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 if (!me->arch.stubs_section || !me->arch.toc_section) { 195 printk("%s: doesn't contain .toc or .stubs.\n", me->name); 196 return -ENOEXEC; 197 } 198 199 /* Override the stubs size */ 200 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); 201 return 0; 202 } 203 204 int apply_relocate(Elf64_Shdr *sechdrs, 205 const char *strtab, 206 unsigned int symindex, 207 unsigned int relsec, 208 struct module *me) 209 { 210 printk(KERN_ERR "%s: Non-ADD RELOCATION unsupported\n", me->name); 211 return -ENOEXEC; 212 } 213 214 /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this 215 gives the value maximum span in an instruction which uses a signed 216 offset) */ 217 static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me) 218 { 219 return sechdrs[me->arch.toc_section].sh_addr + 0x8000; 220 } 221 222 /* Both low and high 16 bits are added as SIGNED additions, so if low 223 16 bits has high bit set, high 16 bits must be adjusted. These 224 macros do that (stolen from binutils). */ 225 #define PPC_LO(v) ((v) & 0xffff) 226 #define PPC_HI(v) (((v) >> 16) & 0xffff) 227 #define PPC_HA(v) PPC_HI ((v) + 0x8000) 228 229 /* Patch stub to reference function and correct r2 value. */ 230 static inline int create_stub(Elf64_Shdr *sechdrs, 231 struct ppc64_stub_entry *entry, 232 struct ppc64_opd_entry *opd, 233 struct module *me) 234 { 235 Elf64_Half *loc1, *loc2; 236 long reladdr; 237 238 *entry = ppc64_stub; 239 240 loc1 = (Elf64_Half *)&entry->jump[2]; 241 loc2 = (Elf64_Half *)&entry->jump[6]; 242 243 /* Stub uses address relative to r2. */ 244 reladdr = (unsigned long)entry - my_r2(sechdrs, me); 245 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 246 printk("%s: Address %p of stub out of range of %p.\n", 247 me->name, (void *)reladdr, (void *)my_r2); 248 return 0; 249 } 250 DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr); 251 252 *loc1 = PPC_HA(reladdr); 253 *loc2 = PPC_LO(reladdr); 254 entry->opd.funcaddr = opd->funcaddr; 255 entry->opd.r2 = opd->r2; 256 return 1; 257 } 258 259 /* Create stub to jump to function described in this OPD: we need the 260 stub to set up the TOC ptr (r2) for the function. */ 261 static unsigned long stub_for_addr(Elf64_Shdr *sechdrs, 262 unsigned long opdaddr, 263 struct module *me) 264 { 265 struct ppc64_stub_entry *stubs; 266 struct ppc64_opd_entry *opd = (void *)opdaddr; 267 unsigned int i, num_stubs; 268 269 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); 270 271 /* Find this stub, or if that fails, the next avail. entry */ 272 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; 273 for (i = 0; stubs[i].opd.funcaddr; i++) { 274 BUG_ON(i >= num_stubs); 275 276 if (stubs[i].opd.funcaddr == opd->funcaddr) 277 return (unsigned long)&stubs[i]; 278 } 279 280 if (!create_stub(sechdrs, &stubs[i], opd, me)) 281 return 0; 282 283 return (unsigned long)&stubs[i]; 284 } 285 286 /* We expect a noop next: if it is, replace it with instruction to 287 restore r2. */ 288 static int restore_r2(u32 *instruction, struct module *me) 289 { 290 if (*instruction != 0x60000000) { 291 printk("%s: Expect noop after relocate, got %08x\n", 292 me->name, *instruction); 293 return 0; 294 } 295 *instruction = 0xe8410028; /* ld r2,40(r1) */ 296 return 1; 297 } 298 299 int apply_relocate_add(Elf64_Shdr *sechdrs, 300 const char *strtab, 301 unsigned int symindex, 302 unsigned int relsec, 303 struct module *me) 304 { 305 unsigned int i; 306 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; 307 Elf64_Sym *sym; 308 unsigned long *location; 309 unsigned long value; 310 311 DEBUGP("Applying ADD relocate section %u to %u\n", relsec, 312 sechdrs[relsec].sh_info); 313 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { 314 /* This is where to make the change */ 315 