1 | /* The analysis "engine". |
2 | Copyright (C) 2019-2024 Free Software Foundation, Inc. |
3 | Contributed by David Malcolm <dmalcolm@redhat.com>. |
4 | |
5 | This file is part of GCC. |
6 | |
7 | GCC is free software; you can redistribute it and/or modify it |
8 | under the terms of the GNU General Public License as published by |
9 | the Free Software Foundation; either version 3, or (at your option) |
10 | any later version. |
11 | |
12 | GCC is distributed in the hope that it will be useful, but |
13 | WITHOUT ANY WARRANTY; without even the implied warranty of |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
15 | General Public License for more details. |
16 | |
17 | You should have received a copy of the GNU General Public License |
18 | along with GCC; see the file COPYING3. If not see |
19 | <http://www.gnu.org/licenses/>. */ |
20 | |
21 | #include "config.h" |
22 | #define INCLUDE_MEMORY |
23 | #include "system.h" |
24 | #include "coretypes.h" |
25 | #include "make-unique.h" |
26 | #include "tree.h" |
27 | #include "fold-const.h" |
28 | #include "gcc-rich-location.h" |
29 | #include "diagnostic-core.h" |
30 | #include "diagnostic-event-id.h" |
31 | #include "diagnostic-path.h" |
32 | #include "function.h" |
33 | #include "pretty-print.h" |
34 | #include "sbitmap.h" |
35 | #include "bitmap.h" |
36 | #include "ordered-hash-map.h" |
37 | #include "analyzer/analyzer.h" |
38 | #include "analyzer/analyzer-logging.h" |
39 | #include "analyzer/call-string.h" |
40 | #include "analyzer/program-point.h" |
41 | #include "analyzer/store.h" |
42 | #include "analyzer/region-model.h" |
43 | #include "analyzer/constraint-manager.h" |
44 | #include "analyzer/sm.h" |
45 | #include "analyzer/pending-diagnostic.h" |
46 | #include "analyzer/diagnostic-manager.h" |
47 | #include "cfg.h" |
48 | #include "basic-block.h" |
49 | #include "gimple.h" |
50 | #include "gimple-iterator.h" |
51 | #include "gimple-pretty-print.h" |
52 | #include "cgraph.h" |
53 | #include "digraph.h" |
54 | #include "analyzer/supergraph.h" |
55 | #include "analyzer/program-state.h" |
56 | #include "analyzer/exploded-graph.h" |
57 | #include "analyzer/analysis-plan.h" |
58 | #include "analyzer/checker-path.h" |
59 | #include "analyzer/state-purge.h" |
60 | #include "analyzer/bar-chart.h" |
61 | #include "analyzer/call-info.h" |
62 | #include <zlib.h> |
63 | #include "plugin.h" |
64 | #include "target.h" |
65 | #include <memory> |
66 | #include "stringpool.h" |
67 | #include "attribs.h" |
68 | #include "tree-dfa.h" |
69 | #include "analyzer/known-function-manager.h" |
70 | #include "analyzer/call-summary.h" |
71 | |
72 | /* For an overview, see gcc/doc/analyzer.texi. */ |
73 | |
74 | #if ENABLE_ANALYZER |
75 | |
76 | namespace ana { |
77 | |
78 | /* class impl_region_model_context : public region_model_context. */ |
79 | |
80 | impl_region_model_context:: |
81 | impl_region_model_context (exploded_graph &eg, |
82 | exploded_node *enode_for_diag, |
83 | const program_state *old_state, |
84 | program_state *new_state, |
85 | uncertainty_t *uncertainty, |
86 | path_context *path_ctxt, |
87 | const gimple *stmt, |
88 | stmt_finder *stmt_finder, |
89 | bool *out_could_have_done_work) |
90 | : m_eg (&eg), m_logger (eg.get_logger ()), |
91 | m_enode_for_diag (enode_for_diag), |
92 | m_old_state (old_state), |
93 | m_new_state (new_state), |
94 | m_stmt (stmt), |
95 | m_stmt_finder (stmt_finder), |
96 | m_ext_state (eg.get_ext_state ()), |
97 | m_uncertainty (uncertainty), |
98 | m_path_ctxt (path_ctxt), |
99 | m_out_could_have_done_work (out_could_have_done_work) |
100 | { |
101 | } |
102 | |
103 | impl_region_model_context:: |
104 | impl_region_model_context (program_state *state, |
105 | const extrinsic_state &ext_state, |
106 | uncertainty_t *uncertainty, |
107 | logger *logger) |
108 | : m_eg (NULL), m_logger (logger), m_enode_for_diag (NULL), |
109 | m_old_state (NULL), |
110 | m_new_state (state), |
111 | m_stmt (NULL), |
112 | m_stmt_finder (NULL), |
113 | m_ext_state (ext_state), |
114 | m_uncertainty (uncertainty), |
115 | m_path_ctxt (NULL), |
116 | m_out_could_have_done_work (nullptr) |
117 | { |
118 | } |
119 | |
120 | bool |
121 | impl_region_model_context::warn (std::unique_ptr<pending_diagnostic> d, |
122 | const stmt_finder *custom_finder) |
123 | { |
124 | LOG_FUNC (get_logger ()); |
125 | auto curr_stmt_finder = custom_finder ? custom_finder : m_stmt_finder; |
126 | if (m_stmt == NULL && curr_stmt_finder == NULL) |
127 | { |
128 | if (get_logger ()) |
129 | get_logger ()->log (fmt: "rejecting diagnostic: no stmt" ); |
130 | return false; |
131 | } |
132 | if (m_eg) |
133 | { |
134 | bool terminate_path = d->terminate_path_p (); |
135 | pending_location ploc (m_enode_for_diag, |
136 | m_enode_for_diag->get_supernode (), |
137 | m_stmt, |
138 | curr_stmt_finder); |
139 | if (m_eg->get_diagnostic_manager ().add_diagnostic (ploc, |
140 | d: std::move (d))) |
141 | { |
142 | if (m_path_ctxt |
143 | && terminate_path |
144 | && flag_analyzer_suppress_followups) |
145 | m_path_ctxt->terminate_path (); |
146 | return true; |
147 | } |
148 | } |
149 | return false; |
150 | } |
151 | |
152 | void |
153 | impl_region_model_context::add_note (std::unique_ptr<pending_note> pn) |
154 | { |
155 | LOG_FUNC (get_logger ()); |
156 | if (m_eg) |
157 | m_eg->get_diagnostic_manager ().add_note (pn: std::move (pn)); |
158 | } |
159 | |
160 | void |
161 | impl_region_model_context::add_event (std::unique_ptr<checker_event> event) |
162 | { |
163 | LOG_FUNC (get_logger ()); |
164 | if (m_eg) |
165 | m_eg->get_diagnostic_manager ().add_event (event: std::move (event)); |
166 | } |
167 | |
168 | void |
169 | impl_region_model_context::on_svalue_leak (const svalue *sval) |
170 | |
171 | { |
172 | for (sm_state_map *smap : m_new_state->m_checker_states) |
173 | smap->on_svalue_leak (sval, ctxt: this); |
174 | } |
175 | |
176 | void |
177 | impl_region_model_context:: |
178 | on_liveness_change (const svalue_set &live_svalues, |
179 | const region_model *model) |
180 | { |
181 | for (sm_state_map *smap : m_new_state->m_checker_states) |
182 | smap->on_liveness_change (live_svalues, model, ext_state: m_ext_state, ctxt: this); |
183 | } |
184 | |
185 | void |
186 | impl_region_model_context::on_unknown_change (const svalue *sval, |
187 | bool is_mutable) |
188 | { |
189 | if (!sval->can_have_associated_state_p ()) |
190 | return; |
191 | for (sm_state_map *smap : m_new_state->m_checker_states) |
192 | smap->on_unknown_change (sval, is_mutable, ext_state: m_ext_state); |
193 | } |
194 | |
195 | void |
196 | impl_region_model_context::on_escaped_function (tree fndecl) |
197 | { |
198 | m_eg->on_escaped_function (fndecl); |
199 | } |
200 | |
201 | uncertainty_t * |
202 | impl_region_model_context::get_uncertainty () |
203 | { |
204 | return m_uncertainty; |
205 | } |
206 | |
207 | /* Purge state involving SVAL. The region_model has already been purged, |
208 | so we only need to purge other state in the program_state: |
209 | the sm-state. */ |
210 | |
211 | void |
212 | impl_region_model_context::purge_state_involving (const svalue *sval) |
213 | { |
214 | int i; |
215 | sm_state_map *smap; |
216 | FOR_EACH_VEC_ELT (m_new_state->m_checker_states, i, smap) |
217 | smap->purge_state_involving (sval, ext_state: m_ext_state); |
218 | } |
219 | |
220 | void |
221 | impl_region_model_context::bifurcate (std::unique_ptr<custom_edge_info> info) |
222 | { |
223 | if (m_path_ctxt) |
224 | m_path_ctxt->bifurcate (info: std::move (info)); |
225 | } |
226 | |
227 | void |
228 | impl_region_model_context::terminate_path () |
229 | { |
230 | if (m_path_ctxt) |
231 | return m_path_ctxt->terminate_path (); |
232 | } |
233 | |
234 | /* struct setjmp_record. */ |
235 | |
236 | int |
237 | setjmp_record::cmp (const setjmp_record &rec1, const setjmp_record &rec2) |
238 | { |
239 | if (int cmp_enode = rec1.m_enode->m_index - rec2.m_enode->m_index) |
240 | return cmp_enode; |
241 | gcc_assert (&rec1 == &rec2); |
242 | return 0; |
243 | } |
244 | |
245 | /* class setjmp_svalue : public svalue. */ |
246 | |
247 | /* Implementation of svalue::accept vfunc for setjmp_svalue. */ |
248 | |
249 | void |
250 | setjmp_svalue::accept (visitor *v) const |
251 | { |
252 | v->visit_setjmp_svalue (this); |
253 | } |
254 | |
255 | /* Implementation of svalue::dump_to_pp vfunc for setjmp_svalue. */ |
256 | |
257 | void |
258 | setjmp_svalue::dump_to_pp (pretty_printer *pp, bool simple) const |
259 | { |
260 | if (simple) |
261 | pp_printf (pp, "SETJMP(EN: %i)" , get_enode_index ()); |
262 | else |
263 | pp_printf (pp, "setjmp_svalue(EN%i)" , get_enode_index ()); |
264 | } |
265 | |
266 | /* Get the index of the stored exploded_node. */ |
267 | |
268 | int |
269 | setjmp_svalue::get_enode_index () const |
270 | { |
271 | return m_setjmp_record.m_enode->m_index; |
272 | } |
273 | |
274 | /* Concrete implementation of sm_context, wiring it up to the rest of this |
275 | file. */ |
276 | |
277 | class impl_sm_context : public sm_context |
278 | { |
279 | public: |
280 | impl_sm_context (exploded_graph &eg, |
281 | int sm_idx, |
282 | const state_machine &sm, |
283 | exploded_node *enode_for_diag, |
284 | const program_state *old_state, |
285 | program_state *new_state, |
286 | const sm_state_map *old_smap, |
287 | sm_state_map *new_smap, |
288 | path_context *path_ctxt, |
289 | const stmt_finder *stmt_finder = NULL, |
290 | bool unknown_side_effects = false) |
291 | : sm_context (sm_idx, sm), |
292 | m_logger (eg.get_logger ()), |
293 | m_eg (eg), m_enode_for_diag (enode_for_diag), |
294 | m_old_state (old_state), m_new_state (new_state), |
295 | m_old_smap (old_smap), m_new_smap (new_smap), |
296 | m_path_ctxt (path_ctxt), |
297 | m_stmt_finder (stmt_finder), |
298 | m_unknown_side_effects (unknown_side_effects) |
299 | { |
300 | } |
301 | |
302 | logger *get_logger () const { return m_logger.get_logger (); } |
303 | |
304 | tree get_fndecl_for_call (const gcall *call) final override |
305 | { |
306 | impl_region_model_context old_ctxt |
307 | (m_eg, m_enode_for_diag, NULL, NULL, NULL/*m_enode->get_state ()*/, |
308 | NULL, call); |
309 | region_model *model = m_new_state->m_region_model; |
310 | return model->get_fndecl_for_call (call, ctxt: &old_ctxt); |
311 | } |
312 | |
313 | state_machine::state_t get_state (const gimple *stmt ATTRIBUTE_UNUSED, |
314 | tree var) final override |
315 | { |
316 | logger * const logger = get_logger (); |
317 | LOG_FUNC (logger); |
318 | /* Use NULL ctxt on this get_rvalue call to avoid triggering |
319 | uninitialized value warnings. */ |
320 | const svalue *var_old_sval |
321 | = m_old_state->m_region_model->get_rvalue (expr: var, NULL); |
322 | |
323 | state_machine::state_t current |
324 | = m_old_smap->get_state (sval: var_old_sval, ext_state: m_eg.get_ext_state ()); |
325 | return current; |
326 | } |
327 | state_machine::state_t get_state (const gimple *stmt ATTRIBUTE_UNUSED, |
328 | const svalue *sval) final override |
329 | { |
330 | logger * const logger = get_logger (); |
331 | LOG_FUNC (logger); |
332 | state_machine::state_t current |
333 | = m_old_smap->get_state (sval, ext_state: m_eg.get_ext_state ()); |
334 | return current; |
335 | } |
336 | |
337 | |
338 | void set_next_state (const gimple *, |
339 | tree var, |
340 | state_machine::state_t to, |
341 | tree origin) final override |
342 | { |
343 | logger * const logger = get_logger (); |
344 | LOG_FUNC (logger); |
345 | const svalue *var_new_sval |
346 | = m_new_state->m_region_model->get_rvalue (expr: var, NULL); |
347 | const svalue *origin_new_sval |
348 | = m_new_state->m_region_model->get_rvalue (expr: origin, NULL); |
349 | |
350 | /* We use the new sval here to avoid issues with uninitialized values. */ |
351 | state_machine::state_t current |
352 | = m_old_smap->get_state (sval: var_new_sval, ext_state: m_eg.get_ext_state ()); |
353 | if (logger) |
354 | logger->log (fmt: "%s: state transition of %qE: %s -> %s" , |
355 | m_sm.get_name (), |
356 | var, |
357 | current->get_name (), |
358 | to->get_name ()); |
359 | m_new_smap->set_state (model: m_new_state->m_region_model, sval: var_new_sval, |
360 | state: to, origin: origin_new_sval, ext_state: m_eg.get_ext_state ()); |
361 | } |
362 | |
363 | void set_next_state (const gimple *stmt, |
364 | const svalue *sval, |
365 | state_machine::state_t to, |
366 | tree origin) final override |
367 | { |
368 | logger * const logger = get_logger (); |
369 | LOG_FUNC (logger); |
370 | impl_region_model_context old_ctxt |
371 | (m_eg, m_enode_for_diag, NULL, NULL, NULL/*m_enode->get_state ()*/, |
372 | NULL, stmt); |
373 | |
374 | const svalue *origin_new_sval |
375 | = m_new_state->m_region_model->get_rvalue (expr: origin, NULL); |
376 | |
377 | state_machine::state_t current |
378 | = m_old_smap->get_state (sval, ext_state: m_eg.get_ext_state ()); |
379 | if (logger) |
380 | { |
381 | logger->start_log_line (); |
382 | logger->log_partial (fmt: "%s: state transition of " , |
383 | m_sm.get_name ()); |
384 | sval->dump_to_pp (pp: logger->get_printer (), simple: true); |
385 | logger->log_partial (fmt: ": %s -> %s" , |
386 | current->get_name (), |
387 | to->get_name ()); |
388 | logger->end_log_line (); |
389 | } |
390 | m_new_smap->set_state (model: m_new_state->m_region_model, sval, |
391 | state: to, origin: origin_new_sval, ext_state: m_eg.get_ext_state ()); |
392 | } |
393 | |
394 | void warn (const supernode *snode, const gimple *stmt, |
395 | tree var, |
396 | std::unique_ptr<pending_diagnostic> d) final override |
397 | { |
398 | LOG_FUNC (get_logger ()); |
399 | gcc_assert (d); |
400 | const svalue *var_old_sval |
401 | = m_old_state->m_region_model->get_rvalue (expr: var, NULL); |
402 | state_machine::state_t current |
403 | = (var |
404 | ? m_old_smap->get_state (sval: var_old_sval, ext_state: m_eg.get_ext_state ()) |
405 | : m_old_smap->get_global_state ()); |
406 | bool terminate_path = d->terminate_path_p (); |
407 | pending_location ploc (m_enode_for_diag, snode, stmt, m_stmt_finder); |
408 | m_eg.get_diagnostic_manager ().add_diagnostic |
409 | (sm: &m_sm, ploc, |
410 | var, sval: var_old_sval, state: current, d: std::move (d)); |
411 | if (m_path_ctxt |
412 | && terminate_path |
413 | && flag_analyzer_suppress_followups) |
414 | m_path_ctxt->terminate_path (); |
415 | } |
416 | |
417 | void warn (const supernode *snode, const gimple *stmt, |
418 | const svalue *sval, |
419 | std::unique_ptr<pending_diagnostic> d) final override |
420 | { |
421 | LOG_FUNC (get_logger ()); |
422 | gcc_assert (d); |
423 | state_machine::state_t current |
424 | = (sval |
425 | ? m_old_smap->get_state (sval, ext_state: m_eg.get_ext_state ()) |
426 | : m_old_smap->get_global_state ()); |
427 | bool terminate_path = d->terminate_path_p (); |
428 | pending_location ploc (m_enode_for_diag, snode, stmt, m_stmt_finder); |
429 | m_eg.get_diagnostic_manager ().add_diagnostic |
430 | (sm: &m_sm, ploc, |
431 | NULL_TREE, sval, state: current, d: std::move (d)); |
432 | if (m_path_ctxt |
433 | && terminate_path |
434 | && flag_analyzer_suppress_followups) |
435 | m_path_ctxt->terminate_path (); |
436 | } |
437 | |
438 | /* Hook for picking more readable trees for SSA names of temporaries, |
439 | so that rather than e.g. |
440 | "double-free of '<unknown>'" |
441 | we can print: |
442 | "double-free of 'inbuf.data'". */ |
443 | |
444 | tree get_diagnostic_tree (tree expr) final override |
445 | { |
446 | /* Only for SSA_NAMEs of temporaries; otherwise, return EXPR, as it's |
447 | likely to be the least surprising tree to report. */ |
448 | if (TREE_CODE (expr) != SSA_NAME) |
449 | return expr; |
450 | if (SSA_NAME_VAR (expr) != NULL) |
451 | return expr; |
452 | |
453 | gcc_assert (m_new_state); |
454 | const svalue *sval = m_new_state->m_region_model->get_rvalue (expr, NULL); |
455 | /* Find trees for all regions storing the value. */ |
456 | if (tree t = m_new_state->m_region_model->get_representative_tree (sval)) |
457 | return t; |
458 | else |
459 | return expr; |
460 | } |
461 | |
462 | tree get_diagnostic_tree (const svalue *sval) final override |
463 | { |
464 | return m_new_state->m_region_model->get_representative_tree (sval); |
465 | } |
466 | |
467 | state_machine::state_t get_global_state () const final override |
468 | { |
469 | return m_old_state->m_checker_states[m_sm_idx]->get_global_state (); |
470 | } |
471 | |
472 | void set_global_state (state_machine::state_t state) final override |
473 | { |
474 | m_new_state->m_checker_states[m_sm_idx]->set_global_state (state); |
475 | } |
476 | |
477 | void clear_all_per_svalue_state () final override |
478 | { |
479 | m_new_state->m_checker_states[m_sm_idx]->clear_all_per_svalue_state (); |
480 | } |
481 | |
482 | void on_custom_transition (custom_transition *transition) final override |
483 | { |
484 | transition->impl_transition (eg: &m_eg, |
485 | src_enode: const_cast<exploded_node *> (m_enode_for_diag), |
486 | sm_idx: m_sm_idx); |
487 | } |
488 | |
489 | tree is_zero_assignment (const gimple *stmt) final override |
490 | { |
491 | const gassign *assign_stmt = dyn_cast <const gassign *> (p: stmt); |
492 | if (!assign_stmt) |
493 | return NULL_TREE; |
494 | impl_region_model_context old_ctxt |
495 | (m_eg, m_enode_for_diag, m_old_state, m_new_state, NULL, NULL, stmt); |
496 | if (const svalue *sval |
497 | = m_new_state->m_region_model->get_gassign_result (assign: assign_stmt, |
498 | ctxt: &old_ctxt)) |
499 | if (tree cst = sval->maybe_get_constant ()) |
500 | if (::zerop(cst)) |
501 | return gimple_assign_lhs (gs: assign_stmt); |
502 | return NULL_TREE; |
503 | } |
504 | |
505 | path_context *get_path_context () const final override |
506 | { |
507 | return m_path_ctxt; |
508 | } |
509 | |
510 | bool unknown_side_effects_p () const final override |
511 | { |
512 | return m_unknown_side_effects; |
513 | } |
514 | |
515 | const program_state *get_old_program_state () const final override |
516 | { |
517 | return m_old_state; |
518 | } |
519 | |
520 | const program_state *get_new_program_state () const final override |
521 | { |
522 | return m_new_state; |
523 | } |
524 | |
525 | log_user m_logger; |
526 | exploded_graph &m_eg; |
527 | exploded_node *m_enode_for_diag; |
528 | const program_state *m_old_state; |
529 | program_state *m_new_state; |
530 | const sm_state_map *m_old_smap; |
531 | sm_state_map *m_new_smap; |
532 | path_context *m_path_ctxt; |
533 | const stmt_finder *m_stmt_finder; |
534 | |
535 | /* Are we handling an external function with unknown side effects? */ |
536 | bool m_unknown_side_effects; |
537 | }; |
538 | |
539 | bool |
540 | impl_region_model_context:: |
541 | get_state_map_by_name (const char *name, |
542 | sm_state_map **out_smap, |
543 | const state_machine **out_sm, |
544 | unsigned *out_sm_idx, |
545 | std::unique_ptr<sm_context> *out_sm_context) |
546 | { |
547 | if (!m_new_state) |
548 | return false; |
549 | |
550 | unsigned sm_idx; |
551 | if (!m_ext_state.get_sm_idx_by_name (name, out: &sm_idx)) |
552 | return false; |
553 | |
554 | const state_machine *sm = &m_ext_state.get_sm (idx: sm_idx); |
555 | sm_state_map *new_smap = m_new_state->m_checker_states[sm_idx]; |
556 | |
557 | *out_smap = new_smap; |
558 | *out_sm = sm; |
559 | if (out_sm_idx) |
560 | *out_sm_idx = sm_idx; |
561 | if (out_sm_context) |
562 | { |
563 | const sm_state_map *old_smap = m_old_state->m_checker_states[sm_idx]; |
564 | *out_sm_context |
565 | = make_unique<impl_sm_context> (args&: *m_eg, |
566 | args&: sm_idx, |
567 | args: *sm, |
568 | args&: m_enode_for_diag, |
569 | args&: m_old_state, |
570 | args&: m_new_state, |
571 | args&: old_smap, |
572 | args&: new_smap, |
573 | args&: m_path_ctxt, |
574 | args&: m_stmt_finder, |
575 | args: false); |
576 | } |
577 | return true; |
578 | } |
579 | |
580 | /* Subclass of stmt_finder for finding the best stmt to report the leak at, |
581 | given the emission path. */ |
582 | |
583 | class leak_stmt_finder : public stmt_finder |
584 | { |
585 | public: |
586 | leak_stmt_finder (const exploded_graph &eg, tree var) |
587 | : m_eg (eg), m_var (var) {} |
588 | |
589 | std::unique_ptr<stmt_finder> clone () const final override |
590 | { |
591 | return make_unique<leak_stmt_finder> (args: m_eg, args: m_var); |
592 | } |
593 | |
594 | const gimple *find_stmt (const exploded_path &epath) |
595 | final override |
596 | { |
597 | logger * const logger = m_eg.get_logger (); |
598 | LOG_FUNC (logger); |
599 | |
600 | if (m_var && TREE_CODE (m_var) == SSA_NAME) |
601 | { |
602 | /* Locate the final write to this SSA name in the path. */ |
603 | const gimple *def_stmt = SSA_NAME_DEF_STMT (m_var); |
604 | |
605 | int idx_of_def_stmt; |
606 | bool found = epath.find_stmt_backwards (search_stmt: def_stmt, out_idx: &idx_of_def_stmt); |
607 | if (!found) |
608 | goto not_found; |
609 | |
610 | /* What was the next write to the underlying var |
611 | after the SSA name was set? (if any). */ |
612 | |
613 | for (unsigned idx = idx_of_def_stmt + 1; |
614 | idx < epath.m_edges.length (); |
615 | ++idx) |
616 | { |
617 | const exploded_edge *eedge = epath.m_edges[idx]; |
618 | if (logger) |
619 | logger->log (fmt: "eedge[%i]: EN %i -> EN %i" , |
620 | idx, |
621 | eedge->m_src->m_index, |
622 | eedge->m_dest->m_index); |
623 | const exploded_node *dst_node = eedge->m_dest; |
624 | const program_point &dst_point = dst_node->get_point (); |
625 | const gimple *stmt = dst_point.get_stmt (); |
626 | if (!stmt) |
627 | continue; |
628 | if (const gassign *assign = dyn_cast <const gassign *> (p: stmt)) |
629 | { |
630 | tree lhs = gimple_assign_lhs (gs: assign); |
631 | if (TREE_CODE (lhs) == SSA_NAME |
632 | && SSA_NAME_VAR (lhs) == SSA_NAME_VAR (m_var)) |
633 | return assign; |
634 | } |
635 | } |
636 | } |
637 | |
638 | not_found: |
639 | |
640 | /* Look backwards for the first statement with a location. */ |
641 | int i; |
642 | const exploded_edge *eedge; |
643 | FOR_EACH_VEC_ELT_REVERSE (epath.m_edges, i, eedge) |
644 | { |
645 | if (logger) |
646 | logger->log (fmt: "eedge[%i]: EN %i -> EN %i" , |
647 | i, |
648 | eedge->m_src->m_index, |
649 | eedge->m_dest->m_index); |
650 | const exploded_node *dst_node = eedge->m_dest; |
651 | const program_point &dst_point = dst_node->get_point (); |
652 | const gimple *stmt = dst_point.get_stmt (); |
653 | if (stmt) |
654 | if (get_pure_location (loc: stmt->location) != UNKNOWN_LOCATION) |
655 | return stmt; |
656 | } |
657 | |
658 | gcc_unreachable (); |
659 | return NULL; |
660 | } |
661 | |
662 | private: |
663 | const exploded_graph &m_eg; |
664 | tree m_var; |
665 | }; |
666 | |
667 | /* A measurement of how good EXPR is for presenting to the user, so |
668 | that e.g. we can say prefer printing |
669 | "leak of 'tmp.m_ptr'" |
670 | over: |
671 | "leak of '<unknown>'". */ |
672 | |
673 | static int |
674 | readability (const_tree expr) |
675 | { |
676 | /* Arbitrarily-chosen "high readability" value. */ |
677 | const int HIGH_READABILITY = 65536; |
678 | |
679 | gcc_assert (expr); |
680 | switch (TREE_CODE (expr)) |
681 | { |
682 | case COMPONENT_REF: |
683 | case MEM_REF: |
684 | /* Impose a slight readability penalty relative to that of |
685 | operand 0. */ |
686 | return readability (TREE_OPERAND (expr, 0)) - 16; |
687 | |
688 | case SSA_NAME: |
689 | { |
690 | if (tree var = SSA_NAME_VAR (expr)) |
691 | { |
692 | if (DECL_ARTIFICIAL (var)) |
693 | { |
694 | /* If we have an SSA name for an artificial var, |
695 | only use it if it has a debug expr associated with |
696 | it that fixup_tree_for_diagnostic can use. */ |
697 | if (VAR_P (var) && DECL_HAS_DEBUG_EXPR_P (var)) |
698 | return readability (DECL_DEBUG_EXPR (var)) - 1; |
699 | } |
700 | else |
701 | { |
702 | /* Slightly favor the underlying var over the SSA name to |
703 | avoid having them compare equal. */ |
704 | return readability (expr: var) - 1; |
705 | } |
706 | } |
707 | /* Avoid printing '<unknown>' for SSA names for temporaries. */ |
708 | return -1; |
709 | } |
710 | break; |
711 | |
712 | case PARM_DECL: |
713 | case VAR_DECL: |
714 | if (DECL_NAME (expr)) |
715 | return HIGH_READABILITY; |
716 | else |
717 | /* We don't want to print temporaries. For example, the C FE |
718 | prints them as e.g. "<Uxxxx>" where "xxxx" is the low 16 bits |
719 | of the tree pointer (see pp_c_tree_decl_identifier). */ |
720 | return -1; |
721 | |
722 | case RESULT_DECL: |
723 | /* Printing "<return-value>" isn't ideal, but is less awful than |
724 | trying to print a temporary. */ |
725 | return HIGH_READABILITY / 2; |
726 | |
727 | case NOP_EXPR: |
728 | { |
729 | /* Impose a moderate readability penalty for casts. */ |
730 | const int CAST_PENALTY = 32; |
731 | return readability (TREE_OPERAND (expr, 0)) - CAST_PENALTY; |
732 | } |
733 | |
734 | case INTEGER_CST: |
735 | return HIGH_READABILITY; |
736 | |
737 | default: |
738 | return 0; |
739 | } |
740 | |
741 | return 0; |
742 | } |
743 | |
744 | /* A qsort comparator for trees to sort them into most user-readable to |
745 | least user-readable. */ |
746 | |
747 | int |
748 | readability_comparator (const void *p1, const void *p2) |
749 | { |
750 | path_var pv1 = *(path_var const *)p1; |
751 | path_var pv2 = *(path_var const *)p2; |
752 | |
753 | const int tree_r1 = readability (expr: pv1.m_tree); |
754 | const int tree_r2 = readability (expr: pv2.m_tree); |
755 | |
756 | /* Favor items that are deeper on the stack and hence more recent; |
757 | this also favors locals over globals. */ |
758 | const int COST_PER_FRAME = 64; |
759 | const int depth_r1 = pv1.m_stack_depth * COST_PER_FRAME; |
760 | const int depth_r2 = pv2.m_stack_depth * COST_PER_FRAME; |
761 | |
762 | /* Combine the scores from the tree and from the stack depth. |
763 | This e.g. lets us have a slightly penalized cast in the most |
764 | recent stack frame "beat" an uncast value in an older stack frame. */ |
765 | const int sum_r1 = tree_r1 + depth_r1; |
766 | const int sum_r2 = tree_r2 + depth_r2; |
767 | if (int cmp = sum_r2 - sum_r1) |
768 | return cmp; |
769 | |
770 | /* Otherwise, more readable trees win. */ |
771 | if (int cmp = tree_r2 - tree_r1) |
772 | return cmp; |
773 | |
774 | /* Otherwise, if they have the same readability, then impose an |
775 | arbitrary deterministic ordering on them. */ |
776 | |
777 | if (int cmp = TREE_CODE (pv1.m_tree) - TREE_CODE (pv2.m_tree)) |
778 | return cmp; |
779 | |
780 | switch (TREE_CODE (pv1.m_tree)) |
781 | { |
782 | default: |
783 | break; |
784 | case SSA_NAME: |
785 | if (int cmp = (SSA_NAME_VERSION (pv1.m_tree) |
786 | - SSA_NAME_VERSION (pv2.m_tree))) |
787 | return cmp; |
788 | break; |
789 | case PARM_DECL: |
790 | case VAR_DECL: |
791 | case RESULT_DECL: |
792 | if (int cmp = DECL_UID (pv1.m_tree) - DECL_UID (pv2.m_tree)) |
793 | return cmp; |
794 | break; |
795 | } |
796 | |
797 | /* TODO: We ought to find ways of sorting such cases. */ |
798 | return 0; |
799 | } |
800 | |
801 | /* Return true is SNODE is the EXIT node of a function, or is one |
802 | of the final snodes within its function. |
803 | |
804 | Specifically, handle the final supernodes before the EXIT node, |
805 | for the case of clobbers that happen immediately before exiting. |
806 | We need a run of snodes leading to the return_p snode, where all edges are |
807 | intraprocedural, and every snode has just one successor. |
808 | |
809 | We use this when suppressing leak reports at the end of "main". */ |
810 | |
811 | static bool |
812 | returning_from_function_p (const supernode *snode) |
813 | { |
814 | if (!snode) |
815 | return false; |
816 | |
817 | unsigned count = 0; |
818 | const supernode *iter = snode; |
819 | while (true) |
820 | { |
821 | if (iter->return_p ()) |
822 | return true; |
823 | if (iter->m_succs.length () != 1) |
824 | return false; |
825 | const superedge *sedge = iter->m_succs[0]; |
826 | if (sedge->get_kind () != SUPEREDGE_CFG_EDGE) |
827 | return false; |
828 | iter = sedge->m_dest; |
829 | |
830 | /* Impose a limit to ensure we terminate for pathological cases. |
831 | |
832 | We only care about the final 3 nodes, due to cases like: |
833 | BB: |
834 | (clobber causing leak) |
835 | |
836 | BB: |
837 | <label>: |
838 | return _val; |
839 | |
840 | EXIT BB.*/ |
841 | if (++count > 3) |
842 | return false; |
843 | } |
844 | } |
845 | |
846 | /* Find the best tree for SVAL and call SM's on_leak vfunc with it. |
847 | If on_leak returns a pending_diagnostic, queue it up to be reported, |
848 | so that we potentially complain about a leak of SVAL in the given STATE. */ |
849 | |
850 | void |
851 | impl_region_model_context::on_state_leak (const state_machine &sm, |
852 | const svalue *sval, |
