1 | /* Subroutines needed for unwinding stack frames for exception handling. */ |
2 | /* Copyright (C) 1997-2022 Free Software Foundation, Inc. |
3 | |
4 | This file is part of the GNU C Library. |
5 | |
6 | The GNU C Library is free software; you can redistribute it and/or |
7 | modify it under the terms of the GNU Lesser General Public |
8 | License as published by the Free Software Foundation; either |
9 | version 2.1 of the License, or (at your option) any later version. |
10 | |
11 | The GNU C Library is distributed in the hope that it will be useful, |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | Lesser General Public License for more details. |
15 | |
16 | You should have received a copy of the GNU Lesser General Public |
17 | License along with the GNU C Library; if not, see |
18 | <https://www.gnu.org/licenses/>. */ |
19 | |
20 | #ifdef _LIBC |
21 | # include <shlib-compat.h> |
22 | #endif |
23 | |
24 | #if !defined _LIBC || SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_2_5) |
25 | |
26 | #ifdef _LIBC |
27 | #include <stdlib.h> |
28 | #include <string.h> |
29 | #include <libc-lock.h> |
30 | #include <dwarf2.h> |
31 | #include <unwind.h> |
32 | #define NO_BASE_OF_ENCODED_VALUE |
33 | #include <unwind-pe.h> |
34 | #include <unwind-dw2-fde.h> |
35 | #else |
36 | #ifndef _Unwind_Find_FDE |
37 | #include "tconfig.h" |
38 | #include "tsystem.h" |
39 | #include "dwarf2.h" |
40 | #include "unwind.h" |
41 | #define NO_BASE_OF_ENCODED_VALUE |
42 | #include "unwind-pe.h" |
43 | #include "unwind-dw2-fde.h" |
44 | #include "gthr.h" |
45 | #endif |
46 | #endif |
47 | |
48 | /* The unseen_objects list contains objects that have been registered |
49 | but not yet categorized in any way. The seen_objects list has had |
50 | it's pc_begin and count fields initialized at minimum, and is sorted |
51 | by decreasing value of pc_begin. */ |
52 | static struct object *unseen_objects; |
53 | static struct object *seen_objects; |
54 | |
55 | #ifdef _LIBC |
56 | |
57 | __libc_lock_define_initialized (static, object_mutex) |
58 | #define init_object_mutex_once() |
59 | #define __gthread_mutex_lock(m) __libc_lock_lock (*(m)) |
60 | #define __gthread_mutex_unlock(m) __libc_lock_unlock (*(m)) |
61 | |
62 | void __register_frame_info_bases (void *begin, struct object *ob, |
63 | void *tbase, void *dbase); |
64 | hidden_proto (__register_frame_info_bases) |
65 | void __register_frame_info_table_bases (void *begin, |
66 | struct object *ob, |
67 | void *tbase, void *dbase); |
68 | hidden_proto (__register_frame_info_table_bases) |
69 | void *__deregister_frame_info_bases (void *begin); |
70 | hidden_proto (__deregister_frame_info_bases) |
71 | |
72 | #else |
73 | |
74 | #ifdef __GTHREAD_MUTEX_INIT |
75 | static __gthread_mutex_t object_mutex = __GTHREAD_MUTEX_INIT; |
76 | #else |
77 | static __gthread_mutex_t object_mutex; |
78 | #endif |
79 | |
80 | #ifdef __GTHREAD_MUTEX_INIT_FUNCTION |
81 | static void |
82 | init_object_mutex (void) |
83 | { |
84 | __GTHREAD_MUTEX_INIT_FUNCTION (&object_mutex); |
85 | } |
86 | |
87 | static void |
88 | init_object_mutex_once (void) |
89 | { |
90 | static __gthread_once_t once = __GTHREAD_ONCE_INIT; |
91 | __gthread_once (&once, init_object_mutex); |
92 | } |
93 | #else |
94 | #define init_object_mutex_once() |
95 | #endif |
96 | |
97 | #endif /* _LIBC */ |
98 | |
99 | /* Called from crtbegin.o to register the unwind info for an object. */ |
100 | |
101 | void |
102 | __register_frame_info_bases (void *begin, struct object *ob, |
103 | void *tbase, void *dbase) |
104 | { |
105 | /* If .eh_frame is empty, don't register at all. */ |
106 | if (*(uword *) begin == 0) |
107 | return; |
108 | |
109 | ob->pc_begin = (void *)-1; |
110 | ob->tbase = tbase; |
111 | ob->dbase = dbase; |
112 | ob->u.single = begin; |
113 | ob->s.i = 0; |
114 | ob->s.b.encoding = DW_EH_PE_omit; |
115 | #ifdef DWARF2_OBJECT_END_PTR_EXTENSION |
116 | ob->fde_end = NULL; |
117 | #endif |
118 | |
119 | init_object_mutex_once (); |
120 | __gthread_mutex_lock (&object_mutex); |
121 | |
122 | ob->next = unseen_objects; |
123 | unseen_objects = ob; |
124 | |
125 | __gthread_mutex_unlock (&object_mutex); |
126 | } |
127 | hidden_def (__register_frame_info_bases) |
128 | |
129 | void |
