Commit | Line | Data |
---|---|---|
551ac1a3 | 1 | /* |
a60dadc5 MD |
2 | * mem.spin: Promela code to validate memory barriers with OOO memory |
3 | * and out-of-order instruction scheduling. | |
551ac1a3 MD |
4 | * |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
18 | * | |
19 | * Copyright (c) 2009 Mathieu Desnoyers | |
20 | */ | |
21 | ||
22 | /* Promela validation variables. */ | |
23 | ||
24 | /* specific defines "included" here */ | |
25 | /* DEFINES file "included" here */ | |
26 | ||
27 | #define NR_READERS 1 | |
28 | #define NR_WRITERS 1 | |
29 | ||
30 | #define NR_PROCS 2 | |
31 | ||
32 | #define get_pid() (_pid) | |
33 | ||
34 | #define get_readerid() (get_pid()) | |
35 | ||
36 | /* | |
37 | * Produced process control and data flow. Updated after each instruction to | |
38 | * show which variables are ready. Using one-hot bit encoding per variable to | |
39 | * save state space. Used as triggers to execute the instructions having those | |
40 | * variables as input. Leaving bits active to inhibit instruction execution. | |
41 | * Scheme used to make instruction disabling and automatic dependency fall-back | |
42 | * automatic. | |
43 | */ | |
44 | ||
45 | #define CONSUME_TOKENS(state, bits, notbits) \ | |
46 | ((!(state & (notbits))) && (state & (bits)) == (bits)) | |
47 | ||
48 | #define PRODUCE_TOKENS(state, bits) \ | |
49 | state = state | (bits); | |
50 | ||
51 | #define CLEAR_TOKENS(state, bits) \ | |
52 | state = state & ~(bits) | |
53 | ||
54 | /* | |
55 | * Types of dependency : | |
56 | * | |
57 | * Data dependency | |
58 | * | |
59 | * - True dependency, Read-after-Write (RAW) | |
60 | * | |
61 | * This type of dependency happens when a statement depends on the result of a | |
62 | * previous statement. This applies to any statement which needs to read a | |
63 | * variable written by a preceding statement. | |
64 | * | |
65 | * - False dependency, Write-after-Read (WAR) | |
66 | * | |
67 | * Typically, variable renaming can ensure that this dependency goes away. | |
68 | * However, if the statements must read and then write from/to the same variable | |
69 | * in the OOO memory model, renaming may be impossible, and therefore this | |
70 | * causes a WAR dependency. | |
71 | * | |
72 | * - Output dependency, Write-after-Write (WAW) | |
73 | * | |
74 | * Two writes to the same variable in subsequent statements. Variable renaming | |
75 | * can ensure this is not needed, but can be required when writing multiple | |
76 | * times to the same OOO mem model variable. | |
77 | * | |
78 | * Control dependency | |
79 | * | |
80 | * Execution of a given instruction depends on a previous instruction evaluating | |
81 | * in a way that allows its execution. E.g. : branches. | |
82 | * | |
83 | * Useful considerations for joining dependencies after branch | |
84 | * | |
85 | * - Pre-dominance | |
86 | * | |
87 | * "We say box i dominates box j if every path (leading from input to output | |
88 | * through the diagram) which passes through box j must also pass through box | |
89 | * i. Thus box i dominates box j if box j is subordinate to box i in the | |
90 | * program." | |
91 | * | |
92 | * http://www.hipersoft.rice.edu/grads/publications/dom14.pdf | |
93 | * Other classic algorithm to calculate dominance : Lengauer-Tarjan (in gcc) | |
94 | * | |
95 | * - Post-dominance | |
96 | * | |
97 | * Just as pre-dominance, but with arcs of the data flow inverted, and input vs | |
98 | * output exchanged. Therefore, i post-dominating j ensures that every path | |
99 | * passing by j will pass by i before reaching the output. | |
100 | * | |
101 | * Other considerations | |
102 | * | |
103 | * Note about "volatile" keyword dependency : The compiler will order volatile | |
104 | * accesses so they appear in the right order on a given CPU. They can be | |
105 | * reordered by the CPU instruction scheduling. This therefore cannot be | |
106 | * considered as a depencency. | |
107 | * | |
108 | * References : | |
109 | * | |
110 | * Cooper, Keith D.; & Torczon, Linda. (2005). Engineering a Compiler. Morgan | |
111 | * Kaufmann. ISBN 1-55860-698-X. | |
112 | * Kennedy, Ken; & Allen, Randy. (2001). Optimizing Compilers for Modern | |
113 | * Architectures: A Dependence-based Approach. Morgan Kaufmann. ISBN | |
114 | * 1-55860-286-0. | |
115 | * Muchnick, Steven S. (1997). Advanced Compiler Design and Implementation. | |
116 | * Morgan Kaufmann. ISBN 1-55860-320-4. | |
117 | */ | |
118 | ||
119 | /* | |
120 | * Note about loops and nested calls | |
121 | * | |
122 | * To keep this model simple, loops expressed in the framework will behave as if | |
123 | * there was a core synchronizing instruction between loops. To see the effect | |
124 | * of loop unrolling, manually unrolling loops is required. Note that if loops | |
125 | * end or start with a core synchronizing instruction, the model is appropriate. | |
126 | * Nested calls are not supported. | |
127 | */ | |
128 | ||
129 | /* | |
130 | * Each process have its own data in cache. Caches are randomly updated. | |
131 | * smp_wmb and smp_rmb forces cache updates (write and read), smp_mb forces | |
132 | * both. | |
133 | */ | |
134 | ||
135 | typedef per_proc_byte { | |
