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 <asm/module.h> 24 #include <asm/uaccess.h> 25 26 /* FIXME: We don't do .init separately. To do this, we'd need to have 27 a separate r2 value in the init and core section, and stub between 28 them, too. 29 30 Using a magic allocator which places modules within 32MB solves 31 this, and makes other things simpler. Anton? 32 --RR. */ 33 #if 0 34 #define DEBUGP printk 35 #else 36 #define DEBUGP(fmt , ...) 37 #endif 38 39 /* There's actually a third entry here, but it's unused */ 40 struct ppc64_opd_entry 41 { 42 unsigned long funcaddr; 43 unsigned long r2; 44 }; 45 46 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into 47 the kernel itself). But on PPC64, these need to be used for every 48 jump, actually, to reset r2 (TOC+0x8000). */ 49 struct ppc64_stub_entry 50 { 51 /* 28 byte jump instruction sequence (7 instructions) */ 52 unsigned char jump[28]; 53 unsigned char unused[4]; 54 /* Data for the above code */ 55 struct ppc64_opd_entry opd; 56 }; 57 58 /* We use a stub to fix up r2 (TOC ptr) and to jump to the (external) 59 function which may be more than 24-bits away. We could simply 60 patch the new r2 value and function pointer into the stub, but it's 61 significantly shorter to put these values at the end of the stub 62 code, and patch the stub address (32-bits relative to the TOC ptr, 63 r2) into the stub. */ 64 static struct ppc64_stub_entry ppc64_stub = 65 { .jump = { 66 0x3d, 0x82, 0x00, 0x00, /* addis r12,r2, <high> */ 67 0x39, 0x8c, 0x00, 0x00, /* addi r12,r12, <low> */ 68 /* Save current r2 value in magic place on the stack. */ 69 0xf8, 0x41, 0x00, 0x28, /* std r2,40(r1) */ 70 0xe9, 0x6c, 0x00, 0x20, /* ld r11,32(r12) */ 71 0xe8, 0x4c, 0x00, 0x28, /* ld r2,40(r12) */ 72 0x7d, 0x69, 0x03, 0xa6, /* mtctr r11 */ 73 0x4e, 0x80, 0x04, 0x20 /* bctr */ 74 } }; 75 76 /* Count how many different 24-bit relocations (different symbol, 77 different addend) */ 78 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num) 79 { 80 unsigned int i, j, ret = 0; 81 82 /* FIXME: Only count external ones --RR */ 83 /* Sure, this is order(n^2), but it's usually short, and not 84 time critical */ 85 for (i = 0; i < num; i++) { 86 /* Only count 24-bit relocs, others don't need stubs */ 87 if (ELF64_R_TYPE(rela[i].r_info) != R_PPC_REL24) 88 continue; 89 for (j = 0; j < i; j++) { 90 /* If this addend appeared before, it's 91 already been counted */ 92 if (rela[i].r_info == rela[j].r_info 93 && rela[i].r_addend == rela[j].r_addend) 94 break; 95 } 96 if (j == i) ret++; 97 } 98 return ret; 99 } 100 101 void *module_alloc(unsigned long size) 102 { 103 if (size == 0) 104 return NULL; 105 106 return vmalloc_exec(size); 107 } 108 109 /* Free memory returned from module_alloc */ 110 void module_free(struct module *mod, void *module_region) 111 { 112 vfree(module_region); 113 /* FIXME: If module_region == mod->init_region, trim exception 114 table entries. */ 115 } 116 117 /* Get size of potential trampolines required. */ 118 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, 119 const Elf64_Shdr *sechdrs) 120 { 121 /* One extra reloc so it's always 0-funcaddr terminated */ 122 unsigned long relocs = 1; 123 unsigned i; 124 125 /* Every relocated section... */ 126 for (i = 1; i < hdr->e_shnum; i++) { 127 