]>
Commit | Line | Data |
---|---|---|
22e40925 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
64db4cff | 2 | /* |
65bb0dc4 | 3 | * Read-Copy Update mechanism for mutual exclusion (tree-based version) |
64db4cff | 4 | * |
64db4cff PM |
5 | * Copyright IBM Corporation, 2008 |
6 | * | |
7 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | |
8 | * Manfred Spraul <manfred@colorfullife.com> | |
65bb0dc4 | 9 | * Paul E. McKenney <paulmck@linux.ibm.com> |
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> | |
f39650de AS |
35 | #include <linux/panic.h> |
36 | #include <linux/panic_notifier.h> | |
64db4cff PM |
37 | #include <linux/percpu.h> |
38 | #include <linux/notifier.h> | |
39 | #include <linux/cpu.h> | |
40 | #include <linux/mutex.h> | |
41 | #include <linux/time.h> | |
bbad9379 | 42 | #include <linux/kernel_stat.h> |
a26ac245 PM |
43 | #include <linux/wait.h> |
44 | #include <linux/kthread.h> | |
ae7e81c0 | 45 | #include <uapi/linux/sched/types.h> |
268bb0ce | 46 | #include <linux/prefetch.h> |
3d3b7db0 | 47 | #include <linux/delay.h> |
661a85dc | 48 | #include <linux/random.h> |
af658dca | 49 | #include <linux/trace_events.h> |
d1d74d14 | 50 | #include <linux/suspend.h> |
a278d471 | 51 | #include <linux/ftrace.h> |
d3052109 | 52 | #include <linux/tick.h> |
2ccaff10 | 53 | #include <linux/sysrq.h> |
c13324a5 | 54 | #include <linux/kprobes.h> |
48d07c04 SAS |
55 | #include <linux/gfp.h> |
56 | #include <linux/oom.h> | |
57 | #include <linux/smpboot.h> | |
58 | #include <linux/jiffies.h> | |
77a40f97 | 59 | #include <linux/slab.h> |
48d07c04 | 60 | #include <linux/sched/isolation.h> |
cfcdef5e | 61 | #include <linux/sched/clock.h> |
5f3c8d62 URS |
62 | #include <linux/vmalloc.h> |
63 | #include <linux/mm.h> | |
26e760c9 | 64 | #include <linux/kasan.h> |
17211455 | 65 | #include <linux/context_tracking.h> |
48d07c04 | 66 | #include "../time/tick-internal.h" |
64db4cff | 67 | |
4102adab | 68 | #include "tree.h" |
29c00b4a | 69 | #include "rcu.h" |
9f77da9f | 70 | |
4102adab PM |
71 | #ifdef MODULE_PARAM_PREFIX |
72 | #undef MODULE_PARAM_PREFIX | |
73 | #endif | |
74 | #define MODULE_PARAM_PREFIX "rcutree." | |
75 | ||
64db4cff PM |
76 | /* Data structures. */ |
77 | ||
4c5273bf | 78 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, rcu_data) = { |
a5d1b0b6 | 79 | .gpwrap = true, |
8d346d43 | 80 | #ifdef CONFIG_RCU_NOCB_CPU |
213d56bf | 81 | .cblist.flags = SEGCBLIST_RCU_CORE, |
8d346d43 | 82 | #endif |
4c5273bf | 83 | }; |
c30fe541 | 84 | static struct rcu_state rcu_state = { |
358be2d3 | 85 | .level = { &rcu_state.node[0] }, |
358be2d3 PM |
86 | .gp_state = RCU_GP_IDLE, |
87 | .gp_seq = (0UL - 300UL) << RCU_SEQ_CTR_SHIFT, | |
88 | .barrier_mutex = __MUTEX_INITIALIZER(rcu_state.barrier_mutex), | |
80b3fd47 | 89 | .barrier_lock = __RAW_SPIN_LOCK_UNLOCKED(rcu_state.barrier_lock), |
358be2d3 PM |
90 | .name = RCU_NAME, |
91 | .abbr = RCU_ABBR, | |
92 | .exp_mutex = __MUTEX_INITIALIZER(rcu_state.exp_mutex), | |
93 | .exp_wake_mutex = __MUTEX_INITIALIZER(rcu_state.exp_wake_mutex), | |
82980b16 | 94 | .ofl_lock = __ARCH_SPIN_LOCK_UNLOCKED, |
358be2d3 | 95 | }; |
27f4d280 | 96 | |
a3dc2948 PM |
97 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
98 | static bool dump_tree; | |
99 | module_param(dump_tree, bool, 0444); | |
48d07c04 | 100 | /* By default, use RCU_SOFTIRQ instead of rcuc kthreads. */ |
8b9a0ecc SW |
101 | static bool use_softirq = !IS_ENABLED(CONFIG_PREEMPT_RT); |
102 | #ifndef CONFIG_PREEMPT_RT | |
48d07c04 | 103 | module_param(use_softirq, bool, 0444); |
8b9a0ecc | 104 | #endif |
7fa27001 PM |
105 | /* Control rcu_node-tree auto-balancing at boot time. */ |
106 | static bool rcu_fanout_exact; | |
107 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
108 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
109 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 110 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 111 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 | 112 | /* Number of rcu_nodes at specified level. */ |
e95d68d2 | 113 | int num_rcu_lvl[] = NUM_RCU_LVL_INIT; |
f885b7f2 PM |
114 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
115 | ||
b0d30417 | 116 | /* |
52d7e48b PM |
117 | * The rcu_scheduler_active variable is initialized to the value |
118 | * RCU_SCHEDULER_INACTIVE and transitions RCU_SCHEDULER_INIT just before the | |
119 | * first task is spawned. So when this variable is RCU_SCHEDULER_INACTIVE, | |
120 | * RCU can assume that there is but one task, allowing RCU to (for example) | |
0d95092c | 121 | * optimize synchronize_rcu() to a simple barrier(). When this variable |
52d7e48b PM |
122 | * is RCU_SCHEDULER_INIT, RCU must actually do all the hard work required |
123 | * to detect real grace periods. This variable is also used to suppress | |
124 | * boot-time false positives from lockdep-RCU error checking. Finally, it | |
125 | * transitions from RCU_SCHEDULER_INIT to RCU_SCHEDULER_RUNNING after RCU | |
126 | * is fully initialized, including all of its kthreads having been spawned. | |
b0d30417 | 127 | */ |
bbad9379 PM |
128 | int rcu_scheduler_active __read_mostly; |
129 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
130 | ||
b0d30417 PM |
131 | /* |
132 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
133 | * during the early_initcall() processing, which is after the scheduler | |
134 | * is capable of creating new tasks. So RCU processing (for example, | |
135 | * creating tasks for RCU priority boosting) must be delayed until after | |
136 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
137 | * currently delay invocation of any RCU callbacks until after this point. | |
138 | * | |
139 | * It might later prove better for people registering RCU callbacks during | |
140 | * early boot to take responsibility for these callbacks, but one step at | |
141 | * a time. | |
142 | */ | |
143 | static int rcu_scheduler_fully_active __read_mostly; | |
144 | ||
b50912d0 PM |
145 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
146 | unsigned long gps, unsigned long flags); | |
0aa04b05 PM |
147 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
148 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 149 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 | 150 | static void invoke_rcu_core(void); |
63d4c8c9 | 151 | static void rcu_report_exp_rdp(struct rcu_data *rdp); |
3549c2bc | 152 | static void sync_sched_exp_online_cleanup(int cpu); |
b2b00ddf | 153 | static void check_cb_ovld_locked(struct rcu_data *rdp, struct rcu_node *rnp); |
3820b513 | 154 | static bool rcu_rdp_is_offloaded(struct rcu_data *rdp); |
a26ac245 | 155 | |
8f489b4d URS |
156 | /* |
157 | * rcuc/rcub/rcuop kthread realtime priority. The "rcuop" | |
158 | * real-time priority(enabling/disabling) is controlled by | |
159 | * the extra CONFIG_RCU_NOCB_CPU_CB_BOOST configuration. | |
160 | */ | |
26730f55 | 161 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; |
3ffe3d1a | 162 | module_param(kthread_prio, int, 0444); |
a94844b2 | 163 | |
8d7dc928 | 164 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd | 165 | |
90040c9e PM |
166 | static int gp_preinit_delay; |
167 | module_param(gp_preinit_delay, int, 0444); | |
168 | static int gp_init_delay; | |
169 | module_param(gp_init_delay, int, 0444); | |
170 | static int gp_cleanup_delay; | |
171 | module_param(gp_cleanup_delay, int, 0444); | |
0f41c0dd | 172 | |
aa40c138 PM |
173 | // Add delay to rcu_read_unlock() for strict grace periods. |
174 | static int rcu_unlock_delay; | |
175 | #ifdef CONFIG_RCU_STRICT_GRACE_PERIOD | |
176 | module_param(rcu_unlock_delay, int, 0444); | |
177 | #endif | |
178 | ||
53c72b59 URS |
179 | /* |
180 | * This rcu parameter is runtime-read-only. It reflects | |
181 | * a minimum allowed number of objects which can be cached | |
182 | * per-CPU. Object size is equal to one page. This value | |
183 | * can be changed at boot time. | |
184 | */ | |
56292e86 | 185 | static int rcu_min_cached_objs = 5; |
53c72b59 URS |
186 | module_param(rcu_min_cached_objs, int, 0444); |
187 | ||
d0bfa8b3 ZQ |
188 | // A page shrinker can ask for pages to be freed to make them |
189 | // available for other parts of the system. This usually happens | |
190 | // under low memory conditions, and in that case we should also | |
191 | // defer page-cache filling for a short time period. | |
192 | // | |
193 | // The default value is 5 seconds, which is long enough to reduce | |
194 | // interference with the shrinker while it asks other systems to | |
195 | // drain their caches. | |
196 | static int rcu_delay_page_cache_fill_msec = 5000; | |
197 | module_param(rcu_delay_page_cache_fill_msec, int, 0444); | |
198 | ||
4cf439a2 | 199 | /* Retrieve RCU kthreads priority for rcutorture */ |
4babd855 JFG |
200 | int rcu_get_gp_kthreads_prio(void) |
201 | { | |
202 | return kthread_prio; | |
203 | } | |
204 | EXPORT_SYMBOL_GPL(rcu_get_gp_kthreads_prio); | |
205 | ||
eab128e8 PM |
206 | /* |
207 | * Number of grace periods between delays, normalized by the duration of | |
bfd090be | 208 | * the delay. The longer the delay, the more the grace periods between |
eab128e8 PM |
209 | * each delay. The reason for this normalization is that it means that, |
210 | * for non-zero delays, the overall slowdown of grace periods is constant | |
211 | * regardless of the duration of the delay. This arrangement balances | |
212 | * the need for long delays to increase some race probabilities with the | |
213 | * need for fast grace periods to increase other race probabilities. | |
214 | */ | |
277ffe1b | 215 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays for debugging. */ |
37745d28 | 216 | |
0aa04b05 PM |
217 | /* |
218 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
219 | * This will not be stable unless the rcu_node structure's ->lock is | |
220 | * held, but the bit corresponding to the current CPU will be stable | |
221 | * in most contexts. | |
222 | */ | |
c30fe541 | 223 | static unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) |
0aa04b05 | 224 | { |
7d0ae808 | 225 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
226 | } |
227 | ||
5ae0f1b5 PM |
228 | /* |
229 | * Is the CPU corresponding to the specified rcu_data structure online | |
230 | * from RCU's perspective? This perspective is given by that structure's | |
231 | * ->qsmaskinitnext field rather than by the global cpu_online_mask. | |
232 | */ | |
233 | static bool rcu_rdp_cpu_online(struct rcu_data *rdp) | |
234 | { | |
235 | return !!(rdp->grpmask & rcu_rnp_online_cpus(rdp->mynode)); | |
236 | } | |
237 | ||
fc2219d4 | 238 | /* |
7d0ae808 | 239 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
240 | * permit this function to be invoked without holding the root rcu_node |
241 | * structure's ->lock, but of course results can be subject to change. | |
242 | */ | |
de8e8730 | 243 | static int rcu_gp_in_progress(void) |
fc2219d4 | 244 | { |
de8e8730 | 245 | return rcu_seq_state(rcu_seq_current(&rcu_state.gp_seq)); |
b1f77b05 IM |
246 | } |
247 | ||
903ee83d PM |
248 | /* |
249 | * Return the number of callbacks queued on the specified CPU. | |
250 | * Handles both the nocbs and normal cases. | |
251 | */ | |
252 | static long rcu_get_n_cbs_cpu(int cpu) | |
253 | { | |
254 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); | |
255 | ||
c035280f | 256 | if (rcu_segcblist_is_enabled(&rdp->cblist)) |
903ee83d | 257 | return rcu_segcblist_n_cbs(&rdp->cblist); |
c035280f | 258 | return 0; |
903ee83d PM |
259 | } |
260 | ||
d28139c4 | 261 | void rcu_softirq_qs(void) |
b1f77b05 | 262 | { |
45975c7d | 263 | rcu_qs(); |
d28139c4 | 264 | rcu_preempt_deferred_qs(current); |
cf868c2a | 265 | rcu_tasks_qs(current, false); |
b1f77b05 | 266 | } |
64db4cff | 267 | |
2625d469 PM |
268 | /* |
269 | * Reset the current CPU's ->dynticks counter to indicate that the | |
270 | * newly onlined CPU is no longer in an extended quiescent state. | |
271 | * This will either leave the counter unchanged, or increment it | |
272 | * to the next non-quiescent value. | |
273 | * | |
274 | * The non-atomic test/increment sequence works because the upper bits | |
275 | * of the ->dynticks counter are manipulated only by the corresponding CPU, | |
276 | * or when the corresponding CPU is offline. | |
277 | */ | |
278 | static void rcu_dynticks_eqs_online(void) | |
279 | { | |
17147677 | 280 | if (ct_dynticks() & RCU_DYNTICKS_IDX) |
2625d469 | 281 | return; |
17147677 | 282 | ct_state_inc(RCU_DYNTICKS_IDX); |
02a5c550 PM |
283 | } |
284 | ||
8b2f63ab PM |
285 | /* |
286 | * Snapshot the ->dynticks counter with full ordering so as to allow | |
287 | * stable comparison of this counter with past and future snapshots. | |
288 | */ | |
62e2412d | 289 | static int rcu_dynticks_snap(int cpu) |
8b2f63ab | 290 | { |
2be57f73 | 291 | smp_mb(); // Fundamental RCU ordering guarantee. |
62e2412d | 292 | return ct_dynticks_cpu_acquire(cpu); |
8b2f63ab PM |
293 | } |
294 | ||
02a5c550 PM |
295 | /* |
296 | * Return true if the snapshot returned from rcu_dynticks_snap() | |
297 | * indicates that RCU is in an extended quiescent state. | |
298 | */ | |
299 | static bool rcu_dynticks_in_eqs(int snap) | |
300 | { | |
17147677 | 301 | return !(snap & RCU_DYNTICKS_IDX); |
02a5c550 PM |
302 | } |
303 | ||
3fcd6a23 PM |
304 | /* Return true if the specified CPU is currently idle from an RCU viewpoint. */ |
305 | bool rcu_is_idle_cpu(int cpu) | |
306 | { | |
62e2412d | 307 | return rcu_dynticks_in_eqs(rcu_dynticks_snap(cpu)); |
3fcd6a23 PM |
308 | } |
309 | ||
02a5c550 | 310 | /* |
dc5a4f29 | 311 | * Return true if the CPU corresponding to the specified rcu_data |
02a5c550 PM |
312 | * structure has spent some time in an extended quiescent state since |
313 | * rcu_dynticks_snap() returned the specified snapshot. | |
314 | */ | |
dc5a4f29 | 315 | static bool rcu_dynticks_in_eqs_since(struct rcu_data *rdp, int snap) |
02a5c550 | 316 | { |
62e2412d | 317 | return snap != rcu_dynticks_snap(rdp->cpu); |
02a5c550 PM |
318 | } |
319 | ||
7d0c9c50 PM |
320 | /* |
321 | * Return true if the referenced integer is zero while the specified | |
322 | * CPU remains within a single extended quiescent state. | |
323 | */ | |
324 | bool rcu_dynticks_zero_in_eqs(int cpu, int *vp) | |
325 | { | |
7d0c9c50 PM |
326 | int snap; |
327 | ||
328 | // If not quiescent, force back to earlier extended quiescent state. | |
17147677 | 329 | snap = ct_dynticks_cpu(cpu) & ~RCU_DYNTICKS_IDX; |
7d0c9c50 PM |
330 | smp_rmb(); // Order ->dynticks and *vp reads. |
331 | if (READ_ONCE(*vp)) | |
332 | return false; // Non-zero, so report failure; | |
333 | smp_rmb(); // Order *vp read and ->dynticks re-read. | |
334 | ||
335 | // If still in the same extended quiescent state, we are good! | |
62e2412d | 336 | return snap == ct_dynticks_cpu(cpu); |
6563de9d | 337 | } |
5cd37193 | 338 | |
4a81e832 PM |
339 | /* |
340 | * Let the RCU core know that this CPU has gone through the scheduler, | |
341 | * which is a quiescent state. This is called when the need for a | |
342 | * quiescent state is urgent, so we burn an atomic operation and full | |
343 | * memory barriers to let the RCU core know about it, regardless of what | |
344 | * this CPU might (or might not) do in the near future. | |
345 | * | |
0f9be8ca | 346 | * We inform the RCU core by emulating a zero-duration dyntick-idle period. |
46a5d164 | 347 | * |
3b57a399 | 348 | * The caller must have disabled interrupts and must not be idle. |
4a81e832 | 349 | */ |
4230e2de | 350 | notrace void rcu_momentary_dyntick_idle(void) |
4a81e832 | 351 | { |
2be57f73 | 352 | int seq; |
3b57a399 | 353 | |
2dba13f0 | 354 | raw_cpu_write(rcu_data.rcu_need_heavy_qs, false); |
17147677 | 355 | seq = ct_state_inc(2 * RCU_DYNTICKS_IDX); |
3b57a399 | 356 | /* It is illegal to call this from idle state. */ |
17147677 | 357 | WARN_ON_ONCE(!(seq & RCU_DYNTICKS_IDX)); |
3e310098 | 358 | rcu_preempt_deferred_qs(current); |
4a81e832 | 359 | } |
79ba7ff5 | 360 | EXPORT_SYMBOL_GPL(rcu_momentary_dyntick_idle); |
4a81e832 | 361 | |
45975c7d | 362 | /** |
806f04e9 | 363 | * rcu_is_cpu_rrupt_from_idle - see if 'interrupted' from idle |
bb73c52b | 364 | * |
eddded80 | 365 | * If the current CPU is idle and running at a first-level (not nested) |
806f04e9 PZ |
366 | * interrupt, or directly, from idle, return true. |
367 | * | |
368 | * The caller must have at least disabled IRQs. | |
5cd37193 | 369 | */ |
45975c7d | 370 | static int rcu_is_cpu_rrupt_from_idle(void) |
5cd37193 | 371 | { |
806f04e9 PZ |
372 | long nesting; |
373 | ||
374 | /* | |
375 | * Usually called from the tick; but also used from smp_function_call() | |
376 | * for expedited grace periods. This latter can result in running from | |
377 | * the idle task, instead of an actual IPI. | |
378 | */ | |
379 | lockdep_assert_irqs_disabled(); | |
eddded80 JFG |
380 | |
381 | /* Check for counter underflows */ | |
904e600e | 382 | RCU_LOCKDEP_WARN(ct_dynticks_nesting() < 0, |
eddded80 | 383 | "RCU dynticks_nesting counter underflow!"); |
95e04f48 | 384 | RCU_LOCKDEP_WARN(ct_dynticks_nmi_nesting() <= 0, |
eddded80 JFG |
385 | "RCU dynticks_nmi_nesting counter underflow/zero!"); |
386 | ||
387 | /* Are we at first interrupt nesting level? */ | |
95e04f48 | 388 | nesting = ct_dynticks_nmi_nesting(); |
806f04e9 | 389 | if (nesting > 1) |
eddded80 JFG |
390 | return false; |
391 | ||
806f04e9 PZ |
392 | /* |
393 | * If we're not in an interrupt, we must be in the idle task! | |
394 | */ | |
395 | WARN_ON_ONCE(!nesting && !is_idle_task(current)); | |
396 | ||
eddded80 | 397 | /* Does CPU appear to be idle from an RCU standpoint? */ |
904e600e | 398 | return ct_dynticks_nesting() == 0; |
5cd37193 | 399 | } |
5cd37193 | 400 | |
29fc5f93 PM |
401 | #define DEFAULT_RCU_BLIMIT (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) ? 1000 : 10) |
402 | // Maximum callbacks per rcu_do_batch ... | |
403 | #define DEFAULT_MAX_RCU_BLIMIT 10000 // ... even during callback flood. | |
17c7798b | 404 | static long blimit = DEFAULT_RCU_BLIMIT; |
29fc5f93 | 405 | #define DEFAULT_RCU_QHIMARK 10000 // If this many pending, ignore blimit. |
17c7798b | 406 | static long qhimark = DEFAULT_RCU_QHIMARK; |
29fc5f93 | 407 | #define DEFAULT_RCU_QLOMARK 100 // Once only this many pending, use blimit. |
17c7798b | 408 | static long qlowmark = DEFAULT_RCU_QLOMARK; |
b2b00ddf PM |
409 | #define DEFAULT_RCU_QOVLD_MULT 2 |
410 | #define DEFAULT_RCU_QOVLD (DEFAULT_RCU_QOVLD_MULT * DEFAULT_RCU_QHIMARK) | |
29fc5f93 PM |
411 | static long qovld = DEFAULT_RCU_QOVLD; // If this many pending, hammer QS. |
412 | static long qovld_calc = -1; // No pre-initialization lock acquisitions! | |
64db4cff | 413 | |
878d7439 ED |
414 | module_param(blimit, long, 0444); |
415 | module_param(qhimark, long, 0444); | |
416 | module_param(qlowmark, long, 0444); | |
b2b00ddf | 417 | module_param(qovld, long, 0444); |
3d76c082 | 418 | |
aecd34b9 | 419 | static ulong jiffies_till_first_fqs = IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) ? 0 : ULONG_MAX; |
026ad283 | 420 | static ulong jiffies_till_next_fqs = ULONG_MAX; |
8c7c4829 | 421 | static bool rcu_kick_kthreads; |
cfcdef5e ED |
422 | static int rcu_divisor = 7; |
423 | module_param(rcu_divisor, int, 0644); | |
424 | ||
425 | /* Force an exit from rcu_do_batch() after 3 milliseconds. */ | |
426 | static long rcu_resched_ns = 3 * NSEC_PER_MSEC; | |
427 | module_param(rcu_resched_ns, long, 0644); | |
d40011f6 | 428 | |
c06aed0e PM |
429 | /* |
430 | * How long the grace period must be before we start recruiting | |
431 | * quiescent-state help from rcu_note_context_switch(). | |
432 | */ | |
433 | static ulong jiffies_till_sched_qs = ULONG_MAX; | |
434 | module_param(jiffies_till_sched_qs, ulong, 0444); | |
85f2b60c | 435 | static ulong jiffies_to_sched_qs; /* See adjust_jiffies_till_sched_qs(). */ |
c06aed0e PM |
436 | module_param(jiffies_to_sched_qs, ulong, 0444); /* Display only! */ |
437 | ||
438 | /* | |
439 | * Make sure that we give the grace-period kthread time to detect any | |
440 | * idle CPUs before taking active measures to force quiescent states. | |
441 | * However, don't go below 100 milliseconds, adjusted upwards for really | |
442 | * large systems. | |
443 | */ | |
444 | static void adjust_jiffies_till_sched_qs(void) | |
445 | { | |
446 | unsigned long j; | |
447 | ||
448 | /* If jiffies_till_sched_qs was specified, respect the request. */ | |
449 | if (jiffies_till_sched_qs != ULONG_MAX) { | |
450 | WRITE_ONCE(jiffies_to_sched_qs, jiffies_till_sched_qs); | |
451 | return; | |
452 | } | |
85f2b60c | 453 | /* Otherwise, set to third fqs scan, but bound below on large system. */ |
c06aed0e PM |
454 | j = READ_ONCE(jiffies_till_first_fqs) + |
455 | 2 * READ_ONCE(jiffies_till_next_fqs); | |
456 | if (j < HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV) | |
457 | j = HZ / 10 + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
458 | pr_info("RCU calculated value of scheduler-enlistment delay is %ld jiffies.\n", j); | |
459 | WRITE_ONCE(jiffies_to_sched_qs, j); | |
460 | } | |
461 | ||
67abb96c BP |
462 | static int param_set_first_fqs_jiffies(const char *val, const struct kernel_param *kp) |
463 | { | |
464 | ulong j; | |
465 | int ret = kstrtoul(val, 0, &j); | |
466 | ||
c06aed0e | 467 | if (!ret) { |
67abb96c | 468 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : j); |
c06aed0e PM |
469 | adjust_jiffies_till_sched_qs(); |
470 | } | |
67abb96c BP |
471 | return ret; |
472 | } | |
473 | ||
474 | static int param_set_next_fqs_jiffies(const char *val, const struct kernel_param *kp) | |
475 | { | |
476 | ulong j; | |
477 | int ret = kstrtoul(val, 0, &j); | |
478 | ||
c06aed0e | 479 | if (!ret) { |
67abb96c | 480 | WRITE_ONCE(*(ulong *)kp->arg, (j > HZ) ? HZ : (j ?: 1)); |
c06aed0e PM |
481 | adjust_jiffies_till_sched_qs(); |
482 | } | |
67abb96c BP |
483 | return ret; |
484 | } | |
485 | ||
7c47ee5a | 486 | static const struct kernel_param_ops first_fqs_jiffies_ops = { |
67abb96c BP |
487 | .set = param_set_first_fqs_jiffies, |
488 | .get = param_get_ulong, | |
489 | }; | |
490 | ||
7c47ee5a | 491 | static const struct kernel_param_ops next_fqs_jiffies_ops = { |
67abb96c BP |
492 | .set = param_set_next_fqs_jiffies, |
493 | .get = param_get_ulong, | |
494 | }; | |
495 | ||
496 | module_param_cb(jiffies_till_first_fqs, &first_fqs_jiffies_ops, &jiffies_till_first_fqs, 0644); | |
497 | module_param_cb(jiffies_till_next_fqs, &next_fqs_jiffies_ops, &jiffies_till_next_fqs, 0644); | |
8c7c4829 | 498 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 499 | |
8ff0b907 | 500 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)); |
dd7dafd1 | 501 | static int rcu_pending(int user); |
64db4cff PM |
502 | |
503 | /* | |
17ef2fe9 | 504 | * Return the number of RCU GPs completed thus far for debug & stats. |
64db4cff | 505 | */ |
17ef2fe9 | 506 | unsigned long rcu_get_gp_seq(void) |
917963d0 | 507 | { |
16fc9c60 | 508 | return READ_ONCE(rcu_state.gp_seq); |
917963d0 | 509 | } |
17ef2fe9 | 510 | EXPORT_SYMBOL_GPL(rcu_get_gp_seq); |
917963d0 | 511 | |
291783b8 PM |
512 | /* |
513 | * Return the number of RCU expedited batches completed thus far for | |
514 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
515 | * numbers mean idle. The value returned will thus be roughly double | |
516 | * the cumulative batches since boot. | |
517 | */ | |
518 | unsigned long rcu_exp_batches_completed(void) | |
519 | { | |
16fc9c60 | 520 | return rcu_state.expedited_sequence; |
291783b8 PM |
521 | } |
522 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
523 | ||
fd897573 PM |
524 | /* |
525 | * Return the root node of the rcu_state structure. | |
526 | */ | |
527 | static struct rcu_node *rcu_get_root(void) | |
528 | { | |
529 | return &rcu_state.node[0]; | |
530 | } | |
531 | ||
ad0dc7f9 PM |
532 | /* |
533 | * Send along grace-period-related data for rcutorture diagnostics. | |
534 | */ | |
535 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
aebc8264 | 536 | unsigned long *gp_seq) |
ad0dc7f9 | 537 | { |
ad0dc7f9 PM |
538 | switch (test_type) { |
539 | case RCU_FLAVOR: | |
f7dd7d44 PM |
540 | *flags = READ_ONCE(rcu_state.gp_flags); |
541 | *gp_seq = rcu_seq_current(&rcu_state.gp_seq); | |
ad0dc7f9 PM |
542 | break; |
543 | default: | |
544 | break; | |
545 | } | |
ad0dc7f9 PM |
546 | } |
547 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
548 | ||
17211455 | 549 | #if defined(CONFIG_NO_HZ_FULL) && (!defined(CONFIG_GENERIC_ENTRY) || !defined(CONFIG_KVM_XFER_TO_GUEST_WORK)) |
f8bb5cae FW |
550 | /* |
551 | * An empty function that will trigger a reschedule on | |
4ae7dc97 | 552 | * IRQ tail once IRQs get re-enabled on userspace/guest resume. |
f8bb5cae FW |
553 | */ |
554 | static void late_wakeup_func(struct irq_work *work) | |
555 | { | |
556 | } | |
557 | ||
558 | static DEFINE_PER_CPU(struct irq_work, late_wakeup_work) = | |
559 | IRQ_WORK_INIT(late_wakeup_func); | |
560 | ||
4ae7dc97 FW |
561 | /* |
562 | * If either: | |
563 | * | |
564 | * 1) the task is about to enter in guest mode and $ARCH doesn't support KVM generic work | |
565 | * 2) the task is about to enter in user mode and $ARCH doesn't support generic entry. | |
566 | * | |
567 | * In these cases the late RCU wake ups aren't supported in the resched loops and our | |
568 | * last resort is to fire a local irq_work that will trigger a reschedule once IRQs | |
569 | * get re-enabled again. | |
570 | */ | |
56450649 | 571 | noinstr void rcu_irq_work_resched(void) |
4ae7dc97 FW |
572 | { |
573 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); | |
574 | ||
575 | if (IS_ENABLED(CONFIG_GENERIC_ENTRY) && !(current->flags & PF_VCPU)) | |
576 | return; | |
577 | ||
578 | if (IS_ENABLED(CONFIG_KVM_XFER_TO_GUEST_WORK) && (current->flags & PF_VCPU)) | |
579 | return; | |
580 | ||
581 | instrumentation_begin(); | |
582 | if (do_nocb_deferred_wakeup(rdp) && need_resched()) { | |
583 | irq_work_queue(this_cpu_ptr(&late_wakeup_work)); | |
584 | } | |
585 | instrumentation_end(); | |
586 | } | |
17211455 | 587 | #endif /* #if defined(CONFIG_NO_HZ_FULL) && (!defined(CONFIG_GENERIC_ENTRY) || !defined(CONFIG_KVM_XFER_TO_GUEST_WORK)) */ |
7c9906ca | 588 | |
07325d4a TG |
589 | #ifdef CONFIG_PROVE_RCU |
590 | /** | |
591 | * rcu_irq_exit_check_preempt - Validate that scheduling is possible | |
592 | */ | |
593 | void rcu_irq_exit_check_preempt(void) | |
594 | { | |
595 | lockdep_assert_irqs_disabled(); | |
596 | ||
904e600e | 597 | RCU_LOCKDEP_WARN(ct_dynticks_nesting() <= 0, |
07325d4a | 598 | "RCU dynticks_nesting counter underflow/zero!"); |
95e04f48 | 599 | RCU_LOCKDEP_WARN(ct_dynticks_nmi_nesting() != |
07325d4a TG |
600 | DYNTICK_IRQ_NONIDLE, |
601 | "Bad RCU dynticks_nmi_nesting counter\n"); | |
602 | RCU_LOCKDEP_WARN(rcu_dynticks_curr_cpu_in_eqs(), | |
603 | "RCU in extended quiescent state!"); | |
604 | } | |
605 | #endif /* #ifdef CONFIG_PROVE_RCU */ | |
606 | ||
d1ec4c34 | 607 | #ifdef CONFIG_NO_HZ_FULL |
aaf2bc50 PM |
608 | /** |
609 | * __rcu_irq_enter_check_tick - Enable scheduler tick on CPU if RCU needs it. | |
610 | * | |
611 | * The scheduler tick is not normally enabled when CPUs enter the kernel | |
612 | * from nohz_full userspace execution. After all, nohz_full userspace | |
613 | * execution is an RCU quiescent state and the time executing in the kernel | |
614 | * is quite short. Except of course when it isn't. And it is not hard to | |
615 | * cause a large system to spend tens of seconds or even minutes looping | |
616 | * in the kernel, which can cause a number of problems, include RCU CPU | |
617 | * stall warnings. | |
618 | * | |
619 | * Therefore, if a nohz_full CPU fails to report a quiescent state | |
620 | * in a timely manner, the RCU grace-period kthread sets that CPU's | |
621 | * ->rcu_urgent_qs flag with the expectation that the next interrupt or | |
622 | * exception will invoke this function, which will turn on the scheduler | |
623 | * tick, which will enable RCU to detect that CPU's quiescent states, | |
