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 /* 135 * If SEV is active then set the encryption mask in the page tables. 136 * This will insure that when the kernel is copied and decompressed 137 * it will be done so encrypted. 138 */ 139 call get_sev_encryption_bit 140 xorl %edx, %edx 141 testl %eax, %eax 142 jz 1f 143 subl $32, %eax /* Encryption bit is always above bit 31 */ 144 bts %eax, %edx /* Set encryption mask for page tables */ 1451: 146 147 /* Initialize Page tables to 0 */ 148 leal pgtable(%ebx), %edi 149 xorl %eax, %eax 150 movl $(BOOT_INIT_PGT_SIZE/4), %ecx 151 rep stosl 152 153 /* Build Level 4 */ 154 leal pgtable + 0(%ebx), %edi 155 leal 0x1007 (%edi), %eax 156 movl %eax, 0(%edi) 157 addl %edx, 4(%edi) 158 159 /* Build Level 3 */ 160 leal pgtable + 0x1000(%ebx), %edi 161 leal 0x1007(%edi), %eax 162 movl $4, %ecx 1631: movl %eax, 0x00(%edi) 164 addl %edx, 0x04(%edi) 165 addl $0x00001000, %eax 166 addl $8, %edi 167 decl %ecx 168 jnz 1b 169 170 /* Build Level 2 */ 171 leal pgtable + 0x2000(%ebx), %edi 172 movl $0x00000183, %eax 173 movl $2048, %ecx 1741: movl %eax, 0(%edi) 175 addl %edx, 4(%edi) 176 addl $0x00200000, %eax 177 addl $8, %edi 178 decl %ecx 179 jnz 1b 180 181 /* Enable the boot page tables */ 182 leal pgtable(%ebx), %eax 183 movl %eax, %cr3 184 185 /* Enable Long mode in EFER (Extended Feature Enable Register) */ 186 movl $MSR_EFER, %ecx 187 rdmsr 188 btsl $_EFER_LME, %eax 189 wrmsr 190 191 /* After gdt is loaded */ 192 xorl %eax, %eax 193 lldt %ax 194 movl $__BOOT_TSS, %eax 195 ltr %ax 196 197 /* 198 * Setup for the jump to 64bit mode 199 * 200 * When the jump is performend we will be in long mode but 201 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1 202 * (and in turn EFER.LMA = 1). To jump into 64bit mode we use 203 * the new gdt/idt that has __KERNEL_CS with CS.L = 1. 204 * We place all of the values on our mini stack so lret can 205 * used to perform that far jump. 206 */ 207 pushl $__KERNEL_CS 208 leal startup_64(%ebp), %eax 209#ifdef CONFIG_EFI_MIXED 210 movl efi32_config(%ebp), %ebx 211 cmp $0, %ebx 212 jz 1f 213 leal handover_entry(%ebp), %eax 2141: 215#endif 216 pushl %eax 217 218 /* Enter paged protected Mode, activating Long Mode */ 219 movl $(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */ 220 movl %eax, %cr0 221 222 /* Jump from 32bit compatibility mode into 64bit mode. */ 223 lret 224ENDPROC(startup_32) 225 226#ifdef CONFIG_EFI_MIXED 227 .org 0x190 228ENTRY(efi32_stub_entry) 229 add $0x4, %esp /* Discard return address */ 230 popl %ecx 231 popl %edx 232 popl %esi 233 234 leal (BP_scratch+4)(%esi), %esp 235 call 1f 2361: pop %ebp 237 subl $1b, %ebp 238 239 movl %ecx, efi32_config(%ebp) 240 movl %edx, efi32_config+8(%ebp) 241 sgdtl efi32_boot_gdt(%ebp) 242 243 leal efi32_config(%ebp), %eax 244 movl %eax, efi_config(%ebp) 245 246 jmp startup_32 247ENDPROC(efi32_stub_entry) 248#endif 249 250 .code64 251 .org 0x200 252ENTRY(startup_64) 253 /* 254 * 64bit entry is 0x200 and it is ABI so immutable! 255 * We come here either from startup_32 or directly from a 256 * 64bit bootloader. 257 * If we come here from a bootloader, kernel(text+data+bss+brk), 258 * ramdisk, zero_page, command line could be above 4G. 259 * We depend on an identity mapped page table being provided 260 * that maps our entire kernel(text+data+bss+brk), zero page 261 * and command line. 262 */ 263 264 /* Setup data segments. */ 265 xorl %eax, %eax 266 movl %eax, %ds 267 movl %eax, %es 268 movl %eax, %ss 269 movl %eax, %fs 270 movl %eax, %gs 271 272 /* 273 * Compute the decompressed kernel start address. It is where 274 * we were loaded at aligned to a 2M boundary. %rbp contains the 275 * decompressed kernel start address. 