xref: /illumos-gate/usr/src/lib/libdwarf/common/dwarf_util.h (revision 142d813a06c6f9a6142e2c276b62129a17a31a65)
1 #ifndef DWARF_UTIL_H
2 #define DWARF_UTIL_H
3 /*
4 
5   Copyright (C) 2000,2003,2004 Silicon Graphics, Inc.  All Rights Reserved.
6   Portions Copyright (C) 2007-2010 David Anderson. All Rights Reserved.
7 
8   This program is free software; you can redistribute it and/or modify it
9   under the terms of version 2.1 of the GNU Lesser General Public License
10   as published by the Free Software Foundation.
11 
12   This program is distributed in the hope that it would be useful, but
13   WITHOUT ANY WARRANTY; without even the implied warranty of
14   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
15 
16   Further, this software is distributed without any warranty that it is
17   free of the rightful claim of any third person regarding infringement
18   or the like.  Any license provided herein, whether implied or
19   otherwise, applies only to this software file.  Patent licenses, if
20   any, provided herein do not apply to combinations of this program with
21   other software, or any other product whatsoever.
22 
23   You should have received a copy of the GNU Lesser General Public
24   License along with this program; if not, write the Free Software
25   Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston MA 02110-1301,
26   USA.
27 
28   Contact information:  Silicon Graphics, Inc., 1500 Crittenden Lane,
29   Mountain View, CA 94043, or:
30 
31   http://www.sgi.com
32 
33   For further information regarding this notice, see:
34 
35   http://oss.sgi.com/projects/GenInfo/NoticeExplan
36 
37 */
38 /* The address of the Free Software Foundation is
39    Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
40    Boston, MA 02110-1301, USA.
41    SGI has moved from the Crittenden Lane address.
42 */
43 
44 
45 
46 
47 /*
48     Decodes unsigned leb128 encoded numbers.
49     Make sure ptr is a pointer to a 1-byte type.
50     In 2003 and earlier this was a hand-inlined
51     version of _dwarf_decode_u_leb128() which did
52     not work correctly if Dwarf_Word was 64 bits.
53 */
54 #define DECODE_LEB128_UWORD(ptr, value) \
55     do { \
56        Dwarf_Word uleblen; \
57 	value = _dwarf_decode_u_leb128(ptr,&uleblen); \
58         ptr += uleblen; \
59     } while (0)
60 
61 /*
62     Decodes signed leb128 encoded numbers.
63     Make sure ptr is a pointer to a 1-byte type.
64     In 2003 and earlier this was a hand-inlined
65     version of _dwarf_decode_s_leb128() which did
66     not work correctly if Dwarf_Word was 64 bits.
67 
68 */
69 #define DECODE_LEB128_SWORD(ptr, value) \
70     do { \
71        Dwarf_Word sleblen; \
72 	value = _dwarf_decode_s_leb128(ptr,&sleblen); \
73         ptr += sleblen; \
74     } while(0)
75 
76 
77 /*
78     Skips leb128_encoded numbers that are guaranteed
79     to be no more than 4 bytes long.  Same for both
80     signed and unsigned numbers.
81 */
82 #define SKIP_LEB128_WORD(ptr) \
83     do{ if ((*(ptr++) & 0x80) != 0) { \
84         if ((*(ptr++) & 0x80) != 0) { \
85             if ((*(ptr++) & 0x80) != 0) { \
86 	        if ((*(ptr++) & 0x80) != 0) { \
87 	        } \
88 	    } \
89         } \
90     } } while (0)
91 
92 
93 #define CHECK_DIE(die, error_ret_value) \
94 do {if (die == NULL) { \
95 	_dwarf_error(NULL, error, DW_DLE_DIE_NULL); \
96 	return(error_ret_value); \
97     } \
98     if (die->di_cu_context == NULL) { \
99 	_dwarf_error(NULL, error, DW_DLE_DIE_NO_CU_CONTEXT); \
100 	return(error_ret_value); \
101     } \
102     if (die->di_cu_context->cc_dbg == NULL) { \
103 	_dwarf_error(NULL, error, DW_DLE_DBG_NULL); \
104 	return(error_ret_value); \
105     }  \
106 } while (0)
107 
108 
109 /*
110    Reads 'source' for 'length' bytes from unaligned addr.
111 
112    Avoids any constant-in-conditional warnings and
113    avoids a test in the generated code (for non-const cases,
114 	which are in the majority.)
