Commit | Line | Data |
---|---|---|
953192ba MD |
1 | /* |
2 | * filter-visitor-generate-bytecode.c | |
3 | * | |
4 | * LTTng filter bytecode generation | |
5 | * | |
6 | * Copyright 2012 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
7 | * | |
8 | * This library is free software; you can redistribute it and/or modify it | |
9 | * under the terms of the GNU Lesser General Public License, version 2.1 only, | |
10 | * as published by the Free Software Foundation. | |
11 | * | |
12 | * This library is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | * Lesser General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU Lesser General Public License | |
18 | * along with this library; if not, write to the Free Software Foundation, | |
19 | * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
20 | */ | |
21 | ||
22 | #include <stdlib.h> | |
23 | #include <string.h> | |
24 | #include <errno.h> | |
25 | #include "align.h" | |
26 | #include "filter-bytecode.h" | |
27 | #include "filter-ir.h" | |
28 | #include "filter-ast.h" | |
29 | ||
a187da1a DG |
30 | #include <common/macros.h> |
31 | ||
953192ba MD |
32 | #ifndef max_t |
33 | #define max_t(type, a, b) ((type) ((a) > (b) ? (a) : (b))) | |
34 | #endif | |
35 | ||
36 | //#define INIT_ALLOC_SIZE PAGE_SIZE | |
37 | #define INIT_ALLOC_SIZE 4 | |
38 | ||
39 | static | |
40 | int recursive_visit_gen_bytecode(struct filter_parser_ctx *ctx, | |
41 | struct ir_op *node); | |
42 | ||
01a204f0 CB |
43 | static inline int fls(unsigned int x) |
44 | { | |
45 | int r = 32; | |
46 | ||
47 | if (!x) | |
48 | return 0; | |
49 | if (!(x & 0xFFFF0000U)) { | |
50 | x <<= 16; | |
51 | r -= 16; | |
52 | } | |
53 | if (!(x & 0xFF000000U)) { | |
54 | x <<= 8; | |
55 | r -= 8; | |
56 | } | |
57 | if (!(x & 0xF0000000U)) { | |
58 | x <<= 4; | |
59 | r -= 4; | |
60 | } | |
61 | if (!(x & 0xC0000000U)) { | |
62 | x <<= 2; | |
63 | r -= 2; | |
64 | } | |
65 | if (!(x & 0x80000000U)) { | |
66 | x <<= 1; | |
67 | r -= 1; | |
68 | } | |
69 | return r; | |
70 | } | |
71 | ||
72 | static inline int get_count_order(unsigned int count) | |
73 | { | |
74 | int order; | |
75 | ||
76 | order = fls(count) - 1; | |
77 | if (count & (count - 1)) | |
78 | order++; | |
79 | return order; | |
80 | } | |
81 | ||
953192ba | 82 | static |
53a80697 | 83 | int bytecode_init(struct lttng_filter_bytecode_alloc **fb) |
953192ba | 84 | { |
1029587a MD |
85 | uint32_t alloc_len; |
86 | ||
87 | alloc_len = sizeof(struct lttng_filter_bytecode_alloc) + INIT_ALLOC_SIZE; | |
88 | *fb = calloc(alloc_len, 1); | |
953192ba MD |
89 | if (!*fb) { |
90 | return -ENOMEM; | |
91 | } else { | |
1029587a | 92 | (*fb)->alloc_len = alloc_len; |
953192ba MD |
93 | return 0; |
94 | } | |
95 | } | |
96 | ||
97 | static | |
53a80697 | 98 | int32_t bytecode_reserve(struct lttng_filter_bytecode_alloc **fb, uint32_t align, uint32_t len) |
953192ba MD |
99 | { |
100 | int32_t ret; | |
101 | uint32_t padding = offset_align((*fb)->b.len, align); | |
ec96a8f6 | 102 | uint32_t new_len = (*fb)->b.len + padding + len; |
