1 /* Kernel module help for PPC64.
2 Copyright (C) 2001, 2003 Rusty Russell IBM Corporation.
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/module.h>
22 #include <linux/elf.h>
23 #include <linux/moduleloader.h>
24 #include <linux/err.h>
25 #include <linux/vmalloc.h>
26 #include <linux/ftrace.h>
27 #include <linux/bug.h>
28 #include <linux/uaccess.h>
29 #include <asm/module.h>
30 #include <asm/firmware.h>
31 #include <asm/code-patching.h>
32 #include <linux/sort.h>
33 #include <asm/setup.h>
35 /* FIXME: We don't do .init separately. To do this, we'd need to have
36 a separate r2 value in the init and core section, and stub between
39 Using a magic allocator which places modules within 32MB solves
40 this, and makes other things simpler. Anton?
43 #if defined(_CALL_ELF) && _CALL_ELF == 2
44 #define R2_STACK_OFFSET 24
46 /* An address is simply the address of the function. */
47 typedef unsigned long func_desc_t;
49 static func_desc_t func_desc(unsigned long addr)
53 static unsigned long func_addr(unsigned long addr)
57 static unsigned long stub_func_addr(func_desc_t func)
62 /* PowerPC64 specific values for the Elf64_Sym st_other field. */
63 #define STO_PPC64_LOCAL_BIT 5
64 #define STO_PPC64_LOCAL_MASK (7 << STO_PPC64_LOCAL_BIT)
65 #define PPC64_LOCAL_ENTRY_OFFSET(other) \
66 (((1 << (((other) & STO_PPC64_LOCAL_MASK) >> STO_PPC64_LOCAL_BIT)) >> 2) << 2)
68 static unsigned int local_entry_offset(const Elf64_Sym *sym)
70 /* sym->st_other indicates offset to local entry point
71 * (otherwise it will assume r12 is the address of the start
72 * of function and try to derive r2 from it). */
73 return PPC64_LOCAL_ENTRY_OFFSET(sym->st_other);
76 #define R2_STACK_OFFSET 40
78 /* An address is address of the OPD entry, which contains address of fn. */
79 typedef struct ppc64_opd_entry func_desc_t;
81 static func_desc_t func_desc(unsigned long addr)
83 return *(struct ppc64_opd_entry *)addr;
85 static unsigned long func_addr(unsigned long addr)
87 return func_desc(addr).funcaddr;
89 static unsigned long stub_func_addr(func_desc_t func)
93 static unsigned int local_entry_offset(const Elf64_Sym *sym)
99 #define STUB_MAGIC 0x73747562 /* stub */
101 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into
102 the kernel itself). But on PPC64, these need to be used for every
103 jump, actually, to reset r2 (TOC+0x8000). */
104 struct ppc64_stub_entry
106 /* 28 byte jump instruction sequence (7 instructions). We only
107 * need 6 instructions on ABIv2 but we always allocate 7 so
108 * so we don't have to modify the trampoline load instruction. */
110 /* Used by ftrace to identify stubs */
112 /* Data for the above code */
113 func_desc_t funcdata;
117 * PPC64 uses 24 bit jumps, but we need to jump into other modules or
118 * the kernel which may be further. So we jump to a stub.
120 * For ELFv1 we need to use this to set up the new r2 value (aka TOC
121 * pointer). For ELFv2 it's the callee's responsibility to set up the
122 * new r2, but for both we need to save the old r2.
124 * We could simply patch the new r2 value and function pointer into
125 * the stub, but it's significantly shorter to put these values at the
126 * end of the stub code, and patch the stub address (32-bits relative
127 * to the TOC ptr, r2) into the stub.
