source: mainline/uspace/lib/c/generic/rcu.c@ 42f5860

lfn serial ticket/834-toolchain-update topic/msim-upgrade topic/simplify-dev-export
Last change on this file since 42f5860 was 42f5860, checked in by Jiří Zárevúcky <jiri.zarevucky@…>, 7 years ago

Clean up userspace RCU.

  • Property mode set to 100644
File size: 12.6 KB
Line 
1/*
2 * Copyright (c) 2012 Adam Hraska
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *
9 * - Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * - Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * - The name of the author may not be used to endorse or promote products
15 * derived from this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29/** @addtogroup liburcu
30 * @{
31 */
32/**
33 * @file
34 *
35 * User space RCU is based on URCU utilizing signals [1]. This
36 * implementation does not however signal each thread of the process
37 * to issue a memory barrier. Instead, we introduced a syscall that
38 * issues memory barriers (via IPIs) on cpus that are running threads
39 * of the current process. First, it does not require us to schedule
40 * and run every thread of the process. Second, IPIs are less intrusive
41 * than switching contexts and entering user space.
42 *
43 * This algorithm is further modified to require a single instead of
44 * two reader group changes per grace period. Signal-URCU flips
45 * the reader group and waits for readers of the previous group
46 * twice in succession in order to wait for new readers that were
47 * delayed and mistakenly associated with the previous reader group.
48 * The modified algorithm ensures that the new reader group is
49 * always empty (by explicitly waiting for it to become empty).
50 * Only then does it flip the reader group and wait for preexisting
51 * readers of the old reader group (invariant of SRCU [2, 3]).
52 *
53 *
54 * [1] User-level implementations of read-copy update,
55 * 2012, appendix
56 * http://www.rdrop.com/users/paulmck/RCU/urcu-supp-accepted.2011.08.30a.pdf
57 *
58 * [2] linux/kernel/srcu.c in Linux 3.5-rc2,
59 * 2012
60 * http://tomoyo.sourceforge.jp/cgi-bin/lxr/source/kernel/srcu.c?v=linux-3.5-rc2-ccs-1.8.3
61 *
62 * [3] [RFC PATCH 5/5 single-thread-version] implement
63 * per-domain single-thread state machine,
64 * 2012, Lai
65 * https://lkml.org/lkml/2012/3/6/586
66 */
67
68#include "rcu.h"
69#include <fibril_synch.h>
70#include <fibril.h>
71#include <stdio.h>
72#include <stddef.h>
73#include <compiler/barrier.h>
74#include <libarch/barrier.h>
75#include <futex.h>
76#include <macros.h>
77#include <async.h>
78#include <adt/list.h>
79#include <smp_memory_barrier.h>
80#include <assert.h>
81#include <time.h>
82
83#include "private/fibril.h"
84
85
86/** RCU sleeps for RCU_SLEEP_MS before polling an active RCU reader again. */
87#define RCU_SLEEP_MS 10
88
89#define RCU_NESTING_SHIFT 1
90#define RCU_NESTING_INC (1 << RCU_NESTING_SHIFT)
91#define RCU_GROUP_BIT_MASK (size_t)(RCU_NESTING_INC - 1)
92#define RCU_GROUP_A (size_t)(0 | RCU_NESTING_INC)
93#define RCU_GROUP_B (size_t)(1 | RCU_NESTING_INC)