853 | state_machine::state_t state) |
854 | { |
855 | logger * const logger = get_logger (); |
856 | LOG_SCOPE (logger); |
857 | if (logger) |
858 | { |
859 | logger->start_log_line (); |
860 | logger->log_partial (fmt: "considering leak of " ); |
861 | sval->dump_to_pp (pp: logger->get_printer (), simple: true); |
862 | logger->end_log_line (); |
863 | } |
864 | |
865 | if (!m_eg) |
866 | return; |
867 | |
868 | /* m_old_state also needs to be non-NULL so that the sm_ctxt can look |
869 | up the old state of SVAL. */ |
870 | gcc_assert (m_old_state); |
871 | |
872 | /* SVAL has leaked within the new state: it is not used by any reachable |
873 | regions. |
874 | We need to convert it back to a tree, but since it's likely no regions |
875 | use it, we have to find the "best" tree for it in the old_state. */ |
876 | svalue_set visited; |
877 | path_var leaked_pv |
878 | = m_old_state->m_region_model->get_representative_path_var (sval, |
879 | visited: &visited); |
880 | |
881 | /* Strip off top-level casts */ |
882 | if (leaked_pv.m_tree && TREE_CODE (leaked_pv.m_tree) == NOP_EXPR) |
883 | leaked_pv.m_tree = TREE_OPERAND (leaked_pv.m_tree, 0); |
884 | |
885 | /* This might be NULL; the pending_diagnostic subclasses need to cope |
886 | with this. */ |
887 | tree leaked_tree = leaked_pv.m_tree; |
888 | if (logger) |
889 | { |
890 | if (leaked_tree) |
891 | logger->log (fmt: "best leaked_tree: %qE" , leaked_tree); |
892 | else |
893 | logger->log (fmt: "best leaked_tree: NULL" ); |
894 | } |
895 | |
896 | leak_stmt_finder stmt_finder (*m_eg, leaked_tree); |
897 | gcc_assert (m_enode_for_diag); |
898 | |
899 | /* Don't complain about leaks when returning from "main". */ |
900 | if (returning_from_function_p (snode: m_enode_for_diag->get_supernode ())) |
901 | { |
902 | tree fndecl = m_enode_for_diag->get_function ()->decl; |
903 | if (id_equal (DECL_NAME (fndecl), str: "main" )) |
904 | { |
905 | if (logger) |
906 | logger->log (fmt: "not reporting leak from main" ); |
907 | return; |
908 | } |
909 | } |
910 | |
911 | tree leaked_tree_for_diag = fixup_tree_for_diagnostic (leaked_tree); |
912 | std::unique_ptr<pending_diagnostic> pd = sm.on_leak (var: leaked_tree_for_diag); |
913 | if (pd) |
914 | { |
915 | pending_location ploc (m_enode_for_diag, |
916 | m_enode_for_diag->get_supernode (), |
917 | m_stmt, |
918 | &stmt_finder); |
919 | m_eg->get_diagnostic_manager ().add_diagnostic |
920 | (sm: &sm, ploc, |
921 | var: leaked_tree_for_diag, sval, state, d: std::move (pd)); |
922 | } |
923 | } |
924 | |
925 | /* Implementation of region_model_context::on_condition vfunc. |
926 | Notify all state machines about the condition, which could lead to |
927 | state transitions. */ |
928 | |
929 | void |
930 | impl_region_model_context::on_condition (const svalue *lhs, |
931 | enum tree_code op, |
932 | const svalue *rhs) |
933 | { |
934 | int sm_idx; |
935 | sm_state_map *smap; |
936 | FOR_EACH_VEC_ELT (m_new_state->m_checker_states, sm_idx, smap) |
937 | { |
938 | const state_machine &sm = m_ext_state.get_sm (idx: sm_idx); |
939 | impl_sm_context sm_ctxt (*m_eg, sm_idx, sm, m_enode_for_diag, |
940 | m_old_state, m_new_state, |
941 | m_old_state->m_checker_states[sm_idx], |
942 | m_new_state->m_checker_states[sm_idx], |
943 | m_path_ctxt); |
944 | sm.on_condition (sm_ctxt: &sm_ctxt, |
945 | node: (m_enode_for_diag |
946 | ? m_enode_for_diag->get_supernode () |
947 | : NULL), |
948 | stmt: m_stmt, |
949 | lhs, op, rhs); |
950 | } |
951 | } |
952 | |
953 | /* Implementation of region_model_context::on_bounded_ranges vfunc. |
954 | Notify all state machines about the ranges, which could lead to |
955 | state transitions. */ |
956 | |
957 | void |
958 | impl_region_model_context::on_bounded_ranges (const svalue &sval, |
959 | const bounded_ranges &ranges) |
960 | { |
961 | int sm_idx; |
962 | sm_state_map *smap; |
963 | FOR_EACH_VEC_ELT (m_new_state->m_checker_states, sm_idx, smap) |
964 | { |
965 | const state_machine &sm = m_ext_state.get_sm (idx: sm_idx); |
966 | impl_sm_context sm_ctxt (*m_eg, sm_idx, sm, m_enode_for_diag, |
967 | m_old_state, m_new_state, |
968 | m_old_state->m_checker_states[sm_idx], |
969 | m_new_state->m_checker_states[sm_idx], |
970 | m_path_ctxt); |
971 | sm.on_bounded_ranges (sm_ctxt: &sm_ctxt, |
972 | node: (m_enode_for_diag |
973 | ? m_enode_for_diag->get_supernode () |
974 | : NULL), |
975 | stmt: m_stmt, sval, ranges); |
976 | } |
977 | } |
978 | |
979 | /* Implementation of region_model_context::on_pop_frame vfunc. |
980 | Notify all state machines about the frame being popped, which |
981 | could lead to states being discarded. */ |
982 | |
983 | void |
984 | impl_region_model_context::on_pop_frame (const frame_region *frame_reg) |
985 | { |
986 | int sm_idx; |
987 | sm_state_map *smap; |
988 | FOR_EACH_VEC_ELT (m_new_state->m_checker_states, sm_idx, smap) |
989 | { |
990 | const state_machine &sm = m_ext_state.get_sm (idx: sm_idx); |
991 | sm.on_pop_frame (smap, frame_reg); |
992 | } |
993 | } |
994 | |
995 | /* Implementation of region_model_context::on_phi vfunc. |
996 | Notify all state machines about the phi, which could lead to |
997 | state transitions. */ |
998 | |
999 | void |
1000 | impl_region_model_context::on_phi (const gphi *phi, tree rhs) |
1001 | { |
1002 | int sm_idx; |
1003 | sm_state_map *smap; |
1004 | FOR_EACH_VEC_ELT (m_new_state->m_checker_states, sm_idx, smap) |
1005 | { |
1006 | const state_machine &sm = m_ext_state.get_sm (idx: sm_idx); |
1007 | impl_sm_context sm_ctxt (*m_eg, sm_idx, sm, m_enode_for_diag, |
1008 | m_old_state, m_new_state, |
1009 | m_old_state->m_checker_states[sm_idx], |
1010 | m_new_state->m_checker_states[sm_idx], |
1011 | m_path_ctxt); |
1012 | sm.on_phi (sm_ctxt: &sm_ctxt, node: m_enode_for_diag->get_supernode (), phi, rhs); |
1013 | } |
1014 | } |
1015 | |
1016 | /* Implementation of region_model_context::on_unexpected_tree_code vfunc. |
1017 | Mark the new state as being invalid for further exploration. |
1018 | TODO(stage1): introduce a warning for when this occurs. */ |
1019 | |
1020 | void |
1021 | impl_region_model_context::on_unexpected_tree_code (tree t, |
1022 | const dump_location_t &loc) |
1023 | { |
1024 | logger * const logger = get_logger (); |
1025 | if (logger) |
1026 | logger->log (fmt: "unhandled tree code: %qs in %qs at %s:%i" , |
1027 | get_tree_code_name (TREE_CODE (t)), |
1028 | loc.get_impl_location ().m_function, |
1029 | loc.get_impl_location ().m_file, |
1030 | loc.get_impl_location ().m_line); |
1031 | if (m_new_state) |
1032 | m_new_state->m_valid = false; |
1033 | } |
1034 | |
1035 | /* Implementation of region_model_context::maybe_did_work vfunc. |
1036 | Mark that "externally visible work" has occurred, and thus we |
1037 | shouldn't report an infinite loop here. */ |
1038 | |
1039 | void |
1040 | impl_region_model_context::maybe_did_work () |
1041 | { |
1042 | if (m_out_could_have_done_work) |
1043 | *m_out_could_have_done_work = true; |
1044 | } |
1045 | |
1046 | /* struct point_and_state. */ |
1047 | |
1048 | /* Assert that this object is sane. */ |
1049 | |
1050 | void |
1051 | point_and_state::validate (const extrinsic_state &ext_state) const |
1052 | { |
1053 | /* Skip this in a release build. */ |
1054 | #if !CHECKING_P |
1055 | return; |
1056 | #endif |
1057 | |
1058 | m_point.validate (); |
1059 | |
1060 | m_state.validate (ext_state); |
1061 | |
1062 | /* Verify that the callstring's model of the stack corresponds to that |
1063 | of the region_model. */ |
1064 | /* They should have the same depth. */ |
1065 | gcc_assert (m_point.get_stack_depth () |
1066 | == m_state.m_region_model->get_stack_depth ()); |
1067 | /* Check the functions in the callstring vs those in the frames |
1068 | at each depth. */ |
1069 | for (const frame_region *iter_frame |
1070 | = m_state.m_region_model->get_current_frame (); |
1071 | iter_frame; iter_frame = iter_frame->get_calling_frame ()) |
1072 | { |
1073 | int index = iter_frame->get_index (); |
1074 | gcc_assert (m_point.get_function_at_depth (index) |
1075 | == &iter_frame->get_function ()); |
1076 | } |
1077 | } |
1078 | |
1079 | /* Subroutine of print_enode_indices: print a run of indices from START_IDX |
1080 | to END_IDX to PP, using and updating *FIRST_RUN. */ |
1081 | |
1082 | static void |
1083 | print_run (pretty_printer *pp, int start_idx, int end_idx, |
1084 | bool *first_run) |
1085 | { |
1086 | if (!(*first_run)) |
1087 | pp_string (pp, ", " ); |
1088 | *first_run = false; |
1089 | if (start_idx == end_idx) |
1090 | pp_printf (pp, "EN: %i" , start_idx); |
1091 | else |
1092 | pp_printf (pp, "EN: %i-%i" , start_idx, end_idx); |
1093 | } |
1094 | |
1095 | /* Print the indices within ENODES to PP, collecting them as |
1096 | runs/singletons e.g. "EN: 4-7, EN: 20-23, EN: 42". */ |
1097 | |
1098 | static void |
1099 | print_enode_indices (pretty_printer *pp, |
1100 | const auto_vec<exploded_node *> &enodes) |
1101 | { |
1102 | int cur_start_idx = -1; |
1103 | int cur_finish_idx = -1; |
1104 | bool first_run = true; |
1105 | unsigned i; |
1106 | exploded_node *enode; |
1107 | FOR_EACH_VEC_ELT (enodes, i, enode) |
1108 | { |
1109 | if (cur_start_idx == -1) |
1110 | { |
1111 | gcc_assert (cur_finish_idx == -1); |
1112 | cur_start_idx = cur_finish_idx = enode->m_index; |
1113 | } |
1114 | else |
1115 | { |
1116 | if (enode->m_index == cur_finish_idx + 1) |
1117 | /* Continuation of a run. */ |
1118 | cur_finish_idx = enode->m_index; |
1119 | else |
1120 | { |
1121 | /* Finish existing run, start a new one. */ |
1122 | gcc_assert (cur_start_idx >= 0); |
1123 | gcc_assert (cur_finish_idx >= 0); |
1124 | print_run (pp, start_idx: cur_start_idx, end_idx: cur_finish_idx, |
1125 | first_run: &first_run); |
1126 | cur_start_idx = cur_finish_idx = enode->m_index; |
1127 | } |
1128 | } |
1129 | } |
1130 | /* Finish any existing run. */ |
1131 | if (cur_start_idx >= 0) |
1132 | { |
1133 | gcc_assert (cur_finish_idx >= 0); |
1134 | print_run (pp, start_idx: cur_start_idx, end_idx: cur_finish_idx, |
1135 | first_run: &first_run); |
1136 | } |
1137 | } |
1138 | |
1139 | /* struct eg_traits::dump_args_t. */ |
1140 | |
1141 | /* The <FILENAME>.eg.dot output can quickly become unwieldy if we show |
1142 | full details for all enodes (both in terms of CPU time to render it, |
1143 | and in terms of being meaningful to a human viewing it). |
1144 | |
1145 | If we show just the IDs then the resulting graph is usually viewable, |
1146 | but then we have to keep switching back and forth between the .dot |
1147 | view and other dumps. |
1148 | |
1149 | This function implements a heuristic for showing detail at the enodes |
1150 | that (we hope) matter, and just the ID at other enodes, fixing the CPU |
1151 | usage of the .dot viewer, and drawing the attention of the viewer |
1152 | to these enodes. |
1153 | |
1154 | Return true if ENODE should be shown in detail in .dot output. |
1155 | Return false if no detail should be shown for ENODE. */ |
1156 | |
1157 | bool |
1158 | eg_traits::dump_args_t::show_enode_details_p (const exploded_node &enode) const |
1159 | { |
1160 | /* If the number of exploded nodes isn't too large, we may as well show |
1161 | all enodes in full detail in the .dot output. */ |
1162 | if (m_eg.m_nodes.length () |
1163 | <= (unsigned) param_analyzer_max_enodes_for_full_dump) |
1164 | return true; |
1165 | |
1166 | /* Otherwise, assume that what's most interesting are state explosions, |
1167 | and thus the places where this happened. |
1168 | Expand enodes at program points where we hit the per-enode limit, so we |
1169 | can investigate what exploded. */ |
1170 | const per_program_point_data *per_point_data |
1171 | = m_eg.get_per_program_point_data (enode.get_point ()); |
1172 | return per_point_data->m_excess_enodes > 0; |
1173 | } |
1174 | |
1175 | /* class exploded_node : public dnode<eg_traits>. */ |
1176 | |
1177 | const char * |
1178 | exploded_node::status_to_str (enum status s) |
1179 | { |
1180 | switch (s) |
1181 | { |
1182 | default: gcc_unreachable (); |
1183 | case STATUS_WORKLIST: return "WORKLIST" ; |
1184 | case STATUS_PROCESSED: return "PROCESSED" ; |
1185 | case STATUS_MERGER: return "MERGER" ; |
1186 | case STATUS_BULK_MERGED: return "BULK_MERGED" ; |
1187 | } |
1188 | } |
1189 | |
1190 | /* exploded_node's ctor. */ |
1191 | |
1192 | exploded_node::exploded_node (const point_and_state &ps, |
1193 | int index) |
1194 | : m_ps (ps), m_status (STATUS_WORKLIST), m_index (index), |
1195 | m_num_processed_stmts (0) |
1196 | { |
1197 | gcc_checking_assert (ps.get_state ().m_region_model->canonicalized_p ()); |
1198 | } |
1199 | |
1200 | /* Get the stmt that was processed in this enode at index IDX. |
1201 | IDX is an index within the stmts processed at this enode, rather |
1202 | than within those of the supernode. */ |
1203 | |
1204 | const gimple * |
1205 | exploded_node::get_processed_stmt (unsigned idx) const |
1206 | { |
1207 | gcc_assert (idx < m_num_processed_stmts); |
1208 | const program_point &point = get_point (); |
1209 | gcc_assert (point.get_kind () == PK_BEFORE_STMT); |
1210 | const supernode *snode = get_supernode (); |
1211 | const unsigned int point_stmt_idx = point.get_stmt_idx (); |
1212 | const unsigned int idx_within_snode = point_stmt_idx + idx; |
1213 | const gimple *stmt = snode->m_stmts[idx_within_snode]; |
1214 | return stmt; |
1215 | } |
1216 | |
1217 | /* For use by dump_dot, get a value for the .dot "fillcolor" attribute. |
1218 | Colorize by sm-state, to make it easier to see how sm-state propagates |
1219 | through the exploded_graph. */ |
1220 | |
1221 | const char * |
1222 | exploded_node::get_dot_fillcolor () const |
1223 | { |
1224 | const program_state &state = get_state (); |
1225 | |
1226 | /* We want to be able to easily distinguish the no-sm-state case, |
1227 | and to be able to distinguish cases where there's a single state |
1228 | from each other. |
1229 | |
1230 | Sum the sm_states, and use the result to choose from a table, |
1231 | modulo table-size, special-casing the "no sm-state" case. */ |
1232 | int total_sm_state = 0; |
1233 | int i; |
1234 | sm_state_map *smap; |
1235 | FOR_EACH_VEC_ELT (state.m_checker_states, i, smap) |
1236 | { |
1237 | for (sm_state_map::iterator_t iter = smap->begin (); |
1238 | iter != smap->end (); |
1239 | ++iter) |
1240 | total_sm_state += (*iter).second.m_state->get_id (); |
1241 | total_sm_state += smap->get_global_state ()->get_id (); |
1242 | } |
1243 | |
1244 | if (total_sm_state > 0) |
1245 | { |
1246 | /* An arbitrarily-picked collection of light colors. */ |
1247 | const char * const colors[] |
1248 | = {"azure" , "coral" , "cornsilk" , "lightblue" , "yellow" , |
1249 | "honeydew" , "lightpink" , "lightsalmon" , "palegreen1" , |
1250 | "wheat" , "seashell" }; |
1251 | const int num_colors = ARRAY_SIZE (colors); |
1252 | return colors[total_sm_state % num_colors]; |
1253 | } |
1254 | else |
1255 | /* No sm-state. */ |
1256 | return "lightgrey" ; |
1257 | } |
1258 | |
1259 | /* Implementation of dnode::dump_dot vfunc for exploded_node. */ |
1260 | |
1261 | void |
1262 | exploded_node::dump_dot (graphviz_out *gv, const dump_args_t &args) const |
1263 | { |
1264 | pretty_printer *pp = gv->get_pp (); |
1265 | |
1266 | dump_dot_id (pp); |
1267 | pp_printf (pp, " [shape=none,margin=0,style=filled,fillcolor=%s,label=\"" , |
1268 | get_dot_fillcolor ()); |
1269 | pp_write_text_to_stream (pp); |
1270 | |
1271 | pp_printf (pp, "EN: %i" , m_index); |
1272 | if (m_status == STATUS_MERGER) |
1273 | pp_string (pp, " (merger)" ); |
1274 | else if (m_status == STATUS_BULK_MERGED) |
1275 | pp_string (pp, " (bulk merged)" ); |
1276 | pp_newline (pp); |
1277 | |
1278 | if (args.show_enode_details_p (enode: *this)) |
1279 | { |
1280 | format f (true); |
1281 | m_ps.get_point ().print (pp, f); |
1282 | pp_newline (pp); |
1283 | |
1284 | const extrinsic_state &ext_state = args.m_eg.get_ext_state (); |
1285 | const program_state &state = m_ps.get_state (); |
1286 | state.dump_to_pp (ext_state, simple: false, multiline: true, pp); |
1287 | pp_newline (pp); |
1288 | |
1289 | dump_processed_stmts (pp); |
1290 | } |
1291 | |
1292 | dump_saved_diagnostics (pp); |
1293 | |
1294 | args.dump_extra_info (this, pp); |
1295 | |
1296 | pp_write_text_as_dot_label_to_stream (pp, /*for_record=*/true); |
1297 | |
1298 | pp_string (pp, "\"];\n\n" ); |
1299 | |
1300 | /* It can be hard to locate the saved diagnostics as text within the |
1301 | enode nodes, so add extra nodes to the graph for each saved_diagnostic, |
1302 | highlighted in red. |
1303 | Compare with dump_saved_diagnostics. */ |
1304 | { |
1305 | unsigned i; |
1306 | const saved_diagnostic *sd; |
1307 | FOR_EACH_VEC_ELT (m_saved_diagnostics, i, sd) |
1308 | { |
1309 | sd->dump_as_dot_node (pp); |
1310 | |
1311 | /* Add edge connecting this enode to the saved_diagnostic. */ |
1312 | dump_dot_id (pp); |
1313 | pp_string (pp, " -> " ); |
1314 | sd->dump_dot_id (pp); |
1315 | pp_string (pp, " [style=\"dotted\" arrowhead=\"none\"];" ); |
1316 | pp_newline (pp); |
1317 | } |
1318 | } |
1319 | |
1320 | pp_flush (pp); |
1321 | } |
1322 | |
1323 | /* Show any stmts that were processed within this enode, |
1324 | and their index within the supernode. */ |
1325 | void |
1326 | exploded_node::dump_processed_stmts (pretty_printer *pp) const |
1327 | { |
1328 | if (m_num_processed_stmts > 0) |
1329 | { |
1330 | const program_point &point = get_point (); |
1331 | gcc_assert (point.get_kind () == PK_BEFORE_STMT); |
1332 | const supernode *snode = get_supernode (); |
1333 | const unsigned int point_stmt_idx = point.get_stmt_idx (); |
1334 | |
1335 | pp_printf (pp, "stmts: %i" , m_num_processed_stmts); |
1336 | pp_newline (pp); |
1337 | for (unsigned i = 0; i < m_num_processed_stmts; i++) |
1338 | { |
1339 | const unsigned int idx_within_snode = point_stmt_idx + i; |
1340 | const gimple *stmt = snode->m_stmts[idx_within_snode]; |
1341 | pp_printf (pp, " %i: " , idx_within_snode); |
1342 | pp_gimple_stmt_1 (pp, stmt, 0, (dump_flags_t)0); |
1343 | pp_newline (pp); |
1344 | } |
1345 | } |
1346 | } |
1347 | |
1348 | /* Dump any saved_diagnostics at this enode to PP. */ |
1349 | |
1350 | void |
1351 | exploded_node::dump_saved_diagnostics (pretty_printer *pp) const |
1352 | { |
1353 | unsigned i; |
1354 | const saved_diagnostic *sd; |
1355 | FOR_EACH_VEC_ELT (m_saved_diagnostics, i, sd) |
1356 | { |
1357 | pp_printf (pp, "DIAGNOSTIC: %s (sd: %i)" , |
1358 | sd->m_d->get_kind (), sd->get_index ()); |
1359 | pp_newline (pp); |
1360 | } |
1361 | } |
1362 | |
1363 | /* Dump this to PP in a form suitable for use as an id in .dot output. */ |
1364 | |
1365 | void |
1366 | exploded_node::dump_dot_id (pretty_printer *pp) const |
1367 | { |
1368 | pp_printf (pp, "exploded_node_%i" , m_index); |
1369 | } |
1370 | |
1371 | /* Dump a multiline representation of this node to PP. */ |
1372 | |
1373 | void |
1374 | exploded_node::dump_to_pp (pretty_printer *pp, |
1375 | const extrinsic_state &ext_state) const |
1376 | { |
1377 | pp_printf (pp, "EN: %i" , m_index); |
1378 | pp_newline (pp); |
1379 | |
1380 | format f (true); |
1381 | m_ps.get_point ().print (pp, f); |
1382 | pp_newline (pp); |
1383 | |
1384 | m_ps.get_state ().dump_to_pp (ext_state, simple: false, multiline: true, pp); |
1385 | pp_newline (pp); |
1386 | } |
1387 | |
1388 | /* Dump a multiline representation of this node to FILE. */ |
1389 | |
1390 | void |
1391 | exploded_node::dump (FILE *fp, |
1392 | const extrinsic_state &ext_state) const |
1393 | { |
1394 | pretty_printer pp; |
1395 | pp_format_decoder (&pp) = default_tree_printer; |
1396 | pp_show_color (&pp) = pp_show_color (global_dc->printer); |
1397 | pp.buffer->stream = fp; |
1398 | dump_to_pp (pp: &pp, ext_state); |
1399 | pp_flush (&pp); |
1400 | } |
1401 | |
1402 | /* Dump a multiline representation of this node to stderr. */ |
1403 | |
1404 | DEBUG_FUNCTION void |
1405 | exploded_node::dump (const extrinsic_state &ext_state) const |
1406 | { |
1407 | dump (stderr, ext_state); |
1408 | } |
1409 | |
1410 | /* Return a new json::object of the form |
1411 | {"point" : object for program_point, |
1412 | "state" : object for program_state, |
1413 | "status" : str, |
1414 | "idx" : int, |
1415 | "processed_stmts" : int}. */ |
1416 | |
1417 | json::object * |
1418 | exploded_node::to_json (const extrinsic_state &ext_state) const |
1419 | { |
1420 | json::object *enode_obj = new json::object (); |
1421 | |
1422 | enode_obj->set (key: "point" , v: get_point ().to_json ()); |
1423 | enode_obj->set (key: "state" , v: get_state ().to_json (ext_state)); |
1424 | enode_obj->set (key: "status" , v: new json::string (status_to_str (s: m_status))); |
1425 | enode_obj->set (key: "idx" , v: new json::integer_number (m_index)); |
1426 | enode_obj->set (key: "processed_stmts" , |
1427 | v: new json::integer_number (m_num_processed_stmts)); |
1428 | |
1429 | return enode_obj; |
1430 | } |
1431 | |
1432 | } // namespace ana |
1433 | |
1434 | /* Return true if FNDECL has a gimple body. */ |
1435 | // TODO: is there a pre-canned way to do this? |
1436 | |
1437 | bool |
1438 | fndecl_has_gimple_body_p (tree fndecl) |
1439 | { |
1440 | if (fndecl == NULL_TREE) |
1441 | return false; |
1442 | |
1443 | cgraph_node *n = cgraph_node::get (decl: fndecl); |
1444 | if (!n) |
1445 | return false; |
1446 | |
1447 | return n->has_gimple_body_p (); |
1448 | } |
1449 | |
1450 | namespace ana { |
1451 | |
1452 | /* Modify STATE in place, applying the effects of the stmt at this node's |
1453 | point. */ |
1454 | |
1455 | exploded_node::on_stmt_flags |
1456 | exploded_node::on_stmt (exploded_graph &eg, |
1457 | const supernode *snode, |
1458 | const gimple *stmt, |
1459 | program_state *state, |
1460 | uncertainty_t *uncertainty, |
1461 | bool *out_could_have_done_work, |
1462 | path_context *path_ctxt) |
1463 | { |
1464 | logger *logger = eg.get_logger (); |
1465 | LOG_SCOPE (logger); |
1466 | if (logger) |
1467 | { |
1468 | logger->start_log_line (); |
1469 | pp_gimple_stmt_1 (logger->get_printer (), stmt, 0, (dump_flags_t)0); |
1470 | logger->end_log_line (); |
1471 | } |
1472 | |
1473 | /* Update input_location in case of ICE: make it easier to track down which |
1474 | source construct we're failing to handle. */ |
1475 | input_location = stmt->location; |
1476 | |
1477 | gcc_assert (state->m_region_model); |
1478 | |
1479 | /* Preserve the old state. It is used here for looking |
1480 | up old checker states, for determining state transitions, and |
1481 | also within impl_region_model_context and impl_sm_context for |
1482 | going from tree to svalue_id. */ |
1483 | const program_state old_state (*state); |
1484 | |
1485 | impl_region_model_context ctxt (eg, this, |
1486 | &old_state, state, uncertainty, |
1487 | path_ctxt, stmt, nullptr, |
1488 | out_could_have_done_work); |
1489 | |
1490 | /* Handle call summaries here. */ |
1491 | if (cgraph_edge *cgedge |
1492 | = supergraph_call_edge (fun: snode->get_function (), stmt)) |
1493 | if (eg.get_analysis_plan ().use_summary_p (edge: cgedge)) |
1494 | { |
1495 | function *called_fn = get_ultimate_function_for_cgraph_edge (edge: cgedge); |
1496 | per_function_data *called_fn_data |
1497 | = eg.get_per_function_data (called_fn); |
1498 | if (called_fn_data) |
1499 | { |
1500 | gcc_assert (called_fn); |
1501 | return replay_call_summaries (eg, |
1502 | snode, |
1503 | call_stmt: as_a <const gcall *> (p: stmt), |
1504 | state, |
1505 | path_ctxt, |
1506 | called_fn: *called_fn, |
1507 | called_fn_data&: *called_fn_data, |
1508 | ctxt: &ctxt); |
1509 | } |
1510 | } |
1511 | |
1512 | bool unknown_side_effects = false; |
1513 | bool terminate_path = false; |
1514 | |
1515 | on_stmt_pre (eg, stmt, state, out_terminate_path: &terminate_path, |
1516 | out_unknown_side_effects: &unknown_side_effects, ctxt: &ctxt); |
1517 | |
1518 | if (terminate_path) |
1519 | return on_stmt_flags::terminate_path (); |
1520 | |
1521 | int sm_idx; |
1522 | sm_state_map *smap; |
1523 | FOR_EACH_VEC_ELT (old_state.m_checker_states, sm_idx, smap) |
1524 | { |
1525 | const state_machine &sm = eg.get_ext_state ().get_sm (idx: sm_idx); |
1526 | const sm_state_map *old_smap |
1527 | = old_state.m_checker_states[sm_idx]; |
1528 | sm_state_map *new_smap = state->m_checker_states[sm_idx]; |
1529 | impl_sm_context sm_ctxt (eg, sm_idx, sm, this, &old_state, state, |
1530 | old_smap, new_smap, path_ctxt, NULL, |
1531 | unknown_side_effects); |
1532 | |
1533 | /* Allow the state_machine to handle the stmt. */ |
1534 | if (sm.on_stmt (sm_ctxt: &sm_ctxt, node: snode, stmt)) |
1535 | unknown_side_effects = false; |
1536 | } |
1537 | |
1538 | if (path_ctxt->terminate_path_p ()) |
1539 | return on_stmt_flags::terminate_path (); |
1540 | |
1541 | on_stmt_post (stmt, state, unknown_side_effects, ctxt: &ctxt); |
1542 | |
1543 | return on_stmt_flags (); |
1544 | } |
1545 | |
1546 | /* Handle the pre-sm-state part of STMT, modifying STATE in-place. |
1547 | Write true to *OUT_TERMINATE_PATH if the path should be terminated. |
1548 | Write true to *OUT_UNKNOWN_SIDE_EFFECTS if the stmt has unknown |
1549 | side effects. */ |
1550 | |
1551 | void |
1552 | exploded_node::on_stmt_pre (exploded_graph &eg, |
1553 | const gimple *stmt, |
1554 | program_state *state, |
1555 | bool *out_terminate_path, |
1556 | bool *out_unknown_side_effects, |
1557 | region_model_context *ctxt) |
1558 | { |
1559 | /* Handle special-case calls that require the full program_state. */ |
1560 | if (const gcall *call = dyn_cast <const gcall *> (p: stmt)) |
1561 | { |
1562 | if (is_special_named_call_p (call, funcname: "__analyzer_dump" , num_args: 0)) |
1563 | { |
1564 | /* Handle the builtin "__analyzer_dump" by dumping state |
1565 | to stderr. */ |
1566 | state->dump (ext_state: eg.get_ext_state (), simple: true); |
1567 | return; |
1568 | } |
1569 | else if (is_special_named_call_p (call, funcname: "__analyzer_dump_state" , num_args: 2)) |
1570 | { |
1571 | state->impl_call_analyzer_dump_state (call, ext_state: eg.get_ext_state (), |
1572 | ctxt); |
1573 | return; |
1574 | } |
1575 | else if (is_setjmp_call_p (call)) |
1576 | { |
1577 | state->m_region_model->on_setjmp (stmt: call, enode: this, ctxt); |
1578 | if (ctxt) |
1579 | ctxt->maybe_did_work (); |
1580 | return; |
1581 | } |
1582 | else if (is_longjmp_call_p (call)) |
1583 | { |
1584 | on_longjmp (eg, call, new_state: state, ctxt); |
1585 | *out_terminate_path = true; |
1586 | if (ctxt) |
1587 | ctxt->maybe_did_work (); |
1588 | return; |
1589 | } |
1590 | } |
1591 | |
1592 | /* Otherwise, defer to m_region_model. */ |
1593 | state->m_region_model->on_stmt_pre (stmt, |
1594 | out_unknown_side_effects, |
1595 | ctxt); |
1596 | } |
1597 | |
1598 | /* Handle the post-sm-state part of STMT, modifying STATE in-place. */ |
1599 | |
1600 | void |
1601 | exploded_node::on_stmt_post (const gimple *stmt, |
1602 | program_state *state, |
1603 | bool unknown_side_effects, |
1604 | region_model_context *ctxt) |
1605 | { |
1606 | if (const gcall *call = dyn_cast <const gcall *> (p: stmt)) |
1607 | state->m_region_model->on_call_post (stmt: call, unknown_side_effects, ctxt); |
1608 | } |
1609 | |
1610 | /* A concrete call_info subclass representing a replay of a call summary. */ |
1611 | |
1612 | class call_summary_edge_info : public call_info |
1613 | { |
1614 | public: |
1615 | call_summary_edge_info (const call_details &cd, |
1616 | const function &called_fn, |
1617 | call_summary *summary, |
1618 | const extrinsic_state &ext_state) |
1619 | : call_info (cd, called_fn), |
1620 | m_called_fn (called_fn), |
1621 | m_summary (summary), |
1622 | m_ext_state (ext_state) |
1623 | {} |
1624 | |
1625 | bool update_state (program_state *state, |
1626 | const exploded_edge *, |
1627 | region_model_context *ctxt) const final override |
1628 | { |
1629 | /* Update STATE based on summary_end_state. */ |
1630 | call_details cd (get_call_details (model: state->m_region_model, ctxt)); |
1631 | call_summary_replay r (cd, m_called_fn, m_summary, m_ext_state); |
1632 | const program_state &summary_end_state = m_summary->get_state (); |
1633 | return state->replay_call_summary (r, summary: summary_end_state); |
1634 | } |
1635 | |
1636 | bool update_model (region_model *model, |
1637 | const exploded_edge *, |
1638 | region_model_context *ctxt) const final override |
1639 | { |
1640 | /* Update STATE based on summary_end_state. */ |
1641 | call_details cd (get_call_details (model, ctxt)); |
1642 | call_summary_replay r (cd, m_called_fn, m_summary, m_ext_state); |
1643 | const program_state &summary_end_state = m_summary->get_state (); |
1644 | model->replay_call_summary (r, summary: *summary_end_state.m_region_model); |