624 | * for example, due to cond_resched() calls in CONFIG_PREEMPT=n kernels. | |
625 | * The tick will be disabled once a quiescent state is reported for | |
626 | * this CPU. | |
627 | * | |
628 | * Of course, in carefully tuned systems, there might never be an | |
629 | * interrupt or exception. In that case, the RCU grace-period kthread | |
630 | * will eventually cause one to happen. However, in less carefully | |
631 | * controlled environments, this function allows RCU to get what it | |
632 | * needs without creating otherwise useless interruptions. | |
633 | */ | |
634 | void __rcu_irq_enter_check_tick(void) | |
635 | { | |
636 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); | |
637 | ||
6dbce04d PZ |
638 | // If we're here from NMI there's nothing to do. |
639 | if (in_nmi()) | |
aaf2bc50 PM |
640 | return; |
641 | ||
642 | RCU_LOCKDEP_WARN(rcu_dynticks_curr_cpu_in_eqs(), | |
643 | "Illegal rcu_irq_enter_check_tick() from extended quiescent state"); | |
644 | ||
645 | if (!tick_nohz_full_cpu(rdp->cpu) || | |
646 | !READ_ONCE(rdp->rcu_urgent_qs) || | |
647 | READ_ONCE(rdp->rcu_forced_tick)) { | |
648 | // RCU doesn't need nohz_full help from this CPU, or it is | |
649 | // already getting that help. | |
650 | return; | |
651 | } | |
652 | ||
653 | // We get here only when not in an extended quiescent state and | |
654 | // from interrupts (as opposed to NMIs). Therefore, (1) RCU is | |
655 | // already watching and (2) The fact that we are in an interrupt | |
656 | // handler and that the rcu_node lock is an irq-disabled lock | |
657 | // prevents self-deadlock. So we can safely recheck under the lock. | |
658 | // Note that the nohz_full state currently cannot change. | |
659 | raw_spin_lock_rcu_node(rdp->mynode); | |
660 | if (rdp->rcu_urgent_qs && !rdp->rcu_forced_tick) { | |
661 | // A nohz_full CPU is in the kernel and RCU needs a | |
662 | // quiescent state. Turn on the tick! | |
663 | WRITE_ONCE(rdp->rcu_forced_tick, true); | |
664 | tick_dep_set_cpu(rdp->cpu, TICK_DEP_BIT_RCU); | |
665 | } | |
666 | raw_spin_unlock_rcu_node(rdp->mynode); | |
667 | } | |
d1ec4c34 | 668 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 669 | |
bc849e91 PM |
670 | /* |
671 | * Check to see if any future non-offloaded RCU-related work will need | |
672 | * to be done by the current CPU, even if none need be done immediately, | |
673 | * returning 1 if so. This function is part of the RCU implementation; | |
674 | * it is -not- an exported member of the RCU API. This is used by | |
675 | * the idle-entry code to figure out whether it is safe to disable the | |
676 | * scheduler-clock interrupt. | |
677 | * | |
678 | * Just check whether or not this CPU has non-offloaded RCU callbacks | |
679 | * queued. | |
680 | */ | |
29845399 | 681 | int rcu_needs_cpu(void) |
bc849e91 | 682 | { |
bc849e91 PM |
683 | return !rcu_segcblist_empty(&this_cpu_ptr(&rcu_data)->cblist) && |
684 | !rcu_rdp_is_offloaded(this_cpu_ptr(&rcu_data)); | |
685 | } | |
686 | ||
66e4c33b | 687 | /* |
516e5ae0 JFG |
688 | * If any sort of urgency was applied to the current CPU (for example, |
689 | * the scheduler-clock interrupt was enabled on a nohz_full CPU) in order | |
690 | * to get to a quiescent state, disable it. | |
66e4c33b | 691 | */ |
516e5ae0 | 692 | static void rcu_disable_urgency_upon_qs(struct rcu_data *rdp) |
66e4c33b | 693 | { |
5b14557b | 694 | raw_lockdep_assert_held_rcu_node(rdp->mynode); |
516e5ae0 JFG |
695 | WRITE_ONCE(rdp->rcu_urgent_qs, false); |
696 | WRITE_ONCE(rdp->rcu_need_heavy_qs, false); | |
66e4c33b PM |
697 | if (tick_nohz_full_cpu(rdp->cpu) && rdp->rcu_forced_tick) { |
698 | tick_dep_clear_cpu(rdp->cpu, TICK_DEP_BIT_RCU); | |
2a2ae872 | 699 | WRITE_ONCE(rdp->rcu_forced_tick, false); |
66e4c33b PM |
700 | } |
701 | } | |
702 | ||
5c173eb8 | 703 | /** |
2320bda2 | 704 | * rcu_is_watching - see if RCU thinks that the current CPU is not idle |
64db4cff | 705 | * |
791875d1 PM |
706 | * Return true if RCU is watching the running CPU, which means that this |
707 | * CPU can safely enter RCU read-side critical sections. In other words, | |
2320bda2 ZZ |
708 | * if the current CPU is not in its idle loop or is in an interrupt or |
709 | * NMI handler, return true. | |
d2098b44 PZ |
710 | * |
711 | * Make notrace because it can be called by the internal functions of | |
712 | * ftrace, and making this notrace removes unnecessary recursion calls. | |
64db4cff | 713 | */ |
d2098b44 | 714 | notrace bool rcu_is_watching(void) |
64db4cff | 715 | { |
f534ed1f | 716 | bool ret; |
34240697 | 717 | |
46f00d18 | 718 | preempt_disable_notrace(); |
791875d1 | 719 | ret = !rcu_dynticks_curr_cpu_in_eqs(); |
46f00d18 | 720 | preempt_enable_notrace(); |
34240697 | 721 | return ret; |
64db4cff | 722 | } |
5c173eb8 | 723 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 724 | |
bcbfdd01 PM |
725 | /* |
726 | * If a holdout task is actually running, request an urgent quiescent | |
727 | * state from its CPU. This is unsynchronized, so migrations can cause | |
728 | * the request to go to the wrong CPU. Which is OK, all that will happen | |
729 | * is that the CPU's next context switch will be a bit slower and next | |
730 | * time around this task will generate another request. | |
731 | */ | |
732 | void rcu_request_urgent_qs_task(struct task_struct *t) | |
733 | { | |
734 | int cpu; | |
735 | ||
736 | barrier(); | |
737 | cpu = task_cpu(t); | |
738 | if (!task_curr(t)) | |
739 | return; /* This task is not running on that CPU. */ | |
2dba13f0 | 740 | smp_store_release(per_cpu_ptr(&rcu_data.rcu_urgent_qs, cpu), true); |
bcbfdd01 PM |
741 | } |
742 | ||
62fde6ed | 743 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
744 | |
745 | /* | |
5554788e | 746 | * Is the current CPU online as far as RCU is concerned? |
2036d94a | 747 | * |
5554788e PM |
748 | * Disable preemption to avoid false positives that could otherwise |
749 | * happen due to the current CPU number being sampled, this task being | |
750 | * preempted, its old CPU being taken offline, resuming on some other CPU, | |
49918a54 | 751 | * then determining that its old CPU is now offline. |
c0d6d01b | 752 | * |
5554788e PM |
753 | * Disable checking if in an NMI handler because we cannot safely |
754 | * report errors from NMI handlers anyway. In addition, it is OK to use | |
755 | * RCU on an offline processor during initial boot, hence the check for | |
756 | * rcu_scheduler_fully_active. | |
c0d6d01b PM |
757 | */ |
758 | bool rcu_lockdep_current_cpu_online(void) | |
759 | { | |
2036d94a | 760 | struct rcu_data *rdp; |
b97d23c5 | 761 | bool ret = false; |
c0d6d01b | 762 | |
5554788e | 763 | if (in_nmi() || !rcu_scheduler_fully_active) |
f6f7ee9a | 764 | return true; |
ff5c4f5c | 765 | preempt_disable_notrace(); |
b97d23c5 | 766 | rdp = this_cpu_ptr(&rcu_data); |
82980b16 DW |
767 | /* |
768 | * Strictly, we care here about the case where the current CPU is | |
769 | * in rcu_cpu_starting() and thus has an excuse for rdp->grpmask | |
770 | * not being up to date. So arch_spin_is_locked() might have a | |
771 | * false positive if it's held by some *other* CPU, but that's | |
772 | * OK because that just means a false *negative* on the warning. | |
773 | */ | |
5ae0f1b5 | 774 | if (rcu_rdp_cpu_online(rdp) || arch_spin_is_locked(&rcu_state.ofl_lock)) |
b97d23c5 | 775 | ret = true; |
ff5c4f5c | 776 | preempt_enable_notrace(); |
b97d23c5 | 777 | return ret; |
c0d6d01b PM |
778 | } |
779 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
780 | ||
62fde6ed | 781 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 782 | |
9b9500da | 783 | /* |
277ffe1b | 784 | * When trying to report a quiescent state on behalf of some other CPU, |
9b9500da | 785 | * it is our responsibility to check for and handle potential overflow |
a66ae8ae | 786 | * of the rcu_node ->gp_seq counter with respect to the rcu_data counters. |
9b9500da PM |
787 | * After all, the CPU might be in deep idle state, and thus executing no |
788 | * code whatsoever. | |
789 | */ | |
790 | static void rcu_gpnum_ovf(struct rcu_node *rnp, struct rcu_data *rdp) | |
791 | { | |
a32e01ee | 792 | raw_lockdep_assert_held_rcu_node(rnp); |
a66ae8ae PM |
793 | if (ULONG_CMP_LT(rcu_seq_current(&rdp->gp_seq) + ULONG_MAX / 4, |
794 | rnp->gp_seq)) | |
9b9500da | 795 | WRITE_ONCE(rdp->gpwrap, true); |
8aa670cd PM |
796 | if (ULONG_CMP_LT(rdp->rcu_iw_gp_seq + ULONG_MAX / 4, rnp->gp_seq)) |
797 | rdp->rcu_iw_gp_seq = rnp->gp_seq + ULONG_MAX / 4; | |
9b9500da PM |
798 | } |
799 | ||
64db4cff PM |
800 | /* |
801 | * Snapshot the specified CPU's dynticks counter so that we can later | |
802 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 803 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 804 | */ |
fe5ac724 | 805 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
64db4cff | 806 | { |
62e2412d | 807 | rdp->dynticks_snap = rcu_dynticks_snap(rdp->cpu); |
02a5c550 | 808 | if (rcu_dynticks_in_eqs(rdp->dynticks_snap)) { |
88d1bead | 809 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 810 | rcu_gpnum_ovf(rdp->mynode, rdp); |
23a9bacd | 811 | return 1; |
7941dbde | 812 | } |
23a9bacd | 813 | return 0; |
64db4cff PM |
814 | } |
815 | ||
816 | /* | |
817 | * Return true if the specified CPU has passed through a quiescent | |
818 | * state by virtue of being in or having passed through an dynticks | |
819 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 820 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 821 | */ |
fe5ac724 | 822 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
64db4cff | 823 | { |
3a19b46a | 824 | unsigned long jtsq; |
9b9500da | 825 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
826 | |
827 | /* | |
828 | * If the CPU passed through or entered a dynticks idle phase with | |
829 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
830 | * already acknowledged the request to pass through a quiescent | |
831 | * state. Either way, that CPU cannot possibly be in an RCU | |
832 | * read-side critical section that started before the beginning | |
833 | * of the current RCU grace period. | |
834 | */ | |
dc5a4f29 | 835 | if (rcu_dynticks_in_eqs_since(rdp, rdp->dynticks_snap)) { |
88d1bead | 836 | trace_rcu_fqs(rcu_state.name, rdp->gp_seq, rdp->cpu, TPS("dti")); |
9b9500da | 837 | rcu_gpnum_ovf(rnp, rdp); |
3a19b46a PM |
838 | return 1; |
839 | } | |
840 | ||
666ca290 JFG |
841 | /* |
842 | * Complain if a CPU that is considered to be offline from RCU's | |
843 | * perspective has not yet reported a quiescent state. After all, | |
844 | * the offline CPU should have reported a quiescent state during | |
845 | * the CPU-offline process, or, failing that, by rcu_gp_init() | |
846 | * if it ran concurrently with either the CPU going offline or the | |
847 | * last task on a leaf rcu_node structure exiting its RCU read-side | |
848 | * critical section while all CPUs corresponding to that structure | |
849 | * are offline. This added warning detects bugs in any of these | |
850 | * code paths. | |
851 | * | |
852 | * The rcu_node structure's ->lock is held here, which excludes | |
853 | * the relevant portions the CPU-hotplug code, the grace-period | |
854 | * initialization code, and the rcu_read_unlock() code paths. | |
855 | * | |
856 | * For more detail, please refer to the "Hotplug CPU" section | |
857 | * of RCU's Requirements documentation. | |
858 | */ | |
5ae0f1b5 | 859 | if (WARN_ON_ONCE(!rcu_rdp_cpu_online(rdp))) { |
f2e2df59 PM |
860 | struct rcu_node *rnp1; |
861 | ||
f2e2df59 PM |
862 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", |
863 | __func__, rnp->grplo, rnp->grphi, rnp->level, | |
864 | (long)rnp->gp_seq, (long)rnp->completedqs); | |
865 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
866 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx ->rcu_gp_init_mask %#lx\n", | |
867 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext, rnp1->rcu_gp_init_mask); | |
f2e2df59 | 868 | pr_info("%s %d: %c online: %ld(%d) offline: %ld(%d)\n", |
5ae0f1b5 | 869 | __func__, rdp->cpu, ".o"[rcu_rdp_cpu_online(rdp)], |
f2e2df59 PM |
870 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, |
871 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
872 | return 1; /* Break things loose after complaining. */ | |
873 | } | |
874 | ||
65d798f0 | 875 | /* |
4a81e832 | 876 | * A CPU running for an extended time within the kernel can |
c06aed0e PM |
877 | * delay RCU grace periods: (1) At age jiffies_to_sched_qs, |
878 | * set .rcu_urgent_qs, (2) At age 2*jiffies_to_sched_qs, set | |
7e28c5af PM |
879 | * both .rcu_need_heavy_qs and .rcu_urgent_qs. Note that the |
880 | * unsynchronized assignments to the per-CPU rcu_need_heavy_qs | |
881 | * variable are safe because the assignments are repeated if this | |
882 | * CPU failed to pass through a quiescent state. This code | |
c06aed0e | 883 | * also checks .jiffies_resched in case jiffies_to_sched_qs |
7e28c5af | 884 | * is set way high. |
6193c76a | 885 | */ |
c06aed0e | 886 | jtsq = READ_ONCE(jiffies_to_sched_qs); |
88ee23ef | 887 | if (!READ_ONCE(rdp->rcu_need_heavy_qs) && |
7e28c5af | 888 | (time_after(jiffies, rcu_state.gp_start + jtsq * 2) || |
b2b00ddf PM |
889 | time_after(jiffies, rcu_state.jiffies_resched) || |
890 | rcu_state.cbovld)) { | |
88ee23ef | 891 | WRITE_ONCE(rdp->rcu_need_heavy_qs, true); |
9226b10d | 892 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ |
9424b867 | 893 | smp_store_release(&rdp->rcu_urgent_qs, true); |
7e28c5af | 894 | } else if (time_after(jiffies, rcu_state.gp_start + jtsq)) { |
9424b867 | 895 | WRITE_ONCE(rdp->rcu_urgent_qs, true); |
6193c76a PM |
896 | } |
897 | ||
28053bc7 | 898 | /* |
c98cac60 | 899 | * NO_HZ_FULL CPUs can run in-kernel without rcu_sched_clock_irq! |
d3052109 PM |
900 | * The above code handles this, but only for straight cond_resched(). |
901 | * And some in-kernel loops check need_resched() before calling | |
902 | * cond_resched(), which defeats the above code for CPUs that are | |
903 | * running in-kernel with scheduling-clock interrupts disabled. | |
904 | * So hit them over the head with the resched_cpu() hammer! | |
28053bc7 | 905 | */ |
d3052109 | 906 | if (tick_nohz_full_cpu(rdp->cpu) && |
b2b00ddf PM |
907 | (time_after(jiffies, READ_ONCE(rdp->last_fqs_resched) + jtsq * 3) || |
908 | rcu_state.cbovld)) { | |
9424b867 | 909 | WRITE_ONCE(rdp->rcu_urgent_qs, true); |
28053bc7 | 910 | resched_cpu(rdp->cpu); |
d3052109 PM |
911 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); |
912 | } | |
913 | ||
914 | /* | |
915 | * If more than halfway to RCU CPU stall-warning time, invoke | |
916 | * resched_cpu() more frequently to try to loosen things up a bit. | |
917 | * Also check to see if the CPU is getting hammered with interrupts, | |
918 | * but only once per grace period, just to keep the IPIs down to | |
919 | * a dull roar. | |
920 | */ | |
921 | if (time_after(jiffies, rcu_state.jiffies_resched)) { | |
922 | if (time_after(jiffies, | |
923 | READ_ONCE(rdp->last_fqs_resched) + jtsq)) { | |
924 | resched_cpu(rdp->cpu); | |
925 | WRITE_ONCE(rdp->last_fqs_resched, jiffies); | |
926 | } | |
9b9500da | 927 | if (IS_ENABLED(CONFIG_IRQ_WORK) && |
8aa670cd | 928 | !rdp->rcu_iw_pending && rdp->rcu_iw_gp_seq != rnp->gp_seq && |
9b9500da | 929 | (rnp->ffmask & rdp->grpmask)) { |
9b9500da | 930 | rdp->rcu_iw_pending = true; |
8aa670cd | 931 | rdp->rcu_iw_gp_seq = rnp->gp_seq; |
9b9500da PM |
932 | irq_work_queue_on(&rdp->rcu_iw, rdp->cpu); |
933 | } | |
934 | } | |
4914950a | 935 | |
a82dcc76 | 936 | return 0; |
64db4cff PM |
937 | } |
938 | ||
41e80595 PM |
939 | /* Trace-event wrapper function for trace_rcu_future_grace_period. */ |
940 | static void trace_rcu_this_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
b73de91d | 941 | unsigned long gp_seq_req, const char *s) |
0446be48 | 942 | { |
0937d045 PM |
943 | trace_rcu_future_grace_period(rcu_state.name, READ_ONCE(rnp->gp_seq), |
944 | gp_seq_req, rnp->level, | |
945 | rnp->grplo, rnp->grphi, s); | |
0446be48 PM |
946 | } |
947 | ||
948 | /* | |
b73de91d | 949 | * rcu_start_this_gp - Request the start of a particular grace period |
df2bf8f7 | 950 | * @rnp_start: The leaf node of the CPU from which to start. |
b73de91d JF |
951 | * @rdp: The rcu_data corresponding to the CPU from which to start. |
952 | * @gp_seq_req: The gp_seq of the grace period to start. | |
953 | * | |
41e80595 | 954 | * Start the specified grace period, as needed to handle newly arrived |
0446be48 | 955 | * callbacks. The required future grace periods are recorded in each |
7a1d0f23 | 956 | * rcu_node structure's ->gp_seq_needed field. Returns true if there |
48a7639c | 957 | * is reason to awaken the grace-period kthread. |
0446be48 | 958 | * |
d5cd9685 PM |
959 | * The caller must hold the specified rcu_node structure's ->lock, which |
960 | * is why the caller is responsible for waking the grace-period kthread. | |
b73de91d JF |
961 | * |
962 | * Returns true if the GP thread needs to be awakened else false. | |
0446be48 | 963 | */ |
df2bf8f7 | 964 | static bool rcu_start_this_gp(struct rcu_node *rnp_start, struct rcu_data *rdp, |
b73de91d | 965 | unsigned long gp_seq_req) |
0446be48 | 966 | { |
48a7639c | 967 | bool ret = false; |
df2bf8f7 | 968 | struct rcu_node *rnp; |
0446be48 PM |
969 | |
970 | /* | |
360e0da6 PM |
971 | * Use funnel locking to either acquire the root rcu_node |
972 | * structure's lock or bail out if the need for this grace period | |
df2bf8f7 JFG |
973 | * has already been recorded -- or if that grace period has in |
974 | * fact already started. If there is already a grace period in | |
975 | * progress in a non-leaf node, no recording is needed because the | |
976 | * end of the grace period will scan the leaf rcu_node structures. | |
977 | * Note that rnp_start->lock must not be released. | |
0446be48 | 978 | */ |
df2bf8f7 JFG |
979 | raw_lockdep_assert_held_rcu_node(rnp_start); |
980 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, TPS("Startleaf")); | |
981 | for (rnp = rnp_start; 1; rnp = rnp->parent) { | |
982 | if (rnp != rnp_start) | |
983 | raw_spin_lock_rcu_node(rnp); | |
984 | if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req) || | |
985 | rcu_seq_started(&rnp->gp_seq, gp_seq_req) || | |
986 | (rnp != rnp_start && | |
987 | rcu_seq_state(rcu_seq_current(&rnp->gp_seq)))) { | |
988 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, | |
b73de91d | 989 | TPS("Prestarted")); |
360e0da6 PM |
990 | goto unlock_out; |
991 | } | |
8ff37290 | 992 | WRITE_ONCE(rnp->gp_seq_needed, gp_seq_req); |
226ca5e7 | 993 | if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) { |
a2165e41 | 994 | /* |
226ca5e7 JFG |
995 | * We just marked the leaf or internal node, and a |
996 | * grace period is in progress, which means that | |
997 | * rcu_gp_cleanup() will see the marking. Bail to | |
998 | * reduce contention. | |
a2165e41 | 999 | */ |
df2bf8f7 | 1000 | trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, |
b73de91d | 1001 | TPS("Startedleaf")); |
a2165e41 PM |
1002 | goto unlock_out; |
1003 | } | |
df2bf8f7 JFG |
1004 | if (rnp != rnp_start && rnp->parent != NULL) |
1005 | raw_spin_unlock_rcu_node(rnp); | |
1006 | if (!rnp->parent) | |
360e0da6 | 1007 | break; /* At root, and perhaps also leaf. */ |
0446be48 PM |
1008 | } |
1009 | ||
360e0da6 | 1010 | /* If GP already in progress, just leave, otherwise start one. */ |
de8e8730 | 1011 | if (rcu_gp_in_progress()) { |
df2bf8f7 | 1012 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedleafroot")); |
0446be48 PM |
1013 | goto unlock_out; |
1014 | } | |
df2bf8f7 | 1015 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedroot")); |
9cbc5b97 | 1016 | WRITE_ONCE(rcu_state.gp_flags, rcu_state.gp_flags | RCU_GP_FLAG_INIT); |
2906d215 | 1017 | WRITE_ONCE(rcu_state.gp_req_activity, jiffies); |
5648d659 | 1018 | if (!READ_ONCE(rcu_state.gp_kthread)) { |
df2bf8f7 | 1019 | trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("NoGPkthread")); |
360e0da6 | 1020 | goto unlock_out; |
0446be48 | 1021 | } |
62ae1951 | 1022 | trace_rcu_grace_period(rcu_state.name, data_race(rcu_state.gp_seq), TPS("newreq")); |
360e0da6 | 1023 | ret = true; /* Caller must wake GP kthread. */ |
0446be48 | 1024 | unlock_out: |
ab5e869c | 1025 | /* Push furthest requested GP to leaf node and rcu_data structure. */ |
df2bf8f7 | 1026 | if (ULONG_CMP_LT(gp_seq_req, rnp->gp_seq_needed)) { |
8ff37290 PM |
1027 | WRITE_ONCE(rnp_start->gp_seq_needed, rnp->gp_seq_needed); |
1028 | WRITE_ONCE(rdp->gp_seq_needed, rnp->gp_seq_needed); | |
ab5e869c | 1029 | } |
df2bf8f7 JFG |
1030 | if (rnp != rnp_start) |
1031 | raw_spin_unlock_rcu_node(rnp); | |
48a7639c | 1032 | return ret; |
0446be48 PM |
1033 | } |
1034 | ||
1035 | /* | |
1036 | * Clean up any old requests for the just-ended grace period. Also return | |
d1e4f01d | 1037 | * whether any additional grace periods have been requested. |
0446be48 | 1038 | */ |
3481f2ea | 1039 | static bool rcu_future_gp_cleanup(struct rcu_node *rnp) |
0446be48 | 1040 | { |
fb31340f | 1041 | bool needmore; |
da1df50d | 1042 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
0446be48 | 1043 | |
7a1d0f23 PM |
1044 | needmore = ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed); |
1045 | if (!needmore) | |
1046 | rnp->gp_seq_needed = rnp->gp_seq; /* Avoid counter wrap. */ | |
b73de91d | 1047 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq, |
41e80595 | 1048 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); |
0446be48 PM |
1049 | return needmore; |
1050 | } | |
1051 | ||
48a7639c | 1052 | /* |
5648d659 PM |
1053 | * Awaken the grace-period kthread. Don't do a self-awaken (unless in an |
1054 | * interrupt or softirq handler, in which case we just might immediately | |
1055 | * sleep upon return, resulting in a grace-period hang), and don't bother | |
1056 | * awakening when there is nothing for the grace-period kthread to do | |
1057 | * (as in several CPUs raced to awaken, we lost), and finally don't try | |
1058 | * to awaken a kthread that has not yet been created. If all those checks | |
1059 | * are passed, track some debug information and awaken. | |
1d1f898d ZJ |
1060 | * |
1061 | * So why do the self-wakeup when in an interrupt or softirq handler | |
1062 | * in the grace-period kthread's context? Because the kthread might have | |
1063 | * been interrupted just as it was going to sleep, and just after the final | |
1064 | * pre-sleep check of the awaken condition. In this case, a wakeup really | |
1065 | * is required, and is therefore supplied. | |
48a7639c | 1066 | */ |
532c00c9 | 1067 | static void rcu_gp_kthread_wake(void) |
48a7639c | 1068 | { |
5648d659 PM |
1069 | struct task_struct *t = READ_ONCE(rcu_state.gp_kthread); |
1070 | ||
2407a64f | 1071 | if ((current == t && !in_hardirq() && !in_serving_softirq()) || |
5648d659 | 1072 | !READ_ONCE(rcu_state.gp_flags) || !t) |
48a7639c | 1073 | return; |
fd897573 PM |
1074 | WRITE_ONCE(rcu_state.gp_wake_time, jiffies); |
1075 | WRITE_ONCE(rcu_state.gp_wake_seq, READ_ONCE(rcu_state.gp_seq)); | |
532c00c9 | 1076 | swake_up_one(&rcu_state.gp_wq); |
48a7639c PM |
1077 | } |
1078 | ||
dc35c893 | 1079 | /* |
29365e56 PM |
1080 | * If there is room, assign a ->gp_seq number to any callbacks on this |
1081 | * CPU that have not already been assigned. Also accelerate any callbacks | |
1082 | * that were previously assigned a ->gp_seq number that has since proven | |
1083 | * to be too conservative, which can happen if callbacks get assigned a | |
1084 | * ->gp_seq number while RCU is idle, but with reference to a non-root | |
1085 | * rcu_node structure. This function is idempotent, so it does not hurt | |
1086 | * to call it repeatedly. Returns an flag saying that we should awaken | |
1087 | * the RCU grace-period kthread. | |
dc35c893 PM |
1088 | * |
1089 | * The caller must hold rnp->lock with interrupts disabled. | |
1090 | */ | |
02f50142 | 1091 | static bool rcu_accelerate_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1092 | { |
b73de91d | 1093 | unsigned long gp_seq_req; |
15fecf89 | 1094 | bool ret = false; |
dc35c893 | 1095 | |
d1b222c6 | 1096 | rcu_lockdep_assert_cblist_protected(rdp); |
a32e01ee | 1097 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1098 | |
15fecf89 PM |
1099 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1100 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1101 | return false; |
dc35c893 | 1102 | |
3afe7fa5 JFG |
1103 | trace_rcu_segcb_stats(&rdp->cblist, TPS("SegCbPreAcc")); |
1104 | ||
dc35c893 | 1105 | /* |
15fecf89 PM |
1106 | * Callbacks are often registered with incomplete grace-period |
1107 | * information. Something about the fact that getting exact | |
1108 | * information requires acquiring a global lock... RCU therefore | |
1109 | * makes a conservative estimate of the grace period number at which | |
1110 | * a given callback will become ready to invoke. The following | |
1111 | * code checks this estimate and improves it when possible, thus | |
1112 | * accelerating callback invocation to an earlier grace-period | |
1113 | * number. | |
dc35c893 | 1114 | */ |
9cbc5b97 | 1115 | gp_seq_req = rcu_seq_snap(&rcu_state.gp_seq); |
b73de91d JF |
1116 | if (rcu_segcblist_accelerate(&rdp->cblist, gp_seq_req)) |
1117 | ret = rcu_start_this_gp(rnp, rdp, gp_seq_req); | |
6d4b418c PM |
1118 | |
1119 | /* Trace depending on how much we were able to accelerate. */ | |
15fecf89 | 1120 | if (rcu_segcblist_restempty(&rdp->cblist, RCU_WAIT_TAIL)) |
a7886e89 | 1121 | trace_rcu_grace_period(rcu_state.name, gp_seq_req, TPS("AccWaitCB")); |
6d4b418c | 1122 | else |
a7886e89 JFG |
1123 | trace_rcu_grace_period(rcu_state.name, gp_seq_req, TPS("AccReadyCB")); |
1124 | ||
3afe7fa5 JFG |
1125 | trace_rcu_segcb_stats(&rdp->cblist, TPS("SegCbPostAcc")); |
1126 | ||
48a7639c | 1127 | return ret; |
dc35c893 PM |
1128 | } |
1129 | ||
e44e73ca PM |
1130 | /* |
1131 | * Similar to rcu_accelerate_cbs(), but does not require that the leaf | |
1132 | * rcu_node structure's ->lock be held. It consults the cached value | |
1133 | * of ->gp_seq_needed in the rcu_data structure, and if that indicates | |
1134 | * that a new grace-period request be made, invokes rcu_accelerate_cbs() | |
1135 | * while holding the leaf rcu_node structure's ->lock. | |
1136 | */ | |
c6e09b97 | 1137 | static void rcu_accelerate_cbs_unlocked(struct rcu_node *rnp, |
e44e73ca PM |
1138 | struct rcu_data *rdp) |
1139 | { | |
1140 | unsigned long c; | |
1141 | bool needwake; | |
1142 | ||
d1b222c6 | 1143 | rcu_lockdep_assert_cblist_protected(rdp); |
c6e09b97 | 1144 | c = rcu_seq_snap(&rcu_state.gp_seq); |
a5b89501 | 1145 | if (!READ_ONCE(rdp->gpwrap) && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { |
e44e73ca PM |
1146 | /* Old request still live, so mark recent callbacks. */ |
1147 | (void)rcu_segcblist_accelerate(&rdp->cblist, c); | |
1148 | return; | |
1149 | } | |
1150 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
02f50142 | 1151 | needwake = rcu_accelerate_cbs(rnp, rdp); |
e44e73ca PM |
1152 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
1153 | if (needwake) | |
532c00c9 | 1154 | rcu_gp_kthread_wake(); |
e44e73ca PM |
1155 | } |
1156 | ||
dc35c893 PM |
1157 | /* |
1158 | * Move any callbacks whose grace period has completed to the | |
1159 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
29365e56 | 1160 | * assign ->gp_seq numbers to any callbacks in the RCU_NEXT_TAIL |
dc35c893 PM |
1161 | * sublist. This function is idempotent, so it does not hurt to |
1162 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1163 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1164 | * |
1165 | * The caller must hold rnp->lock with interrupts disabled. | |
1166 | */ | |
834f56bf | 1167 | static bool rcu_advance_cbs(struct rcu_node *rnp, struct rcu_data *rdp) |
dc35c893 | 1168 | { |
d1b222c6 | 1169 | rcu_lockdep_assert_cblist_protected(rdp); |
a32e01ee | 1170 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1171 | |
15fecf89 PM |
1172 | /* If no pending (not yet ready to invoke) callbacks, nothing to do. */ |