if (sechdrs[i].sh_type == SHT_RELA) { 128 DEBUGP("Found relocations in section %u\n", i); 129 DEBUGP("Ptr: %p. Number: %lu\n", 130 (void *)sechdrs[i].sh_addr, 131 sechdrs[i].sh_size / sizeof(Elf64_Rela)); 132 relocs += count_relocs((void *)sechdrs[i].sh_addr, 133 sechdrs[i].sh_size 134 / sizeof(Elf64_Rela)); 135 } 136 } 137 138 DEBUGP("Looks like a total of %lu stubs, max\n", relocs); 139 return relocs * sizeof(struct ppc64_stub_entry); 140 } 141 142 static void dedotify_versions(struct modversion_info *vers, 143 unsigned long size) 144 { 145 struct modversion_info *end; 146 147 for (end = (void *)vers + size; vers < end; vers++) 148 if (vers->name[0] == '.') 149 memmove(vers->name, vers->name+1, strlen(vers->name)); 150 } 151 152 /* Undefined symbols which refer to .funcname, hack to funcname */ 153 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) 154 { 155 unsigned int i; 156 157 for (i = 1; i < numsyms; i++) { 158 if (syms[i].st_shndx == SHN_UNDEF) { 159 char *name = strtab + syms[i].st_name; 160 if (name[0] == '.') 161 memmove(name, name+1, strlen(name)); 162 } 163 } 164 } 165 166 int module_frob_arch_sections(Elf64_Ehdr *hdr, 167 Elf64_Shdr *sechdrs, 168 char *secstrings, 169 struct module *me) 170 { 171 unsigned int i; 172 173 /* Find .toc and .stubs sections, symtab and strtab */ 174 for (i = 1; i < hdr->e_shnum; i++) { 175 char *p; 176 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) 177 me->arch.stubs_section = i; 178 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) 179 me->arch.toc_section = i; 180 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) 181 dedotify_versions((void *)hdr + sechdrs[i].sh_offset, 182 sechdrs[i].sh_size); 183 184 /* We don't handle .init for the moment: rename to _init */ 185 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init"))) 186 p[0] = '_'; 187 188 if (sechdrs[i].sh_type == SHT_SYMTAB) 189 dedotify((void *)hdr + sechdrs[i].sh_offset, 190 sechdrs[i].sh_size / sizeof(Elf64_Sym), 191 (void *)hdr 192 + sechdrs[sechdrs[i].sh_link].sh_offset); 193 } 194 195 if (!me->arch.stubs_section) { 196 printk("%s: doesn't contain .stubs.\n", me->name); 197 return -ENOEXEC; 198 } 199 200 /* If we don't have a .toc, just use .stubs. We need to set r2 201 to some reasonable value in case the module calls out to 202 other functions via a stub, or if a function pointer escapes 203 the module by some means. */ 204 if (!me->arch.toc_section) 205 me->arch.toc_section = me->arch.stubs_section; 206 207 /* Override the stubs size */ 208 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); 209 return 0; 210 } 211 212 int apply_relocate(Elf64_Shdr *sechdrs, 213 const char *strtab, 214 unsigned int symindex, 215 unsigned int relsec, 216 struct module *me) 217 { 218 printk(KERN_ERR "%s: Non-ADD RELOCATION unsupported\n", me->name); 219 return -ENOEXEC; 220 } 221 222 /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this 223 gives the value maximum span in an instruction which uses a signed 224 offset) */ 225 static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me) 226 { 227 return sechdrs[me->arch.toc_section].sh_addr + 0x8000; 228 } 229 230 /* Both low and high 16 bits are added as SIGNED additions, so if low 231 16 bits has high bit set, high 16 bits must be adjusted. These 232 macros