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/cpufunc.h>
42 #include <machine/md_var.h>
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)
51 #define min(a,b) (((a) < (b)) ? (a) : (b))
52 #define max(a,b) (((a) > (b)) ? (a) : (b))
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))
59 * Process the R_PPC_COPY relocations
62 do_copy_relocations(Obj_Entry *dstobj)
64 const Elf_Rela *relalim;
68 * COPY relocs are invalid outside of the main program
70 assert(dstobj->mainprog);
72 relalim = (const Elf_Rela *) ((caddr_t) dstobj->rela +
74 for (rela = dstobj->rela; rela < relalim; rela++) {
76 const Elf_Sym *dstsym;
81 const Elf_Sym *srcsym = NULL;
85 if (ELF_R_TYPE(rela->r_info) != R_PPC_COPY) {
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 hash = elf_hash(name);
93 size = dstsym->st_size;
94 ve = fetch_ventry(dstobj, ELF_R_SYM(rela->r_info));
96 for (srcobj = dstobj->next; srcobj != NULL;
97 srcobj = srcobj->next) {
98 if ((srcsym = symlook_obj(name, hash, srcobj, ve, 0))
104 if (srcobj == NULL) {
105 _rtld_error("Undefined symbol \"%s\" "
106 " referenced from COPY"
107 " relocation in %s", name, dstobj->path);
111 srcaddr = (const void *) (srcobj->relocbase+srcsym->st_value);
112 memcpy(dstaddr, srcaddr, size);
113 dbg("copy_reloc: src=%p,dst=%p,size=%d\n",srcaddr,dstaddr,size);
121 * Perform early relocation of the run-time linker image
124 reloc_non_plt_self(Elf_Dyn *dynp, Elf_Addr relocbase)
126 const Elf_Rela *rela = 0, *relalim;
131 * Extract the rela/relasz values from the dynamic section
133 for (; dynp->d_tag != DT_NULL; dynp++) {
134 switch (dynp->d_tag) {
136 rela = (const Elf_Rela *)(relocbase+dynp->d_un.d_ptr);
139 relasz = dynp->d_un.d_val;
145 * Relocate these values
147 relalim = (const Elf_Rela *)((caddr_t)rela + relasz);
148 for (; rela < relalim; rela++) {
149 where = (Elf_Addr *)(relocbase + rela->r_offset);
150 *where = (Elf_Addr)(relocbase + rela->r_addend);
156 * Relocate a non-PLT object with addend.
159 reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela,
162 Elf_Addr *where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
164 const Obj_Entry *defobj;
167 switch (ELF_R_TYPE(rela->r_info)) {
172 case R_PPC_ADDR32: /* word32 S + A */
173 case R_PPC_GLOB_DAT: /* word32 S + A */
174 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
180 tmp = (Elf_Addr)(defobj->relocbase + def->st_value +
183 /* Don't issue write if unnecessary; avoid COW page fault */
189 case R_PPC_RELATIVE: /* word32 B + A */
190 tmp = (Elf_Addr)(obj->relocbase + rela->r_addend);
192 /* As above, don't issue write unnecessarily */
200 * These are deferred until all other relocations
201 * have been done. All we do here is make sure
202 * that the COPY relocation is not in a shared
203 * library. They are allowed only in executable
206 if (!obj->mainprog) {
207 _rtld_error("%s: Unexpected R_COPY "
208 " relocation in shared library",
216 * These will be handled by the plt/jmpslot routines
221 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
227 *where = (Elf_Addr) defobj->tlsindex;
232 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
239 * We lazily allocate offsets for static TLS as we
240 * see the first relocation that references the
241 * TLS block. This allows us to support (small
242 * amounts of) static TLS in dynamically loaded
243 * modules. If we run out of space, we generate an
246 if (!defobj->tls_done) {
247 if (!allocate_tls_offset((Obj_Entry*) defobj)) {
248 _rtld_error("%s: No space available for static "
249 "Thread Local Storage", obj->path);
254 *(Elf_Addr **)where = *where * sizeof(Elf_Addr)
255 + (Elf_Addr *)(def->st_value + rela->r_addend
256 + defobj->tlsoffset - TLS_TP_OFFSET);
261 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
267 *where += (Elf_Addr)(def->st_value + rela->r_addend
273 _rtld_error("%s: Unsupported relocation type %d"
274 " in non-PLT relocations\n", obj->path,
275 ELF_R_TYPE(rela->r_info));
283 * Process non-PLT relocations
286 reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld)
288 const Elf_Rela *relalim;
289 const Elf_Rela *rela;
291 int bytes = obj->nchains * sizeof(SymCache);
295 * The dynamic loader may be called from a thread, we have
296 * limited amounts of stack available so we cannot use alloca().
