1 /* 2 * ARM kernel loader. 3 * 4 * Copyright (c) 2006-2007 CodeSourcery. 5 * Written by Paul Brook 6 * 7 * This code is licensed under the GPL. 8 */ 9 10 #include "qemu/osdep.h" 11 #include "qemu/datadir.h" 12 #include "qemu/error-report.h" 13 #include "qapi/error.h" 14 #include <libfdt.h> 15 #include "hw/arm/boot.h" 16 #include "hw/arm/linux-boot-if.h" 17 #include "sysemu/kvm.h" 18 #include "sysemu/sysemu.h" 19 #include "sysemu/numa.h" 20 #include "hw/boards.h" 21 #include "sysemu/reset.h" 22 #include "hw/loader.h" 23 #include "elf.h" 24 #include "sysemu/device_tree.h" 25 #include "qemu/config-file.h" 26 #include "qemu/option.h" 27 #include "qemu/units.h" 28 29 /* Kernel boot protocol is specified in the kernel docs 30 * Documentation/arm/Booting and Documentation/arm64/booting.txt 31 * They have different preferred image load offsets from system RAM base. 32 */ 33 #define KERNEL_ARGS_ADDR 0x100 34 #define KERNEL_NOLOAD_ADDR 0x02000000 35 #define KERNEL_LOAD_ADDR 0x00010000 36 #define KERNEL64_LOAD_ADDR 0x00080000 37 38 #define ARM64_TEXT_OFFSET_OFFSET 8 39 #define ARM64_MAGIC_OFFSET 56 40 41 #define BOOTLOADER_MAX_SIZE (4 * KiB) 42 43 AddressSpace *arm_boot_address_space(ARMCPU *cpu, 44 const struct arm_boot_info *info) 45 { 46 /* Return the address space to use for bootloader reads and writes. 47 * We prefer the secure address space if the CPU has it and we're 48 * going to boot the guest into it. 49 */ 50 int asidx; 51 CPUState *cs = CPU(cpu); 52 53 if (arm_feature(&cpu->env, ARM_FEATURE_EL3) && info->secure_boot) { 54 asidx = ARMASIdx_S; 55 } else { 56 asidx = ARMASIdx_NS; 57 } 58 59 return cpu_get_address_space(cs, asidx); 60 } 61 62 typedef enum { 63 FIXUP_NONE = 0, /* do nothing */ 64 FIXUP_TERMINATOR, /* end of insns */ 65 FIXUP_BOARDID, /* overwrite with board ID number */ 66 FIXUP_BOARD_SETUP, /* overwrite with board specific setup code address */ 67 FIXUP_ARGPTR_LO, /* overwrite with pointer to kernel args */ 68 FIXUP_ARGPTR_HI, /* overwrite with pointer to kernel args (high half) */ 69 FIXUP_ENTRYPOINT_LO, /* overwrite with kernel entry point */ 70 FIXUP_ENTRYPOINT_HI, /* overwrite with kernel entry point (high half) */ 71 FIXUP_GIC_CPU_IF, /* overwrite with GIC CPU interface address */ 72 FIXUP_BOOTREG, /* overwrite with boot register address */ 73 FIXUP_DSB, /* overwrite with correct DSB insn for cpu */ 74 FIXUP_MAX, 75 } FixupType; 76 77 typedef struct ARMInsnFixup { 78 uint32_t insn; 79 FixupType fixup; 80 } ARMInsnFixup; 81 82 static const ARMInsnFixup bootloader_aarch64[] = { 83 { 0x580000c0 }, /* ldr x0, arg ; Load the lower 32-bits of DTB */ 84 { 0xaa1f03e1 }, /* mov x1, xzr */ 85 { 0xaa1f03e2 }, /* mov x2, xzr */ 86 { 0xaa1f03e3 }, /* mov x3, xzr */ 87 { 0x58000084 }, /* ldr x4, entry ; Load the lower 32-bits of kernel entry */ 88 { 0xd61f0080 }, /* br x4 ; Jump to the kernel entry point */ 89 { 0, FIXUP_ARGPTR_LO }, /* arg: .word @DTB Lower 32-bits */ 90 { 0, FIXUP_ARGPTR_HI}, /* .word @DTB Higher 32-bits */ 91 { 0, FIXUP_ENTRYPOINT_LO }, /* entry: .word @Kernel Entry Lower 32-bits */ 92 { 0, FIXUP_ENTRYPOINT_HI }, /* .word @Kernel Entry Higher 32-bits */ 93 { 0, FIXUP_TERMINATOR } 94 }; 95 96 /* A very small bootloader: call the board-setup code (if needed), 97 * set r0-r2, then jump to the kernel. 98 * If we're not calling boot setup code then we don't copy across 99 * the first BOOTLOADER_NO_BOARD_SETUP_OFFSET insns in this array. 100 */ 101 102 static const ARMInsnFixup bootloader[] = { 103 { 0xe28fe004 }, /* add lr, pc, #4 */ 104 { 0xe51ff004 }, /* ldr pc, [pc, #-4] */ 105 { 0, FIXUP_BOARD_SETUP }, 106 #define BOOTLOADER_NO_BOARD_SETUP_OFFSET 3 107 { 0xe3a00000 }, /* mov r0, #0 */ 108 { 0xe59f1004 }, /* ldr r1, [pc, #4] */ 109 { 0xe59f2004 }, /* ldr r2, [pc, #4] */ 110 { 0xe59ff004 }, /* ldr pc, [pc, #4] */ 111 { 0, FIXUP_BOARDID }, 112 { 0, FIXUP_ARGPTR_LO }, 113 { 0, FIXUP_ENTRYPOINT_LO }, 114 { 0, FIXUP_TERMINATOR } 115 }; 116 117 /* Handling for secondary CPU boot in a multicore system. 118 * Unlike the uniprocessor/primary CPU boot, this is platform 119 * dependent. The default code here is based on the secondary 120 * CPU boot protocol used on realview/vexpress boards, with 121 * some parameterisation to increase its flexibility. 122 * QEMU platform models for which this code is not appropriate 123 * should override write_secondary_boot and secondary_cpu_reset_hook 124 * instead. 125 * 126 * This code enables the interrupt controllers for the secondary 127 * CPUs and then puts all the secondary CPUs into a loop waiting 128 * for an interprocessor interrupt and polling a configurable 129 * location for the kernel secondary CPU entry point. 130 */ 131 #define DSB_INSN 0xf57ff04f 132 #define CP15_DSB_INSN 0xee070f9a /* mcr cp15, 0, r0, c7, c10, 4 */ 133 134 static const ARMInsnFixup smpboot[] = { 135 { 0xe59f2028 }, /* ldr r2, gic_cpu_if */ 136 { 0xe59f0028 }, /* ldr r0, bootreg_addr */ 137 { 0xe3a01001 }, /* mov r1, #1 */ 138 { 0xe5821000 }, /* str r1, [r2] - set GICC_CTLR.Enable */ 139 { 0xe3a010ff }, /* mov r1, #0xff */ 140 { 0xe5821004 }, /* str r1, [r2, 4] - set GIC_PMR.Priority to 0xff */ 141 { 0, FIXUP_DSB }, /* dsb */ 142 { 0xe320f003 }, /* wfi */ 143 { 0xe5901000 }, /* ldr r1, [r0] */ 144 { 0xe1110001 }, /* tst r1, r1 */ 145 { 0x0afffffb }, /* beq <wfi> */ 146 { 0xe12fff11 }, /* bx r1 */ 147 { 0, FIXUP_GIC_CPU_IF }, /* gic_cpu_if: .word 0x.... */ 148 { 0, FIXUP_BOOTREG }, /* bootreg_addr: .word 0x.... */ 149 { 0, FIXUP_TERMINATOR } 150 }; 151 152 static void write_bootloader(const char *name, hwaddr addr, 153 const ARMInsnFixup *insns, uint32_t *fixupcontext, 154 AddressSpace *as) 155 { 156 /* Fix up the specified bootloader fragment and write it into 157 * guest memory using rom_add_blob_fixed(). fixupcontext is 158 * an array giving the values to write in for the fixup types 159 * which write a value into the code array. 