1 /* $NetBSD: ppc_reloc.c,v 1.10 2001/09/10 06:09:41 mycroft Exp $ */
2
3 /*-
4 * Copyright (C) 1998 Tsubai Masanari
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. The name of the author may not be used to endorse or promote products
16 * derived from this software without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * $FreeBSD$
30 */
31
32 #include <sys/param.h>
33 #include <sys/mman.h>
34
35 #include <errno.h>
36 #include <stdio.h>
37 #include <stdlib.h>
38 #include <string.h>
39 #include <unistd.h>
40 #include <machine/cpu.h>
41 #include <machine/atomic.h>
42 #include <machine/md_var.h>
43
44 #include "debug.h"
45 #include "rtld.h"
46
47 #define _ppc_ha(x) ((((u_int32_t)(x) & 0x8000) ? \
48 ((u_int32_t)(x) + 0x10000) : (u_int32_t)(x)) >> 16)
49 #define _ppc_la(x) ((u_int32_t)(x) & 0xffff)
50
51 #define min(a,b) (((a) < (b)) ? (a) : (b))
52 #define max(a,b) (((a) > (b)) ? (a) : (b))
53
54 #define PLT_EXTENDED_BEGIN (1 << 13)
55 #define JMPTAB_BASE(N) (18 + N*2 + ((N > PLT_EXTENDED_BEGIN) ? \
56 (N - PLT_EXTENDED_BEGIN)*2 : 0))
57
58 /*
59 * Process the R_PPC_COPY relocations
60 */
61 int
do_copy_relocations(Obj_Entry * dstobj)62 do_copy_relocations(Obj_Entry *dstobj)
63 {
64 const Elf_Rela *relalim;
65 const Elf_Rela *rela;
66
67 /*
68 * COPY relocs are invalid outside of the main program
69 */
70 assert(dstobj->mainprog);
71
72 relalim = (const Elf_Rela *) ((caddr_t) dstobj->rela +
73 dstobj->relasize);
74 for (rela = dstobj->rela; rela < relalim; rela++) {
75 void *dstaddr;
76 const Elf_Sym *dstsym;
77 const char *name;
78 size_t size;
79 const void *srcaddr;
80 const Elf_Sym *srcsym = NULL;
81 const Obj_Entry *srcobj, *defobj;
82 SymLook req;
83 int res;
84
85 if (ELF_R_TYPE(rela->r_info) != R_PPC_COPY) {
86 continue;
87 }
88
89 dstaddr = (void *) (dstobj->relocbase + rela->r_offset);
90 dstsym = dstobj->symtab + ELF_R_SYM(rela->r_info);
91 name = dstobj->strtab + dstsym->st_name;
92 size = dstsym->st_size;
93 symlook_init(&req, name);
94 req.ventry = fetch_ventry(dstobj, ELF_R_SYM(rela->r_info));
95 req.flags = SYMLOOK_EARLY;
96
97 for (srcobj = dstobj->next; srcobj != NULL;
98 srcobj = srcobj->next) {
99 res = symlook_obj(&req, srcobj);
100 if (res == 0) {
101 srcsym = req.sym_out;
102 defobj = req.defobj_out;
103 break;
104 }
105 }
106
107 if (srcobj == NULL) {
108 _rtld_error("Undefined symbol \"%s\" "
109 " referenced from COPY"
110 " relocation in %s", name, dstobj->path);
111 return (-1);
112 }
113
114 srcaddr = (const void *) (defobj->relocbase+srcsym->st_value);
115 memcpy(dstaddr, srcaddr, size);
116 dbg("copy_reloc: src=%p,dst=%p,size=%d\n",srcaddr,dstaddr,size);
117 }
118
119 return (0);
120 }
121
122
123 /*
124 * Perform early relocation of the run-time linker image
125 */
126 void
reloc_non_plt_self(Elf_Dyn * dynp,Elf_Addr relocbase)127 reloc_non_plt_self(Elf_Dyn *dynp, Elf_Addr relocbase)
128 {
129 const Elf_Rela *rela = 0, *relalim;
130 Elf_Addr relasz = 0;
131 Elf_Addr *where;
132
133 /*
134 * Extract the rela/relasz values from the dynamic section
135 */
136 for (; dynp->d_tag != DT_NULL; dynp++) {
137 switch (dynp->d_tag) {
138 case DT_RELA:
139 rela = (const Elf_Rela *)(relocbase+dynp->d_un.d_ptr);
140 break;
141 case DT_RELASZ:
142 relasz = dynp->d_un.d_val;
143 break;
144 }
145 }
146
147 /*
148 * Relocate these values
149 */
150 relalim = (const Elf_Rela *)((caddr_t)rela + relasz);
151 for (; rela < relalim; rela++) {
152 where = (Elf_Addr *)(relocbase + rela->r_offset);
153 *where = (Elf_Addr)(relocbase + rela->r_addend);
154 }
155 }
156
157
158 /*
159 * Relocate a non-PLT object with addend.