1173 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) | |
48a7639c | 1174 | return false; |
dc35c893 PM |
1175 | |
1176 | /* | |
29365e56 | 1177 | * Find all callbacks whose ->gp_seq numbers indicate that they |
dc35c893 PM |
1178 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. |
1179 | */ | |
29365e56 | 1180 | rcu_segcblist_advance(&rdp->cblist, rnp->gp_seq); |
dc35c893 PM |
1181 | |
1182 | /* Classify any remaining callbacks. */ | |
02f50142 | 1183 | return rcu_accelerate_cbs(rnp, rdp); |
dc35c893 PM |
1184 | } |
1185 | ||
7f36ef82 PM |
1186 | /* |
1187 | * Move and classify callbacks, but only if doing so won't require | |
1188 | * that the RCU grace-period kthread be awakened. | |
1189 | */ | |
1190 | static void __maybe_unused rcu_advance_cbs_nowake(struct rcu_node *rnp, | |
1191 | struct rcu_data *rdp) | |
1192 | { | |
d1b222c6 | 1193 | rcu_lockdep_assert_cblist_protected(rdp); |
614ddad1 | 1194 | if (!rcu_seq_state(rcu_seq_current(&rnp->gp_seq)) || !raw_spin_trylock_rcu_node(rnp)) |
7f36ef82 | 1195 | return; |
614ddad1 PM |
1196 | // The grace period cannot end while we hold the rcu_node lock. |
1197 | if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) | |
1198 | WARN_ON_ONCE(rcu_advance_cbs(rnp, rdp)); | |
6608c3a0 | 1199 | raw_spin_unlock_rcu_node(rnp); |
7f36ef82 PM |
1200 | } |
1201 | ||
1a2f5d57 PM |
1202 | /* |
1203 | * In CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels, attempt to generate a | |
1204 | * quiescent state. This is intended to be invoked when the CPU notices | |
1205 | * a new grace period. | |
1206 | */ | |
1207 | static void rcu_strict_gp_check_qs(void) | |
1208 | { | |
1209 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) { | |
1210 | rcu_read_lock(); | |
1211 | rcu_read_unlock(); | |
1212 | } | |
1213 | } | |
1214 | ||
d09b62df | 1215 | /* |
ba9fbe95 PM |
1216 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1217 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1218 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1219 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1220 | */ |
c7e48f7b | 1221 | static bool __note_gp_changes(struct rcu_node *rnp, struct rcu_data *rdp) |
d09b62df | 1222 | { |
5d6742b3 | 1223 | bool ret = false; |
b5ea0370 | 1224 | bool need_qs; |
3820b513 | 1225 | const bool offloaded = rcu_rdp_is_offloaded(rdp); |
48a7639c | 1226 | |
a32e01ee | 1227 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1228 | |
67e14c1e PM |
1229 | if (rdp->gp_seq == rnp->gp_seq) |
1230 | return false; /* Nothing to do. */ | |
d09b62df | 1231 | |
67e14c1e PM |
1232 | /* Handle the ends of any preceding grace periods first. */ |
1233 | if (rcu_seq_completed_gp(rdp->gp_seq, rnp->gp_seq) || | |
1234 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
5d6742b3 PM |
1235 | if (!offloaded) |
1236 | ret = rcu_advance_cbs(rnp, rdp); /* Advance CBs. */ | |
b5ea0370 | 1237 | rdp->core_needs_qs = false; |
9cbc5b97 | 1238 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuend")); |
67e14c1e | 1239 | } else { |
5d6742b3 PM |
1240 | if (!offloaded) |
1241 | ret = rcu_accelerate_cbs(rnp, rdp); /* Recent CBs. */ | |
b5ea0370 PM |
1242 | if (rdp->core_needs_qs) |
1243 | rdp->core_needs_qs = !!(rnp->qsmask & rdp->grpmask); | |
d09b62df | 1244 | } |
398ebe60 | 1245 | |
67e14c1e PM |
1246 | /* Now handle the beginnings of any new-to-this-CPU grace periods. */ |
1247 | if (rcu_seq_new_gp(rdp->gp_seq, rnp->gp_seq) || | |
1248 | unlikely(READ_ONCE(rdp->gpwrap))) { | |
6eaef633 PM |
1249 | /* |
1250 | * If the current grace period is waiting for this CPU, | |
1251 | * set up to detect a quiescent state, otherwise don't | |
1252 | * go looking for one. | |
1253 | */ | |
9cbc5b97 | 1254 | trace_rcu_grace_period(rcu_state.name, rnp->gp_seq, TPS("cpustart")); |
b5ea0370 PM |
1255 | need_qs = !!(rnp->qsmask & rdp->grpmask); |
1256 | rdp->cpu_no_qs.b.norm = need_qs; | |
1257 | rdp->core_needs_qs = need_qs; | |
6eaef633 PM |
1258 | zero_cpu_stall_ticks(rdp); |
1259 | } | |
67e14c1e | 1260 | rdp->gp_seq = rnp->gp_seq; /* Remember new grace-period state. */ |
13dc7d0c | 1261 | if (ULONG_CMP_LT(rdp->gp_seq_needed, rnp->gp_seq_needed) || rdp->gpwrap) |
8ff37290 | 1262 | WRITE_ONCE(rdp->gp_seq_needed, rnp->gp_seq_needed); |
c708b08c PM |
1263 | if (IS_ENABLED(CONFIG_PROVE_RCU) && READ_ONCE(rdp->gpwrap)) |
1264 | WRITE_ONCE(rdp->last_sched_clock, jiffies); | |
3d18469a PM |
1265 | WRITE_ONCE(rdp->gpwrap, false); |
1266 | rcu_gpnum_ovf(rnp, rdp); | |
48a7639c | 1267 | return ret; |
6eaef633 PM |
1268 | } |
1269 | ||
15cabdff | 1270 | static void note_gp_changes(struct rcu_data *rdp) |
6eaef633 PM |
1271 | { |
1272 | unsigned long flags; | |
48a7639c | 1273 | bool needwake; |
6eaef633 PM |
1274 | struct rcu_node *rnp; |
1275 | ||
1276 | local_irq_save(flags); | |
1277 | rnp = rdp->mynode; | |
67e14c1e | 1278 | if ((rdp->gp_seq == rcu_seq_current(&rnp->gp_seq) && |
7d0ae808 | 1279 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ |
2a67e741 | 1280 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
1281 | local_irq_restore(flags); |
1282 | return; | |
1283 | } | |
c7e48f7b | 1284 | needwake = __note_gp_changes(rnp, rdp); |
67c583a7 | 1285 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1a2f5d57 | 1286 | rcu_strict_gp_check_qs(); |
48a7639c | 1287 | if (needwake) |
532c00c9 | 1288 | rcu_gp_kthread_wake(); |
6eaef633 PM |
1289 | } |
1290 | ||
99d6a2ac PM |
1291 | static atomic_t *rcu_gp_slow_suppress; |
1292 | ||
1293 | /* Register a counter to suppress debugging grace-period delays. */ | |
1294 | void rcu_gp_slow_register(atomic_t *rgssp) | |
1295 | { | |
1296 | WARN_ON_ONCE(rcu_gp_slow_suppress); | |
1297 | ||
1298 | WRITE_ONCE(rcu_gp_slow_suppress, rgssp); | |
1299 | } | |
1300 | EXPORT_SYMBOL_GPL(rcu_gp_slow_register); | |
1301 | ||
1302 | /* Unregister a counter, with NULL for not caring which. */ | |
1303 | void rcu_gp_slow_unregister(atomic_t *rgssp) | |
1304 | { | |
1305 | WARN_ON_ONCE(rgssp && rgssp != rcu_gp_slow_suppress); | |
1306 | ||
1307 | WRITE_ONCE(rcu_gp_slow_suppress, NULL); | |
1308 | } | |
1309 | EXPORT_SYMBOL_GPL(rcu_gp_slow_unregister); | |
1310 | ||
1311 | static bool rcu_gp_slow_is_suppressed(void) | |
1312 | { | |
1313 | atomic_t *rgssp = READ_ONCE(rcu_gp_slow_suppress); | |
1314 | ||
1315 | return rgssp && atomic_read(rgssp); | |
1316 | } | |
1317 | ||
22212332 | 1318 | static void rcu_gp_slow(int delay) |
0f41c0dd | 1319 | { |
99d6a2ac PM |
1320 | if (!rcu_gp_slow_is_suppressed() && delay > 0 && |
1321 | !(rcu_seq_ctr(rcu_state.gp_seq) % (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) | |
77865dea | 1322 | schedule_timeout_idle(delay); |
0f41c0dd PM |
1323 | } |
1324 | ||
55b2dcf5 PM |
1325 | static unsigned long sleep_duration; |
1326 | ||
1327 | /* Allow rcutorture to stall the grace-period kthread. */ | |
1328 | void rcu_gp_set_torture_wait(int duration) | |
1329 | { | |
1330 | if (IS_ENABLED(CONFIG_RCU_TORTURE_TEST) && duration > 0) | |
1331 | WRITE_ONCE(sleep_duration, duration); | |
1332 | } | |
1333 | EXPORT_SYMBOL_GPL(rcu_gp_set_torture_wait); | |
1334 | ||
1335 | /* Actually implement the aforementioned wait. */ | |
1336 | static void rcu_gp_torture_wait(void) | |
1337 | { | |
1338 | unsigned long duration; | |
1339 | ||
1340 | if (!IS_ENABLED(CONFIG_RCU_TORTURE_TEST)) | |
1341 | return; | |
1342 | duration = xchg(&sleep_duration, 0UL); | |
1343 | if (duration > 0) { | |
1344 | pr_alert("%s: Waiting %lu jiffies\n", __func__, duration); | |
77865dea | 1345 | schedule_timeout_idle(duration); |
55b2dcf5 PM |
1346 | pr_alert("%s: Wait complete\n", __func__); |
1347 | } | |
1348 | } | |
1349 | ||
933ada2c PM |
1350 | /* |
1351 | * Handler for on_each_cpu() to invoke the target CPU's RCU core | |
1352 | * processing. | |
1353 | */ | |
1354 | static void rcu_strict_gp_boundary(void *unused) | |
1355 | { | |
1356 | invoke_rcu_core(); | |
1357 | } | |
1358 | ||
bf95b2bc PM |
1359 | // Has rcu_init() been invoked? This is used (for example) to determine |
1360 | // whether spinlocks may be acquired safely. | |
1361 | static bool rcu_init_invoked(void) | |
1362 | { | |
1363 | return !!rcu_state.n_online_cpus; | |
1364 | } | |
1365 | ||
1366 | // Make the polled API aware of the beginning of a grace period. | |
1367 | static void rcu_poll_gp_seq_start(unsigned long *snap) | |
1368 | { | |
1369 | struct rcu_node *rnp = rcu_get_root(); | |
1370 | ||
1371 | if (rcu_init_invoked()) | |
1372 | raw_lockdep_assert_held_rcu_node(rnp); | |
1373 | ||
1374 | // If RCU was idle, note beginning of GP. | |
1375 | if (!rcu_seq_state(rcu_state.gp_seq_polled)) | |
1376 | rcu_seq_start(&rcu_state.gp_seq_polled); | |
1377 | ||
1378 | // Either way, record current state. | |
1379 | *snap = rcu_state.gp_seq_polled; | |
1380 | } | |
1381 | ||
1382 | // Make the polled API aware of the end of a grace period. | |
1383 | static void rcu_poll_gp_seq_end(unsigned long *snap) | |
1384 | { | |
1385 | struct rcu_node *rnp = rcu_get_root(); | |
1386 | ||
1387 | if (rcu_init_invoked()) | |
1388 | raw_lockdep_assert_held_rcu_node(rnp); | |
1389 | ||
1390 | // If the previously noted GP is still in effect, record the | |
1391 | // end of that GP. Either way, zero counter to avoid counter-wrap | |
1392 | // problems. | |
1393 | if (*snap && *snap == rcu_state.gp_seq_polled) { | |
1394 | rcu_seq_end(&rcu_state.gp_seq_polled); | |
1395 | rcu_state.gp_seq_polled_snap = 0; | |
dd041405 | 1396 | rcu_state.gp_seq_polled_exp_snap = 0; |
bf95b2bc PM |
1397 | } else { |
1398 | *snap = 0; | |
1399 | } | |
1400 | } | |
1401 | ||
1402 | // Make the polled API aware of the beginning of a grace period, but | |
1403 | // where caller does not hold the root rcu_node structure's lock. | |
1404 | static void rcu_poll_gp_seq_start_unlocked(unsigned long *snap) | |
1405 | { | |
1406 | struct rcu_node *rnp = rcu_get_root(); | |
1407 | ||
1408 | if (rcu_init_invoked()) { | |
1409 | lockdep_assert_irqs_enabled(); | |
1410 | raw_spin_lock_irq_rcu_node(rnp); | |
1411 | } | |
1412 | rcu_poll_gp_seq_start(snap); | |
1413 | if (rcu_init_invoked()) | |
1414 | raw_spin_unlock_irq_rcu_node(rnp); | |
1415 | } | |
1416 | ||
1417 | // Make the polled API aware of the end of a grace period, but where | |
1418 | // caller does not hold the root rcu_node structure's lock. | |
1419 | static void rcu_poll_gp_seq_end_unlocked(unsigned long *snap) | |
1420 | { | |
1421 | struct rcu_node *rnp = rcu_get_root(); | |
1422 | ||
1423 | if (rcu_init_invoked()) { | |
1424 | lockdep_assert_irqs_enabled(); | |
1425 | raw_spin_lock_irq_rcu_node(rnp); | |
1426 | } | |
1427 | rcu_poll_gp_seq_end(snap); | |
1428 | if (rcu_init_invoked()) | |
1429 | raw_spin_unlock_irq_rcu_node(rnp); | |
1430 | } | |
1431 | ||
b3dbec76 | 1432 | /* |
45fed3e7 | 1433 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 1434 | */ |
f74126dc | 1435 | static noinline_for_stack bool rcu_gp_init(void) |
b3dbec76 | 1436 | { |
ec2c2976 | 1437 | unsigned long flags; |
0aa04b05 | 1438 | unsigned long oldmask; |
ec2c2976 | 1439 | unsigned long mask; |
b3dbec76 | 1440 | struct rcu_data *rdp; |
336a4f6c | 1441 | struct rcu_node *rnp = rcu_get_root(); |
b3dbec76 | 1442 | |
9cbc5b97 | 1443 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 1444 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 | 1445 | if (!READ_ONCE(rcu_state.gp_flags)) { |
f7be8209 | 1446 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 1447 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1448 | return false; |
f7be8209 | 1449 | } |
9cbc5b97 | 1450 | WRITE_ONCE(rcu_state.gp_flags, 0); /* Clear all flags: New GP. */ |
b3dbec76 | 1451 | |
de8e8730 | 1452 | if (WARN_ON_ONCE(rcu_gp_in_progress())) { |
f7be8209 PM |
1453 | /* |
1454 | * Grace period already in progress, don't start another. | |
1455 | * Not supposed to be able to happen. | |
1456 | */ | |
67c583a7 | 1457 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1458 | return false; |
7fdefc10 PM |
1459 | } |
1460 | ||
7fdefc10 | 1461 | /* Advance to a new grace period and initialize state. */ |
ad3832e9 | 1462 | record_gp_stall_check_time(); |
ff3bb6f4 | 1463 | /* Record GP times before starting GP, hence rcu_seq_start(). */ |
9cbc5b97 | 1464 | rcu_seq_start(&rcu_state.gp_seq); |
62ae1951 | 1465 | ASSERT_EXCLUSIVE_WRITER(rcu_state.gp_seq); |
9cbc5b97 | 1466 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("start")); |
bf95b2bc | 1467 | rcu_poll_gp_seq_start(&rcu_state.gp_seq_polled_snap); |
67c583a7 | 1468 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 1469 | |
0aa04b05 | 1470 | /* |
f37599e6 JFG |
1471 | * Apply per-leaf buffered online and offline operations to |
1472 | * the rcu_node tree. Note that this new grace period need not | |
1473 | * wait for subsequent online CPUs, and that RCU hooks in the CPU | |
1474 | * offlining path, when combined with checks in this function, | |
1475 | * will handle CPUs that are currently going offline or that will | |
1476 | * go offline later. Please also refer to "Hotplug CPU" section | |
1477 | * of RCU's Requirements documentation. | |
0aa04b05 | 1478 | */ |
683954e5 | 1479 | WRITE_ONCE(rcu_state.gp_state, RCU_GP_ONOFF); |
82980b16 | 1480 | /* Exclude CPU hotplug operations. */ |
aedf4ba9 | 1481 | rcu_for_each_leaf_node(rnp) { |
82980b16 DW |
1482 | local_irq_save(flags); |
1483 | arch_spin_lock(&rcu_state.ofl_lock); | |
1484 | raw_spin_lock_rcu_node(rnp); | |
0aa04b05 PM |
1485 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
1486 | !rnp->wait_blkd_tasks) { | |
1487 | /* Nothing to do on this leaf rcu_node structure. */ | |
82980b16 DW |
1488 | raw_spin_unlock_rcu_node(rnp); |
1489 | arch_spin_unlock(&rcu_state.ofl_lock); | |
1490 | local_irq_restore(flags); | |
0aa04b05 PM |
1491 | continue; |
1492 | } | |
1493 | ||
1494 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
1495 | oldmask = rnp->qsmaskinit; | |
1496 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
1497 | ||
1498 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
1499 | if (!oldmask != !rnp->qsmaskinit) { | |
962aff03 PM |
1500 | if (!oldmask) { /* First online CPU for rcu_node. */ |
1501 | if (!rnp->wait_blkd_tasks) /* Ever offline? */ | |
1502 | rcu_init_new_rnp(rnp); | |
1503 | } else if (rcu_preempt_has_tasks(rnp)) { | |
1504 | rnp->wait_blkd_tasks = true; /* blocked tasks */ | |
1505 | } else { /* Last offline CPU and can propagate. */ | |
0aa04b05 | 1506 | rcu_cleanup_dead_rnp(rnp); |
962aff03 | 1507 | } |
0aa04b05 PM |
1508 | } |
1509 | ||
1510 | /* | |
1511 | * If all waited-on tasks from prior grace period are | |
1512 | * done, and if all this rcu_node structure's CPUs are | |
1513 | * still offline, propagate up the rcu_node tree and | |
1514 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
1515 | * rcu_node structure's CPUs has since come back online, | |
962aff03 | 1516 | * simply clear ->wait_blkd_tasks. |
0aa04b05 PM |
1517 | */ |
1518 | if (rnp->wait_blkd_tasks && | |
962aff03 | 1519 | (!rcu_preempt_has_tasks(rnp) || rnp->qsmaskinit)) { |
0aa04b05 | 1520 | rnp->wait_blkd_tasks = false; |
962aff03 PM |
1521 | if (!rnp->qsmaskinit) |
1522 | rcu_cleanup_dead_rnp(rnp); | |
0aa04b05 PM |
1523 | } |
1524 | ||
82980b16 DW |
1525 | raw_spin_unlock_rcu_node(rnp); |
1526 | arch_spin_unlock(&rcu_state.ofl_lock); | |
1527 | local_irq_restore(flags); | |
0aa04b05 | 1528 | } |
22212332 | 1529 | rcu_gp_slow(gp_preinit_delay); /* Races with CPU hotplug. */ |
7fdefc10 PM |
1530 | |
1531 | /* | |
1532 | * Set the quiescent-state-needed bits in all the rcu_node | |
9cbc5b97 PM |
1533 | * structures for all currently online CPUs in breadth-first |
1534 | * order, starting from the root rcu_node structure, relying on the | |
1535 | * layout of the tree within the rcu_state.node[] array. Note that | |
1536 | * other CPUs will access only the leaves of the hierarchy, thus | |
1537 | * seeing that no grace period is in progress, at least until the | |
1538 | * corresponding leaf node has been initialized. | |
7fdefc10 PM |
1539 | * |
1540 | * The grace period cannot complete until the initialization | |
1541 | * process finishes, because this kthread handles both. | |
1542 | */ | |
683954e5 | 1543 | WRITE_ONCE(rcu_state.gp_state, RCU_GP_INIT); |
aedf4ba9 | 1544 | rcu_for_each_node_breadth_first(rnp) { |
22212332 | 1545 | rcu_gp_slow(gp_init_delay); |
ec2c2976 | 1546 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
da1df50d | 1547 | rdp = this_cpu_ptr(&rcu_data); |
81ab59a3 | 1548 | rcu_preempt_check_blocked_tasks(rnp); |
7fdefc10 | 1549 | rnp->qsmask = rnp->qsmaskinit; |
9cbc5b97 | 1550 | WRITE_ONCE(rnp->gp_seq, rcu_state.gp_seq); |
7fdefc10 | 1551 | if (rnp == rdp->mynode) |
c7e48f7b | 1552 | (void)__note_gp_changes(rnp, rdp); |
7fdefc10 | 1553 | rcu_preempt_boost_start_gp(rnp); |
9cbc5b97 | 1554 | trace_rcu_grace_period_init(rcu_state.name, rnp->gp_seq, |
7fdefc10 PM |
1555 | rnp->level, rnp->grplo, |
1556 | rnp->grphi, rnp->qsmask); | |
ec2c2976 PM |
1557 | /* Quiescent states for tasks on any now-offline CPUs. */ |
1558 | mask = rnp->qsmask & ~rnp->qsmaskinitnext; | |
f2e2df59 | 1559 | rnp->rcu_gp_init_mask = mask; |
ec2c2976 | 1560 | if ((mask || rnp->wait_blkd_tasks) && rcu_is_leaf_node(rnp)) |
b50912d0 | 1561 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
ec2c2976 PM |
1562 | else |
1563 | raw_spin_unlock_irq_rcu_node(rnp); | |
cee43939 | 1564 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1565 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
7fdefc10 | 1566 | } |
b3dbec76 | 1567 | |
933ada2c PM |
1568 | // If strict, make all CPUs aware of new grace period. |
1569 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) | |
1570 | on_each_cpu(rcu_strict_gp_boundary, NULL, 0); | |
1571 | ||
45fed3e7 | 1572 | return true; |
7fdefc10 | 1573 | } |
b3dbec76 | 1574 | |
b9a425cf | 1575 | /* |
b3dae109 | 1576 | * Helper function for swait_event_idle_exclusive() wakeup at force-quiescent-state |
d5374226 | 1577 | * time. |
b9a425cf | 1578 | */ |
0854a05c | 1579 | static bool rcu_gp_fqs_check_wake(int *gfp) |
b9a425cf | 1580 | { |
336a4f6c | 1581 | struct rcu_node *rnp = rcu_get_root(); |
b9a425cf | 1582 | |
1fca4d12 PM |
1583 | // If under overload conditions, force an immediate FQS scan. |
1584 | if (*gfp & RCU_GP_FLAG_OVLD) | |
1585 | return true; | |
1586 | ||
1587 | // Someone like call_rcu() requested a force-quiescent-state scan. | |
0854a05c | 1588 | *gfp = READ_ONCE(rcu_state.gp_flags); |
b9a425cf PM |
1589 | if (*gfp & RCU_GP_FLAG_FQS) |
1590 | return true; | |
1591 | ||
1fca4d12 | 1592 | // The current grace period has completed. |
b9a425cf PM |
1593 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) |
1594 | return true; | |
1595 | ||
1596 | return false; | |
1597 | } | |
1598 | ||
4cdfc175 PM |
1599 | /* |
1600 | * Do one round of quiescent-state forcing. | |
1601 | */ | |
0854a05c | 1602 | static void rcu_gp_fqs(bool first_time) |
4cdfc175 | 1603 | { |
336a4f6c | 1604 | struct rcu_node *rnp = rcu_get_root(); |
4cdfc175 | 1605 | |
9cbc5b97 | 1606 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2431774f | 1607 | WRITE_ONCE(rcu_state.n_force_qs, rcu_state.n_force_qs + 1); |
77f81fe0 | 1608 | if (first_time) { |
4cdfc175 | 1609 | /* Collect dyntick-idle snapshots. */ |
e9ecb780 | 1610 | force_qs_rnp(dyntick_save_progress_counter); |
4cdfc175 PM |
1611 | } else { |
1612 | /* Handle dyntick-idle and offline CPUs. */ | |
e9ecb780 | 1613 | force_qs_rnp(rcu_implicit_dynticks_qs); |
4cdfc175 PM |
1614 | } |
1615 | /* Clear flag to prevent immediate re-entry. */ | |
9cbc5b97 | 1616 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 1617 | raw_spin_lock_irq_rcu_node(rnp); |
9cbc5b97 PM |
1618 | WRITE_ONCE(rcu_state.gp_flags, |
1619 | READ_ONCE(rcu_state.gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 1620 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 1621 | } |
4cdfc175 PM |
1622 | } |
1623 | ||
c3854a05 PM |
1624 | /* |
1625 | * Loop doing repeated quiescent-state forcing until the grace period ends. | |
1626 | */ | |
f74126dc | 1627 | static noinline_for_stack void rcu_gp_fqs_loop(void) |
c3854a05 | 1628 | { |
9bdb5b3a | 1629 | bool first_gp_fqs = true; |
1fca4d12 | 1630 | int gf = 0; |
c3854a05 PM |
1631 | unsigned long j; |
1632 | int ret; | |
1633 | struct rcu_node *rnp = rcu_get_root(); | |
1634 | ||
c06aed0e | 1635 | j = READ_ONCE(jiffies_till_first_fqs); |
1fca4d12 PM |
1636 | if (rcu_state.cbovld) |
1637 | gf = RCU_GP_FLAG_OVLD; | |
c3854a05 PM |
1638 | ret = 0; |
1639 | for (;;) { | |
fb77dccf PM |
1640 | if (rcu_state.cbovld) { |
1641 | j = (j + 2) / 3; | |
1642 | if (j <= 0) | |
1643 | j = 1; | |
1644 | } | |
1645 | if (!ret || time_before(jiffies + j, rcu_state.jiffies_force_qs)) { | |
683954e5 NU |
1646 | WRITE_ONCE(rcu_state.jiffies_force_qs, jiffies + j); |
1647 | /* | |
1648 | * jiffies_force_qs before RCU_GP_WAIT_FQS state | |
1649 | * update; required for stall checks. | |
1650 | */ | |
1651 | smp_wmb(); | |
c3854a05 | 1652 | WRITE_ONCE(rcu_state.jiffies_kick_kthreads, |
9cf422a8 | 1653 | jiffies + (j ? 3 * j : 2)); |
c3854a05 | 1654 | } |
0f11ad32 | 1655 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, |
c3854a05 | 1656 | TPS("fqswait")); |
683954e5 | 1657 | WRITE_ONCE(rcu_state.gp_state, RCU_GP_WAIT_FQS); |
eb880949 LS |
1658 | (void)swait_event_idle_timeout_exclusive(rcu_state.gp_wq, |
1659 | rcu_gp_fqs_check_wake(&gf), j); | |
55b2dcf5 | 1660 | rcu_gp_torture_wait(); |
683954e5 | 1661 | WRITE_ONCE(rcu_state.gp_state, RCU_GP_DOING_FQS); |
c3854a05 | 1662 | /* Locking provides needed memory barriers. */ |
a03ae49c NU |
1663 | /* |
1664 | * Exit the loop if the root rcu_node structure indicates that the grace period | |
1665 | * has ended, leave the loop. The rcu_preempt_blocked_readers_cgp(rnp) check | |
1666 | * is required only for single-node rcu_node trees because readers blocking | |
1667 | * the current grace period are queued only on leaf rcu_node structures. | |
1668 | * For multi-node trees, checking the root node's ->qsmask suffices, because a | |
1669 | * given root node's ->qsmask bit is cleared only when all CPUs and tasks from | |
1670 | * the corresponding leaf nodes have passed through their quiescent state. | |
1671 | */ | |
c3854a05 PM |
1672 | if (!READ_ONCE(rnp->qsmask) && |
1673 | !rcu_preempt_blocked_readers_cgp(rnp)) | |
1674 | break; | |
1675 | /* If time for quiescent-state forcing, do it. */ | |
29ffebc5 | 1676 | if (!time_after(rcu_state.jiffies_force_qs, jiffies) || |
9c392453 | 1677 | (gf & (RCU_GP_FLAG_FQS | RCU_GP_FLAG_OVLD))) { |
0f11ad32 | 1678 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, |
c3854a05 PM |
1679 | TPS("fqsstart")); |
1680 | rcu_gp_fqs(first_gp_fqs); | |
1fca4d12 PM |
1681 | gf = 0; |
1682 | if (first_gp_fqs) { | |
1683 | first_gp_fqs = false; | |
1684 | gf = rcu_state.cbovld ? RCU_GP_FLAG_OVLD : 0; | |
1685 | } | |
0f11ad32 | 1686 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, |
c3854a05 PM |
1687 | TPS("fqsend")); |
1688 | cond_resched_tasks_rcu_qs(); | |
1689 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1690 | ret = 0; /* Force full wait till next FQS. */ | |
c06aed0e | 1691 | j = READ_ONCE(jiffies_till_next_fqs); |
c3854a05 PM |
1692 | } else { |
1693 | /* Deal with stray signal. */ | |
1694 | cond_resched_tasks_rcu_qs(); | |
1695 | WRITE_ONCE(rcu_state.gp_activity, jiffies); | |
1696 | WARN_ON(signal_pending(current)); | |
0f11ad32 | 1697 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, |
c3854a05 PM |
1698 | TPS("fqswaitsig")); |
1699 | ret = 1; /* Keep old FQS timing. */ | |
1700 | j = jiffies; | |
1701 | if (time_after(jiffies, rcu_state.jiffies_force_qs)) | |
1702 | j = 1; | |
1703 | else | |
1704 | j = rcu_state.jiffies_force_qs - j; | |
1fca4d12 | 1705 | gf = 0; |
c3854a05 PM |
1706 | } |
1707 | } | |
1708 | } | |
1709 | ||
7fdefc10 PM |
1710 | /* |
1711 | * Clean up after the old grace period. | |
1712 | */ | |
2f20de99 | 1713 | static noinline void rcu_gp_cleanup(void) |
7fdefc10 | 1714 | { |
b2b00ddf | 1715 | int cpu; |
48a7639c | 1716 | bool needgp = false; |
b2b00ddf | 1717 | unsigned long gp_duration; |
de30ad51 | 1718 | unsigned long new_gp_seq; |
5d6742b3 | 1719 | bool offloaded; |
7fdefc10 | 1720 | struct rcu_data *rdp; |
336a4f6c | 1721 | struct rcu_node *rnp = rcu_get_root(); |
abedf8e2 | 1722 | struct swait_queue_head *sq; |
b3dbec76 | 1723 | |
9cbc5b97 | 1724 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
2a67e741 | 1725 | raw_spin_lock_irq_rcu_node(rnp); |
c51d7b5e PM |
1726 | rcu_state.gp_end = jiffies; |
1727 | gp_duration = rcu_state.gp_end - rcu_state.gp_start; | |
9cbc5b97 PM |
1728 | if (gp_duration > rcu_state.gp_max) |
1729 | rcu_state.gp_max = gp_duration; | |
b3dbec76 | 1730 | |
7fdefc10 PM |
1731 | /* |
1732 | * We know the grace period is complete, but to everyone else | |
1733 | * it appears to still be ongoing. But it is also the case | |
1734 | * that to everyone else it looks like there is nothing that | |
1735 | * they can do to advance the grace period. It is therefore | |
1736 | * safe for us to drop the lock in order to mark the grace | |
1737 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 1738 | */ |
bf95b2bc | 1739 | rcu_poll_gp_seq_end(&rcu_state.gp_seq_polled_snap); |
67c583a7 | 1740 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 1741 | |
5d4b8659 | 1742 | /* |
ff3bb6f4 PM |
1743 | * Propagate new ->gp_seq value to rcu_node structures so that |
1744 | * other CPUs don't have to wait until the start of the next grace | |
1745 | * period to process their callbacks. This also avoids some nasty | |
1746 | * RCU grace-period initialization races by forcing the end of | |
1747 | * the current grace period to be completely recorded in all of | |
1748 | * the rcu_node structures before the beginning of the next grace | |
1749 | * period is recorded in any of the rcu_node structures. | |
5d4b8659 | 1750 | */ |
9cbc5b97 | 1751 | new_gp_seq = rcu_state.gp_seq; |
de30ad51 | 1752 | rcu_seq_end(&new_gp_seq); |
aedf4ba9 | 1753 | rcu_for_each_node_breadth_first(rnp) { |
2a67e741 | 1754 | raw_spin_lock_irq_rcu_node(rnp); |
4bc8d555 | 1755 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
81ab59a3 | 1756 | dump_blkd_tasks(rnp, 10); |
5c60d25f | 1757 | WARN_ON_ONCE(rnp->qsmask); |
de30ad51 | 1758 | WRITE_ONCE(rnp->gp_seq, new_gp_seq); |
3fdefca9 PM |
1759 | if (!rnp->parent) |
1760 | smp_mb(); // Order against failing poll_state_synchronize_rcu_full(). | |
da1df50d | 1761 | rdp = this_cpu_ptr(&rcu_data); |
b11cc576 | 1762 | if (rnp == rdp->mynode) |
c7e48f7b | 1763 | needgp = __note_gp_changes(rnp, rdp) || needgp; |
78e4bc34 | 1764 | /* smp_mb() provided by prior unlock-lock pair. */ |
3481f2ea | 1765 | needgp = rcu_future_gp_cleanup(rnp) || needgp; |
b2b00ddf PM |
1766 | // Reset overload indication for CPUs no longer overloaded |
1767 | if (rcu_is_leaf_node(rnp)) | |
1768 | for_each_leaf_node_cpu_mask(rnp, cpu, rnp->cbovldmask) { | |
1769 | rdp = per_cpu_ptr(&rcu_data, cpu); | |
1770 | check_cb_ovld_locked(rdp, rnp); | |
1771 | } | |
065bb78c | 1772 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 1773 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 1774 | rcu_nocb_gp_cleanup(sq); |
cee43939 | 1775 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1776 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
22212332 | 1777 | rcu_gp_slow(gp_cleanup_delay); |
7fdefc10 | 1778 | } |
336a4f6c | 1779 | rnp = rcu_get_root(); |
9cbc5b97 | 1780 | raw_spin_lock_irq_rcu_node(rnp); /* GP before ->gp_seq update. */ |
7fdefc10 | 1781 | |
0a89e5a4 | 1782 | /* Declare grace period done, trace first to use old GP number. */ |
9cbc5b97 | 1783 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("end")); |
0a89e5a4 | 1784 | rcu_seq_end(&rcu_state.gp_seq); |
62ae1951 | 1785 | ASSERT_EXCLUSIVE_WRITER(rcu_state.gp_seq); |
683954e5 | 1786 | WRITE_ONCE(rcu_state.gp_state, RCU_GP_IDLE); |
fb31340f | 1787 | /* Check for GP requests since above loop. */ |
da1df50d | 1788 | rdp = this_cpu_ptr(&rcu_data); |
5b55072f | 1789 | if (!needgp && ULONG_CMP_LT(rnp->gp_seq, rnp->gp_seq_needed)) { |
