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