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/debugreg.h>
  26 #include <asm/kexec-bzimage64.h>
  27
  28 static struct kexec_file_ops *kexec_file_loaders[] = {
  29                 &kexec_bzImage64_ops,
  30 };
  31
  32 static void free_transition_pgtable(struct kimage *image)
  33 {
  34         free_page((unsigned long)image->arch.pud);
  35         free_page((unsigned long)image->arch.pmd);
  36         free_page((unsigned long)image->arch.pte);
  37 }
  38
  39 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
  40 {
  41         pud_t *pud;
  42         pmd_t *pmd;
  43         pte_t *pte;
  44         unsigned long vaddr, paddr;
  45         int result = -ENOMEM;
  46
  47         vaddr = (unsigned long)relocate_kernel;
  48         paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
  49         pgd += pgd_index(vaddr);
  50         if (!pgd_present(*pgd)) {
  51                 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
  52                 if (!pud)
  53                         goto err;
  54                 image->arch.pud = pud;
  55                 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
  56         }
  57         pud = pud_offset(pgd, vaddr);
  58         if (!pud_present(*pud)) {
  59                 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
  60                 if (!pmd)
  61                         goto err;
  62                 image->arch.pmd = pmd;
  63                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
  64         }
  65         pmd = pmd_offset(pud, vaddr);
  66         if (!pmd_present(*pmd)) {
  67                 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
  68                 if (!pte)
  69                         goto err;
  70                 image->arch.pte = pte;
  71                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
  72         }
  73         pte = pte_offset_kernel(pmd, vaddr);
  74         set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
  75         return 0;
  76 err:
  77         free_transition_pgtable(image);
  78         return result;
  79 }
  80
  81 static void *alloc_pgt_page(void *data)
  82 {
  83         struct kimage *image = (struct kimage *)data;
  84         struct page *page;
  85         void *p = NULL;
  86
  87         page = kimage_alloc_control_pages(image, 0);
  88         if (page) {
  89                 p = page_address(page);
  90                 clear_page(p);
  91         }
  92
  93         return p;
  94 }
  95
  96 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
  97 {
  98         struct x86_mapping_info info = {
  99                 .alloc_pgt_page = alloc_pgt_page,
 100                 .context        = image,
 101                 .pmd_flag       = __PAGE_KERNEL_LARGE_EXEC,
 102         };
 103         unsigned long mstart, mend;
 104         pgd_t *level4p;
 105         int result;
 106         int i;
 107
 108         level4p = (pgd_t *)__va(start_pgtable);
 109         clear_page(level4p);
 110         for (i = 0; i < nr_pfn_mapped; i++) {
 111                 mstart = pfn_mapped[i].start << PAGE_SHIFT;
 112                 mend   = pfn_mapped[i].end << PAGE_SHIFT;
 113
 114                 result = kernel_ident_mapping_init(&info,
 115                                                  level4p, mstart, mend);
 116                 if (result)
 117                         return result;
 118         }
 119
 120         /*
 121          * segments's mem ranges could be outside 0 ~ max_pfn,
 122          * for example when jump back to original kernel from kexeced kernel.
 123          * or first kernel is booted with user mem map, and second kernel
 124          * could be loaded out of that range.
 125          */
 126         for (i = 0; i < image->nr_segments; i++) {
 127                 mstart = image->segment[i].mem;
 128                 mend   = mstart + image->segment[i].memsz;
 129
 130                 result = kernel_ident_mapping_init(&info,
 131                                                  level4p, mstart, mend);
 132
 133                 if (result)
 134                         return result;
 135         }
 136
 137         return init_transition_pgtable(image, level4p);
 138 }
 139
 140 static void set_idt(void *newidt, u16 limit)
 141 {
 142         struct desc_ptr curidt;
 143
 144         /* x86-64 supports unaliged loads & stores */
 145         curidt.size    = limit;
 146         curidt.address = (unsigned long)newidt;
 147
 148         __asm__ __volatile__ (
 149                 "lidtq %0\n"
 150                 : : "m" (curidt)
 151                 );
 152 };
 153
 154
 155 static void set_gdt(void *newgdt, u16 limit)
 156 {
 157         struct desc_ptr curgdt;
 158
 159         /* x86-64 supports unaligned loads & stores */
 160         curgdt.size    = limit;
 161         curgdt.address = (unsigned long)newgdt;
 162
 163         __asm__ __volatile__ (
 164                 "lgdtq %0\n"
 165                 : : "m" (curgdt)
 166                 );
 167 };
 168
 169 static void load_segments(void)
 170 {
 171         __asm__ __volatile__ (
 172                 "\tmovl %0,%%ds\n"
 173                 "\tmovl %0,%%es\n"
 174                 "\tmovl %0,%%ss\n"
 175                 "\tmovl %0,%%fs\n"
 176                 "\tmovl %0,%%gs\n"
 177                 : : "a" (__KERNEL_DS) : "memory"
 178                 );
 179 }
 180
 181 int machine_kexec_prepare(struct kimage *image)
 182 {
 183         unsigned long start_pgtable;
 184         int result;
 185
 186         /* Calculate the offsets */
 187         start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
 188
 189         /* Setup the identity mapped 64bit page table */
 190         result = init_pgtable(image, start_pgtable);
 191         if (result)
 192                 return result;
 193
 194         return 0;
 195 }
 196
 197 void machine_kexec_cleanup(struct kimage *image)
 198 {
 199         free_transition_pgtable(image);
 200 }
 201
 202 /*
 203  * Do not allocate memory (or fail in any way) in machine_kexec().
