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 jmp startup_32 231ENDPROC(efi32_stub_entry) 232#endif 233 234 .code64 235 .org 0x200 236ENTRY(startup_64) 237 /* 238 * 64bit entry is 0x200 and it is ABI so immutable! 239 * We come here either from startup_32 or directly from a 240 * 64bit bootloader. 241 * If we come here from a bootloader, kernel(text+data+bss+brk), 242 * ramdisk, zero_page, command line could be above 4G. 243 * We depend on an identity mapped page table being provided 244 * that maps our entire kernel(text+data+bss+brk), zero page 245 * and command line. 246 */ 247 248 /* Setup data segments. */ 249 xorl %eax, %eax 250 movl %eax, %ds 251 movl %eax, %es 252 movl %eax, %ss 253 movl %eax, %fs 254 movl %eax, %gs 255 256 /* 257 * Compute the decompressed kernel start address. It is where 258 * we were loaded at aligned to a 2M boundary. %rbp contains the 259 * decompressed kernel start address. 260 * 261 * If it is a relocatable kernel then decompress and run the kernel 262 * from load address aligned to 2MB addr, otherwise decompress and 263 * run the kernel from LOAD_PHYSICAL_ADDR 264 * 265 * We cannot rely on the calculation done in 32-bit mode, since we 266 * may have been invoked via the 64-bit entry point. 267 */ 268 269 /* Start with the delta to where the kernel will run at. */ 270#ifdef CONFIG_RELOCATABLE 271 leaq startup_32(%rip) /* - $startup_32 */, %rbp 272 movl BP_kernel_alignment(%rsi), %eax 273 decl %eax 274 addq %rax, %rbp 275 notq %rax 276 andq %rax, %rbp 277 cmpq $LOAD_PHYSICAL_ADDR, %rbp 278 jge 1f 279#endif 280 movq $LOAD_PHYSICAL_ADDR, %rbp 2811: 282 283 /* Target address to relocate to for decompression */ 284 movl BP_init_size(%rsi), %ebx 285 subl $_end, %ebx 286 addq %rbp, %rbx 287 288 /* Set up the stack */ 289 leaq boot_stack_end(%rbx), %rsp 290 291#ifdef CONFIG_X86_5LEVEL 292 /* Check if 5-level paging has already enabled */ 293 movq %cr4, %rax 294 testl $X86_CR4_LA57, %eax 295 jnz lvl5 296 297 /* 298 * At this point we are in long mode with 4-level paging enabled, 299 * but we want to enable 5-level paging. 300 * 301 * The problem is that we cannot do it directly. Setting LA57 in 302 * long mode would trigger #GP. So we need to switch off long mode 303 * first. 304 * 305 * NOTE: This is not going to work if bootloader put us above 4G 306 * limit. 307 * 308 * The first step is go into compatibility mode. 309 */ 310 311 /* Clear additional page table */ 312 leaq lvl5_pgtable(%rbx), %rdi 313 xorq %rax, %rax 314 movq $(PAGE_SIZE/8), %rcx 315 rep stosq 316 317 /* 318 * Setup current CR3 as the first and only entry in a new top level 319 * page table. 320 */ 321 movq %cr3, %rdi 322 leaq 0x7 (%rdi), %rax 323 movq %rax, lvl5_pgtable(%rbx) 324 325 /* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */ 326 pushq $__KERNEL32_CS 327 leaq compatible_mode(%rip), %rax 328 pushq %rax 329 lretq 330lvl5: 331#endif 332 333 /* Zero EFLAGS */ 334 pushq $0 335 popfq 336 337/* 338 * Copy the compressed kernel to the end of our buffer 339 * where decompression in place becomes safe. 340 */ 341 pushq %rsi 342 leaq (_bss-8)(%rip), %rsi 343 leaq (_bss-8)(%rbx), %rdi 344 movq $_bss /* - $startup_32 */, %rcx 345 shrq $3, %rcx 346 std 347 rep movsq 348 cld 349 popq %rsi 350 351/* 352 * Jump to the relocated address. 353 */ 354 leaq relocated(%rbx), %rax 355 jmp *%rax 356 357#ifdef CONFIG_EFI_STUB 358 359/* The entry point for the PE/COFF executable is efi_pe_entry. */ 360ENTRY(efi_pe_entry) 361 movq %rcx, efi64_config(%rip) /* Handle */ 362 movq %rdx, efi64_config+8(%rip) /* EFI System table pointer */ 363 364 leaq efi64_config(%rip), %rax 365 movq %rax, efi_config(%rip) 366 367 call 1f 3681: popq %rbp 369 subq $1b, %rbp 370 371 /* 372 * Relocate efi_config->call(). 