a7639c PM |
1648 | !rsp->gp_kthread) |
1649 | return; | |
b3dae109 | 1650 | swake_up_one(&rsp->gp_wq); |
48a7639c PM |
1651 | } |
1652 | ||
dc35c893 | 1653 | /* |
29365e56 PM |
1654 | * If there is room, assign a ->gp_seq number to any callbacks on this |
1655 | * CPU that have not already been assigned. Also accelerate any callbacks | |
1656 | * that were previously assigned a ->gp_seq number that has since proven | |
1657 | * to be too conservative, which can happen if callbacks get assigned a | |
1658 | * ->gp_seq number while RCU is idle, but with reference to a non-root | |
1659 | * rcu_node structure. This function is idempotent, so it does not hurt | |
1660 | * to call it repeatedly. Returns an flag saying that we should awaken | |
1661 | * the RCU grace-period kthread. | |
dc35c893 PM |
1662 | * |
1663 | * The caller must hold rnp->lock with interrupts disabled. | |
1664 | */ | |
48a7639c | 1665 | static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1666 | struct rcu_data *rdp) |
1667 | { | |
b73de91d | 1668 | unsigned long gp_seq_req; |
15fecf89 | 1669 | bool ret = false; |
dc35c893 | 1670 | |
a32e01ee | 1671 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1672 | |
15fecf89 PM |
1673 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1674 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1675 | return false; |
dc35c893 PM |
1676 | |
1677 | /* | |
15fecf89 PM |
1678 | * Callbacks are often registered with incomplete grace-period |
1679 | * information. Something about the fact that getting exact | |
1680 | * information requires acquiring a global lock... RCU therefore | |
1681 | * makes a conservative estimate of the grace period number at which | |
1682 | * a given callback will become ready to invoke. The following | |
1683 | * code checks this estimate and improves it when possible, thus | |
1684 | * accelerating callback invocation to an earlier grace-period | |
1685 | * number. | |
dc35c893 | 1686 | */ |
b73de91d JF |
1687 | gp_seq_req = rcu_seq_snap(&rsp->gp_seq); |
1688 | if (rcu_segcblist_accelerate(&rdp->cblist, gp_seq_req)) | |
1689 | ret = rcu_start_this_gp(rnp, rdp, gp_seq_req); | |
6d4b418c PM |
1690 | |
1691 | /* Trace depending on how much we were able to accelerate. */ | |
15fecf89 | 1692 | if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL)) |
477351f7 | 1693 | trace_rcu_grace_period(rsp->name, rdp->gp_seq, TPS("AccWaitCB")); |
6d4b418c | 1694 | else |
477351f7 | 1695 | trace_rcu_grace_period(rsp->name, rdp->gp_seq, TPS("AccReadyCB")); |
48a7639c | 1696 | return ret; |
dc35c893 PM |
1697 | } |
1698 | ||
e44e73ca PM |
1699 | /* |
1700 | * Similar to rcu_accelerate_cbs(), but does not require that the leaf | |
1701 | * rcu_node structure's ->lock be held. It consults the cached value | |
1702 | * of ->gp_seq_needed in the rcu_data structure, and if that indicates | |
1703 | * that a new grace-period request be made, invokes rcu_accelerate_cbs() | |
1704 | * while holding the leaf rcu_node structure's ->lock. | |
1705 | */ | |
1706 | static void rcu_accelerate_cbs_unlocked(struct rcu_state *rsp, | |
1707 | struct rcu_node *rnp, | |
1708 | struct rcu_data *rdp) | |
1709 | { | |
1710 | unsigned long c; | |
1711 | bool needwake; | |
1712 | ||
1713 | lockdep_assert_irqs_disabled(); | |
1714 | c = rcu_seq_snap(&rsp->gp_seq); | |
1715 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { | |
1716 | /* Old request still live, so mark recent callbacks. */ | |
1717 | (void)rcu_segcblist_accelerate(&rdp->cblist, c); | |
1718 | return; | |
1719 | } | |
1720 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
1721 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); | |
1722 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ | |
1723 | if (needwake) | |
1724 | rcu_gp_kthread_wake(rsp); | |
1725 | } | |
1726 | ||
dc35c893 PM |
1727 | /* |
1728 | * Move any callbacks whose grace period has completed to the | |
1729 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
29365e56 | 1730 | * assign ->gp_seq numbers to any callbacks in the RCU_NEXT_TAIL |
dc35c893 PM |
1731 | * sublist. This function is idempotent, so it does not hurt to |
1732 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1733 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1734 | * |
1735 | * The caller must hold rnp->lock with interrupts disabled. | |
1736 | */ | |
48a7639c | 1737 | static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1738 | struct rcu_data *rdp) |
1739 | { | |
a32e01ee | 1740 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1741 | |
15fecf89 PM |
1742 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1743 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1744 | return false; |
dc35c893 PM |
1745 | |
1746 | /* | |
29365e56 | 1747 | * Find all callbacks whose ->gp_seq numbers indicate that they |
dc35c893 PM |
1748 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. |
1749 | */ | |
29365e56 | 1750 | rcu_segcblist_advance(&rdp->cblist, rnp->gp_seq); |
dc35c893 PM |
1751 | |
1752 | /* Classify any remaining callbacks. */ | |
48a7639c | 1753 | return rcu_accelerate_cbs(rsp, rnp, rdp); |
dc35c893 PM |
1754 | } |
1755 | ||
d09b62df | 1756 | /* |
ba9fbe95 PM |
1757 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1758 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1759 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1760 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1761 | */ |
48a7639c PM |
1762 | static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, |
1763 | struct rcu_data *rdp) | |
d09b62df | 1764 | { |
48a7639c | 1765 | bool ret; |
3563a438 | 1766 | bool need_gp; |
48a7639c | 1767 | |
a32e01ee | 1768 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1769 | |
67e14c1e PM |
1770 | if (rdp->gp_seq == rnp->gp_seq) |
1771 | return false; /* Nothing to do. */ | |
d09b62df | 1772 | |
67e14c1e PM |
1773 | /* Handle the ends of any preceding grace periods first. */ |
1774 | if (rcu_seq_completed_gp(rdp->gp_seq, rnp->gp_seq) || | |
1775 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
1776 | ret = rcu_advance_cbs(rsp, rnp, rdp); /* Advance callbacks. */ | |
67e14c1e PM |
1777 | trace_rcu_grace_period(rsp->name, rdp->gp_seq, TPS("cpuend")); |
1778 | } else { | |
1779 | ret = rcu_accelerate_cbs(rsp, rnp, rdp); /* Recent callbacks. */ | |
d09b62df | 1780 | } |
398ebe60 | 1781 | |
67e14c1e PM |
1782 | /* Now handle the beginnings of any new-to-this-CPU grace periods. */ |
1783 | if (rcu_seq_new_gp(rdp->gp_seq, rnp->gp_seq) || | |
1784 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
6eaef633 PM |
1785 | /* |
1786 | * If the current grace period is waiting for this CPU, | |
1787 | * set up to detect a quiescent state, otherwise don't | |
1788 | * go looking for one. | |
1789 | */ | |
5ca0905f | 1790 | trace_rcu_grace_period(rsp->name, rnp->gp_seq, TPS("cpustart")); |
3563a438 PM |
1791 | need_gp = !!(rnp->qsmask & rdp->grpmask); |
1792 | rdp->cpu_no_qs.b.norm = need_gp; | |
9577df9a | 1793 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
3563a438 | 1794 | rdp->core_needs_qs = need_gp; |
6eaef633 PM |
1795 | zero_cpu_stall_ticks(rdp); |
1796 | } | |
67e14c1e | 1797 | rdp->gp_seq = rnp->gp_seq; /* Remember new grace-period state. */ |
3d18469a PM |
1798 | if (ULONG_CMP_GE(rnp->gp_seq_needed, rdp->gp_seq_needed) || rdp->gpwrap) |
1799 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
1800 | WRITE_ONCE(rdp->gpwrap, false); | |
1801 | rcu_gpnum_ovf(rnp, rdp); | |
48a7639c | 1802 | return ret; |
6eaef633 PM |
1803 | } |
1804 | ||
d34ea322 | 1805 | static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp) |
6eaef633 PM |
1806 | { |
1807 | unsigned long flags; | |
48a7639c | 1808 | bool needwake; |
6eaef633 PM |
1809 | struct rcu_node *rnp; |
1810 | ||
1811 | local_irq_save(flags); | |
1812 | rnp = rdp->mynode; | |
67e14c1e | 1813 | if ((rdp->gp_seq == rcu_seq_current(&rnp->gp_seq) && |
7d0ae808 | 1814 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ |
2a67e741 | 1815 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
1816 | local_irq_restore(flags); |
1817 | return; | |
1818 | } | |
48a7639c | 1819 | needwake = __note_gp_changes(rsp, rnp, rdp); |
67c583a7 | 1820 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c PM |
1821 | if (needwake) |
1822 | rcu_gp_kthread_wake(rsp); | |
6eaef633 PM |
1823 | } |
1824 | ||
0f41c0dd PM |
1825 | static void rcu_gp_slow(struct rcu_state *rsp, int delay) |
1826 | { | |
1827 | if (delay > 0 && | |
dee4f422 PM |
1828 | !(rcu_seq_ctr(rsp->gp_seq) % |
1829 | (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) | |
0f41c0dd PM |
1830 | schedule_timeout_uninterruptible(delay); |
1831 | } | |
1832 | ||
b3dbec76 | 1833 | /* |
45fed3e7 | 1834 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 1835 | */ |
45fed3e7 | 1836 | static bool rcu_gp_init(struct rcu_state *rsp) |
b3dbec76 | 1837 | { |
ec2c2976 | 1838 | unsigned long flags; |
0aa04b05 | 1839 | unsigned long oldmask; |
ec2c2976 | 1840 | unsigned long mask; |
b3dbec76 | 1841 | struct rcu_data *rdp; |
7fdefc10 | 1842 | struct rcu_node *rnp = rcu_get_root(rsp); |
b3dbec76 | 1843 | |
7d0ae808 | 1844 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 1845 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 | 1846 | if (!READ_ONCE(rsp->gp_flags)) { |
f7be8209 | 1847 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 1848 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1849 | return false; |
f7be8209 | 1850 | } |
7d0ae808 | 1851 | WRITE_ONCE(rsp->gp_flags, 0); /* Clear all flags: New grace period. */ |
b3dbec76 | 1852 | |
f7be8209 PM |
1853 | if (WARN_ON_ONCE(rcu_gp_in_progress(rsp))) { |
1854 | /* | |
1855 | * Grace period already in progress, don't start another. | |
1856 | * Not supposed to be able to happen. | |
1857 | */ | |
67c583a7 | 1858 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1859 | return false; |
7fdefc10 PM |
1860 | } |
1861 | ||
7fdefc10 | 1862 | /* Advance to a new grace period and initialize state. */ |
26cdfedf | 1863 | record_gp_stall_check_time(rsp); |
ff3bb6f4 | 1864 | /* Record GP times before starting GP, hence rcu_seq_start(). */ |
de30ad51 | 1865 | rcu_seq_start(&rsp->gp_seq); |
477351f7 | 1866 | trace_rcu_grace_period(rsp->name, rsp->gp_seq, TPS("start")); |
67c583a7 | 1867 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 1868 | |
0aa04b05 PM |
1869 | /* |
1870 | * Apply per-leaf buffered online and offline operations to the | |
1871 | * rcu_node tree. Note that this new grace period need not wait | |
1872 | * for subsequent online CPUs, and that quiescent-state forcing | |
1873 | * will handle subsequent offline CPUs. | |
1874 | */ | |
fea3f222 | 1875 | rsp->gp_state = RCU_GP_ONOFF; |
0aa04b05 | 1876 | rcu_for_each_leaf_node(rsp, rnp) { |
1e64b15a | 1877 | spin_lock(&rsp->ofl_lock); |
2a67e741 | 1878 | raw_spin_lock_irq_rcu_node(rnp); |
0aa04b05 PM |
1879 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
1880 | !rnp->wait_blkd_tasks) { | |
1881 | /* Nothing to do on this leaf rcu_node structure. */ | |
67c583a7 | 1882 | raw_spin_unlock_irq_rcu_node(rnp); |
1e64b15a | 1883 | spin_unlock(&rsp->ofl_lock); |
0aa04b05 PM |
1884 | continue; |
1885 | } | |
1886 | ||
1887 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
1888 | oldmask = rnp->qsmaskinit; | |
1889 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
1890 | ||
1891 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
1892 | if (!oldmask != !rnp->qsmaskinit) { | |
962aff03 PM |
1893 | if (!oldmask) { /* First online CPU for rcu_node. */ |
1894 | if (!rnp->wait_blkd_tasks) /* Ever offline? */ | |
1895 | rcu_init_new_rnp(rnp); | |
1896 | } else if (rcu_preempt_has_tasks(rnp)) { | |
1897 | rnp->wait_blkd_tasks = true; /* blocked tasks */ | |
1898 | } else { /* Last offline CPU and can propagate. */ | |
0aa04b05 | 1899 | rcu_cleanup_dead_rnp(rnp); |
962aff03 | 1900 | } |
0aa04b05 PM |
1901 | } |
1902 | ||
1903 | /* | |
1904 | * If all waited-on tasks from prior grace period are | |
1905 | * done, and if all this rcu_node structure's CPUs are | |
1906 | * still offline, propagate up the rcu_node tree and | |
1907 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
1908 | * rcu_node structure's CPUs has since come back online, | |
962aff03 | 1909 | * simply clear ->wait_blkd_tasks. |
0aa04b05 PM |
1910 | */ |
1911 | if (rnp->wait_blkd_tasks && | |
962aff03 | 1912 | (!rcu_preempt_has_tasks(rnp) || rnp->qsmaskinit)) { |
0aa04b05 | 1913 | rnp->wait_blkd_tasks = false; |
962aff03 PM |
1914 | if (!rnp->qsmaskinit) |
1915 | rcu_cleanup_dead_rnp(rnp); | |
0aa04b05 PM |
1916 | } |
1917 | ||
67c583a7 | 1918 | raw_spin_unlock_irq_rcu_node(rnp); |
1e64b15a | 1919 | spin_unlock(&rsp->ofl_lock); |
0aa04b05 | 1920 | } |
f34f2f58 | 1921 | rcu_gp_slow(rsp, gp_preinit_delay); /* Races with CPU hotplug. */ |
7fdefc10 PM |
1922 | |
1923 | /* | |
1924 | * Set the quiescent-state-needed bits in all the rcu_node | |
1925 | * structures for all currently online CPUs in breadth-first order, | |
1926 | * starting from the root rcu_node structure, relying on the layout | |
1927 | * of the tree within the rsp->node[] array. Note that other CPUs | |
1928 | * will access only the leaves of the hierarchy, thus seeing that no | |