160 */ 161 int i, len; 162 uint32_t *code; 163 164 len = 0; 165 while (insns[len].fixup != FIXUP_TERMINATOR) { 166 len++; 167 } 168 169 code = g_new0(uint32_t, len); 170 171 for (i = 0; i < len; i++) { 172 uint32_t insn = insns[i].insn; 173 FixupType fixup = insns[i].fixup; 174 175 switch (fixup) { 176 case FIXUP_NONE: 177 break; 178 case FIXUP_BOARDID: 179 case FIXUP_BOARD_SETUP: 180 case FIXUP_ARGPTR_LO: 181 case FIXUP_ARGPTR_HI: 182 case FIXUP_ENTRYPOINT_LO: 183 case FIXUP_ENTRYPOINT_HI: 184 case FIXUP_GIC_CPU_IF: 185 case FIXUP_BOOTREG: 186 case FIXUP_DSB: 187 insn = fixupcontext[fixup]; 188 break; 189 default: 190 abort(); 191 } 192 code[i] = tswap32(insn); 193 } 194 195 assert((len * sizeof(uint32_t)) < BOOTLOADER_MAX_SIZE); 196 197 rom_add_blob_fixed_as(name, code, len * sizeof(uint32_t), addr, as); 198 199 g_free(code); 200 } 201 202 static void default_write_secondary(ARMCPU *cpu, 203 const struct arm_boot_info *info) 204 { 205 uint32_t fixupcontext[FIXUP_MAX]; 206 AddressSpace *as = arm_boot_address_space(cpu, info); 207 208 fixupcontext[FIXUP_GIC_CPU_IF] = info->gic_cpu_if_addr; 209 fixupcontext[FIXUP_BOOTREG] = info->smp_bootreg_addr; 210 if (arm_feature(&cpu->env, ARM_FEATURE_V7)) { 211 fixupcontext[FIXUP_DSB] = DSB_INSN; 212 } else { 213 fixupcontext[FIXUP_DSB] = CP15_DSB_INSN; 214 } 215 216 write_bootloader("smpboot", info->smp_loader_start, 217 smpboot, fixupcontext, as); 218 } 219 220 void arm_write_secure_board_setup_dummy_smc(ARMCPU *cpu, 221 const struct arm_boot_info *info, 222 hwaddr mvbar_addr) 223 { 224 AddressSpace *as = arm_boot_address_space(cpu, info); 225 int n; 226 uint32_t mvbar_blob[] = { 227 /* mvbar_addr: secure monitor vectors 228 * Default unimplemented and unused vectors to spin. Makes it 229 * easier to debug (as opposed to the CPU running away). 230 */ 231 0xeafffffe, /* (spin) */ 232 0xeafffffe, /* (spin) */ 233 0xe1b0f00e, /* movs pc, lr ;SMC exception return */ 234 0xeafffffe, /* (spin) */ 235 0xeafffffe, /* (spin) */ 236 0xeafffffe, /* (spin) */ 237 0xeafffffe, /* (spin) */ 238 0xeafffffe, /* (spin) */ 239 }; 240 uint32_t board_setup_blob[] = { 241 /* board setup addr */ 242 0xee110f51, /* mrc p15, 0, r0, c1, c1, 2 ;read NSACR */ 243 0xe3800b03, /* orr r0, #0xc00 ;set CP11, CP10 */ 244 0xee010f51, /* mcr p15, 0, r0, c1, c1, 2 ;write NSACR */ 245 0xe3a00e00 + (mvbar_addr >> 4), /* mov r0, #mvbar_addr */ 246 0xee0c0f30, /* mcr p15, 0, r0, c12, c0, 1 ;set MVBAR */ 247 0xee110f11, /* mrc p15, 0, r0, c1 , c1, 0 ;read SCR */ 248 0xe3800031, /* orr r0, #0x31 ;enable AW, FW, NS */ 249 0xee010f11, /* mcr p15, 0, r0, c1, c1, 0 ;write SCR */ 250 0xe1a0100e, /* mov r1, lr ;save LR across SMC */ 251 0xe1600070, /* smc #0 ;call monitor to flush SCR */ 252 0xe1a0f001, /* mov pc, r1 ;return */ 253 }; 254 255 /* check that mvbar_addr is correctly aligned and relocatable (using MOV) */ 256 assert((mvbar_addr & 0x1f) == 0 && (mvbar_addr >> 4) < 0x100); 257 258 /* check that these blobs don't overlap */ 259 assert((mvbar_addr + sizeof(mvbar_blob) <= info->board_setup_addr) 260 || (info->board_setup_addr + sizeof(board_setup_blob) <= mvbar_addr)); 261 262 for (n = 0; n < ARRAY_SIZE(mvbar_blob); n++) { 263 mvbar_blob[n] = tswap32(mvbar_blob[n]); 264 } 265 rom_add_blob_fixed_as("board-setup-mvbar", mvbar_blob, sizeof(mvbar_blob), 266 mvbar_addr, as); 267 268 for (n = 0; n < ARRAY_SIZE(board_setup_blob); n++) { 269 board_setup_blob[n] = tswap32(board_setup_blob[n]); 270 } 271 rom_add_blob_fixed_as("board-setup", board_setup_blob, 272 sizeof(board_setup_blob), info->board_setup_addr, as); 273 } 274 275 static void default_reset_secondary(ARMCPU *cpu, 276 const struct arm_boot_info *info) 277 { 278 AddressSpace *as = arm_boot_address_space(cpu, info); 279 CPUState *cs = CPU(cpu); 280 281 address_space_stl_notdirty(as, info->smp_bootreg_addr, 282 0, MEMTXATTRS_UNSPECIFIED, NULL); 283 cpu_set_pc(cs, info->smp_loader_start); 284 } 285 286 static inline bool have_dtb(const struct arm_boot_info *info) 287 { 288 return info->dtb_filename || info->get_dtb; 289 } 290 291 #define WRITE_WORD(p, value) do { \ 292 address_space_stl_notdirty(as, p, value, \ 293 MEMTXATTRS_UNSPECIFIED, NULL); \ 294 p += 4; \ 295 } while (0) 296 297 static void set_kernel_args(const struct arm_boot_info *info, AddressSpace *as) 298 { 299 int initrd_size = info->initrd_size; 300 hwaddr base = info->loader_start; 301 hwaddr p; 302 303 p = base + KERNEL_ARGS_ADDR; 304 /* ATAG_CORE */ 305 WRITE_WORD(p, 5); 306 WRITE_WORD(p, 0x54410001); 307 WRITE_WORD(p, 1); 308 WRITE_WORD(p, 0x1000); 309 WRITE_WORD(p, 0); 310 /* ATAG_MEM */ 311 /* TODO: handle multiple chips on one ATAG list */ 312 WRITE_WORD(p, 4); 313 WRITE_WORD(p, 0x54410002); 314 WRITE_WORD(p, info->ram_size); 315 WRITE_WORD(p, info->loader_start); 316 if (initrd_size) { 317 /* ATAG_INITRD2 */ 318 WRITE_WORD(p, 4); 319 WRITE_WORD(p, 0x54420005); 320 WRITE_WORD(p, info->initrd_start); 321 WRITE_WORD(p, initrd_size); 322 } 323 if (info->kernel_cmdline && *info->kernel_cmdline) { 324 /* ATAG_CMDLINE */ 325 int cmdline_size; 326 327 cmdline_size = strlen(info->kernel_cmdline); 328 address_space_write(as, p + 8, MEMTXATTRS_UNSPECIFIED, 329 info->kernel_cmdline, cmdline_size + 1); 330 cmdline_size = (cmdline_size >> 2) + 1; 331 WRITE_WORD(p, cmdline_size + 2); 332 WRITE_WORD(p, 0x54410009); 333 p += cmdline_size * 4; 334 } 335 if (info->atag_board) { 336 /* ATAG_BOARD */ 337 int atag_board_len; 338 