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>
34 #include <asm/sections.h>
36 /* FIXME: We don't do .init separately. To do this, we'd need to have
37 a separate r2 value in the init and core section, and stub between
40 Using a magic allocator which places modules within 32MB solves
41 this, and makes other things simpler. Anton?
44 #ifdef PPC64_ELF_ABI_v2
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);
77 /* An address is address of the OPD entry, which contains address of fn. */
78 typedef struct ppc64_opd_entry func_desc_t;
80 static func_desc_t func_desc(unsigned long addr)
82 return *(struct ppc64_opd_entry *)addr;
84 static unsigned long func_addr(unsigned long addr)
86 return func_desc(addr).funcaddr;
88 static unsigned long stub_func_addr(func_desc_t func)
92 static unsigned int local_entry_offset(const Elf64_Sym *sym)
98 #define STUB_MAGIC 0x73747562 /* stub */
100 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into
101 the kernel itself). But on PPC64, these need to be used for every
102 jump, actually, to reset r2 (TOC+0x8000). */
103 struct ppc64_stub_entry
105 /* 28 byte jump instruction sequence (7 instructions). We only
106 * need 6 instructions on ABIv2 but we always allocate 7 so
107 * so we don't have to modify the trampoline load instruction. */
109 /* Used by ftrace to identify stubs */
111 /* Data for the above code */
112 func_desc_t funcdata;
116 * PPC64 uses 24 bit jumps, but we need to jump into other modules or
117 * the kernel which may be further. So we jump to a stub.
119 * For ELFv1 we need to use this to set up the new r2 value (aka TOC
120 * pointer). For ELFv2 it's the callee's responsibility to set up the
121 * new r2, but for both we need to save the old r2.
123 * We could simply patch the new r2 value and function pointer into
124 * the stub, but it's significantly shorter to put these values at the
125 * end of the stub code, and patch the stub address (32-bits relative
126 * to the TOC ptr, r2) into the stub.
129 static u32 ppc64_stub_insns[] = {
130 0x3d620000, /* addis r11,r2, <high> */
131 0x396b0000, /* addi r11,r11, <low> */
132 /* Save current r2 value in magic place on the stack. */
133 0xf8410000|R2_STACK_OFFSET, /* std r2,R2_STACK_OFFSET(r1) */
134 0xe98b0020, /* ld r12,32(r11) */
135 #ifdef PPC64_ELF_ABI_v1
136 /* Set up new r2 from function descriptor */
137 0xe84b0028, /* ld r2,40(r11) */
139 0x7d8903a6, /* mtctr r12 */
140 0x4e800420 /* bctr */
143 #ifdef CONFIG_DYNAMIC_FTRACE
144 int module_trampoline_target(struct module *mod, unsigned long addr,
145 unsigned long *target)
147 struct ppc64_stub_entry *stub;
148 func_desc_t funcdata;
151 if (!within_module_core(addr, mod)) {
152 pr_err("%s: stub %lx not in module %s\n", __func__, addr, mod->name);
156 stub = (struct ppc64_stub_entry *)addr;
158 if (probe_kernel_read(&magic, &stub->magic, sizeof(magic))) {
159 pr_err("%s: fault reading magic for stub %lx for %s\n", __func__, addr, mod->name);
163 if (magic != STUB_MAGIC) {
164 pr_err("%s: bad magic for stub %lx for %s\n", __func__, addr, mod->name);
168 if (probe_kernel_read(&funcdata, &stub->funcdata, sizeof(funcdata))) {
169 pr_err("%s: fault reading funcdata for stub %lx for %s\n", __func__, addr, mod->name);
173 *target = stub_func_addr(funcdata);
179 /* Count how many different 24-bit relocations (different symbol,
181 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num)
183 unsigned int i, r_info, r_addend, _count_relocs;
185 /* FIXME: Only count external ones --RR */
189 for (i = 0; i < num; i++)
190 /* Only count 24-bit relocs, others don't need stubs */
191 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
192 (r_info != ELF64_R_SYM(rela[i].r_info) ||
193 r_addend != rela[i].r_addend)) {
