arch/x86/kernel/machine_kexec_64.c

   1 /*
   2  * handle transition of Linux booting another kernel
   3  * Copyright (C) 2002-2005 Eric Biederman  <ebiederm@xmission.com>
   4  *
   5  * This source code is licensed under the GNU General Public License,
   6  * Version 2.  See the file COPYING for more details.
   7  */
   8
   9 #define pr_fmt(fmt)     "kexec: " fmt
  10
  11 #include <linux/mm.h>
  12 #include <linux/kexec.h>
  13 #include <linux/string.h>
  14 #include <linux/gfp.h>
  15 #include <linux/reboot.h>
  16 #include <linux/numa.h>
  17 #include <linux/ftrace.h>
  18 #include <linux/io.h>
  19 #include <linux/suspend.h>
  20 #include <linux/vmalloc.h>
  21
  22 #include <asm/init.h>
  23 #include <asm/pgtable.h>
  24 #include <asm/tlbflush.h>
  25 #include <asm/mmu_context.h>
  26 #include <asm/io_apic.h>
  27 #include <asm/debugreg.h>
  28 #include <asm/kexec-bzimage64.h>
  29 #include <asm/setup.h>
  30 #include <asm/set_memory.h>
  31
  32 #ifdef CONFIG_KEXEC_FILE
  33 static struct kexec_file_ops *kexec_file_loaders[] = {
  34                 &kexec_bzImage64_ops,
  35 };
  36 #endif
  37
  38 static void free_transition_pgtable(struct kimage *image)
  39 {
  40         free_page((unsigned long)image->arch.p4d);
  41         free_page((unsigned long)image->arch.pud);
  42         free_page((unsigned long)image->arch.pmd);
  43         free_page((unsigned long)image->arch.pte);
  44 }
  45
  46 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
  47 {
  48         p4d_t *p4d;
  49         pud_t *pud;
  50         pmd_t *pmd;
  51         pte_t *pte;
  52         unsigned long vaddr, paddr;
  53         int result = -ENOMEM;
  54
  55         vaddr = (unsigned long)relocate_kernel;
  56         paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
  57         pgd += pgd_index(vaddr);
  58         if (!pgd_present(*pgd)) {
  59                 p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
  60                 if (!p4d)
  61                         goto err;
  62                 image->arch.p4d = p4d;
  63                 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
  64         }
  65         p4d = p4d_offset(pgd, vaddr);
  66         if (!p4d_present(*p4d)) {
  67                 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
  68                 if (!pud)
  69                         goto err;
  70                 image->arch.pud = pud;
  71                 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
  72         }
  73         pud = pud_offset(p4d, vaddr);
  74         if (!pud_present(*pud)) {
  75                 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
  76                 if (!pmd)
  77                         goto err;
  78                 image->arch.pmd = pmd;
  79                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
  80         }
  81         pmd = pmd_offset(pud, vaddr);
  82         if (!pmd_present(*pmd)) {
  83                 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
  84                 if (!pte)
  85                         goto err;
  86                 image->arch.pte = pte;
  87                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
  88         }
  89         pte = pte_offset_kernel(pmd, vaddr);
  90         set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC_NOENC));
  91         return 0;
  92 err:
  93         free_transition_pgtable(image);
  94         return result;
  95 }
  96
  97 static void *alloc_pgt_page(void *data)
  98 {
  99         struct kimage *image = (struct kimage *)data;
 100         struct page *page;
 101         void *p = NULL;
 102
 103         page = kimage_alloc_control_pages(image, 0);
 104         if (page) {
 105                 p = page_address(page);
 106                 clear_page(p);
 107         }
 108
 109         return p;
 110 }
 111
 112 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
 113 {
 114         struct x86_mapping_info info = {
 115                 .alloc_pgt_page = alloc_pgt_page,
 116                 .context        = image,
 117                 .page_flag      = __PAGE_KERNEL_LARGE_EXEC,
 118                 .kernpg_flag    = _KERNPG_TABLE_NOENC,
 119         };
 120         unsigned long mstart, mend;
 121         pgd_t *level4p;
 122         int result;
 123         int i;
 124
 125         level4p = (pgd_t *)__va(start_pgtable);
 126         clear_page(level4p);
 127
 128         if (direct_gbpages)
 129                 info.direct_gbpages = true;
 130
 131         for (i = 0; i < nr_pfn_mapped; i++) {
 132                 mstart = pfn_mapped[i].start << PAGE_SHIFT;
 133                 mend   = pfn_mapped[i].end << PAGE_SHIFT;
 134
 135                 result = kernel_ident_mapping_init(&info,
 136                                                  level4p, mstart, mend);
 137                 if (result)
 138                         return result;
 139         }
 140
 141         /*
 142          * segments's mem ranges could be outside 0 ~ max_pfn,
 143          * for example when jump back to original kernel from kexeced kernel.