uint8_t atag_board_buf[0x1000]; 339 340 atag_board_len = (info->atag_board(info, atag_board_buf) + 3) & ~3; 341 WRITE_WORD(p, (atag_board_len + 8) >> 2); 342 WRITE_WORD(p, 0x414f4d50); 343 address_space_write(as, p, MEMTXATTRS_UNSPECIFIED, 344 atag_board_buf, atag_board_len); 345 p += atag_board_len; 346 } 347 /* ATAG_END */ 348 WRITE_WORD(p, 0); 349 WRITE_WORD(p, 0); 350 } 351 352 static void set_kernel_args_old(const struct arm_boot_info *info, 353 AddressSpace *as) 354 { 355 hwaddr p; 356 const char *s; 357 int initrd_size = info->initrd_size; 358 hwaddr base = info->loader_start; 359 360 /* see linux/include/asm-arm/setup.h */ 361 p = base + KERNEL_ARGS_ADDR; 362 /* page_size */ 363 WRITE_WORD(p, 4096); 364 /* nr_pages */ 365 WRITE_WORD(p, info->ram_size / 4096); 366 /* ramdisk_size */ 367 WRITE_WORD(p, 0); 368 #define FLAG_READONLY 1 369 #define FLAG_RDLOAD 4 370 #define FLAG_RDPROMPT 8 371 /* flags */ 372 WRITE_WORD(p, FLAG_READONLY | FLAG_RDLOAD | FLAG_RDPROMPT); 373 /* rootdev */ 374 WRITE_WORD(p, (31 << 8) | 0); /* /dev/mtdblock0 */ 375 /* video_num_cols */ 376 WRITE_WORD(p, 0); 377 /* video_num_rows */ 378 WRITE_WORD(p, 0); 379 /* video_x */ 380 WRITE_WORD(p, 0); 381 /* video_y */ 382 WRITE_WORD(p, 0); 383 /* memc_control_reg */ 384 WRITE_WORD(p, 0); 385 /* unsigned char sounddefault */ 386 /* unsigned char adfsdrives */ 387 /* unsigned char bytes_per_char_h */ 388 /* unsigned char bytes_per_char_v */ 389 WRITE_WORD(p, 0); 390 /* pages_in_bank[4] */ 391 WRITE_WORD(p, 0); 392 WRITE_WORD(p, 0); 393 WRITE_WORD(p, 0); 394 WRITE_WORD(p, 0); 395 /* pages_in_vram */ 396 WRITE_WORD(p, 0); 397 /* initrd_start */ 398 if (initrd_size) { 399 WRITE_WORD(p, info->initrd_start); 400 } else { 401 WRITE_WORD(p, 0); 402 } 403 /* initrd_size */ 404 WRITE_WORD(p, initrd_size); 405 /* rd_start */ 406 WRITE_WORD(p, 0); 407 /* system_rev */ 408 WRITE_WORD(p, 0); 409 /* system_serial_low */ 410 WRITE_WORD(p, 0); 411 /* system_serial_high */ 412 WRITE_WORD(p, 0); 413 /* mem_fclk_21285 */ 414 WRITE_WORD(p, 0); 415 /* zero unused fields */ 416 while (p < base + KERNEL_ARGS_ADDR + 256 + 1024) { 417 WRITE_WORD(p, 0); 418 } 419 s = info->kernel_cmdline; 420 if (s) { 421 address_space_write(as, p, MEMTXATTRS_UNSPECIFIED, s, strlen(s) + 1); 422 } else { 423 WRITE_WORD(p, 0); 424 } 425 } 426 427 static int fdt_add_memory_node(void *fdt, uint32_t acells, hwaddr mem_base, 428 uint32_t scells, hwaddr mem_len, 429 int numa_node_id) 430 { 431 char *nodename; 432 int ret; 433 434 nodename = g_strdup_printf("/memory@%" PRIx64, mem_base); 435 qemu_fdt_add_subnode(fdt, nodename); 436 qemu_fdt_setprop_string(fdt, nodename, "device_type", "memory"); 437 ret = qemu_fdt_setprop_sized_cells(fdt, nodename, "reg", acells, mem_base, 438 scells, mem_len); 439 if (ret < 0) { 440 goto out; 441 } 442 443 /* only set the NUMA ID if it is specified */ 444 if (numa_node_id >= 0) { 445 ret = qemu_fdt_setprop_cell(fdt, nodename, 446 "numa-node-id", numa_node_id); 447 } 448 out: 449 g_free(nodename); 450 return ret; 451 } 452 453 static void fdt_add_psci_node(void *fdt) 454 { 455 uint32_t cpu_suspend_fn; 456 uint32_t cpu_off_fn; 457 uint32_t cpu_on_fn; 458 uint32_t migrate_fn; 459 ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(0)); 460 const char *psci_method; 461 int64_t psci_conduit; 462 int rc; 463 464 psci_conduit = object_property_get_int(OBJECT(armcpu), 465 "psci-conduit", 466 &error_abort); 467 switch (psci_conduit) { 468 case QEMU_PSCI_CONDUIT_DISABLED: 469 return; 470 case QEMU_PSCI_CONDUIT_HVC: 471 psci_method = "hvc"; 472 break; 473 case QEMU_PSCI_CONDUIT_SMC: 474 psci_method = "smc"; 475 break; 476 default: 477 g_assert_not_reached(); 478 } 479 480 /* 481 * A pre-existing /psci node might specify function ID values 482 * that don't match QEMU's PSCI implementation. Delete the whole 483 * node and put our own in instead. 484 */ 485 rc = fdt_path_offset(fdt, "/psci"); 486 if (rc >= 0) { 487 qemu_fdt_nop_node(fdt, "/psci"); 488 } 489 490 qemu_fdt_add_subnode(fdt, "/psci"); 491 if (armcpu->psci_version == 2) { 492 const char comp[] = "arm,psci-0.2\0arm,psci"; 493 qemu_fdt_setprop(fdt, "/psci", "compatible", comp, sizeof(comp)); 494 495 cpu_off_fn = QEMU_PSCI_0_2_FN_CPU_OFF; 496 if (arm_feature(&armcpu->env, ARM_FEATURE_AARCH64)) { 497 cpu_suspend_fn = QEMU_PSCI_0_2_FN64_CPU_SUSPEND; 498 cpu_on_fn = QEMU_PSCI_0_2_FN64_CPU_ON; 499 migrate_fn = QEMU_PSCI_0_2_FN64_MIGRATE; 500 } else { 501 cpu_suspend_fn = QEMU_PSCI_0_2_FN_CPU_SUSPEND; 502 cpu_on_fn = QEMU_PSCI_0_2_FN_CPU_ON; 503 migrate_fn = QEMU_PSCI_0_2_FN_MIGRATE; 504 } 505 } else { 506 qemu_fdt_setprop_string(fdt, "/psci", "compatible", "arm,psci"); 507 508 cpu_suspend_fn = QEMU_PSCI_0_1_FN_CPU_SUSPEND; 509 cpu_off_fn = QEMU_PSCI_0_1_FN_CPU_OFF; 510 cpu_on_fn = QEMU_PSCI_0_1_FN_CPU_ON; 511 migrate_fn = QEMU_PSCI_0_1_FN_MIGRATE; 512 } 513 514 /* We adopt the PSCI spec's nomenclature, and use 'conduit' to refer 515 * to the instruction that should be used to invoke PSCI functions. 