xref: /openbmc/linux/kernel/rcu/rcu_segcblist.c (revision b830f94f)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * RCU segmented callback lists, function definitions
4  *
5  * Copyright IBM Corporation, 2017
6  *
7  * Authors: Paul E. McKenney <paulmck@linux.ibm.com>
8  */
9 
10 #include <linux/types.h>
11 #include <linux/kernel.h>
12 #include <linux/interrupt.h>
13 #include <linux/rcupdate.h>
14 
15 #include "rcu_segcblist.h"
16 
17 /* Initialize simple callback list. */
18 void rcu_cblist_init(struct rcu_cblist *rclp)
19 {
20 	rclp->head = NULL;
21 	rclp->tail = &rclp->head;
22 	rclp->len = 0;
23 	rclp->len_lazy = 0;
24 }
25 
26 /*
27  * Dequeue the oldest rcu_head structure from the specified callback
28  * list.  This function assumes that the callback is non-lazy, but
29  * the caller can later invoke rcu_cblist_dequeued_lazy() if it
30  * finds otherwise (and if it cares about laziness).  This allows
31  * different users to have different ways of determining laziness.
32  */
33 struct rcu_head *rcu_cblist_dequeue(struct rcu_cblist *rclp)
34 {
35 	struct rcu_head *rhp;
36 
37 	rhp = rclp->head;
38 	if (!rhp)
39 		return NULL;
40 	rclp->len--;
41 	rclp->head = rhp->next;
42 	if (!rclp->head)
43 		rclp->tail = &rclp->head;
44 	return rhp;
45 }
46 
47 /*
48  * Initialize an rcu_segcblist structure.
49  */
50 void rcu_segcblist_init(struct rcu_segcblist *rsclp)
51 {
52 	int i;
53 
54 	BUILD_BUG_ON(RCU_NEXT_TAIL + 1 != ARRAY_SIZE(rsclp->gp_seq));
55 	BUILD_BUG_ON(ARRAY_SIZE(rsclp->tails) != ARRAY_SIZE(rsclp->gp_seq));
56 	rsclp->head = NULL;
57 	for (i = 0; i < RCU_CBLIST_NSEGS; i++)
58 		rsclp->tails[i] = &rsclp->head;
59 	rsclp->len = 0;
60 	rsclp->len_lazy = 0;
61 }
62 
63 /*
64  * Disable the specified rcu_segcblist structure, so that callbacks can
65  * no longer be posted to it.  This structure must be empty.
66  */
67 void rcu_segcblist_disable(struct rcu_segcblist *rsclp)
68 {
69 	WARN_ON_ONCE(!rcu_segcblist_empty(rsclp));
70 	WARN_ON_ONCE(rcu_segcblist_n_cbs(rsclp));
71 	WARN_ON_ONCE(rcu_segcblist_n_lazy_cbs(rsclp));
72 	rsclp->tails[RCU_NEXT_TAIL] = NULL;
73 }
74 
75 /*
76  * Does the specified rcu_segcblist structure contain callbacks that
77  * are ready to be invoked?
78  */
79 bool rcu_segcblist_ready_cbs(struct rcu_segcblist *rsclp)
80 {
81 	return rcu_segcblist_is_enabled(rsclp) &&
82 	       &rsclp->head != rsclp->tails[RCU_DONE_TAIL];
83 }
84 
85 /*
86  * Does the specified rcu_segcblist structure contain callbacks that
87  * are still pending, that is, not yet ready to be invoked?
88  */
89 bool rcu_segcblist_pend_cbs(struct rcu_segcblist *rsclp)
90 {
91 	return rcu_segcblist_is_enabled(rsclp) &&
92 	       !rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL);
93 }
94 
95 /*
96  * Return a pointer to the first callback in the specified rcu_segcblist
97  * structure.  This is useful for diagnostics.
98  */
99 struct rcu_head *rcu_segcblist_first_cb(struct rcu_segcblist *rsclp)
100 {
101 	if (rcu_segcblist_is_enabled(rsclp))
102 		return rsclp->head;
103 	return NULL;
104 }
105 
106 /*
107  * Return a pointer to the first pending callback in the specified
108  * rcu_segcblist structure.  This is useful just after posting a given
109  * callback -- if that callback is the first pending callback, then
110  * you cannot rely on someone else having already started up the required
111  * grace period.
112  */
113 struct rcu_head *rcu_segcblist_first_pend_cb(struct rcu_segcblist *rsclp)
114 {
115 	if (rcu_segcblist_is_enabled(rsclp))
116 		return *rsclp->tails[RCU_DONE_TAIL];
117 	return NULL;
118 }
119 
120 /*
121  * Enqueue the specified callback onto the specified rcu_segcblist
122  * structure, updating accounting as needed.  Note that the ->len
123  * field may be accessed locklessly, hence the WRITE_ONCE().