1029587a | 103 | uint32_t new_alloc_len = sizeof(struct lttng_filter_bytecode_alloc) + new_len; |
ec96a8f6 | 104 | uint32_t old_alloc_len = (*fb)->alloc_len; |
953192ba | 105 | |
ec96a8f6 | 106 | if (new_len > LTTNG_FILTER_MAX_LEN) |
5ddb0a08 CB |
107 | return -EINVAL; |
108 | ||
ec96a8f6 | 109 | if (new_alloc_len > old_alloc_len) { |
d0b96690 DG |
110 | struct lttng_filter_bytecode_alloc *newptr; |
111 | ||
ec96a8f6 MD |
112 | new_alloc_len = |
113 | max_t(uint32_t, 1U << get_count_order(new_alloc_len), old_alloc_len << 1); | |
d0b96690 DG |
114 | newptr = realloc(*fb, new_alloc_len); |
115 | if (!newptr) | |
953192ba | 116 | return -ENOMEM; |
d0b96690 | 117 | *fb = newptr; |
1029587a | 118 | /* We zero directly the memory from start of allocation. */ |
ec96a8f6 MD |
119 | memset(&((char *) *fb)[old_alloc_len], 0, new_alloc_len - old_alloc_len); |
120 | (*fb)->alloc_len = new_alloc_len; | |
953192ba MD |
121 | } |
122 | (*fb)->b.len += padding; | |
123 | ret = (*fb)->b.len; | |
124 | (*fb)->b.len += len; | |
125 | return ret; | |
126 | } | |
127 | ||
128 | static | |
53a80697 | 129 | int bytecode_push(struct lttng_filter_bytecode_alloc **fb, const void *data, |
953192ba MD |
130 | uint32_t align, uint32_t len) |
131 | { | |
132 | int32_t offset; | |
133 | ||
134 | offset = bytecode_reserve(fb, align, len); | |
135 | if (offset < 0) | |
136 | return offset; | |
137 | memcpy(&(*fb)->b.data[offset], data, len); | |
138 | return 0; | |
139 | } | |
140 | ||
141 | static | |
53a80697 | 142 | int bytecode_push_logical(struct lttng_filter_bytecode_alloc **fb, |
953192ba MD |
143 | struct logical_op *data, |
144 | uint32_t align, uint32_t len, | |
145 | uint16_t *skip_offset) | |
146 | { | |
147 | int32_t offset; | |
148 | ||
149 | offset = bytecode_reserve(fb, align, len); | |
150 | if (offset < 0) | |
151 | return offset; | |
152 | memcpy(&(*fb)->b.data[offset], data, len); | |
153 | *skip_offset = | |
154 | (void *) &((struct logical_op *) &(*fb)->b.data[offset])->skip_offset | |
155 | - (void *) &(*fb)->b.data[0]; | |
156 | return 0; | |
157 | } | |
158 | ||
159 | static | |
53a80697 | 160 | int bytecode_patch(struct lttng_filter_bytecode_alloc **fb, |
953192ba MD |
161 | const void *data, |
162 | uint16_t offset, | |
163 | uint32_t len) | |
164 | { | |
165 | if (offset >= (*fb)->b.len) { | |
166 | return -EINVAL; | |
167 | } | |
168 | memcpy(&(*fb)->b.data[offset], data, len); | |
169 | return 0; | |
170 | } | |
171 | ||
172 | static | |
173 | int visit_node_root(struct filter_parser_ctx *ctx, struct ir_op *node) | |
174 | { | |
175 | int ret; | |
176 | struct return_op insn; | |
177 | ||
178 | /* Visit child */ | |
179 | ret = recursive_visit_gen_bytecode(ctx, node->u.root.child); | |
180 | if (ret) | |
181 | return ret; | |
182 | ||
183 | /* Generate end of bytecode instruction */ | |
184 | insn.op = FILTER_OP_RETURN; | |
185 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); | |
186 | } | |
187 | ||
953192ba MD |
188 | static |
