1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Load ELF vmlinux file for the kexec_file_load syscall. 4 * 5 * Copyright (C) 2021 Huawei Technologies Co, Ltd. 6 * 7 * Author: Liao Chang (liaochang1@huawei.com) 8 * 9 * Based on kexec-tools' kexec-elf-riscv.c, heavily modified 10 * for kernel. 11 */ 12 13 #define pr_fmt(fmt) "kexec_image: " fmt 14 15 #include <linux/elf.h> 16 #include <linux/kexec.h> 17 #include <linux/slab.h> 18 #include <linux/of.h> 19 #include <linux/libfdt.h> 20 #include <linux/types.h> 21 #include <linux/memblock.h> 22 #include <asm/setup.h> 23 24 int arch_kimage_file_post_load_cleanup(struct kimage *image) 25 { 26 kvfree(image->arch.fdt); 27 image->arch.fdt = NULL; 28 29 vfree(image->elf_headers); 30 image->elf_headers = NULL; 31 image->elf_headers_sz = 0; 32 33 return kexec_image_post_load_cleanup_default(image); 34 } 35 36 static int riscv_kexec_elf_load(struct kimage *image, struct elfhdr *ehdr, 37 struct kexec_elf_info *elf_info, unsigned long old_pbase, 38 unsigned long new_pbase) 39 { 40 int i; 41 int ret = 0; 42 size_t size; 43 struct kexec_buf kbuf; 44 const struct elf_phdr *phdr; 45 46 kbuf.image = image; 47 48 for (i = 0; i < ehdr->e_phnum; i++) { 49 phdr = &elf_info->proghdrs[i]; 50 if (phdr->p_type != PT_LOAD) 51 continue; 52 53 size = phdr->p_filesz; 54 if (size > phdr->p_memsz) 55 size = phdr->p_memsz; 56 57 kbuf.buffer = (void *) elf_info->buffer + phdr->p_offset; 58 kbuf.bufsz = size; 59 kbuf.buf_align = phdr->p_align; 60 kbuf.mem = phdr->p_paddr - old_pbase + new_pbase; 61 kbuf.memsz = phdr->p_memsz; 62 kbuf.top_down = false; 63 ret = kexec_add_buffer(&kbuf); 64 if (ret) 65 break; 66 } 67 68 return ret; 69 } 70 71 /* 72 * Go through the available phsyical memory regions and find one that hold 73 * an image of the specified size. 74 */ 75 static int elf_find_pbase(struct kimage *image, unsigned long kernel_len, 76 struct elfhdr *ehdr, struct kexec_elf_info *elf_info, 77 unsigned long *old_pbase, unsigned long *new_pbase) 78 { 79 int i; 80 int ret; 81 struct kexec_buf kbuf; 82 const struct elf_phdr *phdr; 83 unsigned long lowest_paddr = ULONG_MAX; 84 unsigned long lowest_vaddr = ULONG_MAX; 85 86 for (i = 0; i < ehdr->e_phnum; i++) { 87 phdr = &elf_info->proghdrs[i]; 88 if (phdr->p_type != PT_LOAD) 89 continue; 90 91 if (lowest_paddr > phdr->p_paddr) 92 lowest_paddr = phdr->p_paddr; 93 94 if (lowest_vaddr > phdr->p_vaddr) 95 lowest_vaddr = phdr->p_vaddr; 96 } 97 98 kbuf.image = image; 99 kbuf.buf_min = lowest_paddr; 100 kbuf.buf_max = ULONG_MAX; 101 102 /* 103 * Current riscv boot protocol requires 2MB alignment for 104 * RV64 and 4MB alignment for RV32 105 * 106 */ 107 kbuf.buf_align = PMD_SIZE; 108 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN; 109 kbuf.memsz = ALIGN(kernel_len, PAGE_SIZE); 110 kbuf.top_down = false; 111 ret = arch_kexec_locate_mem_hole(&kbuf); 112 if (!ret) { 113 *old_pbase = lowest_paddr; 114 *new_pbase = kbuf.mem; 115 image->start = ehdr->e_entry - lowest_vaddr + kbuf.mem; 116 } 117 return ret; 118 } 119 120 static int get_nr_ram_ranges_callback(struct resource *res, void *arg) 121 { 122 unsigned int *nr_ranges = arg; 123 124 (*nr_ranges)++; 125 return 0; 126 } 127 128 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg) 129 { 130 struct crash_mem *cmem = arg; 131 132 cmem->ranges[cmem->nr_ranges].start = res->start; 133 cmem->ranges[cmem->nr_ranges].end = res->end; 134 