abd13fdd | 1790 | trace_rcu_this_gp(rnp, rdp, rnp->gp_seq_needed, |
41e80595 | 1791 | TPS("CleanupMore")); |
fb31340f PM |
1792 | needgp = true; |
1793 | } | |
48a7639c | 1794 | /* Advance CBs to reduce false positives below. */ |
3820b513 | 1795 | offloaded = rcu_rdp_is_offloaded(rdp); |
5d6742b3 | 1796 | if ((offloaded || !rcu_accelerate_cbs(rnp, rdp)) && needgp) { |
75182a4e PM |
1797 | |
1798 | // We get here if a grace period was needed (“needgp”) | |
1799 | // and the above call to rcu_accelerate_cbs() did not set | |
1800 | // the RCU_GP_FLAG_INIT bit in ->gp_state (which records | |
1801 | // the need for another grace period). The purpose | |
1802 | // of the “offloaded” check is to avoid invoking | |
1803 | // rcu_accelerate_cbs() on an offloaded CPU because we do not | |
1804 | // hold the ->nocb_lock needed to safely access an offloaded | |
1805 | // ->cblist. We do not want to acquire that lock because | |
1806 | // it can be heavily contended during callback floods. | |
1807 | ||
9cbc5b97 | 1808 | WRITE_ONCE(rcu_state.gp_flags, RCU_GP_FLAG_INIT); |
2906d215 | 1809 | WRITE_ONCE(rcu_state.gp_req_activity, jiffies); |
75182a4e | 1810 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("newreq")); |
18390aea | 1811 | } else { |
75182a4e PM |
1812 | |
1813 | // We get here either if there is no need for an | |
1814 | // additional grace period or if rcu_accelerate_cbs() has | |
1815 | // already set the RCU_GP_FLAG_INIT bit in ->gp_flags. | |
1816 | // So all we need to do is to clear all of the other | |
1817 | // ->gp_flags bits. | |
1818 | ||
1819 | WRITE_ONCE(rcu_state.gp_flags, rcu_state.gp_flags & RCU_GP_FLAG_INIT); | |
bb311ecc | 1820 | } |
67c583a7 | 1821 | raw_spin_unlock_irq_rcu_node(rnp); |
4e025f52 PM |
1822 | |
1823 | // If strict, make all CPUs aware of the end of the old grace period. | |
1824 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) | |
1825 | on_each_cpu(rcu_strict_gp_boundary, NULL, 0); | |
7fdefc10 PM |
1826 | } |
1827 | ||
1828 | /* | |
1829 | * Body of kthread that handles grace periods. | |
1830 | */ | |
0854a05c | 1831 | static int __noreturn rcu_gp_kthread(void *unused) |
7fdefc10 | 1832 | { |
5871968d | 1833 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
1834 | for (;;) { |
1835 | ||
1836 | /* Handle grace-period start. */ | |
1837 | for (;;) { | |
0f11ad32 | 1838 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, |
63c4db78 | 1839 | TPS("reqwait")); |
683954e5 | 1840 | WRITE_ONCE(rcu_state.gp_state, RCU_GP_WAIT_GPS); |
9cbc5b97 PM |
1841 | swait_event_idle_exclusive(rcu_state.gp_wq, |
1842 | READ_ONCE(rcu_state.gp_flags) & | |
1843 | RCU_GP_FLAG_INIT); | |
55b2dcf5 | 1844 | rcu_gp_torture_wait(); |
683954e5 | 1845 | WRITE_ONCE(rcu_state.gp_state, RCU_GP_DONE_GPS); |
78e4bc34 | 1846 | /* Locking provides needed memory barrier. */ |
0854a05c | 1847 | if (rcu_gp_init()) |
7fdefc10 | 1848 | break; |
cee43939 | 1849 | cond_resched_tasks_rcu_qs(); |
9cbc5b97 | 1850 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
73a860cd | 1851 | WARN_ON(signal_pending(current)); |
0f11ad32 | 1852 | trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, |
63c4db78 | 1853 | TPS("reqwaitsig")); |
7fdefc10 | 1854 | } |
cabc49c1 | 1855 | |
4cdfc175 | 1856 | /* Handle quiescent-state forcing. */ |
c3854a05 | 1857 | rcu_gp_fqs_loop(); |
4cdfc175 PM |
1858 | |
1859 | /* Handle grace-period end. */ | |
683954e5 | 1860 | WRITE_ONCE(rcu_state.gp_state, RCU_GP_CLEANUP); |
0854a05c | 1861 | rcu_gp_cleanup(); |
683954e5 | 1862 | WRITE_ONCE(rcu_state.gp_state, RCU_GP_CLEANED); |
b3dbec76 | 1863 | } |
b3dbec76 PM |
1864 | } |
1865 | ||
f41d911f | 1866 | /* |
49918a54 PM |
1867 | * Report a full set of quiescent states to the rcu_state data structure. |
1868 | * Invoke rcu_gp_kthread_wake() to awaken the grace-period kthread if | |
1869 | * another grace period is required. Whether we wake the grace-period | |
1870 | * kthread or it awakens itself for the next round of quiescent-state | |
1871 | * forcing, that kthread will clean up after the just-completed grace | |
1872 | * period. Note that the caller must hold rnp->lock, which is released | |
1873 | * before return. | |
f41d911f | 1874 | */ |
aff4e9ed | 1875 | static void rcu_report_qs_rsp(unsigned long flags) |
336a4f6c | 1876 | __releases(rcu_get_root()->lock) |
f41d911f | 1877 | { |
336a4f6c | 1878 | raw_lockdep_assert_held_rcu_node(rcu_get_root()); |
de8e8730 | 1879 | WARN_ON_ONCE(!rcu_gp_in_progress()); |
9cbc5b97 PM |
1880 | WRITE_ONCE(rcu_state.gp_flags, |
1881 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
336a4f6c | 1882 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(), flags); |
532c00c9 | 1883 | rcu_gp_kthread_wake(); |
f41d911f PM |
1884 | } |
1885 | ||
64db4cff | 1886 | /* |
d3f6bad3 PM |
1887 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
1888 | * Allows quiescent states for a group of CPUs to be reported at one go | |
1889 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
1890 | * must be represented by the same rcu_node structure (which need not be a |
1891 | * leaf rcu_node structure, though it often will be). The gps parameter | |
1892 | * is the grace-period snapshot, which means that the quiescent states | |
c9a24e2d | 1893 | * are valid only if rnp->gp_seq is equal to gps. That structure's lock |
654e9533 | 1894 | * must be held upon entry, and it is released before return. |
ec2c2976 PM |
1895 | * |
1896 | * As a special case, if mask is zero, the bit-already-cleared check is | |
1897 | * disabled. This allows propagating quiescent state due to resumed tasks | |
1898 | * during grace-period initialization. | |
64db4cff | 1899 | */ |
b50912d0 PM |
1900 | static void rcu_report_qs_rnp(unsigned long mask, struct rcu_node *rnp, |
1901 | unsigned long gps, unsigned long flags) | |
64db4cff PM |
1902 | __releases(rnp->lock) |
1903 | { | |
654e9533 | 1904 | unsigned long oldmask = 0; |
28ecd580 PM |
1905 | struct rcu_node *rnp_c; |
1906 | ||
a32e01ee | 1907 | raw_lockdep_assert_held_rcu_node(rnp); |
c0b334c5 | 1908 | |
64db4cff PM |
1909 | /* Walk up the rcu_node hierarchy. */ |
1910 | for (;;) { | |
ec2c2976 | 1911 | if ((!(rnp->qsmask & mask) && mask) || rnp->gp_seq != gps) { |
64db4cff | 1912 | |
654e9533 PM |
1913 | /* |
1914 | * Our bit has already been cleared, or the | |
1915 | * relevant grace period is already over, so done. | |
1916 | */ | |
67c583a7 | 1917 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1918 | return; |
1919 | } | |
654e9533 | 1920 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
5b4c11d5 | 1921 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp) && |
2dee9404 | 1922 | rcu_preempt_blocked_readers_cgp(rnp)); |
7672d647 | 1923 | WRITE_ONCE(rnp->qsmask, rnp->qsmask & ~mask); |
67a0edbf | 1924 | trace_rcu_quiescent_state_report(rcu_state.name, rnp->gp_seq, |
d4c08f2a PM |
1925 | mask, rnp->qsmask, rnp->level, |
1926 | rnp->grplo, rnp->grphi, | |
1927 | !!rnp->gp_tasks); | |
27f4d280 | 1928 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
1929 | |
1930 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 1931 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1932 | return; |
1933 | } | |
d43a5d32 | 1934 | rnp->completedqs = rnp->gp_seq; |
64db4cff PM |
1935 | mask = rnp->grpmask; |
1936 | if (rnp->parent == NULL) { | |
1937 | ||
1938 | /* No more levels. Exit loop holding root lock. */ | |
1939 | ||
1940 | break; | |
1941 | } | |
67c583a7 | 1942 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 1943 | rnp_c = rnp; |
64db4cff | 1944 | rnp = rnp->parent; |
2a67e741 | 1945 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
0937d045 | 1946 | oldmask = READ_ONCE(rnp_c->qsmask); |
64db4cff PM |
1947 | } |
1948 | ||
1949 | /* | |
1950 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 1951 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 1952 | * to clean up and start the next grace period if one is needed. |
64db4cff | 1953 | */ |
aff4e9ed | 1954 | rcu_report_qs_rsp(flags); /* releases rnp->lock. */ |
64db4cff PM |
1955 | } |
1956 | ||
cc99a310 PM |
1957 | /* |
1958 | * Record a quiescent state for all tasks that were previously queued | |
1959 | * on the specified rcu_node structure and that were blocking the current | |
49918a54 | 1960 | * RCU grace period. The caller must hold the corresponding rnp->lock with |
cc99a310 PM |
1961 | * irqs disabled, and this lock is released upon return, but irqs remain |
1962 | * disabled. | |
1963 | */ | |
17a8212b | 1964 | static void __maybe_unused |
139ad4da | 1965 | rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
cc99a310 PM |
1966 | __releases(rnp->lock) |
1967 | { | |
654e9533 | 1968 | unsigned long gps; |
cc99a310 PM |
1969 | unsigned long mask; |
1970 | struct rcu_node *rnp_p; | |
1971 | ||
a32e01ee | 1972 | raw_lockdep_assert_held_rcu_node(rnp); |
c130d2dc | 1973 | if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPT_RCU)) || |
c74859d1 PM |
1974 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)) || |
1975 | rnp->qsmask != 0) { | |
67c583a7 | 1976 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
1977 | return; /* Still need more quiescent states! */ |
1978 | } | |
1979 | ||
77cfc7bf | 1980 | rnp->completedqs = rnp->gp_seq; |
cc99a310 PM |
1981 | rnp_p = rnp->parent; |
1982 | if (rnp_p == NULL) { | |
1983 | /* | |
a77da14c PM |
1984 | * Only one rcu_node structure in the tree, so don't |
1985 | * try to report up to its nonexistent parent! | |
cc99a310 | 1986 | */ |
aff4e9ed | 1987 | rcu_report_qs_rsp(flags); |
cc99a310 PM |
1988 | return; |
1989 | } | |
1990 | ||
c9a24e2d PM |
1991 | /* Report up the rest of the hierarchy, tracking current ->gp_seq. */ |
1992 | gps = rnp->gp_seq; | |
cc99a310 | 1993 | mask = rnp->grpmask; |
67c583a7 | 1994 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 1995 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
b50912d0 | 1996 | rcu_report_qs_rnp(mask, rnp_p, gps, flags); |
cc99a310 PM |
1997 | } |
1998 | ||
64db4cff | 1999 | /* |
d3f6bad3 | 2000 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 2001 | * structure. This must be called from the specified CPU. |
64db4cff PM |
2002 | */ |
2003 | static void | |
cfeac397 | 2004 | rcu_report_qs_rdp(struct rcu_data *rdp) |
64db4cff PM |
2005 | { |
2006 | unsigned long flags; | |
2007 | unsigned long mask; | |
5d6742b3 | 2008 | bool needwake = false; |
b3bb02fe | 2009 | bool needacc = false; |
64db4cff PM |
2010 | struct rcu_node *rnp; |
2011 | ||
cfeac397 | 2012 | WARN_ON_ONCE(rdp->cpu != smp_processor_id()); |
64db4cff | 2013 | rnp = rdp->mynode; |
2a67e741 | 2014 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
c9a24e2d PM |
2015 | if (rdp->cpu_no_qs.b.norm || rdp->gp_seq != rnp->gp_seq || |
2016 | rdp->gpwrap) { | |
64db4cff PM |
2017 | |
2018 | /* | |
e4cc1f22 PM |
2019 | * The grace period in which this quiescent state was |
2020 | * recorded has ended, so don't report it upwards. | |
2021 | * We will instead need a new quiescent state that lies | |
2022 | * within the current grace period. | |
64db4cff | 2023 | */ |
5b74c458 | 2024 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
67c583a7 | 2025 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2026 | return; |
2027 | } | |
2028 | mask = rdp->grpmask; | |
cfeac397 | 2029 | rdp->core_needs_qs = false; |
64db4cff | 2030 | if ((rnp->qsmask & mask) == 0) { |
67c583a7 | 2031 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2032 | } else { |
64db4cff PM |
2033 | /* |
2034 | * This GP can't end until cpu checks in, so all of our | |
2035 | * callbacks can be processed during the next GP. | |
24ee940d | 2036 | * |
b3bb02fe | 2037 | * NOCB kthreads have their own way to deal with that... |
64db4cff | 2038 | */ |
b3bb02fe | 2039 | if (!rcu_rdp_is_offloaded(rdp)) { |
5d6742b3 | 2040 | needwake = rcu_accelerate_cbs(rnp, rdp); |
b3bb02fe FW |
2041 | } else if (!rcu_segcblist_completely_offloaded(&rdp->cblist)) { |
2042 | /* | |
2043 | * ...but NOCB kthreads may miss or delay callbacks acceleration | |
2044 | * if in the middle of a (de-)offloading process. | |
2045 | */ | |
2046 | needacc = true; | |
2047 | } | |
64db4cff | 2048 | |
516e5ae0 | 2049 | rcu_disable_urgency_upon_qs(rdp); |
b50912d0 | 2050 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
654e9533 | 2051 | /* ^^^ Released rnp->lock */ |
48a7639c | 2052 | if (needwake) |
532c00c9 | 2053 | rcu_gp_kthread_wake(); |
b3bb02fe FW |
2054 | |
2055 | if (needacc) { | |
2056 | rcu_nocb_lock_irqsave(rdp, flags); | |
2057 | rcu_accelerate_cbs_unlocked(rnp, rdp); | |
2058 | rcu_nocb_unlock_irqrestore(rdp, flags); | |
2059 | } | |
64db4cff PM |
2060 | } |
2061 | } | |
2062 | ||
2063 | /* | |
2064 | * Check to see if there is a new grace period of which this CPU | |
2065 | * is not yet aware, and if so, set up local rcu_data state for it. | |
2066 | * Otherwise, see if this CPU has just passed through its first | |
2067 | * quiescent state for this grace period, and record that fact if so. | |
2068 | */ | |
2069 | static void | |
8087d3e3 | 2070 | rcu_check_quiescent_state(struct rcu_data *rdp) |
64db4cff | 2071 | { |
05eb552b | 2072 | /* Check for grace-period ends and beginnings. */ |
15cabdff | 2073 | note_gp_changes(rdp); |
64db4cff PM |
2074 | |
2075 | /* | |
2076 | * Does this CPU still need to do its part for current grace period? | |
2077 | * If no, return and let the other CPUs do their part as well. | |
2078 | */ | |
97c668b8 | 2079 | if (!rdp->core_needs_qs) |
64db4cff PM |
2080 | return; |
2081 | ||
2082 | /* | |
2083 | * Was there a quiescent state since the beginning of the grace | |
2084 | * period? If no, then exit and wait for the next call. | |
2085 | */ | |
3a19b46a | 2086 | if (rdp->cpu_no_qs.b.norm) |
64db4cff PM |
2087 | return; |
2088 | ||
d3f6bad3 PM |
2089 | /* |
2090 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
2091 | * judge of that). | |
2092 | */ | |
cfeac397 | 2093 | rcu_report_qs_rdp(rdp); |
64db4cff PM |
2094 | } |
2095 | ||
b1420f1c | 2096 | /* |
780cd590 PM |
2097 | * Near the end of the offline process. Trace the fact that this CPU |
2098 | * is going offline. | |
b1420f1c | 2099 | */ |
780cd590 | 2100 | int rcutree_dying_cpu(unsigned int cpu) |
b1420f1c | 2101 | { |
4f5fbd78 | 2102 | bool blkd; |
4aa846f9 | 2103 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
4f5fbd78 | 2104 | struct rcu_node *rnp = rdp->mynode; |
b1420f1c | 2105 | |
ea46351c | 2106 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 2107 | return 0; |
ea46351c | 2108 | |
4f5fbd78 | 2109 | blkd = !!(rnp->qsmask & rdp->grpmask); |
0937d045 | 2110 | trace_rcu_grace_period(rcu_state.name, READ_ONCE(rnp->gp_seq), |
47fcbc8d | 2111 | blkd ? TPS("cpuofl-bgp") : TPS("cpuofl")); |
780cd590 | 2112 | return 0; |
64db4cff PM |
2113 | } |
2114 | ||
8af3a5e7 PM |
2115 | /* |
2116 | * All CPUs for the specified rcu_node structure have gone offline, | |
2117 | * and all tasks that were preempted within an RCU read-side critical | |
2118 | * section while running on one of those CPUs have since exited their RCU | |
2119 | * read-side critical section. Some other CPU is reporting this fact with | |
2120 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2121 | * This function therefore goes up the tree of rcu_node structures, | |
2122 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2123 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
c50cbe53 | 2124 | * updated. |
8af3a5e7 PM |
2125 | * |
2126 | * This function does check that the specified rcu_node structure has | |
2127 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2128 | * prematurely. That said, invoking it after the fact will cost you | |
2129 | * a needless lock acquisition. So once it has done its work, don't | |
2130 | * invoke it again. | |
2131 | */ | |
2132 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2133 | { | |
2134 | long mask; | |
2135 | struct rcu_node *rnp = rnp_leaf; | |
2136 | ||
962aff03 | 2137 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
ea46351c | 2138 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
962aff03 PM |
2139 | WARN_ON_ONCE(rnp_leaf->qsmaskinit) || |
2140 | WARN_ON_ONCE(rcu_preempt_has_tasks(rnp_leaf))) | |
8af3a5e7 PM |
2141 | return; |
2142 | for (;;) { | |
2143 | mask = rnp->grpmask; | |
2144 | rnp = rnp->parent; | |
2145 | if (!rnp) | |
2146 | break; | |
2a67e741 | 2147 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2148 | rnp->qsmaskinit &= ~mask; |
962aff03 PM |
2149 | /* Between grace periods, so better already be zero! */ |
2150 | WARN_ON_ONCE(rnp->qsmask); | |
8af3a5e7 | 2151 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2152 | raw_spin_unlock_rcu_node(rnp); |
2153 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2154 | return; |
2155 | } | |
67c583a7 | 2156 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2157 | } |
2158 | } | |
2159 | ||
64db4cff | 2160 | /* |
e5601400 | 2161 | * The CPU has been completely removed, and some other CPU is reporting |
a58163d8 PM |
2162 | * this fact from process context. Do the remainder of the cleanup. |
2163 | * There can only be one CPU hotplug operation at a time, so no need for | |
2164 | * explicit locking. | |
64db4cff | 2165 | */ |
780cd590 | 2166 | int rcutree_dead_cpu(unsigned int cpu) |
64db4cff | 2167 | { |
da1df50d | 2168 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
b1420f1c | 2169 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2170 | |
ea46351c | 2171 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
780cd590 | 2172 | return 0; |
ea46351c | 2173 | |
ed73860c | 2174 | WRITE_ONCE(rcu_state.n_online_cpus, rcu_state.n_online_cpus - 1); |
2036d94a | 2175 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2176 | rcu_boost_kthread_setaffinity(rnp, -1); |
96926686 PM |
2177 | // Stop-machine done, so allow nohz_full to disable tick. |
2178 | tick_dep_clear(TICK_DEP_BIT_RCU); | |
780cd590 | 2179 | return 0; |
64db4cff PM |
2180 | } |
2181 | ||
64db4cff PM |
2182 | /* |
2183 | * Invoke any RCU callbacks that have made it to the end of their grace | |
a616aec9 | 2184 | * period. Throttle as specified by rdp->blimit. |
64db4cff | 2185 | */ |
5bb5d09c | 2186 | static void rcu_do_batch(struct rcu_data *rdp) |
64db4cff | 2187 | { |
b5374b2d | 2188 | int div; |
b4e6039e | 2189 | bool __maybe_unused empty; |
64db4cff | 2190 | unsigned long flags; |
15fecf89 PM |
2191 | struct rcu_head *rhp; |
2192 | struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl); | |
6bc33582 | 2193 | long bl, count = 0; |
cfcdef5e | 2194 | long pending, tlimit = 0; |
64db4cff | 2195 | |
dc35c893 | 2196 | /* If no callbacks are ready, just return. */ |
15fecf89 | 2197 | if (!rcu_segcblist_ready_cbs(&rdp->cblist)) { |
3c779dfe | 2198 | trace_rcu_batch_start(rcu_state.name, |
15fecf89 | 2199 | rcu_segcblist_n_cbs(&rdp->cblist), 0); |
3c779dfe | 2200 | trace_rcu_batch_end(rcu_state.name, 0, |
15fecf89 | 2201 | !rcu_segcblist_empty(&rdp->cblist), |
4968c300 | 2202 | need_resched(), is_idle_task(current), |
51038506 | 2203 | rcu_is_callbacks_kthread(rdp)); |
64db4cff | 2204 | return; |
29c00b4a | 2205 | } |
64db4cff PM |
2206 | |
2207 | /* | |
7b65dfa3 | 2208 | * Extract the list of ready callbacks, disabling IRQs to prevent |
15fecf89 PM |
2209 | * races with call_rcu() from interrupt handlers. Leave the |
2210 | * callback counts, as rcu_barrier() needs to be conservative. | |
64db4cff | 2211 | */ |
7b65dfa3 | 2212 | rcu_nocb_lock_irqsave(rdp, flags); |
8146c4e2 | 2213 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
cfcdef5e | 2214 | pending = rcu_segcblist_n_cbs(&rdp->cblist); |
b5374b2d PM |
2215 | div = READ_ONCE(rcu_divisor); |
2216 | div = div < 0 ? 7 : div > sizeof(long) * 8 - 2 ? sizeof(long) * 8 - 2 : div; | |
2217 | bl = max(rdp->blimit, pending >> div); | |
a554ba28 | 2218 | if (in_serving_softirq() && unlikely(bl > 100)) { |
a2b354b9 PM |
2219 | long rrn = READ_ONCE(rcu_resched_ns); |
2220 | ||
2221 | rrn = rrn < NSEC_PER_MSEC ? NSEC_PER_MSEC : rrn > NSEC_PER_SEC ? NSEC_PER_SEC : rrn; | |
2222 | tlimit = local_clock() + rrn; | |
2223 | } | |
3c779dfe | 2224 | trace_rcu_batch_start(rcu_state.name, |
15fecf89 PM |
2225 | rcu_segcblist_n_cbs(&rdp->cblist), bl); |
2226 | rcu_segcblist_extract_done_cbs(&rdp->cblist, &rcl); | |
344e219d | 2227 | if (rcu_rdp_is_offloaded(rdp)) |
7f36ef82 | 2228 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); |
3afe7fa5 JFG |
2229 | |
2230 | trace_rcu_segcb_stats(&rdp->cblist, TPS("SegCbDequeued")); | |
5d6742b3 | 2231 | rcu_nocb_unlock_irqrestore(rdp, flags); |
64db4cff PM |
2232 | |
2233 | /* Invoke callbacks. */ | |
6a949b7a | 2234 | tick_dep_set_task(current, TICK_DEP_BIT_RCU); |
15fecf89 | 2235 | rhp = rcu_cblist_dequeue(&rcl); |
3afe7fa5 | 2236 | |
15fecf89 | 2237 | for (; rhp; rhp = rcu_cblist_dequeue(&rcl)) { |
77a40f97 JFG |
2238 | rcu_callback_t f; |
2239 | ||
6bc33582 | 2240 | count++; |
15fecf89 | 2241 | debug_rcu_head_unqueue(rhp); |
77a40f97 JFG |
2242 | |
2243 | rcu_lock_acquire(&rcu_callback_map); | |
2244 | trace_rcu_invoke_callback(rcu_state.name, rhp); | |
2245 | ||
2246 | f = rhp->func; | |
2247 | WRITE_ONCE(rhp->func, (rcu_callback_t)0L); | |
2248 | f(rhp); | |
2249 | ||
2250 | rcu_lock_release(&rcu_callback_map); | |
2251 | ||
15fecf89 PM |
2252 | /* |
2253 | * Stop only if limit reached and CPU has something to do. | |
15fecf89 | 2254 | */ |
3e61e95e FW |
2255 | if (in_serving_softirq()) { |
2256 | if (count >= bl && (need_resched() || !is_idle_task(current))) | |
2257 | break; | |
a554ba28 FW |
2258 | /* |
2259 | * Make sure we don't spend too much time here and deprive other | |
2260 | * softirq vectors of CPU cycles. | |
2261 | */ | |
2262 | if (unlikely(tlimit)) { | |
2263 | /* only call local_clock() every 32 callbacks */ | |
2264 | if (likely((count & 31) || local_clock() < tlimit)) | |
2265 | continue; | |
2266 | /* Exceeded the time limit, so leave. */ | |
2267 | break; | |
2268 | } | |
3e61e95e | 2269 | } else { |
5d6742b3 PM |
2270 | local_bh_enable(); |
2271 | lockdep_assert_irqs_enabled(); | |
2272 | cond_resched_tasks_rcu_qs(); | |
2273 | lockdep_assert_irqs_enabled(); | |
2274 | local_bh_disable(); | |
2275 | } | |
64db4cff PM |
2276 | } |
2277 | ||
7b65dfa3 | 2278 | rcu_nocb_lock_irqsave(rdp, flags); |
e816d56f | 2279 | rdp->n_cbs_invoked += count; |
3c779dfe | 2280 | trace_rcu_batch_end(rcu_state.name, count, !!rcl.head, need_resched(), |
51038506 | 2281 | is_idle_task(current), rcu_is_callbacks_kthread(rdp)); |
64db4cff | 2282 | |
15fecf89 PM |
2283 | /* Update counts and requeue any remaining callbacks. */ |
2284 | rcu_segcblist_insert_done_cbs(&rdp->cblist, &rcl); | |
6bc33582 | 2285 | rcu_segcblist_add_len(&rdp->cblist, -count); |
64db4cff PM |
2286 | |
2287 | /* Reinstate batch limit if we have worked down the excess. */ | |
15fecf89 | 2288 | count = rcu_segcblist_n_cbs(&rdp->cblist); |
d5a9a8c3 | 2289 | if (rdp->blimit >= DEFAULT_MAX_RCU_BLIMIT && count <= qlowmark) |
64db4cff PM |
2290 | rdp->blimit = blimit; |
2291 | ||
37c72e56 | 2292 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
15fecf89 | 2293 | if (count == 0 && rdp->qlen_last_fqs_check != 0) { |
37c72e56 | 2294 | rdp->qlen_last_fqs_check = 0; |
2431774f | 2295 | rdp->n_force_qs_snap = READ_ONCE(rcu_state.n_force_qs); |
15fecf89 PM |
2296 | } else if (count < rdp->qlen_last_fqs_check - qhimark) |
2297 | rdp->qlen_last_fqs_check = count; | |
efd88b02 PM |
2298 | |
2299 | /* | |
2300 | * The following usually indicates a double call_rcu(). To track | |
2301 | * this down, try building with CONFIG_DEBUG_OBJECTS_RCU_HEAD=y. | |
2302 | */ | |
b4e6039e JFG |
2303 | empty = rcu_segcblist_empty(&rdp->cblist); |
2304 | WARN_ON_ONCE(count == 0 && !empty); | |
d1b222c6 | 2305 | WARN_ON_ONCE(!IS_ENABLED(CONFIG_RCU_NOCB_CPU) && |
b4e6039e JFG |
2306 | count != 0 && empty); |
2307 | WARN_ON_ONCE(count == 0 && rcu_segcblist_n_segment_cbs(&rdp->cblist) != 0); | |
2308 | WARN_ON_ONCE(!empty && rcu_segcblist_n_segment_cbs(&rdp->cblist) == 0); | |
37c72e56 | 2309 | |
5d6742b3 | 2310 | rcu_nocb_unlock_irqrestore(rdp, flags); |
64db4cff | 2311 | |
6a949b7a | 2312 | tick_dep_clear_task(current, TICK_DEP_BIT_RCU); |
64db4cff PM |
2313 | } |
2314 | ||
2315 | /* | |
c98cac60 PM |
2316 | * This function is invoked from each scheduling-clock interrupt, |
2317 | * and checks to see if this CPU is in a non-context-switch quiescent | |
2318 | * state, for example, user mode or idle loop. It also schedules RCU | |
2319 | * core processing. If the current grace period has gone on too long, | |
2320 | * it will ask the scheduler to manufacture a context switch for the sole | |
277ffe1b | 2321 | * purpose of providing the needed quiescent state. |
64db4cff | 2322 | */ |
c98cac60 | 2323 | void rcu_sched_clock_irq(int user) |
64db4cff | 2324 | { |
c708b08c PM |
2325 | unsigned long j; |
2326 | ||
2327 | if (IS_ENABLED(CONFIG_PROVE_RCU)) { | |
2328 | j = jiffies; | |
2329 | WARN_ON_ONCE(time_before(j, __this_cpu_read(rcu_data.last_sched_clock))); | |
2330 | __this_cpu_write(rcu_data.last_sched_clock, j); | |
2331 | } | |
f7f7bac9 | 2332 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
a649d25d | 2333 | lockdep_assert_irqs_disabled(); |
4e95020c | 2334 | raw_cpu_inc(rcu_data.ticks_this_gp); |
92aa39e9 | 2335 | /* The load-acquire pairs with the store-release setting to true. */ |
2dba13f0 | 2336 | if (smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) { |
92aa39e9 | 2337 | /* Idle and userspace execution already are quiescent states. */ |
a0ef9ec2 | 2338 | if (!rcu_is_cpu_rrupt_from_idle() && !user) { |
92aa39e9 PM |
2339 | set_tsk_need_resched(current); |
2340 | set_preempt_need_resched(); | |
2341 | } | |
2dba13f0 | 2342 | __this_cpu_write(rcu_data.rcu_urgent_qs, false); |
64db4cff | 2343 | } |
c98cac60 | 2344 | rcu_flavor_sched_clock_irq(user); |
dd7dafd1 | 2345 | if (rcu_pending(user)) |
a46e0899 | 2346 | invoke_rcu_core(); |
528262f5 Z |
2347 | if (user || rcu_is_cpu_rrupt_from_idle()) |
2348 | rcu_note_voluntary_context_switch(current); | |
a649d25d | 2349 | lockdep_assert_irqs_disabled(); |
07f27570 | 2350 | |
f7f7bac9 | 2351 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2352 | } |
2353 | ||
64db4cff | 2354 | /* |
5d8a752e ZZ |
2355 | * Scan the leaf rcu_node structures. For each structure on which all |
2356 | * CPUs have reported a quiescent state and on which there are tasks | |
2357 | * blocking the current grace period, initiate RCU priority boosting. | |
2358 | * Otherwise, invoke the specified function to check dyntick state for | |
2359 | * each CPU that has not yet reported a quiescent state. | |
64db4cff | 2360 | */ |
8ff0b907 | 2361 | static void force_qs_rnp(int (*f)(struct rcu_data *rdp)) |
64db4cff | 2362 | { |
64db4cff PM |
2363 | int cpu; |
2364 | unsigned long flags; | |
2365 | unsigned long mask; | |
66e4c33b | 2366 | struct rcu_data *rdp; |
a0b6c9a7 | 2367 | struct rcu_node *rnp; |
64db4cff | 2368 | |
b2b00ddf PM |
2369 | rcu_state.cbovld = rcu_state.cbovldnext; |
2370 | rcu_state.cbovldnext = false; | |
aedf4ba9 | 2371 | rcu_for_each_leaf_node(rnp) { |
cee43939 | 2372 | cond_resched_tasks_rcu_qs(); |
64db4cff | 2373 | mask = 0; |
2a67e741 | 2374 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b2b00ddf | 2375 | rcu_state.cbovldnext |= !!rnp->cbovldmask; |
a0b6c9a7 | 2376 | if (rnp->qsmask == 0) { |
9b1ce0ac | 2377 | if (rcu_preempt_blocked_readers_cgp(rnp)) { |
a77da14c PM |
2378 | /* |
2379 | * No point in scanning bits because they | |
2380 | * are all zero. But we might need to | |
2381 | * priority-boost blocked readers. | |
2382 | */ | |
2383 | rcu_initiate_boost(rnp, flags); | |
2384 | /* rcu_initiate_boost() releases rnp->lock */ | |
2385 | continue; | |
2386 | } | |
92816435 PM |
2387 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
2388 | continue; | |
64db4cff | 2389 | } |
7441e766 PM |
2390 | for_each_leaf_node_cpu_mask(rnp, cpu, rnp->qsmask) { |
2391 | rdp = per_cpu_ptr(&rcu_data, cpu); | |
2392 | if (f(rdp)) { | |
2393 | mask |= rdp->grpmask; | |
2394 | rcu_disable_urgency_upon_qs(rdp); | |
0edd1b17 | 2395 | } |
