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
18 #include <linux/module.h>
19 #include <linux/elf.h>
20 #include <linux/moduleloader.h>
21 #include <linux/err.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ftrace.h>
24 #include <linux/bug.h>
25 #include <asm/module.h>
26 #include <asm/firmware.h>
27 #include <asm/code-patching.h>
28 #include <linux/sort.h>
29 #include <asm/setup.h>
31 /* FIXME: We don't do .init separately. To do this, we'd need to have
32 a separate r2 value in the init and core section, and stub between
35 Using a magic allocator which places modules within 32MB solves
36 this, and makes other things simpler. Anton?
41 #define DEBUGP(fmt , ...)
44 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into
45 the kernel itself). But on PPC64, these need to be used for every
46 jump, actually, to reset r2 (TOC+0x8000). */
47 struct ppc64_stub_entry
49 /* 28 byte jump instruction sequence (7 instructions) */
52 /* Data for the above code */
53 struct ppc64_opd_entry opd;
56 /* We use a stub to fix up r2 (TOC ptr) and to jump to the (external)
57 function which may be more than 24-bits away. We could simply
58 patch the new r2 value and function pointer into the stub, but it's
59 significantly shorter to put these values at the end of the stub
60 code, and patch the stub address (32-bits relative to the TOC ptr,
62 static struct ppc64_stub_entry ppc64_stub =
64 0x3d820000, /* addis r12,r2, <high> */
65 0x398c0000, /* addi r12,r12, <low> */
66 /* Save current r2 value in magic place on the stack. */
67 0xf8410028, /* std r2,40(r1) */
68 0xe96c0020, /* ld r11,32(r12) */
69 0xe84c0028, /* ld r2,40(r12) */
70 0x7d6903a6, /* mtctr r11 */
74 /* Count how many different 24-bit relocations (different symbol,
76 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num)
78 unsigned int i, r_info, r_addend, _count_relocs;
80 /* FIXME: Only count external ones --RR */
84 for (i = 0; i < num; i++)
85 /* Only count 24-bit relocs, others don't need stubs */
86 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
87 (r_info != ELF64_R_SYM(rela[i].r_info) ||
88 r_addend != rela[i].r_addend)) {
90 r_info = ELF64_R_SYM(rela[i].r_info);
91 r_addend = rela[i].r_addend;
97 static int relacmp(const void *_x, const void *_y)
99 const Elf64_Rela *x, *y;
101 y = (Elf64_Rela *)_x;
102 x = (Elf64_Rela *)_y;
104 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to
105 * make the comparison cheaper/faster. It won't affect the sorting or
106 * the counting algorithms' performance
108 if (x->r_info < y->r_info)
110 else if (x->r_info > y->r_info)
112 else if (x->r_addend < y->r_addend)
114 else if (x->r_addend > y->r_addend)
120 static void relaswap(void *_x, void *_y, int size)
122 uint64_t *x, *y, tmp;
128 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) {
135 /* Get size of potential trampolines required. */
136 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr,
137 const Elf64_Shdr *sechdrs)
139 /* One extra reloc so it's always 0-funcaddr terminated */
140 unsigned long relocs = 1;
143 /* Every relocated section... */
144 for (i = 1; i < hdr->e_shnum; i++) {
145 if (sechdrs[i].sh_type == SHT_RELA) {
146 DEBUGP("Found relocations in section %u\n", i);
147 DEBUGP("Ptr: %p. Number: %lu\n",
148 (void *)sechdrs[i].sh_addr,
149 sechdrs[i].sh_size / sizeof(Elf64_Rela));
151 /* Sort the relocation information based on a symbol and
152 * addend key. This is a stable O(n*log n) complexity
153 * alogrithm but it will reduce the complexity of
154 * count_relocs() to linear complexity O(n)
156 sort((void *)sechdrs[i].sh_addr,
157 sechdrs[i].sh_size / sizeof(Elf64_Rela),
158 sizeof(Elf64_Rela), relacmp, relaswap);
160 relocs += count_relocs((void *)sechdrs[i].sh_addr,
162 / sizeof(Elf64_Rela));
166 #ifdef CONFIG_DYNAMIC_FTRACE
167 /* make the trampoline to the ftrace_caller */
171 DEBUGP("Looks like a total of %lu stubs, max\n", relocs);
172 return relocs * sizeof(struct ppc64_stub_entry);
175 /* Still needed for ELFv2, for .TOC. */
176 static void dedotify_versions(struct modversion_info *vers,
179 struct modversion_info *end;
181 for (end = (void *)vers + size; vers < end; vers++)
182 if (vers->name[0] == '.')
