1/* Kernel module help for PPC64. 2 Copyright (C) 2001, 2003 Rusty Russell IBM Corporation. 3 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. 8 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. 13 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 17*/ 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 30#include "setup.h" 31 32#define DEBUGP(fmt , ...) 33 34/* Like PPC32, we need little trampolines to do > 24-bit jumps (into 35 the kernel itself). But on PPC64, these need to be used for every 36 jump, actually, to reset r2 (TOC+0x8000). */ 37struct ppc64_stub_entry 38{ 39 /* 28 byte jump instruction sequence (7 instructions) */ 40 unsigned char jump[28]; 41 unsigned char unused[4]; 42 /* Data for the above code */ 43 struct ppc64_opd_entry opd; 44}; 45 46/* We use a stub to fix up r2 (TOC ptr) and to jump to the (external) 47 function which may be more than 24-bits away. We could simply 48 patch the new r2 value and function pointer into the stub, but it's 49 significantly shorter to put these values at the end of the stub 50 code, and patch the stub address (32-bits relative to the TOC ptr, 51 r2) into the stub. */ 52static struct ppc64_stub_entry ppc64_stub = 53{ .jump = { 54 0x3d, 0x82, 0x00, 0x00, /* addis r12,r2, <high> */ 55 0x39, 0x8c, 0x00, 0x00, /* addi r12,r12, <low> */ 56 /* Save current r2 value in magic place on the stack. */ 57 0xf8, 0x41, 0x00, 0x28, /* std r2,40(r1) */ 58 0xe9, 0x6c, 0x00, 0x20, /* ld r11,32(r12) */ 59 0xe8, 0x4c, 0x00, 0x28, /* ld r2,40(r12) */ 60 0x7d, 0x69, 0x03, 0xa6, /* mtctr r11 */ 61 0x4e, 0x80, 0x04, 0x20 /* bctr */ 62} }; 63 64/* Count how many different 24-bit relocations (different symbol, 65 different addend) */ 66static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num) 67{ 68 unsigned int i, r_info, r_addend, _count_relocs; 69 70 _count_relocs = 0; 71 r_info = 0; 72 r_addend = 0; 73 for (i = 0; i < num; i++) 74 /* Only count 24-bit relocs, others don't need stubs */ 75 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 && 76 (r_info != ELF64_R_SYM(rela[i].r_info) || 77 r_addend != rela[i].r_addend)) { 78 _count_relocs++; 79 r_info = ELF64_R_SYM(rela[i].r_info); 80 r_addend = rela[i].r_addend; 81 } 82 83 return _count_relocs; 84} 85 86static int relacmp(const void *_x, const void *_y) 87{ 88 const Elf64_Rela *x, *y; 89 90 y = (Elf64_Rela *)_x; 91 x = (Elf64_Rela *)_y; 92 93 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to 94 * make the comparison cheaper/faster. It won't affect the sorting or 95 * the counting algorithms' performance 96 */ 97 if (x->r_info < y->r_info) 98 return -1; 99 else if (x->r_info > y->r_info) 100 return 1; 101 else if (x->r_addend < y->r_addend) 102 return -1; 103 else if (x->r_addend > y->r_addend) 104 return 1; 105 else 106 return 0; 107} 108 109static void relaswap(void *_x, void *_y, int size) 110{ 111 uint64_t *x, *y, tmp; 112 int i; 113 114 y = (uint64_t *)_x; 115 x = (uint64_t *)_y; 116 117 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) { 118 tmp = x[i]; 119 x[i] = y[i]; 120 y[i] = tmp; 121 } 122} 123 124/* Get size of potential trampolines required. */ 125static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, 126 const Elf64_Shdr *sechdrs) 127{ 128 /* One extra reloc so it's always 0-funcaddr terminated */ 129 unsigned long relocs = 1; 130 unsigned i; 131 132 /* Every relocated section... */ 133 for (i = 1; i < hdr->e_shnum; i++) { 134 if (sechdrs[i].sh_type == SHT_RELA) { 135 DEBUGP("Found relocations in section %u\n", i); 136 DEBUGP("Ptr: %p. Number: %lu\n", 137 (void *)sechdrs[i].sh_addr, 138 sechdrs[i].sh_size / sizeof(Elf64_Rela)); 139 140 /* Sort the relocation information based on a symbol and 141 * addend key. This is a stable O(n*log n) complexity 142 * alogrithm but it will reduce the complexity of 143 * count_relocs() to linear complexity O(n) 144 */ 145 sort((void *)sechdrs[i].sh_addr, 146 sechdrs[i].sh_size / sizeof(Elf64_Rela), 147 sizeof(Elf64_Rela), relacmp, relaswap); 148 149 relocs += count_relocs((void *)sechdrs[i].sh_addr, 150 sechdrs[i].sh_size 151 / sizeof(Elf64_Rela)); 152 } 153 } 154 155#ifdef CONFIG_DYNAMIC_FTRACE 156 /* make the trampoline to the ftrace_caller */ 157 relocs++; 158#endif 159 160 DEBUGP("Looks like a total of %lu stubs, max\n", relocs); 161 return relocs * sizeof(struct ppc64_stub_entry); 162} 163 164static void dedotify_versions(struct modversion_info *vers, 165 unsigned long size) 166{ 167 struct modversion_info *end; 168 169 for (end = (void *)vers + size; vers < end; vers++) 170 if (vers->name[0] == '.') 