94
95
96/** Fibril local RCU data. */
97typedef struct fibril_rcu_data {
98 size_t nesting_cnt;
99 link_t link;
100 bool registered;
101} fibril_rcu_data_t;
102
103/** Process global RCU data. */
104typedef struct rcu_data {
105 size_t cur_gp;
106 size_t reader_group;
107 fibril_rmutex_t list_mutex;
108 list_t fibrils_list;
109 struct {
110 fibril_rmutex_t mutex;
111 bool locked;
112 list_t blocked_fibrils;
113 } sync_lock;
114} rcu_data_t;
115
116typedef struct blocked_fibril {
117 fid_t id;
118 link_t link;
119 bool is_ready;
120} blocked_fibril_t;
121
122
123/** Fibril local RCU data. */
124static fibril_local fibril_rcu_data_t fibril_rcu = {
125 .nesting_cnt = 0,
126 .link = {
127 .next = NULL,
128 .prev = NULL
129 },
130 .registered = false
131};
132
133/** Process global RCU data. */
134static rcu_data_t rcu = {
135 .cur_gp = 0,
136 .reader_group = RCU_GROUP_A,
137 .list_mutex = FIBRIL_RMUTEX_INITIALIZER(rcu.list_mutex),
138 .fibrils_list = LIST_INITIALIZER(rcu.fibrils_list),
139 .sync_lock = {
140 .mutex = FIBRIL_RMUTEX_INITIALIZER(rcu.sync_lock.mutex),
141 .locked = false,
142 .blocked_fibrils = LIST_INITIALIZER(rcu.sync_lock.blocked_fibrils),
143 },
144};
145
146
147static void wait_for_readers(size_t reader_group);
148static void force_mb_in_all_threads(void);
149static bool is_preexisting_reader(const fibril_rcu_data_t *fib, size_t group);
150
151static void lock_sync(void);
152static void unlock_sync(void);
153static void sync_sleep(void);
154
155static bool is_in_group(size_t nesting_cnt, size_t group);
156static bool is_in_reader_section(size_t nesting_cnt);
157static size_t get_other_group(size_t group);
158
159
160/** Registers a fibril so it may start using RCU read sections.
161 *
162 * A fibril must be registered with rcu before it can enter RCU critical
163 * sections delineated by rcu_read_lock() and rcu_read_unlock().
164 */
165void rcu_register_fibril(void)
166{
167 assert(!fibril_rcu.registered);
168
169 fibril_rmutex_lock(&rcu.list_mutex);
170 list_append(&fibril_rcu.link, &rcu.fibrils_list);
171 fibril_rmutex_unlock(&rcu.list_mutex);
172
173 fibril_rcu.registered = true;
174}
175
176/** Deregisters a fibril that had been using RCU read sections.
177 *
178 * A fibril must be deregistered before it exits if it had
179 * been registered with rcu via rcu_register_fibril().
180 */
181void rcu_deregister_fibril(void)
182{
183 assert(fibril_rcu.registered);
184
185 /*
186 * Forcefully unlock any reader sections. The fibril is exiting
187 * so it is not holding any references to data protected by the
188 * rcu section. Therefore, it is safe to unlock. Otherwise,
189 * rcu_synchronize() would wait indefinitely.
190 */
191 memory_barrier();
192 fibril_rcu.nesting_cnt = 0;
193
194 fibril_rmutex_lock(&rcu.list_mutex);
195 list_remove(&fibril_rcu.link);
196 fibril_rmutex_unlock(&rcu.list_mutex);
197
198 fibril_rcu.registered = false;
199}
200
201/** Delimits the start of an RCU reader critical section.
202 *
203 * RCU reader sections may be nested.
