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 | ||
f24274b9 MD |
129 | /* |
130 | * Only Alpha has out-of-order cache bank loads. Other architectures (intel, | |
131 | * powerpc, arm) ensure that dependent reads won't be reordered. c.f. | |
132 | * http://www.linuxjournal.com/article/8212) | |
133 | #ifdef ARCH_ALPHA | |
134 | #define HAVE_OOO_CACHE_READ | |
135 | #endif | |
136 | ||
551ac1a3 MD |
137 | /* |
138 | * Each process have its own data in cache. Caches are randomly updated. | |
139 | * smp_wmb and smp_rmb forces cache updates (write and read), smp_mb forces | |
140 | * both. | |
141 | */ | |
142 | ||
143 | typedef per_proc_byte { | |
144 | byte val[NR_PROCS]; | |
145 | }; | |
146 | ||
551ac1a3 | 147 | typedef per_proc_bit { |
caeea74c MD |
148 | bit val[NR_PROCS]; |
149 | }; | |
150 | ||
151 | /* Bitfield has a maximum of 8 procs */ | |
152 | typedef per_proc_bitfield { | |
551ac1a3 MD |
153 | byte bitfield; |
154 | }; | |
155 | ||
156 | #define DECLARE_CACHED_VAR(type, x) \ | |
157 | type mem_##x; \ | |
158 | per_proc_##type cached_##x; \ | |
caeea74c | 159 | per_proc_bitfield cache_dirty_##x; |
551ac1a3 MD |
160 | |
161 | #define INIT_CACHED_VAR(x, v, j) \ | |
162 | mem_##x = v; \ | |
163 | cache_dirty_##x.bitfield = 0; \ | |
164 | j = 0; \ | |
165 | do \ | |
166 | :: j < NR_PROCS -> \ | |
167 | cached_##x.val[j] = v; \ | |
168 | j++ \ | |
169 | :: j >= NR_PROCS -> break \ | |
170 | od; | |
171 | ||
172 | #define IS_CACHE_DIRTY(x, id) (cache_dirty_##x.bitfield & (1 << id)) | |
173 | ||
174 | #define READ_CACHED_VAR(x) (cached_##x.val[get_pid()]) | |
175 | ||
176 | #define WRITE_CACHED_VAR(x, v) \ | |
177 | atomic { \ | |
178 | cached_##x.val[get_pid()] = v; \ | |
179 | cache_dirty_##x.bitfield = \ | |
180 | cache_dirty_##x.bitfield | (1 << get_pid()); \ | |
181 | } | |
182 | ||
183 | #define CACHE_WRITE_TO_MEM(x, id) \ | |
184 | if \ | |
185 | :: IS_CACHE_DIRTY(x, id) -> \ | |
186 | mem_##x = cached_##x.val[id]; \ | |
187 | cache_dirty_##x.bitfield = \ | |
188 | cache_dirty_##x.bitfield & (~(1 << id)); \ | |
189 | :: else -> \ | |
190 | skip \ | |
191 | fi; | |
192 | ||
193 | #define CACHE_READ_FROM_MEM(x, id) \ | |
194 | if \ | |
195 | :: !IS_CACHE_DIRTY(x, id) -> \ | |
196 | cached_##x.val[id] = mem_##x;\ | |
197 | :: else -> \ | |
198 | skip \ | |
199 | fi; | |
200 | ||
201 | /* | |
202 | * May update other caches if cache is dirty, or not. | |
203 | */ | |
204 | #define RANDOM_CACHE_WRITE_TO_MEM(x, id)\ | |
205 | if \ | |
206 | :: 1 -> CACHE_WRITE_TO_MEM(x, id); \ | |
207 | :: 1 -> skip \ | |
208 | fi; | |
209 | ||
210 | #define RANDOM_CACHE_READ_FROM_MEM(x, id)\ | |
211 | if \ | |
212 | :: 1 -> CACHE_READ_FROM_MEM(x, id); \ | |
213 | :: 1 -> skip \ | |
214 | fi; | |
215 | ||
216 | /* Must consume all prior read tokens. All subsequent reads depend on it. */ | |
f24274b9 | 217 | inline smp_rmb(i) |
551ac1a3 MD |
218 | { |
219 | atomic { | |
220 | CACHE_READ_FROM_MEM(urcu_gp_ctr, get_pid()); | |
221 | i = 0; | |
222 | do | |
223 | :: i < NR_READERS -> | |
224 | CACHE_READ_FROM_MEM(urcu_active_readers[i], get_pid()); | |
225 | i++ | |
226 | :: i >= NR_READERS -> break | |
227 | od; | |
6af482a9 MD |
228 | CACHE_READ_FROM_MEM(rcu_ptr, get_pid()); |
229 | i = 0; | |
230 | do | |
231 | :: i < SLAB_SIZE -> | |
232 | CACHE_READ_FROM_MEM(rcu_data[i], get_pid()); | |
233 | i++ | |
234 | :: i >= SLAB_SIZE -> break | |
235 | od; | |
551ac1a3 MD |
236 | } |
237 | } | |
238 | ||
239 | /* Must consume all prior write tokens. All subsequent writes depend on it. */ | |
f24274b9 | 240 | inline smp_wmb(i) |
551ac1a3 MD |
241 | { |
242 | atomic { | |
243 | CACHE_WRITE_TO_MEM(urcu_gp_ctr, get_pid()); | |
244 | i = 0; | |
245 | do | |
246 | :: i < NR_READERS -> | |
247 | CACHE_WRITE_TO_MEM(urcu_active_readers[i], get_pid()); | |
248 | i++ | |
249 | :: i >= NR_READERS -> break | |
250 | od; | |
6af482a9 MD |
251 | CACHE_WRITE_TO_MEM(rcu_ptr, get_pid()); |
252 | i = 0; | |
253 | do | |
254 | :: i < SLAB_SIZE -> | |
255 | CACHE_WRITE_TO_MEM(rcu_data[i], get_pid()); | |
256 | i++ | |
257 | :: i >= SLAB_SIZE -> break | |
258 | od; | |
551ac1a3 MD |
259 | } |
260 | } | |
261 | ||
262 | /* Synchronization point. Must consume all prior read and write tokens. All | |
263 | * subsequent reads and writes depend on it. */ | |
f24274b9 | 264 | inline smp_mb(i) |
551ac1a3 MD |
265 | { |
266 | atomic { | |
f24274b9 MD |
267 | smp_wmb(i); |
268 | smp_rmb(i); | |
551ac1a3 MD |
269 | } |
270 | } | |
271 | ||
551ac1a3 MD |
272 | #ifdef REMOTE_BARRIERS |
273 | ||
274 | bit reader_barrier[NR_READERS]; | |
275 | ||
276 | /* | |
277 | * We cannot leave the barriers dependencies in place in REMOTE_BARRIERS mode | |
278 | * because they would add unexisting core synchronization and would therefore | |
279 | * create an incomplete model. | |
280 | * Therefore, we model the read-side memory barriers by completely disabling the | |
281 | * memory barriers and their dependencies from the read-side. One at a time | |
282 | * (different verification runs), we make a different instruction listen for | |
283 | * signals. | |
284 | */ | |
285 | ||
286 | #define smp_mb_reader(i, j) | |
287 | ||
288 | /* | |
289 | * Service 0, 1 or many barrier requests. | |
290 | */ | |
291 | inline smp_mb_recv(i, j) | |
292 | { | |
293 | do | |