130 | __register_frame_info (void *begin, struct object *ob) |
131 | { |
132 | __register_frame_info_bases (begin, ob, tbase: 0, dbase: 0); |
133 | } |
134 | |
135 | void |
136 | __register_frame (void *begin) |
137 | { |
138 | struct object *ob; |
139 | |
140 | /* If .eh_frame is empty, don't register at all. */ |
141 | if (*(uword *) begin == 0) |
142 | return; |
143 | |
144 | ob = (struct object *) malloc (size: sizeof (struct object)); |
145 | __register_frame_info_bases (begin, ob, tbase: 0, dbase: 0); |
146 | } |
147 | |
148 | /* Similar, but BEGIN is actually a pointer to a table of unwind entries |
149 | for different translation units. Called from the file generated by |
150 | collect2. */ |
151 | |
152 | void |
153 | __register_frame_info_table_bases (void *begin, struct object *ob, |
154 | void *tbase, void *dbase) |
155 | { |
156 | ob->pc_begin = (void *)-1; |
157 | ob->tbase = tbase; |
158 | ob->dbase = dbase; |
159 | ob->u.array = begin; |
160 | ob->s.i = 0; |
161 | ob->s.b.from_array = 1; |
162 | ob->s.b.encoding = DW_EH_PE_omit; |
163 | |
164 | init_object_mutex_once (); |
165 | __gthread_mutex_lock (&object_mutex); |
166 | |
167 | ob->next = unseen_objects; |
168 | unseen_objects = ob; |
169 | |
170 | __gthread_mutex_unlock (&object_mutex); |
171 | } |
172 | hidden_def (__register_frame_info_table_bases) |
173 | |
174 | void |
175 | __register_frame_info_table (void *begin, struct object *ob) |
176 | { |
177 | __register_frame_info_table_bases (begin, ob, tbase: 0, dbase: 0); |
178 | } |
179 | |
180 | void |
181 | __register_frame_table (void *begin) |
182 | { |
183 | struct object *ob = (struct object *) malloc (size: sizeof (struct object)); |
184 | __register_frame_info_table_bases (begin, ob, tbase: 0, dbase: 0); |
185 | } |
186 | |
187 | /* Called from crtbegin.o to deregister the unwind info for an object. */ |
188 | /* ??? Glibc has for a while now exported __register_frame_info and |
189 | __deregister_frame_info. If we call __register_frame_info_bases |
190 | from crtbegin (wherein it is declared weak), and this object does |
191 | not get pulled from libgcc.a for other reasons, then the |
192 | invocation of __deregister_frame_info will be resolved from glibc. |
193 | Since the registration did not happen there, we'll abort. |
194 | |
195 | Therefore, declare a new deregistration entry point that does the |
196 | exact same thing, but will resolve to the same library as |
197 | implements __register_frame_info_bases. */ |
198 | |
199 | void * |
200 | __deregister_frame_info_bases (void *begin) |
201 | { |
202 | struct object **p; |
203 | struct object *ob = 0; |
204 | struct fde_vector *tofree = NULL; |
205 | |
206 | /* If .eh_frame is empty, we haven't registered. */ |
207 | if (*(uword *) begin == 0) |
208 | return ob; |
209 | |
210 | init_object_mutex_once (); |
211 | __gthread_mutex_lock (&object_mutex); |
212 | |
213 | for (p = &unseen_objects; *p ; p = &(*p)->next) |
214 | if ((*p)->u.single == begin) |
215 | { |
216 | ob = *p; |
217 | *p = ob->next; |
218 | goto out; |
219 | } |
220 | |
221 | for (p = &seen_objects; *p ; p = &(*p)->next) |
222 | if ((*p)->s.b.sorted) |
223 | { |
224 | if ((*p)->u.sort->orig_data == begin) |
225 | { |
226 | ob = *p; |
227 | *p = ob->next; |
228 | tofree = ob->u.sort; |
229 | goto out; |
230 | } |
231 | } |
232 | else |
233 | { |
234 | if ((*p)->u.single == begin) |
235 | { |
236 | ob = *p; |
237 | *p = ob->next; |
238 | goto out; |
239 | } |
240 | } |
241 | |
242 | __gthread_mutex_unlock (&object_mutex); |
243 | abort (); |
244 | |
245 | out: |
246 | __gthread_mutex_unlock (&object_mutex); |
247 | free (ptr: tofree); |
248 | return (void *) ob; |
249 | } |
250 | hidden_def (__deregister_frame_info_bases) |
251 | |
252 | void * |
253 | __deregister_frame_info (void *begin) |
254 | { |
255 | return __deregister_frame_info_bases (begin); |
256 | } |
257 | |
258 | void |
259 | __deregister_frame (void *begin) |
260 | { |
261 | /* If .eh_frame is empty, we haven't registered. */ |
262 | if (*(uword *) begin != 0) |
263 | free (ptr: __deregister_frame_info_bases (begin)); |
264 | } |
265 | |
266 | |
267 | /* Like base_of_encoded_value, but take the base from a struct object |
268 | instead of an _Unwind_Context. */ |
269 | |
270 | static _Unwind_Ptr |
