]>
Commit | Line | Data |
---|---|---|
ffd652c3 ILT |
1 | /* ehopt.c--optimize gcc exception frame information. |
2 | Copyright (C) 1998 Free Software Foundation, Inc. | |
3 | Written by Ian Lance Taylor <[email protected]>. | |
4 | ||
5 | This file is part of GAS, the GNU Assembler. | |
6 | ||
7 | GAS is free software; you can redistribute it and/or modify | |
8 | it under the terms of the GNU General Public License as published by | |
9 | the Free Software Foundation; either version 2, or (at your option) | |
10 | any later version. | |
11 | ||
12 | GAS is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | GNU General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
18 | along with GAS; see the file COPYING. If not, write to the Free | |
19 | Software Foundation, 59 Temple Place - Suite 330, Boston, MA | |
20 | 02111-1307, USA. */ | |
21 | ||
22 | #include "as.h" | |
23 | #include "subsegs.h" | |
24 | ||
25 | /* We include this ELF file, even though we may not be assembling for | |
26 | ELF, since the exception frame information is always in a format | |
27 | derived from DWARF. */ | |
28 | ||
29 | #include "elf/dwarf2.h" | |
30 | ||
31 | /* Try to optimize gcc 2.8 exception frame information. | |
32 | ||
33 | Exception frame information is emitted for every function in the | |
34 | .eh_frame section. Simple information for a function with no | |
35 | exceptions looks like this: | |
36 | ||
37 | __FRAME_BEGIN__: | |
38 | .4byte .LLCIE1 / Length of Common Information Entry | |
39 | .LSCIE1: | |
40 | .4byte 0x0 / CIE Identifier Tag | |
41 | .byte 0x1 / CIE Version | |
42 | .byte 0x0 / CIE Augmentation (none) | |
43 | .byte 0x1 / ULEB128 0x1 (CIE Code Alignment Factor) | |
44 | .byte 0x7c / SLEB128 -4 (CIE Data Alignment Factor) | |
45 | .byte 0x8 / CIE RA Column | |
46 | .byte 0xc / DW_CFA_def_cfa | |
47 | .byte 0x4 / ULEB128 0x4 | |
48 | .byte 0x4 / ULEB128 0x4 | |
49 | .byte 0x88 / DW_CFA_offset, column 0x8 | |
50 | .byte 0x1 / ULEB128 0x1 | |
51 | .align 4 | |
52 | .LECIE1: | |
53 | .set .LLCIE1,.LECIE1-.LSCIE1 / CIE Length Symbol | |
54 | .4byte .LLFDE1 / FDE Length | |
55 | .LSFDE1: | |
56 | .4byte .LSFDE1-__FRAME_BEGIN__ / FDE CIE offset | |
57 | .4byte .LFB1 / FDE initial location | |
58 | .4byte .LFE1-.LFB1 / FDE address range | |
59 | .byte 0x4 / DW_CFA_advance_loc4 | |
60 | .4byte .LCFI0-.LFB1 | |
61 | .byte 0xe / DW_CFA_def_cfa_offset | |
62 | .byte 0x8 / ULEB128 0x8 | |
63 | .byte 0x85 / DW_CFA_offset, column 0x5 | |
64 | .byte 0x2 / ULEB128 0x2 | |
65 | .byte 0x4 / DW_CFA_advance_loc4 | |
66 | .4byte .LCFI1-.LCFI0 | |
67 | .byte 0xd / DW_CFA_def_cfa_register | |
68 | .byte 0x5 / ULEB128 0x5 | |
69 | .byte 0x4 / DW_CFA_advance_loc4 | |
70 | .4byte .LCFI2-.LCFI1 | |
71 | .byte 0x2e / DW_CFA_GNU_args_size | |
72 | .byte 0x4 / ULEB128 0x4 | |
73 | .byte 0x4 / DW_CFA_advance_loc4 | |
74 | .4byte .LCFI3-.LCFI2 | |
75 | .byte 0x2e / DW_CFA_GNU_args_size | |
76 | .byte 0x0 / ULEB128 0x0 | |
77 | .align 4 | |
78 | .LEFDE1: | |
79 | .set .LLFDE1,.LEFDE1-.LSFDE1 / FDE Length Symbol | |
80 | ||
81 | The immediate issue we can address in the assembler is the | |
82 | DW_CFA_advance_loc4 followed by a four byte value. The value is | |
83 | the difference of two addresses in the function. Since gcc does | |
84 | not know this value, it always uses four bytes. We will know the | |
85 | value at the end of assembly, so we can do better. */ | |
86 | ||
87 | static int eh_frame_code_alignment PARAMS ((void)); | |
88 | ||
89 | /* Get the code alignment factor from the CIE. */ | |