1929 | * grace period is in progress, at least until the corresponding | |
590d1757 | 1930 | * leaf node has been initialized. |
7fdefc10 PM |
1931 | * |
1932 | * The grace period cannot complete until the initialization | |
1933 | * process finishes, because this kthread handles both. | |
1934 | */ | |
fea3f222 | 1935 | rsp->gp_state = RCU_GP_INIT; |
7fdefc10 | 1936 | rcu_for_each_node_breadth_first(rsp, rnp) { |
0f41c0dd | 1937 | rcu_gp_slow(rsp, gp_init_delay); |
ec2c2976 | 1938 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b3dbec76 | 1939 | rdp = this_cpu_ptr(rsp->rda); |
57738942 | 1940 | rcu_preempt_check_blocked_tasks(rsp, rnp); |
7fdefc10 | 1941 | rnp->qsmask = rnp->qsmaskinit; |
de30ad51 | 1942 | WRITE_ONCE(rnp->gp_seq, rsp->gp_seq); |
7fdefc10 | 1943 | if (rnp == rdp->mynode) |
48a7639c | 1944 | (void)__note_gp_changes(rsp, rnp, rdp); |
7fdefc10 | 1945 | rcu_preempt_boost_start_gp(rnp); |
477351f7 | 1946 | trace_rcu_grace_period_init(rsp->name, rnp->gp_seq, |
7fdefc10 PM |
1947 | rnp->level, rnp->grplo, |
1948 | rnp->grphi, rnp->qsmask); | |
ec2c2976 PM |
1949 | /* Quiescent states for tasks on any now-offline CPUs. */ |
1950 | mask = rnp->qsmask & ~rnp->qsmaskinitnext; | |
f2e2df59 | 1951 | rnp->rcu_gp_init_mask = mask; |
ec2c2976 PM |
1952 | if ((mask || rnp->wait_blkd_tasks) && rcu_is_leaf_node(rnp)) |
1953 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gp_seq, flags); | |
1954 | else | |
1955 | raw_spin_unlock_irq_rcu_node(rnp); | |
cee43939 | 1956 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 1957 | WRITE_ONCE(rsp->gp_activity, jiffies); |
7fdefc10 | 1958 | } |
b3dbec76 | 1959 | |
45fed3e7 | 1960 | return true; |
7fdefc10 | 1961 | } |
b3dbec76 | 1962 | |
b9a425cf | 1963 | /* |
b3dae109 | 1964 | * Helper function for swait_event_idle_exclusive() wakeup at force-quiescent-state |
d5374226 | 1965 | * time. |
b9a425cf PM |
1966 | */ |
1967 | static bool rcu_gp_fqs_check_wake(struct rcu_state *rsp, int *gfp) | |
1968 | { | |
1969 | struct rcu_node *rnp = rcu_get_root(rsp); | |
1970 | ||
1971 | /* Someone like call_rcu() requested a force-quiescent-state scan. */ | |
1972 | *gfp = READ_ONCE(rsp->gp_flags); | |
1973 | if (*gfp & RCU_GP_FLAG_FQS) | |
1974 | return true; | |
1975 | ||
1976 | /* The current grace period has completed. */ | |
1977 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) | |
1978 | return true; | |
1979 | ||
1980 | return false; | |
1981 | } | |
1982 | ||
4cdfc175 PM |
1983 | /* |
1984 | * Do one round of quiescent-state forcing. | |
1985 | */ | |
77f81fe0 | 1986 | static void rcu_gp_fqs(struct rcu_state *rsp, bool first_time) |
4cdfc175 | 1987 | { |
4cdfc175 PM |
1988 | struct rcu_node *rnp = rcu_get_root(rsp); |
1989 | ||
7d0ae808 | 1990 | WRITE_ONCE(rsp->gp_activity, jiffies); |
4cdfc175 | 1991 | rsp->n_force_qs++; |
77f81fe0 | 1992 | if (first_time) { |
4cdfc175 | 1993 | /* Collect dyntick-idle snapshots. */ |
fe5ac724 | 1994 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
4cdfc175 PM |
1995 | } else { |
1996 | /* Handle dyntick-idle and offline CPUs. */ | |
fe5ac724 | 1997 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
4cdfc175 PM |
1998 | } |
1999 | /* Clear flag to prevent immediate re-entry. */ | |
7d0ae808 | 2000 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 2001 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 PM |
2002 | WRITE_ONCE(rsp->gp_flags, |
2003 | READ_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 2004 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 2005 | } |
4cdfc175 PM |
2006 | } |
2007 | ||
7fdefc10 PM |
2008 | /* |
2009 | * Clean up after the old grace period. | |
2010 | */ | |
4cdfc175 | 2011 | static void rcu_gp_cleanup(struct rcu_state *rsp) |
7fdefc10 PM |
2012 | { |
2013 | unsigned long gp_duration; | |
48a7639c | 2014 | bool needgp = false; |
de30ad51 | 2015 | unsigned long new_gp_seq; |
7fdefc10 PM |
2016 | struct rcu_data *rdp; |
2017 | struct rcu_node *rnp = rcu_get_root(rsp); | |
abedf8e2 | 2018 | struct swait_queue_head *sq; |
b3dbec76 | 2019 | |
7d0ae808 | 2020 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 2021 | raw_spin_lock_irq_rcu_node(rnp); |
7fdefc10 PM |
2022 | gp_duration = jiffies - rsp->gp_start; |
2023 | if (gp_duration > rsp->gp_max) | |
2024 | rsp->gp_max = gp_duration; | |
b3dbec76 | 2025 | |
7fdefc10 PM |
2026 | /* |
2027 | * We know the grace period is complete, but to everyone else | |
2028 | * it appears to still be ongoing. But it is also the case | |
2029 | * that to everyone else it looks like there is nothing that | |
2030 | * they can do to advance the grace period. It is therefore | |
2031 | * safe for us to drop the lock in order to mark the grace | |
2032 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 2033 | */ |
67c583a7 | 2034 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 2035 | |
5d4b8659 | 2036 | /* |
ff3bb6f4 PM |
2037 | * Propagate new ->gp_seq value to rcu_node structures so that |
2038 | * other CPUs don't have to wait until the start of the next grace | |
2039 | * period to process their callbacks. This also avoids some nasty | |
2040 | * RCU grace-period initialization races by forcing the end of | |
2041 | * the current grace period to be completely recorded in all of | |
2042 | * the rcu_node structures before the beginning of the next grace | |
2043 | * period is recorded in any of the rcu_node structures. | |
5d4b8659 | 2044 | */ |
de30ad51 PM |
2045 | new_gp_seq = rsp->gp_seq; |
2046 | rcu_seq_end(&new_gp_seq); | |
5d4b8659 | 2047 | rcu_for_each_node_breadth_first(rsp, rnp) { |
2a67e741 | 2048 | raw_spin_lock_irq_rcu_node(rnp); |
4bc8d555 | 2049 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
57738942 | 2050 | dump_blkd_tasks(rsp, rnp, 10); |
5c60d25f | 2051 | WARN_ON_ONCE(rnp->qsmask); |
de30ad51 | 2052 | WRITE_ONCE(rnp->gp_seq, new_gp_seq); |
b11cc576 PM |
2053 | rdp = this_cpu_ptr(rsp->rda); |
2054 | if (rnp == rdp->mynode) | |
48a7639c | 2055 | needgp = __note_gp_changes(rsp, rnp, rdp) || needgp; |
78e4bc34 | 2056 | /* smp_mb() provided by prior unlock-lock pair. */ |
fb31340f | 2057 | needgp = rcu_future_gp_cleanup(rsp, rnp) || needgp; |
065bb78c | 2058 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 2059 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 2060 | rcu_nocb_gp_cleanup(sq); |
cee43939 | 2061 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 2062 | WRITE_ONCE(rsp->gp_activity, jiffies); |
0f41c0dd | 2063 | rcu_gp_slow(rsp, gp_cleanup_delay); |
7fdefc10 | 2064 | } |
5d4b8659 | 2065 | rnp = rcu_get_root(rsp); |
de30ad51 | 2066 | raw_spin_lock_irq_rcu_node(rnp); /* GP before rsp->gp_seq update. */ |
7fdefc10 | 2067 | |
765a3f4f | 2068 | /* Declare grace period done. */ |
de30ad51 | 2069 | rcu_seq_end(&rsp->gp_seq); |
477351f7 | 2070 | trace_rcu_grace_period(rsp->name, rsp->gp_seq, TPS("end")); |
77f81fe0 | 2071 | rsp->gp_state = RCU_GP_IDLE; |
fb31340f | 2072 | /* Check for GP requests since above loop. */ |
5d4b8659 | 2073 | rdp = this_cpu_ptr(rsp->rda); |
5b55072f | 2074 | if (!needgp && ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed)) { |
abd13fdd | 2075 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq_needed, |
41e80595 | 2076 | TPS("CleanupMore")); |
fb31340f PM |
2077 | needgp = true; |
2078 | } | |
48a7639c | 2079 | /* Advance CBs to reduce false positives below. */ |
384f77f4 | 2080 | if (!rcu_accelerate_cbs(rsp, rnp, rdp) && needgp) { |
7d0ae808 | 2081 | WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT); |
26d950a9 | 2082 | rsp->gp_req_activity = jiffies; |
477351f7 | 2083 | trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gp_seq), |
bb311ecc | 2084 | TPS("newreq")); |
18390aea PM |
2085 | } else { |
2086 | WRITE_ONCE(rsp->gp_flags, rsp->gp_flags & RCU_GP_FLAG_INIT); | |
bb311ecc | 2087 | } |
67c583a7 | 2088 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 PM |
2089 | } |
2090 | ||
2091 | /* | |
2092 | * Body of kthread that handles grace periods. | |
2093 | */ | |
2094 | static int __noreturn rcu_gp_kthread(void *arg) | |
2095 | { | |
77f81fe0 | 2096 | bool first_gp_fqs; |
88d6df61 | 2097 | int gf; |
d40011f6 | 2098 | unsigned long j; |
4cdfc175 | 2099 | int ret; |
7fdefc10 PM |
2100 | struct rcu_state *rsp = arg; |
2101 | struct rcu_node *rnp = rcu_get_root(rsp); | |
2102 | ||
5871968d | 2103 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
2104 | for (;;) { |
2105 | ||
2106 | /* Handle grace-period start. */ | |
2107 | for (;;) { | |
63c4db78 | 2108 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2109 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2110 | TPS("reqwait")); |
afea227f | 2111 | rsp->gp_state = RCU_GP_WAIT_GPS; |
b3dae109 | 2112 | swait_event_idle_exclusive(rsp->gp_wq, READ_ONCE(rsp->gp_flags) & |
d5374226 | 2113 | RCU_GP_FLAG_INIT); |
319362c9 | 2114 | rsp->gp_state = RCU_GP_DONE_GPS; |
78e4bc34 | 2115 | /* Locking provides needed memory barrier. */ |
f7be8209 | 2116 | if (rcu_gp_init(rsp)) |
7fdefc10 | 2117 | break; |
cee43939 | 2118 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 2119 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2120 | WARN_ON(signal_pending(current)); |
63c4db78 | 2121 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2122 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2123 | TPS("reqwaitsig")); |
7fdefc10 | 2124 | } |
cabc49c1 | 2125 | |
4cdfc175 | 2126 | /* Handle quiescent-state forcing. */ |
77f81fe0 | 2127 | first_gp_fqs = true; |
d40011f6 | 2128 | j = jiffies_till_first_fqs; |
88d6df61 | 2129 | ret = 0; |
cabc49c1 | 2130 | for (;;) { |
8c7c4829 | 2131 | if (!ret) { |
88d6df61 | 2132 | rsp->jiffies_force_qs = jiffies + j; |
8c7c4829 PM |
2133 | WRITE_ONCE(rsp->jiffies_kick_kthreads, |
2134 | jiffies + 3 * j); | |
2135 | } | |
63c4db78 | 2136 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2137 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2138 | TPS("fqswait")); |
afea227f | 2139 | rsp->gp_state = RCU_GP_WAIT_FQS; |
b3dae109 | 2140 | ret = swait_event_idle_timeout_exclusive(rsp->gp_wq, |
b9a425cf | 2141 | rcu_gp_fqs_check_wake(rsp, &gf), j); |
32bb1c79 | 2142 | rsp->gp_state = RCU_GP_DOING_FQS; |
78e4bc34 | 2143 | /* Locking provides needed memory barriers. */ |
4cdfc175 | 2144 | /* If grace period done, leave loop. */ |
7d0ae808 | 2145 | if (!READ_ONCE(rnp->qsmask) && |
4cdfc175 | 2146 | !rcu_preempt_blocked_readers_cgp(rnp)) |
cabc49c1 | 2147 | break; |
4cdfc175 | 2148 | /* If time for quiescent-state forcing, do it. */ |
88d6df61 PM |
2149 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) || |
2150 | (gf & RCU_GP_FLAG_FQS)) { | |
63c4db78 | 2151 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2152 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2153 | TPS("fqsstart")); |
77f81fe0 PM |
2154 | rcu_gp_fqs(rsp, first_gp_fqs); |
2155 | first_gp_fqs = false; | |
63c4db78 | 2156 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2157 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2158 | TPS("fqsend")); |
cee43939 | 2159 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 2160 | WRITE_ONCE(rsp->gp_activity, jiffies); |
fcfd0a23 PM |
2161 | ret = 0; /* Force full wait till next FQS. */ |
2162 | j = jiffies_till_next_fqs; | |
4cdfc175 PM |
2163 | } else { |
2164 | /* Deal with stray signal. */ | |
cee43939 | 2165 | cond_resched_tasks_rcu_qs(); |
7d0ae808 | 2166 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2167 | WARN_ON(signal_pending(current)); |
63c4db78 | 2168 | trace_rcu_grace_period(rsp->name, |
477351f7 | 2169 | READ_ONCE(rsp->gp_seq), |
63c4db78 | 2170 | TPS("fqswaitsig")); |
fcfd0a23 PM |
2171 | ret = 1; /* Keep old FQS timing. */ |
2172 | j = jiffies; | |
2173 | if (time_after(jiffies, rsp->jiffies_force_qs)) | |
2174 | j = 1; | |
2175 | else | |
2176 | j = rsp->jiffies_force_qs - j; | |
d40011f6 | 2177 | } |
cabc49c1 | 2178 | } |
4cdfc175 PM |
2179 | |
2180 | /* Handle grace-period end. */ | |
319362c9 | 2181 | rsp->gp_state = RCU_GP_CLEANUP; |
4cdfc175 | 2182 | rcu_gp_cleanup(rsp); |
319362c9 | 2183 | rsp->gp_state = RCU_GP_CLEANED; |
b3dbec76 | 2184 | } |
b3dbec76 PM |
2185 | } |
2186 | ||
f41d911f | 2187 | /* |
8994515c PM |
2188 | * Report a full set of quiescent states to the specified rcu_state data |
2189 | * structure. Invoke rcu_gp_kthread_wake() to awaken the grace-period | |
2190 | * kthread if another grace period is required. Whether we wake | |
2191 | * the grace-period kthread or it awakens itself for the next round | |
2192 | * of quiescent-state forcing, that kthread will clean up after the | |
2193 | * just-completed grace period. Note that the caller must hold rnp->lock, | |
2194 | * which is released before return. | |
f41d911f | 2195 | */ |
d3f6bad3 | 2196 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 2197 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 2198 | { |
a32e01ee | 2199 | raw_lockdep_assert_held_rcu_node(rcu_get_root(rsp)); |
fc2219d4 | 2200 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
cd73ca21 | 2201 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 2202 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags); |
94d44776 | 2203 | rcu_gp_kthread_wake(rsp); |
f41d911f PM |
2204 | } |
2205 | ||
64db4cff | 2206 | /* |