64db4cff | 2396 | } |
45f014c5 | 2397 | if (mask != 0) { |
c9a24e2d | 2398 | /* Idle/offline CPUs, report (releases rnp->lock). */ |
b50912d0 | 2399 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
0aa04b05 PM |
2400 | } else { |
2401 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 2402 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2403 | } |
64db4cff | 2404 | } |
64db4cff PM |
2405 | } |
2406 | ||
2407 | /* | |
2408 | * Force quiescent states on reluctant CPUs, and also detect which | |
2409 | * CPUs are in dyntick-idle mode. | |
2410 | */ | |
cd920e5a | 2411 | void rcu_force_quiescent_state(void) |
64db4cff PM |
2412 | { |
2413 | unsigned long flags; | |
394f2769 PM |
2414 | bool ret; |
2415 | struct rcu_node *rnp; | |
2416 | struct rcu_node *rnp_old = NULL; | |
2417 | ||
2418 | /* Funnel through hierarchy to reduce memory contention. */ | |
da1df50d | 2419 | rnp = __this_cpu_read(rcu_data.mynode); |
394f2769 | 2420 | for (; rnp != NULL; rnp = rnp->parent) { |
67a0edbf | 2421 | ret = (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) || |
66e4c33b | 2422 | !raw_spin_trylock(&rnp->fqslock); |
394f2769 PM |
2423 | if (rnp_old != NULL) |
2424 | raw_spin_unlock(&rnp_old->fqslock); | |
d62df573 | 2425 | if (ret) |
394f2769 | 2426 | return; |
394f2769 PM |
2427 | rnp_old = rnp; |
2428 | } | |
336a4f6c | 2429 | /* rnp_old == rcu_get_root(), rnp == NULL. */ |
64db4cff | 2430 | |
394f2769 | 2431 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 2432 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 2433 | raw_spin_unlock(&rnp_old->fqslock); |
67a0edbf | 2434 | if (READ_ONCE(rcu_state.gp_flags) & RCU_GP_FLAG_FQS) { |
67c583a7 | 2435 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 2436 | return; /* Someone beat us to it. */ |
46a1e34e | 2437 | } |
67a0edbf PM |
2438 | WRITE_ONCE(rcu_state.gp_flags, |
2439 | READ_ONCE(rcu_state.gp_flags) | RCU_GP_FLAG_FQS); | |
67c583a7 | 2440 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
532c00c9 | 2441 | rcu_gp_kthread_wake(); |
64db4cff | 2442 | } |
cd920e5a | 2443 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); |
64db4cff | 2444 | |
a657f261 PM |
2445 | // Workqueue handler for an RCU reader for kernels enforcing struct RCU |
2446 | // grace periods. | |
2447 | static void strict_work_handler(struct work_struct *work) | |
2448 | { | |
2449 | rcu_read_lock(); | |
2450 | rcu_read_unlock(); | |
2451 | } | |
2452 | ||
fb60e533 | 2453 | /* Perform RCU core processing work for the current CPU. */ |
48d07c04 | 2454 | static __latent_entropy void rcu_core(void) |
64db4cff PM |
2455 | { |
2456 | unsigned long flags; | |
da1df50d | 2457 | struct rcu_data *rdp = raw_cpu_ptr(&rcu_data); |
26d950a9 | 2458 | struct rcu_node *rnp = rdp->mynode; |
fbb94cbd FW |
2459 | /* |
2460 | * On RT rcu_core() can be preempted when IRQs aren't disabled. | |
2461 | * Therefore this function can race with concurrent NOCB (de-)offloading | |
2462 | * on this CPU and the below condition must be considered volatile. | |
2463 | * However if we race with: | |
2464 | * | |
2465 | * _ Offloading: In the worst case we accelerate or process callbacks | |
2466 | * concurrently with NOCB kthreads. We are guaranteed to | |
2467 | * call rcu_nocb_lock() if that happens. | |
2468 | * | |
2469 | * _ Deoffloading: In the worst case we miss callbacks acceleration or | |
2470 | * processing. This is fine because the early stage | |
2471 | * of deoffloading invokes rcu_core() after setting | |
2472 | * SEGCBLIST_RCU_CORE. So we guarantee that we'll process | |
2473 | * what could have been dismissed without the need to wait | |
2474 | * for the next rcu_pending() check in the next jiffy. | |
2475 | */ | |
32aa2f41 | 2476 | const bool do_batch = !rcu_segcblist_completely_offloaded(&rdp->cblist); |
64db4cff | 2477 | |
b049fdf8 PM |
2478 | if (cpu_is_offline(smp_processor_id())) |
2479 | return; | |
2480 | trace_rcu_utilization(TPS("Start RCU core")); | |
50dc7def | 2481 | WARN_ON_ONCE(!rdp->beenonline); |
2e597558 | 2482 | |
3e310098 | 2483 | /* Report any deferred quiescent states if preemption enabled. */ |
790da248 | 2484 | if (IS_ENABLED(CONFIG_PREEMPT_COUNT) && (!(preempt_count() & PREEMPT_MASK))) { |
3e310098 | 2485 | rcu_preempt_deferred_qs(current); |
fced9c8c PM |
2486 | } else if (rcu_preempt_need_deferred_qs(current)) { |
2487 | set_tsk_need_resched(current); | |
2488 | set_preempt_need_resched(); | |
2489 | } | |
3e310098 | 2490 | |
64db4cff | 2491 | /* Update RCU state based on any recent quiescent states. */ |
8087d3e3 | 2492 | rcu_check_quiescent_state(rdp); |
64db4cff | 2493 | |
bd7af846 | 2494 | /* No grace period and unregistered callbacks? */ |
de8e8730 | 2495 | if (!rcu_gp_in_progress() && |
634954c2 FW |
2496 | rcu_segcblist_is_enabled(&rdp->cblist) && do_batch) { |
2497 | rcu_nocb_lock_irqsave(rdp, flags); | |
e44e73ca | 2498 | if (!rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
c6e09b97 | 2499 | rcu_accelerate_cbs_unlocked(rnp, rdp); |
634954c2 | 2500 | rcu_nocb_unlock_irqrestore(rdp, flags); |
64db4cff PM |
2501 | } |
2502 | ||
791416c4 | 2503 | rcu_check_gp_start_stall(rnp, rdp, rcu_jiffies_till_stall_check()); |
26d950a9 | 2504 | |
64db4cff | 2505 | /* If there are callbacks ready, invoke them. */ |
32aa2f41 | 2506 | if (do_batch && rcu_segcblist_ready_cbs(&rdp->cblist) && |
0598a4d4 | 2507 | likely(READ_ONCE(rcu_scheduler_fully_active))) { |
43e903ad | 2508 | rcu_do_batch(rdp); |
0598a4d4 FW |
2509 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
2510 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) | |
2511 | invoke_rcu_core(); | |
2512 | } | |
96d3fd0d PM |
2513 | |
2514 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
2515 | do_nocb_deferred_wakeup(rdp); | |
f7f7bac9 | 2516 | trace_rcu_utilization(TPS("End RCU core")); |
a657f261 PM |
2517 | |
2518 | // If strict GPs, schedule an RCU reader in a clean environment. | |
2519 | if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) | |
2520 | queue_work_on(rdp->cpu, rcu_gp_wq, &rdp->strict_work); | |
64db4cff PM |
2521 | } |
2522 | ||
48d07c04 SAS |
2523 | static void rcu_core_si(struct softirq_action *h) |
2524 | { | |
2525 | rcu_core(); | |
2526 | } | |
2527 | ||
2528 | static void rcu_wake_cond(struct task_struct *t, int status) | |
2529 | { | |
2530 | /* | |
2531 | * If the thread is yielding, only wake it when this | |
2532 | * is invoked from idle | |
2533 | */ | |
2534 | if (t && (status != RCU_KTHREAD_YIELDING || is_idle_task(current))) | |
2535 | wake_up_process(t); | |
2536 | } | |
2537 | ||
2538 | static void invoke_rcu_core_kthread(void) | |
2539 | { | |
2540 | struct task_struct *t; | |
2541 | unsigned long flags; | |
2542 | ||
2543 | local_irq_save(flags); | |
2544 | __this_cpu_write(rcu_data.rcu_cpu_has_work, 1); | |
2545 | t = __this_cpu_read(rcu_data.rcu_cpu_kthread_task); | |
2546 | if (t != NULL && t != current) | |
2547 | rcu_wake_cond(t, __this_cpu_read(rcu_data.rcu_cpu_kthread_status)); | |
2548 | local_irq_restore(flags); | |
2549 | } | |
2550 | ||
48d07c04 SAS |
2551 | /* |
2552 | * Wake up this CPU's rcuc kthread to do RCU core processing. | |
2553 | */ | |
a46e0899 | 2554 | static void invoke_rcu_core(void) |
09223371 | 2555 | { |
48d07c04 SAS |
2556 | if (!cpu_online(smp_processor_id())) |
2557 | return; | |
2558 | if (use_softirq) | |
b0f74036 | 2559 | raise_softirq(RCU_SOFTIRQ); |
48d07c04 SAS |
2560 | else |
2561 | invoke_rcu_core_kthread(); | |
2562 | } | |
2563 | ||
2564 | static void rcu_cpu_kthread_park(unsigned int cpu) | |
2565 | { | |
2566 | per_cpu(rcu_data.rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; | |
2567 | } | |
2568 | ||
2569 | static int rcu_cpu_kthread_should_run(unsigned int cpu) | |
2570 | { | |
2571 | return __this_cpu_read(rcu_data.rcu_cpu_has_work); | |
2572 | } | |
2573 | ||
2574 | /* | |
2575 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces | |
2576 | * the RCU softirq used in configurations of RCU that do not support RCU | |
2577 | * priority boosting. | |
2578 | */ | |
2579 | static void rcu_cpu_kthread(unsigned int cpu) | |
2580 | { | |
2581 | unsigned int *statusp = this_cpu_ptr(&rcu_data.rcu_cpu_kthread_status); | |
2582 | char work, *workp = this_cpu_ptr(&rcu_data.rcu_cpu_has_work); | |
c9515875 | 2583 | unsigned long *j = this_cpu_ptr(&rcu_data.rcuc_activity); |
48d07c04 SAS |
2584 | int spincnt; |
2585 | ||
2488a5e6 | 2586 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_run")); |
48d07c04 | 2587 | for (spincnt = 0; spincnt < 10; spincnt++) { |
c9515875 | 2588 | WRITE_ONCE(*j, jiffies); |
48d07c04 SAS |
2589 | local_bh_disable(); |
2590 | *statusp = RCU_KTHREAD_RUNNING; | |
2591 | local_irq_disable(); | |
2592 | work = *workp; | |
2593 | *workp = 0; | |
2594 | local_irq_enable(); | |
2595 | if (work) | |
2596 | rcu_core(); | |
2597 | local_bh_enable(); | |
2598 | if (*workp == 0) { | |
2599 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); | |
2600 | *statusp = RCU_KTHREAD_WAITING; | |
2601 | return; | |
2602 | } | |
2603 | } | |
2604 | *statusp = RCU_KTHREAD_YIELDING; | |
2605 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); | |
77865dea | 2606 | schedule_timeout_idle(2); |
48d07c04 SAS |
2607 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); |
2608 | *statusp = RCU_KTHREAD_WAITING; | |
c9515875 | 2609 | WRITE_ONCE(*j, jiffies); |
48d07c04 SAS |
2610 | } |
2611 | ||
2612 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { | |
2613 | .store = &rcu_data.rcu_cpu_kthread_task, | |
2614 | .thread_should_run = rcu_cpu_kthread_should_run, | |
2615 | .thread_fn = rcu_cpu_kthread, | |
2616 | .thread_comm = "rcuc/%u", | |
2617 | .setup = rcu_cpu_kthread_setup, | |
2618 | .park = rcu_cpu_kthread_park, | |
2619 | }; | |
2620 | ||
2621 | /* | |
2622 | * Spawn per-CPU RCU core processing kthreads. | |
2623 | */ | |
2624 | static int __init rcu_spawn_core_kthreads(void) | |
2625 | { | |
2626 | int cpu; | |
2627 | ||
2628 | for_each_possible_cpu(cpu) | |
2629 | per_cpu(rcu_data.rcu_cpu_has_work, cpu) = 0; | |
4b4399b2 | 2630 | if (use_softirq) |
48d07c04 SAS |
2631 | return 0; |
2632 | WARN_ONCE(smpboot_register_percpu_thread(&rcu_cpu_thread_spec), | |
2633 | "%s: Could not start rcuc kthread, OOM is now expected behavior\n", __func__); | |
2634 | return 0; | |
09223371 SL |
2635 | } |
2636 | ||
29154c57 PM |
2637 | /* |
2638 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
2639 | */ | |
5c7d8967 PM |
2640 | static void __call_rcu_core(struct rcu_data *rdp, struct rcu_head *head, |
2641 | unsigned long flags) | |
64db4cff | 2642 | { |
62fde6ed PM |
2643 | /* |
2644 | * If called from an extended quiescent state, invoke the RCU | |
2645 | * core in order to force a re-evaluation of RCU's idleness. | |
2646 | */ | |
9910affa | 2647 | if (!rcu_is_watching()) |
62fde6ed PM |
2648 | invoke_rcu_core(); |
2649 | ||
a16b7a69 | 2650 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 2651 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 2652 | return; |
64db4cff | 2653 | |
37c72e56 PM |
2654 | /* |
2655 | * Force the grace period if too many callbacks or too long waiting. | |
cd920e5a | 2656 | * Enforce hysteresis, and don't invoke rcu_force_quiescent_state() |
37c72e56 | 2657 | * if some other CPU has recently done so. Also, don't bother |
cd920e5a | 2658 | * invoking rcu_force_quiescent_state() if the newly enqueued callback |
37c72e56 PM |
2659 | * is the only one waiting for a grace period to complete. |
2660 | */ | |
15fecf89 PM |
2661 | if (unlikely(rcu_segcblist_n_cbs(&rdp->cblist) > |
2662 | rdp->qlen_last_fqs_check + qhimark)) { | |
b52573d2 PM |
2663 | |
2664 | /* Are we ignoring a completed grace period? */ | |
15cabdff | 2665 | note_gp_changes(rdp); |
b52573d2 PM |
2666 | |
2667 | /* Start a new grace period if one not already started. */ | |
de8e8730 | 2668 | if (!rcu_gp_in_progress()) { |
c6e09b97 | 2669 | rcu_accelerate_cbs_unlocked(rdp->mynode, rdp); |
b52573d2 PM |
2670 | } else { |
2671 | /* Give the grace period a kick. */ | |
d5a9a8c3 | 2672 | rdp->blimit = DEFAULT_MAX_RCU_BLIMIT; |
2431774f | 2673 | if (READ_ONCE(rcu_state.n_force_qs) == rdp->n_force_qs_snap && |
15fecf89 | 2674 | rcu_segcblist_first_pend_cb(&rdp->cblist) != head) |
cd920e5a | 2675 | rcu_force_quiescent_state(); |
2431774f | 2676 | rdp->n_force_qs_snap = READ_ONCE(rcu_state.n_force_qs); |
15fecf89 | 2677 | rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist); |
b52573d2 | 2678 | } |
4cdfc175 | 2679 | } |
29154c57 PM |
2680 | } |
2681 | ||
ae150184 PM |
2682 | /* |
2683 | * RCU callback function to leak a callback. | |
2684 | */ | |
2685 | static void rcu_leak_callback(struct rcu_head *rhp) | |
2686 | { | |
2687 | } | |
2688 | ||
3fbfbf7a | 2689 | /* |
b2b00ddf PM |
2690 | * Check and if necessary update the leaf rcu_node structure's |
2691 | * ->cbovldmask bit corresponding to the current CPU based on that CPU's | |
2692 | * number of queued RCU callbacks. The caller must hold the leaf rcu_node | |
2693 | * structure's ->lock. | |
3fbfbf7a | 2694 | */ |
b2b00ddf PM |
2695 | static void check_cb_ovld_locked(struct rcu_data *rdp, struct rcu_node *rnp) |
2696 | { | |
2697 | raw_lockdep_assert_held_rcu_node(rnp); | |
2698 | if (qovld_calc <= 0) | |
2699 | return; // Early boot and wildcard value set. | |
2700 | if (rcu_segcblist_n_cbs(&rdp->cblist) >= qovld_calc) | |
2701 | WRITE_ONCE(rnp->cbovldmask, rnp->cbovldmask | rdp->grpmask); | |
2702 | else | |
2703 | WRITE_ONCE(rnp->cbovldmask, rnp->cbovldmask & ~rdp->grpmask); | |
2704 | } | |
2705 | ||
2706 | /* | |
2707 | * Check and if necessary update the leaf rcu_node structure's | |
2708 | * ->cbovldmask bit corresponding to the current CPU based on that CPU's | |
2709 | * number of queued RCU callbacks. No locks need be held, but the | |
2710 | * caller must have disabled interrupts. | |
2711 | * | |
2712 | * Note that this function ignores the possibility that there are a lot | |
2713 | * of callbacks all of which have already seen the end of their respective | |
2714 | * grace periods. This omission is due to the need for no-CBs CPUs to | |
2715 | * be holding ->nocb_lock to do this check, which is too heavy for a | |
2716 | * common-case operation. | |
3fbfbf7a | 2717 | */ |
b2b00ddf PM |
2718 | static void check_cb_ovld(struct rcu_data *rdp) |
2719 | { | |
2720 | struct rcu_node *const rnp = rdp->mynode; | |
2721 | ||
2722 | if (qovld_calc <= 0 || | |
2723 | ((rcu_segcblist_n_cbs(&rdp->cblist) >= qovld_calc) == | |
2724 | !!(READ_ONCE(rnp->cbovldmask) & rdp->grpmask))) | |
2725 | return; // Early boot wildcard value or already set correctly. | |
2726 | raw_spin_lock_rcu_node(rnp); | |
2727 | check_cb_ovld_locked(rdp, rnp); | |
2728 | raw_spin_unlock_rcu_node(rnp); | |
2729 | } | |
2730 | ||
1fe09ebe PM |
2731 | /** |
2732 | * call_rcu() - Queue an RCU callback for invocation after a grace period. | |
2733 | * @head: structure to be used for queueing the RCU updates. | |
2734 | * @func: actual callback function to be invoked after the grace period | |
2735 | * | |
2736 | * The callback function will be invoked some time after a full grace | |
2737 | * period elapses, in other words after all pre-existing RCU read-side | |
2738 | * critical sections have completed. However, the callback function | |
2739 | * might well execute concurrently with RCU read-side critical sections | |
2740 | * that started after call_rcu() was invoked. | |
2741 | * | |
2742 | * RCU read-side critical sections are delimited by rcu_read_lock() | |
2743 | * and rcu_read_unlock(), and may be nested. In addition, but only in | |
2744 | * v5.0 and later, regions of code across which interrupts, preemption, | |
2745 | * or softirqs have been disabled also serve as RCU read-side critical | |
2746 | * sections. This includes hardware interrupt handlers, softirq handlers, | |
2747 | * and NMI handlers. | |
2748 | * | |
2749 | * Note that all CPUs must agree that the grace period extended beyond | |
2750 | * all pre-existing RCU read-side critical section. On systems with more | |
2751 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
2752 | * guaranteed to have executed a full memory barrier since the end of its | |
2753 | * last RCU read-side critical section whose beginning preceded the call | |
2754 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
2755 | * critical section that continues beyond the start of "func()" must have | |
2756 | * executed a memory barrier after the call_rcu() but before the beginning | |
2757 | * of that RCU read-side critical section. Note that these guarantees | |
2758 | * include CPUs that are offline, idle, or executing in user mode, as | |
2759 | * well as CPUs that are executing in the kernel. | |
2760 | * | |
2761 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
2762 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
2763 | * guaranteed to execute a full memory barrier during the time interval | |
2764 | * between the call to call_rcu() and the invocation of "func()" -- even | |
2765 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
2766 | * more than one CPU). | |
2767 | * | |
2768 | * Implementation of these memory-ordering guarantees is described here: | |
2769 | * Documentation/RCU/Design/Memory-Ordering/Tree-RCU-Memory-Ordering.rst. | |
2770 | */ | |
2771 | void call_rcu(struct rcu_head *head, rcu_callback_t func) | |
64db4cff | 2772 | { |
b4b7914a | 2773 | static atomic_t doublefrees; |
64db4cff PM |
2774 | unsigned long flags; |
2775 | struct rcu_data *rdp; | |
5d6742b3 | 2776 | bool was_alldone; |
64db4cff | 2777 | |
b8f2ed53 PM |
2778 | /* Misaligned rcu_head! */ |
2779 | WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1)); | |
2780 | ||
ae150184 | 2781 | if (debug_rcu_head_queue(head)) { |
fa3c6647 PM |
2782 | /* |
2783 | * Probable double call_rcu(), so leak the callback. | |
2784 | * Use rcu:rcu_callback trace event to find the previous | |
1fe09ebe | 2785 | * time callback was passed to call_rcu(). |
fa3c6647 | 2786 | */ |
b4b7914a PM |
2787 | if (atomic_inc_return(&doublefrees) < 4) { |
2788 | pr_err("%s(): Double-freed CB %p->%pS()!!! ", __func__, head, head->func); | |
2789 | mem_dump_obj(head); | |
2790 | } | |
7d0ae808 | 2791 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
2792 | return; |
2793 | } | |
64db4cff PM |
2794 | head->func = func; |
2795 | head->next = NULL; | |
300c0c5e | 2796 | kasan_record_aux_stack_noalloc(head); |
d818cc76 | 2797 | local_irq_save(flags); |
da1df50d | 2798 | rdp = this_cpu_ptr(&rcu_data); |
64db4cff PM |
2799 | |
2800 | /* Add the callback to our list. */ | |
5d6742b3 PM |
2801 | if (unlikely(!rcu_segcblist_is_enabled(&rdp->cblist))) { |
2802 | // This can trigger due to call_rcu() from offline CPU: | |
2803 | WARN_ON_ONCE(rcu_scheduler_active != RCU_SCHEDULER_INACTIVE); | |
34404ca8 | 2804 | WARN_ON_ONCE(!rcu_is_watching()); |
5d6742b3 PM |
2805 | // Very early boot, before rcu_init(). Initialize if needed |
2806 | // and then drop through to queue the callback. | |
15fecf89 PM |
2807 | if (rcu_segcblist_empty(&rdp->cblist)) |
2808 | rcu_segcblist_init(&rdp->cblist); | |
0d8ee37e | 2809 | } |
77a40f97 | 2810 | |
b2b00ddf | 2811 | check_cb_ovld(rdp); |
d1b222c6 PM |
2812 | if (rcu_nocb_try_bypass(rdp, head, &was_alldone, flags)) |
2813 | return; // Enqueued onto ->nocb_bypass, so just leave. | |
b692dc4a | 2814 | // If no-CBs CPU gets here, rcu_nocb_try_bypass() acquired ->nocb_lock. |
77a40f97 | 2815 | rcu_segcblist_enqueue(&rdp->cblist, head); |
c408b215 URS |
2816 | if (__is_kvfree_rcu_offset((unsigned long)func)) |
2817 | trace_rcu_kvfree_callback(rcu_state.name, head, | |
3c779dfe | 2818 | (unsigned long)func, |
15fecf89 | 2819 | rcu_segcblist_n_cbs(&rdp->cblist)); |
d4c08f2a | 2820 | else |
3c779dfe | 2821 | trace_rcu_callback(rcu_state.name, head, |
15fecf89 | 2822 | rcu_segcblist_n_cbs(&rdp->cblist)); |
d4c08f2a | 2823 | |
3afe7fa5 JFG |
2824 | trace_rcu_segcb_stats(&rdp->cblist, TPS("SegCBQueued")); |
2825 | ||
29154c57 | 2826 | /* Go handle any RCU core processing required. */ |
3820b513 | 2827 | if (unlikely(rcu_rdp_is_offloaded(rdp))) { |
5d6742b3 PM |
2828 | __call_rcu_nocb_wake(rdp, was_alldone, flags); /* unlocks */ |
2829 | } else { | |
2830 | __call_rcu_core(rdp, head, flags); | |
2831 | local_irq_restore(flags); | |
2832 | } | |
64db4cff | 2833 | } |
45975c7d | 2834 | EXPORT_SYMBOL_GPL(call_rcu); |
64db4cff | 2835 | |
a35d1690 BP |
2836 | |
2837 | /* Maximum number of jiffies to wait before draining a batch. */ | |
51824b78 | 2838 | #define KFREE_DRAIN_JIFFIES (5 * HZ) |
0392bebe | 2839 | #define KFREE_N_BATCHES 2 |
5f3c8d62 | 2840 | #define FREE_N_CHANNELS 2 |
34c88174 URS |
2841 | |
2842 | /** | |
5f3c8d62 | 2843 | * struct kvfree_rcu_bulk_data - single block to store kvfree_rcu() pointers |
34c88174 | 2844 | * @nr_records: Number of active pointers in the array |
34c88174 | 2845 | * @next: Next bulk object in the block chain |
5f3c8d62 | 2846 | * @records: Array of the kvfree_rcu() pointers |
34c88174 | 2847 | */ |
5f3c8d62 | 2848 | struct kvfree_rcu_bulk_data { |
34c88174 | 2849 | unsigned long nr_records; |
5f3c8d62 | 2850 | struct kvfree_rcu_bulk_data *next; |
3af84862 | 2851 | void *records[]; |
34c88174 URS |
2852 | }; |
2853 | ||
3af84862 URS |
2854 | /* |
2855 | * This macro defines how many entries the "records" array | |
2856 | * will contain. It is based on the fact that the size of | |
5f3c8d62 | 2857 | * kvfree_rcu_bulk_data structure becomes exactly one page. |
3af84862 | 2858 | */ |
5f3c8d62 URS |
2859 | #define KVFREE_BULK_MAX_ENTR \ |
2860 | ((PAGE_SIZE - sizeof(struct kvfree_rcu_bulk_data)) / sizeof(void *)) | |
3af84862 | 2861 | |
a35d1690 | 2862 | /** |
0392bebe | 2863 | * struct kfree_rcu_cpu_work - single batch of kfree_rcu() requests |
a35d1690 | 2864 | * @rcu_work: Let queue_rcu_work() invoke workqueue handler after grace period |
0392bebe | 2865 | * @head_free: List of kfree_rcu() objects waiting for a grace period |
5f3c8d62 | 2866 | * @bkvhead_free: Bulk-List of kvfree_rcu() objects waiting for a grace period |
0392bebe JFG |
2867 | * @krcp: Pointer to @kfree_rcu_cpu structure |
2868 | */ | |
2869 | ||
2870 | struct kfree_rcu_cpu_work { | |
2871 | struct rcu_work rcu_work; | |
2872 | struct rcu_head *head_free; | |
5f3c8d62 | 2873 | struct kvfree_rcu_bulk_data *bkvhead_free[FREE_N_CHANNELS]; |
0392bebe JFG |
2874 | struct kfree_rcu_cpu *krcp; |
2875 | }; | |
2876 | ||
2877 | /** | |
2878 | * struct kfree_rcu_cpu - batch up kfree_rcu() requests for RCU grace period | |
a35d1690 | 2879 | * @head: List of kfree_rcu() objects not yet waiting for a grace period |
5f3c8d62 | 2880 | * @bkvhead: Bulk-List of kvfree_rcu() objects not yet waiting for a grace period |
0392bebe | 2881 | * @krw_arr: Array of batches of kfree_rcu() objects waiting for a grace period |
a35d1690 BP |
2882 | * @lock: Synchronize access to this structure |
2883 | * @monitor_work: Promote @head to @head_free after KFREE_DRAIN_JIFFIES | |
69f08d39 | 2884 | * @initialized: The @rcu_work fields have been initialized |
8e11690d | 2885 | * @count: Number of objects for which GP not started |
72a2fbda MCC |
2886 | * @bkvcache: |
2887 | * A simple cache list that contains objects for reuse purpose. | |
2888 | * In order to save some per-cpu space the list is singular. | |
2889 | * Even though it is lockless an access has to be protected by the | |
2890 | * per-cpu lock. | |
56292e86 | 2891 | * @page_cache_work: A work to refill the cache when it is empty |
d0bfa8b3 | 2892 | * @backoff_page_cache_fill: Delay cache refills |
56292e86 URS |
2893 | * @work_in_progress: Indicates that page_cache_work is running |
2894 | * @hrtimer: A hrtimer for scheduling a page_cache_work | |
72a2fbda | 2895 | * @nr_bkv_objs: number of allocated objects at @bkvcache. |
a35d1690 BP |
2896 | * |
2897 | * This is a per-CPU structure. The reason that it is not included in | |
2898 | * the rcu_data structure is to permit this code to be extracted from | |
2899 | * the RCU files. Such extraction could allow further optimization of | |
2900 | * the interactions with the slab allocators. | |
2901 | */ | |
2902 | struct kfree_rcu_cpu { | |
a35d1690 | 2903 | struct rcu_head *head; |
5f3c8d62 | 2904 | struct kvfree_rcu_bulk_data *bkvhead[FREE_N_CHANNELS]; |
0392bebe | 2905 | struct kfree_rcu_cpu_work krw_arr[KFREE_N_BATCHES]; |
8ac88f71 | 2906 | raw_spinlock_t lock; |
a35d1690 | 2907 | struct delayed_work monitor_work; |
a35d1690 | 2908 | bool initialized; |
9154244c | 2909 | int count; |
56292e86 | 2910 | |
d0bfa8b3 ZQ |
2911 | struct delayed_work page_cache_work; |
2912 | atomic_t backoff_page_cache_fill; | |
56292e86 URS |
2913 | atomic_t work_in_progress; |
2914 | struct hrtimer hrtimer; | |
2915 | ||
53c72b59 URS |
2916 | struct llist_head bkvcache; |
2917 | int nr_bkv_objs; | |
a35d1690 BP |
2918 | }; |
2919 | ||
69f08d39 SAS |
2920 | static DEFINE_PER_CPU(struct kfree_rcu_cpu, krc) = { |
2921 | .lock = __RAW_SPIN_LOCK_UNLOCKED(krc.lock), | |
2922 | }; | |
a35d1690 | 2923 | |
34c88174 | 2924 | static __always_inline void |
5f3c8d62 | 2925 | debug_rcu_bhead_unqueue(struct kvfree_rcu_bulk_data *bhead) |
34c88174 URS |
2926 | { |
2927 | #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD | |
446044eb JFG |
2928 | int i; |
2929 | ||
2930 | for (i = 0; i < bhead->nr_records; i++) | |
2931 | debug_rcu_head_unqueue((struct rcu_head *)(bhead->records[i])); | |
34c88174 URS |
2932 | #endif |
2933 | } | |
2934 | ||
952371d6 URS |
2935 | static inline struct kfree_rcu_cpu * |
2936 | krc_this_cpu_lock(unsigned long *flags) | |
2937 | { | |
2938 | struct kfree_rcu_cpu *krcp; | |
2939 | ||
2940 | local_irq_save(*flags); // For safely calling this_cpu_ptr(). | |
2941 | krcp = this_cpu_ptr(&krc); | |
69f08d39 | 2942 | raw_spin_lock(&krcp->lock); |
952371d6 URS |
2943 | |
2944 | return krcp; | |
2945 | } | |
2946 | ||
2947 | static inline void | |
2948 | krc_this_cpu_unlock(struct kfree_rcu_cpu *krcp, unsigned long flags) | |
2949 | { | |
7ffc9ec8 | 2950 | raw_spin_unlock_irqrestore(&krcp->lock, flags); |
952371d6 URS |
2951 | } |
2952 | ||
5f3c8d62 | 2953 | static inline struct kvfree_rcu_bulk_data * |
53c72b59 URS |
2954 | get_cached_bnode(struct kfree_rcu_cpu *krcp) |
2955 | { | |
2956 | if (!krcp->nr_bkv_objs) | |
2957 | return NULL; | |
2958 | ||
ac7625eb | 2959 | WRITE_ONCE(krcp->nr_bkv_objs, krcp->nr_bkv_objs - 1); |
5f3c8d62 | 2960 | return (struct kvfree_rcu_bulk_data *) |
53c72b59 URS |
2961 | llist_del_first(&krcp->bkvcache); |
2962 | } | |
2963 | ||
2964 | static inline bool | |
2965 | put_cached_bnode(struct kfree_rcu_cpu *krcp, | |
5f3c8d62 | 2966 | struct kvfree_rcu_bulk_data *bnode) |
53c72b59 URS |
2967 | { |
2968 | // Check the limit. | |
2969 | if (krcp->nr_bkv_objs >= rcu_min_cached_objs) | |
2970 | return false; | |
2971 | ||
2972 | llist_add((struct llist_node *) bnode, &krcp->bkvcache); | |
ac7625eb | 2973 | WRITE_ONCE(krcp->nr_bkv_objs, krcp->nr_bkv_objs + 1); |
53c72b59 | 2974 | return true; |
53c72b59 URS |
2975 | } |
2976 | ||
d0bfa8b3 ZQ |
2977 | static int |
2978 | drain_page_cache(struct kfree_rcu_cpu *krcp) | |
2979 | { | |
2980 | unsigned long flags; | |
2981 | struct llist_node *page_list, *pos, *n; | |
2982 | int freed = 0; | |
53c72b59 | 2983 | |
d0bfa8b3 ZQ |
2984 | raw_spin_lock_irqsave(&krcp->lock, flags); |
2985 | page_list = llist_del_all(&krcp->bkvcache); | |
ac7625eb | 2986 | WRITE_ONCE(krcp->nr_bkv_objs, 0); |
d0bfa8b3 | 2987 | raw_spin_unlock_irqrestore(&krcp->lock, flags); |
53c72b59 | 2988 | |
d0bfa8b3 ZQ |
2989 | llist_for_each_safe(pos, n, page_list) { |
2990 | free_page((unsigned long)pos); | |
2991 | freed++; | |
2992 | } | |
2993 | ||
2994 | return freed; | |
53c72b59 URS |
2995 | } |
2996 | ||
495aa969 | 2997 | /* |
a35d1690 | 2998 | * This function is invoked in workqueue context after a grace period. |
277ffe1b | 2999 | * It frees all the objects queued on ->bkvhead_free or ->head_free. |
495aa969 | 3000 | */ |
a35d1690 BP |
3001 | static void kfree_rcu_work(struct work_struct *work) |
3002 | { | |
3003 | unsigned long flags; | |
5f3c8d62 | 3004 | struct kvfree_rcu_bulk_data *bkvhead[FREE_N_CHANNELS], *bnext; |