171 memmove(vers->name, vers->name+1, strlen(vers->name)); 172} 173 174/* Undefined symbols which refer to .funcname, hack to funcname */ 175static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) 176{ 177 unsigned int i; 178 179 for (i = 1; i < numsyms; i++) { 180 if (syms[i].st_shndx == SHN_UNDEF) { 181 char *name = strtab + syms[i].st_name; 182 if (name[0] == '.') 183 memmove(name, name+1, strlen(name)); 184 } 185 } 186} 187 188int module_frob_arch_sections(Elf64_Ehdr *hdr, 189 Elf64_Shdr *sechdrs, 190 char *secstrings, 191 struct module *me) 192{ 193 unsigned int i; 194 195 /* Find .toc and .stubs sections, symtab and strtab */ 196 for (i = 1; i < hdr->e_shnum; i++) { 197 char *p; 198 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) 199 me->arch.stubs_section = i; 200 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) 201 me->arch.toc_section = i; 202 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) 203 dedotify_versions((void *)hdr + sechdrs[i].sh_offset, 204 sechdrs[i].sh_size); 205 206 /* We don't handle .init for the moment: rename to _init */ 207 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init"))) 208 p[0] = '_'; 209 210 if (sechdrs[i].sh_type == SHT_SYMTAB) 211 dedotify((void *)hdr + sechdrs[i].sh_offset, 212 sechdrs[i].sh_size / sizeof(Elf64_Sym), 213 (void *)hdr 214 + sechdrs[sechdrs[i].sh_link].sh_offset); 215 } 216 217 if (!me->arch.stubs_section) { 218 printk("%s: doesn't contain .stubs.\n", me->name); 219 return -ENOEXEC; 220 } 221 222 /* If we don't have a .toc, just use .stubs. We need to set r2 223 to some reasonable value in case the module calls out to 224 other functions via a stub, or if a function pointer escapes 225 the module by some means. */ 226 if (!me->arch.toc_section) 227 me->arch.toc_section = me->arch.stubs_section; 228 229 /* Override the stubs size */ 230 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); 231 return 0; 232} 233 234int apply_relocate(Elf64_Shdr *sechdrs, 235 const char *strtab, 236 unsigned int symindex, 237 unsigned int relsec, 238 struct module *me) 239{ 240 printk(KERN_ERR "%s: Non-ADD RELOCATION unsupported\n", me->name); 241 return -ENOEXEC; 242} 243 244/* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this 245 gives the value maximum span in an instruction which uses a signed 246 offset) */ 247static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me) 248{ 249 return sechdrs[me->arch.toc_section].sh_addr + 0x8000; 250} 251 252/* Both low and high 16 bits are added as SIGNED additions, so if low 253 16 bits has high bit set, high 16 bits must be adjusted. These 254 macros do that (stolen from binutils). */ 255#define PPC_LO(v) ((v) & 0xffff) 256#define PPC_HI(v) (((v) >> 16) & 0xffff) 257#define PPC_HA(v) PPC_HI ((v) + 0x8000) 258 259/* Patch stub to reference function and correct r2 value. */ 260static inline int create_stub(Elf64_Shdr *sechdrs, 261 struct ppc64_stub_entry *entry, 262 struct ppc64_opd_entry *opd, 263 struct module *me) 264{ 265 Elf64_Half *loc1, *loc2; 266 long reladdr; 267 268 *entry = ppc64_stub; 269 270 loc1 = (Elf64_Half *)&entry->jump[2]; 271 loc2 = (Elf64_Half *)&entry->jump[6]; 272 273 /* Stub uses address relative to r2. */ 274 reladdr = (unsigned long)entry - my_r2(sechdrs, me); 275 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 276 printk("%s: Address %p of stub out of range of %p.