136 | byte val[NR_PROCS]; | |
137 | }; | |
138 | ||
139 | /* Bitfield has a maximum of 8 procs */ | |
140 | typedef per_proc_bit { | |
141 | byte bitfield; | |
142 | }; | |
143 | ||
144 | #define DECLARE_CACHED_VAR(type, x) \ | |
145 | type mem_##x; \ | |
146 | per_proc_##type cached_##x; \ | |
147 | per_proc_bit cache_dirty_##x; | |
148 | ||
149 | #define INIT_CACHED_VAR(x, v, j) \ | |
150 | mem_##x = v; \ | |
151 | cache_dirty_##x.bitfield = 0; \ | |
152 | j = 0; \ | |
153 | do \ | |
154 | :: j < NR_PROCS -> \ | |
155 | cached_##x.val[j] = v; \ | |
156 | j++ \ | |
157 | :: j >= NR_PROCS -> break \ | |
158 | od; | |
159 | ||
160 | #define IS_CACHE_DIRTY(x, id) (cache_dirty_##x.bitfield & (1 << id)) | |
161 | ||
162 | #define READ_CACHED_VAR(x) (cached_##x.val[get_pid()]) | |
163 | ||
164 | #define WRITE_CACHED_VAR(x, v) \ | |
165 | atomic { \ | |
166 | cached_##x.val[get_pid()] = v; \ | |
167 | cache_dirty_##x.bitfield = \ | |
168 | cache_dirty_##x.bitfield | (1 << get_pid()); \ | |
169 | } | |
170 | ||
171 | #define CACHE_WRITE_TO_MEM(x, id) \ | |
172 | if \ | |
173 | :: IS_CACHE_DIRTY(x, id) -> \ | |
174 | mem_##x = cached_##x.val[id]; \ | |
175 | cache_dirty_##x.bitfield = \ | |
176 | cache_dirty_##x.bitfield & (~(1 << id)); \ | |
177 | :: else -> \ | |
178 | skip \ | |
179 | fi; | |
180 | ||
181 | #define CACHE_READ_FROM_MEM(x, id) \ | |
182 | if \ | |
183 | :: !IS_CACHE_DIRTY(x, id) -> \ | |
184 | cached_##x.val[id] = mem_##x;\ | |
185 | :: else -> \ | |
186 | skip \ | |
187 | fi; | |
188 | ||
189 | /* | |
190 | * May update other caches if cache is dirty, or not. | |
191 | */ | |
192 | #define RANDOM_CACHE_WRITE_TO_MEM(x, id)\ | |
193 | if \ | |
194 | :: 1 -> CACHE_WRITE_TO_MEM(x, id); \ | |
195 | :: 1 -> skip \ | |
196 | fi; | |
197 | ||
198 | #define RANDOM_CACHE_READ_FROM_MEM(x, id)\ | |
199 | if \ | |
200 | :: 1 -> CACHE_READ_FROM_MEM(x, id); \ | |
201 | :: 1 -> skip \ | |
202 | fi; | |
203 | ||
204 | /* Must consume all prior read tokens. All subsequent reads depend on it. */ | |
205 | inline smp_rmb(i, j) | |
206 | { | |
207 | atomic { | |
208 | CACHE_READ_FROM_MEM(urcu_gp_ctr, get_pid()); | |
209 | i = 0; | |
210 | do | |
211 | :: i < NR_READERS -> | |
212 | CACHE_READ_FROM_MEM(urcu_active_readers[i], get_pid()); | |
213 | i++ | |
214 | :: i >= NR_READERS -> break | |
215 | od; | |
216 | CACHE_READ_FROM_MEM(generation_ptr, get_pid()); | |
217 | } | |
218 | } | |
219 | ||
220 | /* Must consume all prior write tokens. All subsequent writes depend on it. */ | |
221 | inline smp_wmb(i, j) | |
222 | { | |
223 | atomic { | |
224 | CACHE_WRITE_TO_MEM(urcu_gp_ctr, get_pid()); | |
225 | i = 0; | |
226 | do | |
227 | :: i < NR_READERS -> | |
228 | CACHE_WRITE_TO_MEM(urcu_active_readers[i], get_pid()); | |
229 | i++ | |
230 | :: i >= NR_READERS -> break | |
231 | od; | |
232 | CACHE_WRITE_TO_MEM(generation_ptr, get_pid()); | |
233 | } | |
234 | } | |
235 | ||
236 | /* Synchronization point. Must consume all prior read and write tokens. All | |
237 | * subsequent reads and writes depend on it. */ | |
238 | inline smp_mb(i, j) | |
239 | { | |
240 | atomic { | |
241 | smp_wmb(i, j); | |
242 | smp_rmb(i, j); | |
243 | } | |
244 | } | |
245 | ||
551ac1a3 MD |
246 | #ifdef REMOTE_BARRIERS |
247 | ||
248 | bit reader_barrier[NR_READERS]; | |
249 | ||
250 | /* | |
251 | * We cannot leave the barriers dependencies in place in REMOTE_BARRIERS mode | |
252 | * because they would add unexisting core synchronization and would therefore | |
253 | * create an incomplete model. | |
254 | * Therefore, we model the read-side memory barriers by completely disabling the | |
255 | * memory barriers and their dependencies from the read-side. One at a time | |
256 | * (different verification runs), we make a different instruction listen for | |
257 | * signals. | |
258 | */ | |
259 | ||
260 | #define smp_mb_reader(i, j) | |
261 | ||
262 | /* | |
263 | * Service 0, 1 or many barrier requests. | |
264 | */ | |
265 | inline smp_mb_recv(i, j) | |
266 | { | |
267 | do | |
268 | :: (reader_barrier[get_readerid()] == 1) -> | |
269 | smp_mb(i, j); | |
270 | reader_barrier[get_readerid()] = 0; | |
30193782 MD |
271 | :: 1 -> |
272 | /* | |
273 | * Busy-looping waiting for other barrier requests are not considered as | |
274 | * non-progress. | |
275 | */ | |
276 | #ifdef READER_PROGRESS | |
277 | progress_reader2: | |
278 | #endif | |
279 | skip; | |
551ac1a3 MD |
280 | :: 1 -> break; |
281 | od; | |
282 | } | |
283 | ||
30193782 MD |
284 | #ifdef WRITER_PROGRESS |
285 | #define PROGRESS_LABEL(progressid) progress_writer_progid_##progressid: | |
286 | #else | |
287 | #define PROGRESS_LABEL(progressid) | |
288 | #endif | |
289 | ||
290 | #define smp_mb_send(i, j, progressid) \ | |
291 | { \ | |
292 | smp_mb(i, j); \ | |
293 | i = 0; \ | |
294 | do \ | |
295 | :: i < NR_READERS -> \ | |
296 | reader_barrier[i] = 1; \ | |
297 | do \ | |
298 | :: (reader_barrier[i] == 1) -> \ | |
299 | /* \ | |
300 | * Busy-looping waiting for reader barrier handling is of little\ | |
301 | * interest, given the reader has the ability to totally ignore \ | |
302 | * barrier requests. \ | |
303 | */ \ | |
304 | PROGRESS_LABEL(progressid) \ | |
305 | skip; \ | |
