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 19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 20 21 #include <linux/module.h> 22 #include <linux/elf.h> 23 #include <linux/moduleloader.h> 24 #include <linux/err.h> 25 #include <linux/vmalloc.h> 26 #include <linux/ftrace.h> 27 #include <linux/bug.h> 28 #include <linux/uaccess.h> 29 #include <asm/module.h> 30 #include <asm/firmware.h> 31 #include <asm/code-patching.h> 32 #include <linux/sort.h> 33 #include <asm/setup.h> 34 #include <asm/sections.h> 35 36 /* FIXME: We don't do .init separately. To do this, we'd need to have 37 a separate r2 value in the init and core section, and stub between 38 them, too. 39 40 Using a magic allocator which places modules within 32MB solves 41 this, and makes other things simpler. Anton? 42 --RR. */ 43 44 #ifdef PPC64_ELF_ABI_v2 45 46 /* An address is simply the address of the function. */ 47 typedef unsigned long func_desc_t; 48 49 static func_desc_t func_desc(unsigned long addr) 50 { 51 return addr; 52 } 53 static unsigned long func_addr(unsigned long addr) 54 { 55 return addr; 56 } 57 static unsigned long stub_func_addr(func_desc_t func) 58 { 59 return func; 60 } 61 62 /* PowerPC64 specific values for the Elf64_Sym st_other field. */ 63 #define STO_PPC64_LOCAL_BIT 5 64 #define STO_PPC64_LOCAL_MASK (7 << STO_PPC64_LOCAL_BIT) 65 #define PPC64_LOCAL_ENTRY_OFFSET(other) \ 66 (((1 << (((other) & STO_PPC64_LOCAL_MASK) >> STO_PPC64_LOCAL_BIT)) >> 2) << 2) 67 68 static unsigned int local_entry_offset(const Elf64_Sym *sym) 69 { 70 /* sym->st_other indicates offset to local entry point 71 * (otherwise it will assume r12 is the address of the start 72 * of function and try to derive r2 from it). */ 73 return PPC64_LOCAL_ENTRY_OFFSET(sym->st_other); 74 } 75 #else 76 77 /* An address is address of the OPD entry, which contains address of fn. */ 78 typedef struct ppc64_opd_entry func_desc_t; 79 80 static func_desc_t func_desc(unsigned long addr) 81 { 82 return *(struct ppc64_opd_entry *)addr; 83 } 84 static unsigned long func_addr(unsigned long addr) 85 { 86 return func_desc(addr).funcaddr; 87 } 88 static unsigned long stub_func_addr(func_desc_t func) 89 { 90 return func.funcaddr; 91 } 92 static unsigned int local_entry_offset(const Elf64_Sym *sym) 93 { 94 return 0; 95 } 96 97 void *dereference_module_function_descriptor(struct module *mod, void *ptr) 98 { 99 if (ptr < (void *)mod->arch.start_opd || 100 ptr >= (void *)mod->arch.end_opd) 101 return ptr; 102 103 return dereference_function_descriptor(ptr); 104 } 105 #endif 106 107 #define STUB_MAGIC 0x73747562 /* stub */ 108 109 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into 110 the kernel itself). But on PPC64, these need to be used for every 111 jump, actually, to reset r2 (TOC+0x8000). */ 112 struct ppc64_stub_entry 113 { 114 /* 28 byte jump instruction sequence (7 instructions). We only 115 * need 6 instructions on ABIv2 but we always allocate 7 so 116 * so we don't have to modify the trampoline load instruction. */ 117 u32 jump[7]; 118 /* Used by ftrace to identify stubs */ 119 u32 magic; 120 /* Data for the above code */ 121 func_desc_t funcdata; 122 }; 123 124 /* 125 * PPC64 uses 24 bit jumps, but we need to jump into other modules or 126 * the kernel which may be further. So we jump to a stub. 127 * 128 * For ELFv1 we need to use this to set up the new r2 value (aka TOC 129 * pointer). For ELFv2 it's the callee's responsibility to set up the 130 * new r2, but for both we need to save the old r2. 