516 * However, the device tree binding uses 'method' instead, so that is 517 * what we should use here. 518 */ 519 qemu_fdt_setprop_string(fdt, "/psci", "method", psci_method); 520 521 qemu_fdt_setprop_cell(fdt, "/psci", "cpu_suspend", cpu_suspend_fn); 522 qemu_fdt_setprop_cell(fdt, "/psci", "cpu_off", cpu_off_fn); 523 qemu_fdt_setprop_cell(fdt, "/psci", "cpu_on", cpu_on_fn); 524 qemu_fdt_setprop_cell(fdt, "/psci", "migrate", migrate_fn); 525 } 526 527 int arm_load_dtb(hwaddr addr, const struct arm_boot_info *binfo, 528 hwaddr addr_limit, AddressSpace *as, MachineState *ms) 529 { 530 void *fdt = NULL; 531 int size, rc, n = 0; 532 uint32_t acells, scells; 533 unsigned int i; 534 hwaddr mem_base, mem_len; 535 char **node_path; 536 Error *err = NULL; 537 538 if (binfo->dtb_filename) { 539 char *filename; 540 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, binfo->dtb_filename); 541 if (!filename) { 542 fprintf(stderr, "Couldn't open dtb file %s\n", binfo->dtb_filename); 543 goto fail; 544 } 545 546 fdt = load_device_tree(filename, &size); 547 if (!fdt) { 548 fprintf(stderr, "Couldn't open dtb file %s\n", filename); 549 g_free(filename); 550 goto fail; 551 } 552 g_free(filename); 553 } else { 554 fdt = binfo->get_dtb(binfo, &size); 555 if (!fdt) { 556 fprintf(stderr, "Board was unable to create a dtb blob\n"); 557 goto fail; 558 } 559 } 560 561 if (addr_limit > addr && size > (addr_limit - addr)) { 562 /* Installing the device tree blob at addr would exceed addr_limit. 563 * Whether this constitutes failure is up to the caller to decide, 564 * so just return 0 as size, i.e., no error. 565 */ 566 g_free(fdt); 567 return 0; 568 } 569 570 acells = qemu_fdt_getprop_cell(fdt, "/", "#address-cells", 571 NULL, &error_fatal); 572 scells = qemu_fdt_getprop_cell(fdt, "/", "#size-cells", 573 NULL, &error_fatal); 574 if (acells == 0 || scells == 0) { 575 fprintf(stderr, "dtb file invalid (#address-cells or #size-cells 0)\n"); 576 goto fail; 577 } 578 579 if (scells < 2 && binfo->ram_size >= 4 * GiB) { 580 /* This is user error so deserves a friendlier error message 581 * than the failure of setprop_sized_cells would provide 582 */ 583 fprintf(stderr, "qemu: dtb file not compatible with " 584 "RAM size > 4GB\n"); 585 goto fail; 586 } 587 588 /* nop all root nodes matching /memory or /memory@unit-address */ 589 node_path = qemu_fdt_node_unit_path(fdt, "memory", &err); 590 if (err) { 591 error_report_err(err); 592 goto fail; 593 } 594 while (node_path[n]) { 595 if (g_str_has_prefix(node_path[n], "/memory")) { 596 qemu_fdt_nop_node(fdt, node_path[n]); 597 } 598 n++; 599 } 600 g_strfreev(node_path); 601 602 /* 603 * We drop all the memory nodes which correspond to empty NUMA nodes 604 * from the device tree, because the Linux NUMA binding document 605 * states they should not be generated. Linux will get the NUMA node 606 * IDs of the empty NUMA nodes from the distance map if they are needed. 607 * This means QEMU users may be obliged to provide command lines which 608 * configure distance maps when the empty NUMA node IDs are needed and 609 * Linux's default distance map isn't sufficient. 610 */ 611 if (ms->numa_state != NULL && ms->numa_state->num_nodes > 0) { 612 mem_base = binfo->loader_start; 613 for (i = 0; i < ms->numa_state->num_nodes; i++) { 614 mem_len = ms->numa_state->nodes[i].node_mem; 615 if (!mem_len) { 616 continue; 617 } 618 619 rc = fdt_add_memory_node(fdt, acells, mem_base, 620 scells, mem_len, i); 621 if (rc < 0) { 622 fprintf(stderr, "couldn't add /memory@%"PRIx64" node\n", 623 mem_base); 624 goto fail; 625 } 626 627 mem_base += mem_len; 628 } 629 } else { 630 rc = fdt_add_memory_node(fdt, acells, binfo->loader_start, 631 scells, binfo->ram_size, -1); 632 if (rc < 0) { 633 fprintf(stderr, "couldn't add /memory@%"PRIx64" node\n", 634 binfo->loader_start); 635 goto fail; 636 } 637 } 638 639 rc = fdt_path_offset(fdt, "/chosen"); 640 if (rc < 0) { 641 qemu_fdt_add_subnode(fdt, "/chosen"); 642 } 643 644 if (ms->kernel_cmdline && *ms->kernel_cmdline) { 645 rc = qemu_fdt_setprop_string(fdt, "/chosen", "bootargs", 646 ms->kernel_cmdline); 647 if (rc < 0) { 648 fprintf(stderr, "couldn't set /chosen/bootargs\n"); 649 goto fail; 650 } 651 } 652 653 if (binfo->initrd_size) { 654 rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start", 655 binfo->initrd_start); 656 if (rc < 0) { 657 fprintf(stderr, "couldn't set /chosen/linux,initrd-start\n"); 658 goto fail; 659 } 660 661 rc = qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end", 662 binfo->initrd_start + binfo->initrd_size); 663 if (rc < 0) { 664 fprintf(stderr, "couldn't set /chosen/linux,initrd-end\n"); 665 goto fail; 666 } 667 } 668 669 fdt_add_psci_node(fdt); 670 671 if (binfo->modify_dtb) { 672 binfo->modify_dtb(binfo, fdt); 673 } 674 675 qemu_fdt_dumpdtb(fdt, size); 676 677 /* Put the DTB into the memory map as a ROM image: this will ensure 678 * the DTB is copied again upon reset, even if addr points into RAM. 679 */ 680 rom_add_blob_fixed_as("dtb", fdt, size, addr, as); 681 682 g_free(fdt); 683 684 return size; 685 686 fail: 687 g_free(fdt); 688 return -1; 689 } 690 691 static void do_cpu_reset(void *opaque) 692 { 693 ARMCPU *cpu = opaque; 694 CPUState *cs = CPU(cpu); 695 CPUARMState *env = &cpu->env; 696 const struct arm_boot_info *info = env->boot_info; 697 698 cpu_reset(cs); 699 if (info) { 700 if (!info->is_linux) { 701 int i; 702 /* Jump to the entry point. */ 703 uint64_t entry = info->entry; 704 705 switch (info->endianness) { 706 case ARM_ENDIANNESS_LE: 707 env->cp15.sctlr_el[1] &= ~SCTLR_E0E; 708 for (i = 1; i < 4; ++i) { 709 env->cp15.sctlr_el[i] &= ~SCTLR_EE; 710 } 711 env->uncached_cpsr &= ~CPSR_E; 712 break; 713 case ARM_ENDIANNESS_BE8: 714 env->cp15.sctlr_el[1] |= SCTLR_E0E; 715 for (i = 1; i < 4; ++i) { 716 env->cp15.sctlr_el[i] |= SCTLR_EE; 717 } 718 env->uncached_cpsr |= CPSR_E; 719 break; 720 case ARM_ENDIANNESS_BE32: 721 env->cp15.sctlr_el[1] |= SCTLR_B; 722 break; 723 case ARM_ENDIANNESS_UNKNOWN: 724 break; /* Board's decision */ 725 default: 726 g_assert_not_reached(); 727 } 728 729 cpu_set_pc(cs, entry); 730 } else { 731 /* If we are booting Linux then we need to check whether we are 732 * booting into secure or non-secure state and adjust the state 733 * accordingly. Out of reset, ARM is defined to be in secure state 734 * (SCR.NS = 0), we change that here if non-secure boot has been 735 * requested. 