195 r_info = ELF64_R_SYM(rela[i].r_info);
196 r_addend = rela[i].r_addend;
199 return _count_relocs;
202 static int relacmp(const void *_x, const void *_y)
204 const Elf64_Rela *x, *y;
206 y = (Elf64_Rela *)_x;
207 x = (Elf64_Rela *)_y;
209 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to
210 * make the comparison cheaper/faster. It won't affect the sorting or
211 * the counting algorithms' performance
213 if (x->r_info < y->r_info)
215 else if (x->r_info > y->r_info)
217 else if (x->r_addend < y->r_addend)
219 else if (x->r_addend > y->r_addend)
225 static void relaswap(void *_x, void *_y, int size)
227 uint64_t *x, *y, tmp;
233 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) {
240 /* Get size of potential trampolines required. */
241 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr,
242 const Elf64_Shdr *sechdrs)
244 /* One extra reloc so it's always 0-funcaddr terminated */
245 unsigned long relocs = 1;
248 /* Every relocated section... */
249 for (i = 1; i < hdr->e_shnum; i++) {
250 if (sechdrs[i].sh_type == SHT_RELA) {
251 pr_debug("Found relocations in section %u\n", i);
252 pr_debug("Ptr: %p. Number: %Lu\n",
253 (void *)sechdrs[i].sh_addr,
254 sechdrs[i].sh_size / sizeof(Elf64_Rela));
256 /* Sort the relocation information based on a symbol and
257 * addend key. This is a stable O(n*log n) complexity
258 * alogrithm but it will reduce the complexity of
259 * count_relocs() to linear complexity O(n)
261 sort((void *)sechdrs[i].sh_addr,
262 sechdrs[i].sh_size / sizeof(Elf64_Rela),
263 sizeof(Elf64_Rela), relacmp, relaswap);
265 relocs += count_relocs((void *)sechdrs[i].sh_addr,
267 / sizeof(Elf64_Rela));
271 #ifdef CONFIG_DYNAMIC_FTRACE
272 /* make the trampoline to the ftrace_caller */
276 pr_debug("Looks like a total of %lu stubs, max\n", relocs);
277 return relocs * sizeof(struct ppc64_stub_entry);
280 /* Still needed for ELFv2, for .TOC. */
281 static void dedotify_versions(struct modversion_info *vers,
284 struct modversion_info *end;
286 for (end = (void *)vers + size; vers < end; vers++)
287 if (vers->name[0] == '.') {
288 memmove(vers->name, vers->name+1, strlen(vers->name));
293 * Undefined symbols which refer to .funcname, hack to funcname. Make .TOC.
294 * seem to be defined (value set later).
296 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab)
300 for (i = 1; i < numsyms; i++) {
301 if (syms[i].st_shndx == SHN_UNDEF) {
302 char *name = strtab + syms[i].st_name;
303 if (name[0] == '.') {
304 if (strcmp(name+1, "TOC.") == 0)
305 syms[i].st_shndx = SHN_ABS;
312 static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs,
314 unsigned int symindex)
316 unsigned int i, numsyms;
319 syms = (Elf64_Sym *)sechdrs[symindex].sh_addr;
320 numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym);
322 for (i = 1; i < numsyms; i++) {
323 if (syms[i].st_shndx == SHN_ABS
324 && strcmp(strtab + syms[i].st_name, "TOC.") == 0)
330 int module_frob_arch_sections(Elf64_Ehdr *hdr,
337 /* Find .toc and .stubs sections, symtab and strtab */
338 for (i = 1; i < hdr->e_shnum; i++) {
340 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0)
341 me->arch.stubs_section = i;
342 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0)
343 me->arch.toc_section = i;
344 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0)
345 dedotify_versions((void *)hdr + sechdrs[i].sh_offset,
348 /* We don't handle .init for the moment: rename to _init */
349 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init")))
352 if (sechdrs[i].sh_type == SHT_SYMTAB)
353 dedotify((void *)hdr + sechdrs[i].sh_offset,
354 sechdrs[i].sh_size / sizeof(Elf64_Sym),
356 + sechdrs[sechdrs[i].sh_link].sh_offset);
359 if (!me->arch.stubs_section) {
360 pr_err("%s: doesn't contain .stubs.\n", me->name);
