1 /* 2 * Kexec bzImage loader 3 * 4 * Copyright (C) 2014 Red Hat Inc. 5 * Authors: 6 * Vivek Goyal <vgoyal@redhat.com> 7 * 8 * This source code is licensed under the GNU General Public License, 9 * Version 2. See the file COPYING for more details. 10 */ 11 12 #define pr_fmt(fmt) "kexec-bzImage64: " fmt 13 14 #include <linux/string.h> 15 #include <linux/printk.h> 16 #include <linux/errno.h> 17 #include <linux/slab.h> 18 #include <linux/kexec.h> 19 #include <linux/kernel.h> 20 #include <linux/mm.h> 21 #include <linux/efi.h> 22 #include <linux/verification.h> 23 24 #include <asm/bootparam.h> 25 #include <asm/setup.h> 26 #include <asm/crash.h> 27 #include <asm/efi.h> 28 #include <asm/e820/api.h> 29 #include <asm/kexec-bzimage64.h> 30 31 #define MAX_ELFCOREHDR_STR_LEN 30 /* elfcorehdr=0x<64bit-value> */ 32 33 /* 34 * Defines lowest physical address for various segments. Not sure where 35 * exactly these limits came from. Current bzimage64 loader in kexec-tools 36 * uses these so I am retaining it. It can be changed over time as we gain 37 * more insight. 38 */ 39 #define MIN_PURGATORY_ADDR 0x3000 40 #define MIN_BOOTPARAM_ADDR 0x3000 41 #define MIN_KERNEL_LOAD_ADDR 0x100000 42 #define MIN_INITRD_LOAD_ADDR 0x1000000 43 44 /* 45 * This is a place holder for all boot loader specific data structure which 46 * gets allocated in one call but gets freed much later during cleanup 47 * time. Right now there is only one field but it can grow as need be. 48 */ 49 struct bzimage64_data { 50 /* 51 * Temporary buffer to hold bootparams buffer. This should be 52 * freed once the bootparam segment has been loaded. 53 */ 54 void *bootparams_buf; 55 }; 56 57 static int setup_initrd(struct boot_params *params, 58 unsigned long initrd_load_addr, unsigned long initrd_len) 59 { 60 params->hdr.ramdisk_image = initrd_load_addr & 0xffffffffUL; 61 params->hdr.ramdisk_size = initrd_len & 0xffffffffUL; 62 63 params->ext_ramdisk_image = initrd_load_addr >> 32; 64 params->ext_ramdisk_size = initrd_len >> 32; 65 66 return 0; 67 } 68 69 static int setup_cmdline(struct kimage *image, struct boot_params *params, 70 unsigned long bootparams_load_addr, 71 unsigned long cmdline_offset, char *cmdline, 72 unsigned long cmdline_len) 73 { 74 char *cmdline_ptr = ((char *)params) + cmdline_offset; 75 unsigned long cmdline_ptr_phys, len = 0; 76 uint32_t cmdline_low_32, cmdline_ext_32; 77 78 if (image->type == KEXEC_TYPE_CRASH) { 79 len = sprintf(cmdline_ptr, 80 "elfcorehdr=0x%lx ", image->arch.elf_load_addr); 81 } 82 memcpy(cmdline_ptr + len, cmdline, cmdline_len); 83 cmdline_len += len; 84 85 cmdline_ptr[cmdline_len - 1] = '\0'; 86 87 pr_debug("Final command line is: %s\n", cmdline_ptr); 88 cmdline_ptr_phys = bootparams_load_addr + cmdline_offset; 89 cmdline_low_32 = cmdline_ptr_phys & 0xffffffffUL; 90 cmdline_ext_32 = cmdline_ptr_phys >> 32; 91 92 params->hdr.cmd_line_ptr = cmdline_low_32; 93 if (cmdline_ext_32) 94 params->ext_cmd_line_ptr = cmdline_ext_32; 95 96 return 0; 97 } 98 99 static int setup_e820_entries(struct boot_params *params) 100 { 101 unsigned int nr_e820_entries; 102 103 nr_e820_entries = e820_table_saved->nr_entries; 104 105 /* TODO: Pass entries more than E820MAX in bootparams setup data */ 106 if (nr_e820_entries > E820MAX) 107 nr_e820_entries = E820MAX; 108 109 params->e820_entries = nr_e820_entries; 110 