160 */
161 static int
reloc_nonplt_object(Obj_Entry * obj_rtld,Obj_Entry * obj,const Elf_Rela * rela,SymCache * cache,int flags,RtldLockState * lockstate)162 reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
163 SymCache *cache, int flags, RtldLockState *lockstate)
164 {
165 Elf_Addr *where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
166 const Elf_Sym *def;
167 const Obj_Entry *defobj;
168 Elf_Addr tmp;
169
170 switch (ELF_R_TYPE(rela->r_info)) {
171
172 case R_PPC_NONE:
173 break;
174
175 case R_PPC_ADDR32: /* word32 S + A */
176 case R_PPC_GLOB_DAT: /* word32 S + A */
177 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
178 flags, cache, lockstate);
179 if (def == NULL) {
180 return (-1);
181 }
182
183 tmp = (Elf_Addr)(defobj->relocbase + def->st_value +
184 rela->r_addend);
185
186 /* Don't issue write if unnecessary; avoid COW page fault */
187 if (*where != tmp) {
188 *where = tmp;
189 }
190 break;
191
192 case R_PPC_RELATIVE: /* word32 B + A */
193 tmp = (Elf_Addr)(obj->relocbase + rela->r_addend);
194
195 /* As above, don't issue write unnecessarily */
196 if (*where != tmp) {
197 *where = tmp;
198 }
199 break;
200
201 case R_PPC_COPY:
202 /*
203 * These are deferred until all other relocations
204 * have been done. All we do here is make sure
205 * that the COPY relocation is not in a shared
206 * library. They are allowed only in executable
207 * files.
208 */
209 if (!obj->mainprog) {
210 _rtld_error("%s: Unexpected R_COPY "
211 " relocation in shared library",
212 obj->path);
213 return (-1);
214 }
215 break;
216
217 case R_PPC_JMP_SLOT:
218 /*
219 * These will be handled by the plt/jmpslot routines
220 */
221 break;
222
223 case R_PPC_DTPMOD32:
224 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
225 flags, cache, lockstate);
226
227 if (def == NULL)
228 return (-1);
229
230 *where = (Elf_Addr) defobj->tlsindex;
231
232 break;
233
234 case R_PPC_TPREL32:
235 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
236 flags, cache, lockstate);
237
238 if (def == NULL)
239 return (-1);
240
241 /*
242 * We lazily allocate offsets for static TLS as we
243 * see the first relocation that references the
244 * TLS block. This allows us to support (small
245 * amounts of) static TLS in dynamically loaded
246 * modules. If we run out of space, we generate an
247 * error.
248 */
249 if (!defobj->tls_done) {
250 if (!allocate_tls_offset((Obj_Entry*) defobj)) {
251 _rtld_error("%s: No space available for static "
252 "Thread Local Storage", obj->path);
253 return (-1);
254 }
255 }
256
257 *(Elf_Addr **)where = *where * sizeof(Elf_Addr)
258 + (Elf_Addr *)(def->st_value + rela->r_addend
259 + defobj->tlsoffset - TLS_TP_OFFSET);
260
261 break;
262
263 case R_PPC_DTPREL32:
264 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
265 flags, cache, lockstate);
266
267 if (def == NULL)
268 return (-1);
269
270 *where += (Elf_Addr)(def->st_value + rela->r_addend
271 - TLS_DTV_OFFSET);
272
273 break;
274
275 default:
276 _rtld_error("%s: Unsupported relocation type %d"
277 " in non-PLT relocations\n", obj->path,
278 ELF_R_TYPE(rela->r_info));
279 return (-1);
280 }
281 return (0);
282 }
283
284
285 /*
286 * Process non-PLT relocations
287 */
288 int
reloc_non_plt(Obj_Entry * obj,Obj_Entry * obj_rtld,int flags,RtldLockState * lockstate)289 reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, int flags,
290 RtldLockState *lockstate)
291 {
292 const Elf_Rela *relalim;
293 const Elf_Rela *rela;
294 SymCache *cache;
295 int r = -1;
296
297 if ((flags & SYMLOOK_IFUNC) != 0)
298 /* XXX not implemented */
299 return (0);
300
301 /*
302 * The dynamic loader may be called from a thread, we have
303 * limited amounts of stack available so we cannot use alloca().