736 */ 737 if (arm_feature(env, ARM_FEATURE_EL3)) { 738 /* AArch64 is defined to come out of reset into EL3 if enabled. 739 * If we are booting Linux then we need to adjust our EL as 740 * Linux expects us to be in EL2 or EL1. AArch32 resets into 741 * SVC, which Linux expects, so no privilege/exception level to 742 * adjust. 743 */ 744 if (env->aarch64) { 745 env->cp15.scr_el3 |= SCR_RW; 746 if (arm_feature(env, ARM_FEATURE_EL2)) { 747 env->cp15.hcr_el2 |= HCR_RW; 748 env->pstate = PSTATE_MODE_EL2h; 749 } else { 750 env->pstate = PSTATE_MODE_EL1h; 751 } 752 if (cpu_isar_feature(aa64_pauth, cpu)) { 753 env->cp15.scr_el3 |= SCR_API | SCR_APK; 754 } 755 if (cpu_isar_feature(aa64_mte, cpu)) { 756 env->cp15.scr_el3 |= SCR_ATA; 757 } 758 if (cpu_isar_feature(aa64_sve, cpu)) { 759 env->cp15.cptr_el[3] |= CPTR_EZ; 760 } 761 /* AArch64 kernels never boot in secure mode */ 762 assert(!info->secure_boot); 763 /* This hook is only supported for AArch32 currently: 764 * bootloader_aarch64[] will not call the hook, and 765 * the code above has already dropped us into EL2 or EL1. 766 */ 767 assert(!info->secure_board_setup); 768 } 769 770 if (arm_feature(env, ARM_FEATURE_EL2)) { 771 /* If we have EL2 then Linux expects the HVC insn to work */ 772 env->cp15.scr_el3 |= SCR_HCE; 773 } 774 775 /* Set to non-secure if not a secure boot */ 776 if (!info->secure_boot && 777 (cs != first_cpu || !info->secure_board_setup)) { 778 /* Linux expects non-secure state */ 779 env->cp15.scr_el3 |= SCR_NS; 780 /* Set NSACR.{CP11,CP10} so NS can access the FPU */ 781 env->cp15.nsacr |= 3 << 10; 782 } 783 } 784 785 if (!env->aarch64 && !info->secure_boot && 786 arm_feature(env, ARM_FEATURE_EL2)) { 787 /* 788 * This is an AArch32 boot not to Secure state, and 789 * we have Hyp mode available, so boot the kernel into 790 * Hyp mode. This is not how the CPU comes out of reset, 791 * so we need to manually put it there. 792 */ 793 cpsr_write(env, ARM_CPU_MODE_HYP, CPSR_M, CPSRWriteRaw); 794 } 795 796 if (cs == first_cpu) { 797 AddressSpace *as = arm_boot_address_space(cpu, info); 798 799 cpu_set_pc(cs, info->loader_start); 800 801 if (!have_dtb(info)) { 802 if (old_param) { 803 set_kernel_args_old(info, as); 804 } else { 805 set_kernel_args(info, as); 806 } 807 } 808 } else if (info->secondary_cpu_reset_hook) { 809 info->secondary_cpu_reset_hook(cpu, info); 810 } 811 } 812 arm_rebuild_hflags(env); 813 } 814 } 815 816 /** 817 * load_image_to_fw_cfg() - Load an image file into an fw_cfg entry identified 818 * by key. 819 * @fw_cfg: The firmware config instance to store the data in. 820 * @size_key: The firmware config key to store the size of the loaded 821 * data under, with fw_cfg_add_i32(). 822 * @data_key: The firmware config key to store the loaded data under, 823 * with fw_cfg_add_bytes(). 824 * @image_name: The name of the image file to load. If it is NULL, the 825 * function returns without doing anything. 826 * @try_decompress: Whether the image should be decompressed (gunzipped) before 827 * adding it to fw_cfg. If decompression fails, the image is 828 * loaded as-is. 829 * 830 * In case of failure, the function prints an error message to stderr and the 831 * process exits with status 1. 832 */ 833 static void load_image_to_fw_cfg(FWCfgState *fw_cfg, uint16_t size_key, 834 uint16_t data_key, const char *image_name, 835 bool try_decompress) 836 { 837 size_t size = -1; 838 uint8_t *data; 839 840 if (image_name == NULL) { 841 return; 842 } 843 844 if (try_decompress) { 845 size = load_image_gzipped_buffer(image_name, 846 LOAD_IMAGE_MAX_GUNZIP_BYTES, &data); 847 } 848 849 if (size == (size_t)-1) { 850 gchar *contents; 851 gsize length; 852 853 if (!g_file_get_contents(image_name, &contents, &length, NULL)) { 854 error_report("failed to load \"%s\"", image_name); 855 exit(1); 856 } 857 size = length; 858 data = (uint8_t *)contents; 859 } 860 861 fw_cfg_add_i32(fw_cfg, size_key, size); 862 fw_cfg_add_bytes(fw_cfg, data_key, data, size); 863 } 864 865 static int do_arm_linux_init(Object *obj, void *opaque) 866 { 867 if (object_dynamic_cast(obj, TYPE_ARM_LINUX_BOOT_IF)) { 868 ARMLinuxBootIf *albif = ARM_LINUX_BOOT_IF(obj); 869 ARMLinuxBootIfClass *albifc = ARM_LINUX_BOOT_IF_GET_CLASS(obj); 870 struct arm_boot_info *info = opaque; 871 872 if (albifc->arm_linux_init) { 873 albifc->arm_linux_init(albif, info->secure_boot); 874 } 875 } 876 return 0; 877 } 878 879 static int64_t arm_load_elf(struct arm_boot_info *info, uint64_t *pentry, 880 uint64_t *lowaddr, uint64_t *highaddr, 881 int elf_machine, AddressSpace *as) 882 { 883 bool elf_is64; 884 union { 885 Elf32_Ehdr h32; 886 Elf64_Ehdr h64; 887 } elf_header; 888 int data_swab = 0; 889 bool big_endian; 890 int64_t ret = -1; 891 Error *err = NULL; 892 893 894 load_elf_hdr(info->kernel_filename, &elf_header, &elf_is64, &err); 895 if (err) { 896 error_free(err); 897 return ret; 898 } 899 900 if (elf_is64) { 901 big_endian = elf_header.h64.e_ident[EI_DATA] == ELFDATA2MSB; 902 info->endianness = big_endian ? ARM_ENDIANNESS_BE8 903 : ARM_ENDIANNESS_LE; 904 } else { 905 big_endian = elf_header.h32.e_ident[EI_DATA] == ELFDATA2MSB; 906 if (big_endian) { 907 if (bswap32(elf_header.h32.e_flags) & EF_ARM_BE8) { 908 info->endianness = ARM_ENDIANNESS_BE8; 909 } else { 910 info->endianness = ARM_ENDIANNESS_BE32; 911 /* In BE32, the CPU has a different view of the per-byte 912 * address map than the rest of the system. BE32 ELF files 913 * are organised such that they can be programmed through 914 * the CPU's per-word byte-reversed view of the world. QEMU 915 * however loads