271 | base_from_object (unsigned char encoding, struct object *ob) |
272 | { |
273 | if (encoding == DW_EH_PE_omit) |
274 | return 0; |
275 | |
276 | switch (encoding & 0x70) |
277 | { |
278 | case DW_EH_PE_absptr: |
279 | case DW_EH_PE_pcrel: |
280 | case DW_EH_PE_aligned: |
281 | return 0; |
282 | |
283 | case DW_EH_PE_textrel: |
284 | return (_Unwind_Ptr) ob->tbase; |
285 | case DW_EH_PE_datarel: |
286 | return (_Unwind_Ptr) ob->dbase; |
287 | } |
288 | abort (); |
289 | } |
290 | |
291 | /* Return the FDE pointer encoding from the CIE. */ |
292 | /* ??? This is a subset of extract_cie_info from unwind-dw2.c. */ |
293 | |
294 | static int |
295 | get_cie_encoding (struct dwarf_cie *cie) |
296 | { |
297 | const unsigned char *aug, *p; |
298 | _Unwind_Ptr dummy; |
299 | _Unwind_Word utmp; |
300 | _Unwind_Sword stmp; |
301 | |
302 | aug = cie->augmentation; |
303 | if (aug[0] != 'z') |
304 | return DW_EH_PE_absptr; |
305 | |
306 | /* Skip the augmentation string. */ |
307 | p = aug + strlen ((const char *) aug) + 1; |
308 | p = read_uleb128 (p, val: &utmp); /* Skip code alignment. */ |
309 | p = read_sleb128 (p, val: &stmp); /* Skip data alignment. */ |
310 | p++; /* Skip return address column. */ |
311 | |
312 | aug++; /* Skip 'z' */ |
313 | p = read_uleb128 (p, val: &utmp); /* Skip augmentation length. */ |
314 | while (1) |
315 | { |
316 | /* This is what we're looking for. */ |
317 | if (*aug == 'R') |
318 | return *p; |
319 | /* Personality encoding and pointer. */ |
320 | else if (*aug == 'P') |
321 | { |
322 | /* ??? Avoid dereferencing indirect pointers, since we're |
323 | faking the base address. Gotta keep DW_EH_PE_aligned |
324 | intact, however. */ |
325 | p = read_encoded_value_with_base (encoding: *p & 0x7F, base: 0, p: p + 1, val: &dummy); |
326 | } |
327 | /* LSDA encoding. */ |
328 | else if (*aug == 'L') |
329 | p++; |
330 | /* Otherwise end of string, or unknown augmentation. */ |
331 | else |
332 | return DW_EH_PE_absptr; |
333 | aug++; |
334 | } |
335 | } |
336 | |
337 | static inline int |
338 | get_fde_encoding (struct dwarf_fde *f) |
339 | { |
340 | return get_cie_encoding (cie: get_cie (f)); |
341 | } |
342 | |
343 | |
344 | /* Sorting an array of FDEs by address. |
345 | (Ideally we would have the linker sort the FDEs so we don't have to do |
346 | it at run time. But the linkers are not yet prepared for this.) */ |
347 | |
348 | /* Return the Nth pc_begin value from FDE x. */ |
349 | |
350 | static inline _Unwind_Ptr |
351 | get_pc_begin (fde *x, size_t n) |
352 | { |
353 | _Unwind_Ptr p; |
354 | memcpy (&p, x->pc_begin + n * sizeof (_Unwind_Ptr), sizeof (_Unwind_Ptr)); |
355 | return p; |
356 | } |
357 | |
358 | /* Comparison routines. Three variants of increasing complexity. */ |
359 | |
360 | static int |
361 | fde_unencoded_compare (struct object *ob __attribute__((unused)), |
362 | fde *x, fde *y) |
363 | { |
364 | _Unwind_Ptr x_ptr = get_pc_begin (x, n: 0); |
365 | _Unwind_Ptr y_ptr = get_pc_begin (x: y, n: 0); |
366 | |
367 | if (x_ptr > y_ptr) |
368 | return 1; |
369 | if (x_ptr < y_ptr) |
370 | return -1; |
371 | return 0; |
372 | } |
373 | |
374 | static int |
375 | fde_single_encoding_compare (struct object *ob, fde *x, fde *y) |
376 | { |
377 | _Unwind_Ptr base, x_ptr, y_ptr; |
378 | |
379 | base = base_from_object (encoding: ob->s.b.encoding, ob); |
380 | read_encoded_value_with_base (encoding: ob->s.b.encoding, base, p: x->pc_begin, val: &x_ptr); |
381 | read_encoded_value_with_base (encoding: ob->s.b.encoding, base, p: y->pc_begin, val: &y_ptr); |
382 | |
383 | if (x_ptr > y_ptr) |
384 | return 1; |
385 | if (x_ptr < y_ptr) |
386 | return -1; |
387 | return 0; |
388 | } |
389 | |
390 | static int |
391 | fde_mixed_encoding_compare (struct object *ob, fde *x, fde *y) |
392 | { |
393 | int x_encoding, y_encoding; |
394 | _Unwind_Ptr x_ptr, y_ptr; |
395 | |
396 | x_encoding = get_fde_encoding (f: x); |
397 | read_encoded_value_with_base (encoding: x_encoding, base: base_from_object (encoding: x_encoding, ob), |
398 | p: x->pc_begin, val: &x_ptr); |
399 | |
400 | y_encoding = get_fde_encoding (f: y); |
401 | read_encoded_value_with_base (encoding: y_encoding, base: base_from_object (encoding: y_encoding, ob), |
402 | p: y->pc_begin, val: &y_ptr); |