a35d1690 BP |
3005 | struct rcu_head *head, *next; |
3006 | struct kfree_rcu_cpu *krcp; | |
0392bebe | 3007 | struct kfree_rcu_cpu_work *krwp; |
5f3c8d62 | 3008 | int i, j; |
a35d1690 | 3009 | |
0392bebe JFG |
3010 | krwp = container_of(to_rcu_work(work), |
3011 | struct kfree_rcu_cpu_work, rcu_work); | |
3012 | krcp = krwp->krcp; | |
34c88174 | 3013 | |
8ac88f71 | 3014 | raw_spin_lock_irqsave(&krcp->lock, flags); |
5f3c8d62 URS |
3015 | // Channels 1 and 2. |
3016 | for (i = 0; i < FREE_N_CHANNELS; i++) { | |
3017 | bkvhead[i] = krwp->bkvhead_free[i]; | |
3018 | krwp->bkvhead_free[i] = NULL; | |
3019 | } | |
34c88174 | 3020 | |
5f3c8d62 | 3021 | // Channel 3. |
0392bebe JFG |
3022 | head = krwp->head_free; |
3023 | krwp->head_free = NULL; | |
8ac88f71 | 3024 | raw_spin_unlock_irqrestore(&krcp->lock, flags); |
a35d1690 | 3025 | |
277ffe1b | 3026 | // Handle the first two channels. |
5f3c8d62 URS |
3027 | for (i = 0; i < FREE_N_CHANNELS; i++) { |
3028 | for (; bkvhead[i]; bkvhead[i] = bnext) { | |
3029 | bnext = bkvhead[i]->next; | |
3030 | debug_rcu_bhead_unqueue(bkvhead[i]); | |
3031 | ||
3032 | rcu_lock_acquire(&rcu_callback_map); | |
3033 | if (i == 0) { // kmalloc() / kfree(). | |
3034 | trace_rcu_invoke_kfree_bulk_callback( | |
3035 | rcu_state.name, bkvhead[i]->nr_records, | |
3036 | bkvhead[i]->records); | |
3037 | ||
3038 | kfree_bulk(bkvhead[i]->nr_records, | |
3039 | bkvhead[i]->records); | |
3040 | } else { // vmalloc() / vfree(). | |
3041 | for (j = 0; j < bkvhead[i]->nr_records; j++) { | |
c408b215 | 3042 | trace_rcu_invoke_kvfree_callback( |
5f3c8d62 URS |
3043 | rcu_state.name, |
3044 | bkvhead[i]->records[j], 0); | |
3045 | ||
3046 | vfree(bkvhead[i]->records[j]); | |
3047 | } | |
3048 | } | |
3049 | rcu_lock_release(&rcu_callback_map); | |
61370792 | 3050 | |
56292e86 | 3051 | raw_spin_lock_irqsave(&krcp->lock, flags); |
5f3c8d62 URS |
3052 | if (put_cached_bnode(krcp, bkvhead[i])) |
3053 | bkvhead[i] = NULL; | |
56292e86 | 3054 | raw_spin_unlock_irqrestore(&krcp->lock, flags); |
34c88174 | 3055 | |
5f3c8d62 URS |
3056 | if (bkvhead[i]) |
3057 | free_page((unsigned long) bkvhead[i]); | |
34c88174 | 3058 | |
5f3c8d62 URS |
3059 | cond_resched_tasks_rcu_qs(); |
3060 | } | |
34c88174 URS |
3061 | } |
3062 | ||
3063 | /* | |
d8628f35 URS |
3064 | * This is used when the "bulk" path can not be used for the |
3065 | * double-argument of kvfree_rcu(). This happens when the | |
3066 | * page-cache is empty, which means that objects are instead | |
3067 | * queued on a linked list through their rcu_head structures. | |
3068 | * This list is named "Channel 3". | |
34c88174 | 3069 | */ |
a35d1690 | 3070 | for (; head; head = next) { |
77a40f97 | 3071 | unsigned long offset = (unsigned long)head->func; |
446044eb | 3072 | void *ptr = (void *)head - offset; |
77a40f97 | 3073 | |
a35d1690 | 3074 | next = head->next; |
446044eb | 3075 | debug_rcu_head_unqueue((struct rcu_head *)ptr); |
77a40f97 | 3076 | rcu_lock_acquire(&rcu_callback_map); |
c408b215 | 3077 | trace_rcu_invoke_kvfree_callback(rcu_state.name, head, offset); |
77a40f97 | 3078 | |
c408b215 | 3079 | if (!WARN_ON_ONCE(!__is_kvfree_rcu_offset(offset))) |
5f3c8d62 | 3080 | kvfree(ptr); |
77a40f97 JFG |
3081 | |
3082 | rcu_lock_release(&rcu_callback_map); | |
a35d1690 BP |
3083 | cond_resched_tasks_rcu_qs(); |
3084 | } | |
3085 | } | |
3086 | ||
82d26c36 JFG |
3087 | static bool |
3088 | need_offload_krc(struct kfree_rcu_cpu *krcp) | |
3089 | { | |
3090 | int i; | |
3091 | ||
3092 | for (i = 0; i < FREE_N_CHANNELS; i++) | |
3093 | if (krcp->bkvhead[i]) | |
3094 | return true; | |
3095 | ||
3096 | return !!krcp->head; | |
3097 | } | |
3098 | ||
51824b78 URS |
3099 | static void |
3100 | schedule_delayed_monitor_work(struct kfree_rcu_cpu *krcp) | |
3101 | { | |
3102 | long delay, delay_left; | |
3103 | ||
3104 | delay = READ_ONCE(krcp->count) >= KVFREE_BULK_MAX_ENTR ? 1:KFREE_DRAIN_JIFFIES; | |
3105 | if (delayed_work_pending(&krcp->monitor_work)) { | |
3106 | delay_left = krcp->monitor_work.timer.expires - jiffies; | |
3107 | if (delay < delay_left) | |
3108 | mod_delayed_work(system_wq, &krcp->monitor_work, delay); | |
3109 | return; | |
3110 | } | |
3111 | queue_delayed_work(system_wq, &krcp->monitor_work, delay); | |
3112 | } | |
3113 | ||
495aa969 | 3114 | /* |
a78d4a2a | 3115 | * This function is invoked after the KFREE_DRAIN_JIFFIES timeout. |
a35d1690 | 3116 | */ |
a78d4a2a | 3117 | static void kfree_rcu_monitor(struct work_struct *work) |
a35d1690 | 3118 | { |
a78d4a2a URS |
3119 | struct kfree_rcu_cpu *krcp = container_of(work, |
3120 | struct kfree_rcu_cpu, monitor_work.work); | |
3121 | unsigned long flags; | |
5f3c8d62 | 3122 | int i, j; |
0392bebe | 3123 | |
a78d4a2a | 3124 | raw_spin_lock_irqsave(&krcp->lock, flags); |
a35d1690 | 3125 | |
a78d4a2a | 3126 | // Attempt to start a new batch. |
34c88174 | 3127 | for (i = 0; i < KFREE_N_BATCHES; i++) { |
a78d4a2a | 3128 | struct kfree_rcu_cpu_work *krwp = &(krcp->krw_arr[i]); |
a35d1690 | 3129 | |
a78d4a2a URS |
3130 | // Try to detach bkvhead or head and attach it over any |
3131 | // available corresponding free channel. It can be that | |
3132 | // a previous RCU batch is in progress, it means that | |
3133 | // immediately to queue another one is not possible so | |
3134 | // in that case the monitor work is rearmed. | |
5f3c8d62 URS |
3135 | if ((krcp->bkvhead[0] && !krwp->bkvhead_free[0]) || |
3136 | (krcp->bkvhead[1] && !krwp->bkvhead_free[1]) || | |
34c88174 | 3137 | (krcp->head && !krwp->head_free)) { |
d8628f35 URS |
3138 | // Channel 1 corresponds to the SLAB-pointer bulk path. |
3139 | // Channel 2 corresponds to vmalloc-pointer bulk path. | |
5f3c8d62 URS |
3140 | for (j = 0; j < FREE_N_CHANNELS; j++) { |
3141 | if (!krwp->bkvhead_free[j]) { | |
3142 | krwp->bkvhead_free[j] = krcp->bkvhead[j]; | |
3143 | krcp->bkvhead[j] = NULL; | |
3144 | } | |
34c88174 URS |
3145 | } |
3146 | ||
d8628f35 URS |
3147 | // Channel 3 corresponds to both SLAB and vmalloc |
3148 | // objects queued on the linked list. | |
34c88174 URS |
3149 | if (!krwp->head_free) { |
3150 | krwp->head_free = krcp->head; | |
3151 | krcp->head = NULL; | |
3152 | } | |
3153 | ||
a6a82ce1 | 3154 | WRITE_ONCE(krcp->count, 0); |
9154244c | 3155 | |
a78d4a2a URS |
3156 | // One work is per one batch, so there are three |
3157 | // "free channels", the batch can handle. It can | |
3158 | // be that the work is in the pending state when | |
3159 | // channels have been detached following by each | |
3160 | // other. | |
34c88174 | 3161 | queue_rcu_work(system_wq, &krwp->rcu_work); |
34c88174 URS |
3162 | } |
3163 | } | |
3164 | ||
a78d4a2a URS |
3165 | // If there is nothing to detach, it means that our job is |
3166 | // successfully done here. In case of having at least one | |
3167 | // of the channels that is still busy we should rearm the | |
3168 | // work to repeat an attempt. Because previous batches are | |
3169 | // still in progress. | |
82d26c36 | 3170 | if (need_offload_krc(krcp)) |
51824b78 | 3171 | schedule_delayed_monitor_work(krcp); |
a35d1690 | 3172 | |
8ac88f71 | 3173 | raw_spin_unlock_irqrestore(&krcp->lock, flags); |
a35d1690 BP |
3174 | } |
3175 | ||
56292e86 URS |
3176 | static enum hrtimer_restart |
3177 | schedule_page_work_fn(struct hrtimer *t) | |
3178 | { | |
3179 | struct kfree_rcu_cpu *krcp = | |
3180 | container_of(t, struct kfree_rcu_cpu, hrtimer); | |
3181 | ||
d0bfa8b3 | 3182 | queue_delayed_work(system_highpri_wq, &krcp->page_cache_work, 0); |
56292e86 URS |
3183 | return HRTIMER_NORESTART; |
3184 | } | |
3185 | ||
3186 | static void fill_page_cache_func(struct work_struct *work) | |
3187 | { | |
3188 | struct kvfree_rcu_bulk_data *bnode; | |
3189 | struct kfree_rcu_cpu *krcp = | |
3190 | container_of(work, struct kfree_rcu_cpu, | |
d0bfa8b3 | 3191 | page_cache_work.work); |
56292e86 | 3192 | unsigned long flags; |
d0bfa8b3 | 3193 | int nr_pages; |
56292e86 URS |
3194 | bool pushed; |
3195 | int i; | |
3196 | ||
d0bfa8b3 ZQ |
3197 | nr_pages = atomic_read(&krcp->backoff_page_cache_fill) ? |
3198 | 1 : rcu_min_cached_objs; | |
3199 | ||
3200 | for (i = 0; i < nr_pages; i++) { | |
56292e86 | 3201 | bnode = (struct kvfree_rcu_bulk_data *) |
ee6ddf58 | 3202 | __get_free_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN); |
56292e86 | 3203 | |
093590c1 MH |
3204 | if (!bnode) |
3205 | break; | |
56292e86 | 3206 | |
093590c1 MH |
3207 | raw_spin_lock_irqsave(&krcp->lock, flags); |
3208 | pushed = put_cached_bnode(krcp, bnode); | |
3209 | raw_spin_unlock_irqrestore(&krcp->lock, flags); | |
3210 | ||
3211 | if (!pushed) { | |
3212 | free_page((unsigned long) bnode); | |
3213 | break; | |
56292e86 URS |
3214 | } |
3215 | } | |
3216 | ||
3217 | atomic_set(&krcp->work_in_progress, 0); | |
d0bfa8b3 | 3218 | atomic_set(&krcp->backoff_page_cache_fill, 0); |
56292e86 URS |
3219 | } |
3220 | ||
3221 | static void | |
3222 | run_page_cache_worker(struct kfree_rcu_cpu *krcp) | |
3223 | { | |
3224 | if (rcu_scheduler_active == RCU_SCHEDULER_RUNNING && | |
3225 | !atomic_xchg(&krcp->work_in_progress, 1)) { | |
d0bfa8b3 ZQ |
3226 | if (atomic_read(&krcp->backoff_page_cache_fill)) { |
3227 | queue_delayed_work(system_wq, | |
3228 | &krcp->page_cache_work, | |
3229 | msecs_to_jiffies(rcu_delay_page_cache_fill_msec)); | |
3230 | } else { | |
3231 | hrtimer_init(&krcp->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
3232 | krcp->hrtimer.function = schedule_page_work_fn; | |
3233 | hrtimer_start(&krcp->hrtimer, 0, HRTIMER_MODE_REL); | |
3234 | } | |
56292e86 URS |
3235 | } |
3236 | } | |
3237 | ||
148e3731 URS |
3238 | // Record ptr in a page managed by krcp, with the pre-krc_this_cpu_lock() |
3239 | // state specified by flags. If can_alloc is true, the caller must | |
3240 | // be schedulable and not be holding any locks or mutexes that might be | |
3241 | // acquired by the memory allocator or anything that it might invoke. | |
3242 | // Returns true if ptr was successfully recorded, else the caller must | |
3243 | // use a fallback. | |
34c88174 | 3244 | static inline bool |
148e3731 URS |
3245 | add_ptr_to_bulk_krc_lock(struct kfree_rcu_cpu **krcp, |
3246 | unsigned long *flags, void *ptr, bool can_alloc) | |
34c88174 | 3247 | { |
5f3c8d62 URS |
3248 | struct kvfree_rcu_bulk_data *bnode; |
3249 | int idx; | |
34c88174 | 3250 | |
148e3731 URS |
3251 | *krcp = krc_this_cpu_lock(flags); |
3252 | if (unlikely(!(*krcp)->initialized)) | |
34c88174 URS |
3253 | return false; |
3254 | ||
5f3c8d62 | 3255 | idx = !!is_vmalloc_addr(ptr); |
34c88174 URS |
3256 | |
3257 | /* Check if a new block is required. */ | |
148e3731 URS |
3258 | if (!(*krcp)->bkvhead[idx] || |
3259 | (*krcp)->bkvhead[idx]->nr_records == KVFREE_BULK_MAX_ENTR) { | |
3260 | bnode = get_cached_bnode(*krcp); | |
3261 | if (!bnode && can_alloc) { | |
3262 | krc_this_cpu_unlock(*krcp, *flags); | |
3e7ce7a1 URS |
3263 | |
3264 | // __GFP_NORETRY - allows a light-weight direct reclaim | |
3265 | // what is OK from minimizing of fallback hitting point of | |
3266 | // view. Apart of that it forbids any OOM invoking what is | |
3267 | // also beneficial since we are about to release memory soon. | |
3268 | // | |
3269 | // __GFP_NOMEMALLOC - prevents from consuming of all the | |
3270 | // memory reserves. Please note we have a fallback path. | |
3271 | // | |
3272 | // __GFP_NOWARN - it is supposed that an allocation can | |
3273 | // be failed under low memory or high memory pressure | |
3274 | // scenarios. | |
148e3731 | 3275 | bnode = (struct kvfree_rcu_bulk_data *) |
3e7ce7a1 | 3276 | __get_free_page(GFP_KERNEL | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN); |
148e3731 URS |
3277 | *krcp = krc_this_cpu_lock(flags); |
3278 | } | |
3279 | ||
56292e86 | 3280 | if (!bnode) |
34c88174 URS |
3281 | return false; |
3282 | ||
3283 | /* Initialize the new block. */ | |
3284 | bnode->nr_records = 0; | |
148e3731 | 3285 | bnode->next = (*krcp)->bkvhead[idx]; |
34c88174 URS |
3286 | |
3287 | /* Attach it to the head. */ | |
148e3731 | 3288 | (*krcp)->bkvhead[idx] = bnode; |
34c88174 URS |
3289 | } |
3290 | ||
34c88174 | 3291 | /* Finally insert. */ |
148e3731 URS |
3292 | (*krcp)->bkvhead[idx]->records |
3293 | [(*krcp)->bkvhead[idx]->nr_records++] = ptr; | |
34c88174 URS |
3294 | |
3295 | return true; | |
3296 | } | |
3297 | ||
a35d1690 | 3298 | /* |
277ffe1b ZZ |
3299 | * Queue a request for lazy invocation of the appropriate free routine |
3300 | * after a grace period. Please note that three paths are maintained, | |
3301 | * two for the common case using arrays of pointers and a third one that | |
3302 | * is used only when the main paths cannot be used, for example, due to | |
3303 | * memory pressure. | |
a35d1690 | 3304 | * |
c408b215 | 3305 | * Each kvfree_call_rcu() request is added to a batch. The batch will be drained |
34c88174 URS |
3306 | * every KFREE_DRAIN_JIFFIES number of jiffies. All the objects in the batch will |
3307 | * be free'd in workqueue context. This allows us to: batch requests together to | |
5f3c8d62 | 3308 | * reduce the number of grace periods during heavy kfree_rcu()/kvfree_rcu() load. |
495aa969 | 3309 | */ |
c408b215 | 3310 | void kvfree_call_rcu(struct rcu_head *head, rcu_callback_t func) |
495aa969 | 3311 | { |
a35d1690 BP |
3312 | unsigned long flags; |
3313 | struct kfree_rcu_cpu *krcp; | |
3042f83f | 3314 | bool success; |
446044eb | 3315 | void *ptr; |
a35d1690 | 3316 | |
3042f83f URS |
3317 | if (head) { |
3318 | ptr = (void *) head - (unsigned long) func; | |
3319 | } else { | |
3320 | /* | |
3321 | * Please note there is a limitation for the head-less | |
3322 | * variant, that is why there is a clear rule for such | |
3323 | * objects: it can be used from might_sleep() context | |
3324 | * only. For other places please embed an rcu_head to | |
3325 | * your data. | |
3326 | */ | |
3327 | might_sleep(); | |
3328 | ptr = (unsigned long *) func; | |
3329 | } | |
3330 | ||
a35d1690 | 3331 | // Queue the object but don't yet schedule the batch. |
446044eb | 3332 | if (debug_rcu_head_queue(ptr)) { |
e99637be JFG |
3333 | // Probable double kfree_rcu(), just leak. |
3334 | WARN_ONCE(1, "%s(): Double-freed call. rcu_head %p\n", | |
3335 | __func__, head); | |
3042f83f URS |
3336 | |
3337 | // Mark as success and leave. | |
148e3731 | 3338 | return; |
e99637be | 3339 | } |
34c88174 | 3340 | |
300c0c5e | 3341 | kasan_record_aux_stack_noalloc(ptr); |
148e3731 | 3342 | success = add_ptr_to_bulk_krc_lock(&krcp, &flags, ptr, !head); |
3042f83f | 3343 | if (!success) { |
56292e86 URS |
3344 | run_page_cache_worker(krcp); |
3345 | ||
3042f83f URS |
3346 | if (head == NULL) |
3347 | // Inline if kvfree_rcu(one_arg) call. | |
3348 | goto unlock_return; | |
3349 | ||
34c88174 URS |
3350 | head->func = func; |
3351 | head->next = krcp->head; | |
3352 | krcp->head = head; | |
3042f83f | 3353 | success = true; |
34c88174 | 3354 | } |
a35d1690 | 3355 | |
a6a82ce1 | 3356 | WRITE_ONCE(krcp->count, krcp->count + 1); |
9154244c | 3357 | |
a35d1690 | 3358 | // Set timer to drain after KFREE_DRAIN_JIFFIES. |
82d26c36 | 3359 | if (rcu_scheduler_active == RCU_SCHEDULER_RUNNING) |
51824b78 | 3360 | schedule_delayed_monitor_work(krcp); |
a35d1690 | 3361 | |
e99637be | 3362 | unlock_return: |
952371d6 | 3363 | krc_this_cpu_unlock(krcp, flags); |
3042f83f URS |
3364 | |
3365 | /* | |
3366 | * Inline kvfree() after synchronize_rcu(). We can do | |
3367 | * it from might_sleep() context only, so the current | |
3368 | * CPU can pass the QS state. | |
3369 | */ | |
3370 | if (!success) { | |
3371 | debug_rcu_head_unqueue((struct rcu_head *) ptr); | |
3372 | synchronize_rcu(); | |
3373 | kvfree(ptr); | |
3374 | } | |
495aa969 | 3375 | } |
c408b215 | 3376 | EXPORT_SYMBOL_GPL(kvfree_call_rcu); |
495aa969 | 3377 | |
9154244c JFG |
3378 | static unsigned long |
3379 | kfree_rcu_shrink_count(struct shrinker *shrink, struct shrink_control *sc) | |
3380 | { | |
3381 | int cpu; | |
a6a82ce1 | 3382 | unsigned long count = 0; |
9154244c JFG |
3383 | |
3384 | /* Snapshot count of all CPUs */ | |
70060b87 | 3385 | for_each_possible_cpu(cpu) { |
9154244c JFG |
3386 | struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu); |
3387 | ||
a6a82ce1 | 3388 | count += READ_ONCE(krcp->count); |
ac7625eb | 3389 | count += READ_ONCE(krcp->nr_bkv_objs); |
d0bfa8b3 | 3390 | atomic_set(&krcp->backoff_page_cache_fill, 1); |
9154244c JFG |
3391 | } |
3392 | ||
38269096 | 3393 | return count == 0 ? SHRINK_EMPTY : count; |
9154244c JFG |
3394 | } |
3395 | ||
3396 | static unsigned long | |
3397 | kfree_rcu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc) | |
3398 | { | |
3399 | int cpu, freed = 0; | |
9154244c | 3400 | |
70060b87 | 3401 | for_each_possible_cpu(cpu) { |
9154244c JFG |
3402 | int count; |
3403 | struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu); | |
3404 | ||
3405 | count = krcp->count; | |
d0bfa8b3 | 3406 | count += drain_page_cache(krcp); |
7fe1da33 | 3407 | kfree_rcu_monitor(&krcp->monitor_work.work); |
9154244c JFG |
3408 | |
3409 | sc->nr_to_scan -= count; | |
3410 | freed += count; | |
3411 | ||
3412 | if (sc->nr_to_scan <= 0) | |
3413 | break; | |
3414 | } | |
3415 | ||
c6dfd72b | 3416 | return freed == 0 ? SHRINK_STOP : freed; |
9154244c JFG |
3417 | } |
3418 | ||
3419 | static struct shrinker kfree_rcu_shrinker = { | |
3420 | .count_objects = kfree_rcu_shrink_count, | |
3421 | .scan_objects = kfree_rcu_shrink_scan, | |
3422 | .batch = 0, | |
3423 | .seeks = DEFAULT_SEEKS, | |
3424 | }; | |
3425 | ||
a35d1690 BP |
3426 | void __init kfree_rcu_scheduler_running(void) |
3427 | { | |
3428 | int cpu; | |
3429 | unsigned long flags; | |
3430 | ||
70060b87 | 3431 | for_each_possible_cpu(cpu) { |
a35d1690 BP |
3432 | struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu); |
3433 | ||
8ac88f71 | 3434 | raw_spin_lock_irqsave(&krcp->lock, flags); |
82d26c36 | 3435 | if (need_offload_krc(krcp)) |
51824b78 | 3436 | schedule_delayed_monitor_work(krcp); |
8ac88f71 | 3437 | raw_spin_unlock_irqrestore(&krcp->lock, flags); |
a35d1690 BP |
3438 | } |
3439 | } | |
3440 | ||
e5bc3af7 PM |
3441 | /* |
3442 | * During early boot, any blocking grace-period wait automatically | |
258f887a | 3443 | * implies a grace period. |
e5bc3af7 | 3444 | * |
258f887a PM |
3445 | * Later on, this could in theory be the case for kernels built with |
3446 | * CONFIG_SMP=y && CONFIG_PREEMPTION=y running on a single CPU, but this | |
3447 | * is not a common case. Furthermore, this optimization would cause | |
3448 | * the rcu_gp_oldstate structure to expand by 50%, so this potential | |
3449 | * grace-period optimization is ignored once the scheduler is running. | |
e5bc3af7 PM |
3450 | */ |
3451 | static int rcu_blocking_is_gp(void) | |
3452 | { | |
258f887a PM |
3453 | if (rcu_scheduler_active != RCU_SCHEDULER_INACTIVE) |
3454 | return false; | |
e5bc3af7 | 3455 | might_sleep(); /* Check for RCU read-side critical section. */ |
258f887a | 3456 | return true; |
e5bc3af7 PM |
3457 | } |
3458 | ||
3459 | /** | |
3460 | * synchronize_rcu - wait until a grace period has elapsed. | |
3461 | * | |
3462 | * Control will return to the caller some time after a full grace | |
3463 | * period has elapsed, in other words after all currently executing RCU | |
3464 | * read-side critical sections have completed. Note, however, that | |
3465 | * upon return from synchronize_rcu(), the caller might well be executing | |
3466 | * concurrently with new RCU read-side critical sections that began while | |
1893afd6 PM |
3467 | * synchronize_rcu() was waiting. |
3468 | * | |
3469 | * RCU read-side critical sections are delimited by rcu_read_lock() | |
3470 | * and rcu_read_unlock(), and may be nested. In addition, but only in | |
3471 | * v5.0 and later, regions of code across which interrupts, preemption, | |
3472 | * or softirqs have been disabled also serve as RCU read-side critical | |
e5bc3af7 PM |
3473 | * sections. This includes hardware interrupt handlers, softirq handlers, |
3474 | * and NMI handlers. | |
3475 | * | |
3476 | * Note that this guarantee implies further memory-ordering guarantees. | |
3477 | * On systems with more than one CPU, when synchronize_rcu() returns, | |
3478 | * each CPU is guaranteed to have executed a full memory barrier since | |
3479 | * the end of its last RCU read-side critical section whose beginning | |
3480 | * preceded the call to synchronize_rcu(). In addition, each CPU having | |
3481 | * an RCU read-side critical section that extends beyond the return from | |
3482 | * synchronize_rcu() is guaranteed to have executed a full memory barrier | |
3483 | * after the beginning of synchronize_rcu() and before the beginning of | |
3484 | * that RCU read-side critical section. Note that these guarantees include | |
3485 | * CPUs that are offline, idle, or executing in user mode, as well as CPUs | |
3486 | * that are executing in the kernel. | |
3487 | * | |
3488 | * Furthermore, if CPU A invoked synchronize_rcu(), which returned | |
3489 | * to its caller on CPU B, then both CPU A and CPU B are guaranteed | |
3490 | * to have executed a full memory barrier during the execution of | |
3491 | * synchronize_rcu() -- even if CPU A and CPU B are the same CPU (but | |
3492 | * again only if the system has more than one CPU). | |
3d3a0d1b PM |
3493 | * |
3494 | * Implementation of these memory-ordering guarantees is described here: | |
3495 | * Documentation/RCU/Design/Memory-Ordering/Tree-RCU-Memory-Ordering.rst. | |
e5bc3af7 PM |
3496 | */ |
3497 | void synchronize_rcu(void) | |
3498 | { | |
910e1209 PM |
3499 | unsigned long flags; |
3500 | struct rcu_node *rnp; | |
3501 | ||
e5bc3af7 PM |
3502 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
3503 | lock_is_held(&rcu_lock_map) || | |
3504 | lock_is_held(&rcu_sched_lock_map), | |
3505 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
910e1209 PM |
3506 | if (!rcu_blocking_is_gp()) { |
3507 | if (rcu_gp_is_expedited()) | |
3508 | synchronize_rcu_expedited(); | |
3509 | else | |
3510 | wait_rcu_gp(call_rcu); | |
3511 | return; | |
bf95b2bc | 3512 | } |
910e1209 PM |
3513 | |
3514 | // Context allows vacuous grace periods. | |
3515 | // Note well that this code runs with !PREEMPT && !SMP. | |
3516 | // In addition, all code that advances grace periods runs at | |
3517 | // process level. Therefore, this normal GP overlaps with other | |
3518 | // normal GPs only by being fully nested within them, which allows | |
3519 | // reuse of ->gp_seq_polled_snap. | |
3520 | rcu_poll_gp_seq_start_unlocked(&rcu_state.gp_seq_polled_snap); | |
3521 | rcu_poll_gp_seq_end_unlocked(&rcu_state.gp_seq_polled_snap); | |
3522 | ||
d761de8a PM |
3523 | // Update the normal grace-period counters to record |
3524 | // this grace period, but only those used by the boot CPU. | |
3525 | // The rcu_scheduler_starting() will take care of the rest of | |
3526 | // these counters. | |
910e1209 PM |
3527 | local_irq_save(flags); |
3528 | WARN_ON_ONCE(num_online_cpus() > 1); | |
3529 | rcu_state.gp_seq += (1 << RCU_SEQ_CTR_SHIFT); | |
d761de8a | 3530 | for (rnp = this_cpu_ptr(&rcu_data)->mynode; rnp; rnp = rnp->parent) |
910e1209 PM |
3531 | rnp->gp_seq_needed = rnp->gp_seq = rcu_state.gp_seq; |
3532 | local_irq_restore(flags); | |
e5bc3af7 PM |
3533 | } |
3534 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
3535 | ||
91a967fd PM |
3536 | /** |
3537 | * get_completed_synchronize_rcu_full - Return a full pre-completed polled state cookie | |
3538 | * @rgosp: Place to put state cookie | |
3539 | * | |
3540 | * Stores into @rgosp a value that will always be treated by functions | |
3541 | * like poll_state_synchronize_rcu_full() as a cookie whose grace period | |
3542 | * has already completed. | |
3543 | */ | |
3544 | void get_completed_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp) | |
3545 | { | |
3546 | rgosp->rgos_norm = RCU_GET_STATE_COMPLETED; | |
3547 | rgosp->rgos_exp = RCU_GET_STATE_COMPLETED; | |
91a967fd PM |
3548 | } |
3549 | EXPORT_SYMBOL_GPL(get_completed_synchronize_rcu_full); | |
3550 | ||
765a3f4f PM |
3551 | /** |
3552 | * get_state_synchronize_rcu - Snapshot current RCU state | |
3553 | * | |
3554 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
7abb18bd PM |
3555 | * or poll_state_synchronize_rcu() to determine whether or not a full |
3556 | * grace period has elapsed in the meantime. | |
765a3f4f PM |
3557 | */ |
3558 | unsigned long get_state_synchronize_rcu(void) | |
3559 | { | |
3560 | /* | |
3561 | * Any prior manipulation of RCU-protected data must happen | |
e4be81a2 | 3562 | * before the load from ->gp_seq. |
765a3f4f PM |
3563 | */ |
3564 | smp_mb(); /* ^^^ */ | |
bf95b2bc | 3565 | return rcu_seq_snap(&rcu_state.gp_seq_polled); |
765a3f4f PM |
3566 | } |
3567 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
3568 | ||
7abb18bd | 3569 | /** |
3fdefca9 PM |
3570 | * get_state_synchronize_rcu_full - Snapshot RCU state, both normal and expedited |
3571 | * @rgosp: location to place combined normal/expedited grace-period state | |
7abb18bd | 3572 | * |
3fdefca9 PM |
3573 | * Places the normal and expedited grace-period states in @rgosp. This |
3574 | * state value can be passed to a later call to cond_synchronize_rcu_full() | |
3575 | * or poll_state_synchronize_rcu_full() to determine whether or not a | |
3576 | * grace period (whether normal or expedited) has elapsed in the meantime. | |
3577 | * The rcu_gp_oldstate structure takes up twice the memory of an unsigned | |
3578 | * long, but is guaranteed to see all grace periods. In contrast, the | |
3579 | * combined state occupies less memory, but can sometimes fail to take | |
3580 | * grace periods into account. | |
7abb18bd | 3581 | * |
3fdefca9 PM |
3582 | * This does not guarantee that the needed grace period will actually |
3583 | * start. | |
7abb18bd | 3584 | */ |
3fdefca9 PM |
3585 | void get_state_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp) |
3586 | { | |
3587 | struct rcu_node *rnp = rcu_get_root(); | |
3588 | ||
3589 | /* | |
3590 | * Any prior manipulation of RCU-protected data must happen | |
3591 | * before the loads from ->gp_seq and ->expedited_sequence. | |
3592 | */ | |
3593 | smp_mb(); /* ^^^ */ | |
3594 | rgosp->rgos_norm = rcu_seq_snap(&rnp->gp_seq); | |
3595 | rgosp->rgos_exp = rcu_seq_snap(&rcu_state.expedited_sequence); | |
3fdefca9 PM |
3596 | } |
3597 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu_full); | |
3598 | ||
76ea3641 PM |
3599 | /* |
3600 | * Helper function for start_poll_synchronize_rcu() and | |
3601 | * start_poll_synchronize_rcu_full(). | |
7abb18bd | 3602 | */ |
76ea3641 | 3603 | static void start_poll_synchronize_rcu_common(void) |
7abb18bd PM |
3604 | { |
3605 | unsigned long flags; | |
7abb18bd PM |
3606 | bool needwake; |
3607 | struct rcu_data *rdp; | |
3608 | struct rcu_node *rnp; | |
3609 | ||
3610 | lockdep_assert_irqs_enabled(); | |
3611 | local_irq_save(flags); | |
3612 | rdp = this_cpu_ptr(&rcu_data); | |
3613 | rnp = rdp->mynode; | |
3614 | raw_spin_lock_rcu_node(rnp); // irqs already disabled. | |
bf95b2bc PM |
3615 | // Note it is possible for a grace period to have elapsed between |
3616 | // the above call to get_state_synchronize_rcu() and the below call | |
3617 | // to rcu_seq_snap. This is OK, the worst that happens is that we | |
3618 | // get a grace period that no one needed. These accesses are ordered | |
3619 | // by smp_mb(), and we are accessing them in the opposite order | |
3620 | // from which they are updated at grace-period start, as required. | |
3621 | needwake = rcu_start_this_gp(rnp, rdp, rcu_seq_snap(&rcu_state.gp_seq)); | |
7abb18bd PM |
3622 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
3623 | if (needwake) | |
3624 | rcu_gp_kthread_wake(); | |
76ea3641 PM |
3625 | } |
3626 | ||
3627 | /** | |
3628 | * start_poll_synchronize_rcu - Snapshot and start RCU grace period | |
3629 | * | |
3630 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
3631 | * or poll_state_synchronize_rcu() to determine whether or not a full | |