294 | :: (reader_barrier[get_readerid()] == 1) -> | |
6b0de963 MD |
295 | /* |
296 | * We choose to ignore cycles caused by writer busy-looping, | |
297 | * waiting for the reader, sending barrier requests, and the | |
298 | * reader always services them without continuing execution. | |
299 | */ | |
300 | progress_ignoring_mb1: | |
f24274b9 | 301 | smp_mb(i); |
551ac1a3 | 302 | reader_barrier[get_readerid()] = 0; |
6af482a9 | 303 | :: 1 -> |
6b0de963 MD |
304 | /* |
305 | * We choose to ignore writer's non-progress caused by the | |
306 | * reader ignoring the writer's mb() requests. | |
307 | */ | |
308 | progress_ignoring_mb2: | |
6af482a9 | 309 | break; |
551ac1a3 MD |
310 | od; |
311 | } | |
312 | ||
6b0de963 | 313 | #define PROGRESS_LABEL(progressid) progress_writer_progid_##progressid: |
30193782 MD |
314 | |
315 | #define smp_mb_send(i, j, progressid) \ | |
316 | { \ | |
f24274b9 | 317 | smp_mb(i); \ |
30193782 MD |
318 | i = 0; \ |
319 | do \ | |
320 | :: i < NR_READERS -> \ | |
321 | reader_barrier[i] = 1; \ | |
30193782 MD |
322 | /* \ |
323 | * Busy-looping waiting for reader barrier handling is of little\ | |
324 | * interest, given the reader has the ability to totally ignore \ | |
325 | * barrier requests. \ | |
326 | */ \ | |
6af482a9 | 327 | do \ |
6b0de963 MD |
328 | :: (reader_barrier[i] == 1) -> \ |
329 | PROGRESS_LABEL(progressid) \ | |
330 | skip; \ | |
30193782 MD |
331 | :: (reader_barrier[i] == 0) -> break; \ |
332 | od; \ | |
333 | i++; \ | |
334 | :: i >= NR_READERS -> \ | |
335 | break \ | |
336 | od; \ | |
f24274b9 | 337 | smp_mb(i); \ |
551ac1a3 MD |
338 | } |
339 | ||
340 | #else | |
341 | ||
f24274b9 MD |
342 | #define smp_mb_send(i, j, progressid) smp_mb(i) |
343 | #define smp_mb_reader smp_mb(i) | |
551ac1a3 MD |
344 | #define smp_mb_recv(i, j) |
345 | ||
346 | #endif | |
347 | ||
6af482a9 | 348 | /* Keep in sync manually with smp_rmb, smp_wmb, ooo_mem and init() */ |
551ac1a3 | 349 | DECLARE_CACHED_VAR(byte, urcu_gp_ctr); |
caeea74c | 350 | /* Note ! currently only one reader */ |
551ac1a3 | 351 | DECLARE_CACHED_VAR(byte, urcu_active_readers[NR_READERS]); |
caeea74c MD |
352 | /* RCU data */ |
353 | DECLARE_CACHED_VAR(bit, rcu_data[SLAB_SIZE]); | |
354 | ||
6af482a9 | 355 | /* RCU pointer */ |
caeea74c MD |
356 | #if (SLAB_SIZE == 2) |
357 | DECLARE_CACHED_VAR(bit, rcu_ptr); | |
358 | bit ptr_read_first[NR_READERS]; | |
359 | bit ptr_read_second[NR_READERS]; | |
360 | #else | |
6af482a9 | 361 | DECLARE_CACHED_VAR(byte, rcu_ptr); |
caeea74c MD |
362 | byte ptr_read_first[NR_READERS]; |
363 | byte ptr_read_second[NR_READERS]; | |
364 | #endif | |
551ac1a3 | 365 | |
caeea74c MD |
366 | bit data_read_first[NR_READERS]; |
367 | bit data_read_second[NR_READERS]; | |
551ac1a3 MD |
368 | |
369 | bit init_done = 0; | |
370 | ||
551ac1a3 MD |
371 | inline wait_init_done() |
372 | { | |
373 | do | |
374 | :: init_done == 0 -> skip; | |
375 | :: else -> break; | |
376 | od; | |
377 | } | |
378 | ||
379 | inline ooo_mem(i) | |
380 | { | |
381 | atomic { | |
382 | RANDOM_CACHE_WRITE_TO_MEM(urcu_gp_ctr, get_pid()); | |
383 | i = 0; | |
384 | do | |
385 | :: i < NR_READERS -> | |
386 | RANDOM_CACHE_WRITE_TO_MEM(urcu_active_readers[i], | |
387 | get_pid()); | |
388 | i++ | |
389 | :: i >= NR_READERS -> break | |
390 | od; | |
6af482a9 MD |
391 | RANDOM_CACHE_WRITE_TO_MEM(rcu_ptr, get_pid()); |
392 | i = 0; | |
393 | do | |
394 | :: i < SLAB_SIZE -> | |
395 | RANDOM_CACHE_WRITE_TO_MEM(rcu_data[i], get_pid()); | |
396 | i++ | |
397 | :: i >= SLAB_SIZE -> break | |
398 | od; | |
f24274b9 | 399 | #ifdef HAVE_OOO_CACHE_READ |
551ac1a3 MD |
400 | RANDOM_CACHE_READ_FROM_MEM(urcu_gp_ctr, get_pid()); |
401 | i = 0; | |
402 | do | |
403 | :: i < NR_READERS -> | |
404 | RANDOM_CACHE_READ_FROM_MEM(urcu_active_readers[i], | |
405 | get_pid()); | |
406 | i++ | |
407 | :: i >= NR_READERS -> break | |
408 | od; | |
6af482a9 MD |
409 | RANDOM_CACHE_READ_FROM_MEM(rcu_ptr, get_pid()); |
410 | i = 0; | |
411 | do | |
412 | :: i < SLAB_SIZE -> | |
413 | RANDOM_CACHE_READ_FROM_MEM(rcu_data[i], get_pid()); | |
414 | i++ | |
415 | :: i >= SLAB_SIZE -> break | |
416 | od; | |
f24274b9 MD |
417 | #else |
418 | smp_rmb(i); | |
419 | #endif /* HAVE_OOO_CACHE_READ */ | |
551ac1a3 MD |
420 | } |
421 | } | |
422 | ||
423 | /* | |
424 | * Bit encoding, urcu_reader : | |
425 | */ | |
426 | ||
427 | int _proc_urcu_reader; | |
428 | #define proc_urcu_reader _proc_urcu_reader | |
429 | ||
430 | /* Body of PROCEDURE_READ_LOCK */ | |
431 | #define READ_PROD_A_READ (1 << 0) | |
432 | #define READ_PROD_B_IF_TRUE (1 << 1) | |
433 | #define READ_PROD_B_IF_FALSE (1 << 2) | |
434 | #define READ_PROD_C_IF_TRUE_READ (1 << 3) | |
435 | ||
436 | #define PROCEDURE_READ_LOCK(base, consumetoken, producetoken) \ | |
437 | :: CONSUME_TOKENS(proc_urcu_reader, consumetoken, READ_PROD_A_READ << base) -> \ | |
438 | ooo_mem(i); \ | |
439 | tmp = READ_CACHED_VAR(urcu_active_readers[get_readerid()]); \ | |
440 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_A_READ << base); \ | |
441 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
442 | READ_PROD_A_READ << base, /* RAW, pre-dominant */ \ | |
443 | (READ_PROD_B_IF_TRUE | READ_PROD_B_IF_FALSE) << base) -> \ | |
444 | if \ | |
445 | :: (!(tmp & RCU_GP_CTR_NEST_MASK)) -> \ | |
446 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_B_IF_TRUE << base); \ | |