 204  * We are past the point of no return, committed to rebooting now.
 205  */
 206 void machine_kexec(struct kimage *image)
 207 {
 208         unsigned long page_list[PAGES_NR];
 209         void *control_page;
 210         int save_ftrace_enabled;
 211
 212 #ifdef CONFIG_KEXEC_JUMP
 213         if (image->preserve_context)
 214                 save_processor_state();
 215 #endif
 216
 217         save_ftrace_enabled = __ftrace_enabled_save();
 218
 219         /* Interrupts aren't acceptable while we reboot */
 220         local_irq_disable();
 221         hw_breakpoint_disable();
 222
 223         if (image->preserve_context) {
 224 #ifdef CONFIG_X86_IO_APIC
 225                 /*
 226                  * We need to put APICs in legacy mode so that we can
 227                  * get timer interrupts in second kernel. kexec/kdump
 228                  * paths already have calls to disable_IO_APIC() in
 229                  * one form or other. kexec jump path also need
 230                  * one.
 231                  */
 232                 disable_IO_APIC();
 233 #endif
 234         }
 235
 236         control_page = page_address(image->control_code_page) + PAGE_SIZE;
 237         memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
 238
 239         page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
 240         page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
 241         page_list[PA_TABLE_PAGE] =
 242           (unsigned long)__pa(page_address(image->control_code_page));
 243
 244         if (image->type == KEXEC_TYPE_DEFAULT)
 245                 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
 246                                                 << PAGE_SHIFT);
 247
 248         /*
 249          * The segment registers are funny things, they have both a
 250          * visible and an invisible part.  Whenever the visible part is
 251          * set to a specific selector, the invisible part is loaded
 252          * with from a table in memory.  At no other time is the
 253          * descriptor table in memory accessed.
 254          *
 255          * I take advantage of this here by force loading the
 256          * segments, before I zap the gdt with an invalid value.
 257          */
 258         load_segments();
 259         /*
 260          * The gdt & idt are now invalid.
 261          * If you want to load them you must set up your own idt & gdt.
 262          */
 263         set_gdt(phys_to_virt(0), 0);
 264         set_idt(phys_to_virt(0), 0);
 265
 266         /* now call it */
 267         image->start = relocate_kernel((unsigned long)image->head,
 268                                        (unsigned long)page_list,
 269                                        image->start,
 270                                        image->preserve_context);
 271
 272 #ifdef CONFIG_KEXEC_JUMP
 273         if (image->preserve_context)
 274                 restore_processor_state();
 275 #endif
 276
 277         __ftrace_enabled_restore(save_ftrace_enabled);
 278 }
 279
 280 void arch_crash_save_vmcoreinfo(void)
 281 {
 282         VMCOREINFO_SYMBOL(phys_base);
 283         VMCOREINFO_SYMBOL(init_level4_pgt);
 284
 285 #ifdef CONFIG_NUMA
 286         VMCOREINFO_SYMBOL(node_data);
 287         VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
 288 #endif
 289         vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
 290                               (unsigned long)&_text - __START_KERNEL);
 291 }
 292
 293 /* arch-dependent functionality related to kexec file-based syscall */
 294
 295 int arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
 296                                   unsigned long buf_len)
 297 {
 298         int i, ret = -ENOEXEC;
 299         struct kexec_file_ops *fops;
 300
 301         for (i = 0; i < ARRAY_SIZE(kexec_file_loaders); i++) {
 302                 fops = kexec_file_loaders[i];
 303                 if (!fops || !fops->probe)
 304                         continue;
 305
 306                 ret = fops->probe(buf, buf_len);
 307                 if (!ret) {
 308                         image->fops = fops;
 309                         return ret;
 310                 }
 311         }
 312
 313         return ret;
 314 }
 315
 316 void *arch_kexec_kernel_image_load(struct kimage *image)
 317 {
 318         if (!image->fops || !image->fops->load)
 319                 return ERR_PTR(-ENOEXEC);
 320
 321         return image->fops->load(image, image->kernel_buf,
 322                                  image->kernel_buf_len, image->initrd_buf,
 323                                  image->initrd_buf_len, image->cmdline_buf,
 324                                  image->cmdline_buf_len);
 325 }
 326
 327 int arch_kimage_file_post_load_cleanup(struct kimage *image)
 328 {
 329         if (!image->fops || !image->fops->cleanup)
 330                 return 0;
 331
 332         return image->fops->cleanup(image->image_loader_data);
 333 }
 334
 335 /*
 336  * Apply purgatory relocations.
 337  *
 338  * ehdr: Pointer to elf headers
 339  * sechdrs: Pointer to section headers.