189 | int visit_node_load(struct filter_parser_ctx *ctx, struct ir_op *node) | |
190 | { | |
191 | int ret; | |
192 | ||
193 | switch (node->data_type) { | |
194 | case IR_DATA_UNKNOWN: | |
195 | default: | |
196 | fprintf(stderr, "[error] Unknown data type in %s\n", | |
197 | __func__); | |
198 | return -EINVAL; | |
199 | ||
200 | case IR_DATA_STRING: | |
201 | { | |
202 | struct load_op *insn; | |
203 | uint32_t insn_len = sizeof(struct load_op) | |
204 | + strlen(node->u.load.u.string) + 1; | |
205 | ||
206 | insn = calloc(insn_len, 1); | |
207 | if (!insn) | |
208 | return -ENOMEM; | |
209 | insn->op = FILTER_OP_LOAD_STRING; | |
953192ba MD |
210 | strcpy(insn->data, node->u.load.u.string); |
211 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); | |
212 | free(insn); | |
213 | return ret; | |
214 | } | |
215 | case IR_DATA_NUMERIC: | |
216 | { | |
217 | struct load_op *insn; | |
218 | uint32_t insn_len = sizeof(struct load_op) | |
219 | + sizeof(struct literal_numeric); | |
220 | ||
221 | insn = calloc(insn_len, 1); | |
222 | if (!insn) | |
223 | return -ENOMEM; | |
224 | insn->op = FILTER_OP_LOAD_S64; | |
953192ba MD |
225 | *(int64_t *) insn->data = node->u.load.u.num; |
226 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); | |
227 | free(insn); | |
228 | return ret; | |
229 | } | |
e90d8561 MD |
230 | case IR_DATA_FLOAT: |
231 | { | |
232 | struct load_op *insn; | |
233 | uint32_t insn_len = sizeof(struct load_op) | |
234 | + sizeof(struct literal_double); | |
235 | ||
236 | insn = calloc(insn_len, 1); | |
237 | if (!insn) | |
238 | return -ENOMEM; | |
239 | insn->op = FILTER_OP_LOAD_DOUBLE; | |
e90d8561 MD |
240 | *(double *) insn->data = node->u.load.u.flt; |
241 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); | |
242 | free(insn); | |
243 | return ret; | |
244 | } | |
586dc72f MD |
245 | case IR_DATA_FIELD_REF: /* fall-through */ |
246 | case IR_DATA_GET_CONTEXT_REF: | |
953192ba MD |
247 | { |
248 | struct load_op *insn; | |
249 | uint32_t insn_len = sizeof(struct load_op) | |
250 | + sizeof(struct field_ref); | |
251 | struct field_ref ref_offset; | |
ec96a8f6 MD |
252 | uint32_t reloc_offset_u32; |
253 | uint16_t reloc_offset; | |
953192ba MD |
254 | |
255 | insn = calloc(insn_len, 1); | |
256 | if (!insn) | |
257 | return -ENOMEM; | |
586dc72f MD |
258 | switch(node->data_type) { |
259 | case IR_DATA_FIELD_REF: | |
260 | insn->op = FILTER_OP_LOAD_FIELD_REF; | |
261 | break; | |
262 | case IR_DATA_GET_CONTEXT_REF: | |
263 | insn->op = FILTER_OP_GET_CONTEXT_REF; | |
264 | break; | |
265 | default: | |
266 | return -EINVAL; | |
267 | } | |
953192ba MD |
268 | ref_offset.offset = (uint16_t) -1U; |
269 | memcpy(insn->data, &ref_offset, sizeof(ref_offset)); | |
65775683 | 270 | /* reloc_offset points to struct load_op */ |
ec96a8f6 MD |
271 | reloc_offset_u32 = bytecode_get_len(&ctx->bytecode->b); |
272 | if (reloc_offset_u32 > LTTNG_FILTER_MAX_LEN - 1) { | |
273 | free(insn); | |
274 | return -EINVAL; | |
275 | } | |
276 | reloc_offset = (uint16_t) reloc_offset_u32; | |