ELF files independently of the CPU. So 916 * tell the ELF loader to byte reverse the data for us. 917 */ 918 data_swab = 2; 919 } 920 } else { 921 info->endianness = ARM_ENDIANNESS_LE; 922 } 923 } 924 925 ret = load_elf_as(info->kernel_filename, NULL, NULL, NULL, 926 pentry, lowaddr, highaddr, NULL, big_endian, elf_machine, 927 1, data_swab, as); 928 if (ret <= 0) { 929 /* The header loaded but the image didn't */ 930 exit(1); 931 } 932 933 return ret; 934 } 935 936 static uint64_t load_aarch64_image(const char *filename, hwaddr mem_base, 937 hwaddr *entry, AddressSpace *as) 938 { 939 hwaddr kernel_load_offset = KERNEL64_LOAD_ADDR; 940 uint64_t kernel_size = 0; 941 uint8_t *buffer; 942 int size; 943 944 /* On aarch64, it's the bootloader's job to uncompress the kernel. */ 945 size = load_image_gzipped_buffer(filename, LOAD_IMAGE_MAX_GUNZIP_BYTES, 946 &buffer); 947 948 if (size < 0) { 949 gsize len; 950 951 /* Load as raw file otherwise */ 952 if (!g_file_get_contents(filename, (char **)&buffer, &len, NULL)) { 953 return -1; 954 } 955 size = len; 956 } 957 958 /* check the arm64 magic header value -- very old kernels may not have it */ 959 if (size > ARM64_MAGIC_OFFSET + 4 && 960 memcmp(buffer + ARM64_MAGIC_OFFSET, "ARM\x64", 4) == 0) { 961 uint64_t hdrvals[2]; 962 963 /* The arm64 Image header has text_offset and image_size fields at 8 and 964 * 16 bytes into the Image header, respectively. The text_offset field 965 * is only valid if the image_size is non-zero. 966 */ 967 memcpy(&hdrvals, buffer + ARM64_TEXT_OFFSET_OFFSET, sizeof(hdrvals)); 968 969 kernel_size = le64_to_cpu(hdrvals[1]); 970 971 if (kernel_size != 0) { 972 kernel_load_offset = le64_to_cpu(hdrvals[0]); 973 974 /* 975 * We write our startup "bootloader" at the very bottom of RAM, 976 * so that bit can't be used for the image. Luckily the Image 977 * format specification is that the image requests only an offset 978 * from a 2MB boundary, not an absolute load address. So if the 979 * image requests an offset that might mean it overlaps with the 980 * bootloader, we can just load it starting at 2MB+offset rather 981 * than 0MB + offset. 982 */ 983 if (kernel_load_offset < BOOTLOADER_MAX_SIZE) { 984 kernel_load_offset += 2 * MiB; 985 } 986 } 987 } 988 989 /* 990 * Kernels before v3.17 don't populate the image_size field, and 991 * raw images have no header. For those our best guess at the size 992 * is the size of the Image file itself. 993 */ 994 if (kernel_size == 0) { 995 kernel_size = size; 996 } 997 998 *entry = mem_base + kernel_load_offset; 999 rom_add_blob_fixed_as(filename, buffer, size, *entry, as); 1000 1001 g_free(buffer); 1002 1003 return kernel_size; 1004 } 1005 1006 static void arm_setup_direct_kernel_boot(ARMCPU *cpu, 1007 struct arm_boot_info *info) 1008 { 1009 /* Set up for a direct boot of a kernel image file. */ 1010 CPUState *cs; 1011 AddressSpace *as = arm_boot_address_space(cpu, info); 1012 int kernel_size; 1013 int initrd_size; 1014 int is_linux = 0; 1015 uint64_t elf_entry; 1016 /* Addresses of first byte used and first byte not used by the image */ 1017 uint64_t image_low_addr = 0, image_high_addr = 0; 1018 int elf_machine; 1019 hwaddr entry; 1020 static const ARMInsnFixup *primary_loader; 1021 uint64_t ram_end = info->loader_start + info->ram_size; 1022 1023 if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) { 1024 primary_loader = bootloader_aarch64; 1025 elf_machine = EM_AARCH64; 1026 } else { 1027 primary_loader = bootloader; 1028 if (!info->write_board_setup) { 1029 primary_loader += BOOTLOADER_NO_BOARD_SETUP_OFFSET; 1030 } 1031 elf_machine = EM_ARM; 1032 } 1033 1034 /* Assume that raw images are linux kernels, and ELF images are not. */ 1035 kernel_size = arm_load_elf(info, &elf_entry, &image_low_addr, 1036 &image_high_addr, elf_machine, as); 1037 if (kernel_size > 0 && have_dtb(info)) { 1038 /* 1039 * If there is still some room left at the base of RAM, try and put 1040 * the DTB there like we do for images loaded with -bios or -pflash. 1041 */ 1042 if (image_low_addr > info->loader_start 1043 || image_high_addr < info->loader_start) { 1044 /* 1045 * Set image_low_addr as address limit for arm_load_dtb if it may be 1046 * pointing into RAM, otherwise pass '0' (no limit) 1047 */ 1048 if (image_low_addr < info->loader_start) { 1049 image_low_addr = 0; 1050 } 1051 info->dtb_start = info->loader_start; 1052 info->dtb_limit = image_low_addr; 1053 } 1054 } 1055 entry = elf_entry; 1056 if (kernel_size < 0) { 1057 uint64_t loadaddr = info->loader_start + KERNEL_NOLOAD_ADDR; 1058 kernel_size = load_uimage_as(info->kernel_filename, &entry, &loadaddr, 1059 &is_linux, NULL, NULL, as); 1060 if (kernel_size >= 0) { 1061 image_low_addr = loadaddr; 1062 image_high_addr = image_low_addr + kernel_size; 1063 } 1064 } 1065 if (arm_feature(&cpu->env, ARM_FEATURE_AARCH64) && kernel_size < 0) { 1066 kernel_size = load_aarch64_image(info->kernel_filename, 1067 info->loader_start, &entry, as); 1068 is_linux = 1; 1069 if (kernel_size >= 0) { 1070 image_low_addr = entry; 1071 image_high_addr = image_low_addr + kernel_size; 1072 } 1073 } else if (kernel_size < 0) { 1074 /* 32-bit ARM */ 1075 entry = info->loader_start + KERNEL_LOAD_ADDR; 1076 kernel_size = load_image_targphys_as(info->kernel_filename, entry, 1077 ram_end - KERNEL_LOAD_ADDR, as); 1078 is_linux = 1; 1079 if (kernel_size >= 0) { 1080 image_low_addr = entry; 1081 image_high_addr = image_low_addr + kernel_size; 1082 } 1083 } 1084 if (kernel_size < 0) { 1085 error_report("could not load kernel '%s'", info->kernel_filename); 1086 exit(1); 1087 } 1088 1089 if (kernel_size > info->ram_size) { 1090 error_report("kernel '%s' is too large to fit in RAM " 1091 "(kernel size %d, RAM size %" PRId64 ")", 1092 info->kernel_filename, kernel_size, info->ram_size); 1093 exit(1); 1094 } 1095 1096 info->entry = entry; 1097 1098 /* 1099 * We want to put the initrd far enough into RAM that when the 1100 * kernel is uncompressed it will not clobber the initrd. However 1101 * on boards without much RAM we must ensure that we still leave 1102 * enough room for a decent sized initrd, and on boards with large 1103 * amounts of RAM we must avoid the initrd being so far up in RAM 1104 * that it is outside lowmem and inaccessible to the kernel. 