306 | :: (reader_barrier[i] == 0) -> break; \ | |
307 | od; \ | |
308 | i++; \ | |
309 | :: i >= NR_READERS -> \ | |
310 | break \ | |
311 | od; \ | |
312 | smp_mb(i, j); \ | |
551ac1a3 MD |
313 | } |
314 | ||
315 | #else | |
316 | ||
794a737e | 317 | #define smp_mb_send(i, j, progressid) smp_mb(i, j) |
551ac1a3 MD |
318 | #define smp_mb_reader smp_mb |
319 | #define smp_mb_recv(i, j) | |
320 | ||
321 | #endif | |
322 | ||
323 | /* Keep in sync manually with smp_rmb, wmp_wmb, ooo_mem and init() */ | |
324 | DECLARE_CACHED_VAR(byte, urcu_gp_ctr); | |
325 | /* Note ! currently only two readers */ | |
326 | DECLARE_CACHED_VAR(byte, urcu_active_readers[NR_READERS]); | |
327 | /* pointer generation */ | |
328 | DECLARE_CACHED_VAR(byte, generation_ptr); | |
329 | ||
330 | byte last_free_gen = 0; | |
331 | bit free_done = 0; | |
332 | byte read_generation[NR_READERS]; | |
333 | bit data_access[NR_READERS]; | |
334 | ||
335 | bit write_lock = 0; | |
336 | ||
337 | bit init_done = 0; | |
338 | ||
339 | bit sighand_exec = 0; | |
340 | ||
341 | inline wait_init_done() | |
342 | { | |
343 | do | |
344 | :: init_done == 0 -> skip; | |
345 | :: else -> break; | |
346 | od; | |
347 | } | |
348 | ||
349 | inline ooo_mem(i) | |
350 | { | |
351 | atomic { | |
352 | RANDOM_CACHE_WRITE_TO_MEM(urcu_gp_ctr, get_pid()); | |
353 | i = 0; | |
354 | do | |
355 | :: i < NR_READERS -> | |
356 | RANDOM_CACHE_WRITE_TO_MEM(urcu_active_readers[i], | |
357 | get_pid()); | |
358 | i++ | |
359 | :: i >= NR_READERS -> break | |
360 | od; | |
361 | RANDOM_CACHE_WRITE_TO_MEM(generation_ptr, get_pid()); | |
362 | RANDOM_CACHE_READ_FROM_MEM(urcu_gp_ctr, get_pid()); | |
363 | i = 0; | |
364 | do | |
365 | :: i < NR_READERS -> | |
366 | RANDOM_CACHE_READ_FROM_MEM(urcu_active_readers[i], | |
367 | get_pid()); | |
368 | i++ | |
369 | :: i >= NR_READERS -> break | |
370 | od; | |
371 | RANDOM_CACHE_READ_FROM_MEM(generation_ptr, get_pid()); | |
372 | } | |
373 | } | |
374 | ||
375 | /* | |
376 | * Bit encoding, urcu_reader : | |
377 | */ | |
378 | ||
379 | int _proc_urcu_reader; | |
380 | #define proc_urcu_reader _proc_urcu_reader | |
381 | ||
382 | /* Body of PROCEDURE_READ_LOCK */ | |
383 | #define READ_PROD_A_READ (1 << 0) | |
384 | #define READ_PROD_B_IF_TRUE (1 << 1) | |
385 | #define READ_PROD_B_IF_FALSE (1 << 2) | |
386 | #define READ_PROD_C_IF_TRUE_READ (1 << 3) | |
387 | ||
388 | #define PROCEDURE_READ_LOCK(base, consumetoken, producetoken) \ | |
389 | :: CONSUME_TOKENS(proc_urcu_reader, consumetoken, READ_PROD_A_READ << base) -> \ | |
390 | ooo_mem(i); \ | |
391 | tmp = READ_CACHED_VAR(urcu_active_readers[get_readerid()]); \ | |
392 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_A_READ << base); \ | |
393 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
394 | READ_PROD_A_READ << base, /* RAW, pre-dominant */ \ | |
395 | (READ_PROD_B_IF_TRUE | READ_PROD_B_IF_FALSE) << base) -> \ | |
396 | if \ | |
397 | :: (!(tmp & RCU_GP_CTR_NEST_MASK)) -> \ | |
398 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_B_IF_TRUE << base); \ | |
399 | :: else -> \ | |
400 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_B_IF_FALSE << base); \ | |
401 | fi; \ | |
402 | /* IF TRUE */ \ | |
403 | :: CONSUME_TOKENS(proc_urcu_reader, READ_PROD_B_IF_TRUE << base, \ | |
404 | READ_PROD_C_IF_TRUE_READ << base) -> \ | |
405 | ooo_mem(i); \ | |
406 | tmp2 = READ_CACHED_VAR(urcu_gp_ctr); \ | |
407 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_C_IF_TRUE_READ << base); \ | |
408 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
409 | (READ_PROD_C_IF_TRUE_READ /* pre-dominant */ \ | |
410 | | READ_PROD_A_READ) << base, /* WAR */ \ | |
411 | producetoken) -> \ | |
412 | ooo_mem(i); \ | |
413 | WRITE_CACHED_VAR(urcu_active_readers[get_readerid()], tmp2); \ | |
414 | PRODUCE_TOKENS(proc_urcu_reader, producetoken); \ | |
415 | /* IF_MERGE implies \ | |
416 | * post-dominance */ \ | |
417 | /* ELSE */ \ | |
418 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
419 | (READ_PROD_B_IF_FALSE /* pre-dominant */ \ | |
420 | | READ_PROD_A_READ) << base, /* WAR */ \ | |
421 | producetoken) -> \ | |
422 | ooo_mem(i); \ | |
423 | WRITE_CACHED_VAR(urcu_active_readers[get_readerid()], \ | |
424 | tmp + 1); \ | |
425 | PRODUCE_TOKENS(proc_urcu_reader, producetoken); \ | |
426 | /* IF_MERGE implies \ | |
427 | * post-dominance */ \ | |
428 | /* ENDIF */ \ | |
429 | skip | |
430 | ||
431 | /* Body of PROCEDURE_READ_LOCK */ | |
432 | #define READ_PROC_READ_UNLOCK (1 << 0) | |
433 | ||
434 | #define PROCEDURE_READ_UNLOCK(base, consumetoken, producetoken) \ | |
435 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
436 | consumetoken, \ | |
437 | READ_PROC_READ_UNLOCK << base) -> \ | |
438 | ooo_mem(i); \ | |
439 | tmp2 = READ_CACHED_VAR(urcu_active_readers[get_readerid()]); \ | |
440 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_READ_UNLOCK << base); \ | |
441 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
442 | consumetoken \ | |
443 | | (READ_PROC_READ_UNLOCK << base), /* WAR */ \ | |
444 | producetoken) -> \ | |
445 | ooo_mem(i); \ | |
446 | WRITE_CACHED_VAR(urcu_active_readers[get_readerid()], tmp2 - 1); \ | |
447 | PRODUCE_TOKENS(proc_urcu_reader, producetoken); \ | |
