1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* Kernel module help for PPC64. 3 Copyright (C) 2001, 2003 Rusty Russell IBM Corporation. 4 5 */ 6 7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 8 9 #include <linux/module.h> 10 #include <linux/elf.h> 11 #include <linux/moduleloader.h> 12 #include <linux/err.h> 13 #include <linux/vmalloc.h> 14 #include <linux/ftrace.h> 15 #include <linux/bug.h> 16 #include <linux/uaccess.h> 17 #include <linux/kernel.h> 18 #include <asm/module.h> 19 #include <asm/firmware.h> 20 #include <asm/text-patching.h> 21 #include <linux/sort.h> 22 #include <asm/setup.h> 23 #include <asm/sections.h> 24 #include <asm/inst.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 34 bool module_elf_check_arch(Elf_Ehdr *hdr) 35 { 36 unsigned long abi_level = hdr->e_flags & 0x3; 37 38 if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2)) 39 return abi_level == 2; 40 else 41 return abi_level < 2; 42 } 43 44 #ifdef CONFIG_PPC64_ELF_ABI_V2 45 46 static func_desc_t func_desc(unsigned long addr) 47 { 48 func_desc_t desc = { 49 .addr = addr, 50 }; 51 52 return desc; 53 } 54 55 /* PowerPC64 specific values for the Elf64_Sym st_other field. */ 56 #define STO_PPC64_LOCAL_BIT 5 57 #define STO_PPC64_LOCAL_MASK (7 << STO_PPC64_LOCAL_BIT) 58 #define PPC64_LOCAL_ENTRY_OFFSET(other) \ 59 (((1 << (((other) & STO_PPC64_LOCAL_MASK) >> STO_PPC64_LOCAL_BIT)) >> 2) << 2) 60 61 static unsigned int local_entry_offset(const Elf64_Sym *sym) 62 { 63 /* sym->st_other indicates offset to local entry point 64 * (otherwise it will assume r12 is the address of the start 65 * of function and try to derive r2 from it). */ 66 return PPC64_LOCAL_ENTRY_OFFSET(sym->st_other); 67 } 68 #else 69 70 static func_desc_t func_desc(unsigned long addr) 71 { 72 return *(struct func_desc *)addr; 73 } 74 static unsigned int local_entry_offset(const Elf64_Sym *sym) 75 { 76 return 0; 77 } 78 79 void *dereference_module_function_descriptor(struct module *mod, void *ptr) 80 { 81 if (ptr < (void *)mod->arch.start_opd || 82 ptr >= (void *)mod->arch.end_opd) 83 return ptr; 84 85 return dereference_function_descriptor(ptr); 86 } 87 #endif 88 89 static unsigned long func_addr(unsigned long addr) 90 { 91 return func_desc(addr).addr; 92 } 93 94 static unsigned long stub_func_addr(func_desc_t func) 95 { 96 return func.addr; 97 } 98 99 #define STUB_MAGIC 0x73747562 /* stub */ 100 101 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into 102 the kernel itself). But on PPC64, these need to be used for every 103 jump, actually, to reset r2 (TOC+0x8000). */ 104 struct ppc64_stub_entry { 105 /* 106 * 28 byte jump instruction sequence (7 instructions) that can 107 * hold ppc64_stub_insns or stub_insns. Must be 8-byte aligned 108 * with PCREL kernels that use prefix instructions in the stub. 109 */ 110 u32 jump[7]; 111 /* Used by ftrace to identify stubs */ 112 u32 magic; 113 /* Data for the above code */ 114 func_desc_t funcdata; 115 } __aligned(8); 116 117 struct ppc64_got_entry { 118 u64 addr; 119 }; 120 121 /* 122 * PPC64 uses 24 bit jumps, but we need to jump into other modules or 123 * the kernel which may be further. So we jump to a stub. 124 * 125 * Target address and TOC are loaded from function descriptor in the 126 * ppc64_stub_entry. 127 * 128 * r12 is used to generate the target address, which is required for the 129 * ELFv2 global entry point calling convention. 130 * 131 * TOC handling: 132 * - PCREL does not have a TOC. 133 * - ELFv2 non-PCREL just has to save r2, the callee is responsible for 134 * setting its own TOC pointer at the global entry address. 135 * - ELFv1 must load the new TOC pointer from the function descriptor. 136 */ 137 static u32 ppc64_stub_insns[] = { 138 #ifdef CONFIG_PPC_KERNEL_PCREL 139 /* pld r12,addr */ 140 PPC_PREFIX_8LS | __PPC_PRFX_R(1), 141 PPC_INST_PLD | ___PPC_RT(_R12), 142 #else 143 PPC_RAW_ADDIS(_R11, _R2, 0), 144 PPC_RAW_ADDI(_R11, _R11, 0), 145 /* Save current r2 value in magic place on the stack. */ 146 PPC_RAW_STD(_R2, _R1, R2_STACK_OFFSET), 147 PPC_RAW_LD(_R12, _R11, 32), 148 #ifdef CONFIG_PPC64_ELF_ABI_V1 149 /* Set up new r2 from function descriptor */ 150 PPC_RAW_LD(_R2, _R11, 40), 151 #endif 152 #endif 153 PPC_RAW_MTCTR(_R12), 154 PPC_RAW_BCTR(), 155 }; 156 157 /* 158 * Count how many different r_type relocations (different symbol, 159 * different addend). 