373 */ 374 addq %rbp, efi64_config+40(%rip) 375 376 movq %rax, %rdi 377 call make_boot_params 378 cmpq $0,%rax 379 je fail 380 mov %rax, %rsi 381 leaq startup_32(%rip), %rax 382 movl %eax, BP_code32_start(%rsi) 383 jmp 2f /* Skip the relocation */ 384 385handover_entry: 386 call 1f 3871: popq %rbp 388 subq $1b, %rbp 389 390 /* 391 * Relocate efi_config->call(). 392 */ 393 movq efi_config(%rip), %rax 394 addq %rbp, 40(%rax) 3952: 396 movq efi_config(%rip), %rdi 397 call efi_main 398 movq %rax,%rsi 399 cmpq $0,%rax 400 jne 2f 401fail: 402 /* EFI init failed, so hang. */ 403 hlt 404 jmp fail 4052: 406 movl BP_code32_start(%esi), %eax 407 leaq startup_64(%rax), %rax 408 jmp *%rax 409ENDPROC(efi_pe_entry) 410 411 .org 0x390 412ENTRY(efi64_stub_entry) 413 movq %rdi, efi64_config(%rip) /* Handle */ 414 movq %rsi, efi64_config+8(%rip) /* EFI System table pointer */ 415 416 leaq efi64_config(%rip), %rax 417 movq %rax, efi_config(%rip) 418 419 movq %rdx, %rsi 420 jmp handover_entry 421ENDPROC(efi64_stub_entry) 422#endif 423 424 .text 425relocated: 426 427/* 428 * Clear BSS (stack is currently empty) 429 */ 430 xorl %eax, %eax 431 leaq _bss(%rip), %rdi 432 leaq _ebss(%rip), %rcx 433 subq %rdi, %rcx 434 shrq $3, %rcx 435 rep stosq 436 437/* 438 * Adjust our own GOT 439 */ 440 leaq _got(%rip), %rdx 441 leaq _egot(%rip), %rcx 4421: 443 cmpq %rcx, %rdx 444 jae 2f 445 addq %rbx, (%rdx) 446 addq $8, %rdx 447 jmp 1b 4482: 449 450/* 451 * Do the extraction, and jump to the new kernel.. 452 */ 453 pushq %rsi /* Save the real mode argument */ 454 movq %rsi, %rdi /* real mode address */ 455 leaq boot_heap(%rip), %rsi /* malloc area for uncompression */ 456 leaq input_data(%rip), %rdx /* input_data */ 457 movl $z_input_len, %ecx /* input_len */ 458 movq %rbp, %r8 /* output target address */ 459 movq $z_output_len, %r9 /* decompressed length, end of relocs */ 460 call extract_kernel /* returns kernel location in %rax */ 461 popq %rsi 462 463/* 464 * Jump to the decompressed kernel. 465 */ 466 jmp *%rax 467 468 .code32 469#ifdef CONFIG_X86_5LEVEL 470compatible_mode: 471 /* Setup data and stack segments */ 472 movl $__KERNEL_DS, %eax 473 movl %eax, %ds 474 movl %eax, %ss 475 476 /* Disable paging */ 477 movl %cr0, %eax 478 btrl $X86_CR0_PG_BIT, %eax 479 movl %eax, %cr0 480 481 /* Point CR3 to 5-level paging */ 482 leal lvl5_pgtable(%ebx), %eax 483 movl %eax, %cr3 484 485 /* Enable PAE and LA57 mode */ 486 movl %cr4, %eax 487 orl $(X86_CR4_PAE | X86_CR4_LA57), %eax 488 movl %eax, %cr4 489 490 /* Calculate address we are running at */ 491 call 1f 4921: popl %edi 493 subl $1b, %edi 494 495 /* Prepare stack for far return to Long Mode */ 496 pushl $__KERNEL_CS 497 leal lvl5(%edi), %eax 498 push %eax 499 500 /* Enable paging back */ 501 movl $(X86_CR0_PG | X86_CR0_PE), %eax 502 movl %eax, %cr0 503 504 lret 505#endif 506 507no_longmode: 508 /* This isn't an x86-64 CPU so hang */ 5091: 510 hlt 511 jmp 1b 512 513#include "../../kernel/verify_cpu.S" 514 515 .data 516gdt: 517 .word gdt_end - gdt 518 .long gdt 519 .word 0 520 .quad 0x00cf9a000000ffff /* __KERNEL32_CS */ 521 .quad 0x00af9a000000ffff /* __KERNEL_CS */ 522 .quad 0x00cf92000000ffff /* __KERNEL_DS */ 523 .quad 0x0080890000000000 /* TS descriptor */ 524 .quad 0x0000000000000000 /* TS continued */ 525gdt_end: 526 527#ifdef CONFIG_EFI_STUB 528efi_config: 529 .quad 0 530 531#ifdef CONFIG_EFI_MIXED 532 .global efi32_config 533efi32_config: 534 .fill 5,8,0 535 .quad efi64_thunk 536 .byte 0 537#endif 538 539 .global efi64_config 540efi64_config: 541 .fill 5,8,0 542 .quad efi_call 543 .byte 1 544#endif /* CONFIG_EFI_STUB */ 545 546/* 547 * Stack and heap for uncompression 548 */ 549 .bss 550 .balign 4 551boot_heap: 552 .fill BOOT_HEAP_SIZE, 1, 0 553boot_stack: 554 .fill BOOT_STACK_SIZE, 1, 0 555boot_stack_end: 556 557/* 558 * Space for page tables (not in .bss so not zeroed) 559 */ 560 .section ".pgtable","a",@nobits 561 .balign 4096 562pgtable: 563 .fill BOOT_PGT_SIZE, 1, 0 564#ifdef CONFIG_X86_5LEVEL 565lvl5_pgtable: 566 .fill PAGE_SIZE, 1, 0 567#endif 568