1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * linux/boot/head.S 4 * 5 * Copyright (C) 1991, 1992, 1993 Linus Torvalds 6 */ 7 8/* 9 * head.S contains the 32-bit startup code. 10 * 11 * NOTE!!! Startup happens at absolute address 0x00001000, which is also where 12 * the page directory will exist. The startup code will be overwritten by 13 * the page directory. [According to comments etc elsewhere on a compressed 14 * kernel it will end up at 0x1000 + 1Mb I hope so as I assume this. - AC] 15 * 16 * Page 0 is deliberately kept safe, since System Management Mode code in 17 * laptops may need to access the BIOS data stored there. This is also 18 * useful for future device drivers that either access the BIOS via VM86 19 * mode. 20 */ 21 22/* 23 * High loaded stuff by Hans Lermen & Werner Almesberger, Feb. 1996 24 */ 25 .code32 26 .text 27 28#include <linux/init.h> 29#include <linux/linkage.h> 30#include <asm/segment.h> 31#include <asm/boot.h> 32#include <asm/msr.h> 33#include <asm/processor-flags.h> 34#include <asm/asm-offsets.h> 35#include <asm/bootparam.h> 36 37/* 38 * Locally defined symbols should be marked hidden: 39 */ 40 .hidden _bss 41 .hidden _ebss 42 .hidden _got 43 .hidden _egot 44 45 __HEAD 46 .code32 47ENTRY(startup_32) 48 /* 49 * 32bit entry is 0 and it is ABI so immutable! 50 * If we come here directly from a bootloader, 51 * kernel(text+data+bss+brk) ramdisk, zero_page, command line 52 * all need to be under the 4G limit. 53 */ 54 cld 55 /* 56 * Test KEEP_SEGMENTS flag to see if the bootloader is asking 57 * us to not reload segments 58 */ 59 testb $KEEP_SEGMENTS, BP_loadflags(%esi) 60 jnz 1f 61 62 cli 63 movl $(__BOOT_DS), %eax 64 movl %eax, %ds 65 movl %eax, %es 66 movl %eax, %ss 671: 68 69/* 70 * Calculate the delta between where we were compiled to run 71 * at and where we were actually loaded at. This can only be done 72 * with a short local call on x86. Nothing else will tell us what 73 * address we are running at. The reserved chunk of the real-mode 74 * data at 0x1e4 (defined as a scratch field) are used as the stack 75 * for this calculation. Only 4 bytes are needed. 76 */ 77 leal (BP_scratch+4)(%esi), %esp 78 call 1f 791: popl %ebp 80 subl $1b, %ebp 81 82/* setup a stack and make sure cpu supports long mode. */ 83 movl $boot_stack_end, %eax 84 addl %ebp, %eax 85 movl %eax, %esp 86 87 call verify_cpu 88 testl %eax, %eax 89 jnz no_longmode 90 91/* 92 * Compute the delta between where we were compiled to run at 93 * and where the code will actually run at. 94 * 95 * %ebp contains the address we are loaded at by the boot loader and %ebx 96 * contains the address where we should move the kernel image temporarily 97 * for safe in-place decompression. 98 */ 99 100#ifdef CONFIG_RELOCATABLE 101 movl %ebp, %ebx 102 movl BP_kernel_alignment(%esi), %eax 103 decl %eax 104 addl %eax, %ebx 105 notl %eax 106 andl %eax, %ebx 107 cmpl $LOAD_PHYSICAL_ADDR, %ebx 108 jge 1f 109#endif 110 movl $LOAD_PHYSICAL_ADDR, %ebx 1111: 112 113 /* Target address to relocate to for decompression */ 114 movl BP_init_size(%esi), %eax 115 subl $_end, %eax 116 addl %eax, %ebx 117 118/* 119 * Prepare for entering 64 bit mode 120 */ 121 122 /* Load new GDT with the 64bit segments using 32bit descriptor */ 123 addl %ebp, gdt+2(%ebp) 124 lgdt gdt(%ebp) 125 126 /* Enable PAE mode */ 127 movl %cr4, %eax 128 orl $X86_CR4_PAE, %eax 129 movl %eax, %cr4 130 131 /* 132 * Build early 4G boot pagetable 133 */ 