953192ba MD |
277 | ret = bytecode_push(&ctx->bytecode, insn, 1, insn_len); |
278 | if (ret) { | |
279 | free(insn); | |
280 | return ret; | |
281 | } | |
282 | /* append reloc */ | |
283 | ret = bytecode_push(&ctx->bytecode_reloc, &reloc_offset, | |
284 | 1, sizeof(reloc_offset)); | |
285 | if (ret) { | |
286 | free(insn); | |
287 | return ret; | |
288 | } | |
289 | ret = bytecode_push(&ctx->bytecode_reloc, node->u.load.u.ref, | |
290 | 1, strlen(node->u.load.u.ref) + 1); | |
291 | free(insn); | |
292 | return ret; | |
293 | } | |
294 | } | |
295 | } | |
296 | ||
297 | static | |
298 | int visit_node_unary(struct filter_parser_ctx *ctx, struct ir_op *node) | |
299 | { | |
300 | int ret; | |
301 | struct unary_op insn; | |
302 | ||
303 | /* Visit child */ | |
304 | ret = recursive_visit_gen_bytecode(ctx, node->u.unary.child); | |
305 | if (ret) | |
306 | return ret; | |
307 | ||
308 | /* Generate end of bytecode instruction */ | |
309 | switch (node->u.unary.type) { | |
310 | case AST_UNARY_UNKNOWN: | |
311 | default: | |
312 | fprintf(stderr, "[error] Unknown unary node type in %s\n", | |
313 | __func__); | |
314 | return -EINVAL; | |
315 | case AST_UNARY_PLUS: | |
316 | /* Nothing to do. */ | |
317 | return 0; | |
318 | case AST_UNARY_MINUS: | |
319 | insn.op = FILTER_OP_UNARY_MINUS; | |
953192ba MD |
320 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
321 | case AST_UNARY_NOT: | |
322 | insn.op = FILTER_OP_UNARY_NOT; | |
953192ba MD |
323 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); |
324 | } | |
325 | } | |
326 | ||
327 | /* | |
328 | * Binary comparator nesting is disallowed. This allows fitting into | |
329 | * only 2 registers. | |
330 | */ | |
331 | static | |
332 | int visit_node_binary(struct filter_parser_ctx *ctx, struct ir_op *node) | |
333 | { | |
334 | int ret; | |
335 | struct binary_op insn; | |
336 | ||
337 | /* Visit child */ | |
338 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.left); | |
339 | if (ret) | |
340 | return ret; | |
341 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.right); | |
342 | if (ret) | |
343 | return ret; | |
344 | ||
345 | switch (node->u.binary.type) { | |
346 | case AST_OP_UNKNOWN: | |
347 | default: | |
348 | fprintf(stderr, "[error] Unknown unary node type in %s\n", | |
349 | __func__); | |
350 | return -EINVAL; | |
351 | ||
352 | case AST_OP_AND: | |
353 | case AST_OP_OR: | |
354 | fprintf(stderr, "[error] Unexpected logical node type in %s\n", | |
355 | __func__); | |
356 | return -EINVAL; | |
357 | ||
358 | case AST_OP_MUL: | |
359 | insn.op = FILTER_OP_MUL; | |
360 | break; | |
361 | case AST_OP_DIV: | |
362 | insn.op = FILTER_OP_DIV; | |
363 | break; | |
364 | case AST_OP_MOD: | |
365 | insn.op = FILTER_OP_MOD; | |
366 | break; | |
367 | case AST_OP_PLUS: | |
368 | insn.op = FILTER_OP_PLUS; | |
369 | break; | |
370 | case AST_OP_MINUS: | |
371 | insn.op = FILTER_OP_MINUS; | |
372 | break; | |
373 | case AST_OP_RSHIFT: | |
374 | insn.op = FILTER_OP_RSHIFT; | |
375 | break; | |
