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/bug.h> 24 #include <asm/module.h> 25 #include <asm/uaccess.h> 26 #include <asm/firmware.h> 27 #include <linux/sort.h> 28 29 #include "setup.h" 30 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 33 them, too. 34 35 Using a magic allocator which places modules within 32MB solves 36 this, and makes other things simpler. Anton? 37 --RR. */ 38 #if 0 39 #define DEBUGP printk 40 #else 41 #define DEBUGP(fmt , ...) 42 #endif 43 44 /* There's actually a third entry here, but it's unused */ 45 struct ppc64_opd_entry 46 { 47 unsigned long funcaddr; 48 unsigned long r2; 49 }; 50 51 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into 52 the kernel itself). But on PPC64, these need to be used for every 53 jump, actually, to reset r2 (TOC+0x8000). */ 54 struct ppc64_stub_entry 55 { 56 /* 28 byte jump instruction sequence (7 instructions) */ 57 unsigned char jump[28]; 58 unsigned char unused[4]; 59 /* Data for the above code */ 60 struct ppc64_opd_entry opd; 61 }; 62 63 /* We use a stub to fix up r2 (TOC ptr) and to jump to the (external) 64 function which may be more than 24-bits away. We could simply 65 patch the new r2 value and function pointer into the stub, but it's 66 significantly shorter to put these values at the end of the stub 67 code, and patch the stub address (32-bits relative to the TOC ptr, 68 r2) into the stub. */ 69 static struct ppc64_stub_entry ppc64_stub = 70 { .jump = { 71 0x3d, 0x82, 0x00, 0x00, /* addis r12,r2, <high> */ 72 0x39, 0x8c, 0x00, 0x00, /* addi r12,r12, <low> */ 73 /* Save current r2 value in magic place on the stack. */ 74 0xf8, 0x41, 0x00, 0x28, /* std r2,40(r1) */ 75 0xe9, 0x6c, 0x00, 0x20, /* ld r11,32(r12) */ 76 0xe8, 0x4c, 0x00, 0x28, /* ld r2,40(r12) */ 77 0x7d, 0x69, 0x03, 0xa6, /* mtctr r11 */ 78 0x4e, 0x80, 0x04, 0x20 /* bctr */ 79 } }; 80 81 /* Count how many different 24-bit relocations (different symbol, 82 different addend) */ 83 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num) 84 { 85 unsigned int i, r_info, r_addend, _count_relocs; 86 87 /* FIXME: Only count external ones --RR */ 88 _count_relocs = 0; 89 r_info = 0; 90 r_addend = 0; 91 for (i = 0; i < num; i++) 92 /* Only count 24-bit relocs, others don't need stubs */ 93 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 && 94 (r_info != ELF64_R_SYM(rela[i].r_info) || 95 r_addend != rela[i].r_addend)) { 96 _count_relocs++; 97 r_info = ELF64_R_SYM(rela[i].r_info); 98 r_addend = rela[i].r_addend; 99 } 100 101 return _count_relocs; 102 } 103 104 void *module_alloc(unsigned long size) 105 { 106 if (size == 0) 107 return NULL; 108 109 return vmalloc_exec(size); 110 } 111 112 /* Free memory returned from module_alloc */ 113 void module_free(struct module *mod, void *module_region) 114 { 115 vfree(module_region); 116 /* FIXME: If module_region == mod->init_region, trim exception 117 table entries. */ 118 } 119 120 static int relacmp(const void *_x, const void *_y) 121 { 122 const Elf64_Rela *x, *y; 123 124 y = (Elf64_Rela *)_x; 125 x = (Elf64_Rela *)_y; 126 127 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to 128 * make the comparison cheaper/faster. It won't affect the sorting or 129 * the counting algorithms' performance 130 */ 131 if (x->r_info < y->r_info) 132 return -1; 133 else if (x->r_info > y->r_info) 134 return 1; 135 else if (x->r_addend < y->r_addend) 136 return -1; 137 else if (x->r_addend > y->r_addend) 138 return 1; 139 else 140 return 0; 141 } 142 143 static void relaswap(void *_x, void *_y, int size) 144 { 145 uint64_t *x, *y, tmp; 146 int i; 147 148 y = (uint64_t *)_x; 149 x = (uint64_t *)_y; 150 151 