1645 | return true; |
1646 | } |
1647 | |
1648 | label_text get_desc (bool /*can_colorize*/) const final override |
1649 | { |
1650 | return m_summary->get_desc (); |
1651 | } |
1652 | |
1653 | private: |
1654 | const function &m_called_fn; |
1655 | call_summary *m_summary; |
1656 | const extrinsic_state &m_ext_state; |
1657 | }; |
1658 | |
1659 | /* Use PATH_CTXT to bifurcate, which when handled will add custom edges |
1660 | for a replay of the various feasible summaries in CALLED_FN_DATA. */ |
1661 | |
1662 | exploded_node::on_stmt_flags |
1663 | exploded_node::replay_call_summaries (exploded_graph &eg, |
1664 | const supernode *snode, |
1665 | const gcall *call_stmt, |
1666 | program_state *state, |
1667 | path_context *path_ctxt, |
1668 | const function &called_fn, |
1669 | per_function_data &called_fn_data, |
1670 | region_model_context *ctxt) |
1671 | { |
1672 | logger *logger = eg.get_logger (); |
1673 | LOG_SCOPE (logger); |
1674 | |
1675 | /* Each summary will call bifurcate on the PATH_CTXT. */ |
1676 | for (auto summary : called_fn_data.m_summaries) |
1677 | replay_call_summary (eg, snode, call_stmt, state, |
1678 | path_ctxt, called_fn, summary, ctxt); |
1679 | path_ctxt->terminate_path (); |
1680 | |
1681 | return on_stmt_flags (); |
1682 | } |
1683 | |
1684 | /* Use PATH_CTXT to bifurcate, which when handled will add a |
1685 | custom edge for a replay of SUMMARY, if the summary's |
1686 | conditions are feasible based on the current state. */ |
1687 | |
1688 | void |
1689 | exploded_node::replay_call_summary (exploded_graph &eg, |
1690 | const supernode *snode, |
1691 | const gcall *call_stmt, |
1692 | program_state *old_state, |
1693 | path_context *path_ctxt, |
1694 | const function &called_fn, |
1695 | call_summary *summary, |
1696 | region_model_context *ctxt) |
1697 | { |
1698 | logger *logger = eg.get_logger (); |
1699 | LOG_SCOPE (logger); |
1700 | gcc_assert (snode); |
1701 | gcc_assert (call_stmt); |
1702 | gcc_assert (old_state); |
1703 | gcc_assert (summary); |
1704 | |
1705 | if (logger) |
1706 | logger->log (fmt: "using %s as summary for call to %qE from %qE" , |
1707 | summary->get_desc ().get (), |
1708 | called_fn.decl, |
1709 | snode->get_function ()->decl); |
1710 | const extrinsic_state &ext_state = eg.get_ext_state (); |
1711 | const program_state &summary_end_state = summary->get_state (); |
1712 | if (logger) |
1713 | { |
1714 | pretty_printer *pp = logger->get_printer (); |
1715 | |
1716 | logger->start_log_line (); |
1717 | pp_string (pp, "callsite state: " ); |
1718 | old_state->dump_to_pp (ext_state, simple: true, multiline: false, pp); |
1719 | logger->end_log_line (); |
1720 | |
1721 | logger->start_log_line (); |
1722 | pp_string (pp, "summary end state: " ); |
1723 | summary_end_state.dump_to_pp (ext_state, simple: true, multiline: false, pp); |
1724 | logger->end_log_line (); |
1725 | } |
1726 | |
1727 | program_state new_state (*old_state); |
1728 | |
1729 | call_details cd (call_stmt, new_state.m_region_model, ctxt); |
1730 | call_summary_replay r (cd, called_fn, summary, ext_state); |
1731 | |
1732 | if (path_ctxt) |
1733 | path_ctxt->bifurcate (info: make_unique<call_summary_edge_info> (args&: cd, |
1734 | args: called_fn, |
1735 | args&: summary, |
1736 | args: ext_state)); |
1737 | } |
1738 | |
1739 | |
1740 | /* Consider the effect of following superedge SUCC from this node. |
1741 | |
1742 | Return true if it's feasible to follow the edge, or false |
1743 | if it's infeasible. |
1744 | |
1745 | Examples: if it's the "true" branch within |
1746 | a CFG and we know the conditional is false, we know it's infeasible. |
1747 | If it's one of multiple interprocedual "return" edges, then only |
1748 | the edge back to the most recent callsite is feasible. |
1749 | |
1750 | Update NEXT_STATE accordingly (e.g. to record that a condition was |
1751 | true or false, or that the NULL-ness of a pointer has been checked, |
1752 | pushing/popping stack frames, etc). |
1753 | |
1754 | Update NEXT_POINT accordingly (updating the call string). */ |
1755 | |
1756 | bool |
1757 | exploded_node::on_edge (exploded_graph &eg, |
1758 | const superedge *succ, |
1759 | program_point *next_point, |
1760 | program_state *next_state, |
1761 | uncertainty_t *uncertainty) |
1762 | { |
1763 | LOG_FUNC (eg.get_logger ()); |
1764 | |
1765 | if (!next_point->on_edge (eg, succ)) |
1766 | return false; |
1767 | |
1768 | if (!next_state->on_edge (eg, enode: this, succ, uncertainty)) |
1769 | return false; |
1770 | |
1771 | return true; |
1772 | } |
1773 | |
1774 | /* Verify that the stack at LONGJMP_POINT is still valid, given a call |
1775 | to "setjmp" at SETJMP_POINT - the stack frame that "setjmp" was |
1776 | called in must still be valid. |
1777 | |
1778 | Caveat: this merely checks the call_strings in the points; it doesn't |
1779 | detect the case where a frame returns and is then called again. */ |
1780 | |
1781 | static bool |
1782 | valid_longjmp_stack_p (const program_point &longjmp_point, |
1783 | const program_point &setjmp_point) |
1784 | { |
1785 | const call_string &cs_at_longjmp = longjmp_point.get_call_string (); |
1786 | const call_string &cs_at_setjmp = setjmp_point.get_call_string (); |
1787 | |
1788 | if (cs_at_longjmp.length () < cs_at_setjmp.length ()) |
1789 | return false; |
1790 | |
1791 | /* Check that the call strings match, up to the depth of the |
1792 | setjmp point. */ |
1793 | for (unsigned depth = 0; depth < cs_at_setjmp.length (); depth++) |
1794 | if (cs_at_longjmp[depth] != cs_at_setjmp[depth]) |
1795 | return false; |
1796 | |
1797 | return true; |
1798 | } |
1799 | |
1800 | /* A pending_diagnostic subclass for complaining about bad longjmps, |
1801 | where the enclosing function of the "setjmp" has returned (and thus |
1802 | the stack frame no longer exists). */ |
1803 | |
1804 | class stale_jmp_buf : public pending_diagnostic_subclass<stale_jmp_buf> |
1805 | { |
1806 | public: |
1807 | stale_jmp_buf (const gcall *setjmp_call, const gcall *longjmp_call, |
1808 | const program_point &setjmp_point) |
1809 | : m_setjmp_call (setjmp_call), m_longjmp_call (longjmp_call), |
1810 | m_setjmp_point (setjmp_point), m_stack_pop_event (NULL) |
1811 | {} |
1812 | |
1813 | int get_controlling_option () const final override |
1814 | { |
1815 | return OPT_Wanalyzer_stale_setjmp_buffer; |
1816 | } |
1817 | |
1818 | bool emit (diagnostic_emission_context &ctxt) final override |
1819 | { |
1820 | return ctxt.warn ("%qs called after enclosing function of %qs has returned" , |
1821 | get_user_facing_name (call: m_longjmp_call), |
1822 | get_user_facing_name (call: m_setjmp_call)); |
1823 | } |
1824 | |
1825 | const char *get_kind () const final override |
1826 | { return "stale_jmp_buf" ; } |
1827 | |
1828 | bool operator== (const stale_jmp_buf &other) const |
1829 | { |
1830 | return (m_setjmp_call == other.m_setjmp_call |
1831 | && m_longjmp_call == other.m_longjmp_call); |
1832 | } |
1833 | |
1834 | bool |
1835 | maybe_add_custom_events_for_superedge (const exploded_edge &eedge, |
1836 | checker_path *emission_path) |
1837 | final override |
1838 | { |
1839 | /* Detect exactly when the stack first becomes invalid, |
1840 | and issue an event then. */ |
1841 | if (m_stack_pop_event) |
1842 | return false; |
1843 | const exploded_node *src_node = eedge.m_src; |
1844 | const program_point &src_point = src_node->get_point (); |
1845 | const exploded_node *dst_node = eedge.m_dest; |
1846 | const program_point &dst_point = dst_node->get_point (); |
1847 | if (valid_longjmp_stack_p (longjmp_point: src_point, setjmp_point: m_setjmp_point) |
1848 | && !valid_longjmp_stack_p (longjmp_point: dst_point, setjmp_point: m_setjmp_point)) |
1849 | { |
1850 | /* Compare with diagnostic_manager::add_events_for_superedge. */ |
1851 | const int src_stack_depth = src_point.get_stack_depth (); |
1852 | m_stack_pop_event = new precanned_custom_event |
1853 | (event_loc_info (src_point.get_location (), |
1854 | src_point.get_fndecl (), |
1855 | src_stack_depth), |
1856 | "stack frame is popped here, invalidating saved environment" ); |
1857 | emission_path->add_event |
1858 | (event: std::unique_ptr<custom_event> (m_stack_pop_event)); |
1859 | return false; |
1860 | } |
1861 | return false; |
1862 | } |
1863 | |
1864 | label_text describe_final_event (const evdesc::final_event &ev) final override |
1865 | { |
1866 | if (m_stack_pop_event) |
1867 | return ev.formatted_print |
1868 | (fmt: "%qs called after enclosing function of %qs returned at %@" , |
1869 | get_user_facing_name (call: m_longjmp_call), |
1870 | get_user_facing_name (call: m_setjmp_call), |
1871 | m_stack_pop_event->get_id_ptr ()); |
1872 | else |
1873 | return ev.formatted_print |
1874 | (fmt: "%qs called after enclosing function of %qs has returned" , |
1875 | get_user_facing_name (call: m_longjmp_call), |
1876 | get_user_facing_name (call: m_setjmp_call));; |
1877 | } |
1878 | |
1879 | |
1880 | private: |
1881 | const gcall *m_setjmp_call; |
1882 | const gcall *m_longjmp_call; |
1883 | program_point m_setjmp_point; |
1884 | custom_event *m_stack_pop_event; |
1885 | }; |
1886 | |
1887 | /* Handle LONGJMP_CALL, a call to longjmp or siglongjmp. |
1888 | |
1889 | Attempt to locate where setjmp/sigsetjmp was called on the jmp_buf and build |
1890 | an exploded_node and exploded_edge to it representing a rewind to that frame, |
1891 | handling the various kinds of failure that can occur. */ |
1892 | |
1893 | void |
1894 | exploded_node::on_longjmp (exploded_graph &eg, |
1895 | const gcall *longjmp_call, |
1896 | program_state *new_state, |
1897 | region_model_context *ctxt) |
1898 | { |
1899 | tree buf_ptr = gimple_call_arg (gs: longjmp_call, index: 0); |
1900 | gcc_assert (POINTER_TYPE_P (TREE_TYPE (buf_ptr))); |
1901 | |
1902 | region_model *new_region_model = new_state->m_region_model; |
1903 | const svalue *buf_ptr_sval = new_region_model->get_rvalue (expr: buf_ptr, ctxt); |
1904 | const region *buf = new_region_model->deref_rvalue (ptr_sval: buf_ptr_sval, ptr_tree: buf_ptr, |
1905 | ctxt); |
1906 | |
1907 | const svalue *buf_content_sval |
1908 | = new_region_model->get_store_value (reg: buf, ctxt); |
1909 | const setjmp_svalue *setjmp_sval |
1910 | = buf_content_sval->dyn_cast_setjmp_svalue (); |
1911 | if (!setjmp_sval) |
1912 | return; |
1913 | |
1914 | const setjmp_record tmp_setjmp_record = setjmp_sval->get_setjmp_record (); |
1915 | |
1916 | /* Build a custom enode and eedge for rewinding from the longjmp/siglongjmp |
1917 | call back to the setjmp/sigsetjmp. */ |
1918 | rewind_info_t rewind_info (tmp_setjmp_record, longjmp_call); |
1919 | |
1920 | const gcall *setjmp_call = rewind_info.get_setjmp_call (); |
1921 | const program_point &setjmp_point = rewind_info.get_setjmp_point (); |
1922 | |
1923 | const program_point &longjmp_point = get_point (); |
1924 | |
1925 | /* Verify that the setjmp's call_stack hasn't been popped. */ |
1926 | if (!valid_longjmp_stack_p (longjmp_point, setjmp_point)) |
1927 | { |
1928 | ctxt->warn (d: make_unique<stale_jmp_buf> (args&: setjmp_call, |
1929 | args&: longjmp_call, |
1930 | args: setjmp_point)); |
1931 | return; |
1932 | } |
1933 | |
1934 | gcc_assert (longjmp_point.get_stack_depth () |
1935 | >= setjmp_point.get_stack_depth ()); |
1936 | |
1937 | /* Update the state for use by the destination node. */ |
1938 | |
1939 | /* Stash the current number of diagnostics so that we can update |
1940 | any that this adds to show where the longjmp is rewinding to. */ |
1941 | |
1942 | diagnostic_manager *dm = &eg.get_diagnostic_manager (); |
1943 | unsigned prev_num_diagnostics = dm->get_num_diagnostics (); |
1944 | |
1945 | new_region_model->on_longjmp (longjmp_call, setjmp_call, |
1946 | setjmp_stack_depth: setjmp_point.get_stack_depth (), ctxt); |
1947 | |
1948 | /* Detect leaks in the new state relative to the old state. */ |
1949 | program_state::detect_leaks (src_state: get_state (), dest_state: *new_state, NULL, |
1950 | ext_state: eg.get_ext_state (), ctxt); |
1951 | |
1952 | program_point next_point |
1953 | = program_point::after_supernode (supernode: setjmp_point.get_supernode (), |
1954 | call_string: setjmp_point.get_call_string ()); |
1955 | |
1956 | exploded_node *next |
1957 | = eg.get_or_create_node (point: next_point, state: *new_state, enode_for_diag: this); |
1958 | |
1959 | /* Create custom exploded_edge for a longjmp. */ |
1960 | if (next) |
1961 | { |
1962 | exploded_edge *eedge |
1963 | = eg.add_edge (src: const_cast<exploded_node *> (this), dest: next, NULL, could_do_work: true, |
1964 | custom: make_unique<rewind_info_t> (args: tmp_setjmp_record, |
1965 | args&: longjmp_call)); |
1966 | |
1967 | /* For any diagnostics that were queued here (such as leaks) we want |
1968 | the checker_path to show the rewinding events after the "final event" |
1969 | so that the user sees where the longjmp is rewinding to (otherwise the |
1970 | path is meaningless). |
1971 | |
1972 | For example, we want to emit something like: |
1973 | | NN | { |
1974 | | NN | longjmp (env, 1); |
1975 | | | ~~~~~~~~~~~~~~~~ |
1976 | | | | |
1977 | | | (10) 'ptr' leaks here; was allocated at (7) |
1978 | | | (11) rewinding from 'longjmp' in 'inner'... |
1979 | | |
1980 | <-------------+ |
1981 | | |
1982 | 'outer': event 12 |
1983 | | |
1984 | | NN | i = setjmp(env); |
1985 | | | ^~~~~~ |
1986 | | | | |
1987 | | | (12) ...to 'setjmp' in 'outer' (saved at (2)) |
1988 | |
1989 | where the "final" event above is event (10), but we want to append |
1990 | events (11) and (12) afterwards. |
1991 | |
1992 | Do this by setting m_trailing_eedge on any diagnostics that were |
1993 | just saved. */ |
1994 | unsigned num_diagnostics = dm->get_num_diagnostics (); |
1995 | for (unsigned i = prev_num_diagnostics; i < num_diagnostics; i++) |
1996 | { |
1997 | saved_diagnostic *sd = dm->get_saved_diagnostic (idx: i); |
1998 | sd->m_trailing_eedge = eedge; |
1999 | } |
2000 | } |
2001 | } |
2002 | |
2003 | /* Subroutine of exploded_graph::process_node for finding the successors |
2004 | of the supernode for a function exit basic block. |
2005 | |
2006 | Ensure that pop_frame is called, potentially queuing diagnostics about |
2007 | leaks. */ |
2008 | |
2009 | void |
2010 | exploded_node::detect_leaks (exploded_graph &eg) |
2011 | { |
2012 | LOG_FUNC_1 (eg.get_logger (), "EN: %i" , m_index); |
2013 | |
2014 | gcc_assert (get_point ().get_supernode ()->return_p ()); |
2015 | |
2016 | /* If we're not a "top-level" function, do nothing; pop_frame |
2017 | will be called when handling the return superedge. */ |
2018 | if (get_point ().get_stack_depth () > 1) |
2019 | return; |
2020 | |
2021 | /* We have a "top-level" function. */ |
2022 | gcc_assert (get_point ().get_stack_depth () == 1); |
2023 | |
2024 | const program_state &old_state = get_state (); |
2025 | |
2026 | /* Work with a temporary copy of the state: pop the frame, and see |
2027 | what leaks (via purge_unused_svalues). */ |
2028 | program_state new_state (old_state); |
2029 | |
2030 | gcc_assert (new_state.m_region_model); |
2031 | |
2032 | uncertainty_t uncertainty; |
2033 | impl_region_model_context ctxt (eg, this, |
2034 | &old_state, &new_state, &uncertainty, NULL, |
2035 | get_stmt ()); |
2036 | const svalue *result = NULL; |
2037 | new_state.m_region_model->pop_frame (NULL, out_result: &result, ctxt: &ctxt); |
2038 | program_state::detect_leaks (src_state: old_state, dest_state: new_state, extra_sval: result, |
2039 | ext_state: eg.get_ext_state (), ctxt: &ctxt); |
2040 | } |
2041 | |
2042 | /* Dump the successors and predecessors of this enode to OUTF. */ |
2043 | |
2044 | void |
2045 | exploded_node::dump_succs_and_preds (FILE *outf) const |
2046 | { |
2047 | unsigned i; |
2048 | exploded_edge *e; |
2049 | { |
2050 | auto_vec<exploded_node *> preds (m_preds.length ()); |
2051 | FOR_EACH_VEC_ELT (m_preds, i, e) |
2052 | preds.quick_push (obj: e->m_src); |
2053 | pretty_printer pp; |
2054 | print_enode_indices (pp: &pp, enodes: preds); |
2055 | fprintf (stream: outf, format: "preds: %s\n" , |
2056 | pp_formatted_text (&pp)); |
2057 | } |
2058 | { |
2059 | auto_vec<exploded_node *> succs (m_succs.length ()); |
2060 | FOR_EACH_VEC_ELT (m_succs, i, e) |
2061 | succs.quick_push (obj: e->m_dest); |
2062 | pretty_printer pp; |
2063 | print_enode_indices (pp: &pp, enodes: succs); |
2064 | fprintf (stream: outf, format: "succs: %s\n" , |
2065 | pp_formatted_text (&pp)); |
2066 | } |
2067 | } |
2068 | |
2069 | /* class dynamic_call_info_t : public custom_edge_info. */ |
2070 | |
2071 | /* Implementation of custom_edge_info::update_model vfunc |
2072 | for dynamic_call_info_t. |
2073 | |
2074 | Update state for a dynamically discovered call (or return), by pushing |
2075 | or popping the a frame for the appropriate function. */ |
2076 | |
2077 | bool |
2078 | dynamic_call_info_t::update_model (region_model *model, |
2079 | const exploded_edge *eedge, |
2080 | region_model_context *ctxt) const |
2081 | { |
2082 | gcc_assert (eedge); |
2083 | if (m_is_returning_call) |
2084 | model->update_for_return_gcall (call_stmt: m_dynamic_call, ctxt); |
2085 | else |
2086 | { |
2087 | function *callee = eedge->m_dest->get_function (); |
2088 | model->update_for_gcall (call_stmt: m_dynamic_call, ctxt, callee); |
2089 | } |
2090 | return true; |
2091 | } |
2092 | |
2093 | /* Implementation of custom_edge_info::add_events_to_path vfunc |
2094 | for dynamic_call_info_t. */ |
2095 | |
2096 | void |
2097 | dynamic_call_info_t::add_events_to_path (checker_path *emission_path, |
2098 | const exploded_edge &eedge) const |
2099 | { |
2100 | const exploded_node *src_node = eedge.m_src; |
2101 | const program_point &src_point = src_node->get_point (); |
2102 | const int src_stack_depth = src_point.get_stack_depth (); |
2103 | const exploded_node *dest_node = eedge.m_dest; |
2104 | const program_point &dest_point = dest_node->get_point (); |
2105 | const int dest_stack_depth = dest_point.get_stack_depth (); |
2106 | |
2107 | if (m_is_returning_call) |
2108 | emission_path->add_event |
2109 | (event: make_unique<return_event> (args: eedge, |
2110 | args: event_loc_info (m_dynamic_call |
2111 | ? m_dynamic_call->location |
2112 | : UNKNOWN_LOCATION, |
2113 | dest_point.get_fndecl (), |
2114 | dest_stack_depth))); |
2115 | else |
2116 | emission_path->add_event |
2117 | (event: make_unique<call_event> (args: eedge, |
2118 | args: event_loc_info (m_dynamic_call |
2119 | ? m_dynamic_call->location |
2120 | : UNKNOWN_LOCATION, |
2121 | src_point.get_fndecl (), |
2122 | src_stack_depth))); |
2123 | } |
2124 | |
2125 | /* class rewind_info_t : public custom_edge_info. */ |
2126 | |
2127 | /* Implementation of custom_edge_info::update_model vfunc |
2128 | for rewind_info_t. |
2129 | |
2130 | Update state for the special-case of a rewind of a longjmp |
2131 | to a setjmp (which doesn't have a superedge, but does affect |
2132 | state). */ |
2133 | |
2134 | bool |
2135 | rewind_info_t::update_model (region_model *model, |
2136 | const exploded_edge *eedge, |
2137 | region_model_context *) const |
2138 | { |
2139 | gcc_assert (eedge); |
2140 | const program_point &longjmp_point = eedge->m_src->get_point (); |
2141 | const program_point &setjmp_point = eedge->m_dest->get_point (); |
2142 | |
2143 | gcc_assert (longjmp_point.get_stack_depth () |
2144 | >= setjmp_point.get_stack_depth ()); |
2145 | |
2146 | model->on_longjmp (longjmp_call: get_longjmp_call (), |
2147 | setjmp_call: get_setjmp_call (), |
2148 | setjmp_stack_depth: setjmp_point.get_stack_depth (), NULL); |
2149 | return true; |
2150 | } |
2151 | |
2152 | /* Implementation of custom_edge_info::add_events_to_path vfunc |
2153 | for rewind_info_t. */ |
2154 | |
2155 | void |
2156 | rewind_info_t::add_events_to_path (checker_path *emission_path, |
2157 | const exploded_edge &eedge) const |
2158 | { |
2159 | const exploded_node *src_node = eedge.m_src; |
2160 | const program_point &src_point = src_node->get_point (); |
2161 | const int src_stack_depth = src_point.get_stack_depth (); |
2162 | const exploded_node *dst_node = eedge.m_dest; |
2163 | const program_point &dst_point = dst_node->get_point (); |
2164 | const int dst_stack_depth = dst_point.get_stack_depth (); |
2165 | |
2166 | emission_path->add_event |
2167 | (event: make_unique<rewind_from_longjmp_event> |
2168 | (args: &eedge, |
2169 | args: event_loc_info (get_longjmp_call ()->location, |
2170 | src_point.get_fndecl (), |
2171 | src_stack_depth), |
2172 | args: this)); |
2173 | emission_path->add_event |
2174 | (event: make_unique<rewind_to_setjmp_event> |
2175 | (args: &eedge, |
2176 | args: event_loc_info (get_setjmp_call ()->location, |
2177 | dst_point.get_fndecl (), |
2178 | dst_stack_depth), |
2179 | args: this)); |
2180 | } |
2181 | |
2182 | /* class exploded_edge : public dedge<eg_traits>. */ |
2183 | |
2184 | /* exploded_edge's ctor. */ |
2185 | |
2186 | exploded_edge::exploded_edge (exploded_node *src, exploded_node *dest, |
2187 | const superedge *sedge, bool could_do_work, |
2188 | std::unique_ptr<custom_edge_info> custom_info) |
2189 | : dedge<eg_traits> (src, dest), m_sedge (sedge), |
2190 | m_custom_info (std::move (custom_info)), |
2191 | m_could_do_work_p (could_do_work) |
2192 | { |
2193 | } |
2194 | |
2195 | /* Implementation of dedge::dump_dot vfunc for exploded_edge. |
2196 | Use the label of the underlying superedge, if any. */ |
2197 | |
2198 | void |
2199 | exploded_edge::dump_dot (graphviz_out *gv, const dump_args_t &) const |
2200 | { |
2201 | pretty_printer *pp = gv->get_pp (); |
2202 | |
2203 | m_src->dump_dot_id (pp); |
2204 | pp_string (pp, " -> " ); |
2205 | m_dest->dump_dot_id (pp); |
2206 | dump_dot_label (pp); |
2207 | } |
2208 | |
2209 | /* Second half of exploded_edge::dump_dot. This is split out |
2210 | for use by trimmed_graph::dump_dot and base_feasible_edge::dump_dot. */ |
2211 | |
2212 | void |
2213 | exploded_edge::dump_dot_label (pretty_printer *pp) const |
2214 | { |
2215 | const char *style = "\"solid,bold\"" ; |
2216 | const char *color = "black" ; |
2217 | int weight = 10; |
2218 | const char *constraint = "true" ; |
2219 | |
2220 | if (m_sedge) |
2221 | switch (m_sedge->m_kind) |
2222 | { |
2223 | default: |
2224 | gcc_unreachable (); |
2225 | case SUPEREDGE_CFG_EDGE: |
2226 | break; |
2227 | case SUPEREDGE_CALL: |
2228 | color = "red" ; |
2229 | //constraint = "false"; |
2230 | break; |
2231 | case SUPEREDGE_RETURN: |
2232 | color = "green" ; |
2233 | //constraint = "false"; |
2234 | break; |
2235 | case SUPEREDGE_INTRAPROCEDURAL_CALL: |
2236 | style = "\"dotted\"" ; |
2237 | break; |
2238 | } |
2239 | if (m_custom_info) |
2240 | { |
2241 | color = "red" ; |
2242 | style = "\"dotted\"" ; |
2243 | } |
2244 | |
2245 | pp_printf (pp, |
2246 | (" [style=%s, color=%s, weight=%d, constraint=%s," |
2247 | " headlabel=\"" ), |
2248 | style, color, weight, constraint); |
2249 | |
2250 | if (m_sedge) |
2251 | m_sedge->dump_label_to_pp (pp, user_facing: false); |
2252 | else if (m_custom_info) |
2253 | m_custom_info->print (pp); |
2254 | |
2255 | pp_printf (pp, "%s" , |
2256 | could_do_work_p () ? "(could do work)" : "DOES NO WORK" ); |
2257 | |
2258 | //pp_write_text_as_dot_label_to_stream (pp, /*for_record=*/false); |
2259 | |
2260 | pp_printf (pp, "\"];\n" ); |
2261 | } |
2262 | |
2263 | /* Return a new json::object of the form |
2264 | {"src_idx": int, the index of the source exploded edge, |
2265 | "dst_idx": int, the index of the destination exploded edge, |
2266 | "sedge": (optional) object for the superedge, if any, |
2267 | "custom": (optional) str, a description, if this is a custom edge}. */ |
2268 | |
2269 | json::object * |
2270 | exploded_edge::to_json () const |
2271 | { |
2272 | json::object *eedge_obj = new json::object (); |
2273 | eedge_obj->set (key: "src_idx" , v: new json::integer_number (m_src->m_index)); |
2274 | eedge_obj->set (key: "dst_idx" , v: new json::integer_number (m_dest->m_index)); |
2275 | if (m_sedge) |
2276 | eedge_obj->set (key: "sedge" , v: m_sedge->to_json ()); |
2277 | if (m_custom_info) |
2278 | { |
2279 | pretty_printer pp; |
2280 | pp_format_decoder (&pp) = default_tree_printer; |
2281 | m_custom_info->print (pp: &pp); |
2282 | eedge_obj->set (key: "custom" , v: new json::string (pp_formatted_text (&pp))); |
2283 | } |
2284 | return eedge_obj; |
2285 | } |
2286 | |
2287 | /* struct stats. */ |
2288 | |
2289 | /* stats' ctor. */ |
2290 | |
2291 | stats::stats (int num_supernodes) |
2292 | : m_node_reuse_count (0), |
2293 | m_node_reuse_after_merge_count (0), |
2294 | m_num_supernodes (num_supernodes) |
2295 | { |
2296 | for (int i = 0; i < NUM_POINT_KINDS; i++) |
2297 | m_num_nodes[i] = 0; |
2298 | } |
2299 | |
2300 | /* Log these stats in multiline form to LOGGER. */ |
2301 | |
2302 | void |
2303 | stats::log (logger *logger) const |
2304 | { |
2305 | gcc_assert (logger); |
2306 | for (int i = 0; i < NUM_POINT_KINDS; i++) |
2307 | if (m_num_nodes[i] > 0) |
2308 | logger->log (fmt: "m_num_nodes[%s]: %i" , |
2309 | point_kind_to_string (pk: static_cast <enum point_kind> (i)), |
2310 | m_num_nodes[i]); |
2311 | logger->log (fmt: "m_node_reuse_count: %i" , m_node_reuse_count); |
2312 | logger->log (fmt: "m_node_reuse_after_merge_count: %i" , |
2313 | m_node_reuse_after_merge_count); |
2314 | } |
2315 | |
2316 | /* Dump these stats in multiline form to OUT. */ |
2317 | |
2318 | void |
2319 | stats::dump (FILE *out) const |
2320 | { |
2321 | for (int i = 0; i < NUM_POINT_KINDS; i++) |
2322 | if (m_num_nodes[i] > 0) |
2323 | fprintf (stream: out, format: "m_num_nodes[%s]: %i\n" , |
2324 | point_kind_to_string (pk: static_cast <enum point_kind> (i)), |
2325 | m_num_nodes[i]); |
2326 | fprintf (stream: out, format: "m_node_reuse_count: %i\n" , m_node_reuse_count); |
2327 | fprintf (stream: out, format: "m_node_reuse_after_merge_count: %i\n" , |
2328 | m_node_reuse_after_merge_count); |
2329 | |
2330 | if (m_num_supernodes > 0) |
2331 | fprintf (stream: out, format: "PK_AFTER_SUPERNODE nodes per supernode: %.2f\n" , |
2332 | (float)m_num_nodes[PK_AFTER_SUPERNODE] / (float)m_num_supernodes); |
2333 | } |
2334 | |
2335 | /* Return the total number of enodes recorded within this object. */ |
2336 | |
2337 | int |
2338 | stats::get_total_enodes () const |
2339 | { |
2340 | int result = 0; |
2341 | for (int i = 0; i < NUM_POINT_KINDS; i++) |
2342 | result += m_num_nodes[i]; |
2343 | return result; |
2344 | } |
2345 | |
2346 | /* struct per_function_data. */ |
2347 | |
2348 | per_function_data::~per_function_data () |
2349 | { |
2350 | for (auto iter : m_summaries) |
2351 | delete iter; |
2352 | } |
2353 | |
2354 | void |
2355 | per_function_data::add_call_summary (exploded_node *node) |
2356 | { |
2357 | m_summaries.safe_push (obj: new call_summary (this, node)); |
2358 | } |
2359 | |
2360 | /* strongly_connected_components's ctor. Tarjan's SCC algorithm. */ |
2361 | |
2362 | strongly_connected_components:: |
2363 | strongly_connected_components (const supergraph &sg, logger *logger) |
2364 | : m_sg (sg), m_per_node (m_sg.num_nodes ()) |
2365 | { |
2366 | LOG_SCOPE (logger); |
2367 | auto_timevar tv (TV_ANALYZER_SCC); |
2368 | |
2369 | for (int i = 0; i < m_sg.num_nodes (); i++) |
2370 | m_per_node.quick_push (obj: per_node_data ()); |
2371 | |
2372 | for (int i = 0; i < m_sg.num_nodes (); i++) |
2373 | if (m_per_node[i].m_index == -1) |
2374 | strong_connect (index: i); |
2375 | |
2376 | if (0) |
2377 | dump (); |
2378 | } |
2379 | |
2380 | /* Dump this object to stderr. */ |
2381 | |
2382 | DEBUG_FUNCTION void |
2383 | strongly_connected_components::dump () const |
2384 | { |
2385 | for (int i = 0; i < m_sg.num_nodes (); i++) |
2386 | { |
2387 | const per_node_data &v = m_per_node[i]; |
2388 | fprintf (stderr, format: "SN %i: index: %i lowlink: %i on_stack: %i\n" , |
2389 | i, v.m_index, v.m_lowlink, v.m_on_stack); |
2390 | } |
2391 | } |
2392 | |
2393 | /* Return a new json::array of per-snode SCC ids. */ |
2394 | |
2395 | json::array * |
2396 | strongly_connected_components::to_json () const |
2397 | { |
2398 | json::array *scc_arr = new json::array (); |
2399 | for (int i = 0; i < m_sg.num_nodes (); i++) |
2400 | scc_arr->append (v: new json::integer_number (get_scc_id (node_index: i))); |
2401 | return scc_arr; |
2402 | } |
2403 | |
2404 | /* Subroutine of strongly_connected_components's ctor, part of Tarjan's |
2405 | SCC algorithm. */ |
2406 | |
2407 | void |
2408 | strongly_connected_components::strong_connect (unsigned index) |
2409 | { |
2410 | supernode *v_snode = m_sg.get_node_by_index (idx: index); |
2411 | |
2412 | /* Set the depth index for v to the smallest unused index. */ |