memcpy(¶ms->e820_table, &e820_table_saved->entries, nr_e820_entries*sizeof(struct e820_entry)); 111 112 return 0; 113 } 114 115 #ifdef CONFIG_EFI 116 static int setup_efi_info_memmap(struct boot_params *params, 117 unsigned long params_load_addr, 118 unsigned int efi_map_offset, 119 unsigned int efi_map_sz) 120 { 121 void *efi_map = (void *)params + efi_map_offset; 122 unsigned long efi_map_phys_addr = params_load_addr + efi_map_offset; 123 struct efi_info *ei = ¶ms->efi_info; 124 125 if (!efi_map_sz) 126 return 0; 127 128 efi_runtime_map_copy(efi_map, efi_map_sz); 129 130 ei->efi_memmap = efi_map_phys_addr & 0xffffffff; 131 ei->efi_memmap_hi = efi_map_phys_addr >> 32; 132 ei->efi_memmap_size = efi_map_sz; 133 134 return 0; 135 } 136 137 static int 138 prepare_add_efi_setup_data(struct boot_params *params, 139 unsigned long params_load_addr, 140 unsigned int efi_setup_data_offset) 141 { 142 unsigned long setup_data_phys; 143 struct setup_data *sd = (void *)params + efi_setup_data_offset; 144 struct efi_setup_data *esd = (void *)sd + sizeof(struct setup_data); 145 146 esd->fw_vendor = efi.fw_vendor; 147 esd->runtime = efi.runtime; 148 esd->tables = efi.config_table; 149 esd->smbios = efi.smbios; 150 151 sd->type = SETUP_EFI; 152 sd->len = sizeof(struct efi_setup_data); 153 154 /* Add setup data */ 155 setup_data_phys = params_load_addr + efi_setup_data_offset; 156 sd->next = params->hdr.setup_data; 157 params->hdr.setup_data = setup_data_phys; 158 159 return 0; 160 } 161 162 static int 163 setup_efi_state(struct boot_params *params, unsigned long params_load_addr, 164 unsigned int efi_map_offset, unsigned int efi_map_sz, 165 unsigned int efi_setup_data_offset) 166 { 167 struct efi_info *current_ei = &boot_params.efi_info; 168 struct efi_info *ei = ¶ms->efi_info; 169 170 if (!current_ei->efi_memmap_size) 171 return 0; 172 173 /* 174 * If 1:1 mapping is not enabled, second kernel can not setup EFI 175 * and use EFI run time services. User space will have to pass 176 * acpi_rsdp=<addr> on kernel command line to make second kernel boot 177 * without efi. 178 */ 179 if (efi_enabled(EFI_OLD_MEMMAP)) 180 return 0; 181 182 ei->efi_loader_signature = current_ei->efi_loader_signature; 183 ei->efi_systab = current_ei->efi_systab; 184 ei->efi_systab_hi = current_ei->efi_systab_hi; 185 186 ei->efi_memdesc_version = current_ei->efi_memdesc_version; 187 ei->efi_memdesc_size = efi_get_runtime_map_desc_size(); 188 189 setup_efi_info_memmap(params, params_load_addr, efi_map_offset, 190 efi_map_sz); 191 prepare_add_efi_setup_data(params, params_load_addr, 192 efi_setup_data_offset); 193 return 0; 194 } 195 #endif /* CONFIG_EFI */ 196 197 static int 198 setup_boot_parameters(struct kimage *image, struct boot_params *params, 199 unsigned long params_load_addr, 200 unsigned int efi_map_offset, unsigned int efi_map_sz, 201 unsigned int efi_setup_data_offset) 202 { 203 unsigned int nr_e820_entries; 204 unsigned long long mem_k, start, end; 205 int i, ret = 0; 206 207 /* Get subarch from existing bootparams */ 208 params->hdr.hardware_subarch = boot_params.hdr.hardware_subarch; 209 210 /* Copying screen_info will do? */ 211 memcpy(¶ms->screen_info, &boot_params.screen_info, 212 sizeof(struct screen_info)); 213 214 /* Fill in memsize later */ 215 params->screen_info.ext_mem_k = 0; 216 params->alt_mem_k = 0; 217 218 /* Default