447 | :: else -> \ | |
448 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_B_IF_FALSE << base); \ | |
449 | fi; \ | |
450 | /* IF TRUE */ \ | |
451 | :: CONSUME_TOKENS(proc_urcu_reader, READ_PROD_B_IF_TRUE << base, \ | |
452 | READ_PROD_C_IF_TRUE_READ << base) -> \ | |
453 | ooo_mem(i); \ | |
454 | tmp2 = READ_CACHED_VAR(urcu_gp_ctr); \ | |
455 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_C_IF_TRUE_READ << base); \ | |
456 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
457 | (READ_PROD_C_IF_TRUE_READ /* pre-dominant */ \ | |
458 | | READ_PROD_A_READ) << base, /* WAR */ \ | |
459 | producetoken) -> \ | |
460 | ooo_mem(i); \ | |
461 | WRITE_CACHED_VAR(urcu_active_readers[get_readerid()], tmp2); \ | |
462 | PRODUCE_TOKENS(proc_urcu_reader, producetoken); \ | |
463 | /* IF_MERGE implies \ | |
464 | * post-dominance */ \ | |
465 | /* ELSE */ \ | |
466 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
467 | (READ_PROD_B_IF_FALSE /* pre-dominant */ \ | |
468 | | READ_PROD_A_READ) << base, /* WAR */ \ | |
469 | producetoken) -> \ | |
470 | ooo_mem(i); \ | |
471 | WRITE_CACHED_VAR(urcu_active_readers[get_readerid()], \ | |
472 | tmp + 1); \ | |
473 | PRODUCE_TOKENS(proc_urcu_reader, producetoken); \ | |
474 | /* IF_MERGE implies \ | |
475 | * post-dominance */ \ | |
476 | /* ENDIF */ \ | |
477 | skip | |
478 | ||
479 | /* Body of PROCEDURE_READ_LOCK */ | |
480 | #define READ_PROC_READ_UNLOCK (1 << 0) | |
481 | ||
482 | #define PROCEDURE_READ_UNLOCK(base, consumetoken, producetoken) \ | |
483 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
484 | consumetoken, \ | |
485 | READ_PROC_READ_UNLOCK << base) -> \ | |
486 | ooo_mem(i); \ | |
487 | tmp2 = READ_CACHED_VAR(urcu_active_readers[get_readerid()]); \ | |
488 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_READ_UNLOCK << base); \ | |
489 | :: CONSUME_TOKENS(proc_urcu_reader, \ | |
490 | consumetoken \ | |
491 | | (READ_PROC_READ_UNLOCK << base), /* WAR */ \ | |
492 | producetoken) -> \ | |
493 | ooo_mem(i); \ | |
494 | WRITE_CACHED_VAR(urcu_active_readers[get_readerid()], tmp2 - 1); \ | |
495 | PRODUCE_TOKENS(proc_urcu_reader, producetoken); \ | |
496 | skip | |
497 | ||
498 | ||
499 | #define READ_PROD_NONE (1 << 0) | |
500 | ||
501 | /* PROCEDURE_READ_LOCK base = << 1 : 1 to 5 */ | |
502 | #define READ_LOCK_BASE 1 | |
503 | #define READ_LOCK_OUT (1 << 5) | |
504 | ||
505 | #define READ_PROC_FIRST_MB (1 << 6) | |
506 | ||
507 | /* PROCEDURE_READ_LOCK (NESTED) base : << 7 : 7 to 11 */ | |
508 | #define READ_LOCK_NESTED_BASE 7 | |
509 | #define READ_LOCK_NESTED_OUT (1 << 11) | |
510 | ||
511 | #define READ_PROC_READ_GEN (1 << 12) | |
19d8de31 | 512 | #define READ_PROC_ACCESS_GEN (1 << 13) |
551ac1a3 | 513 | |
19d8de31 MD |
514 | /* PROCEDURE_READ_UNLOCK (NESTED) base = << 14 : 14 to 15 */ |
515 | #define READ_UNLOCK_NESTED_BASE 14 | |
516 | #define READ_UNLOCK_NESTED_OUT (1 << 15) | |
551ac1a3 | 517 | |
19d8de31 | 518 | #define READ_PROC_SECOND_MB (1 << 16) |
551ac1a3 | 519 | |
19d8de31 MD |
520 | /* PROCEDURE_READ_UNLOCK base = << 17 : 17 to 18 */ |
521 | #define READ_UNLOCK_BASE 17 | |
522 | #define READ_UNLOCK_OUT (1 << 18) | |
551ac1a3 | 523 | |
19d8de31 MD |
524 | /* PROCEDURE_READ_LOCK_UNROLL base = << 19 : 19 to 23 */ |
525 | #define READ_LOCK_UNROLL_BASE 19 | |
526 | #define READ_LOCK_OUT_UNROLL (1 << 23) | |
551ac1a3 | 527 | |
19d8de31 | 528 | #define READ_PROC_THIRD_MB (1 << 24) |
551ac1a3 | 529 | |
19d8de31 MD |
530 | #define READ_PROC_READ_GEN_UNROLL (1 << 25) |
531 | #define READ_PROC_ACCESS_GEN_UNROLL (1 << 26) | |
551ac1a3 | 532 | |
19d8de31 | 533 | #define READ_PROC_FOURTH_MB (1 << 27) |
551ac1a3 | 534 | |
19d8de31 MD |
535 | /* PROCEDURE_READ_UNLOCK_UNROLL base = << 28 : 28 to 29 */ |
536 | #define READ_UNLOCK_UNROLL_BASE 28 | |
537 | #define READ_UNLOCK_OUT_UNROLL (1 << 29) | |
551ac1a3 MD |
538 | |
539 | ||
540 | /* Should not include branches */ | |
541 | #define READ_PROC_ALL_TOKENS (READ_PROD_NONE \ | |
542 | | READ_LOCK_OUT \ | |
543 | | READ_PROC_FIRST_MB \ | |
544 | | READ_LOCK_NESTED_OUT \ | |
545 | | READ_PROC_READ_GEN \ | |
19d8de31 | 546 | | READ_PROC_ACCESS_GEN \ |
551ac1a3 MD |
547 | | READ_UNLOCK_NESTED_OUT \ |
548 | | READ_PROC_SECOND_MB \ | |
549 | | READ_UNLOCK_OUT \ | |
550 | | READ_LOCK_OUT_UNROLL \ | |
551 | | READ_PROC_THIRD_MB \ | |
552 | | READ_PROC_READ_GEN_UNROLL \ | |
19d8de31 | 553 | | READ_PROC_ACCESS_GEN_UNROLL \ |
551ac1a3 MD |
554 | | READ_PROC_FOURTH_MB \ |
555 | | READ_UNLOCK_OUT_UNROLL) | |
556 | ||
557 | /* Must clear all tokens, including branches */ | |
19d8de31 | 558 | #define READ_PROC_ALL_TOKENS_CLEAR ((1 << 30) - 1) |
551ac1a3 MD |
559 | |
560 | inline urcu_one_read(i, j, nest_i, tmp, tmp2) | |
561 | { | |
562 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROD_NONE); | |
563 | ||
564 | #ifdef NO_MB | |
565 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FIRST_MB); | |
566 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_SECOND_MB); | |
567 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_THIRD_MB); | |
568 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FOURTH_MB); | |
569 | #endif | |
570 | ||
571 | #ifdef REMOTE_BARRIERS | |
572 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FIRST_MB); | |