364 /* If we don't have a .toc, just use .stubs. We need to set r2
365 to some reasonable value in case the module calls out to
366 other functions via a stub, or if a function pointer escapes
367 the module by some means. */
368 if (!me->arch.toc_section)
369 me->arch.toc_section = me->arch.stubs_section;
371 /* Override the stubs size */
372 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs);
376 /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this
377 gives the value maximum span in an instruction which uses a signed
379 static inline unsigned long my_r2(const Elf64_Shdr *sechdrs, struct module *me)
381 return sechdrs[me->arch.toc_section].sh_addr + 0x8000;
384 /* Both low and high 16 bits are added as SIGNED additions, so if low
385 16 bits has high bit set, high 16 bits must be adjusted. These
386 macros do that (stolen from binutils). */
387 #define PPC_LO(v) ((v) & 0xffff)
388 #define PPC_HI(v) (((v) >> 16) & 0xffff)
389 #define PPC_HA(v) PPC_HI ((v) + 0x8000)
391 /* Patch stub to reference function and correct r2 value. */
392 static inline int create_stub(const Elf64_Shdr *sechdrs,
393 struct ppc64_stub_entry *entry,
399 memcpy(entry->jump, ppc64_stub_insns, sizeof(ppc64_stub_insns));
401 /* Stub uses address relative to r2. */
402 reladdr = (unsigned long)entry - my_r2(sechdrs, me);
403 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
404 pr_err("%s: Address %p of stub out of range of %p.\n",
405 me->name, (void *)reladdr, (void *)my_r2);
408 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr);
410 entry->jump[0] |= PPC_HA(reladdr);
411 entry->jump[1] |= PPC_LO(reladdr);
412 entry->funcdata = func_desc(addr);
413 entry->magic = STUB_MAGIC;
418 /* Create stub to jump to function described in this OPD/ptr: we need the
419 stub to set up the TOC ptr (r2) for the function. */
420 static unsigned long stub_for_addr(const Elf64_Shdr *sechdrs,
424 struct ppc64_stub_entry *stubs;
425 unsigned int i, num_stubs;
427 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs);
429 /* Find this stub, or if that fails, the next avail. entry */
430 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr;
431 for (i = 0; stub_func_addr(stubs[i].funcdata); i++) {
432 BUG_ON(i >= num_stubs);
434 if (stub_func_addr(stubs[i].funcdata) == func_addr(addr))
435 return (unsigned long)&stubs[i];
438 if (!create_stub(sechdrs, &stubs[i], addr, me))
441 return (unsigned long)&stubs[i];
444 #ifdef CC_USING_MPROFILE_KERNEL
445 static bool is_early_mcount_callsite(u32 *instruction)
448 * Check if this is one of the -mprofile-kernel sequences.
450 if (instruction[-1] == PPC_INST_STD_LR &&
451 instruction[-2] == PPC_INST_MFLR)
454 if (instruction[-1] == PPC_INST_MFLR)
461 * In case of _mcount calls, do not save the current callee's TOC (in r2) into
462 * the original caller's stack frame. If we did we would clobber the saved TOC
463 * value of the original caller.
465 static void squash_toc_save_inst(const char *name, unsigned long addr)
467 struct ppc64_stub_entry *stub = (struct ppc64_stub_entry *)addr;
469 /* Only for calls to _mcount */
470 if (strcmp("_mcount", name) != 0)
473 stub->jump[2] = PPC_INST_NOP;
476 static void squash_toc_save_inst(const char *name, unsigned long addr) { }
478 /* without -mprofile-kernel, mcount calls are never early */
479 static bool is_early_mcount_callsite(u32 *instruction)
485 /* We expect a noop next: if it is, replace it with instruction to
487 static int restore_r2(u32 *instruction, struct module *me)
489 u32 *prev_insn = instruction - 1;
491 if (is_early_mcount_callsite(prev_insn))
495 * Make sure the branch isn't a sibling call. Sibling calls aren't
496 * "link" branches and they don't return, so they don't need the r2
497 * restore afterwards.