do that (stolen from binutils). */ 233 #define PPC_LO(v) ((v) & 0xffff) 234 #define PPC_HI(v) (((v) >> 16) & 0xffff) 235 #define PPC_HA(v) PPC_HI ((v) + 0x8000) 236 237 /* Patch stub to reference function and correct r2 value. */ 238 static inline int create_stub(Elf64_Shdr *sechdrs, 239 struct ppc64_stub_entry *entry, 240 struct ppc64_opd_entry *opd, 241 struct module *me) 242 { 243 Elf64_Half *loc1, *loc2; 244 long reladdr; 245 246 *entry = ppc64_stub; 247 248 loc1 = (Elf64_Half *)&entry->jump[2]; 249 loc2 = (Elf64_Half *)&entry->jump[6]; 250 251 /* Stub uses address relative to r2. */ 252 reladdr = (unsigned long)entry - my_r2(sechdrs, me); 253 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 254 printk("%s: Address %p of stub out of range of %p.\n", 255 me->name, (void *)reladdr, (void *)my_r2); 256 return 0; 257 } 258 DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr); 259 260 *loc1 = PPC_HA(reladdr); 261 *loc2 = PPC_LO(reladdr); 262 entry->opd.funcaddr = opd->funcaddr; 263 entry->opd.r2 = opd->r2; 264 return 1; 265 } 266 267 /* Create stub to jump to function described in this OPD: we need the 268 stub to set up the TOC ptr (r2) for the function. */ 269 static unsigned long stub_for_addr(Elf64_Shdr *sechdrs, 270 unsigned long opdaddr, 271 struct module *me) 272 { 273 struct ppc64_stub_entry *stubs; 274 struct ppc64_opd_entry *opd = (void *)opdaddr; 275 unsigned int i, num_stubs; 276 277 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); 278 279 /* Find this stub, or if that fails, the next avail. entry */ 280 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; 281 for (i = 0; stubs[i].opd.funcaddr; i++) { 282 BUG_ON(i >= num_stubs); 283 284 if (stubs[i].opd.funcaddr == opd->funcaddr) 285 return (unsigned long)&stubs[i]; 286 } 287 288 if (!create_stub(sechdrs, &stubs[i], opd, me)) 289 return 0; 290 291 return (unsigned long)&stubs[i]; 292 } 293 294 /* We expect a noop next: if it is, replace it with instruction to 295 restore r2. */ 296 static int restore_r2(u32 *instruction, struct module *me) 297 { 298 if (*instruction != 0x60000000) { 299 printk("%s: Expect noop after relocate, got %08x\n", 300 me->name, *instruction); 301 return 0; 302 } 303 *instruction = 0xe8410028; /* ld r2,40(r1) */ 304 return 1; 305 } 306 307 int apply_relocate_add(Elf64_Shdr *sechdrs, 308 const char *strtab, 309 unsigned int symindex, 310 unsigned int relsec, 311 struct module *me) 312 { 313 unsigned int i; 314 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; 315 Elf64_Sym *sym; 316 unsigned long *location; 317 unsigned long value; 318 319 DEBUGP("Applying ADD relocate section %u to %u\n", relsec, 320 sechdrs[relsec].sh_info); 321 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { 322 /* This is where to make the change */ 323 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr 324 + rela[i].r_offset; 325 /* This is the symbol it is referring to */ 326 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr 327 + ELF64_R_SYM(rela[i].r_info); 328 329 DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n", 330 location, (long)ELF64_R_TYPE(rela[i].r_info), 331 strtab + sym->st_name, (unsigned long)sym->st_value, 332 (long)rela[i].r_addend); 333 334 /* `Everything is relative'. */ 335 value = sym->st_value + rela[i].r_addend; 