cmem->nr_ranges++; 135 136 return 0; 137 } 138 139 static int prepare_elf_headers(void **addr, unsigned long *sz) 140 { 141 struct crash_mem *cmem; 142 unsigned int nr_ranges; 143 int ret; 144 145 nr_ranges = 1; /* For exclusion of crashkernel region */ 146 walk_system_ram_res(0, -1, &nr_ranges, get_nr_ram_ranges_callback); 147 148 cmem = kmalloc(struct_size(cmem, ranges, nr_ranges), GFP_KERNEL); 149 if (!cmem) 150 return -ENOMEM; 151 152 cmem->max_nr_ranges = nr_ranges; 153 cmem->nr_ranges = 0; 154 ret = walk_system_ram_res(0, -1, cmem, prepare_elf64_ram_headers_callback); 155 if (ret) 156 goto out; 157 158 /* Exclude crashkernel region */ 159 ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end); 160 if (!ret) 161 ret = crash_prepare_elf64_headers(cmem, true, addr, sz); 162 163 out: 164 kfree(cmem); 165 return ret; 166 } 167 168 static char *setup_kdump_cmdline(struct kimage *image, char *cmdline, 169 unsigned long cmdline_len) 170 { 171 int elfcorehdr_strlen; 172 char *cmdline_ptr; 173 174 cmdline_ptr = kzalloc(COMMAND_LINE_SIZE, GFP_KERNEL); 175 if (!cmdline_ptr) 176 return NULL; 177 178 elfcorehdr_strlen = sprintf(cmdline_ptr, "elfcorehdr=0x%lx ", 179 image->elf_load_addr); 180 181 if (elfcorehdr_strlen + cmdline_len > COMMAND_LINE_SIZE) { 182 pr_err("Appending elfcorehdr=<addr> exceeds cmdline size\n"); 183 kfree(cmdline_ptr); 184 return NULL; 185 } 186 187 memcpy(cmdline_ptr + elfcorehdr_strlen, cmdline, cmdline_len); 188 /* Ensure it's nul terminated */ 189 cmdline_ptr[COMMAND_LINE_SIZE - 1] = '\0'; 190 return cmdline_ptr; 191 } 192 193 static void *elf_kexec_load(struct kimage *image, char *kernel_buf, 194 unsigned long kernel_len, char *initrd, 195 unsigned long initrd_len, char *cmdline, 196 unsigned long cmdline_len) 197 { 198 int ret; 199 unsigned long old_kernel_pbase = ULONG_MAX; 200 unsigned long new_kernel_pbase = 0UL; 201 unsigned long initrd_pbase = 0UL; 202 unsigned long headers_sz; 203 unsigned long kernel_start; 204 void *fdt, *headers; 205 struct elfhdr ehdr; 206 struct kexec_buf kbuf; 207 struct kexec_elf_info elf_info; 208 char *modified_cmdline = NULL; 209 210 ret = kexec_build_elf_info(kernel_buf, kernel_len, &ehdr, &elf_info); 211 if (ret) 212 return ERR_PTR(ret); 213 214 ret = elf_find_pbase(image, kernel_len, &ehdr, &elf_info, 215 &old_kernel_pbase, &new_kernel_pbase); 216 if (ret) 217 goto out; 218 kernel_start = image->start; 219 pr_notice("The entry point of kernel at 0x%lx\n", image->start); 220 221 /* Add the kernel binary to the image */ 222 ret = riscv_kexec_elf_load(image, &ehdr, &elf_info, 223 old_kernel_pbase, new_kernel_pbase); 224 if (ret) 225 goto out; 226 227 kbuf.image = image; 228 kbuf.buf_min = new_kernel_pbase + kernel_len; 229 kbuf.buf_max = ULONG_MAX; 230 231 /* Add elfcorehdr */ 232 if (image->type == KEXEC_TYPE_CRASH) { 233 ret = prepare_elf_headers(&headers, &headers_sz); 234 if (ret) { 235 pr_err("Preparing elf core header failed\n"); 236 goto out; 237 } 238 239 kbuf.buffer = headers; 240 kbuf.bufsz = headers_sz; 241 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN; 242 kbuf.memsz = headers_sz; 243 kbuf.buf_align = ELF_CORE_HEADER_ALIGN; 244 kbuf.top_down = true; 245 246 ret = kexec_add_buffer(&kbuf); 247 if (ret) { 248 vfree(headers); 249 goto out; 250 } 251 image->elf_headers = headers; 252 image->elf_load_addr = kbuf.mem; 253 image->elf_headers_sz = headers_sz; 254 255 pr_debug("Loaded elf