134 /* Initialize Page tables to 0 */ 135 leal pgtable(%ebx), %edi 136 xorl %eax, %eax 137 movl $(BOOT_INIT_PGT_SIZE/4), %ecx 138 rep stosl 139 140 /* Build Level 4 */ 141 leal pgtable + 0(%ebx), %edi 142 leal 0x1007 (%edi), %eax 143 movl %eax, 0(%edi) 144 145 /* Build Level 3 */ 146 leal pgtable + 0x1000(%ebx), %edi 147 leal 0x1007(%edi), %eax 148 movl $4, %ecx 1491: movl %eax, 0x00(%edi) 150 addl $0x00001000, %eax 151 addl $8, %edi 152 decl %ecx 153 jnz 1b 154 155 /* Build Level 2 */ 156 leal pgtable + 0x2000(%ebx), %edi 157 movl $0x00000183, %eax 158 movl $2048, %ecx 1591: movl %eax, 0(%edi) 160 addl $0x00200000, %eax 161 addl $8, %edi 162 decl %ecx 163 jnz 1b 164 165 /* Enable the boot page tables */ 166 leal pgtable(%ebx), %eax 167 movl %eax, %cr3 168 169 /* Enable Long mode in EFER (Extended Feature Enable Register) */ 170 movl $MSR_EFER, %ecx 171 rdmsr 172 btsl $_EFER_LME, %eax 173 wrmsr 174 175 /* After gdt is loaded */ 176 xorl %eax, %eax 177 lldt %ax 178 movl $__BOOT_TSS, %eax 179 ltr %ax 180 181 /* 182 * Setup for the jump to 64bit mode 183 * 184 * When the jump is performend we will be in long mode but 185 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1 186 * (and in turn EFER.LMA = 1). To jump into 64bit mode we use 187 * the new gdt/idt that has __KERNEL_CS with CS.L = 1. 188 * We place all of the values on our mini stack so lret can 189 * used to perform that far jump. 190 */ 191 pushl $__KERNEL_CS 192 leal startup_64(%ebp), %eax 193#ifdef CONFIG_EFI_MIXED 194 movl efi32_config(%ebp), %ebx 195 cmp $0, %ebx 196 jz 1f 197 leal handover_entry(%ebp), %eax 1981: 199#endif 200 pushl %eax 201 202 /* Enter paged protected Mode, activating Long Mode */ 203 movl $(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */ 204 movl %eax, %cr0 205 206 /* Jump from 32bit compatibility mode into 64bit mode. */ 207 lret 208ENDPROC(startup_32) 209 210#ifdef CONFIG_EFI_MIXED 211 .org 0x190 212ENTRY(efi32_stub_entry) 213 add $0x4, %esp /* Discard return address */ 214 popl %ecx 215 popl %edx 216 popl %esi 217 218 leal (BP_scratch+4)(%esi), %esp 219 call 1f 2201: pop %ebp 221 subl $1b, %ebp 222 223 movl %ecx, efi32_config(%ebp) 224 movl %edx, efi32_config+8(%ebp) 225 sgdtl efi32_boot_gdt(%ebp) 226 227 leal efi32_config(%ebp), %eax 228 movl %eax, efi_config(%ebp) 229 230 /* Disable paging */ 231 movl %cr0, %eax 232 btrl $X86_CR0_PG_BIT, %eax 233 movl %eax, %cr0 234 235 jmp startup_32 236ENDPROC(efi32_stub_entry) 237#endif 238 239 .code64 240 .org 0x200 241ENTRY(startup_64) 242 /* 243 * 64bit entry is 0x200 and it is ABI so immutable! 244 * We come here either from startup_32 or directly from a 245 * 64bit bootloader. 246 * If we come here from a bootloader, kernel(text+data+bss+brk), 247 * ramdisk, zero_page, command line could be above 4G. 248 * We depend on an identity mapped page table being provided 249 * that maps our entire kernel(text+data+bss+brk), zero page 250 * and command line. 251 */ 252 253 /* Setup data segments. */ 254 xorl %eax, %eax 255 movl %eax, %ds 256 movl %eax, %es 257 movl %eax, %ss 258 movl %eax, %fs 259 movl %eax, %gs 260 261 /* 262 * Compute the decompressed kernel start address. It is where 263 * we were loaded at aligned to a 2M boundary. %rbp contains the 264 * decompressed kernel start address. 265 * 266 * If it is a relocatable kernel then decompress and run the kernel 267 * from load address aligned to 2MB addr, otherwise decompress and 268 * run the kernel from LOAD_PHYSICAL_ADDR 269 * 270 * We cannot rely on the calculation done in 32-bit mode, since we 271 * may have been invoked via the 64-bit entry point. 