Commit | Line | Data |
---|---|---|
22e40925 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
64db4cff PM |
2 | /* |
3 | * Read-Copy Update mechanism for mutual exclusion | |
4 | * | |
64db4cff PM |
5 | * Copyright IBM Corporation, 2008 |
6 | * | |
7 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | |
8 | * Manfred Spraul <manfred@colorfullife.com> | |
22e40925 | 9 | * Paul E. McKenney <paulmck@linux.ibm.com> Hierarchical version |
64db4cff | 10 | * |
22e40925 | 11 | * Based on the original work by Paul McKenney <paulmck@linux.ibm.com> |
64db4cff PM |
12 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. |
13 | * | |
14 | * For detailed explanation of Read-Copy Update mechanism see - | |
a71fca58 | 15 | * Documentation/RCU |
64db4cff | 16 | */ |
a7538352 JP |
17 | |
18 | #define pr_fmt(fmt) "rcu: " fmt | |
19 | ||
64db4cff PM |
20 | #include <linux/types.h> |
21 | #include <linux/kernel.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/spinlock.h> | |
24 | #include <linux/smp.h> | |
f9411ebe | 25 | #include <linux/rcupdate_wait.h> |
64db4cff PM |
26 | #include <linux/interrupt.h> |
27 | #include <linux/sched.h> | |
b17b0153 | 28 | #include <linux/sched/debug.h> |
c1dc0b9c | 29 | #include <linux/nmi.h> |
8826f3b0 | 30 | #include <linux/atomic.h> |
64db4cff | 31 | #include <linux/bitops.h> |
9984de1a | 32 | #include <linux/export.h> |
64db4cff PM |
33 | #include <linux/completion.h> |
34 | #include <linux/moduleparam.h> | |
35 | #include <linux/percpu.h> | |
36 | #include <linux/notifier.h> | |
37 | #include <linux/cpu.h> | |
38 | #include <linux/mutex.h> | |
39 | #include <linux/time.h> | |
bbad9379 | 40 | #include <linux/kernel_stat.h> |
a26ac245 PM |
41 | #include <linux/wait.h> |
42 | #include <linux/kthread.h> | |
ae7e81c0 | 43 | #include <uapi/linux/sched/types.h> |
268bb0ce | 44 | #include <linux/prefetch.h> |
3d3b7db0 PM |
45 | #include <linux/delay.h> |
46 | #include <linux/stop_machine.h> | |
661a85dc | 47 | #include <linux/random.h> |
af658dca | 48 | #include <linux/trace_events.h> |
d1d74d14 | 49 | #include <linux/suspend.h> |
a278d471 | 50 | #include <linux/ftrace.h> |
d3052109 | 51 | #include <linux/tick.h> |
2ccaff10 | 52 | #include <linux/sysrq.h> |
c13324a5 | 53 | #include <linux/kprobes.h> |
48d07c04 SAS |
54 | #include <linux/gfp.h> |
55 | #include <linux/oom.h> | |
56 | #include <linux/smpboot.h> | |
57 | #include <linux/jiffies.h> | |
58 | #include <linux/sched/isolation.h> | |
59 | #include "../time/tick-internal.h" | |
64db4cff | 60 | |
4102adab | 61 | #include "tree.h" |
29c00b4a | 62 | #include "rcu.h" |
9f77da9f | 63 | |
4102adab PM |
64 | #ifdef MODULE_PARAM_PREFIX |
65 | #undef MODULE_PARAM_PREFIX | |
66 | #endif | |
67 | #define MODULE_PARAM_PREFIX "rcutree." | |
68 | ||
64db4cff PM |
69 | /* Data structures. */ |
70 | ||
f7f7bac9 | 71 | /* |
dc5a4f29 PM |
72 | * Steal a bit from the bottom of ->dynticks for idle entry/exit |
73 | * control. Initially this is for TLB flushing. | |
f7f7bac9 | 74 | */ |
dc5a4f29 PM |
75 | #define RCU_DYNTICK_CTRL_MASK 0x1 |
76 | #define RCU_DYNTICK_CTRL_CTR (RCU_DYNTICK_CTRL_MASK + 1) | |
77 | #ifndef rcu_eqs_special_exit | |
78 | #define rcu_eqs_special_exit() do { } while (0) | |
a8a29b3b AB |
79 | #endif |
80 | ||
4c5273bf PM |
81 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, rcu_data) = { |
82 | .dynticks_nesting = 1, | |
83 | .dynticks_nmi_nesting = DYNTICK_IRQ_NONIDLE, | |
dc5a4f29 | 84 | .dynticks = ATOMIC_INIT(RCU_DYNTICK_CTRL_CTR), |
4c5273bf | 85 | }; |
358be2d3 PM |
86 | struct rcu_state rcu_state = { |
87 | .level = { &rcu_state.node[0] }, | |
358be2d3 PM |
88 | .gp_state = RCU_GP_IDLE, |
89 | .gp_seq = (0UL - 300UL) << RCU_SEQ_CTR_SHIFT, | |
90 | .barrier_mutex = __MUTEX_INITIALIZER(rcu_state.barrier_mutex), | |
91 | .name = RCU_NAME, | |
92 | .abbr = RCU_ABBR, | |
93 | .exp_mutex = __MUTEX_INITIALIZER(rcu_state.exp_mutex), | |
94 | .exp_wake_mutex = __MUTEX_INITIALIZER(rcu_state.exp_wake_mutex), | |
894d45bb | 95 | .ofl_lock = __RAW_SPIN_LOCK_UNLOCKED(rcu_state.ofl_lock), |
358be2d3 | 96 | }; |
27f4d280 | 97 | |
a3dc2948 PM |
98 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
99 | static bool dump_tree; | |
100 | module_param(dump_tree, bool, 0444); | |
48d07c04 SAS |
101 | /* By default, use RCU_SOFTIRQ instead of rcuc kthreads. */ |
102 | static bool use_softirq = 1; | |
103 | module_param(use_softirq, bool, 0444); | |
7fa27001 PM |
104 | /* Control rcu_node-tree auto-balancing at boot time. */ |
105 | static bool rcu_fanout_exact; | |
106 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
107 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
108 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 109 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 110 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 | 111 | /* Number of rcu_nodes at specified level. */ |
e95d68d2 | 112 | int num_rcu_lvl[] = NUM_RCU_LVL_INIT; |
f885b7f2 PM |
113 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
114 | ||
b0d30417 | 115 | /* |
52d7e48b PM |
116 | * The rcu_scheduler_active variable is initialized to the value |
117 | * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the | |
118 | * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE, | |
119 | * RCU can assume that there is but one task, allowing RCU to (for example) | |
0d95092c | 120 | * optimize synchronize_rcu() to a simple barrier(). When this variable |
52d7e48b PM |
121 | * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required |
122 | * to detect real grace periods. This variable is also used to suppress | |
123 | * boot-time false positives from lockdep-RCU error checking. Finally, it | |
124 | * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU | |
125 | * is fully initialized, including all of its kthreads having been spawned. | |
b0d30417 | 126 | */ |
bbad9379 PM |
127 | int rcu_scheduler_active __read_mostly; |
128 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
129 | ||
b0d30417 PM |
130 | /* |
131 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
132 | * during the early_initcall() processing, which is after the scheduler | |
133 | * is capable of creating new tasks. So RCU processing (for example, | |
134 | * creating tasks for RCU priority boosting) must be delayed until after | |
135 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
136 | * currently delay invocation of any RCU callbacks until after this point. | |
137 | * | |
138 | * It might later prove better for people registering RCU callbacks during | |
139 | * early boot to take responsibility for these callbacks, but one step at | |
140 | * a time. | |
141 | */ | |
142 | static int rcu_scheduler_fully_active __read_mostly; | |
143 | ||
b50912d0 PM |
144 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
145 | unsigned long gps, unsigned long flags); | |
0aa04b05 PM |
146 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
147 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 148 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 | 149 | static void invoke_rcu_core(void); |
aff4e9ed | 150 | static void invoke_rcu_callbacks(struct rcu_data *rdp); |
63d4c8c9 | 151 | static void rcu_report_exp_rdp(struct rcu_data *rdp); |
3549c2bc | 152 | static void sync_sched_exp_online_cleanup(int cpu); |
a26ac245 | 153 | |
a94844b2 | 154 | /* rcuc/rcub kthread realtime priority */ |
26730f55 | 155 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; |
3ffe3d1a | 156 | module_param(kthread_prio, int, 0444); |
a94844b2 | 157 | |
8d7dc928 | 158 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd | 159 | |
90040c9e PM |
160 | static int gp_preinit_delay; |
161 | module_param(gp_preinit_delay, int, 0444); | |
162 | static int gp_init_delay; | |
163 | module_param(gp_init_delay, int, 0444); | |
164 | static int gp_cleanup_delay; | |
165 | module_param(gp_cleanup_delay, int, 0444); | |
0f41c0dd | 166 | |
4cf439a2 | 167 | /* Retrieve RCU kthreads priority for rcutorture */ |
4babd855 JFG |
168 | int rcu_get_gp_kthreads_prio(void) |
169 | { | |
170 | return kthread_prio; | |
171 | } | |
172 | EXPORT_SYMBOL_GPL(rcu_get_gp_kthreads_prio); | |
173 | ||
eab128e8 PM |
174 | /* |
175 | * Number of grace periods between delays, normalized by the duration of | |
bfd090be | 176 | * the delay. The longer the delay, the more the grace periods between |
eab128e8 PM |
177 | * each delay. The reason for this normalization is that it means that, |
178 | * for non-zero delays, the overall slowdown of grace periods is constant | |
179 | * regardless of the duration of the delay. This arrangement balances | |
180 | * the need for long delays to increase some race probabilities with the | |
181 | * need for fast grace periods to increase other race probabilities. | |
182 | */ | |
183 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */ | |
37745d28 | 184 | |
0aa04b05 PM |
185 | /* |
186 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
187 | * This will not be stable unless the rcu_node structure's ->lock is | |
188 | * held, but the bit corresponding to the current CPU will be stable | |
189 | * in most contexts. | |
190 | */ | |
191 | unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) | |
192 | { | |
7d0ae808 | 193 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
194 | } |
195 | ||
fc2219d4 | 196 | /* |
7d0ae808 | 197 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
198 | * permit this function to be invoked without holding the root rcu_node |
199 | * structure's ->lock, but of course results can be subject to change. | |
200 | */ | |
de8e8730 | 201 | static int rcu_gp_in_progress(void) |
fc2219d4 | 202 | { |
de8e8730 | 203 | return rcu_seq_state(rcu_seq_current(&rcu_state.gp_seq)); |
b1f77b05 IM |
204 | } |
205 | ||
903ee83d PM |
206 | /* |
207 | * Return the number of callbacks queued on the specified CPU. | |
208 | * Handles both the nocbs and normal cases. | |
209 | */ | |
210 | static long rcu_get_n_cbs_cpu(int cpu) | |
211 | { | |
212 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); | |
213 | ||
214 | if (rcu_segcblist_is_enabled(&rdp->cblist)) /* Online normal CPU? */ | |
215 | return rcu_segcblist_n_cbs(&rdp->cblist); | |
216 | return rcu_get_n_cbs_nocb_cpu(rdp); /* Works for offline, too. */ | |
217 | } | |
218 | ||
d28139c4 | 219 | void rcu_softirq_qs(void) |
b1f77b05 | 220 | { |
45975c7d | 221 | rcu_qs(); |
d28139c4 | 222 | rcu_preempt_deferred_qs(current); |
b1f77b05 | 223 | } |
64db4cff | 224 | |
2625d469 PM |
225 | /* |
226 | * Record entry into an extended quiescent state. This is only to be | |
227 | * called when not already in an extended quiescent state. | |
228 | */ | |
229 | static void rcu_dynticks_eqs_enter(void) | |
230 | { | |
dc5a4f29 | 231 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
b8c17e66 | 232 | int seq; |
2625d469 PM |
233 | |
234 | /* | |
b8c17e66 | 235 | * CPUs seeing atomic_add_return() must see prior RCU read-side |
2625d469 PM |
236 | * critical sections, and we also must force ordering with the |
237 | * next idle sojourn. | |
238 | */ | |
dc5a4f29 | 239 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks); |
b8c17e66 PM |
240 | /* Better be in an extended quiescent state! */ |
241 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
242 | (seq & RCU_DYNTICK_CTRL_CTR)); | |
243 | /* Better not have special action (TLB flush) pending! */ | |
244 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && | |
245 | (seq & RCU_DYNTICK_CTRL_MASK)); | |
2625d469 PM |
246 | } |
247 | ||
248 | /* | |
249 | * Record exit from an extended quiescent state. This is only to be | |
250 | * called from an extended quiescent state. | |
251 | */ | |
252 | static void rcu_dynticks_eqs_exit(void) | |
253 | { | |
dc5a4f29 | 254 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
b8c17e66 | 255 | int seq; |
2625d469 PM |
256 | |
257 | /* | |
b8c17e66 | 258 | * CPUs seeing atomic_add_return() must see prior idle sojourns, |
2625d469 PM |
259 | * and we also must force ordering with the next RCU read-side |
260 | * critical section. | |
261 | */ | |
dc5a4f29 | 262 | seq = atomic_add_return(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks); |
b8c17e66 PM |
263 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
264 | !(seq & RCU_DYNTICK_CTRL_CTR)); | |
265 | if (seq & RCU_DYNTICK_CTRL_MASK) { | |
dc5a4f29 | 266 | atomic_andnot(RCU_DYNTICK_CTRL_MASK, &rdp->dynticks); |
b8c17e66 PM |
267 | smp_mb__after_atomic(); /* _exit after clearing mask. */ |
268 | /* Prefer duplicate flushes to losing a flush. */ | |
269 | rcu_eqs_special_exit(); | |
270 | } | |
2625d469 PM |
271 | } |
272 | ||
273 | /* | |
274 | * Reset the current CPU's ->dynticks counter to indicate that the | |
275 | * newly onlined CPU is no longer in an extended quiescent state. | |
276 | * This will either leave the counter unchanged, or increment it | |
277 | * to the next non-quiescent value. | |
278 | * | |
279 | * The non-atomic test/increment sequence works because the upper bits | |
280 | * of the ->dynticks counter are manipulated only by the corresponding CPU, | |
281 | * or when the corresponding CPU is offline. | |
282 | */ | |
283 | static void rcu_dynticks_eqs_online(void) | |
284 | { | |
dc5a4f29 | 285 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
2625d469 | 286 | |
dc5a4f29 | 287 | if (atomic_read(&rdp->dynticks) & RCU_DYNTICK_CTRL_CTR) |
2625d469 | 288 | return; |
dc5a4f29 | 289 | atomic_add(RCU_DYNTICK_CTRL_CTR, &rdp->dynticks); |
2625d469 PM |
290 | } |
291 | ||
02a5c550 PM |
292 | /* |
293 | * Is the current CPU in an extended quiescent state? | |
294 | * | |
295 | * No ordering, as we are sampling CPU-local information. | |
296 | */ | |
297 | bool rcu_dynticks_curr_cpu_in_eqs(void) | |
298 | { | |
dc5a4f29 | 299 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
02a5c550 | 300 | |
dc5a4f29 | 301 | return !(atomic_read(&rdp->dynticks) & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
302 | } |
303 | ||
8b2f63ab PM |
304 | /* |
305 | * Snapshot the ->dynticks counter with full ordering so as to allow | |
306 | * stable comparison of this counter with past and future snapshots. | |
307 | */ | |
dc5a4f29 | 308 | int rcu_dynticks_snap(struct rcu_data *rdp) |
8b2f63ab | 309 | { |
dc5a4f29 | 310 | int snap = atomic_add_return(0, &rdp->dynticks); |
8b2f63ab | 311 | |
b8c17e66 | 312 | return snap & ~RCU_DYNTICK_CTRL_MASK; |
8b2f63ab PM |
313 | } |
314 | ||
02a5c550 PM |
315 | /* |
316 | * Return true if the snapshot returned from rcu_dynticks_snap() | |
317 | * indicates that RCU is in an extended quiescent state. | |
318 | */ | |
319 | static bool rcu_dynticks_in_eqs(int snap) | |
320 | { | |
b8c17e66 | 321 | return !(snap & RCU_DYNTICK_CTRL_CTR); |
02a5c550 PM |
322 | } |
323 | ||
324 | /* | |
dc5a4f29 | 325 | * Return true if the CPU corresponding to the specified rcu_data |
02a5c550 PM |
326 | * structure has spent some time in an extended quiescent state since |
327 | * rcu_dynticks_snap() returned the specified snapshot. | |
328 | */ | |
dc5a4f29 | 329 | static bool rcu_dynticks_in_eqs_since(struct rcu_data *rdp, int snap) |
02a5c550 | 330 | { |
dc5a4f29 | 331 | return snap != rcu_dynticks_snap(rdp); |
02a5c550 PM |
332 | } |
333 | ||
b8c17e66 PM |
334 | /* |
335 | * Set the special (bottom) bit of the specified CPU so that it | |
336 | * will take special action (such as flushing its TLB) on the | |
337 | * next exit from an extended quiescent state. Returns true if | |
338 | * the bit was successfully set, or false if the CPU was not in | |
339 | * an extended quiescent state. | |
340 | */ | |
341 | bool rcu_eqs_special_set(int cpu) | |
342 | { | |
343 | int old; | |
344 | int new; | |
dc5a4f29 | 345 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
b8c17e66 PM |
346 | |
347 | do { | |
dc5a4f29 | 348 | old = atomic_read(&rdp->dynticks); |
b8c17e66 PM |
349 | if (old & RCU_DYNTICK_CTRL_CTR) |
350 | return false; | |
351 | new = old | RCU_DYNTICK_CTRL_MASK; | |
dc5a4f29 | 352 | } while (atomic_cmpxchg(&rdp->dynticks, old, new) != old); |
b8c17e66 | 353 | return true; |
6563de9d | 354 | } |
5cd37193 | 355 | |
4a81e832 PM |
356 | /* |
357 | * Let the RCU core know that this CPU has gone through the scheduler, | |
358 | * which is a quiescent state. This is called when the need for a | |
359 | * quiescent state is urgent, so we burn an atomic operation and full | |
360 | * memory barriers to let the RCU core know about it, regardless of what | |
361 | * this CPU might (or might not) do in the near future. | |
362 | * | |
0f9be8ca | 363 | * We inform the RCU core by emulating a zero-duration dyntick-idle period. |
46a5d164 | 364 | * |
3b57a399 | 365 | * The caller must have disabled interrupts and must not be idle. |
4a81e832 | 366 | */ |
395a2f09 | 367 | static void __maybe_unused rcu_momentary_dyntick_idle(void) |
4a81e832 | 368 | { |
3b57a399 PM |
369 | int special; |
370 | ||
2dba13f0 | 371 | raw_cpu_write(rcu_data.rcu_need_heavy_qs, false); |
dc5a4f29 PM |
372 | special = atomic_add_return(2 * RCU_DYNTICK_CTRL_CTR, |
373 | &this_cpu_ptr(&rcu_data)->dynticks); | |
3b57a399 PM |
374 | /* It is illegal to call this from idle state. */ |
375 | WARN_ON_ONCE(!(special & RCU_DYNTICK_CTRL_CTR)); | |
3e310098 | 376 | rcu_preempt_deferred_qs(current); |
4a81e832 PM |
377 | } |
378 | ||
45975c7d PM |
379 | /** |
380 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle | |
bb73c52b | 381 | * |
45975c7d PM |
382 | * If the current CPU is idle or running at a first-level (not nested) |
383 | * interrupt from idle, return true. The caller must have at least | |
384 | * disabled preemption. | |
5cd37193 | 385 | */ |
45975c7d | 386 | static int rcu_is_cpu_rrupt_from_idle(void) |
5cd37193 | 387 | { |
4c5273bf PM |
388 | return __this_cpu_read(rcu_data.dynticks_nesting) <= 0 && |
389 | __this_cpu_read(rcu_data.dynticks_nmi_nesting) <= 1; | |
5cd37193 | 390 | } |
5cd37193 | 391 | |
17c7798b PM |
392 | #define DEFAULT_RCU_BLIMIT 10 /* Maximum callbacks per rcu_do_batch. */ |
393 | static long blimit = DEFAULT_RCU_BLIMIT; | |
394 | #define DEFAULT_RCU_QHIMARK 10000 /* If this many pending, ignore blimit. */ | |
395 | static long qhimark = DEFAULT_RCU_QHIMARK; | |
396 | #define DEFAULT_RCU_QLOMARK 100 /* Once only this many pending, use blimit. */ | |
397 | static long qlowmark = DEFAULT_RCU_QLOMARK; | |
64db4cff | 398 | |
878d7439 ED |
399 | module_param(blimit, long, 0444); |
400 | module_param(qhimark, long, 0444); | |
401 | module_param(qlowmark, long, 0444); | |
3d76c082 | 402 | |
026ad283 PM |
403 | static ulong jiffies_till_first_fqs = ULONG_MAX; |
404 | static ulong jiffies_till_next_fqs = ULONG_MAX; | |
8c7c4829 | 405 | static bool rcu_kick_kthreads; |
d40011f6 | 406 | |
c06aed0e PM |
407 | /* |
408 | * How long the grace period must be before we start recruiting | |
409 | * quiescent-state help from rcu_note_context_switch(). | |
410 | */ | |
411 | static ulong jiffies_till_sched_qs = ULONG_MAX; | |
412 | module_param(jiffies_till_sched_qs, ulong, 0444); | |
85f2b60c | 413 | static ulong jiffies_to_sched_qs; /* See adjust_jiffies_till_sched_qs(). */ |
c06aed0e PM |
414 | module_param(jiffies_to_sched_qs, ulong, 0444); /* Display only! */ |
415 | ||
416 | /* | |
417 | * Make sure that we give the grace-period kthread time to detect any | |
418 | * idle CPUs before taking active measures to force quiescent states. | |
419 | * However, don't go below 100 milliseconds, adjusted upwards for really | |
420 | * large systems. | |
421 | */ | |
422 | static void adjust_jiffies_till_sched_qs(void) | |
423 | { | |
424 | unsigned long j; | |
425 | ||
426 | /* If jiffies_till_sched_qs was specified, respect the request. */ | |
427 | if (jiffies_till_sched_qs != ULONG_MAX) { | |
428 | WRITE_ONCE(jiffies_to_sched_qs, jiffies_till_sched_qs); | |
429 | return; | |
430 | } | |
85f2b60c | 431 | /* Otherwise, set to third fqs scan, but bound below on large system. */ |
c06aed0e PM |
432 | j = READ_ONCE(jiffies_till_first_fqs) + |
433 | 2 * READ_ONCE(jiffies_till_next_fqs); | |
434 | if (j < HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV) | |