448 | skip | |
449 | ||
450 | ||
451 | #define READ_PROD_NONE (1 << 0) | |
452 | ||
453 | /* PROCEDURE_READ_LOCK base = << 1 : 1 to 5 */ | |
454 | #define READ_LOCK_BASE 1 | |
455 | #define READ_LOCK_OUT (1 << 5) | |
456 | ||
457 | #define READ_PROC_FIRST_MB (1 << 6) | |
458 | ||
459 | /* PROCEDURE_READ_LOCK (NESTED) base : << 7 : 7 to 11 */ | |
460 | #define READ_LOCK_NESTED_BASE 7 | |
461 | #define READ_LOCK_NESTED_OUT (1 << 11) | |
462 | ||
463 | #define READ_PROC_READ_GEN (1 << 12) | |
19d8de31 | 464 | #define READ_PROC_ACCESS_GEN (1 << 13) |
551ac1a3 | 465 | |
19d8de31 MD |
466 | /* PROCEDURE_READ_UNLOCK (NESTED) base = << 14 : 14 to 15 */ |
467 | #define READ_UNLOCK_NESTED_BASE 14 | |
468 | #define READ_UNLOCK_NESTED_OUT (1 << 15) | |
551ac1a3 | 469 | |
19d8de31 | 470 | #define READ_PROC_SECOND_MB (1 << 16) |
551ac1a3 | 471 | |
19d8de31 MD |
472 | /* PROCEDURE_READ_UNLOCK base = << 17 : 17 to 18 */ |
473 | #define READ_UNLOCK_BASE 17 | |
474 | #define READ_UNLOCK_OUT (1 << 18) | |
551ac1a3 | 475 | |
19d8de31 MD |
476 | /* PROCEDURE_READ_LOCK_UNROLL base = << 19 : 19 to 23 */ |
477 | #define READ_LOCK_UNROLL_BASE 19 | |
478 | #define READ_LOCK_OUT_UNROLL (1 << 23) | |
551ac1a3 | 479 | |
19d8de31 | 480 | #define READ_PROC_THIRD_MB (1 << 24) |
551ac1a3 | 481 | |
19d8de31 MD |
482 | #define READ_PROC_READ_GEN_UNROLL (1 << 25) |
483 | #define READ_PROC_ACCESS_GEN_UNROLL (1 << 26) | |
551ac1a3 | 484 | |
19d8de31 | 485 | #define READ_PROC_FOURTH_MB (1 << 27) |
551ac1a3 | 486 | |
19d8de31 MD |
487 | /* PROCEDURE_READ_UNLOCK_UNROLL base = << 28 : 28 to 29 */ |
488 | #define READ_UNLOCK_UNROLL_BASE 28 | |
489 | #define READ_UNLOCK_OUT_UNROLL (1 << 29) | |
551ac1a3 MD |
490 | |
491 | ||
492 | /* Should not include branches */ | |
493 | #define READ_PROC_ALL_TOKENS (READ_PROD_NONE \ | |
494 | | READ_LOCK_OUT \ | |
495 | | READ_PROC_FIRST_MB \ | |
496 | | READ_LOCK_NESTED_OUT \ | |
497 | | READ_PROC_READ_GEN \ | |
19d8de31 | 498 | | READ_PROC_ACCESS_GEN \ |
551ac1a3 MD |
499 | | READ_UNLOCK_NESTED_OUT \ |
500 | | READ_PROC_SECOND_MB \ | |
501 | | READ_UNLOCK_OUT \ | |
502 | | READ_LOCK_OUT_UNROLL \ | |
503 | | READ_PROC_THIRD_MB \ | |
504 | | READ_PROC_READ_GEN_UNROLL \ | |
19d8de31 | 505 | | READ_PROC_ACCESS_GEN_UNROLL \ |
551ac1a3 MD |
506 | | READ_PROC_FOURTH_MB \ |
507 | | READ_UNLOCK_OUT_UNROLL) | |
508 | ||
509 | /* Must clear all tokens, including branches */ | |
19d8de31 | 510 | #define READ_PROC_ALL_TOKENS_CLEAR ((1 << 30) - 1) |
551ac1a3 MD |
511 | |
512 | inline urcu_one_read(i, j, nest_i, tmp, tmp2) | |
513 | { | |
514 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_NONE); | |
515 | ||
516 | #ifdef NO_MB | |
517 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FIRST_MB); | |
518 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_SECOND_MB); | |
519 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_THIRD_MB); | |
520 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FOURTH_MB); | |
521 | #endif | |
522 | ||
523 | #ifdef REMOTE_BARRIERS | |
524 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FIRST_MB); | |
525 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_SECOND_MB); | |
526 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_THIRD_MB); | |
527 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FOURTH_MB); | |
528 | #endif | |
529 | ||
530 | do | |
531 | :: 1 -> | |
532 | ||
533 | #ifdef REMOTE_BARRIERS | |
534 | /* | |
535 | * Signal-based memory barrier will only execute when the | |
536 | * execution order appears in program order. | |
537 | */ | |
538 | if | |
539 | :: 1 -> | |
540 | atomic { | |
541 | if | |
542 | :: CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE, | |
543 | READ_LOCK_OUT | READ_LOCK_NESTED_OUT | |
19d8de31 | 544 | | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
551ac1a3 MD |
545 | | READ_UNLOCK_OUT |
546 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 547 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 MD |
548 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT, |
549 | READ_LOCK_NESTED_OUT | |
19d8de31 | 550 | | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
551ac1a3 MD |
551 | | READ_UNLOCK_OUT |
552 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 553 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 554 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT | READ_LOCK_NESTED_OUT, |
19d8de31 | 555 | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
551ac1a3 MD |
556 | | READ_UNLOCK_OUT |
557 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 558 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 MD |
559 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
560 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN, | |
19d8de31 MD |
561 | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
562 | | READ_UNLOCK_OUT | |
563 | | READ_LOCK_OUT_UNROLL | |
564 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) | |
565 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT | |
566 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN, | |
551ac1a3 MD |
567 | READ_UNLOCK_NESTED_OUT |
568 | | READ_UNLOCK_OUT | |
569 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 570 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 571 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