573 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_SECOND_MB); | |
574 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_THIRD_MB); | |
575 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FOURTH_MB); | |
576 | #endif | |
577 | ||
578 | do | |
579 | :: 1 -> | |
580 | ||
581 | #ifdef REMOTE_BARRIERS | |
582 | /* | |
583 | * Signal-based memory barrier will only execute when the | |
584 | * execution order appears in program order. | |
585 | */ | |
586 | if | |
587 | :: 1 -> | |
588 | atomic { | |
589 | if | |
590 | :: CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE, | |
591 | READ_LOCK_OUT | READ_LOCK_NESTED_OUT | |
19d8de31 | 592 | | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
551ac1a3 MD |
593 | | READ_UNLOCK_OUT |
594 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 595 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 MD |
596 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT, |
597 | READ_LOCK_NESTED_OUT | |
19d8de31 | 598 | | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
551ac1a3 MD |
599 | | READ_UNLOCK_OUT |
600 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 601 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 602 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT | READ_LOCK_NESTED_OUT, |
19d8de31 | 603 | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
551ac1a3 MD |
604 | | READ_UNLOCK_OUT |
605 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 606 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 MD |
607 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
608 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN, | |
19d8de31 MD |
609 | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
610 | | READ_UNLOCK_OUT | |
611 | | READ_LOCK_OUT_UNROLL | |
612 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) | |
613 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT | |
614 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN, | |
551ac1a3 MD |
615 | READ_UNLOCK_NESTED_OUT |
616 | | READ_UNLOCK_OUT | |
617 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 618 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 619 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
19d8de31 MD |
620 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN |
621 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT, | |
551ac1a3 MD |
622 | READ_UNLOCK_OUT |
623 | | READ_LOCK_OUT_UNROLL | |
19d8de31 | 624 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 625 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
19d8de31 MD |
626 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN |
627 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT | |
551ac1a3 MD |
628 | | READ_UNLOCK_OUT, |
629 | READ_LOCK_OUT_UNROLL | |
19d8de31 | 630 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 631 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
19d8de31 MD |
632 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN |
633 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT | |
551ac1a3 | 634 | | READ_UNLOCK_OUT | READ_LOCK_OUT_UNROLL, |
19d8de31 | 635 | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
551ac1a3 | 636 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT |
19d8de31 MD |
637 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN |
638 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT | |
551ac1a3 MD |
639 | | READ_UNLOCK_OUT | READ_LOCK_OUT_UNROLL |
640 | | READ_PROC_READ_GEN_UNROLL, | |
19d8de31 MD |
641 | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL) |
642 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT | |
643 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN | |
644 | | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT | |
645 | | READ_UNLOCK_OUT | READ_LOCK_OUT_UNROLL | |
646 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL, | |
551ac1a3 MD |
647 | READ_UNLOCK_OUT_UNROLL) |
648 | || CONSUME_TOKENS(proc_urcu_reader, READ_PROD_NONE | READ_LOCK_OUT | |
19d8de31 | 649 | | READ_LOCK_NESTED_OUT | READ_PROC_READ_GEN | READ_PROC_ACCESS_GEN | READ_UNLOCK_NESTED_OUT |
551ac1a3 | 650 | | READ_UNLOCK_OUT | READ_LOCK_OUT_UNROLL |
19d8de31 | 651 | | READ_PROC_READ_GEN_UNROLL | READ_PROC_ACCESS_GEN_UNROLL | READ_UNLOCK_OUT_UNROLL, |
551ac1a3 MD |
652 | 0) -> |
653 | goto non_atomic3; | |
654 | non_atomic3_end: | |
655 | skip; | |
656 | fi; | |
657 | } | |
551ac1a3 MD |
658 | fi; |
659 | ||
660 | goto non_atomic3_skip; | |
661 | non_atomic3: | |
f089ec24 | 662 | smp_mb_recv(i, j); |
551ac1a3 MD |
663 | goto non_atomic3_end; |
664 | non_atomic3_skip: | |
665 | ||
666 | #endif /* REMOTE_BARRIERS */ | |
667 | ||
668 | atomic { | |
669 | if | |
670 | PROCEDURE_READ_LOCK(READ_LOCK_BASE, READ_PROD_NONE, READ_LOCK_OUT); | |
671 | ||
672 | :: CONSUME_TOKENS(proc_urcu_reader, | |
673 | READ_LOCK_OUT, /* post-dominant */ | |
674 | READ_PROC_FIRST_MB) -> | |
675 | smp_mb_reader(i, j); | |
676 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FIRST_MB); | |
677 | ||
678 | PROCEDURE_READ_LOCK(READ_LOCK_NESTED_BASE, READ_PROC_FIRST_MB | READ_LOCK_OUT, | |
679 | READ_LOCK_NESTED_OUT); | |
680 | ||
681 | :: CONSUME_TOKENS(proc_urcu_reader, | |
682 | READ_PROC_FIRST_MB, /* mb() orders reads */ | |
683 | READ_PROC_READ_GEN) -> | |
684 | ooo_mem(i); | |
caeea74c | 685 | ptr_read_first[get_readerid()] = READ_CACHED_VAR(rcu_ptr); |