435 | j = HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
436 | pr_info("RCU calculated value of scheduler-enlistment delay is %ld jiffies.\n", j); | |
437 | WRITE_ONCE(jiffies_to_sched_qs, j); | |
438 | } | |
439 | ||
67abb96c BP |
440 | static int param_set_first_fqs_jiffies(const char *val, const struct kernel_param *kp) |
441 | { | |
442 | ulong j; | |
443 | int ret = kstrtoul(val, 0, &j); | |
444 | ||
c06aed0e | 445 | if (!ret) { |
67abb96c | 446 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : j); |
c06aed0e PM |
447 | adjust_jiffies_till_sched_qs(); |
448 | } | |
67abb96c BP |
449 | return ret; |
450 | } | |
451 | ||
452 | static int param_set_next_fqs_jiffies(const char *val, const struct kernel_param *kp) | |
453 | { | |
454 | ulong j; | |
455 | int ret = kstrtoul(val, 0, &j); | |
456 | ||
c06aed0e | 457 | if (!ret) { |
67abb96c | 458 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : (j ?: 1)); |
c06aed0e PM |
459 | adjust_jiffies_till_sched_qs(); |
460 | } | |
67abb96c BP |
461 | return ret; |
462 | } | |
463 | ||
464 | static struct kernel_param_ops first_fqs_jiffies_ops = { | |
465 | .set = param_set_first_fqs_jiffies, | |
466 | .get = param_get_ulong, | |
467 | }; | |
468 | ||
469 | static struct kernel_param_ops next_fqs_jiffies_ops = { | |
470 | .set = param_set_next_fqs_jiffies, | |
471 | .get = param_get_ulong, | |
472 | }; | |
473 | ||
474 | module_param_cb(jiffies_till_first_fqs, &first_fqs_jiffies_ops, &jiffies_till_first_fqs, 0644); | |
475 | module_param_cb(jiffies_till_next_fqs, &next_fqs_jiffies_ops, &jiffies_till_next_fqs, 0644); | |
8c7c4829 | 476 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 477 | |
8ff0b907 | 478 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)); |
e3950ecd | 479 | static int rcu_pending(void); |
64db4cff PM |
480 | |
481 | /* | |
17ef2fe9 | 482 | * Return the number of RCU GPs completed thus far for debug & stats. |
64db4cff | 483 | */ |
17ef2fe9 | 484 | unsigned long rcu_get_gp_seq(void) |
917963d0 | 485 | { |
16fc9c60 | 486 | return READ_ONCE(rcu_state.gp_seq); |
917963d0 | 487 | } |
17ef2fe9 | 488 | EXPORT_SYMBOL_GPL(rcu_get_gp_seq); |
917963d0 | 489 | |
291783b8 PM |
490 | /* |
491 | * Return the number of RCU expedited batches completed thus far for | |
492 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
493 | * numbers mean idle. The value returned will thus be roughly double | |
494 | * the cumulative batches since boot. | |
495 | */ | |
496 | unsigned long rcu_exp_batches_completed(void) | |
497 | { | |
16fc9c60 | 498 | return rcu_state.expedited_sequence; |
291783b8 PM |
499 | } |
500 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
501 | ||
fd897573 PM |
502 | /* |
503 | * Return the root node of the rcu_state structure. | |
504 | */ | |
505 | static struct rcu_node *rcu_get_root(void) | |
506 | { | |
507 | return &rcu_state.node[0]; | |
508 | } | |
509 | ||
69196019 PM |
510 | /* |
511 | * Convert a ->gp_state value to a character string. | |
512 | */ | |
513 | static const char *gp_state_getname(short gs) | |
514 | { | |
515 | if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names)) | |
516 | return "???"; | |
517 | return gp_state_names[gs]; | |
518 | } | |
519 | ||
ad0dc7f9 PM |
520 | /* |
521 | * Send along grace-period-related data for rcutorture diagnostics. | |
522 | */ | |
523 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
aebc8264 | 524 | unsigned long *gp_seq) |
ad0dc7f9 | 525 | { |
ad0dc7f9 PM |
526 | switch (test_type) { |
527 | case RCU_FLAVOR: | |
f7dd7d44 PM |
528 | *flags = READ_ONCE(rcu_state.gp_flags); |
529 | *gp_seq = rcu_seq_current(&rcu_state.gp_seq); | |
ad0dc7f9 PM |
530 | break; |
531 | default: | |
532 | break; | |
533 | } | |
ad0dc7f9 PM |
534 | } |
535 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
536 | ||
9b2e4f18 | 537 | /* |
215bba9f PM |
538 | * Enter an RCU extended quiescent state, which can be either the |
539 | * idle loop or adaptive-tickless usermode execution. | |
9b2e4f18 | 540 | * |
215bba9f PM |
541 | * We crowbar the ->dynticks_nmi_nesting field to zero to allow for |
542 | * the possibility of usermode upcalls having messed up our count | |
543 | * of interrupt nesting level during the prior busy period. | |
9b2e4f18 | 544 | */ |
215bba9f | 545 | static void rcu_eqs_enter(bool user) |
9b2e4f18 | 546 | { |
4c5273bf | 547 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
96d3fd0d | 548 | |
4c5273bf PM |
549 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting != DYNTICK_IRQ_NONIDLE); |
550 | WRITE_ONCE(rdp->dynticks_nmi_nesting, 0); | |
215bba9f | 551 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
4c5273bf PM |
552 | rdp->dynticks_nesting == 0); |
553 | if (rdp->dynticks_nesting != 1) { | |
554 | rdp->dynticks_nesting--; | |
215bba9f | 555 | return; |
9b2e4f18 | 556 | } |
96d3fd0d | 557 | |
b04db8e1 | 558 | lockdep_assert_irqs_disabled(); |
dc5a4f29 | 559 | trace_rcu_dyntick(TPS("Start"), rdp->dynticks_nesting, 0, rdp->dynticks); |
e68bbb26 | 560 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
b97d23c5 PM |
561 | rdp = this_cpu_ptr(&rcu_data); |
562 | do_nocb_deferred_wakeup(rdp); | |
198bbf81 | 563 | rcu_prepare_for_idle(); |
3e310098 | 564 | rcu_preempt_deferred_qs(current); |
4c5273bf | 565 | WRITE_ONCE(rdp->dynticks_nesting, 0); /* Avoid irq-access tearing. */ |
844ccdd7 | 566 | rcu_dynticks_eqs_enter(); |
176f8f7a | 567 | rcu_dynticks_task_enter(); |
64db4cff | 568 | } |
adf5091e FW |
569 | |
570 | /** | |
571 | * rcu_idle_enter - inform RCU that current CPU is entering idle | |
572 | * | |
573 | * Enter idle mode, in other words, -leave- the mode in which RCU | |
574 | * read-side critical sections can occur. (Though RCU read-side | |
575 | * critical sections can occur in irq handlers in idle, a possibility | |
576 | * handled by irq_enter() and irq_exit().) | |
577 | * | |
c0da313e PM |
578 | * If you add or remove a call to rcu_idle_enter(), be sure to test with |
579 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
580 | */ |
581 | void rcu_idle_enter(void) | |
582 | { | |
b04db8e1 | 583 | lockdep_assert_irqs_disabled(); |
cb349ca9 | 584 | rcu_eqs_enter(false); |
adf5091e | 585 | } |
64db4cff | 586 | |
d1ec4c34 | 587 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
588 | /** |
589 | * rcu_user_enter - inform RCU that we are resuming userspace. | |
590 | * | |
591 | * Enter RCU idle mode right before resuming userspace. No use of RCU | |
592 | * is permitted between this call and rcu_user_exit(). This way the | |
593 | * CPU doesn't need to maintain the tick for RCU maintenance purposes | |
594 | * when the CPU runs in userspace. | |
c0da313e PM |
595 | * |
596 | * If you add or remove a call to rcu_user_enter(), be sure to test with | |
597 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
598 | */ |
599 | void rcu_user_enter(void) | |
600 | { | |
b04db8e1 | 601 | lockdep_assert_irqs_disabled(); |
d4db30af | 602 | rcu_eqs_enter(true); |
adf5091e | 603 | } |
d1ec4c34 | 604 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 605 | |
cf7614e1 | 606 | /* |
fd581a91 | 607 | * If we are returning from the outermost NMI handler that interrupted an |
dc5a4f29 | 608 | * RCU-idle period, update rdp->dynticks and rdp->dynticks_nmi_nesting |
fd581a91 PM |
609 | * to let the RCU grace-period handling know that the CPU is back to |
610 | * being RCU-idle. | |
611 | * | |
cf7614e1 | 612 | * If you add or remove a call to rcu_nmi_exit_common(), be sure to test |
fd581a91 PM |
613 | * with CONFIG_RCU_EQS_DEBUG=y. |
614 | */ | |
cf7614e1 | 615 | static __always_inline void rcu_nmi_exit_common(bool irq) |
fd581a91 | 616 | { |
4c5273bf | 617 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
fd581a91 PM |
618 | |
619 | /* | |
620 | * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks. | |
621 | * (We are exiting an NMI handler, so RCU better be paying attention | |
622 | * to us!) | |
623 | */ | |
4c5273bf | 624 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting <= 0); |
fd581a91 PM |
625 | WARN_ON_ONCE(rcu_dynticks_curr_cpu_in_eqs()); |
626 | ||
627 | /* | |
628 | * If the nesting level is not 1, the CPU wasn't RCU-idle, so | |
629 | * leave it in non-RCU-idle state. | |
630 | */ | |
4c5273bf | 631 | if (rdp->dynticks_nmi_nesting != 1) { |
dc5a4f29 | 632 | trace_rcu_dyntick(TPS("--="), rdp->dynticks_nmi_nesting, rdp->dynticks_nmi_nesting - 2, rdp->dynticks); |
4c5273bf PM |
633 | WRITE_ONCE(rdp->dynticks_nmi_nesting, /* No store tearing. */ |
634 | rdp->dynticks_nmi_nesting - 2); | |
fd581a91 PM |
635 | return; |
636 | } | |
637 | ||
638 | /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */ | |
dc5a4f29 | 639 | trace_rcu_dyntick(TPS("Startirq"), rdp->dynticks_nmi_nesting, 0, rdp->dynticks); |
4c5273bf | 640 | WRITE_ONCE(rdp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */ |
cf7614e1 BP |
641 | |
642 | if (irq) | |
643 | rcu_prepare_for_idle(); | |
644 | ||
fd581a91 | 645 | rcu_dynticks_eqs_enter(); |
cf7614e1 BP |
646 | |
647 | if (irq) | |
648 | rcu_dynticks_task_enter(); | |
649 | } | |
650 | ||
651 | /** | |
652 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
cf7614e1 BP |
653 | * |
654 | * If you add or remove a call to rcu_nmi_exit(), be sure to test | |
655 | * with CONFIG_RCU_EQS_DEBUG=y. | |
656 | */ | |
657 | void rcu_nmi_exit(void) | |
658 | { | |
659 | rcu_nmi_exit_common(false); | |
fd581a91 PM |
660 | } |
661 | ||
9b2e4f18 PM |
662 | /** |
663 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
664 | * | |
665 | * Exit from an interrupt handler, which might possibly result in entering | |
666 | * idle mode, in other words, leaving the mode in which read-side critical | |
7c9906ca | 667 | * sections can occur. The caller must have disabled interrupts. |
64db4cff | 668 | * |
9b2e4f18 PM |
669 | * This code assumes that the idle loop never does anything that might |
670 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
58721f5d PM |
671 | * architecture's idle loop violates this assumption, RCU will give you what |
672 | * you deserve, good and hard. But very infrequently and irreproducibly. | |
9b2e4f18 PM |
673 | * |
674 | * Use things like work queues to work around this limitation. | |
675 | * | |
676 | * You have been warned. | |
c0da313e PM |
677 | * |
678 | * If you add or remove a call to rcu_irq_exit(), be sure to test with | |
679 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 680 | */ |
9b2e4f18 | 681 | void rcu_irq_exit(void) |
64db4cff | 682 | { |
b04db8e1 | 683 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 684 | rcu_nmi_exit_common(true); |
7c9906ca PM |
685 | } |
686 | ||
687 | /* | |
688 | * Wrapper for rcu_irq_exit() where interrupts are enabled. | |
c0da313e PM |
689 | * |
690 | * If you add or remove a call to rcu_irq_exit_irqson(), be sure to test | |
691 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
692 | */ |
693 | void rcu_irq_exit_irqson(void) | |
694 | { | |
695 | unsigned long flags; | |
696 | ||
697 | local_irq_save(flags); | |
698 | rcu_irq_exit(); | |
9b2e4f18 PM |
699 | local_irq_restore(flags); |
700 | } | |
701 | ||
adf5091e FW |
702 | /* |
703 | * Exit an RCU extended quiescent state, which can be either the | |
704 | * idle loop or adaptive-tickless usermode execution. | |
51a1fd30 PM |
705 | * |
706 | * We crowbar the ->dynticks_nmi_nesting field to DYNTICK_IRQ_NONIDLE to | |
707 | * allow for the possibility of usermode upcalls messing up our count of | |
708 | * interrupt nesting level during the busy period that is just now starting. | |
9b2e4f18 | 709 | */ |
adf5091e | 710 | static void rcu_eqs_exit(bool user) |
9b2e4f18 | 711 | { |
4c5273bf | 712 | struct rcu_data *rdp; |
84585aa8 | 713 | long oldval; |
9b2e4f18 | 714 | |
b04db8e1 | 715 | lockdep_assert_irqs_disabled(); |
4c5273bf PM |
716 | rdp = this_cpu_ptr(&rcu_data); |
717 | oldval = rdp->dynticks_nesting; | |
1ce46ee5 | 718 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0); |
51a1fd30 | 719 | if (oldval) { |
4c5273bf | 720 | rdp->dynticks_nesting++; |
9dd238e2 | 721 | return; |
3a592405 | 722 | } |
9dd238e2 PM |
723 | rcu_dynticks_task_exit(); |
724 | rcu_dynticks_eqs_exit(); | |
725 | rcu_cleanup_after_idle(); | |
dc5a4f29 | 726 | trace_rcu_dyntick(TPS("End"), rdp->dynticks_nesting, 1, rdp->dynticks); |
e68bbb26 | 727 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current)); |
4c5273bf PM |
728 | WRITE_ONCE(rdp->dynticks_nesting, 1); |
729 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting); | |
730 | WRITE_ONCE(rdp->dynticks_nmi_nesting, DYNTICK_IRQ_NONIDLE); | |
9b2e4f18 | 731 | } |
adf5091e FW |
732 | |
733 | /** | |
734 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
735 | * | |
736 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
737 | * read-side critical sections can occur. | |
738 | * | |
c0da313e PM |
739 | * If you add or remove a call to rcu_idle_exit(), be sure to test with |
740 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
741 | */ |
742 | void rcu_idle_exit(void) | |
743 | { | |
c5d900bf FW |
744 | unsigned long flags; |
745 | ||
746 | local_irq_save(flags); | |
cb349ca9 | 747 | rcu_eqs_exit(false); |
c5d900bf | 748 | local_irq_restore(flags); |
adf5091e | 749 | } |
9b2e4f18 | 750 | |
d1ec4c34 | 751 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
752 | /** |
753 | * rcu_user_exit - inform RCU that we are exiting userspace. | |
754 | * | |
755 | * Exit RCU idle mode while entering the kernel because it can | |
756 | * run a RCU read side critical section anytime. | |
c0da313e PM |
757 | * |
758 | * If you add or remove a call to rcu_user_exit(), be sure to test with | |
759 | * CONFIG_RCU_EQS_DEBUG=y. | |
adf5091e FW |
760 | */ |
761 | void rcu_user_exit(void) | |
762 | { | |
91d1aa43 | 763 | rcu_eqs_exit(1); |
adf5091e | 764 | } |
d1ec4c34 | 765 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 766 | |
64db4cff | 767 | /** |
cf7614e1 BP |
768 | * rcu_nmi_enter_common - inform RCU of entry to NMI context |
769 | * @irq: Is this call from rcu_irq_enter? | |
64db4cff | 770 | * |
dc5a4f29 | 771 | * If the CPU was idle from RCU's viewpoint, update rdp->dynticks and |
4c5273bf | 772 | * rdp->dynticks_nmi_nesting to let the RCU grace-period handling know |
734d1680 PM |
773 | * that the CPU is active. This implementation permits nested NMIs, as |
774 | * long as the nesting level does not overflow an int. (You will probably | |
775 | * run out of stack space first.) | |
c0da313e | 776 | * |
cf7614e1 | 777 | * If you add or remove a call to rcu_nmi_enter_common(), be sure to test |
c0da313e | 778 | * with CONFIG_RCU_EQS_DEBUG=y. |
64db4cff | 779 | */ |
cf7614e1 | 780 | static __always_inline void rcu_nmi_enter_common(bool irq) |
64db4cff | 781 | { |
4c5273bf | 782 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
84585aa8 | 783 | long incby = 2; |
64db4cff | 784 | |
734d1680 | 785 | /* Complain about underflow. */ |
4c5273bf | 786 | WARN_ON_ONCE(rdp->dynticks_nmi_nesting < 0); |
734d1680 PM |
787 | |
788 | /* | |
789 | * If idle from RCU viewpoint, atomically increment ->dynticks | |
790 | * to mark non-idle and increment ->dynticks_nmi_nesting by one. | |
791 | * Otherwise, increment ->dynticks_nmi_nesting by two. This means | |
792 | * if ->dynticks_nmi_nesting is equal to one, we are guaranteed | |
793 | * to be in the outermost NMI handler that interrupted an RCU-idle | |
794 | * period (observation due to Andy Lutomirski). | |
795 | */ | |
02a5c550 | 796 | if (rcu_dynticks_curr_cpu_in_eqs()) { |
cf7614e1 BP |
797 | |
798 | if (irq) | |
799 | rcu_dynticks_task_exit(); | |
800 | ||
2625d469 | 801 | rcu_dynticks_eqs_exit(); |
cf7614e1 BP |
802 | |
803 | if (irq) | |
804 | rcu_cleanup_after_idle(); | |
805 | ||
734d1680 PM |
806 | incby = 1; |
807 | } | |
bd2b879a | 808 | trace_rcu_dyntick(incby == 1 ? TPS("Endirq") : TPS("++="), |
4c5273bf | 809 | rdp->dynticks_nmi_nesting, |
dc5a4f29 | 810 | rdp->dynticks_nmi_nesting + incby, rdp->dynticks); |
4c5273bf PM |
811 | WRITE_ONCE(rdp->dynticks_nmi_nesting, /* Prevent store tearing. */ |
812 | rdp->dynticks_nmi_nesting + incby); | |
734d1680 | 813 | barrier(); |
64db4cff PM |
814 | } |
815 | ||
cf7614e1 BP |
816 | /** |
817 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
818 | */ | |
819 | void rcu_nmi_enter(void) | |
820 | { | |
821 | rcu_nmi_enter_common(false); | |
822 | } | |
c13324a5 | 823 | NOKPROBE_SYMBOL(rcu_nmi_enter); |
cf7614e1 | 824 | |
64db4cff | 825 | /** |
9b2e4f18 | 826 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle |
64db4cff | 827 | * |
9b2e4f18 PM |
828 | * Enter an interrupt handler, which might possibly result in exiting |
829 | * idle mode, in other words, entering the mode in which read-side critical | |
7c9906ca | 830 | * sections can occur. The caller must have disabled interrupts. |
c0da313e | 831 | * |
9b2e4f18 | 832 | * Note that the Linux kernel is fully capable of entering an interrupt |
58721f5d PM |
833 | * handler that it never exits, for example when doing upcalls to user mode! |
834 | * This code assumes that the idle loop never does upcalls to user mode. | |
835 | * If your architecture's idle loop does do upcalls to user mode (or does | |
836 | * anything else that results in unbalanced calls to the irq_enter() and | |
837 | * irq_exit() functions), RCU will give you what you deserve, good and hard. | |
838 | * But very infrequently and irreproducibly. | |
9b2e4f18 PM |
839 | * |
840 | * Use things like work queues to work around this limitation. | |
841 | * | |
842 | * You have been warned. | |
c0da313e PM |
843 | * |
844 | * If you add or remove a call to rcu_irq_enter(), be sure to test with | |
845 | * CONFIG_RCU_EQS_DEBUG=y. | |
64db4cff | 846 | */ |
9b2e4f18 | 847 | void rcu_irq_enter(void) |
64db4cff | 848 | { |
b04db8e1 | 849 | lockdep_assert_irqs_disabled(); |
cf7614e1 | 850 | rcu_nmi_enter_common(true); |
7c9906ca | 851 | } |
734d1680 | 852 | |
7c9906ca PM |
853 | /* |
854 | * Wrapper for rcu_irq_enter() where interrupts are enabled. | |
c0da313e PM |
855 | * |
856 | * If you add or remove a call to rcu_irq_enter_irqson(), be sure to test | |
857 | * with CONFIG_RCU_EQS_DEBUG=y. | |
7c9906ca PM |
858 | */ |
859 | void rcu_irq_enter_irqson(void) | |
860 | { | |
861 | unsigned long flags; | |
734d1680 | 862 | |
7c9906ca PM |
863 | local_irq_save(flags); |
864 | rcu_irq_enter(); | |
64db4cff | 865 | local_irq_restore(flags); |
64db4cff PM |
866 | } |
867 | ||
5c173eb8 | 868 | /** |
2320bda2 | 869 | * rcu_is_watching - see if RCU thinks that the current CPU is not idle |
64db4cff | 870 | * |
791875d1 PM |
871 | * Return true if RCU is watching the running CPU, which means that this |
872 | * CPU can safely enter RCU read-side critical sections. In other words, | |
2320bda2 ZZ |
873 | * if the current CPU is not in its idle loop or is in an interrupt or |
874 | * NMI handler, return true. | |
64db4cff | 875 | */ |
9418fb20 | 876 | bool notrace rcu_is_watching(void) |
64db4cff | 877 | { |
f534ed1f | 878 | bool ret; |
34240697 | 879 | |
46f00d18 | 880 | preempt_disable_notrace(); |
791875d1 | 881 | ret = !rcu_dynticks_curr_cpu_in_eqs(); |
46f00d18 | 882 | preempt_enable_notrace(); |
34240697 | 883 | return ret; |
64db4cff | 884 | } |
5c173eb8 | 885 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 886 | |
bcbfdd01 PM |
887 | /* |
888 | * If a holdout task is actually running, request an urgent quiescent | |
889 | * state from its CPU. This is unsynchronized, so migrations can cause | |
890 | * the request to go to the wrong CPU. Which is OK, all that will happen | |
891 | * is that the CPU's next context switch will be a bit slower and next | |
892 | * time around this task will generate another request. | |
893 | */ | |
894 | void rcu_request_urgent_qs_task(struct task_struct *t) | |
895 | { | |
896 | int cpu; | |
897 | ||
898 | barrier(); | |
899 | cpu = task_cpu(t); | |
900 | if (!task_curr(t)) | |
901 | return; /* This task is not running on that CPU. */ | |
2dba13f0 | 902 | smp_store_release(per_cpu_ptr(&rcu_data.rcu_urgent_qs, cpu), true); |
bcbfdd01 PM |
903 | } |
904 | ||