APM info */ 219 memset(¶ms->apm_bios_info, 0, sizeof(params->apm_bios_info)); 220 221 /* Default drive info */ 222 memset(¶ms->hd0_info, 0, sizeof(params->hd0_info)); 223 memset(¶ms->hd1_info, 0, sizeof(params->hd1_info)); 224 225 if (image->type == KEXEC_TYPE_CRASH) { 226 ret = crash_setup_memmap_entries(image, params); 227 if (ret) 228 return ret; 229 } else 230 setup_e820_entries(params); 231 232 nr_e820_entries = params->e820_entries; 233 234 for (i = 0; i < nr_e820_entries; i++) { 235 if (params->e820_table[i].type != E820_RAM) 236 continue; 237 start = params->e820_table[i].addr; 238 end = params->e820_table[i].addr + params->e820_table[i].size - 1; 239 240 if ((start <= 0x100000) && end > 0x100000) { 241 mem_k = (end >> 10) - (0x100000 >> 10); 242 params->screen_info.ext_mem_k = mem_k; 243 params->alt_mem_k = mem_k; 244 if (mem_k > 0xfc00) 245 params->screen_info.ext_mem_k = 0xfc00; /* 64M*/ 246 if (mem_k > 0xffffffff) 247 params->alt_mem_k = 0xffffffff; 248 } 249 } 250 251 #ifdef CONFIG_EFI 252 /* Setup EFI state */ 253 setup_efi_state(params, params_load_addr, efi_map_offset, efi_map_sz, 254 efi_setup_data_offset); 255 #endif 256 257 /* Setup EDD info */ 258 memcpy(params->eddbuf, boot_params.eddbuf, 259 EDDMAXNR * sizeof(struct edd_info)); 260 params->eddbuf_entries = boot_params.eddbuf_entries; 261 262 memcpy(params->edd_mbr_sig_buffer, boot_params.edd_mbr_sig_buffer, 263 EDD_MBR_SIG_MAX * sizeof(unsigned int)); 264 265 return ret; 266 } 267 268 static int bzImage64_probe(const char *buf, unsigned long len) 269 { 270 int ret = -ENOEXEC; 271 struct setup_header *header; 272 273 /* kernel should be at least two sectors long */ 274 if (len < 2 * 512) { 275 pr_err("File is too short to be a bzImage\n"); 276 return ret; 277 } 278 279 header = (struct setup_header *)(buf + offsetof(struct boot_params, hdr)); 280 if (memcmp((char *)&header->header, "HdrS", 4) != 0) { 281 pr_err("Not a bzImage\n"); 282 return ret; 283 } 284 285 if (header->boot_flag != 0xAA55) { 286 pr_err("No x86 boot sector present\n"); 287 return ret; 288 } 289 290 if (header->version < 0x020C) { 291 pr_err("Must be at least protocol version 2.12\n"); 292 return ret; 293 } 294 295 if (!(header->loadflags & LOADED_HIGH)) { 296 pr_err("zImage not a bzImage\n"); 297 return ret; 298 } 299 300 if (!(header->xloadflags & XLF_KERNEL_64)) { 301 pr_err("Not a bzImage64. XLF_KERNEL_64 is not set.\n"); 302 return ret; 303 } 304 305 if (!(header->xloadflags & XLF_CAN_BE_LOADED_ABOVE_4G)) { 306 pr_err("XLF_CAN_BE_LOADED_ABOVE_4G is not set.\n"); 307 return ret; 308 } 309 310 /* 311 * Can't handle 32bit EFI as it does not allow loading kernel 312 * above 4G. This should be handled by 32bit bzImage loader 313 */ 314 if (efi_enabled(EFI_RUNTIME_SERVICES) && !efi_enabled(EFI_64BIT)) { 315 pr_debug("EFI is 32 bit. Can't load kernel above 4G.