19d8de31 MD |
572 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN |
573 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT, | |
551ac1a3 MD |
574 | READ_UNLOCK_OUT |
575 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 576 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 577 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
19d8de31 MD |
578 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN |
579 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT | |
551ac1a3 MD |
580 | | READ_UNLOCK_OUT, |
581 | READ_LOCK_OUT_UNROLL | |
19d8de31 | 582 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 583 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
19d8de31 MD |
584 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN |
585 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT | |
551ac1a3 | 586 | | READ_UNLOCK_OUT | READ_LOCK_OUT_UNROLL, |
19d8de31 | 587 | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 588 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
19d8de31 MD |
589 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN |
590 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT | |
551ac1a3 MD |
591 | | READ_UNLOCK_OUT | READ_LOCK_OUT_UNROLL |
592 | | READ_PROC_READ_GEN_UNROLL, | |
19d8de31 MD |
593 | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
594 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT | |
595 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN | |
596 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT | |
597 | | READ_UNLOCK_OUT | READ_LOCK_OUT_UNROLL | |
598 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL, | |
551ac1a3 MD |
599 | READ_UNLOCK_OUT_UNROLL) |
600 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT | |
19d8de31 | 601 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
551ac1a3 | 602 | | READ_UNLOCK_OUT | READ_LOCK_OUT_UNROLL |
19d8de31 | 603 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL, |
551ac1a3 MD |
604 | 0) -> |
605 | goto non_atomic3; | |
606 | non_atomic3_end: | |
607 | skip; | |
608 | fi; | |
609 | } | |
610 | :: 1 -> skip; | |
611 | fi; | |
612 | ||
613 | goto non_atomic3_skip; | |
614 | non_atomic3: | |
f089ec24 | 615 | smp_mb_recv(i, j); |
551ac1a3 MD |
616 | goto non_atomic3_end; |
617 | non_atomic3_skip: | |
618 | ||
619 | #endif /* REMOTE_BARRIERS */ | |
620 | ||
621 | atomic { | |
622 | if | |
623 | PROCEDURE_READ_LOCK(READ_LOCK_BASE, READ_PROD_NONE, READ_LOCK_OUT); | |
624 | ||
625 | :: CONSUME_TOKENS(proc_urcu_reader, | |
626 | READ_LOCK_OUT, /* post-dominant */ | |
627 | READ_PROC_FIRST_MB) -> | |
628 | smp_mb_reader(i, j); | |
629 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FIRST_MB); | |
630 | ||
631 | PROCEDURE_READ_LOCK(READ_LOCK_NESTED_BASE, READ_PROC_FIRST_MB | READ_LOCK_OUT, | |
632 | READ_LOCK_NESTED_OUT); | |
633 | ||
634 | :: CONSUME_TOKENS(proc_urcu_reader, | |
635 | READ_PROC_FIRST_MB, /* mb() orders reads */ | |
636 | READ_PROC_READ_GEN) -> | |
637 | ooo_mem(i); | |
638 | read_generation[get_readerid()] = | |
639 | READ_CACHED_VAR(generation_ptr); | |
19d8de31 MD |
640 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_READ_GEN); |
641 | ||
642 | :: CONSUME_TOKENS(proc_urcu_reader, | |
643 | READ_PROC_FIRST_MB /* mb() orders reads */ | |
644 | | READ_PROC_READ_GEN, | |
645 | READ_PROC_ACCESS_GEN) -> | |
646 | ooo_mem(i); | |
551ac1a3 MD |
647 | goto non_atomic; |
648 | non_atomic_end: | |
19d8de31 MD |
649 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_ACCESS_GEN); |
650 | ||
551ac1a3 MD |
651 | |
652 | /* Note : we remove the nested memory barrier from the read unlock | |
653 | * model, given it is not usually needed. The implementation has the barrier | |
654 | * because the performance impact added by a branch in the common case does not | |
655 | * justify it. | |
656 | */ | |
657 | ||
658 | PROCEDURE_READ_UNLOCK(READ_UNLOCK_NESTED_BASE, | |
659 | READ_PROC_FIRST_MB | |
660 | | READ_LOCK_OUT | |
661 | | READ_LOCK_NESTED_OUT, | |
662 | READ_UNLOCK_NESTED_OUT); | |
663 | ||
664 | ||
665 | :: CONSUME_TOKENS(proc_urcu_reader, | |
19d8de31 MD |
666 | READ_PROC_ACCESS_GEN /* mb() orders reads */ |
667 | | READ_PROC_READ_GEN /* mb() orders reads */ | |
551ac1a3 MD |
668 | | READ_PROC_FIRST_MB /* mb() ordered */ |
669 | | READ_LOCK_OUT /* post-dominant */ | |
670 | | READ_LOCK_NESTED_OUT /* post-dominant */ | |
671 | | READ_UNLOCK_NESTED_OUT, | |
672 | READ_PROC_SECOND_MB) -> | |
673 | smp_mb_reader(i, j); | |
674 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_SECOND_MB); | |
675 | ||
676 | PROCEDURE_READ_UNLOCK(READ_UNLOCK_BASE, | |
677 | READ_PROC_SECOND_MB /* mb() orders reads */ | |
678 | | READ_PROC_FIRST_MB /* mb() orders reads */ | |
679 | | READ_LOCK_NESTED_OUT /* RAW */ | |
680 | | READ_LOCK_OUT /* RAW */ | |
681 | | READ_UNLOCK_NESTED_OUT, /* RAW */ | |
682 | READ_UNLOCK_OUT); | |
683 | ||
684 | /* Unrolling loop : second consecutive lock */ | |
685 | /* reading urcu_active_readers, which have been written by | |
686 | * READ_UNLOCK_OUT : RAW */ | |
687 | PROCEDURE_READ_LOCK(READ_LOCK_UNROLL_BASE, | |
688 | READ_UNLOCK_OUT /* RAW */ | |
689 | | READ_PROC_SECOND_MB /* mb() orders reads */ | |