62fde6ed | 905 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
906 | |
907 | /* | |
5554788e | 908 | * Is the current CPU online as far as RCU is concerned? |
2036d94a | 909 | * |
5554788e PM |
910 | * Disable preemption to avoid false positives that could otherwise |
911 | * happen due to the current CPU number being sampled, this task being | |
912 | * preempted, its old CPU being taken offline, resuming on some other CPU, | |
49918a54 | 913 | * then determining that its old CPU is now offline. |
c0d6d01b | 914 | * |
5554788e PM |
915 | * Disable checking if in an NMI handler because we cannot safely |
916 | * report errors from NMI handlers anyway. In addition, it is OK to use | |
917 | * RCU on an offline processor during initial boot, hence the check for | |
918 | * rcu_scheduler_fully_active. | |
c0d6d01b PM |
919 | */ |
920 | bool rcu_lockdep_current_cpu_online(void) | |
921 | { | |
2036d94a PM |
922 | struct rcu_data *rdp; |
923 | struct rcu_node *rnp; | |
b97d23c5 | 924 | bool ret = false; |
c0d6d01b | 925 | |
5554788e | 926 | if (in_nmi() || !rcu_scheduler_fully_active) |
f6f7ee9a | 927 | return true; |
c0d6d01b | 928 | preempt_disable(); |
b97d23c5 PM |
929 | rdp = this_cpu_ptr(&rcu_data); |
930 | rnp = rdp->mynode; | |
931 | if (rdp->grpmask & rcu_rnp_online_cpus(rnp)) | |
932 | ret = true; | |
c0d6d01b | 933 | preempt_enable(); |
b97d23c5 | 934 | return ret; |
c0d6d01b PM |
935 | } |
936 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
937 | ||
62fde6ed | 938 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 939 | |
9b9500da PM |
940 | /* |
941 | * We are reporting a quiescent state on behalf of some other CPU, so | |
942 | * it is our responsibility to check for and handle potential overflow | |
a66ae8ae | 943 | * of the rcu_node ->gp_seq counter with respect to the rcu_data counters. |
9b9500da PM |
944 | * After all, the CPU might be in deep idle state, and thus executing no |
945 | * code whatsoever. | |
946 | */ | |
947 | static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp) | |
948 | { | |
a32e01ee | 949 | raw_lockdep_assert_held_rcu_node(rnp); |
a66ae8ae PM |
950 | if (ULONG_CMP_LT(rcu_seq_current(&rdp->gp_seq) + ULONG_MAX / 4, |
951 | rnp->gp_seq)) | |
9b9500da | 952 | WRITE_ONCE(rdp->gpwrap, true); |
8aa670cd PM |
953 | if (ULONG_CMP_LT(rdp->rcu_iw_gp_seq + ULONG_MAX / 4, rnp->gp_seq)) |
954 | rdp->rcu_iw_gp_seq = rnp->gp_seq + ULONG_MAX / 4; | |
9b9500da PM |
955 | } |
956 | ||
64db4cff PM |
957 | /* |
958 | * Snapshot the specified CPU's dynticks counter so that we can later | |
959 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 960 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 961 | */ |
fe5ac724 | 962 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
64db4cff | 963 | { |
dc5a4f29 | 964 | rdp->dynticks_snap = rcu_dynticks_snap(rdp); |
02a5c550 | 965 | if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) { |
88d1bead | 966 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 967 | rcu_gpnum_ovf(rdp->mynode, rdp); |
23a9bacd | 968 | return 1; |
7941dbde | 969 | } |
23a9bacd | 970 | return 0; |
64db4cff PM |
971 | } |
972 | ||
973 | /* | |
974 | * Return true if the specified CPU has passed through a quiescent | |
975 | * state by virtue of being in or having passed through an dynticks | |
976 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 977 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 978 | */ |
fe5ac724 | 979 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
64db4cff | 980 | { |
3a19b46a | 981 | unsigned long jtsq; |
0f9be8ca | 982 | bool *rnhqp; |
9226b10d | 983 | bool *ruqp; |
9b9500da | 984 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
985 | |
986 | /* | |
987 | * If the CPU passed through or entered a dynticks idle phase with | |
988 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
989 | * already acknowledged the request to pass through a quiescent | |
990 | * state. Either way, that CPU cannot possibly be in an RCU | |
991 | * read-side critical section that started before the beginning | |
992 | * of the current RCU grace period. | |
993 | */ | |
dc5a4f29 | 994 | if (rcu_dynticks_in_eqs_since(rdp, rdp->dynticks_snap)) { |
88d1bead | 995 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 996 | rcu_gpnum_ovf(rnp, rdp); |
3a19b46a PM |
997 | return 1; |
998 | } | |
999 | ||
f2e2df59 PM |
1000 | /* If waiting too long on an offline CPU, complain. */ |
1001 | if (!(rdp->grpmask & rcu_rnp_online_cpus(rnp)) && | |
88d1bead | 1002 | time_after(jiffies, rcu_state.gp_start + HZ)) { |
f2e2df59 PM |
1003 | bool onl; |
1004 | struct rcu_node *rnp1; | |
1005 | ||
1006 | WARN_ON(1); /* Offline CPUs are supposed to report QS! */ | |
1007 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", | |
1008 | __func__, rnp->grplo, rnp->grphi, rnp->level, | |
1009 | (long)rnp->gp_seq, (long)rnp->completedqs); | |
1010 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
1011 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx ->rcu_gp_init_mask %#lx\n", | |
1012 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext, rnp1->rcu_gp_init_mask); | |
1013 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); | |
1014 | pr_info("%s %d: %c online: %ld(%d) offline: %ld(%d)\n", | |
1015 | __func__, rdp->cpu, ".o"[onl], | |
1016 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
1017 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
1018 | return 1; /* Break things loose after complaining. */ | |
1019 | } | |
1020 | ||
65d798f0 | 1021 | /* |
4a81e832 | 1022 | * A CPU running for an extended time within the kernel can |
c06aed0e PM |
1023 | * delay RCU grace periods: (1) At age jiffies_to_sched_qs, |
1024 | * set .rcu_urgent_qs, (2) At age 2*jiffies_to_sched_qs, set | |
7e28c5af PM |
1025 | * both .rcu_need_heavy_qs and .rcu_urgent_qs. Note that the |
1026 | * unsynchronized assignments to the per-CPU rcu_need_heavy_qs | |
1027 | * variable are safe because the assignments are repeated if this | |
1028 | * CPU failed to pass through a quiescent state. This code | |
c06aed0e | 1029 | * also checks .jiffies_resched in case jiffies_to_sched_qs |
7e28c5af | 1030 | * is set way high. |
6193c76a | 1031 | */ |
c06aed0e | 1032 | jtsq = READ_ONCE(jiffies_to_sched_qs); |
2dba13f0 PM |
1033 | ruqp = per_cpu_ptr(&rcu_data.rcu_urgent_qs, rdp->cpu); |
1034 | rnhqp = &per_cpu(rcu_data.rcu_need_heavy_qs, rdp->cpu); | |
0f9be8ca | 1035 | if (!READ_ONCE(*rnhqp) && |
7e28c5af | 1036 | (time_after(jiffies, rcu_state.gp_start + jtsq * 2) || |
88d1bead | 1037 | time_after(jiffies, rcu_state.jiffies_resched))) { |
0f9be8ca | 1038 | WRITE_ONCE(*rnhqp, true); |
9226b10d PM |
1039 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ |
1040 | smp_store_release(ruqp, true); | |
7e28c5af PM |
1041 | } else if (time_after(jiffies, rcu_state.gp_start + jtsq)) { |
1042 | WRITE_ONCE(*ruqp, true); | |
6193c76a PM |
1043 | } |
1044 | ||
28053bc7 | 1045 | /* |
c98cac60 | 1046 | * NO_HZ_FULL CPUs can run in-kernel without rcu_sched_clock_irq! |
d3052109 PM |
1047 | * The above code handles this, but only for straight cond_resched(). |
1048 | * And some in-kernel loops check need_resched() before calling | |
1049 | * cond_resched(), which defeats the above code for CPUs that are | |
1050 | * running in-kernel with scheduling-clock interrupts disabled. | |
1051 | * So hit them over the head with the resched_cpu() hammer! | |
28053bc7 | 1052 | */ |
d3052109 PM |
1053 | if (tick_nohz_full_cpu(rdp->cpu) && |
1054 | time_after(jiffies, | |
1055 | READ_ONCE(rdp->last_fqs_resched) + jtsq * 3)) { | |
28053bc7 | 1056 | resched_cpu(rdp->cpu); |
d3052109 PM |
1057 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); |
1058 | } | |
1059 | ||
1060 | /* | |
1061 | * If more than halfway to RCU CPU stall-warning time, invoke | |
1062 | * resched_cpu() more frequently to try to loosen things up a bit. | |
1063 | * Also check to see if the CPU is getting hammered with interrupts, | |
1064 | * but only once per grace period, just to keep the IPIs down to | |
1065 | * a dull roar. | |
1066 | */ | |
1067 | if (time_after(jiffies, rcu_state.jiffies_resched)) { | |
1068 | if (time_after(jiffies, | |
1069 | READ_ONCE(rdp->last_fqs_resched) + jtsq)) { | |
1070 | resched_cpu(rdp->cpu); | |
1071 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); | |
1072 | } | |
9b9500da | 1073 | if (IS_ENABLED(CONFIG_IRQ_WORK) && |
8aa670cd | 1074 | !rdp->rcu_iw_pending && rdp->rcu_iw_gp_seq != rnp->gp_seq && |
9b9500da PM |
1075 | (rnp->ffmask & rdp->grpmask)) { |
1076 | init_irq_work(&rdp->rcu_iw, rcu_iw_handler); | |
1077 | rdp->rcu_iw_pending = true; | |
8aa670cd | 1078 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
1079 | irq_work_queue_on(&rdp->rcu_iw, rdp->cpu); |
1080 | } | |
1081 | } | |
4914950a | 1082 | |
a82dcc76 | 1083 | return 0; |
64db4cff PM |
1084 | } |
1085 | ||
41e80595 PM |
1086 | /* Trace-event wrapper function for trace_rcu_future_grace_period. */ |
1087 | static void trace_rcu_this_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
b73de91d | 1088 | unsigned long gp_seq_req, const char *s) |
0446be48 | 1089 | { |
88d1bead | 1090 | trace_rcu_future_grace_period(rcu_state.name, rnp->gp_seq, gp_seq_req, |
abd13fdd | 1091 | rnp->level, rnp->grplo, rnp->grphi, s); |
0446be48 PM |
1092 | } |
1093 | ||
1094 | /* | |
b73de91d | 1095 | * rcu_start_this_gp - Request the start of a particular grace period |
df2bf8f7 | 1096 | * @rnp_start: The leaf node of the CPU from which to start. |
b73de91d JF |
1097 | * @rdp: The rcu_data corresponding to the CPU from which to start. |
1098 | * @gp_seq_req: The gp_seq of the grace period to start. | |
1099 | * | |
41e80595 | 1100 | * Start the specified grace period, as needed to handle newly arrived |
0446be48 | 1101 | * callbacks. The required future grace periods are recorded in each |
7a1d0f23 | 1102 | * rcu_node structure's ->gp_seq_needed field. Returns true if there |
48a7639c | 1103 | * is reason to awaken the grace-period kthread. |
0446be48 | 1104 | * |
d5cd9685 PM |
1105 | * The caller must hold the specified rcu_node structure's ->lock, which |
1106 | * is why the caller is responsible for waking the grace-period kthread. | |
b73de91d JF |
1107 | * |
1108 | * Returns true if the GP thread needs to be awakened else false. | |
0446be48 | 1109 | */ |
df2bf8f7 | 1110 | static bool rcu_start_this_gp(struct rcu_node *rnp_start, struct rcu_data *rdp, |
b73de91d | 1111 | unsigned long gp_seq_req) |
0446be48 | 1112 | { |
48a7639c | 1113 | bool ret = false; |
df2bf8f7 | 1114 | struct rcu_node *rnp; |
0446be48 PM |
1115 | |
1116 | /* | |
360e0da6 PM |
1117 | * Use funnel locking to either acquire the root rcu_node |
1118 | * structure's lock or bail out if the need for this grace period | |
df2bf8f7 JFG |
1119 | * has already been recorded -- or if that grace period has in |
1120 | * fact already started. If there is already a grace period in | |
1121 | * progress in a non-leaf node, no recording is needed because the | |
1122 | * end of the grace period will scan the leaf rcu_node structures. | |
1123 | * Note that rnp_start->lock must not be released. | |
0446be48 | 1124 | */ |
df2bf8f7 JFG |
1125 | raw_lockdep_assert_held_rcu_node(rnp_start); |
1126 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, TPS("Startleaf")); | |
1127 | for (rnp = rnp_start; 1; rnp = rnp->parent) { | |
1128 | if (rnp != rnp_start) | |
1129 | raw_spin_lock_rcu_node(rnp); | |
1130 | if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req) || | |
1131 | rcu_seq_started(&rnp->gp_seq, gp_seq_req) || | |
1132 | (rnp != rnp_start && | |
1133 | rcu_seq_state(rcu_seq_current(&rnp->gp_seq)))) { | |
1134 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, | |
b73de91d | 1135 | TPS("Prestarted")); |
360e0da6 PM |
1136 | goto unlock_out; |
1137 | } | |
df2bf8f7 | 1138 | rnp->gp_seq_needed = gp_seq_req; |
226ca5e7 | 1139 | if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) { |
a2165e41 | 1140 | /* |
226ca5e7 JFG |
1141 | * We just marked the leaf or internal node, and a |
1142 | * grace period is in progress, which means that | |
1143 | * rcu_gp_cleanup() will see the marking. Bail to | |
1144 | * reduce contention. | |
a2165e41 | 1145 | */ |
df2bf8f7 | 1146 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, |
b73de91d | 1147 | TPS("Startedleaf")); |
a2165e41 PM |
1148 | goto unlock_out; |
1149 | } | |
df2bf8f7 JFG |
1150 | if (rnp != rnp_start && rnp->parent != NULL) |
1151 | raw_spin_unlock_rcu_node(rnp); | |
1152 | if (!rnp->parent) | |
360e0da6 | 1153 | break; /* At root, and perhaps also leaf. */ |
0446be48 PM |
1154 | } |
1155 | ||
360e0da6 | 1156 | /* If GP already in progress, just leave, otherwise start one. */ |
de8e8730 | 1157 | if (rcu_gp_in_progress()) { |
df2bf8f7 | 1158 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedleafroot")); |
0446be48 PM |
1159 | goto unlock_out; |
1160 | } | |
df2bf8f7 | 1161 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedroot")); |
9cbc5b97 PM |
1162 | WRITE_ONCE(rcu_state.gp_flags, rcu_state.gp_flags | RCU_GP_FLAG_INIT); |
1163 | rcu_state.gp_req_activity = jiffies; | |
1164 | if (!rcu_state.gp_kthread) { | |
df2bf8f7 | 1165 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("NoGPkthread")); |
360e0da6 | 1166 | goto unlock_out; |
0446be48 | 1167 | } |
9cbc5b97 | 1168 | trace_rcu_grace_period(rcu_state.name, READ_ONCE(rcu_state.gp_seq), TPS("newreq")); |
360e0da6 | 1169 | ret = true; /* Caller must wake GP kthread. */ |
0446be48 | 1170 | unlock_out: |
ab5e869c | 1171 | /* Push furthest requested GP to leaf node and rcu_data structure. */ |
df2bf8f7 JFG |
1172 | if (ULONG_CMP_LT(gp_seq_req, rnp->gp_seq_needed)) { |
1173 | rnp_start->gp_seq_needed = rnp->gp_seq_needed; | |
1174 | rdp->gp_seq_needed = rnp->gp_seq_needed; | |
ab5e869c | 1175 | } |
df2bf8f7 JFG |
1176 | if (rnp != rnp_start) |
1177 | raw_spin_unlock_rcu_node(rnp); | |
48a7639c | 1178 | return ret; |
0446be48 PM |
1179 | } |
1180 | ||
1181 | /* | |
1182 | * Clean up any old requests for the just-ended grace period. Also return | |
d1e4f01d | 1183 | * whether any additional grace periods have been requested. |
0446be48 | 1184 | */ |
3481f2ea | 1185 | static bool rcu_future_gp_cleanup(struct rcu_node *rnp) |
0446be48 | 1186 | { |
fb31340f | 1187 | bool needmore; |
da1df50d | 1188 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
0446be48 | 1189 | |
7a1d0f23 PM |
1190 | needmore = ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed); |
1191 | if (!needmore) | |
1192 | rnp->gp_seq_needed = rnp->gp_seq; /* Avoid counter wrap. */ | |
b73de91d | 1193 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq, |
41e80595 | 1194 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); |
0446be48 PM |
1195 | return needmore; |
1196 | } | |
1197 | ||
48a7639c | 1198 | /* |
1d1f898d ZJ |
1199 | * Awaken the grace-period kthread. Don't do a self-awaken (unless in |
1200 | * an interrupt or softirq handler), and don't bother awakening when there | |
1201 | * is nothing for the grace-period kthread to do (as in several CPUs raced | |
1202 | * to awaken, and we lost), and finally don't try to awaken a kthread that | |
1203 | * has not yet been created. If all those checks are passed, track some | |
1204 | * debug information and awaken. | |
1205 | * | |
1206 | * So why do the self-wakeup when in an interrupt or softirq handler | |
1207 | * in the grace-period kthread's context? Because the kthread might have | |
1208 | * been interrupted just as it was going to sleep, and just after the final | |
1209 | * pre-sleep check of the awaken condition. In this case, a wakeup really | |
1210 | * is required, and is therefore supplied. | |
48a7639c | 1211 | */ |
532c00c9 | 1212 | static void rcu_gp_kthread_wake(void) |
48a7639c | 1213 | { |
1d1f898d | 1214 | if ((current == rcu_state.gp_kthread && |
0f58d2ac | 1215 | !in_irq() && !in_serving_softirq()) || |
532c00c9 PM |
1216 | !READ_ONCE(rcu_state.gp_flags) || |
1217 | !rcu_state.gp_kthread) | |
48a7639c | 1218 | return; |
fd897573 PM |
1219 | WRITE_ONCE(rcu_state.gp_wake_time, jiffies); |
1220 | WRITE_ONCE(rcu_state.gp_wake_seq, READ_ONCE(rcu_state.gp_seq)); | |
532c00c9 | 1221 | swake_up_one(&rcu_state.gp_wq); |
48a7639c PM |
1222 | } |
1223 | ||
dc35c893 | 1224 | /* |
29365e56 PM |
1225 | * If there is room, assign a ->gp_seq number to any callbacks on this |
1226 | * CPU that have not already been assigned. Also accelerate any callbacks | |
1227 | * that were previously assigned a ->gp_seq number that has since proven | |
1228 | * to be too conservative, which can happen if callbacks get assigned a | |
1229 | * ->gp_seq number while RCU is idle, but with reference to a non-root | |
1230 | * rcu_node structure. This function is idempotent, so it does not hurt | |
1231 | * to call it repeatedly. Returns an flag saying that we should awaken | |
1232 | * the RCU grace-period kthread. | |
dc35c893 PM |
1233 | * |
1234 | * The caller must hold rnp->lock with interrupts disabled. | |
1235 | */ | |
02f50142 | 1236 | static bool rcu_accelerate_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1237 | { |
b73de91d | 1238 | unsigned long gp_seq_req; |
15fecf89 | 1239 | bool ret = false; |
dc35c893 | 1240 | |
a32e01ee | 1241 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1242 | |
15fecf89 PM |
1243 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1244 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1245 | return false; |
dc35c893 PM |
1246 | |
1247 | /* | |
15fecf89 PM |
1248 | * Callbacks are often registered with incomplete grace-period |
1249 | * information. Something about the fact that getting exact | |
1250 | * information requires acquiring a global lock... RCU therefore | |
1251 | * makes a conservative estimate of the grace period number at which | |
1252 | * a given callback will become ready to invoke. The following | |
1253 | * code checks this estimate and improves it when possible, thus | |
1254 | * accelerating callback invocation to an earlier grace-period | |
1255 | * number. | |
dc35c893 | 1256 | */ |
9cbc5b97 | 1257 | gp_seq_req = rcu_seq_snap(&rcu_state.gp_seq); |
b73de91d JF |
1258 | if (rcu_segcblist_accelerate(&rdp->cblist, gp_seq_req)) |
1259 | ret = rcu_start_this_gp(rnp, rdp, gp_seq_req); | |
6d4b418c PM |
1260 | |
1261 | /* Trace depending on how much we were able to accelerate. */ | |
15fecf89 | 1262 | if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL)) |
9cbc5b97 | 1263 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccWaitCB")); |
6d4b418c | 1264 | else |
9cbc5b97 | 1265 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("AccReadyCB")); |
48a7639c | 1266 | return ret; |
dc35c893 PM |
1267 | } |
1268 | ||
e44e73ca PM |
1269 | /* |
1270 | * Similar to rcu_accelerate_cbs(), but does not require that the leaf | |
1271 | * rcu_node structure's ->lock be held. It consults the cached value | |
1272 | * of ->gp_seq_needed in the rcu_data structure, and if that indicates | |
1273 | * that a new grace-period request be made, invokes rcu_accelerate_cbs() | |
1274 | * while holding the leaf rcu_node structure's ->lock. | |
1275 | */ | |
c6e09b97 | 1276 | static void rcu_accelerate_cbs_unlocked(struct rcu_node *rnp, |
e44e73ca PM |
1277 | struct rcu_data *rdp) |
1278 | { | |
1279 | unsigned long c; | |
1280 | bool needwake; | |
1281 | ||
1282 | lockdep_assert_irqs_disabled(); | |
c6e09b97 | 1283 | c = rcu_seq_snap(&rcu_state.gp_seq); |
e44e73ca PM |
1284 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { |
1285 | /* Old request still live, so mark recent callbacks. */ | |
1286 | (void)rcu_segcblist_accelerate(&rdp->cblist, c); | |
1287 | return; | |
1288 | } | |
1289 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
02f50142 | 1290 | needwake = rcu_accelerate_cbs(rnp, rdp); |
e44e73ca PM |
1291 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
1292 | if (needwake) | |
532c00c9 | 1293 | rcu_gp_kthread_wake(); |
e44e73ca PM |
1294 | } |
1295 | ||
dc35c893 PM |
1296 | /* |
1297 | * Move any callbacks whose grace period has completed to the | |