124  * The ->len field is used by rcu_barrier() and friends to determine
125  * if it must post a callback on this structure, and it is OK
126  * for rcu_barrier() to sometimes post callbacks needlessly, but
127  * absolutely not OK for it to ever miss posting a callback.
128  */
129 void rcu_segcblist_enqueue(struct rcu_segcblist *rsclp,
130 			   struct rcu_head *rhp, bool lazy)
131 {
132 	WRITE_ONCE(rsclp->len, rsclp->len + 1); /* ->len sampled locklessly. */
133 	if (lazy)
134 		rsclp->len_lazy++;
135 	smp_mb(); /* Ensure counts are updated before callback is enqueued. */
136 	rhp->next = NULL;
137 	*rsclp->tails[RCU_NEXT_TAIL] = rhp;
138 	rsclp->tails[RCU_NEXT_TAIL] = &rhp->next;
139 }
140 
141 /*
142  * Entrain the specified callback onto the specified rcu_segcblist at
143  * the end of the last non-empty segment.  If the entire rcu_segcblist
144  * is empty, make no change, but return false.
145  *
146  * This is intended for use by rcu_barrier()-like primitives, -not-
147  * for normal grace-period use.  IMPORTANT:  The callback you enqueue
148  * will wait for all prior callbacks, NOT necessarily for a grace
149  * period.  You have been warned.
150  */
151 bool rcu_segcblist_entrain(struct rcu_segcblist *rsclp,
152 			   struct rcu_head *rhp, bool lazy)
153 {
154 	int i;
155 
156 	if (rcu_segcblist_n_cbs(rsclp) == 0)
157 		return false;
158 	WRITE_ONCE(rsclp->len, rsclp->len + 1);
159 	if (lazy)
160 		rsclp->len_lazy++;
161 	smp_mb(); /* Ensure counts are updated before callback is entrained. */
162 	rhp->next = NULL;
163 	for (i = RCU_NEXT_TAIL; i > RCU_DONE_TAIL; i--)
164 		if (rsclp->tails[i] != rsclp->tails[i - 1])
165 			break;
166 	*rsclp->tails[i] = rhp;
167 	for (; i <= RCU_NEXT_TAIL; i++)
168 		rsclp->tails[i] = &rhp->next;
169 	return true;
170 }
171 
172 /*
173  * Extract only the counts from the specified rcu_segcblist structure,
174  * and place them in the specified rcu_cblist structure.  This function
175  * supports both callback orphaning and invocation, hence the separation
176  * of counts and callbacks.  (Callbacks ready for invocation must be
177  * orphaned and adopted separately from pending callbacks, but counts
178  * apply to all callbacks.  Locking must be used to make sure that
179  * both orphaned-callbacks lists are consistent.)
180  */
181 void rcu_segcblist_extract_count(struct rcu_segcblist *rsclp,
182 					       struct rcu_cblist *rclp)
183 {
184 	rclp->len_lazy += rsclp->len_lazy;
185 	rclp->len += rsclp->len;
186 	rsclp->len_lazy = 0;
187 	WRITE_ONCE(rsclp->len, 0); /* ->len sampled locklessly. */
188 }
189 
190 /*
191  * Extract only those callbacks ready to be invoked from the specified
192  * rcu_segcblist structure and place them in the specified rcu_cblist
193  * structure.
194  */
195 void rcu_segcblist_extract_done_cbs(struct rcu_segcblist *rsclp,
196 				    struct rcu_cblist *rclp)
197 {
198 	int i;
199 
200 	if (!rcu_segcblist_ready_cbs(rsclp))
201 		return; /* Nothing to do. */
202 	*rclp->tail = rsclp->head;
203 	rsclp->head = *rsclp->tails[RCU_DONE_TAIL];
204 	*rsclp->tails[RCU_DONE_TAIL] = NULL;
205 	rclp->tail = rsclp->tails[RCU_DONE_TAIL];
206 	for (i = RCU_CBLIST_NSEGS - 1; i >= RCU_DONE_TAIL; i--)
207 		if (rsclp->tails[i] == rsclp->tails[RCU_DONE_TAIL])
208 			rsclp->tails[i] = &rsclp->head;
209 }
210 
211 /*
212  * Extract only those callbacks still pending (not yet ready to be
213  * invoked) from the specified rcu_segcblist structure and place them in
214  * the specified rcu_cblist structure.  Note that this loses information
215  * about any callbacks that might have been partway done waiting for
216  * their grace period.  Too bad!  They will have to start over.
