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 #include <asm/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 /* 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 48 { 49 /* 28 byte jump instruction sequence (7 instructions) */ 50 unsigned char jump[28]; 51 unsigned char unused[4]; 52 /* Data for the above code */ 53 struct ppc64_opd_entry opd; 54 }; 55 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, 61 r2) into the stub. */ 62 static struct ppc64_stub_entry ppc64_stub = 63 { .jump = { 64 #ifdef __LITTLE_ENDIAN__ 65 0x00, 0x00, 0x82, 0x3d, /* addis r12,r2, <high> */ 66 0x00, 0x00, 0x8c, 0x39, /* addi r12,r12, <low> */ 67 /* Save current r2 value in magic place on the stack. */ 68 0x28, 0x00, 0x41, 0xf8, /* std r2,40(r1) */ 69 0x20, 0x00, 0x6c, 0xe9, /* ld r11,32(r12) */ 70 0x28, 0x00, 0x4c, 0xe8, /* ld r2,40(r12) */ 71 0xa6, 0x03, 0x69, 0x7d, /* mtctr r11 */ 72 0x20, 0x04, 0x80, 0x4e /* bctr */ 73 #else 74 0x3d, 0x82, 0x00, 0x00, /* addis r12,r2, <high> */ 75 0x39, 0x8c, 0x00, 0x00, /* addi r12,r12, <low> */ 76 /* Save current r2 value in magic place on the stack. */ 77 0xf8, 0x41, 0x00, 0x28, /* std r2,40(r1) */ 78 0xe9, 0x6c, 0x00, 0x20, /* ld r11,32(r12) */ 79 0xe8, 0x4c, 0x00, 0x28, /* ld r2,40(r12) */ 80 0x7d, 0x69, 0x03, 0xa6, /* mtctr r11 */ 81 0x4e, 0x80, 0x04, 0x20 /* bctr */ 82 #endif 83 } }; 84 85 /* Count how many different 24-bit relocations (different symbol, 86 different addend) */ 87 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num) 88 { 89 unsigned int i, r_info, r_addend, _count_relocs; 90 91 /* FIXME: Only count external ones --RR */ 92 _count_relocs = 0; 93 r_info = 0; 94 r_addend = 0; 95 for (i = 0; i < num; i++) 96 /* Only count 24-bit relocs, others don't need stubs */ 97 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 && 98 (r_info != ELF64_R_SYM(rela[i].r_info) || 99 r_addend != rela[i].r_addend)) { 100 _count_relocs++; 101 r_info = ELF64_R_SYM(rela[i].r_info); 102 r_addend = rela[i].r_addend; 103 } 104 105 return _count_relocs; 106 } 107 108 static int relacmp(const void *_x, const void *_y) 109 { 110 const Elf64_Rela *x, *y; 111 112 y = (Elf64_Rela *)_x; 113 x = (Elf64_Rela *)_y; 114 115 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to 116 * make the comparison cheaper/faster. It won't affect the sorting or 117 * the counting algorithms' performance 118 */ 119 if (x->r_info < y->r_info) 120 return -1; 121 else if (x->r_info > y->r_info) 122 return 1; 123 else if (x->r_addend < y->r_addend) 124 return -1; 125 else if (x->r_addend > y->r_addend) 126 return 1; 127 else 128 return 0; 129 } 130 131 static void relaswap(void *_x, void *_y, int size) 132 { 133 uint64_t *x, *y, tmp; 134 int i; 135 136 y = (uint64_t *)_x; 137 x = (uint64_t *)_y; 138 139 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) { 140 tmp = x[i]; 141 x[i] = y[i]; 142 y[i] = tmp; 143 } 144 } 145 146 /* Get size of potential trampolines required. */ 147 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, 148 const Elf64_Shdr *sechdrs) 149 { 150 /* One extra reloc so it's always 0-funcaddr terminated */ 151 unsigned long relocs = 1; 152 unsigned i; 153 154 /* Every relocated section... */ 155 for (i = 1; i < hdr->e_shnum; i++) { 156 if (sechdrs[i].sh_type == SHT_RELA) { 157 DEBUGP("Found