376 | case AST_OP_LSHIFT: | |
377 | insn.op = FILTER_OP_LSHIFT; | |
378 | break; | |
379 | case AST_OP_BIN_AND: | |
380 | insn.op = FILTER_OP_BIN_AND; | |
381 | break; | |
382 | case AST_OP_BIN_OR: | |
383 | insn.op = FILTER_OP_BIN_OR; | |
384 | break; | |
385 | case AST_OP_BIN_XOR: | |
386 | insn.op = FILTER_OP_BIN_XOR; | |
387 | break; | |
388 | ||
389 | case AST_OP_EQ: | |
390 | insn.op = FILTER_OP_EQ; | |
391 | break; | |
392 | case AST_OP_NE: | |
393 | insn.op = FILTER_OP_NE; | |
394 | break; | |
395 | case AST_OP_GT: | |
396 | insn.op = FILTER_OP_GT; | |
397 | break; | |
398 | case AST_OP_LT: | |
399 | insn.op = FILTER_OP_LT; | |
400 | break; | |
401 | case AST_OP_GE: | |
402 | insn.op = FILTER_OP_GE; | |
403 | break; | |
404 | case AST_OP_LE: | |
405 | insn.op = FILTER_OP_LE; | |
406 | break; | |
407 | } | |
408 | return bytecode_push(&ctx->bytecode, &insn, 1, sizeof(insn)); | |
409 | } | |
410 | ||
8cf9540a MD |
411 | /* |
412 | * A logical op always return a s64 (1 or 0). | |
413 | */ | |
953192ba MD |
414 | static |
415 | int visit_node_logical(struct filter_parser_ctx *ctx, struct ir_op *node) | |
416 | { | |
417 | int ret; | |
418 | struct logical_op insn; | |
419 | uint16_t skip_offset_loc; | |
420 | uint16_t target_loc; | |
421 | ||
422 | /* Visit left child */ | |
423 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.left); | |
424 | if (ret) | |
425 | return ret; | |
8cf9540a | 426 | /* Cast to s64 if float or field ref */ |
586dc72f MD |
427 | if ((node->u.binary.left->data_type == IR_DATA_FIELD_REF |
428 | || node->u.binary.left->data_type == IR_DATA_GET_CONTEXT_REF) | |
8cf9540a MD |
429 | || node->u.binary.left->data_type == IR_DATA_FLOAT) { |
430 | struct cast_op cast_insn; | |
431 | ||
586dc72f MD |
432 | if (node->u.binary.left->data_type == IR_DATA_FIELD_REF |
433 | || node->u.binary.left->data_type == IR_DATA_GET_CONTEXT_REF) { | |
29fefef8 MD |
434 | cast_insn.op = FILTER_OP_CAST_TO_S64; |
435 | } else { | |
436 | cast_insn.op = FILTER_OP_CAST_DOUBLE_TO_S64; | |
437 | } | |
8cf9540a MD |
438 | ret = bytecode_push(&ctx->bytecode, &cast_insn, |
439 | 1, sizeof(cast_insn)); | |
440 | if (ret) | |
441 | return ret; | |
442 | } | |
953192ba MD |
443 | switch (node->u.logical.type) { |
444 | default: | |
445 | fprintf(stderr, "[error] Unknown node type in %s\n", | |
446 | __func__); | |
447 | return -EINVAL; | |
448 | ||
449 | case AST_OP_AND: | |
450 | insn.op = FILTER_OP_AND; | |
451 | break; | |
452 | case AST_OP_OR: | |
453 | insn.op = FILTER_OP_OR; | |
454 | break; | |
455 | } | |
456 | insn.skip_offset = (uint16_t) -1UL; /* Temporary */ | |
457 | ret = bytecode_push_logical(&ctx->bytecode, &insn, 1, sizeof(insn), | |
458 | &skip_offset_loc); | |
459 | if (ret) | |
460 | return ret; | |
461 | /* Visit right child */ | |
462 | ret = recursive_visit_gen_bytecode(ctx, node->u.binary.right); | |
463 | if (ret) | |
464 | return ret; | |
8cf9540a | 465 | /* Cast to s64 if float or field ref */ |
586dc72f MD |
466 | if ((node->u.binary.right->data_type == IR_DATA_FIELD_REF |