3632 | * grace period has elapsed in the meantime. If the needed grace period | |
3633 | * is not already slated to start, notifies RCU core of the need for that | |
3634 | * grace period. | |
3635 | * | |
3636 | * Interrupts must be enabled for the case where it is necessary to awaken | |
3637 | * the grace-period kthread. | |
3638 | */ | |
3639 | unsigned long start_poll_synchronize_rcu(void) | |
3640 | { | |
3641 | unsigned long gp_seq = get_state_synchronize_rcu(); | |
3642 | ||
3643 | start_poll_synchronize_rcu_common(); | |
7abb18bd PM |
3644 | return gp_seq; |
3645 | } | |
3646 | EXPORT_SYMBOL_GPL(start_poll_synchronize_rcu); | |
3647 | ||
3648 | /** | |
76ea3641 PM |
3649 | * start_poll_synchronize_rcu_full - Take a full snapshot and start RCU grace period |
3650 | * @rgosp: value from get_state_synchronize_rcu_full() or start_poll_synchronize_rcu_full() | |
7abb18bd | 3651 | * |
76ea3641 PM |
3652 | * Places the normal and expedited grace-period states in *@rgos. This |
3653 | * state value can be passed to a later call to cond_synchronize_rcu_full() | |
3654 | * or poll_state_synchronize_rcu_full() to determine whether or not a | |
3655 | * grace period (whether normal or expedited) has elapsed in the meantime. | |
3656 | * If the needed grace period is not already slated to start, notifies | |
3657 | * RCU core of the need for that grace period. | |
3658 | * | |
3659 | * Interrupts must be enabled for the case where it is necessary to awaken | |
3660 | * the grace-period kthread. | |
3661 | */ | |
3662 | void start_poll_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp) | |
3663 | { | |
3664 | get_state_synchronize_rcu_full(rgosp); | |
3665 | ||
3666 | start_poll_synchronize_rcu_common(); | |
3667 | } | |
3668 | EXPORT_SYMBOL_GPL(start_poll_synchronize_rcu_full); | |
3669 | ||
7abb18bd | 3670 | /** |
91a967fd | 3671 | * poll_state_synchronize_rcu - Has the specified RCU grace period completed? |
3d3a0d1b | 3672 | * @oldstate: value from get_state_synchronize_rcu() or start_poll_synchronize_rcu() |
7abb18bd PM |
3673 | * |
3674 | * If a full RCU grace period has elapsed since the earlier call from | |
f21e0143 | 3675 | * which @oldstate was obtained, return @true, otherwise return @false. |
a616aec9 | 3676 | * If @false is returned, it is the caller's responsibility to invoke this |
7abb18bd PM |
3677 | * function later on until it does return @true. Alternatively, the caller |
3678 | * can explicitly wait for a grace period, for example, by passing @oldstate | |
3679 | * to cond_synchronize_rcu() or by directly invoking synchronize_rcu(). | |
3680 | * | |
3681 | * Yes, this function does not take counter wrap into account. | |
3682 | * But counter wrap is harmless. If the counter wraps, we have waited for | |
2403e804 | 3683 | * more than a billion grace periods (and way more on a 64-bit system!). |
f21e0143 | 3684 | * Those needing to keep old state values for very long time periods |
91a967fd PM |
3685 | * (many hours even on 32-bit systems) should check them occasionally and |
3686 | * either refresh them or set a flag indicating that the grace period has | |
3687 | * completed. Alternatively, they can use get_completed_synchronize_rcu() | |
3688 | * to get a guaranteed-completed grace-period state. | |
3d3a0d1b PM |
3689 | * |
3690 | * This function provides the same memory-ordering guarantees that | |
3691 | * would be provided by a synchronize_rcu() that was invoked at the call | |
3692 | * to the function that provided @oldstate, and that returned at the end | |
3693 | * of this function. | |
7abb18bd PM |
3694 | */ |
3695 | bool poll_state_synchronize_rcu(unsigned long oldstate) | |
3696 | { | |
414c1238 | 3697 | if (oldstate == RCU_GET_STATE_COMPLETED || |
bf95b2bc | 3698 | rcu_seq_done_exact(&rcu_state.gp_seq_polled, oldstate)) { |
7abb18bd PM |
3699 | smp_mb(); /* Ensure GP ends before subsequent accesses. */ |
3700 | return true; | |
3701 | } | |
3702 | return false; | |
3703 | } | |
3704 | EXPORT_SYMBOL_GPL(poll_state_synchronize_rcu); | |
3705 | ||
765a3f4f | 3706 | /** |
91a967fd PM |
3707 | * poll_state_synchronize_rcu_full - Has the specified RCU grace period completed? |
3708 | * @rgosp: value from get_state_synchronize_rcu_full() or start_poll_synchronize_rcu_full() | |
765a3f4f | 3709 | * |
91a967fd PM |
3710 | * If a full RCU grace period has elapsed since the earlier call from |
3711 | * which *rgosp was obtained, return @true, otherwise return @false. | |
3712 | * If @false is returned, it is the caller's responsibility to invoke this | |
3713 | * function later on until it does return @true. Alternatively, the caller | |
3714 | * can explicitly wait for a grace period, for example, by passing @rgosp | |
3715 | * to cond_synchronize_rcu() or by directly invoking synchronize_rcu(). | |
3716 | * | |
3717 | * Yes, this function does not take counter wrap into account. | |
3718 | * But counter wrap is harmless. If the counter wraps, we have waited | |
3719 | * for more than a billion grace periods (and way more on a 64-bit | |
3720 | * system!). Those needing to keep rcu_gp_oldstate values for very | |
3721 | * long time periods (many hours even on 32-bit systems) should check | |
3722 | * them occasionally and either refresh them or set a flag indicating | |
3723 | * that the grace period has completed. Alternatively, they can use | |
3724 | * get_completed_synchronize_rcu_full() to get a guaranteed-completed | |
3725 | * grace-period state. | |
765a3f4f | 3726 | * |
91a967fd PM |
3727 | * This function provides the same memory-ordering guarantees that would |
3728 | * be provided by a synchronize_rcu() that was invoked at the call to | |
3729 | * the function that provided @rgosp, and that returned at the end of this | |
3730 | * function. And this guarantee requires that the root rcu_node structure's | |
3731 | * ->gp_seq field be checked instead of that of the rcu_state structure. | |
3732 | * The problem is that the just-ending grace-period's callbacks can be | |
3733 | * invoked between the time that the root rcu_node structure's ->gp_seq | |
3734 | * field is updated and the time that the rcu_state structure's ->gp_seq | |
3735 | * field is updated. Therefore, if a single synchronize_rcu() is to | |
3736 | * cause a subsequent poll_state_synchronize_rcu_full() to return @true, | |
3737 | * then the root rcu_node structure is the one that needs to be polled. | |
3738 | */ | |
3739 | bool poll_state_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp) | |
3740 | { | |
3741 | struct rcu_node *rnp = rcu_get_root(); | |
3742 | ||
3743 | smp_mb(); // Order against root rcu_node structure grace-period cleanup. | |
3744 | if (rgosp->rgos_norm == RCU_GET_STATE_COMPLETED || | |
3745 | rcu_seq_done_exact(&rnp->gp_seq, rgosp->rgos_norm) || | |
3746 | rgosp->rgos_exp == RCU_GET_STATE_COMPLETED || | |
7ecef087 | 3747 | rcu_seq_done_exact(&rcu_state.expedited_sequence, rgosp->rgos_exp)) { |
91a967fd PM |
3748 | smp_mb(); /* Ensure GP ends before subsequent accesses. */ |
3749 | return true; | |
3750 | } | |
3751 | return false; | |
3752 | } | |
3753 | EXPORT_SYMBOL_GPL(poll_state_synchronize_rcu_full); | |
3754 | ||
765a3f4f PM |
3755 | /** |
3756 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
d96c52fe | 3757 | * @oldstate: value from get_state_synchronize_rcu(), start_poll_synchronize_rcu(), or start_poll_synchronize_rcu_expedited() |
765a3f4f PM |
3758 | * |
3759 | * If a full RCU grace period has elapsed since the earlier call to | |
7abb18bd PM |
3760 | * get_state_synchronize_rcu() or start_poll_synchronize_rcu(), just return. |
3761 | * Otherwise, invoke synchronize_rcu() to wait for a full grace period. | |
765a3f4f | 3762 | * |
d96c52fe PM |
3763 | * Yes, this function does not take counter wrap into account. |
3764 | * But counter wrap is harmless. If the counter wraps, we have waited for | |
765a3f4f | 3765 | * more than 2 billion grace periods (and way more on a 64-bit system!), |
d96c52fe | 3766 | * so waiting for a couple of additional grace periods should be just fine. |
3d3a0d1b PM |
3767 | * |
3768 | * This function provides the same memory-ordering guarantees that | |
3769 | * would be provided by a synchronize_rcu() that was invoked at the call | |
d96c52fe | 3770 | * to the function that provided @oldstate and that returned at the end |
3d3a0d1b | 3771 | * of this function. |
765a3f4f PM |
3772 | */ |
3773 | void cond_synchronize_rcu(unsigned long oldstate) | |
3774 | { | |
7abb18bd | 3775 | if (!poll_state_synchronize_rcu(oldstate)) |
765a3f4f PM |
3776 | synchronize_rcu(); |
3777 | } | |
3778 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
3779 | ||
b6fe4917 PM |
3780 | /** |
3781 | * cond_synchronize_rcu_full - Conditionally wait for an RCU grace period | |
3782 | * @rgosp: value from get_state_synchronize_rcu_full(), start_poll_synchronize_rcu_full(), or start_poll_synchronize_rcu_expedited_full() | |
3783 | * | |
3784 | * If a full RCU grace period has elapsed since the call to | |
3785 | * get_state_synchronize_rcu_full(), start_poll_synchronize_rcu_full(), | |
3786 | * or start_poll_synchronize_rcu_expedited_full() from which @rgosp was | |
3787 | * obtained, just return. Otherwise, invoke synchronize_rcu() to wait | |
3788 | * for a full grace period. | |
3789 | * | |
3790 | * Yes, this function does not take counter wrap into account. | |
3791 | * But counter wrap is harmless. If the counter wraps, we have waited for | |
3792 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3793 | * so waiting for a couple of additional grace periods should be just fine. | |
3794 | * | |
3795 | * This function provides the same memory-ordering guarantees that | |
3796 | * would be provided by a synchronize_rcu() that was invoked at the call | |
3797 | * to the function that provided @rgosp and that returned at the end of | |
3798 | * this function. | |
3799 | */ | |
3800 | void cond_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp) | |
3801 | { | |
3802 | if (!poll_state_synchronize_rcu_full(rgosp)) | |
3803 | synchronize_rcu(); | |
3804 | } | |
3805 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu_full); | |
3806 | ||
64db4cff | 3807 | /* |
98ece508 | 3808 | * Check to see if there is any immediate RCU-related work to be done by |
49918a54 PM |
3809 | * the current CPU, returning 1 if so and zero otherwise. The checks are |
3810 | * in order of increasing expense: checks that can be carried out against | |
3811 | * CPU-local state are performed first. However, we must check for CPU | |
3812 | * stalls first, else we might not get a chance. | |
64db4cff | 3813 | */ |
dd7dafd1 | 3814 | static int rcu_pending(int user) |
64db4cff | 3815 | { |
ed93dfc6 | 3816 | bool gp_in_progress; |
98ece508 | 3817 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
2f51f988 PM |
3818 | struct rcu_node *rnp = rdp->mynode; |
3819 | ||
a649d25d PM |
3820 | lockdep_assert_irqs_disabled(); |
3821 | ||
64db4cff | 3822 | /* Check for CPU stalls, if enabled. */ |
ea12ff2b | 3823 | check_cpu_stall(rdp); |
64db4cff | 3824 | |
85f69b32 | 3825 | /* Does this CPU need a deferred NOCB wakeup? */ |
87090516 | 3826 | if (rcu_nocb_need_deferred_wakeup(rdp, RCU_NOCB_WAKE)) |
85f69b32 PM |
3827 | return 1; |
3828 | ||
dd7dafd1 PM |
3829 | /* Is this a nohz_full CPU in userspace or idle? (Ignore RCU if so.) */ |
3830 | if ((user || rcu_is_cpu_rrupt_from_idle()) && rcu_nohz_full_cpu()) | |
a096932f PM |
3831 | return 0; |
3832 | ||
64db4cff | 3833 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
ed93dfc6 PM |
3834 | gp_in_progress = rcu_gp_in_progress(); |
3835 | if (rdp->core_needs_qs && !rdp->cpu_no_qs.b.norm && gp_in_progress) | |
64db4cff PM |
3836 | return 1; |
3837 | ||
3838 | /* Does this CPU have callbacks ready to invoke? */ | |
3820b513 | 3839 | if (!rcu_rdp_is_offloaded(rdp) && |
bd56e0a4 | 3840 | rcu_segcblist_ready_cbs(&rdp->cblist)) |
64db4cff PM |
3841 | return 1; |
3842 | ||
3843 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
ed93dfc6 | 3844 | if (!gp_in_progress && rcu_segcblist_is_enabled(&rdp->cblist) && |
3820b513 | 3845 | !rcu_rdp_is_offloaded(rdp) && |
c1935209 | 3846 | !rcu_segcblist_restempty(&rdp->cblist, RCU_NEXT_READY_TAIL)) |
64db4cff PM |
3847 | return 1; |
3848 | ||
67e14c1e PM |
3849 | /* Have RCU grace period completed or started? */ |
3850 | if (rcu_seq_current(&rnp->gp_seq) != rdp->gp_seq || | |
01c495f7 | 3851 | unlikely(READ_ONCE(rdp->gpwrap))) /* outside lock */ |
64db4cff PM |
3852 | return 1; |
3853 | ||
64db4cff PM |
3854 | /* nothing to do */ |
3855 | return 0; | |
3856 | } | |
3857 | ||
a83eff0a | 3858 | /* |
dd46a788 | 3859 | * Helper function for rcu_barrier() tracing. If tracing is disabled, |
a83eff0a PM |
3860 | * the compiler is expected to optimize this away. |
3861 | */ | |
dd46a788 | 3862 | static void rcu_barrier_trace(const char *s, int cpu, unsigned long done) |
a83eff0a | 3863 | { |
8344b871 PM |
3864 | trace_rcu_barrier(rcu_state.name, s, cpu, |
3865 | atomic_read(&rcu_state.barrier_cpu_count), done); | |
a83eff0a PM |
3866 | } |
3867 | ||
b1420f1c | 3868 | /* |
dd46a788 PM |
3869 | * RCU callback function for rcu_barrier(). If we are last, wake |
3870 | * up the task executing rcu_barrier(). | |
aa24f937 PM |
3871 | * |
3872 | * Note that the value of rcu_state.barrier_sequence must be captured | |
3873 | * before the atomic_dec_and_test(). Otherwise, if this CPU is not last, | |
3874 | * other CPUs might count the value down to zero before this CPU gets | |
3875 | * around to invoking rcu_barrier_trace(), which might result in bogus | |
3876 | * data from the next instance of rcu_barrier(). | |
b1420f1c | 3877 | */ |
24ebbca8 | 3878 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 3879 | { |
aa24f937 PM |
3880 | unsigned long __maybe_unused s = rcu_state.barrier_sequence; |
3881 | ||
ec9f5835 | 3882 | if (atomic_dec_and_test(&rcu_state.barrier_cpu_count)) { |
aa24f937 | 3883 | rcu_barrier_trace(TPS("LastCB"), -1, s); |
ec9f5835 | 3884 | complete(&rcu_state.barrier_completion); |
a83eff0a | 3885 | } else { |
aa24f937 | 3886 | rcu_barrier_trace(TPS("CB"), -1, s); |
a83eff0a | 3887 | } |
d0ec774c PM |
3888 | } |
3889 | ||
3890 | /* | |
a16578dd | 3891 | * If needed, entrain an rcu_barrier() callback on rdp->cblist. |
d0ec774c | 3892 | */ |
a16578dd | 3893 | static void rcu_barrier_entrain(struct rcu_data *rdp) |
d0ec774c | 3894 | { |
a16578dd PM |
3895 | unsigned long gseq = READ_ONCE(rcu_state.barrier_sequence); |
3896 | unsigned long lseq = READ_ONCE(rdp->barrier_seq_snap); | |
d0ec774c | 3897 | |
80b3fd47 | 3898 | lockdep_assert_held(&rcu_state.barrier_lock); |
a16578dd PM |
3899 | if (rcu_seq_state(lseq) || !rcu_seq_state(gseq) || rcu_seq_ctr(lseq) != rcu_seq_ctr(gseq)) |
3900 | return; | |
dd46a788 | 3901 | rcu_barrier_trace(TPS("IRQ"), -1, rcu_state.barrier_sequence); |
f92c734f PM |
3902 | rdp->barrier_head.func = rcu_barrier_callback; |
3903 | debug_rcu_head_queue(&rdp->barrier_head); | |
5d6742b3 | 3904 | rcu_nocb_lock(rdp); |
d1b222c6 | 3905 | WARN_ON_ONCE(!rcu_nocb_flush_bypass(rdp, NULL, jiffies)); |
77a40f97 | 3906 | if (rcu_segcblist_entrain(&rdp->cblist, &rdp->barrier_head)) { |
ec9f5835 | 3907 | atomic_inc(&rcu_state.barrier_cpu_count); |
f92c734f PM |
3908 | } else { |
3909 | debug_rcu_head_unqueue(&rdp->barrier_head); | |
a16578dd | 3910 | rcu_barrier_trace(TPS("IRQNQ"), -1, rcu_state.barrier_sequence); |
f92c734f | 3911 | } |
5d6742b3 | 3912 | rcu_nocb_unlock(rdp); |
a16578dd PM |
3913 | smp_store_release(&rdp->barrier_seq_snap, gseq); |
3914 | } | |
3915 | ||
3916 | /* | |
3917 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
3918 | */ | |
3919 | static void rcu_barrier_handler(void *cpu_in) | |
3920 | { | |
3921 | uintptr_t cpu = (uintptr_t)cpu_in; | |
3922 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); | |
3923 | ||
3924 | lockdep_assert_irqs_disabled(); | |
3925 | WARN_ON_ONCE(cpu != rdp->cpu); | |
3926 | WARN_ON_ONCE(cpu != smp_processor_id()); | |
80b3fd47 | 3927 | raw_spin_lock(&rcu_state.barrier_lock); |
a16578dd | 3928 | rcu_barrier_entrain(rdp); |
80b3fd47 | 3929 | raw_spin_unlock(&rcu_state.barrier_lock); |
d0ec774c PM |
3930 | } |
3931 | ||
dd46a788 PM |
3932 | /** |
3933 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
3934 | * | |
3935 | * Note that this primitive does not necessarily wait for an RCU grace period | |
3936 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
3937 | * in the system, then rcu_barrier() is within its rights to return | |
3938 | * immediately, without waiting for anything, much less an RCU grace period. | |
d0ec774c | 3939 | */ |
dd46a788 | 3940 | void rcu_barrier(void) |
d0ec774c | 3941 | { |
127e2981 | 3942 | uintptr_t cpu; |
a16578dd PM |
3943 | unsigned long flags; |
3944 | unsigned long gseq; | |
b1420f1c | 3945 | struct rcu_data *rdp; |
ec9f5835 | 3946 | unsigned long s = rcu_seq_snap(&rcu_state.barrier_sequence); |
b1420f1c | 3947 | |
dd46a788 | 3948 | rcu_barrier_trace(TPS("Begin"), -1, s); |
b1420f1c | 3949 | |
e74f4c45 | 3950 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
ec9f5835 | 3951 | mutex_lock(&rcu_state.barrier_mutex); |
b1420f1c | 3952 | |
4f525a52 | 3953 | /* Did someone else do our work for us? */ |
ec9f5835 | 3954 | if (rcu_seq_done(&rcu_state.barrier_sequence, s)) { |
0cabb47a | 3955 | rcu_barrier_trace(TPS("EarlyExit"), -1, rcu_state.barrier_sequence); |
cf3a9c48 | 3956 | smp_mb(); /* caller's subsequent code after above check. */ |
ec9f5835 | 3957 | mutex_unlock(&rcu_state.barrier_mutex); |
cf3a9c48 PM |
3958 | return; |
3959 | } | |
3960 | ||
4f525a52 | 3961 | /* Mark the start of the barrier operation. */ |
80b3fd47 | 3962 | raw_spin_lock_irqsave(&rcu_state.barrier_lock, flags); |
ec9f5835 | 3963 | rcu_seq_start(&rcu_state.barrier_sequence); |
a16578dd | 3964 | gseq = rcu_state.barrier_sequence; |
dd46a788 | 3965 | rcu_barrier_trace(TPS("Inc1"), -1, rcu_state.barrier_sequence); |
b1420f1c | 3966 | |
d0ec774c | 3967 | /* |
127e2981 PM |
3968 | * Initialize the count to two rather than to zero in order |
3969 | * to avoid a too-soon return to zero in case of an immediate | |
3970 | * invocation of the just-enqueued callback (or preemption of | |
3971 | * this task). Exclude CPU-hotplug operations to ensure that no | |
3972 | * offline non-offloaded CPU has callbacks queued. | |
d0ec774c | 3973 | */ |
ec9f5835 | 3974 | init_completion(&rcu_state.barrier_completion); |
127e2981 | 3975 | atomic_set(&rcu_state.barrier_cpu_count, 2); |
80b3fd47 | 3976 | raw_spin_unlock_irqrestore(&rcu_state.barrier_lock, flags); |
b1420f1c PM |
3977 | |
3978 | /* | |
1331e7a1 PM |
3979 | * Force each CPU with callbacks to register a new callback. |
3980 | * When that callback is invoked, we will know that all of the | |
3981 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 3982 | */ |
3fbfbf7a | 3983 | for_each_possible_cpu(cpu) { |
da1df50d | 3984 | rdp = per_cpu_ptr(&rcu_data, cpu); |
a16578dd PM |
3985 | retry: |
3986 | if (smp_load_acquire(&rdp->barrier_seq_snap) == gseq) | |
ce5215c1 | 3987 | continue; |
80b3fd47 | 3988 | raw_spin_lock_irqsave(&rcu_state.barrier_lock, flags); |
0cabb47a | 3989 | if (!rcu_segcblist_n_cbs(&rdp->cblist)) { |
a16578dd | 3990 | WRITE_ONCE(rdp->barrier_seq_snap, gseq); |
80b3fd47 | 3991 | raw_spin_unlock_irqrestore(&rcu_state.barrier_lock, flags); |
0cabb47a | 3992 | rcu_barrier_trace(TPS("NQ"), cpu, rcu_state.barrier_sequence); |
ce5215c1 | 3993 | continue; |
0cabb47a | 3994 | } |
a16578dd PM |
3995 | if (!rcu_rdp_cpu_online(rdp)) { |
3996 | rcu_barrier_entrain(rdp); | |
3997 | WARN_ON_ONCE(READ_ONCE(rdp->barrier_seq_snap) != gseq); | |
80b3fd47 | 3998 | raw_spin_unlock_irqrestore(&rcu_state.barrier_lock, flags); |
0cabb47a | 3999 | rcu_barrier_trace(TPS("OfflineNoCBQ"), cpu, rcu_state.barrier_sequence); |
a16578dd | 4000 | continue; |
b1420f1c | 4001 | } |
80b3fd47 | 4002 | raw_spin_unlock_irqrestore(&rcu_state.barrier_lock, flags); |
a16578dd PM |
4003 | if (smp_call_function_single(cpu, rcu_barrier_handler, (void *)cpu, 1)) { |
4004 | schedule_timeout_uninterruptible(1); | |
4005 | goto retry; | |
b1420f1c | 4006 | } |
a16578dd PM |
4007 | WARN_ON_ONCE(READ_ONCE(rdp->barrier_seq_snap) != gseq); |
4008 | rcu_barrier_trace(TPS("OnlineQ"), cpu, rcu_state.barrier_sequence); | |
b1420f1c | 4009 | } |
b1420f1c PM |
4010 | |
4011 | /* | |
4012 | * Now that we have an rcu_barrier_callback() callback on each | |
4013 | * CPU, and thus each counted, remove the initial count. | |
4014 | */ | |
127e2981 | 4015 | if (atomic_sub_and_test(2, &rcu_state.barrier_cpu_count)) |
ec9f5835 | 4016 | complete(&rcu_state.barrier_completion); |
b1420f1c PM |
4017 | |
4018 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
ec9f5835 | 4019 | wait_for_completion(&rcu_state.barrier_completion); |
b1420f1c | 4020 | |
4f525a52 | 4021 | /* Mark the end of the barrier operation. */ |
dd46a788 | 4022 | rcu_barrier_trace(TPS("Inc2"), -1, rcu_state.barrier_sequence); |
ec9f5835 | 4023 | rcu_seq_end(&rcu_state.barrier_sequence); |
a16578dd PM |
4024 | gseq = rcu_state.barrier_sequence; |
4025 | for_each_possible_cpu(cpu) { | |
4026 | rdp = per_cpu_ptr(&rcu_data, cpu); | |
4027 | ||
4028 | WRITE_ONCE(rdp->barrier_seq_snap, gseq); | |
4029 | } | |
4f525a52 | 4030 | |
b1420f1c | 4031 | /* Other rcu_barrier() invocations can now safely proceed. */ |
ec9f5835 | 4032 | mutex_unlock(&rcu_state.barrier_mutex); |
d0ec774c | 4033 | } |
45975c7d | 4034 | EXPORT_SYMBOL_GPL(rcu_barrier); |
d0ec774c | 4035 | |
0aa04b05 PM |
4036 | /* |
4037 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
4038 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
a616aec9 | 4039 | * must hold the corresponding leaf rcu_node ->lock with interrupts |
0aa04b05 PM |
4040 | * disabled. |
4041 | */ | |
4042 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
4043 | { | |
4044 | long mask; | |
8d672fa6 | 4045 | long oldmask; |
0aa04b05 PM |
4046 | struct rcu_node *rnp = rnp_leaf; |
4047 | ||
8d672fa6 | 4048 | raw_lockdep_assert_held_rcu_node(rnp_leaf); |
962aff03 | 4049 | WARN_ON_ONCE(rnp->wait_blkd_tasks); |
0aa04b05 PM |
4050 | for (;;) { |
4051 | mask = rnp->grpmask; | |
4052 | rnp = rnp->parent; | |
4053 | if (rnp == NULL) | |
4054 | return; | |
6cf10081 | 4055 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
8d672fa6 | 4056 | oldmask = rnp->qsmaskinit; |
0aa04b05 | 4057 | rnp->qsmaskinit |= mask; |
67c583a7 | 4058 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
8d672fa6 PM |
4059 | if (oldmask) |
4060 | return; | |
0aa04b05 PM |
4061 | } |
4062 | } | |
4063 | ||
64db4cff | 4064 | /* |
27569620 | 4065 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 4066 | */ |
27569620 | 4067 | static void __init |
53b46303 | 4068 | rcu_boot_init_percpu_data(int cpu) |
64db4cff | 4069 | { |
904e600e | 4070 | struct context_tracking *ct = this_cpu_ptr(&context_tracking); |
da1df50d | 4071 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27569620 PM |
4072 | |
4073 | /* Set up local state, ensuring consistent view of global state. */ | |
bc75e999 | 4074 | rdp->grpmask = leaf_node_cpu_bit(rdp->mynode, cpu); |
a657f261 | 4075 | INIT_WORK(&rdp->strict_work, strict_work_handler); |
904e600e | 4076 | WARN_ON_ONCE(ct->dynticks_nesting != 1); |
62e2412d | 4077 | WARN_ON_ONCE(rcu_dynticks_in_eqs(rcu_dynticks_snap(cpu))); |
a16578dd | 4078 | rdp->barrier_seq_snap = rcu_state.barrier_sequence; |
53b46303 | 4079 | rdp->rcu_ofl_gp_seq = rcu_state.gp_seq; |
57738942 | 4080 | rdp->rcu_ofl_gp_flags = RCU_GP_CLEANED; |
53b46303 | 4081 | rdp->rcu_onl_gp_seq = rcu_state.gp_seq; |
57738942 | 4082 | rdp->rcu_onl_gp_flags = RCU_GP_CLEANED; |
c708b08c | 4083 | rdp->last_sched_clock = jiffies; |
27569620 | 4084 | rdp->cpu = cpu; |
3fbfbf7a | 4085 | rcu_boot_init_nocb_percpu_data(rdp); |
27569620 PM |
4086 | } |
4087 | ||
4088 | /* | |
53b46303 PM |
4089 | * Invoked early in the CPU-online process, when pretty much all services |
4090 | * are available. The incoming CPU is not present. | |
4091 | * | |
4092 | * Initializes a CPU's per-CPU RCU data. Note that only one online or | |
ff3bb6f4 PM |
4093 | * offline event can be happening at a given time. Note also that we can |
4094 | * accept some slop in the rsp->gp_seq access due to the fact that this | |
e83e73f5 PM |
4095 | * CPU cannot possibly have any non-offloaded RCU callbacks in flight yet. |
4096 | * And any offloaded callbacks are being numbered elsewhere. | |
64db4cff | 4097 | */ |
53b46303 | 4098 | int rcutree_prepare_cpu(unsigned int cpu) |
64db4cff PM |
4099 | { |
4100 | unsigned long flags; | |
904e600e | 4101 | struct context_tracking *ct = per_cpu_ptr(&context_tracking, cpu); |
da1df50d | 4102 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
336a4f6c | 4103 | struct rcu_node *rnp = rcu_get_root(); |
64db4cff PM |
4104 | |
4105 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 4106 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 | 4107 | rdp->qlen_last_fqs_check = 0; |
2431774f | 4108 | rdp->n_force_qs_snap = READ_ONCE(rcu_state.n_force_qs); |
64db4cff | 4109 | rdp->blimit = blimit; |
904e600e | 4110 | ct->dynticks_nesting = 1; /* CPU not up, no tearing. */ |
67c583a7 | 4111 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
ec711bc1 | 4112 | |
126d9d49 | 4113 | /* |
ec711bc1 FW |
4114 | * Only non-NOCB CPUs that didn't have early-boot callbacks need to be |
4115 | * (re-)initialized. | |
126d9d49 | 4116 | */ |
ec711bc1 | 4117 | if (!rcu_segcblist_is_enabled(&rdp->cblist)) |
126d9d49 | 4118 | rcu_segcblist_init(&rdp->cblist); /* Re-enable callbacks. */ |
64db4cff | 4119 | |
0aa04b05 PM |
4120 | /* |
4121 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
4122 | * propagation up the rcu_node tree will happen at the beginning | |
4123 | * of the next grace period. | |
4124 | */ | |
64db4cff | 4125 | rnp = rdp->mynode; |
2a67e741 | 4126 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
b9585e94 | 4127 | rdp->beenonline = true; /* We have now been online. */ |
8ff37290 PM |
4128 | rdp->gp_seq = READ_ONCE(rnp->gp_seq); |
4129 | rdp->gp_seq_needed = rdp->gp_seq; | |
5b74c458 | 4130 | rdp->cpu_no_qs.b.norm = true; |
97c668b8 | 4131 | rdp->core_needs_qs = false; |
9b9500da | 4132 | rdp->rcu_iw_pending = false; |
7a9f50a0 | 4133 | rdp->rcu_iw = IRQ_WORK_INIT_HARD(rcu_iw_handler); |
8ff37290 | 4134 | rdp->rcu_iw_gp_seq = rdp->gp_seq - 1; |
53b46303 | 4135 | trace_rcu_grace_period(rcu_state.name, rdp->gp_seq, TPS("cpuonl")); |
67c583a7 | 4136 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
3ef5a1c3 | 4137 | rcu_spawn_one_boost_kthread(rnp); |
ad368d15 | 4138 | rcu_spawn_cpu_nocb_kthread(cpu); |
ed73860c | 4139 | WRITE_ONCE(rcu_state.n_online_cpus, rcu_state.n_online_cpus + 1); |
4df83742 TG |
4140 | |
4141 | return 0; | |
4142 | } | |
4143 | ||
deb34f36 PM |
4144 | /* |
4145 | * Update RCU priority boot kthread affinity for CPU-hotplug changes. | |
4146 | */ | |
4df83742 TG |
4147 | static void rcutree_affinity_setting(unsigned int cpu, int outgoing) |
4148 | { | |
da1df50d | 4149 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
4df83742 TG |
4150 | |
4151 | rcu_boost_kthread_setaffinity(rdp->mynode, outgoing); | |
4152 | } | |
4153 | ||
deb34f36 PM |
4154 | /* |
4155 | * Near the end of the CPU-online process. Pretty much all services | |
4156 | * enabled, and the CPU is now very much alive. | |
4157 | */ | |
4df83742 TG |
4158 | int rcutree_online_cpu(unsigned int cpu) |
4159 | { | |
9b9500da PM |
4160 | unsigned long flags; |
4161 | struct rcu_data *rdp; | |
4162 | struct rcu_node *rnp; | |
9b9500da | 4163 | |
b97d23c5 PM |
4164 | rdp = per_cpu_ptr(&rcu_data, cpu); |
4165 | rnp = rdp->mynode; | |
4166 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
4167 | rnp->ffmask |= rdp->grpmask; | |
4168 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
9b9500da PM |
4169 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
4170 | return 0; /* Too early in boot for scheduler work. */ | |
4171 | sync_sched_exp_online_cleanup(cpu); | |
4172 | rcutree_affinity_setting(cpu, -1); | |