relocations in section %u\n", i); 158 DEBUGP("Ptr: %p. Number: %lu\n", 159 (void *)sechdrs[i].sh_addr, 160 sechdrs[i].sh_size / sizeof(Elf64_Rela)); 161 162 /* Sort the relocation information based on a symbol and 163 * addend key. This is a stable O(n*log n) complexity 164 * alogrithm but it will reduce the complexity of 165 * count_relocs() to linear complexity O(n) 166 */ 167 sort((void *)sechdrs[i].sh_addr, 168 sechdrs[i].sh_size / sizeof(Elf64_Rela), 169 sizeof(Elf64_Rela), relacmp, relaswap); 170 171 relocs += count_relocs((void *)sechdrs[i].sh_addr, 172 sechdrs[i].sh_size 173 / sizeof(Elf64_Rela)); 174 } 175 } 176 177 #ifdef CONFIG_DYNAMIC_FTRACE 178 /* make the trampoline to the ftrace_caller */ 179 relocs++; 180 #endif 181 182 DEBUGP("Looks like a total of %lu stubs, max\n", relocs); 183 return relocs * sizeof(struct ppc64_stub_entry); 184 } 185 186 static void dedotify_versions(struct modversion_info *vers, 187 unsigned long size) 188 { 189 struct modversion_info *end; 190 191 for (end = (void *)vers + size; vers < end; vers++) 192 if (vers->name[0] == '.') 193 memmove(vers->name, vers->name+1, strlen(vers->name)); 194 } 195 196 /* Undefined symbols which refer to .funcname, hack to funcname */ 197 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) 198 { 199 unsigned int i; 200 201 for (i = 1; i < numsyms; i++) { 202 if (syms[i].st_shndx == SHN_UNDEF) { 203 char *name = strtab + syms[i].st_name; 204 if (name[0] == '.') 205 memmove(name, name+1, strlen(name)); 206 } 207 } 208 } 209 210 int module_frob_arch_sections(Elf64_Ehdr *hdr, 211 Elf64_Shdr *sechdrs, 212 char *secstrings, 213 struct module *me) 214 { 215 unsigned int i; 216 217 /* Find .toc and .stubs sections, symtab and strtab */ 218 for (i = 1; i < hdr->e_shnum; i++) { 219 char *p; 220 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) 221 me->arch.stubs_section = i; 222 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) 223 me->arch.toc_section = i; 224 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) 225 dedotify_versions((void *)hdr + sechdrs[i].sh_offset, 226 sechdrs[i].sh_size); 227 228 /* We don't handle .init for the moment: rename to _init */ 229 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init"))) 230 p[0] = '_'; 231 232 if (sechdrs[i].sh_type == SHT_SYMTAB) 233 dedotify((void *)hdr + sechdrs[i].sh_offset, 234 sechdrs[i].sh_size / sizeof(Elf64_Sym), 235 (void *)hdr 236 + sechdrs[sechdrs[i].sh_link].sh_offset); 237 } 238 239 if (!me->arch.stubs_section) { 240 printk("%s: doesn't contain .stubs.\n", me->name); 241 return -ENOEXEC; 242 } 243 244 /* If we don't have a .toc, just use .stubs. We need to set r2 245 to some reasonable value in case the module calls out to 246 other functions via a stub, or if a function pointer escapes 247 the module by some means. */ 248 if (!me->arch.toc_section) 249 me->arch.toc_section = me->arch.stubs_section; 250 251 /* Override the stubs size */ 252 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); 253 return 0; 254 } 255 256 /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this 257 gives the value maximum span in an instruction which uses a signed 258 offset) */ 259 static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me) 260 { 261 return sechdrs[me->arch.toc_section].sh_addr + 0x8000; 