160 */ 161 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num, 162 unsigned long r_type) 163 { 164 unsigned int i, r_info, r_addend, _count_relocs; 165 166 /* FIXME: Only count external ones --RR */ 167 _count_relocs = 0; 168 r_info = 0; 169 r_addend = 0; 170 for (i = 0; i < num; i++) 171 /* Only count r_type relocs, others don't need stubs */ 172 if (ELF64_R_TYPE(rela[i].r_info) == r_type && 173 (r_info != ELF64_R_SYM(rela[i].r_info) || 174 r_addend != rela[i].r_addend)) { 175 _count_relocs++; 176 r_info = ELF64_R_SYM(rela[i].r_info); 177 r_addend = rela[i].r_addend; 178 } 179 180 return _count_relocs; 181 } 182 183 static int relacmp(const void *_x, const void *_y) 184 { 185 const Elf64_Rela *x, *y; 186 187 y = (Elf64_Rela *)_x; 188 x = (Elf64_Rela *)_y; 189 190 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to 191 * make the comparison cheaper/faster. It won't affect the sorting or 192 * the counting algorithms' performance 193 */ 194 if (x->r_info < y->r_info) 195 return -1; 196 else if (x->r_info > y->r_info) 197 return 1; 198 else if (x->r_addend < y->r_addend) 199 return -1; 200 else if (x->r_addend > y->r_addend) 201 return 1; 202 else 203 return 0; 204 } 205 206 /* Get size of potential trampolines required. */ 207 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, 208 const Elf64_Shdr *sechdrs, 209 char *secstrings, 210 struct module *me) 211 { 212 /* One extra reloc so it's always 0-addr terminated */ 213 unsigned long relocs = 1; 214 unsigned i; 215 216 /* Every relocated section... */ 217 for (i = 1; i < hdr->e_shnum; i++) { 218 if (sechdrs[i].sh_type == SHT_RELA) { 219 pr_debug("Found relocations in section %u\n", i); 220 pr_debug("Ptr: %p. Number: %Lu\n", 221 (void *)sechdrs[i].sh_addr, 222 sechdrs[i].sh_size / sizeof(Elf64_Rela)); 223 224 /* Sort the relocation information based on a symbol and 225 * addend key. This is a stable O(n*log n) complexity 226 * algorithm but it will reduce the complexity of 227 * count_relocs() to linear complexity O(n) 228 */ 229 sort((void *)sechdrs[i].sh_addr, 230 sechdrs[i].sh_size / sizeof(Elf64_Rela), 231 sizeof(Elf64_Rela), relacmp, NULL); 232 233 relocs += count_relocs((void *)sechdrs[i].sh_addr, 234 sechdrs[i].sh_size 235 / sizeof(Elf64_Rela), 236 R_PPC_REL24); 237 #ifdef CONFIG_PPC_KERNEL_PCREL 238 relocs += count_relocs((void *)sechdrs[i].sh_addr, 239 sechdrs[i].sh_size 240 / sizeof(Elf64_Rela), 241 R_PPC64_REL24_NOTOC); 242 #endif 243 } 244 } 245 246 /* stubs for ftrace_caller and ftrace_regs_caller */ 247 relocs += IS_ENABLED(CONFIG_DYNAMIC_FTRACE) + IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_REGS); 248 249 #ifdef CONFIG_PPC_FTRACE_OUT_OF_LINE 250 /* stubs for the function tracer */ 251 for (i = 1; i < hdr->e_shnum; i++) { 252 if (!strcmp(secstrings + sechdrs[i].sh_name, "__patchable_function_entries")) { 253 me->arch.ool_stub_count = sechdrs[i].sh_size / sizeof(unsigned long); 254 me->arch.ool_stub_index = 0; 255 relocs += roundup(me->arch.ool_stub_count * sizeof(struct ftrace_ool_stub), 256 sizeof(struct ppc64_stub_entry)) / 257 sizeof(struct ppc64_stub_entry); 258 break; 259 } 260 } 261 if (i == hdr->e_shnum) { 262 pr_err("%s: doesn't contain __patchable_function_entries.\n", me->name); 263 return -ENOEXEC; 264 } 265 #endif 266 267 pr_debug("Looks like a total of %lu stubs, max\n", relocs); 268 return relocs * sizeof(struct ppc64_stub_entry); 269 } 270 271 #ifdef CONFIG_PPC_KERNEL_PCREL 272 static int count_pcpu_relocs(const Elf64_Shdr *sechdrs, 273 const Elf64_Rela *rela, unsigned int num, 274 unsigned int symindex, unsigned int pcpu) 275 { 276 unsigned int i, r_info, r_addend, _count_relocs; 277 278 _count_relocs = 0; 279 r_info = 0; 280 r_addend = 0; 281 282 for (i = 0; i < num; i++) { 283 Elf64_Sym *sym; 284 285 /* This is the symbol it is referring to */ 286 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr 287 + ELF64_R_SYM(rela[i].r_info); 288 289 if (sym->st_shndx == pcpu && 290 (r_info != ELF64_R_SYM(rela[i].r_info) || 291 r_addend != rela[i].r_addend)) { 292 _count_relocs++; 293 r_info = ELF64_R_SYM(rela[i].r_info); 294 r_addend = rela[i].r_addend; 295 } 296 } 297 298 return _count_relocs; 299 } 300 301 /* Get size of potential GOT required. */ 302 static unsigned long get_got_size(const Elf64_Ehdr *hdr, 303 const Elf64_Shdr *sechdrs, 304 struct module *me) 305 { 306 /* One extra reloc so it's always 0-addr terminated */ 307 unsigned long relocs = 1; 308 unsigned int i, symindex = 0; 309 310 for (i = 1; i < hdr->e_shnum; i++) { 311 if (sechdrs[i].sh_type == SHT_SYMTAB) { 312 symindex = i; 313 break; 314 } 315 } 316 WARN_ON_ONCE(!symindex); 317 318 /* Every relocated section... */ 319 for (i = 1; i < hdr->e_shnum; i++) { 320 if (sechdrs[i].sh_type == SHT_RELA) { 321 pr_debug("Found relocations in section %u\n", i); 322 pr_debug("Ptr: %p. Number: %llu\n", (void *)sechdrs[i].sh_addr, 323 sechdrs[i].sh_size / sizeof(Elf64_Rela)); 324 325 /* 326 * Sort the relocation information based on a symbol and 327 * addend key. This is a stable O(n*log n) complexity 328 * algorithm but it will reduce the complexity of 329 * count_relocs() to linear complexity O(n) 330 */ 331 sort((void *)sechdrs[i].sh_addr, 332 sechdrs[i].sh_size / sizeof(Elf64_Rela), 333 sizeof(Elf64_Rela), relacmp, NULL); 334 335 relocs += count_relocs((void *)sechdrs[i].sh_addr, 336 sechdrs[i].sh_size 337 / sizeof(Elf64_Rela), 338 R_PPC64_GOT_PCREL34); 339 340 /* 341 * Percpu data access typically gets linked with 342 * REL34 relocations, but the percpu section gets 343 * moved at load time and requires that to be 344 * converted to GOT linkage. 345 */ 346 if (IS_ENABLED(CONFIG_SMP) && symindex) 347 relocs += count_pcpu_relocs(sechdrs, 348 (void *)sechdrs[i].sh_addr, 349 sechdrs[i].sh_size 350 / sizeof(Elf64_Rela), 351 symindex, me->arch.pcpu_section); 352 } 353 } 354 355 pr_debug("Looks like a total of %lu GOT entries, max\n", relocs); 356 return relocs * sizeof(struct ppc64_got_entry); 357 } 358 #else /* CONFIG_PPC_KERNEL_PCREL */ 359 360 /* Still needed for ELFv2, for .TOC. */ 361 static void dedotify_versions(struct modversion_info *vers, 362 unsigned long size) 363 { 364 struct modversion_info *end; 365 366 for (end = (void *)vers + size; vers < end; vers++) 367 if (vers->name[0] == '.') { 368 memmove(vers->name, vers->name+1, strlen(vers->name)); 369 } 370 } 371 372 /* 373 * Undefined symbols which refer to .funcname, hack to funcname. Make .TOC. 374 * seem to be defined (value set later). 375 */ 376 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) 377 { 378 unsigned int i; 379 380 for (i = 1; i < numsyms; i++) { 381 if (syms[i].st_shndx == SHN_UNDEF) { 382 char *name = strtab + syms[i].st_name; 383 if (name[0] == '.') { 384 if (strcmp(name+1, "TOC.") == 0) 385 syms[i].st_shndx = SHN_ABS; 386 syms[i].st_name++; 387 } 388 } 389 } 390 } 391 392 static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs, 393 const char *strtab, 394 unsigned int symindex) 395 { 396 unsigned int i, numsyms; 397 Elf64_Sym *syms; 398 399 syms = (Elf64_Sym *)sechdrs[symindex].sh_addr; 400 numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym); 401 402 for (i = 1; i < numsyms; i++) { 403 if (syms[i].st_shndx == SHN_ABS 404 && strcmp(strtab + syms[i].st_name, "TOC.") == 0) 405 return &syms[i]; 406 } 407 return NULL; 408 } 409 #endif /* CONFIG_PPC_KERNEL_PCREL */ 410 411 bool module_init_section(const char *name) 412 { 413 /* We don't handle .init for the moment: always return false. */ 414 return false; 415 } 416 417 int module_frob_arch_sections(Elf64_Ehdr *hdr, 418 Elf64_Shdr *sechdrs, 419 char *secstrings, 420 struct module *me) 421 { 422 unsigned int i; 423 424 /* Find .toc and .stubs sections, symtab and strtab */ 425 for (i = 1; i < hdr->e_shnum; i++) { 426 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) 427 me->arch.stubs_section = i; 428 #ifdef CONFIG_PPC_KERNEL_PCREL 429 else if (strcmp(secstrings + sechdrs[i].sh_name, ".data..percpu") == 0) 430 me->arch.pcpu_section = i; 431 else if (strcmp(secstrings + sechdrs[i].sh_name, ".mygot") == 0) { 432 me->arch.got_section = i; 433 if (sechdrs[i].sh_addralign < 8) 434 sechdrs[i].sh_addralign = 8; 435 } 436 #else 437 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) { 438 me->arch.toc_section = i; 439 if (sechdrs[i].sh_addralign < 8) 440 sechdrs[i].sh_addralign = 8; 441 } 442 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) 443 dedotify_versions((void *)hdr + sechdrs[i].sh_offset, 444 sechdrs[i].sh_size); 445 446 if (sechdrs[i].sh_type == SHT_SYMTAB) 447 dedotify((void *)hdr + sechdrs[i].sh_offset, 448 sechdrs[i].sh_size / sizeof(Elf64_Sym), 449 (void *)hdr 450 + sechdrs[sechdrs[i].sh_link].sh_offset); 451 #endif 452 } 453 454 if (!me->arch.stubs_section) { 455 pr_err("%s: doesn't contain .stubs.