272 */ 273 274 /* Start with the delta to where the kernel will run at. */ 275#ifdef CONFIG_RELOCATABLE 276 leaq startup_32(%rip) /* - $startup_32 */, %rbp 277 movl BP_kernel_alignment(%rsi), %eax 278 decl %eax 279 addq %rax, %rbp 280 notq %rax 281 andq %rax, %rbp 282 cmpq $LOAD_PHYSICAL_ADDR, %rbp 283 jge 1f 284#endif 285 movq $LOAD_PHYSICAL_ADDR, %rbp 2861: 287 288 /* Target address to relocate to for decompression */ 289 movl BP_init_size(%rsi), %ebx 290 subl $_end, %ebx 291 addq %rbp, %rbx 292 293 /* Set up the stack */ 294 leaq boot_stack_end(%rbx), %rsp 295 296#ifdef CONFIG_X86_5LEVEL 297 /* 298 * Check if we need to enable 5-level paging. 299 * RSI holds real mode data and need to be preserved across 300 * a function call. 301 */ 302 pushq %rsi 303 call l5_paging_required 304 popq %rsi 305 306 /* If l5_paging_required() returned zero, we're done here. */ 307 cmpq $0, %rax 308 je lvl5 309 310 /* 311 * At this point we are in long mode with 4-level paging enabled, 312 * but we want to enable 5-level paging. 313 * 314 * The problem is that we cannot do it directly. Setting LA57 in 315 * long mode would trigger #GP. So we need to switch off long mode 316 * first. 317 * 318 * NOTE: This is not going to work if bootloader put us above 4G 319 * limit. 320 * 321 * The first step is go into compatibility mode. 322 */ 323 324 /* Clear additional page table */ 325 leaq lvl5_pgtable(%rbx), %rdi 326 xorq %rax, %rax 327 movq $(PAGE_SIZE/8), %rcx 328 rep stosq 329 330 /* 331 * Setup current CR3 as the first and only entry in a new top level 332 * page table. 333 */ 334 movq %cr3, %rdi 335 leaq 0x7 (%rdi), %rax 336 movq %rax, lvl5_pgtable(%rbx) 337 338 /* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */ 339 pushq $__KERNEL32_CS 340 leaq compatible_mode(%rip), %rax 341 pushq %rax 342 lretq 343lvl5: 344#endif 345 346 /* Zero EFLAGS */ 347 pushq $0 348 popfq 349 350/* 351 * Copy the compressed kernel to the end of our buffer 352 * where decompression in place becomes safe. 353 */ 354 pushq %rsi 355 leaq (_bss-8)(%rip), %rsi 356 leaq (_bss-8)(%rbx), %rdi 357 movq $_bss /* - $startup_32 */, %rcx 358 shrq $3, %rcx 359 std 360 rep movsq 361 cld 362 popq %rsi 363 364/* 365 * Jump to the relocated address. 366 */ 367 leaq relocated(%rbx), %rax 368 jmp *%rax 369 370#ifdef CONFIG_EFI_STUB 371 372/* The entry point for the PE/COFF executable is efi_pe_entry. */ 373ENTRY(efi_pe_entry) 374 movq %rcx, efi64_config(%rip) /* Handle */ 375 movq %rdx, efi64_config+8(%rip) /* EFI System table pointer */ 376 377 leaq efi64_config(%rip), %rax 378 movq %rax, efi_config(%rip) 379 380 call 1f 3811: popq %rbp 382 subq $1b, %rbp 383 384 /* 385 * Relocate efi_config->call(). 386 */ 387 addq %rbp, efi64_config+40(%rip) 388 389 movq %rax, %rdi 390 call make_boot_params 391 cmpq $0,%rax 392 je fail 393 mov %rax, %rsi 394 leaq startup_32(%rip), %rax 395 movl %eax, BP_code32_start(%rsi) 396 jmp 2f /* Skip the relocation */ 397 398handover_entry: 399 call 1f 4001: popq %rbp 401 subq $1b, %rbp 402 403 /* 404 * Relocate efi_config->call(). 