2413 | per_node_data *v = &m_per_node[index]; |
2414 | v->m_index = index; |
2415 | v->m_lowlink = index; |
2416 | m_stack.safe_push (obj: index); |
2417 | v->m_on_stack = true; |
2418 | index++; |
2419 | |
2420 | /* Consider successors of v. */ |
2421 | unsigned i; |
2422 | superedge *sedge; |
2423 | FOR_EACH_VEC_ELT (v_snode->m_succs, i, sedge) |
2424 | { |
2425 | if (sedge->get_kind () != SUPEREDGE_CFG_EDGE |
2426 | && sedge->get_kind () != SUPEREDGE_INTRAPROCEDURAL_CALL) |
2427 | continue; |
2428 | supernode *w_snode = sedge->m_dest; |
2429 | per_node_data *w = &m_per_node[w_snode->m_index]; |
2430 | if (w->m_index == -1) |
2431 | { |
2432 | /* Successor w has not yet been visited; recurse on it. */ |
2433 | strong_connect (index: w_snode->m_index); |
2434 | v->m_lowlink = MIN (v->m_lowlink, w->m_lowlink); |
2435 | } |
2436 | else if (w->m_on_stack) |
2437 | { |
2438 | /* Successor w is in stack S and hence in the current SCC |
2439 | If w is not on stack, then (v, w) is a cross-edge in the DFS |
2440 | tree and must be ignored. */ |
2441 | v->m_lowlink = MIN (v->m_lowlink, w->m_index); |
2442 | } |
2443 | } |
2444 | |
2445 | /* If v is a root node, pop the stack and generate an SCC. */ |
2446 | |
2447 | if (v->m_lowlink == v->m_index) |
2448 | { |
2449 | per_node_data *w; |
2450 | do { |
2451 | int idx = m_stack.pop (); |
2452 | w = &m_per_node[idx]; |
2453 | w->m_on_stack = false; |
2454 | } while (w != v); |
2455 | } |
2456 | } |
2457 | |
2458 | /* worklist's ctor. */ |
2459 | |
2460 | worklist::worklist (const exploded_graph &eg, const analysis_plan &plan) |
2461 | : m_scc (eg.get_supergraph (), eg.get_logger ()), |
2462 | m_plan (plan), |
2463 | m_queue (key_t (*this, NULL)) |
2464 | { |
2465 | } |
2466 | |
2467 | /* Return the number of nodes in the worklist. */ |
2468 | |
2469 | unsigned |
2470 | worklist::length () const |
2471 | { |
2472 | return m_queue.nodes (); |
2473 | } |
2474 | |
2475 | /* Return the next node in the worklist, removing it. */ |
2476 | |
2477 | exploded_node * |
2478 | worklist::take_next () |
2479 | { |
2480 | return m_queue.extract_min (); |
2481 | } |
2482 | |
2483 | /* Return the next node in the worklist without removing it. */ |
2484 | |
2485 | exploded_node * |
2486 | worklist::peek_next () |
2487 | { |
2488 | return m_queue.min (); |
2489 | } |
2490 | |
2491 | /* Add ENODE to the worklist. */ |
2492 | |
2493 | void |
2494 | worklist::add_node (exploded_node *enode) |
2495 | { |
2496 | gcc_assert (enode->get_status () == exploded_node::STATUS_WORKLIST); |
2497 | m_queue.insert (key: key_t (*this, enode), data: enode); |
2498 | } |
2499 | |
2500 | /* Comparator for implementing worklist::key_t comparison operators. |
2501 | Return negative if KA is before KB |
2502 | Return positive if KA is after KB |
2503 | Return 0 if they are equal. |
2504 | |
2505 | The ordering of the worklist is critical for performance and for |
2506 | avoiding node explosions. Ideally we want all enodes at a CFG join-point |
2507 | with the same callstring to be sorted next to each other in the worklist |
2508 | so that a run of consecutive enodes can be merged and processed "in bulk" |
2509 | rather than individually or pairwise, minimizing the number of new enodes |
2510 | created. */ |
2511 | |
2512 | int |
2513 | worklist::key_t::cmp (const worklist::key_t &ka, const worklist::key_t &kb) |
2514 | { |
2515 | const program_point &point_a = ka.m_enode->get_point (); |
2516 | const program_point &point_b = kb.m_enode->get_point (); |
2517 | const call_string &call_string_a = point_a.get_call_string (); |
2518 | const call_string &call_string_b = point_b.get_call_string (); |
2519 | |
2520 | /* Order empty-callstring points with different functions based on the |
2521 | analysis_plan, so that we generate summaries before they are used. */ |
2522 | if (flag_analyzer_call_summaries |
2523 | && call_string_a.empty_p () |
2524 | && call_string_b.empty_p () |
2525 | && point_a.get_function () != NULL |
2526 | && point_b.get_function () != NULL |
2527 | && point_a.get_function () != point_b.get_function ()) |
2528 | { |
2529 | if (int cmp = ka.m_worklist.m_plan.cmp_function (fun_a: point_a.get_function (), |
2530 | fun_b: point_b.get_function ())) |
2531 | return cmp; |
2532 | } |
2533 | |
2534 | /* Sort by callstring, so that nodes with deeper call strings are processed |
2535 | before those with shallower call strings. |
2536 | If we have |
2537 | splitting BB |
2538 | / \ |
2539 | / \ |
2540 | fn call no fn call |
2541 | \ / |
2542 | \ / |
2543 | join BB |
2544 | then we want the path inside the function call to be fully explored up |
2545 | to the return to the join BB before we explore on the "no fn call" path, |
2546 | so that both enodes at the join BB reach the front of the worklist at |
2547 | the same time and thus have a chance of being merged. */ |
2548 | int cs_cmp = call_string::cmp (a: call_string_a, b: call_string_b); |
2549 | if (cs_cmp) |
2550 | return cs_cmp; |
2551 | |
2552 | /* Order by SCC. */ |
2553 | int scc_id_a = ka.get_scc_id (enode: ka.m_enode); |
2554 | int scc_id_b = kb.get_scc_id (enode: kb.m_enode); |
2555 | if (scc_id_a != scc_id_b) |
2556 | return scc_id_a - scc_id_b; |
2557 | |
2558 | /* If in same SCC, order by supernode index (an arbitrary but stable |
2559 | ordering). */ |
2560 | const supernode *snode_a = ka.m_enode->get_supernode (); |
2561 | const supernode *snode_b = kb.m_enode->get_supernode (); |
2562 | if (snode_a == NULL) |
2563 | { |
2564 | if (snode_b != NULL) |
2565 | /* One is NULL. */ |
2566 | return -1; |
2567 | else |
2568 | /* Both are NULL. */ |
2569 | return 0; |
2570 | } |
2571 | if (snode_b == NULL) |
2572 | /* One is NULL. */ |
2573 | return 1; |
2574 | /* Neither are NULL. */ |
2575 | gcc_assert (snode_a && snode_b); |
2576 | if (snode_a->m_index != snode_b->m_index) |
2577 | return snode_a->m_index - snode_b->m_index; |
2578 | |
2579 | gcc_assert (snode_a == snode_b); |
2580 | |
2581 | /* Order within supernode via program point. */ |
2582 | int within_snode_cmp |
2583 | = function_point::cmp_within_supernode (point_a: point_a.get_function_point (), |
2584 | point_b: point_b.get_function_point ()); |
2585 | if (within_snode_cmp) |
2586 | return within_snode_cmp; |
2587 | |
2588 | /* Otherwise, we ought to have the same program_point. */ |
2589 | gcc_assert (point_a == point_b); |
2590 | |
2591 | const program_state &state_a = ka.m_enode->get_state (); |
2592 | const program_state &state_b = kb.m_enode->get_state (); |
2593 | |
2594 | /* Sort by sm-state, so that identical sm-states are grouped |
2595 | together in the worklist. */ |
2596 | for (unsigned sm_idx = 0; sm_idx < state_a.m_checker_states.length (); |
2597 | ++sm_idx) |
2598 | { |
2599 | sm_state_map *smap_a = state_a.m_checker_states[sm_idx]; |
2600 | sm_state_map *smap_b = state_b.m_checker_states[sm_idx]; |
2601 | |
2602 | if (int smap_cmp = sm_state_map::cmp (smap_a: *smap_a, smap_b: *smap_b)) |
2603 | return smap_cmp; |
2604 | } |
2605 | |
2606 | /* Otherwise, we have two enodes at the same program point but with |
2607 | different states. We don't have a good total ordering on states, |
2608 | so order them by enode index, so that we have at least have a |
2609 | stable sort. */ |
2610 | return ka.m_enode->m_index - kb.m_enode->m_index; |
2611 | } |
2612 | |
2613 | /* Return a new json::object of the form |
2614 | {"scc" : [per-snode-IDs]}, */ |
2615 | |
2616 | json::object * |
2617 | worklist::to_json () const |
2618 | { |
2619 | json::object *worklist_obj = new json::object (); |
2620 | |
2621 | worklist_obj->set (key: "scc" , v: m_scc.to_json ()); |
2622 | |
2623 | /* The following field isn't yet being JSONified: |
2624 | queue_t m_queue; */ |
2625 | |
2626 | return worklist_obj; |
2627 | } |
2628 | |
2629 | /* exploded_graph's ctor. */ |
2630 | |
2631 | exploded_graph::exploded_graph (const supergraph &sg, logger *logger, |
2632 | const extrinsic_state &ext_state, |
2633 | const state_purge_map *purge_map, |
2634 | const analysis_plan &plan, |
2635 | int verbosity) |
2636 | : m_sg (sg), m_logger (logger), |
2637 | m_worklist (*this, plan), |
2638 | m_ext_state (ext_state), |
2639 | m_purge_map (purge_map), |
2640 | m_plan (plan), |
2641 | m_diagnostic_manager (logger, ext_state.get_engine (), verbosity), |
2642 | m_global_stats (m_sg.num_nodes ()), |
2643 | m_functionless_stats (m_sg.num_nodes ()), |
2644 | m_PK_AFTER_SUPERNODE_per_snode (m_sg.num_nodes ()) |
2645 | { |
2646 | m_origin = get_or_create_node |
2647 | (point: program_point::origin (mgr: *ext_state.get_model_manager ()), |
2648 | state: program_state (ext_state), NULL); |
2649 | for (int i = 0; i < m_sg.num_nodes (); i++) |
2650 | m_PK_AFTER_SUPERNODE_per_snode.quick_push (obj: i); |
2651 | } |
2652 | |
2653 | /* exploded_graph's dtor. */ |
2654 | |
2655 | exploded_graph::~exploded_graph () |
2656 | { |
2657 | for (auto iter : m_per_point_data) |
2658 | delete iter.second; |
2659 | for (auto iter : m_per_function_data) |
2660 | delete iter.second; |
2661 | for (auto iter : m_per_function_stats) |
2662 | delete iter.second; |
2663 | for (auto iter : m_per_call_string_data) |
2664 | delete iter.second; |
2665 | } |
2666 | |
2667 | /* Subroutine for use when implementing __attribute__((tainted_args)) |
2668 | on functions and on function pointer fields in structs. |
2669 | |
2670 | Called on STATE representing a call to FNDECL. |
2671 | Mark all params of FNDECL in STATE as "tainted". Mark the value of all |
2672 | regions pointed to by params of FNDECL as "tainted". |
2673 | |
2674 | Return true if successful; return false if the "taint" state machine |
2675 | was not found. */ |
2676 | |
2677 | static bool |
2678 | mark_params_as_tainted (program_state *state, tree fndecl, |
2679 | const extrinsic_state &ext_state) |
2680 | { |
2681 | unsigned taint_sm_idx; |
2682 | if (!ext_state.get_sm_idx_by_name (name: "taint" , out: &taint_sm_idx)) |
2683 | return false; |
2684 | sm_state_map *smap = state->m_checker_states[taint_sm_idx]; |
2685 | |
2686 | const state_machine &sm = ext_state.get_sm (idx: taint_sm_idx); |
2687 | state_machine::state_t tainted = sm.get_state_by_name (name: "tainted" ); |
2688 | |
2689 | region_model_manager *mgr = ext_state.get_model_manager (); |
2690 | |
2691 | function *fun = DECL_STRUCT_FUNCTION (fndecl); |
2692 | gcc_assert (fun); |
2693 | |
2694 | for (tree iter_parm = DECL_ARGUMENTS (fndecl); iter_parm; |
2695 | iter_parm = DECL_CHAIN (iter_parm)) |
2696 | { |
2697 | tree param = iter_parm; |
2698 | if (tree parm_default_ssa = ssa_default_def (fun, iter_parm)) |
2699 | param = parm_default_ssa; |
2700 | const region *param_reg = state->m_region_model->get_lvalue (expr: param, NULL); |
2701 | const svalue *init_sval = mgr->get_or_create_initial_value (reg: param_reg); |
2702 | smap->set_state (model: state->m_region_model, sval: init_sval, |
2703 | state: tainted, NULL /*origin_new_sval*/, ext_state); |
2704 | if (POINTER_TYPE_P (TREE_TYPE (param))) |
2705 | { |
2706 | const region *pointee_reg = mgr->get_symbolic_region (sval: init_sval); |
2707 | /* Mark "*param" as tainted. */ |
2708 | const svalue *init_pointee_sval |
2709 | = mgr->get_or_create_initial_value (reg: pointee_reg); |
2710 | smap->set_state (model: state->m_region_model, sval: init_pointee_sval, |
2711 | state: tainted, NULL /*origin_new_sval*/, ext_state); |
2712 | } |
2713 | } |
2714 | |
2715 | return true; |
2716 | } |
2717 | |
2718 | /* Custom event for use by tainted_args_function_info when a function |
2719 | has been marked with __attribute__((tainted_args)). */ |
2720 | |
2721 | class tainted_args_function_custom_event : public custom_event |
2722 | { |
2723 | public: |
2724 | tainted_args_function_custom_event (const event_loc_info &loc_info) |
2725 | : custom_event (loc_info), |
2726 | m_fndecl (loc_info.m_fndecl) |
2727 | { |
2728 | } |
2729 | |
2730 | label_text get_desc (bool can_colorize) const final override |
2731 | { |
2732 | return make_label_text |
2733 | (can_colorize, |
2734 | fmt: "function %qE marked with %<__attribute__((tainted_args))%>" , |
2735 | m_fndecl); |
2736 | } |
2737 | |
2738 | private: |
2739 | tree m_fndecl; |
2740 | }; |
2741 | |
2742 | /* Custom exploded_edge info for top-level calls to a function |
2743 | marked with __attribute__((tainted_args)). */ |
2744 | |
2745 | class tainted_args_function_info : public custom_edge_info |
2746 | { |
2747 | public: |
2748 | tainted_args_function_info (tree fndecl) |
2749 | : m_fndecl (fndecl) |
2750 | {} |
2751 | |
2752 | void print (pretty_printer *pp) const final override |
2753 | { |
2754 | pp_string (pp, "call to tainted_args function" ); |
2755 | }; |
2756 | |
2757 | bool update_model (region_model *, |
2758 | const exploded_edge *, |
2759 | region_model_context *) const final override |
2760 | { |
2761 | /* No-op. */ |
2762 | return true; |
2763 | } |
2764 | |
2765 | void add_events_to_path (checker_path *emission_path, |
2766 | const exploded_edge &) const final override |
2767 | { |
2768 | emission_path->add_event |
2769 | (event: make_unique<tainted_args_function_custom_event> |
2770 | (args: event_loc_info (DECL_SOURCE_LOCATION (m_fndecl), m_fndecl, 0))); |
2771 | } |
2772 | |
2773 | private: |
2774 | tree m_fndecl; |
2775 | }; |
2776 | |
2777 | /* Ensure that there is an exploded_node representing an external call to |
2778 | FUN, adding it to the worklist if creating it. |
2779 | |
2780 | Add an edge from the origin exploded_node to the function entrypoint |
2781 | exploded_node. |
2782 | |
2783 | Return the exploded_node for the entrypoint to the function. */ |
2784 | |
2785 | exploded_node * |
2786 | exploded_graph::add_function_entry (const function &fun) |
2787 | { |
2788 | gcc_assert (gimple_has_body_p (fun.decl)); |
2789 | |
2790 | /* Be idempotent. */ |
2791 | function *key = const_cast<function *> (&fun); |
2792 | if (m_functions_with_enodes.contains (k: key)) |
2793 | { |
2794 | logger * const logger = get_logger (); |
2795 | if (logger) |
2796 | logger->log (fmt: "entrypoint for %qE already exists" , fun.decl); |
2797 | return NULL; |
2798 | } |
2799 | |
2800 | program_point point |
2801 | = program_point::from_function_entry (mgr: *m_ext_state.get_model_manager (), |
2802 | sg: m_sg, fun); |
2803 | program_state state (m_ext_state); |
2804 | state.push_frame (ext_state: m_ext_state, fun); |
2805 | |
2806 | std::unique_ptr<custom_edge_info> edge_info = NULL; |
2807 | |
2808 | if (lookup_attribute (attr_name: "tainted_args" , DECL_ATTRIBUTES (fun.decl))) |
2809 | { |
2810 | if (mark_params_as_tainted (state: &state, fndecl: fun.decl, ext_state: m_ext_state)) |
2811 | edge_info = make_unique<tainted_args_function_info> (args: fun.decl); |
2812 | } |
2813 | |
2814 | if (!state.m_valid) |
2815 | return NULL; |
2816 | |
2817 | exploded_node *enode = get_or_create_node (point, state, NULL); |
2818 | if (!enode) |
2819 | return NULL; |
2820 | |
2821 | add_edge (src: m_origin, dest: enode, NULL, could_do_work: false, custom: std::move (edge_info)); |
2822 | |
2823 | m_functions_with_enodes.add (k: key); |
2824 | |
2825 | return enode; |
2826 | } |
2827 | |
2828 | /* Get or create an exploded_node for (POINT, STATE). |
2829 | If a new node is created, it is added to the worklist. |
2830 | |
2831 | Use ENODE_FOR_DIAG, a pre-existing enode, for any diagnostics |
2832 | that need to be emitted (e.g. when purging state *before* we have |
2833 | a new enode). */ |
2834 | |
2835 | exploded_node * |
2836 | exploded_graph::get_or_create_node (const program_point &point, |
2837 | const program_state &state, |
2838 | exploded_node *enode_for_diag) |
2839 | { |
2840 | logger * const logger = get_logger (); |
2841 | LOG_FUNC (logger); |
2842 | if (logger) |
2843 | { |
2844 | format f (false); |
2845 | pretty_printer *pp = logger->get_printer (); |
2846 | logger->start_log_line (); |
2847 | pp_string (pp, "point: " ); |
2848 | point.print (pp, f); |
2849 | logger->end_log_line (); |
2850 | logger->start_log_line (); |
2851 | pp_string (pp, "state: " ); |
2852 | state.dump_to_pp (ext_state: m_ext_state, simple: true, multiline: false, pp); |
2853 | logger->end_log_line (); |
2854 | } |
2855 | |
2856 | /* Stop exploring paths for which we don't know how to effectively |
2857 | model the state. */ |
2858 | if (!state.m_valid) |
2859 | { |
2860 | if (logger) |
2861 | logger->log (fmt: "invalid state; not creating node" ); |
2862 | return NULL; |
2863 | } |
2864 | |
2865 | auto_cfun sentinel (point.get_function ()); |
2866 | |
2867 | state.validate (ext_state: get_ext_state ()); |
2868 | |
2869 | //state.dump (get_ext_state ()); |
2870 | |
2871 | /* Prune state to try to improve the chances of a cache hit, |
2872 | avoiding generating redundant nodes. */ |
2873 | uncertainty_t uncertainty; |
2874 | program_state pruned_state |
2875 | = state.prune_for_point (eg&: *this, point, enode_for_diag, uncertainty: &uncertainty); |
2876 | |
2877 | pruned_state.validate (ext_state: get_ext_state ()); |
2878 | |
2879 | //pruned_state.dump (get_ext_state ()); |
2880 | |
2881 | if (logger) |
2882 | { |
2883 | pretty_printer *pp = logger->get_printer (); |
2884 | logger->start_log_line (); |
2885 | pp_string (pp, "pruned_state: " ); |
2886 | pruned_state.dump_to_pp (ext_state: m_ext_state, simple: true, multiline: false, pp); |
2887 | logger->end_log_line (); |
2888 | pruned_state.m_region_model->dump_to_pp (pp: logger->get_printer (), simple: true, |
2889 | multiline: false); |
2890 | } |
2891 | |
2892 | stats *per_fn_stats = get_or_create_function_stats (fn: point.get_function ()); |
2893 | |
2894 | stats *per_cs_stats |
2895 | = &get_or_create_per_call_string_data (point.get_call_string ())->m_stats; |
2896 | |
2897 | point_and_state ps (point, pruned_state); |
2898 | ps.validate (ext_state: m_ext_state); |
2899 | if (exploded_node **slot = m_point_and_state_to_node.get (k: &ps)) |
2900 | { |
2901 | /* An exploded_node for PS already exists. */ |
2902 | if (logger) |
2903 | logger->log (fmt: "reused EN: %i" , (*slot)->m_index); |
2904 | m_global_stats.m_node_reuse_count++; |
2905 | per_fn_stats->m_node_reuse_count++; |
2906 | per_cs_stats->m_node_reuse_count++; |
2907 | return *slot; |
2908 | } |
2909 | |
2910 | per_program_point_data *per_point_data |
2911 | = get_or_create_per_program_point_data (point); |
2912 | |
2913 | /* Consider merging state with another enode at this program_point. */ |
2914 | if (flag_analyzer_state_merge) |
2915 | { |
2916 | exploded_node *existing_enode; |
2917 | unsigned i; |
2918 | FOR_EACH_VEC_ELT (per_point_data->m_enodes, i, existing_enode) |
2919 | { |
2920 | if (logger) |
2921 | logger->log (fmt: "considering merging with existing EN: %i for point" , |
2922 | existing_enode->m_index); |
2923 | gcc_assert (existing_enode->get_point () == point); |
2924 | const program_state &existing_state = existing_enode->get_state (); |
2925 | |
2926 | /* This merges successfully within the loop. */ |
2927 | |
2928 | program_state merged_state (m_ext_state); |
2929 | if (pruned_state.can_merge_with_p (other: existing_state, ext_state: m_ext_state, point, |
2930 | out: &merged_state)) |
2931 | { |
2932 | merged_state.validate (ext_state: m_ext_state); |
2933 | if (logger) |
2934 | logger->log (fmt: "merging new state with that of EN: %i" , |
2935 | existing_enode->m_index); |
2936 | |
2937 | /* Try again for a cache hit. |
2938 | Whether we get one or not, merged_state's value_ids have no |
2939 | relationship to those of the input state, and thus to those |
2940 | of CHANGE, so we must purge any svalue_ids from *CHANGE. */ |
2941 | ps.set_state (merged_state); |
2942 | |
2943 | if (exploded_node **slot = m_point_and_state_to_node.get (k: &ps)) |
2944 | { |
2945 | /* An exploded_node for PS already exists. */ |
2946 | if (logger) |
2947 | logger->log (fmt: "reused EN: %i" , (*slot)->m_index); |
2948 | m_global_stats.m_node_reuse_after_merge_count++; |
2949 | per_fn_stats->m_node_reuse_after_merge_count++; |
2950 | per_cs_stats->m_node_reuse_after_merge_count++; |
2951 | return *slot; |
2952 | } |
2953 | } |
2954 | else |
2955 | if (logger) |
2956 | logger->log (fmt: "not merging new state with that of EN: %i" , |
2957 | existing_enode->m_index); |
2958 | } |
2959 | } |
2960 | |
2961 | /* Impose a limit on the number of enodes per program point, and |
2962 | simply stop if we exceed it. */ |
2963 | if ((int)per_point_data->m_enodes.length () |
2964 | >= param_analyzer_max_enodes_per_program_point) |
2965 | { |
2966 | pretty_printer pp; |
2967 | point.print (pp: &pp, f: format (false)); |
2968 | print_enode_indices (pp: &pp, enodes: per_point_data->m_enodes); |
2969 | if (logger) |
2970 | logger->log (fmt: "not creating enode; too many at program point: %s" , |
2971 | pp_formatted_text (&pp)); |
2972 | warning_at (point.get_location (), OPT_Wanalyzer_too_complex, |
2973 | "terminating analysis for this program point: %s" , |
2974 | pp_formatted_text (&pp)); |
2975 | per_point_data->m_excess_enodes++; |
2976 | return NULL; |
2977 | } |
2978 | |
2979 | ps.validate (ext_state: m_ext_state); |
2980 | |
2981 | /* An exploded_node for "ps" doesn't already exist; create one. */ |
2982 | exploded_node *node = new exploded_node (ps, m_nodes.length ()); |
2983 | add_node (node); |
2984 | m_point_and_state_to_node.put (k: node->get_ps_key (), v: node); |
2985 | |
2986 | /* Update per-program_point data. */ |
2987 | per_point_data->m_enodes.safe_push (obj: node); |
2988 | |
2989 | const enum point_kind node_pk = node->get_point ().get_kind (); |
2990 | m_global_stats.m_num_nodes[node_pk]++; |
2991 | per_fn_stats->m_num_nodes[node_pk]++; |
2992 | per_cs_stats->m_num_nodes[node_pk]++; |
2993 | |
2994 | if (node_pk == PK_AFTER_SUPERNODE) |
2995 | m_PK_AFTER_SUPERNODE_per_snode[point.get_supernode ()->m_index]++; |
2996 | |
2997 | if (logger) |
2998 | { |
2999 | format f (false); |
3000 | pretty_printer *pp = logger->get_printer (); |
3001 | logger->log (fmt: "created EN: %i" , node->m_index); |
3002 | logger->start_log_line (); |
3003 | pp_string (pp, "point: " ); |
3004 | point.print (pp, f); |
3005 | logger->end_log_line (); |
3006 | logger->start_log_line (); |
3007 | pp_string (pp, "pruned_state: " ); |
3008 | pruned_state.dump_to_pp (ext_state: m_ext_state, simple: true, multiline: false, pp); |
3009 | logger->end_log_line (); |
3010 | } |
3011 | |
3012 | /* Add the new node to the worlist. */ |
3013 | m_worklist.add_node (enode: node); |
3014 | return node; |
3015 | } |
3016 | |
3017 | /* Add an exploded_edge from SRC to DEST, recording its association |
3018 | with SEDGE (which may be NULL), and, if non-NULL, taking ownership |
3019 | of CUSTOM_INFO. COULD_DO_WORK is used for detecting infinite loops. |
3020 | Return the newly-created eedge. */ |
3021 | |
3022 | exploded_edge * |
3023 | exploded_graph::add_edge (exploded_node *src, exploded_node *dest, |
3024 | const superedge *sedge, bool could_do_work, |
3025 | std::unique_ptr<custom_edge_info> custom_info) |
3026 | { |
3027 | if (get_logger ()) |
3028 | get_logger ()->log (fmt: "creating edge EN: %i -> EN: %i" , |
3029 | src->m_index, dest->m_index); |
3030 | exploded_edge *e |
3031 | = new exploded_edge (src, dest, sedge, could_do_work, |
3032 | std::move (custom_info)); |
3033 | digraph<eg_traits>::add_edge (edge: e); |
3034 | return e; |
3035 | } |
3036 | |
3037 | /* Ensure that this graph has per-program_point-data for POINT; |
3038 | borrow a pointer to it. */ |
3039 | |
3040 | per_program_point_data * |
3041 | exploded_graph:: |
3042 | get_or_create_per_program_point_data (const program_point &point) |
3043 | { |
3044 | if (per_program_point_data **slot = m_per_point_data.get (k: &point)) |
3045 | return *slot; |
3046 | |
3047 | per_program_point_data *per_point_data = new per_program_point_data (point); |
3048 | m_per_point_data.put (k: &per_point_data->m_key, v: per_point_data); |
3049 | return per_point_data; |
3050 | } |
3051 | |
3052 | /* Get this graph's per-program-point-data for POINT if there is any, |
3053 | otherwise NULL. */ |
3054 | |
3055 | per_program_point_data * |
3056 | exploded_graph::get_per_program_point_data (const program_point &point) const |
3057 | { |
3058 | if (per_program_point_data **slot |
3059 | = const_cast <point_map_t &> (m_per_point_data).get (k: &point)) |
3060 | return *slot; |
3061 | |
3062 | return NULL; |
3063 | } |
3064 | |
3065 | /* Ensure that this graph has per-call_string-data for CS; |
3066 | borrow a pointer to it. */ |
3067 | |
3068 | per_call_string_data * |
3069 | exploded_graph::get_or_create_per_call_string_data (const call_string &cs) |
3070 | { |
3071 | if (per_call_string_data **slot = m_per_call_string_data.get (k: &cs)) |
3072 | return *slot; |
3073 | |
3074 | per_call_string_data *data = new per_call_string_data (cs, m_sg.num_nodes ()); |
3075 | m_per_call_string_data.put (k: &data->m_key, |
3076 | v: data); |
3077 | return data; |
3078 | } |
3079 | |
3080 | /* Ensure that this graph has per-function-data for FUN; |
3081 | borrow a pointer to it. */ |
3082 | |
3083 | per_function_data * |
3084 | exploded_graph::get_or_create_per_function_data (function *fun) |
3085 | { |
3086 | if (per_function_data **slot = m_per_function_data.get (k: fun)) |
3087 | return *slot; |
3088 | |
3089 | per_function_data *data = new per_function_data (); |
3090 | m_per_function_data.put (k: fun, v: data); |
3091 | return data; |
3092 | } |
3093 | |
3094 | /* Get this graph's per-function-data for FUN if there is any, |
3095 | otherwise NULL. */ |
3096 | |
3097 | per_function_data * |
3098 | exploded_graph::get_per_function_data (function *fun) const |
3099 | { |
3100 | if (per_function_data **slot |
3101 | = const_cast <per_function_data_t &> (m_per_function_data).get (k: fun)) |
3102 | return *slot; |
3103 | |
3104 | return NULL; |
3105 | } |
3106 | |
3107 | /* Return true if FUN should be traversed directly, rather than only as |
3108 | called via other functions. */ |
3109 | |
3110 | static bool |
3111 | toplevel_function_p (const function &fun, logger *logger) |
3112 | { |
3113 | /* Don't directly traverse into functions that have an "__analyzer_" |
3114 | prefix. Doing so is useful for the analyzer test suite, allowing |
3115 | us to have functions that are called in traversals, but not directly |
3116 | explored, thus testing how the analyzer handles calls and returns. |
3117 | With this, we can have DejaGnu directives that cover just the case |
3118 | of where a function is called by another function, without generating |
3119 | excess messages from the case of the first function being traversed |
3120 | directly. */ |
3121 | #define ANALYZER_PREFIX "__analyzer_" |
3122 | if (!strncmp (IDENTIFIER_POINTER (DECL_NAME (fun.decl)), ANALYZER_PREFIX, |
3123 | n: strlen (ANALYZER_PREFIX))) |
3124 | { |
3125 | if (logger) |
3126 | logger->log (fmt: "not traversing %qE (starts with %qs)" , |
3127 | fun.decl, ANALYZER_PREFIX); |
3128 | return false; |
3129 | } |
3130 | |
3131 | if (logger) |
3132 | logger->log (fmt: "traversing %qE (all checks passed)" , fun.decl); |
3133 | |
3134 | return true; |
3135 | } |
3136 | |
3137 | /* Custom event for use by tainted_call_info when a callback field has been |
3138 | marked with __attribute__((tainted_args)), for labelling the field. */ |
3139 | |
3140 | class tainted_args_field_custom_event : public custom_event |
3141 | { |
3142 | public: |
3143 | tainted_args_field_custom_event (tree field) |
3144 | : custom_event (event_loc_info (DECL_SOURCE_LOCATION (field), NULL_TREE, 0)), |
3145 | m_field (field) |
3146 | { |
3147 | } |
3148 | |
3149 | label_text get_desc (bool can_colorize) const final override |
3150 | { |
3151 | return make_label_text (can_colorize, |
3152 | fmt: "field %qE of %qT" |
3153 | " is marked with %<__attribute__((tainted_args))%>" , |
3154 | m_field, DECL_CONTEXT (m_field)); |
3155 | } |
3156 | |
3157 | private: |
3158 | tree m_field; |
3159 | }; |
3160 | |
3161 | /* Custom event for use by tainted_call_info when a callback field has been |
3162 | marked with __attribute__((tainted_args)), for labelling the function used |
3163 | in that callback. */ |
3164 | |
3165 | class tainted_args_callback_custom_event : public custom_event |
3166 | { |
3167 | public: |
3168 | tainted_args_callback_custom_event (const event_loc_info &loc_info, |
3169 | tree field) |
3170 | : custom_event (loc_info), |
3171 | m_field (field) |
3172 | { |
3173 | } |
3174 | |
3175 | label_text get_desc (bool can_colorize) const final override |
3176 | { |
3177 | return make_label_text (can_colorize, |
3178 | fmt: "function %qE used as initializer for field %qE" |
3179 | " marked with %<__attribute__((tainted_args))%>" , |