d3f6bad3 PM |
2207 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
2208 | * Allows quiescent states for a group of CPUs to be reported at one go | |
2209 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
2210 | * must be represented by the same rcu_node structure (which need not be a |
2211 | * leaf rcu_node structure, though it often will be). The gps parameter | |
2212 | * is the grace-period snapshot, which means that the quiescent states | |
c9a24e2d | 2213 | * are valid only if rnp->gp_seq is equal to gps. That structure's lock |
654e9533 | 2214 | * must be held upon entry, and it is released before return. |
ec2c2976 PM |
2215 | * |
2216 | * As a special case, if mask is zero, the bit-already-cleared check is | |
2217 | * disabled. This allows propagating quiescent state due to resumed tasks | |
2218 | * during grace-period initialization. | |
64db4cff PM |
2219 | */ |
2220 | static void | |
d3f6bad3 | 2221 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
654e9533 | 2222 | struct rcu_node *rnp, unsigned long gps, unsigned long flags) |
64db4cff PM |
2223 | __releases(rnp->lock) |
2224 | { | |
654e9533 | 2225 | unsigned long oldmask = 0; |
28ecd580 PM |
2226 | struct rcu_node *rnp_c; |
2227 | ||
a32e01ee | 2228 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 2229 | |
64db4cff PM |
2230 | /* Walk up the rcu_node hierarchy. */ |
2231 | for (;;) { | |
ec2c2976 | 2232 | if ((!(rnp->qsmask & mask) && mask) || rnp->gp_seq != gps) { |
64db4cff | 2233 | |
654e9533 PM |
2234 | /* |
2235 | * Our bit has already been cleared, or the | |
2236 | * relevant grace period is already over, so done. | |
2237 | */ | |
67c583a7 | 2238 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2239 | return; |
2240 | } | |
654e9533 | 2241 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
5b4c11d5 | 2242 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp) && |
2dee9404 | 2243 | rcu_preempt_blocked_readers_cgp(rnp)); |
64db4cff | 2244 | rnp->qsmask &= ~mask; |
db023296 | 2245 | trace_rcu_quiescent_state_report(rsp->name, rnp->gp_seq, |
d4c08f2a PM |
2246 | mask, rnp->qsmask, rnp->level, |
2247 | rnp->grplo, rnp->grphi, | |
2248 | !!rnp->gp_tasks); | |
27f4d280 | 2249 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
2250 | |
2251 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 2252 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2253 | return; |
2254 | } | |
d43a5d32 | 2255 | rnp->completedqs = rnp->gp_seq; |
64db4cff PM |
2256 | mask = rnp->grpmask; |
2257 | if (rnp->parent == NULL) { | |
2258 | ||
2259 | /* No more levels. Exit loop holding root lock. */ | |
2260 | ||
2261 | break; | |
2262 | } | |
67c583a7 | 2263 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 2264 | rnp_c = rnp; |
64db4cff | 2265 | rnp = rnp->parent; |
2a67e741 | 2266 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
654e9533 | 2267 | oldmask = rnp_c->qsmask; |
64db4cff PM |
2268 | } |
2269 | ||
2270 | /* | |
2271 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 2272 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 2273 | * to clean up and start the next grace period if one is needed. |
64db4cff | 2274 | */ |
d3f6bad3 | 2275 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
2276 | } |
2277 | ||
cc99a310 PM |
2278 | /* |
2279 | * Record a quiescent state for all tasks that were previously queued | |
2280 | * on the specified rcu_node structure and that were blocking the current | |
2281 | * RCU grace period. The caller must hold the specified rnp->lock with | |
2282 | * irqs disabled, and this lock is released upon return, but irqs remain | |
2283 | * disabled. | |
2284 | */ | |
17a8212b PM |
2285 | static void __maybe_unused |
2286 | rcu_report_unblock_qs_rnp(struct rcu_state *rsp, | |
2287 | struct rcu_node *rnp, unsigned long flags) | |
cc99a310 PM |
2288 | __releases(rnp->lock) |
2289 | { | |
654e9533 | 2290 | unsigned long gps; |
cc99a310 PM |
2291 | unsigned long mask; |
2292 | struct rcu_node *rnp_p; | |
2293 | ||
a32e01ee | 2294 | raw_lockdep_assert_held_rcu_node(rnp); |
45975c7d | 2295 | if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPT)) || |
c74859d1 PM |
2296 | WARN_ON_ONCE(rsp != rcu_state_p) || |
2297 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)) || | |
2298 | rnp->qsmask != 0) { | |
67c583a7 | 2299 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
2300 | return; /* Still need more quiescent states! */ |
2301 | } | |
2302 | ||
77cfc7bf | 2303 | rnp->completedqs = rnp->gp_seq; |
cc99a310 PM |
2304 | rnp_p = rnp->parent; |
2305 | if (rnp_p == NULL) { | |
2306 | /* | |
a77da14c PM |
2307 | * Only one rcu_node structure in the tree, so don't |
2308 | * try to report up to its nonexistent parent! | |
cc99a310 PM |
2309 | */ |
2310 | rcu_report_qs_rsp(rsp, flags); | |
2311 | return; | |
2312 | } | |
2313 | ||
c9a24e2d PM |
2314 | /* Report up the rest of the hierarchy, tracking current ->gp_seq. */ |
2315 | gps = rnp->gp_seq; | |
cc99a310 | 2316 | mask = rnp->grpmask; |
67c583a7 | 2317 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 2318 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
654e9533 | 2319 | rcu_report_qs_rnp(mask, rsp, rnp_p, gps, flags); |
cc99a310 PM |
2320 | } |
2321 | ||
64db4cff | 2322 | /* |
d3f6bad3 | 2323 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 2324 | * structure. This must be called from the specified CPU. |
64db4cff PM |
2325 | */ |
2326 | static void | |
d7d6a11e | 2327 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2328 | { |
2329 | unsigned long flags; | |
2330 | unsigned long mask; | |
48a7639c | 2331 | bool needwake; |
64db4cff PM |
2332 | struct rcu_node *rnp; |
2333 | ||
2334 | rnp = rdp->mynode; | |
2a67e741 | 2335 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
c9a24e2d PM |
2336 | if (rdp->cpu_no_qs.b.norm || rdp->gp_seq != rnp->gp_seq || |
2337 | rdp->gpwrap) { | |
64db4cff PM |
2338 | |
2339 | /* | |
e4cc1f22 PM |
2340 | * The grace period in which this quiescent state was |
2341 | * recorded has ended, so don't report it upwards. | |
2342 | * We will instead need a new quiescent state that lies | |
2343 | * within the current grace period. | |
64db4cff | 2344 | */ |
5b74c458 | 2345 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
9577df9a | 2346 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_dynticks.rcu_qs_ctr); |
67c583a7 | 2347 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2348 | return; |
2349 | } | |
2350 | mask = rdp->grpmask; | |
2351 | if ((rnp->qsmask & mask) == 0) { | |
67c583a7 | 2352 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2353 | } else { |
bb53e416 | 2354 | rdp->core_needs_qs = false; |
64db4cff PM |
2355 | |
2356 | /* | |
2357 | * This GP can't end until cpu checks in, so all of our | |
2358 | * callbacks can be processed during the next GP. | |
2359 | */ | |
48a7639c | 2360 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
64db4cff | 2361 | |
c9a24e2d | 2362 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gp_seq, flags); |
654e9533 | 2363 | /* ^^^ Released rnp->lock */ |
48a7639c PM |
2364 | if (needwake) |
2365 | rcu_gp_kthread_wake(rsp); | |
64db4cff PM |
2366 | } |
2367 | } | |
2368 | ||
2369 | /* | |
2370 | * Check to see if there is a new grace period of which this CPU | |
2371 | * is not yet aware, and if so, set up local rcu_data state for it. | |
2372 | * Otherwise, see if this CPU has just passed through its first | |
2373 | * quiescent state for this grace period, and record that fact if so. | |
2374 | */ | |
2375 | static void | |
2376 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
2377 | { | |
05eb552b PM |
2378 | /* Check for grace-period ends and beginnings. */ |
2379 | note_gp_changes(rsp, rdp); | |
64db4cff PM |
2380 | |
2381 | /* | |
2382 | * Does this CPU still need to do its part for current grace period? | |
2383 | * If no, return and let the other CPUs do their part as well. | |
2384 | */ | |
97c668b8 | 2385 | if (!rdp->core_needs_qs) |
64db4cff PM |
2386 | return; |
2387 | ||
2388 | /* | |
2389 | * Was there a quiescent state since the beginning of the grace | |
2390 | * period? If no, then exit and wait for the next call. | |
2391 | */ | |
3a19b46a | 2392 | if (rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2393 | return; |
2394 | ||
d3f6bad3 PM |
2395 | /* |
2396 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
2397 | * judge of that). | |
2398 | */ | |
d7d6a11e | 2399 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp); |
64db4cff PM |
2400 | } |
2401 | ||
b1420f1c PM |
2402 | /* |
2403 | * Trace the fact that this CPU is going offline. | |
2404 | */ | |
2405 | static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) | |
2406 | { | |
477351f7 | 2407 | RCU_TRACE(bool blkd;) |
88a4976d PM |
2408 | RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda);) |
2409 | RCU_TRACE(struct rcu_node *rnp = rdp->mynode;) | |
b1420f1c | 2410 | |
ea46351c PM |
2411 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2412 | return; | |
2413 | ||
477351f7 PM |
2414 | RCU_TRACE(blkd = !!(rnp->qsmask & rdp->grpmask);) |
2415 | trace_rcu_grace_period(rsp->name, rnp->gp_seq, | |
2416 | blkd ? TPS("cpuofl") : TPS("cpuofl-bgp")); | |
64db4cff PM |
2417 | } |
2418 | ||
8af3a5e7 PM |
2419 | /* |
2420 | * All CPUs for the specified rcu_node structure have gone offline, | |
2421 | * and all tasks that were preempted within an RCU read-side critical | |
2422 | * section while running on one of those CPUs have since exited their RCU | |
2423 | * read-side critical section. Some other CPU is reporting this fact with | |
2424 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2425 | * This function therefore goes up the tree of rcu_node structures, | |
2426 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2427 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
c50cbe53 | 2428 | * updated. |
8af3a5e7 PM |
2429 | * |
2430 | * This function does check that the specified rcu_node structure has | |
2431 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2432 | * prematurely. That said, invoking it after the fact will cost you | |
2433 | * a needless lock acquisition. So once it has done its work, don't | |
2434 | * invoke it again. | |
2435 | */ | |
2436 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2437 | { | |
2438 | long mask; | |
2439 | struct rcu_node *rnp = rnp_leaf; | |
2440 | ||
962aff03 | 2441 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
ea46351c | 2442 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
962aff03 PM |
2443 | WARN_ON_ONCE(rnp_leaf->qsmaskinit) || |
2444 | WARN_ON_ONCE(rcu_preempt_has_tasks(rnp_leaf))) | |
8af3a5e7 PM |
2445 | return; |
2446 | for (;;) { | |
2447 | mask = rnp->grpmask; | |
2448 | rnp = rnp->parent; | |
2449 | if (!rnp) | |
2450 | break; | |
2a67e741 | 2451 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2452 | rnp->qsmaskinit &= ~mask; |
962aff03 PM |
2453 | /* Between grace periods, so better already be zero! */ |
2454 | WARN_ON_ONCE(rnp->qsmask); | |
8af3a5e7 | 2455 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2456 | raw_spin_unlock_rcu_node(rnp); |
2457 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2458 | return; |
2459 | } | |
67c583a7 | 2460 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2461 | } |
2462 | } | |
2463 | ||
64db4cff | 2464 | /* |
e5601400 | 2465 | * The CPU has been completely removed, and some other CPU is reporting |
a58163d8 PM |
2466 | * this fact from process context. Do the remainder of the cleanup. |
2467 | * There can only be one CPU hotplug operation at a time, so no need for | |
2468 | * explicit locking. | |
64db4cff | 2469 | */ |
e5601400 | 2470 | static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) |
64db4cff | 2471 | { |
e5601400 | 2472 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
b1420f1c | 2473 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2474 | |
ea46351c PM |
2475 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2476 | return; | |
2477 | ||
2036d94a | 2478 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2479 | rcu_boost_kthread_setaffinity(rnp, -1); |
64db4cff PM |
2480 | } |
2481 | ||
64db4cff PM |
2482 | /* |
2483 | * Invoke any RCU callbacks that have made it to the end of their grace | |
2484 | * period. Thottle as specified by rdp->blimit. | |
2485 | */ | |
37c72e56 | 2486 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2487 | { |
2488 | unsigned long flags; | |
15fecf89 PM |
2489 | struct rcu_head *rhp; |
2490 | struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl); | |
2491 | long bl, count; | |
64db4cff | 2492 | |
dc35c893 | 2493 | /* If no callbacks are ready, just return. */ |
15fecf89 PM |
2494 | if (!rcu_segcblist_ready_cbs(&rdp->cblist)) { |
2495 | trace_rcu_batch_start(rsp->name, | |
2496 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
2497 | rcu_segcblist_n_cbs(&rdp->cblist), 0); | |
2498 | trace_rcu_batch_end(rsp->name, 0, | |
2499 | !rcu_segcblist_empty(&rdp->cblist), | |
4968c300 PM |
2500 | need_resched(), is_idle_task(current), |
2501 | rcu_is_callbacks_kthread()); | |
64db4cff | 2502 | return; |
29c00b4a | 2503 | } |
64db4cff PM |
2504 | |
2505 | /* | |