217  */
218 void rcu_segcblist_extract_pend_cbs(struct rcu_segcblist *rsclp,
219 				    struct rcu_cblist *rclp)
220 {
221 	int i;
222 
223 	if (!rcu_segcblist_pend_cbs(rsclp))
224 		return; /* Nothing to do. */
225 	*rclp->tail = *rsclp->tails[RCU_DONE_TAIL];
226 	rclp->tail = rsclp->tails[RCU_NEXT_TAIL];
227 	*rsclp->tails[RCU_DONE_TAIL] = NULL;
228 	for (i = RCU_DONE_TAIL + 1; i < RCU_CBLIST_NSEGS; i++)
229 		rsclp->tails[i] = rsclp->tails[RCU_DONE_TAIL];
230 }
231 
232 /*
233  * Insert counts from the specified rcu_cblist structure in the
234  * specified rcu_segcblist structure.
235  */
236 void rcu_segcblist_insert_count(struct rcu_segcblist *rsclp,
237 				struct rcu_cblist *rclp)
238 {
239 	rsclp->len_lazy += rclp->len_lazy;
240 	/* ->len sampled locklessly. */
241 	WRITE_ONCE(rsclp->len, rsclp->len + rclp->len);
242 	rclp->len_lazy = 0;
243 	rclp->len = 0;
244 }
245 
246 /*
247  * Move callbacks from the specified rcu_cblist to the beginning of the
248  * done-callbacks segment of the specified rcu_segcblist.
249  */
250 void rcu_segcblist_insert_done_cbs(struct rcu_segcblist *rsclp,
251 				   struct rcu_cblist *rclp)
252 {
253 	int i;
254 
255 	if (!rclp->head)
256 		return; /* No callbacks to move. */
257 	*rclp->tail = rsclp->head;
258 	rsclp->head = rclp->head;
259 	for (i = RCU_DONE_TAIL; i < RCU_CBLIST_NSEGS; i++)
260 		if (&rsclp->head == rsclp->tails[i])
261 			rsclp->tails[i] = rclp->tail;
262 		else
263 			break;
264 	rclp->head = NULL;
265 	rclp->tail = &rclp->head;
266 }
267 
268 /*
269  * Move callbacks from the specified rcu_cblist to the end of the
270  * new-callbacks segment of the specified rcu_segcblist.
271  */
272 void rcu_segcblist_insert_pend_cbs(struct rcu_segcblist *rsclp,
273 				   struct rcu_cblist *rclp)
274 {
275 	if (!rclp->head)
276 		return; /* Nothing to do. */
277 	*rsclp->tails[RCU_NEXT_TAIL] = rclp->head;
278 	rsclp->tails[RCU_NEXT_TAIL] = rclp->tail;
279 	rclp->head = NULL;
280 	rclp->tail = &rclp->head;
281 }
282 
283 /*
284  * Advance the callbacks in the specified rcu_segcblist structure based
285  * on the current value passed in for the grace-period counter.
286  */
287 void rcu_segcblist_advance(struct rcu_segcblist *rsclp, unsigned long seq)
288 {
289 	int i, j;
290 
291 	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
292 	if (rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL))
293 		return;
294 
295 	/*
296 	 * Find all callbacks whose ->gp_seq numbers indicate that they
297 	 * are ready to invoke, and put them into the RCU_DONE_TAIL segment.
298 	 */
299 	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) {
300 		if (ULONG_CMP_LT(seq, rsclp->gp_seq[i]))
301 			break;
302 		rsclp->tails[RCU_DONE_TAIL] = rsclp->tails[i];
303 	}
304 
305 	/* If no callbacks moved, nothing more need be done. */
306 	if (i == RCU_WAIT_TAIL)
307 		return;
308 
309 	/* Clean up tail pointers that might have been misordered above. */
310 	for (j = RCU_WAIT_TAIL; j < i; j++)
311 		rsclp->tails[j] = rsclp->tails[RCU_DONE_TAIL];
312 
313 	/*
314 	 * Callbacks moved, so clean up the misordered ->tails[] pointers
315 	 * that now point into the middle of the list of ready-to-invoke
316 	 * callbacks.  The overall effect is to copy down the later pointers
317 	 * into the gap that was created by the now-ready segments.
