mdreloc.c revision 1.39 1 /* $NetBSD: mdreloc.c,v 1.39 2006/05/10 21:53:15 mrg Exp $ */
2
3 /*-
4 * Copyright (c) 2000 Eduardo Horvath.
5 * Copyright (c) 1999, 2002 The NetBSD Foundation, Inc.
6 * All rights reserved.
7 *
8 * This code is derived from software contributed to The NetBSD Foundation
9 * by Paul Kranenburg and by Charles M. Hannum.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the NetBSD
22 * Foundation, Inc. and its contributors.
23 * 4. Neither the name of The NetBSD Foundation nor the names of its
24 * contributors may be used to endorse or promote products derived
25 * from this software without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 * POSSIBILITY OF SUCH DAMAGE.
38 */
39
40 #include <sys/cdefs.h>
41 #ifndef lint
42 __RCSID("$NetBSD: mdreloc.c,v 1.39 2006/05/10 21:53:15 mrg Exp $");
43 #endif /* not lint */
44
45 #include <errno.h>
46 #include <stdio.h>
47 #include <stdlib.h>
48 #include <string.h>
49 #include <unistd.h>
50 #include <sys/stat.h>
51
52 #include "rtldenv.h"
53 #include "debug.h"
54 #include "rtld.h"
55
56 /*
57 * The following table holds for each relocation type:
58 * - the width in bits of the memory location the relocation
59 * applies to (not currently used)
60 * - the number of bits the relocation value must be shifted to the
61 * right (i.e. discard least significant bits) to fit into
62 * the appropriate field in the instruction word.
63 * - flags indicating whether
64 * * the relocation involves a symbol
65 * * the relocation is relative to the current position
66 * * the relocation is for a GOT entry
67 * * the relocation is relative to the load address
68 *
69 */
70 #define _RF_S 0x80000000 /* Resolve symbol */
71 #define _RF_A 0x40000000 /* Use addend */
72 #define _RF_P 0x20000000 /* Location relative */
73 #define _RF_G 0x10000000 /* GOT offset */
74 #define _RF_B 0x08000000 /* Load address relative */
75 #define _RF_U 0x04000000 /* Unaligned */
76 #define _RF_SZ(s) (((s) & 0xff) << 8) /* memory target size */
77 #define _RF_RS(s) ( (s) & 0xff) /* right shift */
78 static const int reloc_target_flags[] = {
79 0, /* NONE */
80 _RF_S|_RF_A| _RF_SZ(8) | _RF_RS(0), /* RELOC_8 */
81 _RF_S|_RF_A| _RF_SZ(16) | _RF_RS(0), /* RELOC_16 */
82 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* RELOC_32 */
83 _RF_S|_RF_A|_RF_P| _RF_SZ(8) | _RF_RS(0), /* DISP_8 */
84 _RF_S|_RF_A|_RF_P| _RF_SZ(16) | _RF_RS(0), /* DISP_16 */
85 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(0), /* DISP_32 */
86 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(2), /* WDISP_30 */
87 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(2), /* WDISP_22 */
88 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(10), /* HI22 */
89 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* 22 */
90 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* 13 */
91 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* LO10 */
92 _RF_G| _RF_SZ(32) | _RF_RS(0), /* GOT10 */
93 _RF_G| _RF_SZ(32) | _RF_RS(0), /* GOT13 */
94 _RF_G| _RF_SZ(32) | _RF_RS(10), /* GOT22 */
95 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(0), /* PC10 */
96 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(10), /* PC22 */
97 _RF_A|_RF_P| _RF_SZ(32) | _RF_RS(2), /* WPLT30 */
98 _RF_SZ(32) | _RF_RS(0), /* COPY */
99 _RF_S|_RF_A| _RF_SZ(64) | _RF_RS(0), /* GLOB_DAT */