690 | | READ_PROC_FIRST_MB /* mb() orders reads */ | |
691 | | READ_LOCK_NESTED_OUT /* RAW */ | |
692 | | READ_LOCK_OUT /* RAW */ | |
693 | | READ_UNLOCK_NESTED_OUT, /* RAW */ | |
694 | READ_LOCK_OUT_UNROLL); | |
695 | ||
696 | ||
697 | :: CONSUME_TOKENS(proc_urcu_reader, | |
698 | READ_PROC_FIRST_MB /* mb() ordered */ | |
699 | | READ_PROC_SECOND_MB /* mb() ordered */ | |
700 | | READ_LOCK_OUT_UNROLL /* post-dominant */ | |
701 | | READ_LOCK_NESTED_OUT | |
702 | | READ_LOCK_OUT | |
703 | | READ_UNLOCK_NESTED_OUT | |
704 | | READ_UNLOCK_OUT, | |
705 | READ_PROC_THIRD_MB) -> | |
706 | smp_mb_reader(i, j); | |
707 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_THIRD_MB); | |
708 | ||
709 | :: CONSUME_TOKENS(proc_urcu_reader, | |
710 | READ_PROC_FIRST_MB /* mb() orders reads */ | |
711 | | READ_PROC_SECOND_MB /* mb() orders reads */ | |
712 | | READ_PROC_THIRD_MB, /* mb() orders reads */ | |
713 | READ_PROC_READ_GEN_UNROLL) -> | |
714 | ooo_mem(i); | |
715 | read_generation[get_readerid()] = | |
716 | READ_CACHED_VAR(generation_ptr); | |
19d8de31 MD |
717 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_READ_GEN_UNROLL); |
718 | ||
719 | :: CONSUME_TOKENS(proc_urcu_reader, | |
720 | READ_PROC_READ_GEN_UNROLL | |
721 | | READ_PROC_FIRST_MB /* mb() orders reads */ | |
722 | | READ_PROC_SECOND_MB /* mb() orders reads */ | |
723 | | READ_PROC_THIRD_MB, /* mb() orders reads */ | |
724 | READ_PROC_ACCESS_GEN_UNROLL) -> | |
725 | ooo_mem(i); | |
551ac1a3 MD |
726 | goto non_atomic2; |
727 | non_atomic2_end: | |
19d8de31 | 728 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_ACCESS_GEN_UNROLL); |
551ac1a3 MD |
729 | |
730 | :: CONSUME_TOKENS(proc_urcu_reader, | |
731 | READ_PROC_READ_GEN_UNROLL /* mb() orders reads */ | |
19d8de31 | 732 | | READ_PROC_ACCESS_GEN_UNROLL /* mb() orders reads */ |
551ac1a3 MD |
733 | | READ_PROC_FIRST_MB /* mb() ordered */ |
734 | | READ_PROC_SECOND_MB /* mb() ordered */ | |
735 | | READ_PROC_THIRD_MB /* mb() ordered */ | |
736 | | READ_LOCK_OUT_UNROLL /* post-dominant */ | |
737 | | READ_LOCK_NESTED_OUT | |
738 | | READ_LOCK_OUT | |
739 | | READ_UNLOCK_NESTED_OUT | |
740 | | READ_UNLOCK_OUT, | |
741 | READ_PROC_FOURTH_MB) -> | |
742 | smp_mb_reader(i, j); | |
743 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FOURTH_MB); | |
744 | ||
745 | PROCEDURE_READ_UNLOCK(READ_UNLOCK_UNROLL_BASE, | |
746 | READ_PROC_FOURTH_MB /* mb() orders reads */ | |
747 | | READ_PROC_THIRD_MB /* mb() orders reads */ | |
748 | | READ_LOCK_OUT_UNROLL /* RAW */ | |
749 | | READ_PROC_SECOND_MB /* mb() orders reads */ | |
750 | | READ_PROC_FIRST_MB /* mb() orders reads */ | |
751 | | READ_LOCK_NESTED_OUT /* RAW */ | |
752 | | READ_LOCK_OUT /* RAW */ | |
753 | | READ_UNLOCK_NESTED_OUT, /* RAW */ | |
754 | READ_UNLOCK_OUT_UNROLL); | |
755 | :: CONSUME_TOKENS(proc_urcu_reader, READ_PROC_ALL_TOKENS, 0) -> | |
756 | CLEAR_TOKENS(proc_urcu_reader, READ_PROC_ALL_TOKENS_CLEAR); | |
757 | break; | |
758 | fi; | |
759 | } | |
760 | od; | |
761 | /* | |
762 | * Dependency between consecutive loops : | |
763 | * RAW dependency on | |
764 | * WRITE_CACHED_VAR(urcu_active_readers[get_readerid()], tmp2 - 1) | |
765 | * tmp = READ_CACHED_VAR(urcu_active_readers[get_readerid()]); | |
766 | * between loops. | |
767 | * _WHEN THE MB()s are in place_, they add full ordering of the | |
768 | * generation pointer read wrt active reader count read, which ensures | |
769 | * execution will not spill across loop execution. | |
770 | * However, in the event mb()s are removed (execution using signal | |
771 | * handler to promote barrier()() -> smp_mb()), nothing prevents one loop | |
772 | * to spill its execution on other loop's execution. | |
773 | */ | |
774 | goto end; | |
775 | non_atomic: | |
776 | data_access[get_readerid()] = 1; | |
777 | data_access[get_readerid()] = 0; | |
778 | goto non_atomic_end; | |
779 | non_atomic2: | |
780 | data_access[get_readerid()] = 1; | |
781 | data_access[get_readerid()] = 0; | |
782 | goto non_atomic2_end; | |
783 | end: | |
784 | skip; | |
785 | } | |
786 | ||
787 | ||
788 | ||
789 | active proctype urcu_reader() | |
790 | { | |
791 | byte i, j, nest_i; | |
792 | byte tmp, tmp2; | |
793 | ||
794 | wait_init_done(); | |
795 | ||
796 | assert(get_pid() < NR_PROCS); | |
797 | ||
798 | end_reader: | |
799 | do | |
800 | :: 1 -> | |
801 | /* | |
802 | * We do not test reader's progress here, because we are mainly | |
803 | * interested in writer's progress. The reader never blocks | |
804 | * anyway. We have to test for reader/writer's progress | |
805 | * separately, otherwise we could think the writer is doing | |
806 | * progress when it's blocked by an always progressing reader. | |
807 | */ | |
808 | #ifdef READER_PROGRESS | |
809 | progress_reader: | |
810 | #endif | |
811 | urcu_one_read(i, j, nest_i, tmp, tmp2); | |
812 | od; | |
813 | } | |
814 | ||
815 | /* no name clash please */ | |
816 | #undef proc_urcu_reader | |
817 | ||
818 | ||
819 | /* Model the RCU update process. */ | |
820 | ||
821 | /* | |
822 | * Bit encoding, urcu_writer : | |
823 | * Currently only supports one reader. | |
824 | */ | |
825 | ||
826 | int _proc_urcu_writer; | |
827 | #define proc_urcu_writer _proc_urcu_writer | |
