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
  21 #include <asm/init.h>
  22 #include <asm/pgtable.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
  29 #ifdef CONFIG_KEXEC_FILE
  30 static struct kexec_file_ops *kexec_file_loaders[] = {
  31                 &kexec_bzImage64_ops,
  32 };
  33 #endif
  34
  35 static void free_transition_pgtable(struct kimage *image)
  36 {
  37         free_page((unsigned long)image->arch.pud);
  38         free_page((unsigned long)image->arch.pmd);
  39         free_page((unsigned long)image->arch.pte);
  40 }
  41
  42 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
  43 {
  44         pud_t *pud;
  45         pmd_t *pmd;
  46         pte_t *pte;
  47         unsigned long vaddr, paddr;
  48         int result = -ENOMEM;
  49
  50         vaddr = (unsigned long)relocate_kernel;
  51         paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
  52         pgd += pgd_index(vaddr);
  53         if (!pgd_present(*pgd)) {
  54                 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
  55                 if (!pud)
  56                         goto err;
  57                 image->arch.pud = pud;
  58                 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
  59         }
  60         pud = pud_offset(pgd, vaddr);
  61         if (!pud_present(*pud)) {
  62                 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
  63                 if (!pmd)
  64                         goto err;
  65                 image->arch.pmd = pmd;
  66                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
  67         }
  68         pmd = pmd_offset(pud, vaddr);
  69         if (!pmd_present(*pmd)) {
  70                 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
  71                 if (!pte)
  72                         goto err;
  73                 image->arch.pte = pte;
  74                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
  75         }
  76         pte = pte_offset_kernel(pmd, vaddr);
  77         set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
  78         return 0;
  79 err:
  80         free_transition_pgtable(image);
  81         return result;
  82 }
  83
  84 static void *alloc_pgt_page(void *data)
  85 {
  86         struct kimage *image = (struct kimage *)data;
  87         struct page *page;
  88         void *p = NULL;
  89
  90         page = kimage_alloc_control_pages(image, 0);
  91         if (page) {
  92                 p = page_address(page);
  93                 clear_page(p);
  94         }
  95
  96         return p;
  97 }
  98
  99 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
 100 {
 101         struct x86_mapping_info info = {
 102                 .alloc_pgt_page = alloc_pgt_page,
 103                 .context        = image,
 104                 .pmd_flag       = __PAGE_KERNEL_LARGE_EXEC,
 105         };
 106         unsigned long mstart, mend;
 107         pgd_t *level4p;
 108         int result;
 109         int i;
 110
 111         level4p = (pgd_t *)__va(start_pgtable);
 112         clear_page(level4p);
 113         for (i = 0; i < nr_pfn_mapped; i++) {
 114                 mstart = pfn_mapped[i].start << PAGE_SHIFT;
 115                 mend   = pfn_mapped[i].end << PAGE_SHIFT;
 116
 117                 result = kernel_ident_mapping_init(&info,
 118                                                  level4p, mstart, mend);
 119                 if (result)
 120                         return result;
 121         }
 122
 123         /*
 124          * segments's mem ranges could be outside 0 ~ max_pfn,
 125          * for example when jump back to original kernel from kexeced kernel.
 126          * or first kernel is booted with user mem map, and second kernel
 127          * could be loaded out of that range.