100 _RF_SZ(32) | _RF_RS(0), /* JMP_SLOT */
101 _RF_A| _RF_B| _RF_SZ(64) | _RF_RS(0), /* RELATIVE */
102 _RF_S|_RF_A| _RF_U| _RF_SZ(32) | _RF_RS(0), /* UA_32 */
103
104 _RF_A| _RF_SZ(32) | _RF_RS(0), /* PLT32 */
105 _RF_A| _RF_SZ(32) | _RF_RS(10), /* HIPLT22 */
106 _RF_A| _RF_SZ(32) | _RF_RS(0), /* LOPLT10 */
107 _RF_A|_RF_P| _RF_SZ(32) | _RF_RS(0), /* PCPLT32 */
108 _RF_A|_RF_P| _RF_SZ(32) | _RF_RS(10), /* PCPLT22 */
109 _RF_A|_RF_P| _RF_SZ(32) | _RF_RS(0), /* PCPLT10 */
110 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* 10 */
111 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* 11 */
112 _RF_S|_RF_A| _RF_SZ(64) | _RF_RS(0), /* 64 */
113 _RF_S|_RF_A|/*extra*/ _RF_SZ(32) | _RF_RS(0), /* OLO10 */
114 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(42), /* HH22 */
115 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(32), /* HM10 */
116 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(10), /* LM22 */
117 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(42), /* PC_HH22 */
118 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(32), /* PC_HM10 */
119 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(10), /* PC_LM22 */
120 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(2), /* WDISP16 */
121 _RF_S|_RF_A|_RF_P| _RF_SZ(32) | _RF_RS(2), /* WDISP19 */
122 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* GLOB_JMP */
123 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* 7 */
124 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* 5 */
125 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* 6 */
126 _RF_S|_RF_A|_RF_P| _RF_SZ(64) | _RF_RS(0), /* DISP64 */
127 _RF_A| _RF_SZ(64) | _RF_RS(0), /* PLT64 */
128 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(10), /* HIX22 */
129 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* LOX10 */
130 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(22), /* H44 */
131 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(12), /* M44 */
132 _RF_S|_RF_A| _RF_SZ(32) | _RF_RS(0), /* L44 */
133 _RF_S|_RF_A| _RF_SZ(64) | _RF_RS(0), /* REGISTER */
134 _RF_S|_RF_A| _RF_U| _RF_SZ(64) | _RF_RS(0), /* UA64 */
135 _RF_S|_RF_A| _RF_U| _RF_SZ(16) | _RF_RS(0), /* UA16 */
136 };
137
138 #ifdef RTLD_DEBUG_RELOC
139 static const char *reloc_names[] = {
140 "NONE", "RELOC_8", "RELOC_16", "RELOC_32", "DISP_8",
141 "DISP_16", "DISP_32", "WDISP_30", "WDISP_22", "HI22",
142 "22", "13", "LO10", "GOT10", "GOT13",
143 "GOT22", "PC10", "PC22", "WPLT30", "COPY",
144 "GLOB_DAT", "JMP_SLOT", "RELATIVE", "UA_32", "PLT32",
145 "HIPLT22", "LOPLT10", "LOPLT10", "PCPLT22", "PCPLT32",
146 "10", "11", "64", "OLO10", "HH22",
147 "HM10", "LM22", "PC_HH22", "PC_HM10", "PC_LM22",
148 "WDISP16", "WDISP19", "GLOB_JMP", "7", "5", "6",
149 "DISP64", "PLT64", "HIX22", "LOX10", "H44", "M44",
150 "L44", "REGISTER", "UA64", "UA16"
151 };
152 #endif
153
154 #define RELOC_RESOLVE_SYMBOL(t) ((reloc_target_flags[t] & _RF_S) != 0)
155 #define RELOC_PC_RELATIVE(t) ((reloc_target_flags[t] & _RF_P) != 0)
156 #define RELOC_BASE_RELATIVE(t) ((reloc_target_flags[t] & _RF_B) != 0)
157 #define RELOC_UNALIGNED(t) ((reloc_target_flags[t] & _RF_U) != 0)
158 #define RELOC_USE_ADDEND(t) ((reloc_target_flags[t] & _RF_A) != 0)
159 #define RELOC_TARGET_SIZE(t) ((reloc_target_flags[t] >> 8) & 0xff)