2506 | * Extract the list of ready callbacks, disabling to prevent | |
15fecf89 PM |
2507 | * races with call_rcu() from interrupt handlers. Leave the |
2508 | * callback counts, as rcu_barrier() needs to be conservative. | |
64db4cff PM |
2509 | */ |
2510 | local_irq_save(flags); | |
8146c4e2 | 2511 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 2512 | bl = rdp->blimit; |
15fecf89 PM |
2513 | trace_rcu_batch_start(rsp->name, rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2514 | rcu_segcblist_n_cbs(&rdp->cblist), bl); | |
2515 | rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl); | |
64db4cff PM |
2516 | local_irq_restore(flags); |
2517 | ||
2518 | /* Invoke callbacks. */ | |
15fecf89 PM |
2519 | rhp = rcu_cblist_dequeue(&rcl); |
2520 | for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) { | |
2521 | debug_rcu_head_unqueue(rhp); | |
2522 | if (__rcu_reclaim(rsp->name, rhp)) | |
2523 | rcu_cblist_dequeued_lazy(&rcl); | |
2524 | /* | |
2525 | * Stop only if limit reached and CPU has something to do. | |
2526 | * Note: The rcl structure counts down from zero. | |
2527 | */ | |
4b27f20b | 2528 | if (-rcl.len >= bl && |
dff1672d PM |
2529 | (need_resched() || |
2530 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
2531 | break; |
2532 | } | |
2533 | ||
2534 | local_irq_save(flags); | |
4b27f20b | 2535 | count = -rcl.len; |
8ef0f37e PM |
2536 | trace_rcu_batch_end(rsp->name, count, !!rcl.head, need_resched(), |
2537 | is_idle_task(current), rcu_is_callbacks_kthread()); | |
64db4cff | 2538 | |
15fecf89 PM |
2539 | /* Update counts and requeue any remaining callbacks. */ |
2540 | rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl); | |
b1420f1c | 2541 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
15fecf89 | 2542 | rcu_segcblist_insert_count(&rdp->cblist, &rcl); |
64db4cff PM |
2543 | |
2544 | /* Reinstate batch limit if we have worked down the excess. */ | |
15fecf89 PM |
2545 | count = rcu_segcblist_n_cbs(&rdp->cblist); |
2546 | if (rdp->blimit == LONG_MAX && count <= qlowmark) | |
64db4cff PM |
2547 | rdp->blimit = blimit; |
2548 | ||
37c72e56 | 2549 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
15fecf89 | 2550 | if (count == 0 && rdp->qlen_last_fqs_check != 0) { |
37c72e56 PM |
2551 | rdp->qlen_last_fqs_check = 0; |
2552 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
15fecf89 PM |
2553 | } else if (count < rdp->qlen_last_fqs_check - qhimark) |
2554 | rdp->qlen_last_fqs_check = count; | |
efd88b02 PM |
2555 | |
2556 | /* | |
2557 | * The following usually indicates a double call_rcu(). To track | |
2558 | * this down, try building with CONFIG_DEBUG_OBJECTS_RCU_HEAD=y. | |
2559 | */ | |
15fecf89 | 2560 | WARN_ON_ONCE(rcu_segcblist_empty(&rdp->cblist) != (count == 0)); |
37c72e56 | 2561 | |
64db4cff PM |
2562 | local_irq_restore(flags); |
2563 | ||
e0f23060 | 2564 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
15fecf89 | 2565 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2566 | invoke_rcu_core(); |
64db4cff PM |
2567 | } |
2568 | ||
2569 | /* | |
2570 | * Check to see if this CPU is in a non-context-switch quiescent state | |
2571 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 2572 | * Also schedule RCU core processing. |
64db4cff | 2573 | * |
9b2e4f18 | 2574 | * This function must be called from hardirq context. It is normally |
5403d367 | 2575 | * invoked from the scheduling-clock interrupt. |
64db4cff | 2576 | */ |
c3377c2d | 2577 | void rcu_check_callbacks(int user) |
64db4cff | 2578 | { |
f7f7bac9 | 2579 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
a858af28 | 2580 | increment_cpu_stall_ticks(); |
45975c7d | 2581 | rcu_flavor_check_callbacks(user); |
e3950ecd | 2582 | if (rcu_pending()) |
a46e0899 | 2583 | invoke_rcu_core(); |
07f27570 | 2584 | |
f7f7bac9 | 2585 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2586 | } |
2587 | ||
64db4cff PM |
2588 | /* |
2589 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
2590 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
2591 | * Also initiate boosting for any threads blocked on the root rcu_node. |
2592 | * | |
ee47eb9f | 2593 | * The caller must have suppressed start of new grace periods. |
64db4cff | 2594 | */ |
fe5ac724 | 2595 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *rsp)) |
64db4cff | 2596 | { |
64db4cff PM |
2597 | int cpu; |
2598 | unsigned long flags; | |
2599 | unsigned long mask; | |
a0b6c9a7 | 2600 | struct rcu_node *rnp; |
64db4cff | 2601 | |
a0b6c9a7 | 2602 | rcu_for_each_leaf_node(rsp, rnp) { |
cee43939 | 2603 | cond_resched_tasks_rcu_qs(); |
64db4cff | 2604 | mask = 0; |
2a67e741 | 2605 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
a0b6c9a7 | 2606 | if (rnp->qsmask == 0) { |
45975c7d | 2607 | if (!IS_ENABLED(CONFIG_PREEMPT) || |
a77da14c PM |
2608 | rsp != rcu_state_p || |
2609 | rcu_preempt_blocked_readers_cgp(rnp)) { | |
2610 | /* | |
2611 | * No point in scanning bits because they | |
2612 | * are all zero. But we might need to | |
2613 | * priority-boost blocked readers. | |
2614 | */ | |
2615 | rcu_initiate_boost(rnp, flags); | |
2616 | /* rcu_initiate_boost() releases rnp->lock */ | |
2617 | continue; | |
2618 | } | |
92816435 PM |
2619 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
2620 | continue; | |
64db4cff | 2621 | } |
bc75e999 MR |
2622 | for_each_leaf_node_possible_cpu(rnp, cpu) { |
2623 | unsigned long bit = leaf_node_cpu_bit(rnp, cpu); | |
0edd1b17 | 2624 | if ((rnp->qsmask & bit) != 0) { |
fe5ac724 | 2625 | if (f(per_cpu_ptr(rsp->rda, cpu))) |
0edd1b17 PM |
2626 | mask |= bit; |
2627 | } | |
64db4cff | 2628 | } |
45f014c5 | 2629 | if (mask != 0) { |
c9a24e2d PM |
2630 | /* Idle/offline CPUs, report (releases rnp->lock). */ |
2631 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gp_seq, flags); | |
0aa04b05 PM |
2632 | } else { |
2633 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 2634 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2635 | } |
64db4cff | 2636 | } |
64db4cff PM |
2637 | } |
2638 | ||
2639 | /* | |
2640 | * Force quiescent states on reluctant CPUs, and also detect which | |
2641 | * CPUs are in dyntick-idle mode. | |
2642 | */ | |
4cdfc175 | 2643 | static void force_quiescent_state(struct rcu_state *rsp) |
64db4cff PM |
2644 | { |
2645 | unsigned long flags; | |
394f2769 PM |
2646 | bool ret; |
2647 | struct rcu_node *rnp; | |
2648 | struct rcu_node *rnp_old = NULL; | |
2649 | ||
2650 | /* Funnel through hierarchy to reduce memory contention. */ | |
d860d403 | 2651 | rnp = __this_cpu_read(rsp->rda->mynode); |
394f2769 | 2652 | for (; rnp != NULL; rnp = rnp->parent) { |
7d0ae808 | 2653 | ret = (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) || |
394f2769 PM |
2654 | !raw_spin_trylock(&rnp->fqslock); |
2655 | if (rnp_old != NULL) | |
2656 | raw_spin_unlock(&rnp_old->fqslock); | |
d62df573 | 2657 | if (ret) |
394f2769 | 2658 | return; |
394f2769 PM |
2659 | rnp_old = rnp; |
2660 | } | |
2661 | /* rnp_old == rcu_get_root(rsp), rnp == NULL. */ | |
64db4cff | 2662 | |
394f2769 | 2663 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 2664 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 2665 | raw_spin_unlock(&rnp_old->fqslock); |
7d0ae808 | 2666 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
67c583a7 | 2667 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 2668 | return; /* Someone beat us to it. */ |
46a1e34e | 2669 | } |
7d0ae808 | 2670 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 2671 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
94d44776 | 2672 | rcu_gp_kthread_wake(rsp); |
64db4cff PM |
2673 | } |
2674 | ||
26d950a9 PM |
2675 | /* |
2676 | * This function checks for grace-period requests that fail to motivate | |
2677 | * RCU to come out of its idle mode. | |
2678 | */ | |
2679 | static void | |
2680 | rcu_check_gp_start_stall(struct rcu_state *rsp, struct rcu_node *rnp, | |
2681 | struct rcu_data *rdp) | |
2682 | { | |
b06ae25a | 2683 | const unsigned long gpssdelay = rcu_jiffies_till_stall_check() * HZ; |
26d950a9 PM |
2684 | unsigned long flags; |
2685 | unsigned long j; | |
2686 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
2687 | static atomic_t warned = ATOMIC_INIT(0); | |
2688 | ||
7a1d0f23 PM |
2689 | if (!IS_ENABLED(CONFIG_PROVE_RCU) || rcu_gp_in_progress(rsp) || |
2690 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed)) | |
26d950a9 PM |
2691 | return; |
2692 | j = jiffies; /* Expensive access, and in common case don't get here. */ | |
b06ae25a PM |
2693 | if (time_before(j, READ_ONCE(rsp->gp_req_activity) + gpssdelay) || |
2694 | time_before(j, READ_ONCE(rsp->gp_activity) + gpssdelay) || | |
26d950a9 PM |
2695 | atomic_read(&warned)) |
2696 | return; | |
2697 | ||
2698 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
2699 | j = jiffies; | |
7a1d0f23 PM |
2700 | if (rcu_gp_in_progress(rsp) || |
2701 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) || | |
b06ae25a PM |
2702 | time_before(j, READ_ONCE(rsp->gp_req_activity) + gpssdelay) || |
2703 | time_before(j, READ_ONCE(rsp->gp_activity) + gpssdelay) || | |
26d950a9 PM |
2704 | atomic_read(&warned)) { |
2705 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2706 | return; | |
2707 | } | |
2708 | /* Hold onto the leaf lock to make others see warned==1. */ | |
2709 | ||
2710 | if (rnp_root != rnp) | |
2711 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ | |
2712 | j = jiffies; | |
7a1d0f23 PM |
2713 | if (rcu_gp_in_progress(rsp) || |
2714 | ULONG_CMP_GE(rnp_root->gp_seq, rnp_root->gp_seq_needed) || | |
b06ae25a PM |
2715 | time_before(j, rsp->gp_req_activity + gpssdelay) || |
2716 | time_before(j, rsp->gp_activity + gpssdelay) || | |
26d950a9 PM |
2717 | atomic_xchg(&warned, 1)) { |
2718 | raw_spin_unlock_rcu_node(rnp_root); /* irqs remain disabled. */ | |
2719 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2720 | return; | |
2721 | } | |
b06ae25a | 2722 | pr_alert("%s: g%ld->%ld gar:%lu ga:%lu f%#x gs:%d %s->state:%#lx\n", |
7a1d0f23 PM |
2723 | __func__, (long)READ_ONCE(rsp->gp_seq), |
2724 | (long)READ_ONCE(rnp_root->gp_seq_needed), | |
26d950a9 | 2725 | j - rsp->gp_req_activity, j - rsp->gp_activity, |
b06ae25a | 2726 | rsp->gp_flags, rsp->gp_state, rsp->name, |
26d950a9 PM |
2727 | rsp->gp_kthread ? rsp->gp_kthread->state : 0x1ffffL); |
2728 | WARN_ON(1); | |
2729 | if (rnp_root != rnp) | |
2730 | raw_spin_unlock_rcu_node(rnp_root); | |
2731 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2732 | } | |
2733 | ||
64db4cff | 2734 | /* |
e0f23060 PM |
2735 | * This does the RCU core processing work for the specified rcu_state |
2736 | * and rcu_data structures. This may be called only from the CPU to | |
2737 | * whom the rdp belongs. | |
64db4cff PM |
2738 | */ |
2739 | static void | |
1bca8cf1 | 2740 | __rcu_process_callbacks(struct rcu_state *rsp) |
64db4cff PM |
2741 | { |
2742 | unsigned long flags; | |
fa07a58f | 2743 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
26d950a9 | 2744 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 2745 | |
50dc7def | 2746 | WARN_ON_ONCE(!rdp->beenonline); |
2e597558 | 2747 | |
3e310098 PM |
2748 | /* Report any deferred quiescent states if preemption enabled. */ |
2749 | if (!(preempt_count() & PREEMPT_MASK)) | |
2750 | rcu_preempt_deferred_qs(current); | |
2751 | else if (rcu_preempt_need_deferred_qs(current)) | |
2752 | resched_cpu(rdp->cpu); /* Provoke future context switch. */ | |
2753 | ||
64db4cff PM |
2754 | /* Update RCU state based on any recent quiescent states. */ |
2755 | rcu_check_quiescent_state(rsp, rdp); | |
2756 | ||
bd7af846 PM |
2757 | /* No grace period and unregistered callbacks? */ |
2758 | if (!rcu_gp_in_progress(rsp) && | |
2759 | rcu_segcblist_is_enabled(&rdp->cblist)) { | |
2760 | local_irq_save(flags); | |
e44e73ca PM |
2761 | if (!rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
2762 | rcu_accelerate_cbs_unlocked(rsp, rnp, rdp); | |
2763 | local_irq_restore(flags); | |
64db4cff PM |
2764 | } |
2765 | ||
26d950a9 PM |
2766 | rcu_check_gp_start_stall(rsp, rnp, rdp); |
2767 | ||
64db4cff | 2768 | /* If there are callbacks ready, invoke them. */ |
15fecf89 | 2769 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2770 | invoke_rcu_callbacks(rsp, rdp); |
96d3fd0d PM |
2771 | |
2772 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
2773 | do_nocb_deferred_wakeup(rdp); | |
09223371 SL |
2774 | } |
2775 | ||
64db4cff | 2776 | /* |
e0f23060 | 2777 | * Do RCU core processing for the current CPU. |
64db4cff | 2778 | */ |
0766f788 | 2779 | static __latent_entropy void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 2780 | { |
6ce75a23 PM |
2781 | struct rcu_state *rsp; |
2782 | ||
bfa00b4c PM |
2783 | if (cpu_is_offline(smp_processor_id())) |
2784 | return; | |
f7f7bac9 | 2785 | trace_rcu_utilization(TPS("Start RCU core")); |
6ce75a23 PM |
2786 | for_each_rcu_flavor(rsp) |
2787 | __rcu_process_callbacks(rsp); | |
f7f7bac9 | 2788 | trace_rcu_utilization(TPS("End RCU core")); |
64db4cff PM |
2789 | } |
2790 | ||
a26ac245 | 2791 | /* |
e0f23060 PM |
2792 | * Schedule RCU callback invocation. If the specified type of RCU |
2793 | * does not support RCU priority boosting, just do a direct call, | |
2794 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