 128          */
 129         for (i = 0; i < image->nr_segments; i++) {
 130                 mstart = image->segment[i].mem;
 131                 mend   = mstart + image->segment[i].memsz;
 132
 133                 result = kernel_ident_mapping_init(&info,
 134                                                  level4p, mstart, mend);
 135
 136                 if (result)
 137                         return result;
 138         }
 139
 140         return init_transition_pgtable(image, level4p);
 141 }
 142
 143 static void set_idt(void *newidt, u16 limit)
 144 {
 145         struct desc_ptr curidt;
 146
 147         /* x86-64 supports unaliged loads & stores */
 148         curidt.size    = limit;
 149         curidt.address = (unsigned long)newidt;
 150
 151         __asm__ __volatile__ (
 152                 "lidtq %0\n"
 153                 : : "m" (curidt)
 154                 );
 155 };
 156
 157
 158 static void set_gdt(void *newgdt, u16 limit)
 159 {
 160         struct desc_ptr curgdt;
 161
 162         /* x86-64 supports unaligned loads & stores */
 163         curgdt.size    = limit;
 164         curgdt.address = (unsigned long)newgdt;
 165
 166         __asm__ __volatile__ (
 167                 "lgdtq %0\n"
 168                 : : "m" (curgdt)
 169                 );
 170 };
 171
 172 static void load_segments(void)
 173 {
 174         __asm__ __volatile__ (
 175                 "\tmovl %0,%%ds\n"
 176                 "\tmovl %0,%%es\n"
 177                 "\tmovl %0,%%ss\n"
 178                 "\tmovl %0,%%fs\n"
 179                 "\tmovl %0,%%gs\n"
 180                 : : "a" (__KERNEL_DS) : "memory"
 181                 );
 182 }
 183
 184 #ifdef CONFIG_KEXEC_FILE
 185 /* Update purgatory as needed after various image segments have been prepared */
 186 static int arch_update_purgatory(struct kimage *image)
 187 {
 188         int ret = 0;
 189
 190         if (!image->file_mode)
 191                 return 0;
 192
 193         /* Setup copying of backup region */
 194         if (image->type == KEXEC_TYPE_CRASH) {
 195                 ret = kexec_purgatory_get_set_symbol(image, "backup_dest",
 196                                 &image->arch.backup_load_addr,
 197                                 sizeof(image->arch.backup_load_addr), 0);
 198                 if (ret)
 199                         return ret;
 200
 201                 ret = kexec_purgatory_get_set_symbol(image, "backup_src",
 202                                 &image->arch.backup_src_start,
 203                                 sizeof(image->arch.backup_src_start), 0);
 204                 if (ret)
 205                         return ret;
 206
 207                 ret = kexec_purgatory_get_set_symbol(image, "backup_sz",
 208                                 &image->arch.backup_src_sz,
 209                                 sizeof(image->arch.backup_src_sz), 0);
 210                 if (ret)
 211                         return ret;
 212         }
 213
 214         return ret;
 215 }
 216 #else /* !CONFIG_KEXEC_FILE */
 217 static inline int arch_update_purgatory(struct kimage *image)
 218 {
 219         return 0;
 220 }
 221 #endif /* CONFIG_KEXEC_FILE */
 222
 223 int machine_kexec_prepare(struct kimage *image)
 224 {
 225         unsigned long start_pgtable;
 226         int result;
 227
 228         /* Calculate the offsets */
 229         start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
 230
 231         /* Setup the identity mapped 64bit page table */
 232         result = init_pgtable(image, start_pgtable);
 233         if (result)
 234                 return result;
 235
 236         /* update purgatory as needed */
 237         result = arch_update_purgatory(image);
 238         if (result)
 239                 return result;
 240
 241         return 0;
 242 }
 243
 244 void machine_kexec_cleanup(struct kimage *image)
 245 {
 246         free_transition_pgtable(image);
 247 }
 248
 249 /*
 250  * Do not allocate memory (or fail in any way) in machine_kexec().
 251  * We are past the point of no return, committed to rebooting now.
 252  */
 253 void machine_kexec(struct kimage *image)
 254 {
 255         unsigned long page_list[PAGES_NR];
 256         void *control_page;
 257         int save_ftrace_enabled;
 258
 259 #ifdef CONFIG_KEXEC_JUMP
 260         if (image->preserve_context)
 261                 save_processor_state();
 262 #endif
 263
 264         save_ftrace_enabled = __ftrace_enabled_save();
 265
 266         /* Interrupts aren't acceptable while we reboot */
 267         local_irq_disable();
 268         hw_breakpoint_disable();
 269
 270         if (image->preserve_context) {
 271 #ifdef CONFIG_X86_IO_APIC
 272                 /*
 273                  * We need to put APICs in legacy mode so that we can
 274                  * get timer interrupts in second kernel. kexec/kdump
 275                  * paths already have calls to disable_IO_APIC() in
 276                  * one form or other. kexec jump path also need
 277                  * one.
 278                  */
 279                 disable_IO_APIC();
 280 #endif
 281         }
 282
 283         control_page = page_address(image->control_code_page) + PAGE_SIZE;
 284         memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
 285
 286         page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
 287         page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
 288         page_list[PA_TABLE_PAGE] =
 289           (unsigned long)__pa(page_address(image->control_code_page));
 290
 291         if (image->type == KEXEC_TYPE_DEFAULT)
 292                 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
 293                                                 << PAGE_SHIFT);
 294
 295         /*
 296          * The segment registers are funny things, they have both a
 297          * visible and an invisible part.  Whenever the visible part is
 298          * set to a specific selector, the invisible part is loaded
 299          * with from a table in memory.  At no other time is the
 300          * descriptor table in memory accessed.