160 #define RELOC_VALUE_RIGHTSHIFT(t) (reloc_target_flags[t] & 0xff)
161
162 static const long reloc_target_bitmask[] = {
163 #define _BM(x) (~(-(1ULL << (x))))
164 0, /* NONE */
165 _BM(8), _BM(16), _BM(32), /* RELOC_8, _16, _32 */
166 _BM(8), _BM(16), _BM(32), /* DISP8, DISP16, DISP32 */
167 _BM(30), _BM(22), /* WDISP30, WDISP22 */
168 _BM(22), _BM(22), /* HI22, _22 */
169 _BM(13), _BM(10), /* RELOC_13, _LO10 */
170 _BM(10), _BM(13), _BM(22), /* GOT10, GOT13, GOT22 */
171 _BM(10), _BM(22), /* _PC10, _PC22 */
172 _BM(30), 0, /* _WPLT30, _COPY */
173 _BM(32), _BM(32), _BM(32), /* _GLOB_DAT, JMP_SLOT, _RELATIVE */
174 _BM(32), _BM(32), /* _UA32, PLT32 */
175 _BM(22), _BM(10), /* _HIPLT22, LOPLT10 */
176 _BM(32), _BM(22), _BM(10), /* _PCPLT32, _PCPLT22, _PCPLT10 */
177 _BM(10), _BM(11), -1, /* _10, _11, _64 */
178 _BM(10), _BM(22), /* _OLO10, _HH22 */
179 _BM(10), _BM(22), /* _HM10, _LM22 */
180 _BM(22), _BM(10), _BM(22), /* _PC_HH22, _PC_HM10, _PC_LM22 */
181 _BM(16), _BM(19), /* _WDISP16, _WDISP19 */
182 -1, /* GLOB_JMP */
183 _BM(7), _BM(5), _BM(6) /* _7, _5, _6 */
184 -1, -1, /* DISP64, PLT64 */
185 _BM(22), _BM(13), /* HIX22, LOX10 */
186 _BM(22), _BM(10), _BM(13), /* H44, M44, L44 */
187 -1, -1, _BM(16), /* REGISTER, UA64, UA16 */
188 #undef _BM
189 };
190 #define RELOC_VALUE_BITMASK(t) (reloc_target_bitmask[t])
191
192 /*
193 * Instruction templates:
194 */
195 #define BAA 0x10400000 /* ba,a %xcc, 0 */
196 #define SETHI 0x03000000 /* sethi %hi(0), %g1 */
197 #define JMP 0x81c06000 /* jmpl %g1+%lo(0), %g0 */
198 #define NOP 0x01000000 /* sethi %hi(0), %g0 */
199 #define OR 0x82806000 /* or %g1, 0, %g1 */
200 #define XOR 0x82c06000 /* xor %g1, 0, %g1 */
201 #define MOV71 0x8283a000 /* or %o7, 0, %g1 */
202 #define MOV17 0x9c806000 /* or %g1, 0, %o7 */
203 #define CALL 0x40000000 /* call 0 */
204 #define SLLX 0x8b407000 /* sllx %g1, 0, %g1 */
205 #define SETHIG5 0x0b000000 /* sethi %hi(0), %g5 */
206 #define ORG5 0x82804005 /* or %g1, %g5, %g1 */
207
208
209 /* %hi(v)/%lo(v) with variable shift */
210 #define HIVAL(v, s) (((v) >> (s)) & 0x003fffff)
211 #define LOVAL(v, s) (((v) >> (s)) & 0x000003ff)
212
213 void _rtld_bind_start_0(long, long);
214 void _rtld_bind_start_1(long, long);
215 void _rtld_relocate_nonplt_self(Elf_Dyn *, Elf_Addr);
216 caddr_t _rtld_bind(const Obj_Entry *, Elf_Word);
217
218 /*
219 * Install rtld function call into this PLT slot.
220 */
221 #define SAVE 0x9de3bf50 /* i.e. `save %sp,-176,%sp' */
222 #define SETHI_l0 0x21000000
223 #define SETHI_l1 0x23000000
224 #define OR_l0_l0 0xa0142000
225 #define SLLX_l0_32_l0 0xa12c3020
226 #define OR_l0_l1_l0 0xa0140011
227 #define JMPL_l0_o0 0x91c42000
228 #define MOV_g1_o1 0x92100001
229
230 void _rtld_install_plt(Elf_Word *, Elf_Addr);
231 static inline int _rtld_relocate_plt_object(const Obj_Entry *,
232 const Elf_Rela *, Elf_Addr *);
233
234 void
235 _rtld_install_plt(Elf_Word *pltgot, Elf_Addr proc)
236 {
237 pltgot[0] = SAVE;
238 pltgot[1] = SETHI_l0 | HIVAL(proc, 42);
239 pltgot[2] = SETHI_l1 | HIVAL(proc, 10);
240 pltgot[3] = OR_l0_l0 | LOVAL(proc, 32);
241 pltgot[4] = SLLX_l0_32_l0;
242 pltgot[5] = OR_l0_l1_l0;
243 pltgot[6] = JMPL_l0_o0 | LOVAL(proc, 0);
244 pltgot[7] = MOV_g1_o1;
245 }
246
247 void
248 _rtld_setup_pltgot(const Obj_Entry *obj)
249 {
250 /*
251 * On sparc64 we got troubles.