115    Uses a temp to avoid the test.
116    The decl here should avoid any problem of size in the temp.
117    This code is ENDIAN DEPENDENT
118    The memcpy args are the endian issue.
119 */
120 typedef Dwarf_Unsigned BIGGEST_UINT;
121 
122 #ifdef WORDS_BIGENDIAN
123 #define READ_UNALIGNED(dbg,dest,desttype, source, length) \
124     do { \
125       BIGGEST_UINT _ltmp = 0;  \
126       dbg->de_copy_word( (((char *)(&_ltmp)) + sizeof(_ltmp) - length), \
127 			source, length) ; \
128       dest = (desttype)_ltmp;  \
129     } while (0)
130 
131 
132 /*
133     This macro sign-extends a variable depending on the length.
134     It fills the bytes between the size of the destination and
135     the length with appropriate padding.
136     This code is ENDIAN DEPENDENT but dependent only
137     on host endianness, not object file endianness.
138     The memcpy args are the issue.
139 */
140 #define SIGN_EXTEND(dest, length) \
141     do {if (*(Dwarf_Sbyte *)((char *)&dest + sizeof(dest) - length) < 0) {\
142 	memcpy((char *)&dest, "\xff\xff\xff\xff\xff\xff\xff\xff", \
143 	    sizeof(dest) - length);  \
144         } \
145      } while (0)
146 #else /* LITTLE ENDIAN */
147 
148 #define READ_UNALIGNED(dbg,dest,desttype, source, length) \
149     do  { \
150       BIGGEST_UINT _ltmp = 0;  \
151       dbg->de_copy_word( (char *)(&_ltmp) , \
152                         source, length) ; \
153       dest = (desttype)_ltmp;  \
154      } while (0)
155 
156 
157 /*
158     This macro sign-extends a variable depending on the length.
159     It fills the bytes between the size of the destination and
160     the length with appropriate padding.
161     This code is ENDIAN DEPENDENT but dependent only
162     on host endianness, not object file endianness.
163     The memcpy args are the issue.
164 */
165 #define SIGN_EXTEND(dest, length) \
166     do {if (*(Dwarf_Sbyte *)((char *)&dest + (length-1)) < 0) {\
167         memcpy((char *)&dest+length,    \
168                 "\xff\xff\xff\xff\xff\xff\xff\xff", \
169             sizeof(dest) - length); \
170         }  \
171     } while (0)
172 
173 #endif /* ! LITTLE_ENDIAN */
174 
175 
176 
177 /*
178    READ_AREA LENGTH reads the length (the older way
179    of pure 32 or 64 bit
180    or the new proposed dwarfv2.1 64bit-extension way)
181 
182    It reads the bits from where rw_src_data_p  points to
183    and updates the rw_src_data_p to point past what was just read.
184 
185    It updates w_length_size (to the size of an offset, either 4 or 8)
186    and w_exten_size (set 0 unless this frame has the DWARF3,4 64bit
187    extension, in which case w_exten_size is set to 4).
188 
189    r_dbg is just the current dbg pointer.
190    w_target is the output length field.
191    r_targtype is the output type. Always Dwarf_Unsigned so far.
192 
193 */
194 /* This one handles the v2.1 64bit extension
195    and 32bit (and   MIPS fixed 64  bit via the
196 	dwarf_init-set r_dbg->de_length_size)..
197    It does not recognize any but the one distingushed value
198    (the only one with defined meaning).
199    It assumes that no CU will have a length
200 	0xffffffxx  (32bit length)
201 	or
202 	0xffffffxx xxxxxxxx (64bit length)
203    which makes possible auto-detection of the extension.
204 
205    This depends on knowing that only a non-zero length
206    is legitimate (AFAICT), and for IRIX non-standard -64
207    dwarf that the first 32 bits of the 64bit offset will be
208    zero (because the compiler could not handle a truly large
209    value as of Jan 2003 and because no app has that much debug
210    info anyway, at least not in the IRIX case).
211 
212    At present not testing for '64bit elf' here as that
213    does not seem necessary (none of the 64bit length seems
214    appropriate unless it's  ident[EI_CLASS] == ELFCLASS64).