90 | ||
91 | static int | |
92 | eh_frame_code_alignment () | |
93 | { | |
94 | static int code_alignment; | |
95 | segT current_seg; | |
96 | subsegT current_subseg; | |
97 | fragS *f; | |
1430b6ed | 98 | fixS *fix; |
ffd652c3 | 99 | int offset; |
eb0a94a1 ILT |
100 | char augmentation[10]; |
101 | int iaug; | |
ffd652c3 ILT |
102 | |
103 | if (code_alignment != 0) | |
104 | return code_alignment; | |
105 | ||
106 | /* We should find the CIE at the start of the .eh_frame section. */ | |
107 | ||
108 | current_seg = now_seg; | |
109 | current_subseg = now_subseg; | |
110 | subseg_new (".eh_frame", 0); | |
111 | f = seg_info (now_seg)->frchainP->frch_root; | |
1430b6ed | 112 | fix = seg_info (now_seg)->frchainP->fix_root; |
ffd652c3 ILT |
113 | subseg_set (current_seg, current_subseg); |
114 | ||
115 | /* Look through the frags of the section to find the code alignment. */ | |
116 | ||
117 | /* First make sure that the CIE Identifier Tag is 0. */ | |
118 | ||
119 | offset = 4; | |
120 | while (f != NULL && offset >= f->fr_fix) | |
121 | { | |
122 | offset -= f->fr_fix; | |
123 | f = f->fr_next; | |
124 | } | |
125 | if (f == NULL | |
126 | || f->fr_fix - offset < 4 | |
127 | || f->fr_literal[offset] != 0 | |
128 | || f->fr_literal[offset + 1] != 0 | |
129 | || f->fr_literal[offset + 2] != 0 | |
130 | || f->fr_literal[offset + 3] != 0) | |
131 | { | |
132 | code_alignment = -1; | |
133 | return -1; | |
134 | } | |
135 | ||
136 | /* Next make sure the CIE version number is 1. */ | |
137 | ||
138 | offset += 4; | |
139 | while (f != NULL && offset >= f->fr_fix) | |
140 | { | |
141 | offset -= f->fr_fix; | |
142 | f = f->fr_next; | |
143 | } | |
144 | if (f == NULL | |
145 | || f->fr_fix - offset < 1 | |
146 | || f->fr_literal[offset] != 1) | |
147 | { | |
148 | code_alignment = -1; | |
149 | return -1; | |
150 | } | |
151 | ||
152 | /* Skip the augmentation (a null terminated string). */ | |
153 | ||
eb0a94a1 | 154 | iaug = 0; |
ffd652c3 ILT |
155 | ++offset; |
156 | while (1) | |
157 | { | |
158 | while (f != NULL && offset >= f->fr_fix) | |
159 | { | |
160 | offset -= f->fr_fix; | |
161 | f = f->fr_next; | |
162 | } | |
163 | if (f == NULL) | |
164 | { | |
165 | code_alignment = -1; | |
166 | return -1; | |
167 | } | |
168 | while (offset < f->fr_fix && f->fr_literal[offset] != '\0') | |
1430b6ed | 169 | { |
eb0a94a1 | 170 | if (iaug < (sizeof augmentation) - 1) |
1430b6ed | 171 | { |
eb0a94a1 ILT |
172 | augmentation[iaug] = f->fr_literal[offset]; |
173 | ++iaug; | |
1430b6ed ILT |
174 | } |
175 | ++offset; | |
176 | } | |
ffd652c3 ILT |
177 | if (offset < f->fr_fix) |
178 | break; | |
179 | } | |
180 | ++offset; | |
181 | while (f != NULL && offset >= f->fr_fix) | |
182 | { | |
183 | offset -= f->fr_fix; | |
184 | f = f->fr_next; | |
185 | } | |
186 | if (f == NULL) | |
187 | { | |
188 | code_alignment = -1; | |
189 | return -1; | |
190 | } | |
191 | ||
eb0a94a1 ILT |
192 | augmentation[iaug] = '\0'; |
193 | if (augmentation[0] == '\0') | |
1430b6ed | 194 | { |
eb0a94a1 ILT |
195 | /* No augmentation. */ |
196 | } | |
197 | else if (strcmp (augmentation, "eh") == 0) | |
198 | { | |
199 | /* We have to skip a pointer. Unfortunately, we don't know how | |
200 | large it is. We find out by looking for a matching fixup. */ | |
1430b6ed ILT |
201 | while (fix != NULL |
202 | && (fix->fx_frag != f || fix->fx_where != offset)) | |
203 | fix = fix->fx_next; | |
204 | if (fix == NULL) | |
205 | offset += 4; | |
206 | else | |
207 | offset += fix->fx_size; | |
