1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * handle transition of Linux booting another kernel 4 * Copyright (C) 2002-2005 Eric Biederman <ebiederm@xmission.com> 5 */ 6 7 #define pr_fmt(fmt) "kexec: " fmt 8 9 #include <linux/mm.h> 10 #include <linux/kexec.h> 11 #include <linux/string.h> 12 #include <linux/gfp.h> 13 #include <linux/reboot.h> 14 #include <linux/numa.h> 15 #include <linux/ftrace.h> 16 #include <linux/io.h> 17 #include <linux/suspend.h> 18 #include <linux/vmalloc.h> 19 #include <linux/efi.h> 20 #include <linux/cc_platform.h> 21 22 #include <asm/init.h> 23 #include <asm/tlbflush.h> 24 #include <asm/mmu_context.h> 25 #include <asm/io_apic.h> 26 #include <asm/debugreg.h> 27 #include <asm/kexec-bzimage64.h> 28 #include <asm/setup.h> 29 #include <asm/set_memory.h> 30 #include <asm/cpu.h> 31 32 #ifdef CONFIG_ACPI 33 /* 34 * Used while adding mapping for ACPI tables. 35 * Can be reused when other iomem regions need be mapped 36 */ 37 struct init_pgtable_data { 38 struct x86_mapping_info *info; 39 pgd_t *level4p; 40 }; 41 42 static int mem_region_callback(struct resource *res, void *arg) 43 { 44 struct init_pgtable_data *data = arg; 45 unsigned long mstart, mend; 46 47 mstart = res->start; 48 mend = mstart + resource_size(res) - 1; 49 50 return kernel_ident_mapping_init(data->info, data->level4p, mstart, mend); 51 } 52 53 static int 54 map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) 55 { 56 struct init_pgtable_data data; 57 unsigned long flags; 58 int ret; 59 60 data.info = info; 61 data.level4p = level4p; 62 flags = IORESOURCE_MEM | IORESOURCE_BUSY; 63 64 ret = walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, 65 &data, mem_region_callback); 66 if (ret && ret != -EINVAL) 67 return ret; 68 69 /* ACPI tables could be located in ACPI Non-volatile Storage region */ 70 ret = walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, 71 &data, mem_region_callback); 72 if (ret && ret != -EINVAL) 73 return ret; 74 75 return 0; 76 } 77 #else 78 static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; } 79 #endif 80 81 #ifdef CONFIG_KEXEC_FILE 82 const struct kexec_file_ops * const kexec_file_loaders[] = { 83 &kexec_bzImage64_ops, 84 NULL 85 }; 86 #endif 87 88 static int 89 map_efi_systab(struct x86_mapping_info *info, pgd_t *level4p) 90 { 91 #ifdef CONFIG_EFI 92 unsigned long mstart, mend; 93 94 if (!efi_enabled(EFI_BOOT)) 95 return 0; 96 97 mstart = (boot_params.efi_info.efi_systab | 98 ((u64)boot_params.efi_info.efi_systab_hi<<32)); 99 100 if (efi_enabled(EFI_64BIT)) 101 mend = mstart + sizeof(efi_system_table_64_t); 102 else 103 mend = mstart + sizeof(efi_system_table_32_t); 104 105 if (!mstart) 106 return 0; 107 108 return kernel_ident_mapping_init(info, level4p, mstart, mend); 109 #endif 110 return 0; 111 } 112 113 static void free_transition_pgtable(struct kimage *image) 114 { 115 free_page((unsigned long)image->arch.p4d); 116 image->arch.p4d = NULL; 117 free_page((unsigned long)image->arch.pud); 118 image->arch.pud = NULL; 119 free_page((unsigned long)image->arch.pmd); 120 image->arch.pmd = NULL; 121 free_page((unsigned long)image->arch.pte); 122 image->arch.pte = NULL; 123 } 124 125 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd) 126 { 127 pgprot_t prot = PAGE_KERNEL_EXEC_NOENC; 128 unsigned long vaddr, paddr; 129 int result = -ENOMEM; 130 p4d_t *p4d; 131 pud_t *pud; 132 pmd_t *pmd; 133 pte_t *pte; 134 135 vaddr = (unsigned long)relocate_kernel; 136 paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE); 137 pgd += pgd_index(vaddr); 138 if (!pgd_present(*pgd)) { 139 p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL); 140 if (!p4d) 141 goto err; 142 image->arch.p4d = p4d; 143 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE)); 144 } 145 p4d = p4d_offset(pgd, vaddr); 146 if (!p4d_present(*p4d)) { 147 pud = (pud_t *)get_zeroed_page(GFP_KERNEL); 148 if (!pud) 149 goto err; 150 image->arch.pud = pud; 151 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE)); 152 } 153 pud = pud_offset(p4d, vaddr); 154 if (!pud_present(*pud)) { 155 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL); 156 if (!pmd) 157 goto err; 158 image->arch.pmd = pmd; 159 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE)); 160 } 161 pmd = pmd_offset(pud, vaddr); 162 if (!pmd_present(*pmd)) { 163 pte = (pte_t *)get_zeroed_page(GFP_KERNEL); 164 if (!pte) 165 goto err; 166 image->arch.pte = pte; 167 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE)); 168 } 169 pte = pte_offset_kernel(pmd, vaddr); 170 171 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) 172 prot = PAGE_KERNEL_EXEC; 173 174 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot)); 175 return 0; 176 err: 177 return result; 178 } 179 180 static void *alloc_pgt_page(void *data) 181 { 182 struct kimage *image = (struct kimage *)data; 183 struct page *page; 184 void *p = NULL; 185 186 page = kimage_alloc_control_pages(image, 0); 187 if (page) { 188 p = page_address(page); 189 clear_page(p); 190 } 191 192 return p; 193 } 194 195 static int init_pgtable(struct kimage *image, unsigned long start_pgtable) 196 { 197 struct x86_mapping_info info = { 198 .alloc_pgt_page = alloc_pgt_page, 199 .context = image, 200 .page_flag = __PAGE_KERNEL_LARGE_EXEC, 201 .kernpg_flag = _KERNPG_TABLE_NOENC, 202 }; 203 unsigned long mstart, mend; 204 pgd_t *level4p; 205 int result; 206 int i; 207 208 level4p = (pgd_t *)__va(start_pgtable); 209 clear_page(level4p); 210 211 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) { 212 info.page_flag |= _PAGE_ENC; 213 info.kernpg_flag |= _PAGE_ENC; 214 } 215 216 if (direct_gbpages) 217 info.direct_gbpages = true; 218 219 for (i = 0; i < nr_pfn_mapped; i++) { 220 mstart = pfn_mapped[i].start << PAGE_SHIFT; 221 mend = pfn_mapped[i].end << PAGE_SHIFT; 222 223 result = kernel_ident_mapping_init(&info, 224 level4p, mstart, mend); 225 if (result) 226 return result; 227 } 228 229 /* 230 * segments's mem ranges could be outside 0 ~ max_pfn, 231 * for example when jump back to original kernel from kexeced kernel. 232 * or first kernel is booted with user mem map, and second kernel 233 * could be loaded out of that range. 234 */ 235 for (i = 0; i < image->nr_segments; i++) { 236 mstart = image->segment[i].mem; 237 mend = mstart + image->segment[i].memsz; 238 239 result = kernel_ident_mapping_init(&info, 240 level4p, mstart, mend); 241 242 if (result) 243 return result; 244 } 245 246 /* 247 * Prepare EFI systab and ACPI tables for kexec kernel since they are 248 * not covered by pfn_mapped. 249 */ 250 result = map_efi_systab(&info, level4p); 251 if (result) 252 return result; 253 254 result = map_acpi_tables(&info, level4p); 255 if (result) 256 return result; 257 258 return init_transition_pgtable(image, level4p); 259 } 260 261 static void load_segments(void) 262 { 263 __asm__ __volatile__ ( 264 "\tmovl %0,%%ds\n" 265 "\tmovl %0,%%es\n" 266 "\tmovl %0,%%ss\n" 267 "\tmovl %0,%%fs\n" 268 "\tmovl %0,%%gs\n" 269 : : "a" (__KERNEL_DS) : "memory" 270 ); 271 } 272 273 int machine_kexec_prepare(struct kimage *image) 274 { 275 unsigned long start_pgtable; 276 int result; 277 278 /* Calculate the offsets */ 279 start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT; 280 281 /* Setup the identity mapped 64bit page table */ 282 result = init_pgtable(image, start_pgtable); 283 if (result) 284 return result; 285 286 return 0; 287 } 288 289 void machine_kexec_cleanup(struct kimage *image) 290 { 291 free_transition_pgtable(image); 292 } 293 294 /* 295 * Do not allocate memory (or fail in any way) in machine_kexec(). 296 * We are past the point of no return, committed to rebooting now. 297 */ 298 void machine_kexec(struct kimage *image) 299 { 300 unsigned long page_list[PAGES_NR]; 301 void *control_page; 302 int save_ftrace_enabled; 303 304 #ifdef CONFIG_KEXEC_JUMP 305 if (image->preserve_context) 306 save_processor_state(); 307 #endif 308 309 save_ftrace_enabled = __ftrace_enabled_save(); 310 311 /* Interrupts aren't acceptable while we reboot */ 312 local_irq_disable(); 313 hw_breakpoint_disable(); 314 cet_disable(); 315 316 if (image->preserve_context) { 317 #ifdef CONFIG_X86_IO_APIC 318 /* 319 * We need to put APICs in legacy mode so that we can 320 * get timer interrupts in second kernel. kexec/kdump 321 * paths already have calls to restore_boot_irq_mode() 