403 | |
404 | if (x_ptr > y_ptr) |
405 | return 1; |
406 | if (x_ptr < y_ptr) |
407 | return -1; |
408 | return 0; |
409 | } |
410 | |
411 | typedef int (*fde_compare_t) (struct object *, fde *, fde *); |
412 | |
413 | |
414 | /* This is a special mix of insertion sort and heap sort, optimized for |
415 | the data sets that actually occur. They look like |
416 | 101 102 103 127 128 105 108 110 190 111 115 119 125 160 126 129 130. |
417 | I.e. a linearly increasing sequence (coming from functions in the text |
418 | section), with additionally a few unordered elements (coming from functions |
419 | in gnu_linkonce sections) whose values are higher than the values in the |
420 | surrounding linear sequence (but not necessarily higher than the values |
421 | at the end of the linear sequence!). |
422 | The worst-case total run time is O(N) + O(n log (n)), where N is the |
423 | total number of FDEs and n is the number of erratic ones. */ |
424 | |
425 | struct fde_accumulator |
426 | { |
427 | struct fde_vector *linear; |
428 | struct fde_vector *erratic; |
429 | }; |
430 | |
431 | static int |
432 | start_fde_sort (struct fde_accumulator *accu, size_t count) |
433 | { |
434 | size_t size; |
435 | if (! count) |
436 | return 0; |
437 | |
438 | size = sizeof (struct fde_vector) + sizeof (fde *) * count; |
439 | if ((accu->linear = (struct fde_vector *) malloc (size: size))) |
440 | { |
441 | accu->linear->count = 0; |
442 | if ((accu->erratic = (struct fde_vector *) malloc (size: size))) |
443 | accu->erratic->count = 0; |
444 | return 1; |
445 | } |
446 | else |
447 | return 0; |
448 | } |
449 | |
450 | static inline void |
451 | fde_insert (struct fde_accumulator *accu, fde *this_fde) |
452 | { |
453 | if (accu->linear) |
454 | accu->linear->array[accu->linear->count++] = this_fde; |
455 | } |
456 | |
457 | /* Split LINEAR into a linear sequence with low values and an erratic |
458 | sequence with high values, put the linear one (of longest possible |
459 | length) into LINEAR and the erratic one into ERRATIC. This is O(N). |
460 | |
461 | Because the longest linear sequence we are trying to locate within the |
462 | incoming LINEAR array can be interspersed with (high valued) erratic |
463 | entries. We construct a chain indicating the sequenced entries. |
464 | To avoid having to allocate this chain, we overlay it onto the space of |
465 | the ERRATIC array during construction. A final pass iterates over the |
466 | chain to determine what should be placed in the ERRATIC array, and |
467 | what is the linear sequence. This overlay is safe from aliasing. */ |
468 | |
469 | static void |
470 | fde_split (struct object *ob, fde_compare_t fde_compare, |
471 | struct fde_vector *linear, struct fde_vector *erratic) |
472 | { |
473 | static fde *marker; |
474 | size_t count = linear->count; |
475 | fde **chain_end = ▮ |
476 | size_t i, j, k; |
477 | |
478 | /* This should optimize out, but it is wise to make sure this assumption |
479 | is correct. Should these have different sizes, we cannot cast between |
480 | them and the overlaying onto ERRATIC will not work. */ |
481 | if (sizeof (fde *) != sizeof (fde **)) |
482 | abort (); |
483 | |
484 | for (i = 0; i < count; i++) |
485 | { |
486 | fde **probe; |
487 | |
488 | for (probe = chain_end; |
489 | probe != &marker && fde_compare (ob, linear->array[i], *probe) < 0; |
490 | probe = chain_end) |
491 | { |
492 | chain_end = (fde **) erratic->array[probe - linear->array]; |
493 | erratic->array[probe - linear->array] = NULL; |
494 | } |
495 | erratic->array[i] = (fde *) chain_end; |
496 | chain_end = &linear->array[i]; |
497 | } |
498 | |
499 | /* Each entry in LINEAR which is part of the linear sequence we have |
500 | discovered will correspond to a non-NULL entry in the chain we built in |
501 | the ERRATIC array. */ |
502 | for (i = j = k = 0; i < count; i++) |
503 | if (erratic->array[i]) |
504 | linear->array[j++] = linear->array[i]; |
505 | else |
506 | erratic->array[k++] = linear->array[i]; |
507 | linear->count = j; |
508 | erratic->count = k; |
509 | } |
510 | |
511 | /* This is O(n log(n)). BSD/OS defines heapsort in stdlib.h, so we must |
512 | use a name that does not conflict. */ |
513 | |
514 | static void |