\n", me->name); 456 return -ENOEXEC; 457 } 458 459 #ifdef CONFIG_PPC_KERNEL_PCREL 460 if (!me->arch.got_section) { 461 pr_err("%s: doesn't contain .mygot.\n", me->name); 462 return -ENOEXEC; 463 } 464 465 /* Override the got size */ 466 sechdrs[me->arch.got_section].sh_size = get_got_size(hdr, sechdrs, me); 467 #else 468 /* If we don't have a .toc, just use .stubs. We need to set r2 469 to some reasonable value in case the module calls out to 470 other functions via a stub, or if a function pointer escapes 471 the module by some means. */ 472 if (!me->arch.toc_section) 473 me->arch.toc_section = me->arch.stubs_section; 474 #endif 475 476 /* Override the stubs size */ 477 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs, secstrings, me); 478 479 return 0; 480 } 481 482 #if defined(CONFIG_MPROFILE_KERNEL) || defined(CONFIG_ARCH_USING_PATCHABLE_FUNCTION_ENTRY) 483 484 static u32 stub_insns[] = { 485 #ifdef CONFIG_PPC_KERNEL_PCREL 486 PPC_RAW_LD(_R12, _R13, offsetof(struct paca_struct, kernelbase)), 487 PPC_RAW_NOP(), /* align the prefix insn */ 488 /* paddi r12,r12,addr */ 489 PPC_PREFIX_MLS | __PPC_PRFX_R(0), 490 PPC_INST_PADDI | ___PPC_RT(_R12) | ___PPC_RA(_R12), 491 PPC_RAW_MTCTR(_R12), 492 PPC_RAW_BCTR(), 493 #else 494 PPC_RAW_LD(_R12, _R13, offsetof(struct paca_struct, kernel_toc)), 495 PPC_RAW_ADDIS(_R12, _R12, 0), 496 PPC_RAW_ADDI(_R12, _R12, 0), 497 PPC_RAW_MTCTR(_R12), 498 PPC_RAW_BCTR(), 499 #endif 500 }; 501 502 /* 503 * For mprofile-kernel we use a special stub for ftrace_caller() because we 504 * can't rely on r2 containing this module's TOC when we enter the stub. 505 * 506 * That can happen if the function calling us didn't need to use the toc. In 507 * that case it won't have setup r2, and the r2 value will be either the 508 * kernel's toc, or possibly another modules toc. 509 * 510 * To deal with that this stub uses the kernel toc, which is always accessible 511 * via the paca (in r13). The target (ftrace_caller()) is responsible for 512 * saving and restoring the toc before returning. 513 */ 514 static inline int create_ftrace_stub(struct ppc64_stub_entry *entry, 515 unsigned long addr, 516 struct module *me) 517 { 518 long reladdr; 519 520 if ((unsigned long)entry->jump % 8 != 0) { 521 pr_err("%s: Address of stub entry is not 8-byte aligned\n", me->name); 522 return 0; 523 } 524 525 BUILD_BUG_ON(sizeof(stub_insns) > sizeof(entry->jump)); 526 memcpy(entry->jump, stub_insns, sizeof(stub_insns)); 527 528 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) { 529 /* Stub uses address relative to kernel base (from the paca) */ 530 reladdr = addr - local_paca->kernelbase; 531 if (reladdr > 0x1FFFFFFFFL || reladdr < -0x200000000L) { 532 pr_err("%s: Address of %ps out of range of 34-bit relative address.\n", 533 me->name, (void *)addr); 534 return 0; 535 } 536 537 entry->jump[2] |= IMM_H18(reladdr); 538 entry->jump[3] |= IMM_L(reladdr); 539 } else { 540 /* Stub uses address relative to kernel toc (from the paca) */ 541 reladdr = addr - kernel_toc_addr(); 542 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 543 pr_err("%s: Address of %ps out of range of kernel_toc.\n", 544 me->name, (void *)addr); 545 return 0; 546 } 547 548 entry->jump[1] |= PPC_HA(reladdr); 549 entry->jump[2] |= PPC_LO(reladdr); 550 } 551 552 /* Even though we don't use funcdata in the stub, it's needed elsewhere. */ 553 entry->funcdata = func_desc(addr); 554 entry->magic = STUB_MAGIC; 555 556 return 1; 557 } 558 559 static bool is_mprofile_ftrace_call(const char *name) 560 { 561 if (!strcmp("_mcount", name)) 562 return true; 563 #ifdef CONFIG_DYNAMIC_FTRACE 564 if (!strcmp("ftrace_caller", name)) 565 return true; 566 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS 567 if (!strcmp("ftrace_regs_caller", name)) 568 return true; 569 #endif 570 #endif 571 572 return false; 573 } 574 #else 575 static inline int create_ftrace_stub(struct ppc64_stub_entry *entry, 576 unsigned long addr, 577 struct module *me) 578 { 579 return 0; 580 } 581 582 static bool is_mprofile_ftrace_call(const char *name) 583 { 584 return false; 585 } 586 #endif 587 588 /* 589 * r2 is the TOC pointer: it actually points 0x8000 into the TOC (this gives the 590 * value maximum span in an instruction which uses a signed offset). Round down 591 * to a 256 byte boundary for the odd case where we are setting up r2 without a 592 * .toc section. 