183 memmove(vers->name, vers->name+1, strlen(vers->name));
186 /* Undefined symbols which refer to .funcname, hack to funcname (or .TOC.) */
187 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab)
191 for (i = 1; i < numsyms; i++) {
192 if (syms[i].st_shndx == SHN_UNDEF) {
193 char *name = strtab + syms[i].st_name;
195 memmove(name, name+1, strlen(name));
200 static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs,
202 unsigned int symindex)
204 unsigned int i, numsyms;
207 syms = (Elf64_Sym *)sechdrs[symindex].sh_addr;
208 numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym);
210 for (i = 1; i < numsyms; i++) {
211 if (syms[i].st_shndx == SHN_UNDEF
212 && strcmp(strtab + syms[i].st_name, ".TOC.") == 0)
218 int module_frob_arch_sections(Elf64_Ehdr *hdr,
225 /* Find .toc and .stubs sections, symtab and strtab */
226 for (i = 1; i < hdr->e_shnum; i++) {
228 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0)
229 me->arch.stubs_section = i;
230 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0)
231 me->arch.toc_section = i;
232 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0)
233 dedotify_versions((void *)hdr + sechdrs[i].sh_offset,
236 /* We don't handle .init for the moment: rename to _init */
237 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init")))
240 if (sechdrs[i].sh_type == SHT_SYMTAB)
241 dedotify((void *)hdr + sechdrs[i].sh_offset,
242 sechdrs[i].sh_size / sizeof(Elf64_Sym),
244 + sechdrs[sechdrs[i].sh_link].sh_offset);
247 if (!me->arch.stubs_section) {
248 printk("%s: doesn't contain .stubs.\n", me->name);
252 /* If we don't have a .toc, just use .stubs. We need to set r2
253 to some reasonable value in case the module calls out to
254 other functions via a stub, or if a function pointer escapes
255 the module by some means. */
256 if (!me->arch.toc_section)
257 me->arch.toc_section = me->arch.stubs_section;
259 /* Override the stubs size */
260 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs);
264 /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this
265 gives the value maximum span in an instruction which uses a signed
267 static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me)
269 return sechdrs[me->arch.toc_section].sh_addr + 0x8000;
272 /* Both low and high 16 bits are added as SIGNED additions, so if low
273 16 bits has high bit set, high 16 bits must be adjusted. These
274 macros do that (stolen from binutils). */
275 #define PPC_LO(v) ((v) & 0xffff)
276 #define PPC_HI(v) (((v) >> 16) & 0xffff)
277 #define PPC_HA(v) PPC_HI ((v) + 0x8000)
279 /* Patch stub to reference function and correct r2 value. */
280 static inline int create_stub(Elf64_Shdr *sechdrs,
281 struct ppc64_stub_entry *entry,
282 struct ppc64_opd_entry *opd,
289 /* Stub uses address relative to r2. */
290 reladdr = (unsigned long)entry - my_r2(sechdrs, me);
291 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
292 printk("%s: Address %p of stub out of range of %p.\n",
293 me->name, (void *)reladdr, (void *)my_r2);
296 DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr);
298 entry->jump[0] |= PPC_HA(reladdr);
299 entry->jump[1] |= PPC_LO(reladdr);
300 entry->opd.funcaddr = opd->funcaddr;
301 entry->opd.r2 = opd->r2;
305 /* Create stub to jump to function described in this OPD: we need the
306 stub to set up the TOC ptr (r2) for the function. */
307 static unsigned long stub_for_addr(Elf64_Shdr *sechdrs,
308 unsigned long opdaddr,
311 struct ppc64_stub_entry *stubs;
312 struct ppc64_opd_entry *opd = (void *)opdaddr;
313 unsigned int i, num_stubs;
315 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs);
317 /* Find this stub, or if that fails, the next avail. entry */
318 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr;
319 for (i = 0; stubs[i].opd.funcaddr; i++) {
320 BUG_ON(i >= num_stubs);
322 if (stubs[i].opd.funcaddr == opd->funcaddr)
323 return (unsigned long)&stubs[i];
326 if (!create_stub(sechdrs, &stubs[i], opd, me))
329 return (unsigned long)&stubs[i];
332 /* We expect a noop next: if it is, replace it with instruction to