262 } 263 264 /* Both low and high 16 bits are added as SIGNED additions, so if low 265 16 bits has high bit set, high 16 bits must be adjusted. These 266 macros do that (stolen from binutils). */ 267 #define PPC_LO(v) ((v) & 0xffff) 268 #define PPC_HI(v) (((v) >> 16) & 0xffff) 269 #define PPC_HA(v) PPC_HI ((v) + 0x8000) 270 271 /* Patch stub to reference function and correct r2 value. */ 272 static inline int create_stub(Elf64_Shdr *sechdrs, 273 struct ppc64_stub_entry *entry, 274 struct ppc64_opd_entry *opd, 275 struct module *me) 276 { 277 Elf64_Half *loc1, *loc2; 278 long reladdr; 279 280 *entry = ppc64_stub; 281 282 #ifdef __LITTLE_ENDIAN__ 283 loc1 = (Elf64_Half *)&entry->jump[0]; 284 loc2 = (Elf64_Half *)&entry->jump[4]; 285 #else 286 loc1 = (Elf64_Half *)&entry->jump[2]; 287 loc2 = (Elf64_Half *)&entry->jump[6]; 288 #endif 289 290 /* Stub uses address relative to r2. */ 291 reladdr = (unsigned long)entry - my_r2(sechdrs, me); 292 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 293 printk("%s: Address %p of stub out of range of %p.\n", 294 me->name, (void *)reladdr, (void *)my_r2); 295 return 0; 296 } 297 DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr); 298 299 *loc1 = PPC_HA(reladdr); 300 *loc2 = PPC_LO(reladdr); 301 entry->opd.funcaddr = opd->funcaddr; 302 entry->opd.r2 = opd->r2; 303 return 1; 304 } 305 306 /* Create stub to jump to function described in this OPD: we need the 307 stub to set up the TOC ptr (r2) for the function. */ 308 static unsigned long stub_for_addr(Elf64_Shdr *sechdrs, 309 unsigned long opdaddr, 310 struct module *me) 311 { 312 struct ppc64_stub_entry *stubs; 313 struct ppc64_opd_entry *opd = (void *)opdaddr; 314 unsigned int i, num_stubs; 315 316 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); 317 318 /* Find this stub, or if that fails, the next avail. entry */ 319 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; 320 for (i = 0; stubs[i].opd.funcaddr; i++) { 321 BUG_ON(i >= num_stubs); 322 323 if (stubs[i].opd.funcaddr == opd->funcaddr) 324 return (unsigned long)&stubs[i]; 325 } 326 327 if (!create_stub(sechdrs, &stubs[i], opd, me)) 328 return 0; 329 330 return (unsigned long)&stubs[i]; 331 } 332 333 /* We expect a noop next: if it is, replace it with instruction to 334 restore r2. */ 335 static int restore_r2(u32 *instruction, struct module *me) 336 { 337 if (*instruction != PPC_INST_NOP) { 338 printk("%s: Expect noop after relocate, got %08x\n", 339 me->name, *instruction); 340 return 0; 341 } 342 *instruction = 0xe8410028; /* ld r2,40(r1) */ 343 return 1; 344 } 345 346 int apply_relocate_add(Elf64_Shdr *sechdrs, 347 const char *strtab, 348 unsigned int symindex, 349 unsigned int relsec, 350 struct module *me) 351 { 352 unsigned int i; 353 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; 354 Elf64_Sym *sym; 355 unsigned long *location; 356 unsigned long value; 357 358 DEBUGP("Applying ADD relocate section %u to %u\n", relsec, 359 sechdrs[relsec].sh_info); 360 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { 361 /* This is where to make the change */ 362 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr 363 + rela[i].r_offset; 364 /* This is the symbol it is referring to */ 365 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr 366 + ELF64_R_SYM(rela[i].r_info); 