1298 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
29365e56 | 1299 | * assign ->gp_seq numbers to any callbacks in the RCU_NEXT_TAIL |
dc35c893 PM |
1300 | * sublist. This function is idempotent, so it does not hurt to |
1301 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1302 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1303 | * |
1304 | * The caller must hold rnp->lock with interrupts disabled. | |
1305 | */ | |
834f56bf | 1306 | static bool rcu_advance_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1307 | { |
a32e01ee | 1308 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1309 | |
15fecf89 PM |
1310 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1311 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1312 | return false; |
dc35c893 PM |
1313 | |
1314 | /* | |
29365e56 | 1315 | * Find all callbacks whose ->gp_seq numbers indicate that they |
dc35c893 PM |
1316 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. |
1317 | */ | |
29365e56 | 1318 | rcu_segcblist_advance(&rdp->cblist, rnp->gp_seq); |
dc35c893 PM |
1319 | |
1320 | /* Classify any remaining callbacks. */ | |
02f50142 | 1321 | return rcu_accelerate_cbs(rnp, rdp); |
dc35c893 PM |
1322 | } |
1323 | ||
d09b62df | 1324 | /* |
ba9fbe95 PM |
1325 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1326 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1327 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1328 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1329 | */ |
c7e48f7b | 1330 | static bool __note_gp_changes(struct rcu_node *rnp, struct rcu_data *rdp) |
d09b62df | 1331 | { |
48a7639c | 1332 | bool ret; |
3563a438 | 1333 | bool need_gp; |
48a7639c | 1334 | |
a32e01ee | 1335 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1336 | |
67e14c1e PM |
1337 | if (rdp->gp_seq == rnp->gp_seq) |
1338 | return false; /* Nothing to do. */ | |
d09b62df | 1339 | |
67e14c1e PM |
1340 | /* Handle the ends of any preceding grace periods first. */ |
1341 | if (rcu_seq_completed_gp(rdp->gp_seq, rnp->gp_seq) || | |
1342 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
834f56bf | 1343 | ret = rcu_advance_cbs(rnp, rdp); /* Advance callbacks. */ |
9cbc5b97 | 1344 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuend")); |
67e14c1e | 1345 | } else { |
02f50142 | 1346 | ret = rcu_accelerate_cbs(rnp, rdp); /* Recent callbacks. */ |
d09b62df | 1347 | } |
398ebe60 | 1348 | |
67e14c1e PM |
1349 | /* Now handle the beginnings of any new-to-this-CPU grace periods. */ |
1350 | if (rcu_seq_new_gp(rdp->gp_seq, rnp->gp_seq) || | |
1351 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
6eaef633 PM |
1352 | /* |
1353 | * If the current grace period is waiting for this CPU, | |
1354 | * set up to detect a quiescent state, otherwise don't | |
1355 | * go looking for one. | |
1356 | */ | |
9cbc5b97 | 1357 | trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, TPS("cpustart")); |
3563a438 PM |
1358 | need_gp = !!(rnp->qsmask & rdp->grpmask); |
1359 | rdp->cpu_no_qs.b.norm = need_gp; | |
3563a438 | 1360 | rdp->core_needs_qs = need_gp; |
6eaef633 PM |
1361 | zero_cpu_stall_ticks(rdp); |
1362 | } | |
67e14c1e | 1363 | rdp->gp_seq = rnp->gp_seq; /* Remember new grace-period state. */ |
13dc7d0c | 1364 | if (ULONG_CMP_LT(rdp->gp_seq_needed, rnp->gp_seq_needed) || rdp->gpwrap) |
3d18469a PM |
1365 | rdp->gp_seq_needed = rnp->gp_seq_needed; |
1366 | WRITE_ONCE(rdp->gpwrap, false); | |
1367 | rcu_gpnum_ovf(rnp, rdp); | |
48a7639c | 1368 | return ret; |
6eaef633 PM |
1369 | } |
1370 | ||
15cabdff | 1371 | static void note_gp_changes(struct rcu_data *rdp) |
6eaef633 PM |
1372 | { |
1373 | unsigned long flags; | |
48a7639c | 1374 | bool needwake; |
6eaef633 PM |
1375 | struct rcu_node *rnp; |
1376 | ||
1377 | local_irq_save(flags); | |
1378 | rnp = rdp->mynode; | |
67e14c1e | 1379 | if ((rdp->gp_seq == rcu_seq_current(&rnp->gp_seq) && |
7d0ae808 | 1380 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ |
2a67e741 | 1381 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
1382 | local_irq_restore(flags); |
1383 | return; | |
1384 | } | |
c7e48f7b | 1385 | needwake = __note_gp_changes(rnp, rdp); |
67c583a7 | 1386 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c | 1387 | if (needwake) |
532c00c9 | 1388 | rcu_gp_kthread_wake(); |
6eaef633 PM |
1389 | } |
1390 | ||
22212332 | 1391 | static void rcu_gp_slow(int delay) |
0f41c0dd PM |
1392 | { |
1393 | if (delay > 0 && | |
22212332 | 1394 | !(rcu_seq_ctr(rcu_state.gp_seq) % |
dee4f422 | 1395 | (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) |
0f41c0dd PM |
1396 | schedule_timeout_uninterruptible(delay); |
1397 | } | |
1398 | ||
b3dbec76 | 1399 | /* |
45fed3e7 | 1400 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 1401 | */ |
0854a05c | 1402 | static bool rcu_gp_init(void) |
b3dbec76 | 1403 | { |
ec2c2976 | 1404 | unsigned long flags; |
0aa04b05 | 1405 | unsigned long oldmask; |
ec2c2976 | 1406 | unsigned long mask; |
b3dbec76 | 1407 | struct rcu_data *rdp; |
336a4f6c | 1408 | struct rcu_node *rnp = rcu_get_root(); |
b3dbec76 | 1409 | |
9cbc5b97 | 1410 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 1411 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 | 1412 | if (!READ_ONCE(rcu_state.gp_flags)) { |
f7be8209 | 1413 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 1414 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1415 | return false; |
f7be8209 | 1416 | } |
9cbc5b97 | 1417 | WRITE_ONCE(rcu_state.gp_flags, 0); /* Clear all flags: New GP. */ |
b3dbec76 | 1418 | |
de8e8730 | 1419 | if (WARN_ON_ONCE(rcu_gp_in_progress())) { |
f7be8209 PM |
1420 | /* |
1421 | * Grace period already in progress, don't start another. | |
1422 | * Not supposed to be able to happen. | |
1423 | */ | |
67c583a7 | 1424 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1425 | return false; |
7fdefc10 PM |
1426 | } |
1427 | ||
7fdefc10 | 1428 | /* Advance to a new grace period and initialize state. */ |
ad3832e9 | 1429 | record_gp_stall_check_time(); |
ff3bb6f4 | 1430 | /* Record GP times before starting GP, hence rcu_seq_start(). */ |
9cbc5b97 PM |
1431 | rcu_seq_start(&rcu_state.gp_seq); |
1432 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("start")); | |
67c583a7 | 1433 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 1434 | |
0aa04b05 PM |
1435 | /* |
1436 | * Apply per-leaf buffered online and offline operations to the | |
1437 | * rcu_node tree. Note that this new grace period need not wait | |
1438 | * for subsequent online CPUs, and that quiescent-state forcing | |
1439 | * will handle subsequent offline CPUs. | |
1440 | */ | |
9cbc5b97 | 1441 | rcu_state.gp_state = RCU_GP_ONOFF; |
aedf4ba9 | 1442 | rcu_for_each_leaf_node(rnp) { |
894d45bb | 1443 | raw_spin_lock(&rcu_state.ofl_lock); |
2a67e741 | 1444 | raw_spin_lock_irq_rcu_node(rnp); |
0aa04b05 PM |
1445 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
1446 | !rnp->wait_blkd_tasks) { | |
1447 | /* Nothing to do on this leaf rcu_node structure. */ | |
67c583a7 | 1448 | raw_spin_unlock_irq_rcu_node(rnp); |
894d45bb | 1449 | raw_spin_unlock(&rcu_state.ofl_lock); |
0aa04b05 PM |
1450 | continue; |
1451 | } | |
1452 | ||
1453 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
1454 | oldmask = rnp->qsmaskinit; | |
1455 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
1456 | ||
1457 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
1458 | if (!oldmask != !rnp->qsmaskinit) { | |
962aff03 PM |
1459 | if (!oldmask) { /* First online CPU for rcu_node. */ |
1460 | if (!rnp->wait_blkd_tasks) /* Ever offline? */ | |
1461 | rcu_init_new_rnp(rnp); | |
1462 | } else if (rcu_preempt_has_tasks(rnp)) { | |
1463 | rnp->wait_blkd_tasks = true; /* blocked tasks */ | |
1464 | } else { /* Last offline CPU and can propagate. */ | |
0aa04b05 | 1465 | rcu_cleanup_dead_rnp(rnp); |
962aff03 | 1466 | } |
0aa04b05 PM |
1467 | } |
1468 | ||
1469 | /* | |
1470 | * If all waited-on tasks from prior grace period are | |
1471 | * done, and if all this rcu_node structure's CPUs are | |
1472 | * still offline, propagate up the rcu_node tree and | |
1473 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
1474 | * rcu_node structure's CPUs has since come back online, | |
962aff03 | 1475 | * simply clear ->wait_blkd_tasks. |
0aa04b05 PM |
1476 | */ |
1477 | if (rnp->wait_blkd_tasks && | |
962aff03 | 1478 | (!rcu_preempt_has_tasks(rnp) || rnp->qsmaskinit)) { |
0aa04b05 | 1479 | rnp->wait_blkd_tasks = false; |
962aff03 PM |
1480 | if (!rnp->qsmaskinit) |
1481 | rcu_cleanup_dead_rnp(rnp); | |
0aa04b05 PM |
1482 | } |
1483 | ||
67c583a7 | 1484 | raw_spin_unlock_irq_rcu_node(rnp); |
894d45bb | 1485 | raw_spin_unlock(&rcu_state.ofl_lock); |
0aa04b05 | 1486 | } |
22212332 | 1487 | rcu_gp_slow(gp_preinit_delay); /* Races with CPU hotplug. */ |
7fdefc10 PM |
1488 | |
1489 | /* | |
1490 | * Set the quiescent-state-needed bits in all the rcu_node | |
9cbc5b97 PM |
1491 | * structures for all currently online CPUs in breadth-first |
1492 | * order, starting from the root rcu_node structure, relying on the | |
1493 | * layout of the tree within the rcu_state.node[] array. Note that | |
1494 | * other CPUs will access only the leaves of the hierarchy, thus | |
1495 | * seeing that no grace period is in progress, at least until the | |
1496 | * corresponding leaf node has been initialized. | |
7fdefc10 PM |
1497 | * |
1498 | * The grace period cannot complete until the initialization | |
1499 | * process finishes, because this kthread handles both. | |
1500 | */ | |
9cbc5b97 | 1501 | rcu_state.gp_state = RCU_GP_INIT; |
aedf4ba9 | 1502 | rcu_for_each_node_breadth_first(rnp) { |
22212332 | 1503 | rcu_gp_slow(gp_init_delay); |
ec2c2976 | 1504 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
da1df50d | 1505 | rdp = this_cpu_ptr(&rcu_data); |
81ab59a3 | 1506 | rcu_preempt_check_blocked_tasks(rnp); |
7fdefc10 | 1507 | rnp->qsmask = rnp->qsmaskinit; |
9cbc5b97 | 1508 | WRITE_ONCE(rnp->gp_seq, rcu_state.gp_seq); |
7fdefc10 | 1509 | if (rnp == rdp->mynode) |
c7e48f7b | 1510 | (void)__note_gp_changes(rnp, rdp); |
7fdefc10 | 1511 | rcu_preempt_boost_start_gp(rnp); |
9cbc5b97 | 1512 | trace_rcu_grace_period_init(rcu_state.name, rnp->gp_seq, |
7fdefc10 PM |
1513 | rnp->level, rnp->grplo, |
1514 | rnp->grphi, rnp->qsmask); | |
ec2c2976 PM |
1515 | /* Quiescent states for tasks on any now-offline CPUs. */ |
1516 | mask = rnp->qsmask & ~rnp->qsmaskinitnext; | |
f2e2df59 | 1517 | rnp->rcu_gp_init_mask = mask; |
ec2c2976 | 1518 | if ((mask || rnp->wait_blkd_tasks) && rcu_is_leaf_node(rnp)) |
b50912d0 | 1519 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
ec2c2976 PM |
1520 | else |
1521 | raw_spin_unlock_irq_rcu_node(rnp); | |
cee43939 | 1522 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1523 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
7fdefc10 | 1524 | } |
b3dbec76 | 1525 | |
45fed3e7 | 1526 | return true; |
7fdefc10 | 1527 | } |
b3dbec76 | 1528 | |
b9a425cf | 1529 | /* |
b3dae109 | 1530 | * Helper function for swait_event_idle_exclusive() wakeup at force-quiescent-state |
d5374226 | 1531 | * time. |
b9a425cf | 1532 | */ |
0854a05c | 1533 | static bool rcu_gp_fqs_check_wake(int *gfp) |
b9a425cf | 1534 | { |
336a4f6c | 1535 | struct rcu_node *rnp = rcu_get_root(); |
b9a425cf PM |
1536 | |
1537 | /* Someone like call_rcu() requested a force-quiescent-state scan. */ | |
0854a05c | 1538 | *gfp = READ_ONCE(rcu_state.gp_flags); |
b9a425cf PM |
1539 | if (*gfp & RCU_GP_FLAG_FQS) |
1540 | return true; | |
1541 | ||
1542 | /* The current grace period has completed. */ | |
1543 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) | |
1544 | return true; | |
1545 | ||
1546 | return false; | |
1547 | } | |
1548 | ||
4cdfc175 PM |
1549 | /* |
1550 | * Do one round of quiescent-state forcing. | |
1551 | */ | |
0854a05c | 1552 | static void rcu_gp_fqs(bool first_time) |
4cdfc175 | 1553 | { |
336a4f6c | 1554 | struct rcu_node *rnp = rcu_get_root(); |
4cdfc175 | 1555 | |
9cbc5b97 PM |
1556 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
1557 | rcu_state.n_force_qs++; | |
77f81fe0 | 1558 | if (first_time) { |
4cdfc175 | 1559 | /* Collect dyntick-idle snapshots. */ |
e9ecb780 | 1560 | force_qs_rnp(dyntick_save_progress_counter); |
4cdfc175 PM |
1561 | } else { |
1562 | /* Handle dyntick-idle and offline CPUs. */ | |
e9ecb780 | 1563 | force_qs_rnp(rcu_implicit_dynticks_qs); |
4cdfc175 PM |
1564 | } |
1565 | /* Clear flag to prevent immediate re-entry. */ | |
9cbc5b97 | 1566 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 1567 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 PM |
1568 | WRITE_ONCE(rcu_state.gp_flags, |
1569 | READ_ONCE(rcu_state.gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 1570 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 1571 | } |
4cdfc175 PM |
1572 | } |
1573 | ||
c3854a05 PM |
1574 | /* |
1575 | * Loop doing repeated quiescent-state forcing until the grace period ends. | |
1576 | */ | |
1577 | static void rcu_gp_fqs_loop(void) | |
1578 | { | |
1579 | bool first_gp_fqs; | |
1580 | int gf; | |
1581 | unsigned long j; | |
1582 | int ret; | |
1583 | struct rcu_node *rnp = rcu_get_root(); | |
1584 | ||
1585 | first_gp_fqs = true; | |
c06aed0e | 1586 | j = READ_ONCE(jiffies_till_first_fqs); |
c3854a05 PM |
1587 | ret = 0; |
1588 | for (;;) { | |
1589 | if (!ret) { | |
1590 | rcu_state.jiffies_force_qs = jiffies + j; | |
1591 | WRITE_ONCE(rcu_state.jiffies_kick_kthreads, | |
9cf422a8 | 1592 | jiffies + (j ? 3 * j : 2)); |
c3854a05 PM |
1593 | } |
1594 | trace_rcu_grace_period(rcu_state.name, | |
1595 | READ_ONCE(rcu_state.gp_seq), | |
1596 | TPS("fqswait")); | |
1597 | rcu_state.gp_state = RCU_GP_WAIT_FQS; | |
1598 | ret = swait_event_idle_timeout_exclusive( | |
1599 | rcu_state.gp_wq, rcu_gp_fqs_check_wake(&gf), j); | |
1600 | rcu_state.gp_state = RCU_GP_DOING_FQS; | |
1601 | /* Locking provides needed memory barriers. */ | |
1602 | /* If grace period done, leave loop. */ | |
1603 | if (!READ_ONCE(rnp->qsmask) && | |
1604 | !rcu_preempt_blocked_readers_cgp(rnp)) | |
1605 | break; | |
1606 | /* If time for quiescent-state forcing, do it. */ | |
1607 | if (ULONG_CMP_GE(jiffies, rcu_state.jiffies_force_qs) || | |
1608 | (gf & RCU_GP_FLAG_FQS)) { | |
1609 | trace_rcu_grace_period(rcu_state.name, | |
1610 | READ_ONCE(rcu_state.gp_seq), | |
1611 | TPS("fqsstart")); | |
1612 | rcu_gp_fqs(first_gp_fqs); | |
1613 | first_gp_fqs = false; | |
1614 | trace_rcu_grace_period(rcu_state.name, | |
1615 | READ_ONCE(rcu_state.gp_seq), | |
1616 | TPS("fqsend")); | |
1617 | cond_resched_tasks_rcu_qs(); | |
1618 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1619 | ret = 0; /* Force full wait till next FQS. */ | |
c06aed0e | 1620 | j = READ_ONCE(jiffies_till_next_fqs); |
c3854a05 PM |
1621 | } else { |
1622 | /* Deal with stray signal. */ | |
1623 | cond_resched_tasks_rcu_qs(); | |
1624 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1625 | WARN_ON(signal_pending(current)); | |
1626 | trace_rcu_grace_period(rcu_state.name, | |
1627 | READ_ONCE(rcu_state.gp_seq), | |
1628 | TPS("fqswaitsig")); | |
1629 | ret = 1; /* Keep old FQS timing. */ | |
1630 | j = jiffies; | |
1631 | if (time_after(jiffies, rcu_state.jiffies_force_qs)) | |
1632 | j = 1; | |
1633 | else | |
1634 | j = rcu_state.jiffies_force_qs - j; | |
1635 | } | |
1636 | } | |
1637 | } | |
1638 | ||
7fdefc10 PM |
1639 | /* |
1640 | * Clean up after the old grace period. | |
1641 | */ | |
0854a05c | 1642 | static void rcu_gp_cleanup(void) |
7fdefc10 PM |
1643 | { |
1644 | unsigned long gp_duration; | |
48a7639c | 1645 | bool needgp = false; |
de30ad51 | 1646 | unsigned long new_gp_seq; |
7fdefc10 | 1647 | struct rcu_data *rdp; |
336a4f6c | 1648 | struct rcu_node *rnp = rcu_get_root(); |
abedf8e2 | 1649 | struct swait_queue_head *sq; |
b3dbec76 | 1650 | |
9cbc5b97 | 1651 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 1652 | raw_spin_lock_irq_rcu_node(rnp); |
c51d7b5e PM |
1653 | rcu_state.gp_end = jiffies; |
1654 | gp_duration = rcu_state.gp_end - rcu_state.gp_start; | |
9cbc5b97 PM |
1655 | if (gp_duration > rcu_state.gp_max) |
1656 | rcu_state.gp_max = gp_duration; | |
b3dbec76 | 1657 | |
7fdefc10 PM |
1658 | /* |
1659 | * We know the grace period is complete, but to everyone else | |
1660 | * it appears to still be ongoing. But it is also the case | |
1661 | * that to everyone else it looks like there is nothing that | |
1662 | * they can do to advance the grace period. It is therefore | |
1663 | * safe for us to drop the lock in order to mark the grace | |
1664 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 1665 | */ |
67c583a7 | 1666 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 1667 | |
5d4b8659 | 1668 | /* |
ff3bb6f4 PM |
1669 | * Propagate new ->gp_seq value to rcu_node structures so that |
1670 | * other CPUs don't have to wait until the start of the next grace | |
1671 | * period to process their callbacks. This also avoids some nasty | |
1672 | * RCU grace-period initialization races by forcing the end of | |
1673 | * the current grace period to be completely recorded in all of | |
1674 | * the rcu_node structures before the beginning of the next grace | |
1675 | * period is recorded in any of the rcu_node structures. | |
5d4b8659 | 1676 | */ |
9cbc5b97 | 1677 | new_gp_seq = rcu_state.gp_seq; |
de30ad51 | 1678 | rcu_seq_end(&new_gp_seq); |
aedf4ba9 | 1679 | rcu_for_each_node_breadth_first(rnp) { |
2a67e741 | 1680 | raw_spin_lock_irq_rcu_node(rnp); |
4bc8d555 | 1681 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 1682 | dump_blkd_tasks(rnp, 10); |
5c60d25f | 1683 | WARN_ON_ONCE(rnp->qsmask); |
de30ad51 | 1684 | WRITE_ONCE(rnp->gp_seq, new_gp_seq); |
da1df50d | 1685 | rdp = this_cpu_ptr(&rcu_data); |
b11cc576 | 1686 | if (rnp == rdp->mynode) |
c7e48f7b | 1687 | needgp = __note_gp_changes(rnp, rdp) || needgp; |
78e4bc34 | 1688 | /* smp_mb() provided by prior unlock-lock pair. */ |
3481f2ea | 1689 | needgp = rcu_future_gp_cleanup(rnp) || needgp; |
065bb78c | 1690 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 1691 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 1692 | rcu_nocb_gp_cleanup(sq); |
cee43939 | 1693 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1694 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
22212332 | 1695 | rcu_gp_slow(gp_cleanup_delay); |
7fdefc10 | 1696 | } |
336a4f6c | 1697 | rnp = rcu_get_root(); |
9cbc5b97 | 1698 | raw_spin_lock_irq_rcu_node(rnp); /* GP before ->gp_seq update. */ |
7fdefc10 | 1699 | |
0a89e5a4 | 1700 | /* Declare grace period done, trace first to use old GP number. */ |
9cbc5b97 | 1701 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("end")); |
0a89e5a4 | 1702 | rcu_seq_end(&rcu_state.gp_seq); |
9cbc5b97 | 1703 | rcu_state.gp_state = RCU_GP_IDLE; |
fb31340f | 1704 | /* Check for GP requests since above loop. */ |
da1df50d | 1705 | rdp = this_cpu_ptr(&rcu_data); |
5b55072f | 1706 | if (!needgp && ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed)) { |
abd13fdd | 1707 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq_needed, |
41e80595 | 1708 | TPS("CleanupMore")); |
fb31340f PM |
1709 | needgp = true; |
1710 | } | |
48a7639c | 1711 | /* Advance CBs to reduce false positives below. */ |
02f50142 | 1712 | if (!rcu_accelerate_cbs(rnp, rdp) && needgp) { |
9cbc5b97 PM |
1713 | WRITE_ONCE(rcu_state.gp_flags, RCU_GP_FLAG_INIT); |
1714 | rcu_state.gp_req_activity = jiffies; | |
1715 | trace_rcu_grace_period(rcu_state.name, | |
1716 | READ_ONCE(rcu_state.gp_seq), | |
bb311ecc | 1717 | TPS("newreq")); |
18390aea | 1718 | } else { |
9cbc5b97 PM |
1719 | WRITE_ONCE(rcu_state.gp_flags, |
1720 | rcu_state.gp_flags & RCU_GP_FLAG_INIT); | |
bb311ecc | 1721 | } |
67c583a7 | 1722 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 PM |
1723 | } |
1724 | ||
1725 | /* | |
1726 | * Body of kthread that handles grace periods. | |
1727 | */ | |
0854a05c | 1728 | static int __noreturn rcu_gp_kthread(void *unused) |
7fdefc10 | 1729 | { |
5871968d | 1730 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
1731 | for (;;) { |
1732 | ||
1733 | /* Handle grace-period start. */ | |