 301          *
 302          * I take advantage of this here by force loading the
 303          * segments, before I zap the gdt with an invalid value.
 304          */
 305         load_segments();
 306         /*
 307          * The gdt & idt are now invalid.
 308          * If you want to load them you must set up your own idt & gdt.
 309          */
 310         set_gdt(phys_to_virt(0), 0);
 311         set_idt(phys_to_virt(0), 0);
 312
 313         /* now call it */
 314         image->start = relocate_kernel((unsigned long)image->head,
 315                                        (unsigned long)page_list,
 316                                        image->start,
 317                                        image->preserve_context);
 318
 319 #ifdef CONFIG_KEXEC_JUMP
 320         if (image->preserve_context)
 321                 restore_processor_state();
 322 #endif
 323
 324         __ftrace_enabled_restore(save_ftrace_enabled);
 325 }
 326
 327 void arch_crash_save_vmcoreinfo(void)
 328 {
 329         VMCOREINFO_SYMBOL(phys_base);
 330         VMCOREINFO_SYMBOL(init_level4_pgt);
 331
 332 #ifdef CONFIG_NUMA
 333         VMCOREINFO_SYMBOL(node_data);
 334         VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
 335 #endif
 336         vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
 337                               (unsigned long)&_text - __START_KERNEL);
 338 }
 339
 340 /* arch-dependent functionality related to kexec file-based syscall */
 341
 342 #ifdef CONFIG_KEXEC_FILE
 343 int arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
 344                                   unsigned long buf_len)
 345 {
 346         int i, ret = -ENOEXEC;
 347         struct kexec_file_ops *fops;
 348
 349         for (i = 0; i < ARRAY_SIZE(kexec_file_loaders); i++) {
 350                 fops = kexec_file_loaders[i];
 351                 if (!fops || !fops->probe)
 352                         continue;
 353
 354                 ret = fops->probe(buf, buf_len);
 355                 if (!ret) {
 356                         image->fops = fops;
 357                         return ret;
 358                 }
 359         }
 360
 361         return ret;
 362 }
 363
 364 void *arch_kexec_kernel_image_load(struct kimage *image)
 365 {
 366         vfree(image->arch.elf_headers);
 367         image->arch.elf_headers = NULL;
 368
 369         if (!image->fops || !image->fops->load)
 370                 return ERR_PTR(-ENOEXEC);
 371
 372         return image->fops->load(image, image->kernel_buf,
 373                                  image->kernel_buf_len, image->initrd_buf,
 374                                  image->initrd_buf_len, image->cmdline_buf,
 375                                  image->cmdline_buf_len);
 376 }
 377
 378 int arch_kimage_file_post_load_cleanup(struct kimage *image)
 379 {
 380         if (!image->fops || !image->fops->cleanup)
 381                 return 0;
 382
 383         return image->fops->cleanup(image->image_loader_data);
 384 }
 385
 386 int arch_kexec_kernel_verify_sig(struct kimage *image, void *kernel,
 387                                  unsigned long kernel_len)
 388 {
 389         if (!image->fops || !image->fops->verify_sig) {
 390                 pr_debug("kernel loader does not support signature verification.");
 391                 return -EKEYREJECTED;
 392         }
 393
 394         return image->fops->verify_sig(kernel, kernel_len);
 395 }
 396
 397 /*
 398  * Apply purgatory relocations.
 399  *
 400  * ehdr: Pointer to elf headers
 401  * sechdrs: Pointer to section headers.
 402  * relsec: section index of SHT_RELA section.
 403  *
 404  * TODO: Some of the code belongs to generic code. Move that in kexec.c.