318 	 */
319 	for (j = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++, j++) {
320 		if (rsclp->tails[j] == rsclp->tails[RCU_NEXT_TAIL])
321 			break;  /* No more callbacks. */
322 		rsclp->tails[j] = rsclp->tails[i];
323 		rsclp->gp_seq[j] = rsclp->gp_seq[i];
324 	}
325 }
326 
327 /*
328  * "Accelerate" callbacks based on more-accurate grace-period information.
329  * The reason for this is that RCU does not synchronize the beginnings and
330  * ends of grace periods, and that callbacks are posted locally.  This in
331  * turn means that the callbacks must be labelled conservatively early
332  * on, as getting exact information would degrade both performance and
333  * scalability.  When more accurate grace-period information becomes
334  * available, previously posted callbacks can be "accelerated", marking
335  * them to complete at the end of the earlier grace period.
336  *
337  * This function operates on an rcu_segcblist structure, and also the
338  * grace-period sequence number seq at which new callbacks would become
339  * ready to invoke.  Returns true if there are callbacks that won't be
340  * ready to invoke until seq, false otherwise.
341  */
342 bool rcu_segcblist_accelerate(struct rcu_segcblist *rsclp, unsigned long seq)
343 {
344 	int i;
345 
346 	WARN_ON_ONCE(!rcu_segcblist_is_enabled(rsclp));
347 	if (rcu_segcblist_restempty(rsclp, RCU_DONE_TAIL))
348 		return false;
349 
350 	/*
351 	 * Find the segment preceding the oldest segment of callbacks
352 	 * whose ->gp_seq[] completion is at or after that passed in via
353 	 * "seq", skipping any empty segments.  This oldest segment, along
354 	 * with any later segments, can be merged in with any newly arrived
355 	 * callbacks in the RCU_NEXT_TAIL segment, and assigned "seq"
356 	 * as their ->gp_seq[] grace-period completion sequence number.
357 	 */
358 	for (i = RCU_NEXT_READY_TAIL; i > RCU_DONE_TAIL; i--)
359 		if (rsclp->tails[i] != rsclp->tails[i - 1] &&
360 		    ULONG_CMP_LT(rsclp->gp_seq[i], seq))
361 			break;
362 
363 	/*
364 	 * If all the segments contain callbacks that correspond to
365 	 * earlier grace-period sequence numbers than "seq", leave.
366 	 * Assuming that the rcu_segcblist structure has enough
367 	 * segments in its arrays, this can only happen if some of
368 	 * the non-done segments contain callbacks that really are
369 	 * ready to invoke.  This situation will get straightened
370 	 * out by the next call to rcu_segcblist_advance().
371 	 *
372 	 * Also advance to the oldest segment of callbacks whose
373 	 * ->gp_seq[] completion is at or after that passed in via "seq",
374 	 * skipping any empty segments.
375 	 */
376 	if (++i >= RCU_NEXT_TAIL)
377 		return false;
378 
379 	/*
380 	 * Merge all later callbacks, including newly arrived callbacks,
381 	 * into the segment located by the for-loop above.  Assign "seq"
382 	 * as the ->gp_seq[] value in order to correctly handle the case
383 	 * where there were no pending callbacks in the rcu_segcblist
384 	 * structure other than in the RCU_NEXT_TAIL segment.
385 	 */
386 	for (; i < RCU_NEXT_TAIL; i++) {
387 		rsclp->tails[i] = rsclp->tails[RCU_NEXT_TAIL];
388 		rsclp->gp_seq[i] = seq;
389 	}
390 	return true;
391 }
392 
393 /*
394  * Merge the source rcu_segcblist structure into the destination
395  * rcu_segcblist structure, then initialize the source.  Any pending
396  * callbacks from the source get to start over.  It is best to
397  * advance and accelerate both the destination and the source
398  * before merging.
399  */
400 void rcu_segcblist_merge(struct rcu_segcblist *dst_rsclp,
401 			 struct rcu_segcblist *src_rsclp)
402 {
403 	struct rcu_cblist donecbs;
404 	struct rcu_cblist pendcbs;
405 
406 	rcu_cblist_init(&donecbs);
407 	rcu_cblist_init(&pendcbs);
408 	rcu_segcblist_extract_count(src_rsclp, &donecbs);
409 	rcu_segcblist_extract_done_cbs(src_rsclp, &donecbs);
410 	rcu_segcblist_extract_pend_cbs(src_rsclp, &pendcbs);
411 	rcu_segcblist_insert_count(dst_rsclp, &donecbs);
412 	rcu_segcblist_insert_done_cbs(dst_rsclp, &donecbs);
413 	rcu_segcblist_insert_pend_cbs(dst_rsclp, &pendcbs);
414 	rcu_segcblist_init(src_rsclp);
415 }
416