1734 | for (;;) { | |
9cbc5b97 PM |
1735 | trace_rcu_grace_period(rcu_state.name, |
1736 | READ_ONCE(rcu_state.gp_seq), | |
63c4db78 | 1737 | TPS("reqwait")); |
9cbc5b97 PM |
1738 | rcu_state.gp_state = RCU_GP_WAIT_GPS; |
1739 | swait_event_idle_exclusive(rcu_state.gp_wq, | |
1740 | READ_ONCE(rcu_state.gp_flags) & | |
1741 | RCU_GP_FLAG_INIT); | |
1742 | rcu_state.gp_state = RCU_GP_DONE_GPS; | |
78e4bc34 | 1743 | /* Locking provides needed memory barrier. */ |
0854a05c | 1744 | if (rcu_gp_init()) |
7fdefc10 | 1745 | break; |
cee43939 | 1746 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1747 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
73a860cd | 1748 | WARN_ON(signal_pending(current)); |
9cbc5b97 PM |
1749 | trace_rcu_grace_period(rcu_state.name, |
1750 | READ_ONCE(rcu_state.gp_seq), | |
63c4db78 | 1751 | TPS("reqwaitsig")); |
7fdefc10 | 1752 | } |
cabc49c1 | 1753 | |
4cdfc175 | 1754 | /* Handle quiescent-state forcing. */ |
c3854a05 | 1755 | rcu_gp_fqs_loop(); |
4cdfc175 PM |
1756 | |
1757 | /* Handle grace-period end. */ | |
9cbc5b97 | 1758 | rcu_state.gp_state = RCU_GP_CLEANUP; |
0854a05c | 1759 | rcu_gp_cleanup(); |
9cbc5b97 | 1760 | rcu_state.gp_state = RCU_GP_CLEANED; |
b3dbec76 | 1761 | } |
b3dbec76 PM |
1762 | } |
1763 | ||
f41d911f | 1764 | /* |
49918a54 PM |
1765 | * Report a full set of quiescent states to the rcu_state data structure. |
1766 | * Invoke rcu_gp_kthread_wake() to awaken the grace-period kthread if | |
1767 | * another grace period is required. Whether we wake the grace-period | |
1768 | * kthread or it awakens itself for the next round of quiescent-state | |
1769 | * forcing, that kthread will clean up after the just-completed grace | |
1770 | * period. Note that the caller must hold rnp->lock, which is released | |
1771 | * before return. | |
f41d911f | 1772 | */ |
aff4e9ed | 1773 | static void rcu_report_qs_rsp(unsigned long flags) |
336a4f6c | 1774 | __releases(rcu_get_root()->lock) |
f41d911f | 1775 | { |
336a4f6c | 1776 | raw_lockdep_assert_held_rcu_node(rcu_get_root()); |
de8e8730 | 1777 | WARN_ON_ONCE(!rcu_gp_in_progress()); |
9cbc5b97 PM |
1778 | WRITE_ONCE(rcu_state.gp_flags, |
1779 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
336a4f6c | 1780 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(), flags); |
532c00c9 | 1781 | rcu_gp_kthread_wake(); |
f41d911f PM |
1782 | } |
1783 | ||
64db4cff | 1784 | /* |
d3f6bad3 PM |
1785 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
1786 | * Allows quiescent states for a group of CPUs to be reported at one go | |
1787 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
1788 | * must be represented by the same rcu_node structure (which need not be a |
1789 | * leaf rcu_node structure, though it often will be). The gps parameter | |
1790 | * is the grace-period snapshot, which means that the quiescent states | |
c9a24e2d | 1791 | * are valid only if rnp->gp_seq is equal to gps. That structure's lock |
654e9533 | 1792 | * must be held upon entry, and it is released before return. |
ec2c2976 PM |
1793 | * |
1794 | * As a special case, if mask is zero, the bit-already-cleared check is | |
1795 | * disabled. This allows propagating quiescent state due to resumed tasks | |
1796 | * during grace-period initialization. | |
64db4cff | 1797 | */ |
b50912d0 PM |
1798 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
1799 | unsigned long gps, unsigned long flags) | |
64db4cff PM |
1800 | __releases(rnp->lock) |
1801 | { | |
654e9533 | 1802 | unsigned long oldmask = 0; |
28ecd580 PM |
1803 | struct rcu_node *rnp_c; |
1804 | ||
a32e01ee | 1805 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1806 | |
64db4cff PM |
1807 | /* Walk up the rcu_node hierarchy. */ |
1808 | for (;;) { | |
ec2c2976 | 1809 | if ((!(rnp->qsmask & mask) && mask) || rnp->gp_seq != gps) { |
64db4cff | 1810 | |
654e9533 PM |
1811 | /* |
1812 | * Our bit has already been cleared, or the | |
1813 | * relevant grace period is already over, so done. | |
1814 | */ | |
67c583a7 | 1815 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1816 | return; |
1817 | } | |
654e9533 | 1818 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
5b4c11d5 | 1819 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp) && |
2dee9404 | 1820 | rcu_preempt_blocked_readers_cgp(rnp)); |
64db4cff | 1821 | rnp->qsmask &= ~mask; |
67a0edbf | 1822 | trace_rcu_quiescent_state_report(rcu_state.name, rnp->gp_seq, |
d4c08f2a PM |
1823 | mask, rnp->qsmask, rnp->level, |
1824 | rnp->grplo, rnp->grphi, | |
1825 | !!rnp->gp_tasks); | |
27f4d280 | 1826 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
1827 | |
1828 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 1829 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1830 | return; |
1831 | } | |
d43a5d32 | 1832 | rnp->completedqs = rnp->gp_seq; |
64db4cff PM |
1833 | mask = rnp->grpmask; |
1834 | if (rnp->parent == NULL) { | |
1835 | ||
1836 | /* No more levels. Exit loop holding root lock. */ | |
1837 | ||
1838 | break; | |
1839 | } | |
67c583a7 | 1840 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 1841 | rnp_c = rnp; |
64db4cff | 1842 | rnp = rnp->parent; |
2a67e741 | 1843 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
654e9533 | 1844 | oldmask = rnp_c->qsmask; |
64db4cff PM |
1845 | } |
1846 | ||
1847 | /* | |
1848 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 1849 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 1850 | * to clean up and start the next grace period if one is needed. |
64db4cff | 1851 | */ |
aff4e9ed | 1852 | rcu_report_qs_rsp(flags); /* releases rnp->lock. */ |
64db4cff PM |
1853 | } |
1854 | ||
cc99a310 PM |
1855 | /* |
1856 | * Record a quiescent state for all tasks that were previously queued | |
1857 | * on the specified rcu_node structure and that were blocking the current | |
49918a54 | 1858 | * RCU grace period. The caller must hold the corresponding rnp->lock with |
cc99a310 PM |
1859 | * irqs disabled, and this lock is released upon return, but irqs remain |
1860 | * disabled. | |
1861 | */ | |
17a8212b | 1862 | static void __maybe_unused |
139ad4da | 1863 | rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
cc99a310 PM |
1864 | __releases(rnp->lock) |
1865 | { | |
654e9533 | 1866 | unsigned long gps; |
cc99a310 PM |
1867 | unsigned long mask; |
1868 | struct rcu_node *rnp_p; | |
1869 | ||
a32e01ee | 1870 | raw_lockdep_assert_held_rcu_node(rnp); |
45975c7d | 1871 | if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPT)) || |
c74859d1 PM |
1872 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)) || |
1873 | rnp->qsmask != 0) { | |
67c583a7 | 1874 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
1875 | return; /* Still need more quiescent states! */ |
1876 | } | |
1877 | ||
77cfc7bf | 1878 | rnp->completedqs = rnp->gp_seq; |
cc99a310 PM |
1879 | rnp_p = rnp->parent; |
1880 | if (rnp_p == NULL) { | |
1881 | /* | |
a77da14c PM |
1882 | * Only one rcu_node structure in the tree, so don't |
1883 | * try to report up to its nonexistent parent! | |
cc99a310 | 1884 | */ |
aff4e9ed | 1885 | rcu_report_qs_rsp(flags); |
cc99a310 PM |
1886 | return; |
1887 | } | |
1888 | ||
c9a24e2d PM |
1889 | /* Report up the rest of the hierarchy, tracking current ->gp_seq. */ |
1890 | gps = rnp->gp_seq; | |
cc99a310 | 1891 | mask = rnp->grpmask; |
67c583a7 | 1892 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 1893 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
b50912d0 | 1894 | rcu_report_qs_rnp(mask, rnp_p, gps, flags); |
cc99a310 PM |
1895 | } |
1896 | ||
64db4cff | 1897 | /* |
d3f6bad3 | 1898 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 1899 | * structure. This must be called from the specified CPU. |
64db4cff PM |
1900 | */ |
1901 | static void | |
33085c46 | 1902 | rcu_report_qs_rdp(int cpu, struct rcu_data *rdp) |
64db4cff PM |
1903 | { |
1904 | unsigned long flags; | |
1905 | unsigned long mask; | |
48a7639c | 1906 | bool needwake; |
64db4cff PM |
1907 | struct rcu_node *rnp; |
1908 | ||
1909 | rnp = rdp->mynode; | |
2a67e741 | 1910 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
c9a24e2d PM |
1911 | if (rdp->cpu_no_qs.b.norm || rdp->gp_seq != rnp->gp_seq || |
1912 | rdp->gpwrap) { | |
64db4cff PM |
1913 | |
1914 | /* | |
e4cc1f22 PM |
1915 | * The grace period in which this quiescent state was |
1916 | * recorded has ended, so don't report it upwards. | |
1917 | * We will instead need a new quiescent state that lies | |
1918 | * within the current grace period. | |
64db4cff | 1919 | */ |
5b74c458 | 1920 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
67c583a7 | 1921 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1922 | return; |
1923 | } | |
1924 | mask = rdp->grpmask; | |
b2eb85b4 | 1925 | rdp->core_needs_qs = false; |
64db4cff | 1926 | if ((rnp->qsmask & mask) == 0) { |
67c583a7 | 1927 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1928 | } else { |
64db4cff PM |
1929 | /* |
1930 | * This GP can't end until cpu checks in, so all of our | |
1931 | * callbacks can be processed during the next GP. | |
1932 | */ | |
02f50142 | 1933 | needwake = rcu_accelerate_cbs(rnp, rdp); |
64db4cff | 1934 | |
b50912d0 | 1935 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
654e9533 | 1936 | /* ^^^ Released rnp->lock */ |
48a7639c | 1937 | if (needwake) |
532c00c9 | 1938 | rcu_gp_kthread_wake(); |
64db4cff PM |
1939 | } |
1940 | } | |
1941 | ||
1942 | /* | |
1943 | * Check to see if there is a new grace period of which this CPU | |
1944 | * is not yet aware, and if so, set up local rcu_data state for it. | |
1945 | * Otherwise, see if this CPU has just passed through its first | |
1946 | * quiescent state for this grace period, and record that fact if so. | |
1947 | */ | |
1948 | static void | |
8087d3e3 | 1949 | rcu_check_quiescent_state(struct rcu_data *rdp) |
64db4cff | 1950 | { |
05eb552b | 1951 | /* Check for grace-period ends and beginnings. */ |
15cabdff | 1952 | note_gp_changes(rdp); |
64db4cff PM |
1953 | |
1954 | /* | |
1955 | * Does this CPU still need to do its part for current grace period? | |
1956 | * If no, return and let the other CPUs do their part as well. | |
1957 | */ | |
97c668b8 | 1958 | if (!rdp->core_needs_qs) |
64db4cff PM |
1959 | return; |
1960 | ||
1961 | /* | |
1962 | * Was there a quiescent state since the beginning of the grace | |
1963 | * period? If no, then exit and wait for the next call. | |
1964 | */ | |
3a19b46a | 1965 | if (rdp->cpu_no_qs.b.norm) |
64db4cff PM |
1966 | return; |
1967 | ||
d3f6bad3 PM |
1968 | /* |
1969 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
1970 | * judge of that). | |
1971 | */ | |
33085c46 | 1972 | rcu_report_qs_rdp(rdp->cpu, rdp); |
64db4cff PM |
1973 | } |
1974 | ||
b1420f1c | 1975 | /* |
780cd590 PM |
1976 | * Near the end of the offline process. Trace the fact that this CPU |
1977 | * is going offline. | |
b1420f1c | 1978 | */ |
780cd590 | 1979 | int rcutree_dying_cpu(unsigned int cpu) |
b1420f1c | 1980 | { |
4f5fbd78 YS |
1981 | bool blkd; |
1982 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); | |
1983 | struct rcu_node *rnp = rdp->mynode; | |
b1420f1c | 1984 | |
ea46351c | 1985 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 1986 | return 0; |
ea46351c | 1987 | |
4f5fbd78 | 1988 | blkd = !!(rnp->qsmask & rdp->grpmask); |
780cd590 | 1989 | trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, |
477351f7 | 1990 | blkd ? TPS("cpuofl") : TPS("cpuofl-bgp")); |
780cd590 | 1991 | return 0; |
64db4cff PM |
1992 | } |
1993 | ||
8af3a5e7 PM |
1994 | /* |
1995 | * All CPUs for the specified rcu_node structure have gone offline, | |
1996 | * and all tasks that were preempted within an RCU read-side critical | |
1997 | * section while running on one of those CPUs have since exited their RCU | |
1998 | * read-side critical section. Some other CPU is reporting this fact with | |
1999 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2000 | * This function therefore goes up the tree of rcu_node structures, | |
2001 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2002 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
c50cbe53 | 2003 | * updated. |
8af3a5e7 PM |
2004 | * |
2005 | * This function does check that the specified rcu_node structure has | |
2006 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2007 | * prematurely. That said, invoking it after the fact will cost you | |
2008 | * a needless lock acquisition. So once it has done its work, don't | |
2009 | * invoke it again. | |
2010 | */ | |
2011 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2012 | { | |
2013 | long mask; | |
2014 | struct rcu_node *rnp = rnp_leaf; | |
2015 | ||
962aff03 | 2016 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
ea46351c | 2017 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
962aff03 PM |
2018 | WARN_ON_ONCE(rnp_leaf->qsmaskinit) || |
2019 | WARN_ON_ONCE(rcu_preempt_has_tasks(rnp_leaf))) | |
8af3a5e7 PM |
2020 | return; |
2021 | for (;;) { | |
2022 | mask = rnp->grpmask; | |
2023 | rnp = rnp->parent; | |
2024 | if (!rnp) | |
2025 | break; | |
2a67e741 | 2026 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2027 | rnp->qsmaskinit &= ~mask; |
962aff03 PM |
2028 | /* Between grace periods, so better already be zero! */ |
2029 | WARN_ON_ONCE(rnp->qsmask); | |
8af3a5e7 | 2030 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2031 | raw_spin_unlock_rcu_node(rnp); |
2032 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2033 | return; |
2034 | } | |
67c583a7 | 2035 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2036 | } |
2037 | } | |
2038 | ||
64db4cff | 2039 | /* |
e5601400 | 2040 | * The CPU has been completely removed, and some other CPU is reporting |
a58163d8 PM |
2041 | * this fact from process context. Do the remainder of the cleanup. |
2042 | * There can only be one CPU hotplug operation at a time, so no need for | |
2043 | * explicit locking. | |
64db4cff | 2044 | */ |
780cd590 | 2045 | int rcutree_dead_cpu(unsigned int cpu) |
64db4cff | 2046 | { |
da1df50d | 2047 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
b1420f1c | 2048 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2049 | |
ea46351c | 2050 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 2051 | return 0; |
ea46351c | 2052 | |
2036d94a | 2053 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2054 | rcu_boost_kthread_setaffinity(rnp, -1); |
780cd590 PM |
2055 | /* Do any needed no-CB deferred wakeups from this CPU. */ |
2056 | do_nocb_deferred_wakeup(per_cpu_ptr(&rcu_data, cpu)); | |
2057 | return 0; | |
64db4cff PM |
2058 | } |
2059 | ||
64db4cff PM |
2060 | /* |
2061 | * Invoke any RCU callbacks that have made it to the end of their grace | |
2062 | * period. Thottle as specified by rdp->blimit. | |
2063 | */ | |
5bb5d09c | 2064 | static void rcu_do_batch(struct rcu_data *rdp) |
64db4cff PM |
2065 | { |
2066 | unsigned long flags; | |
15fecf89 PM |
2067 | struct rcu_head *rhp; |
2068 | struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl); | |
2069 | long bl, count; | |
64db4cff | 2070 | |
dc35c893 | 2071 | /* If no callbacks are ready, just return. */ |
15fecf89 | 2072 | if (!rcu_segcblist_ready_cbs(&rdp->cblist)) { |
3c779dfe | 2073 | trace_rcu_batch_start(rcu_state.name, |
15fecf89 PM |
2074 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2075 | rcu_segcblist_n_cbs(&rdp->cblist), 0); | |
3c779dfe | 2076 | trace_rcu_batch_end(rcu_state.name, 0, |
15fecf89 | 2077 | !rcu_segcblist_empty(&rdp->cblist), |
4968c300 PM |
2078 | need_resched(), is_idle_task(current), |
2079 | rcu_is_callbacks_kthread()); | |
64db4cff | 2080 | return; |
29c00b4a | 2081 | } |
64db4cff PM |
2082 | |
2083 | /* | |
2084 | * Extract the list of ready callbacks, disabling to prevent | |
15fecf89 PM |
2085 | * races with call_rcu() from interrupt handlers. Leave the |
2086 | * callback counts, as rcu_barrier() needs to be conservative. | |
64db4cff PM |
2087 | */ |
2088 | local_irq_save(flags); | |
8146c4e2 | 2089 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 2090 | bl = rdp->blimit; |
3c779dfe PM |
2091 | trace_rcu_batch_start(rcu_state.name, |
2092 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), | |
15fecf89 PM |
2093 | rcu_segcblist_n_cbs(&rdp->cblist), bl); |
2094 | rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl); | |
64db4cff PM |
2095 | local_irq_restore(flags); |
2096 | ||
2097 | /* Invoke callbacks. */ | |
15fecf89 PM |
2098 | rhp = rcu_cblist_dequeue(&rcl); |
2099 | for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) { | |
2100 | debug_rcu_head_unqueue(rhp); | |
3c779dfe | 2101 | if (__rcu_reclaim(rcu_state.name, rhp)) |
15fecf89 PM |
2102 | rcu_cblist_dequeued_lazy(&rcl); |
2103 | /* | |
2104 | * Stop only if limit reached and CPU has something to do. | |
2105 | * Note: The rcl structure counts down from zero. | |
2106 | */ | |
4b27f20b | 2107 | if (-rcl.len >= bl && |
dff1672d PM |
2108 | (need_resched() || |
2109 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
2110 | break; |
2111 | } | |
2112 | ||
2113 | local_irq_save(flags); | |
4b27f20b | 2114 | count = -rcl.len; |
3c779dfe | 2115 | trace_rcu_batch_end(rcu_state.name, count, !!rcl.head, need_resched(), |
8ef0f37e | 2116 | is_idle_task(current), rcu_is_callbacks_kthread()); |
64db4cff | 2117 | |
15fecf89 PM |
2118 | /* Update counts and requeue any remaining callbacks. */ |
2119 | rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl); | |
b1420f1c | 2120 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
15fecf89 | 2121 | rcu_segcblist_insert_count(&rdp->cblist, &rcl); |
64db4cff PM |
2122 | |
2123 | /* Reinstate batch limit if we have worked down the excess. */ | |
15fecf89 PM |
2124 | count = rcu_segcblist_n_cbs(&rdp->cblist); |
2125 | if (rdp->blimit == LONG_MAX && count <= qlowmark) | |
64db4cff PM |
2126 | rdp->blimit = blimit; |
2127 | ||
37c72e56 | 2128 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
15fecf89 | 2129 | if (count == 0 && rdp->qlen_last_fqs_check != 0) { |
37c72e56 | 2130 | rdp->qlen_last_fqs_check = 0; |
3c779dfe | 2131 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
15fecf89 PM |
2132 | } else if (count < rdp->qlen_last_fqs_check - qhimark) |
2133 | rdp->qlen_last_fqs_check = count; | |
efd88b02 PM |
2134 | |
2135 | /* | |
2136 | * The following usually indicates a double call_rcu(). To track | |
2137 | * this down, try building with CONFIG_DEBUG_OBJECTS_RCU_HEAD=y. | |
2138 | */ | |
15fecf89 | 2139 | WARN_ON_ONCE(rcu_segcblist_empty(&rdp->cblist) != (count == 0)); |
37c72e56 | 2140 | |
64db4cff PM |
2141 | local_irq_restore(flags); |
2142 | ||
e0f23060 | 2143 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
15fecf89 | 2144 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
a46e0899 | 2145 | invoke_rcu_core(); |
64db4cff PM |
2146 | } |
2147 | ||
2148 | /* | |
c98cac60 PM |
2149 | * This function is invoked from each scheduling-clock interrupt, |
2150 | * and checks to see if this CPU is in a non-context-switch quiescent | |
2151 | * state, for example, user mode or idle loop. It also schedules RCU | |
2152 | * core processing. If the current grace period has gone on too long, | |
2153 | * it will ask the scheduler to manufacture a context switch for the sole | |
2154 | * purpose of providing a providing the needed quiescent state. | |
64db4cff | 2155 | */ |
c98cac60 | 2156 | void rcu_sched_clock_irq(int user) |
64db4cff | 2157 | { |
f7f7bac9 | 2158 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
4e95020c | 2159 | raw_cpu_inc(rcu_data.ticks_this_gp); |
92aa39e9 | 2160 | /* The load-acquire pairs with the store-release setting to true. */ |
2dba13f0 | 2161 | if (smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
92aa39e9 | 2162 | /* Idle and userspace execution already are quiescent states. */ |
a0ef9ec2 | 2163 | if (!rcu_is_cpu_rrupt_from_idle() && !user) { |
92aa39e9 PM |
2164 | set_tsk_need_resched(current); |
2165 | set_preempt_need_resched(); | |
2166 | } | |
2dba13f0 | 2167 | __this_cpu_write(rcu_data.rcu_urgent_qs, false); |
64db4cff | 2168 | } |
c98cac60 | 2169 | rcu_flavor_sched_clock_irq(user); |