322 * in one form or other. kexec jump path also need one. 323 */ 324 clear_IO_APIC(); 325 restore_boot_irq_mode(); 326 #endif 327 } 328 329 control_page = page_address(image->control_code_page) + PAGE_SIZE; 330 __memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE); 331 332 page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page); 333 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page; 334 page_list[PA_TABLE_PAGE] = 335 (unsigned long)__pa(page_address(image->control_code_page)); 336 337 if (image->type == KEXEC_TYPE_DEFAULT) 338 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page) 339 << PAGE_SHIFT); 340 341 /* 342 * The segment registers are funny things, they have both a 343 * visible and an invisible part. Whenever the visible part is 344 * set to a specific selector, the invisible part is loaded 345 * with from a table in memory. At no other time is the 346 * descriptor table in memory accessed. 347 * 348 * I take advantage of this here by force loading the 349 * segments, before I zap the gdt with an invalid value. 350 */ 351 load_segments(); 352 /* 353 * The gdt & idt are now invalid. 354 * If you want to load them you must set up your own idt & gdt. 355 */ 356 native_idt_invalidate(); 357 native_gdt_invalidate(); 358 359 /* now call it */ 360 image->start = relocate_kernel((unsigned long)image->head, 361 (unsigned long)page_list, 362 image->start, 363 image->preserve_context, 364 cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT)); 365 366 #ifdef CONFIG_KEXEC_JUMP 367 if (image->preserve_context) 368 restore_processor_state(); 369 #endif 370 371 __ftrace_enabled_restore(save_ftrace_enabled); 372 } 373 374 /* arch-dependent functionality related to kexec file-based syscall */ 375 376 #ifdef CONFIG_KEXEC_FILE 377 void *arch_kexec_kernel_image_load(struct kimage *image) 378 { 379 vfree(image->elf_headers); 380 image->elf_headers = NULL; 381 382 if (!image->fops || !image->fops->load) 383 return ERR_PTR(-ENOEXEC); 384 385 return image->fops->load(image, image->kernel_buf, 386 image->kernel_buf_len, image->initrd_buf, 387 image->initrd_buf_len, image->cmdline_buf, 388 image->cmdline_buf_len); 389 } 390 391 /* 392 * Apply purgatory relocations. 393 * 394 * @pi: Purgatory to be relocated. 395 * @section: Section relocations applying to. 396 * @relsec: Section containing RELAs. 397 * @symtabsec: Corresponding symtab. 398 * 399 * TODO: Some of the code belongs to generic code. Move that in kexec.c. 400 */ 401 int arch_kexec_apply_relocations_add(struct purgatory_info *pi, 402 Elf_Shdr *section, const Elf_Shdr *relsec, 403 const Elf_Shdr *symtabsec) 404 { 405 unsigned int i; 406 Elf64_Rela *rel; 407 Elf64_Sym *sym; 408 void *location; 409 unsigned long address, sec_base, value; 410 const char *strtab, *name, *shstrtab; 411 const Elf_Shdr *sechdrs; 412 413 /* String & section header string table */ 414 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 415 strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset; 416 shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset; 417 418 rel = (void *)pi->ehdr + relsec->sh_offset; 419 420 pr_debug("Applying relocate section %s to %u\n", 421 shstrtab + relsec->sh_name, relsec->sh_info); 422 423 for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) { 424 425 /* 426 * rel[i].r_offset contains byte offset from beginning 427 * of section to the storage unit affected. 428 * 429 * This is location to update. This is temporary buffer 430 * where section is currently loaded. This will finally be 431 * loaded to a different address later, pointed to by 432 * ->sh_addr. kexec takes care of moving it 433 * (kexec_load_segment()). 434 */ 435 location = pi->purgatory_buf; 436 location += section->sh_offset; 437 location += rel[i].r_offset; 438 439 /* Final address of the location */ 440 address = section->sh_addr + rel[i].r_offset; 441 442 /* 443 * rel[i].r_info contains information about symbol table index 444 * w.r.t which relocation must be made and type of relocation 445 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get 446 * these respectively. 