252 *
253 * Instructions are 4 bytes long.
254 * Elf[64]_Addr is 8 bytes long, so are our pltglot[]
255 * array entries.
256 * Each PLT entry jumps to PLT0 to enter the dynamic
257 * linker.
258 * Loading an arbitrary 64-bit pointer takes 6
259 * instructions and 2 registers.
260 *
261 * Somehow we need to issue a save to get a new stack
262 * frame, load the address of the dynamic linker, and
263 * jump there, in 8 instructions or less.
264 *
265 * Oh, we need to fill out both PLT0 and PLT1.
266 */
267 {
268 Elf_Word *entry = (Elf_Word *)obj->pltgot;
269
270 /* Install in entries 0 and 1 */
271 _rtld_install_plt(&entry[0], (Elf_Addr) &_rtld_bind_start_0);
272 _rtld_install_plt(&entry[8], (Elf_Addr) &_rtld_bind_start_1);
273
274 /*
275 * Install the object reference in first slot
276 * of entry 2.
277 */
278 obj->pltgot[8] = (Elf_Addr) obj;
279 }
280 }
281
282 void
283 _rtld_relocate_nonplt_self(Elf_Dyn *dynp, Elf_Addr relocbase)
284 {
285 const Elf_Rela *rela = 0, *relalim;
286 Elf_Addr relasz = 0;
287 Elf_Addr *where;
288
289 for (; dynp->d_tag != DT_NULL; dynp++) {
290 switch (dynp->d_tag) {
291 case DT_RELA:
292 rela = (const Elf_Rela *)(relocbase + dynp->d_un.d_ptr);
293 break;
294 case DT_RELASZ:
295 relasz = dynp->d_un.d_val;
296 break;
297 }
298 }
299 relalim = (const Elf_Rela *)((caddr_t)rela + relasz);
300 for (; rela < relalim; rela++) {
301 where = (Elf_Addr *)(relocbase + rela->r_offset);
302 *where = (Elf_Addr)(relocbase + rela->r_addend);
303 }
304 }
305
306 int
307 _rtld_relocate_nonplt_objects(const Obj_Entry *obj)
308 {
309 const Elf_Rela *rela;
310
311 for (rela = obj->rela; rela < obj->relalim; rela++) {
312 Elf_Addr *where;
313 Elf_Word type;
314 Elf_Addr value = 0, mask;
315 const Elf_Sym *def = NULL;
316 const Obj_Entry *defobj = NULL;
317 unsigned long symnum;
318
319 where = (Elf_Addr *) (obj->relocbase + rela->r_offset);
320 symnum = ELF_R_SYM(rela->r_info);
321
322 type = ELF_R_TYPE(rela->r_info);
323 if (type == R_TYPE(NONE))
324 continue;
325
326 /* We do JMP_SLOTs in _rtld_bind() below */
327 if (type == R_TYPE(JMP_SLOT))
328 continue;
329
330 /* COPY relocs are also handled elsewhere */
331 if (type == R_TYPE(COPY))
332 continue;
333
334 /*
335 * We use the fact that relocation types are an `enum'
336 * Note: R_SPARC_UA16 is currently numerically largest.
337 */
338 if (type > R_TYPE(UA16))
339 return (-1);
340
341 value = rela->r_addend;
342
343 /*
344 * Handle relative relocs here, as an optimization.
345 */
346 if (type == R_TYPE(RELATIVE)) {
347 *where = (Elf_Addr)(obj->relocbase + value);
348 rdbg(("RELATIVE in %s --> %p", obj->path,
349 (void *)*where));
350 continue;
351 }
352
353 if (RELOC_RESOLVE_SYMBOL(type)) {
354
355 /* Find the symbol */
356 def = _rtld_find_symdef(symnum, obj, &defobj, false);
357 if (def == NULL)
358 return (-1);
359
360 /* Add in the symbol's absolute address */
361 value += (Elf_Addr)(defobj->relocbase + def->st_value);
362 }
363
364 if (RELOC_PC_RELATIVE(type)) {
365 value -= (Elf_Addr)where;
366 }
367
368 if (RELOC_BASE_RELATIVE(type)) {
369 /*
370 * Note that even though sparcs use `Elf_rela'
371 * exclusively we still need the implicit memory addend
372 * in relocations referring to GOT entries.