204 */
205void rcu_read_lock(void)
206{
207 assert(fibril_rcu.registered);
208
209 size_t nesting_cnt = ACCESS_ONCE(fibril_rcu.nesting_cnt);
210
211 if (0 == (nesting_cnt >> RCU_NESTING_SHIFT)) {
212 ACCESS_ONCE(fibril_rcu.nesting_cnt) = ACCESS_ONCE(rcu.reader_group);
213 /* Required by MB_FORCE_L */
214 compiler_barrier(); /* CC_BAR_L */
215 } else {
216 ACCESS_ONCE(fibril_rcu.nesting_cnt) = nesting_cnt + RCU_NESTING_INC;
217 }
218}
219
220/** Delimits the start of an RCU reader critical section. */
221void rcu_read_unlock(void)
222{
223 assert(fibril_rcu.registered);
224 assert(rcu_read_locked());
225
226 /* Required by MB_FORCE_U */
227 compiler_barrier(); /* CC_BAR_U */
228 /* todo: ACCESS_ONCE(nesting_cnt) ? */
229 fibril_rcu.nesting_cnt -= RCU_NESTING_INC;
230}
231
232/** Returns true if the current fibril is in an RCU reader section. */
233bool rcu_read_locked(void)
234{
235 return 0 != (fibril_rcu.nesting_cnt >> RCU_NESTING_SHIFT);
236}
237
238/** Blocks until all preexisting readers exit their critical sections. */
239void rcu_synchronize(void)
240{
241 assert(!rcu_read_locked());
242
243 /* Contain load of rcu.cur_gp. */
244 memory_barrier();
245
246 /* Approximately the number of the GP in progress. */
247 size_t gp_in_progress = ACCESS_ONCE(rcu.cur_gp);
248
249 lock_sync();
250
251 /*
252 * Exit early if we were stuck waiting for the mutex for a full grace
253 * period. Started waiting during gp_in_progress (or gp_in_progress + 1
254 * if the value propagated to this cpu too late) so wait for the next
255 * full GP, gp_in_progress + 1, to finish. Ie don't wait if the GP
256 * after that, gp_in_progress + 2, already started.
257 */
258 /* rcu.cur_gp >= gp_in_progress + 2, but tolerates overflows. */
259 if (rcu.cur_gp != gp_in_progress && rcu.cur_gp + 1 != gp_in_progress) {
260 unlock_sync();
261 return;
262 }
263
264 ++ACCESS_ONCE(rcu.cur_gp);
265
266 /*
267 * Pairs up with MB_FORCE_L (ie CC_BAR_L). Makes changes prior
268 * to rcu_synchronize() visible to new readers.
269 */
270 memory_barrier(); /* MB_A */
271
272 /*
273 * Pairs up with MB_A.
274 *
275 * If the memory barrier is issued before CC_BAR_L in the target
276 * thread, it pairs up with MB_A and the thread sees all changes
277 * prior to rcu_synchronize(). Ie any reader sections are new
278 * rcu readers.
279 *
280 * If the memory barrier is issued after CC_BAR_L, it pairs up
281 * with MB_B and it will make the most recent nesting_cnt visible
282 * in this thread. Since the reader may have already accessed
283 * memory protected by RCU (it ran instructions passed CC_BAR_L),
284 * it is a preexisting reader. Seeing the most recent nesting_cnt
285 * ensures the thread will be identified as a preexisting reader
286 * and we will wait for it in wait_for_readers(old_reader_group).
287 */
288 force_mb_in_all_threads(); /* MB_FORCE_L */
289
290 /*
291 * Pairs with MB_FORCE_L (ie CC_BAR_L, CC_BAR_U) and makes the most
292 * current fibril.nesting_cnt visible to this cpu.
293 */
294 read_barrier(); /* MB_B */
295
296 size_t new_reader_group = get_other_group(rcu.reader_group);
297 wait_for_readers(new_reader_group);
298
299 /* Separates waiting for readers in new_reader_group from group flip. */
300 memory_barrier();
301
302 /* Flip the group new readers should associate with. */
303 size_t old_reader_group = rcu.reader_group;
304 rcu.reader_group = new_reader_group;
305
306 /* Flip the group before waiting for preexisting readers in the old group.*/
307 memory_barrier();
308
309 wait_for_readers(old_reader_group);
310
311 /* MB_FORCE_U */
312 force_mb_in_all_threads(); /* MB_FORCE_U */
313
314 unlock_sync();
315}
316
317/** Issues a memory barrier in each thread of this process. */
318static void force_mb_in_all_threads(void)
319{
320 /*
321 * Only issue barriers in running threads. The scheduler will
322 * execute additional memory barriers when switching to threads
323 * of the process that are currently not running.