593 */ 594 static inline unsigned long my_r2(const Elf64_Shdr *sechdrs, struct module *me) 595 { 596 #ifndef CONFIG_PPC_KERNEL_PCREL 597 return (sechdrs[me->arch.toc_section].sh_addr & ~0xfful) + 0x8000; 598 #else 599 return -1; 600 #endif 601 } 602 603 /* Patch stub to reference function and correct r2 value. */ 604 static inline int create_stub(const Elf64_Shdr *sechdrs, 605 struct ppc64_stub_entry *entry, 606 unsigned long addr, 607 struct module *me, 608 const char *name) 609 { 610 long reladdr; 611 func_desc_t desc; 612 int i; 613 614 if (is_mprofile_ftrace_call(name)) 615 return create_ftrace_stub(entry, addr, me); 616 617 if ((unsigned long)entry->jump % 8 != 0) { 618 pr_err("%s: Address of stub entry is not 8-byte aligned\n", me->name); 619 return 0; 620 } 621 622 BUILD_BUG_ON(sizeof(ppc64_stub_insns) > sizeof(entry->jump)); 623 for (i = 0; i < ARRAY_SIZE(ppc64_stub_insns); i++) { 624 if (patch_instruction(&entry->jump[i], 625 ppc_inst(ppc64_stub_insns[i]))) 626 return 0; 627 } 628 629 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) { 630 /* Stub uses address relative to itself! */ 631 reladdr = 0 + offsetof(struct ppc64_stub_entry, funcdata); 632 BUILD_BUG_ON(reladdr != 32); 633 if (reladdr > 0x1FFFFFFFFL || reladdr < -0x200000000L) { 634 pr_err("%s: Address of %p out of range of 34-bit relative address.\n", 635 me->name, (void *)reladdr); 636 return 0; 637 } 638 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr); 639 640 /* May not even need this if we're relative to 0 */ 641 if (patch_instruction(&entry->jump[0], 642 ppc_inst_prefix(entry->jump[0] | IMM_H18(reladdr), 643 entry->jump[1] | IMM_L(reladdr)))) 644 return 0; 645 646 } else { 647 /* Stub uses address relative to r2. */ 648 reladdr = (unsigned long)entry - my_r2(sechdrs, me); 649 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 650 pr_err("%s: Address %p of stub out of range of %p.\n", 651 me->name, (void *)reladdr, (void *)my_r2); 652 return 0; 653 } 654 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr); 655 656 if (patch_instruction(&entry->jump[0], 657 ppc_inst(entry->jump[0] | PPC_HA(reladdr)))) 658 return 0; 659 660 if (patch_instruction(&entry->jump[1], 661 ppc_inst(entry->jump[1] | PPC_LO(reladdr)))) 662 return 0; 663 } 664 665 // func_desc_t is 8 bytes if ABIv2, else 16 bytes 666 desc = func_desc(addr); 667 for (i = 0; i < sizeof(func_desc_t) / sizeof(u32); i++) { 668 if (patch_u32(((u32 *)&entry->funcdata) + i, ((u32 *)&desc)[i])) 669 return 0; 670 } 671 672 if (patch_u32(&entry->magic, STUB_MAGIC)) 673 return 0; 674 675 return 1; 676 } 677 678 /* Create stub to jump to function described in this OPD/ptr: we need the 679 stub to set up the TOC ptr (r2) for the function. */ 680 static unsigned long stub_for_addr(const Elf64_Shdr *sechdrs, 681 unsigned long addr, 682 struct module *me, 683 const char *name) 684 { 685 struct ppc64_stub_entry *stubs; 686 unsigned int i, num_stubs; 687 688 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); 689 690 /* Find this stub, or if that fails, the next avail. entry */ 691 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; 692 for (i = 0; stub_func_addr(stubs[i].funcdata); i++) { 693 if (WARN_ON(i >= num_stubs)) 694 return 0; 695 696 if (stub_func_addr(stubs[i].funcdata) == func_addr(addr)) 697 return (unsigned long)&stubs[i]; 698 } 699 700 if (!create_stub(sechdrs, &stubs[i], addr, me, name)) 701 return 0; 702 703 return (unsigned long)&stubs[i]; 704 } 705 706 #ifdef CONFIG_PPC_KERNEL_PCREL 707 /* Create GOT to load the location described in this ptr */ 708 static unsigned long got_for_addr(const Elf64_Shdr *sechdrs, 709 unsigned long