334 static int restore_r2(u32 *instruction, struct module *me)
336 if (*instruction != PPC_INST_NOP) {
337 printk("%s: Expect noop after relocate, got %08x\n",
338 me->name, *instruction);
341 *instruction = 0xe8410028; /* ld r2,40(r1) */
345 int apply_relocate_add(Elf64_Shdr *sechdrs,
347 unsigned int symindex,
352 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr;
354 unsigned long *location;
357 DEBUGP("Applying ADD relocate section %u to %u\n", relsec,
358 sechdrs[relsec].sh_info);
360 /* First time we're called, we can fix up .TOC. */
361 if (!me->arch.toc_fixed) {
362 sym = find_dot_toc(sechdrs, strtab, symindex);
363 /* It's theoretically possible that a module doesn't want a
364 * .TOC. so don't fail it just for that. */
366 sym->st_value = my_r2(sechdrs, me);
367 me->arch.toc_fixed = true;
370 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
371 /* This is where to make the change */
372 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
374 /* This is the symbol it is referring to */
375 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
376 + ELF64_R_SYM(rela[i].r_info);
378 DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n",
379 location, (long)ELF64_R_TYPE(rela[i].r_info),
380 strtab + sym->st_name, (unsigned long)sym->st_value,
381 (long)rela[i].r_addend);
383 /* `Everything is relative'. */
384 value = sym->st_value + rela[i].r_addend;
386 switch (ELF64_R_TYPE(rela[i].r_info)) {
389 *(u32 *)location = value;
394 *(unsigned long *)location = value;
398 *(unsigned long *)location = my_r2(sechdrs, me);
402 /* Subtract TOC pointer */
403 value -= my_r2(sechdrs, me);
404 if (value + 0x8000 > 0xffff) {
405 printk("%s: bad TOC16 relocation (%lu)\n",
409 *((uint16_t *) location)
410 = (*((uint16_t *) location) & ~0xffff)
414 case R_PPC64_TOC16_LO:
415 /* Subtract TOC pointer */
416 value -= my_r2(sechdrs, me);
417 *((uint16_t *) location)
418 = (*((uint16_t *) location) & ~0xffff)
422 case R_PPC64_TOC16_DS:
423 /* Subtract TOC pointer */
424 value -= my_r2(sechdrs, me);
425 if ((value & 3) != 0 || value + 0x8000 > 0xffff) {
426 printk("%s: bad TOC16_DS relocation (%lu)\n",
430 *((uint16_t *) location)
431 = (*((uint16_t *) location) & ~0xfffc)
435 case R_PPC64_TOC16_LO_DS:
436 /* Subtract TOC pointer */
437 value -= my_r2(sechdrs, me);
438 if ((value & 3) != 0) {
439 printk("%s: bad TOC16_LO_DS relocation (%lu)\n",
443 *((uint16_t *) location)
444 = (*((uint16_t *) location) & ~0xfffc)
448 case R_PPC64_TOC16_HA:
449 /* Subtract TOC pointer */
450 value -= my_r2(sechdrs, me);
451 value = ((value + 0x8000) >> 16);
452 *((uint16_t *) location)
453 = (*((uint16_t *) location) & ~0xffff)
458 /* FIXME: Handle weak symbols here --RR */
459 if (sym->st_shndx == SHN_UNDEF) {
460 /* External: go via stub */
461 value = stub_for_addr(sechdrs, value, me);
464 if (!restore_r2((u32 *)location + 1, me))
468 /* Convert value to relative */
469 value -= (unsigned long)location;
470 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){
471 printk("%s: REL24 %li out of range!\n",
472 me->name, (long int)value);
476 /* Only replace bits 2 through 26 */
477 *(uint32_t *)location
478 = (*(uint32_t *)location & ~0x03fffffc)
479 | (value & 0x03fffffc);
483 /* 64 bits relative (used by features fixups) */
484 *location = value - (unsigned long)location;
487 case R_PPC64_TOCSAVE:
489 * Marker reloc indicates we don't have to save r2.
490 * That would only save us one instruction, so ignore
495 case R_PPC64_REL16_HA:
496 /* Subtract location pointer */
497 value -= (unsigned long)location;
498 value = ((value + 0x8000) >> 16);
499 *((uint16_t *) location)
500 = (*((uint16_t *) location) & ~0xffff)
504 case R_PPC64_REL16_LO:
505 /* Subtract location pointer */
506 value -= (unsigned long)location;
507 *((uint16_t *) location)
508 = (*((uint16_t *) location) & ~0xffff)
513 printk("%s: Unknown ADD relocation: %lu\n",
515 (unsigned long)ELF64_R_TYPE(rela[i].r_info));
520 #ifdef CONFIG_DYNAMIC_FTRACE
521 me->arch.toc = my_r2(sechdrs, me);
522 me->arch.tramp = stub_for_addr(sechdrs,
523 (unsigned long)ftrace_caller,