324 */
325 smp_memory_barrier();
326}
327
328/** Waits for readers of reader_group to exit their readers sections. */
329static void wait_for_readers(size_t reader_group)
330{
331 fibril_rmutex_lock(&rcu.list_mutex);
332
333 list_t quiescent_fibrils;
334 list_initialize(&quiescent_fibrils);
335
336 while (!list_empty(&rcu.fibrils_list)) {
337 list_foreach_safe(rcu.fibrils_list, fibril_it, next_fibril) {
338 fibril_rcu_data_t *fib = member_to_inst(fibril_it,
339 fibril_rcu_data_t, link);
340
341 if (is_preexisting_reader(fib, reader_group)) {
342 fibril_rmutex_unlock(&rcu.list_mutex);
343 sync_sleep();
344 fibril_rmutex_lock(&rcu.list_mutex);
345 /* Break to while loop. */
346 break;
347 } else {
348 list_remove(fibril_it);
349 list_append(fibril_it, &quiescent_fibrils);
350 }
351 }
352 }
353
354 list_concat(&rcu.fibrils_list, &quiescent_fibrils);
355 fibril_rmutex_unlock(&rcu.list_mutex);
356}
357
358static void lock_sync(void)
359{
360 fibril_rmutex_lock(&rcu.sync_lock.mutex);
361 if (rcu.sync_lock.locked) {
362 blocked_fibril_t blocked_fib;
363 blocked_fib.id = fibril_get_id();
364
365 list_append(&blocked_fib.link, &rcu.sync_lock.blocked_fibrils);
366
367 do {
368 blocked_fib.is_ready = false;
369 fibril_rmutex_unlock(&rcu.sync_lock.mutex);
370 futex_lock(&async_futex);
371 fibril_switch(FIBRIL_FROM_BLOCKED);
372 futex_unlock(&async_futex);
373 fibril_rmutex_lock(&rcu.sync_lock.mutex);
374 } while (rcu.sync_lock.locked);
375
376 list_remove(&blocked_fib.link);
377 rcu.sync_lock.locked = true;
378 } else {
379 rcu.sync_lock.locked = true;
380 }
381}
382
383static void unlock_sync(void)
384{
385 assert(rcu.sync_lock.locked);
386
387 /* Unlock but wake up any fibrils waiting for the lock. */
388
389 if (!list_empty(&rcu.sync_lock.blocked_fibrils)) {
390 blocked_fibril_t *blocked_fib = member_to_inst(
391 list_first(&rcu.sync_lock.blocked_fibrils), blocked_fibril_t, link);
392
393 if (!blocked_fib->is_ready) {
394 blocked_fib->is_ready = true;
395 fibril_add_ready(blocked_fib->id);
396 }
397 }
398
399 rcu.sync_lock.locked = false;
400 fibril_rmutex_unlock(&rcu.sync_lock.mutex);
401}
402
403static void sync_sleep(void)
404{
405 assert(rcu.sync_lock.locked);
406 /*
407 * Release the futex to avoid deadlocks in singlethreaded apps
408 * but keep sync locked.
409 */
410 fibril_rmutex_unlock(&rcu.sync_lock.mutex);
411 fibril_usleep(RCU_SLEEP_MS * 1000);
412 fibril_rmutex_lock(&rcu.sync_lock.mutex);
413}
414
415
416static bool is_preexisting_reader(const fibril_rcu_data_t *fib, size_t group)
417{
418 size_t nesting_cnt = ACCESS_ONCE(fib->nesting_cnt);
419
420 return is_in_group(nesting_cnt, group) && is_in_reader_section(nesting_cnt);
421}
422
423static size_t get_other_group(size_t group)
424{
425 if (group == RCU_GROUP_A)
426 return RCU_GROUP_B;
427 else
428 return RCU_GROUP_A;
429}
430
431static bool is_in_reader_section(size_t nesting_cnt)
432{
433 return RCU_NESTING_INC <= nesting_cnt;
434}
435
436static bool is_in_group(size_t nesting_cnt, size_t group)
437{
438 return (nesting_cnt & RCU_GROUP_BIT_MASK) == (group & RCU_GROUP_BIT_MASK);
439}
440
441
442
443/** @}
444 */
Note: See TracBrowser for help on using the repository browser.