304 */
305 if (obj != obj_rtld) {
306 cache = calloc(obj->dynsymcount, sizeof(SymCache));
307 /* No need to check for NULL here */
308 } else
309 cache = NULL;
310
311 /*
312 * From the SVR4 PPC ABI:
313 * "The PowerPC family uses only the Elf32_Rela relocation
314 * entries with explicit addends."
315 */
316 relalim = (const Elf_Rela *)((caddr_t)obj->rela + obj->relasize);
317 for (rela = obj->rela; rela < relalim; rela++) {
318 if (reloc_nonplt_object(obj_rtld, obj, rela, cache, flags,
319 lockstate) < 0)
320 goto done;
321 }
322 r = 0;
323 done:
324 if (cache != NULL)
325 free(cache);
326
327 /* Synchronize icache for text seg in case we made any changes */
328 __syncicache(obj->mapbase, obj->textsize);
329
330 return (r);
331 }
332
333 /*
334 * Initialise a PLT slot to the resolving trampoline
335 */
336 static int
reloc_plt_object(Obj_Entry * obj,const Elf_Rela * rela)337 reloc_plt_object(Obj_Entry *obj, const Elf_Rela *rela)
338 {
339 Elf_Word *where = (Elf_Word *)(obj->relocbase + rela->r_offset);
340 Elf_Addr *pltresolve, *pltlongresolve, *jmptab;
341 Elf_Addr distance;
342 int N = obj->pltrelasize / sizeof(Elf_Rela);
343 int reloff;
344
345 reloff = rela - obj->pltrela;
346
347 if (reloff < 0)
348 return (-1);
349
350 pltlongresolve = obj->pltgot + 5;
351 pltresolve = pltlongresolve + 5;
352
353 distance = (Elf_Addr)pltresolve - (Elf_Addr)(where + 1);
354
355 dbg(" reloc_plt_object: where=%p,pltres=%p,reloff=%x,distance=%x",
356 (void *)where, (void *)pltresolve, reloff, distance);
357
358 if (reloff < PLT_EXTENDED_BEGIN) {
359 /* li r11,reloff */
360 /* b pltresolve */
361 where[0] = 0x39600000 | reloff;
362 where[1] = 0x48000000 | (distance & 0x03fffffc);
363 } else {
364 jmptab = obj->pltgot + JMPTAB_BASE(N);
365 jmptab[reloff] = (u_int)pltlongresolve;
366
367 /* lis r11,jmptab[reloff]@ha */
368 /* lwzu r12,jmptab[reloff]@l(r11) */
369 /* mtctr r12 */
370 /* bctr */
371 where[0] = 0x3d600000 | _ppc_ha(&jmptab[reloff]);
372 where[1] = 0x858b0000 | _ppc_la(&jmptab[reloff]);
373 where[2] = 0x7d8903a6;
374 where[3] = 0x4e800420;
375 }
376
377
378 /*
379 * The icache will be sync'd in reloc_plt, which is called
380 * after all the slots have been updated
381 */
382
383 return (0);
384 }
385
386
387 /*
388 * Process the PLT relocations.
389 */
390 int
reloc_plt(Obj_Entry * obj)391 reloc_plt(Obj_Entry *obj)
392 {
393 const Elf_Rela *relalim;
394 const Elf_Rela *rela;
395 int N = obj->pltrelasize / sizeof(Elf_Rela);
396
397 if (obj->pltrelasize != 0) {
398
399 relalim = (const Elf_Rela *)((char *)obj->pltrela +
400 obj->pltrelasize);
401 for (rela = obj->pltrela; rela < relalim; rela++) {
402 assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT);
403
404 if (reloc_plt_object(obj, rela) < 0) {
405 return (-1);
406 }
407 }
408 }
409
410 /*
411 * Sync the icache for the byte range represented by the
412 * trampoline routines and call slots.