828 | ||
829 | #define WRITE_PROD_NONE (1 << 0) | |
830 | ||
831 | #define WRITE_PROC_FIRST_MB (1 << 1) | |
832 | ||
833 | /* first flip */ | |
834 | #define WRITE_PROC_FIRST_READ_GP (1 << 2) | |
835 | #define WRITE_PROC_FIRST_WRITE_GP (1 << 3) | |
836 | #define WRITE_PROC_FIRST_WAIT (1 << 4) | |
837 | #define WRITE_PROC_FIRST_WAIT_LOOP (1 << 5) | |
838 | ||
839 | /* second flip */ | |
840 | #define WRITE_PROC_SECOND_READ_GP (1 << 6) | |
841 | #define WRITE_PROC_SECOND_WRITE_GP (1 << 7) | |
842 | #define WRITE_PROC_SECOND_WAIT (1 << 8) | |
843 | #define WRITE_PROC_SECOND_WAIT_LOOP (1 << 9) | |
844 | ||
845 | #define WRITE_PROC_SECOND_MB (1 << 10) | |
846 | ||
847 | #define WRITE_PROC_ALL_TOKENS (WRITE_PROD_NONE \ | |
848 | | WRITE_PROC_FIRST_MB \ | |
849 | | WRITE_PROC_FIRST_READ_GP \ | |
850 | | WRITE_PROC_FIRST_WRITE_GP \ | |
851 | | WRITE_PROC_FIRST_WAIT \ | |
852 | | WRITE_PROC_SECOND_READ_GP \ | |
853 | | WRITE_PROC_SECOND_WRITE_GP \ | |
854 | | WRITE_PROC_SECOND_WAIT \ | |
855 | | WRITE_PROC_SECOND_MB) | |
856 | ||
857 | #define WRITE_PROC_ALL_TOKENS_CLEAR ((1 << 11) - 1) | |
858 | ||
859 | active proctype urcu_writer() | |
860 | { | |
861 | byte i, j; | |
862 | byte tmp, tmp2, tmpa; | |
863 | byte old_gen; | |
f089ec24 MD |
864 | byte cur_gp_val = 0; /* |
865 | * Keep a local trace of the current parity so | |
866 | * we don't add non-existing dependencies on the global | |
867 | * GP update. Needed to test single flip case. | |
868 | */ | |
551ac1a3 MD |
869 | |
870 | wait_init_done(); | |
871 | ||
872 | assert(get_pid() < NR_PROCS); | |
873 | ||
874 | do | |
875 | :: (READ_CACHED_VAR(generation_ptr) < 5) -> | |
876 | #ifdef WRITER_PROGRESS | |
877 | progress_writer1: | |
878 | #endif | |
879 | ooo_mem(i); | |
880 | atomic { | |
881 | old_gen = READ_CACHED_VAR(generation_ptr); | |
882 | WRITE_CACHED_VAR(generation_ptr, old_gen + 1); | |
883 | } | |
884 | ooo_mem(i); | |
885 | ||
886 | do | |
887 | :: 1 -> | |
888 | atomic { | |
889 | if | |
890 | :: write_lock == 0 -> | |
891 | write_lock = 1; | |
892 | break; | |
893 | :: else -> | |
894 | skip; | |
895 | fi; | |
896 | } | |
897 | od; | |
898 | ||
899 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROD_NONE); | |
900 | ||
901 | #ifdef NO_MB | |
902 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_MB); | |
903 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_MB); | |
904 | #endif | |
905 | ||
906 | #ifdef SINGLE_FLIP | |
907 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_READ_GP); | |
908 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WRITE_GP); | |
909 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WAIT); | |
794a737e MD |
910 | /* For single flip, we need to know the current parity */ |
911 | cur_gp_val = cur_gp_val ^ RCU_GP_CTR_BIT; | |
551ac1a3 MD |
912 | #endif |
913 | ||
f089ec24 MD |
914 | do :: 1 -> |
915 | atomic { | |
916 | if | |
551ac1a3 MD |
917 | :: CONSUME_TOKENS(proc_urcu_writer, |
918 | WRITE_PROD_NONE, | |
919 | WRITE_PROC_FIRST_MB) -> | |
f089ec24 MD |
920 | goto smp_mb_send1; |
921 | smp_mb_send1_end: | |
551ac1a3 MD |
922 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_MB); |
923 | ||
924 | /* first flip */ | |
925 | :: CONSUME_TOKENS(proc_urcu_writer, | |
926 | WRITE_PROC_FIRST_MB, | |
927 | WRITE_PROC_FIRST_READ_GP) -> | |
928 | tmpa = READ_CACHED_VAR(urcu_gp_ctr); | |
929 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_READ_GP); | |
930 | :: CONSUME_TOKENS(proc_urcu_writer, | |
931 | WRITE_PROC_FIRST_MB | WRITE_PROC_FIRST_READ_GP, | |
932 | WRITE_PROC_FIRST_WRITE_GP) -> | |
933 | ooo_mem(i); | |
934 | WRITE_CACHED_VAR(urcu_gp_ctr, tmpa ^ RCU_GP_CTR_BIT); | |
935 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_WRITE_GP); | |
936 | ||
937 | :: CONSUME_TOKENS(proc_urcu_writer, | |
938 | //WRITE_PROC_FIRST_WRITE_GP /* TEST ADDING SYNC CORE */ | |
939 | WRITE_PROC_FIRST_MB, /* can be reordered before/after flips */ | |
940 | WRITE_PROC_FIRST_WAIT | WRITE_PROC_FIRST_WAIT_LOOP) -> | |
941 | ooo_mem(i); | |
942 | /* ONLY WAITING FOR READER 0 */ | |
943 | tmp2 = READ_CACHED_VAR(urcu_active_readers[0]); | |
794a737e MD |
944 | #ifndef SINGLE_FLIP |
945 | /* In normal execution, we are always starting by | |
946 | * waiting for the even parity. | |
947 | */ | |
948 | cur_gp_val = RCU_GP_CTR_BIT; | |
949 | #endif | |
551ac1a3 MD |
950 | if |
951 | :: (tmp2 & RCU_GP_CTR_NEST_MASK) | |
f089ec24 | 952 | && ((tmp2 ^ cur_gp_val) & RCU_GP_CTR_BIT) -> |
551ac1a3 MD |
953 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_WAIT_LOOP); |
954 | :: else -> | |
955 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_WAIT); | |
956 | fi; | |
957 | ||
958 | :: CONSUME_TOKENS(proc_urcu_writer, | |
959 | //WRITE_PROC_FIRST_WRITE_GP /* TEST ADDING SYNC CORE */ | |
960 | WRITE_PROC_FIRST_WRITE_GP | |
961 | | WRITE_PROC_FIRST_READ_GP | |
962 | | WRITE_PROC_FIRST_WAIT_LOOP | |
963 | | WRITE_PROC_FIRST_MB, /* can be reordered before/after flips */ | |
964 | 0) -> | |
965 | #ifndef GEN_ERROR_WRITER_PROGRESS | |
f089ec24 MD |
966 | goto smp_mb_send2; |
967 | smp_mb_send2_end: | |
551ac1a3 MD |
968 | #else |
969 | ooo_mem(i); | |
970 | #endif | |
971 | /* This instruction loops to WRITE_PROC_FIRST_WAIT */ | |