3180 | get_fndecl (), m_field); |
3181 | } |
3182 | |
3183 | private: |
3184 | tree m_field; |
3185 | }; |
3186 | |
3187 | /* Custom edge info for use when adding a function used by a callback field |
3188 | marked with '__attribute__((tainted_args))'. */ |
3189 | |
3190 | class tainted_args_call_info : public custom_edge_info |
3191 | { |
3192 | public: |
3193 | tainted_args_call_info (tree field, tree fndecl, location_t loc) |
3194 | : m_field (field), m_fndecl (fndecl), m_loc (loc) |
3195 | {} |
3196 | |
3197 | void print (pretty_printer *pp) const final override |
3198 | { |
3199 | pp_string (pp, "call to tainted field" ); |
3200 | }; |
3201 | |
3202 | bool update_model (region_model *, |
3203 | const exploded_edge *, |
3204 | region_model_context *) const final override |
3205 | { |
3206 | /* No-op. */ |
3207 | return true; |
3208 | } |
3209 | |
3210 | void add_events_to_path (checker_path *emission_path, |
3211 | const exploded_edge &) const final override |
3212 | { |
3213 | /* Show the field in the struct declaration, e.g. |
3214 | "(1) field 'store' is marked with '__attribute__((tainted_args))'" */ |
3215 | emission_path->add_event |
3216 | (event: make_unique<tainted_args_field_custom_event> (args: m_field)); |
3217 | |
3218 | /* Show the callback in the initializer |
3219 | e.g. |
3220 | "(2) function 'gadget_dev_desc_UDC_store' used as initializer |
3221 | for field 'store' marked with '__attribute__((tainted_args))'". */ |
3222 | emission_path->add_event |
3223 | (event: make_unique<tainted_args_callback_custom_event> |
3224 | (args: event_loc_info (m_loc, m_fndecl, 0), |
3225 | args: m_field)); |
3226 | } |
3227 | |
3228 | private: |
3229 | tree m_field; |
3230 | tree m_fndecl; |
3231 | location_t m_loc; |
3232 | }; |
3233 | |
3234 | /* Given an initializer at LOC for FIELD marked with |
3235 | '__attribute__((tainted_args))' initialized with FNDECL, add an |
3236 | entrypoint to FNDECL to EG (and to its worklist) where the params to |
3237 | FNDECL are marked as tainted. */ |
3238 | |
3239 | static void |
3240 | add_tainted_args_callback (exploded_graph *eg, tree field, tree fndecl, |
3241 | location_t loc) |
3242 | { |
3243 | logger *logger = eg->get_logger (); |
3244 | |
3245 | LOG_SCOPE (logger); |
3246 | |
3247 | if (!gimple_has_body_p (fndecl)) |
3248 | return; |
3249 | |
3250 | const extrinsic_state &ext_state = eg->get_ext_state (); |
3251 | |
3252 | function *fun = DECL_STRUCT_FUNCTION (fndecl); |
3253 | gcc_assert (fun); |
3254 | |
3255 | program_point point |
3256 | = program_point::from_function_entry (mgr: *ext_state.get_model_manager (), |
3257 | sg: eg->get_supergraph (), fun: *fun); |
3258 | program_state state (ext_state); |
3259 | state.push_frame (ext_state, fun: *fun); |
3260 | |
3261 | if (!mark_params_as_tainted (state: &state, fndecl, ext_state)) |
3262 | return; |
3263 | |
3264 | if (!state.m_valid) |
3265 | return; |
3266 | |
3267 | exploded_node *enode = eg->get_or_create_node (point, state, NULL); |
3268 | if (logger) |
3269 | { |
3270 | if (enode) |
3271 | logger->log (fmt: "created EN %i for tainted_args %qE entrypoint" , |
3272 | enode->m_index, fndecl); |
3273 | else |
3274 | { |
3275 | logger->log (fmt: "did not create enode for tainted_args %qE entrypoint" , |
3276 | fndecl); |
3277 | return; |
3278 | } |
3279 | } |
3280 | |
3281 | eg->add_edge (src: eg->get_origin (), dest: enode, NULL, could_do_work: false, |
3282 | custom_info: make_unique<tainted_args_call_info> (args&: field, args&: fndecl, args&: loc)); |
3283 | } |
3284 | |
3285 | /* Callback for walk_tree for finding callbacks within initializers; |
3286 | ensure that any callback initializer where the corresponding field is |
3287 | marked with '__attribute__((tainted_args))' is treated as an entrypoint |
3288 | to the analysis, special-casing that the inputs to the callback are |
3289 | untrustworthy. */ |
3290 | |
3291 | static tree |
3292 | add_any_callbacks (tree *tp, int *, void *data) |
3293 | { |
3294 | exploded_graph *eg = (exploded_graph *)data; |
3295 | if (TREE_CODE (*tp) == CONSTRUCTOR) |
3296 | { |
3297 | /* Find fields with the "tainted_args" attribute. |
3298 | walk_tree only walks the values, not the index values; |
3299 | look at the index values. */ |
3300 | unsigned HOST_WIDE_INT idx; |
3301 | constructor_elt *ce; |
3302 | |
3303 | for (idx = 0; vec_safe_iterate (CONSTRUCTOR_ELTS (*tp), ix: idx, ptr: &ce); |
3304 | idx++) |
3305 | if (ce->index && TREE_CODE (ce->index) == FIELD_DECL) |
3306 | if (lookup_attribute (attr_name: "tainted_args" , DECL_ATTRIBUTES (ce->index))) |
3307 | { |
3308 | tree value = ce->value; |
3309 | if (TREE_CODE (value) == ADDR_EXPR |
3310 | && TREE_CODE (TREE_OPERAND (value, 0)) == FUNCTION_DECL) |
3311 | add_tainted_args_callback (eg, field: ce->index, |
3312 | TREE_OPERAND (value, 0), |
3313 | EXPR_LOCATION (value)); |
3314 | } |
3315 | } |
3316 | |
3317 | return NULL_TREE; |
3318 | } |
3319 | |
3320 | /* Add initial nodes to EG, with entrypoints for externally-callable |
3321 | functions. */ |
3322 | |
3323 | void |
3324 | exploded_graph::build_initial_worklist () |
3325 | { |
3326 | logger * const logger = get_logger (); |
3327 | LOG_SCOPE (logger); |
3328 | |
3329 | cgraph_node *node; |
3330 | FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node) |
3331 | { |
3332 | function *fun = node->get_fun (); |
3333 | gcc_assert (fun); |
3334 | if (!toplevel_function_p (fun: *fun, logger)) |
3335 | continue; |
3336 | exploded_node *enode = add_function_entry (fun: *fun); |
3337 | if (logger) |
3338 | { |
3339 | if (enode) |
3340 | logger->log (fmt: "created EN %i for %qE entrypoint" , |
3341 | enode->m_index, fun->decl); |
3342 | else |
3343 | logger->log (fmt: "did not create enode for %qE entrypoint" , fun->decl); |
3344 | } |
3345 | } |
3346 | |
3347 | /* Find callbacks that are reachable from global initializers. */ |
3348 | varpool_node *vpnode; |
3349 | FOR_EACH_VARIABLE (vpnode) |
3350 | { |
3351 | tree decl = vpnode->decl; |
3352 | tree init = DECL_INITIAL (decl); |
3353 | if (!init) |
3354 | continue; |
3355 | walk_tree (&init, add_any_callbacks, this, NULL); |
3356 | } |
3357 | } |
3358 | |
3359 | /* The main loop of the analysis. |
3360 | Take freshly-created exploded_nodes from the worklist, calling |
3361 | process_node on them to explore the <point, state> graph. |
3362 | Add edges to their successors, potentially creating new successors |
3363 | (which are also added to the worklist). */ |
3364 | |
3365 | void |
3366 | exploded_graph::process_worklist () |
3367 | { |
3368 | logger * const logger = get_logger (); |
3369 | LOG_SCOPE (logger); |
3370 | auto_timevar tv (TV_ANALYZER_WORKLIST); |
3371 | |
3372 | while (m_worklist.length () > 0) |
3373 | { |
3374 | exploded_node *node = m_worklist.take_next (); |
3375 | gcc_assert (node->get_status () == exploded_node::STATUS_WORKLIST); |
3376 | gcc_assert (node->m_succs.length () == 0 |
3377 | || node == m_origin); |
3378 | |
3379 | if (logger) |
3380 | logger->log (fmt: "next to process: EN: %i" , node->m_index); |
3381 | |
3382 | /* If we have a run of nodes that are before-supernode, try merging and |
3383 | processing them together, rather than pairwise or individually. */ |
3384 | if (flag_analyzer_state_merge && node != m_origin) |
3385 | if (maybe_process_run_of_before_supernode_enodes (node)) |
3386 | goto handle_limit; |
3387 | |
3388 | /* Avoid exponential explosions of nodes by attempting to merge |
3389 | nodes that are at the same program point and which have |
3390 | sufficiently similar state. */ |
3391 | if (flag_analyzer_state_merge && node != m_origin) |
3392 | if (exploded_node *node_2 = m_worklist.peek_next ()) |
3393 | { |
3394 | gcc_assert (node_2->get_status () |
3395 | == exploded_node::STATUS_WORKLIST); |
3396 | gcc_assert (node->m_succs.length () == 0); |
3397 | gcc_assert (node_2->m_succs.length () == 0); |
3398 | |
3399 | gcc_assert (node != node_2); |
3400 | |
3401 | if (logger) |
3402 | logger->log (fmt: "peek worklist: EN: %i" , node_2->m_index); |
3403 | |
3404 | if (node->get_point () == node_2->get_point ()) |
3405 | { |
3406 | const program_point &point = node->get_point (); |
3407 | if (logger) |
3408 | { |
3409 | format f (false); |
3410 | pretty_printer *pp = logger->get_printer (); |
3411 | logger->start_log_line (); |
3412 | logger->log_partial |
3413 | (fmt: "got potential merge EN: %i and EN: %i at " , |
3414 | node->m_index, node_2->m_index); |
3415 | point.print (pp, f); |
3416 | logger->end_log_line (); |
3417 | } |
3418 | const program_state &state = node->get_state (); |
3419 | const program_state &state_2 = node_2->get_state (); |
3420 | |
3421 | /* They shouldn't be equal, or we wouldn't have two |
3422 | separate nodes. */ |
3423 | gcc_assert (state != state_2); |
3424 | |
3425 | program_state merged_state (m_ext_state); |
3426 | if (state.can_merge_with_p (other: state_2, ext_state: m_ext_state, |
3427 | point, out: &merged_state)) |
3428 | { |
3429 | if (logger) |
3430 | logger->log (fmt: "merging EN: %i and EN: %i" , |
3431 | node->m_index, node_2->m_index); |
3432 | |
3433 | if (merged_state == state) |
3434 | { |
3435 | /* Then merge node_2 into node by adding an edge. */ |
3436 | add_edge (src: node_2, dest: node, NULL, could_do_work: false); |
3437 | |
3438 | /* Remove node_2 from the worklist. */ |
3439 | m_worklist.take_next (); |
3440 | node_2->set_status (exploded_node::STATUS_MERGER); |
3441 | |
3442 | /* Continue processing "node" below. */ |
3443 | } |
3444 | else if (merged_state == state_2) |
3445 | { |
3446 | /* Then merge node into node_2, and leave node_2 |
3447 | in the worklist, to be processed on the next |
3448 | iteration. */ |
3449 | add_edge (src: node, dest: node_2, NULL, could_do_work: false); |
3450 | node->set_status (exploded_node::STATUS_MERGER); |
3451 | continue; |
3452 | } |
3453 | else |
3454 | { |
3455 | /* We have a merged state that differs from |
3456 | both state and state_2. */ |
3457 | |
3458 | /* Remove node_2 from the worklist. */ |
3459 | m_worklist.take_next (); |
3460 | |
3461 | /* Create (or get) an exploded node for the merged |
3462 | states, adding to the worklist. */ |
3463 | exploded_node *merged_enode |
3464 | = get_or_create_node (point: node->get_point (), |
3465 | state: merged_state, enode_for_diag: node); |
3466 | if (merged_enode == NULL) |
3467 | continue; |
3468 | |
3469 | if (logger) |
3470 | logger->log (fmt: "merged EN: %i and EN: %i into EN: %i" , |
3471 | node->m_index, node_2->m_index, |
3472 | merged_enode->m_index); |
3473 | |
3474 | /* "node" and "node_2" have both now been removed |
3475 | from the worklist; we should not process them. |
3476 | |
3477 | "merged_enode" may be a new node; if so it will be |
3478 | processed in a subsequent iteration. |
3479 | Alternatively, "merged_enode" could be an existing |
3480 | node; one way the latter can |
3481 | happen is if we end up merging a succession of |
3482 | similar nodes into one. */ |
3483 | |
3484 | /* If merged_node is one of the two we were merging, |
3485 | add it back to the worklist to ensure it gets |
3486 | processed. |
3487 | |
3488 | Add edges from the merged nodes to it (but not a |
3489 | self-edge). */ |
3490 | if (merged_enode == node) |
3491 | m_worklist.add_node (enode: merged_enode); |
3492 | else |
3493 | { |
3494 | add_edge (src: node, dest: merged_enode, NULL, could_do_work: false); |
3495 | node->set_status (exploded_node::STATUS_MERGER); |
3496 | } |
3497 | |
3498 | if (merged_enode == node_2) |
3499 | m_worklist.add_node (enode: merged_enode); |
3500 | else |
3501 | { |
3502 | add_edge (src: node_2, dest: merged_enode, NULL, could_do_work: false); |
3503 | node_2->set_status (exploded_node::STATUS_MERGER); |
3504 | } |
3505 | |
3506 | continue; |
3507 | } |
3508 | } |
3509 | |
3510 | /* TODO: should we attempt more than two nodes, |
3511 | or just do pairs of nodes? (and hope that we get |
3512 | a cascade of mergers). */ |
3513 | } |
3514 | } |
3515 | |
3516 | process_node (node); |
3517 | |
3518 | handle_limit: |
3519 | /* Impose a hard limit on the number of exploded nodes, to ensure |
3520 | that the analysis terminates in the face of pathological state |
3521 | explosion (or bugs). |
3522 | |
3523 | Specifically, the limit is on the number of PK_AFTER_SUPERNODE |
3524 | exploded nodes, looking at supernode exit events. |
3525 | |
3526 | We use exit rather than entry since there can be multiple |
3527 | entry ENs, one per phi; the number of PK_AFTER_SUPERNODE ought |
3528 | to be equivalent to the number of supernodes multiplied by the |
3529 | number of states. */ |
3530 | const int limit = m_sg.num_nodes () * param_analyzer_bb_explosion_factor; |
3531 | if (m_global_stats.m_num_nodes[PK_AFTER_SUPERNODE] > limit) |
3532 | { |
3533 | if (logger) |
3534 | logger->log (fmt: "bailing out; too many nodes" ); |
3535 | warning_at (node->get_point ().get_location (), |
3536 | OPT_Wanalyzer_too_complex, |
3537 | "analysis bailed out early" |
3538 | " (%i 'after-snode' enodes; %i enodes)" , |
3539 | m_global_stats.m_num_nodes[PK_AFTER_SUPERNODE], |
3540 | m_nodes.length ()); |
3541 | return; |
3542 | } |
3543 | } |
3544 | } |
3545 | |
3546 | /* Attempt to process a consecutive run of sufficiently-similar nodes in |
3547 | the worklist at a CFG join-point (having already popped ENODE from the |
3548 | head of the worklist). |
3549 | |
3550 | If ENODE's point is of the form (before-supernode, SNODE) and the next |
3551 | nodes in the worklist are a consecutive run of enodes of the same form, |
3552 | for the same supernode as ENODE (but potentially from different in-edges), |
3553 | process them all together, setting their status to STATUS_BULK_MERGED, |
3554 | and return true. |
3555 | Otherwise, return false, in which case ENODE must be processed in the |
3556 | normal way. |
3557 | |
3558 | When processing them all together, generate successor states based |
3559 | on phi nodes for the appropriate CFG edges, and then attempt to merge |
3560 | these states into a minimal set of merged successor states, partitioning |
3561 | the inputs by merged successor state. |
3562 | |
3563 | Create new exploded nodes for all of the merged states, and add edges |
3564 | connecting the input enodes to the corresponding merger exploded nodes. |
3565 | |
3566 | We hope we have a much smaller number of merged successor states |
3567 | compared to the number of input enodes - ideally just one, |
3568 | if all successor states can be merged. |
3569 | |
3570 | Processing and merging many together as one operation rather than as |
3571 | pairs avoids scaling issues where per-pair mergers could bloat the |
3572 | graph with merger nodes (especially so after switch statements). */ |
3573 | |
3574 | bool |
3575 | exploded_graph:: |
3576 | maybe_process_run_of_before_supernode_enodes (exploded_node *enode) |
3577 | { |
3578 | /* A struct for tracking per-input state. */ |
3579 | struct item |
3580 | { |
3581 | item (exploded_node *input_enode) |
3582 | : m_input_enode (input_enode), |
3583 | m_processed_state (input_enode->get_state ()), |
3584 | m_merger_idx (-1) |
3585 | {} |
3586 | |
3587 | exploded_node *m_input_enode; |
3588 | program_state m_processed_state; |
3589 | int m_merger_idx; |
3590 | }; |
3591 | |
3592 | gcc_assert (enode->get_status () == exploded_node::STATUS_WORKLIST); |
3593 | gcc_assert (enode->m_succs.length () == 0); |
3594 | |
3595 | const program_point &point = enode->get_point (); |
3596 | |
3597 | if (point.get_kind () != PK_BEFORE_SUPERNODE) |
3598 | return false; |
3599 | |
3600 | const supernode *snode = point.get_supernode (); |
3601 | |
3602 | logger * const logger = get_logger (); |
3603 | LOG_SCOPE (logger); |
3604 | |
3605 | /* Find a run of enodes in the worklist that are before the same supernode, |
3606 | but potentially from different in-edges. */ |
3607 | auto_vec <exploded_node *> enodes; |
3608 | enodes.safe_push (obj: enode); |
3609 | while (exploded_node *enode_2 = m_worklist.peek_next ()) |
3610 | { |
3611 | gcc_assert (enode_2->get_status () |
3612 | == exploded_node::STATUS_WORKLIST); |
3613 | gcc_assert (enode_2->m_succs.length () == 0); |
3614 | |
3615 | const program_point &point_2 = enode_2->get_point (); |
3616 | |
3617 | if (point_2.get_kind () == PK_BEFORE_SUPERNODE |
3618 | && point_2.get_supernode () == snode |
3619 | && &point_2.get_call_string () == &point.get_call_string ()) |
3620 | { |
3621 | enodes.safe_push (obj: enode_2); |
3622 | m_worklist.take_next (); |
3623 | } |
3624 | else |
3625 | break; |
3626 | } |
3627 | |
3628 | /* If the only node is ENODE, then give up. */ |
3629 | if (enodes.length () == 1) |
3630 | return false; |
3631 | |
3632 | if (logger) |
3633 | logger->log (fmt: "got run of %i enodes for SN: %i" , |
3634 | enodes.length (), snode->m_index); |
3635 | |
3636 | /* All of these enodes have a shared successor point (even if they |
3637 | were for different in-edges). */ |
3638 | program_point next_point (point.get_next ()); |
3639 | |
3640 | /* Calculate the successor state for each enode in enodes. */ |
3641 | auto_delete_vec<item> items (enodes.length ()); |
3642 | unsigned i; |
3643 | exploded_node *iter_enode; |
3644 | FOR_EACH_VEC_ELT (enodes, i, iter_enode) |
3645 | { |
3646 | item *it = new item (iter_enode); |
3647 | items.quick_push (obj: it); |
3648 | const program_state &state = iter_enode->get_state (); |
3649 | program_state *next_state = &it->m_processed_state; |
3650 | next_state->validate (ext_state: m_ext_state); |
3651 | const program_point &iter_point = iter_enode->get_point (); |
3652 | if (const superedge *iter_sedge = iter_point.get_from_edge ()) |
3653 | { |
3654 | uncertainty_t uncertainty; |
3655 | impl_region_model_context ctxt (*this, iter_enode, |
3656 | &state, next_state, |
3657 | &uncertainty, NULL, NULL); |
3658 | const cfg_superedge *last_cfg_superedge |
3659 | = iter_sedge->dyn_cast_cfg_superedge (); |
3660 | if (last_cfg_superedge) |
3661 | next_state->m_region_model->update_for_phis |
3662 | (snode, last_cfg_superedge, ctxt: &ctxt); |
3663 | } |
3664 | next_state->validate (ext_state: m_ext_state); |
3665 | } |
3666 | |
3667 | /* Attempt to partition the items into a set of merged states. |
3668 | We hope we have a much smaller number of merged states |
3669 | compared to the number of input enodes - ideally just one, |
3670 | if all can be merged. */ |
3671 | auto_delete_vec <program_state> merged_states; |
3672 | auto_vec<item *> first_item_for_each_merged_state; |
3673 | item *it; |
3674 | FOR_EACH_VEC_ELT (items, i, it) |
3675 | { |
3676 | const program_state &it_state = it->m_processed_state; |
3677 | program_state *merged_state; |
3678 | unsigned iter_merger_idx; |
3679 | FOR_EACH_VEC_ELT (merged_states, iter_merger_idx, merged_state) |
3680 | { |
3681 | merged_state->validate (ext_state: m_ext_state); |
3682 | program_state merge (m_ext_state); |
3683 | if (it_state.can_merge_with_p (other: *merged_state, ext_state: m_ext_state, |
3684 | point: next_point, out: &merge)) |
3685 | { |
3686 | *merged_state = merge; |
3687 | merged_state->validate (ext_state: m_ext_state); |
3688 | it->m_merger_idx = iter_merger_idx; |
3689 | if (logger) |
3690 | logger->log (fmt: "reusing merger state %i for item %i (EN: %i)" , |
3691 | it->m_merger_idx, i, it->m_input_enode->m_index); |
3692 | goto got_merger; |
3693 | } |
3694 | } |
3695 | /* If it couldn't be merged with any existing merged_states, |
3696 | create a new one. */ |
3697 | if (it->m_merger_idx == -1) |
3698 | { |
3699 | it->m_merger_idx = merged_states.length (); |
3700 | merged_states.safe_push (obj: new program_state (it_state)); |
3701 | first_item_for_each_merged_state.safe_push (obj: it); |
3702 | if (logger) |
3703 | logger->log (fmt: "using new merger state %i for item %i (EN: %i)" , |
3704 | it->m_merger_idx, i, it->m_input_enode->m_index); |
3705 | } |
3706 | got_merger: |
3707 | gcc_assert (it->m_merger_idx >= 0); |
3708 | gcc_assert ((unsigned)it->m_merger_idx < merged_states.length ()); |
3709 | } |
3710 | |
3711 | /* Create merger nodes. */ |
3712 | auto_vec<exploded_node *> next_enodes (merged_states.length ()); |
3713 | program_state *merged_state; |
3714 | FOR_EACH_VEC_ELT (merged_states, i, merged_state) |
3715 | { |
3716 | exploded_node *src_enode |
3717 | = first_item_for_each_merged_state[i]->m_input_enode; |
3718 | exploded_node *next |
3719 | = get_or_create_node (point: next_point, state: *merged_state, enode_for_diag: src_enode); |
3720 | /* "next" could be NULL; we handle that when adding the edges below. */ |
3721 | next_enodes.quick_push (obj: next); |
3722 | if (logger) |
3723 | { |
3724 | if (next) |
3725 | logger->log (fmt: "using EN: %i for merger state %i" , next->m_index, i); |
3726 | else |
3727 | logger->log (fmt: "using NULL enode for merger state %i" , i); |
3728 | } |
3729 | } |
3730 | |
3731 | /* Create edges from each input enode to the appropriate successor enode. |
3732 | Update the status of the now-processed input enodes. */ |
3733 | FOR_EACH_VEC_ELT (items, i, it) |
3734 | { |
3735 | exploded_node *next = next_enodes[it->m_merger_idx]; |
3736 | if (next) |
3737 | add_edge (src: it->m_input_enode, dest: next, NULL, |
3738 | could_do_work: false); /* no "work" is done during merger. */ |
3739 | it->m_input_enode->set_status (exploded_node::STATUS_BULK_MERGED); |
3740 | } |
3741 | |
3742 | if (logger) |
3743 | logger->log (fmt: "merged %i in-enodes into %i out-enode(s) at SN: %i" , |
3744 | items.length (), merged_states.length (), snode->m_index); |
3745 | |
3746 | return true; |
3747 | } |
3748 | |
3749 | /* Return true if STMT must appear at the start of its exploded node, and |
3750 | thus we can't consolidate its effects within a run of other statements, |
3751 | where PREV_STMT was the previous statement. */ |
3752 | |
3753 | static bool |
3754 | stmt_requires_new_enode_p (const gimple *stmt, |
3755 | const gimple *prev_stmt) |
3756 | { |
3757 | if (const gcall *call = dyn_cast <const gcall *> (p: stmt)) |
3758 | { |
3759 | /* Stop consolidating at calls to |
3760 | "__analyzer_dump_exploded_nodes", so they always appear at the |
3761 | start of an exploded_node. */ |
3762 | if (is_special_named_call_p (call, funcname: "__analyzer_dump_exploded_nodes" , |
3763 | num_args: 1)) |
3764 | return true; |
3765 | |
3766 | /* sm-signal.cc injects an additional custom eedge at "signal" calls |
3767 | from the registration enode to the handler enode, separate from the |
3768 | regular next state, which defeats the "detect state change" logic |
3769 | in process_node. Work around this via special-casing, to ensure |
3770 | we split the enode immediately before any "signal" call. */ |
3771 | if (is_special_named_call_p (call, funcname: "signal" , num_args: 2)) |
3772 | return true; |
3773 | } |
3774 | |
3775 | /* If we had a PREV_STMT with an unknown location, and this stmt |
3776 | has a known location, then if a state change happens here, it |
3777 | could be consolidated into PREV_STMT, giving us an event with |
3778 | no location. Ensure that STMT gets its own exploded_node to |
3779 | avoid this. */ |
3780 | if (get_pure_location (loc: prev_stmt->location) == UNKNOWN_LOCATION |
3781 | && get_pure_location (loc: stmt->location) != UNKNOWN_LOCATION) |
3782 | return true; |
3783 | |
3784 | return false; |
3785 | } |
3786 | |
3787 | /* Return true if OLD_STATE and NEW_STATE are sufficiently different that |
3788 | we should split enodes and create an exploded_edge separating them |
3789 | (which makes it easier to identify state changes of intereset when |
3790 | constructing checker_paths). */ |
3791 | |
3792 | static bool |
3793 | state_change_requires_new_enode_p (const program_state &old_state, |
3794 | const program_state &new_state) |
3795 | { |
3796 | /* Changes in dynamic extents signify creations of heap/alloca regions |
3797 | and resizings of heap regions; likely to be of interest in |
3798 | diagnostic paths. */ |
3799 | if (old_state.m_region_model->get_dynamic_extents () |
3800 | != new_state.m_region_model->get_dynamic_extents ()) |
3801 | return true; |
3802 | |
3803 | /* Changes in sm-state are of interest. */ |
3804 | int sm_idx; |
3805 | sm_state_map *smap; |
3806 | FOR_EACH_VEC_ELT (old_state.m_checker_states, sm_idx, smap) |
3807 | { |
3808 | const sm_state_map *old_smap = old_state.m_checker_states[sm_idx]; |
3809 | const sm_state_map *new_smap = new_state.m_checker_states[sm_idx]; |
3810 | if (*old_smap != *new_smap) |
3811 | return true; |
3812 | } |
3813 | |
3814 | return false; |
3815 | } |
3816 | |
3817 | /* Create enodes and eedges for the function calls that doesn't have an |
3818 | underlying call superedge. |
3819 | |
3820 | Such case occurs when GCC's middle end didn't know which function to |
3821 | call but the analyzer does (with the help of current state). |
3822 | |
3823 | Some example such calls are dynamically dispatched calls to virtual |
3824 | functions or calls that happen via function pointer. */ |
3825 | |
3826 | bool |
3827 | exploded_graph::maybe_create_dynamic_call (const gcall *call, |
3828 | tree fn_decl, |
3829 | exploded_node *node, |
3830 | program_state next_state, |
3831 | program_point &next_point, |
3832 | uncertainty_t *uncertainty, |
3833 | logger *logger) |
3834 | { |
3835 | LOG_FUNC (logger); |
3836 | |
3837 | const program_point *this_point = &node->get_point (); |
3838 | function *fun = DECL_STRUCT_FUNCTION (fn_decl); |
3839 | if (fun) |
3840 | { |
3841 | const supergraph &sg = this->get_supergraph (); |
3842 | supernode *sn_entry = sg.get_node_for_function_entry (fun: *fun); |
3843 | supernode *sn_exit = sg.get_node_for_function_exit (fun: *fun); |
3844 | |
3845 | program_point new_point |
3846 | = program_point::before_supernode (supernode: sn_entry, |
3847 | NULL, |
3848 | call_string: this_point->get_call_string ()); |
3849 | |
3850 | new_point.push_to_call_stack (caller: sn_exit, |
3851 | callee: next_point.get_supernode()); |
3852 | |
3853 | /* Impose a maximum recursion depth and don't analyze paths |
3854 | that exceed it further. |
3855 | This is something of a blunt workaround, but it only |
3856 | applies to recursion (and mutual recursion), not to |
3857 | general call stacks. */ |
3858 | if (new_point.get_call_string ().calc_recursion_depth () |
3859 | > param_analyzer_max_recursion_depth) |
3860 | { |
3861 | if (logger) |
3862 | logger->log (fmt: "rejecting call edge: recursion limit exceeded" ); |
3863 | return false; |
3864 | } |
3865 | |
3866 | next_state.push_call (eg&: *this, enode: node, call_stmt: call, uncertainty); |
3867 | |
3868 | if (next_state.m_valid) |
3869 | { |
3870 | if (logger) |
3871 | logger->log (fmt: "Discovered call to %s [SN: %i -> SN: %i]" , |
3872 | function_name(fun), |
3873 | this_point->get_supernode ()->m_index, |
3874 | sn_entry->m_index); |
3875 | |
3876 | exploded_node *enode = get_or_create_node (point: new_point, |
3877 | state: next_state, |
3878 | enode_for_diag: node); |
3879 | if (enode) |
3880 | add_edge (src: node,dest: enode, NULL, |
3881 | could_do_work: false, /* No work is done by the call itself. */ |
3882 | custom_info: make_unique<dynamic_call_info_t> (args&: call)); |
3883 | return true; |
3884 | } |
3885 | } |
3886 | return false; |
3887 | } |
3888 | |
3889 | /* Subclass of path_context for use within exploded_graph::process_node, |
3890 | so that we can split states e.g. at "realloc" calls. */ |
3891 | |
3892 | class impl_path_context : public path_context |
3893 | { |
3894 | public: |
3895 | impl_path_context (const program_state *cur_state, |
3896 | logger *logger) |
3897 | : m_cur_state (cur_state), |
3898 | m_logger (logger), |
3899 | m_terminate_path (false) |
3900 | { |
3901 | } |
3902 | |
3903 | bool bifurcation_p () const |
3904 | { |
3905 | return m_custom_eedge_infos.length () > 0; |
3906 | } |
3907 | |
3908 | const program_state &get_state_at_bifurcation () const |
3909 | { |
3910 | gcc_assert (m_state_at_bifurcation); |
3911 | return *m_state_at_bifurcation; |
3912 | } |
3913 | |
3914 | void |
3915 | bifurcate (std::unique_ptr<custom_edge_info> info) final override |
3916 | { |
3917 | if (m_logger) |
3918 | m_logger->log (fmt: "bifurcating path" ); |
3919 | |
3920 | if (m_state_at_bifurcation) |
3921 | /* Verify that the state at bifurcation is consistent when we |
3922 | split into multiple out-edges. */ |
3923 | gcc_assert (*m_state_at_bifurcation == *m_cur_state); |
3924 | else |
3925 | /* Take a copy of the cur_state at the moment when bifurcation |
3926 | happens. */ |
3927 | m_state_at_bifurcation |
3928 | = std::unique_ptr<program_state> (new program_state (*m_cur_state)); |