924df8a0 | 2795 | * are running on the current CPU with softirqs disabled, the |
e0f23060 | 2796 | * rcu_cpu_kthread_task cannot disappear out from under us. |
a26ac245 | 2797 | */ |
a46e0899 | 2798 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 2799 | { |
7d0ae808 | 2800 | if (unlikely(!READ_ONCE(rcu_scheduler_fully_active))) |
b0d30417 | 2801 | return; |
a46e0899 PM |
2802 | if (likely(!rsp->boost)) { |
2803 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
2804 | return; |
2805 | } | |
a46e0899 | 2806 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
2807 | } |
2808 | ||
a46e0899 | 2809 | static void invoke_rcu_core(void) |
09223371 | 2810 | { |
b0f74036 PM |
2811 | if (cpu_online(smp_processor_id())) |
2812 | raise_softirq(RCU_SOFTIRQ); | |
09223371 SL |
2813 | } |
2814 | ||
29154c57 PM |
2815 | /* |
2816 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
2817 | */ | |
2818 | static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp, | |
2819 | struct rcu_head *head, unsigned long flags) | |
64db4cff | 2820 | { |
62fde6ed PM |
2821 | /* |
2822 | * If called from an extended quiescent state, invoke the RCU | |
2823 | * core in order to force a re-evaluation of RCU's idleness. | |
2824 | */ | |
9910affa | 2825 | if (!rcu_is_watching()) |
62fde6ed PM |
2826 | invoke_rcu_core(); |
2827 | ||
a16b7a69 | 2828 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 2829 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 2830 | return; |
64db4cff | 2831 | |
37c72e56 PM |
2832 | /* |
2833 | * Force the grace period if too many callbacks or too long waiting. | |
2834 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
2835 | * if some other CPU has recently done so. Also, don't bother | |
2836 | * invoking force_quiescent_state() if the newly enqueued callback | |
2837 | * is the only one waiting for a grace period to complete. | |
2838 | */ | |
15fecf89 PM |
2839 | if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) > |
2840 | rdp->qlen_last_fqs_check + qhimark)) { | |
b52573d2 PM |
2841 | |
2842 | /* Are we ignoring a completed grace period? */ | |
470716fc | 2843 | note_gp_changes(rsp, rdp); |
b52573d2 PM |
2844 | |
2845 | /* Start a new grace period if one not already started. */ | |
2846 | if (!rcu_gp_in_progress(rsp)) { | |
e44e73ca | 2847 | rcu_accelerate_cbs_unlocked(rsp, rdp->mynode, rdp); |
b52573d2 PM |
2848 | } else { |
2849 | /* Give the grace period a kick. */ | |
2850 | rdp->blimit = LONG_MAX; | |
2851 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
15fecf89 | 2852 | rcu_segcblist_first_pend_cb(&rdp->cblist) != head) |
4cdfc175 | 2853 | force_quiescent_state(rsp); |
b52573d2 | 2854 | rdp->n_force_qs_snap = rsp->n_force_qs; |
15fecf89 | 2855 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); |
b52573d2 | 2856 | } |
4cdfc175 | 2857 | } |
29154c57 PM |
2858 | } |
2859 | ||
ae150184 PM |
2860 | /* |
2861 | * RCU callback function to leak a callback. | |
2862 | */ | |
2863 | static void rcu_leak_callback(struct rcu_head *rhp) | |
2864 | { | |
2865 | } | |
2866 | ||
3fbfbf7a PM |
2867 | /* |
2868 | * Helper function for call_rcu() and friends. The cpu argument will | |
2869 | * normally be -1, indicating "currently running CPU". It may specify | |
2870 | * a CPU only if that CPU is a no-CBs CPU. Currently, only _rcu_barrier() | |
2871 | * is expected to specify a CPU. | |
2872 | */ | |
64db4cff | 2873 | static void |
b6a4ae76 | 2874 | __call_rcu(struct rcu_head *head, rcu_callback_t func, |
3fbfbf7a | 2875 | struct rcu_state *rsp, int cpu, bool lazy) |
64db4cff PM |
2876 | { |
2877 | unsigned long flags; | |
2878 | struct rcu_data *rdp; | |
2879 | ||
b8f2ed53 PM |
2880 | /* Misaligned rcu_head! */ |
2881 | WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1)); | |
2882 | ||
ae150184 | 2883 | if (debug_rcu_head_queue(head)) { |
fa3c6647 PM |
2884 | /* |
2885 | * Probable double call_rcu(), so leak the callback. | |
2886 | * Use rcu:rcu_callback trace event to find the previous | |
2887 | * time callback was passed to __call_rcu(). | |
2888 | */ | |
2889 | WARN_ONCE(1, "__call_rcu(): Double-freed CB %p->%pF()!!!\n", | |
2890 | head, head->func); | |
7d0ae808 | 2891 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
2892 | return; |
2893 | } | |
64db4cff PM |
2894 | head->func = func; |
2895 | head->next = NULL; | |
64db4cff | 2896 | local_irq_save(flags); |
394f99a9 | 2897 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
2898 | |
2899 | /* Add the callback to our list. */ | |
15fecf89 | 2900 | if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist)) || cpu != -1) { |
3fbfbf7a PM |
2901 | int offline; |
2902 | ||
2903 | if (cpu != -1) | |
2904 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
143da9c2 PM |
2905 | if (likely(rdp->mynode)) { |
2906 | /* Post-boot, so this should be for a no-CBs CPU. */ | |
2907 | offline = !__call_rcu_nocb(rdp, head, lazy, flags); | |
2908 | WARN_ON_ONCE(offline); | |
2909 | /* Offline CPU, _call_rcu() illegal, leak callback. */ | |
2910 | local_irq_restore(flags); | |
2911 | return; | |
2912 | } | |
2913 | /* | |
2914 | * Very early boot, before rcu_init(). Initialize if needed | |
2915 | * and then drop through to queue the callback. | |
2916 | */ | |
2917 | BUG_ON(cpu != -1); | |
34404ca8 | 2918 | WARN_ON_ONCE(!rcu_is_watching()); |
15fecf89 PM |
2919 | if (rcu_segcblist_empty(&rdp->cblist)) |
2920 | rcu_segcblist_init(&rdp->cblist); | |
0d8ee37e | 2921 | } |
15fecf89 PM |
2922 | rcu_segcblist_enqueue(&rdp->cblist, head, lazy); |
2923 | if (!lazy) | |
c57afe80 | 2924 | rcu_idle_count_callbacks_posted(); |
2655d57e | 2925 | |
d4c08f2a PM |
2926 | if (__is_kfree_rcu_offset((unsigned long)func)) |
2927 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
15fecf89 PM |
2928 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2929 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2930 | else |
15fecf89 PM |
2931 | trace_rcu_callback(rsp->name, head, |
2932 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
2933 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2934 | |
29154c57 PM |
2935 | /* Go handle any RCU core processing required. */ |
2936 | __call_rcu_core(rsp, rdp, head, flags); | |
64db4cff PM |
2937 | local_irq_restore(flags); |
2938 | } | |
2939 | ||
a68a2bb2 | 2940 | /** |
45975c7d | 2941 | * call_rcu() - Queue an RCU callback for invocation after a grace period. |
a68a2bb2 PM |
2942 | * @head: structure to be used for queueing the RCU updates. |
2943 | * @func: actual callback function to be invoked after the grace period | |
2944 | * | |
2945 | * The callback function will be invoked some time after a full grace | |
45975c7d PM |
2946 | * period elapses, in other words after all pre-existing RCU read-side |
2947 | * critical sections have completed. However, the callback function | |
2948 | * might well execute concurrently with RCU read-side critical sections | |
2949 | * that started after call_rcu() was invoked. RCU read-side critical | |
2950 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), and | |
2951 | * may be nested. In addition, regions of code across which interrupts, | |
2952 | * preemption, or softirqs have been disabled also serve as RCU read-side | |
2953 | * critical sections. This includes hardware interrupt handlers, softirq | |
2954 | * handlers, and NMI handlers. | |
2955 | * | |
2956 | * Note that all CPUs must agree that the grace period extended beyond | |
2957 | * all pre-existing RCU read-side critical section. On systems with more | |
2958 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
2959 | * guaranteed to have executed a full memory barrier since the end of its | |
2960 | * last RCU read-side critical section whose beginning preceded the call | |
2961 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
2962 | * critical section that continues beyond the start of "func()" must have | |
2963 | * executed a memory barrier after the call_rcu() but before the beginning | |
2964 | * of that RCU read-side critical section. Note that these guarantees | |
2965 | * include CPUs that are offline, idle, or executing in user mode, as | |
2966 | * well as CPUs that are executing in the kernel. | |
2967 | * | |
2968 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
2969 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
2970 | * guaranteed to execute a full memory barrier during the time interval | |
2971 | * between the call to call_rcu() and the invocation of "func()" -- even | |
2972 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
2973 | * more than one CPU). | |
2974 | */ | |
2975 | void call_rcu(struct rcu_head *head, rcu_callback_t func) | |
2976 | { | |
2977 | __call_rcu(head, func, rcu_state_p, -1, 0); | |
2978 | } | |
2979 | EXPORT_SYMBOL_GPL(call_rcu); | |
2980 | ||
2981 | /** | |
2982 | * call_rcu_sched() - Queue an RCU for invocation after sched grace period. | |
2983 | * @head: structure to be used for queueing the RCU updates. | |
2984 | * @func: actual callback function to be invoked after the grace period | |
a68a2bb2 | 2985 | * |
45975c7d | 2986 | * This is transitional. |
64db4cff | 2987 | */ |
b6a4ae76 | 2988 | void call_rcu_sched(struct rcu_head *head, rcu_callback_t func) |
64db4cff | 2989 | { |
45975c7d | 2990 | call_rcu(head, func); |
64db4cff | 2991 | } |
d6714c22 | 2992 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff | 2993 | |
495aa969 ACB |
2994 | /* |
2995 | * Queue an RCU callback for lazy invocation after a grace period. | |
2996 | * This will likely be later named something like "call_rcu_lazy()", | |
2997 | * but this change will require some way of tagging the lazy RCU | |
2998 | * callbacks in the list of pending callbacks. Until then, this | |
2999 | * function may only be called from __kfree_rcu(). | |
3000 | */ | |
3001 | void kfree_call_rcu(struct rcu_head *head, | |
b6a4ae76 | 3002 | rcu_callback_t func) |
495aa969 | 3003 | { |
e534165b | 3004 | __call_rcu(head, func, rcu_state_p, -1, 1); |
495aa969 ACB |
3005 | } |
3006 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
3007 | ||
6ebb237b PM |
3008 | /** |
3009 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
3010 | * | |
45975c7d | 3011 | * This is transitional. |
6ebb237b PM |
3012 | */ |
3013 | void synchronize_sched(void) | |
3014 | { | |
45975c7d | 3015 | synchronize_rcu(); |
6ebb237b PM |
3016 | } |
3017 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
3018 | ||
765a3f4f PM |
3019 | /** |
3020 | * get_state_synchronize_rcu - Snapshot current RCU state | |
3021 | * | |
3022 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
3023 | * to determine whether or not a full grace period has elapsed in the | |
3024 | * meantime. | |
3025 | */ | |
3026 | unsigned long get_state_synchronize_rcu(void) | |
3027 | { | |
3028 | /* | |
3029 | * Any prior manipulation of RCU-protected data must happen | |
e4be81a2 | 3030 | * before the load from ->gp_seq. |
765a3f4f PM |
3031 | */ |
3032 | smp_mb(); /* ^^^ */ | |
e4be81a2 | 3033 | return rcu_seq_snap(&rcu_state_p->gp_seq); |
765a3f4f PM |
3034 | } |
3035 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
3036 | ||
3037 | /** | |
3038 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
3039 | * | |
3040 | * @oldstate: return value from earlier call to get_state_synchronize_rcu() | |
3041 | * | |
3042 | * If a full RCU grace period has elapsed since the earlier call to | |
3043 | * get_state_synchronize_rcu(), just return. Otherwise, invoke | |
3044 | * synchronize_rcu() to wait for a full grace period. | |
3045 | * | |
3046 | * Yes, this function does not take counter wrap into account. But | |
3047 | * counter wrap is harmless. If the counter wraps, we have waited for | |
3048 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3049 | * so waiting for one additional grace period should be just fine. | |
3050 | */ | |
3051 | void cond_synchronize_rcu(unsigned long oldstate) | |
3052 | { | |
e4be81a2 | 3053 | if (!rcu_seq_done(&rcu_state_p->gp_seq, oldstate)) |
765a3f4f | 3054 | synchronize_rcu(); |
e4be81a2 PM |
3055 | else |
3056 | smp_mb(); /* Ensure GP ends before subsequent accesses. */ | |
765a3f4f PM |
3057 | } |
3058 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
3059 | ||
24560056 PM |
3060 | /** |
3061 | * get_state_synchronize_sched - Snapshot current RCU-sched state | |
3062 | * | |
45975c7d | 3063 | * This is transitional, and only used by rcutorture. |
24560056 PM |
3064 | */ |
3065 | unsigned long get_state_synchronize_sched(void) | |
3066 | { | |
45975c7d | 3067 | return get_state_synchronize_rcu(); |
24560056 PM |
3068 | } |
3069 | EXPORT_SYMBOL_GPL(get_state_synchronize_sched); | |
3070 | ||
3071 | /** | |
3072 | * cond_synchronize_sched - Conditionally wait for an RCU-sched grace period | |
24560056 PM |
3073 | * @oldstate: return value from earlier call to get_state_synchronize_sched() |
3074 | * | |
45975c7d | 3075 | * This is transitional and only used by rcutorture. |
24560056 PM |
3076 | */ |
3077 | void cond_synchronize_sched(unsigned long oldstate) | |
3078 | { | |
45975c7d | 3079 | cond_synchronize_rcu(oldstate); |
24560056 PM |
3080 | } |
3081 | EXPORT_SYMBOL_GPL(cond_synchronize_sched); | |
3082 | ||
64db4cff PM |
3083 | /* |