addr, 710 struct module *me, 711 const char *name) 712 { 713 struct ppc64_got_entry *got; 714 unsigned int i, num_got; 715 716 if (!IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) 717 return addr; 718 719 num_got = sechdrs[me->arch.got_section].sh_size / sizeof(*got); 720 721 /* Find this stub, or if that fails, the next avail. entry */ 722 got = (void *)sechdrs[me->arch.got_section].sh_addr; 723 for (i = 0; got[i].addr; i++) { 724 if (WARN_ON(i >= num_got)) 725 return 0; 726 727 if (got[i].addr == addr) 728 return (unsigned long)&got[i]; 729 } 730 731 got[i].addr = addr; 732 733 return (unsigned long)&got[i]; 734 } 735 #endif 736 737 /* We expect a noop next: if it is, replace it with instruction to 738 restore r2. */ 739 static int restore_r2(const char *name, u32 *instruction, struct module *me) 740 { 741 u32 *prev_insn = instruction - 1; 742 u32 insn_val = *instruction; 743 744 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) 745 return 0; 746 747 if (is_mprofile_ftrace_call(name)) 748 return 0; 749 750 /* 751 * Make sure the branch isn't a sibling call. Sibling calls aren't 752 * "link" branches and they don't return, so they don't need the r2 753 * restore afterwards. 754 */ 755 if (!instr_is_relative_link_branch(ppc_inst(*prev_insn))) 756 return 0; 757 758 /* 759 * For livepatch, the restore r2 instruction might have already been 760 * written previously, if the referenced symbol is in a previously 761 * unloaded module which is now being loaded again. In that case, skip 762 * the warning and the instruction write. 763 */ 764 if (insn_val == PPC_INST_LD_TOC) 765 return 0; 766 767 if (insn_val != PPC_RAW_NOP()) { 768 pr_err("%s: Expected nop after call, got %08x at %pS\n", 769 me->name, insn_val, instruction); 770 return -ENOEXEC; 771 } 772 773 /* ld r2,R2_STACK_OFFSET(r1) */ 774 return patch_instruction(instruction, ppc_inst(PPC_INST_LD_TOC)); 775 } 776 777 int apply_relocate_add(Elf64_Shdr *sechdrs, 778 const char *strtab, 779 unsigned int symindex, 780 unsigned int relsec, 781 struct module *me) 782 { 783 unsigned int i; 784 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; 785 Elf64_Sym *sym; 786 unsigned long *location; 787 unsigned long value; 788 789 pr_debug("Applying ADD relocate section %u to %u\n", relsec, 790 sechdrs[relsec].sh_info); 791 792 #ifndef CONFIG_PPC_KERNEL_PCREL 793 /* First time we're called, we can fix up .TOC. */ 794 if (!me->arch.toc_fixed) { 795 sym = find_dot_toc(sechdrs, strtab, symindex); 796 /* It's theoretically possible that a module doesn't want a 797 * .TOC. so don't fail it just for that. */ 798 if (sym) 799 sym->st_value = my_r2(sechdrs, me); 800 me->arch.toc_fixed = true; 801 } 802 #endif 803 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { 804 /* This is where to make the change */ 805 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr 806 + rela[i].r_offset; 807 /* This is the symbol it is referring to */ 808 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr 809 + ELF64_R_SYM(rela[i].r_info); 810 811 pr_debug("RELOC at %p: %li-type as %s (0x%lx) + %li\n", 812 location, (long)ELF64_R_TYPE(rela[i].r_info), 813 strtab + sym->st_name, (unsigned long)sym->st_value, 814 (long)rela[i].r_addend); 815 816 /* `Everything is relative'. */ 817 value = sym->st_value + rela[i].r_addend; 818 819 switch (ELF64_R_TYPE(rela[i].r_info)) { 820 case R_PPC64_ADDR32: 821 /* Simply set it */ 822 *(u32 *)location = value; 823 break; 824 825 case R_PPC64_ADDR64: 826 /* Simply set it */ 827 *(unsigned long *)location = value; 828 break; 829 830 #ifndef CONFIG_PPC_KERNEL_PCREL 831 case R_PPC64_TOC: 832 *(unsigned long *)location = my_r2(sechdrs, me); 833 break; 834 835 case R_PPC64_TOC16: 836 /* Subtract TOC pointer */ 837 value -= my_r2(sechdrs, me); 838 if (value + 0x8000 > 0xffff) { 839 pr_err("%s: bad TOC16 relocation (0x%lx)\n", 840 me->name, value); 841 return -ENOEXEC; 842 } 843 *((uint16_t *) location) 844 = (*((uint16_t *) location) & ~0xffff) 845 | (value & 0xffff); 846 break; 847 848 case R_PPC64_TOC16_LO: 849 /* Subtract TOC pointer */ 850 value -= my_r2(sechdrs, me); 851 *((uint16_t *) location) 852 = (*((uint16_t *) location) & ~0xffff) 853 | (value & 0xffff); 854 break; 855 856 case