131 * 132 * We could simply patch the new r2 value and function pointer into 133 * the stub, but it's significantly shorter to put these values at the 134 * end of the stub code, and patch the stub address (32-bits relative 135 * to the TOC ptr, r2) into the stub. 136 */ 137 138 static u32 ppc64_stub_insns[] = { 139 0x3d620000, /* addis r11,r2, <high> */ 140 0x396b0000, /* addi r11,r11, <low> */ 141 /* Save current r2 value in magic place on the stack. */ 142 0xf8410000|R2_STACK_OFFSET, /* std r2,R2_STACK_OFFSET(r1) */ 143 0xe98b0020, /* ld r12,32(r11) */ 144 #ifdef PPC64_ELF_ABI_v1 145 /* Set up new r2 from function descriptor */ 146 0xe84b0028, /* ld r2,40(r11) */ 147 #endif 148 0x7d8903a6, /* mtctr r12 */ 149 0x4e800420 /* bctr */ 150 }; 151 152 #ifdef CONFIG_DYNAMIC_FTRACE 153 int module_trampoline_target(struct module *mod, unsigned long addr, 154 unsigned long *target) 155 { 156 struct ppc64_stub_entry *stub; 157 func_desc_t funcdata; 158 u32 magic; 159 160 if (!within_module_core(addr, mod)) { 161 pr_err("%s: stub %lx not in module %s\n", __func__, addr, mod->name); 162 return -EFAULT; 163 } 164 165 stub = (struct ppc64_stub_entry *)addr; 166 167 if (probe_kernel_read(&magic, &stub->magic, sizeof(magic))) { 168 pr_err("%s: fault reading magic for stub %lx for %s\n", __func__, addr, mod->name); 169 return -EFAULT; 170 } 171 172 if (magic != STUB_MAGIC) { 173 pr_err("%s: bad magic for stub %lx for %s\n", __func__, addr, mod->name); 174 return -EFAULT; 175 } 176 177 if (probe_kernel_read(&funcdata, &stub->funcdata, sizeof(funcdata))) { 178 pr_err("%s: fault reading funcdata for stub %lx for %s\n", __func__, addr, mod->name); 179 return -EFAULT; 180 } 181 182 *target = stub_func_addr(funcdata); 183 184 return 0; 185 } 186 #endif 187 188 /* Count how many different 24-bit relocations (different symbol, 189 different addend) */ 190 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num) 191 { 192 unsigned int i, r_info, r_addend, _count_relocs; 193 194 /* FIXME: Only count external ones --RR */ 195 _count_relocs = 0; 196 r_info = 0; 197 r_addend = 0; 198 for (i = 0; i < num; i++) 199 /* Only count 24-bit relocs, others don't need stubs */ 200 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 && 201 (r_info != ELF64_R_SYM(rela[i].r_info) || 202 r_addend != rela[i].r_addend)) { 203 _count_relocs++; 204 r_info = ELF64_R_SYM(rela[i].r_info); 205 r_addend = rela[i].r_addend; 206 } 207 208 return _count_relocs; 209 } 210 211 static int relacmp(const void *_x, const void *_y) 212 { 213 const Elf64_Rela *x, *y; 214 215 y = (Elf64_Rela *)_x; 216 x = (Elf64_Rela *)_y; 217 218 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to 219 * make the comparison cheaper/faster. It won't affect the sorting or 220 * the counting algorithms' performance 221 */ 222 if (x->r_info < y->r_info) 223 return -1; 224 else if (x->r_info > y->r_info) 225 return 1; 226 else if (x->r_addend < y->r_addend) 227 return -1; 228 else if (x->r_addend > y->r_addend) 229 return 1; 230 else 231 return 0; 232 } 233 234 static void relaswap(void *_x, void *_y, int size) 235 { 236 uint64_t *x, *y, tmp; 237 int i; 238 239 y = (uint64_t *)_x; 240 x = (uint64_t *)_y; 241 242 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) { 243 tmp = x[i]; 244 x[i] = y[i]; 245 y[i] = tmp; 246 } 247 } 248 249 /* Get size of potential trampolines required. */ 250 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, 251 const Elf64_Shdr *sechdrs) 252 { 253 /* One extra reloc so