276 * 277 * If it is a relocatable kernel then decompress and run the kernel 278 * from load address aligned to 2MB addr, otherwise decompress and 279 * run the kernel from LOAD_PHYSICAL_ADDR 280 * 281 * We cannot rely on the calculation done in 32-bit mode, since we 282 * may have been invoked via the 64-bit entry point. 283 */ 284 285 /* Start with the delta to where the kernel will run at. */ 286#ifdef CONFIG_RELOCATABLE 287 leaq startup_32(%rip) /* - $startup_32 */, %rbp 288 movl BP_kernel_alignment(%rsi), %eax 289 decl %eax 290 addq %rax, %rbp 291 notq %rax 292 andq %rax, %rbp 293 cmpq $LOAD_PHYSICAL_ADDR, %rbp 294 jge 1f 295#endif 296 movq $LOAD_PHYSICAL_ADDR, %rbp 2971: 298 299 /* Target address to relocate to for decompression */ 300 movl BP_init_size(%rsi), %ebx 301 subl $_end, %ebx 302 addq %rbp, %rbx 303 304 /* Set up the stack */ 305 leaq boot_stack_end(%rbx), %rsp 306 307#ifdef CONFIG_X86_5LEVEL 308 /* 309 * Check if we need to enable 5-level paging. 310 * RSI holds real mode data and need to be preserved across 311 * a function call. 312 */ 313 pushq %rsi 314 call l5_paging_required 315 popq %rsi 316 317 /* If l5_paging_required() returned zero, we're done here. */ 318 cmpq $0, %rax 319 je lvl5 320 321 /* 322 * At this point we are in long mode with 4-level paging enabled, 323 * but we want to enable 5-level paging. 324 * 325 * The problem is that we cannot do it directly. Setting LA57 in 326 * long mode would trigger #GP. So we need to switch off long mode 327 * first. 328 * 329 * NOTE: This is not going to work if bootloader put us above 4G 330 * limit. 331 * 332 * The first step is go into compatibility mode. 333 */ 334 335 /* Clear additional page table */ 336 leaq lvl5_pgtable(%rbx), %rdi 337 xorq %rax, %rax 338 movq $(PAGE_SIZE/8), %rcx 339 rep stosq 340 341 /* 342 * Setup current CR3 as the first and only entry in a new top level 343 * page table. 344 */ 345 movq %cr3, %rdi 346 leaq 0x7 (%rdi), %rax 347 movq %rax, lvl5_pgtable(%rbx) 348 349 /* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */ 350 pushq $__KERNEL32_CS 351 leaq compatible_mode(%rip), %rax 352 pushq %rax 353 lretq 354lvl5: 355#endif 356 357 /* Zero EFLAGS */ 358 pushq $0 359 popfq 360 361/* 362 * Copy the compressed kernel to the end of our buffer 363 * where decompression in place becomes safe. 364 */ 365 pushq %rsi 366 leaq (_bss-8)(%rip), %rsi 367 leaq (_bss-8)(%rbx), %rdi 368 movq $_bss /* - $startup_32 */, %rcx 369 shrq $3, %rcx 370 std 371 rep movsq 372 cld 373 popq %rsi 374 375/* 376 * Jump to the relocated address. 377 */ 378 leaq relocated(%rbx), %rax 379 jmp *%rax 380 381#ifdef CONFIG_EFI_STUB 382 383/* The entry point for the PE/COFF executable is efi_pe_entry. */ 384ENTRY(efi_pe_entry) 385 movq %rcx, efi64_config(%rip) /* Handle */ 386 movq %rdx, efi64_config+8(%rip) /* EFI System table pointer */ 387 388 leaq efi64_config(%rip), %rax 389 movq %rax, efi_config(%rip) 390 391 call 1f 3921: popq %rbp 393 subq $1b, %rbp 394 395 /* 396 * Relocate efi_config->call(). 397 */ 398 addq %rbp, efi64_config+40(%rip) 399 400 movq %rax, %rdi 401 call make_boot_params 402 cmpq $0,%rax 403 je fail 404 mov %rax, %rsi 405 leaq startup_32(%rip), %rax 406 movl %eax, BP_code32_start(%rsi) 407 jmp 2f /* Skip the relocation */ 408 409handover_entry: 410 call 1f 4111: popq %rbp 412 subq $1b, %rbp 413 414 /* 415 * Relocate efi_config->call(). 