373 * Undoubtedly, someone f*cked this up in the distant
374 * past, and now we're stuck with it in the name of
375 * compatibility for all eternity..
376 *
377 * In any case, the implicit and explicit should be
378 * mutually exclusive. We provide a check for that
379 * here.
380 */
381 #ifdef DIAGNOSTIC
382 if (value != 0 && *where != 0) {
383 xprintf("BASE_REL(%s): where=%p, *where 0x%lx, "
384 "addend=0x%lx, base %p\n",
385 obj->path, where, *where,
386 rela->r_addend, obj->relocbase);
387 }
388 #endif
389 /* XXXX -- apparently we ignore the preexisting value */
390 value += (Elf_Addr)(obj->relocbase);
391 }
392
393 mask = RELOC_VALUE_BITMASK(type);
394 value >>= RELOC_VALUE_RIGHTSHIFT(type);
395 value &= mask;
396
397 if (RELOC_UNALIGNED(type)) {
398 /* Handle unaligned relocations. */
399 Elf_Addr tmp = 0;
400 char *ptr = (char *)where;
401 int i, size = RELOC_TARGET_SIZE(type)/8;
402
403 /* Read it in one byte at a time. */
404 for (i=0; i<size; i++)
405 tmp = (tmp << 8) | ptr[i];
406
407 tmp &= ~mask;
408 tmp |= value;
409
410 /* Write it back out. */
411 for (i=0; i<size; i++)
412 ptr[i] = ((tmp >> (8*i)) & 0xff);
413 #ifdef RTLD_DEBUG_RELOC
414 value = (Elf_Addr)tmp;
415 #endif
416
417 } else if (RELOC_TARGET_SIZE(type) > 32) {
418 *where &= ~mask;
419 *where |= value;
420 #ifdef RTLD_DEBUG_RELOC
421 value = (Elf_Addr)*where;
422 #endif
423 } else {
424 Elf32_Addr *where32 = (Elf32_Addr *)where;
425
426 *where32 &= ~mask;
427 *where32 |= value;
428 #ifdef RTLD_DEBUG_RELOC
429 value = (Elf_Addr)*where32;
430 #endif
431 }
432
433 #ifdef RTLD_DEBUG_RELOC
434 if (RELOC_RESOLVE_SYMBOL(type)) {
435 rdbg(("%s %s in %s --> %p in %s", reloc_names[type],
436 obj->strtab + obj->symtab[symnum].st_name,
437 obj->path, (void *)value, defobj->path));
438 } else {
439 rdbg(("%s in %s --> %p", reloc_names[type],
440 obj->path, (void *)value));
441 }
442 #endif
443 }
444 return (0);
445 }
446
447 int
448 _rtld_relocate_plt_lazy(const Obj_Entry *obj)
449 {
450 return (0);
451 }
452
453 caddr_t
454 _rtld_bind(const Obj_Entry *obj, Elf_Word reloff)
455 {
456 const Elf_Rela *rela = obj->pltrela + reloff;
457 Elf_Addr result;
458 int err;
459
460 result = 0; /* XXX gcc */
461
462 if (ELF_R_TYPE(obj->pltrela->r_info) == R_TYPE(JMP_SLOT)) {
463 /*
464 * XXXX
465 *
466 * The first four PLT entries are reserved. There is some
467 * disagreement whether they should have associated relocation
468 * entries. Both the SPARC 32-bit and 64-bit ELF
469 * specifications say that they should have relocation entries,
470 * but the 32-bit SPARC binutils do not generate them, and now
471 * the 64-bit SPARC binutils have stopped generating them too.
472 *
473 * So, to provide binary compatibility, we will check the first
474 * entry, if it is reserved it should not be of the type
475 * JMP_SLOT. If it is JMP_SLOT, then the 4 reserved entries
476 * were not generated and our index is 4 entries too far.
477 */
478 rela -= 4;
479 }
480
481 err = _rtld_relocate_plt_object(obj, rela, &result);
482 if (err)
483 _rtld_die();
484
485 return (caddr_t)result;
486 }
487
488 int
489 _rtld_relocate_plt_objects(const Obj_Entry *obj)
490 {
491 const Elf_Rela *rela;
492
493 rela = obj->pltrela;
494
495 /*
496 * Check for first four reserved entries - and skip them.