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) { 152 tmp = x[i]; 153 x[i] = y[i]; 154 y[i] = tmp; 155 } 156 } 157 158 /* Get size of potential trampolines required. */ 159 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, 160 const Elf64_Shdr *sechdrs) 161 { 162 /* One extra reloc so it's always 0-funcaddr terminated */ 163 unsigned long relocs = 1; 164 unsigned i; 165 166 /* Every relocated section... */ 167 for (i = 1; i < hdr->e_shnum; i++) { 168 if (sechdrs[i].sh_type == SHT_RELA) { 169 DEBUGP("Found relocations in section %u\n", i); 170 DEBUGP("Ptr: %p. Number: %lu\n", 171 (void *)sechdrs[i].sh_addr, 172 sechdrs[i].sh_size / sizeof(Elf64_Rela)); 173 174 /* Sort the relocation information based on a symbol and 175 * addend key. This is a stable O(n*log n) complexity 176 * alogrithm but it will reduce the complexity of 177 * count_relocs() to linear complexity O(n) 178 */ 179 sort((void *)sechdrs[i].sh_addr, 180 sechdrs[i].sh_size / sizeof(Elf64_Rela), 181 sizeof(Elf64_Rela), relacmp, relaswap); 182 183 relocs += count_relocs((void *)sechdrs[i].sh_addr, 184 sechdrs[i].sh_size 185 / sizeof(Elf64_Rela)); 186 } 187 } 188 189 DEBUGP("Looks like a total of %lu stubs, max\n", relocs); 190 return relocs * sizeof(struct ppc64_stub_entry); 191 } 192 193 static void dedotify_versions(struct modversion_info *vers, 194 unsigned long size) 195 { 196 struct modversion_info *end; 197 198 for (end = (void *)vers + size; vers < end; vers++) 199 if (vers->name[0] == '.') 200 memmove(vers->name, vers->name+1, strlen(vers->name)); 201 } 202 203 /* Undefined symbols which refer to .funcname, hack to funcname */ 204 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) 205 { 206 unsigned int i; 207 208 for (i = 1; i < numsyms; i++) { 209 if (syms[i].st_shndx == SHN_UNDEF) { 210 char *name = strtab + syms[i].st_name; 211 if (name[0] == '.') 212 memmove(name, name+1, strlen(name)); 213 } 214 } 215 } 216 217 int module_frob_arch_sections(Elf64_Ehdr *hdr, 218 Elf64_Shdr *sechdrs, 219 char *secstrings, 220 struct module *me) 221 { 222 unsigned int i; 223 224 /* Find .toc and .stubs sections, symtab and strtab */ 225 for (i = 1; i < hdr->e_shnum; i++) { 226 char *p; 227 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) 228 me->arch.stubs_section = i; 229 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) 230 me->arch.toc_section = i; 231 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) 232 dedotify_versions((void *)hdr + sechdrs[i].sh_offset, 233 sechdrs[i].sh_size); 234 235 /* We don't handle .init for the moment: rename to _init */ 236 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init"))) 237 p[0] = '_'; 238 239 if (sechdrs[i].sh_type == SHT_SYMTAB) 240 dedotify((void *)hdr + sechdrs[i].sh_offset, 241 sechdrs[i].sh_size / sizeof(Elf64_Sym), 242 (void *)hdr 243 + sechdrs[sechdrs[i].sh_link].sh_offset); 244 } 245 246 if (!me->arch.stubs_section) { 247 printk("%s: doesn't contain .stubs.\n", me->name); 248 return -ENOEXEC; 249 } 250 251 /* If we don't have a .toc, just use .stubs. We need to set r2 252 to some reasonable value in case the module calls out to 253 other functions via a stub, or if a function pointer escapes 254 the module by some means. */ 255 if (!me->arch.toc_section) 256 me->arch.toc_section = me->arch.stubs_section; 257 258 /* Override the stubs size */ 259 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); 260 return 0; 261 } 262 263 int apply_relocate(Elf64_Shdr *sechdrs, 264 const char *strtab, 265 unsigned int symindex, 266 unsigned int relsec, 267 struct module *me) 268 { 269 printk(KERN_ERR "%s: Non-ADD RELOCATION unsupported\n", me->name); 270 return -ENOEXEC; 271 } 272 273 /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this 274 gives the value maximum span in an instruction which uses a signed 275 offset) */ 276 static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me) 277 { 278 return sechdrs[me->arch.toc_section].sh_addr + 0x8000; 279 } 280 281 /* Both low and high 16 bits are added as SIGNED additions, so if low 282 16 bits has high bit set, high 16 bits must be adjusted. These 283 macros do that (stolen from binutils). */ 284 #define PPC_LO(v) ((v) & 0xffff) 285 #define PPC_HI(v) (((v) >> 16) & 0xffff) 286 #define PPC_HA(v) PPC_HI ((v) + 0x8000) 287 288 /* Patch stub to reference function and correct r2 value. */ 289 static inline int create_stub(Elf64_Shdr *sechdrs, 290 struct ppc64_stub_entry *entry, 291 struct ppc64_opd_entry *opd, 292 struct module *me) 293 { 294 Elf64_Half *loc1, *loc2; 295 long reladdr; 296 297 *entry = ppc64_stub; 298 299 loc1 = (Elf64_Half *)&entry->jump[2]; 300 loc2 = (Elf64_Half *)&entry->jump[6]; 301 302 /* Stub uses address relative to r2. */ 303 reladdr = (unsigned long)entry - my_r2(sechdrs, me); 304 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 305 printk("%s: Address %p of stub out of range of %p.\n", 306 me->name, (void *)reladdr, (void *)my_r2); 307 return 0; 308 } 309 DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr); 310 311 *loc1 = PPC_HA(reladdr); 312 *loc2 = PPC_LO(reladdr); 313 entry->opd.funcaddr = opd->funcaddr; 314 entry->opd.r2 = opd->r2; 315 return 1; 316 } 317 318 /* Create stub to jump to function described in this OPD: we need the 319 stub to set up the TOC ptr (r2) for the function. */ 320 static unsigned long stub_for_addr(Elf64_Shdr *sechdrs, 321 unsigned long opdaddr, 322 struct module *me) 323 { 324 struct ppc64_stub_entry *stubs; 325 struct ppc64_opd_entry *opd = (void *)opdaddr; 326 unsigned int i, num_stubs; 327 328 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); 329 330 /* Find this stub, or if that fails, the next avail. entry */ 331 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; 332 for (i = 0; stubs[i].opd.funcaddr; i++) { 333 BUG_ON(i >= num_stubs); 334 335 if (stubs[i].opd.funcaddr == opd->funcaddr) 336 return (unsigned long)&stubs[i]; 337 } 338 339 if (!create_stub(sechdrs, &stubs[i], opd, me)) 340 return 0; 341 342 return (unsigned long)&stubs[i]; 343 } 344 345 /* We expect a noop next: if it is, replace it with instruction to 346 restore r2. */ 347 static int restore_r2(u32 *instruction, struct module *me) 348 { 349 if (*instruction != 0x60000000) { 350 printk("%s: Expect noop after relocate, got %08x\n", 351 me->name, *instruction); 352 return 0; 353 } 354 *instruction = 0xe8410028; /* ld r2,40(r1) */ 355 return 1; 356 } 357 358 int apply_relocate_add(Elf64_Shdr *sechdrs, 359 const char *strtab, 360 unsigned int symindex, 361 unsigned int relsec, 362 struct module *me) 363 { 364 unsigned int i; 365 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; 366 Elf64_Sym *sym; 367 unsigned long *location; 368 unsigned long value; 369 370 DEBUGP("Applying ADD relocate section %u to %u\n", relsec, 371 sechdrs[relsec].sh_info); 372 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { 373 /* This is where to make the change */ 374 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr 375 + rela[i].r_offset; 376 /* This is the symbol it is referring to */ 377 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr 378 + ELF64_R_SYM(rela[i].r_info); 379 380 DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n", 381 location, (long)ELF64_R_TYPE(rela[i].r_info), 382 strtab + sym->st_name, (unsigned long)sym->st_value, 383 (long)rela[i].r_addend); 384 385 /* `Everything