1/* 2 * linux/boot/head.S 3 * 4 * Copyright (C) 1991, 1992, 1993 Linus Torvalds 5 */ 6 7/* 8 * head.S contains the 32-bit startup code. 9 * 10 * NOTE!!! Startup happens at absolute address 0x00001000, which is also where 11 * the page directory will exist. The startup code will be overwritten by 12 * the page directory. [According to comments etc elsewhere on a compressed 13 * kernel it will end up at 0x1000 + 1Mb I hope so as I assume this. - AC] 14 * 15 * Page 0 is deliberately kept safe, since System Management Mode code in 16 * laptops may need to access the BIOS data stored there. This is also 17 * useful for future device drivers that either access the BIOS via VM86 18 * mode. 19 */ 20 21/* 22 * High loaded stuff by Hans Lermen & Werner Almesberger, Feb. 1996 23 */ 24 .code32 25 .text 26 27#include <linux/init.h> 28#include <linux/linkage.h> 29#include <asm/segment.h> 30#include <asm/boot.h> 31#include <asm/msr.h> 32#include <asm/processor-flags.h> 33#include <asm/asm-offsets.h> 34#include <asm/bootparam.h> 35 36/* 37 * Locally defined symbols should be marked hidden: 38 */ 39 .hidden _bss 40 .hidden _ebss 41 .hidden _got 42 .hidden _egot 43 44 __HEAD 45 .code32 46ENTRY(startup_32) 47 /* 48 * 32bit entry is 0 and it is ABI so immutable! 49 * If we come here directly from a bootloader, 50 * kernel(text+data+bss+brk) ramdisk, zero_page, command line 51 * all need to be under the 4G limit. 52 */ 53 cld 54 /* 55 * Test KEEP_SEGMENTS flag to see if the bootloader is asking 56 * us to not reload segments 57 */ 58 testb $KEEP_SEGMENTS, BP_loadflags(%esi) 59 jnz 1f 60 61 cli 62 movl $(__BOOT_DS), %eax 63 movl %eax, %ds 64 movl %eax, %es 65 movl %eax, %ss 661: 67 68/* 69 * Calculate the delta between where we were compiled to run 70 * at and where we were actually loaded at. This can only be done 71 * with a short local call on x86. Nothing else will tell us what 72 * address we are running at. The reserved chunk of the real-mode 73 * data at 0x1e4 (defined as a scratch field) are used as the stack 74 * for this calculation. Only 4 bytes are needed. 75 */ 76 leal (BP_scratch+4)(%esi), %esp 77 call 1f 781: popl %ebp 79 subl $1b, %ebp 80 81/* setup a stack and make sure cpu supports long mode. */ 82 movl $boot_stack_end, %eax 83 addl %ebp, %eax 84 movl %eax, %esp 85 86 call verify_cpu 87 testl %eax, %eax 88 jnz no_longmode 89 90/* 91 * Compute the delta between where we were compiled to run at 92 * and where the code will actually run at. 93 * 94 * %ebp contains the address we are loaded at by the boot loader and %ebx 95 * contains the address where we should move the kernel image temporarily 96 * for safe in-place decompression. 97 */ 98 99#ifdef CONFIG_RELOCATABLE 100 movl %ebp, %ebx 101 movl BP_kernel_alignment(%esi), %eax 102 decl %eax 103 addl %eax, %ebx 104 notl %eax 105 andl %eax, %ebx 106 cmpl $LOAD_PHYSICAL_ADDR, %ebx 107 jge 1f 108#endif 109 movl $LOAD_PHYSICAL_ADDR, %ebx 1101: 111 112 /* Target address to relocate to for decompression */ 113 movl BP_init_size(%esi), %eax 114 subl $_end, %eax 115 addl %eax, %ebx 116 117/* 118 * Prepare for entering 64 bit mode 119 */ 120 121 /* Load new GDT with the 64bit segments using 32bit descriptor */ 122 addl %ebp, gdt+2(%ebp) 123 lgdt gdt(%ebp) 124 125 /* Enable PAE mode */ 126 movl %cr4, %eax 127 orl $X86_CR4_PAE, %eax 128 movl %eax, %cr4 129 130 /* 131 * Build