19d8de31 MD |
686 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_READ_GEN); |
687 | ||
688 | :: CONSUME_TOKENS(proc_urcu_reader, | |
689 | READ_PROC_FIRST_MB /* mb() orders reads */ | |
690 | | READ_PROC_READ_GEN, | |
691 | READ_PROC_ACCESS_GEN) -> | |
6af482a9 MD |
692 | /* smp_read_barrier_depends */ |
693 | goto rmb1; | |
694 | rmb1_end: | |
caeea74c MD |
695 | data_read_first[get_readerid()] = |
696 | READ_CACHED_VAR(rcu_data[ptr_read_first[get_readerid()]]); | |
19d8de31 MD |
697 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_ACCESS_GEN); |
698 | ||
551ac1a3 MD |
699 | |
700 | /* Note : we remove the nested memory barrier from the read unlock | |
701 | * model, given it is not usually needed. The implementation has the barrier | |
702 | * because the performance impact added by a branch in the common case does not | |
703 | * justify it. | |
704 | */ | |
705 | ||
706 | PROCEDURE_READ_UNLOCK(READ_UNLOCK_NESTED_BASE, | |
707 | READ_PROC_FIRST_MB | |
708 | | READ_LOCK_OUT | |
709 | | READ_LOCK_NESTED_OUT, | |
710 | READ_UNLOCK_NESTED_OUT); | |
711 | ||
712 | ||
713 | :: CONSUME_TOKENS(proc_urcu_reader, | |
19d8de31 MD |
714 | READ_PROC_ACCESS_GEN /* mb() orders reads */ |
715 | | READ_PROC_READ_GEN /* mb() orders reads */ | |
551ac1a3 MD |
716 | | READ_PROC_FIRST_MB /* mb() ordered */ |
717 | | READ_LOCK_OUT /* post-dominant */ | |
718 | | READ_LOCK_NESTED_OUT /* post-dominant */ | |
719 | | READ_UNLOCK_NESTED_OUT, | |
720 | READ_PROC_SECOND_MB) -> | |
721 | smp_mb_reader(i, j); | |
722 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_SECOND_MB); | |
723 | ||
724 | PROCEDURE_READ_UNLOCK(READ_UNLOCK_BASE, | |
725 | READ_PROC_SECOND_MB /* mb() orders reads */ | |
726 | | READ_PROC_FIRST_MB /* mb() orders reads */ | |
727 | | READ_LOCK_NESTED_OUT /* RAW */ | |
728 | | READ_LOCK_OUT /* RAW */ | |
729 | | READ_UNLOCK_NESTED_OUT, /* RAW */ | |
730 | READ_UNLOCK_OUT); | |
731 | ||
732 | /* Unrolling loop : second consecutive lock */ | |
733 | /* reading urcu_active_readers, which have been written by | |
734 | * READ_UNLOCK_OUT : RAW */ | |
735 | PROCEDURE_READ_LOCK(READ_LOCK_UNROLL_BASE, | |
736 | READ_UNLOCK_OUT /* RAW */ | |
737 | | READ_PROC_SECOND_MB /* mb() orders reads */ | |
738 | | READ_PROC_FIRST_MB /* mb() orders reads */ | |
739 | | READ_LOCK_NESTED_OUT /* RAW */ | |
740 | | READ_LOCK_OUT /* RAW */ | |
741 | | READ_UNLOCK_NESTED_OUT, /* RAW */ | |
742 | READ_LOCK_OUT_UNROLL); | |
743 | ||
744 | ||
745 | :: CONSUME_TOKENS(proc_urcu_reader, | |
746 | READ_PROC_FIRST_MB /* mb() ordered */ | |
747 | | READ_PROC_SECOND_MB /* mb() ordered */ | |
748 | | READ_LOCK_OUT_UNROLL /* post-dominant */ | |
749 | | READ_LOCK_NESTED_OUT | |
750 | | READ_LOCK_OUT | |
751 | | READ_UNLOCK_NESTED_OUT | |
752 | | READ_UNLOCK_OUT, | |
753 | READ_PROC_THIRD_MB) -> | |
754 | smp_mb_reader(i, j); | |
755 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_THIRD_MB); | |
756 | ||
757 | :: CONSUME_TOKENS(proc_urcu_reader, | |
758 | READ_PROC_FIRST_MB /* mb() orders reads */ | |
759 | | READ_PROC_SECOND_MB /* mb() orders reads */ | |
760 | | READ_PROC_THIRD_MB, /* mb() orders reads */ | |
761 | READ_PROC_READ_GEN_UNROLL) -> | |
762 | ooo_mem(i); | |
caeea74c | 763 | ptr_read_second[get_readerid()] = READ_CACHED_VAR(rcu_ptr); |
19d8de31 MD |
764 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_READ_GEN_UNROLL); |
765 | ||
766 | :: CONSUME_TOKENS(proc_urcu_reader, | |
767 | READ_PROC_READ_GEN_UNROLL | |
768 | | READ_PROC_FIRST_MB /* mb() orders reads */ | |
769 | | READ_PROC_SECOND_MB /* mb() orders reads */ | |
770 | | READ_PROC_THIRD_MB, /* mb() orders reads */ | |
771 | READ_PROC_ACCESS_GEN_UNROLL) -> | |
6af482a9 MD |
772 | /* smp_read_barrier_depends */ |
773 | goto rmb2; | |
774 | rmb2_end: | |
caeea74c MD |
775 | data_read_second[get_readerid()] = |
776 | READ_CACHED_VAR(rcu_data[ptr_read_second[get_readerid()]]); | |
19d8de31 | 777 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_ACCESS_GEN_UNROLL); |
551ac1a3 MD |
778 | |
779 | :: CONSUME_TOKENS(proc_urcu_reader, | |
780 | READ_PROC_READ_GEN_UNROLL /* mb() orders reads */ | |
19d8de31 | 781 | | READ_PROC_ACCESS_GEN_UNROLL /* mb() orders reads */ |
551ac1a3 MD |
782 | | READ_PROC_FIRST_MB /* mb() ordered */ |
783 | | READ_PROC_SECOND_MB /* mb() ordered */ | |
784 | | READ_PROC_THIRD_MB /* mb() ordered */ | |
785 | | READ_LOCK_OUT_UNROLL /* post-dominant */ | |
786 | | READ_LOCK_NESTED_OUT | |
787 | | READ_LOCK_OUT | |
788 | | READ_UNLOCK_NESTED_OUT | |
789 | | READ_UNLOCK_OUT, | |
790 | READ_PROC_FOURTH_MB) -> | |
791 | smp_mb_reader(i, j); | |
792 | PRODUCE_TOKENS(proc_urcu_reader, READ_PROC_FOURTH_MB); | |
793 | ||
794 | PROCEDURE_READ_UNLOCK(READ_UNLOCK_UNROLL_BASE, | |
795 | READ_PROC_FOURTH_MB /* mb() orders reads */ | |
796 | | READ_PROC_THIRD_MB /* mb() orders reads */ | |
797 | | READ_LOCK_OUT_UNROLL /* RAW */ | |
798 | | READ_PROC_SECOND_MB /* mb() orders reads */ | |
799 | | READ_PROC_FIRST_MB /* mb() orders reads */ | |
800 | | READ_LOCK_NESTED_OUT /* RAW */ | |
801 | | READ_LOCK_OUT /* RAW */ | |
802 | | READ_UNLOCK_NESTED_OUT, /* RAW */ | |
803 | READ_UNLOCK_OUT_UNROLL); | |
804 | :: CONSUME_TOKENS(proc_urcu_reader, READ_PROC_ALL_TOKENS, 0) -> | |
805 | CLEAR_TOKENS(proc_urcu_reader, READ_PROC_ALL_TOKENS_CLEAR); | |
806 | break; | |