515 | frame_heapsort (struct object *ob, fde_compare_t fde_compare, |
516 | struct fde_vector *erratic) |
517 | { |
518 | /* For a description of this algorithm, see: |
519 | Samuel P. Harbison, Guy L. Steele Jr.: C, a reference manual, 2nd ed., |
520 | p. 60-61. */ |
521 | fde ** a = erratic->array; |
522 | /* A portion of the array is called a "heap" if for all i>=0: |
523 | If i and 2i+1 are valid indices, then a[i] >= a[2i+1]. |
524 | If i and 2i+2 are valid indices, then a[i] >= a[2i+2]. */ |
525 | #define SWAP(x,y) do { fde * tmp = x; x = y; y = tmp; } while (0) |
526 | size_t n = erratic->count; |
527 | size_t m = n; |
528 | size_t i; |
529 | |
530 | while (m > 0) |
531 | { |
532 | /* Invariant: a[m..n-1] is a heap. */ |
533 | m--; |
534 | for (i = m; 2*i+1 < n; ) |
535 | { |
536 | if (2*i+2 < n |
537 | && fde_compare (ob, a[2*i+2], a[2*i+1]) > 0 |
538 | && fde_compare (ob, a[2*i+2], a[i]) > 0) |
539 | { |
540 | SWAP (a[i], a[2*i+2]); |
541 | i = 2*i+2; |
542 | } |
543 | else if (fde_compare (ob, a[2*i+1], a[i]) > 0) |
544 | { |
545 | SWAP (a[i], a[2*i+1]); |
546 | i = 2*i+1; |
547 | } |
548 | else |
549 | break; |
550 | } |
551 | } |
552 | while (n > 1) |
553 | { |
554 | /* Invariant: a[0..n-1] is a heap. */ |
555 | n--; |
556 | SWAP (a[0], a[n]); |
557 | for (i = 0; 2*i+1 < n; ) |
558 | { |
559 | if (2*i+2 < n |
560 | && fde_compare (ob, a[2*i+2], a[2*i+1]) > 0 |
561 | && fde_compare (ob, a[2*i+2], a[i]) > 0) |
562 | { |
563 | SWAP (a[i], a[2*i+2]); |
564 | i = 2*i+2; |
565 | } |
566 | else if (fde_compare (ob, a[2*i+1], a[i]) > 0) |
567 | { |
568 | SWAP (a[i], a[2*i+1]); |
569 | i = 2*i+1; |
570 | } |
571 | else |
572 | break; |
573 | } |
574 | } |
575 | #undef SWAP |
576 | } |
577 | |
578 | /* Merge V1 and V2, both sorted, and put the result into V1. */ |
579 | static void |
580 | fde_merge (struct object *ob, fde_compare_t fde_compare, |
581 | struct fde_vector *v1, struct fde_vector *v2) |
582 | { |
583 | size_t i1, i2; |
584 | fde * fde2; |
585 | |
586 | i2 = v2->count; |
587 | if (i2 > 0) |
588 | { |
589 | i1 = v1->count; |
590 | do |
591 | { |
592 | i2--; |
593 | fde2 = v2->array[i2]; |
594 | while (i1 > 0 && fde_compare (ob, v1->array[i1-1], fde2) > 0) |
595 | { |
596 | v1->array[i1+i2] = v1->array[i1-1]; |
597 | i1--; |
598 | } |
599 | v1->array[i1+i2] = fde2; |
600 | } |
601 | while (i2 > 0); |
602 | v1->count += v2->count; |
603 | } |
604 | } |
605 | |
606 | static void |
607 | end_fde_sort (struct object *ob, struct fde_accumulator *accu, size_t count) |
608 | { |
609 | fde_compare_t fde_compare; |
610 | |
611 | if (accu->linear->count != count) |
612 | abort (); |
613 | |
614 | if (ob->s.b.mixed_encoding) |
615 | fde_compare = fde_mixed_encoding_compare; |
616 | else if (ob->s.b.encoding == DW_EH_PE_absptr) |
617 | fde_compare = fde_unencoded_compare; |
618 | else |
619 | fde_compare = fde_single_encoding_compare; |
620 | |
621 | if (accu->erratic) |
622 | { |
623 | fde_split (ob, fde_compare, linear: accu->linear, erratic: accu->erratic); |
624 | if (accu->linear->count + accu->erratic->count != count) |
625 | abort (); |
626 | frame_heapsort (ob, fde_compare, erratic: accu->erratic); |
627 | fde_merge (ob, fde_compare, v1: accu->linear, v2: accu->erratic); |
628 | free (ptr: accu->erratic); |
629 | } |
630 | else |
631 | { |
632 | /* We've not managed to malloc an erratic array, |
633 | so heap sort in the linear one. */ |
634 | frame_heapsort (ob, fde_compare, erratic: accu->linear); |
635 | } |
636 | } |
637 | |
638 | |
639 | /* Update encoding, mixed_encoding, and pc_begin for OB for the |
640 | fde array beginning at THIS_FDE. Return the number of fdes |
641 | encountered along the way. */ |
642 | |
643 | static size_t |
644 | classify_object_over_fdes (struct object *ob, fde *this_fde) |
645 | { |
646 | struct dwarf_cie *last_cie = 0; |
647 | size_t count = 0; |
648 | int encoding = DW_EH_PE_absptr; |
649 | _Unwind_Ptr base = 0; |
650 | |
651 | for (; ! last_fde (obj: ob, f: this_fde); this_fde = next_fde (f: this_fde)) |
652 | { |
653 | struct dwarf_cie *this_cie; |
654 | _Unwind_Ptr mask, pc_begin; |
655 | |
656 | /* Skip CIEs. */ |
657 | if (this_fde->CIE_delta == 0) |
658 | continue; |
659 | |
660 | /* Determine the encoding for this FDE. Note mixed encoded |
661 | objects for later. */ |