215 */
216 #   define    READ_AREA_LENGTH(r_dbg,w_target,r_targtype,         \
217 	rw_src_data_p,w_length_size,w_exten_size)                 \
218 do {    READ_UNALIGNED(r_dbg,w_target,r_targtype,                 \
219                 rw_src_data_p, ORIGINAL_DWARF_OFFSET_SIZE);       \
220     if(w_target == DISTINGUISHED_VALUE) {                         \
221 	     /* dwarf3 64bit extension */                         \
222              w_length_size  = DISTINGUISHED_VALUE_OFFSET_SIZE;    \
223              rw_src_data_p += ORIGINAL_DWARF_OFFSET_SIZE;         \
224              w_exten_size   = ORIGINAL_DWARF_OFFSET_SIZE;         \
225              READ_UNALIGNED(r_dbg,w_target,r_targtype,            \
226                   rw_src_data_p, DISTINGUISHED_VALUE_OFFSET_SIZE);\
227              rw_src_data_p += DISTINGUISHED_VALUE_OFFSET_SIZE;    \
228     } else {                                                      \
229 	if(w_target == 0 && r_dbg->de_big_endian_object) {        \
230 	     /* IRIX 64 bit, big endian.  This test */            \
231 	     /* is not a truly precise test, a precise test */    \
232              /* would check if the target was IRIX.  */           \
233              READ_UNALIGNED(r_dbg,w_target,r_targtype,            \
234                 rw_src_data_p, DISTINGUISHED_VALUE_OFFSET_SIZE);  \
235 	     w_length_size  = DISTINGUISHED_VALUE_OFFSET_SIZE;    \
236 	     rw_src_data_p += DISTINGUISHED_VALUE_OFFSET_SIZE;    \
237 	     w_exten_size = 0;                                    \
238 	} else {                                                  \
239 	     /* standard 32 bit dwarf2/dwarf3 */                  \
240 	     w_exten_size   = 0;                                  \
241              w_length_size  = ORIGINAL_DWARF_OFFSET_SIZE;         \
242              rw_src_data_p += w_length_size;                      \
243 	}                                                         \
244     } } while(0)
245 
246 Dwarf_Unsigned
247 _dwarf_decode_u_leb128(Dwarf_Small * leb128,
248 		       Dwarf_Word * leb128_length);
249 
250 Dwarf_Signed
251 _dwarf_decode_s_leb128(Dwarf_Small * leb128,
252 		       Dwarf_Word * leb128_length);
253 
254 Dwarf_Unsigned
255 _dwarf_get_size_of_val(Dwarf_Debug dbg,
256     Dwarf_Unsigned form,
257     Dwarf_Half address_size,
258     Dwarf_Small * val_ptr,
259     int v_length_size);
260 
261 struct Dwarf_Hash_Table_Entry_s;
262 /* This single struct is the base for the hash table.
263    The intent is that once the total_abbrev_count across
264    all the entries is greater than  10*current_table_entry_count
265    one should build a new Dwarf_Hash_Table_Base_s, rehash
266    all the existing entries, and delete the old table and entries.
267    (10 is a heuristic, nothing magic about it, but once the
268    count gets to 30 or 40 times current_table_entry_count
269    things really slow down a lot. One (500MB) application had
270    127000 abbreviations in one compilation unit)
271    The incoming 'code' is an abbrev number and those simply
272    increase linearly so the hashing is perfect always.
273 */
274 struct Dwarf_Hash_Table_s {
275       unsigned long       tb_table_entry_count;
276       unsigned long       tb_total_abbrev_count;
277       /* Each table entry is a list of abbreviations. */
278       struct  Dwarf_Hash_Table_Entry_s *tb_entries;
279 };
280 
281 /*
282     This struct is used to build a hash table for the
283     abbreviation codes for a compile-unit.
284 */
285 struct Dwarf_Hash_Table_Entry_s {
286     Dwarf_Abbrev_List at_head;
287 };
288 
289 
290 
291 Dwarf_Abbrev_List
292 _dwarf_get_abbrev_for_code(Dwarf_CU_Context cu_context,
293 			   Dwarf_Unsigned code);
294 
295 
296 /* return 1 if string ends before 'endptr' else
297 ** return 0 meaning string is not properly terminated.
298 ** Presumption is the 'endptr' pts to end of some dwarf section data.
299 */
300 int _dwarf_string_valid(void *startptr, void *endptr);
301 
302 Dwarf_Unsigned _dwarf_length_of_cu_header(Dwarf_Debug,
303 					  Dwarf_Unsigned offset);
304 Dwarf_Unsigned _dwarf_length_of_cu_header_simple(Dwarf_Debug);
305 
306 int  _dwarf_load_debug_info(Dwarf_Debug dbg, Dwarf_Error *error);
307 void _dwarf_free_abbrev_hash_table_contents(Dwarf_Debug dbg,
308     struct Dwarf_Hash_Table_s* hash_table);
309 int _dwarf_get_address_size(Dwarf_Debug dbg, Dwarf_Die die);
310 
311 #endif /* DWARF_UTIL_H */
312