367 368 DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n", 369 location, (long)ELF64_R_TYPE(rela[i].r_info), 370 strtab + sym->st_name, (unsigned long)sym->st_value, 371 (long)rela[i].r_addend); 372 373 /* `Everything is relative'. */ 374 value = sym->st_value + rela[i].r_addend; 375 376 switch (ELF64_R_TYPE(rela[i].r_info)) { 377 case R_PPC64_ADDR32: 378 /* Simply set it */ 379 *(u32 *)location = value; 380 break; 381 382 case R_PPC64_ADDR64: 383 /* Simply set it */ 384 *(unsigned long *)location = value; 385 break; 386 387 case R_PPC64_TOC: 388 *(unsigned long *)location = my_r2(sechdrs, me); 389 break; 390 391 case R_PPC64_TOC16: 392 /* Subtract TOC pointer */ 393 value -= my_r2(sechdrs, me); 394 if (value + 0x8000 > 0xffff) { 395 printk("%s: bad TOC16 relocation (%lu)\n", 396 me->name, value); 397 return -ENOEXEC; 398 } 399 *((uint16_t *) location) 400 = (*((uint16_t *) location) & ~0xffff) 401 | (value & 0xffff); 402 break; 403 404 case R_PPC64_TOC16_LO: 405 /* Subtract TOC pointer */ 406 value -= my_r2(sechdrs, me); 407 *((uint16_t *) location) 408 = (*((uint16_t *) location) & ~0xffff) 409 | (value & 0xffff); 410 break; 411 412 case R_PPC64_TOC16_DS: 413 /* Subtract TOC pointer */ 414 value -= my_r2(sechdrs, me); 415 if ((value & 3) != 0 || value + 0x8000 > 0xffff) { 416 printk("%s: bad TOC16_DS relocation (%lu)\n", 417 me->name, value); 418 return -ENOEXEC; 419 } 420 *((uint16_t *) location) 421 = (*((uint16_t *) location) & ~0xfffc) 422 | (value & 0xfffc); 423 break; 424 425 case R_PPC64_TOC16_LO_DS: 426 /* Subtract TOC pointer */ 427 value -= my_r2(sechdrs, me); 428 if ((value & 3) != 0) { 429 printk("%s: bad TOC16_LO_DS relocation (%lu)\n", 430 me->name, value); 431 return -ENOEXEC; 432 } 433 *((uint16_t *) location) 434 = (*((uint16_t *) location) & ~0xfffc) 435 | (value & 0xfffc); 436 break; 437 438 case R_PPC64_TOC16_HA: 439 /* Subtract TOC pointer */ 440 value -= my_r2(sechdrs, me); 441 value = ((value + 0x8000) >> 16); 442 *((uint16_t *) location) 443 = (*((uint16_t *) location) & ~0xffff) 444 | (value & 0xffff); 445 break; 446 447 case R_PPC_REL24: 448 /* FIXME: Handle weak symbols here --RR */ 449 if (sym->st_shndx == SHN_UNDEF) { 450 /* External: go via stub */ 451 value = stub_for_addr(sechdrs, value, me); 452 if (!value) 453 return -ENOENT; 454 if (!restore_r2((u32 *)location + 1, me)) 455 return -ENOEXEC; 456 } 457 458 /* Convert value to relative */ 459 value -= (unsigned long)location; 460 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ 461 printk("%s: REL24 %li out of range!\n", 462 me->name, (long int)value); 463 return -ENOEXEC; 464 } 465 466 /* Only replace bits 2 through 26 */ 467 *(uint32_t *)location 468 = (*(uint32_t *)location & ~0x03fffffc) 469 | (value & 0x03fffffc); 470 break; 471 472 case R_PPC64_REL64: 473 /* 64 bits relative (used by features fixups) */ 474 *location = value - (unsigned long)location; 475 break; 476 477 default: 478 printk("%s: Unknown ADD relocation: %lu\n", 479 me->name, 480 (unsigned long)ELF64_R_TYPE(rela[i].r_info)); 481 return -ENOEXEC; 482 } 483 } 484 485 #ifdef CONFIG_DYNAMIC_FTRACE 486 me->arch.toc = my_r2(sechdrs, me); 487 me->arch.tramp = stub_for_addr(sechdrs, 488 (unsigned long)ftrace_caller, 489 me); 490 #endif 491 492 return 0; 493 } 494