416 */ 417 movq efi_config(%rip), %rax 418 addq %rbp, 40(%rax) 4192: 420 movq efi_config(%rip), %rdi 421 call efi_main 422 movq %rax,%rsi 423 cmpq $0,%rax 424 jne 2f 425fail: 426 /* EFI init failed, so hang. */ 427 hlt 428 jmp fail 4292: 430 movl BP_code32_start(%esi), %eax 431 leaq startup_64(%rax), %rax 432 jmp *%rax 433ENDPROC(efi_pe_entry) 434 435 .org 0x390 436ENTRY(efi64_stub_entry) 437 movq %rdi, efi64_config(%rip) /* Handle */ 438 movq %rsi, efi64_config+8(%rip) /* EFI System table pointer */ 439 440 leaq efi64_config(%rip), %rax 441 movq %rax, efi_config(%rip) 442 443 movq %rdx, %rsi 444 jmp handover_entry 445ENDPROC(efi64_stub_entry) 446#endif 447 448 .text 449relocated: 450 451/* 452 * Clear BSS (stack is currently empty) 453 */ 454 xorl %eax, %eax 455 leaq _bss(%rip), %rdi 456 leaq _ebss(%rip), %rcx 457 subq %rdi, %rcx 458 shrq $3, %rcx 459 rep stosq 460 461/* 462 * Adjust our own GOT 463 */ 464 leaq _got(%rip), %rdx 465 leaq _egot(%rip), %rcx 4661: 467 cmpq %rcx, %rdx 468 jae 2f 469 addq %rbx, (%rdx) 470 addq $8, %rdx 471 jmp 1b 4722: 473 474/* 475 * Do the extraction, and jump to the new kernel.. 476 */ 477 pushq %rsi /* Save the real mode argument */ 478 movq %rsi, %rdi /* real mode address */ 479 leaq boot_heap(%rip), %rsi /* malloc area for uncompression */ 480 leaq input_data(%rip), %rdx /* input_data */ 481 movl $z_input_len, %ecx /* input_len */ 482 movq %rbp, %r8 /* output target address */ 483 movq $z_output_len, %r9 /* decompressed length, end of relocs */ 484 call extract_kernel /* returns kernel location in %rax */ 485 popq %rsi 486 487/* 488 * Jump to the decompressed kernel. 489 */ 490 jmp *%rax 491 492 .code32 493#ifdef CONFIG_X86_5LEVEL 494compatible_mode: 495 /* Setup data and stack segments */ 496 movl $__KERNEL_DS, %eax 497 movl %eax, %ds 498 movl %eax, %ss 499 500 /* Disable paging */ 501 movl %cr0, %eax 502 btrl $X86_CR0_PG_BIT, %eax 503 movl %eax, %cr0 504 505 /* Point CR3 to 5-level paging */ 506 leal lvl5_pgtable(%ebx), %eax 507 movl %eax, %cr3 508 509 /* Enable PAE and LA57 mode */ 510 movl %cr4, %eax 511 orl $(X86_CR4_PAE | X86_CR4_LA57), %eax 512 movl %eax, %cr4 513 514 /* Calculate address we are running at */ 515 call 1f 5161: popl %edi 517 subl $1b, %edi 518 519 /* Prepare stack for far return to Long Mode */ 520 pushl $__KERNEL_CS 521 leal lvl5(%edi), %eax 522 push %eax 523 524 /* Enable paging back */ 525 movl $(X86_CR0_PG | X86_CR0_PE), %eax 526 movl %eax, %cr0 527 528 lret 529#endif 530 531no_longmode: 532 /* This isn't an x86-64 CPU so hang */ 5331: 534 hlt 535 jmp 1b 536 537#include "../../kernel/verify_cpu.S" 538 539 .data 540gdt: 541 .word gdt_end - gdt 542 .long gdt 543 .word 0 544 .quad 0x00cf9a000000ffff /* __KERNEL32_CS */ 545 .quad 0x00af9a000000ffff /* __KERNEL_CS */ 546 .quad 0x00cf92000000ffff /* __KERNEL_DS */ 547 .quad 0x0080890000000000 /* TS descriptor */ 548 .quad 0x0000000000000000 /* TS continued */ 549gdt_end: 550 551#ifdef CONFIG_EFI_STUB 552efi_config: 553 .quad 0 554 555#ifdef CONFIG_EFI_MIXED 556 .global efi32_config 557efi32_config: 558 .fill 5,8,0 559 .quad efi64_thunk 560 .byte 0 561#endif 562 563 .global efi64_config 564efi64_config: 565 .fill 5,8,0 566 .quad efi_call 567 .byte 1 568#endif /* CONFIG_EFI_STUB */ 569 570/* 571 * Stack and heap for uncompression 572 */ 573 .bss 574 .balign 4 575boot_heap: 576 .fill BOOT_HEAP_SIZE, 1, 0 577boot_stack: 578 .fill BOOT_STACK_SIZE, 1, 0 579boot_stack_end: 580 581/* 582 * Space for page tables (not in .bss so not zeroed) 583 */ 584 .section ".pgtable","a",@nobits 585 .balign 4096 586pgtable: 587 .fill BOOT_PGT_SIZE, 1, 0 588#ifdef CONFIG_X86_5LEVEL 589lvl5_pgtable: 590 .fill PAGE_SIZE, 1, 0 591#endif 592