208 | while (f != NULL && offset >= f->fr_fix) | |
209 | { | |
210 | offset -= f->fr_fix; | |
211 | f = f->fr_next; | |
212 | } | |
213 | if (f == NULL) | |
214 | { | |
215 | code_alignment = -1; | |
216 | return -1; | |
217 | } | |
218 | } | |
eb0a94a1 ILT |
219 | else |
220 | { | |
221 | code_alignment = -1; | |
222 | return -1; | |
223 | } | |
1430b6ed | 224 | |
ffd652c3 ILT |
225 | /* We're now at the code alignment factor, which is a ULEB128. If |
226 | it isn't a single byte, forget it. */ | |
227 | ||
228 | code_alignment = f->fr_literal[offset] & 0xff; | |
229 | if ((code_alignment & 0x80) != 0 || code_alignment == 0) | |
230 | { | |
231 | code_alignment = -1; | |
232 | return -1; | |
233 | } | |
234 | ||
235 | return code_alignment; | |
236 | } | |
237 | ||
238 | /* This function is called from emit_expr. It looks for cases which | |
239 | we can optimize. | |
240 | ||
241 | Rather than try to parse all this information as we read it, we | |
242 | look for a single byte DW_CFA_advance_loc4 followed by a 4 byte | |
243 | difference. We turn that into a rs_cfa_advance frag, and handle | |
244 | those frags at the end of the assembly. If the gcc output changes | |
245 | somewhat, this optimization may stop working. | |
246 | ||
247 | This function returns non-zero if it handled the expression and | |
248 | emit_expr should not do anything, or zero otherwise. It can also | |
249 | change *EXP and *PNBYTES. */ | |
250 | ||
251 | int | |
252 | check_eh_frame (exp, pnbytes) | |
253 | expressionS *exp; | |
254 | unsigned int *pnbytes; | |
255 | { | |
256 | static int saw_advance_loc4; | |
257 | static fragS *loc4_frag; | |
258 | static int loc4_fix; | |
259 | ||
260 | if (flag_traditional_format) | |
261 | { | |
262 | /* Don't optimize. */ | |
263 | } | |
264 | else if (strcmp (segment_name (now_seg), ".eh_frame") != 0) | |
265 | saw_advance_loc4 = 0; | |
266 | else if (*pnbytes == 1 | |
267 | && exp->X_op == O_constant | |
268 | && exp->X_add_number == DW_CFA_advance_loc4) | |
269 | { | |
270 | /* This might be a DW_CFA_advance_loc4. Record the frag and the | |
271 | position within the frag, so that we can change it later. */ | |
272 | saw_advance_loc4 = 1; | |
273 | frag_grow (1); | |
274 | loc4_frag = frag_now; | |
275 | loc4_fix = frag_now_fix (); | |
276 | } | |
277 | else if (saw_advance_loc4 | |
278 | && *pnbytes == 4 | |
279 | && exp->X_op == O_constant) | |
280 | { | |
281 | int ca; | |
282 | ||
283 | /* This is a case which we can optimize. The two symbols being | |
284 | subtracted were in the same frag and the expression was | |
285 | reduced to a constant. We can do the optimization entirely | |
286 | in this function. */ | |
287 | ||
288 | saw_advance_loc4 = 0; | |
289 | ||
290 | ca = eh_frame_code_alignment (); | |
291 | if (ca < 0) | |
292 | { | |
293 | /* Don't optimize. */ | |
294 | } | |
295 | else if (exp->X_add_number % ca == 0 | |
1430b6ed | 296 | && exp->X_add_number / ca < 0x40) |
ffd652c3 ILT |
297 | { |
298 | loc4_frag->fr_literal[loc4_fix] | |
299 | = DW_CFA_advance_loc | (exp->X_add_number / ca); | |
300 | /* No more bytes needed. */ | |
301 | return 1; | |
302 | } | |
303 | else if (exp->X_add_number < 0x100) | |
304 | { | |
305 | loc4_frag->fr_literal[loc4_fix] = DW_CFA_advance_loc1; | |
306 | *pnbytes = 1; | |
307 | } | |
308 | else if (exp->X_add_number < 0x10000) | |
309 | { | |
310 | loc4_frag->fr_literal[loc4_fix] = DW_CFA_advance_loc2; | |
311 | *pnbytes = 2; | |
312 | } | |
313 | } | |
314 | else if (saw_advance_loc4 | |