130 static u32 ppc64_stub_insns[] = {
131 0x3d620000, /* addis r11,r2, <high> */
132 0x396b0000, /* addi r11,r11, <low> */
133 /* Save current r2 value in magic place on the stack. */
134 0xf8410000|R2_STACK_OFFSET, /* std r2,R2_STACK_OFFSET(r1) */
135 0xe98b0020, /* ld r12,32(r11) */
136 #if !defined(_CALL_ELF) || _CALL_ELF != 2
137 /* Set up new r2 from function descriptor */
138 0xe84b0028, /* ld r2,40(r11) */
140 0x7d8903a6, /* mtctr r12 */
141 0x4e800420 /* bctr */
144 #ifdef CONFIG_DYNAMIC_FTRACE
145 int module_trampoline_target(struct module *mod, unsigned long addr,
146 unsigned long *target)
148 struct ppc64_stub_entry *stub;
149 func_desc_t funcdata;
152 if (!within_module_core(addr, mod)) {
153 pr_err("%s: stub %lx not in module %s\n", __func__, addr, mod->name);
157 stub = (struct ppc64_stub_entry *)addr;
159 if (probe_kernel_read(&magic, &stub->magic, sizeof(magic))) {
160 pr_err("%s: fault reading magic for stub %lx for %s\n", __func__, addr, mod->name);
164 if (magic != STUB_MAGIC) {
165 pr_err("%s: bad magic for stub %lx for %s\n", __func__, addr, mod->name);
169 if (probe_kernel_read(&funcdata, &stub->funcdata, sizeof(funcdata))) {
170 pr_err("%s: fault reading funcdata for stub %lx for %s\n", __func__, addr, mod->name);
174 *target = stub_func_addr(funcdata);
180 /* Count how many different 24-bit relocations (different symbol,
182 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num)
184 unsigned int i, r_info, r_addend, _count_relocs;
186 /* FIXME: Only count external ones --RR */
190 for (i = 0; i < num; i++)
191 /* Only count 24-bit relocs, others don't need stubs */
192 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
193 (r_info != ELF64_R_SYM(rela[i].r_info) ||
194 r_addend != rela[i].r_addend)) {
196 r_info = ELF64_R_SYM(rela[i].r_info);
197 r_addend = rela[i].r_addend;
200 return _count_relocs;
203 static int relacmp(const void *_x, const void *_y)
205 const Elf64_Rela *x, *y;
207 y = (Elf64_Rela *)_x;
208 x = (Elf64_Rela *)_y;
210 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to
211 * make the comparison cheaper/faster. It won't affect the sorting or
212 * the counting algorithms' performance
214 if (x->r_info < y->r_info)
216 else if (x->r_info > y->r_info)
218 else if (x->r_addend < y->r_addend)
220 else if (x->r_addend > y->r_addend)
226 static void relaswap(void *_x, void *_y, int size)
228 uint64_t *x, *y, tmp;
234 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) {
241 /* Get size of potential trampolines required. */
242 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr,
243 const Elf64_Shdr *sechdrs)
245 /* One extra reloc so it's always 0-funcaddr terminated */
246 unsigned long relocs = 1;
249 /* Every relocated section... */
250 for (i = 1; i < hdr->e_shnum; i++) {
251 if (sechdrs[i].sh_type == SHT_RELA) {
252 pr_debug("Found relocations in section %u\n", i);
253 pr_debug("Ptr: %p. Number: %Lu\n",
254 (void *)sechdrs[i].sh_addr,
255 sechdrs[i].sh_size / sizeof(Elf64_Rela));
257 /* Sort the relocation information based on a symbol and
258 * addend key. This is a stable O(n*log n) complexity
259 * alogrithm but it will reduce the complexity of
260 * count_relocs() to linear complexity O(n)
262 sort((void *)sechdrs[i].sh_addr,
263 sechdrs[i].sh_size / sizeof(Elf64_Rela),
264 sizeof(Elf64_Rela), relacmp, relaswap);
266 relocs += count_relocs((void *)sechdrs[i].sh_addr,
268 / sizeof(Elf64_Rela));
272 #ifdef CONFIG_DYNAMIC_FTRACE
273 /* make the trampoline to the ftrace_caller */
277 pr_debug("Looks like a total of %lu stubs, max\n", relocs);
278 return relocs * sizeof(struct ppc64_stub_entry);
281 /* Still needed for ELFv2, for .TOC. */
282 static void dedotify_versions(struct modversion_info *vers,
285 struct modversion_info *end;
287 for (end = (void *)vers + size; vers < end; vers++)
288 if (vers->name[0] == '.') {
289 memmove(vers->name, vers->name+1, strlen(vers->name));
290 #ifdef ARCH_RELOCATES_KCRCTAB
291 /* The TOC symbol has no CRC computed. To avoid CRC
292 * check failing, we must force it to the expected
293 * value (see CRC check in module.c).
295 if (!strcmp(vers->name, "TOC."))
296 vers->crc = -(unsigned long)reloc_start;
302 * Undefined symbols which refer to .funcname, hack to funcname. Make .TOC.
303 * seem to be defined (value set later).