447 */ 448 sym = (void *)pi->ehdr + symtabsec->sh_offset; 449 sym += ELF64_R_SYM(rel[i].r_info); 450 451 if (sym->st_name) 452 name = strtab + sym->st_name; 453 else 454 name = shstrtab + sechdrs[sym->st_shndx].sh_name; 455 456 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n", 457 name, sym->st_info, sym->st_shndx, sym->st_value, 458 sym->st_size); 459 460 if (sym->st_shndx == SHN_UNDEF) { 461 pr_err("Undefined symbol: %s\n", name); 462 return -ENOEXEC; 463 } 464 465 if (sym->st_shndx == SHN_COMMON) { 466 pr_err("symbol '%s' in common section\n", name); 467 return -ENOEXEC; 468 } 469 470 if (sym->st_shndx == SHN_ABS) 471 sec_base = 0; 472 else if (sym->st_shndx >= pi->ehdr->e_shnum) { 473 pr_err("Invalid section %d for symbol %s\n", 474 sym->st_shndx, name); 475 return -ENOEXEC; 476 } else 477 sec_base = pi->sechdrs[sym->st_shndx].sh_addr; 478 479 value = sym->st_value; 480 value += sec_base; 481 value += rel[i].r_addend; 482 483 switch (ELF64_R_TYPE(rel[i].r_info)) { 484 case R_X86_64_NONE: 485 break; 486 case R_X86_64_64: 487 *(u64 *)location = value; 488 break; 489 case R_X86_64_32: 490 *(u32 *)location = value; 491 if (value != *(u32 *)location) 492 goto overflow; 493 break; 494 case R_X86_64_32S: 495 *(s32 *)location = value; 496 if ((s64)value != *(s32 *)location) 497 goto overflow; 498 break; 499 case R_X86_64_PC32: 500 case R_X86_64_PLT32: 501 value -= (u64)address; 502 *(u32 *)location = value; 503 break; 504 default: 505 pr_err("Unknown rela relocation: %llu\n", 506 ELF64_R_TYPE(rel[i].r_info)); 507 return -ENOEXEC; 508 } 509 } 510 return 0; 511 512 overflow: 513 pr_err("Overflow in relocation type %d value 0x%lx\n", 514 (int)ELF64_R_TYPE(rel[i].r_info), value); 515 return -ENOEXEC; 516 } 517 #endif /* CONFIG_KEXEC_FILE */ 518 519 static int 520 kexec_mark_range(unsigned long start, unsigned long end, bool protect) 521 { 522 struct page *page; 523 unsigned int nr_pages; 524 525 /* 526 * For physical range: [start, end]. We must skip the unassigned 527 * crashk resource with zero-valued "end" member. 528 */ 529 if (!end || start > end) 530 return 0; 531 532 page = pfn_to_page(start >> PAGE_SHIFT); 533 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; 534 if (protect) 535 return set_pages_ro(page, nr_pages); 536 else 537 return set_pages_rw(page, nr_pages); 538 } 539 540 static void kexec_mark_crashkres(bool protect) 541 { 542 unsigned long control; 543 544 kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect); 545 546 /* Don't touch the control code page used in crash_kexec().*/ 547 control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page)); 548 /* Control code page is located in the 2nd page. */ 549 kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect); 550 control += KEXEC_CONTROL_PAGE_SIZE; 551 kexec_mark_range(control, crashk_res.end, protect); 552 } 553 554 void arch_kexec_protect_crashkres(void) 555 { 556 kexec_mark_crashkres(true); 557 } 558 559 void arch_kexec_unprotect_crashkres(void) 560 { 561 kexec_mark_crashkres(false); 562 } 563 564 /* 565 * During a traditional boot under SME, SME will encrypt the kernel, 566 * so the SME kexec kernel also needs to be un-encrypted in order to 567 * replicate a normal SME boot. 568 * 569 * During a traditional boot under SEV, the kernel has already been 570 * loaded encrypted, so the SEV kexec kernel needs to be encrypted in 571 * order to replicate a normal SEV boot. 572 */ 573 int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp) 574 { 575 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT)) 576 return 0; 577 578 /* 579 * If host memory encryption is active we need to be sure that kexec 580 * pages are not encrypted because when we boot to the new kernel the 581 * pages won't be accessed encrypted (initially). 582 */ 583 return set_memory_decrypted((unsigned long)vaddr, pages); 584 } 585 586 void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages) 587 { 588 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT)) 589 return; 590 591 /* 592 * If host memory encryption is active we need to reset the pages back 593 * to being an encrypted mapping before freeing them. 594 */ 595 set_memory_encrypted((unsigned long)vaddr, pages); 596 } 597