 144          * or first kernel is booted with user mem map, and second kernel
 145          * could be loaded out of that range.
 146          */
 147         for (i = 0; i < image->nr_segments; i++) {
 148                 mstart = image->segment[i].mem;
 149                 mend   = mstart + image->segment[i].memsz;
 150
 151                 result = kernel_ident_mapping_init(&info,
 152                                                  level4p, mstart, mend);
 153
 154                 if (result)
 155                         return result;
 156         }
 157
 158         return init_transition_pgtable(image, level4p);
 159 }
 160
 161 static void set_idt(void *newidt, u16 limit)
 162 {
 163         struct desc_ptr curidt;
 164
 165         /* x86-64 supports unaliged loads & stores */
 166         curidt.size    = limit;
 167         curidt.address = (unsigned long)newidt;
 168
 169         __asm__ __volatile__ (
 170                 "lidtq %0\n"
 171                 : : "m" (curidt)
 172                 );
 173 };
 174
 175
 176 static void set_gdt(void *newgdt, u16 limit)
 177 {
 178         struct desc_ptr curgdt;
 179
 180         /* x86-64 supports unaligned loads & stores */
 181         curgdt.size    = limit;
 182         curgdt.address = (unsigned long)newgdt;
 183
 184         __asm__ __volatile__ (
 185                 "lgdtq %0\n"
 186                 : : "m" (curgdt)
 187                 );
 188 };
 189
 190 static void load_segments(void)
 191 {
 192         __asm__ __volatile__ (
 193                 "\tmovl %0,%%ds\n"
 194                 "\tmovl %0,%%es\n"
 195                 "\tmovl %0,%%ss\n"
 196                 "\tmovl %0,%%fs\n"
 197                 "\tmovl %0,%%gs\n"
 198                 : : "a" (__KERNEL_DS) : "memory"
 199                 );
 200 }
 201
 202 #ifdef CONFIG_KEXEC_FILE
 203 /* Update purgatory as needed after various image segments have been prepared */
 204 static int arch_update_purgatory(struct kimage *image)
 205 {
 206         int ret = 0;
 207
 208         if (!image->file_mode)
 209                 return 0;
 210
 211         /* Setup copying of backup region */
 212         if (image->type == KEXEC_TYPE_CRASH) {
 213                 ret = kexec_purgatory_get_set_symbol(image,
 214                                 "purgatory_backup_dest",
 215                                 &image->arch.backup_load_addr,
 216                                 sizeof(image->arch.backup_load_addr), 0);
 217                 if (ret)
 218                         return ret;
 219
 220                 ret = kexec_purgatory_get_set_symbol(image,
 221                                 "purgatory_backup_src",
 222                                 &image->arch.backup_src_start,
 223                                 sizeof(image->arch.backup_src_start), 0);
 224                 if (ret)
 225                         return ret;
 226
 227                 ret = kexec_purgatory_get_set_symbol(image,
 228                                 "purgatory_backup_sz",
 229                                 &image->arch.backup_src_sz,
 230                                 sizeof(image->arch.backup_src_sz), 0);
 231                 if (ret)
 232                         return ret;
 233         }
 234
 235         return ret;
 236 }
 237 #else /* !CONFIG_KEXEC_FILE */
 238 static inline int arch_update_purgatory(struct kimage *image)
 239 {
 240         return 0;
 241 }
 242 #endif /* CONFIG_KEXEC_FILE */
 243
 244 int machine_kexec_prepare(struct kimage *image)
 245 {
 246         unsigned long start_pgtable;
 247         int result;
 248
 249         /* Calculate the offsets */
 250         start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
 251
 252         /* Setup the identity mapped 64bit page table */
 253         result = init_pgtable(image, start_pgtable);
 254         if (result)
 255                 return result;
 256
 257         /* update purgatory as needed */
 258         result = arch_update_purgatory(image);
 259         if (result)
 260                 return result;
 261
 262         return 0;
 263 }
 264
 265 void machine_kexec_cleanup(struct kimage *image)
 266 {
 267         free_transition_pgtable(image);
 268 }
 269
 270 /*
 271  * Do not allocate memory (or fail in any way) in machine_kexec().