\n"); 316 return ret; 317 } 318 319 /* I've got a bzImage */ 320 pr_debug("It's a relocatable bzImage64\n"); 321 ret = 0; 322 323 return ret; 324 } 325 326 static void *bzImage64_load(struct kimage *image, char *kernel, 327 unsigned long kernel_len, char *initrd, 328 unsigned long initrd_len, char *cmdline, 329 unsigned long cmdline_len) 330 { 331 332 struct setup_header *header; 333 int setup_sects, kern16_size, ret = 0; 334 unsigned long setup_header_size, params_cmdline_sz; 335 struct boot_params *params; 336 unsigned long bootparam_load_addr, kernel_load_addr, initrd_load_addr; 337 unsigned long purgatory_load_addr; 338 struct bzimage64_data *ldata; 339 struct kexec_entry64_regs regs64; 340 void *stack; 341 unsigned int setup_hdr_offset = offsetof(struct boot_params, hdr); 342 unsigned int efi_map_offset, efi_map_sz, efi_setup_data_offset; 343 struct kexec_buf kbuf = { .image = image, .buf_max = ULONG_MAX, 344 .top_down = true }; 345 346 header = (struct setup_header *)(kernel + setup_hdr_offset); 347 setup_sects = header->setup_sects; 348 if (setup_sects == 0) 349 setup_sects = 4; 350 351 kern16_size = (setup_sects + 1) * 512; 352 if (kernel_len < kern16_size) { 353 pr_err("bzImage truncated\n"); 354 return ERR_PTR(-ENOEXEC); 355 } 356 357 if (cmdline_len > header->cmdline_size) { 358 pr_err("Kernel command line too long\n"); 359 return ERR_PTR(-EINVAL); 360 } 361 362 /* 363 * In case of crash dump, we will append elfcorehdr=<addr> to 364 * command line. Make sure it does not overflow 365 */ 366 if (cmdline_len + MAX_ELFCOREHDR_STR_LEN > header->cmdline_size) { 367 pr_debug("Appending elfcorehdr=<addr> to command line exceeds maximum allowed length\n"); 368 return ERR_PTR(-EINVAL); 369 } 370 371 /* Allocate and load backup region */ 372 if (image->type == KEXEC_TYPE_CRASH) { 373 ret = crash_load_segments(image); 374 if (ret) 375 return ERR_PTR(ret); 376 } 377 378 /* 379 * Load purgatory. For 64bit entry point, purgatory code can be 380 * anywhere. 381 */ 382 ret = kexec_load_purgatory(image, MIN_PURGATORY_ADDR, ULONG_MAX, 1, 383 &purgatory_load_addr); 384 if (ret) { 385 pr_err("Loading purgatory failed\n"); 386 return ERR_PTR(ret); 387 } 388 389 pr_debug("Loaded purgatory at 0x%lx\n", purgatory_load_addr); 390 391 392 /* 393 * Load Bootparams and cmdline and space for efi stuff. 394 * 395 * Allocate memory together for multiple data structures so 396 * that they all can go in single area/segment and we don't 397 * have to create separate segment for each. Keeps things 398 * little bit simple 399 */ 400 efi_map_sz = efi_get_runtime_map_size(); 401 efi_map_sz = ALIGN(efi_map_sz, 16); 402 params_cmdline_sz = sizeof(struct boot_params) + cmdline_len + 403 MAX_ELFCOREHDR_STR_LEN; 404 params_cmdline_sz = ALIGN(params_cmdline_sz, 16); 405 kbuf.bufsz = params_cmdline_sz + efi_map_sz + 406 sizeof(struct setup_data) + 407 sizeof(struct efi_setup_data); 408 409 params = kzalloc(kbuf.bufsz, GFP_KERNEL); 410 if (!params) 411 return ERR_PTR(-ENOMEM); 412 efi_map_offset = params_cmdline_sz; 413 efi_setup_data_offset = efi_map_offset + efi_map_sz; 414 415 /* Copy setup header onto bootparams. Documentation/x86/boot.txt */ 416 setup_header_size = 0x0202 + kernel[0x0201] - setup_hdr_offset; 417 418 /* Is there a limit on setup header size? */ 419 memcpy(¶ms->hdr, (kernel + setup_hdr_offset), setup_header_size); 420 421 kbuf.buffer = params; 422 kbuf.memsz = kbuf.bufsz; 423 kbuf.buf_align = 16; 424 kbuf.buf_min = MIN_BOOTPARAM_ADDR; 425 ret = kexec_add_buffer(&kbuf); 426 if (ret) 427 goto out_free_params; 428 bootparam_load_addr = kbuf.mem; 429 pr_debug("Loaded boot_param, command line and misc at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 430 bootparam_load_addr, kbuf.bufsz, kbuf.bufsz); 431 432 /* Load kernel */ 433 kbuf.buffer = kernel + kern16_size; 434 kbuf.bufsz = kernel_len - kern16_size; 435 kbuf.memsz = PAGE_ALIGN(header->init_size); 436 kbuf.buf_align = header->kernel_alignment; 437 kbuf.buf_min = MIN_KERNEL_LOAD_ADDR; 438 ret = kexec_add_buffer(&kbuf); 439 if (ret) 440 goto out_free_params; 441 kernel_load_addr = kbuf.mem; 442 443 pr_debug("Loaded 64bit kernel at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 444 kernel_load_addr, kbuf.bufsz, kbuf.memsz); 445 446 /* Load initrd high */ 447 if (initrd) { 448 kbuf.buffer = initrd; 449 kbuf.bufsz = kbuf.memsz = initrd_len; 450 kbuf.buf_align = PAGE_SIZE; 451 kbuf.buf_min = MIN_INITRD_LOAD_ADDR; 452 ret = kexec_add_buffer(&kbuf); 453 if (ret) 454 goto out_free_params; 455 initrd_load_addr = kbuf.mem; 456 457 pr_debug("Loaded initrd at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 458 initrd_load_addr, initrd_len, initrd_len); 459 460 setup_initrd(params, initrd_load_addr, initrd_len); 461 } 462 463 setup_cmdline(image, params, bootparam_load_addr, 464 sizeof(struct boot_params), cmdline, cmdline_len); 465 466 /* bootloader info. Do we need a separate ID for kexec kernel loader? */ 467 params->hdr.type_of_loader = 0x0D << 4; 468 params->hdr.loadflags = 0; 469 470 /* Setup purgatory regs for entry */ 471 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", ®s64, 472 sizeof(regs64), 1); 473 if (ret) 474 goto out_free_params; 475 476 regs64.rbx = 0; /* Bootstrap Processor */ 477 regs64.rsi = bootparam_load_addr; 478 regs64.rip = kernel_load_addr + 0x200; 479 stack = kexec_purgatory_get_symbol_addr(image, "stack_end"); 480 if (IS_ERR(stack)) { 481 pr_err("Could not find address of symbol stack_end\n"); 482 ret = -EINVAL; 483 goto out_free_params; 484 } 485 486 regs64.rsp = (unsigned long)stack; 487 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", ®s64, 488 sizeof(regs64), 0); 489 if (ret) 490 goto out_free_params; 491 492 ret = setup_boot_parameters(image, params, bootparam_load_addr, 493 efi_map_offset, efi_map_sz, 494 efi_setup_data_offset); 495 if (ret) 496 goto out_free_params; 497 498 /* Allocate loader specific data */ 499 ldata = kzalloc(sizeof(struct bzimage64_data), GFP_KERNEL); 500 if (!ldata) { 501 ret = -ENOMEM; 502 goto out_free_params; 503 } 504 505 /* 506 * Store pointer to params so that it could be freed after loading 507 * params segment has been loaded and contents have been copied 508 * somewhere else. 509 */ 510 ldata->bootparams_buf = params; 511 return ldata; 512 513 out_free_params: 514 kfree(params); 515 return ERR_PTR(ret); 516 } 517 518 /* This cleanup function is called after various segments have been loaded */ 519 static int bzImage64_cleanup(void *loader_data) 520 { 521 struct bzimage64_data *ldata = loader_data; 522 523 if (!ldata) 524 return 0; 525 526 kfree(ldata->bootparams_buf); 527 ldata->bootparams_buf = NULL; 528 529 return 0; 530 } 531 532 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG 533 static int bzImage64_verify_sig(const char *kernel, unsigned long kernel_len) 534 { 535 return verify_pefile_signature(kernel, kernel_len, 536 NULL, 537 VERIFYING_KEXEC_PE_SIGNATURE); 538 } 539 #endif 540 541 struct kexec_file_ops kexec_bzImage64_ops = { 542 .probe = bzImage64_probe, 543 .load = bzImage64_load, 544 .cleanup = bzImage64_cleanup, 545 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG 546 .verify_sig = bzImage64_verify_sig, 547 #endif 548 }; 549