it's always 0-funcaddr terminated */ 254 unsigned long relocs = 1; 255 unsigned i; 256 257 /* Every relocated section... */ 258 for (i = 1; i < hdr->e_shnum; i++) { 259 if (sechdrs[i].sh_type == SHT_RELA) { 260 pr_debug("Found relocations in section %u\n", i); 261 pr_debug("Ptr: %p. Number: %Lu\n", 262 (void *)sechdrs[i].sh_addr, 263 sechdrs[i].sh_size / sizeof(Elf64_Rela)); 264 265 /* Sort the relocation information based on a symbol and 266 * addend key. This is a stable O(n*log n) complexity 267 * alogrithm but it will reduce the complexity of 268 * count_relocs() to linear complexity O(n) 269 */ 270 sort((void *)sechdrs[i].sh_addr, 271 sechdrs[i].sh_size / sizeof(Elf64_Rela), 272 sizeof(Elf64_Rela), relacmp, relaswap); 273 274 relocs += count_relocs((void *)sechdrs[i].sh_addr, 275 sechdrs[i].sh_size 276 / sizeof(Elf64_Rela)); 277 } 278 } 279 280 #ifdef CONFIG_DYNAMIC_FTRACE 281 /* make the trampoline to the ftrace_caller */ 282 relocs++; 283 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS 284 /* an additional one for ftrace_regs_caller */ 285 relocs++; 286 #endif 287 #endif 288 289 pr_debug("Looks like a total of %lu stubs, max\n", relocs); 290 return relocs * sizeof(struct ppc64_stub_entry); 291 } 292 293 /* Still needed for ELFv2, for .TOC. */ 294 static void dedotify_versions(struct modversion_info *vers, 295 unsigned long size) 296 { 297 struct modversion_info *end; 298 299 for (end = (void *)vers + size; vers < end; vers++) 300 if (vers->name[0] == '.') { 301 memmove(vers->name, vers->name+1, strlen(vers->name)); 302 } 303 } 304 305 /* 306 * Undefined symbols which refer to .funcname, hack to funcname. Make .TOC. 307 * seem to be defined (value set later). 308 */ 309 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) 310 { 311 unsigned int i; 312 313 for (i = 1; i < numsyms; i++) { 314 if (syms[i].st_shndx == SHN_UNDEF) { 315 char *name = strtab + syms[i].st_name; 316 if (name[0] == '.') { 317 if (strcmp(name+1, "TOC.") == 0) 318 syms[i].st_shndx = SHN_ABS; 319 syms[i].st_name++; 320 } 321 } 322 } 323 } 324 325 static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs, 326 const char *strtab, 327 unsigned int symindex) 328 { 329 unsigned int i, numsyms; 330 Elf64_Sym *syms; 331 332 syms = (Elf64_Sym *)sechdrs[symindex].sh_addr; 333 numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym); 334 335 for (i = 1; i < numsyms; i++) { 336 if (syms[i].st_shndx == SHN_ABS 337 && strcmp(strtab + syms[i].st_name, "TOC.") == 0) 338 return &syms[i]; 339 } 340 return NULL; 341 } 342 343 int module_frob_arch_sections(Elf64_Ehdr *hdr, 344 Elf64_Shdr *sechdrs, 345 char *secstrings, 346 struct module *me) 347 { 348 unsigned int i; 349 350 /* Find .toc and .stubs sections, symtab and strtab */ 351 for (i = 1; i < hdr->e_shnum; i++) { 352 char *p; 353 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) 354 me->arch.stubs_section = i; 355 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) { 356 me->arch.toc_section = i; 357 if (sechdrs[i].sh_addralign < 8) 358 sechdrs[i].sh_addralign = 8; 359 } 360 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) 361 dedotify_versions((void *)hdr + sechdrs[i].sh_offset, 362 sechdrs[i].sh_size); 363 364 /* We don't handle .init for the moment: rename to _init */ 365 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init"))) 366 p[0] = '_'; 367 368 if (sechdrs[i].sh_type == SHT_SYMTAB) 369 dedotify((void *)hdr + sechdrs[i].sh_offset, 370 sechdrs[i].sh_size / sizeof(Elf64_Sym), 371 (void *)hdr 372 + sechdrs[sechdrs[i].sh_link].sh_offset); 373 } 374 375 if (!me->arch.stubs_section) { 376 pr_err("%s: doesn't contain .stubs.