467 | || node->u.binary.right->data_type == IR_DATA_GET_CONTEXT_REF) | |
8cf9540a MD |
468 | || node->u.binary.right->data_type == IR_DATA_FLOAT) { |
469 | struct cast_op cast_insn; | |
470 | ||
586dc72f MD |
471 | if (node->u.binary.right->data_type == IR_DATA_FIELD_REF |
472 | || node->u.binary.right->data_type == IR_DATA_GET_CONTEXT_REF) { | |
29fefef8 MD |
473 | cast_insn.op = FILTER_OP_CAST_TO_S64; |
474 | } else { | |
475 | cast_insn.op = FILTER_OP_CAST_DOUBLE_TO_S64; | |
476 | } | |
8cf9540a MD |
477 | ret = bytecode_push(&ctx->bytecode, &cast_insn, |
478 | 1, sizeof(cast_insn)); | |
479 | if (ret) | |
480 | return ret; | |
481 | } | |
953192ba MD |
482 | /* We now know where the logical op can skip. */ |
483 | target_loc = (uint16_t) bytecode_get_len(&ctx->bytecode->b); | |
484 | ret = bytecode_patch(&ctx->bytecode, | |
485 | &target_loc, /* Offset to jump to */ | |
486 | skip_offset_loc, /* Where to patch */ | |
487 | sizeof(uint16_t)); | |
488 | return ret; | |
489 | } | |
490 | ||
491 | /* | |
492 | * Postorder traversal of the tree. We need the children result before | |
493 | * we can evaluate the parent. | |
494 | */ | |
495 | static | |
496 | int recursive_visit_gen_bytecode(struct filter_parser_ctx *ctx, | |
497 | struct ir_op *node) | |
498 | { | |
499 | switch (node->op) { | |
500 | case IR_OP_UNKNOWN: | |
501 | default: | |
502 | fprintf(stderr, "[error] Unknown node type in %s\n", | |
503 | __func__); | |
504 | return -EINVAL; | |
505 | ||
506 | case IR_OP_ROOT: | |
507 | return visit_node_root(ctx, node); | |
508 | case IR_OP_LOAD: | |
509 | return visit_node_load(ctx, node); | |
510 | case IR_OP_UNARY: | |
511 | return visit_node_unary(ctx, node); | |
512 | case IR_OP_BINARY: | |
513 | return visit_node_binary(ctx, node); | |
514 | case IR_OP_LOGICAL: | |
515 | return visit_node_logical(ctx, node); | |
516 | } | |
517 | } | |
518 | ||
a187da1a | 519 | LTTNG_HIDDEN |
953192ba MD |
520 | void filter_bytecode_free(struct filter_parser_ctx *ctx) |
521 | { | |
522 | free(ctx->bytecode); | |
523 | ctx->bytecode = NULL; | |
524 | free(ctx->bytecode_reloc); | |
525 | ctx->bytecode_reloc = NULL; | |
526 | } | |
527 | ||
a187da1a | 528 | LTTNG_HIDDEN |
953192ba MD |
529 | int filter_visitor_bytecode_generate(struct filter_parser_ctx *ctx) |
530 | { | |
531 | int ret; | |
532 | ||
533 | ret = bytecode_init(&ctx->bytecode); | |
534 | if (ret) | |
535 | return ret; | |
536 | ret = bytecode_init(&ctx->bytecode_reloc); | |
537 | if (ret) | |
538 | goto error; | |
539 | ret = recursive_visit_gen_bytecode(ctx, ctx->ir_root); | |
540 | if (ret) | |
541 | goto error; | |
542 | ||
543 | /* Finally, append symbol table to bytecode */ | |
544 | ctx->bytecode->b.reloc_table_offset = bytecode_get_len(&ctx->bytecode->b); | |
545 | return bytecode_push(&ctx->bytecode, ctx->bytecode_reloc->b.data, | |
546 | 1, bytecode_get_len(&ctx->bytecode_reloc->b)); | |
547 | ||
548 | error: | |
549 | filter_bytecode_free(ctx); | |
550 | return ret; | |
551 | } |