 272  * We are past the point of no return, committed to rebooting now.
 273  */
 274 void machine_kexec(struct kimage *image)
 275 {
 276         unsigned long page_list[PAGES_NR];
 277         void *control_page;
 278         int save_ftrace_enabled;
 279
 280 #ifdef CONFIG_KEXEC_JUMP
 281         if (image->preserve_context)
 282                 save_processor_state();
 283 #endif
 284
 285         save_ftrace_enabled = __ftrace_enabled_save();
 286
 287         /* Interrupts aren't acceptable while we reboot */
 288         local_irq_disable();
 289         hw_breakpoint_disable();
 290
 291         if (image->preserve_context) {
 292 #ifdef CONFIG_X86_IO_APIC
 293                 /*
 294                  * We need to put APICs in legacy mode so that we can
 295                  * get timer interrupts in second kernel. kexec/kdump
 296                  * paths already have calls to disable_IO_APIC() in
 297                  * one form or other. kexec jump path also need
 298                  * one.
 299                  */
 300                 disable_IO_APIC();
 301 #endif
 302         }
 303
 304         control_page = page_address(image->control_code_page) + PAGE_SIZE;
 305         memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
 306
 307         page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
 308         page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
 309         page_list[PA_TABLE_PAGE] =
 310           (unsigned long)__pa(page_address(image->control_code_page));
 311
 312         if (image->type == KEXEC_TYPE_DEFAULT)
 313                 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
 314                                                 << PAGE_SHIFT);
 315
 316         /*
 317          * The segment registers are funny things, they have both a
 318          * visible and an invisible part.  Whenever the visible part is
 319          * set to a specific selector, the invisible part is loaded
 320          * with from a table in memory.  At no other time is the
 321          * descriptor table in memory accessed.
 322          *
 323          * I take advantage of this here by force loading the
 324          * segments, before I zap the gdt with an invalid value.
 325          */
 326         load_segments();
 327         /*
 328          * The gdt & idt are now invalid.
 329          * If you want to load them you must set up your own idt & gdt.
 330          */
 331         set_gdt(phys_to_virt(0), 0);
 332         set_idt(phys_to_virt(0), 0);
 333
 334         /* now call it */
 335         image->start = relocate_kernel((unsigned long)image->head,
 336                                        (unsigned long)page_list,
 337                                        image->start,
 338                                        image->preserve_context,
 339                                        sme_active());
 340
 341 #ifdef CONFIG_KEXEC_JUMP
 342         if (image->preserve_context)
 343                 restore_processor_state();
 344 #endif
 345
 346         __ftrace_enabled_restore(save_ftrace_enabled);
 347 }
 348
 349 void arch_crash_save_vmcoreinfo(void)
 350 {
 351         VMCOREINFO_NUMBER(phys_base);
 352         VMCOREINFO_SYMBOL(init_top_pgt);
 353
 354 #ifdef CONFIG_NUMA
 355         VMCOREINFO_SYMBOL(node_data);
 356         VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
 357 #endif
 358         vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
 359                               kaslr_offset());
 360         VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
 361 }
 362
 363 /* arch-dependent functionality related to kexec file-based syscall */
 364
 365 #ifdef CONFIG_KEXEC_FILE
 366 int arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
 367                                   unsigned long buf_len)
 368 {
 369         int i, ret = -ENOEXEC;
 370         struct kexec_file_ops *fops;
 371
 372         for (i = 0; i < ARRAY_SIZE(kexec_file_loaders); i++) {
 373                 fops = kexec_file_loaders[i];
 374                 if (!fops || !fops->probe)
 375                         continue;
 376
 377                 ret = fops->probe(buf, buf_len);
 378                 if (!ret) {
 379                         image->fops = fops;
 380                         return ret;