807 | fi; | |
808 | } | |
809 | od; | |
810 | /* | |
811 | * Dependency between consecutive loops : | |
812 | * RAW dependency on | |
813 | * WRITE_CACHED_VAR(urcu_active_readers[get_readerid()], tmp2 - 1) | |
814 | * tmp = READ_CACHED_VAR(urcu_active_readers[get_readerid()]); | |
815 | * between loops. | |
816 | * _WHEN THE MB()s are in place_, they add full ordering of the | |
817 | * generation pointer read wrt active reader count read, which ensures | |
818 | * execution will not spill across loop execution. | |
819 | * However, in the event mb()s are removed (execution using signal | |
820 | * handler to promote barrier()() -> smp_mb()), nothing prevents one loop | |
821 | * to spill its execution on other loop's execution. | |
822 | */ | |
823 | goto end; | |
6af482a9 MD |
824 | rmb1: |
825 | #ifndef NO_RMB | |
f24274b9 | 826 | smp_rmb(i); |
6af482a9 | 827 | #else |
caeea74c | 828 | ooo_mem(i); |
6af482a9 MD |
829 | #endif |
830 | goto rmb1_end; | |
831 | rmb2: | |
832 | #ifndef NO_RMB | |
f24274b9 | 833 | smp_rmb(i); |
6af482a9 | 834 | #else |
caeea74c | 835 | ooo_mem(i); |
6af482a9 MD |
836 | #endif |
837 | goto rmb2_end; | |
551ac1a3 MD |
838 | end: |
839 | skip; | |
840 | } | |
841 | ||
842 | ||
843 | ||
844 | active proctype urcu_reader() | |
845 | { | |
846 | byte i, j, nest_i; | |
847 | byte tmp, tmp2; | |
848 | ||
849 | wait_init_done(); | |
850 | ||
851 | assert(get_pid() < NR_PROCS); | |
852 | ||
853 | end_reader: | |
854 | do | |
855 | :: 1 -> | |
856 | /* | |
857 | * We do not test reader's progress here, because we are mainly | |
858 | * interested in writer's progress. The reader never blocks | |
859 | * anyway. We have to test for reader/writer's progress | |
860 | * separately, otherwise we could think the writer is doing | |
861 | * progress when it's blocked by an always progressing reader. | |
862 | */ | |
863 | #ifdef READER_PROGRESS | |
864 | progress_reader: | |
865 | #endif | |
866 | urcu_one_read(i, j, nest_i, tmp, tmp2); | |
867 | od; | |
868 | } | |
869 | ||
870 | /* no name clash please */ | |
871 | #undef proc_urcu_reader | |
872 | ||
873 | ||
874 | /* Model the RCU update process. */ | |
875 | ||
876 | /* | |
877 | * Bit encoding, urcu_writer : | |
878 | * Currently only supports one reader. | |
879 | */ | |
880 | ||
881 | int _proc_urcu_writer; | |
882 | #define proc_urcu_writer _proc_urcu_writer | |
883 | ||
884 | #define WRITE_PROD_NONE (1 << 0) | |
885 | ||
caeea74c MD |
886 | #define WRITE_DATA (1 << 1) |
887 | #define WRITE_PROC_WMB (1 << 2) | |
888 | #define WRITE_XCHG_PTR (1 << 3) | |
889 | ||
890 | #define WRITE_PROC_FIRST_MB (1 << 4) | |
551ac1a3 MD |
891 | |
892 | /* first flip */ | |
caeea74c MD |
893 | #define WRITE_PROC_FIRST_READ_GP (1 << 5) |
894 | #define WRITE_PROC_FIRST_WRITE_GP (1 << 6) | |
895 | #define WRITE_PROC_FIRST_WAIT (1 << 7) | |
896 | #define WRITE_PROC_FIRST_WAIT_LOOP (1 << 8) | |
551ac1a3 MD |
897 | |
898 | /* second flip */ | |
caeea74c MD |
899 | #define WRITE_PROC_SECOND_READ_GP (1 << 9) |
900 | #define WRITE_PROC_SECOND_WRITE_GP (1 << 10) | |
901 | #define WRITE_PROC_SECOND_WAIT (1 << 11) | |
902 | #define WRITE_PROC_SECOND_WAIT_LOOP (1 << 12) | |
903 | ||
904 | #define WRITE_PROC_SECOND_MB (1 << 13) | |
551ac1a3 | 905 | |
caeea74c | 906 | #define WRITE_FREE (1 << 14) |
551ac1a3 MD |
907 | |
908 | #define WRITE_PROC_ALL_TOKENS (WRITE_PROD_NONE \ | |
caeea74c MD |
909 | | WRITE_DATA \ |
910 | | WRITE_PROC_WMB \ | |
911 | | WRITE_XCHG_PTR \ | |
551ac1a3 MD |
912 | | WRITE_PROC_FIRST_MB \ |
913 | | WRITE_PROC_FIRST_READ_GP \ | |
914 | | WRITE_PROC_FIRST_WRITE_GP \ | |
915 | | WRITE_PROC_FIRST_WAIT \ | |
916 | | WRITE_PROC_SECOND_READ_GP \ | |
917 | | WRITE_PROC_SECOND_WRITE_GP \ | |
918 | | WRITE_PROC_SECOND_WAIT \ | |
caeea74c MD |
919 | | WRITE_PROC_SECOND_MB \ |
920 | | WRITE_FREE) | |
551ac1a3 | 921 | |
caeea74c | 922 | #define WRITE_PROC_ALL_TOKENS_CLEAR ((1 << 15) - 1) |
551ac1a3 | 923 | |
6af482a9 MD |
924 | /* |
925 | * Mutexes are implied around writer execution. A single writer at a time. | |
926 | */ | |
551ac1a3 MD |
927 | active proctype urcu_writer() |
928 | { | |
929 | byte i, j; | |
930 | byte tmp, tmp2, tmpa; | |
6af482a9 | 931 | byte cur_data = 0, old_data, loop_nr = 0; |
f089ec24 MD |
932 | byte cur_gp_val = 0; /* |
933 | * Keep a local trace of the current parity so | |
934 | * we don't add non-existing dependencies on the global | |
935 | * GP update. Needed to test single flip case. | |
936 | */ | |
551ac1a3 MD |
937 | |
938 | wait_init_done(); | |
939 | ||
940 | assert(get_pid() < NR_PROCS); | |
941 | ||
942 | do | |
e5b00154 | 943 | :: (loop_nr < 3) -> |
551ac1a3 MD |
944 | #ifdef WRITER_PROGRESS |
945 | progress_writer1: | |
946 | #endif | |
6af482a9 MD |
947 | loop_nr = loop_nr + 1; |
948 | ||
551ac1a3 MD |
949 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROD_NONE); |
950 | ||
caeea74c MD |
951 | #ifdef NO_WMB |
952 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_WMB); | |
953 | #endif | |
954 | ||
551ac1a3 MD |
955 | #ifdef NO_MB |
956 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_MB); | |
957 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_MB); | |
958 | #endif | |
959 | ||
960 | #ifdef SINGLE_FLIP | |
961 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_READ_GP); | |