413 */
414 if (obj->pltgot != NULL)
415 __syncicache(obj->pltgot, JMPTAB_BASE(N)*4);
416
417 return (0);
418 }
419
420
421 /*
422 * LD_BIND_NOW was set - force relocation for all jump slots
423 */
424 int
reloc_jmpslots(Obj_Entry * obj,int flags,RtldLockState * lockstate)425 reloc_jmpslots(Obj_Entry *obj, int flags, RtldLockState *lockstate)
426 {
427 const Obj_Entry *defobj;
428 const Elf_Rela *relalim;
429 const Elf_Rela *rela;
430 const Elf_Sym *def;
431 Elf_Addr *where;
432 Elf_Addr target;
433
434 relalim = (const Elf_Rela *)((char *)obj->pltrela + obj->pltrelasize);
435 for (rela = obj->pltrela; rela < relalim; rela++) {
436 assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT);
437 where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
438 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
439 SYMLOOK_IN_PLT | flags, NULL, lockstate);
440 if (def == NULL) {
441 dbg("reloc_jmpslots: sym not found");
442 return (-1);
443 }
444
445 target = (Elf_Addr)(defobj->relocbase + def->st_value);
446
447 #if 0
448 /* PG XXX */
449 dbg("\"%s\" in \"%s\" --> %p in \"%s\"",
450 defobj->strtab + def->st_name, basename(obj->path),
451 (void *)target, basename(defobj->path));
452 #endif
453
454 reloc_jmpslot(where, target, defobj, obj,
455 (const Elf_Rel *) rela);
456 }
457
458 obj->jmpslots_done = true;
459
460 return (0);
461 }
462
463
464 /*
465 * Update the value of a PLT jump slot. Branch directly to the target if
466 * it is within +/- 32Mb, otherwise go indirectly via the pltcall
467 * trampoline call and jump table.
468 */
469 Elf_Addr
reloc_jmpslot(Elf_Addr * wherep,Elf_Addr target,const Obj_Entry * defobj,const Obj_Entry * obj,const Elf_Rel * rel)470 reloc_jmpslot(Elf_Addr *wherep, Elf_Addr target, const Obj_Entry *defobj,
471 const Obj_Entry *obj, const Elf_Rel *rel)
472 {
473 Elf_Addr offset;
474 const Elf_Rela *rela = (const Elf_Rela *) rel;
475
476 dbg(" reloc_jmpslot: where=%p, target=%p",
477 (void *)wherep, (void *)target);
478
479 /*
480 * At the PLT entry pointed at by `wherep', construct
481 * a direct transfer to the now fully resolved function
482 * address.
483 */
484 offset = target - (Elf_Addr)wherep;
485
486 if (abs((int)offset) < 32*1024*1024) { /* inside 32MB? */
487 /* b value # branch directly */
488 *wherep = 0x48000000 | (offset & 0x03fffffc);
489 __syncicache(wherep, 4);
490 } else {
491 Elf_Addr *pltcall, *jmptab;
492 int distance;
493 int N = obj->pltrelasize / sizeof(Elf_Rela);
494 int reloff = rela - obj->pltrela;
495
496 if (reloff < 0)
497 return (-1);
498
499 pltcall = obj->pltgot;
500
501 dbg(" reloc_jmpslot: indir, reloff=%x, N=%x\n",
502 reloff, N);
503
504 jmptab = obj->pltgot + JMPTAB_BASE(N);
505 jmptab[reloff] = target;
506 mb(); /* Order jmptab update before next changes */
507
508 if (reloff < PLT_EXTENDED_BEGIN) {
509 /* for extended PLT entries, we keep the old code */
510
511 distance = (Elf_Addr)pltcall - (Elf_Addr)(wherep + 1);
512
513 /* li r11,reloff */
514 /* b pltcall # use indirect pltcall routine */
515
516 /* first instruction same as before */
517 wherep[1] = 0x48000000 | (distance & 0x03fffffc);
518 __syncicache(wherep, 8);
519 }
520 }
521
522 return (target);
523 }
524
525 int
reloc_iresolve(Obj_Entry * obj,struct Struct_RtldLockState * lockstate)526 reloc_iresolve(Obj_Entry *obj, struct Struct_RtldLockState *lockstate)
527 {
528
529 /* XXX not implemented */
530 return (0);
531 }
532
533 int
reloc_gnu_ifunc(Obj_Entry * obj,int flags,struct Struct_RtldLockState * lockstate)534 reloc_gnu_ifunc(Obj_Entry *obj, int flags,
535 struct Struct_RtldLockState *lockstate)
536 {
537
538 /* XXX not implemented */
539 return (0);
540 }
541
542 /*
543 * Setup the plt glue routines.