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr 316 + rela[i].r_offset; 317 /* This is the symbol it is referring to */ 318 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr 319 + ELF64_R_SYM(rela[i].r_info); 320 321 DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n", 322 location, (long)ELF64_R_TYPE(rela[i].r_info), 323 strtab + sym->st_name, (unsigned long)sym->st_value, 324 (long)rela[i].r_addend); 325 326 /* `Everything is relative'. */ 327 value = sym->st_value + rela[i].r_addend; 328 329 switch (ELF64_R_TYPE(rela[i].r_info)) { 330 case R_PPC64_ADDR32: 331 /* Simply set it */ 332 *(u32 *)location = value; 333 break; 334 335 case R_PPC64_ADDR64: 336 /* Simply set it */ 337 *(unsigned long *)location = value; 338 break; 339 340 case R_PPC64_TOC: 341 *(unsigned long *)location = my_r2(sechdrs, me); 342 break; 343 344 case R_PPC64_TOC16: 345 /* Subtact TOC pointer */ 346 value -= my_r2(sechdrs, me); 347 if (value + 0x8000 > 0xffff) { 348 printk("%s: bad TOC16 relocation (%lu)\n", 349 me->name, value); 350 return -ENOEXEC; 351 } 352 *((uint16_t *) location) 353 = (*((uint16_t *) location) & ~0xffff) 354 | (value & 0xffff); 355 break; 356 357 case R_PPC64_TOC16_DS: 358 /* Subtact TOC pointer */ 359 value -= my_r2(sechdrs, me); 360 if ((value & 3) != 0 || value + 0x8000 > 0xffff) { 361 printk("%s: bad TOC16_DS relocation (%lu)\n", 362 me->name, value); 363 return -ENOEXEC; 364 } 365 *((uint16_t *) location) 366 = (*((uint16_t *) location) & ~0xfffc) 367 | (value & 0xfffc); 368 break; 369 370 case R_PPC_REL24: 371 /* FIXME: Handle weak symbols here --RR */ 372 if (sym->st_shndx == SHN_UNDEF) { 373 /* External: go via stub */ 374 value = stub_for_addr(sechdrs, value, me); 375 if (!value) 376 return -ENOENT; 377 if (!restore_r2((u32 *)location + 1, me)) 378 return -ENOEXEC; 379 } 380 381 /* Convert value to relative */ 382 value -= (unsigned long)location; 383 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ 384 printk("%s: REL24 %li out of range!\n", 385 me->name, (long int)value); 386 return -ENOEXEC; 387 } 388 389 /* Only replace bits 2 through 26 */ 390 *(uint32_t *)location 391 = (*(uint32_t *)location & ~0x03fffffc) 392 | (value & 0x03fffffc); 393 break; 394 395 default: 396 printk("%s: Unknown ADD relocation: %lu\n", 397 me->name, 398 (unsigned long)ELF64_R_TYPE(rela[i].r_info)); 399 return -ENOEXEC; 400 } 401 } 402 403 return 0; 404 } 405 406 LIST_HEAD(module_bug_list); 407 408 int module_finalize(const Elf_Ehdr *hdr, 409 const Elf_Shdr *sechdrs, struct module *me) 410 { 411 char *secstrings; 412 unsigned int i; 413 414 me->arch.bug_table = NULL; 415 me->arch.num_bugs = 0; 416 417 /* Find the __bug_table section, if present */ 418 secstrings = (char *)hdr + sechdrs[hdr->e_shstrndx].sh_offset; 419 for (i = 1; i < hdr->e_shnum; i++) { 420 if (strcmp(secstrings+sechdrs[i].sh_name, "__bug_table")) 421 continue; 422 me->arch.bug_table = (void *) sechdrs[i].sh_addr; 423 me->arch.num_bugs = sechdrs[i].sh_size / sizeof(struct bug_entry); 424 break; 425 } 426 427 /* 428 * Strictly speaking this should have a spinlock to protect against 429 * traversals, but since we only traverse on BUG()s, a spinlock 430 * could potentially lead to deadlock and thus be counter-productive. 431 */ 432 list_add(&me->arch.bug_list, &module_bug_list); 433 434 return 0; 435 } 436 437 void module_arch_cleanup(struct module *mod) 438 { 439 list_del(&mod->arch.bug_list); 440 } 441 442 struct bug_entry *module_find_bug(unsigned long bugaddr) 443 { 444 struct mod_arch_specific *mod; 445 unsigned int i; 446 struct bug_entry *bug; 447 448 list_for_each_entry(mod, &module_bug_list, bug_list) { 449 bug = mod->bug_table; 450 for (i = 0; i < mod->num_bugs; ++i, ++bug) 451 if (bugaddr == bug->bug_addr) 452 return bug; 453 } 454 return NULL; 455 } 456