 381                 }
 382         }
 383
 384         return ret;
 385 }
 386
 387 void *arch_kexec_kernel_image_load(struct kimage *image)
 388 {
 389         vfree(image->arch.elf_headers);
 390         image->arch.elf_headers = NULL;
 391
 392         if (!image->fops || !image->fops->load)
 393                 return ERR_PTR(-ENOEXEC);
 394
 395         return image->fops->load(image, image->kernel_buf,
 396                                  image->kernel_buf_len, image->initrd_buf,
 397                                  image->initrd_buf_len, image->cmdline_buf,
 398                                  image->cmdline_buf_len);
 399 }
 400
 401 int arch_kimage_file_post_load_cleanup(struct kimage *image)
 402 {
 403         if (!image->fops || !image->fops->cleanup)
 404                 return 0;
 405
 406         return image->fops->cleanup(image->image_loader_data);
 407 }
 408
 409 #ifdef CONFIG_KEXEC_VERIFY_SIG
 410 int arch_kexec_kernel_verify_sig(struct kimage *image, void *kernel,
 411                                  unsigned long kernel_len)
 412 {
 413         if (!image->fops || !image->fops->verify_sig) {
 414                 pr_debug("kernel loader does not support signature verification.");
 415                 return -EKEYREJECTED;
 416         }
 417
 418         return image->fops->verify_sig(kernel, kernel_len);
 419 }
 420 #endif
 421
 422 /*
 423  * Apply purgatory relocations.
 424  *
 425  * ehdr: Pointer to elf headers
 426  * sechdrs: Pointer to section headers.
 427  * relsec: section index of SHT_RELA section.
 428  *
 429  * TODO: Some of the code belongs to generic code. Move that in kexec.c.
 430  */
 431 int arch_kexec_apply_relocations_add(const Elf64_Ehdr *ehdr,
 432                                      Elf64_Shdr *sechdrs, unsigned int relsec)
 433 {
 434         unsigned int i;
 435         Elf64_Rela *rel;
 436         Elf64_Sym *sym;
 437         void *location;
 438         Elf64_Shdr *section, *symtabsec;
 439         unsigned long address, sec_base, value;
 440         const char *strtab, *name, *shstrtab;
 441
 442         /*
 443          * ->sh_offset has been modified to keep the pointer to section
 444          * contents in memory
 445          */
 446         rel = (void *)sechdrs[relsec].sh_offset;
 447
 448         /* Section to which relocations apply */
 449         section = &sechdrs[sechdrs[relsec].sh_info];
 450
 451         pr_debug("Applying relocate section %u to %u\n", relsec,
 452                  sechdrs[relsec].sh_info);
 453
 454         /* Associated symbol table */
 455         symtabsec = &sechdrs[sechdrs[relsec].sh_link];
 456
 457         /* String table */
 458         if (symtabsec->sh_link >= ehdr->e_shnum) {
 459                 /* Invalid strtab section number */
 460                 pr_err("Invalid string table section index %d\n",
 461                        symtabsec->sh_link);
 462                 return -ENOEXEC;
 463         }
 464
 465         strtab = (char *)sechdrs[symtabsec->sh_link].sh_offset;
 466
 467         /* section header string table */
 468         shstrtab = (char *)sechdrs[ehdr->e_shstrndx].sh_offset;
 469
 470         for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rel); i++) {
 471
 472                 /*
 473                  * rel[i].r_offset contains byte offset from beginning
 474                  * of section to the storage unit affected.
 475                  *
 476                  * This is location to update (->sh_offset). This is temporary
 477                  * buffer where section is currently loaded. This will finally
 478                  * be loaded to a different address later, pointed to by
 479                  * ->sh_addr. kexec takes care of moving it
 480                  *  (kexec_load_segment()).
 481                  */
 482                 location = (void *)(section->sh_offset + rel[i].r_offset);
 483
 484                 /* Final address of the location */
 485                 address = section->sh_addr + rel[i].r_offset;
 486
 487                 /*
 488                  * rel[i].r_info contains information about symbol table index
 489                  * w.r.t which relocation must be made and type of relocation
 490                  * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
 491                  * these respectively.
