1 /* $NetBSD: ppc_reloc.c,v 1.10 2001/09/10 06:09:41 mycroft Exp $ */
4 * Copyright (C) 1998 Tsubai Masanari
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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.
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.
32 #include <sys/param.h>
40 #include <machine/cpu.h>
41 #include <machine/md_var.h>
46 #define _ppc_ha(x) ((((u_int32_t)(x) & 0x8000) ? \
47 ((u_int32_t)(x) + 0x10000) : (u_int32_t)(x)) >> 16)
48 #define _ppc_la(x) ((u_int32_t)(x) & 0xffff)
50 #define min(a,b) (((a) < (b)) ? (a) : (b))
51 #define max(a,b) (((a) > (b)) ? (a) : (b))
53 #define PLT_EXTENDED_BEGIN (1 << 13)
54 #define JMPTAB_BASE(N) (18 + N*2 + ((N > PLT_EXTENDED_BEGIN) ? \
55 (N - PLT_EXTENDED_BEGIN)*2 : 0))
58 * Process the R_PPC_COPY relocations
61 do_copy_relocations(Obj_Entry *dstobj)
63 const Elf_Rela *relalim;
67 * COPY relocs are invalid outside of the main program
69 assert(dstobj->mainprog);
71 relalim = (const Elf_Rela *) ((caddr_t) dstobj->rela +
73 for (rela = dstobj->rela; rela < relalim; rela++) {
75 const Elf_Sym *dstsym;
80 const Elf_Sym *srcsym = NULL;
84 if (ELF_R_TYPE(rela->r_info) != R_PPC_COPY) {
88 dstaddr = (void *) (dstobj->relocbase + rela->r_offset);
89 dstsym = dstobj->symtab + ELF_R_SYM(rela->r_info);
90 name = dstobj->strtab + dstsym->st_name;
91 hash = elf_hash(name);
92 size = dstsym->st_size;
93 ve = fetch_ventry(dstobj, ELF_R_SYM(rela->r_info));
95 for (srcobj = dstobj->next; srcobj != NULL;
96 srcobj = srcobj->next) {
97 if ((srcsym = symlook_obj(name, hash, srcobj, ve, 0))
103 if (srcobj == NULL) {
104 _rtld_error("Undefined symbol \"%s\" "
105 " referenced from COPY"
106 " relocation in %s", name, dstobj->path);
110 srcaddr = (const void *) (srcobj->relocbase+srcsym->st_value);
111 memcpy(dstaddr, srcaddr, size);
112 dbg("copy_reloc: src=%p,dst=%p,size=%d\n",srcaddr,dstaddr,size);
120 * Perform early relocation of the run-time linker image
123 reloc_non_plt_self(Elf_Dyn *dynp, Elf_Addr relocbase)
125 const Elf_Rela *rela = 0, *relalim;
130 * Extract the rela/relasz values from the dynamic section
132 for (; dynp->d_tag != DT_NULL; dynp++) {
133 switch (dynp->d_tag) {
135 rela = (const Elf_Rela *)(relocbase+dynp->d_un.d_ptr);
138 relasz = dynp->d_un.d_val;
144 * Relocate these values
146 relalim = (const Elf_Rela *)((caddr_t)rela + relasz);
147 for (; rela < relalim; rela++) {
148 where = (Elf_Addr *)(relocbase + rela->r_offset);
149 *where = (Elf_Addr)(relocbase + rela->r_addend);
155 * Relocate a non-PLT object with addend.
158 reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
161 Elf_Addr *where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
163 const Obj_Entry *defobj;
166 switch (ELF_R_TYPE(rela->r_info)) {
171 case R_PPC_ADDR32: /* word32 S + A */
172 case R_PPC_GLOB_DAT: /* word32 S + A */
173 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
179 tmp = (Elf_Addr)(defobj->relocbase + def->st_value +
182 /* Don't issue write if unnecessary; avoid COW page fault */
188 case R_PPC_RELATIVE: /* word32 B + A */
189 tmp = (Elf_Addr)(obj->relocbase + rela->r_addend);
191 /* As above, don't issue write unnecessarily */
199 * These are deferred until all other relocations
200 * have been done. All we do here is make sure
201 * that the COPY relocation is not in a shared
202 * library. They are allowed only in executable
205 if (!obj->mainprog) {
206 _rtld_error("%s: Unexpected R_COPY "
207 " relocation in shared library",
215 * These will be handled by the plt/jmpslot routines
220 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
226 *where = (Elf_Addr) defobj->tlsindex;
231 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
238 * We lazily allocate offsets for static TLS as we
239 * see the first relocation that references the
240 * TLS block. This allows us to support (small
241 * amounts of) static TLS in dynamically loaded
242 * modules. If we run out of space, we generate an
245 if (!defobj->tls_done) {
246 if (!allocate_tls_offset((Obj_Entry*) defobj)) {
247 _rtld_error("%s: No space available for static "
248 "Thread Local Storage", obj->path);
253 *(Elf_Addr **)where = *where * sizeof(Elf_Addr)
254 + (Elf_Addr *)(def->st_value + rela->r_addend
255 + defobj->tlsoffset - TLS_TP_OFFSET);
260 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
266 *where += (Elf_Addr)(def->st_value + rela->r_addend
272 _rtld_error("%s: Unsupported relocation type %d"
273 " in non-PLT relocations\n", obj->path,
274 ELF_R_TYPE(rela->r_info));
282 * Process non-PLT relocations
285 reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld)
287 const Elf_Rela *relalim;
288 const Elf_Rela *rela;
290 int bytes = obj->nchains * sizeof(SymCache);
294 * The dynamic loader may be called from a thread, we have
295 * limited amounts of stack available so we cannot use alloca().