298 if (obj != obj_rtld) {
299 cache = mmap(NULL, bytes, PROT_READ|PROT_WRITE, MAP_ANON,
301 if (cache == MAP_FAILED)
307 * From the SVR4 PPC ABI:
308 * "The PowerPC family uses only the Elf32_Rela relocation
309 * entries with explicit addends."
311 relalim = (const Elf_Rela *)((caddr_t)obj->rela + obj->relasize);
312 for (rela = obj->rela; rela < relalim; rela++) {
313 if (reloc_nonplt_object(obj_rtld, obj, rela, cache) < 0)
319 munmap(cache, bytes);
325 * Initialise a PLT slot to the resolving trampoline
328 reloc_plt_object(Obj_Entry *obj, const Elf_Rela *rela)
330 Elf_Word *where = (Elf_Word *)(obj->relocbase + rela->r_offset);
331 Elf_Addr *pltresolve, *pltlongresolve, *jmptab;
333 int N = obj->pltrelasize / sizeof(Elf_Rela);
336 reloff = rela - obj->pltrela;
341 pltlongresolve = obj->pltgot + 5;
342 pltresolve = pltlongresolve + 5;
344 distance = (Elf_Addr)pltresolve - (Elf_Addr)(where + 1);
346 dbg(" reloc_plt_object: where=%p,pltres=%p,reloff=%x,distance=%x",
347 (void *)where, (void *)pltresolve, reloff, distance);
349 if (reloff < PLT_EXTENDED_BEGIN) {
352 where[0] = 0x39600000 | reloff;
353 where[1] = 0x48000000 | (distance & 0x03fffffc);
355 jmptab = obj->pltgot + JMPTAB_BASE(N);
356 jmptab[reloff] = (u_int)pltlongresolve;
358 /* lis r11,jmptab[reloff]@ha */
359 /* lwzu r12,jmptab[reloff]@l(r11) */
362 where[0] = 0x3d600000 | _ppc_ha(&jmptab[reloff]);
363 where[1] = 0x858b0000 | _ppc_la(&jmptab[reloff]);
364 where[2] = 0x7d8903a6;
365 where[3] = 0x4e800420;
370 * The icache will be sync'd in init_pltgot, which is called
371 * after all the slots have been updated
379 * Process the PLT relocations.
382 reloc_plt(Obj_Entry *obj)
384 const Elf_Rela *relalim;
385 const Elf_Rela *rela;
387 if (obj->pltrelasize != 0) {
389 relalim = (const Elf_Rela *)((char *)obj->pltrela +
391 for (rela = obj->pltrela; rela < relalim; rela++) {
392 assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT);
394 if (reloc_plt_object(obj, rela) < 0) {
405 * LD_BIND_NOW was set - force relocation for all jump slots
408 reloc_jmpslots(Obj_Entry *obj)
410 const Obj_Entry *defobj;
411 const Elf_Rela *relalim;
412 const Elf_Rela *rela;
417 relalim = (const Elf_Rela *)((char *)obj->pltrela + obj->pltrelasize);
418 for (rela = obj->pltrela; rela < relalim; rela++) {
419 assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT);
420 where = (Elf_Addr *)(obj->relocbase + rela->r_offset);
421 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj,
424 dbg("reloc_jmpslots: sym not found");
428 target = (Elf_Addr)(defobj->relocbase + def->st_value);
432 dbg("\"%s\" in \"%s\" --> %p in \"%s\"",
433 defobj->strtab + def->st_name, basename(obj->path),
434 (void *)target, basename(defobj->path));
437 reloc_jmpslot(where, target, defobj, obj,
438 (const Elf_Rel *) rela);
441 obj->jmpslots_done = true;
448 * Update the value of a PLT jump slot. Branch directly to the target if
449 * it is within +/- 32Mb, otherwise go indirectly via the pltcall
450 * trampoline call and jump table.
453 reloc_jmpslot(Elf_Addr *wherep, Elf_Addr target, const Obj_Entry *defobj,
454 const Obj_Entry *obj, const Elf_Rel *rel)
457 const Elf_Rela *rela = (const Elf_Rela *) rel;
459 dbg(" reloc_jmpslot: where=%p, target=%p",
460 (void *)wherep, (void *)target);
463 * At the PLT entry pointed at by `wherep', construct
464 * a direct transfer to the now fully resolved function
467 offset = target - (Elf_Addr)wherep;
469 if (abs(offset) < 32*1024*1024) { /* inside 32MB? */
470 /* b value # branch directly */
471 *wherep = 0x48000000 | (offset & 0x03fffffc);
472 __syncicache(wherep, 4);
474 Elf_Addr *pltcall, *jmptab;
476 int N = obj->pltrelasize / sizeof(Elf_Rela);
477 int reloff = rela - obj->pltrela;
482 pltcall = obj->pltgot;
484 dbg(" reloc_jmpslot: indir, reloff=%x, N=%x\n",
487 jmptab = obj->pltgot + JMPTAB_BASE(N);
488 jmptab[reloff] = target;
489 powerpc_mb(); /* Order jmptab update before next changes */
491 if (reloff < PLT_EXTENDED_BEGIN) {
492 /* for extended PLT entries, we keep the old code */
494 distance = (Elf_Addr)pltcall - (Elf_Addr)(wherep + 1);
497 /* b pltcall # use indirect pltcall routine */
499 /* first instruction same as before */
500 wherep[1] = 0x48000000 | (distance & 0x03fffffc);
501 __syncicache(wherep, 8);
510 * Setup the plt glue routines.