962 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WRITE_GP); | |
963 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WAIT); | |
794a737e MD |
964 | /* For single flip, we need to know the current parity */ |
965 | cur_gp_val = cur_gp_val ^ RCU_GP_CTR_BIT; | |
551ac1a3 MD |
966 | #endif |
967 | ||
f089ec24 MD |
968 | do :: 1 -> |
969 | atomic { | |
970 | if | |
caeea74c | 971 | |
551ac1a3 MD |
972 | :: CONSUME_TOKENS(proc_urcu_writer, |
973 | WRITE_PROD_NONE, | |
caeea74c MD |
974 | WRITE_DATA) -> |
975 | ooo_mem(i); | |
976 | cur_data = (cur_data + 1) % SLAB_SIZE; | |
977 | WRITE_CACHED_VAR(rcu_data[cur_data], WINE); | |
978 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_DATA); | |
979 | ||
980 | ||
981 | :: CONSUME_TOKENS(proc_urcu_writer, | |
982 | WRITE_DATA, | |
983 | WRITE_PROC_WMB) -> | |
f24274b9 | 984 | smp_wmb(i); |
caeea74c MD |
985 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_WMB); |
986 | ||
987 | :: CONSUME_TOKENS(proc_urcu_writer, | |
988 | WRITE_PROC_WMB, | |
989 | WRITE_XCHG_PTR) -> | |
990 | /* rcu_xchg_pointer() */ | |
991 | atomic { | |
992 | old_data = READ_CACHED_VAR(rcu_ptr); | |
993 | WRITE_CACHED_VAR(rcu_ptr, cur_data); | |
994 | } | |
995 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_XCHG_PTR); | |
996 | ||
997 | :: CONSUME_TOKENS(proc_urcu_writer, | |
998 | WRITE_DATA | WRITE_PROC_WMB | WRITE_XCHG_PTR, | |
551ac1a3 | 999 | WRITE_PROC_FIRST_MB) -> |
f089ec24 MD |
1000 | goto smp_mb_send1; |
1001 | smp_mb_send1_end: | |
551ac1a3 MD |
1002 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_MB); |
1003 | ||
1004 | /* first flip */ | |
1005 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1006 | WRITE_PROC_FIRST_MB, | |
1007 | WRITE_PROC_FIRST_READ_GP) -> | |
1008 | tmpa = READ_CACHED_VAR(urcu_gp_ctr); | |
1009 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_READ_GP); | |
1010 | :: CONSUME_TOKENS(proc_urcu_writer, | |
caeea74c MD |
1011 | WRITE_PROC_FIRST_MB | WRITE_PROC_WMB |
1012 | | WRITE_PROC_FIRST_READ_GP, | |
551ac1a3 MD |
1013 | WRITE_PROC_FIRST_WRITE_GP) -> |
1014 | ooo_mem(i); | |
1015 | WRITE_CACHED_VAR(urcu_gp_ctr, tmpa ^ RCU_GP_CTR_BIT); | |
1016 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_WRITE_GP); | |
1017 | ||
1018 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1019 | //WRITE_PROC_FIRST_WRITE_GP /* TEST ADDING SYNC CORE */ | |
1020 | WRITE_PROC_FIRST_MB, /* can be reordered before/after flips */ | |
1021 | WRITE_PROC_FIRST_WAIT | WRITE_PROC_FIRST_WAIT_LOOP) -> | |
1022 | ooo_mem(i); | |
1023 | /* ONLY WAITING FOR READER 0 */ | |
1024 | tmp2 = READ_CACHED_VAR(urcu_active_readers[0]); | |
794a737e MD |
1025 | #ifndef SINGLE_FLIP |
1026 | /* In normal execution, we are always starting by | |
1027 | * waiting for the even parity. | |
1028 | */ | |
1029 | cur_gp_val = RCU_GP_CTR_BIT; | |
1030 | #endif | |
551ac1a3 MD |
1031 | if |
1032 | :: (tmp2 & RCU_GP_CTR_NEST_MASK) | |
f089ec24 | 1033 | && ((tmp2 ^ cur_gp_val) & RCU_GP_CTR_BIT) -> |
551ac1a3 MD |
1034 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_WAIT_LOOP); |
1035 | :: else -> | |
1036 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_WAIT); | |
1037 | fi; | |
1038 | ||
1039 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1040 | //WRITE_PROC_FIRST_WRITE_GP /* TEST ADDING SYNC CORE */ | |
1041 | WRITE_PROC_FIRST_WRITE_GP | |
1042 | | WRITE_PROC_FIRST_READ_GP | |
1043 | | WRITE_PROC_FIRST_WAIT_LOOP | |
caeea74c | 1044 | | WRITE_DATA | WRITE_PROC_WMB | WRITE_XCHG_PTR |
551ac1a3 MD |
1045 | | WRITE_PROC_FIRST_MB, /* can be reordered before/after flips */ |
1046 | 0) -> | |
1047 | #ifndef GEN_ERROR_WRITER_PROGRESS | |
f089ec24 MD |
1048 | goto smp_mb_send2; |
1049 | smp_mb_send2_end: | |
551ac1a3 MD |
1050 | #else |
1051 | ooo_mem(i); | |
1052 | #endif | |
1053 | /* This instruction loops to WRITE_PROC_FIRST_WAIT */ | |
1054 | CLEAR_TOKENS(proc_urcu_writer, WRITE_PROC_FIRST_WAIT_LOOP | WRITE_PROC_FIRST_WAIT); | |
1055 | ||
1056 | /* second flip */ | |
1057 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1058 | WRITE_PROC_FIRST_WAIT /* Control dependency : need to branch out of | |
1059 | * the loop to execute the next flip (CHECK) */ | |
1060 | | WRITE_PROC_FIRST_WRITE_GP | |
1061 | | WRITE_PROC_FIRST_READ_GP | |
1062 | | WRITE_PROC_FIRST_MB, | |
1063 | WRITE_PROC_SECOND_READ_GP) -> | |
551ac1a3 MD |
1064 | ooo_mem(i); |
1065 | tmpa = READ_CACHED_VAR(urcu_gp_ctr); | |
1066 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_READ_GP); | |
1067 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1068 | WRITE_PROC_FIRST_MB | |
caeea74c | 1069 | | WRITE_PROC_WMB |
551ac1a3 MD |
1070 | | WRITE_PROC_FIRST_READ_GP |
1071 | | WRITE_PROC_FIRST_WRITE_GP | |
1072 | | WRITE_PROC_SECOND_READ_GP, | |
1073 | WRITE_PROC_SECOND_WRITE_GP) -> | |
1074 | ooo_mem(i); | |
1075 | WRITE_CACHED_VAR(urcu_gp_ctr, tmpa ^ RCU_GP_CTR_BIT); | |
1076 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WRITE_GP); | |
1077 | ||
1078 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1079 | //WRITE_PROC_FIRST_WRITE_GP /* TEST ADDING SYNC CORE */ | |
1080 | WRITE_PROC_FIRST_WAIT | |
1081 | | WRITE_PROC_FIRST_MB, /* can be reordered before/after flips */ | |
1082 | WRITE_PROC_SECOND_WAIT | WRITE_PROC_SECOND_WAIT_LOOP) -> | |
1083 | ooo_mem(i); | |
1084 | /* ONLY WAITING FOR READER 0 */ | |