3084 | * Check to see if there is any immediate RCU-related work to be done | |
3085 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
3086 | * The checks are in order of increasing expense: checks that can be | |
3087 | * carried out against CPU-local state are performed first. However, | |
3088 | * we must check for CPU stalls first, else we might not get a chance. | |
3089 | */ | |
3090 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
3091 | { | |
2f51f988 PM |
3092 | struct rcu_node *rnp = rdp->mynode; |
3093 | ||
64db4cff PM |
3094 | /* Check for CPU stalls, if enabled. */ |
3095 | check_cpu_stall(rsp, rdp); | |
3096 | ||
a096932f PM |
3097 | /* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */ |
3098 | if (rcu_nohz_full_cpu(rsp)) | |
3099 | return 0; | |
3100 | ||
64db4cff | 3101 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
01c495f7 | 3102 | if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) |
64db4cff PM |
3103 | return 1; |
3104 | ||
3105 | /* Does this CPU have callbacks ready to invoke? */ | |
01c495f7 | 3106 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
64db4cff PM |
3107 | return 1; |
3108 | ||
3109 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
c1935209 PM |
3110 | if (!rcu_gp_in_progress(rsp) && |
3111 | rcu_segcblist_is_enabled(&rdp->cblist) && | |
3112 | !rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) | |
64db4cff PM |
3113 | return 1; |
3114 | ||
67e14c1e PM |
3115 | /* Have RCU grace period completed or started? */ |
3116 | if (rcu_seq_current(&rnp->gp_seq) != rdp->gp_seq || | |
01c495f7 | 3117 | unlikely(READ_ONCE(rdp->gpwrap))) /* outside lock */ |
64db4cff PM |
3118 | return 1; |
3119 | ||
96d3fd0d | 3120 | /* Does this CPU need a deferred NOCB wakeup? */ |
01c495f7 | 3121 | if (rcu_nocb_need_deferred_wakeup(rdp)) |
96d3fd0d | 3122 | return 1; |
96d3fd0d | 3123 | |
64db4cff PM |
3124 | /* nothing to do */ |
3125 | return 0; | |
3126 | } | |
3127 | ||
3128 | /* | |
3129 | * Check to see if there is any immediate RCU-related work to be done | |
3130 | * by the current CPU, returning 1 if so. This function is part of the | |
3131 | * RCU implementation; it is -not- an exported member of the RCU API. | |
3132 | */ | |
e3950ecd | 3133 | static int rcu_pending(void) |
64db4cff | 3134 | { |
6ce75a23 PM |
3135 | struct rcu_state *rsp; |
3136 | ||
3137 | for_each_rcu_flavor(rsp) | |
e3950ecd | 3138 | if (__rcu_pending(rsp, this_cpu_ptr(rsp->rda))) |
6ce75a23 PM |
3139 | return 1; |
3140 | return 0; | |
64db4cff PM |
3141 | } |
3142 | ||
3143 | /* | |
c0f4dfd4 PM |
3144 | * Return true if the specified CPU has any callback. If all_lazy is |
3145 | * non-NULL, store an indication of whether all callbacks are lazy. | |
3146 | * (If there are no callbacks, all of them are deemed to be lazy.) | |
64db4cff | 3147 | */ |
51fbb910 | 3148 | static bool rcu_cpu_has_callbacks(bool *all_lazy) |
64db4cff | 3149 | { |
c0f4dfd4 PM |
3150 | bool al = true; |
3151 | bool hc = false; | |
3152 | struct rcu_data *rdp; | |
6ce75a23 PM |
3153 | struct rcu_state *rsp; |
3154 | ||
c0f4dfd4 | 3155 | for_each_rcu_flavor(rsp) { |
aa6da514 | 3156 | rdp = this_cpu_ptr(rsp->rda); |
15fecf89 | 3157 | if (rcu_segcblist_empty(&rdp->cblist)) |
69c8d28c PM |
3158 | continue; |
3159 | hc = true; | |
15fecf89 | 3160 | if (rcu_segcblist_n_nonlazy_cbs(&rdp->cblist) || !all_lazy) { |
c0f4dfd4 | 3161 | al = false; |
69c8d28c PM |
3162 | break; |
3163 | } | |
c0f4dfd4 PM |
3164 | } |
3165 | if (all_lazy) | |
3166 | *all_lazy = al; | |
3167 | return hc; | |
64db4cff PM |
3168 | } |
3169 | ||
a83eff0a PM |
3170 | /* |
3171 | * Helper function for _rcu_barrier() tracing. If tracing is disabled, | |
3172 | * the compiler is expected to optimize this away. | |
3173 | */ | |
e66c33d5 | 3174 | static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s, |
a83eff0a PM |
3175 | int cpu, unsigned long done) |
3176 | { | |
3177 | trace_rcu_barrier(rsp->name, s, cpu, | |
3178 | atomic_read(&rsp->barrier_cpu_count), done); | |
3179 | } | |
3180 | ||
b1420f1c PM |
3181 | /* |
3182 | * RCU callback function for _rcu_barrier(). If we are last, wake | |
3183 | * up the task executing _rcu_barrier(). | |
3184 | */ | |
24ebbca8 | 3185 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 3186 | { |
24ebbca8 PM |
3187 | struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head); |
3188 | struct rcu_state *rsp = rdp->rsp; | |
3189 | ||
a83eff0a | 3190 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) { |
d8db2e86 PM |
3191 | _rcu_barrier_trace(rsp, TPS("LastCB"), -1, |
3192 | rsp->barrier_sequence); | |
7db74df8 | 3193 | complete(&rsp->barrier_completion); |
a83eff0a | 3194 | } else { |
d8db2e86 | 3195 | _rcu_barrier_trace(rsp, TPS("CB"), -1, rsp->barrier_sequence); |
a83eff0a | 3196 | } |
d0ec774c PM |
3197 | } |
3198 | ||
3199 | /* | |
3200 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
3201 | */ | |
3202 | static void rcu_barrier_func(void *type) | |
3203 | { | |
037b64ed | 3204 | struct rcu_state *rsp = type; |
fa07a58f | 3205 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
d0ec774c | 3206 | |
d8db2e86 | 3207 | _rcu_barrier_trace(rsp, TPS("IRQ"), -1, rsp->barrier_sequence); |
f92c734f PM |
3208 | rdp->barrier_head.func = rcu_barrier_callback; |
3209 | debug_rcu_head_queue(&rdp->barrier_head); | |
3210 | if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head, 0)) { | |
3211 | atomic_inc(&rsp->barrier_cpu_count); | |
3212 | } else { | |
3213 | debug_rcu_head_unqueue(&rdp->barrier_head); | |
d8db2e86 PM |
3214 | _rcu_barrier_trace(rsp, TPS("IRQNQ"), -1, |
3215 | rsp->barrier_sequence); | |
f92c734f | 3216 | } |
d0ec774c PM |
3217 | } |
3218 | ||
d0ec774c PM |
3219 | /* |
3220 | * Orchestrate the specified type of RCU barrier, waiting for all | |
3221 | * RCU callbacks of the specified type to complete. | |
3222 | */ | |
037b64ed | 3223 | static void _rcu_barrier(struct rcu_state *rsp) |
d0ec774c | 3224 | { |
b1420f1c | 3225 | int cpu; |
b1420f1c | 3226 | struct rcu_data *rdp; |
4f525a52 | 3227 | unsigned long s = rcu_seq_snap(&rsp->barrier_sequence); |
b1420f1c | 3228 | |
d8db2e86 | 3229 | _rcu_barrier_trace(rsp, TPS("Begin"), -1, s); |
b1420f1c | 3230 | |
e74f4c45 | 3231 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
7be7f0be | 3232 | mutex_lock(&rsp->barrier_mutex); |
b1420f1c | 3233 | |
4f525a52 PM |
3234 | /* Did someone else do our work for us? */ |
3235 | if (rcu_seq_done(&rsp->barrier_sequence, s)) { | |
d8db2e86 PM |
3236 | _rcu_barrier_trace(rsp, TPS("EarlyExit"), -1, |
3237 | rsp->barrier_sequence); | |
cf3a9c48 PM |
3238 | smp_mb(); /* caller's subsequent code after above check. */ |
3239 | mutex_unlock(&rsp->barrier_mutex); | |
3240 | return; | |
3241 | } | |
3242 | ||
4f525a52 PM |
3243 | /* Mark the start of the barrier operation. */ |
3244 | rcu_seq_start(&rsp->barrier_sequence); | |
d8db2e86 | 3245 | _rcu_barrier_trace(rsp, TPS("Inc1"), -1, rsp->barrier_sequence); |
b1420f1c | 3246 | |
d0ec774c | 3247 | /* |
b1420f1c PM |
3248 | * Initialize the count to one rather than to zero in order to |
3249 | * avoid a too-soon return to zero in case of a short grace period | |
1331e7a1 PM |
3250 | * (or preemption of this task). Exclude CPU-hotplug operations |
3251 | * to ensure that no offline CPU has callbacks queued. | |
d0ec774c | 3252 | */ |
7db74df8 | 3253 | init_completion(&rsp->barrier_completion); |
24ebbca8 | 3254 | atomic_set(&rsp->barrier_cpu_count, 1); |
1331e7a1 | 3255 | get_online_cpus(); |
b1420f1c PM |
3256 | |
3257 | /* | |
1331e7a1 PM |
3258 | * Force each CPU with callbacks to register a new callback. |
3259 | * When that callback is invoked, we will know that all of the | |
3260 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 3261 | */ |
3fbfbf7a | 3262 | for_each_possible_cpu(cpu) { |
d1e43fa5 | 3263 | if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu)) |
3fbfbf7a | 3264 | continue; |
b1420f1c | 3265 | rdp = per_cpu_ptr(rsp->rda, cpu); |
d1e43fa5 | 3266 | if (rcu_is_nocb_cpu(cpu)) { |
d7e29933 | 3267 | if (!rcu_nocb_cpu_needs_barrier(rsp, cpu)) { |
d8db2e86 | 3268 | _rcu_barrier_trace(rsp, TPS("OfflineNoCB"), cpu, |
4f525a52 | 3269 | rsp->barrier_sequence); |
d7e29933 | 3270 | } else { |
d8db2e86 | 3271 | _rcu_barrier_trace(rsp, TPS("OnlineNoCB"), cpu, |
4f525a52 | 3272 | rsp->barrier_sequence); |
41050a00 | 3273 | smp_mb__before_atomic(); |
d7e29933 PM |
3274 | atomic_inc(&rsp->barrier_cpu_count); |
3275 | __call_rcu(&rdp->barrier_head, | |
3276 | rcu_barrier_callback, rsp, cpu, 0); | |
3277 | } | |
15fecf89 | 3278 | } else if (rcu_segcblist_n_cbs(&rdp->cblist)) { |
d8db2e86 | 3279 | _rcu_barrier_trace(rsp, TPS("OnlineQ"), cpu, |
4f525a52 | 3280 | rsp->barrier_sequence); |
037b64ed | 3281 | smp_call_function_single(cpu, rcu_barrier_func, rsp, 1); |
b1420f1c | 3282 | } else { |
d8db2e86 | 3283 | _rcu_barrier_trace(rsp, TPS("OnlineNQ"), cpu, |
4f525a52 | 3284 | rsp->barrier_sequence); |
b1420f1c PM |
3285 | } |
3286 | } | |
1331e7a1 | 3287 | put_online_cpus(); |
b1420f1c PM |
3288 | |
3289 | /* | |
3290 | * Now that we have an rcu_barrier_callback() callback on each | |
3291 | * CPU, and thus each counted, remove the initial count. | |
3292 | */ | |
24ebbca8 | 3293 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) |
7db74df8 | 3294 | complete(&rsp->barrier_completion); |
b1420f1c PM |
3295 | |
3296 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
7db74df8 | 3297 | wait_for_completion(&rsp->barrier_completion); |
b1420f1c | 3298 | |
4f525a52 | 3299 | /* Mark the end of the barrier operation. */ |
d8db2e86 | 3300 | _rcu_barrier_trace(rsp, TPS("Inc2"), -1, rsp->barrier_sequence); |
4f525a52 PM |
3301 | rcu_seq_end(&rsp->barrier_sequence); |
3302 | ||
b1420f1c | 3303 | /* Other rcu_barrier() invocations can now safely proceed. */ |
7be7f0be | 3304 | mutex_unlock(&rsp->barrier_mutex); |
d0ec774c | 3305 | } |
d0ec774c PM |
3306 | |
3307 | /** | |
3308 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
3309 | */ | |
3310 | void rcu_barrier_bh(void) | |
3311 | { | |
65cfe358 | 3312 | _rcu_barrier(rcu_state_p); |
d0ec774c PM |
3313 | } |
3314 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
3315 | ||
45975c7d PM |
3316 | /** |
3317 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
3318 | * | |
3319 | * Note that this primitive does not necessarily wait for an RCU grace period | |
3320 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
3321 | * in the system, then rcu_barrier() is within its rights to return | |
3322 | * immediately, without waiting for anything, much less an RCU grace period. | |
3323 | */ | |
3324 | void rcu_barrier(void) | |
3325 | { | |
3326 | _rcu_barrier(rcu_state_p); | |
3327 | } | |
3328 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
3329 | ||
d0ec774c PM |
3330 | /** |
3331 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
45975c7d PM |
3332 | * |
3333 | * This is transitional. | |
d0ec774c PM |
3334 | */ |
3335 | void rcu_barrier_sched(void) | |
3336 | { | |
45975c7d | 3337 | rcu_barrier(); |
d0ec774c PM |
3338 | } |
3339 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
3340 | ||
0aa04b05 PM |
3341 | /* |
3342 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
3343 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
3344 | * must hold the corresponding leaf rcu_node ->lock with interrrupts | |
3345 | * disabled. | |
3346 | */ | |
3347 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
3348 | { | |
3349 | long mask; | |
8d672fa6 | 3350 | long oldmask; |
0aa04b05 PM |
3351 | struct rcu_node *rnp = rnp_leaf; |
3352 | ||
8d672fa6 | 3353 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
962aff03 | 3354 | WARN_ON_ONCE(rnp->wait_blkd_tasks); |
0aa04b05 PM |
3355 | for (;;) { |
3356 | mask = rnp->grpmask; | |
3357 | rnp = rnp->parent; | |
3358 | if (rnp == NULL) | |
3359 | return; | |
6cf10081 | 3360 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
8d672fa6 | 3361 | oldmask = rnp->qsmaskinit; |
0aa04b05 | 3362 | rnp->qsmaskinit |= mask; |
67c583a7 | 3363 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
8d672fa6 PM |
3364 | if (oldmask) |
3365 | return; | |
0aa04b05 PM |
3366 | } |
3367 | } | |
3368 | ||
64db4cff | 3369 | /* |
27569620 | 3370 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 3371 | */ |
27569620 PM |
3372 | static void __init |
3373 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff | 3374 | { |
394f99a9 | 3375 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
3376 | |
3377 | /* Set up local state, ensuring consistent view of global state. */ | |
bc75e999 | 3378 | rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu); |
27569620 | 3379 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
51a1fd30 | 3380 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != 1); |
02a5c550 | 3381 | WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp->dynticks))); |
57738942 PM |
3382 | rdp->rcu_ofl_gp_seq = rsp->gp_seq; |
3383 | rdp->rcu_ofl_gp_flags = RCU_GP_CLEANED; | |
3384 | rdp->rcu_onl_gp_seq = rsp->gp_seq; | |
3385 | rdp->rcu_onl_gp_flags = RCU_GP_CLEANED; | |
27569620 | 3386 | rdp->cpu = cpu; |
d4c08f2a | 3387 | rdp->rsp = rsp; |
3fbfbf7a | 3388 | rcu_boot_init_nocb_percpu_data(rdp); |
27569620 PM |