497 * See above for details.
498 */
499 if (ELF_R_TYPE(obj->pltrela->r_info) != R_TYPE(JMP_SLOT))
500 rela += 4;
501
502 for (; rela < obj->pltrelalim; rela++)
503 if (_rtld_relocate_plt_object(obj, rela, NULL) < 0)
504 return -1;
505
506 return 0;
507 }
508
509 /*
510 * New inline function that is called by _rtld_relocate_plt_object and
511 * _rtld_bind
512 */
513 static inline int
514 _rtld_relocate_plt_object(const Obj_Entry *obj, const Elf_Rela *rela, Elf_Addr *tp)
515 {
516 Elf_Word *where = (Elf_Word *)(obj->relocbase + rela->r_offset);
517 const Elf_Sym *def;
518 const Obj_Entry *defobj;
519 Elf_Addr value, offset;
520
521 /* Fully resolve procedure addresses now */
522
523 assert(ELF_R_TYPE(rela->r_info) == R_TYPE(JMP_SLOT));
524
525 def = _rtld_find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, true);
526 if (def == NULL)
527 return -1;
528
529 value = (Elf_Addr)(defobj->relocbase + def->st_value);
530 rdbg(("bind now/fixup in %s --> new=%p",
531 defobj->strtab + def->st_name, (void *)value));
532
533 /*
534 * At the PLT entry pointed at by `where', we now construct
535 * a direct transfer to the now fully resolved function
536 * address.
537 *
538 * A PLT entry is supposed to start by looking like this:
539 *
540 * sethi %hi(. - .PLT0), %g1
541 * ba,a %xcc, .PLT1
542 * nop
543 * nop
544 * nop
545 * nop
546 * nop
547 * nop
548 *
549 * When we replace these entries we start from the second
550 * entry and do it in reverse order so the last thing we
551 * do is replace the branch. That allows us to change this
552 * atomically.
553 *
554 * We now need to find out how far we need to jump. We
555 * have a choice of several different relocation techniques
556 * which are increasingly expensive.
557 */
558
559 offset = ((Elf_Addr)where) - value;
560 if (rela->r_addend) {
561 Elf_Addr *ptr = (Elf_Addr *)where;
562 /*
563 * This entry is >=32768. The relocations points to a
564 * PC-relative pointer to the bind_0 stub at the top of the
565 * PLT section. Update it to point to the target function.
566 */
567 ptr[0] += value - (Elf_Addr)obj->pltgot;
568
569 } else if (offset <= (1L<<20) && offset >= -(1L<<20)) {
570 /*
571 * We're within 1MB -- we can use a direct branch insn.
572 *
573 * We can generate this pattern:
574 *
575 * sethi %hi(. - .PLT0), %g1
576 * ba,a %xcc, addr
577 * nop
578 * nop
579 * nop
580 * nop
581 * nop
582 * nop
583 *
584 */
585 where[1] = BAA | ((offset >> 2) &0x3fffff);
586 __asm volatile("iflush %0+4" : : "r" (where));
587 } else if (value >= 0 && value < (1L<<32)) {
588 /*
589 * We're within 32-bits of address zero.
590 *
591 * The resulting code in the jump slot is:
592 *
593 * sethi %hi(. - .PLT0), %g1
594 * sethi %hi(addr), %g1
595 * jmp %g1+%lo(addr)
596 * nop
597 * nop
598 * nop
599 * nop
600 * nop
601 *
602 */
603 where[2] = JMP | LOVAL(value, 0);
604 where[1] = SETHI | HIVAL(value, 10);
605 __asm volatile("iflush %0+8" : : "r" (where));
606 __asm volatile("iflush %0+4" : : "r" (where));
607
608 } else if (value <= 0 && value > -(1L<<32)) {
609 /*
610 * We're within 32-bits of address -1.