1105 * So for boards with less than 256MB of RAM we put the initrd 1106 * halfway into RAM, and for boards with 256MB of RAM or more we put 1107 * the initrd at 128MB. 1108 * We also refuse to put the initrd somewhere that will definitely 1109 * overlay the kernel we just loaded, though for kernel formats which 1110 * don't tell us their exact size (eg self-decompressing 32-bit kernels) 1111 * we might still make a bad choice here. 1112 */ 1113 info->initrd_start = info->loader_start + 1114 MIN(info->ram_size / 2, 128 * MiB); 1115 if (image_high_addr) { 1116 info->initrd_start = MAX(info->initrd_start, image_high_addr); 1117 } 1118 info->initrd_start = TARGET_PAGE_ALIGN(info->initrd_start); 1119 1120 if (is_linux) { 1121 uint32_t fixupcontext[FIXUP_MAX]; 1122 1123 if (info->initrd_filename) { 1124 1125 if (info->initrd_start >= ram_end) { 1126 error_report("not enough space after kernel to load initrd"); 1127 exit(1); 1128 } 1129 1130 initrd_size = load_ramdisk_as(info->initrd_filename, 1131 info->initrd_start, 1132 ram_end - info->initrd_start, as); 1133 if (initrd_size < 0) { 1134 initrd_size = load_image_targphys_as(info->initrd_filename, 1135 info->initrd_start, 1136 ram_end - 1137 info->initrd_start, 1138 as); 1139 } 1140 if (initrd_size < 0) { 1141 error_report("could not load initrd '%s'", 1142 info->initrd_filename); 1143 exit(1); 1144 } 1145 if (info->initrd_start + initrd_size > ram_end) { 1146 error_report("could not load initrd '%s': " 1147 "too big to fit into RAM after the kernel", 1148 info->initrd_filename); 1149 exit(1); 1150 } 1151 } else { 1152 initrd_size = 0; 1153 } 1154 info->initrd_size = initrd_size; 1155 1156 fixupcontext[FIXUP_BOARDID] = info->board_id; 1157 fixupcontext[FIXUP_BOARD_SETUP] = info->board_setup_addr; 1158 1159 /* 1160 * for device tree boot, we pass the DTB directly in r2. Otherwise 1161 * we point to the kernel args. 1162 */ 1163 if (have_dtb(info)) { 1164 hwaddr align; 1165 1166 if (elf_machine == EM_AARCH64) { 1167 /* 1168 * Some AArch64 kernels on early bootup map the fdt region as 1169 * 1170 * [ ALIGN_DOWN(fdt, 2MB) ... ALIGN_DOWN(fdt, 2MB) + 2MB ] 1171 * 1172 * Let's play safe and prealign it to 2MB to give us some space. 1173 */ 1174 align = 2 * MiB; 1175 } else { 1176 /* 1177 * Some 32bit kernels will trash anything in the 4K page the 1178 * initrd ends in, so make sure the DTB isn't caught up in that. 1179 */ 1180 align = 4 * KiB; 1181 } 1182 1183 /* Place the DTB after the initrd in memory with alignment. */ 1184 info->dtb_start = QEMU_ALIGN_UP(info->initrd_start + initrd_size, 1185 align); 1186 if (info->dtb_start >= ram_end) { 1187 error_report("Not enough space for DTB after kernel/initrd"); 1188 exit(1); 1189 } 1190 fixupcontext[FIXUP_ARGPTR_LO] = info->dtb_start; 1191 fixupcontext[FIXUP_ARGPTR_HI] = info->dtb_start >> 32; 1192 } else { 1193 fixupcontext[FIXUP_ARGPTR_LO] = 1194 info->loader_start + KERNEL_ARGS_ADDR; 1195 fixupcontext[FIXUP_ARGPTR_HI] = 1196 (info->loader_start + KERNEL_ARGS_ADDR) >> 32; 1197 if (info->ram_size >= 4 * GiB) { 1198 error_report("RAM size must be less than 4GB to boot" 1199 " Linux kernel using ATAGS (try passing a device tree" 1200 " using -dtb)"); 1201 exit(1); 1202 } 1203 } 1204 fixupcontext[FIXUP_ENTRYPOINT_LO] = entry; 1205 fixupcontext[FIXUP_ENTRYPOINT_HI] = entry >> 32; 1206 1207 write_bootloader("bootloader", info->loader_start, 1208 primary_loader, fixupcontext, as); 1209 1210 if (info->write_board_setup) { 1211 info->write_board_setup(cpu, info); 1212 } 1213 1214 /* 1215 * Notify devices which need to fake up firmware initialization 1216 * that we're doing a direct kernel boot. 1217 */ 1218 object_child_foreach_recursive(object_get_root(), 1219 do_arm_linux_init, info); 1220 } 1221 info->is_linux = is_linux; 1222 1223 for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) { 1224 ARM_CPU(cs)->env.boot_info = info; 1225 } 1226 } 1227 1228 static void arm_setup_firmware_boot(ARMCPU *cpu, struct arm_boot_info *info) 1229 { 1230 /* Set up for booting firmware (which might load a kernel via fw_cfg) */ 1231 1232 if (have_dtb(info)) { 1233 /* 1234 * If we have a device tree blob, but no kernel to supply it to (or 1235 * the kernel is supposed to be loaded by the bootloader), copy the 1236 * DTB to the base of RAM for the bootloader to pick up. 1237 */ 1238 info->dtb_start = info->loader_start; 1239 } 1240 1241 if (info->kernel_filename) { 1242 FWCfgState *fw_cfg; 1243 bool try_decompressing_kernel; 1244 1245 fw_cfg = fw_cfg_find(); 1246 1247 if (!fw_cfg) { 1248 error_report("This machine type does not support loading both " 1249 "a guest firmware/BIOS image and a guest kernel at " 1250 "the same time. You should change your QEMU command " 1251 "line to specify one or the other, but not both."); 1252 exit(1); 1253 } 1254 1255 try_decompressing_kernel = arm_feature(&cpu->env, 1256 ARM_FEATURE_AARCH64); 1257 1258 /* 1259 * Expose the kernel, the command line, and the initrd in fw_cfg. 1260 * We don't process them here at all, it's all left to the 1261 * firmware. 