 492                  */
 493                 sym = (Elf64_Sym *)symtabsec->sh_offset +
 494                                 ELF64_R_SYM(rel[i].r_info);
 495
 496                 if (sym->st_name)
 497                         name = strtab + sym->st_name;
 498                 else
 499                         name = shstrtab + sechdrs[sym->st_shndx].sh_name;
 500
 501                 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
 502                          name, sym->st_info, sym->st_shndx, sym->st_value,
 503                          sym->st_size);
 504
 505                 if (sym->st_shndx == SHN_UNDEF) {
 506                         pr_err("Undefined symbol: %s\n", name);
 507                         return -ENOEXEC;
 508                 }
 509
 510                 if (sym->st_shndx == SHN_COMMON) {
 511                         pr_err("symbol '%s' in common section\n", name);
 512                         return -ENOEXEC;
 513                 }
 514
 515                 if (sym->st_shndx == SHN_ABS)
 516                         sec_base = 0;
 517                 else if (sym->st_shndx >= ehdr->e_shnum) {
 518                         pr_err("Invalid section %d for symbol %s\n",
 519                                sym->st_shndx, name);
 520                         return -ENOEXEC;
 521                 } else
 522                         sec_base = sechdrs[sym->st_shndx].sh_addr;
 523
 524                 value = sym->st_value;
 525                 value += sec_base;
 526                 value += rel[i].r_addend;
 527
 528                 switch (ELF64_R_TYPE(rel[i].r_info)) {
 529                 case R_X86_64_NONE:
 530                         break;
 531                 case R_X86_64_64:
 532                         *(u64 *)location = value;
 533                         break;
 534                 case R_X86_64_32:
 535                         *(u32 *)location = value;
 536                         if (value != *(u32 *)location)
 537                                 goto overflow;
 538                         break;
 539                 case R_X86_64_32S:
 540                         *(s32 *)location = value;
 541                         if ((s64)value != *(s32 *)location)
 542                                 goto overflow;
 543                         break;
 544                 case R_X86_64_PC32:
 545                 case R_X86_64_PLT32:
 546                         value -= (u64)address;
 547                         *(u32 *)location = value;
 548                         break;
 549                 default:
 550                         pr_err("Unknown rela relocation: %llu\n",
 551                                ELF64_R_TYPE(rel[i].r_info));
 552                         return -ENOEXEC;
 553                 }
 554         }
 555         return 0;
 556
 557 overflow:
 558         pr_err("Overflow in relocation type %d value 0x%lx\n",
 559                (int)ELF64_R_TYPE(rel[i].r_info), value);
 560         return -ENOEXEC;
 561 }
 562 #endif /* CONFIG_KEXEC_FILE */
 563
 564 static int
 565 kexec_mark_range(unsigned long start, unsigned long end, bool protect)
 566 {
 567         struct page *page;
 568         unsigned int nr_pages;
 569
 570         /*
 571          * For physical range: [start, end]. We must skip the unassigned
 572          * crashk resource with zero-valued "end" member.
 573          */
 574         if (!end || start > end)
 575                 return 0;
 576
 577         page = pfn_to_page(start >> PAGE_SHIFT);
 578         nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
 579         if (protect)
 580                 return set_pages_ro(page, nr_pages);
 581         else
 582                 return set_pages_rw(page, nr_pages);
 583 }
 584
 585 static void kexec_mark_crashkres(bool protect)
 586 {
 587         unsigned long control;
 588
 589         kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
 590
 591         /* Don't touch the control code page used in crash_kexec().*/
 592         control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
 593         /* Control code page is located in the 2nd page. */
 594         kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
 595         control += KEXEC_CONTROL_PAGE_SIZE;
 596         kexec_mark_range(control, crashk_res.end, protect);
 597 }
 598
 599 void arch_kexec_protect_crashkres(void)
 600 {
 601         kexec_mark_crashkres(true);
 602 }
 603
 604 void arch_kexec_unprotect_crashkres(void)
 605 {
 606         kexec_mark_crashkres(false);
 607 }
 608
 609 int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
 610 {
 611         /*
 612          * If SME is active we need to be sure that kexec pages are
 613          * not encrypted because when we boot to the new kernel the
 614          * pages won't be accessed encrypted (initially).
 615          */
 616         return set_memory_decrypted((unsigned long)vaddr, pages);
 617 }
 618
 619 void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
 620 {
 621         /*
 622          * If SME is active we need to reset the pages back to being
 623          * an encrypted mapping before freeing them.
 624          */
 625         set_memory_encrypted((unsigned long)vaddr, pages);
 626 }