3929 | |
3930 | /* Take ownership of INFO. */ |
3931 | m_custom_eedge_infos.safe_push (obj: info.release ()); |
3932 | } |
3933 | |
3934 | void terminate_path () final override |
3935 | { |
3936 | if (m_logger) |
3937 | m_logger->log (fmt: "terminating path" ); |
3938 | m_terminate_path = true; |
3939 | } |
3940 | |
3941 | bool terminate_path_p () const final override |
3942 | { |
3943 | return m_terminate_path; |
3944 | } |
3945 | |
3946 | const vec<custom_edge_info *> & get_custom_eedge_infos () |
3947 | { |
3948 | return m_custom_eedge_infos; |
3949 | } |
3950 | |
3951 | private: |
3952 | const program_state *m_cur_state; |
3953 | |
3954 | logger *m_logger; |
3955 | |
3956 | /* Lazily-created copy of the state before the split. */ |
3957 | std::unique_ptr<program_state> m_state_at_bifurcation; |
3958 | |
3959 | auto_vec <custom_edge_info *> m_custom_eedge_infos; |
3960 | |
3961 | bool m_terminate_path; |
3962 | }; |
3963 | |
3964 | /* A subclass of pending_diagnostic for complaining about jumps through NULL |
3965 | function pointers. */ |
3966 | |
3967 | class jump_through_null : public pending_diagnostic_subclass<jump_through_null> |
3968 | { |
3969 | public: |
3970 | jump_through_null (const gcall *call) |
3971 | : m_call (call) |
3972 | {} |
3973 | |
3974 | const char *get_kind () const final override |
3975 | { |
3976 | return "jump_through_null" ; |
3977 | } |
3978 | |
3979 | bool operator== (const jump_through_null &other) const |
3980 | { |
3981 | return m_call == other.m_call; |
3982 | } |
3983 | |
3984 | int get_controlling_option () const final override |
3985 | { |
3986 | return OPT_Wanalyzer_jump_through_null; |
3987 | } |
3988 | |
3989 | bool emit (diagnostic_emission_context &ctxt) final override |
3990 | { |
3991 | return ctxt.warn ("jump through null pointer" ); |
3992 | } |
3993 | |
3994 | label_text describe_final_event (const evdesc::final_event &ev) final override |
3995 | { |
3996 | return ev.formatted_print (fmt: "jump through null pointer here" ); |
3997 | } |
3998 | |
3999 | private: |
4000 | const gcall *m_call; |
4001 | }; |
4002 | |
4003 | /* The core of exploded_graph::process_worklist (the main analysis loop), |
4004 | handling one node in the worklist. |
4005 | |
4006 | Get successor <point, state> pairs for NODE, calling get_or_create on |
4007 | them, and adding an exploded_edge to each successors. |
4008 | |
4009 | Freshly-created nodes will be added to the worklist. */ |
4010 | |
4011 | void |
4012 | exploded_graph::process_node (exploded_node *node) |
4013 | { |
4014 | logger * const logger = get_logger (); |
4015 | LOG_FUNC_1 (logger, "EN: %i" , node->m_index); |
4016 | |
4017 | node->set_status (exploded_node::STATUS_PROCESSED); |
4018 | |
4019 | const program_point &point = node->get_point (); |
4020 | |
4021 | /* Update cfun and input_location in case of an ICE: make it easier to |
4022 | track down which source construct we're failing to handle. */ |
4023 | auto_cfun sentinel (node->get_function ()); |
4024 | const gimple *stmt = point.get_stmt (); |
4025 | if (stmt) |
4026 | input_location = stmt->location; |
4027 | |
4028 | const program_state &state = node->get_state (); |
4029 | if (logger) |
4030 | { |
4031 | pretty_printer *pp = logger->get_printer (); |
4032 | logger->start_log_line (); |
4033 | pp_string (pp, "point: " ); |
4034 | point.print (pp, f: format (false)); |
4035 | pp_string (pp, ", state: " ); |
4036 | state.dump_to_pp (ext_state: m_ext_state, simple: true, multiline: false, pp); |
4037 | logger->end_log_line (); |
4038 | } |
4039 | |
4040 | switch (point.get_kind ()) |
4041 | { |
4042 | default: |
4043 | gcc_unreachable (); |
4044 | case PK_ORIGIN: |
4045 | /* This node exists to simplify finding the shortest path |
4046 | to an exploded_node. */ |
4047 | break; |
4048 | |
4049 | case PK_BEFORE_SUPERNODE: |
4050 | { |
4051 | program_state next_state (state); |
4052 | uncertainty_t uncertainty; |
4053 | |
4054 | if (point.get_from_edge ()) |
4055 | { |
4056 | impl_region_model_context ctxt (*this, node, |
4057 | &state, &next_state, |
4058 | &uncertainty, NULL, NULL); |
4059 | const cfg_superedge *last_cfg_superedge |
4060 | = point.get_from_edge ()->dyn_cast_cfg_superedge (); |
4061 | if (last_cfg_superedge) |
4062 | next_state.m_region_model->update_for_phis |
4063 | (snode: node->get_supernode (), |
4064 | last_cfg_superedge, |
4065 | ctxt: &ctxt); |
4066 | program_state::detect_leaks (src_state: state, dest_state: next_state, NULL, |
4067 | ext_state: get_ext_state (), ctxt: &ctxt); |
4068 | } |
4069 | |
4070 | program_point next_point (point.get_next ()); |
4071 | exploded_node *next = get_or_create_node (point: next_point, state: next_state, enode_for_diag: node); |
4072 | if (next) |
4073 | add_edge (src: node, dest: next, NULL, |
4074 | could_do_work: false); /* Assume no work is done at phi nodes. */ |
4075 | } |
4076 | break; |
4077 | case PK_BEFORE_STMT: |
4078 | { |
4079 | /* Determine the effect of a run of one or more statements |
4080 | within one supernode, generating an edge to the program_point |
4081 | after the last statement that's processed. |
4082 | |
4083 | Stop iterating statements and thus consolidating into one enode |
4084 | when: |
4085 | - reaching the end of the statements in the supernode |
4086 | - if an sm-state-change occurs (so that it gets its own |
4087 | exploded_node) |
4088 | - if "-fanalyzer-fine-grained" is active |
4089 | - encountering certain statements must appear at the start of |
4090 | their enode (for which stmt_requires_new_enode_p returns true) |
4091 | |
4092 | Update next_state in-place, to get the result of the one |
4093 | or more stmts that are processed. |
4094 | |
4095 | Split the node in-place if an sm-state-change occurs, so that |
4096 | the sm-state-change occurs on an edge where the src enode has |
4097 | exactly one stmt, the one that caused the change. */ |
4098 | program_state next_state (state); |
4099 | |
4100 | impl_path_context path_ctxt (&next_state, logger); |
4101 | |
4102 | bool could_have_done_work = false; |
4103 | uncertainty_t uncertainty; |
4104 | const supernode *snode = point.get_supernode (); |
4105 | unsigned stmt_idx; |
4106 | const gimple *prev_stmt = NULL; |
4107 | for (stmt_idx = point.get_stmt_idx (); |
4108 | stmt_idx < snode->m_stmts.length (); |
4109 | stmt_idx++) |
4110 | { |
4111 | const gimple *stmt = snode->m_stmts[stmt_idx]; |
4112 | |
4113 | if (stmt_idx > point.get_stmt_idx ()) |
4114 | if (stmt_requires_new_enode_p (stmt, prev_stmt)) |
4115 | { |
4116 | stmt_idx--; |
4117 | break; |
4118 | } |
4119 | prev_stmt = stmt; |
4120 | |
4121 | program_state old_state (next_state); |
4122 | |
4123 | /* Process the stmt. */ |
4124 | exploded_node::on_stmt_flags flags |
4125 | = node->on_stmt (eg&: *this, snode, stmt, state: &next_state, uncertainty: &uncertainty, |
4126 | out_could_have_done_work: &could_have_done_work, path_ctxt: &path_ctxt); |
4127 | node->m_num_processed_stmts++; |
4128 | |
4129 | /* If flags.m_terminate_path, stop analyzing; any nodes/edges |
4130 | will have been added by on_stmt (e.g. for handling longjmp). */ |
4131 | if (flags.m_terminate_path) |
4132 | return; |
4133 | |
4134 | if (next_state.m_region_model) |
4135 | { |
4136 | impl_region_model_context ctxt (*this, node, |
4137 | &old_state, &next_state, |
4138 | &uncertainty, NULL, stmt); |
4139 | program_state::detect_leaks (src_state: old_state, dest_state: next_state, NULL, |
4140 | ext_state: get_ext_state (), ctxt: &ctxt); |
4141 | } |
4142 | |
4143 | unsigned next_idx = stmt_idx + 1; |
4144 | program_point next_point |
4145 | = (next_idx < point.get_supernode ()->m_stmts.length () |
4146 | ? program_point::before_stmt (supernode: point.get_supernode (), stmt_idx: next_idx, |
4147 | call_string: point.get_call_string ()) |
4148 | : program_point::after_supernode (supernode: point.get_supernode (), |
4149 | call_string: point.get_call_string ())); |
4150 | next_state = next_state.prune_for_point (eg&: *this, point: next_point, enode_for_diag: node, |
4151 | uncertainty: &uncertainty); |
4152 | |
4153 | if (flag_analyzer_fine_grained |
4154 | || state_change_requires_new_enode_p (old_state, new_state: next_state) |
4155 | || path_ctxt.bifurcation_p () |
4156 | || path_ctxt.terminate_path_p ()) |
4157 | { |
4158 | program_point split_point |
4159 | = program_point::before_stmt (supernode: point.get_supernode (), |
4160 | stmt_idx, |
4161 | call_string: point.get_call_string ()); |
4162 | if (split_point != node->get_point ()) |
4163 | { |
4164 | /* If we're not at the start of NODE, split the enode at |
4165 | this stmt, so we have: |
4166 | node -> split_enode |
4167 | so that when split_enode is processed the next edge |
4168 | we add will be: |
4169 | split_enode -> next |
4170 | and any state change will effectively occur on that |
4171 | latter edge, and split_enode will contain just stmt. */ |
4172 | if (logger) |
4173 | logger->log (fmt: "getting split_enode" ); |
4174 | exploded_node *split_enode |
4175 | = get_or_create_node (point: split_point, state: old_state, enode_for_diag: node); |
4176 | if (!split_enode) |
4177 | return; |
4178 | /* "stmt" will be reprocessed when split_enode is |
4179 | processed. */ |
4180 | node->m_num_processed_stmts--; |
4181 | if (logger) |
4182 | logger->log (fmt: "creating edge to split_enode" ); |
4183 | add_edge (src: node, dest: split_enode, NULL, could_do_work: could_have_done_work); |
4184 | return; |
4185 | } |
4186 | else |
4187 | /* If we're at the start of NODE, stop iterating, |
4188 | so that an edge will be created from NODE to |
4189 | (next_point, next_state) below. */ |
4190 | break; |
4191 | } |
4192 | } |
4193 | unsigned next_idx = stmt_idx + 1; |
4194 | program_point next_point |
4195 | = (next_idx < point.get_supernode ()->m_stmts.length () |
4196 | ? program_point::before_stmt (supernode: point.get_supernode (), stmt_idx: next_idx, |
4197 | call_string: point.get_call_string ()) |
4198 | : program_point::after_supernode (supernode: point.get_supernode (), |
4199 | call_string: point.get_call_string ())); |
4200 | if (path_ctxt.terminate_path_p ()) |
4201 | { |
4202 | if (logger) |
4203 | logger->log (fmt: "not adding node: terminating path" ); |
4204 | } |
4205 | else |
4206 | { |
4207 | exploded_node *next |
4208 | = get_or_create_node (point: next_point, state: next_state, enode_for_diag: node); |
4209 | if (next) |
4210 | add_edge (src: node, dest: next, NULL, could_do_work: could_have_done_work); |
4211 | } |
4212 | |
4213 | /* If we have custom edge infos, "bifurcate" the state |
4214 | accordingly, potentially creating a new state/enode/eedge |
4215 | instances. For example, to handle a "realloc" call, we |
4216 | might split into 3 states, for the "failure", |
4217 | "resizing in place", and "moving to a new buffer" cases. */ |
4218 | for (auto edge_info_iter : path_ctxt.get_custom_eedge_infos ()) |
4219 | { |
4220 | /* Take ownership of the edge infos from the path_ctxt. */ |
4221 | std::unique_ptr<custom_edge_info> edge_info (edge_info_iter); |
4222 | if (logger) |
4223 | { |
4224 | logger->start_log_line (); |
4225 | logger->log_partial (fmt: "bifurcating for edge: " ); |
4226 | edge_info->print (pp: logger->get_printer ()); |
4227 | logger->end_log_line (); |
4228 | } |
4229 | program_state bifurcated_new_state |
4230 | (path_ctxt.get_state_at_bifurcation ()); |
4231 | |
4232 | /* Apply edge_info to state. */ |
4233 | impl_region_model_context |
4234 | bifurcation_ctxt (*this, |
4235 | node, // enode_for_diag |
4236 | &path_ctxt.get_state_at_bifurcation (), |
4237 | &bifurcated_new_state, |
4238 | NULL, // uncertainty_t *uncertainty |
4239 | NULL, // path_context *path_ctxt |
4240 | stmt); |
4241 | if (edge_info->update_state (state: &bifurcated_new_state, |
4242 | NULL, /* no exploded_edge yet. */ |
4243 | ctxt: &bifurcation_ctxt)) |
4244 | { |
4245 | exploded_node *next2 |
4246 | = get_or_create_node (point: next_point, state: bifurcated_new_state, enode_for_diag: node); |
4247 | if (next2) |
4248 | add_edge (src: node, dest: next2, NULL, |
4249 | could_do_work: true /* assume that work could be done */, |
4250 | custom_info: std::move (edge_info)); |
4251 | } |
4252 | else |
4253 | { |
4254 | if (logger) |
4255 | logger->log (fmt: "infeasible state, not adding node" ); |
4256 | } |
4257 | } |
4258 | } |
4259 | break; |
4260 | case PK_AFTER_SUPERNODE: |
4261 | { |
4262 | bool found_a_superedge = false; |
4263 | bool is_an_exit_block = false; |
4264 | /* If this is an EXIT BB, detect leaks, and potentially |
4265 | create a function summary. */ |
4266 | if (point.get_supernode ()->return_p ()) |
4267 | { |
4268 | is_an_exit_block = true; |
4269 | node->detect_leaks (eg&: *this); |
4270 | if (flag_analyzer_call_summaries |
4271 | && point.get_call_string ().empty_p ()) |
4272 | { |
4273 | /* TODO: create function summary |
4274 | There can be more than one; each corresponds to a different |
4275 | final enode in the function. */ |
4276 | if (logger) |
4277 | { |
4278 | pretty_printer *pp = logger->get_printer (); |
4279 | logger->start_log_line (); |
4280 | logger->log_partial |
4281 | (fmt: "would create function summary for %qE; state: " , |
4282 | point.get_fndecl ()); |
4283 | state.dump_to_pp (ext_state: m_ext_state, simple: true, multiline: false, pp); |
4284 | logger->end_log_line (); |
4285 | } |
4286 | per_function_data *per_fn_data |
4287 | = get_or_create_per_function_data (fun: point.get_function ()); |
4288 | per_fn_data->add_call_summary (node); |
4289 | } |
4290 | } |
4291 | /* Traverse into successors of the supernode. */ |
4292 | int i; |
4293 | superedge *succ; |
4294 | FOR_EACH_VEC_ELT (point.get_supernode ()->m_succs, i, succ) |
4295 | { |
4296 | found_a_superedge = true; |
4297 | if (logger) |
4298 | { |
4299 | label_text succ_desc (succ->get_description (user_facing: false)); |
4300 | logger->log (fmt: "considering SN: %i -> SN: %i (%s)" , |
4301 | succ->m_src->m_index, succ->m_dest->m_index, |
4302 | succ_desc.get ()); |
4303 | } |
4304 | |
4305 | program_point next_point |
4306 | = program_point::before_supernode (supernode: succ->m_dest, from_edge: succ, |
4307 | call_string: point.get_call_string ()); |
4308 | program_state next_state (state); |
4309 | uncertainty_t uncertainty; |
4310 | |
4311 | /* Make use the current state and try to discover and analyse |
4312 | indirect function calls (a call that doesn't have an underlying |
4313 | cgraph edge representing call). |
4314 | |
4315 | Some examples of such calls are virtual function calls |
4316 | and calls that happen via a function pointer. */ |
4317 | if (succ->m_kind == SUPEREDGE_INTRAPROCEDURAL_CALL |
4318 | && !(succ->get_any_callgraph_edge ())) |
4319 | { |
4320 | const gcall *call |
4321 | = point.get_supernode ()->get_final_call (); |
4322 | |
4323 | impl_region_model_context ctxt (*this, |
4324 | node, |
4325 | &state, |
4326 | &next_state, |
4327 | &uncertainty, |
4328 | NULL, |
4329 | point.get_stmt()); |
4330 | |
4331 | region_model *model = state.m_region_model; |
4332 | bool call_discovered = false; |
4333 | |
4334 | if (tree fn_decl = model->get_fndecl_for_call (call, ctxt: &ctxt)) |
4335 | call_discovered = maybe_create_dynamic_call (call, |
4336 | fn_decl, |
4337 | node, |
4338 | next_state, |
4339 | next_point, |
4340 | uncertainty: &uncertainty, |
4341 | logger); |
4342 | if (!call_discovered) |
4343 | { |
4344 | /* Check for jump through NULL. */ |
4345 | if (tree fn_ptr = gimple_call_fn (gs: call)) |
4346 | { |
4347 | const svalue *fn_ptr_sval |
4348 | = model->get_rvalue (expr: fn_ptr, ctxt: &ctxt); |
4349 | if (fn_ptr_sval->all_zeroes_p ()) |
4350 | ctxt.warn (d: make_unique<jump_through_null> (args&: call)); |
4351 | } |
4352 | |
4353 | /* An unknown function or a special function was called |
4354 | at this point, in such case, don't terminate the |
4355 | analysis of the current function. |
4356 | |
4357 | The analyzer handles calls to such functions while |
4358 | analysing the stmt itself, so the function call |
4359 | must have been handled by the anlyzer till now. */ |
4360 | exploded_node *next |
4361 | = get_or_create_node (point: next_point, |
4362 | state: next_state, |
4363 | enode_for_diag: node); |
4364 | if (next) |
4365 | add_edge (src: node, dest: next, sedge: succ, |
4366 | could_do_work: true /* assume that work is done */); |
4367 | } |
4368 | } |
4369 | |
4370 | if (!node->on_edge (eg&: *this, succ, next_point: &next_point, next_state: &next_state, |
4371 | uncertainty: &uncertainty)) |
4372 | { |
4373 | if (logger) |
4374 | logger->log (fmt: "skipping impossible edge to SN: %i" , |
4375 | succ->m_dest->m_index); |
4376 | continue; |
4377 | } |
4378 | exploded_node *next = get_or_create_node (point: next_point, state: next_state, |
4379 | enode_for_diag: node); |
4380 | if (next) |
4381 | { |
4382 | add_edge (src: node, dest: next, sedge: succ, could_do_work: false); |
4383 | |
4384 | /* We might have a function entrypoint. */ |
4385 | detect_infinite_recursion (enode: next); |
4386 | } |
4387 | } |
4388 | |
4389 | /* Return from the calls which doesn't have a return superedge. |
4390 | Such case occurs when GCC's middle end didn't knew which function to |
4391 | call but analyzer did. */ |
4392 | if ((is_an_exit_block && !found_a_superedge) |
4393 | && (!point.get_call_string ().empty_p ())) |
4394 | { |
4395 | const call_string &cs = point.get_call_string (); |
4396 | program_point next_point |
4397 | = program_point::before_supernode (supernode: cs.get_caller_node (), |
4398 | NULL, |
4399 | call_string: cs); |
4400 | program_state next_state (state); |
4401 | uncertainty_t uncertainty; |
4402 | |
4403 | const gcall *call |
4404 | = next_point.get_supernode ()->get_returning_call (); |
4405 | |
4406 | if (call) |
4407 | next_state.returning_call (eg&: *this, enode: node, call_stmt: call, uncertainty: &uncertainty); |
4408 | |
4409 | if (next_state.m_valid) |
4410 | { |
4411 | next_point.pop_from_call_stack (); |
4412 | exploded_node *enode = get_or_create_node (point: next_point, |
4413 | state: next_state, |
4414 | enode_for_diag: node); |
4415 | if (enode) |
4416 | add_edge (src: node, dest: enode, NULL, could_do_work: false, |
4417 | custom_info: make_unique<dynamic_call_info_t> (args&: call, args: true)); |
4418 | } |
4419 | } |
4420 | } |
4421 | break; |
4422 | } |
4423 | } |
4424 | |
4425 | /* Ensure that this graph has a stats instance for FN, return it. |
4426 | FN can be NULL, in which case a stats instances is returned covering |
4427 | "functionless" parts of the graph (the origin node). */ |
4428 | |
4429 | stats * |
4430 | exploded_graph::get_or_create_function_stats (function *fn) |
4431 | { |
4432 | if (!fn) |
4433 | return &m_functionless_stats; |
4434 | |
4435 | if (stats **slot = m_per_function_stats.get (k: fn)) |
4436 | return *slot; |
4437 | else |
4438 | { |
4439 | int num_supernodes = fn ? n_basic_blocks_for_fn (fn) : 0; |
4440 | /* not quite the num supernodes, but nearly. */ |
4441 | stats *new_stats = new stats (num_supernodes); |
4442 | m_per_function_stats.put (k: fn, v: new_stats); |
4443 | return new_stats; |
4444 | } |
4445 | } |
4446 | |
4447 | /* Print bar charts to PP showing: |
4448 | - the number of enodes per function, and |
4449 | - for each function: |
4450 | - the number of enodes per supernode/BB |
4451 | - the number of excess enodes per supernode/BB beyond the |
4452 | per-program-point limit, if there were any. */ |
4453 | |
4454 | void |
4455 | exploded_graph::print_bar_charts (pretty_printer *pp) const |
4456 | { |
4457 | cgraph_node *cgnode; |
4458 | |
4459 | pp_string (pp, "enodes per function:" ); |
4460 | pp_newline (pp); |
4461 | bar_chart enodes_per_function; |
4462 | FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (cgnode) |
4463 | { |
4464 | function *fn = cgnode->get_fun (); |
4465 | const stats * const *s_ptr |
4466 | = const_cast <function_stat_map_t &> (m_per_function_stats).get (k: fn); |
4467 | enodes_per_function.add_item (name: function_name (fn), |
4468 | value: s_ptr ? (*s_ptr)->get_total_enodes () : 0); |
4469 | } |
4470 | enodes_per_function.print (pp); |
4471 | |
4472 | /* Accumulate number of enodes per supernode. */ |
4473 | auto_vec<unsigned> enodes_per_supernode (m_sg.num_nodes ()); |
4474 | for (int i = 0; i < m_sg.num_nodes (); i++) |
4475 | enodes_per_supernode.quick_push (obj: 0); |
4476 | int i; |
4477 | exploded_node *enode; |
4478 | FOR_EACH_VEC_ELT (m_nodes, i, enode) |
4479 | { |
4480 | const supernode *iter_snode = enode->get_supernode (); |
4481 | if (!iter_snode) |
4482 | continue; |
4483 | enodes_per_supernode[iter_snode->m_index]++; |
4484 | } |
4485 | |
4486 | /* Accumulate excess enodes per supernode. */ |
4487 | auto_vec<unsigned> excess_enodes_per_supernode (m_sg.num_nodes ()); |
4488 | for (int i = 0; i < m_sg.num_nodes (); i++) |
4489 | excess_enodes_per_supernode.quick_push (obj: 0); |
4490 | for (point_map_t::iterator iter = m_per_point_data.begin (); |
4491 | iter != m_per_point_data.end (); ++iter) |
4492 | { |
4493 | const program_point *point = (*iter).first; |
4494 | const supernode *iter_snode = point->get_supernode (); |
4495 | if (!iter_snode) |
4496 | continue; |
4497 | const per_program_point_data *point_data = (*iter).second; |
4498 | excess_enodes_per_supernode[iter_snode->m_index] |
4499 | += point_data->m_excess_enodes; |
4500 | } |
4501 | |
4502 | /* Show per-function bar_charts of enodes per supernode/BB. */ |
4503 | pp_string (pp, "per-function enodes per supernode/BB:" ); |
4504 | pp_newline (pp); |
4505 | FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (cgnode) |
4506 | { |
4507 | function *fn = cgnode->get_fun (); |
4508 | pp_printf (pp, "function: %qs" , function_name (fn)); |
4509 | pp_newline (pp); |
4510 | |
4511 | bar_chart enodes_per_snode; |
4512 | bar_chart excess_enodes_per_snode; |
4513 | bool have_excess_enodes = false; |
4514 | for (int i = 0; i < m_sg.num_nodes (); i++) |
4515 | { |
4516 | const supernode *iter_snode = m_sg.get_node_by_index (idx: i); |
4517 | if (iter_snode->get_function () != fn) |
4518 | continue; |
4519 | pretty_printer tmp_pp; |
4520 | pp_printf (&tmp_pp, "sn %i (bb %i)" , |
4521 | iter_snode->m_index, iter_snode->m_bb->index); |
4522 | enodes_per_snode.add_item (name: pp_formatted_text (&tmp_pp), |
4523 | value: enodes_per_supernode[iter_snode->m_index]); |
4524 | const int num_excess |
4525 | = excess_enodes_per_supernode[iter_snode->m_index]; |
4526 | excess_enodes_per_snode.add_item (name: pp_formatted_text (&tmp_pp), |
4527 | value: num_excess); |
4528 | if (num_excess) |
4529 | have_excess_enodes = true; |
4530 | } |
4531 | enodes_per_snode.print (pp); |
4532 | if (have_excess_enodes) |
4533 | { |
4534 | pp_printf (pp, "EXCESS ENODES:" ); |
4535 | pp_newline (pp); |
4536 | excess_enodes_per_snode.print (pp); |
4537 | } |
4538 | } |
4539 | } |
4540 | |
4541 | /* Write all stats information to this graph's logger, if any. */ |
4542 | |
4543 | void |
4544 | exploded_graph::log_stats () const |
4545 | { |
4546 | logger * const logger = get_logger (); |
4547 | if (!logger) |
4548 | return; |
4549 | |
4550 | LOG_SCOPE (logger); |
4551 | |
4552 | m_ext_state.get_engine ()->log_stats (logger); |
4553 | |
4554 | logger->log (fmt: "m_sg.num_nodes (): %i" , m_sg.num_nodes ()); |
4555 | logger->log (fmt: "m_nodes.length (): %i" , m_nodes.length ()); |
4556 | logger->log (fmt: "m_edges.length (): %i" , m_edges.length ()); |
4557 | logger->log (fmt: "remaining enodes in worklist: %i" , m_worklist.length ()); |
4558 | |
4559 | logger->log (fmt: "global stats:" ); |
4560 | m_global_stats.log (logger); |
4561 | |
4562 | for (function_stat_map_t::iterator iter = m_per_function_stats.begin (); |
4563 | iter != m_per_function_stats.end (); |
4564 | ++iter) |
4565 | { |
4566 | function *fn = (*iter).first; |
4567 | log_scope s (logger, function_name (fn)); |
4568 | (*iter).second->log (logger); |
4569 | } |
4570 | |
4571 | print_bar_charts (pp: logger->get_printer ()); |
4572 | } |
4573 | |
4574 | /* Dump all stats information to OUT. */ |
4575 | |
4576 | void |
4577 | exploded_graph::dump_stats (FILE *out) const |
4578 | { |
4579 | fprintf (stream: out, format: "m_sg.num_nodes (): %i\n" , m_sg.num_nodes ()); |
4580 | fprintf (stream: out, format: "m_nodes.length (): %i\n" , m_nodes.length ()); |
4581 | fprintf (stream: out, format: "m_edges.length (): %i\n" , m_edges.length ()); |
4582 | fprintf (stream: out, format: "remaining enodes in worklist: %i" , m_worklist.length ()); |
4583 | |
4584 | fprintf (stream: out, format: "global stats:\n" ); |
4585 | m_global_stats.dump (out); |
4586 | |
4587 | for (function_stat_map_t::iterator iter = m_per_function_stats.begin (); |
4588 | iter != m_per_function_stats.end (); |
4589 | ++iter) |
4590 | { |
4591 | function *fn = (*iter).first; |
4592 | fprintf (stream: out, format: "function: %s\n" , function_name (fn)); |
4593 | (*iter).second->dump (out); |
4594 | } |
4595 | |
4596 | fprintf (stream: out, format: "PK_AFTER_SUPERNODE per supernode:\n" ); |
4597 | for (unsigned i = 0; i < m_PK_AFTER_SUPERNODE_per_snode.length (); i++) |
4598 | fprintf (stream: out, format: " SN %i: %3i\n" , i, m_PK_AFTER_SUPERNODE_per_snode[i]); |
4599 | } |
4600 | |
4601 | void |
4602 | exploded_graph::dump_states_for_supernode (FILE *out, |
4603 | const supernode *snode) const |
4604 | { |
4605 | fprintf (stream: out, format: "PK_AFTER_SUPERNODE nodes for SN: %i\n" , snode->m_index); |
4606 | int i; |
4607 | exploded_node *enode; |
4608 | int state_idx = 0; |
4609 | FOR_EACH_VEC_ELT (m_nodes, i, enode) |
4610 | { |
4611 | const supernode *iter_snode = enode->get_supernode (); |
4612 | if (enode->get_point ().get_kind () == PK_AFTER_SUPERNODE |
4613 | && iter_snode == snode) |
4614 | { |
4615 | pretty_printer pp; |
4616 | pp_format_decoder (&pp) = default_tree_printer; |
4617 | enode->get_state ().dump_to_pp (ext_state: m_ext_state, simple: true, multiline: false, pp: &pp); |
4618 | fprintf (stream: out, format: "state %i: EN: %i\n %s\n" , |
4619 | state_idx++, enode->m_index, |
4620 | pp_formatted_text (&pp)); |
4621 | } |
4622 | } |
4623 | fprintf (stream: out, format: "#exploded_node for PK_AFTER_SUPERNODE for SN: %i = %i\n" , |
4624 | snode->m_index, state_idx); |
4625 | } |
4626 | |
4627 | /* Return a new json::object of the form |
4628 | {"nodes" : [objs for enodes], |
4629 | "edges" : [objs for eedges], |
4630 | "ext_state": object for extrinsic_state, |
4631 | "diagnostic_manager": object for diagnostic_manager}. */ |
4632 | |
4633 | json::object * |
4634 | exploded_graph::to_json () const |
4635 | { |
4636 | json::object *egraph_obj = new json::object (); |
4637 | |
4638 | /* Nodes. */ |
4639 | { |
4640 | json::array *nodes_arr = new json::array (); |
4641 | unsigned i; |
4642 | exploded_node *n; |
4643 | FOR_EACH_VEC_ELT (m_nodes, i, n) |
4644 | nodes_arr->append (v: n->to_json (ext_state: m_ext_state)); |
4645 | egraph_obj->set (key: "nodes" , v: nodes_arr); |
4646 | } |
4647 | |
4648 | /* Edges. */ |
4649 | { |
4650 | json::array *edges_arr = new json::array (); |
4651 | unsigned i; |
4652 | exploded_edge *n; |
4653 | FOR_EACH_VEC_ELT (m_edges, i, n) |
4654 | edges_arr->append (v: n->to_json ()); |
4655 | egraph_obj->set (key: "edges" , v: edges_arr); |
4656 | } |
4657 | |
4658 | /* m_sg is JSONified at the top-level. */ |
4659 | |
4660 | egraph_obj->set (key: "ext_state" , v: m_ext_state.to_json ()); |
4661 | egraph_obj->set (key: "worklist" , v: m_worklist.to_json ()); |
4662 | egraph_obj->set (key: "diagnostic_manager" , v: m_diagnostic_manager.to_json ()); |
4663 | |
4664 | /* The following fields aren't yet being JSONified: |
4665 | const state_purge_map *const m_purge_map; |
4666 | const analysis_plan &m_plan; |
4667 | stats m_global_stats; |
4668 | function_stat_map_t m_per_function_stats; |
4669 | stats m_functionless_stats; |
4670 | call_string_data_map_t m_per_call_string_data; |
4671 | auto_vec<int> m_PK_AFTER_SUPERNODE_per_snode; */ |
4672 | |
4673 | return egraph_obj; |
4674 | } |
4675 | |
4676 | /* class exploded_path. */ |
4677 | |
4678 | /* Copy ctor. */ |
4679 | |
4680 | exploded_path::exploded_path (const exploded_path &other) |
4681 | : m_edges (other.m_edges.length ()) |
4682 | { |
4683 | int i; |
4684 | const exploded_edge *eedge; |
4685 | FOR_EACH_VEC_ELT (other.m_edges, i, eedge) |
4686 | m_edges.quick_push (obj: eedge); |
4687 | } |
4688 | |
4689 | /* Look for the last use of SEARCH_STMT within this path. |
4690 | If found write the edge's index to *OUT_IDX and return true, otherwise |
4691 | return false. */ |
4692 | |
4693 | bool |