972 | CLEAR_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_WAIT_LOOP | WRITE_PROC_FIRST_WAIT); | |
973 | ||
974 | /* second flip */ | |
975 | :: CONSUME_TOKENS(proc_urcu_writer, | |
976 | WRITE_PROC_FIRST_WAIT /* Control dependency : need to branch out of | |
977 | * the loop to execute the next flip (CHECK) */ | |
978 | | WRITE_PROC_FIRST_WRITE_GP | |
979 | | WRITE_PROC_FIRST_READ_GP | |
980 | | WRITE_PROC_FIRST_MB, | |
981 | WRITE_PROC_SECOND_READ_GP) -> | |
551ac1a3 MD |
982 | ooo_mem(i); |
983 | tmpa = READ_CACHED_VAR(urcu_gp_ctr); | |
984 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_READ_GP); | |
985 | :: CONSUME_TOKENS(proc_urcu_writer, | |
986 | WRITE_PROC_FIRST_MB | |
987 | | WRITE_PROC_FIRST_READ_GP | |
988 | | WRITE_PROC_FIRST_WRITE_GP | |
989 | | WRITE_PROC_SECOND_READ_GP, | |
990 | WRITE_PROC_SECOND_WRITE_GP) -> | |
991 | ooo_mem(i); | |
992 | WRITE_CACHED_VAR(urcu_gp_ctr, tmpa ^ RCU_GP_CTR_BIT); | |
993 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WRITE_GP); | |
994 | ||
995 | :: CONSUME_TOKENS(proc_urcu_writer, | |
996 | //WRITE_PROC_FIRST_WRITE_GP /* TEST ADDING SYNC CORE */ | |
997 | WRITE_PROC_FIRST_WAIT | |
998 | | WRITE_PROC_FIRST_MB, /* can be reordered before/after flips */ | |
999 | WRITE_PROC_SECOND_WAIT | WRITE_PROC_SECOND_WAIT_LOOP) -> | |
1000 | ooo_mem(i); | |
1001 | /* ONLY WAITING FOR READER 0 */ | |
1002 | tmp2 = READ_CACHED_VAR(urcu_active_readers[0]); | |
1003 | if | |
1004 | :: (tmp2 & RCU_GP_CTR_NEST_MASK) | |
794a737e | 1005 | && ((tmp2 ^ 0) & RCU_GP_CTR_BIT) -> |
551ac1a3 MD |
1006 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WAIT_LOOP); |
1007 | :: else -> | |
1008 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WAIT); | |
1009 | fi; | |
1010 | ||
1011 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1012 | //WRITE_PROC_FIRST_WRITE_GP /* TEST ADDING SYNC CORE */ | |
1013 | WRITE_PROC_SECOND_WRITE_GP | |
1014 | | WRITE_PROC_FIRST_WRITE_GP | |
1015 | | WRITE_PROC_SECOND_READ_GP | |
1016 | | WRITE_PROC_FIRST_READ_GP | |
1017 | | WRITE_PROC_SECOND_WAIT_LOOP | |
1018 | | WRITE_PROC_FIRST_MB, /* can be reordered before/after flips */ | |
1019 | 0) -> | |
1020 | #ifndef GEN_ERROR_WRITER_PROGRESS | |
f089ec24 MD |
1021 | goto smp_mb_send3; |
1022 | smp_mb_send3_end: | |
551ac1a3 MD |
1023 | #else |
1024 | ooo_mem(i); | |
1025 | #endif | |
1026 | /* This instruction loops to WRITE_PROC_SECOND_WAIT */ | |
1027 | CLEAR_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WAIT_LOOP | WRITE_PROC_SECOND_WAIT); | |
1028 | ||
1029 | ||
1030 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1031 | WRITE_PROC_FIRST_WAIT | |
1032 | | WRITE_PROC_SECOND_WAIT | |
1033 | | WRITE_PROC_FIRST_READ_GP | |
1034 | | WRITE_PROC_SECOND_READ_GP | |
1035 | | WRITE_PROC_FIRST_WRITE_GP | |
1036 | | WRITE_PROC_SECOND_WRITE_GP | |
1037 | | WRITE_PROC_FIRST_MB, | |
1038 | WRITE_PROC_SECOND_MB) -> | |
f089ec24 MD |
1039 | goto smp_mb_send4; |
1040 | smp_mb_send4_end: | |
551ac1a3 MD |
1041 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_MB); |
1042 | ||
1043 | :: CONSUME_TOKENS(proc_urcu_writer, WRITE_PROC_ALL_TOKENS, 0) -> | |
1044 | CLEAR_TOKENS(proc_urcu_writer, WRITE_PROC_ALL_TOKENS_CLEAR); | |
1045 | break; | |
f089ec24 MD |
1046 | fi; |
1047 | } | |
551ac1a3 MD |
1048 | od; |
1049 | ||
1050 | write_lock = 0; | |
1051 | /* free-up step, e.g., kfree(). */ | |
1052 | atomic { | |
1053 | last_free_gen = old_gen; | |
1054 | free_done = 1; | |
1055 | } | |
1056 | :: else -> break; | |
1057 | od; | |
1058 | /* | |
1059 | * Given the reader loops infinitely, let the writer also busy-loop | |
1060 | * with progress here so, with weak fairness, we can test the | |
1061 | * writer's progress. | |
1062 | */ | |
1063 | end_writer: | |
1064 | do | |
1065 | :: 1 -> | |
1066 | #ifdef WRITER_PROGRESS | |
1067 | progress_writer2: | |
1068 | #endif | |
1069 | skip; | |
1070 | od; | |
f089ec24 MD |
1071 | |
1072 | /* Non-atomic parts of the loop */ | |
1073 | goto end; | |
1074 | smp_mb_send1: | |
30193782 | 1075 | smp_mb_send(i, j, 1); |
f089ec24 MD |
1076 | goto smp_mb_send1_end; |
1077 | #ifndef GEN_ERROR_WRITER_PROGRESS | |
1078 | smp_mb_send2: | |
30193782 | 1079 | smp_mb_send(i, j, 2); |
f089ec24 MD |
1080 | goto smp_mb_send2_end; |
1081 | smp_mb_send3: | |
30193782 | 1082 | smp_mb_send(i, j, 3); |
f089ec24 MD |
1083 | goto smp_mb_send3_end; |
1084 | #endif | |
1085 | smp_mb_send4: | |
30193782 | 1086 | smp_mb_send(i, j, 4); |
f089ec24 MD |
1087 | goto smp_mb_send4_end; |
1088 | end: | |
1089 | skip; | |
551ac1a3 MD |
1090 | } |
1091 | ||
1092 | /* no name clash please */ | |
1093 | #undef proc_urcu_writer | |
1094 | ||
1095 | ||
1096 | /* Leave after the readers and writers so the pid count is ok. */ | |
1097 | init { | |
1098 | byte i, j; | |
1099 | ||
1100 | atomic { | |
1101 | INIT_CACHED_VAR(urcu_gp_ctr, 1, j); | |
1102 | INIT_CACHED_VAR(generation_ptr, 0, j); | |
1103 | ||
1104 | i = 0; | |
1105 | do | |
1106 | :: i < NR_READERS -> | |
1107 | INIT_CACHED_VAR(urcu_active_readers[i], 0, j); | |
1108 | read_generation[i] = 1; | |
1109 | data_access[i] = 0; | |
1110 | i++; | |
1111 | :: i >= NR_READERS -> break | |
1112 | od; | |
1113 | init_done = 1; | |
1114 | } | |
1115 | } |