662 | this_cie = get_cie (f: this_fde); |
663 | if (this_cie != last_cie) |
664 | { |
665 | last_cie = this_cie; |
666 | encoding = get_cie_encoding (cie: this_cie); |
667 | base = base_from_object (encoding, ob); |
668 | if (ob->s.b.encoding == DW_EH_PE_omit) |
669 | ob->s.b.encoding = encoding; |
670 | else if (ob->s.b.encoding != encoding) |
671 | ob->s.b.mixed_encoding = 1; |
672 | } |
673 | |
674 | read_encoded_value_with_base (encoding, base, p: this_fde->pc_begin, |
675 | val: &pc_begin); |
676 | |
677 | /* Take care to ignore link-once functions that were removed. |
678 | In these cases, the function address will be NULL, but if |
679 | the encoding is smaller than a pointer a true NULL may not |
680 | be representable. Assume 0 in the representable bits is NULL. */ |
681 | mask = size_of_encoded_value (encoding); |
682 | if (mask < sizeof (void *)) |
683 | mask = (1L << (mask << 3)) - 1; |
684 | else |
685 | mask = -1; |
686 | |
687 | if ((pc_begin & mask) == 0) |
688 | continue; |
689 | |
690 | count += 1; |
691 | if ((void *) pc_begin < ob->pc_begin) |
692 | ob->pc_begin = (void *) pc_begin; |
693 | } |
694 | |
695 | return count; |
696 | } |
697 | |
698 | static void |
699 | add_fdes (struct object *ob, struct fde_accumulator *accu, fde *this_fde) |
700 | { |
701 | struct dwarf_cie *last_cie = 0; |
702 | int encoding = ob->s.b.encoding; |
703 | _Unwind_Ptr base = base_from_object (encoding: ob->s.b.encoding, ob); |
704 | |
705 | for (; ! last_fde (obj: ob, f: this_fde); this_fde = next_fde (f: this_fde)) |
706 | { |
707 | struct dwarf_cie *this_cie; |
708 | |
709 | /* Skip CIEs. */ |
710 | if (this_fde->CIE_delta == 0) |
711 | continue; |
712 | |
713 | if (ob->s.b.mixed_encoding) |
714 | { |
715 | /* Determine the encoding for this FDE. Note mixed encoded |
716 | objects for later. */ |
717 | this_cie = get_cie (f: this_fde); |
718 | if (this_cie != last_cie) |
719 | { |
720 | last_cie = this_cie; |
721 | encoding = get_cie_encoding (cie: this_cie); |
722 | base = base_from_object (encoding, ob); |
723 | } |
724 | } |
725 | |
726 | if (encoding == DW_EH_PE_absptr) |
727 | { |
728 | if (get_pc_begin (x: this_fde, n: 0) == 0) |
729 | continue; |
730 | } |
731 | else |
732 | { |
733 | _Unwind_Ptr pc_begin, mask; |
734 | |
735 | read_encoded_value_with_base (encoding, base, p: this_fde->pc_begin, |
736 | val: &pc_begin); |
737 | |
738 | /* Take care to ignore link-once functions that were removed. |
739 | In these cases, the function address will be NULL, but if |
740 | the encoding is smaller than a pointer a true NULL may not |
741 | be representable. Assume 0 in the representable bits is NULL. */ |
742 | mask = size_of_encoded_value (encoding); |
743 | if (mask < sizeof (void *)) |
744 | mask = (1L << (mask << 3)) - 1; |
745 | else |
746 | mask = -1; |
747 | |
748 | if ((pc_begin & mask) == 0) |
749 | continue; |
750 | } |
751 | |
752 | fde_insert (accu, this_fde); |
753 | } |
754 | } |
755 | |
756 | /* Set up a sorted array of pointers to FDEs for a loaded object. We |
757 | count up the entries before allocating the array because it's likely to |
758 | be faster. We can be called multiple times, should we have failed to |
759 | allocate a sorted fde array on a previous occasion. */ |
760 | |
761 | static void |
762 | init_object (struct object* ob) |
763 | { |
764 | struct fde_accumulator accu; |
765 | size_t count; |
766 | |
767 | count = ob->s.b.count; |
768 | if (count == 0) |
769 | { |
770 | if (ob->s.b.from_array) |
771 | { |
772 | fde **p = ob->u.array; |
773 | for (count = 0; *p; ++p) |
774 | count += classify_object_over_fdes (ob, this_fde: *p); |
775 | } |
776 | else |
777 | count = classify_object_over_fdes (ob, this_fde: ob->u.single); |
778 | |
779 | /* The count field we have in the main struct object is somewhat |
780 | limited, but should suffice for virtually all cases. If the |
781 | counted value doesn't fit, re-write a zero. The worst that |
782 | happens is that we re-count next time -- admittedly non-trivial |
783 | in that this implies some 2M fdes, but at least we function. */ |
784 | ob->s.b.count = count; |
785 | if (ob->s.b.count != count) |
786 | ob->s.b.count = 0; |
787 | } |
788 | |
789 | if (!start_fde_sort (accu: &accu, count)) |