4694 | exploded_path::find_stmt_backwards (const gimple *search_stmt, |
4695 | int *out_idx) const |
4696 | { |
4697 | int i; |
4698 | const exploded_edge *eedge; |
4699 | FOR_EACH_VEC_ELT_REVERSE (m_edges, i, eedge) |
4700 | { |
4701 | const exploded_node *dst_node = eedge->m_dest; |
4702 | const program_point &dst_point = dst_node->get_point (); |
4703 | const gimple *stmt = dst_point.get_stmt (); |
4704 | if (stmt == search_stmt) |
4705 | { |
4706 | *out_idx = i; |
4707 | return true; |
4708 | } |
4709 | } |
4710 | return false; |
4711 | } |
4712 | |
4713 | /* Get the final exploded_node in this path, which must be non-empty. */ |
4714 | |
4715 | exploded_node * |
4716 | exploded_path::get_final_enode () const |
4717 | { |
4718 | gcc_assert (m_edges.length () > 0); |
4719 | return m_edges[m_edges.length () - 1]->m_dest; |
4720 | } |
4721 | |
4722 | /* Check state along this path, returning true if it is feasible. |
4723 | If OUT is non-NULL, and the path is infeasible, write a new |
4724 | feasibility_problem to *OUT. */ |
4725 | |
4726 | bool |
4727 | exploded_path::feasible_p (logger *logger, |
4728 | std::unique_ptr<feasibility_problem> *out, |
4729 | engine *eng, const exploded_graph *eg) const |
4730 | { |
4731 | LOG_SCOPE (logger); |
4732 | |
4733 | feasibility_state state (eng->get_model_manager (), |
4734 | eg->get_supergraph ()); |
4735 | |
4736 | /* Traverse the path, updating this state. */ |
4737 | for (unsigned edge_idx = 0; edge_idx < m_edges.length (); edge_idx++) |
4738 | { |
4739 | const exploded_edge *eedge = m_edges[edge_idx]; |
4740 | if (logger) |
4741 | logger->log (fmt: "considering edge %i: EN:%i -> EN:%i" , |
4742 | edge_idx, |
4743 | eedge->m_src->m_index, |
4744 | eedge->m_dest->m_index); |
4745 | |
4746 | std::unique_ptr <rejected_constraint> rc; |
4747 | if (!state.maybe_update_for_edge (logger, eedge, ctxt: nullptr, out_rc: &rc)) |
4748 | { |
4749 | gcc_assert (rc); |
4750 | if (out) |
4751 | { |
4752 | const exploded_node &src_enode = *eedge->m_src; |
4753 | const program_point &src_point = src_enode.get_point (); |
4754 | const gimple *last_stmt |
4755 | = src_point.get_supernode ()->get_last_stmt (); |
4756 | *out = ::make_unique<feasibility_problem> (args&: edge_idx, args: *eedge, |
4757 | args&: last_stmt, |
4758 | args: std::move (rc)); |
4759 | } |
4760 | return false; |
4761 | } |
4762 | |
4763 | if (logger) |
4764 | { |
4765 | logger->log (fmt: "state after edge %i: EN:%i -> EN:%i" , |
4766 | edge_idx, |
4767 | eedge->m_src->m_index, |
4768 | eedge->m_dest->m_index); |
4769 | logger->start_log_line (); |
4770 | state.get_model ().dump_to_pp (pp: logger->get_printer (), simple: true, multiline: false); |
4771 | logger->end_log_line (); |
4772 | } |
4773 | } |
4774 | |
4775 | return true; |
4776 | } |
4777 | |
4778 | /* Dump this path in multiline form to PP. |
4779 | If EXT_STATE is non-NULL, then show the nodes. */ |
4780 | |
4781 | void |
4782 | exploded_path::dump_to_pp (pretty_printer *pp, |
4783 | const extrinsic_state *ext_state) const |
4784 | { |
4785 | for (unsigned i = 0; i < m_edges.length (); i++) |
4786 | { |
4787 | const exploded_edge *eedge = m_edges[i]; |
4788 | pp_printf (pp, "m_edges[%i]: EN %i -> EN %i" , |
4789 | i, |
4790 | eedge->m_src->m_index, |
4791 | eedge->m_dest->m_index); |
4792 | pp_newline (pp); |
4793 | |
4794 | if (ext_state) |
4795 | eedge->m_dest->dump_to_pp (pp, ext_state: *ext_state); |
4796 | } |
4797 | } |
4798 | |
4799 | /* Dump this path in multiline form to FP. */ |
4800 | |
4801 | void |
4802 | exploded_path::dump (FILE *fp, const extrinsic_state *ext_state) const |
4803 | { |
4804 | pretty_printer pp; |
4805 | pp_format_decoder (&pp) = default_tree_printer; |
4806 | pp_show_color (&pp) = pp_show_color (global_dc->printer); |
4807 | pp.buffer->stream = fp; |
4808 | dump_to_pp (pp: &pp, ext_state); |
4809 | pp_flush (&pp); |
4810 | } |
4811 | |
4812 | /* Dump this path in multiline form to stderr. */ |
4813 | |
4814 | DEBUG_FUNCTION void |
4815 | exploded_path::dump (const extrinsic_state *ext_state) const |
4816 | { |
4817 | dump (stderr, ext_state); |
4818 | } |
4819 | |
4820 | /* Dump this path verbosely to FILENAME. */ |
4821 | |
4822 | void |
4823 | exploded_path::dump_to_file (const char *filename, |
4824 | const extrinsic_state &ext_state) const |
4825 | { |
4826 | FILE *fp = fopen (filename: filename, modes: "w" ); |
4827 | if (!fp) |
4828 | return; |
4829 | pretty_printer pp; |
4830 | pp_format_decoder (&pp) = default_tree_printer; |
4831 | pp.buffer->stream = fp; |
4832 | dump_to_pp (pp: &pp, ext_state: &ext_state); |
4833 | pp_flush (&pp); |
4834 | fclose (stream: fp); |
4835 | } |
4836 | |
4837 | /* class feasibility_problem. */ |
4838 | |
4839 | void |
4840 | feasibility_problem::dump_to_pp (pretty_printer *pp) const |
4841 | { |
4842 | pp_printf (pp, "edge from EN: %i to EN: %i" , |
4843 | m_eedge.m_src->m_index, m_eedge.m_dest->m_index); |
4844 | if (m_rc) |
4845 | { |
4846 | pp_string (pp, "; rejected constraint: " ); |
4847 | m_rc->dump_to_pp (pp); |
4848 | pp_string (pp, "; rmodel: " ); |
4849 | m_rc->get_model ().dump_to_pp (pp, simple: true, multiline: false); |
4850 | } |
4851 | } |
4852 | |
4853 | /* class feasibility_state. */ |
4854 | |
4855 | /* Ctor for feasibility_state, at the beginning of a path. */ |
4856 | |
4857 | feasibility_state::feasibility_state (region_model_manager *manager, |
4858 | const supergraph &sg) |
4859 | : m_model (manager), |
4860 | m_snodes_visited (sg.m_nodes.length ()) |
4861 | { |
4862 | bitmap_clear (m_snodes_visited); |
4863 | } |
4864 | |
4865 | /* Copy ctor for feasibility_state, for extending a path. */ |
4866 | |
4867 | feasibility_state::feasibility_state (const feasibility_state &other) |
4868 | : m_model (other.m_model), |
4869 | m_snodes_visited (const_sbitmap (other.m_snodes_visited)->n_bits) |
4870 | { |
4871 | bitmap_copy (m_snodes_visited, other.m_snodes_visited); |
4872 | } |
4873 | |
4874 | feasibility_state::feasibility_state (const region_model &model, |
4875 | const supergraph &sg) |
4876 | : m_model (model), |
4877 | m_snodes_visited (sg.m_nodes.length ()) |
4878 | { |
4879 | bitmap_clear (m_snodes_visited); |
4880 | } |
4881 | |
4882 | feasibility_state & |
4883 | feasibility_state::operator= (const feasibility_state &other) |
4884 | { |
4885 | m_model = other.m_model; |
4886 | bitmap_copy (m_snodes_visited, other.m_snodes_visited); |
4887 | return *this; |
4888 | } |
4889 | |
4890 | /* The heart of feasibility-checking. |
4891 | |
4892 | Attempt to update this state in-place based on traversing EEDGE |
4893 | in a path. |
4894 | Update the model for the stmts in the src enode. |
4895 | Attempt to add constraints for EEDGE. |
4896 | |
4897 | If feasible, return true. |
4898 | Otherwise, return false and write to *OUT_RC. */ |
4899 | |
4900 | bool |
4901 | feasibility_state:: |
4902 | maybe_update_for_edge (logger *logger, |
4903 | const exploded_edge *eedge, |
4904 | region_model_context *ctxt, |
4905 | std::unique_ptr<rejected_constraint> *out_rc) |
4906 | { |
4907 | const exploded_node &src_enode = *eedge->m_src; |
4908 | const program_point &src_point = src_enode.get_point (); |
4909 | if (logger) |
4910 | { |
4911 | logger->start_log_line (); |
4912 | src_point.print (pp: logger->get_printer (), f: format (false)); |
4913 | logger->end_log_line (); |
4914 | } |
4915 | |
4916 | /* Update state for the stmts that were processed in each enode. */ |
4917 | for (unsigned stmt_idx = 0; stmt_idx < src_enode.m_num_processed_stmts; |
4918 | stmt_idx++) |
4919 | { |
4920 | const gimple *stmt = src_enode.get_processed_stmt (idx: stmt_idx); |
4921 | |
4922 | /* Update cfun and input_location in case of ICE: make it easier to |
4923 | track down which source construct we're failing to handle. */ |
4924 | auto_cfun sentinel (src_point.get_function ()); |
4925 | input_location = stmt->location; |
4926 | |
4927 | update_for_stmt (stmt); |
4928 | } |
4929 | |
4930 | const superedge *sedge = eedge->m_sedge; |
4931 | if (sedge) |
4932 | { |
4933 | if (logger) |
4934 | { |
4935 | label_text desc (sedge->get_description (user_facing: false)); |
4936 | logger->log (fmt: " sedge: SN:%i -> SN:%i %s" , |
4937 | sedge->m_src->m_index, |
4938 | sedge->m_dest->m_index, |
4939 | desc.get ()); |
4940 | } |
4941 | |
4942 | const gimple *last_stmt = src_point.get_supernode ()->get_last_stmt (); |
4943 | if (!m_model.maybe_update_for_edge (edge: *sedge, last_stmt, ctxt, out: out_rc)) |
4944 | { |
4945 | if (logger) |
4946 | { |
4947 | logger->start_log_line (); |
4948 | logger->log_partial (fmt: "rejecting due to region model: " ); |
4949 | m_model.dump_to_pp (pp: logger->get_printer (), simple: true, multiline: false); |
4950 | logger->end_log_line (); |
4951 | } |
4952 | return false; |
4953 | } |
4954 | } |
4955 | else |
4956 | { |
4957 | /* Special-case the initial eedge from the origin node to the |
4958 | initial function by pushing a frame for it. */ |
4959 | if (src_point.get_kind () == PK_ORIGIN) |
4960 | { |
4961 | gcc_assert (eedge->m_src->m_index == 0); |
4962 | gcc_assert (eedge->m_dest->get_point ().get_kind () |
4963 | == PK_BEFORE_SUPERNODE); |
4964 | function *fun = eedge->m_dest->get_function (); |
4965 | gcc_assert (fun); |
4966 | m_model.push_frame (fun: *fun, NULL, ctxt); |
4967 | if (logger) |
4968 | logger->log (fmt: " pushing frame for %qD" , fun->decl); |
4969 | } |
4970 | else if (eedge->m_custom_info) |
4971 | { |
4972 | eedge->m_custom_info->update_model (model: &m_model, eedge, ctxt); |
4973 | } |
4974 | } |
4975 | |
4976 | /* Handle phi nodes on an edge leaving a PK_BEFORE_SUPERNODE (to |
4977 | a PK_BEFORE_STMT, or a PK_AFTER_SUPERNODE if no stmts). |
4978 | This will typically not be associated with a superedge. */ |
4979 | if (src_point.get_from_edge ()) |
4980 | { |
4981 | const cfg_superedge *last_cfg_superedge |
4982 | = src_point.get_from_edge ()->dyn_cast_cfg_superedge (); |
4983 | const exploded_node &dst_enode = *eedge->m_dest; |
4984 | const unsigned dst_snode_idx = dst_enode.get_supernode ()->m_index; |
4985 | if (last_cfg_superedge) |
4986 | { |
4987 | if (logger) |
4988 | logger->log (fmt: " update for phis" ); |
4989 | m_model.update_for_phis (snode: src_enode.get_supernode (), |
4990 | last_cfg_superedge, |
4991 | ctxt); |
4992 | /* If we've entering an snode that we've already visited on this |
4993 | epath, then we need do fix things up for loops; see the |
4994 | comment for store::loop_replay_fixup. |
4995 | Perhaps we should probably also verify the callstring, |
4996 | and track program_points, but hopefully doing it by supernode |
4997 | is good enough. */ |
4998 | if (bitmap_bit_p (map: m_snodes_visited, bitno: dst_snode_idx)) |
4999 | m_model.loop_replay_fixup (dst_state: dst_enode.get_state ().m_region_model); |
5000 | } |
5001 | bitmap_set_bit (map: m_snodes_visited, bitno: dst_snode_idx); |
5002 | } |
5003 | return true; |
5004 | } |
5005 | |
5006 | /* Update this object for the effects of STMT. */ |
5007 | |
5008 | void |
5009 | feasibility_state::update_for_stmt (const gimple *stmt) |
5010 | { |
5011 | if (const gassign *assign = dyn_cast <const gassign *> (p: stmt)) |
5012 | m_model.on_assignment (stmt: assign, NULL); |
5013 | else if (const gasm *asm_stmt = dyn_cast <const gasm *> (p: stmt)) |
5014 | m_model.on_asm_stmt (asm_stmt, NULL); |
5015 | else if (const gcall *call = dyn_cast <const gcall *> (p: stmt)) |
5016 | { |
5017 | bool unknown_side_effects = m_model.on_call_pre (stmt: call, NULL); |
5018 | m_model.on_call_post (stmt: call, unknown_side_effects, NULL); |
5019 | } |
5020 | else if (const greturn *return_ = dyn_cast <const greturn *> (p: stmt)) |
5021 | m_model.on_return (stmt: return_, NULL); |
5022 | } |
5023 | |
5024 | /* Dump this object to PP. */ |
5025 | |
5026 | void |
5027 | feasibility_state::dump_to_pp (pretty_printer *pp, |
5028 | bool simple, bool multiline) const |
5029 | { |
5030 | m_model.dump_to_pp (pp, simple, multiline); |
5031 | } |
5032 | |
5033 | /* A family of cluster subclasses for use when generating .dot output for |
5034 | exploded graphs (-fdump-analyzer-exploded-graph), for grouping the |
5035 | enodes into hierarchical boxes. |
5036 | |
5037 | All functionless enodes appear in the top-level graph. |
5038 | Every (function, call_string) pair gets its own cluster. Within that |
5039 | cluster, each supernode gets its own cluster. |
5040 | |
5041 | Hence all enodes relating to a particular function with a particular |
5042 | callstring will be in a cluster together; all enodes for the same |
5043 | function but with a different callstring will be in a different |
5044 | cluster. */ |
5045 | |
5046 | /* Base class of cluster for clustering exploded_node instances in .dot |
5047 | output, based on various subclass-specific criteria. */ |
5048 | |
5049 | class exploded_cluster : public cluster<eg_traits> |
5050 | { |
5051 | }; |
5052 | |
5053 | /* Cluster containing all exploded_node instances for one supernode. */ |
5054 | |
5055 | class supernode_cluster : public exploded_cluster |
5056 | { |
5057 | public: |
5058 | supernode_cluster (const supernode *supernode) : m_supernode (supernode) {} |
5059 | |
5060 | // TODO: dtor? |
5061 | |
5062 | void dump_dot (graphviz_out *gv, const dump_args_t &args) const final override |
5063 | { |
5064 | gv->println (fmt: "subgraph \"cluster_supernode_%i\" {" , m_supernode->m_index); |
5065 | gv->indent (); |
5066 | gv->println (fmt: "style=\"dashed\";" ); |
5067 | gv->println (fmt: "label=\"SN: %i (bb: %i; scc: %i)\";" , |
5068 | m_supernode->m_index, m_supernode->m_bb->index, |
5069 | args.m_eg.get_scc_id (node: *m_supernode)); |
5070 | |
5071 | int i; |
5072 | exploded_node *enode; |
5073 | FOR_EACH_VEC_ELT (m_enodes, i, enode) |
5074 | enode->dump_dot (gv, args); |
5075 | |
5076 | /* Terminate subgraph. */ |
5077 | gv->outdent (); |
5078 | gv->println (fmt: "}" ); |
5079 | } |
5080 | |
5081 | void add_node (exploded_node *en) final override |
5082 | { |
5083 | m_enodes.safe_push (obj: en); |
5084 | } |
5085 | |
5086 | /* Comparator for use by auto_vec<supernode_cluster *>::qsort. */ |
5087 | |
5088 | static int cmp_ptr_ptr (const void *p1, const void *p2) |
5089 | { |
5090 | const supernode_cluster *c1 |
5091 | = *(const supernode_cluster * const *)p1; |
5092 | const supernode_cluster *c2 |
5093 | = *(const supernode_cluster * const *)p2; |
5094 | return c1->m_supernode->m_index - c2->m_supernode->m_index; |
5095 | } |
5096 | |
5097 | private: |
5098 | const supernode *m_supernode; |
5099 | auto_vec <exploded_node *> m_enodes; |
5100 | }; |
5101 | |
5102 | /* Cluster containing all supernode_cluster instances for one |
5103 | (function, call_string) pair. */ |
5104 | |
5105 | class function_call_string_cluster : public exploded_cluster |
5106 | { |
5107 | public: |
5108 | function_call_string_cluster (function *fun, const call_string &cs) |
5109 | : m_fun (fun), m_cs (cs) {} |
5110 | |
5111 | ~function_call_string_cluster () |
5112 | { |
5113 | for (map_t::iterator iter = m_map.begin (); |
5114 | iter != m_map.end (); |
5115 | ++iter) |
5116 | delete (*iter).second; |
5117 | } |
5118 | |
5119 | void dump_dot (graphviz_out *gv, const dump_args_t &args) const final override |
5120 | { |
5121 | const char *funcname = function_name (m_fun); |
5122 | |
5123 | gv->println (fmt: "subgraph \"cluster_function_%s\" {" , |
5124 | IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (m_fun->decl))); |
5125 | gv->indent (); |
5126 | gv->write_indent (); |
5127 | gv->print (fmt: "label=\"call string: " ); |
5128 | m_cs.print (pp: gv->get_pp ()); |
5129 | gv->print (fmt: " function: %s \";" , funcname); |
5130 | gv->print (fmt: "\n" ); |
5131 | |
5132 | /* Dump m_map, sorting it to avoid churn when comparing dumps. */ |
5133 | auto_vec<supernode_cluster *> child_clusters (m_map.elements ()); |
5134 | for (map_t::iterator iter = m_map.begin (); |
5135 | iter != m_map.end (); |
5136 | ++iter) |
5137 | child_clusters.quick_push (obj: (*iter).second); |
5138 | |
5139 | child_clusters.qsort (supernode_cluster::cmp_ptr_ptr); |
5140 | |
5141 | unsigned i; |
5142 | supernode_cluster *child_cluster; |
5143 | FOR_EACH_VEC_ELT (child_clusters, i, child_cluster) |
5144 | child_cluster->dump_dot (gv, args); |
5145 | |
5146 | /* Terminate subgraph. */ |
5147 | gv->outdent (); |
5148 | gv->println (fmt: "}" ); |
5149 | } |
5150 | |
5151 | void add_node (exploded_node *en) final override |
5152 | { |
5153 | const supernode *supernode = en->get_supernode (); |
5154 | gcc_assert (supernode); |
5155 | supernode_cluster **slot = m_map.get (k: supernode); |
5156 | if (slot) |
5157 | (*slot)->add_node (en); |
5158 | else |
5159 | { |
5160 | supernode_cluster *child = new supernode_cluster (supernode); |
5161 | m_map.put (k: supernode, v: child); |
5162 | child->add_node (en); |
5163 | } |
5164 | } |
5165 | |
5166 | /* Comparator for use by auto_vec<function_call_string_cluster *>. */ |
5167 | |
5168 | static int cmp_ptr_ptr (const void *p1, const void *p2) |
5169 | { |
5170 | const function_call_string_cluster *c1 |
5171 | = *(const function_call_string_cluster * const *)p1; |
5172 | const function_call_string_cluster *c2 |
5173 | = *(const function_call_string_cluster * const *)p2; |
5174 | if (int cmp_names |
5175 | = strcmp (IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (c1->m_fun->decl)), |
5176 | IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (c2->m_fun->decl)))) |
5177 | return cmp_names; |
5178 | return call_string::cmp (a: c1->m_cs, b: c2->m_cs); |
5179 | } |
5180 | |
5181 | private: |
5182 | function *m_fun; |
5183 | const call_string &m_cs; |
5184 | typedef ordered_hash_map<const supernode *, supernode_cluster *> map_t; |
5185 | map_t m_map; |
5186 | }; |
5187 | |
5188 | /* Keys for root_cluster. */ |
5189 | |
5190 | struct function_call_string |
5191 | { |
5192 | function_call_string (function *fun, const call_string *cs) |
5193 | : m_fun (fun), m_cs (cs) |
5194 | { |
5195 | gcc_assert (fun); |
5196 | gcc_assert (cs); |
5197 | } |
5198 | |
5199 | function *m_fun; |
5200 | const call_string *m_cs; |
5201 | }; |
5202 | |
5203 | } // namespace ana |
5204 | |
5205 | template <> struct default_hash_traits<function_call_string> |
5206 | : public pod_hash_traits<function_call_string> |
5207 | { |
5208 | static const bool empty_zero_p = false; |
5209 | }; |
5210 | |
5211 | template <> |
5212 | inline hashval_t |
5213 | pod_hash_traits<function_call_string>::hash (value_type v) |
5214 | { |
5215 | return (pointer_hash <function>::hash (candidate: v.m_fun) |
5216 | ^ pointer_hash <const call_string>::hash (candidate: v.m_cs)); |
5217 | } |
5218 | |
5219 | template <> |
5220 | inline bool |
5221 | pod_hash_traits<function_call_string>::equal (const value_type &existing, |
5222 | const value_type &candidate) |
5223 | { |
5224 | return existing.m_fun == candidate.m_fun && &existing.m_cs == &candidate.m_cs; |
5225 | } |
5226 | template <> |
5227 | inline void |
5228 | pod_hash_traits<function_call_string>::mark_deleted (value_type &v) |
5229 | { |
5230 | v.m_fun = reinterpret_cast<function *> (1); |
5231 | } |
5232 | template <> |
5233 | inline void |
5234 | pod_hash_traits<function_call_string>::mark_empty (value_type &v) |
5235 | { |
5236 | v.m_fun = NULL; |
5237 | } |
5238 | template <> |
5239 | inline bool |
5240 | pod_hash_traits<function_call_string>::is_deleted (value_type v) |
5241 | { |
5242 | return v.m_fun == reinterpret_cast<function *> (1); |
5243 | } |
5244 | template <> |
5245 | inline bool |
5246 | pod_hash_traits<function_call_string>::is_empty (value_type v) |
5247 | { |
5248 | return v.m_fun == NULL; |
5249 | } |
5250 | |
5251 | namespace ana { |
5252 | |
5253 | /* Top-level cluster for generating .dot output for exploded graphs, |
5254 | handling the functionless nodes, and grouping the remaining nodes by |
5255 | callstring. */ |
5256 | |
5257 | class root_cluster : public exploded_cluster |
5258 | { |
5259 | public: |
5260 | ~root_cluster () |
5261 | { |
5262 | for (map_t::iterator iter = m_map.begin (); |
5263 | iter != m_map.end (); |
5264 | ++iter) |
5265 | delete (*iter).second; |
5266 | } |
5267 | |
5268 | void dump_dot (graphviz_out *gv, const dump_args_t &args) const final override |
5269 | { |
5270 | int i; |
5271 | exploded_node *enode; |
5272 | FOR_EACH_VEC_ELT (m_functionless_enodes, i, enode) |
5273 | enode->dump_dot (gv, args); |
5274 | |
5275 | /* Dump m_map, sorting it to avoid churn when comparing dumps. */ |
5276 | auto_vec<function_call_string_cluster *> child_clusters (m_map.elements ()); |
5277 | for (map_t::iterator iter = m_map.begin (); |
5278 | iter != m_map.end (); |
5279 | ++iter) |
5280 | child_clusters.quick_push (obj: (*iter).second); |
5281 | |
5282 | child_clusters.qsort (function_call_string_cluster::cmp_ptr_ptr); |
5283 | |
5284 | function_call_string_cluster *child_cluster; |
5285 | FOR_EACH_VEC_ELT (child_clusters, i, child_cluster) |
5286 | child_cluster->dump_dot (gv, args); |
5287 | } |
5288 | |
5289 | void add_node (exploded_node *en) final override |
5290 | { |
5291 | function *fun = en->get_function (); |
5292 | if (!fun) |
5293 | { |
5294 | m_functionless_enodes.safe_push (obj: en); |
5295 | return; |
5296 | } |
5297 | |
5298 | const call_string &cs = en->get_point ().get_call_string (); |
5299 | function_call_string key (fun, &cs); |
5300 | function_call_string_cluster **slot = m_map.get (k: key); |
5301 | if (slot) |
5302 | (*slot)->add_node (en); |
5303 | else |
5304 | { |
5305 | function_call_string_cluster *child |
5306 | = new function_call_string_cluster (fun, cs); |
5307 | m_map.put (k: key, v: child); |
5308 | child->add_node (en); |
5309 | } |
5310 | } |
5311 | |
5312 | private: |
5313 | typedef hash_map<function_call_string, function_call_string_cluster *> map_t; |
5314 | map_t m_map; |
5315 | |
5316 | /* This should just be the origin exploded_node. */ |
5317 | auto_vec <exploded_node *> m_functionless_enodes; |
5318 | }; |
5319 | |
5320 | /* Subclass of range_label for use within |
5321 | exploded_graph::dump_exploded_nodes for implementing |
5322 | -fdump-analyzer-exploded-nodes: a label for a specific |
5323 | exploded_node. */ |
5324 | |
5325 | class enode_label : public range_label |
5326 | { |
5327 | public: |
5328 | enode_label (const extrinsic_state &ext_state, |
5329 | exploded_node *enode) |
5330 | : m_ext_state (ext_state), m_enode (enode) {} |
5331 | |
5332 | label_text get_text (unsigned) const final override |
5333 | { |
5334 | pretty_printer pp; |
5335 | pp_format_decoder (&pp) = default_tree_printer; |
5336 | m_enode->get_state ().dump_to_pp (ext_state: m_ext_state, simple: true, multiline: false, pp: &pp); |
5337 | return make_label_text (can_colorize: false, fmt: "EN: %i: %s" , |
5338 | m_enode->m_index, pp_formatted_text (&pp)); |
5339 | } |
5340 | |
5341 | private: |
5342 | const extrinsic_state &m_ext_state; |
5343 | exploded_node *m_enode; |
5344 | }; |
5345 | |
5346 | /* Postprocessing support for dumping the exploded nodes. |
5347 | Handle -fdump-analyzer-exploded-nodes, |
5348 | -fdump-analyzer-exploded-nodes-2, and the |
5349 | "__analyzer_dump_exploded_nodes" builtin. */ |
5350 | |
5351 | void |
5352 | exploded_graph::dump_exploded_nodes () const |
5353 | { |
5354 | // TODO |
5355 | /* Locate calls to __analyzer_dump_exploded_nodes. */ |
5356 | // Print how many egs there are for them? |
5357 | /* Better: log them as we go, and record the exploded nodes |
5358 | in question. */ |
5359 | |
5360 | /* Show every enode. */ |
5361 | |
5362 | /* Gather them by stmt, so that we can more clearly see the |
5363 | "hotspots" requiring numerous exploded nodes. */ |
5364 | |
5365 | /* Alternatively, simply throw them all into one big rich_location |
5366 | and see if the label-printing will sort it out... |
5367 | This requires them all to be in the same source file. */ |
5368 | |
5369 | if (flag_dump_analyzer_exploded_nodes) |
5370 | { |
5371 | auto_timevar tv (TV_ANALYZER_DUMP); |
5372 | gcc_rich_location richloc (UNKNOWN_LOCATION); |
5373 | unsigned i; |
5374 | exploded_node *enode; |
5375 | FOR_EACH_VEC_ELT (m_nodes, i, enode) |
5376 | { |
5377 | if (const gimple *stmt = enode->get_stmt ()) |
5378 | { |
5379 | if (get_pure_location (loc: richloc.get_loc ()) == UNKNOWN_LOCATION) |
5380 | richloc.set_range (idx: 0, loc: stmt->location, range_display_kind: SHOW_RANGE_WITH_CARET); |
5381 | else |
5382 | richloc.add_range (loc: stmt->location, |
5383 | range_display_kind: SHOW_RANGE_WITHOUT_CARET, |
5384 | label: new enode_label (m_ext_state, enode)); |
5385 | } |
5386 | } |
5387 | warning_at (&richloc, 0, "%i exploded nodes" , m_nodes.length ()); |
5388 | |
5389 | /* Repeat the warning without all the labels, so that message is visible |
5390 | (the other one may well have scrolled past the terminal limit). */ |
5391 | warning_at (richloc.get_loc (), 0, |
5392 | "%i exploded nodes" , m_nodes.length ()); |
5393 | |
5394 | if (m_worklist.length () > 0) |
5395 | warning_at (richloc.get_loc (), 0, |
5396 | "worklist still contains %i nodes" , m_worklist.length ()); |
5397 | } |
5398 | |
5399 | /* Dump the egraph in textual form to a dump file. */ |
5400 | if (flag_dump_analyzer_exploded_nodes_2) |
5401 | { |
5402 | auto_timevar tv (TV_ANALYZER_DUMP); |
5403 | char *filename |
5404 | = concat (dump_base_name, ".eg.txt" , NULL); |
5405 | FILE *outf = fopen (filename: filename, modes: "w" ); |
5406 | if (!outf) |
5407 | error_at (UNKNOWN_LOCATION, "unable to open %qs for writing" , filename); |
5408 | free (ptr: filename); |
5409 | |
5410 | fprintf (stream: outf, format: "exploded graph for %s\n" , dump_base_name); |
5411 | fprintf (stream: outf, format: " nodes: %i\n" , m_nodes.length ()); |
5412 | fprintf (stream: outf, format: " edges: %i\n" , m_edges.length ()); |
5413 | |
5414 | unsigned i; |
5415 | exploded_node *enode; |
5416 | FOR_EACH_VEC_ELT (m_nodes, i, enode) |
5417 | { |
5418 | fprintf (stream: outf, format: "\nEN %i:\n" , enode->m_index); |
5419 | enode->dump_succs_and_preds (outf); |
5420 | pretty_printer pp; |
5421 | enode->get_point ().print (pp: &pp, f: format (true)); |
5422 | fprintf (stream: outf, format: "%s\n" , pp_formatted_text (&pp)); |
5423 | enode->get_state ().dump_to_file (ext_state: m_ext_state, simple: false, multiline: true, outf); |
5424 | } |
5425 | |
5426 | fclose (stream: outf); |
5427 | } |
5428 | |
5429 | /* Dump the egraph in textual form to multiple dump files, one per enode. */ |
5430 | if (flag_dump_analyzer_exploded_nodes_3) |
5431 | { |
5432 | auto_timevar tv (TV_ANALYZER_DUMP); |
5433 | |
5434 | unsigned i; |
5435 | exploded_node *enode; |
5436 | FOR_EACH_VEC_ELT (m_nodes, i, enode) |
5437 | { |
5438 | char *filename |
5439 | = xasprintf ("%s.en-%i.txt" , dump_base_name, i); |
5440 | FILE *outf = fopen (filename: filename, modes: "w" ); |
5441 | if (!outf) |
5442 | error_at (UNKNOWN_LOCATION, "unable to open %qs for writing" , filename); |
5443 | free (ptr: filename); |
5444 | |
5445 | fprintf (stream: outf, format: "EN %i:\n" , enode->m_index); |
5446 | enode->dump_succs_and_preds (outf); |
5447 | pretty_printer pp; |
5448 | enode->get_point ().print (pp: &pp, f: format (true)); |
5449 | fprintf (stream: outf, format: "%s\n" , pp_formatted_text (&pp)); |
5450 | enode->get_state ().dump_to_file (ext_state: m_ext_state, simple: false, multiline: true, outf); |
5451 | |
5452 | fclose (stream: outf); |
5453 | } |
5454 | } |
5455 | |
5456 | /* Emit a warning at any call to "__analyzer_dump_exploded_nodes", |
5457 | giving the number of processed exploded nodes for "before-stmt", |
5458 | and the IDs of processed, merger, and worklist enodes. |
5459 | |
5460 | We highlight the count of *processed* enodes since this is of most |
5461 | interest in DejaGnu tests for ensuring that state merger has |
5462 | happened. |
5463 | |
5464 | We don't show the count of merger and worklist enodes, as this is |
5465 | more of an implementation detail of the merging/worklist that we |
5466 | don't want to bake into our expected DejaGnu messages. */ |
5467 | |
5468 | unsigned i; |
5469 | exploded_node *enode; |
5470 | hash_set<const gimple *> seen; |
5471 | FOR_EACH_VEC_ELT (m_nodes, i, enode) |
5472 | { |
5473 | if (enode->get_point ().get_kind () != PK_BEFORE_STMT) |
5474 | continue; |
5475 | |
5476 | if (const gimple *stmt = enode->get_stmt ()) |
5477 | if (const gcall *call = dyn_cast <const gcall *> (p: stmt)) |
5478 | if (is_special_named_call_p (call, funcname: "__analyzer_dump_exploded_nodes" , |
5479 | num_args: 1)) |
5480 | { |
5481 | if (seen.contains (k: stmt)) |
5482 | continue; |
5483 | |
5484 | auto_vec<exploded_node *> processed_enodes; |
5485 | auto_vec<exploded_node *> merger_enodes; |