336 337 switch (ELF64_R_TYPE(rela[i].r_info)) { 338 case R_PPC64_ADDR32: 339 /* Simply set it */ 340 *(u32 *)location = value; 341 break; 342 343 case R_PPC64_ADDR64: 344 /* Simply set it */ 345 *(unsigned long *)location = value; 346 break; 347 348 case R_PPC64_TOC: 349 *(unsigned long *)location = my_r2(sechdrs, me); 350 break; 351 352 case R_PPC64_TOC16: 353 /* Subtract TOC pointer */ 354 value -= my_r2(sechdrs, me); 355 if (value + 0x8000 > 0xffff) { 356 printk("%s: bad TOC16 relocation (%lu)\n", 357 me->name, value); 358 return -ENOEXEC; 359 } 360 *((uint16_t *) location) 361 = (*((uint16_t *) location) & ~0xffff) 362 | (value & 0xffff); 363 break; 364 365 case R_PPC64_TOC16_DS: 366 /* Subtract TOC pointer */ 367 value -= my_r2(sechdrs, me); 368 if ((value & 3) != 0 || value + 0x8000 > 0xffff) { 369 printk("%s: bad TOC16_DS relocation (%lu)\n", 370 me->name, value); 371 return -ENOEXEC; 372 } 373 *((uint16_t *) location) 374 = (*((uint16_t *) location) & ~0xfffc) 375 | (value & 0xfffc); 376 break; 377 378 case R_PPC_REL24: 379 /* FIXME: Handle weak symbols here --RR */ 380 if (sym->st_shndx == SHN_UNDEF) { 381 /* External: go via stub */ 382 value = stub_for_addr(sechdrs, value, me); 383 if (!value) 384 return -ENOENT; 385 if (!restore_r2((u32 *)location + 1, me)) 386 return -ENOEXEC; 387 } 388 389 /* Convert value to relative */ 390 value -= (unsigned long)location; 391 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ 392 printk("%s: REL24 %li out of range!\n", 393 me->name, (long int)value); 394 return -ENOEXEC; 395 } 396 397 /* Only replace bits 2 through 26 */ 398 *(uint32_t *)location 399 = (*(uint32_t *)location & ~0x03fffffc) 400 | (value & 0x03fffffc); 401 break; 402 403 default: 404 printk("%s: Unknown ADD relocation: %lu\n", 405 me->name, 406 (unsigned long)ELF64_R_TYPE(rela[i].r_info)); 407 return -ENOEXEC; 408 } 409 } 410 411 return 0; 412 } 413 414 LIST_HEAD(module_bug_list); 415 416 int module_finalize(const Elf_Ehdr *hdr, 417 const Elf_Shdr *sechdrs, struct module *me) 418 { 419 char *secstrings; 420 unsigned int i; 421 422 me->arch.bug_table = NULL; 423 me->arch.num_bugs = 0; 424 425 /* Find the __bug_table section, if present */ 426 secstrings = (char *)hdr + sechdrs[hdr->e_shstrndx].sh_offset; 427 for (i = 1; i < hdr->e_shnum; i++) { 428 if (strcmp(secstrings+sechdrs[i].sh_name, "__bug_table")) 429 continue; 430 me->arch.bug_table = (void *) sechdrs[i].sh_addr; 431 me->arch.num_bugs = sechdrs[i].sh_size / sizeof(struct bug_entry); 432 break; 433 } 434 435 /* 436 * Strictly speaking this should have a spinlock to protect against 437 * traversals, but since we only traverse on BUG()s, a spinlock 438 * could potentially lead to deadlock and thus be counter-productive. 439 */ 440 list_add(&me->arch.bug_list, &module_bug_list); 441 442 return 0; 443 } 444 445 void module_arch_cleanup(struct module *mod) 446 { 447 list_del(&mod->arch.bug_list); 448 } 449 450 struct bug_entry *module_find_bug(unsigned long bugaddr) 451 { 452 struct mod_arch_specific *mod; 453 unsigned int i; 454 struct bug_entry *bug; 455 456 list_for_each_entry(mod, &module_bug_list, bug_list) { 457 bug = mod->bug_table; 458 for (i = 0; i < mod->num_bugs; ++i, ++bug) 459 if (bugaddr == bug->bug_addr) 460 return bug; 461 } 462 return NULL; 463 } 464