core header at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 256 image->elf_load_addr, kbuf.bufsz, kbuf.memsz); 257 258 /* Setup cmdline for kdump kernel case */ 259 modified_cmdline = setup_kdump_cmdline(image, cmdline, 260 cmdline_len); 261 if (!modified_cmdline) { 262 pr_err("Setting up cmdline for kdump kernel failed\n"); 263 ret = -EINVAL; 264 goto out; 265 } 266 cmdline = modified_cmdline; 267 } 268 269 #ifdef CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY 270 /* Add purgatory to the image */ 271 kbuf.top_down = true; 272 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN; 273 ret = kexec_load_purgatory(image, &kbuf); 274 if (ret) { 275 pr_err("Error loading purgatory ret=%d\n", ret); 276 goto out; 277 } 278 ret = kexec_purgatory_get_set_symbol(image, "riscv_kernel_entry", 279 &kernel_start, 280 sizeof(kernel_start), 0); 281 if (ret) 282 pr_err("Error update purgatory ret=%d\n", ret); 283 #endif /* CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY */ 284 285 /* Add the initrd to the image */ 286 if (initrd != NULL) { 287 kbuf.buffer = initrd; 288 kbuf.bufsz = kbuf.memsz = initrd_len; 289 kbuf.buf_align = PAGE_SIZE; 290 kbuf.top_down = true; 291 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN; 292 ret = kexec_add_buffer(&kbuf); 293 if (ret) 294 goto out; 295 initrd_pbase = kbuf.mem; 296 pr_notice("Loaded initrd at 0x%lx\n", initrd_pbase); 297 } 298 299 /* Add the DTB to the image */ 300 fdt = of_kexec_alloc_and_setup_fdt(image, initrd_pbase, 301 initrd_len, cmdline, 0); 302 if (!fdt) { 303 pr_err("Error setting up the new device tree.\n"); 304 ret = -EINVAL; 305 goto out; 306 } 307 308 fdt_pack(fdt); 309 kbuf.buffer = fdt; 310 kbuf.bufsz = kbuf.memsz = fdt_totalsize(fdt); 311 kbuf.buf_align = PAGE_SIZE; 312 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN; 313 kbuf.top_down = true; 314 ret = kexec_add_buffer(&kbuf); 315 if (ret) { 316 pr_err("Error add DTB kbuf ret=%d\n", ret); 317 goto out_free_fdt; 318 } 319 /* Cache the fdt buffer address for memory cleanup */ 320 image->arch.fdt = fdt; 321 pr_notice("Loaded device tree at 0x%lx\n", kbuf.mem); 322 goto out; 323 324 out_free_fdt: 325 kvfree(fdt); 326 out: 327 kfree(modified_cmdline); 328 kexec_free_elf_info(&elf_info); 329 return ret ? ERR_PTR(ret) : NULL; 330 } 331 332 #define RV_X(x, s, n) (((x) >> (s)) & ((1 << (n)) - 1)) 333 #define RISCV_IMM_BITS 12 334 #define RISCV_IMM_REACH (1LL << RISCV_IMM_BITS) 335 #define RISCV_CONST_HIGH_PART(x) \ 336 (((x) + (RISCV_IMM_REACH >> 1)) & ~(RISCV_IMM_REACH - 1)) 337 #define RISCV_CONST_LOW_PART(x) ((x) - RISCV_CONST_HIGH_PART(x)) 338 339 #define ENCODE_ITYPE_IMM(x) \ 340 (RV_X(x, 0, 12) << 20) 341 #define ENCODE_BTYPE_IMM(x) \ 342 ((RV_X(x, 1, 4) << 8) | (RV_X(x, 5, 6) << 25) | \ 343 (RV_X(x, 11, 1) << 7) | (RV_X(x, 12, 1) << 31)) 344 #define ENCODE_UTYPE_IMM(x) \ 345 (RV_X(x, 12, 20) << 12) 346 #define ENCODE_JTYPE_IMM(x) \ 347 ((RV_X(x, 1, 10) << 21) | (RV_X(x, 11, 1) << 20) | \ 348 (RV_X(x, 12, 8) << 12) | (RV_X(x, 20, 1) << 31)) 349 #define ENCODE_CBTYPE_IMM(x) \ 350 ((RV_X(x, 1, 2) << 3) | (RV_X(x, 3, 2) << 10) | (RV_X(x, 5, 1) << 2) | \ 351 (RV_X(x, 6, 2) << 5) | (RV_X(x, 8, 1) << 12)) 352 #define ENCODE_CJTYPE_IMM(x) \ 353 ((RV_X(x, 1, 3) << 3) | (RV_X(x, 4, 1) << 11) | (RV_X(x, 5, 1) << 2) | \ 354 (RV_X(x, 6, 1) << 7) | (RV_X(x, 7, 1) << 6) | (RV_X(x, 8, 2) << 9) | \ 355 (RV_X(x, 10, 1) << 8) | (RV_X(x, 