\n", me->name); 377 return -ENOEXEC; 378 } 379 380 /* If we don't have a .toc, just use .stubs. We need to set r2 381 to some reasonable value in case the module calls out to 382 other functions via a stub, or if a function pointer escapes 383 the module by some means. */ 384 if (!me->arch.toc_section) 385 me->arch.toc_section = me->arch.stubs_section; 386 387 /* Override the stubs size */ 388 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); 389 return 0; 390 } 391 392 /* 393 * r2 is the TOC pointer: it actually points 0x8000 into the TOC (this gives the 394 * value maximum span in an instruction which uses a signed offset). Round down 395 * to a 256 byte boundary for the odd case where we are setting up r2 without a 396 * .toc section. 397 */ 398 static inline unsigned long my_r2(const Elf64_Shdr *sechdrs, struct module *me) 399 { 400 return (sechdrs[me->arch.toc_section].sh_addr & ~0xfful) + 0x8000; 401 } 402 403 /* Both low and high 16 bits are added as SIGNED additions, so if low 404 16 bits has high bit set, high 16 bits must be adjusted. These 405 macros do that (stolen from binutils). */ 406 #define PPC_LO(v) ((v) & 0xffff) 407 #define PPC_HI(v) (((v) >> 16) & 0xffff) 408 #define PPC_HA(v) PPC_HI ((v) + 0x8000) 409 410 /* Patch stub to reference function and correct r2 value. */ 411 static inline int create_stub(const Elf64_Shdr *sechdrs, 412 struct ppc64_stub_entry *entry, 413 unsigned long addr, 414 struct module *me) 415 { 416 long reladdr; 417 418 memcpy(entry->jump, ppc64_stub_insns, sizeof(ppc64_stub_insns)); 419 420 /* Stub uses address relative to r2. */ 421 reladdr = (unsigned long)entry - my_r2(sechdrs, me); 422 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 423 pr_err("%s: Address %p of stub out of range of %p.\n", 424 me->name, (void *)reladdr, (void *)my_r2); 425 return 0; 426 } 427 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr); 428 429 entry->jump[0] |= PPC_HA(reladdr); 430 entry->jump[1] |= PPC_LO(reladdr); 431 entry->funcdata = func_desc(addr); 432 entry->magic = STUB_MAGIC; 433 434 return 1; 435 } 436 437 /* Create stub to jump to function described in this OPD/ptr: we need the 438 stub to set up the TOC ptr (r2) for the function. */ 439 static unsigned long stub_for_addr(const Elf64_Shdr *sechdrs, 440 unsigned long addr, 441 struct module *me) 442 { 443 struct ppc64_stub_entry *stubs; 444 unsigned int i, num_stubs; 445 446 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); 447 448 /* Find this stub, or if that fails, the next avail. entry */ 449 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; 450 for (i = 0; stub_func_addr(stubs[i].funcdata); i++) { 451 if (WARN_ON(i >= num_stubs)) 452 return 0; 453 454 if (stub_func_addr(stubs[i].funcdata) == func_addr(addr)) 455 return (unsigned long)&stubs[i]; 456 } 457 458 if (!create_stub(sechdrs, &stubs[i], addr, me)) 459 return 0; 460 461 return (unsigned long)&stubs[i]; 462 } 463 464 #ifdef CONFIG_MPROFILE_KERNEL 465 static bool is_mprofile_mcount_callsite(const char *name, u32 *instruction) 466 { 467 if (strcmp("_mcount", name)) 468 return false; 469 470 /* 471 * Check if this is one of the -mprofile-kernel sequences. 