e3950ecd | 2170 | if (rcu_pending()) |
a46e0899 | 2171 | invoke_rcu_core(); |
07f27570 | 2172 | |
f7f7bac9 | 2173 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2174 | } |
2175 | ||
64db4cff | 2176 | /* |
5d8a752e ZZ |
2177 | * Scan the leaf rcu_node structures. For each structure on which all |
2178 | * CPUs have reported a quiescent state and on which there are tasks | |
2179 | * blocking the current grace period, initiate RCU priority boosting. | |
2180 | * Otherwise, invoke the specified function to check dyntick state for | |
2181 | * each CPU that has not yet reported a quiescent state. | |
64db4cff | 2182 | */ |
8ff0b907 | 2183 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)) |
64db4cff | 2184 | { |
64db4cff PM |
2185 | int cpu; |
2186 | unsigned long flags; | |
2187 | unsigned long mask; | |
a0b6c9a7 | 2188 | struct rcu_node *rnp; |
64db4cff | 2189 | |
aedf4ba9 | 2190 | rcu_for_each_leaf_node(rnp) { |
cee43939 | 2191 | cond_resched_tasks_rcu_qs(); |
64db4cff | 2192 | mask = 0; |
2a67e741 | 2193 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
a0b6c9a7 | 2194 | if (rnp->qsmask == 0) { |
45975c7d | 2195 | if (!IS_ENABLED(CONFIG_PREEMPT) || |
a77da14c PM |
2196 | rcu_preempt_blocked_readers_cgp(rnp)) { |
2197 | /* | |
2198 | * No point in scanning bits because they | |
2199 | * are all zero. But we might need to | |
2200 | * priority-boost blocked readers. | |
2201 | */ | |
2202 | rcu_initiate_boost(rnp, flags); | |
2203 | /* rcu_initiate_boost() releases rnp->lock */ | |
2204 | continue; | |
2205 | } | |
92816435 PM |
2206 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
2207 | continue; | |
64db4cff | 2208 | } |
bc75e999 MR |
2209 | for_each_leaf_node_possible_cpu(rnp, cpu) { |
2210 | unsigned long bit = leaf_node_cpu_bit(rnp, cpu); | |
0edd1b17 | 2211 | if ((rnp->qsmask & bit) != 0) { |
da1df50d | 2212 | if (f(per_cpu_ptr(&rcu_data, cpu))) |
0edd1b17 PM |
2213 | mask |= bit; |
2214 | } | |
64db4cff | 2215 | } |
45f014c5 | 2216 | if (mask != 0) { |
c9a24e2d | 2217 | /* Idle/offline CPUs, report (releases rnp->lock). */ |
b50912d0 | 2218 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
0aa04b05 PM |
2219 | } else { |
2220 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 2221 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2222 | } |
64db4cff | 2223 | } |
64db4cff PM |
2224 | } |
2225 | ||
2226 | /* | |
2227 | * Force quiescent states on reluctant CPUs, and also detect which | |
2228 | * CPUs are in dyntick-idle mode. | |
2229 | */ | |
cd920e5a | 2230 | void rcu_force_quiescent_state(void) |
64db4cff PM |
2231 | { |
2232 | unsigned long flags; | |
394f2769 PM |
2233 | bool ret; |
2234 | struct rcu_node *rnp; | |
2235 | struct rcu_node *rnp_old = NULL; | |
2236 | ||
2237 | /* Funnel through hierarchy to reduce memory contention. */ | |
da1df50d | 2238 | rnp = __this_cpu_read(rcu_data.mynode); |
394f2769 | 2239 | for (; rnp != NULL; rnp = rnp->parent) { |
67a0edbf | 2240 | ret = (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) || |
394f2769 PM |
2241 | !raw_spin_trylock(&rnp->fqslock); |
2242 | if (rnp_old != NULL) | |
2243 | raw_spin_unlock(&rnp_old->fqslock); | |
d62df573 | 2244 | if (ret) |
394f2769 | 2245 | return; |
394f2769 PM |
2246 | rnp_old = rnp; |
2247 | } | |
336a4f6c | 2248 | /* rnp_old == rcu_get_root(), rnp == NULL. */ |
64db4cff | 2249 | |
394f2769 | 2250 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 2251 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 2252 | raw_spin_unlock(&rnp_old->fqslock); |
67a0edbf | 2253 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
67c583a7 | 2254 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 2255 | return; /* Someone beat us to it. */ |
46a1e34e | 2256 | } |
67a0edbf PM |
2257 | WRITE_ONCE(rcu_state.gp_flags, |
2258 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
67c583a7 | 2259 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
532c00c9 | 2260 | rcu_gp_kthread_wake(); |
64db4cff | 2261 | } |
cd920e5a | 2262 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); |
64db4cff | 2263 | |
fb60e533 | 2264 | /* Perform RCU core processing work for the current CPU. */ |
48d07c04 | 2265 | static __latent_entropy void rcu_core(void) |
64db4cff PM |
2266 | { |
2267 | unsigned long flags; | |
da1df50d | 2268 | struct rcu_data *rdp = raw_cpu_ptr(&rcu_data); |
26d950a9 | 2269 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 2270 | |
b049fdf8 PM |
2271 | if (cpu_is_offline(smp_processor_id())) |
2272 | return; | |
2273 | trace_rcu_utilization(TPS("Start RCU core")); | |
50dc7def | 2274 | WARN_ON_ONCE(!rdp->beenonline); |
2e597558 | 2275 | |
3e310098 | 2276 | /* Report any deferred quiescent states if preemption enabled. */ |
fced9c8c | 2277 | if (!(preempt_count() & PREEMPT_MASK)) { |
3e310098 | 2278 | rcu_preempt_deferred_qs(current); |
fced9c8c PM |
2279 | } else if (rcu_preempt_need_deferred_qs(current)) { |
2280 | set_tsk_need_resched(current); | |
2281 | set_preempt_need_resched(); | |
2282 | } | |
3e310098 | 2283 | |
64db4cff | 2284 | /* Update RCU state based on any recent quiescent states. */ |
8087d3e3 | 2285 | rcu_check_quiescent_state(rdp); |
64db4cff | 2286 | |
bd7af846 | 2287 | /* No grace period and unregistered callbacks? */ |
de8e8730 | 2288 | if (!rcu_gp_in_progress() && |
bd7af846 PM |
2289 | rcu_segcblist_is_enabled(&rdp->cblist)) { |
2290 | local_irq_save(flags); | |
e44e73ca | 2291 | if (!rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
c6e09b97 | 2292 | rcu_accelerate_cbs_unlocked(rnp, rdp); |
e44e73ca | 2293 | local_irq_restore(flags); |
64db4cff PM |
2294 | } |
2295 | ||
791416c4 | 2296 | rcu_check_gp_start_stall(rnp, rdp, rcu_jiffies_till_stall_check()); |
26d950a9 | 2297 | |
64db4cff | 2298 | /* If there are callbacks ready, invoke them. */ |
15fecf89 | 2299 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
aff4e9ed | 2300 | invoke_rcu_callbacks(rdp); |
96d3fd0d PM |
2301 | |
2302 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
2303 | do_nocb_deferred_wakeup(rdp); | |
f7f7bac9 | 2304 | trace_rcu_utilization(TPS("End RCU core")); |
64db4cff PM |
2305 | } |
2306 | ||
48d07c04 SAS |
2307 | static void rcu_core_si(struct softirq_action *h) |
2308 | { | |
2309 | rcu_core(); | |
2310 | } | |
2311 | ||
2312 | static void rcu_wake_cond(struct task_struct *t, int status) | |
2313 | { | |
2314 | /* | |
2315 | * If the thread is yielding, only wake it when this | |
2316 | * is invoked from idle | |
2317 | */ | |
2318 | if (t && (status != RCU_KTHREAD_YIELDING || is_idle_task(current))) | |
2319 | wake_up_process(t); | |
2320 | } | |
2321 | ||
2322 | static void invoke_rcu_core_kthread(void) | |
2323 | { | |
2324 | struct task_struct *t; | |
2325 | unsigned long flags; | |
2326 | ||
2327 | local_irq_save(flags); | |
2328 | __this_cpu_write(rcu_data.rcu_cpu_has_work, 1); | |
2329 | t = __this_cpu_read(rcu_data.rcu_cpu_kthread_task); | |
2330 | if (t != NULL && t != current) | |
2331 | rcu_wake_cond(t, __this_cpu_read(rcu_data.rcu_cpu_kthread_status)); | |
2332 | local_irq_restore(flags); | |
2333 | } | |
2334 | ||
a26ac245 | 2335 | /* |
48d07c04 SAS |
2336 | * Do RCU callback invocation. Not that if we are running !use_softirq, |
2337 | * we are already in the rcuc kthread. If callbacks are offloaded, then | |
2338 | * ->cblist is always empty, so we don't get here. Therefore, we only | |
2339 | * ever need to check for the scheduler being operational (some callbacks | |
2340 | * do wakeups, so we do need the scheduler). | |
a26ac245 | 2341 | */ |
aff4e9ed | 2342 | static void invoke_rcu_callbacks(struct rcu_data *rdp) |
a26ac245 | 2343 | { |
7d0ae808 | 2344 | if (unlikely(!READ_ONCE(rcu_scheduler_fully_active))) |
b0d30417 | 2345 | return; |
48d07c04 | 2346 | rcu_do_batch(rdp); |
a26ac245 PM |
2347 | } |
2348 | ||
48d07c04 SAS |
2349 | /* |
2350 | * Wake up this CPU's rcuc kthread to do RCU core processing. | |
2351 | */ | |
a46e0899 | 2352 | static void invoke_rcu_core(void) |
09223371 | 2353 | { |
48d07c04 SAS |
2354 | if (!cpu_online(smp_processor_id())) |
2355 | return; | |
2356 | if (use_softirq) | |
b0f74036 | 2357 | raise_softirq(RCU_SOFTIRQ); |
48d07c04 SAS |
2358 | else |
2359 | invoke_rcu_core_kthread(); | |
2360 | } | |
2361 | ||
2362 | static void rcu_cpu_kthread_park(unsigned int cpu) | |
2363 | { | |
2364 | per_cpu(rcu_data.rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; | |
2365 | } | |
2366 | ||
2367 | static int rcu_cpu_kthread_should_run(unsigned int cpu) | |
2368 | { | |
2369 | return __this_cpu_read(rcu_data.rcu_cpu_has_work); | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces | |
2374 | * the RCU softirq used in configurations of RCU that do not support RCU | |
2375 | * priority boosting. | |
2376 | */ | |
2377 | static void rcu_cpu_kthread(unsigned int cpu) | |
2378 | { | |
2379 | unsigned int *statusp = this_cpu_ptr(&rcu_data.rcu_cpu_kthread_status); | |
2380 | char work, *workp = this_cpu_ptr(&rcu_data.rcu_cpu_has_work); | |
2381 | int spincnt; | |
2382 | ||
2383 | for (spincnt = 0; spincnt < 10; spincnt++) { | |
2384 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); | |
2385 | local_bh_disable(); | |
2386 | *statusp = RCU_KTHREAD_RUNNING; | |
2387 | local_irq_disable(); | |
2388 | work = *workp; | |
2389 | *workp = 0; | |
2390 | local_irq_enable(); | |
2391 | if (work) | |
2392 | rcu_core(); | |
2393 | local_bh_enable(); | |
2394 | if (*workp == 0) { | |
2395 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); | |
2396 | *statusp = RCU_KTHREAD_WAITING; | |
2397 | return; | |
2398 | } | |
2399 | } | |
2400 | *statusp = RCU_KTHREAD_YIELDING; | |
2401 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); | |
2402 | schedule_timeout_interruptible(2); | |
2403 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); | |
2404 | *statusp = RCU_KTHREAD_WAITING; | |
2405 | } | |
2406 | ||
2407 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { | |
2408 | .store = &rcu_data.rcu_cpu_kthread_task, | |
2409 | .thread_should_run = rcu_cpu_kthread_should_run, | |
2410 | .thread_fn = rcu_cpu_kthread, | |
2411 | .thread_comm = "rcuc/%u", | |
2412 | .setup = rcu_cpu_kthread_setup, | |
2413 | .park = rcu_cpu_kthread_park, | |
2414 | }; | |
2415 | ||
2416 | /* | |
2417 | * Spawn per-CPU RCU core processing kthreads. | |
2418 | */ | |
2419 | static int __init rcu_spawn_core_kthreads(void) | |
2420 | { | |
2421 | int cpu; | |
2422 | ||
2423 | for_each_possible_cpu(cpu) | |
2424 | per_cpu(rcu_data.rcu_cpu_has_work, cpu) = 0; | |
2425 | if (!IS_ENABLED(CONFIG_RCU_BOOST) && use_softirq) | |
2426 | return 0; | |
2427 | WARN_ONCE(smpboot_register_percpu_thread(&rcu_cpu_thread_spec), | |
2428 | "%s: Could not start rcuc kthread, OOM is now expected behavior\n", __func__); | |
2429 | return 0; | |
09223371 | 2430 | } |
48d07c04 | 2431 | early_initcall(rcu_spawn_core_kthreads); |
09223371 | 2432 | |
29154c57 PM |
2433 | /* |
2434 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
2435 | */ | |
5c7d8967 PM |
2436 | static void __call_rcu_core(struct rcu_data *rdp, struct rcu_head *head, |
2437 | unsigned long flags) | |
64db4cff | 2438 | { |
62fde6ed PM |
2439 | /* |
2440 | * If called from an extended quiescent state, invoke the RCU | |
2441 | * core in order to force a re-evaluation of RCU's idleness. | |
2442 | */ | |
9910affa | 2443 | if (!rcu_is_watching()) |
62fde6ed PM |
2444 | invoke_rcu_core(); |
2445 | ||
a16b7a69 | 2446 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 2447 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 2448 | return; |
64db4cff | 2449 | |
37c72e56 PM |
2450 | /* |
2451 | * Force the grace period if too many callbacks or too long waiting. | |
cd920e5a | 2452 | * Enforce hysteresis, and don't invoke rcu_force_quiescent_state() |
37c72e56 | 2453 | * if some other CPU has recently done so. Also, don't bother |
cd920e5a | 2454 | * invoking rcu_force_quiescent_state() if the newly enqueued callback |
37c72e56 PM |
2455 | * is the only one waiting for a grace period to complete. |
2456 | */ | |
15fecf89 PM |
2457 | if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) > |
2458 | rdp->qlen_last_fqs_check + qhimark)) { | |
b52573d2 PM |
2459 | |
2460 | /* Are we ignoring a completed grace period? */ | |
15cabdff | 2461 | note_gp_changes(rdp); |
b52573d2 PM |
2462 | |
2463 | /* Start a new grace period if one not already started. */ | |
de8e8730 | 2464 | if (!rcu_gp_in_progress()) { |
c6e09b97 | 2465 | rcu_accelerate_cbs_unlocked(rdp->mynode, rdp); |
b52573d2 PM |
2466 | } else { |
2467 | /* Give the grace period a kick. */ | |
2468 | rdp->blimit = LONG_MAX; | |
5c7d8967 | 2469 | if (rcu_state.n_force_qs == rdp->n_force_qs_snap && |
15fecf89 | 2470 | rcu_segcblist_first_pend_cb(&rdp->cblist) != head) |
cd920e5a | 2471 | rcu_force_quiescent_state(); |
5c7d8967 | 2472 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
15fecf89 | 2473 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); |
b52573d2 | 2474 | } |
4cdfc175 | 2475 | } |
29154c57 PM |
2476 | } |
2477 | ||
ae150184 PM |
2478 | /* |
2479 | * RCU callback function to leak a callback. | |
2480 | */ | |
2481 | static void rcu_leak_callback(struct rcu_head *rhp) | |
2482 | { | |
2483 | } | |
2484 | ||
3fbfbf7a PM |
2485 | /* |
2486 | * Helper function for call_rcu() and friends. The cpu argument will | |
2487 | * normally be -1, indicating "currently running CPU". It may specify | |
dd46a788 | 2488 | * a CPU only if that CPU is a no-CBs CPU. Currently, only rcu_barrier() |
3fbfbf7a PM |
2489 | * is expected to specify a CPU. |
2490 | */ | |
64db4cff | 2491 | static void |
5c7d8967 | 2492 | __call_rcu(struct rcu_head *head, rcu_callback_t func, int cpu, bool lazy) |
64db4cff PM |
2493 | { |
2494 | unsigned long flags; | |
2495 | struct rcu_data *rdp; | |
2496 | ||
b8f2ed53 PM |
2497 | /* Misaligned rcu_head! */ |
2498 | WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1)); | |
2499 | ||
ae150184 | 2500 | if (debug_rcu_head_queue(head)) { |
fa3c6647 PM |
2501 | /* |
2502 | * Probable double call_rcu(), so leak the callback. | |
2503 | * Use rcu:rcu_callback trace event to find the previous | |
2504 | * time callback was passed to __call_rcu(). | |
2505 | */ | |
d75f773c | 2506 | WARN_ONCE(1, "__call_rcu(): Double-freed CB %p->%pS()!!!\n", |
fa3c6647 | 2507 | head, head->func); |
7d0ae808 | 2508 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
2509 | return; |
2510 | } | |
64db4cff PM |
2511 | head->func = func; |
2512 | head->next = NULL; | |
64db4cff | 2513 | local_irq_save(flags); |
da1df50d | 2514 | rdp = this_cpu_ptr(&rcu_data); |
64db4cff PM |
2515 | |
2516 | /* Add the callback to our list. */ | |
15fecf89 | 2517 | if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist)) || cpu != -1) { |
3fbfbf7a PM |
2518 | int offline; |
2519 | ||
2520 | if (cpu != -1) | |
da1df50d | 2521 | rdp = per_cpu_ptr(&rcu_data, cpu); |
143da9c2 PM |
2522 | if (likely(rdp->mynode)) { |
2523 | /* Post-boot, so this should be for a no-CBs CPU. */ | |
2524 | offline = !__call_rcu_nocb(rdp, head, lazy, flags); | |
2525 | WARN_ON_ONCE(offline); | |
2526 | /* Offline CPU, _call_rcu() illegal, leak callback. */ | |
2527 | local_irq_restore(flags); | |
2528 | return; | |
2529 | } | |
2530 | /* | |
2531 | * Very early boot, before rcu_init(). Initialize if needed | |
2532 | * and then drop through to queue the callback. | |
2533 | */ | |
08543bda | 2534 | WARN_ON_ONCE(cpu != -1); |
34404ca8 | 2535 | WARN_ON_ONCE(!rcu_is_watching()); |
15fecf89 PM |
2536 | if (rcu_segcblist_empty(&rdp->cblist)) |
2537 | rcu_segcblist_init(&rdp->cblist); | |
0d8ee37e | 2538 | } |
15fecf89 | 2539 | rcu_segcblist_enqueue(&rdp->cblist, head, lazy); |
d4c08f2a | 2540 | if (__is_kfree_rcu_offset((unsigned long)func)) |
3c779dfe PM |
2541 | trace_rcu_kfree_callback(rcu_state.name, head, |
2542 | (unsigned long)func, | |
15fecf89 PM |
2543 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2544 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2545 | else |
3c779dfe | 2546 | trace_rcu_callback(rcu_state.name, head, |
15fecf89 PM |
2547 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), |
2548 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
d4c08f2a | 2549 | |
29154c57 | 2550 | /* Go handle any RCU core processing required. */ |
5c7d8967 | 2551 | __call_rcu_core(rdp, head, flags); |
64db4cff PM |
2552 | local_irq_restore(flags); |
2553 | } | |
2554 | ||
a68a2bb2 | 2555 | /** |
45975c7d | 2556 | * call_rcu() - Queue an RCU callback for invocation after a grace period. |
a68a2bb2 PM |
2557 | * @head: structure to be used for queueing the RCU updates. |
2558 | * @func: actual callback function to be invoked after the grace period | |
2559 | * | |
2560 | * The callback function will be invoked some time after a full grace | |
45975c7d PM |
2561 | * period elapses, in other words after all pre-existing RCU read-side |
2562 | * critical sections have completed. However, the callback function | |
2563 | * might well execute concurrently with RCU read-side critical sections | |
2564 | * that started after call_rcu() was invoked. RCU read-side critical | |
2565 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), and | |
2566 | * may be nested. In addition, regions of code across which interrupts, | |
2567 | * preemption, or softirqs have been disabled also serve as RCU read-side | |
2568 | * critical sections. This includes hardware interrupt handlers, softirq | |
2569 | * handlers, and NMI handlers. | |
2570 | * | |
2571 | * Note that all CPUs must agree that the grace period extended beyond | |
2572 | * all pre-existing RCU read-side critical section. On systems with more | |
2573 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
2574 | * guaranteed to have executed a full memory barrier since the end of its | |
2575 | * last RCU read-side critical section whose beginning preceded the call | |
2576 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
2577 | * critical section that continues beyond the start of "func()" must have | |
2578 | * executed a memory barrier after the call_rcu() but before the beginning | |
2579 | * of that RCU read-side critical section. Note that these guarantees | |
2580 | * include CPUs that are offline, idle, or executing in user mode, as | |
2581 | * well as CPUs that are executing in the kernel. | |
2582 | * | |
2583 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
2584 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
2585 | * guaranteed to execute a full memory barrier during the time interval | |
2586 | * between the call to call_rcu() and the invocation of "func()" -- even | |
2587 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
2588 | * more than one CPU). | |
2589 | */ | |
2590 | void call_rcu(struct rcu_head *head, rcu_callback_t func) | |
2591 | { | |
5c7d8967 | 2592 | __call_rcu(head, func, -1, 0); |
45975c7d PM |
2593 | } |
2594 | EXPORT_SYMBOL_GPL(call_rcu); | |
64db4cff | 2595 | |
495aa969 ACB |
2596 | /* |
2597 | * Queue an RCU callback for lazy invocation after a grace period. | |
2598 | * This will likely be later named something like "call_rcu_lazy()", | |
2599 | * but this change will require some way of tagging the lazy RCU | |
2600 | * callbacks in the list of pending callbacks. Until then, this | |
2601 | * function may only be called from __kfree_rcu(). | |
2602 | */ | |
98ece508 | 2603 | void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func) |
495aa969 | 2604 | { |
5c7d8967 | 2605 | __call_rcu(head, func, -1, 1); |
495aa969 ACB |
2606 | } |
2607 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
2608 | ||
e5bc3af7 PM |
2609 | /* |
2610 | * During early boot, any blocking grace-period wait automatically | |
2611 | * implies a grace period. Later on, this is never the case for PREEMPT. | |
2612 | * | |
2613 | * Howevr, because a context switch is a grace period for !PREEMPT, any | |
2614 | * blocking grace-period wait automatically implies a grace period if | |
2615 | * there is only one CPU online at any point time during execution of | |
2616 | * either synchronize_rcu() or synchronize_rcu_expedited(). It is OK to | |