512 #define PLTCALL_SIZE 20
513 #define PLTLONGRESOLVE_SIZE 20
514 #define PLTRESOLVE_SIZE 24
517 init_pltgot(Obj_Entry *obj)
519 Elf_Word *pltcall, *pltresolve, *pltlongresolve;
521 int N = obj->pltrelasize / sizeof(Elf_Rela);
523 pltcall = obj->pltgot;
525 if (pltcall == NULL) {
530 * From the SVR4 PPC ABI:
532 * 'The first 18 words (72 bytes) of the PLT are reserved for
533 * use by the dynamic linker.
535 * 'If the executable or shared object requires N procedure
536 * linkage table entries, the link editor shall reserve 3*N
537 * words (12*N bytes) following the 18 reserved words. The
538 * first 2*N of these words are the procedure linkage table
539 * entries themselves. The static linker directs calls to bytes
540 * (72 + (i-1)*8), for i between 1 and N inclusive. The remaining
541 * N words (4*N bytes) are reserved for use by the dynamic linker.'
545 * Copy the absolute-call assembler stub into the first part of
546 * the reserved PLT area.
548 memcpy(pltcall, _rtld_powerpc_pltcall, PLTCALL_SIZE);
551 * Determine the address of the jumptable, which is the dyn-linker
552 * reserved area after the call cells. Write the absolute address
553 * of the jumptable into the absolute-call assembler code so it
554 * can determine this address.
556 jmptab = obj->pltgot + JMPTAB_BASE(N);
557 pltcall[1] |= _ppc_ha(jmptab); /* addis 11,11,jmptab@ha */
558 pltcall[2] |= _ppc_la(jmptab); /* lwz 11,jmptab@l(11) */
561 * Skip down 20 bytes into the initial reserved area and copy
562 * in the standard resolving assembler call. Into this assembler,
563 * insert the absolute address of the _rtld_bind_start routine
564 * and the address of the relocation object.
566 * We place pltlongresolve first, so it can fix up its arguments
567 * and then fall through to the regular PLT resolver.
569 pltlongresolve = obj->pltgot + 5;
571 memcpy(pltlongresolve, _rtld_powerpc_pltlongresolve,
572 PLTLONGRESOLVE_SIZE);
573 pltlongresolve[0] |= _ppc_ha(jmptab); /* lis 12,jmptab@ha */
574 pltlongresolve[1] |= _ppc_la(jmptab); /* addi 12,12,jmptab@l */
576 pltresolve = pltlongresolve + PLTLONGRESOLVE_SIZE/sizeof(uint32_t);
577 memcpy(pltresolve, _rtld_powerpc_pltresolve, PLTRESOLVE_SIZE);
578 pltresolve[0] |= _ppc_ha(_rtld_bind_start);
579 pltresolve[1] |= _ppc_la(_rtld_bind_start);
580 pltresolve[3] |= _ppc_ha(obj);
581 pltresolve[4] |= _ppc_la(obj);
584 * Sync the icache for the byte range represented by the
585 * trampoline routines and call slots.
587 __syncicache(obj->pltgot, JMPTAB_BASE(N)*4);
591 allocate_initial_tls(Obj_Entry *list)
593 register Elf_Addr **tp __asm__("r2");
597 * Fix the size of the static TLS block by using the maximum
598 * offset allocated so far and adding a bit for dynamic modules to
602 tls_static_space = tls_last_offset + tls_last_size + RTLD_STATIC_TLS_EXTRA;
604 _tp = (Elf_Addr **) ((char *) allocate_tls(list, NULL, TLS_TCB_SIZE, 8)
605 + TLS_TP_OFFSET + TLS_TCB_SIZE);
608 * XXX gcc seems to ignore 'tp = _tp;'
611 __asm __volatile("mr %0,%1" : "=r"(tp) : "r"(_tp));
615 __tls_get_addr(tls_index* ti)
617 register Elf_Addr **tp __asm__("r2");
620 p = tls_get_addr_common((Elf_Addr**)((Elf_Addr)tp - TLS_TP_OFFSET
621 - TLS_TCB_SIZE), ti->ti_module, ti->ti_offset);
623 return (p + TLS_DTV_OFFSET);