96926686 PM |
4173 | |
4174 | // Stop-machine done, so allow nohz_full to disable tick. | |
4175 | tick_dep_clear(TICK_DEP_BIT_RCU); | |
4df83742 TG |
4176 | return 0; |
4177 | } | |
4178 | ||
deb34f36 PM |
4179 | /* |
4180 | * Near the beginning of the process. The CPU is still very much alive | |
4181 | * with pretty much all services enabled. | |
4182 | */ | |
4df83742 TG |
4183 | int rcutree_offline_cpu(unsigned int cpu) |
4184 | { | |
9b9500da PM |
4185 | unsigned long flags; |
4186 | struct rcu_data *rdp; | |
4187 | struct rcu_node *rnp; | |
9b9500da | 4188 | |
b97d23c5 PM |
4189 | rdp = per_cpu_ptr(&rcu_data, cpu); |
4190 | rnp = rdp->mynode; | |
4191 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
4192 | rnp->ffmask &= ~rdp->grpmask; | |
4193 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
9b9500da | 4194 | |
4df83742 | 4195 | rcutree_affinity_setting(cpu, cpu); |
96926686 PM |
4196 | |
4197 | // nohz_full CPUs need the tick for stop-machine to work quickly | |
4198 | tick_dep_set(TICK_DEP_BIT_RCU); | |
4df83742 TG |
4199 | return 0; |
4200 | } | |
4201 | ||
7ec99de3 PM |
4202 | /* |
4203 | * Mark the specified CPU as being online so that subsequent grace periods | |
4204 | * (both expedited and normal) will wait on it. Note that this means that | |
4205 | * incoming CPUs are not allowed to use RCU read-side critical sections | |
4206 | * until this function is called. Failing to observe this restriction | |
4207 | * will result in lockdep splats. | |
deb34f36 PM |
4208 | * |
4209 | * Note that this function is special in that it is invoked directly | |
4210 | * from the incoming CPU rather than from the cpuhp_step mechanism. | |
4211 | * This is because this function must be invoked at a precise location. | |
7ec99de3 PM |
4212 | */ |
4213 | void rcu_cpu_starting(unsigned int cpu) | |
4214 | { | |
4215 | unsigned long flags; | |
4216 | unsigned long mask; | |
4217 | struct rcu_data *rdp; | |
4218 | struct rcu_node *rnp; | |
abfce041 | 4219 | bool newcpu; |
7ec99de3 | 4220 | |
c0f97f20 PM |
4221 | rdp = per_cpu_ptr(&rcu_data, cpu); |
4222 | if (rdp->cpu_started) | |
f64c6013 | 4223 | return; |
c0f97f20 | 4224 | rdp->cpu_started = true; |
f64c6013 | 4225 | |
b97d23c5 PM |
4226 | rnp = rdp->mynode; |
4227 | mask = rdp->grpmask; | |
82980b16 DW |
4228 | local_irq_save(flags); |
4229 | arch_spin_lock(&rcu_state.ofl_lock); | |
2caebefb | 4230 | rcu_dynticks_eqs_online(); |
80b3fd47 | 4231 | raw_spin_lock(&rcu_state.barrier_lock); |
82980b16 | 4232 | raw_spin_lock_rcu_node(rnp); |
105abf82 | 4233 | WRITE_ONCE(rnp->qsmaskinitnext, rnp->qsmaskinitnext | mask); |
80b3fd47 | 4234 | raw_spin_unlock(&rcu_state.barrier_lock); |
abfce041 | 4235 | newcpu = !(rnp->expmaskinitnext & mask); |
b97d23c5 | 4236 | rnp->expmaskinitnext |= mask; |
b97d23c5 | 4237 | /* Allow lockless access for expedited grace periods. */ |
abfce041 | 4238 | smp_store_release(&rcu_state.ncpus, rcu_state.ncpus + newcpu); /* ^^^ */ |
2f084695 | 4239 | ASSERT_EXCLUSIVE_WRITER(rcu_state.ncpus); |
b97d23c5 | 4240 | rcu_gpnum_ovf(rnp, rdp); /* Offline-induced counter wrap? */ |
eb7a6653 PM |
4241 | rdp->rcu_onl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
4242 | rdp->rcu_onl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
9f866dac JFG |
4243 | |
4244 | /* An incoming CPU should never be blocking a grace period. */ | |
4245 | if (WARN_ON_ONCE(rnp->qsmask & mask)) { /* RCU waiting on incoming CPU? */ | |
82980b16 DW |
4246 | /* rcu_report_qs_rnp() *really* wants some flags to restore */ |
4247 | unsigned long flags2; | |
4248 | ||
4249 | local_irq_save(flags2); | |
516e5ae0 | 4250 | rcu_disable_urgency_upon_qs(rdp); |
b97d23c5 | 4251 | /* Report QS -after- changing ->qsmaskinitnext! */ |
82980b16 | 4252 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags2); |
b97d23c5 | 4253 | } else { |
82980b16 | 4254 | raw_spin_unlock_rcu_node(rnp); |
7ec99de3 | 4255 | } |
82980b16 DW |
4256 | arch_spin_unlock(&rcu_state.ofl_lock); |
4257 | local_irq_restore(flags); | |
313517fc | 4258 | smp_mb(); /* Ensure RCU read-side usage follows above initialization. */ |
7ec99de3 PM |
4259 | } |
4260 | ||
27d50c7e | 4261 | /* |
53b46303 PM |
4262 | * The outgoing function has no further need of RCU, so remove it from |
4263 | * the rcu_node tree's ->qsmaskinitnext bit masks. | |
4264 | * | |
4265 | * Note that this function is special in that it is invoked directly | |
4266 | * from the outgoing CPU rather than from the cpuhp_step mechanism. | |
4267 | * This is because this function must be invoked at a precise location. | |
27d50c7e | 4268 | */ |
53b46303 | 4269 | void rcu_report_dead(unsigned int cpu) |
27d50c7e | 4270 | { |
82980b16 | 4271 | unsigned long flags, seq_flags; |
27d50c7e | 4272 | unsigned long mask; |
da1df50d | 4273 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
27d50c7e TG |
4274 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
4275 | ||
147c6852 PM |
4276 | // Do any dangling deferred wakeups. |
4277 | do_nocb_deferred_wakeup(rdp); | |
4278 | ||
49918a54 | 4279 | /* QS for any half-done expedited grace period. */ |
768f5d50 | 4280 | rcu_report_exp_rdp(rdp); |
53b46303 PM |
4281 | rcu_preempt_deferred_qs(current); |
4282 | ||
27d50c7e TG |
4283 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ |
4284 | mask = rdp->grpmask; | |
82980b16 DW |
4285 | local_irq_save(seq_flags); |
4286 | arch_spin_lock(&rcu_state.ofl_lock); | |
27d50c7e | 4287 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ |
53b46303 PM |
4288 | rdp->rcu_ofl_gp_seq = READ_ONCE(rcu_state.gp_seq); |
4289 | rdp->rcu_ofl_gp_flags = READ_ONCE(rcu_state.gp_flags); | |
fece2776 PM |
4290 | if (rnp->qsmask & mask) { /* RCU waiting on outgoing CPU? */ |
4291 | /* Report quiescent state -before- changing ->qsmaskinitnext! */ | |
e2bb1288 | 4292 | rcu_disable_urgency_upon_qs(rdp); |
b50912d0 | 4293 | rcu_report_qs_rnp(mask, rnp, rnp->gp_seq, flags); |
fece2776 PM |
4294 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
4295 | } | |
105abf82 | 4296 | WRITE_ONCE(rnp->qsmaskinitnext, rnp->qsmaskinitnext & ~mask); |
710d60cb | 4297 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
82980b16 DW |
4298 | arch_spin_unlock(&rcu_state.ofl_lock); |
4299 | local_irq_restore(seq_flags); | |
f64c6013 | 4300 | |
c0f97f20 | 4301 | rdp->cpu_started = false; |
27d50c7e | 4302 | } |
a58163d8 | 4303 | |
04e613de | 4304 | #ifdef CONFIG_HOTPLUG_CPU |
53b46303 PM |
4305 | /* |
4306 | * The outgoing CPU has just passed through the dying-idle state, and we | |
4307 | * are being invoked from the CPU that was IPIed to continue the offline | |
4308 | * operation. Migrate the outgoing CPU's callbacks to the current CPU. | |
4309 | */ | |
4310 | void rcutree_migrate_callbacks(int cpu) | |
a58163d8 PM |
4311 | { |
4312 | unsigned long flags; | |
b1a2d79f | 4313 | struct rcu_data *my_rdp; |
c00045be | 4314 | struct rcu_node *my_rnp; |
da1df50d | 4315 | struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu); |
ec4eacce | 4316 | bool needwake; |
a58163d8 | 4317 | |
3820b513 | 4318 | if (rcu_rdp_is_offloaded(rdp) || |
ce5215c1 | 4319 | rcu_segcblist_empty(&rdp->cblist)) |
95335c03 PM |
4320 | return; /* No callbacks to migrate. */ |
4321 | ||
80b3fd47 | 4322 | raw_spin_lock_irqsave(&rcu_state.barrier_lock, flags); |
a16578dd PM |
4323 | WARN_ON_ONCE(rcu_rdp_cpu_online(rdp)); |
4324 | rcu_barrier_entrain(rdp); | |
da1df50d | 4325 | my_rdp = this_cpu_ptr(&rcu_data); |
c00045be | 4326 | my_rnp = my_rdp->mynode; |
5d6742b3 | 4327 | rcu_nocb_lock(my_rdp); /* irqs already disabled. */ |
d1b222c6 | 4328 | WARN_ON_ONCE(!rcu_nocb_flush_bypass(my_rdp, NULL, jiffies)); |
c00045be | 4329 | raw_spin_lock_rcu_node(my_rnp); /* irqs already disabled. */ |
ec4eacce | 4330 | /* Leverage recent GPs and set GP for new callbacks. */ |
c00045be PM |
4331 | needwake = rcu_advance_cbs(my_rnp, rdp) || |
4332 | rcu_advance_cbs(my_rnp, my_rdp); | |
f2dbe4a5 | 4333 | rcu_segcblist_merge(&my_rdp->cblist, &rdp->cblist); |
80b3fd47 | 4334 | raw_spin_unlock(&rcu_state.barrier_lock); /* irqs remain disabled. */ |
23651d9b | 4335 | needwake = needwake || rcu_advance_cbs(my_rnp, my_rdp); |
c035280f | 4336 | rcu_segcblist_disable(&rdp->cblist); |
a16578dd | 4337 | WARN_ON_ONCE(rcu_segcblist_empty(&my_rdp->cblist) != !rcu_segcblist_n_cbs(&my_rdp->cblist)); |
52c1d81e | 4338 | check_cb_ovld_locked(my_rdp, my_rnp); |
3820b513 | 4339 | if (rcu_rdp_is_offloaded(my_rdp)) { |
5d6742b3 PM |
4340 | raw_spin_unlock_rcu_node(my_rnp); /* irqs remain disabled. */ |
4341 | __call_rcu_nocb_wake(my_rdp, true, flags); | |
4342 | } else { | |
4343 | rcu_nocb_unlock(my_rdp); /* irqs remain disabled. */ | |
4344 | raw_spin_unlock_irqrestore_rcu_node(my_rnp, flags); | |
4345 | } | |
ec4eacce | 4346 | if (needwake) |
532c00c9 | 4347 | rcu_gp_kthread_wake(); |
5d6742b3 | 4348 | lockdep_assert_irqs_enabled(); |
a58163d8 PM |
4349 | WARN_ONCE(rcu_segcblist_n_cbs(&rdp->cblist) != 0 || |
4350 | !rcu_segcblist_empty(&rdp->cblist), | |
4351 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, 1stCB=%p\n", | |
4352 | cpu, rcu_segcblist_n_cbs(&rdp->cblist), | |
4353 | rcu_segcblist_first_cb(&rdp->cblist)); | |
4354 | } | |
27d50c7e TG |
4355 | #endif |
4356 | ||
deb34f36 PM |
4357 | /* |
4358 | * On non-huge systems, use expedited RCU grace periods to make suspend | |
4359 | * and hibernation run faster. | |
4360 | */ | |
d1d74d14 BP |
4361 | static int rcu_pm_notify(struct notifier_block *self, |
4362 | unsigned long action, void *hcpu) | |
4363 | { | |
4364 | switch (action) { | |
4365 | case PM_HIBERNATION_PREPARE: | |
4366 | case PM_SUSPEND_PREPARE: | |
e85e6a21 | 4367 | rcu_expedite_gp(); |
d1d74d14 BP |
4368 | break; |
4369 | case PM_POST_HIBERNATION: | |
4370 | case PM_POST_SUSPEND: | |
e85e6a21 | 4371 | rcu_unexpedite_gp(); |
d1d74d14 BP |
4372 | break; |
4373 | default: | |
4374 | break; | |
4375 | } | |
4376 | return NOTIFY_OK; | |
4377 | } | |
4378 | ||
9621fbee KS |
4379 | #ifdef CONFIG_RCU_EXP_KTHREAD |
4380 | struct kthread_worker *rcu_exp_gp_kworker; | |
4381 | struct kthread_worker *rcu_exp_par_gp_kworker; | |
4382 | ||
4383 | static void __init rcu_start_exp_gp_kworkers(void) | |
4384 | { | |
4385 | const char *par_gp_kworker_name = "rcu_exp_par_gp_kthread_worker"; | |
4386 | const char *gp_kworker_name = "rcu_exp_gp_kthread_worker"; | |
4387 | struct sched_param param = { .sched_priority = kthread_prio }; | |
4388 | ||
4389 | rcu_exp_gp_kworker = kthread_create_worker(0, gp_kworker_name); | |
4390 | if (IS_ERR_OR_NULL(rcu_exp_gp_kworker)) { | |
4391 | pr_err("Failed to create %s!\n", gp_kworker_name); | |
4392 | return; | |
4393 | } | |
4394 | ||
4395 | rcu_exp_par_gp_kworker = kthread_create_worker(0, par_gp_kworker_name); | |
4396 | if (IS_ERR_OR_NULL(rcu_exp_par_gp_kworker)) { | |
4397 | pr_err("Failed to create %s!\n", par_gp_kworker_name); | |
4398 | kthread_destroy_worker(rcu_exp_gp_kworker); | |
4399 | return; | |
4400 | } | |
4401 | ||
4402 | sched_setscheduler_nocheck(rcu_exp_gp_kworker->task, SCHED_FIFO, ¶m); | |
4403 | sched_setscheduler_nocheck(rcu_exp_par_gp_kworker->task, SCHED_FIFO, | |
4404 | ¶m); | |
4405 | } | |
4406 | ||
4407 | static inline void rcu_alloc_par_gp_wq(void) | |
4408 | { | |
4409 | } | |
4410 | #else /* !CONFIG_RCU_EXP_KTHREAD */ | |
4411 | struct workqueue_struct *rcu_par_gp_wq; | |
4412 | ||
4413 | static void __init rcu_start_exp_gp_kworkers(void) | |
4414 | { | |
4415 | } | |
4416 | ||
4417 | static inline void rcu_alloc_par_gp_wq(void) | |
4418 | { | |
4419 | rcu_par_gp_wq = alloc_workqueue("rcu_par_gp", WQ_MEM_RECLAIM, 0); | |
4420 | WARN_ON(!rcu_par_gp_wq); | |
4421 | } | |
4422 | #endif /* CONFIG_RCU_EXP_KTHREAD */ | |
4423 | ||
b3dbec76 | 4424 | /* |
49918a54 | 4425 | * Spawn the kthreads that handle RCU's grace periods. |
b3dbec76 PM |
4426 | */ |
4427 | static int __init rcu_spawn_gp_kthread(void) | |
4428 | { | |
4429 | unsigned long flags; | |
4430 | struct rcu_node *rnp; | |
a94844b2 | 4431 | struct sched_param sp; |
b3dbec76 | 4432 | struct task_struct *t; |
3352911f | 4433 | struct rcu_data *rdp = this_cpu_ptr(&rcu_data); |
b3dbec76 | 4434 | |
9386c0b7 | 4435 | rcu_scheduler_fully_active = 1; |
b97d23c5 | 4436 | t = kthread_create(rcu_gp_kthread, NULL, "%s", rcu_state.name); |
08543bda PM |
4437 | if (WARN_ONCE(IS_ERR(t), "%s: Could not start grace-period kthread, OOM is now expected behavior\n", __func__)) |
4438 | return 0; | |
b97d23c5 PM |
4439 | if (kthread_prio) { |
4440 | sp.sched_priority = kthread_prio; | |
4441 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
b3dbec76 | 4442 | } |
b97d23c5 PM |
4443 | rnp = rcu_get_root(); |
4444 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
5648d659 PM |
4445 | WRITE_ONCE(rcu_state.gp_activity, jiffies); |
4446 | WRITE_ONCE(rcu_state.gp_req_activity, jiffies); | |
4447 | // Reset .gp_activity and .gp_req_activity before setting .gp_kthread. | |
4448 | smp_store_release(&rcu_state.gp_kthread, t); /* ^^^ */ | |
b97d23c5 PM |
4449 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
4450 | wake_up_process(t); | |
3352911f FW |
4451 | /* This is a pre-SMP initcall, we expect a single CPU */ |
4452 | WARN_ON(num_online_cpus() > 1); | |
87c5adf0 FW |
4453 | /* |
4454 | * Those kthreads couldn't be created on rcu_init() -> rcutree_prepare_cpu() | |
4455 | * due to rcu_scheduler_fully_active. | |
4456 | */ | |
4457 | rcu_spawn_cpu_nocb_kthread(smp_processor_id()); | |
3352911f | 4458 | rcu_spawn_one_boost_kthread(rdp->mynode); |
8e4b1d2b | 4459 | rcu_spawn_core_kthreads(); |
9621fbee KS |
4460 | /* Create kthread worker for expedited GPs */ |
4461 | rcu_start_exp_gp_kworkers(); | |
b3dbec76 PM |
4462 | return 0; |
4463 | } | |
4464 | early_initcall(rcu_spawn_gp_kthread); | |
4465 | ||
bbad9379 | 4466 | /* |
52d7e48b PM |
4467 | * This function is invoked towards the end of the scheduler's |
4468 | * initialization process. Before this is called, the idle task might | |
4469 | * contain synchronous grace-period primitives (during which time, this idle | |
4470 | * task is booting the system, and such primitives are no-ops). After this | |
4471 | * function is called, any synchronous grace-period primitives are run as | |
4472 | * expedited, with the requesting task driving the grace period forward. | |
900b1028 | 4473 | * A later core_initcall() rcu_set_runtime_mode() will switch to full |
52d7e48b | 4474 | * runtime RCU functionality. |
bbad9379 PM |
4475 | */ |
4476 | void rcu_scheduler_starting(void) | |
4477 | { | |
d761de8a PM |
4478 | unsigned long flags; |
4479 | struct rcu_node *rnp; | |
4480 | ||
bbad9379 PM |
4481 | WARN_ON(num_online_cpus() != 1); |
4482 | WARN_ON(nr_context_switches() > 0); | |
52d7e48b | 4483 | rcu_test_sync_prims(); |
d761de8a PM |
4484 | |
4485 | // Fix up the ->gp_seq counters. | |
4486 | local_irq_save(flags); | |
4487 | rcu_for_each_node_breadth_first(rnp) | |
4488 | rnp->gp_seq_needed = rnp->gp_seq = rcu_state.gp_seq; | |
4489 | local_irq_restore(flags); | |
4490 | ||
4491 | // Switch out of early boot mode. | |
52d7e48b PM |
4492 | rcu_scheduler_active = RCU_SCHEDULER_INIT; |
4493 | rcu_test_sync_prims(); | |
bbad9379 PM |
4494 | } |
4495 | ||
64db4cff | 4496 | /* |
49918a54 | 4497 | * Helper function for rcu_init() that initializes the rcu_state structure. |
64db4cff | 4498 | */ |
b8bb1f63 | 4499 | static void __init rcu_init_one(void) |
64db4cff | 4500 | { |
cb007102 AG |
4501 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
4502 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
4503 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
4504 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
199977bf | 4505 | |
199977bf | 4506 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ |
64db4cff PM |
4507 | int cpustride = 1; |
4508 | int i; | |
4509 | int j; | |
4510 | struct rcu_node *rnp; | |
4511 | ||
05b84aec | 4512 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 4513 | |
3eaaaf6c PM |
4514 | /* Silence gcc 4.8 false positive about array index out of range. */ |
4515 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
4516 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 4517 | |
64db4cff PM |
4518 | /* Initialize the level-tracking arrays. */ |
4519 | ||
f885b7f2 | 4520 | for (i = 1; i < rcu_num_lvls; i++) |
eb7a6653 PM |
4521 | rcu_state.level[i] = |
4522 | rcu_state.level[i - 1] + num_rcu_lvl[i - 1]; | |
41f5c631 | 4523 | rcu_init_levelspread(levelspread, num_rcu_lvl); |
64db4cff PM |
4524 | |
4525 | /* Initialize the elements themselves, starting from the leaves. */ | |
4526 | ||
f885b7f2 | 4527 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 4528 | cpustride *= levelspread[i]; |
eb7a6653 | 4529 | rnp = rcu_state.level[i]; |
41f5c631 | 4530 | for (j = 0; j < num_rcu_lvl[i]; j++, rnp++) { |
67c583a7 BF |
4531 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
4532 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 4533 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
4534 | raw_spin_lock_init(&rnp->fqslock); |
4535 | lockdep_set_class_and_name(&rnp->fqslock, | |
4536 | &rcu_fqs_class[i], fqs[i]); | |
eb7a6653 PM |
4537 | rnp->gp_seq = rcu_state.gp_seq; |
4538 | rnp->gp_seq_needed = rcu_state.gp_seq; | |
4539 | rnp->completedqs = rcu_state.gp_seq; | |
64db4cff PM |
4540 | rnp->qsmask = 0; |
4541 | rnp->qsmaskinit = 0; | |
4542 | rnp->grplo = j * cpustride; | |
4543 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
4544 | if (rnp->grphi >= nr_cpu_ids) |
4545 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
4546 | if (i == 0) { |
4547 | rnp->grpnum = 0; | |
4548 | rnp->grpmask = 0; | |
4549 | rnp->parent = NULL; | |
4550 | } else { | |
199977bf | 4551 | rnp->grpnum = j % levelspread[i - 1]; |
df63fa5b | 4552 | rnp->grpmask = BIT(rnp->grpnum); |
eb7a6653 | 4553 | rnp->parent = rcu_state.level[i - 1] + |
199977bf | 4554 | j / levelspread[i - 1]; |
64db4cff PM |
4555 | } |
4556 | rnp->level = i; | |
12f5f524 | 4557 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 4558 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
4559 | init_waitqueue_head(&rnp->exp_wq[0]); |
4560 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
4561 | init_waitqueue_head(&rnp->exp_wq[2]); |
4562 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 4563 | spin_lock_init(&rnp->exp_lock); |
218b957a | 4564 | mutex_init(&rnp->boost_kthread_mutex); |
d96c52fe PM |
4565 | raw_spin_lock_init(&rnp->exp_poll_lock); |
4566 | rnp->exp_seq_poll_rq = RCU_GET_STATE_COMPLETED; | |
4567 | INIT_WORK(&rnp->exp_poll_wq, sync_rcu_do_polled_gp); | |
64db4cff PM |
4568 | } |
4569 | } | |
0c34029a | 4570 | |
eb7a6653 PM |
4571 | init_swait_queue_head(&rcu_state.gp_wq); |
4572 | init_swait_queue_head(&rcu_state.expedited_wq); | |
aedf4ba9 | 4573 | rnp = rcu_first_leaf_node(); |
0c34029a | 4574 | for_each_possible_cpu(i) { |
4a90a068 | 4575 | while (i > rnp->grphi) |
0c34029a | 4576 | rnp++; |
da1df50d | 4577 | per_cpu_ptr(&rcu_data, i)->mynode = rnp; |
53b46303 | 4578 | rcu_boot_init_percpu_data(i); |
0c34029a | 4579 | } |
64db4cff PM |
4580 | } |
4581 | ||
c8db27dd AC |
4582 | /* |
4583 | * Force priority from the kernel command-line into range. | |
4584 | */ | |
4585 | static void __init sanitize_kthread_prio(void) | |
4586 | { | |
4587 | int kthread_prio_in = kthread_prio; | |
4588 | ||
4589 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 2 | |
4590 | && IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)) | |
4591 | kthread_prio = 2; | |
4592 | else if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
4593 | kthread_prio = 1; | |
4594 | else if (kthread_prio < 0) | |
4595 | kthread_prio = 0; | |
4596 | else if (kthread_prio > 99) | |
4597 | kthread_prio = 99; | |
4598 | ||
4599 | if (kthread_prio != kthread_prio_in) | |
4600 | pr_alert("%s: Limited prio to %d from %d\n", | |
4601 | __func__, kthread_prio, kthread_prio_in); | |
4602 | } | |
4603 | ||
f885b7f2 PM |
4604 | /* |
4605 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 4606 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
4607 | * the ->node array in the rcu_state structure. |
4608 | */ | |
b5befe84 | 4609 | void rcu_init_geometry(void) |
f885b7f2 | 4610 | { |
026ad283 | 4611 | ulong d; |
f885b7f2 | 4612 | int i; |
b5befe84 | 4613 | static unsigned long old_nr_cpu_ids; |
05b84aec | 4614 | int rcu_capacity[RCU_NUM_LVLS]; |
b5befe84 FW |
4615 | static bool initialized; |
4616 | ||
4617 | if (initialized) { | |
4618 | /* | |
4619 | * Warn if setup_nr_cpu_ids() had not yet been invoked, | |
4620 | * unless nr_cpus_ids == NR_CPUS, in which case who cares? | |
4621 | */ | |
4622 | WARN_ON_ONCE(old_nr_cpu_ids != nr_cpu_ids); | |
4623 | return; | |
4624 | } | |
4625 | ||
4626 | old_nr_cpu_ids = nr_cpu_ids; | |
4627 | initialized = true; | |
f885b7f2 | 4628 | |
026ad283 PM |
4629 | /* |
4630 | * Initialize any unspecified boot parameters. | |
4631 | * The default values of jiffies_till_first_fqs and | |
4632 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
4633 | * value, which is a function of HZ, then adding one for each | |
4634 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
4635 | */ | |
4636 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
4637 | if (jiffies_till_first_fqs == ULONG_MAX) | |
4638 | jiffies_till_first_fqs = d; | |
4639 | if (jiffies_till_next_fqs == ULONG_MAX) | |
4640 | jiffies_till_next_fqs = d; | |
6973032a | 4641 | adjust_jiffies_till_sched_qs(); |
026ad283 | 4642 | |
f885b7f2 | 4643 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 4644 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 4645 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 4646 | return; |
a7538352 | 4647 | pr_info("Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%u\n", |
39479098 | 4648 | rcu_fanout_leaf, nr_cpu_ids); |
f885b7f2 | 4649 | |
f885b7f2 | 4650 | /* |
ee968ac6 PM |
4651 | * The boot-time rcu_fanout_leaf parameter must be at least two |
4652 | * and cannot exceed the number of bits in the rcu_node masks. | |
4653 | * Complain and fall back to the compile-time values if this | |
4654 | * limit is exceeded. | |
f885b7f2 | 4655 | */ |
ee968ac6 | 4656 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 4657 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 4658 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
4659 | WARN_ON(1); |
4660 | return; | |
4661 | } | |
4662 | ||
f885b7f2 PM |
4663 | /* |
4664 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 4665 | * with the given number of levels. |
f885b7f2 | 4666 | */ |
9618138b | 4667 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 4668 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 4669 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
4670 | |
4671 | /* | |
75cf15a4 | 4672 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 4673 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 4674 | */ |
ee968ac6 PM |
4675 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
4676 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
4677 | WARN_ON(1); | |
4678 | return; | |
4679 | } | |
f885b7f2 | 4680 | |
679f9858 | 4681 | /* Calculate the number of levels in the tree. */ |
9618138b | 4682 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 4683 | } |
9618138b | 4684 | rcu_num_lvls = i + 1; |
679f9858 | 4685 | |
f885b7f2 | 4686 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 4687 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 4688 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
4689 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
4690 | } | |
f885b7f2 PM |
4691 | |
4692 | /* Calculate the total number of rcu_node structures. */ | |
4693 | rcu_num_nodes = 0; | |
679f9858 | 4694 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 4695 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
4696 | } |
4697 | ||
a3dc2948 PM |
4698 | /* |
4699 | * Dump out the structure of the rcu_node combining tree associated | |
49918a54 | 4700 | * with the rcu_state structure. |
a3dc2948 | 4701 | */ |
b8bb1f63 | 4702 | static void __init rcu_dump_rcu_node_tree(void) |
a3dc2948 PM |
4703 | { |
4704 | int level = 0; | |
4705 | struct rcu_node *rnp; | |
4706 | ||
4707 | pr_info("rcu_node tree layout dump\n"); | |
4708 | pr_info(" "); | |
aedf4ba9 | 4709 | rcu_for_each_node_breadth_first(rnp) { |
a3dc2948 PM |
4710 | if (rnp->level != level) { |
4711 | pr_cont("\n"); | |
4712 | pr_info(" "); | |
4713 | level = rnp->level; | |
4714 | } | |
4715 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
4716 | } | |
4717 | pr_cont("\n"); | |
4718 | } | |
4719 | ||
ad7c946b PM |
4720 | struct workqueue_struct *rcu_gp_wq; |
4721 | ||
a35d1690 BP |
4722 | static void __init kfree_rcu_batch_init(void) |
4723 | { | |
4724 | int cpu; | |
0392bebe | 4725 | int i; |
a35d1690 | 4726 | |
d0bfa8b3 ZQ |
4727 | /* Clamp it to [0:100] seconds interval. */ |
4728 | if (rcu_delay_page_cache_fill_msec < 0 || | |
4729 | rcu_delay_page_cache_fill_msec > 100 * MSEC_PER_SEC) { | |
4730 | ||
4731 | rcu_delay_page_cache_fill_msec = | |
4732 | clamp(rcu_delay_page_cache_fill_msec, 0, | |
4733 | (int) (100 * MSEC_PER_SEC)); | |
4734 | ||
4735 | pr_info("Adjusting rcutree.rcu_delay_page_cache_fill_msec to %d ms.\n", | |
4736 | rcu_delay_page_cache_fill_msec); | |
4737 | } | |
4738 | ||
a35d1690 BP |
4739 | for_each_possible_cpu(cpu) { |
4740 | struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu); | |
4741 | ||
34c88174 URS |
4742 | for (i = 0; i < KFREE_N_BATCHES; i++) { |
4743 | INIT_RCU_WORK(&krcp->krw_arr[i].rcu_work, kfree_rcu_work); | |
0392bebe | 4744 | krcp->krw_arr[i].krcp = krcp; |
34c88174 URS |
4745 | } |
4746 | ||
a35d1690 | 4747 | INIT_DELAYED_WORK(&krcp->monitor_work, kfree_rcu_monitor); |
d0bfa8b3 | 4748 | INIT_DELAYED_WORK(&krcp->page_cache_work, fill_page_cache_func); |
a35d1690 BP |
4749 | krcp->initialized = true; |
4750 | } | |
e33c267a | 4751 | if (register_shrinker(&kfree_rcu_shrinker, "rcu-kfree")) |
9154244c | 4752 | pr_err("Failed to register kfree_rcu() shrinker!\n"); |
a35d1690 BP |
4753 | } |
4754 | ||
9f680ab4 | 4755 | void __init rcu_init(void) |
64db4cff | 4756 | { |
2eed973a | 4757 | int cpu = smp_processor_id(); |
9f680ab4 | 4758 | |
47627678 PM |
4759 | rcu_early_boot_tests(); |
4760 | ||
a35d1690 | 4761 | kfree_rcu_batch_init(); |
f41d911f | 4762 | rcu_bootup_announce(); |
c8db27dd | 4763 | sanitize_kthread_prio(); |
f885b7f2 | 4764 | rcu_init_geometry(); |
b8bb1f63 | 4765 | rcu_init_one(); |
a3dc2948 | 4766 | if (dump_tree) |
b8bb1f63 | 4767 | rcu_dump_rcu_node_tree(); |
48d07c04 SAS |
4768 | if (use_softirq) |
4769 | open_softirq(RCU_SOFTIRQ, rcu_core_si); | |
9f680ab4 PM |
4770 | |
4771 | /* | |
4772 | * We don't need protection against CPU-hotplug here because | |
4773 | * this is called early in boot, before either interrupts | |
4774 | * or the scheduler are operational. | |
4775 | */ | |
d1d74d14 | 4776 | pm_notifier(rcu_pm_notify, 0); |
2eed973a FW |
4777 | WARN_ON(num_online_cpus() > 1); // Only one CPU this early in boot. |
4778 | rcutree_prepare_cpu(cpu); | |
4779 | rcu_cpu_starting(cpu); | |
4780 | rcutree_online_cpu(cpu); | |
ad7c946b | 4781 | |
277ffe1b | 4782 | /* Create workqueue for Tree SRCU and for expedited GPs. */ |
ad7c946b PM |
4783 | rcu_gp_wq = alloc_workqueue("rcu_gp", WQ_MEM_RECLAIM, 0); |
4784 | WARN_ON(!rcu_gp_wq); | |
9621fbee | 4785 | rcu_alloc_par_gp_wq(); |
b2b00ddf PM |
4786 | |
4787 | /* Fill in default value for rcutree.qovld boot parameter. */ | |
4788 | /* -After- the rcu_node ->lock fields are initialized! */ | |
4789 | if (qovld < 0) | |
4790 | qovld_calc = DEFAULT_RCU_QOVLD_MULT * qhimark; | |
4791 | else | |
4792 | qovld_calc = qovld; | |
d96c52fe PM |
4793 | |
4794 | // Kick-start any polled grace periods that started early. | |
4795 | if (!(per_cpu_ptr(&rcu_data, cpu)->mynode->exp_seq_poll_rq & 0x1)) | |
4796 | (void)start_poll_synchronize_rcu_expedited(); | |
64db4cff PM |
4797 | } |
4798 | ||
10462d6f | 4799 | #include "tree_stall.h" |
3549c2bc | 4800 | #include "tree_exp.h" |
dfcb2754 | 4801 | #include "tree_nocb.h" |
4102adab | 4802 | #include "tree_plugin.h" |