early 4G boot pagetable 132 */ 133 /* Initialize Page tables to 0 */ 134 leal pgtable(%ebx), %edi 135 xorl %eax, %eax 136 movl $(BOOT_INIT_PGT_SIZE/4), %ecx 137 rep stosl 138 139 /* Build Level 4 */ 140 leal pgtable + 0(%ebx), %edi 141 leal 0x1007 (%edi), %eax 142 movl %eax, 0(%edi) 143 144 /* Build Level 3 */ 145 leal pgtable + 0x1000(%ebx), %edi 146 leal 0x1007(%edi), %eax 147 movl $4, %ecx 1481: movl %eax, 0x00(%edi) 149 addl $0x00001000, %eax 150 addl $8, %edi 151 decl %ecx 152 jnz 1b 153 154 /* Build Level 2 */ 155 leal pgtable + 0x2000(%ebx), %edi 156 movl $0x00000183, %eax 157 movl $2048, %ecx 1581: movl %eax, 0(%edi) 159 addl $0x00200000, %eax 160 addl $8, %edi 161 decl %ecx 162 jnz 1b 163 164 /* Enable the boot page tables */ 165 leal pgtable(%ebx), %eax 166 movl %eax, %cr3 167 168 /* Enable Long mode in EFER (Extended Feature Enable Register) */ 169 movl $MSR_EFER, %ecx 170 rdmsr 171 btsl $_EFER_LME, %eax 172 wrmsr 173 174 /* After gdt is loaded */ 175 xorl %eax, %eax 176 lldt %ax 177 movl $__BOOT_TSS, %eax 178 ltr %ax 179 180 /* 181 * Setup for the jump to 64bit mode 182 * 183 * When the jump is performend we will be in long mode but 184 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1 185 * (and in turn EFER.LMA = 1). To jump into 64bit mode we use 186 * the new gdt/idt that has __KERNEL_CS with CS.L = 1. 187 * We place all of the values on our mini stack so lret can 188 * used to perform that far jump. 189 */ 190 pushl $__KERNEL_CS 191 leal startup_64(%ebp), %eax 192#ifdef CONFIG_EFI_MIXED 193 movl efi32_config(%ebp), %ebx 194 cmp $0, %ebx 195 jz 1f 196 leal handover_entry(%ebp), %eax 1971: 198#endif 199 pushl %eax 200 201 /* Enter paged protected Mode, activating Long Mode */ 202 movl $(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */ 203 movl %eax, %cr0 204 205 /* Jump from 32bit compatibility mode into 64bit mode. */ 206 lret 207ENDPROC(startup_32) 208 209#ifdef CONFIG_EFI_MIXED 210 .org 0x190 211ENTRY(efi32_stub_entry) 212 add $0x4, %esp /* Discard return address */ 213 popl %ecx 214 popl %edx 215 popl %esi 216 217 leal (BP_scratch+4)(%esi), %esp 218 call 1f 2191: pop %ebp 220 subl $1b, %ebp 221 222 movl %ecx, efi32_config(%ebp) 223 movl %edx, efi32_config+8(%ebp) 224 sgdtl efi32_boot_gdt(%ebp) 225 226 leal efi32_config(%ebp), %eax 227 movl %eax, efi_config(%ebp) 228 229 jmp startup_32 230ENDPROC(efi32_stub_entry) 231#endif 232 233 .code64 234 .org 0x200 235ENTRY(startup_64) 236 /* 237 * 64bit entry is 0x200 and it is ABI so immutable! 238 * We come here either from startup_32 or directly from a 239 * 64bit bootloader. 240 * If we come here from a bootloader, kernel(text+data+bss+brk), 241 * ramdisk, zero_page, command line could be above 4G. 242 * We depend on an identity mapped page table being provided 243 * that maps our entire kernel(text+data+bss+brk), zero page 244 * and command line. 245 */ 246 247 /* Setup data segments. */ 248 xorl %eax, %eax 249 movl %eax, %ds 250 movl %eax, %es 251 movl %eax, %ss 252 movl %eax, %fs 253 movl %eax, %gs 254 255 /* 256 * Compute the decompressed kernel start address. It is where 257 * we were loaded at aligned to a 2M boundary. %rbp contains the 258 * decompressed kernel start address. 