297 if (obj != obj_rtld) {
298 cache = mmap(NULL, bytes, PROT_READ|PROT_WRITE, MAP_ANON,
300 if (cache == MAP_FAILED)
306 * From the SVR4 PPC ABI:
307 * "The PowerPC family uses only the Elf32_Rela relocation
308 * entries with explicit addends."
310 relalim = (const Elf_Rela *)((caddr_t)obj->rela + obj->relasize);
311 for (rela = obj->rela; rela < relalim; rela++) {
312 if (reloc_nonplt_object(obj_rtld, obj, rela, cache) < 0)
318 munmap(cache, bytes);
324 * Initialise a PLT slot to the resolving trampoline
327 reloc_plt_object(Obj_Entry *obj, const Elf_Rela *rela)
329 Elf_Word *where = (Elf_Word *)(obj->relocbase + rela->r_offset);
330 Elf_Addr *pltresolve, *pltlongresolve, *jmptab;
332 int N = obj->pltrelasize / sizeof(Elf_Rela);
335 reloff = rela - obj->pltrela;
340 pltlongresolve = obj->pltgot + 5;
341 pltresolve = pltlongresolve + 5;
343 distance = (Elf_Addr)pltresolve - (Elf_Addr)(where + 1);
345 dbg(" reloc_plt_object: where=%p,pltres=%p,reloff=%x,distance=%x",
346 (void *)where, (void *)pltresolve, reloff, distance);
348 if (reloff < PLT_EXTENDED_BEGIN) {
351 where[0] = 0x39600000 | reloff;
352 where[1] = 0x48000000 | (distance & 0x03fffffc);
354 jmptab = obj->pltgot + JMPTAB_BASE(N);
355 jmptab[reloff] = (u_int)pltlongresolve;
357 /* lis r11,jmptab[reloff]@ha */
358 /* lwzu r12,jmptab[reloff]@l(r11) */
361 where[0] = 0x3d600000 | _ppc_ha(&jmptab[reloff]);
362 where[1] = 0x858b0000 | _ppc_la(&jmptab[reloff]);
363 where[2] = 0x7d8903a6;
364 where[3] = 0x4e800420;
369 * The icache will be sync'd in init_pltgot, which is called
370 * after all the slots have been updated
378 * Process the PLT relocations.
381 reloc_plt(Obj_Entry *obj)
383 const Elf_Rela *relalim;
384 const Elf_Rela *rela;
386 if (obj->pltrelasize != 0) {
388 relalim = (const Elf_Rela *)((char *)obj->pltrela +
390 for (rela = obj->pltrela; rela < relalim; rela++) {
391 assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT);
393 if (reloc_plt_object(obj, rela) < 0) {
404 * LD_BIND_NOW was set - force relocation for all jump slots
407 reloc_jmpslots(Obj_Entry *obj)
409 const Obj_Entry *defobj;
410 const Elf_Rela *relalim;
411 const Elf_Rela *rela;
416 relalim = (const Elf_Rela *)((char *)obj->pltrela + obj->pltrelasize);
417 for (rela = obj->pltrela; rela < relalim; rela++) {
418 assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT);
419 where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
420 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
423 dbg("reloc_jmpslots: sym not found");
427 target = (Elf_Addr)(defobj->relocbase + def->st_value);
431 dbg("\"%s\" in \"%s\" --> %p in \"%s\"",
432 defobj->strtab + def->st_name, basename(obj->path),
433 (void *)target, basename(defobj->path));
436 reloc_jmpslot(where, target, defobj, obj,
437 (const Elf_Rel *) rela);
440 obj->jmpslots_done = true;
447 * Update the value of a PLT jump slot. Branch directly to the target if
448 * it is within +/- 32Mb, otherwise go indirectly via the pltcall
449 * trampoline call and jump table.