11, 1) << 12)) 356 #define ENCODE_UJTYPE_IMM(x) \ 357 (ENCODE_UTYPE_IMM(RISCV_CONST_HIGH_PART(x)) | \ 358 (ENCODE_ITYPE_IMM(RISCV_CONST_LOW_PART(x)) << 32)) 359 #define ENCODE_UITYPE_IMM(x) \ 360 (ENCODE_UTYPE_IMM(x) | (ENCODE_ITYPE_IMM(x) << 32)) 361 362 #define CLEAN_IMM(type, x) \ 363 ((~ENCODE_##type##_IMM((uint64_t)(-1))) & (x)) 364 365 int arch_kexec_apply_relocations_add(struct purgatory_info *pi, 366 Elf_Shdr *section, 367 const Elf_Shdr *relsec, 368 const Elf_Shdr *symtab) 369 { 370 const char *strtab, *name, *shstrtab; 371 const Elf_Shdr *sechdrs; 372 Elf64_Rela *relas; 373 int i, r_type; 374 375 /* String & section header string table */ 376 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; 377 strtab = (char *)pi->ehdr + sechdrs[symtab->sh_link].sh_offset; 378 shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset; 379 380 relas = (void *)pi->ehdr + relsec->sh_offset; 381 382 for (i = 0; i < relsec->sh_size / sizeof(*relas); i++) { 383 const Elf_Sym *sym; /* symbol to relocate */ 384 unsigned long addr; /* final location after relocation */ 385 unsigned long val; /* relocated symbol value */ 386 unsigned long sec_base; /* relocated symbol value */ 387 void *loc; /* tmp location to modify */ 388 389 sym = (void *)pi->ehdr + symtab->sh_offset; 390 sym += ELF64_R_SYM(relas[i].r_info); 391 392 if (sym->st_name) 393 name = strtab + sym->st_name; 394 else 395 name = shstrtab + sechdrs[sym->st_shndx].sh_name; 396 397 loc = pi->purgatory_buf; 398 loc += section->sh_offset; 399 loc += relas[i].r_offset; 400 401 if (sym->st_shndx == SHN_ABS) 402 sec_base = 0; 403 else if (sym->st_shndx >= pi->ehdr->e_shnum) { 404 pr_err("Invalid section %d for symbol %s\n", 405 sym->st_shndx, name); 406 return -ENOEXEC; 407 } else 408 sec_base = pi->sechdrs[sym->st_shndx].sh_addr; 409 410 val = sym->st_value; 411 val += sec_base; 412 val += relas[i].r_addend; 413 414 addr = section->sh_addr + relas[i].r_offset; 415 416 r_type = ELF64_R_TYPE(relas[i].r_info); 417 418 switch (r_type) { 419 case R_RISCV_BRANCH: 420 *(u32 *)loc = CLEAN_IMM(BTYPE, *(u32 *)loc) | 421 ENCODE_BTYPE_IMM(val - addr); 422 break; 423 case R_RISCV_JAL: 424 *(u32 *)loc = CLEAN_IMM(JTYPE, *(u32 *)loc) | 425 ENCODE_JTYPE_IMM(val - addr); 426 break; 427 /* 428 * With no R_RISCV_PCREL_LO12_S, R_RISCV_PCREL_LO12_I 429 * sym is expected to be next to R_RISCV_PCREL_HI20 430 * in purgatory relsec. Handle it like R_RISCV_CALL 431 * sym, instead of searching the whole relsec. 432 */ 433 case R_RISCV_PCREL_HI20: 434 case R_RISCV_CALL_PLT: 435 case R_RISCV_CALL: 436 *(u64 *)loc = CLEAN_IMM(UITYPE, *(u64 *)loc) | 437 ENCODE_UJTYPE_IMM(val - addr); 438 break; 439 case R_RISCV_RVC_BRANCH: 440 *(u32 *)loc = CLEAN_IMM(CBTYPE, *(u32 *)loc) | 441 ENCODE_CBTYPE_IMM(val - addr); 442 break; 443 case R_RISCV_RVC_JUMP: 444 *(u32 *)loc = CLEAN_IMM(CJTYPE, *(u32 *)loc) | 445 ENCODE_CJTYPE_IMM(val - addr); 446 break; 447 case R_RISCV_ADD32: 448 *(u32 *)loc += val; 449 break; 450 case R_RISCV_SUB32: 451 *(u32 *)loc -= val; 452 break; 453 /* It has been applied by R_RISCV_PCREL_HI20 sym */ 454 case R_RISCV_PCREL_LO12_I: 455 case R_RISCV_ALIGN: 456 case R_RISCV_RELAX: 457 break; 458 default: 459 pr_err("Unknown rela relocation: %d\n", r_type); 460 return -ENOEXEC; 461 } 462 } 463 return 0; 464 } 465 466 const struct kexec_file_ops elf_kexec_ops = { 467 .probe = kexec_elf_probe, 468 .load = elf_kexec_load, 469 }; 470