259 * 260 * If it is a relocatable kernel then decompress and run the kernel 261 * from load address aligned to 2MB addr, otherwise decompress and 262 * run the kernel from LOAD_PHYSICAL_ADDR 263 * 264 * We cannot rely on the calculation done in 32-bit mode, since we 265 * may have been invoked via the 64-bit entry point. 266 */ 267 268 /* Start with the delta to where the kernel will run at. */ 269#ifdef CONFIG_RELOCATABLE 270 leaq startup_32(%rip) /* - $startup_32 */, %rbp 271 movl BP_kernel_alignment(%rsi), %eax 272 decl %eax 273 addq %rax, %rbp 274 notq %rax 275 andq %rax, %rbp 276 cmpq $LOAD_PHYSICAL_ADDR, %rbp 277 jge 1f 278#endif 279 movq $LOAD_PHYSICAL_ADDR, %rbp 2801: 281 282 /* Target address to relocate to for decompression */ 283 movl BP_init_size(%rsi), %ebx 284 subl $_end, %ebx 285 addq %rbp, %rbx 286 287 /* Set up the stack */ 288 leaq boot_stack_end(%rbx), %rsp 289 290#ifdef CONFIG_X86_5LEVEL 291 /* Check if 5-level paging has already enabled */ 292 movq %cr4, %rax 293 testl $X86_CR4_LA57, %eax 294 jnz lvl5 295 296 /* 297 * At this point we are in long mode with 4-level paging enabled, 298 * but we want to enable 5-level paging. 299 * 300 * The problem is that we cannot do it directly. Setting LA57 in 301 * long mode would trigger #GP. So we need to switch off long mode 302 * first. 303 * 304 * NOTE: This is not going to work if bootloader put us above 4G 305 * limit. 306 * 307 * The first step is go into compatibility mode. 308 */ 309 310 /* Clear additional page table */ 311 leaq lvl5_pgtable(%rbx), %rdi 312 xorq %rax, %rax 313 movq $(PAGE_SIZE/8), %rcx 314 rep stosq 315 316 /* 317 * Setup current CR3 as the first and only entry in a new top level 318 * page table. 319 */ 320 movq %cr3, %rdi 321 leaq 0x7 (%rdi), %rax 322 movq %rax, lvl5_pgtable(%rbx) 323 324 /* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */ 325 pushq $__KERNEL32_CS 326 leaq compatible_mode(%rip), %rax 327 pushq %rax 328 lretq 329lvl5: 330#endif 331 332 /* Zero EFLAGS */ 333 pushq $0 334 popfq 335 336/* 337 * Copy the compressed kernel to the end of our buffer 338 * where decompression in place becomes safe. 339 */ 340 pushq %rsi 341 leaq (_bss-8)(%rip), %rsi 342 leaq (_bss-8)(%rbx), %rdi 343 movq $_bss /* - $startup_32 */, %rcx 344 shrq $3, %rcx 345 std 346 rep movsq 347 cld 348 popq %rsi 349 350/* 351 * Jump to the relocated address. 352 */ 353 leaq relocated(%rbx), %rax 354 jmp *%rax 355 356#ifdef CONFIG_EFI_STUB 357 358/* The entry point for the PE/COFF executable is efi_pe_entry. */ 359ENTRY(efi_pe_entry) 360 movq %rcx, efi64_config(%rip) /* Handle */ 361 movq %rdx, efi64_config+8(%rip) /* EFI System table pointer */ 362 363 leaq efi64_config(%rip), %rax 364 movq %rax, efi_config(%rip) 365 366 call 1f 3671: popq %rbp 368 subq $1b, %rbp 369 370 /* 371 * Relocate efi_config->call(). 372 */ 373 addq %rbp, efi64_config+40(%rip) 374 375 movq %rax, %rdi 376 call make_boot_params 377 cmpq $0,%rax 378 je fail 379 mov %rax, %rsi 380 leaq startup_32(%rip), %rax 381 movl %eax, BP_code32_start(%rsi) 382 jmp 2f /* Skip the relocation */ 383 384handover_entry: 385 call 1f 3861: popq %rbp 387 subq $1b, %rbp 388 389 /* 390 * Relocate efi_config->call(). 