1262 */ 1263 load_image_to_fw_cfg(fw_cfg, 1264 FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA, 1265 info->kernel_filename, 1266 try_decompressing_kernel); 1267 load_image_to_fw_cfg(fw_cfg, 1268 FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA, 1269 info->initrd_filename, false); 1270 1271 if (info->kernel_cmdline) { 1272 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, 1273 strlen(info->kernel_cmdline) + 1); 1274 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, 1275 info->kernel_cmdline); 1276 } 1277 } 1278 1279 /* 1280 * We will start from address 0 (typically a boot ROM image) in the 1281 * same way as hardware. Leave env->boot_info NULL, so that 1282 * do_cpu_reset() knows it does not need to alter the PC on reset. 1283 */ 1284 } 1285 1286 void arm_load_kernel(ARMCPU *cpu, MachineState *ms, struct arm_boot_info *info) 1287 { 1288 CPUState *cs; 1289 AddressSpace *as = arm_boot_address_space(cpu, info); 1290 int boot_el; 1291 CPUARMState *env = &cpu->env; 1292 int nb_cpus = 0; 1293 1294 /* 1295 * CPU objects (unlike devices) are not automatically reset on system 1296 * reset, so we must always register a handler to do so. If we're 1297 * actually loading a kernel, the handler is also responsible for 1298 * arranging that we start it correctly. 1299 */ 1300 for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) { 1301 qemu_register_reset(do_cpu_reset, ARM_CPU(cs)); 1302 nb_cpus++; 1303 } 1304 1305 /* 1306 * The board code is not supposed to set secure_board_setup unless 1307 * running its code in secure mode is actually possible, and KVM 1308 * doesn't support secure. 1309 */ 1310 assert(!(info->secure_board_setup && kvm_enabled())); 1311 info->kernel_filename = ms->kernel_filename; 1312 info->kernel_cmdline = ms->kernel_cmdline; 1313 info->initrd_filename = ms->initrd_filename; 1314 info->dtb_filename = ms->dtb; 1315 info->dtb_limit = 0; 1316 1317 /* Load the kernel. */ 1318 if (!info->kernel_filename || info->firmware_loaded) { 1319 arm_setup_firmware_boot(cpu, info); 1320 } else { 1321 arm_setup_direct_kernel_boot(cpu, info); 1322 } 1323 1324 /* 1325 * Disable the PSCI conduit if it is set up to target the same 1326 * or a lower EL than the one we're going to start the guest code in. 1327 * This logic needs to agree with the code in do_cpu_reset() which 1328 * decides whether we're going to boot the guest in the highest 1329 * supported exception level or in a lower one. 1330 */ 1331 1332 /* 1333 * If PSCI is enabled, then SMC calls all go to the PSCI handler and 1334 * are never emulated to trap into guest code. It therefore does not 1335 * make sense for the board to have a setup code fragment that runs 1336 * in Secure, because this will probably need to itself issue an SMC of some 1337 * kind as part of its operation. 1338 */ 1339 assert(info->psci_conduit == QEMU_PSCI_CONDUIT_DISABLED || 1340 !info->secure_board_setup); 1341 1342 /* Boot into highest supported EL ... */ 1343 if (arm_feature(env, ARM_FEATURE_EL3)) { 1344 boot_el = 3; 1345 } else if (arm_feature(env, ARM_FEATURE_EL2)) { 1346 boot_el = 2; 1347 } else { 1348 boot_el = 1; 1349 } 1350 /* ...except that if we're booting Linux we adjust the EL we boot into */ 1351 if (info->is_linux && !info->secure_boot) { 1352 boot_el = arm_feature(env, ARM_FEATURE_EL2) ? 2 : 1; 1353 } 1354 1355 if ((info->psci_conduit == QEMU_PSCI_CONDUIT_HVC && boot_el >= 2) || 1356 (info->psci_conduit == QEMU_PSCI_CONDUIT_SMC && boot_el == 3)) { 1357 info->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED; 1358 } 1359 1360 if (info->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED) { 1361 for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) { 1362 Object *cpuobj = OBJECT(cs); 1363 1364 object_property_set_int(cpuobj, "psci-conduit", info->psci_conduit, 1365 &error_abort); 1366 /* 1367 * Secondary CPUs start in PSCI powered-down state. Like the 1368 * code in do_cpu_reset(), we assume first_cpu is the primary 1369 * CPU. 1370 */ 1371 if (cs != first_cpu) { 1372 object_property_set_bool(cpuobj, "start-powered-off", true, 1373 &error_abort); 1374 } 1375 } 1376 } 1377 1378 if (info->psci_conduit == QEMU_PSCI_CONDUIT_DISABLED && 1379 info->is_linux && nb_cpus > 1) { 1380 /* 1381 * We're booting Linux but not using PSCI, so for SMP we need 1382 * to write a custom secondary CPU boot loader stub, and arrange 1383 * for the secondary CPU reset to make the accompanying initialization. 1384 */ 1385 if (!info->secondary_cpu_reset_hook) { 1386 info->secondary_cpu_reset_hook = default_reset_secondary; 1387 } 1388 if (!info->write_secondary_boot) { 1389 info->write_secondary_boot = default_write_secondary; 1390 } 1391 info->write_secondary_boot(cpu, info); 1392 } else { 1393 /* 1394 * No secondary boot stub; don't use the reset hook that would 1395 * have set the CPU up to call it 1396 */ 1397 info->write_secondary_boot = NULL; 1398 info->secondary_cpu_reset_hook = NULL; 1399 } 1400 1401 /* 1402 * arm_load_dtb() may add a PSCI node so it must be called after we have 1403 * decided whether to enable PSCI and set the psci-conduit CPU properties. 1404 */ 1405 if (!info->skip_dtb_autoload && have_dtb(info)) { 1406 if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as, ms) < 0) { 1407 exit(1); 1408 } 1409 } 1410 } 1411 1412 static const TypeInfo arm_linux_boot_if_info = { 1413 .name = TYPE_ARM_LINUX_BOOT_IF, 1414 .parent = TYPE_INTERFACE, 1415 .class_size = sizeof(ARMLinuxBootIfClass), 1416 }; 1417 1418 static void arm_linux_boot_register_types(void) 1419 { 1420 type_register_static(&arm_linux_boot_if_info); 1421 } 1422 1423 type_init(arm_linux_boot_register_types) 1424