2617 | * occasionally incorrectly indicate that there are multiple CPUs online | |
2618 | * when there was in fact only one the whole time, as this just adds some | |
2619 | * overhead: RCU still operates correctly. | |
2620 | */ | |
2621 | static int rcu_blocking_is_gp(void) | |
2622 | { | |
2623 | int ret; | |
2624 | ||
2625 | if (IS_ENABLED(CONFIG_PREEMPT)) | |
2626 | return rcu_scheduler_active == RCU_SCHEDULER_INACTIVE; | |
2627 | might_sleep(); /* Check for RCU read-side critical section. */ | |
2628 | preempt_disable(); | |
2629 | ret = num_online_cpus() <= 1; | |
2630 | preempt_enable(); | |
2631 | return ret; | |
2632 | } | |
2633 | ||
2634 | /** | |
2635 | * synchronize_rcu - wait until a grace period has elapsed. | |
2636 | * | |
2637 | * Control will return to the caller some time after a full grace | |
2638 | * period has elapsed, in other words after all currently executing RCU | |
2639 | * read-side critical sections have completed. Note, however, that | |
2640 | * upon return from synchronize_rcu(), the caller might well be executing | |
2641 | * concurrently with new RCU read-side critical sections that began while | |
2642 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
2643 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
2644 | * In addition, regions of code across which interrupts, preemption, or | |
2645 | * softirqs have been disabled also serve as RCU read-side critical | |
2646 | * sections. This includes hardware interrupt handlers, softirq handlers, | |
2647 | * and NMI handlers. | |
2648 | * | |
2649 | * Note that this guarantee implies further memory-ordering guarantees. | |
2650 | * On systems with more than one CPU, when synchronize_rcu() returns, | |
2651 | * each CPU is guaranteed to have executed a full memory barrier since | |
2652 | * the end of its last RCU read-side critical section whose beginning | |
2653 | * preceded the call to synchronize_rcu(). In addition, each CPU having | |
2654 | * an RCU read-side critical section that extends beyond the return from | |
2655 | * synchronize_rcu() is guaranteed to have executed a full memory barrier | |
2656 | * after the beginning of synchronize_rcu() and before the beginning of | |
2657 | * that RCU read-side critical section. Note that these guarantees include | |
2658 | * CPUs that are offline, idle, or executing in user mode, as well as CPUs | |
2659 | * that are executing in the kernel. | |
2660 | * | |
2661 | * Furthermore, if CPU A invoked synchronize_rcu(), which returned | |
2662 | * to its caller on CPU B, then both CPU A and CPU B are guaranteed | |
2663 | * to have executed a full memory barrier during the execution of | |
2664 | * synchronize_rcu() -- even if CPU A and CPU B are the same CPU (but | |
2665 | * again only if the system has more than one CPU). | |
2666 | */ | |
2667 | void synchronize_rcu(void) | |
2668 | { | |
2669 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || | |
2670 | lock_is_held(&rcu_lock_map) || | |
2671 | lock_is_held(&rcu_sched_lock_map), | |
2672 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
2673 | if (rcu_blocking_is_gp()) | |
2674 | return; | |
2675 | if (rcu_gp_is_expedited()) | |
2676 | synchronize_rcu_expedited(); | |
2677 | else | |
2678 | wait_rcu_gp(call_rcu); | |
2679 | } | |
2680 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
2681 | ||
765a3f4f PM |
2682 | /** |
2683 | * get_state_synchronize_rcu - Snapshot current RCU state | |
2684 | * | |
2685 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
2686 | * to determine whether or not a full grace period has elapsed in the | |
2687 | * meantime. | |
2688 | */ | |
2689 | unsigned long get_state_synchronize_rcu(void) | |
2690 | { | |
2691 | /* | |
2692 | * Any prior manipulation of RCU-protected data must happen | |
e4be81a2 | 2693 | * before the load from ->gp_seq. |
765a3f4f PM |
2694 | */ |
2695 | smp_mb(); /* ^^^ */ | |
16fc9c60 | 2696 | return rcu_seq_snap(&rcu_state.gp_seq); |
765a3f4f PM |
2697 | } |
2698 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
2699 | ||
2700 | /** | |
2701 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
2702 | * | |
2703 | * @oldstate: return value from earlier call to get_state_synchronize_rcu() | |
2704 | * | |
2705 | * If a full RCU grace period has elapsed since the earlier call to | |
2706 | * get_state_synchronize_rcu(), just return. Otherwise, invoke | |
2707 | * synchronize_rcu() to wait for a full grace period. | |
2708 | * | |
2709 | * Yes, this function does not take counter wrap into account. But | |
2710 | * counter wrap is harmless. If the counter wraps, we have waited for | |
2711 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
2712 | * so waiting for one additional grace period should be just fine. | |
2713 | */ | |
2714 | void cond_synchronize_rcu(unsigned long oldstate) | |
2715 | { | |
16fc9c60 | 2716 | if (!rcu_seq_done(&rcu_state.gp_seq, oldstate)) |
765a3f4f | 2717 | synchronize_rcu(); |
e4be81a2 PM |
2718 | else |
2719 | smp_mb(); /* Ensure GP ends before subsequent accesses. */ | |
765a3f4f PM |
2720 | } |
2721 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
2722 | ||
64db4cff | 2723 | /* |
98ece508 | 2724 | * Check to see if there is any immediate RCU-related work to be done by |
49918a54 PM |
2725 | * the current CPU, returning 1 if so and zero otherwise. The checks are |
2726 | * in order of increasing expense: checks that can be carried out against | |
2727 | * CPU-local state are performed first. However, we must check for CPU | |
2728 | * stalls first, else we might not get a chance. | |
64db4cff | 2729 | */ |
98ece508 | 2730 | static int rcu_pending(void) |
64db4cff | 2731 | { |
98ece508 | 2732 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
2f51f988 PM |
2733 | struct rcu_node *rnp = rdp->mynode; |
2734 | ||
64db4cff | 2735 | /* Check for CPU stalls, if enabled. */ |
ea12ff2b | 2736 | check_cpu_stall(rdp); |
64db4cff | 2737 | |
a096932f | 2738 | /* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */ |
4580b054 | 2739 | if (rcu_nohz_full_cpu()) |
a096932f PM |
2740 | return 0; |
2741 | ||
64db4cff | 2742 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
01c495f7 | 2743 | if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2744 | return 1; |
2745 | ||
2746 | /* Does this CPU have callbacks ready to invoke? */ | |
01c495f7 | 2747 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
64db4cff PM |
2748 | return 1; |
2749 | ||
2750 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
de8e8730 | 2751 | if (!rcu_gp_in_progress() && |
c1935209 PM |
2752 | rcu_segcblist_is_enabled(&rdp->cblist) && |
2753 | !rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) | |
64db4cff PM |
2754 | return 1; |
2755 | ||
67e14c1e PM |
2756 | /* Have RCU grace period completed or started? */ |
2757 | if (rcu_seq_current(&rnp->gp_seq) != rdp->gp_seq || | |
01c495f7 | 2758 | unlikely(READ_ONCE(rdp->gpwrap))) /* outside lock */ |
64db4cff PM |
2759 | return 1; |
2760 | ||
96d3fd0d | 2761 | /* Does this CPU need a deferred NOCB wakeup? */ |
01c495f7 | 2762 | if (rcu_nocb_need_deferred_wakeup(rdp)) |
96d3fd0d | 2763 | return 1; |
96d3fd0d | 2764 | |
64db4cff PM |
2765 | /* nothing to do */ |
2766 | return 0; | |
2767 | } | |
2768 | ||
a83eff0a | 2769 | /* |
dd46a788 | 2770 | * Helper function for rcu_barrier() tracing. If tracing is disabled, |
a83eff0a PM |
2771 | * the compiler is expected to optimize this away. |
2772 | */ | |
dd46a788 | 2773 | static void rcu_barrier_trace(const char *s, int cpu, unsigned long done) |
a83eff0a | 2774 | { |
8344b871 PM |
2775 | trace_rcu_barrier(rcu_state.name, s, cpu, |
2776 | atomic_read(&rcu_state.barrier_cpu_count), done); | |
a83eff0a PM |
2777 | } |
2778 | ||
b1420f1c | 2779 | /* |
dd46a788 PM |
2780 | * RCU callback function for rcu_barrier(). If we are last, wake |
2781 | * up the task executing rcu_barrier(). | |
b1420f1c | 2782 | */ |
24ebbca8 | 2783 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 2784 | { |
ec9f5835 | 2785 | if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) { |
dd46a788 | 2786 | rcu_barrier_trace(TPS("LastCB"), -1, |
ec9f5835 PM |
2787 | rcu_state.barrier_sequence); |
2788 | complete(&rcu_state.barrier_completion); | |
a83eff0a | 2789 | } else { |
dd46a788 | 2790 | rcu_barrier_trace(TPS("CB"), -1, rcu_state.barrier_sequence); |
a83eff0a | 2791 | } |
d0ec774c PM |
2792 | } |
2793 | ||
2794 | /* | |
2795 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
2796 | */ | |
ec9f5835 | 2797 | static void rcu_barrier_func(void *unused) |
d0ec774c | 2798 | { |
da1df50d | 2799 | struct rcu_data *rdp = raw_cpu_ptr(&rcu_data); |
d0ec774c | 2800 | |
dd46a788 | 2801 | rcu_barrier_trace(TPS("IRQ"), -1, rcu_state.barrier_sequence); |
f92c734f PM |
2802 | rdp->barrier_head.func = rcu_barrier_callback; |
2803 | debug_rcu_head_queue(&rdp->barrier_head); | |
2804 | if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head, 0)) { | |
ec9f5835 | 2805 | atomic_inc(&rcu_state.barrier_cpu_count); |
f92c734f PM |
2806 | } else { |
2807 | debug_rcu_head_unqueue(&rdp->barrier_head); | |
dd46a788 | 2808 | rcu_barrier_trace(TPS("IRQNQ"), -1, |
ec9f5835 | 2809 | rcu_state.barrier_sequence); |
f92c734f | 2810 | } |
d0ec774c PM |
2811 | } |
2812 | ||
dd46a788 PM |
2813 | /** |
2814 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
2815 | * | |
2816 | * Note that this primitive does not necessarily wait for an RCU grace period | |
2817 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
2818 | * in the system, then rcu_barrier() is within its rights to return | |
2819 | * immediately, without waiting for anything, much less an RCU grace period. | |
d0ec774c | 2820 | */ |
dd46a788 | 2821 | void rcu_barrier(void) |
d0ec774c | 2822 | { |
b1420f1c | 2823 | int cpu; |
b1420f1c | 2824 | struct rcu_data *rdp; |
ec9f5835 | 2825 | unsigned long s = rcu_seq_snap(&rcu_state.barrier_sequence); |
b1420f1c | 2826 | |
dd46a788 | 2827 | rcu_barrier_trace(TPS("Begin"), -1, s); |
b1420f1c | 2828 | |
e74f4c45 | 2829 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
ec9f5835 | 2830 | mutex_lock(&rcu_state.barrier_mutex); |
b1420f1c | 2831 | |
4f525a52 | 2832 | /* Did someone else do our work for us? */ |
ec9f5835 | 2833 | if (rcu_seq_done(&rcu_state.barrier_sequence, s)) { |
dd46a788 | 2834 | rcu_barrier_trace(TPS("EarlyExit"), -1, |
ec9f5835 | 2835 | rcu_state.barrier_sequence); |
cf3a9c48 | 2836 | smp_mb(); /* caller's subsequent code after above check. */ |
ec9f5835 | 2837 | mutex_unlock(&rcu_state.barrier_mutex); |
cf3a9c48 PM |
2838 | return; |
2839 | } | |
2840 | ||
4f525a52 | 2841 | /* Mark the start of the barrier operation. */ |
ec9f5835 | 2842 | rcu_seq_start(&rcu_state.barrier_sequence); |
dd46a788 | 2843 | rcu_barrier_trace(TPS("Inc1"), -1, rcu_state.barrier_sequence); |
b1420f1c | 2844 | |
d0ec774c | 2845 | /* |
b1420f1c PM |
2846 | * Initialize the count to one rather than to zero in order to |
2847 | * avoid a too-soon return to zero in case of a short grace period | |
1331e7a1 PM |
2848 | * (or preemption of this task). Exclude CPU-hotplug operations |
2849 | * to ensure that no offline CPU has callbacks queued. | |
d0ec774c | 2850 | */ |
ec9f5835 PM |
2851 | init_completion(&rcu_state.barrier_completion); |
2852 | atomic_set(&rcu_state.barrier_cpu_count, 1); | |
1331e7a1 | 2853 | get_online_cpus(); |
b1420f1c PM |
2854 | |
2855 | /* | |
1331e7a1 PM |
2856 | * Force each CPU with callbacks to register a new callback. |
2857 | * When that callback is invoked, we will know that all of the | |
2858 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 2859 | */ |
3fbfbf7a | 2860 | for_each_possible_cpu(cpu) { |
d1e43fa5 | 2861 | if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu)) |
3fbfbf7a | 2862 | continue; |
da1df50d | 2863 | rdp = per_cpu_ptr(&rcu_data, cpu); |
d1e43fa5 | 2864 | if (rcu_is_nocb_cpu(cpu)) { |
4580b054 | 2865 | if (!rcu_nocb_cpu_needs_barrier(cpu)) { |
dd46a788 | 2866 | rcu_barrier_trace(TPS("OfflineNoCB"), cpu, |
ec9f5835 | 2867 | rcu_state.barrier_sequence); |
d7e29933 | 2868 | } else { |
dd46a788 | 2869 | rcu_barrier_trace(TPS("OnlineNoCB"), cpu, |
ec9f5835 | 2870 | rcu_state.barrier_sequence); |
41050a00 | 2871 | smp_mb__before_atomic(); |
ec9f5835 | 2872 | atomic_inc(&rcu_state.barrier_cpu_count); |
d7e29933 | 2873 | __call_rcu(&rdp->barrier_head, |
5c7d8967 | 2874 | rcu_barrier_callback, cpu, 0); |
d7e29933 | 2875 | } |
15fecf89 | 2876 | } else if (rcu_segcblist_n_cbs(&rdp->cblist)) { |
dd46a788 | 2877 | rcu_barrier_trace(TPS("OnlineQ"), cpu, |
ec9f5835 PM |
2878 | rcu_state.barrier_sequence); |
2879 | smp_call_function_single(cpu, rcu_barrier_func, NULL, 1); | |
b1420f1c | 2880 | } else { |
dd46a788 | 2881 | rcu_barrier_trace(TPS("OnlineNQ"), cpu, |
ec9f5835 | 2882 | rcu_state.barrier_sequence); |
b1420f1c PM |
2883 | } |
2884 | } | |
1331e7a1 | 2885 | put_online_cpus(); |
b1420f1c PM |
2886 | |
2887 | /* | |
2888 | * Now that we have an rcu_barrier_callback() callback on each | |
2889 | * CPU, and thus each counted, remove the initial count. | |
2890 | */ | |
ec9f5835 PM |
2891 | if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) |
2892 | complete(&rcu_state.barrier_completion); | |
b1420f1c PM |
2893 | |
2894 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
ec9f5835 | 2895 | wait_for_completion(&rcu_state.barrier_completion); |
b1420f1c | 2896 | |
4f525a52 | 2897 | /* Mark the end of the barrier operation. */ |
dd46a788 | 2898 | rcu_barrier_trace(TPS("Inc2"), -1, rcu_state.barrier_sequence); |
ec9f5835 | 2899 | rcu_seq_end(&rcu_state.barrier_sequence); |
4f525a52 | 2900 | |
b1420f1c | 2901 | /* Other rcu_barrier() invocations can now safely proceed. */ |
ec9f5835 | 2902 | mutex_unlock(&rcu_state.barrier_mutex); |
d0ec774c | 2903 | } |
45975c7d | 2904 | EXPORT_SYMBOL_GPL(rcu_barrier); |
d0ec774c | 2905 | |
0aa04b05 PM |
2906 | /* |
2907 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
2908 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
2909 | * must hold the corresponding leaf rcu_node ->lock with interrrupts | |
2910 | * disabled. | |
2911 | */ | |
2912 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
2913 | { | |
2914 | long mask; | |
8d672fa6 | 2915 | long oldmask; |
0aa04b05 PM |
2916 | struct rcu_node *rnp = rnp_leaf; |
2917 | ||
8d672fa6 | 2918 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
962aff03 | 2919 | WARN_ON_ONCE(rnp->wait_blkd_tasks); |
0aa04b05 PM |
2920 | for (;;) { |
2921 | mask = rnp->grpmask; | |
2922 | rnp = rnp->parent; | |
2923 | if (rnp == NULL) | |
2924 | return; | |
6cf10081 | 2925 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
8d672fa6 | 2926 | oldmask = rnp->qsmaskinit; |
0aa04b05 | 2927 | rnp->qsmaskinit |= mask; |
67c583a7 | 2928 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
8d672fa6 PM |
2929 | if (oldmask) |
2930 | return; | |
0aa04b05 PM |
2931 | } |
2932 | } | |
2933 | ||
64db4cff | 2934 | /* |
27569620 | 2935 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 2936 | */ |
27569620 | 2937 | static void __init |
53b46303 | 2938 | rcu_boot_init_percpu_data(int cpu) |
64db4cff | 2939 | { |
da1df50d | 2940 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27569620 PM |
2941 | |
2942 | /* Set up local state, ensuring consistent view of global state. */ | |
bc75e999 | 2943 | rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu); |
4c5273bf | 2944 | WARN_ON_ONCE(rdp->dynticks_nesting != 1); |
dc5a4f29 | 2945 | WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp))); |
53b46303 | 2946 | rdp->rcu_ofl_gp_seq = rcu_state.gp_seq; |
57738942 | 2947 | rdp->rcu_ofl_gp_flags = RCU_GP_CLEANED; |
53b46303 | 2948 | rdp->rcu_onl_gp_seq = rcu_state.gp_seq; |
57738942 | 2949 | rdp->rcu_onl_gp_flags = RCU_GP_CLEANED; |
27569620 | 2950 | rdp->cpu = cpu; |
3fbfbf7a | 2951 | rcu_boot_init_nocb_percpu_data(rdp); |
27569620 PM |
2952 | } |
2953 | ||
2954 | /* | |
53b46303 PM |
2955 | * Invoked early in the CPU-online process, when pretty much all services |
2956 | * are available. The incoming CPU is not present. | |
2957 | * | |
2958 | * Initializes a CPU's per-CPU RCU data. Note that only one online or | |
ff3bb6f4 PM |
2959 | * offline event can be happening at a given time. Note also that we can |
2960 | * accept some slop in the rsp->gp_seq access due to the fact that this | |
2961 | * CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 2962 | */ |
53b46303 | 2963 | int rcutree_prepare_cpu(unsigned int cpu) |
64db4cff PM |
2964 | { |
2965 | unsigned long flags; | |
da1df50d | 2966 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
336a4f6c | 2967 | struct rcu_node *rnp = rcu_get_root(); |
64db4cff PM |
2968 | |
2969 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 2970 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 | 2971 | rdp->qlen_last_fqs_check = 0; |
53b46303 | 2972 | rdp->n_force_qs_snap = rcu_state.n_force_qs; |
64db4cff | 2973 | rdp->blimit = blimit; |
15fecf89 PM |
2974 | if (rcu_segcblist_empty(&rdp->cblist) && /* No early-boot CBs? */ |
2975 | !init_nocb_callback_list(rdp)) | |
2976 | rcu_segcblist_init(&rdp->cblist); /* Re-enable callbacks. */ | |
4c5273bf | 2977 | rdp->dynticks_nesting = 1; /* CPU not up, no tearing. */ |
2625d469 | 2978 | rcu_dynticks_eqs_online(); |
67c583a7 | 2979 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
64db4cff | 2980 | |
0aa04b05 PM |
2981 | /* |
2982 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
2983 | * propagation up the rcu_node tree will happen at the beginning | |
2984 | * of the next grace period. | |
2985 | */ | |
64db4cff | 2986 | rnp = rdp->mynode; |
2a67e741 | 2987 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
b9585e94 | 2988 | rdp->beenonline = true; /* We have now been online. */ |
de30ad51 | 2989 | rdp->gp_seq = rnp->gp_seq; |
7a1d0f23 | 2990 | rdp->gp_seq_needed = rnp->gp_seq; |
5b74c458 | 2991 | rdp->cpu_no_qs.b.norm = true; |
97c668b8 | 2992 | rdp->core_needs_qs = false; |
9b9500da | 2993 | rdp->rcu_iw_pending = false; |
8aa670cd | 2994 | rdp->rcu_iw_gp_seq = rnp->gp_seq - 1; |
53b46303 | 2995 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuonl")); |
67c583a7 | 2996 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
4df83742 | 2997 | rcu_prepare_kthreads(cpu); |
ad368d15 | 2998 | rcu_spawn_cpu_nocb_kthread(cpu); |
4df83742 TG |
2999 | |
3000 | return 0; | |
3001 | } | |
3002 | ||
deb34f36 PM |
3003 | /* |
3004 | * Update RCU priority boot kthread affinity for CPU-hotplug changes. | |
3005 | */ | |
4df83742 TG |
3006 | static void rcutree_affinity_setting(unsigned int cpu, int outgoing) |
3007 | { | |
da1df50d | 3008 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
4df83742 TG |
3009 | |
3010 | rcu_boost_kthread_setaffinity(rdp->mynode, outgoing); | |
3011 | } | |
3012 | ||
deb34f36 PM |
3013 | /* |
3014 | * Near the end of the CPU-online process. Pretty much all services | |
3015 | * enabled, and the CPU is now very much alive. | |
3016 | */ | |
4df83742 TG |
3017 | int rcutree_online_cpu(unsigned int cpu) |
3018 | { | |
9b9500da PM |
3019 | unsigned long flags; |
3020 | struct rcu_data *rdp; | |
3021 | struct rcu_node *rnp; | |
9b9500da | 3022 | |
b97d23c5 PM |
3023 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3024 | rnp = rdp->mynode; | |
3025 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3026 | rnp->ffmask |= rdp->grpmask; | |
3027 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
9b9500da PM |
3028 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
3029 | return 0; /* Too early in boot for scheduler work. */ | |
3030 | sync_sched_exp_online_cleanup(cpu); | |
3031 | rcutree_affinity_setting(cpu, -1); | |
4df83742 TG |
3032 | return 0; |
3033 | } | |
3034 | ||
deb34f36 PM |
3035 | /* |
3036 | * Near the beginning of the process. The CPU is still very much alive | |
3037 | * with pretty much all services enabled. | |
3038 | */ | |
4df83742 TG |
3039 | int rcutree_offline_cpu(unsigned int cpu) |
3040 | { | |
9b9500da PM |
3041 | unsigned long flags; |
3042 | struct rcu_data *rdp; | |
3043 | struct rcu_node *rnp; | |