472 */ 473 if (instruction[-1] == PPC_INST_STD_LR && 474 instruction[-2] == PPC_INST_MFLR) 475 return true; 476 477 if (instruction[-1] == PPC_INST_MFLR) 478 return true; 479 480 return false; 481 } 482 483 /* 484 * In case of _mcount calls, do not save the current callee's TOC (in r2) into 485 * the original caller's stack frame. If we did we would clobber the saved TOC 486 * value of the original caller. 487 */ 488 static void squash_toc_save_inst(const char *name, unsigned long addr) 489 { 490 struct ppc64_stub_entry *stub = (struct ppc64_stub_entry *)addr; 491 492 /* Only for calls to _mcount */ 493 if (strcmp("_mcount", name) != 0) 494 return; 495 496 stub->jump[2] = PPC_INST_NOP; 497 } 498 #else 499 static void squash_toc_save_inst(const char *name, unsigned long addr) { } 500 501 static bool is_mprofile_mcount_callsite(const char *name, u32 *instruction) 502 { 503 return false; 504 } 505 #endif 506 507 /* We expect a noop next: if it is, replace it with instruction to 508 restore r2. */ 509 static int restore_r2(const char *name, u32 *instruction, struct module *me) 510 { 511 u32 *prev_insn = instruction - 1; 512 513 if (is_mprofile_mcount_callsite(name, prev_insn)) 514 return 1; 515 516 /* 517 * Make sure the branch isn't a sibling call. Sibling calls aren't 518 * "link" branches and they don't return, so they don't need the r2 519 * restore afterwards. 520 */ 521 if (!instr_is_relative_link_branch(*prev_insn)) 522 return 1; 523 524 if (*instruction != PPC_INST_NOP) { 525 pr_err("%s: Expected nop after call, got %08x at %pS\n", 526 me->name, *instruction, instruction); 527 return 0; 528 } 529 /* ld r2,R2_STACK_OFFSET(r1) */ 530 *instruction = PPC_INST_LD_TOC; 531 return 1; 532 } 533 534 int apply_relocate_add(Elf64_Shdr *sechdrs, 535 const char *strtab, 536 unsigned int symindex, 537 unsigned int relsec, 538 struct module *me) 539 { 540 unsigned int i; 541 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; 542 Elf64_Sym *sym; 543 unsigned long *location; 544 unsigned long value; 545 546 pr_debug("Applying ADD relocate section %u to %u\n", relsec, 547 sechdrs[relsec].sh_info); 548 549 /* First time we're called, we can fix up .TOC. */ 550 if (!me->arch.toc_fixed) { 551 sym = find_dot_toc(sechdrs, strtab, symindex); 552 /* It's theoretically possible that a module doesn't want a 553 * .TOC. so don't fail it just for that. */ 554 if (sym) 555 sym->st_value = my_r2(sechdrs, me); 556 me->arch.toc_fixed = true; 557 } 558 559 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { 560 /* This is where to make the change */ 561 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr 562 + rela[i].r_offset; 563 /* This is the symbol it is referring to */ 564 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr 565 + ELF64_R_SYM(rela[i].r_info); 566 567 pr_debug("RELOC at %p: %li-type as %s (0x%lx) + %li\n", 568 location, (long)ELF64_R_TYPE(rela[i].r_info), 569 strtab + sym->st_name, (unsigned long)sym->st_value, 570 (long)rela[i].r_addend); 571 572 /* `Everything is relative'. */ 573 value = sym->st_value + rela[i].r_addend; 574 575 switch (ELF64_R_TYPE(rela[i].r_info)) { 576 case R_PPC64_ADDR32: 577 /* Simply set it */ 578 *(u32 *)location = value; 579 break; 580 581 case R_PPC64_ADDR64: 582 /* Simply set it */ 583 *(unsigned long *)location = value; 584 break; 585 586 case R_PPC64_TOC: 587 *(unsigned long *)location = my_r2(sechdrs, me); 588 break; 589 590 case R_PPC64_TOC16: 591 /* Subtract TOC pointer */ 592 value -= my_r2(sechdrs, me); 593 if (value + 0x8000 > 0xffff) { 594 pr_err("%s: bad TOC16 relocation (0x%lx)\n", 595 me->name, value); 596 return -ENOEXEC; 597 } 598 *((uint16_t *) location) 599 = (*((uint16_t *) location) & ~0xffff) 600 | (value & 0xffff); 601 break; 602 603 case R_PPC64_TOC16_LO: 604 /* Subtract TOC pointer */ 605 value -= my_r2(sechdrs, me); 606 *((uint16_t *) location) 607 = (*((uint16_t *) location) & ~0xffff) 608 | (value & 0xffff); 609 break; 610 611 case R_PPC64_TOC16_DS: 612 /* Subtract TOC pointer */ 613 value -= my_r2(sechdrs, me); 614 if ((value & 3) != 0 || value + 0x8000 > 0xffff) { 615 pr_err("%s: bad TOC16_DS relocation (0x%lx)\n", 616 me->name, value); 617 return -ENOEXEC; 618 } 619 *((uint16_t *) location) 620 = (*((uint16_t *) location) & ~0xfffc) 621 | (value & 0xfffc); 622 break; 623 624 case R_PPC64_TOC16_LO_DS: 625 /* Subtract TOC pointer */ 626 value -= my_r2(sechdrs, me); 627 if ((value & 3) != 0) { 628 pr_err("%s: bad TOC16_LO_DS relocation (0x%lx)\n", 629 me->name, value); 630 return -ENOEXEC; 631 } 632 *((uint16_t *) location) 633 = (*((uint16_t *) location) & ~0xfffc) 634 | (value & 0xfffc); 635 break; 636 637 case R_PPC64_TOC16_HA: 638 /* Subtract TOC pointer */ 639 value -= my_r2(sechdrs, me); 640 value = ((value + 0x8000) >> 16); 641 *((uint16_t *) location) 642 = (*((uint16_t *) location) & ~0xffff) 643 | (value & 0xffff); 644 break; 645 646 case R_PPC_REL24: 647 /* FIXME: Handle weak symbols here --RR */ 648 if (sym->st_shndx == SHN_UNDEF || 649 sym->st_shndx == SHN_LIVEPATCH) { 650 /* External: go via stub */ 651 value = stub_for_addr(sechdrs, value, me); 652 if (!value) 653 return -ENOENT; 654 if (!restore_r2(strtab + sym->st_name, 655 (u32 *)location + 1, me)) 656 return -ENOEXEC; 657 658 squash_toc_save_inst(strtab + sym->st_name, value); 659 } else 660 value += local_entry_offset(sym); 661 662 /* Convert value to relative */ 663 value -= (unsigned long)location; 664 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ 665 pr_err("%s: REL24 %li out of range!\n", 666 me->name, (long int)value); 667 return -ENOEXEC; 668 } 669 670 /* Only replace bits 2 through 26 */ 671 *(uint32_t *)location 672 = (*(uint32_t *)location & ~0x03fffffc) 673 | (value & 0x03fffffc); 674 break; 675 676 case R_PPC64_REL64: 677 /* 64 bits relative (used by features fixups) */ 678 *location = value - (unsigned long)location; 679 break; 680 681 case R_PPC64_REL32: 682 /* 32 bits relative (used by relative exception tables) */ 683 /* Convert value to relative */ 684 value -= (unsigned long)location; 685 if (value + 0x80000000 > 0xffffffff) { 686 pr_err("%s: REL32 %li out of range!\n", 687 me->name, (long int)value); 688 return -ENOEXEC; 689 } 690 *(u32 *)location = value; 691 break; 692 693 case R_PPC64_TOCSAVE: 694 /* 695 * Marker reloc indicates we don't have to save r2. 696 * That would only save us one instruction, so ignore 697 * it. 698 */ 699 break; 700 701 case R_PPC64_ENTRY: 702 /* 703 * Optimize ELFv2 large code model entry point if 704 * the TOC is within 2GB range of current location. 705 */ 706 value = my_r2(sechdrs, me) - (unsigned long)location; 707 if (value + 0x80008000 > 0xffffffff) 708 break; 709 /* 710 * Check for the large code model prolog sequence: 711 * ld r2, ...(r12) 712 * add r2, r2, r12 713 */ 714 if ((((uint32_t *)location)[0] & ~0xfffc) 715 != 0xe84c0000) 716 break; 717 if (((uint32_t *)location)[1] != 0x7c426214) 718 break; 719 /* 720 * If found, replace it with: 721 * addis r2, r12, (.TOC.-func)@ha 722 * addi r2, r12, (.TOC.