3389 | } |
3390 | ||
3391 | /* | |
3392 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
ff3bb6f4 PM |
3393 | * offline event can be happening at a given time. Note also that we can |
3394 | * accept some slop in the rsp->gp_seq access due to the fact that this | |
3395 | * CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 3396 | */ |
49fb4c62 | 3397 | static void |
9b67122a | 3398 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp) |
64db4cff PM |
3399 | { |
3400 | unsigned long flags; | |
394f99a9 | 3401 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
3402 | struct rcu_node *rnp = rcu_get_root(rsp); |
3403 | ||
3404 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 3405 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 PM |
3406 | rdp->qlen_last_fqs_check = 0; |
3407 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 3408 | rdp->blimit = blimit; |
15fecf89 PM |
3409 | if (rcu_segcblist_empty(&rdp->cblist) && /* No early-boot CBs? */ |
3410 | !init_nocb_callback_list(rdp)) | |
3411 | rcu_segcblist_init(&rdp->cblist); /* Re-enable callbacks. */ | |
2342172f | 3412 | rdp->dynticks->dynticks_nesting = 1; /* CPU not up, no tearing. */ |
2625d469 | 3413 | rcu_dynticks_eqs_online(); |
67c583a7 | 3414 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
64db4cff | 3415 | |
0aa04b05 PM |
3416 | /* |
3417 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
3418 | * propagation up the rcu_node tree will happen at the beginning | |
3419 | * of the next grace period. | |
3420 | */ | |
64db4cff | 3421 | rnp = rdp->mynode; |
2a67e741 | 3422 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
b9585e94 | 3423 | rdp->beenonline = true; /* We have now been online. */ |
de30ad51 | 3424 | rdp->gp_seq = rnp->gp_seq; |
7a1d0f23 | 3425 | rdp->gp_seq_needed = rnp->gp_seq; |
5b74c458 | 3426 | rdp->cpu_no_qs.b.norm = true; |
9577df9a | 3427 | rdp->rcu_qs_ctr_snap = per_cpu(rcu_dynticks.rcu_qs_ctr, cpu); |
97c668b8 | 3428 | rdp->core_needs_qs = false; |
9b9500da | 3429 | rdp->rcu_iw_pending = false; |
8aa670cd | 3430 | rdp->rcu_iw_gp_seq = rnp->gp_seq - 1; |
477351f7 | 3431 | trace_rcu_grace_period(rsp->name, rdp->gp_seq, TPS("cpuonl")); |
67c583a7 | 3432 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
3433 | } |
3434 | ||
deb34f36 PM |
3435 | /* |
3436 | * Invoked early in the CPU-online process, when pretty much all | |
3437 | * services are available. The incoming CPU is not present. | |
3438 | */ | |
4df83742 | 3439 | int rcutree_prepare_cpu(unsigned int cpu) |
64db4cff | 3440 | { |
6ce75a23 PM |
3441 | struct rcu_state *rsp; |
3442 | ||
3443 | for_each_rcu_flavor(rsp) | |
9b67122a | 3444 | rcu_init_percpu_data(cpu, rsp); |
4df83742 TG |
3445 | |
3446 | rcu_prepare_kthreads(cpu); | |
3447 | rcu_spawn_all_nocb_kthreads(cpu); | |
3448 | ||
3449 | return 0; | |
3450 | } | |
3451 | ||
deb34f36 PM |
3452 | /* |
3453 | * Update RCU priority boot kthread affinity for CPU-hotplug changes. | |
3454 | */ | |
4df83742 TG |
3455 | static void rcutree_affinity_setting(unsigned int cpu, int outgoing) |
3456 | { | |
3457 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); | |
3458 | ||
3459 | rcu_boost_kthread_setaffinity(rdp->mynode, outgoing); | |
3460 | } | |
3461 | ||
deb34f36 PM |
3462 | /* |
3463 | * Near the end of the CPU-online process. Pretty much all services | |
3464 | * enabled, and the CPU is now very much alive. | |
3465 | */ | |
4df83742 TG |
3466 | int rcutree_online_cpu(unsigned int cpu) |
3467 | { | |
9b9500da PM |
3468 | unsigned long flags; |
3469 | struct rcu_data *rdp; | |
3470 | struct rcu_node *rnp; | |
3471 | struct rcu_state *rsp; | |
3472 | ||
3473 | for_each_rcu_flavor(rsp) { | |
3474 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
3475 | rnp = rdp->mynode; | |
3476 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3477 | rnp->ffmask |= rdp->grpmask; | |
3478 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
3479 | } | |
da915ad5 PM |
3480 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3481 | srcu_online_cpu(cpu); | |
9b9500da PM |
3482 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
3483 | return 0; /* Too early in boot for scheduler work. */ | |
3484 | sync_sched_exp_online_cleanup(cpu); | |
3485 | rcutree_affinity_setting(cpu, -1); | |
4df83742 TG |
3486 | return 0; |
3487 | } | |
3488 | ||
deb34f36 PM |
3489 | /* |
3490 | * Near the beginning of the process. The CPU is still very much alive | |
3491 | * with pretty much all services enabled. | |
3492 | */ | |
4df83742 TG |
3493 | int rcutree_offline_cpu(unsigned int cpu) |
3494 | { | |
9b9500da PM |
3495 | unsigned long flags; |
3496 | struct rcu_data *rdp; | |
3497 | struct rcu_node *rnp; | |
3498 | struct rcu_state *rsp; | |
3499 | ||
3500 | for_each_rcu_flavor(rsp) { | |
3501 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
3502 | rnp = rdp->mynode; | |
3503 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3504 | rnp->ffmask &= ~rdp->grpmask; | |
3505 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
3506 | } | |
3507 | ||
4df83742 | 3508 | rcutree_affinity_setting(cpu, cpu); |
da915ad5 PM |
3509 | if (IS_ENABLED(CONFIG_TREE_SRCU)) |
3510 | srcu_offline_cpu(cpu); | |
4df83742 TG |
3511 | return 0; |
3512 | } | |
3513 | ||
deb34f36 PM |
3514 | /* |
3515 | * Near the end of the offline process. We do only tracing here. | |
3516 | */ | |
4df83742 TG |
3517 | int rcutree_dying_cpu(unsigned int cpu) |
3518 | { | |
3519 | struct rcu_state *rsp; | |
3520 | ||
3521 | for_each_rcu_flavor(rsp) | |
3522 | rcu_cleanup_dying_cpu(rsp); | |
3523 | return 0; | |
3524 | } | |
3525 | ||
deb34f36 PM |
3526 | /* |
3527 | * The outgoing CPU is gone and we are running elsewhere. | |
3528 | */ | |
4df83742 TG |
3529 | int rcutree_dead_cpu(unsigned int cpu) |
3530 | { | |
3531 | struct rcu_state *rsp; | |
3532 | ||
3533 | for_each_rcu_flavor(rsp) { | |
3534 | rcu_cleanup_dead_cpu(cpu, rsp); | |
3535 | do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu)); | |
3536 | } | |
3537 | return 0; | |
64db4cff PM |
3538 | } |
3539 | ||
f64c6013 PZ |
3540 | static DEFINE_PER_CPU(int, rcu_cpu_started); |
3541 | ||
7ec99de3 PM |
3542 | /* |
3543 | * Mark the specified CPU as being online so that subsequent grace periods | |
3544 | * (both expedited and normal) will wait on it. Note that this means that | |
3545 | * incoming CPUs are not allowed to use RCU read-side critical sections | |
3546 | * until this function is called. Failing to observe this restriction | |
3547 | * will result in lockdep splats. | |
deb34f36 PM |
3548 | * |
3549 | * Note that this function is special in that it is invoked directly | |
3550 | * from the incoming CPU rather than from the cpuhp_step mechanism. | |
3551 | * This is because this function must be invoked at a precise location. | |
7ec99de3 PM |
3552 | */ |
3553 | void rcu_cpu_starting(unsigned int cpu) | |
3554 | { | |
3555 | unsigned long flags; | |
3556 | unsigned long mask; | |
313517fc PM |
3557 | int nbits; |
3558 | unsigned long oldmask; | |
7ec99de3 PM |
3559 | struct rcu_data *rdp; |
3560 | struct rcu_node *rnp; | |
3561 | struct rcu_state *rsp; | |
3562 | ||
f64c6013 PZ |
3563 | if (per_cpu(rcu_cpu_started, cpu)) |
3564 | return; | |
3565 | ||
3566 | per_cpu(rcu_cpu_started, cpu) = 1; | |
3567 | ||
7ec99de3 | 3568 | for_each_rcu_flavor(rsp) { |
fdbb9b31 | 3569 | rdp = per_cpu_ptr(rsp->rda, cpu); |
7ec99de3 PM |
3570 | rnp = rdp->mynode; |
3571 | mask = rdp->grpmask; | |
3572 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3573 | rnp->qsmaskinitnext |= mask; | |
313517fc | 3574 | oldmask = rnp->expmaskinitnext; |
7ec99de3 | 3575 | rnp->expmaskinitnext |= mask; |
313517fc PM |
3576 | oldmask ^= rnp->expmaskinitnext; |
3577 | nbits = bitmap_weight(&oldmask, BITS_PER_LONG); | |
3578 | /* Allow lockless access for expedited grace periods. */ | |
3579 | smp_store_release(&rsp->ncpus, rsp->ncpus + nbits); /* ^^^ */ | |
e05121ba | 3580 | rcu_gpnum_ovf(rnp, rdp); /* Offline-induced counter wrap? */ |
57738942 PM |
3581 | rdp->rcu_onl_gp_seq = READ_ONCE(rsp->gp_seq); |
3582 | rdp->rcu_onl_gp_flags = READ_ONCE(rsp->gp_flags); | |
99990da1 PM |
3583 | if (rnp->qsmask & mask) { /* RCU waiting on incoming CPU? */ |
3584 | /* Report QS -after- changing ->qsmaskinitnext! */ | |
3585 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gp_seq, flags); | |
3586 | } else { | |
3587 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
3588 | } | |
7ec99de3 | 3589 | } |
313517fc | 3590 | smp_mb(); /* Ensure RCU read-side usage follows above initialization. */ |
7ec99de3 PM |
3591 | } |
3592 | ||
27d50c7e TG |
3593 | #ifdef CONFIG_HOTPLUG_CPU |
3594 | /* | |
3595 | * The CPU is exiting the idle loop into the arch_cpu_idle_dead() | |
c50cbe53 | 3596 | * function. We now remove it from the rcu_node tree's ->qsmaskinitnext |
27d50c7e TG |
3597 | * bit masks. |
3598 | */ | |
3599 | static void rcu_cleanup_dying_idle_cpu(int cpu, struct rcu_state *rsp) | |
3600 | { | |
3601 | unsigned long flags; | |
3602 | unsigned long mask; | |
3603 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
3604 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ | |
3605 | ||
27d50c7e TG |
3606 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ |
3607 | mask = rdp->grpmask; | |
1e64b15a | 3608 | spin_lock(&rsp->ofl_lock); |
27d50c7e | 3609 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ |
57738942 PM |
3610 | rdp->rcu_ofl_gp_seq = READ_ONCE(rsp->gp_seq); |
3611 | rdp->rcu_ofl_gp_flags = READ_ONCE(rsp->gp_flags); | |
fece2776 PM |
3612 | if (rnp->qsmask & mask) { /* RCU waiting on outgoing CPU? */ |
3613 | /* Report quiescent state -before- changing ->qsmaskinitnext! */ | |
3614 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gp_seq, flags); | |
3615 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3616 | } | |
27d50c7e | 3617 | rnp->qsmaskinitnext &= ~mask; |
710d60cb | 3618 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1e64b15a | 3619 | spin_unlock(&rsp->ofl_lock); |
27d50c7e TG |
3620 | } |
3621 | ||
deb34f36 PM |
3622 | /* |
3623 | * The outgoing function has no further need of RCU, so remove it from | |
3624 | * the list of CPUs that RCU must track. | |
3625 | * | |
3626 | * Note that this function is special in that it is invoked directly | |
3627 | * from the outgoing CPU rather than from the cpuhp_step mechanism. | |
3628 | * This is because this function must be invoked at a precise location. | |
3629 | */ | |
27d50c7e TG |
3630 | void rcu_report_dead(unsigned int cpu) |
3631 | { | |
3632 | struct rcu_state *rsp; | |
3633 | ||
3634 | /* QS for any half-done expedited RCU-sched GP. */ | |
3635 | preempt_disable(); | |
45975c7d | 3636 | rcu_report_exp_rdp(&rcu_state, this_cpu_ptr(rcu_state.rda)); |
27d50c7e | 3637 | preempt_enable(); |
3e310098 | 3638 | rcu_preempt_deferred_qs(current); |
27d50c7e TG |
3639 | for_each_rcu_flavor(rsp) |
3640 | rcu_cleanup_dying_idle_cpu(cpu, rsp); | |
f64c6013 PZ |
3641 | |
3642 | per_cpu(rcu_cpu_started, cpu) = 0; | |
27d50c7e | 3643 | } |
a58163d8 | 3644 | |
f2dbe4a5 | 3645 | /* Migrate the dead CPU's callbacks to the current CPU. */ |
a58163d8 PM |
3646 | static void rcu_migrate_callbacks(int cpu, struct rcu_state *rsp) |
3647 | { | |
3648 | unsigned long flags; | |
b1a2d79f | 3649 | struct rcu_data *my_rdp; |
a58163d8 | 3650 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
9fa46fb8 | 3651 | struct rcu_node *rnp_root = rcu_get_root(rdp->rsp); |
ec4eacce | 3652 | bool needwake; |
a58163d8 | 3653 | |
95335c03 PM |
3654 | if (rcu_is_nocb_cpu(cpu) || rcu_segcblist_empty(&rdp->cblist)) |
3655 | return; /* No callbacks to migrate. */ | |
3656 | ||
b1a2d79f PM |
3657 | local_irq_save(flags); |
3658 | my_rdp = this_cpu_ptr(rsp->rda); | |
3659 | if (rcu_nocb_adopt_orphan_cbs(my_rdp, rdp, flags)) { | |
3660 | local_irq_restore(flags); | |
3661 | return; | |
3662 | } | |
9fa46fb8 | 3663 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ |
ec4eacce PM |
3664 | /* Leverage recent GPs and set GP for new callbacks. */ |
3665 | needwake = rcu_advance_cbs(rsp, rnp_root, rdp) || | |
3666 | rcu_advance_cbs(rsp, rnp_root, my_rdp); | |
f2dbe4a5 | 3667 | rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist); |
09efeeee PM |
3668 | WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) != |
3669 | !rcu_segcblist_n_cbs(&my_rdp->cblist)); | |
537b85c8 | 3670 | raw_spin_unlock_irqrestore_rcu_node(rnp_root, flags); |
ec4eacce PM |
3671 | if (needwake) |
3672 | rcu_gp_kthread_wake(rsp); | |
a58163d8 PM |
3673 | WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 || |
3674 | !rcu_segcblist_empty(&rdp->cblist), | |
3675 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n", | |
3676 | cpu, rcu_segcblist_n_cbs(&rdp->cblist), | |
3677 | rcu_segcblist_first_cb(&rdp->cblist)); | |
3678 | } | |
3679 | ||
3680 | /* | |
3681 | * The outgoing CPU has just passed through the dying-idle state, | |
3682 | * and we are being invoked from the CPU that was IPIed to continue the | |
3683 | * offline operation. We need to migrate the outgoing CPU's callbacks. | |
3684 | */ | |
3685 | void rcutree_migrate_callbacks(int cpu) | |
3686 | { | |
3687 | struct rcu_state *rsp; | |
3688 | ||
3689 | for_each_rcu_flavor(rsp) | |