405 */ 406 movq efi_config(%rip), %rax 407 addq %rbp, 40(%rax) 4082: 409 movq efi_config(%rip), %rdi 410 call efi_main 411 movq %rax,%rsi 412 cmpq $0,%rax 413 jne 2f 414fail: 415 /* EFI init failed, so hang. */ 416 hlt 417 jmp fail 4182: 419 movl BP_code32_start(%esi), %eax 420 leaq startup_64(%rax), %rax 421 jmp *%rax 422ENDPROC(efi_pe_entry) 423 424 .org 0x390 425ENTRY(efi64_stub_entry) 426 movq %rdi, efi64_config(%rip) /* Handle */ 427 movq %rsi, efi64_config+8(%rip) /* EFI System table pointer */ 428 429 leaq efi64_config(%rip), %rax 430 movq %rax, efi_config(%rip) 431 432 movq %rdx, %rsi 433 jmp handover_entry 434ENDPROC(efi64_stub_entry) 435#endif 436 437 .text 438relocated: 439 440/* 441 * Clear BSS (stack is currently empty) 442 */ 443 xorl %eax, %eax 444 leaq _bss(%rip), %rdi 445 leaq _ebss(%rip), %rcx 446 subq %rdi, %rcx 447 shrq $3, %rcx 448 rep stosq 449 450/* 451 * Adjust our own GOT 452 */ 453 leaq _got(%rip), %rdx 454 leaq _egot(%rip), %rcx 4551: 456 cmpq %rcx, %rdx 457 jae 2f 458 addq %rbx, (%rdx) 459 addq $8, %rdx 460 jmp 1b 4612: 462 463/* 464 * Do the extraction, and jump to the new kernel.. 465 */ 466 pushq %rsi /* Save the real mode argument */ 467 movq %rsi, %rdi /* real mode address */ 468 leaq boot_heap(%rip), %rsi /* malloc area for uncompression */ 469 leaq input_data(%rip), %rdx /* input_data */ 470 movl $z_input_len, %ecx /* input_len */ 471 movq %rbp, %r8 /* output target address */ 472 movq $z_output_len, %r9 /* decompressed length, end of relocs */ 473 call extract_kernel /* returns kernel location in %rax */ 474 popq %rsi 475 476/* 477 * Jump to the decompressed kernel. 478 */ 479 jmp *%rax 480 481 .code32 482#ifdef CONFIG_X86_5LEVEL 483compatible_mode: 484 /* Setup data and stack segments */ 485 movl $__KERNEL_DS, %eax 486 movl %eax, %ds 487 movl %eax, %ss 488 489 /* Disable paging */ 490 movl %cr0, %eax 491 btrl $X86_CR0_PG_BIT, %eax 492 movl %eax, %cr0 493 494 /* Point CR3 to 5-level paging */ 495 leal lvl5_pgtable(%ebx), %eax 496 movl %eax, %cr3 497 498 /* Enable PAE and LA57 mode */ 499 movl %cr4, %eax 500 orl $(X86_CR4_PAE | X86_CR4_LA57), %eax 501 movl %eax, %cr4 502 503 /* Calculate address we are running at */ 504 call 1f 5051: popl %edi 506 subl $1b, %edi 507 508 /* Prepare stack for far return to Long Mode */ 509 pushl $__KERNEL_CS 510 leal lvl5(%edi), %eax 511 push %eax 512 513 /* Enable paging back */ 514 movl $(X86_CR0_PG | X86_CR0_PE), %eax 515 movl %eax, %cr0 516 517 lret 518#endif 519 520no_longmode: 521 /* This isn't an x86-64 CPU so hang */ 5221: 523 hlt 524 jmp 1b 525 526#include "../../kernel/verify_cpu.S" 527 528 .data 529gdt: 530 .word gdt_end - gdt 531 .long gdt 532 .word 0 533 .quad 0x00cf9a000000ffff /* __KERNEL32_CS */ 534 .quad 0x00af9a000000ffff /* __KERNEL_CS */ 535 .quad 0x00cf92000000ffff /* __KERNEL_DS */ 536 .quad 0x0080890000000000 /* TS descriptor */ 537 .quad 0x0000000000000000 /* TS continued */ 538gdt_end: 539 540#ifdef CONFIG_EFI_STUB 541efi_config: 542 .quad 0 543 544#ifdef CONFIG_EFI_MIXED 545 .global efi32_config 546efi32_config: 547 .fill 5,8,0 548 .quad efi64_thunk 549 .byte 0 550#endif 551 552 .global efi64_config 553efi64_config: 554 .fill 5,8,0 555 .quad efi_call 556 .byte 1 557#endif /* CONFIG_EFI_STUB */ 558 559/* 560 * Stack and heap for uncompression 561 */ 562 .bss 563 .balign 4 564boot_heap: 565 .fill BOOT_HEAP_SIZE, 1, 0 566boot_stack: 567 .fill BOOT_STACK_SIZE, 1, 0 568boot_stack_end: 569 570/* 571 * Space for page tables (not in .bss so not zeroed) 572 */ 573 .section ".pgtable","a",@nobits 574 .balign 4096 575pgtable: 576 .fill BOOT_PGT_SIZE, 1, 0 577#ifdef CONFIG_X86_5LEVEL 578lvl5_pgtable: 579 .fill PAGE_SIZE, 1, 0 580#endif 581