391 */ 392 movq efi_config(%rip), %rax 393 addq %rbp, 40(%rax) 3942: 395 movq efi_config(%rip), %rdi 396 call efi_main 397 movq %rax,%rsi 398 cmpq $0,%rax 399 jne 2f 400fail: 401 /* EFI init failed, so hang. */ 402 hlt 403 jmp fail 4042: 405 movl BP_code32_start(%esi), %eax 406 leaq startup_64(%rax), %rax 407 jmp *%rax 408ENDPROC(efi_pe_entry) 409 410 .org 0x390 411ENTRY(efi64_stub_entry) 412 movq %rdi, efi64_config(%rip) /* Handle */ 413 movq %rsi, efi64_config+8(%rip) /* EFI System table pointer */ 414 415 leaq efi64_config(%rip), %rax 416 movq %rax, efi_config(%rip) 417 418 movq %rdx, %rsi 419 jmp handover_entry 420ENDPROC(efi64_stub_entry) 421#endif 422 423 .text 424relocated: 425 426/* 427 * Clear BSS (stack is currently empty) 428 */ 429 xorl %eax, %eax 430 leaq _bss(%rip), %rdi 431 leaq _ebss(%rip), %rcx 432 subq %rdi, %rcx 433 shrq $3, %rcx 434 rep stosq 435 436/* 437 * Adjust our own GOT 438 */ 439 leaq _got(%rip), %rdx 440 leaq _egot(%rip), %rcx 4411: 442 cmpq %rcx, %rdx 443 jae 2f 444 addq %rbx, (%rdx) 445 addq $8, %rdx 446 jmp 1b 4472: 448 449/* 450 * Do the extraction, and jump to the new kernel.. 451 */ 452 pushq %rsi /* Save the real mode argument */ 453 movq %rsi, %rdi /* real mode address */ 454 leaq boot_heap(%rip), %rsi /* malloc area for uncompression */ 455 leaq input_data(%rip), %rdx /* input_data */ 456 movl $z_input_len, %ecx /* input_len */ 457 movq %rbp, %r8 /* output target address */ 458 movq $z_output_len, %r9 /* decompressed length, end of relocs */ 459 call extract_kernel /* returns kernel location in %rax */ 460 popq %rsi 461 462/* 463 * Jump to the decompressed kernel. 464 */ 465 jmp *%rax 466 467 .code32 468#ifdef CONFIG_X86_5LEVEL 469compatible_mode: 470 /* Setup data and stack segments */ 471 movl $__KERNEL_DS, %eax 472 movl %eax, %ds 473 movl %eax, %ss 474 475 /* Disable paging */ 476 movl %cr0, %eax 477 btrl $X86_CR0_PG_BIT, %eax 478 movl %eax, %cr0 479 480 /* Point CR3 to 5-level paging */ 481 leal lvl5_pgtable(%ebx), %eax 482 movl %eax, %cr3 483 484 /* Enable PAE and LA57 mode */ 485 movl %cr4, %eax 486 orl $(X86_CR4_PAE | X86_CR4_LA57), %eax 487 movl %eax, %cr4 488 489 /* Calculate address we are running at */ 490 call 1f 4911: popl %edi 492 subl $1b, %edi 493 494 /* Prepare stack for far return to Long Mode */ 495 pushl $__KERNEL_CS 496 leal lvl5(%edi), %eax 497 push %eax 498 499 /* Enable paging back */ 500 movl $(X86_CR0_PG | X86_CR0_PE), %eax 501 movl %eax, %cr0 502 503 lret 504#endif 505 506no_longmode: 507 /* This isn't an x86-64 CPU so hang */ 5081: 509 hlt 510 jmp 1b 511 512#include "../../kernel/verify_cpu.S" 513 514 .data 515gdt: 516 .word gdt_end - gdt 517 .long gdt 518 .word 0 519 .quad 0x00cf9a000000ffff /* __KERNEL32_CS */ 520 .quad 0x00af9a000000ffff /* __KERNEL_CS */ 521 .quad 0x00cf92000000ffff /* __KERNEL_DS */ 522 .quad 0x0080890000000000 /* TS descriptor */ 523 .quad 0x0000000000000000 /* TS continued */ 524gdt_end: 525 526#ifdef CONFIG_EFI_STUB 527efi_config: 528 .quad 0 529 530#ifdef CONFIG_EFI_MIXED 531 .global efi32_config 532efi32_config: 533 .fill 5,8,0 534 .quad efi64_thunk 535 .byte 0 536#endif 537 538 .global efi64_config 539efi64_config: 540 .fill 5,8,0 541 .quad efi_call 542 .byte 1 543#endif /* CONFIG_EFI_STUB */ 544 545/* 546 * Stack and heap for uncompression 547 */ 548 .bss 549 .balign 4 550boot_heap: 551 .fill BOOT_HEAP_SIZE, 1, 0 552boot_stack: 553 .fill BOOT_STACK_SIZE, 1, 0 554boot_stack_end: 555 556/* 557 * Space for page tables (not in .bss so not zeroed) 558 */ 559 .section ".pgtable","a",@nobits 560 .balign 4096 561pgtable: 562 .fill BOOT_PGT_SIZE, 1, 0 563#ifdef CONFIG_X86_5LEVEL 564lvl5_pgtable: 565 .fill PAGE_SIZE, 1, 0 566#endif 567