452 reloc_jmpslot(Elf_Addr *wherep, Elf_Addr target, const Obj_Entry *defobj,
453 const Obj_Entry *obj, const Elf_Rel *rel)
456 const Elf_Rela *rela = (const Elf_Rela *) rel;
458 dbg(" reloc_jmpslot: where=%p, target=%p",
459 (void *)wherep, (void *)target);
462 * At the PLT entry pointed at by `wherep', construct
463 * a direct transfer to the now fully resolved function
466 offset = target - (Elf_Addr)wherep;
468 if (abs(offset) < 32*1024*1024) { /* inside 32MB? */
469 /* b value # branch directly */
470 *wherep = 0x48000000 | (offset & 0x03fffffc);
471 __syncicache(wherep, 4);
473 Elf_Addr *pltcall, *jmptab;
475 int N = obj->pltrelasize / sizeof(Elf_Rela);
476 int reloff = rela - obj->pltrela;
481 pltcall = obj->pltgot;
483 dbg(" reloc_jmpslot: indir, reloff=%x, N=%x\n",
486 jmptab = obj->pltgot + JMPTAB_BASE(N);
487 jmptab[reloff] = target;
489 if (reloff < PLT_EXTENDED_BEGIN) {
490 /* for extended PLT entries, we keep the old code */
492 distance = (Elf_Addr)pltcall - (Elf_Addr)(wherep + 1);
495 /* b pltcall # use indirect pltcall routine */
496 wherep[0] = 0x39600000 | reloff;
497 wherep[1] = 0x48000000 | (distance & 0x03fffffc);
498 __syncicache(wherep, 8);
507 * Setup the plt glue routines.
509 #define PLTCALL_SIZE 20
510 #define PLTLONGRESOLVE_SIZE 20
511 #define PLTRESOLVE_SIZE 24
514 init_pltgot(Obj_Entry *obj)
516 Elf_Word *pltcall, *pltresolve, *pltlongresolve;
518 int N = obj->pltrelasize / sizeof(Elf_Rela);
520 pltcall = obj->pltgot;
522 if (pltcall == NULL) {
527 * From the SVR4 PPC ABI:
529 * 'The first 18 words (72 bytes) of the PLT are reserved for
530 * use by the dynamic linker.
532 * 'If the executable or shared object requires N procedure
533 * linkage table entries, the link editor shall reserve 3*N
534 * words (12*N bytes) following the 18 reserved words. The
535 * first 2*N of these words are the procedure linkage table
536 * entries themselves. The static linker directs calls to bytes
537 * (72 + (i-1)*8), for i between 1 and N inclusive. The remaining
538 * N words (4*N bytes) are reserved for use by the dynamic linker.'
542 * Copy the absolute-call assembler stub into the first part of
543 * the reserved PLT area.
545 memcpy(pltcall, _rtld_powerpc_pltcall, PLTCALL_SIZE);
548 * Determine the address of the jumptable, which is the dyn-linker
549 * reserved area after the call cells. Write the absolute address
550 * of the jumptable into the absolute-call assembler code so it
551 * can determine this address.
553 jmptab = obj->pltgot + JMPTAB_BASE(N);
554 pltcall[1] |= _ppc_ha(jmptab); /* addis 11,11,jmptab@ha */
555 pltcall[2] |= _ppc_la(jmptab); /* lwz 11,jmptab@l(11) */
558 * Skip down 20 bytes into the initial reserved area and copy
559 * in the standard resolving assembler call. Into this assembler,
560 * insert the absolute address of the _rtld_bind_start routine
561 * and the address of the relocation object.
563 * We place pltlongresolve first, so it can fix up its arguments
564 * and then fall through to the regular PLT resolver.
566 pltlongresolve = obj->pltgot + 5;
568 memcpy(pltlongresolve, _rtld_powerpc_pltlongresolve,
569 PLTLONGRESOLVE_SIZE);
570 pltlongresolve[0] |= _ppc_ha(jmptab); /* lis 12,jmptab@ha */
571 pltlongresolve[1] |= _ppc_la(jmptab); /* addi 12,12,jmptab@l */
573 pltresolve = pltlongresolve + PLTLONGRESOLVE_SIZE/sizeof(uint32_t);
574 memcpy(pltresolve, _rtld_powerpc_pltresolve, PLTRESOLVE_SIZE);
575 pltresolve[0] |= _ppc_ha(_rtld_bind_start);
576 pltresolve[1] |= _ppc_la(_rtld_bind_start);
577 pltresolve[3] |= _ppc_ha(obj);
578 pltresolve[4] |= _ppc_la(obj);
581 * Sync the icache for the byte range represented by the
582 * trampoline routines and call slots.
584 __syncicache(pltcall, 72 + N * 8);
588 allocate_initial_tls(Obj_Entry *list)
590 register Elf_Addr **tp __asm__("r2");
594 * Fix the size of the static TLS block by using the maximum
595 * offset allocated so far and adding a bit for dynamic modules to
599 tls_static_space = tls_last_offset + tls_last_size + RTLD_STATIC_TLS_EXTRA;
601 _tp = (Elf_Addr **) ((char *) allocate_tls(list, NULL, TLS_TCB_SIZE, 8)
602 + TLS_TP_OFFSET + TLS_TCB_SIZE);
605 * XXX gcc seems to ignore 'tp = _tp;'
608 __asm __volatile("mr %0,%1" : "=r"(tp) : "r"(_tp));
612 __tls_get_addr(tls_index* ti)
614 register Elf_Addr **tp __asm__("r2");
617 p = tls_get_addr_common((Elf_Addr**)((Elf_Addr)tp - TLS_TP_OFFSET
618 - TLS_TCB_SIZE), ti->ti_module, ti->ti_offset);
620 return (p + TLS_DTV_OFFSET);