544 */
545 #define PLTCALL_SIZE 20
546 #define PLTLONGRESOLVE_SIZE 20
547 #define PLTRESOLVE_SIZE 24
548
549 void
init_pltgot(Obj_Entry * obj)550 init_pltgot(Obj_Entry *obj)
551 {
552 Elf_Word *pltcall, *pltresolve, *pltlongresolve;
553 Elf_Word *jmptab;
554 int N = obj->pltrelasize / sizeof(Elf_Rela);
555
556 pltcall = obj->pltgot;
557
558 if (pltcall == NULL) {
559 return;
560 }
561
562 /*
563 * From the SVR4 PPC ABI:
564 *
565 * 'The first 18 words (72 bytes) of the PLT are reserved for
566 * use by the dynamic linker.
567 * ...
568 * 'If the executable or shared object requires N procedure
569 * linkage table entries, the link editor shall reserve 3*N
570 * words (12*N bytes) following the 18 reserved words. The
571 * first 2*N of these words are the procedure linkage table
572 * entries themselves. The static linker directs calls to bytes
573 * (72 + (i-1)*8), for i between 1 and N inclusive. The remaining
574 * N words (4*N bytes) are reserved for use by the dynamic linker.'
575 */
576
577 /*
578 * Copy the absolute-call assembler stub into the first part of
579 * the reserved PLT area.
580 */
581 memcpy(pltcall, _rtld_powerpc_pltcall, PLTCALL_SIZE);
582
583 /*
584 * Determine the address of the jumptable, which is the dyn-linker
585 * reserved area after the call cells. Write the absolute address
586 * of the jumptable into the absolute-call assembler code so it
587 * can determine this address.
588 */
589 jmptab = obj->pltgot + JMPTAB_BASE(N);
590 pltcall[1] |= _ppc_ha(jmptab); /* addis 11,11,jmptab@ha */
591 pltcall[2] |= _ppc_la(jmptab); /* lwz 11,jmptab@l(11) */
592
593 /*
594 * Skip down 20 bytes into the initial reserved area and copy
595 * in the standard resolving assembler call. Into this assembler,
596 * insert the absolute address of the _rtld_bind_start routine
597 * and the address of the relocation object.
598 *
599 * We place pltlongresolve first, so it can fix up its arguments
600 * and then fall through to the regular PLT resolver.
601 */
602 pltlongresolve = obj->pltgot + 5;
603
604 memcpy(pltlongresolve, _rtld_powerpc_pltlongresolve,
605 PLTLONGRESOLVE_SIZE);
606 pltlongresolve[0] |= _ppc_ha(jmptab); /* lis 12,jmptab@ha */
607 pltlongresolve[1] |= _ppc_la(jmptab); /* addi 12,12,jmptab@l */
608
609 pltresolve = pltlongresolve + PLTLONGRESOLVE_SIZE/sizeof(uint32_t);
610 memcpy(pltresolve, _rtld_powerpc_pltresolve, PLTRESOLVE_SIZE);
611 pltresolve[0] |= _ppc_ha(_rtld_bind_start);
612 pltresolve[1] |= _ppc_la(_rtld_bind_start);
613 pltresolve[3] |= _ppc_ha(obj);
614 pltresolve[4] |= _ppc_la(obj);
615
616 /*
617 * The icache will be sync'd in reloc_plt, which is called
618 * after all the slots have been updated
619 */
620 }
621
622 void
allocate_initial_tls(Obj_Entry * list)623 allocate_initial_tls(Obj_Entry *list)
624 {
625 Elf_Addr **tp;
626
627 /*
628 * Fix the size of the static TLS block by using the maximum
629 * offset allocated so far and adding a bit for dynamic modules to
630 * use.
631 */
632
633 tls_static_space = tls_last_offset + tls_last_size + RTLD_STATIC_TLS_EXTRA;
634
635 tp = (Elf_Addr **) ((char *) allocate_tls(list, NULL, TLS_TCB_SIZE, 8)
636 + TLS_TP_OFFSET + TLS_TCB_SIZE);
637
638 /*
639 * XXX gcc seems to ignore 'tp = _tp;'
640 */
641
642 __asm __volatile("mr 2,%0" :: "r"(tp));
643 }
644
645 void*
__tls_get_addr(tls_index * ti)646 __tls_get_addr(tls_index* ti)
647 {
648 register Elf_Addr **tp;
649 char *p;
650
651 __asm __volatile("mr %0,2" : "=r"(tp));
652 p = tls_get_addr_common((Elf_Addr**)((Elf_Addr)tp - TLS_TP_OFFSET
653 - TLS_TCB_SIZE), ti->ti_module, ti->ti_offset);
654
655 return (p + TLS_DTV_OFFSET);
656 }
657