R_PPC64_TOC16_DS: 857 /* Subtract TOC pointer */ 858 value -= my_r2(sechdrs, me); 859 if ((value & 3) != 0 || value + 0x8000 > 0xffff) { 860 pr_err("%s: bad TOC16_DS relocation (0x%lx)\n", 861 me->name, value); 862 return -ENOEXEC; 863 } 864 *((uint16_t *) location) 865 = (*((uint16_t *) location) & ~0xfffc) 866 | (value & 0xfffc); 867 break; 868 869 case R_PPC64_TOC16_LO_DS: 870 /* Subtract TOC pointer */ 871 value -= my_r2(sechdrs, me); 872 if ((value & 3) != 0) { 873 pr_err("%s: bad TOC16_LO_DS relocation (0x%lx)\n", 874 me->name, value); 875 return -ENOEXEC; 876 } 877 *((uint16_t *) location) 878 = (*((uint16_t *) location) & ~0xfffc) 879 | (value & 0xfffc); 880 break; 881 882 case R_PPC64_TOC16_HA: 883 /* Subtract TOC pointer */ 884 value -= my_r2(sechdrs, me); 885 value = ((value + 0x8000) >> 16); 886 *((uint16_t *) location) 887 = (*((uint16_t *) location) & ~0xffff) 888 | (value & 0xffff); 889 break; 890 #endif 891 892 case R_PPC_REL24: 893 #ifdef CONFIG_PPC_KERNEL_PCREL 894 /* PCREL still generates REL24 for mcount */ 895 case R_PPC64_REL24_NOTOC: 896 #endif 897 /* FIXME: Handle weak symbols here --RR */ 898 if (sym->st_shndx == SHN_UNDEF || 899 sym->st_shndx == SHN_LIVEPATCH) { 900 /* External: go via stub */ 901 value = stub_for_addr(sechdrs, value, me, 902 strtab + sym->st_name); 903 if (!value) 904 return -ENOENT; 905 if (restore_r2(strtab + sym->st_name, 906 (u32 *)location + 1, me)) 907 return -ENOEXEC; 908 } else 909 value += local_entry_offset(sym); 910 911 /* Convert value to relative */ 912 value -= (unsigned long)location; 913 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ 914 pr_err("%s: REL24 %li out of range!\n", 915 me->name, (long int)value); 916 return -ENOEXEC; 917 } 918 919 /* Only replace bits 2 through 26 */ 920 value = (*(uint32_t *)location & ~PPC_LI_MASK) | PPC_LI(value); 921 922 if (patch_instruction((u32 *)location, ppc_inst(value))) 923 return -EFAULT; 924 925 break; 926 927 case R_PPC64_REL64: 928 /* 64 bits relative (used by features fixups) */ 929 *location = value - (unsigned long)location; 930 break; 931 932 case R_PPC64_REL32: 933 /* 32 bits relative (used by relative exception tables) */ 934 /* Convert value to relative */ 935 value -= (unsigned long)location; 936 if (value + 0x80000000 > 0xffffffff) { 937 pr_err("%s: REL32 %li out of range!\n", 938 me->name, (long int)value); 939 return -ENOEXEC; 940 } 941 *(u32 *)location = value; 942 break; 943 944 #ifdef CONFIG_PPC_KERNEL_PCREL 945 case R_PPC64_PCREL34: { 946 unsigned long absvalue = value; 947 948 /* Convert value to relative */ 949 value -= (unsigned long)location; 950 951 if (value + 0x200000000 > 0x3ffffffff) { 952 if (sym->st_shndx != me->arch.pcpu_section) { 953 pr_err("%s: REL34 %li out of range!\n", 954 me->name, (long)value); 955 return -ENOEXEC; 956 } 957 958 /* 959 * per-cpu section is special cased because 960 * it is moved during loading, so has to be 961 * converted to use GOT. 962 */ 963 value = got_for_addr(sechdrs, absvalue, me, 964 strtab + sym->st_name); 965 if (!value) 966 return -ENOENT; 967 value -= (unsigned long)location; 968 969 /* Turn pla into pld */ 970 if (patch_instruction((u32 *)location, 971 ppc_inst_prefix((*(u32 *)location & ~0x02000000), 972 (*((u32 *)location + 1) & ~0xf8000000) | 0xe4000000))) 973 return -EFAULT; 974 } 975 976 if (patch_instruction((u32 *)location, 977 ppc_inst_prefix((*(u32 *)location & ~0x3ffff) | IMM_H18(value), 978 (*((u32 *)location + 1) & ~0xffff) | IMM_L(value)))) 979 return -EFAULT; 980 981 break; 982 } 983 984 #else 985 case R_PPC64_TOCSAVE: 986 /* 987 * Marker reloc indicates we don't have to save r2. 988 * That would only save us one instruction, so ignore 989 * it. 990 */ 991 break; 992 #endif 993 994 case R_PPC64_ENTRY: 995 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) 996 break; 997 998 /* 999 * Optimize ELFv2 large code model entry point if 1000 * the TOC is within 2GB range of current location. 1001 */ 1002 value = my_r2(sechdrs, me) - (unsigned long)location; 1003 if (value + 0x80008000 > 0xffffffff) 1004 break; 1005 /* 1006 * Check for the large code model prolog sequence: 1007 * ld r2, ...