is relative'. */ 386 value = sym->st_value + rela[i].r_addend; 387 388 switch (ELF64_R_TYPE(rela[i].r_info)) { 389 case R_PPC64_ADDR32: 390 /* Simply set it */ 391 *(u32 *)location = value; 392 break; 393 394 case R_PPC64_ADDR64: 395 /* Simply set it */ 396 *(unsigned long *)location = value; 397 break; 398 399 case R_PPC64_TOC: 400 *(unsigned long *)location = my_r2(sechdrs, me); 401 break; 402 403 case R_PPC64_TOC16: 404 /* Subtract TOC pointer */ 405 value -= my_r2(sechdrs, me); 406 if (value + 0x8000 > 0xffff) { 407 printk("%s: bad TOC16 relocation (%lu)\n", 408 me->name, value); 409 return -ENOEXEC; 410 } 411 *((uint16_t *) location) 412 = (*((uint16_t *) location) & ~0xffff) 413 | (value & 0xffff); 414 break; 415 416 case R_PPC64_TOC16_DS: 417 /* Subtract TOC pointer */ 418 value -= my_r2(sechdrs, me); 419 if ((value & 3) != 0 || value + 0x8000 > 0xffff) { 420 printk("%s: bad TOC16_DS relocation (%lu)\n", 421 me->name, value); 422 return -ENOEXEC; 423 } 424 *((uint16_t *) location) 425 = (*((uint16_t *) location) & ~0xfffc) 426 | (value & 0xfffc); 427 break; 428 429 case R_PPC_REL24: 430 /* FIXME: Handle weak symbols here --RR */ 431 if (sym->st_shndx == SHN_UNDEF) { 432 /* External: go via stub */ 433 value = stub_for_addr(sechdrs, value, me); 434 if (!value) 435 return -ENOENT; 436 if (!restore_r2((u32 *)location + 1, me)) 437 return -ENOEXEC; 438 } 439 440 /* Convert value to relative */ 441 value -= (unsigned long)location; 442 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ 443 printk("%s: REL24 %li out of range!\n", 444 me->name, (long int)value); 445 return -ENOEXEC; 446 } 447 448 /* Only replace bits 2 through 26 */ 449 *(uint32_t *)location 450 = (*(uint32_t *)location & ~0x03fffffc) 451 | (value & 0x03fffffc); 452 break; 453 454 case R_PPC64_REL64: 455 /* 64 bits relative (used by features fixups) */ 456 *location = value - (unsigned long)location; 457 break; 458 459 default: 460 printk("%s: Unknown ADD relocation: %lu\n", 461 me->name, 462 (unsigned long)ELF64_R_TYPE(rela[i].r_info)); 463 return -ENOEXEC; 464 } 465 } 466 467 return 0; 468 } 469 470 LIST_HEAD(module_bug_list); 471 472 static const Elf_Shdr *find_section(const Elf_Ehdr *hdr, 473 const Elf_Shdr *sechdrs, 474 const char *name) 475 { 476 char *secstrings; 477 unsigned int i; 478 479 secstrings = (char *)hdr + sechdrs[hdr->e_shstrndx].sh_offset; 480 for (i = 1; i < hdr->e_shnum; i++) 481 if (strcmp(secstrings+sechdrs[i].sh_name, name) == 0) 482 return &sechdrs[i]; 483 return NULL; 484 } 485 486 int module_finalize(const Elf_Ehdr *hdr, 487 const Elf_Shdr *sechdrs, struct module *me) 488 { 489 const Elf_Shdr *sect; 490 int err; 491 492 err = module_bug_finalize(hdr, sechdrs, me); 493 if (err) 494 return err; 495 496 /* Apply feature fixups */ 497 sect = find_section(hdr, sechdrs, "__ftr_fixup"); 498 if (sect != NULL) 499 do_feature_fixups(cur_cpu_spec->cpu_features, 500 (void *)sect->sh_addr, 501 (void *)sect->sh_addr + sect->sh_size); 502 503 sect = find_section(hdr, sechdrs, "__fw_ftr_fixup"); 504 if (sect != NULL) 505 do_feature_fixups(powerpc_firmware_features, 506 (void *)sect->sh_addr, 507 (void *)sect->sh_addr + sect->sh_size); 508 509 return 0; 510 } 511 512 void module_arch_cleanup(struct module *mod) 513 { 514 module_bug_cleanup(mod); 515 } 516 517 struct bug_entry *module_find_bug(unsigned long bugaddr) 518 { 519 struct mod_arch_specific *mod; 520 unsigned int i; 521 struct bug_entry *bug; 522 523 list_for_each_entry(mod, &module_bug_list, bug_list) { 524 bug = mod->bug_table; 525 for (i = 0; i < mod->num_bugs; ++i, ++bug) 526 if (bugaddr == bug->bug_addr) 527 return bug; 528 } 529 return NULL; 530 } 531