305 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab)
309 for (i = 1; i < numsyms; i++) {
310 if (syms[i].st_shndx == SHN_UNDEF) {
311 char *name = strtab + syms[i].st_name;
312 if (name[0] == '.') {
313 if (strcmp(name+1, "TOC.") == 0)
314 syms[i].st_shndx = SHN_ABS;
315 memmove(name, name+1, strlen(name));
321 static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs,
323 unsigned int symindex)
325 unsigned int i, numsyms;
328 syms = (Elf64_Sym *)sechdrs[symindex].sh_addr;
329 numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym);
331 for (i = 1; i < numsyms; i++) {
332 if (syms[i].st_shndx == SHN_ABS
333 && strcmp(strtab + syms[i].st_name, "TOC.") == 0)
339 int module_frob_arch_sections(Elf64_Ehdr *hdr,
346 /* Find .toc and .stubs sections, symtab and strtab */
347 for (i = 1; i < hdr->e_shnum; i++) {
349 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0)
350 me->arch.stubs_section = i;
351 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0)
352 me->arch.toc_section = i;
353 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0)
354 dedotify_versions((void *)hdr + sechdrs[i].sh_offset,
357 /* We don't handle .init for the moment: rename to _init */
358 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init")))
361 if (sechdrs[i].sh_type == SHT_SYMTAB)
362 dedotify((void *)hdr + sechdrs[i].sh_offset,
363 sechdrs[i].sh_size / sizeof(Elf64_Sym),
365 + sechdrs[sechdrs[i].sh_link].sh_offset);
368 if (!me->arch.stubs_section) {
369 pr_err("%s: doesn't contain .stubs.\n", me->name);
373 /* If we don't have a .toc, just use .stubs. We need to set r2
374 to some reasonable value in case the module calls out to
375 other functions via a stub, or if a function pointer escapes
376 the module by some means. */
377 if (!me->arch.toc_section)
378 me->arch.toc_section = me->arch.stubs_section;
380 /* Override the stubs size */
381 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs);
385 /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this
386 gives the value maximum span in an instruction which uses a signed
388 static inline unsigned long my_r2(const Elf64_Shdr *sechdrs, struct module *me)
390 return sechdrs[me->arch.toc_section].sh_addr + 0x8000;
393 /* Both low and high 16 bits are added as SIGNED additions, so if low
394 16 bits has high bit set, high 16 bits must be adjusted. These
395 macros do that (stolen from binutils). */
396 #define PPC_LO(v) ((v) & 0xffff)
397 #define PPC_HI(v) (((v) >> 16) & 0xffff)
398 #define PPC_HA(v) PPC_HI ((v) + 0x8000)
400 /* Patch stub to reference function and correct r2 value. */
401 static inline int create_stub(const Elf64_Shdr *sechdrs,
402 struct ppc64_stub_entry *entry,
408 memcpy(entry->jump, ppc64_stub_insns, sizeof(ppc64_stub_insns));
410 /* Stub uses address relative to r2. */
411 reladdr = (unsigned long)entry - my_r2(sechdrs, me);
412 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
413 pr_err("%s: Address %p of stub out of range of %p.\n",
414 me->name, (void *)reladdr, (void *)my_r2);
417 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr);
419 entry->jump[0] |= PPC_HA(reladdr);
420 entry->jump[1] |= PPC_LO(reladdr);
421 entry->funcdata = func_desc(addr);
422 entry->magic = STUB_MAGIC;
427 /* Create stub to jump to function described in this OPD/ptr: we need the
428 stub to set up the TOC ptr (r2) for the function. */
429 static unsigned long stub_for_addr(const Elf64_Shdr *sechdrs,
433 struct ppc64_stub_entry *stubs;
434 unsigned int i, num_stubs;
436 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs);
438 /* Find this stub, or if that fails, the next avail. entry */
439 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr;
440 for (i = 0; stub_func_addr(stubs[i].funcdata); i++) {
441 BUG_ON(i >= num_stubs);
443 if (stub_func_addr(stubs[i].funcdata) == func_addr(addr))
444 return (unsigned long)&stubs[i];
447 if (!create_stub(sechdrs, &stubs[i], addr, me))
450 return (unsigned long)&stubs[i];
453 /* We expect a noop next: if it is, replace it with instruction to
455 static int restore_r2(u32 *instruction, struct module *me)
457 if (*instruction != PPC_INST_NOP) {
458 pr_err("%s: Expect noop after relocate, got %08x\n",
459 me->name, *instruction);
462 /* ld r2,R2_STACK_OFFSET(r1) */
463 *instruction = 0xe8410000 | R2_STACK_OFFSET;
467 int apply_relocate_add(Elf64_Shdr *sechdrs,
469 unsigned int symindex,
474 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr;