1085 | tmp2 = READ_CACHED_VAR(urcu_active_readers[0]); | |
1086 | if | |
1087 | :: (tmp2 & RCU_GP_CTR_NEST_MASK) | |
794a737e | 1088 | && ((tmp2 ^ 0) & RCU_GP_CTR_BIT) -> |
551ac1a3 MD |
1089 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WAIT_LOOP); |
1090 | :: else -> | |
1091 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WAIT); | |
1092 | fi; | |
1093 | ||
1094 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1095 | //WRITE_PROC_FIRST_WRITE_GP /* TEST ADDING SYNC CORE */ | |
1096 | WRITE_PROC_SECOND_WRITE_GP | |
1097 | | WRITE_PROC_FIRST_WRITE_GP | |
1098 | | WRITE_PROC_SECOND_READ_GP | |
1099 | | WRITE_PROC_FIRST_READ_GP | |
1100 | | WRITE_PROC_SECOND_WAIT_LOOP | |
caeea74c | 1101 | | WRITE_DATA | WRITE_PROC_WMB | WRITE_XCHG_PTR |
551ac1a3 MD |
1102 | | WRITE_PROC_FIRST_MB, /* can be reordered before/after flips */ |
1103 | 0) -> | |
1104 | #ifndef GEN_ERROR_WRITER_PROGRESS | |
f089ec24 MD |
1105 | goto smp_mb_send3; |
1106 | smp_mb_send3_end: | |
551ac1a3 MD |
1107 | #else |
1108 | ooo_mem(i); | |
1109 | #endif | |
1110 | /* This instruction loops to WRITE_PROC_SECOND_WAIT */ | |
1111 | CLEAR_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_WAIT_LOOP | WRITE_PROC_SECOND_WAIT); | |
1112 | ||
1113 | ||
1114 | :: CONSUME_TOKENS(proc_urcu_writer, | |
1115 | WRITE_PROC_FIRST_WAIT | |
1116 | | WRITE_PROC_SECOND_WAIT | |
1117 | | WRITE_PROC_FIRST_READ_GP | |
1118 | | WRITE_PROC_SECOND_READ_GP | |
1119 | | WRITE_PROC_FIRST_WRITE_GP | |
1120 | | WRITE_PROC_SECOND_WRITE_GP | |
caeea74c | 1121 | | WRITE_DATA | WRITE_PROC_WMB | WRITE_XCHG_PTR |
551ac1a3 MD |
1122 | | WRITE_PROC_FIRST_MB, |
1123 | WRITE_PROC_SECOND_MB) -> | |
f089ec24 MD |
1124 | goto smp_mb_send4; |
1125 | smp_mb_send4_end: | |
551ac1a3 MD |
1126 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_PROC_SECOND_MB); |
1127 | ||
caeea74c MD |
1128 | :: CONSUME_TOKENS(proc_urcu_writer, |
1129 | WRITE_XCHG_PTR | |
1130 | | WRITE_PROC_FIRST_WAIT | |
1131 | | WRITE_PROC_SECOND_WAIT | |
1132 | | WRITE_PROC_WMB /* No dependency on | |
1133 | * WRITE_DATA because we | |
1134 | * write to a | |
1135 | * different location. */ | |
1136 | | WRITE_PROC_SECOND_MB | |
1137 | | WRITE_PROC_FIRST_MB, | |
1138 | WRITE_FREE) -> | |
1139 | WRITE_CACHED_VAR(rcu_data[old_data], POISON); | |
1140 | PRODUCE_TOKENS(proc_urcu_writer, WRITE_FREE); | |
1141 | ||
551ac1a3 MD |
1142 | :: CONSUME_TOKENS(proc_urcu_writer, WRITE_PROC_ALL_TOKENS, 0) -> |
1143 | CLEAR_TOKENS(proc_urcu_writer, WRITE_PROC_ALL_TOKENS_CLEAR); | |
1144 | break; | |
f089ec24 MD |
1145 | fi; |
1146 | } | |
551ac1a3 | 1147 | od; |
caeea74c MD |
1148 | /* |
1149 | * Note : Promela model adds implicit serialization of the | |
1150 | * WRITE_FREE instruction. Normally, it would be permitted to | |
1151 | * spill on the next loop execution. Given the validation we do | |
1152 | * checks for the data entry read to be poisoned, it's ok if | |
1153 | * we do not check "late arriving" memory poisoning. | |
1154 | */ | |
551ac1a3 MD |
1155 | :: else -> break; |
1156 | od; | |
1157 | /* | |
1158 | * Given the reader loops infinitely, let the writer also busy-loop | |
1159 | * with progress here so, with weak fairness, we can test the | |
1160 | * writer's progress. | |
1161 | */ | |
1162 | end_writer: | |
1163 | do | |
1164 | :: 1 -> | |
1165 | #ifdef WRITER_PROGRESS | |
1166 | progress_writer2: | |
6b0de963 MD |
1167 | #endif |
1168 | #ifdef READER_PROGRESS | |
1169 | /* | |
1170 | * Make sure we don't block the reader's progress. | |
1171 | */ | |
1172 | smp_mb_send(i, j, 5); | |
551ac1a3 MD |
1173 | #endif |
1174 | skip; | |
1175 | od; | |
f089ec24 MD |
1176 | |
1177 | /* Non-atomic parts of the loop */ | |
1178 | goto end; | |
1179 | smp_mb_send1: | |
30193782 | 1180 | smp_mb_send(i, j, 1); |
f089ec24 MD |
1181 | goto smp_mb_send1_end; |
1182 | #ifndef GEN_ERROR_WRITER_PROGRESS | |
1183 | smp_mb_send2: | |
30193782 | 1184 | smp_mb_send(i, j, 2); |
f089ec24 MD |
1185 | goto smp_mb_send2_end; |
1186 | smp_mb_send3: | |
30193782 | 1187 | smp_mb_send(i, j, 3); |
f089ec24 MD |
1188 | goto smp_mb_send3_end; |
1189 | #endif | |
1190 | smp_mb_send4: | |
30193782 | 1191 | smp_mb_send(i, j, 4); |
f089ec24 MD |
1192 | goto smp_mb_send4_end; |
1193 | end: | |
1194 | skip; | |
551ac1a3 MD |
1195 | } |
1196 | ||
1197 | /* no name clash please */ | |
1198 | #undef proc_urcu_writer | |
1199 | ||
1200 | ||
1201 | /* Leave after the readers and writers so the pid count is ok. */ | |
1202 | init { | |
1203 | byte i, j; | |
1204 | ||
1205 | atomic { | |
1206 | INIT_CACHED_VAR(urcu_gp_ctr, 1, j); | |
6af482a9 | 1207 | INIT_CACHED_VAR(rcu_ptr, 0, j); |
551ac1a3 MD |
1208 | |
1209 | i = 0; | |
1210 | do | |
1211 | :: i < NR_READERS -> | |
1212 | INIT_CACHED_VAR(urcu_active_readers[i], 0, j); | |
caeea74c MD |
1213 | ptr_read_first[i] = 1; |
1214 | ptr_read_second[i] = 1; | |
1215 | data_read_first[i] = WINE; | |
1216 | data_read_second[i] = WINE; | |
551ac1a3 MD |
1217 | i++; |
1218 | :: i >= NR_READERS -> break | |
1219 | od; | |
6af482a9 MD |
1220 | INIT_CACHED_VAR(rcu_data[0], WINE, j); |
1221 | i = 1; | |
1222 | do | |
1223 | :: i < SLAB_SIZE -> | |
1224 | INIT_CACHED_VAR(rcu_data[i], POISON, j); | |
1225 | i++ | |
1226 | :: i >= SLAB_SIZE -> break | |
1227 | od; | |
1228 | ||
551ac1a3 MD |
1229 | init_done = 1; |
1230 | } | |
1231 | } |