9b9500da | 3044 | |
b97d23c5 PM |
3045 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3046 | rnp = rdp->mynode; | |
3047 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3048 | rnp->ffmask &= ~rdp->grpmask; | |
3049 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
9b9500da | 3050 | |
4df83742 TG |
3051 | rcutree_affinity_setting(cpu, cpu); |
3052 | return 0; | |
3053 | } | |
3054 | ||
f64c6013 PZ |
3055 | static DEFINE_PER_CPU(int, rcu_cpu_started); |
3056 | ||
7ec99de3 PM |
3057 | /* |
3058 | * Mark the specified CPU as being online so that subsequent grace periods | |
3059 | * (both expedited and normal) will wait on it. Note that this means that | |
3060 | * incoming CPUs are not allowed to use RCU read-side critical sections | |
3061 | * until this function is called. Failing to observe this restriction | |
3062 | * will result in lockdep splats. | |
deb34f36 PM |
3063 | * |
3064 | * Note that this function is special in that it is invoked directly | |
3065 | * from the incoming CPU rather than from the cpuhp_step mechanism. | |
3066 | * This is because this function must be invoked at a precise location. | |
7ec99de3 PM |
3067 | */ |
3068 | void rcu_cpu_starting(unsigned int cpu) | |
3069 | { | |
3070 | unsigned long flags; | |
3071 | unsigned long mask; | |
313517fc PM |
3072 | int nbits; |
3073 | unsigned long oldmask; | |
7ec99de3 PM |
3074 | struct rcu_data *rdp; |
3075 | struct rcu_node *rnp; | |
7ec99de3 | 3076 | |
f64c6013 PZ |
3077 | if (per_cpu(rcu_cpu_started, cpu)) |
3078 | return; | |
3079 | ||
3080 | per_cpu(rcu_cpu_started, cpu) = 1; | |
3081 | ||
b97d23c5 PM |
3082 | rdp = per_cpu_ptr(&rcu_data, cpu); |
3083 | rnp = rdp->mynode; | |
3084 | mask = rdp->grpmask; | |
3085 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3086 | rnp->qsmaskinitnext |= mask; | |
3087 | oldmask = rnp->expmaskinitnext; | |
3088 | rnp->expmaskinitnext |= mask; | |
3089 | oldmask ^= rnp->expmaskinitnext; | |
3090 | nbits = bitmap_weight(&oldmask, BITS_PER_LONG); | |
3091 | /* Allow lockless access for expedited grace periods. */ | |
eb7a6653 | 3092 | smp_store_release(&rcu_state.ncpus, rcu_state.ncpus + nbits); /* ^^^ */ |
b97d23c5 | 3093 | rcu_gpnum_ovf(rnp, rdp); /* Offline-induced counter wrap? */ |
eb7a6653 PM |
3094 | rdp->rcu_onl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
3095 | rdp->rcu_onl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
b97d23c5 PM |
3096 | if (rnp->qsmask & mask) { /* RCU waiting on incoming CPU? */ |
3097 | /* Report QS -after- changing ->qsmaskinitnext! */ | |
3098 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); | |
3099 | } else { | |
3100 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
7ec99de3 | 3101 | } |
313517fc | 3102 | smp_mb(); /* Ensure RCU read-side usage follows above initialization. */ |
7ec99de3 PM |
3103 | } |
3104 | ||
27d50c7e TG |
3105 | #ifdef CONFIG_HOTPLUG_CPU |
3106 | /* | |
53b46303 PM |
3107 | * The outgoing function has no further need of RCU, so remove it from |
3108 | * the rcu_node tree's ->qsmaskinitnext bit masks. | |
3109 | * | |
3110 | * Note that this function is special in that it is invoked directly | |
3111 | * from the outgoing CPU rather than from the cpuhp_step mechanism. | |
3112 | * This is because this function must be invoked at a precise location. | |
27d50c7e | 3113 | */ |
53b46303 | 3114 | void rcu_report_dead(unsigned int cpu) |
27d50c7e TG |
3115 | { |
3116 | unsigned long flags; | |
3117 | unsigned long mask; | |
da1df50d | 3118 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27d50c7e TG |
3119 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
3120 | ||
49918a54 | 3121 | /* QS for any half-done expedited grace period. */ |
53b46303 | 3122 | preempt_disable(); |
63d4c8c9 | 3123 | rcu_report_exp_rdp(this_cpu_ptr(&rcu_data)); |
53b46303 PM |
3124 | preempt_enable(); |
3125 | rcu_preempt_deferred_qs(current); | |
3126 | ||
27d50c7e TG |
3127 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ |
3128 | mask = rdp->grpmask; | |
894d45bb | 3129 | raw_spin_lock(&rcu_state.ofl_lock); |
27d50c7e | 3130 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ |
53b46303 PM |
3131 | rdp->rcu_ofl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
3132 | rdp->rcu_ofl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
fece2776 PM |
3133 | if (rnp->qsmask & mask) { /* RCU waiting on outgoing CPU? */ |
3134 | /* Report quiescent state -before- changing ->qsmaskinitnext! */ | |
b50912d0 | 3135 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
fece2776 PM |
3136 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
3137 | } | |
27d50c7e | 3138 | rnp->qsmaskinitnext &= ~mask; |
710d60cb | 3139 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
894d45bb | 3140 | raw_spin_unlock(&rcu_state.ofl_lock); |
f64c6013 PZ |
3141 | |
3142 | per_cpu(rcu_cpu_started, cpu) = 0; | |
27d50c7e | 3143 | } |
a58163d8 | 3144 | |
53b46303 PM |
3145 | /* |
3146 | * The outgoing CPU has just passed through the dying-idle state, and we | |
3147 | * are being invoked from the CPU that was IPIed to continue the offline | |
3148 | * operation. Migrate the outgoing CPU's callbacks to the current CPU. | |
3149 | */ | |
3150 | void rcutree_migrate_callbacks(int cpu) | |
a58163d8 PM |
3151 | { |
3152 | unsigned long flags; | |
b1a2d79f | 3153 | struct rcu_data *my_rdp; |
da1df50d | 3154 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
336a4f6c | 3155 | struct rcu_node *rnp_root = rcu_get_root(); |
ec4eacce | 3156 | bool needwake; |
a58163d8 | 3157 | |
95335c03 PM |
3158 | if (rcu_is_nocb_cpu(cpu) || rcu_segcblist_empty(&rdp->cblist)) |
3159 | return; /* No callbacks to migrate. */ | |
3160 | ||
b1a2d79f | 3161 | local_irq_save(flags); |
da1df50d | 3162 | my_rdp = this_cpu_ptr(&rcu_data); |
b1a2d79f PM |
3163 | if (rcu_nocb_adopt_orphan_cbs(my_rdp, rdp, flags)) { |
3164 | local_irq_restore(flags); | |
3165 | return; | |
3166 | } | |
9fa46fb8 | 3167 | raw_spin_lock_rcu_node(rnp_root); /* irqs already disabled. */ |
ec4eacce | 3168 | /* Leverage recent GPs and set GP for new callbacks. */ |
834f56bf PM |
3169 | needwake = rcu_advance_cbs(rnp_root, rdp) || |
3170 | rcu_advance_cbs(rnp_root, my_rdp); | |
f2dbe4a5 | 3171 | rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist); |
09efeeee PM |
3172 | WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) != |
3173 | !rcu_segcblist_n_cbs(&my_rdp->cblist)); | |
537b85c8 | 3174 | raw_spin_unlock_irqrestore_rcu_node(rnp_root, flags); |
ec4eacce | 3175 | if (needwake) |
532c00c9 | 3176 | rcu_gp_kthread_wake(); |
a58163d8 PM |
3177 | WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 || |
3178 | !rcu_segcblist_empty(&rdp->cblist), | |
3179 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n", | |
3180 | cpu, rcu_segcblist_n_cbs(&rdp->cblist), | |
3181 | rcu_segcblist_first_cb(&rdp->cblist)); | |
3182 | } | |
27d50c7e TG |
3183 | #endif |
3184 | ||
deb34f36 PM |
3185 | /* |
3186 | * On non-huge systems, use expedited RCU grace periods to make suspend | |
3187 | * and hibernation run faster. | |
3188 | */ | |
d1d74d14 BP |
3189 | static int rcu_pm_notify(struct notifier_block *self, |
3190 | unsigned long action, void *hcpu) | |
3191 | { | |
3192 | switch (action) { | |
3193 | case PM_HIBERNATION_PREPARE: | |
3194 | case PM_SUSPEND_PREPARE: | |
e85e6a21 | 3195 | rcu_expedite_gp(); |
d1d74d14 BP |
3196 | break; |
3197 | case PM_POST_HIBERNATION: | |
3198 | case PM_POST_SUSPEND: | |
e85e6a21 | 3199 | rcu_unexpedite_gp(); |
d1d74d14 BP |
3200 | break; |
3201 | default: | |
3202 | break; | |
3203 | } | |
3204 | return NOTIFY_OK; | |
3205 | } | |
3206 | ||
b3dbec76 | 3207 | /* |
49918a54 | 3208 | * Spawn the kthreads that handle RCU's grace periods. |
b3dbec76 PM |
3209 | */ |
3210 | static int __init rcu_spawn_gp_kthread(void) | |
3211 | { | |
3212 | unsigned long flags; | |
a94844b2 | 3213 | int kthread_prio_in = kthread_prio; |
b3dbec76 | 3214 | struct rcu_node *rnp; |
a94844b2 | 3215 | struct sched_param sp; |
b3dbec76 PM |
3216 | struct task_struct *t; |
3217 | ||
a94844b2 | 3218 | /* Force priority into range. */ |
c7cd161e JFG |
3219 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 2 |
3220 | && IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)) | |
3221 | kthread_prio = 2; | |
3222 | else if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
a94844b2 PM |
3223 | kthread_prio = 1; |
3224 | else if (kthread_prio < 0) | |
3225 | kthread_prio = 0; | |
3226 | else if (kthread_prio > 99) | |
3227 | kthread_prio = 99; | |
c7cd161e | 3228 | |
a94844b2 PM |
3229 | if (kthread_prio != kthread_prio_in) |
3230 | pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n", | |
3231 | kthread_prio, kthread_prio_in); | |
3232 | ||
9386c0b7 | 3233 | rcu_scheduler_fully_active = 1; |
b97d23c5 | 3234 | t = kthread_create(rcu_gp_kthread, NULL, "%s", rcu_state.name); |
08543bda PM |
3235 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start grace-period kthread, OOM is now expected behavior\n", __func__)) |
3236 | return 0; | |
b97d23c5 PM |
3237 | rnp = rcu_get_root(); |
3238 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3239 | rcu_state.gp_kthread = t; | |
3240 | if (kthread_prio) { | |
3241 | sp.sched_priority = kthread_prio; | |
3242 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
b3dbec76 | 3243 | } |
b97d23c5 PM |
3244 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
3245 | wake_up_process(t); | |
35ce7f29 | 3246 | rcu_spawn_nocb_kthreads(); |
9386c0b7 | 3247 | rcu_spawn_boost_kthreads(); |
b3dbec76 PM |
3248 | return 0; |
3249 | } | |
3250 | early_initcall(rcu_spawn_gp_kthread); | |
3251 | ||
bbad9379 | 3252 | /* |
52d7e48b PM |
3253 | * This function is invoked towards the end of the scheduler's |
3254 | * initialization process. Before this is called, the idle task might | |
3255 | * contain synchronous grace-period primitives (during which time, this idle | |
3256 | * task is booting the system, and such primitives are no-ops). After this | |
3257 | * function is called, any synchronous grace-period primitives are run as | |
3258 | * expedited, with the requesting task driving the grace period forward. | |
900b1028 | 3259 | * A later core_initcall() rcu_set_runtime_mode() will switch to full |
52d7e48b | 3260 | * runtime RCU functionality. |
bbad9379 PM |
3261 | */ |
3262 | void rcu_scheduler_starting(void) | |
3263 | { | |
3264 | WARN_ON(num_online_cpus() != 1); | |
3265 | WARN_ON(nr_context_switches() > 0); | |
52d7e48b PM |
3266 | rcu_test_sync_prims(); |
3267 | rcu_scheduler_active = RCU_SCHEDULER_INIT; | |
3268 | rcu_test_sync_prims(); | |
bbad9379 PM |
3269 | } |
3270 | ||
64db4cff | 3271 | /* |
49918a54 | 3272 | * Helper function for rcu_init() that initializes the rcu_state structure. |
64db4cff | 3273 | */ |
b8bb1f63 | 3274 | static void __init rcu_init_one(void) |
64db4cff | 3275 | { |
cb007102 AG |
3276 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
3277 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
3278 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
3279 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
199977bf | 3280 | |
199977bf | 3281 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ |
64db4cff PM |
3282 | int cpustride = 1; |
3283 | int i; | |
3284 | int j; | |
3285 | struct rcu_node *rnp; | |
3286 | ||
05b84aec | 3287 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 3288 | |
3eaaaf6c PM |
3289 | /* Silence gcc 4.8 false positive about array index out of range. */ |
3290 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
3291 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 3292 | |
64db4cff PM |
3293 | /* Initialize the level-tracking arrays. */ |
3294 | ||
f885b7f2 | 3295 | for (i = 1; i < rcu_num_lvls; i++) |
eb7a6653 PM |
3296 | rcu_state.level[i] = |
3297 | rcu_state.level[i - 1] + num_rcu_lvl[i - 1]; | |
41f5c631 | 3298 | rcu_init_levelspread(levelspread, num_rcu_lvl); |
64db4cff PM |
3299 | |
3300 | /* Initialize the elements themselves, starting from the leaves. */ | |
3301 | ||
f885b7f2 | 3302 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 3303 | cpustride *= levelspread[i]; |
eb7a6653 | 3304 | rnp = rcu_state.level[i]; |
41f5c631 | 3305 | for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) { |
67c583a7 BF |
3306 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
3307 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 3308 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
3309 | raw_spin_lock_init(&rnp->fqslock); |
3310 | lockdep_set_class_and_name(&rnp->fqslock, | |
3311 | &rcu_fqs_class[i], fqs[i]); | |
eb7a6653 PM |
3312 | rnp->gp_seq = rcu_state.gp_seq; |
3313 | rnp->gp_seq_needed = rcu_state.gp_seq; | |
3314 | rnp->completedqs = rcu_state.gp_seq; | |
64db4cff PM |
3315 | rnp->qsmask = 0; |
3316 | rnp->qsmaskinit = 0; | |
3317 | rnp->grplo = j * cpustride; | |
3318 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
3319 | if (rnp->grphi >= nr_cpu_ids) |
3320 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
3321 | if (i == 0) { |
3322 | rnp->grpnum = 0; | |
3323 | rnp->grpmask = 0; | |
3324 | rnp->parent = NULL; | |
3325 | } else { | |
199977bf | 3326 | rnp->grpnum = j % levelspread[i - 1]; |
df63fa5b | 3327 | rnp->grpmask = BIT(rnp->grpnum); |
eb7a6653 | 3328 | rnp->parent = rcu_state.level[i - 1] + |
199977bf | 3329 | j / levelspread[i - 1]; |
64db4cff PM |
3330 | } |
3331 | rnp->level = i; | |
12f5f524 | 3332 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 3333 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
3334 | init_waitqueue_head(&rnp->exp_wq[0]); |
3335 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
3336 | init_waitqueue_head(&rnp->exp_wq[2]); |
3337 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 3338 | spin_lock_init(&rnp->exp_lock); |
64db4cff PM |
3339 | } |
3340 | } | |
0c34029a | 3341 | |
eb7a6653 PM |
3342 | init_swait_queue_head(&rcu_state.gp_wq); |
3343 | init_swait_queue_head(&rcu_state.expedited_wq); | |
aedf4ba9 | 3344 | rnp = rcu_first_leaf_node(); |
0c34029a | 3345 | for_each_possible_cpu(i) { |
4a90a068 | 3346 | while (i > rnp->grphi) |
0c34029a | 3347 | rnp++; |
da1df50d | 3348 | per_cpu_ptr(&rcu_data, i)->mynode = rnp; |
53b46303 | 3349 | rcu_boot_init_percpu_data(i); |
0c34029a | 3350 | } |
64db4cff PM |
3351 | } |
3352 | ||
f885b7f2 PM |
3353 | /* |
3354 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 3355 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
3356 | * the ->node array in the rcu_state structure. |
3357 | */ | |
3358 | static void __init rcu_init_geometry(void) | |
3359 | { | |
026ad283 | 3360 | ulong d; |
f885b7f2 | 3361 | int i; |
05b84aec | 3362 | int rcu_capacity[RCU_NUM_LVLS]; |
f885b7f2 | 3363 | |
026ad283 PM |
3364 | /* |
3365 | * Initialize any unspecified boot parameters. | |
3366 | * The default values of jiffies_till_first_fqs and | |
3367 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
3368 | * value, which is a function of HZ, then adding one for each | |
3369 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
3370 | */ | |
3371 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
3372 | if (jiffies_till_first_fqs == ULONG_MAX) | |
3373 | jiffies_till_first_fqs = d; | |
3374 | if (jiffies_till_next_fqs == ULONG_MAX) | |
3375 | jiffies_till_next_fqs = d; | |
6973032a | 3376 | adjust_jiffies_till_sched_qs(); |
026ad283 | 3377 | |
f885b7f2 | 3378 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 3379 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 3380 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 3381 | return; |
a7538352 | 3382 | pr_info("Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n", |
39479098 | 3383 | rcu_fanout_leaf, nr_cpu_ids); |
f885b7f2 | 3384 | |
f885b7f2 | 3385 | /* |
ee968ac6 PM |
3386 | * The boot-time rcu_fanout_leaf parameter must be at least two |
3387 | * and cannot exceed the number of bits in the rcu_node masks. | |
3388 | * Complain and fall back to the compile-time values if this | |
3389 | * limit is exceeded. | |
f885b7f2 | 3390 | */ |
ee968ac6 | 3391 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 3392 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 3393 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
3394 | WARN_ON(1); |
3395 | return; | |
3396 | } | |
3397 | ||
f885b7f2 PM |
3398 | /* |
3399 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 3400 | * with the given number of levels. |
f885b7f2 | 3401 | */ |
9618138b | 3402 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 3403 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 3404 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
3405 | |
3406 | /* | |
75cf15a4 | 3407 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 3408 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 3409 | */ |
ee968ac6 PM |
3410 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
3411 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
3412 | WARN_ON(1); | |
3413 | return; | |
3414 | } | |
f885b7f2 | 3415 | |
679f9858 | 3416 | /* Calculate the number of levels in the tree. */ |
9618138b | 3417 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 3418 | } |
9618138b | 3419 | rcu_num_lvls = i + 1; |
679f9858 | 3420 | |
f885b7f2 | 3421 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 3422 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 3423 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
3424 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
3425 | } | |
f885b7f2 PM |
3426 | |
3427 | /* Calculate the total number of rcu_node structures. */ | |
3428 | rcu_num_nodes = 0; | |
679f9858 | 3429 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 3430 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
3431 | } |
3432 | ||
a3dc2948 PM |
3433 | /* |
3434 | * Dump out the structure of the rcu_node combining tree associated | |
49918a54 | 3435 | * with the rcu_state structure. |
a3dc2948 | 3436 | */ |
b8bb1f63 | 3437 | static void __init rcu_dump_rcu_node_tree(void) |
a3dc2948 PM |
3438 | { |
3439 | int level = 0; | |
3440 | struct rcu_node *rnp; | |
3441 | ||
3442 | pr_info("rcu_node tree layout dump\n"); | |
3443 | pr_info(" "); | |
aedf4ba9 | 3444 | rcu_for_each_node_breadth_first(rnp) { |
a3dc2948 PM |
3445 | if (rnp->level != level) { |
3446 | pr_cont("\n"); | |
3447 | pr_info(" "); | |
3448 | level = rnp->level; | |
3449 | } | |
3450 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
3451 | } | |
3452 | pr_cont("\n"); | |
3453 | } | |
3454 | ||
ad7c946b | 3455 | struct workqueue_struct *rcu_gp_wq; |
25f3d7ef | 3456 | struct workqueue_struct *rcu_par_gp_wq; |
ad7c946b | 3457 | |
9f680ab4 | 3458 | void __init rcu_init(void) |
64db4cff | 3459 | { |
017c4261 | 3460 | int cpu; |
9f680ab4 | 3461 | |
47627678 PM |
3462 | rcu_early_boot_tests(); |
3463 | ||
f41d911f | 3464 | rcu_bootup_announce(); |
f885b7f2 | 3465 | rcu_init_geometry(); |
b8bb1f63 | 3466 | rcu_init_one(); |
a3dc2948 | 3467 | if (dump_tree) |
b8bb1f63 | 3468 | rcu_dump_rcu_node_tree(); |
48d07c04 SAS |
3469 | if (use_softirq) |
3470 | open_softirq(RCU_SOFTIRQ, rcu_core_si); | |
9f680ab4 PM |
3471 | |
3472 | /* | |
3473 | * We don't need protection against CPU-hotplug here because | |
3474 | * this is called early in boot, before either interrupts | |
3475 | * or the scheduler are operational. | |
3476 | */ | |
d1d74d14 | 3477 | pm_notifier(rcu_pm_notify, 0); |
7ec99de3 | 3478 | for_each_online_cpu(cpu) { |
4df83742 | 3479 | rcutree_prepare_cpu(cpu); |
7ec99de3 | 3480 | rcu_cpu_starting(cpu); |
9b9500da | 3481 | rcutree_online_cpu(cpu); |
7ec99de3 | 3482 | } |
ad7c946b PM |
3483 | |
3484 | /* Create workqueue for expedited GPs and for Tree SRCU. */ | |
3485 | rcu_gp_wq = alloc_workqueue("rcu_gp", WQ_MEM_RECLAIM, 0); | |
3486 | WARN_ON(!rcu_gp_wq); | |
25f3d7ef PM |
3487 | rcu_par_gp_wq = alloc_workqueue("rcu_par_gp", WQ_MEM_RECLAIM, 0); |
3488 | WARN_ON(!rcu_par_gp_wq); | |
e0fcba9a | 3489 | srcu_init(); |
64db4cff PM |
3490 | } |
3491 | ||
10462d6f | 3492 | #include "tree_stall.h" |
3549c2bc | 3493 | #include "tree_exp.h" |
4102adab | 3494 | #include "tree_plugin.h" |