790 | return; |
791 | |
792 | if (ob->s.b.from_array) |
793 | { |
794 | fde **p; |
795 | for (p = ob->u.array; *p; ++p) |
796 | add_fdes (ob, accu: &accu, this_fde: *p); |
797 | } |
798 | else |
799 | add_fdes (ob, accu: &accu, this_fde: ob->u.single); |
800 | |
801 | end_fde_sort (ob, accu: &accu, count); |
802 | |
803 | /* Save the original fde pointer, since this is the key by which the |
804 | DSO will deregister the object. */ |
805 | accu.linear->orig_data = ob->u.single; |
806 | ob->u.sort = accu.linear; |
807 | |
808 | ob->s.b.sorted = 1; |
809 | } |
810 | |
811 | /* A linear search through a set of FDEs for the given PC. This is |
812 | used when there was insufficient memory to allocate and sort an |
813 | array. */ |
814 | |
815 | static fde * |
816 | linear_search_fdes (struct object *ob, fde *this_fde, void *pc) |
817 | { |
818 | struct dwarf_cie *last_cie = 0; |
819 | int encoding = ob->s.b.encoding; |
820 | _Unwind_Ptr base = base_from_object (encoding: ob->s.b.encoding, ob); |
821 | |
822 | for (; ! last_fde (obj: ob, f: this_fde); this_fde = next_fde (f: this_fde)) |
823 | { |
824 | struct dwarf_cie *this_cie; |
825 | _Unwind_Ptr pc_begin, pc_range; |
826 | |
827 | /* Skip CIEs. */ |
828 | if (this_fde->CIE_delta == 0) |
829 | continue; |
830 | |
831 | if (ob->s.b.mixed_encoding) |
832 | { |
833 | /* Determine the encoding for this FDE. Note mixed encoded |
834 | objects for later. */ |
835 | this_cie = get_cie (f: this_fde); |
836 | if (this_cie != last_cie) |
837 | { |
838 | last_cie = this_cie; |
839 | encoding = get_cie_encoding (cie: this_cie); |
840 | base = base_from_object (encoding, ob); |
841 | } |
842 | } |
843 | |
844 | if (encoding == DW_EH_PE_absptr) |
845 | { |
846 | pc_begin = get_pc_begin (x: this_fde, n: 0); |
847 | pc_range = get_pc_begin (x: this_fde, n: 1); |
848 | if (pc_begin == 0) |
849 | continue; |
850 | } |
851 | else |
852 | { |
853 | _Unwind_Ptr mask; |
854 | const unsigned char *p; |
855 | |
856 | p = read_encoded_value_with_base (encoding, base, |
857 | p: this_fde->pc_begin, val: &pc_begin); |
858 | read_encoded_value_with_base (encoding: encoding & 0x0F, base: 0, p, val: &pc_range); |
859 | |
860 | /* Take care to ignore link-once functions that were removed. |
861 | In these cases, the function address will be NULL, but if |
862 | the encoding is smaller than a pointer a true NULL may not |
863 | be representable. Assume 0 in the representable bits is NULL. */ |
864 | mask = size_of_encoded_value (encoding); |
865 | if (mask < sizeof (void *)) |
866 | mask = (1L << (mask << 3)) - 1; |
867 | else |
868 | mask = -1; |
869 | |
870 | if ((pc_begin & mask) == 0) |
871 | continue; |
872 | } |
873 | |
874 | if ((_Unwind_Ptr) pc - pc_begin < pc_range) |
875 | return this_fde; |
876 | } |
877 | |
878 | return NULL; |
879 | } |
880 | |
881 | /* Binary search for an FDE containing the given PC. Here are three |
882 | implementations of increasing complexity. */ |
883 | |
884 | static fde * |
885 | binary_search_unencoded_fdes (struct object *ob, void *pc) |
886 | { |
887 | struct fde_vector *vec = ob->u.sort; |
888 | size_t lo, hi; |
889 | |
890 | for (lo = 0, hi = vec->count; lo < hi; ) |
891 | { |
892 | size_t i = (lo + hi) / 2; |
893 | fde *f = vec->array[i]; |
894 | void *pc_begin; |
895 | uaddr pc_range; |
896 | |
897 | pc_begin = (void *) get_pc_begin (x: f, n: 0); |
898 | pc_range = (uaddr) get_pc_begin (x: f, n: 1); |
899 | |
900 | if (pc < pc_begin) |
901 | hi = i; |
902 | else if (pc >= pc_begin + pc_range) |
903 | lo = i + 1; |
904 | else |
905 | return f; |
906 | } |
907 | |
908 | return NULL; |
909 | } |
910 | |
911 | static fde * |
912 | binary_search_single_encoding_fdes (struct object *ob, void *pc) |
913 | { |
914 | struct fde_vector *vec = ob->u.sort; |
915 | int encoding = ob->s.b.encoding; |
916 | _Unwind_Ptr base = base_from_object (encoding, ob); |
917 | size_t lo, hi; |
918 | |
919 | for (lo = 0, hi = vec->count; lo < hi; ) |
920 | { |
921 | size_t i = (lo + hi) / 2; |
922 | fde *f = vec->array[i]; |
923 | _Unwind_Ptr pc_begin, pc_range; |
924 | const unsigned char *p; |
925 | |
926 | p = read_encoded_value_with_base (encoding, base, p: f->pc_begin, |
927 | val: &pc_begin); |
928 | read_encoded_value_with_base (encoding: encoding & 0x0F, base: 0, p, val: &pc_range); |