499 if (!instr_is_relative_link_branch(*prev_insn))
502 if (*instruction != PPC_INST_NOP) {
503 pr_err("%s: Expect noop after relocate, got %08x\n",
504 me->name, *instruction);
507 /* ld r2,R2_STACK_OFFSET(r1) */
508 *instruction = PPC_INST_LD_TOC;
512 int apply_relocate_add(Elf64_Shdr *sechdrs,
514 unsigned int symindex,
519 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr;
521 unsigned long *location;
524 pr_debug("Applying ADD relocate section %u to %u\n", relsec,
525 sechdrs[relsec].sh_info);
527 /* First time we're called, we can fix up .TOC. */
528 if (!me->arch.toc_fixed) {
529 sym = find_dot_toc(sechdrs, strtab, symindex);
530 /* It's theoretically possible that a module doesn't want a
531 * .TOC. so don't fail it just for that. */
533 sym->st_value = my_r2(sechdrs, me);
534 me->arch.toc_fixed = true;
537 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
538 /* This is where to make the change */
539 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
541 /* This is the symbol it is referring to */
542 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
543 + ELF64_R_SYM(rela[i].r_info);
545 pr_debug("RELOC at %p: %li-type as %s (0x%lx) + %li\n",
546 location, (long)ELF64_R_TYPE(rela[i].r_info),
547 strtab + sym->st_name, (unsigned long)sym->st_value,
548 (long)rela[i].r_addend);
550 /* `Everything is relative'. */
551 value = sym->st_value + rela[i].r_addend;
553 switch (ELF64_R_TYPE(rela[i].r_info)) {
556 *(u32 *)location = value;
561 *(unsigned long *)location = value;
565 *(unsigned long *)location = my_r2(sechdrs, me);
569 /* Subtract TOC pointer */
570 value -= my_r2(sechdrs, me);
571 if (value + 0x8000 > 0xffff) {
572 pr_err("%s: bad TOC16 relocation (0x%lx)\n",
576 *((uint16_t *) location)
577 = (*((uint16_t *) location) & ~0xffff)
581 case R_PPC64_TOC16_LO:
582 /* Subtract TOC pointer */
583 value -= my_r2(sechdrs, me);
584 *((uint16_t *) location)
585 = (*((uint16_t *) location) & ~0xffff)
589 case R_PPC64_TOC16_DS:
590 /* Subtract TOC pointer */
591 value -= my_r2(sechdrs, me);
592 if ((value & 3) != 0 || value + 0x8000 > 0xffff) {
593 pr_err("%s: bad TOC16_DS relocation (0x%lx)\n",
597 *((uint16_t *) location)
598 = (*((uint16_t *) location) & ~0xfffc)
602 case R_PPC64_TOC16_LO_DS:
603 /* Subtract TOC pointer */
604 value -= my_r2(sechdrs, me);
605 if ((value & 3) != 0) {
606 pr_err("%s: bad TOC16_LO_DS relocation (0x%lx)\n",
610 *((uint16_t *) location)
611 = (*((uint16_t *) location) & ~0xfffc)
615 case R_PPC64_TOC16_HA:
616 /* Subtract TOC pointer */
617 value -= my_r2(sechdrs, me);
618 value = ((value + 0x8000) >> 16);
619 *((uint16_t *) location)
620 = (*((uint16_t *) location) & ~0xffff)
625 /* FIXME: Handle weak symbols here --RR */
626 if (sym->st_shndx == SHN_UNDEF) {
627 /* External: go via stub */
628 value = stub_for_addr(sechdrs, value, me);
631 if (!restore_r2((u32 *)location + 1, me))
634 squash_toc_save_inst(strtab + sym->st_name, value);
636 value += local_entry_offset(sym);
638 /* Convert value to relative */
639 value -= (unsigned long)location;
640 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){
641 pr_err("%s: REL24 %li out of range!\n",
642 me->name, (long int)value);
646 /* Only replace bits 2 through 26 */
647 *(uint32_t *)location
648 = (*(uint32_t *)location & ~0x03fffffc)
649 | (value & 0x03fffffc);
653 /* 64 bits relative (used by features fixups) */
654 *location = value - (unsigned long)location;
658 /* 32 bits relative (used by relative exception tables) */
659 /* Convert value to relative */
660 value -= (unsigned long)location;
661 if (value + 0x80000000 > 0xffffffff) {
662 pr_err("%s: REL32 %li out of range!\n",
663 me->name, (long int)value);
666 *(u32 *)location = value;
669 case R_PPC64_TOCSAVE:
671 * Marker reloc indicates we don't have to save r2.