(r12) 1008 * add r2, r2, r12 1009 */ 1010 if ((((uint32_t *)location)[0] & ~0xfffc) != PPC_RAW_LD(_R2, _R12, 0)) 1011 break; 1012 if (((uint32_t *)location)[1] != PPC_RAW_ADD(_R2, _R2, _R12)) 1013 break; 1014 /* 1015 * If found, replace it with: 1016 * addis r2, r12, (.TOC.-func)@ha 1017 * addi r2, r2, (.TOC.-func)@l 1018 */ 1019 ((uint32_t *)location)[0] = PPC_RAW_ADDIS(_R2, _R12, PPC_HA(value)); 1020 ((uint32_t *)location)[1] = PPC_RAW_ADDI(_R2, _R2, PPC_LO(value)); 1021 break; 1022 1023 case R_PPC64_REL16_HA: 1024 /* Subtract location pointer */ 1025 value -= (unsigned long)location; 1026 value = ((value + 0x8000) >> 16); 1027 *((uint16_t *) location) 1028 = (*((uint16_t *) location) & ~0xffff) 1029 | (value & 0xffff); 1030 break; 1031 1032 case R_PPC64_REL16_LO: 1033 /* Subtract location pointer */ 1034 value -= (unsigned long)location; 1035 *((uint16_t *) location) 1036 = (*((uint16_t *) location) & ~0xffff) 1037 | (value & 0xffff); 1038 break; 1039 1040 #ifdef CONFIG_PPC_KERNEL_PCREL 1041 case R_PPC64_GOT_PCREL34: 1042 value = got_for_addr(sechdrs, value, me, 1043 strtab + sym->st_name); 1044 if (!value) 1045 return -ENOENT; 1046 value -= (unsigned long)location; 1047 ((uint32_t *)location)[0] = (((uint32_t *)location)[0] & ~0x3ffff) | 1048 ((value >> 16) & 0x3ffff); 1049 ((uint32_t *)location)[1] = (((uint32_t *)location)[1] & ~0xffff) | 1050 (value & 0xffff); 1051 break; 1052 #endif 1053 1054 default: 1055 pr_err("%s: Unknown ADD relocation: %lu\n", 1056 me->name, 1057 (unsigned long)ELF64_R_TYPE(rela[i].r_info)); 1058 return -ENOEXEC; 1059 } 1060 } 1061 1062 return 0; 1063 } 1064 1065 #ifdef CONFIG_DYNAMIC_FTRACE 1066 int module_trampoline_target(struct module *mod, unsigned long addr, 1067 unsigned long *target) 1068 { 1069 struct ppc64_stub_entry *stub; 1070 func_desc_t funcdata; 1071 u32 magic; 1072 1073 if (!within_module_core(addr, mod)) { 1074 pr_err("%s: stub %lx not in module %s\n", __func__, addr, mod->name); 1075 return -EFAULT; 1076 } 1077 1078 stub = (struct ppc64_stub_entry *)addr; 1079 1080 if (copy_from_kernel_nofault(&magic, &stub->magic, 1081 sizeof(magic))) { 1082 pr_err("%s: fault reading magic for stub %lx for %s\n", __func__, addr, mod->name); 1083 return -EFAULT; 1084 } 1085 1086 if (magic != STUB_MAGIC) { 1087 pr_err("%s: bad magic for stub %lx for %s\n", __func__, addr, mod->name); 1088 return -EFAULT; 1089 } 1090 1091 if (copy_from_kernel_nofault(&funcdata, &stub->funcdata, 1092 sizeof(funcdata))) { 1093 pr_err("%s: fault reading funcdata for stub %lx for %s\n", __func__, addr, mod->name); 1094 return -EFAULT; 1095 } 1096 1097 *target = stub_func_addr(funcdata); 1098 1099 return 0; 1100 } 1101 1102 static int setup_ftrace_ool_stubs(const Elf64_Shdr *sechdrs, unsigned long addr, struct module *me) 1103 { 1104 #ifdef CONFIG_PPC_FTRACE_OUT_OF_LINE 1105 unsigned int i, total_stubs, num_stubs; 1106 struct ppc64_stub_entry *stub; 1107 1108 total_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stub); 1109 num_stubs = roundup(me->arch.ool_stub_count * sizeof(struct ftrace_ool_stub), 1110 sizeof(struct ppc64_stub_entry)) / sizeof(struct ppc64_stub_entry); 1111 1112 /* Find the next available entry */ 1113 stub = (void *)sechdrs[me->arch.stubs_section].sh_addr; 1114 for (i = 0; stub_func_addr(stub[i].funcdata); i++) 1115 if (WARN_ON(i >= total_stubs)) 1116 return -1; 1117 1118 if (WARN_ON(i + num_stubs > total_stubs)) 1119 return -1; 1120 1121 stub += i; 1122 me->arch.ool_stubs = (struct ftrace_ool_stub *)stub; 1123 1124 /* reserve stubs */ 1125 for (i = 0; i < num_stubs; i++) 1126 if (patch_u32((void *)&stub->funcdata, PPC_RAW_NOP())) 1127 return -1; 1128 #endif 1129 1130 return 0; 1131 } 1132 1133 int module_finalize_ftrace(struct module *mod, const Elf_Shdr *sechdrs) 1134 { 1135 mod->arch.tramp = stub_for_addr(sechdrs, 1136 (unsigned long)ftrace_caller, 1137 mod, 1138 "ftrace_caller"); 1139 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS 1140 mod->arch.tramp_regs = stub_for_addr(sechdrs, 1141 (unsigned long)ftrace_regs_caller, 1142 mod, 1143 "ftrace_regs_caller"); 1144 if (!mod->arch.tramp_regs) 1145 return -ENOENT; 1146 #endif 1147 1148 if (!mod->arch.tramp) 1149 return -ENOENT; 1150 1151 if (setup_ftrace_ool_stubs(sechdrs, mod->arch.tramp, mod)) 1152 return -ENOENT; 1153 1154 return 0; 1155 } 1156 #endif 1157