5486 | auto_vec<exploded_node *> worklist_enodes; |
5487 | /* This is O(N^2). */ |
5488 | unsigned j; |
5489 | exploded_node *other_enode; |
5490 | FOR_EACH_VEC_ELT (m_nodes, j, other_enode) |
5491 | { |
5492 | if (other_enode->get_point ().get_kind () != PK_BEFORE_STMT) |
5493 | continue; |
5494 | if (other_enode->get_stmt () == stmt) |
5495 | switch (other_enode->get_status ()) |
5496 | { |
5497 | default: |
5498 | gcc_unreachable (); |
5499 | case exploded_node::STATUS_WORKLIST: |
5500 | worklist_enodes.safe_push (obj: other_enode); |
5501 | break; |
5502 | case exploded_node::STATUS_PROCESSED: |
5503 | processed_enodes.safe_push (obj: other_enode); |
5504 | break; |
5505 | case exploded_node::STATUS_MERGER: |
5506 | merger_enodes.safe_push (obj: other_enode); |
5507 | break; |
5508 | } |
5509 | } |
5510 | |
5511 | pretty_printer pp; |
5512 | pp_character (&pp, '['); |
5513 | print_enode_indices (pp: &pp, enodes: processed_enodes); |
5514 | if (merger_enodes.length () > 0) |
5515 | { |
5516 | pp_string (&pp, "] merger(s): [" ); |
5517 | print_enode_indices (pp: &pp, enodes: merger_enodes); |
5518 | } |
5519 | if (worklist_enodes.length () > 0) |
5520 | { |
5521 | pp_string (&pp, "] worklist: [" ); |
5522 | print_enode_indices (pp: &pp, enodes: worklist_enodes); |
5523 | } |
5524 | pp_character (&pp, ']'); |
5525 | |
5526 | warning_n (stmt->location, 0, processed_enodes.length (), |
5527 | "%i processed enode: %s" , |
5528 | "%i processed enodes: %s" , |
5529 | processed_enodes.length (), pp_formatted_text (&pp)); |
5530 | seen.add (k: stmt); |
5531 | |
5532 | /* If the argument is non-zero, then print all of the states |
5533 | of the various enodes. */ |
5534 | tree t_arg = fold (gimple_call_arg (gs: call, index: 0)); |
5535 | if (TREE_CODE (t_arg) != INTEGER_CST) |
5536 | { |
5537 | error_at (call->location, |
5538 | "integer constant required for arg 1" ); |
5539 | return; |
5540 | } |
5541 | int i_arg = TREE_INT_CST_LOW (t_arg); |
5542 | if (i_arg) |
5543 | { |
5544 | exploded_node *other_enode; |
5545 | FOR_EACH_VEC_ELT (processed_enodes, j, other_enode) |
5546 | { |
5547 | fprintf (stderr, format: "%i of %i: EN %i:\n" , |
5548 | j + 1, processed_enodes.length (), |
5549 | other_enode->m_index); |
5550 | other_enode->dump_succs_and_preds (stderr); |
5551 | /* Dump state. */ |
5552 | other_enode->get_state ().dump (ext_state: m_ext_state, simple: false); |
5553 | } |
5554 | } |
5555 | } |
5556 | } |
5557 | } |
5558 | |
5559 | DEBUG_FUNCTION exploded_node * |
5560 | exploded_graph::get_node_by_index (int idx) const |
5561 | { |
5562 | exploded_node *enode = m_nodes[idx]; |
5563 | gcc_assert (enode->m_index == idx); |
5564 | return enode; |
5565 | } |
5566 | |
5567 | /* Ensure that there is an exploded_node for a top-level call to FNDECL. */ |
5568 | |
5569 | void |
5570 | exploded_graph::on_escaped_function (tree fndecl) |
5571 | { |
5572 | logger * const logger = get_logger (); |
5573 | LOG_FUNC_1 (logger, "%qE" , fndecl); |
5574 | |
5575 | cgraph_node *cgnode = cgraph_node::get (decl: fndecl); |
5576 | if (!cgnode) |
5577 | return; |
5578 | |
5579 | function *fun = cgnode->get_fun (); |
5580 | if (!fun) |
5581 | return; |
5582 | |
5583 | if (!gimple_has_body_p (fndecl)) |
5584 | return; |
5585 | |
5586 | exploded_node *enode = add_function_entry (fun: *fun); |
5587 | if (logger) |
5588 | { |
5589 | if (enode) |
5590 | logger->log (fmt: "created EN %i for %qE entrypoint" , |
5591 | enode->m_index, fun->decl); |
5592 | else |
5593 | logger->log (fmt: "did not create enode for %qE entrypoint" , fun->decl); |
5594 | } |
5595 | } |
5596 | |
5597 | /* A collection of classes for visualizing the callgraph in .dot form |
5598 | (as represented in the supergraph). */ |
5599 | |
5600 | /* Forward decls. */ |
5601 | class viz_callgraph_node; |
5602 | class viz_callgraph_edge; |
5603 | class viz_callgraph; |
5604 | class viz_callgraph_cluster; |
5605 | |
5606 | /* Traits for using "digraph.h" to visualize the callgraph. */ |
5607 | |
5608 | struct viz_callgraph_traits |
5609 | { |
5610 | typedef viz_callgraph_node node_t; |
5611 | typedef viz_callgraph_edge edge_t; |
5612 | typedef viz_callgraph graph_t; |
5613 | struct dump_args_t |
5614 | { |
5615 | dump_args_t (const exploded_graph *eg) : m_eg (eg) {} |
5616 | const exploded_graph *m_eg; |
5617 | }; |
5618 | typedef viz_callgraph_cluster cluster_t; |
5619 | }; |
5620 | |
5621 | /* Subclass of dnode representing a function within the callgraph. */ |
5622 | |
5623 | class viz_callgraph_node : public dnode<viz_callgraph_traits> |
5624 | { |
5625 | friend class viz_callgraph; |
5626 | |
5627 | public: |
5628 | viz_callgraph_node (function *fun, int index) |
5629 | : m_fun (fun), m_index (index), m_num_supernodes (0), m_num_superedges (0) |
5630 | { |
5631 | gcc_assert (fun); |
5632 | } |
5633 | |
5634 | void dump_dot (graphviz_out *gv, const dump_args_t &args) const final override |
5635 | { |
5636 | pretty_printer *pp = gv->get_pp (); |
5637 | |
5638 | dump_dot_id (pp); |
5639 | pp_printf (pp, " [shape=none,margin=0,style=filled,fillcolor=%s,label=\"" , |
5640 | "lightgrey" ); |
5641 | pp_write_text_to_stream (pp); |
5642 | |
5643 | pp_printf (pp, "VCG: %i: %s" , m_index, function_name (m_fun)); |
5644 | pp_newline (pp); |
5645 | |
5646 | pp_printf (pp, "supernodes: %i\n" , m_num_supernodes); |
5647 | pp_newline (pp); |
5648 | |
5649 | pp_printf (pp, "superedges: %i\n" , m_num_superedges); |
5650 | pp_newline (pp); |
5651 | |
5652 | if (args.m_eg) |
5653 | { |
5654 | unsigned i; |
5655 | exploded_node *enode; |
5656 | unsigned num_enodes = 0; |
5657 | FOR_EACH_VEC_ELT (args.m_eg->m_nodes, i, enode) |
5658 | { |
5659 | if (enode->get_point ().get_function () == m_fun) |
5660 | num_enodes++; |
5661 | } |
5662 | pp_printf (pp, "enodes: %i\n" , num_enodes); |
5663 | pp_newline (pp); |
5664 | |
5665 | // TODO: also show the per-callstring breakdown |
5666 | const exploded_graph::call_string_data_map_t *per_cs_data |
5667 | = args.m_eg->get_per_call_string_data (); |
5668 | for (exploded_graph::call_string_data_map_t::iterator iter |
5669 | = per_cs_data->begin (); |
5670 | iter != per_cs_data->end (); |
5671 | ++iter) |
5672 | { |
5673 | const call_string *cs = (*iter).first; |
5674 | //per_call_string_data *data = (*iter).second; |
5675 | num_enodes = 0; |
5676 | FOR_EACH_VEC_ELT (args.m_eg->m_nodes, i, enode) |
5677 | { |
5678 | if (enode->get_point ().get_function () == m_fun |
5679 | && &enode->get_point ().get_call_string () == cs) |
5680 | num_enodes++; |
5681 | } |
5682 | if (num_enodes > 0) |
5683 | { |
5684 | cs->print (pp); |
5685 | pp_printf (pp, ": %i\n" , num_enodes); |
5686 | } |
5687 | } |
5688 | |
5689 | /* Show any summaries. */ |
5690 | per_function_data *data = args.m_eg->get_per_function_data (fun: m_fun); |
5691 | if (data) |
5692 | { |
5693 | pp_newline (pp); |
5694 | pp_printf (pp, "summaries: %i\n" , data->m_summaries.length ()); |
5695 | for (auto summary : data->m_summaries) |
5696 | { |
5697 | pp_printf (pp, "\nsummary: %s:\n" , summary->get_desc ().get ()); |
5698 | const extrinsic_state &ext_state = args.m_eg->get_ext_state (); |
5699 | const program_state &state = summary->get_state (); |
5700 | state.dump_to_pp (ext_state, simple: false, multiline: true, pp); |
5701 | pp_newline (pp); |
5702 | } |
5703 | } |
5704 | } |
5705 | |
5706 | pp_write_text_as_dot_label_to_stream (pp, /*for_record=*/true); |
5707 | pp_string (pp, "\"];\n\n" ); |
5708 | pp_flush (pp); |
5709 | } |
5710 | |
5711 | void dump_dot_id (pretty_printer *pp) const |
5712 | { |
5713 | pp_printf (pp, "vcg_%i" , m_index); |
5714 | } |
5715 | |
5716 | private: |
5717 | function *m_fun; |
5718 | int m_index; |
5719 | int m_num_supernodes; |
5720 | int m_num_superedges; |
5721 | }; |
5722 | |
5723 | /* Subclass of dedge representing a callgraph edge. */ |
5724 | |
5725 | class viz_callgraph_edge : public dedge<viz_callgraph_traits> |
5726 | { |
5727 | public: |
5728 | viz_callgraph_edge (viz_callgraph_node *src, viz_callgraph_node *dest) |
5729 | : dedge<viz_callgraph_traits> (src, dest) |
5730 | {} |
5731 | |
5732 | void dump_dot (graphviz_out *gv, const dump_args_t &) const |
5733 | final override |
5734 | { |
5735 | pretty_printer *pp = gv->get_pp (); |
5736 | |
5737 | const char *style = "\"solid,bold\"" ; |
5738 | const char *color = "black" ; |
5739 | int weight = 10; |
5740 | const char *constraint = "true" ; |
5741 | |
5742 | m_src->dump_dot_id (pp); |
5743 | pp_string (pp, " -> " ); |
5744 | m_dest->dump_dot_id (pp); |
5745 | pp_printf (pp, |
5746 | (" [style=%s, color=%s, weight=%d, constraint=%s," |
5747 | " headlabel=\"" ), |
5748 | style, color, weight, constraint); |
5749 | pp_printf (pp, "\"];\n" ); |
5750 | } |
5751 | }; |
5752 | |
5753 | /* Subclass of digraph representing the callgraph. */ |
5754 | |
5755 | class viz_callgraph : public digraph<viz_callgraph_traits> |
5756 | { |
5757 | public: |
5758 | viz_callgraph (const supergraph &sg); |
5759 | |
5760 | viz_callgraph_node *get_vcg_node_for_function (function *fun) |
5761 | { |
5762 | return *m_map.get (k: fun); |
5763 | } |
5764 | |
5765 | viz_callgraph_node *get_vcg_node_for_snode (supernode *snode) |
5766 | { |
5767 | return get_vcg_node_for_function (fun: snode->m_fun); |
5768 | } |
5769 | |
5770 | private: |
5771 | hash_map<function *, viz_callgraph_node *> m_map; |
5772 | }; |
5773 | |
5774 | /* Placeholder subclass of cluster. */ |
5775 | |
5776 | class viz_callgraph_cluster : public cluster<viz_callgraph_traits> |
5777 | { |
5778 | }; |
5779 | |
5780 | /* viz_callgraph's ctor. */ |
5781 | |
5782 | viz_callgraph::viz_callgraph (const supergraph &sg) |
5783 | { |
5784 | cgraph_node *node; |
5785 | FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node) |
5786 | { |
5787 | function *fun = node->get_fun (); |
5788 | viz_callgraph_node *vcg_node |
5789 | = new viz_callgraph_node (fun, m_nodes.length ()); |
5790 | m_map.put (k: fun, v: vcg_node); |
5791 | add_node (node: vcg_node); |
5792 | } |
5793 | |
5794 | unsigned i; |
5795 | superedge *sedge; |
5796 | FOR_EACH_VEC_ELT (sg.m_edges, i, sedge) |
5797 | { |
5798 | viz_callgraph_node *vcg_src = get_vcg_node_for_snode (snode: sedge->m_src); |
5799 | if (vcg_src->m_fun) |
5800 | get_vcg_node_for_function (fun: vcg_src->m_fun)->m_num_superedges++; |
5801 | if (sedge->dyn_cast_call_superedge ()) |
5802 | { |
5803 | viz_callgraph_node *vcg_dest = get_vcg_node_for_snode (snode: sedge->m_dest); |
5804 | viz_callgraph_edge *vcg_edge |
5805 | = new viz_callgraph_edge (vcg_src, vcg_dest); |
5806 | add_edge (edge: vcg_edge); |
5807 | } |
5808 | } |
5809 | |
5810 | supernode *snode; |
5811 | FOR_EACH_VEC_ELT (sg.m_nodes, i, snode) |
5812 | { |
5813 | if (snode->m_fun) |
5814 | get_vcg_node_for_function (fun: snode->m_fun)->m_num_supernodes++; |
5815 | } |
5816 | } |
5817 | |
5818 | /* Dump the callgraph to FILENAME. */ |
5819 | |
5820 | static void |
5821 | dump_callgraph (const supergraph &sg, const char *filename, |
5822 | const exploded_graph *eg) |
5823 | { |
5824 | FILE *outf = fopen (filename: filename, modes: "w" ); |
5825 | if (!outf) |
5826 | return; |
5827 | |
5828 | // TODO |
5829 | viz_callgraph vcg (sg); |
5830 | vcg.dump_dot (path: filename, NULL, args: viz_callgraph_traits::dump_args_t (eg)); |
5831 | |
5832 | fclose (stream: outf); |
5833 | } |
5834 | |
5835 | /* Dump the callgraph to "<srcfile>.callgraph.dot". */ |
5836 | |
5837 | static void |
5838 | dump_callgraph (const supergraph &sg, const exploded_graph *eg) |
5839 | { |
5840 | auto_timevar tv (TV_ANALYZER_DUMP); |
5841 | char *filename = concat (dump_base_name, ".callgraph.dot" , NULL); |
5842 | dump_callgraph (sg, filename, eg); |
5843 | free (ptr: filename); |
5844 | } |
5845 | |
5846 | /* Subclass of dot_annotator for implementing |
5847 | DUMP_BASE_NAME.supergraph-eg.dot, a post-analysis dump of the supergraph. |
5848 | |
5849 | Annotate the supergraph nodes by printing the exploded nodes in concise |
5850 | form within them, next to their pertinent statements where appropriate, |
5851 | colorizing the exploded nodes based on sm-state. |
5852 | Also show saved diagnostics within the exploded nodes, giving information |
5853 | on whether they were feasible, and, if infeasible, where the problem |
5854 | was. */ |
5855 | |
5856 | class exploded_graph_annotator : public dot_annotator |
5857 | { |
5858 | public: |
5859 | exploded_graph_annotator (const exploded_graph &eg) |
5860 | : m_eg (eg) |
5861 | { |
5862 | /* Avoid O(N^2) by prepopulating m_enodes_per_snodes. */ |
5863 | unsigned i; |
5864 | supernode *snode; |
5865 | FOR_EACH_VEC_ELT (eg.get_supergraph ().m_nodes, i, snode) |
5866 | m_enodes_per_snodes.safe_push (obj: new auto_vec <exploded_node *> ()); |
5867 | exploded_node *enode; |
5868 | FOR_EACH_VEC_ELT (m_eg.m_nodes, i, enode) |
5869 | if (enode->get_supernode ()) |
5870 | m_enodes_per_snodes[enode->get_supernode ()->m_index]->safe_push (obj: enode); |
5871 | } |
5872 | |
5873 | /* Show exploded nodes for BEFORE_SUPERNODE points before N. */ |
5874 | bool add_node_annotations (graphviz_out *gv, const supernode &n, |
5875 | bool within_table) |
5876 | const final override |
5877 | { |
5878 | if (!within_table) |
5879 | return false; |
5880 | gv->begin_tr (); |
5881 | pretty_printer *pp = gv->get_pp (); |
5882 | |
5883 | gv->begin_td (); |
5884 | pp_string (pp, "BEFORE" ); |
5885 | pp_printf (pp, " (scc: %i)" , m_eg.get_scc_id (node: n)); |
5886 | gv->end_td (); |
5887 | |
5888 | unsigned i; |
5889 | exploded_node *enode; |
5890 | bool had_enode = false; |
5891 | FOR_EACH_VEC_ELT (*m_enodes_per_snodes[n.m_index], i, enode) |
5892 | { |
5893 | gcc_assert (enode->get_supernode () == &n); |
5894 | const program_point &point = enode->get_point (); |
5895 | if (point.get_kind () != PK_BEFORE_SUPERNODE) |
5896 | continue; |
5897 | print_enode (gv, enode); |
5898 | had_enode = true; |
5899 | } |
5900 | if (!had_enode) |
5901 | pp_string (pp, "<TD BGCOLOR=\"red\">UNREACHED</TD>" ); |
5902 | pp_flush (pp); |
5903 | gv->end_tr (); |
5904 | return true; |
5905 | } |
5906 | |
5907 | /* Show exploded nodes for STMT. */ |
5908 | void add_stmt_annotations (graphviz_out *gv, const gimple *stmt, |
5909 | bool within_row) |
5910 | const final override |
5911 | { |
5912 | if (!within_row) |
5913 | return; |
5914 | pretty_printer *pp = gv->get_pp (); |
5915 | |
5916 | const supernode *snode |
5917 | = m_eg.get_supergraph ().get_supernode_for_stmt (stmt); |
5918 | unsigned i; |
5919 | exploded_node *enode; |
5920 | bool had_td = false; |
5921 | FOR_EACH_VEC_ELT (*m_enodes_per_snodes[snode->m_index], i, enode) |
5922 | { |
5923 | const program_point &point = enode->get_point (); |
5924 | if (point.get_kind () != PK_BEFORE_STMT) |
5925 | continue; |
5926 | if (point.get_stmt () != stmt) |
5927 | continue; |
5928 | print_enode (gv, enode); |
5929 | had_td = true; |
5930 | } |
5931 | pp_flush (pp); |
5932 | if (!had_td) |
5933 | { |
5934 | gv->begin_td (); |
5935 | gv->end_td (); |
5936 | } |
5937 | } |
5938 | |
5939 | /* Show exploded nodes for AFTER_SUPERNODE points after N. */ |
5940 | bool add_after_node_annotations (graphviz_out *gv, const supernode &n) |
5941 | const final override |
5942 | { |
5943 | gv->begin_tr (); |
5944 | pretty_printer *pp = gv->get_pp (); |
5945 | |
5946 | gv->begin_td (); |
5947 | pp_string (pp, "AFTER" ); |
5948 | gv->end_td (); |
5949 | |
5950 | unsigned i; |
5951 | exploded_node *enode; |
5952 | FOR_EACH_VEC_ELT (*m_enodes_per_snodes[n.m_index], i, enode) |
5953 | { |
5954 | gcc_assert (enode->get_supernode () == &n); |
5955 | const program_point &point = enode->get_point (); |
5956 | if (point.get_kind () != PK_AFTER_SUPERNODE) |
5957 | continue; |
5958 | print_enode (gv, enode); |
5959 | } |
5960 | pp_flush (pp); |
5961 | gv->end_tr (); |
5962 | return true; |
5963 | } |
5964 | |
5965 | private: |
5966 | /* Concisely print a TD element for ENODE, showing the index, status, |
5967 | and any saved_diagnostics at the enode. Colorize it to show sm-state. |
5968 | |
5969 | Ideally we'd dump ENODE's state here, hidden behind some kind of |
5970 | interactive disclosure method like a tooltip, so that the states |
5971 | can be explored without overwhelming the graph. |
5972 | However, I wasn't able to get graphviz/xdot to show tooltips on |
5973 | individual elements within a HTML-like label. */ |
5974 | void print_enode (graphviz_out *gv, const exploded_node *enode) const |
5975 | { |
5976 | pretty_printer *pp = gv->get_pp (); |
5977 | pp_printf (pp, "<TD BGCOLOR=\"%s\">" , |
5978 | enode->get_dot_fillcolor ()); |
5979 | pp_printf (pp, "<TABLE BORDER=\"0\">" ); |
5980 | gv->begin_trtd (); |
5981 | pp_printf (pp, "EN: %i" , enode->m_index); |
5982 | switch (enode->get_status ()) |
5983 | { |
5984 | default: |
5985 | gcc_unreachable (); |
5986 | case exploded_node::STATUS_WORKLIST: |
5987 | pp_string (pp, "(W)" ); |
5988 | break; |
5989 | case exploded_node::STATUS_PROCESSED: |
5990 | break; |
5991 | case exploded_node::STATUS_MERGER: |
5992 | pp_string (pp, "(M)" ); |
5993 | break; |
5994 | case exploded_node::STATUS_BULK_MERGED: |
5995 | pp_string (pp, "(BM)" ); |
5996 | break; |
5997 | } |
5998 | gv->end_tdtr (); |
5999 | |
6000 | /* Dump any saved_diagnostics at this enode. */ |
6001 | for (unsigned i = 0; i < enode->get_num_diagnostics (); i++) |
6002 | { |
6003 | const saved_diagnostic *sd = enode->get_saved_diagnostic (idx: i); |
6004 | print_saved_diagnostic (gv, sd); |
6005 | } |
6006 | pp_printf (pp, "</TABLE>" ); |
6007 | pp_printf (pp, "</TD>" ); |
6008 | } |
6009 | |
6010 | /* Print a TABLE element for SD, showing the kind, the length of the |
6011 | exploded_path, whether the path was feasible, and if infeasible, |
6012 | what the problem was. */ |
6013 | void print_saved_diagnostic (graphviz_out *gv, |
6014 | const saved_diagnostic *sd) const |
6015 | { |
6016 | pretty_printer *pp = gv->get_pp (); |
6017 | gv->begin_trtd (); |
6018 | pp_printf (pp, "<TABLE BORDER=\"0\">" ); |
6019 | gv->begin_tr (); |
6020 | pp_string (pp, "<TD BGCOLOR=\"green\">" ); |
6021 | pp_printf (pp, "DIAGNOSTIC: %s" , sd->m_d->get_kind ()); |
6022 | gv->end_tdtr (); |
6023 | gv->begin_trtd (); |
6024 | if (sd->get_best_epath ()) |
6025 | pp_printf (pp, "epath length: %i" , sd->get_epath_length ()); |
6026 | else |
6027 | pp_printf (pp, "no best epath" ); |
6028 | gv->end_tdtr (); |
6029 | if (const feasibility_problem *p = sd->get_feasibility_problem ()) |
6030 | { |
6031 | gv->begin_trtd (); |
6032 | pp_printf (pp, "INFEASIBLE at eedge %i: EN:%i -> EN:%i" , |
6033 | p->m_eedge_idx, |
6034 | p->m_eedge.m_src->m_index, |
6035 | p->m_eedge.m_dest->m_index); |
6036 | pp_write_text_as_html_like_dot_to_stream (pp); |
6037 | gv->end_tdtr (); |
6038 | gv->begin_trtd (); |
6039 | p->m_eedge.m_sedge->dump (pp); |
6040 | pp_write_text_as_html_like_dot_to_stream (pp); |
6041 | gv->end_tdtr (); |
6042 | gv->begin_trtd (); |
6043 | pp_gimple_stmt_1 (pp, p->m_last_stmt, 0, (dump_flags_t)0); |
6044 | pp_write_text_as_html_like_dot_to_stream (pp); |
6045 | gv->end_tdtr (); |
6046 | /* Ideally we'd print p->m_model here; see the notes above about |
6047 | tooltips. */ |
6048 | } |
6049 | pp_printf (pp, "</TABLE>" ); |
6050 | gv->end_tdtr (); |
6051 | } |
6052 | |
6053 | const exploded_graph &m_eg; |
6054 | auto_delete_vec<auto_vec <exploded_node *> > m_enodes_per_snodes; |
6055 | }; |
6056 | |
6057 | /* Implement -fdump-analyzer-json. */ |
6058 | |
6059 | static void |
6060 | dump_analyzer_json (const supergraph &sg, |
6061 | const exploded_graph &eg) |
6062 | { |
6063 | auto_timevar tv (TV_ANALYZER_DUMP); |
6064 | char *filename = concat (dump_base_name, ".analyzer.json.gz" , NULL); |
6065 | gzFile output = gzopen (filename, "w" ); |
6066 | if (!output) |
6067 | { |
6068 | error_at (UNKNOWN_LOCATION, "unable to open %qs for writing" , filename); |
6069 | free (ptr: filename); |
6070 | return; |
6071 | } |
6072 | |
6073 | json::object *toplev_obj = new json::object (); |
6074 | toplev_obj->set (key: "sgraph" , v: sg.to_json ()); |
6075 | toplev_obj->set (key: "egraph" , v: eg.to_json ()); |
6076 | |
6077 | pretty_printer pp; |
6078 | toplev_obj->print (pp: &pp, flag_diagnostics_json_formatting); |
6079 | pp_formatted_text (&pp); |
6080 | |
6081 | delete toplev_obj; |
6082 | |
6083 | if (gzputs (file: output, s: pp_formatted_text (&pp)) == EOF |
6084 | || gzclose (file: output)) |
6085 | error_at (UNKNOWN_LOCATION, "error writing %qs" , filename); |
6086 | |
6087 | free (ptr: filename); |
6088 | } |
6089 | |
6090 | /* Concrete subclass of plugin_analyzer_init_iface, allowing plugins |
6091 | to register new state machines. */ |
6092 | |
6093 | class plugin_analyzer_init_impl : public plugin_analyzer_init_iface |
6094 | { |
6095 | public: |
6096 | plugin_analyzer_init_impl (auto_delete_vec <state_machine> *checkers, |
6097 | known_function_manager *known_fn_mgr, |
6098 | logger *logger) |
6099 | : m_checkers (checkers), |
6100 | m_known_fn_mgr (known_fn_mgr), |
6101 | m_logger (logger) |
6102 | {} |
6103 | |
6104 | void register_state_machine (std::unique_ptr<state_machine> sm) final override |
6105 | { |
6106 | LOG_SCOPE (m_logger); |
6107 | m_checkers->safe_push (obj: sm.release ()); |
6108 | } |
6109 | |
6110 | void register_known_function (const char *name, |
6111 | std::unique_ptr<known_function> kf) final override |
6112 | { |
6113 | LOG_SCOPE (m_logger); |
6114 | m_known_fn_mgr->add (name, kf: std::move (kf)); |
6115 | } |
6116 | |
6117 | logger *get_logger () const final override |
6118 | { |
6119 | return m_logger; |
6120 | } |
6121 | |
6122 | private: |
6123 | auto_delete_vec <state_machine> *m_checkers; |
6124 | known_function_manager *m_known_fn_mgr; |
6125 | logger *m_logger; |
6126 | }; |
6127 | |
6128 | /* Run the analysis "engine". */ |
6129 | |
6130 | void |
6131 | impl_run_checkers (logger *logger) |
6132 | { |
6133 | LOG_SCOPE (logger); |
6134 | |
6135 | if (logger) |
6136 | { |
6137 | logger->log (fmt: "BITS_BIG_ENDIAN: %i" , BITS_BIG_ENDIAN ? 1 : 0); |
6138 | logger->log (fmt: "BYTES_BIG_ENDIAN: %i" , BYTES_BIG_ENDIAN ? 1 : 0); |
6139 | logger->log (fmt: "WORDS_BIG_ENDIAN: %i" , WORDS_BIG_ENDIAN ? 1 : 0); |
6140 | log_stashed_constants (logger); |
6141 | } |
6142 | |
6143 | /* If using LTO, ensure that the cgraph nodes have function bodies. */ |
6144 | cgraph_node *node; |
6145 | FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node) |
6146 | node->get_untransformed_body (); |
6147 | |
6148 | /* Create the supergraph. */ |
6149 | supergraph sg (logger); |
6150 | |
6151 | engine eng (&sg, logger); |
6152 | |
6153 | state_purge_map *purge_map = NULL; |
6154 | |
6155 | if (flag_analyzer_state_purge) |
6156 | purge_map = new state_purge_map (sg, eng.get_model_manager (), logger); |
6157 | |
6158 | if (flag_dump_analyzer_supergraph) |
6159 | { |
6160 | /* Dump supergraph pre-analysis. */ |
6161 | auto_timevar tv (TV_ANALYZER_DUMP); |
6162 | char *filename = concat (dump_base_name, ".supergraph.dot" , NULL); |
6163 | supergraph::dump_args_t args ((enum supergraph_dot_flags)0, NULL); |
6164 | sg.dump_dot (path: filename, args); |
6165 | free (ptr: filename); |
6166 | } |
6167 | |
6168 | if (flag_dump_analyzer_state_purge) |
6169 | { |
6170 | auto_timevar tv (TV_ANALYZER_DUMP); |
6171 | state_purge_annotator a (purge_map); |
6172 | char *filename = concat (dump_base_name, ".state-purge.dot" , NULL); |
6173 | supergraph::dump_args_t args ((enum supergraph_dot_flags)0, &a); |
6174 | sg.dump_dot (path: filename, args); |
6175 | free (ptr: filename); |
6176 | } |
6177 | |
6178 | auto_delete_vec <state_machine> checkers; |
6179 | make_checkers (out&: checkers, logger); |
6180 | |
6181 | register_known_functions (kfm&: *eng.get_known_function_manager (), |
6182 | rmm&: *eng.get_model_manager ()); |
6183 | |
6184 | plugin_analyzer_init_impl data (&checkers, |
6185 | eng.get_known_function_manager (), |
6186 | logger); |
6187 | invoke_plugin_callbacks (event: PLUGIN_ANALYZER_INIT, gcc_data: &data); |
6188 | |
6189 | if (logger) |
6190 | { |
6191 | int i; |
6192 | state_machine *sm; |
6193 | FOR_EACH_VEC_ELT (checkers, i, sm) |
6194 | logger->log (fmt: "checkers[%i]: %s" , i, sm->get_name ()); |
6195 | } |
6196 | |
6197 | /* Extrinsic state shared by nodes in the graph. */ |
6198 | const extrinsic_state ext_state (checkers, &eng, logger); |
6199 | |
6200 | const analysis_plan plan (sg, logger); |
6201 | |
6202 | /* The exploded graph. */ |
6203 | exploded_graph eg (sg, logger, ext_state, purge_map, plan, |
6204 | analyzer_verbosity); |
6205 | |
6206 | /* Add entrypoints to the graph for externally-callable functions. */ |
6207 | eg.build_initial_worklist (); |
6208 | |
6209 | /* Now process the worklist, exploring the <point, state> graph. */ |
6210 | eg.process_worklist (); |
6211 | |
6212 | if (warn_analyzer_infinite_loop) |
6213 | eg.detect_infinite_loops (); |
6214 | |
6215 | if (flag_dump_analyzer_exploded_graph) |
6216 | { |
6217 | auto_timevar tv (TV_ANALYZER_DUMP); |
6218 | char *filename |
6219 | = concat (dump_base_name, ".eg.dot" , NULL); |
6220 | exploded_graph::dump_args_t args (eg); |
6221 | root_cluster c; |
6222 | eg.dump_dot (path: filename, root_cluster: &c, args); |
6223 | free (ptr: filename); |
6224 | } |
6225 | |
6226 | /* Now emit any saved diagnostics. */ |
6227 | eg.get_diagnostic_manager ().emit_saved_diagnostics (eg); |
6228 | |
6229 | eg.dump_exploded_nodes (); |
6230 | |
6231 | eg.log_stats (); |
6232 | |
6233 | if (flag_dump_analyzer_callgraph) |
6234 | dump_callgraph (sg, eg: &eg); |
6235 | |
6236 | if (flag_dump_analyzer_supergraph) |
6237 | { |
6238 | /* Dump post-analysis form of supergraph. */ |
6239 | auto_timevar tv (TV_ANALYZER_DUMP); |
6240 | char *filename = concat (dump_base_name, ".supergraph-eg.dot" , NULL); |
6241 | exploded_graph_annotator a (eg); |
6242 | supergraph::dump_args_t args ((enum supergraph_dot_flags)0, &a); |
6243 | sg.dump_dot (path: filename, args); |
6244 | free (ptr: filename); |
6245 | } |
6246 | |
6247 | if (flag_dump_analyzer_json) |
6248 | dump_analyzer_json (sg, eg); |
6249 | |
6250 | if (flag_dump_analyzer_untracked) |
6251 | eng.get_model_manager ()->dump_untracked_regions (); |
6252 | |
6253 | delete purge_map; |
6254 | |
6255 | /* Free up any dominance info that we may have created. */ |
6256 | FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (node) |
6257 | { |
6258 | function *fun = node->get_fun (); |
6259 | free_dominance_info (fun, CDI_DOMINATORS); |
6260 | } |
6261 | } |
6262 | |
6263 | /* Handle -fdump-analyzer and -fdump-analyzer-stderr. */ |
6264 | static FILE *dump_fout = NULL; |
6265 | |
6266 | /* Track if we're responsible for closing dump_fout. */ |
6267 | static bool owns_dump_fout = false; |
6268 | |
6269 | /* If dumping is enabled, attempt to create dump_fout if it hasn't already |
6270 | been opened. Return it. */ |
6271 | |
6272 | FILE * |
6273 | get_or_create_any_logfile () |
6274 | { |
6275 | if (!dump_fout) |
6276 | { |
6277 | if (flag_dump_analyzer_stderr) |
6278 | dump_fout = stderr; |
6279 | else if (flag_dump_analyzer) |
6280 | { |
6281 | char *dump_filename = concat (dump_base_name, ".analyzer.txt" , NULL); |
6282 | dump_fout = fopen (filename: dump_filename, modes: "w" ); |
6283 | free (ptr: dump_filename); |
6284 | if (dump_fout) |
6285 | owns_dump_fout = true; |
6286 | } |
6287 | } |
6288 | return dump_fout; |
6289 | } |
6290 | |
6291 | /* External entrypoint to the analysis "engine". |
6292 | Set up any dumps, then call impl_run_checkers. */ |
6293 | |
6294 | void |
6295 | run_checkers () |
6296 | { |
6297 | /* Save input_location. */ |
6298 | location_t saved_input_location = input_location; |
6299 | |
6300 | { |
6301 | log_user the_logger (NULL); |
6302 | get_or_create_any_logfile (); |
6303 | if (dump_fout) |
6304 | the_logger.set_logger (new logger (dump_fout, 0, 0, |
6305 | *global_dc->printer)); |
6306 | LOG_SCOPE (the_logger.get_logger ()); |
6307 | |
6308 | impl_run_checkers (logger: the_logger.get_logger ()); |
6309 | |
6310 | /* end of lifetime of the_logger (so that dump file is closed after the |
6311 | various dtors run). */ |
6312 | } |
6313 | |
6314 | if (owns_dump_fout) |
6315 | { |
6316 | fclose (stream: dump_fout); |
6317 | owns_dump_fout = false; |
6318 | dump_fout = NULL; |
6319 | } |
6320 | |
6321 | /* Restore input_location. Subsequent passes may assume that input_location |
6322 | is some arbitrary value *not* in the block tree, which might be violated |
6323 | if we didn't restore it. */ |
6324 | input_location = saved_input_location; |
6325 | } |
6326 | |
6327 | } // namespace ana |
6328 | |
6329 | #endif /* #if ENABLE_ANALYZER */ |
6330 | |