929 | |
930 | if ((_Unwind_Ptr) pc < pc_begin) |
931 | hi = i; |
932 | else if ((_Unwind_Ptr) pc >= pc_begin + pc_range) |
933 | lo = i + 1; |
934 | else |
935 | return f; |
936 | } |
937 | |
938 | return NULL; |
939 | } |
940 | |
941 | static fde * |
942 | binary_search_mixed_encoding_fdes (struct object *ob, void *pc) |
943 | { |
944 | struct fde_vector *vec = ob->u.sort; |
945 | size_t lo, hi; |
946 | |
947 | for (lo = 0, hi = vec->count; lo < hi; ) |
948 | { |
949 | size_t i = (lo + hi) / 2; |
950 | fde *f = vec->array[i]; |
951 | _Unwind_Ptr pc_begin, pc_range; |
952 | const unsigned char *p; |
953 | int encoding; |
954 | |
955 | encoding = get_fde_encoding (f); |
956 | p = read_encoded_value_with_base (encoding, |
957 | base: base_from_object (encoding, ob), |
958 | p: f->pc_begin, val: &pc_begin); |
959 | read_encoded_value_with_base (encoding: encoding & 0x0F, base: 0, p, val: &pc_range); |
960 | |
961 | if ((_Unwind_Ptr) pc < pc_begin) |
962 | hi = i; |
963 | else if ((_Unwind_Ptr) pc >= pc_begin + pc_range) |
964 | lo = i + 1; |
965 | else |
966 | return f; |
967 | } |
968 | |
969 | return NULL; |
970 | } |
971 | |
972 | static fde * |
973 | search_object (struct object* ob, void *pc) |
974 | { |
975 | /* If the data hasn't been sorted, try to do this now. We may have |
976 | more memory available than last time we tried. */ |
977 | if (! ob->s.b.sorted) |
978 | { |
979 | init_object (ob); |
980 | |
981 | /* Despite the above comment, the normal reason to get here is |
982 | that we've not processed this object before. A quick range |
983 | check is in order. */ |
984 | if (pc < ob->pc_begin) |
985 | return NULL; |
986 | } |
987 | |
988 | if (ob->s.b.sorted) |
989 | { |
990 | if (ob->s.b.mixed_encoding) |
991 | return binary_search_mixed_encoding_fdes (ob, pc); |
992 | else if (ob->s.b.encoding == DW_EH_PE_absptr) |
993 | return binary_search_unencoded_fdes (ob, pc); |
994 | else |
995 | return binary_search_single_encoding_fdes (ob, pc); |
996 | } |
997 | else |
998 | { |
999 | /* Long slow labourious linear search, cos we've no memory. */ |
1000 | if (ob->s.b.from_array) |
1001 | { |
1002 | fde **p; |
1003 | for (p = ob->u.array; *p ; p++) |
1004 | { |
1005 | fde *f = linear_search_fdes (ob, this_fde: *p, pc); |
1006 | if (f) |
1007 | return f; |
1008 | } |
1009 | return NULL; |
1010 | } |
1011 | else |
1012 | return linear_search_fdes (ob, this_fde: ob->u.single, pc); |
1013 | } |
1014 | } |
1015 | |
1016 | fde * |
1017 | _Unwind_Find_FDE (void *pc, struct dwarf_eh_bases *bases) |
1018 | { |
1019 | struct object *ob; |
1020 | fde *f = NULL; |
1021 | |
1022 | init_object_mutex_once (); |
1023 | __gthread_mutex_lock (&object_mutex); |
1024 | |
1025 | /* Linear search through the classified objects, to find the one |
1026 | containing the pc. Note that pc_begin is sorted descending, and |
1027 | we expect objects to be non-overlapping. */ |
1028 | for (ob = seen_objects; ob; ob = ob->next) |
1029 | if (pc >= ob->pc_begin) |
1030 | { |
1031 | f = search_object (ob, pc); |
1032 | if (f) |
1033 | goto fini; |
1034 | break; |
1035 | } |
1036 | |
1037 | /* Classify and search the objects we've not yet processed. */ |
1038 | while ((ob = unseen_objects)) |
1039 | { |
1040 | struct object **p; |
1041 | |
1042 | unseen_objects = ob->next; |
1043 | f = search_object (ob, pc); |
1044 | |
1045 | /* Insert the object into the classified list. */ |
1046 | for (p = &seen_objects; *p ; p = &(*p)->next) |
1047 | if ((*p)->pc_begin < ob->pc_begin) |
1048 | break; |
1049 | ob->next = *p; |
1050 | *p = ob; |
1051 | |
1052 | if (f) |
1053 | goto fini; |
1054 | } |
1055 | |
1056 | fini: |
1057 | __gthread_mutex_unlock (&object_mutex); |
1058 | |
1059 | if (f) |
1060 | { |
1061 | int encoding; |
1062 | _Unwind_Ptr func; |
1063 | |
1064 | bases->tbase = ob->tbase; |
1065 | bases->dbase = ob->dbase; |
1066 | |
1067 | encoding = ob->s.b.encoding; |
1068 | if (ob->s.b.mixed_encoding) |
1069 | encoding = get_fde_encoding (f); |
1070 | read_encoded_value_with_base (encoding, base: base_from_object (encoding, ob), |
1071 | p: f->pc_begin, val: &func); |
1072 | bases->func = (void *) func; |
1073 | } |
1074 | |
1075 | return f; |
1076 | } |
1077 | |
1078 | #endif |
1079 | |