476 unsigned long *location;
479 pr_debug("Applying ADD relocate section %u to %u\n", relsec,
480 sechdrs[relsec].sh_info);
482 /* First time we're called, we can fix up .TOC. */
483 if (!me->arch.toc_fixed) {
484 sym = find_dot_toc(sechdrs, strtab, symindex);
485 /* It's theoretically possible that a module doesn't want a
486 * .TOC. so don't fail it just for that. */
488 sym->st_value = my_r2(sechdrs, me);
489 me->arch.toc_fixed = true;
492 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
493 /* This is where to make the change */
494 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
496 /* This is the symbol it is referring to */
497 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
498 + ELF64_R_SYM(rela[i].r_info);
500 pr_debug("RELOC at %p: %li-type as %s (0x%lx) + %li\n",
501 location, (long)ELF64_R_TYPE(rela[i].r_info),
502 strtab + sym->st_name, (unsigned long)sym->st_value,
503 (long)rela[i].r_addend);
505 /* `Everything is relative'. */
506 value = sym->st_value + rela[i].r_addend;
508 switch (ELF64_R_TYPE(rela[i].r_info)) {
511 *(u32 *)location = value;
516 *(unsigned long *)location = value;
520 *(unsigned long *)location = my_r2(sechdrs, me);
524 /* Subtract TOC pointer */
525 value -= my_r2(sechdrs, me);
526 if (value + 0x8000 > 0xffff) {
527 pr_err("%s: bad TOC16 relocation (0x%lx)\n",
531 *((uint16_t *) location)
532 = (*((uint16_t *) location) & ~0xffff)
536 case R_PPC64_TOC16_LO:
537 /* Subtract TOC pointer */
538 value -= my_r2(sechdrs, me);
539 *((uint16_t *) location)
540 = (*((uint16_t *) location) & ~0xffff)
544 case R_PPC64_TOC16_DS:
545 /* Subtract TOC pointer */
546 value -= my_r2(sechdrs, me);
547 if ((value & 3) != 0 || value + 0x8000 > 0xffff) {
548 pr_err("%s: bad TOC16_DS relocation (0x%lx)\n",
552 *((uint16_t *) location)
553 = (*((uint16_t *) location) & ~0xfffc)
557 case R_PPC64_TOC16_LO_DS:
558 /* Subtract TOC pointer */
559 value -= my_r2(sechdrs, me);
560 if ((value & 3) != 0) {
561 pr_err("%s: bad TOC16_LO_DS relocation (0x%lx)\n",
565 *((uint16_t *) location)
566 = (*((uint16_t *) location) & ~0xfffc)
570 case R_PPC64_TOC16_HA:
571 /* Subtract TOC pointer */
572 value -= my_r2(sechdrs, me);
573 value = ((value + 0x8000) >> 16);
574 *((uint16_t *) location)
575 = (*((uint16_t *) location) & ~0xffff)
580 /* FIXME: Handle weak symbols here --RR */
581 if (sym->st_shndx == SHN_UNDEF) {
582 /* External: go via stub */
583 value = stub_for_addr(sechdrs, value, me);
586 if (!restore_r2((u32 *)location + 1, me))
589 value += local_entry_offset(sym);
591 /* Convert value to relative */
592 value -= (unsigned long)location;
593 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){
594 pr_err("%s: REL24 %li out of range!\n",
595 me->name, (long int)value);
599 /* Only replace bits 2 through 26 */
600 *(uint32_t *)location
601 = (*(uint32_t *)location & ~0x03fffffc)
602 | (value & 0x03fffffc);
606 /* 64 bits relative (used by features fixups) */
607 *location = value - (unsigned long)location;
610 case R_PPC64_TOCSAVE:
612 * Marker reloc indicates we don't have to save r2.
613 * That would only save us one instruction, so ignore
620 * Optimize ELFv2 large code model entry point if
621 * the TOC is within 2GB range of current location.
623 value = my_r2(sechdrs, me) - (unsigned long)location;
624 if (value + 0x80008000 > 0xffffffff)
627 * Check for the large code model prolog sequence:
631 if ((((uint32_t *)location)[0] & ~0xfffc)
634 if (((uint32_t *)location)[1] != 0x7c426214)
637 * If found, replace it with:
638 * addis r2, r12, (.TOC.-func)@ha
639 * addi r2, r12, (.TOC.-func)@l
641 ((uint32_t *)location)[0] = 0x3c4c0000 + PPC_HA(value);
642 ((uint32_t *)location)[1] = 0x38420000 + PPC_LO(value);
645 case R_PPC64_REL16_HA:
646 /* Subtract location pointer */
647 value -= (unsigned long)location;
648 value = ((value + 0x8000) >> 16);
649 *((uint16_t *) location)
650 = (*((uint16_t *) location) & ~0xffff)
654 case R_PPC64_REL16_LO:
655 /* Subtract location pointer */
656 value -= (unsigned long)location;
657 *((uint16_t *) location)
658 = (*((uint16_t *) location) & ~0xffff)
663 pr_err("%s: Unknown ADD relocation: %lu\n",
665 (unsigned long)ELF64_R_TYPE(rela[i].r_info));
673 #ifdef CONFIG_DYNAMIC_FTRACE
674 int module_finalize_ftrace(struct module *mod, const Elf_Shdr *sechdrs)
676 mod->arch.toc = my_r2(sechdrs, mod);
677 mod->arch.tramp = stub_for_addr(sechdrs, (unsigned long)ftrace_caller, mod);
679 if (!mod->arch.tramp)