 340  * relsec: section index of SHT_RELA section.
 341  *
 342  * TODO: Some of the code belongs to generic code. Move that in kexec.c.
 343  */
 344 int arch_kexec_apply_relocations_add(const Elf64_Ehdr *ehdr,
 345                                      Elf64_Shdr *sechdrs, unsigned int relsec)
 346 {
 347         unsigned int i;
 348         Elf64_Rela *rel;
 349         Elf64_Sym *sym;
 350         void *location;
 351         Elf64_Shdr *section, *symtabsec;
 352         unsigned long address, sec_base, value;
 353         const char *strtab, *name, *shstrtab;
 354
 355         /*
 356          * ->sh_offset has been modified to keep the pointer to section
 357          * contents in memory
 358          */
 359         rel = (void *)sechdrs[relsec].sh_offset;
 360
 361         /* Section to which relocations apply */
 362         section = &sechdrs[sechdrs[relsec].sh_info];
 363
 364         pr_debug("Applying relocate section %u to %u\n", relsec,
 365                  sechdrs[relsec].sh_info);
 366
 367         /* Associated symbol table */
 368         symtabsec = &sechdrs[sechdrs[relsec].sh_link];
 369
 370         /* String table */
 371         if (symtabsec->sh_link >= ehdr->e_shnum) {
 372                 /* Invalid strtab section number */
 373                 pr_err("Invalid string table section index %d\n",
 374                        symtabsec->sh_link);
 375                 return -ENOEXEC;
 376         }
 377
 378         strtab = (char *)sechdrs[symtabsec->sh_link].sh_offset;
 379
 380         /* section header string table */
 381         shstrtab = (char *)sechdrs[ehdr->e_shstrndx].sh_offset;
 382
 383         for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rel); i++) {
 384
 385                 /*
 386                  * rel[i].r_offset contains byte offset from beginning
 387                  * of section to the storage unit affected.
 388                  *
 389                  * This is location to update (->sh_offset). This is temporary
 390                  * buffer where section is currently loaded. This will finally
 391                  * be loaded to a different address later, pointed to by
 392                  * ->sh_addr. kexec takes care of moving it
 393                  *  (kexec_load_segment()).
 394                  */
 395                 location = (void *)(section->sh_offset + rel[i].r_offset);
 396
 397                 /* Final address of the location */
 398                 address = section->sh_addr + rel[i].r_offset;
 399
 400                 /*
 401                  * rel[i].r_info contains information about symbol table index
 402                  * w.r.t which relocation must be made and type of relocation
 403                  * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
 404                  * these respectively.
 405                  */
 406                 sym = (Elf64_Sym *)symtabsec->sh_offset +
 407                                 ELF64_R_SYM(rel[i].r_info);
 408
 409                 if (sym->st_name)
 410                         name = strtab + sym->st_name;
 411                 else
 412                         name = shstrtab + sechdrs[sym->st_shndx].sh_name;
 413
 414                 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
 415                          name, sym->st_info, sym->st_shndx, sym->st_value,
 416                          sym->st_size);
 417
 418                 if (sym->st_shndx == SHN_UNDEF) {
 419                         pr_err("Undefined symbol: %s\n", name);
 420                         return -ENOEXEC;
 421                 }
 422
 423                 if (sym->st_shndx == SHN_COMMON) {
 424                         pr_err("symbol '%s' in common section\n", name);
 425                         return -ENOEXEC;
 426                 }
 427
 428                 if (sym->st_shndx == SHN_ABS)
 429                         sec_base = 0;
 430                 else if (sym->st_shndx >= ehdr->e_shnum) {
 431                         pr_err("Invalid section %d for symbol %s\n",
 432                                sym->st_shndx, name);
 433                         return -ENOEXEC;
 434                 } else
 435                         sec_base = sechdrs[sym->st_shndx].sh_addr;
 436
 437                 value = sym->st_value;
 438                 value += sec_base;
 439                 value += rel[i].r_addend;
 440
 441                 switch (ELF64_R_TYPE(rel[i].r_info)) {
 442                 case R_X86_64_NONE:
 443                         break;
 444                 case R_X86_64_64:
 445                         *(u64 *)location = value;
 446                         break;
 447                 case R_X86_64_32:
 448                         *(u32 *)location = value;
 449                         if (value != *(u32 *)location)
 450                                 goto overflow;
 451                         break;
 452                 case R_X86_64_32S:
 453                         *(s32 *)location = value;
 454                         if ((s64)value != *(s32 *)location)
 455                                 goto overflow;
 456                         break;
 457                 case R_X86_64_PC32:
 458                         value -= (u64)address;
 459                         *(u32 *)location = value;
 460                         break;
 461                 default:
 462                         pr_err("Unknown rela relocation: %llu\n",
 463                                ELF64_R_TYPE(rel[i].r_info));
 464                         return -ENOEXEC;
 465                 }
 466         }
 467         return 0;
 468
 469 overflow:
 470         pr_err("Overflow in relocation type %d value 0x%lx\n",
 471                (int)ELF64_R_TYPE(rel[i].r_info), value);
 472         return -ENOEXEC;
 473 }