\n", 277 me->name, (void *)reladdr, (void *)my_r2); 278 return 0; 279 } 280 DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr); 281 282 *loc1 = PPC_HA(reladdr); 283 *loc2 = PPC_LO(reladdr); 284 entry->opd.funcaddr = opd->funcaddr; 285 entry->opd.r2 = opd->r2; 286 return 1; 287} 288 289/* Create stub to jump to function described in this OPD: we need the 290 stub to set up the TOC ptr (r2) for the function. */ 291static unsigned long stub_for_addr(Elf64_Shdr *sechdrs, 292 unsigned long opdaddr, 293 struct module *me) 294{ 295 struct ppc64_stub_entry *stubs; 296 struct ppc64_opd_entry *opd = (void *)opdaddr; 297 unsigned int i, num_stubs; 298 299 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); 300 301 /* Find this stub, or if that fails, the next avail. entry */ 302 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; 303 for (i = 0; stubs[i].opd.funcaddr; i++) { 304 BUG_ON(i >= num_stubs); 305 306 if (stubs[i].opd.funcaddr == opd->funcaddr) 307 return (unsigned long)&stubs[i]; 308 } 309 310 if (!create_stub(sechdrs, &stubs[i], opd, me)) 311 return 0; 312 313 return (unsigned long)&stubs[i]; 314} 315 316/* We expect a noop next: if it is, replace it with instruction to 317 restore r2. */ 318static int restore_r2(u32 *instruction, struct module *me) 319{ 320 if (*instruction != PPC_INST_NOP) { 321 printk("%s: Expect noop after relocate, got %08x\n", 322 me->name, *instruction); 323 return 0; 324 } 325 *instruction = 0xe8410028; /* ld r2,40(r1) */ 326 return 1; 327} 328 329int apply_relocate_add(Elf64_Shdr *sechdrs, 330 const char *strtab, 331 unsigned int symindex, 332 unsigned int relsec, 333 struct module *me) 334{ 335 unsigned int i; 336 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; 337 Elf64_Sym *sym; 338 unsigned long *location; 339 unsigned long value; 340 341 DEBUGP("Applying ADD relocate section %u to %u\n", relsec, 342 sechdrs[relsec].sh_info); 343 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { 344 /* This is where to make the change */ 345 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr 346 + rela[i].r_offset; 347 /* This is the symbol it is referring to */ 348 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr 349 + ELF64_R_SYM(rela[i].r_info); 350 351 DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n", 352 location, (long)ELF64_R_TYPE(rela[i].r_info), 353 strtab + sym->st_name, (unsigned long)sym->st_value, 354 (long)rela[i].r_addend); 355 356 /* `Everything is relative'. */ 357 value = sym->st_value + rela[i].r_addend; 358 359 switch (ELF64_R_TYPE(rela[i].r_info)) { 360 case R_PPC64_ADDR32: 361 /* Simply set it */ 362 *(u32 *)location = value; 363 break; 364 365 case R_PPC64_ADDR64: 366 /* Simply set it */ 367 *(unsigned long *)location = value; 368 break; 369 370 case R_PPC64_TOC: 371 *(unsigned long *)location = my_r2(sechdrs, me); 372 break; 373 374 case R_PPC64_TOC16: 375 /* Subtract TOC pointer */ 376 value -= my_r2(sechdrs, me); 377 if (value + 0x8000 > 0xffff) { 378 printk("%s: bad TOC16 relocation (%lu)\n", 379 me->name, value); 380 return -ENOEXEC; 381 } 382 *((uint16_t *) location) 383 = (*((uint16_t *) location) & ~0xffff) 384 | (value & 0xffff); 385 break; 386 387 case R_PPC64_TOC16_DS: 388 /* Subtract TOC pointer */ 389 value -= my_r2(sechdrs, me); 390 if ((value & 3) != 0 || value + 0x8000 > 0xffff) { 391 printk("%s: bad TOC16_DS relocation (%lu)\n", 392 me->name, value); 393 return -ENOEXEC; 394 } 395 *((uint16_t *) location) 396 = (*((uint16_t *) location) & ~0xfffc) 397 | (value & 0xfffc); 398 break; 399 400 case R_PPC_REL24: 401 if (sym->st_shndx == SHN_UNDEF) { 402 /* External: go via stub */ 403 value = stub_for_addr(sechdrs, value, me); 404 if (!value) 405 return -ENOENT; 406 if (!restore_r2((u32 *)location + 1, me)) 407 return -ENOEXEC; 408 } 409 410 /* Convert value to relative */ 411 value -= (unsigned long)location; 412 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ 413 printk("%s: REL24 %li out of range!\n", 414 me->name, (long int)value); 415 return -ENOEXEC; 416 } 417 418 /* Only replace bits 2 through 26 */ 419 *(uint32_t *)location 420 = (*(uint32_t *)location & ~0x03fffffc) 421 | (value & 0x03fffffc); 422 break; 423 424 case R_PPC64_REL64: 425 /* 64 bits relative (used by features fixups) */ 426 *location = value - (unsigned long)location; 427 break; 428 429 default: 430 printk("%s: Unknown ADD relocation: %lu\n", 431 me->name, 432 (unsigned long)ELF64_R_TYPE(rela[i].r_info)); 433 return -ENOEXEC; 434 } 435 } 436 437#ifdef CONFIG_DYNAMIC_FTRACE 438 me->arch.toc = my_r2(sechdrs, me); 439 me->arch.tramp = stub_for_addr(sechdrs, 440 (unsigned long)ftrace_caller, 441 me); 442#endif 443 444 return 0; 445} 446