-func)@l 723 */ 724 ((uint32_t *)location)[0] = 0x3c4c0000 + PPC_HA(value); 725 ((uint32_t *)location)[1] = 0x38420000 + PPC_LO(value); 726 break; 727 728 case R_PPC64_REL16_HA: 729 /* Subtract location pointer */ 730 value -= (unsigned long)location; 731 value = ((value + 0x8000) >> 16); 732 *((uint16_t *) location) 733 = (*((uint16_t *) location) & ~0xffff) 734 | (value & 0xffff); 735 break; 736 737 case R_PPC64_REL16_LO: 738 /* Subtract location pointer */ 739 value -= (unsigned long)location; 740 *((uint16_t *) location) 741 = (*((uint16_t *) location) & ~0xffff) 742 | (value & 0xffff); 743 break; 744 745 default: 746 pr_err("%s: Unknown ADD relocation: %lu\n", 747 me->name, 748 (unsigned long)ELF64_R_TYPE(rela[i].r_info)); 749 return -ENOEXEC; 750 } 751 } 752 753 return 0; 754 } 755 756 #ifdef CONFIG_DYNAMIC_FTRACE 757 758 #ifdef CONFIG_MPROFILE_KERNEL 759 760 #define PACATOC offsetof(struct paca_struct, kernel_toc) 761 762 /* 763 * For mprofile-kernel we use a special stub for ftrace_caller() because we 764 * can't rely on r2 containing this module's TOC when we enter the stub. 765 * 766 * That can happen if the function calling us didn't need to use the toc. In 767 * that case it won't have setup r2, and the r2 value will be either the 768 * kernel's toc, or possibly another modules toc. 769 * 770 * To deal with that this stub uses the kernel toc, which is always accessible 771 * via the paca (in r13). The target (ftrace_caller()) is responsible for 772 * saving and restoring the toc before returning. 773 */ 774 static unsigned long create_ftrace_stub(const Elf64_Shdr *sechdrs, 775 struct module *me, unsigned long addr) 776 { 777 struct ppc64_stub_entry *entry; 778 unsigned int i, num_stubs; 779 static u32 stub_insns[] = { 780 0xe98d0000 | PACATOC, /* ld r12,PACATOC(r13) */ 781 0x3d8c0000, /* addis r12,r12,<high> */ 782 0x398c0000, /* addi r12,r12,<low> */ 783 0x7d8903a6, /* mtctr r12 */ 784 0x4e800420, /* bctr */ 785 }; 786 long reladdr; 787 788 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*entry); 789 790 /* Find the next available stub entry */ 791 entry = (void *)sechdrs[me->arch.stubs_section].sh_addr; 792 for (i = 0; i < num_stubs && stub_func_addr(entry->funcdata); i++, entry++); 793 794 if (i >= num_stubs) { 795 pr_err("%s: Unable to find a free slot for ftrace stub.\n", me->name); 796 return 0; 797 } 798 799 memcpy(entry->jump, stub_insns, sizeof(stub_insns)); 800 801 /* Stub uses address relative to kernel toc (from the paca) */ 802 reladdr = addr - kernel_toc_addr(); 803 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 804 pr_err("%s: Address of %ps out of range of kernel_toc.\n", 805 me->name, (void *)addr); 806 return 0; 807 } 808 809 entry->jump[1] |= PPC_HA(reladdr); 810 entry->jump[2] |= PPC_LO(reladdr); 811 812 /* Eventhough we don't use funcdata in the stub, it's needed elsewhere. */ 813 entry->funcdata = func_desc(addr); 814 entry->magic = STUB_MAGIC; 815 816 return (unsigned long)entry; 817 } 818 #else 819 static unsigned long create_ftrace_stub(const Elf64_Shdr *sechdrs, 820 struct module *me, unsigned long addr) 821 { 822 return stub_for_addr(sechdrs, addr, me); 823 } 824 #endif 825 826 int module_finalize_ftrace(struct module *mod, const Elf_Shdr *sechdrs) 827 { 828 mod->arch.tramp = create_ftrace_stub(sechdrs, mod, 829 (unsigned long)ftrace_caller); 830 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS 831 mod->arch.tramp_regs = create_ftrace_stub(sechdrs, mod, 832 (unsigned long)ftrace_regs_caller); 833 if (!mod->arch.tramp_regs) 834 return -ENOENT; 835 #endif 836 837 if (!mod->arch.tramp) 838 return -ENOENT; 839 840 return 0; 841 } 842 #endif 843