315 | && *pnbytes == 4 | |
316 | && exp->X_op == O_subtract) | |
317 | { | |
318 | ||
319 | /* This is a case we can optimize. The expression was not | |
320 | reduced, so we can not finish the optimization until the end | |
321 | of the assembly. We set up a variant frag which we handle | |
322 | later. */ | |
323 | ||
324 | saw_advance_loc4 = 0; | |
325 | ||
326 | frag_var (rs_cfa, 4, 0, 0, make_expr_symbol (exp), | |
327 | loc4_fix, (char *) loc4_frag); | |
1430b6ed ILT |
328 | |
329 | return 1; | |
ffd652c3 ILT |
330 | } |
331 | else | |
332 | saw_advance_loc4 = 0; | |
333 | ||
334 | return 0; | |
335 | } | |
336 | ||
337 | /* The function estimates the size of a rs_cfa variant frag based on | |
338 | the current values of the symbols. It is called before the | |
339 | relaxation loop. We set fr_subtype to the expected length. */ | |
340 | ||
341 | int | |
342 | eh_frame_estimate_size_before_relax (frag) | |
343 | fragS *frag; | |
344 | { | |
345 | int ca; | |
346 | offsetT diff; | |
347 | int ret; | |
348 | ||
349 | ca = eh_frame_code_alignment (); | |
350 | diff = resolve_symbol_value (frag->fr_symbol, 0); | |
351 | ||
352 | if (ca < 0) | |
353 | ret = 4; | |
354 | else if (diff % ca == 0 && diff / ca < 0x40) | |
355 | ret = 0; | |
356 | else if (diff < 0x100) | |
357 | ret = 1; | |
358 | else if (diff < 0x10000) | |
359 | ret = 2; | |
360 | else | |
361 | ret = 4; | |
362 | ||
363 | frag->fr_subtype = ret; | |
364 | ||
365 | return ret; | |
366 | } | |
367 | ||
368 | /* This function relaxes a rs_cfa variant frag based on the current | |
369 | values of the symbols. fr_subtype is the current length of the | |
370 | frag. This returns the change in frag length. */ | |
371 | ||
372 | int | |
373 | eh_frame_relax_frag (frag) | |
374 | fragS *frag; | |
375 | { | |
376 | int oldsize, newsize; | |
377 | ||
378 | oldsize = frag->fr_subtype; | |
379 | newsize = eh_frame_estimate_size_before_relax (frag); | |
380 | return newsize - oldsize; | |
381 | } | |
382 | ||
383 | /* This function converts a rs_cfa variant frag into a normal fill | |
384 | frag. This is called after all relaxation has been done. | |
385 | fr_subtype will be the desired length of the frag. */ | |
386 | ||
387 | void | |
388 | eh_frame_convert_frag (frag) | |
389 | fragS *frag; | |
390 | { | |
391 | offsetT diff; | |
392 | fragS *loc4_frag; | |
393 | int loc4_fix; | |
394 | ||
395 | loc4_frag = (fragS *) frag->fr_opcode; | |
396 | loc4_fix = (int) frag->fr_offset; | |
397 | ||
398 | diff = resolve_symbol_value (frag->fr_symbol, 1); | |
399 | ||
400 | if (frag->fr_subtype == 0) | |
401 | { | |
402 | int ca; | |
403 | ||
404 | ca = eh_frame_code_alignment (); | |
405 | assert (ca > 0 && diff % ca == 0 && diff / ca < 0x40); | |
406 | loc4_frag->fr_literal[loc4_fix] = DW_CFA_advance_loc | (diff / ca); | |
407 | } | |
408 | else if (frag->fr_subtype == 1) | |
409 | { | |
410 | assert (diff < 0x100); | |
411 | loc4_frag->fr_literal[loc4_fix] = DW_CFA_advance_loc1; | |
412 | frag->fr_literal[frag->fr_fix] = diff; | |
413 | } | |
414 | else if (frag->fr_subtype == 2) | |
415 | { | |
416 | assert (diff < 0x10000); | |
417 | loc4_frag->fr_literal[loc4_fix] = DW_CFA_advance_loc2; | |
418 | md_number_to_chars (frag->fr_literal + frag->fr_fix, diff, 2); | |
419 | } | |
420 | else | |
421 | md_number_to_chars (frag->fr_literal + frag->fr_fix, diff, 4); | |
422 | ||
423 | frag->fr_fix += frag->fr_subtype; | |
424 | frag->fr_type = rs_fill; | |
1430b6ed | 425 | frag->fr_offset = 0; |
ffd652c3 | 426 | } |