 405  */
 406 int arch_kexec_apply_relocations_add(const Elf64_Ehdr *ehdr,
 407                                      Elf64_Shdr *sechdrs, unsigned int relsec)
 408 {
 409         unsigned int i;
 410         Elf64_Rela *rel;
 411         Elf64_Sym *sym;
 412         void *location;
 413         Elf64_Shdr *section, *symtabsec;
 414         unsigned long address, sec_base, value;
 415         const char *strtab, *name, *shstrtab;
 416
 417         /*
 418          * ->sh_offset has been modified to keep the pointer to section
 419          * contents in memory
 420          */
 421         rel = (void *)sechdrs[relsec].sh_offset;
 422
 423         /* Section to which relocations apply */
 424         section = &sechdrs[sechdrs[relsec].sh_info];
 425
 426         pr_debug("Applying relocate section %u to %u\n", relsec,
 427                  sechdrs[relsec].sh_info);
 428
 429         /* Associated symbol table */
 430         symtabsec = &sechdrs[sechdrs[relsec].sh_link];
 431
 432         /* String table */
 433         if (symtabsec->sh_link >= ehdr->e_shnum) {
 434                 /* Invalid strtab section number */
 435                 pr_err("Invalid string table section index %d\n",
 436                        symtabsec->sh_link);
 437                 return -ENOEXEC;
 438         }
 439
 440         strtab = (char *)sechdrs[symtabsec->sh_link].sh_offset;
 441
 442         /* section header string table */
 443         shstrtab = (char *)sechdrs[ehdr->e_shstrndx].sh_offset;
 444
 445         for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rel); i++) {
 446
 447                 /*
 448                  * rel[i].r_offset contains byte offset from beginning
 449                  * of section to the storage unit affected.
 450                  *
 451                  * This is location to update (->sh_offset). This is temporary
 452                  * buffer where section is currently loaded. This will finally
 453                  * be loaded to a different address later, pointed to by
 454                  * ->sh_addr. kexec takes care of moving it
 455                  *  (kexec_load_segment()).
 456                  */
 457                 location = (void *)(section->sh_offset + rel[i].r_offset);
 458
 459                 /* Final address of the location */
 460                 address = section->sh_addr + rel[i].r_offset;
 461
 462                 /*
 463                  * rel[i].r_info contains information about symbol table index
 464                  * w.r.t which relocation must be made and type of relocation
 465                  * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
 466                  * these respectively.
 467                  */
 468                 sym = (Elf64_Sym *)symtabsec->sh_offset +
 469                                 ELF64_R_SYM(rel[i].r_info);
 470
 471                 if (sym->st_name)
 472                         name = strtab + sym->st_name;
 473                 else
 474                         name = shstrtab + sechdrs[sym->st_shndx].sh_name;
 475
 476                 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
 477                          name, sym->st_info, sym->st_shndx, sym->st_value,
 478                          sym->st_size);
 479
 480                 if (sym->st_shndx == SHN_UNDEF) {
 481                         pr_err("Undefined symbol: %s\n", name);
 482                         return -ENOEXEC;
 483                 }
 484
 485                 if (sym->st_shndx == SHN_COMMON) {
 486                         pr_err("symbol '%s' in common section\n", name);
 487                         return -ENOEXEC;
 488                 }
 489
 490                 if (sym->st_shndx == SHN_ABS)
 491                         sec_base = 0;
 492                 else if (sym->st_shndx >= ehdr->e_shnum) {
 493                         pr_err("Invalid section %d for symbol %s\n",
 494                                sym->st_shndx, name);
 495                         return -ENOEXEC;
 496                 } else
 497                         sec_base = sechdrs[sym->st_shndx].sh_addr;
 498
 499                 value = sym->st_value;
 500                 value += sec_base;
 501                 value += rel[i].r_addend;
 502
 503                 switch (ELF64_R_TYPE(rel[i].r_info)) {
 504                 case R_X86_64_NONE:
 505                         break;
 506                 case R_X86_64_64:
 507                         *(u64 *)location = value;
 508                         break;
 509                 case R_X86_64_32:
 510                         *(u32 *)location = value;
 511                         if (value != *(u32 *)location)
 512                                 goto overflow;
 513                         break;
 514                 case R_X86_64_32S:
 515                         *(s32 *)location = value;
 516                         if ((s64)value != *(s32 *)location)
 517                                 goto overflow;
 518                         break;
 519                 case R_X86_64_PC32:
 520                         value -= (u64)address;
 521                         *(u32 *)location = value;
 522                         break;
 523                 default:
 524                         pr_err("Unknown rela relocation: %llu\n",
 525                                ELF64_R_TYPE(rel[i].r_info));
 526                         return -ENOEXEC;
 527                 }
 528         }
 529         return 0;
 530
 531 overflow:
 532         pr_err("Overflow in relocation type %d value 0x%lx\n",
 533                (int)ELF64_R_TYPE(rel[i].r_info), value);
 534         return -ENOEXEC;
 535 }
 536 #endif /* CONFIG_KEXEC_FILE */