611 *
612 * The resulting code in the jump slot is:
613 *
614 * sethi %hi(. - .PLT0), %g1
615 * sethi %hix(addr), %g1
616 * xor %g1, %lox(addr), %g1
617 * jmp %g1
618 * nop
619 * nop
620 * nop
621 * nop
622 *
623 */
624 where[3] = JMP;
625 where[2] = XOR | ((~value) & 0x00001fff);
626 where[1] = SETHI | HIVAL(~value, 10);
627 __asm volatile("iflush %0+12" : : "r" (where));
628 __asm volatile("iflush %0+8" : : "r" (where));
629 __asm volatile("iflush %0+4" : : "r" (where));
630
631 } else if (offset <= (1L<<32) && offset >= -((1L<<32) - 4)) {
632 /*
633 * We're within 32-bits -- we can use a direct call insn
634 *
635 * The resulting code in the jump slot is:
636 *
637 * sethi %hi(. - .PLT0), %g1
638 * mov %o7, %g1
639 * call (.+offset)
640 * mov %g1, %o7
641 * nop
642 * nop
643 * nop
644 * nop
645 *
646 */
647 where[3] = MOV17;
648 where[2] = CALL | ((offset >> 4) & 0x3fffffff);
649 where[1] = MOV71;
650 __asm volatile("iflush %0+12" : : "r" (where));
651 __asm volatile("iflush %0+8" : : "r" (where));
652 __asm volatile("iflush %0+4" : : "r" (where));
653
654 } else if (offset >= 0 && offset < (1L<<44)) {
655 /*
656 * We're within 44 bits. We can generate this pattern:
657 *
658 * The resulting code in the jump slot is:
659 *
660 * sethi %hi(. - .PLT0), %g1
661 * sethi %h44(addr), %g1
662 * or %g1, %m44(addr), %g1
663 * sllx %g1, 12, %g1
664 * jmp %g1+%l44(addr)
665 * nop
666 * nop
667 * nop
668 *
669 */
670 where[4] = JMP | LOVAL(offset, 0);
671 where[3] = SLLX | 12;
672 where[2] = OR | (((offset) >> 12) & 0x00001fff);
673 where[1] = SETHI | HIVAL(offset, 22);
674 __asm volatile("iflush %0+16" : : "r" (where));
675 __asm volatile("iflush %0+12" : : "r" (where));
676 __asm volatile("iflush %0+8" : : "r" (where));
677 __asm volatile("iflush %0+4" : : "r" (where));
678
679 } else if (offset < 0 && offset > -(1L<<44)) {
680 /*
681 * We're within 44 bits. We can generate this pattern:
682 *
683 * The resulting code in the jump slot is:
684 *
685 * sethi %hi(. - .PLT0), %g1
686 * sethi %h44(-addr), %g1
687 * xor %g1, %m44(-addr), %g1
688 * sllx %g1, 12, %g1
689 * jmp %g1+%l44(addr)
690 * nop
691 * nop
692 * nop
693 *
694 */
695 where[4] = JMP | LOVAL(offset, 0);
696 where[3] = SLLX | 12;
697 where[2] = XOR | (((~offset) >> 12) & 0x00001fff);
698 where[1] = SETHI | HIVAL(~offset, 22);
699 __asm volatile("iflush %0+16" : : "r" (where));
700 __asm volatile("iflush %0+12" : : "r" (where));
701 __asm volatile("iflush %0+8" : : "r" (where));
702 __asm volatile("iflush %0+4" : : "r" (where));
703
704 } else {
705 /*
706 * We need to load all 64-bits
707 *
708 * The resulting code in the jump slot is:
709 *
710 * sethi %hi(. - .PLT0), %g1
711 * sethi %hh(addr), %g1
712 * sethi %lm(addr), %g5
713 * or %g1, %hm(addr), %g1
714 * sllx %g1, 32, %g1
715 * or %g1, %g5, %g1
716 * jmp %g1+%lo(addr)
717 * nop
718 *
719 */
720 where[6] = JMP | LOVAL(value, 0);
721 where[5] = ORG5;
722 where[4] = SLLX | 32;
723 where[3] = OR | LOVAL(value, 32);
724 where[2] = SETHIG5 | HIVAL(value, 10);
725 where[1] = SETHI | HIVAL(value, 42);
726 __asm volatile("iflush %0+24" : : "r" (where));
727 __asm volatile("iflush %0+20" : : "r" (where));
728 __asm volatile("iflush %0+16" : : "r" (where));
729 __asm volatile("iflush %0+12" : : "r" (where));
730 __asm volatile("iflush %0+8" : : "r" (where));
731 __asm volatile("iflush %0+4" : : "r" (where));
732
733 }
734
735 if (tp)
736 *tp = value;
737
738 return 0;
739 }
740