672 * That would only save us one instruction, so ignore
679 * Optimize ELFv2 large code model entry point if
680 * the TOC is within 2GB range of current location.
682 value = my_r2(sechdrs, me) - (unsigned long)location;
683 if (value + 0x80008000 > 0xffffffff)
686 * Check for the large code model prolog sequence:
690 if ((((uint32_t *)location)[0] & ~0xfffc)
693 if (((uint32_t *)location)[1] != 0x7c426214)
696 * If found, replace it with:
697 * addis r2, r12, (.TOC.-func)@ha
698 * addi r2, r2, (.TOC.-func)@l
700 ((uint32_t *)location)[0] = 0x3c4c0000 + PPC_HA(value);
701 ((uint32_t *)location)[1] = 0x38420000 + PPC_LO(value);
704 case R_PPC64_REL16_HA:
705 /* Subtract location pointer */
706 value -= (unsigned long)location;
707 value = ((value + 0x8000) >> 16);
708 *((uint16_t *) location)
709 = (*((uint16_t *) location) & ~0xffff)
713 case R_PPC64_REL16_LO:
714 /* Subtract location pointer */
715 value -= (unsigned long)location;
716 *((uint16_t *) location)
717 = (*((uint16_t *) location) & ~0xffff)
722 pr_err("%s: Unknown ADD relocation: %lu\n",
724 (unsigned long)ELF64_R_TYPE(rela[i].r_info));
732 #ifdef CONFIG_DYNAMIC_FTRACE
734 #ifdef CC_USING_MPROFILE_KERNEL
736 #define PACATOC offsetof(struct paca_struct, kernel_toc)
739 * For mprofile-kernel we use a special stub for ftrace_caller() because we
740 * can't rely on r2 containing this module's TOC when we enter the stub.
742 * That can happen if the function calling us didn't need to use the toc. In
743 * that case it won't have setup r2, and the r2 value will be either the
744 * kernel's toc, or possibly another modules toc.
746 * To deal with that this stub uses the kernel toc, which is always accessible
747 * via the paca (in r13). The target (ftrace_caller()) is responsible for
748 * saving and restoring the toc before returning.
750 static unsigned long create_ftrace_stub(const Elf64_Shdr *sechdrs, struct module *me)
752 struct ppc64_stub_entry *entry;
753 unsigned int i, num_stubs;
754 static u32 stub_insns[] = {
755 0xe98d0000 | PACATOC, /* ld r12,PACATOC(r13) */
756 0x3d8c0000, /* addis r12,r12,<high> */
757 0x398c0000, /* addi r12,r12,<low> */
758 0x7d8903a6, /* mtctr r12 */
759 0x4e800420, /* bctr */
763 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*entry);
765 /* Find the next available stub entry */
766 entry = (void *)sechdrs[me->arch.stubs_section].sh_addr;
767 for (i = 0; i < num_stubs && stub_func_addr(entry->funcdata); i++, entry++);
769 if (i >= num_stubs) {
770 pr_err("%s: Unable to find a free slot for ftrace stub.\n", me->name);
774 memcpy(entry->jump, stub_insns, sizeof(stub_insns));
776 /* Stub uses address relative to kernel toc (from the paca) */
777 reladdr = (unsigned long)ftrace_caller - kernel_toc_addr();
778 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
779 pr_err("%s: Address of ftrace_caller out of range of kernel_toc.\n", me->name);
783 entry->jump[1] |= PPC_HA(reladdr);
784 entry->jump[2] |= PPC_LO(reladdr);
786 /* Eventhough we don't use funcdata in the stub, it's needed elsewhere. */
787 entry->funcdata = func_desc((unsigned long)ftrace_caller);
788 entry->magic = STUB_MAGIC;
790 return (unsigned long)entry;
793 static unsigned long create_ftrace_stub(const Elf64_Shdr *sechdrs, struct module *me)
795 return stub_for_addr(sechdrs, (unsigned long)ftrace_caller, me);
799 int module_finalize_ftrace(struct module *mod, const Elf_Shdr *sechdrs)
801 mod->arch.toc = my_r2(sechdrs, mod);
802 mod->arch.tramp = create_ftrace_stub(sechdrs, mod);
804 if (!mod->arch.tramp)