3690 | rcu_migrate_callbacks(cpu, rsp); | |
3691 | } | |
27d50c7e TG |
3692 | #endif |
3693 | ||
deb34f36 PM |
3694 | /* |
3695 | * On non-huge systems, use expedited RCU grace periods to make suspend | |
3696 | * and hibernation run faster. | |
3697 | */ | |
d1d74d14 BP |
3698 | static int rcu_pm_notify(struct notifier_block *self, |
3699 | unsigned long action, void *hcpu) | |
3700 | { | |
3701 | switch (action) { | |
3702 | case PM_HIBERNATION_PREPARE: | |
3703 | case PM_SUSPEND_PREPARE: | |
3704 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ | |
5afff48b | 3705 | rcu_expedite_gp(); |
d1d74d14 BP |
3706 | break; |
3707 | case PM_POST_HIBERNATION: | |
3708 | case PM_POST_SUSPEND: | |
5afff48b PM |
3709 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ |
3710 | rcu_unexpedite_gp(); | |
d1d74d14 BP |
3711 | break; |
3712 | default: | |
3713 | break; | |
3714 | } | |
3715 | return NOTIFY_OK; | |
3716 | } | |
3717 | ||
b3dbec76 | 3718 | /* |
9386c0b7 | 3719 | * Spawn the kthreads that handle each RCU flavor's grace periods. |
b3dbec76 PM |
3720 | */ |
3721 | static int __init rcu_spawn_gp_kthread(void) | |
3722 | { | |
3723 | unsigned long flags; | |
a94844b2 | 3724 | int kthread_prio_in = kthread_prio; |
b3dbec76 PM |
3725 | struct rcu_node *rnp; |
3726 | struct rcu_state *rsp; | |
a94844b2 | 3727 | struct sched_param sp; |
b3dbec76 PM |
3728 | struct task_struct *t; |
3729 | ||
a94844b2 | 3730 | /* Force priority into range. */ |
c7cd161e JFG |
3731 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 2 |
3732 | && IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)) | |
3733 | kthread_prio = 2; | |
3734 | else if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
a94844b2 PM |
3735 | kthread_prio = 1; |
3736 | else if (kthread_prio < 0) | |
3737 | kthread_prio = 0; | |
3738 | else if (kthread_prio > 99) | |
3739 | kthread_prio = 99; | |
c7cd161e | 3740 | |
a94844b2 PM |
3741 | if (kthread_prio != kthread_prio_in) |
3742 | pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n", | |
3743 | kthread_prio, kthread_prio_in); | |
3744 | ||
9386c0b7 | 3745 | rcu_scheduler_fully_active = 1; |
b3dbec76 | 3746 | for_each_rcu_flavor(rsp) { |
a94844b2 | 3747 | t = kthread_create(rcu_gp_kthread, rsp, "%s", rsp->name); |
b3dbec76 PM |
3748 | BUG_ON(IS_ERR(t)); |
3749 | rnp = rcu_get_root(rsp); | |
6cf10081 | 3750 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b3dbec76 | 3751 | rsp->gp_kthread = t; |
a94844b2 PM |
3752 | if (kthread_prio) { |
3753 | sp.sched_priority = kthread_prio; | |
3754 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
3755 | } | |
67c583a7 | 3756 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
e11f1335 | 3757 | wake_up_process(t); |
b3dbec76 | 3758 | } |
35ce7f29 | 3759 | rcu_spawn_nocb_kthreads(); |
9386c0b7 | 3760 | rcu_spawn_boost_kthreads(); |
b3dbec76 PM |
3761 | return 0; |
3762 | } | |
3763 | early_initcall(rcu_spawn_gp_kthread); | |
3764 | ||
bbad9379 | 3765 | /* |
52d7e48b PM |
3766 | * This function is invoked towards the end of the scheduler's |
3767 | * initialization process. Before this is called, the idle task might | |
3768 | * contain synchronous grace-period primitives (during which time, this idle | |
3769 | * task is booting the system, and such primitives are no-ops). After this | |
3770 | * function is called, any synchronous grace-period primitives are run as | |
3771 | * expedited, with the requesting task driving the grace period forward. | |
900b1028 | 3772 | * A later core_initcall() rcu_set_runtime_mode() will switch to full |
52d7e48b | 3773 | * runtime RCU functionality. |
bbad9379 PM |
3774 | */ |
3775 | void rcu_scheduler_starting(void) | |
3776 | { | |
3777 | WARN_ON(num_online_cpus() != 1); | |
3778 | WARN_ON(nr_context_switches() > 0); | |
52d7e48b PM |
3779 | rcu_test_sync_prims(); |
3780 | rcu_scheduler_active = RCU_SCHEDULER_INIT; | |
3781 | rcu_test_sync_prims(); | |
bbad9379 PM |
3782 | } |
3783 | ||
64db4cff PM |
3784 | /* |
3785 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
3786 | */ | |
a87f203e | 3787 | static void __init rcu_init_one(struct rcu_state *rsp) |
64db4cff | 3788 | { |
cb007102 AG |
3789 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
3790 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
3791 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
3792 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
199977bf | 3793 | |
199977bf | 3794 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ |
64db4cff PM |
3795 | int cpustride = 1; |
3796 | int i; | |
3797 | int j; | |
3798 | struct rcu_node *rnp; | |
3799 | ||
05b84aec | 3800 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 3801 | |
3eaaaf6c PM |
3802 | /* Silence gcc 4.8 false positive about array index out of range. */ |
3803 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
3804 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 3805 | |
64db4cff PM |
3806 | /* Initialize the level-tracking arrays. */ |
3807 | ||
f885b7f2 | 3808 | for (i = 1; i < rcu_num_lvls; i++) |
41f5c631 PM |
3809 | rsp->level[i] = rsp->level[i - 1] + num_rcu_lvl[i - 1]; |
3810 | rcu_init_levelspread(levelspread, num_rcu_lvl); | |
64db4cff PM |
3811 | |
3812 | /* Initialize the elements themselves, starting from the leaves. */ | |
3813 | ||
f885b7f2 | 3814 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 3815 | cpustride *= levelspread[i]; |
64db4cff | 3816 | rnp = rsp->level[i]; |
41f5c631 | 3817 | for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) { |
67c583a7 BF |
3818 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
3819 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 3820 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
3821 | raw_spin_lock_init(&rnp->fqslock); |
3822 | lockdep_set_class_and_name(&rnp->fqslock, | |
3823 | &rcu_fqs_class[i], fqs[i]); | |
de30ad51 | 3824 | rnp->gp_seq = rsp->gp_seq; |
7a1d0f23 | 3825 | rnp->gp_seq_needed = rsp->gp_seq; |
d43a5d32 | 3826 | rnp->completedqs = rsp->gp_seq; |
64db4cff PM |
3827 | rnp->qsmask = 0; |
3828 | rnp->qsmaskinit = 0; | |
3829 | rnp->grplo = j * cpustride; | |
3830 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
3831 | if (rnp->grphi >= nr_cpu_ids) |
3832 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
3833 | if (i == 0) { |
3834 | rnp->grpnum = 0; | |
3835 | rnp->grpmask = 0; | |
3836 | rnp->parent = NULL; | |
3837 | } else { | |
199977bf | 3838 | rnp->grpnum = j % levelspread[i - 1]; |
64db4cff PM |
3839 | rnp->grpmask = 1UL << rnp->grpnum; |
3840 | rnp->parent = rsp->level[i - 1] + | |
199977bf | 3841 | j / levelspread[i - 1]; |
64db4cff PM |
3842 | } |
3843 | rnp->level = i; | |
12f5f524 | 3844 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 3845 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
3846 | init_waitqueue_head(&rnp->exp_wq[0]); |
3847 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
3848 | init_waitqueue_head(&rnp->exp_wq[2]); |
3849 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 3850 | spin_lock_init(&rnp->exp_lock); |
64db4cff PM |
3851 | } |
3852 | } | |
0c34029a | 3853 | |
abedf8e2 PG |
3854 | init_swait_queue_head(&rsp->gp_wq); |
3855 | init_swait_queue_head(&rsp->expedited_wq); | |
5b4c11d5 | 3856 | rnp = rcu_first_leaf_node(rsp); |
0c34029a | 3857 | for_each_possible_cpu(i) { |
4a90a068 | 3858 | while (i > rnp->grphi) |
0c34029a | 3859 | rnp++; |
394f99a9 | 3860 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
3861 | rcu_boot_init_percpu_data(i, rsp); |
3862 | } | |
6ce75a23 | 3863 | list_add(&rsp->flavors, &rcu_struct_flavors); |
64db4cff PM |
3864 | } |
3865 | ||
f885b7f2 PM |
3866 | /* |
3867 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 3868 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
3869 | * the ->node array in the rcu_state structure. |
3870 | */ | |
3871 | static void __init rcu_init_geometry(void) | |
3872 | { | |
026ad283 | 3873 | ulong d; |
f885b7f2 | 3874 | int i; |
05b84aec | 3875 | int rcu_capacity[RCU_NUM_LVLS]; |
f885b7f2 | 3876 | |
026ad283 PM |
3877 | /* |
3878 | * Initialize any unspecified boot parameters. | |
3879 | * The default values of jiffies_till_first_fqs and | |
3880 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
3881 | * value, which is a function of HZ, then adding one for each | |
3882 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
3883 | */ | |
3884 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
3885 | if (jiffies_till_first_fqs == ULONG_MAX) | |
3886 | jiffies_till_first_fqs = d; | |
3887 | if (jiffies_till_next_fqs == ULONG_MAX) | |
3888 | jiffies_till_next_fqs = d; | |
3889 | ||
f885b7f2 | 3890 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 3891 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 3892 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 3893 | return; |
a7538352 | 3894 | pr_info("Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n", |
39479098 | 3895 | rcu_fanout_leaf, nr_cpu_ids); |
f885b7f2 | 3896 | |
f885b7f2 | 3897 | /* |
ee968ac6 PM |
3898 | * The boot-time rcu_fanout_leaf parameter must be at least two |
3899 | * and cannot exceed the number of bits in the rcu_node masks. | |
3900 | * Complain and fall back to the compile-time values if this | |
3901 | * limit is exceeded. | |
f885b7f2 | 3902 | */ |
ee968ac6 | 3903 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 3904 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 3905 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
3906 | WARN_ON(1); |
3907 | return; | |
3908 | } | |
3909 | ||
f885b7f2 PM |
3910 | /* |
3911 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 3912 | * with the given number of levels. |
f885b7f2 | 3913 | */ |
9618138b | 3914 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 3915 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 3916 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
3917 | |
3918 | /* | |
75cf15a4 | 3919 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 3920 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 3921 | */ |
ee968ac6 PM |
3922 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
3923 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
3924 | WARN_ON(1); | |
3925 | return; | |
3926 | } | |
f885b7f2 | 3927 | |
679f9858 | 3928 | /* Calculate the number of levels in the tree. */ |
9618138b | 3929 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 3930 | } |
9618138b | 3931 | rcu_num_lvls = i + 1; |
679f9858 | 3932 | |
f885b7f2 | 3933 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 3934 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 3935 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
3936 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
3937 | } | |
f885b7f2 PM |
3938 | |
3939 | /* Calculate the total number of rcu_node structures. */ | |
3940 | rcu_num_nodes = 0; | |
679f9858 | 3941 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 3942 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
3943 | } |
3944 | ||
a3dc2948 PM |
3945 | /* |
3946 | * Dump out the structure of the rcu_node combining tree associated | |
3947 | * with the rcu_state structure referenced by rsp. | |
3948 | */ | |
3949 | static void __init rcu_dump_rcu_node_tree(struct rcu_state *rsp) | |
3950 | { | |
3951 | int level = 0; | |
3952 | struct rcu_node *rnp; | |
3953 | ||
3954 | pr_info("rcu_node tree layout dump\n"); | |
3955 | pr_info(" "); | |
3956 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
3957 | if (rnp->level != level) { | |
3958 | pr_cont("\n"); | |
3959 | pr_info(" "); | |
3960 | level = rnp->level; | |
3961 | } | |
3962 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
3963 | } | |
3964 | pr_cont("\n"); | |
3965 | } | |
3966 | ||
ad7c946b | 3967 | struct workqueue_struct *rcu_gp_wq; |
25f3d7ef | 3968 | struct workqueue_struct *rcu_par_gp_wq; |
ad7c946b | 3969 | |
9f680ab4 | 3970 | void __init rcu_init(void) |
64db4cff | 3971 | { |
017c4261 | 3972 | int cpu; |
9f680ab4 | 3973 | |
47627678 PM |
3974 | rcu_early_boot_tests(); |
3975 | ||
f41d911f | 3976 | rcu_bootup_announce(); |
f885b7f2 | 3977 | rcu_init_geometry(); |
45975c7d | 3978 | rcu_init_one(&rcu_state); |
a3dc2948 | 3979 | if (dump_tree) |
45975c7d | 3980 | rcu_dump_rcu_node_tree(&rcu_state); |
b5b39360 | 3981 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
3982 | |
3983 | /* | |
3984 | * We don't need protection against CPU-hotplug here because | |
3985 | * this is called early in boot, before either interrupts | |
3986 | * or the scheduler are operational. | |
3987 | */ | |
d1d74d14 | 3988 | pm_notifier(rcu_pm_notify, 0); |
7ec99de3 | 3989 | for_each_online_cpu(cpu) { |
4df83742 | 3990 | rcutree_prepare_cpu(cpu); |
7ec99de3 | 3991 | rcu_cpu_starting(cpu); |
9b9500da | 3992 | rcutree_online_cpu(cpu); |
7ec99de3 | 3993 | } |
ad7c946b PM |
3994 | |
3995 | /* Create workqueue for expedited GPs and for Tree SRCU. */ | |
3996 | rcu_gp_wq = alloc_workqueue("rcu_gp", WQ_MEM_RECLAIM, 0); | |
3997 | WARN_ON(!rcu_gp_wq); | |
25f3d7ef PM |
3998 | rcu_par_gp_wq = alloc_workqueue("rcu_par_gp", WQ_MEM_RECLAIM, 0); |
3999 | WARN_ON(!rcu_par_gp_wq); | |
64db4cff PM |
4000 | } |
4001 | ||
3549c2bc | 4002 | #include "tree_exp.h" |
4102adab | 4003 | #include "tree_plugin.h" |