Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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#include "pgtable.h"
37
38/*
39 * Locally defined symbols should be marked hidden:
40 */
41 .hidden _bss
42 .hidden _ebss
43 .hidden _got
44 .hidden _egot
45
46 __HEAD
47 .code32
48ENTRY(startup_32)
49 /*
50 * 32bit entry is 0 and it is ABI so immutable!
51 * If we come here directly from a bootloader,
52 * kernel(text+data+bss+brk) ramdisk, zero_page, command line
53 * all need to be under the 4G limit.
54 */
55 cld
56 /*
57 * Test KEEP_SEGMENTS flag to see if the bootloader is asking
58 * us to not reload segments
59 */
60 testb $KEEP_SEGMENTS, BP_loadflags(%esi)
61 jnz 1f
62
63 cli
64 movl $(__BOOT_DS), %eax
65 movl %eax, %ds
66 movl %eax, %es
67 movl %eax, %ss
681:
69
70/*
71 * Calculate the delta between where we were compiled to run
72 * at and where we were actually loaded at. This can only be done
73 * with a short local call on x86. Nothing else will tell us what
74 * address we are running at. The reserved chunk of the real-mode
75 * data at 0x1e4 (defined as a scratch field) are used as the stack
76 * for this calculation. Only 4 bytes are needed.
77 */
78 leal (BP_scratch+4)(%esi), %esp
79 call 1f
801: popl %ebp
81 subl $1b, %ebp
82
83/* setup a stack and make sure cpu supports long mode. */
84 movl $boot_stack_end, %eax
85 addl %ebp, %eax
86 movl %eax, %esp
87
88 call verify_cpu
89 testl %eax, %eax
90 jnz no_longmode
91
92/*
93 * Compute the delta between where we were compiled to run at
94 * and where the code will actually run at.
95 *
96 * %ebp contains the address we are loaded at by the boot loader and %ebx
97 * contains the address where we should move the kernel image temporarily
98 * for safe in-place decompression.
99 */
100
101#ifdef CONFIG_RELOCATABLE
102 movl %ebp, %ebx
103 movl BP_kernel_alignment(%esi), %eax
104 decl %eax
105 addl %eax, %ebx
106 notl %eax
107 andl %eax, %ebx
108 cmpl $LOAD_PHYSICAL_ADDR, %ebx
109 jge 1f
110#endif
111 movl $LOAD_PHYSICAL_ADDR, %ebx
1121:
113
114 /* Target address to relocate to for decompression */
115 movl BP_init_size(%esi), %eax
116 subl $_end, %eax
117 addl %eax, %ebx
118
119/*
120 * Prepare for entering 64 bit mode
121 */
122
123 /* Load new GDT with the 64bit segments using 32bit descriptor */
124 addl %ebp, gdt+2(%ebp)
125 lgdt gdt(%ebp)
126
127 /* Enable PAE mode */
128 movl %cr4, %eax
129 orl $X86_CR4_PAE, %eax
130 movl %eax, %cr4
131
132 /*
133 * Build early 4G boot pagetable
134 */
135 /*
136 * If SEV is active then set the encryption mask in the page tables.
137 * This will insure that when the kernel is copied and decompressed
138 * it will be done so encrypted.
139 */
140 call get_sev_encryption_bit
141 xorl %edx, %edx
142 testl %eax, %eax
143 jz 1f
144 subl $32, %eax /* Encryption bit is always above bit 31 */
145 bts %eax, %edx /* Set encryption mask for page tables */
1461:
147
148 /* Initialize Page tables to 0 */
149 leal pgtable(%ebx), %edi
150 xorl %eax, %eax
151 movl $(BOOT_INIT_PGT_SIZE/4), %ecx
152 rep stosl
153
154 /* Build Level 4 */
155 leal pgtable + 0(%ebx), %edi
156 leal 0x1007 (%edi), %eax
157 movl %eax, 0(%edi)
158 addl %edx, 4(%edi)
159
160 /* Build Level 3 */
161 leal pgtable + 0x1000(%ebx), %edi
162 leal 0x1007(%edi), %eax
163 movl $4, %ecx
1641: movl %eax, 0x00(%edi)
165 addl %edx, 0x04(%edi)
166 addl $0x00001000, %eax
167 addl $8, %edi
168 decl %ecx
169 jnz 1b
170
171 /* Build Level 2 */
172 leal pgtable + 0x2000(%ebx), %edi
173 movl $0x00000183, %eax
174 movl $2048, %ecx
1751: movl %eax, 0(%edi)
176 addl %edx, 4(%edi)
177 addl $0x00200000, %eax
178 addl $8, %edi
179 decl %ecx
180 jnz 1b
181
182 /* Enable the boot page tables */
183 leal pgtable(%ebx), %eax
184 movl %eax, %cr3
185
186 /* Enable Long mode in EFER (Extended Feature Enable Register) */
187 movl $MSR_EFER, %ecx
188 rdmsr
189 btsl $_EFER_LME, %eax
190 wrmsr
191
192 /* After gdt is loaded */
193 xorl %eax, %eax
194 lldt %ax
195 movl $__BOOT_TSS, %eax
196 ltr %ax
197
198 /*
199 * Setup for the jump to 64bit mode
200 *
201 * When the jump is performend we will be in long mode but
202 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1
203 * (and in turn EFER.LMA = 1). To jump into 64bit mode we use
204 * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
205 * We place all of the values on our mini stack so lret can
206 * used to perform that far jump.
207 */
208 pushl $__KERNEL_CS
209 leal startup_64(%ebp), %eax
210#ifdef CONFIG_EFI_MIXED
211 movl efi32_config(%ebp), %ebx
212 cmp $0, %ebx
213 jz 1f
214 leal handover_entry(%ebp), %eax
2151:
216#endif
217 pushl %eax
218
219 /* Enter paged protected Mode, activating Long Mode */
220 movl $(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */
221 movl %eax, %cr0
222
223 /* Jump from 32bit compatibility mode into 64bit mode. */
224 lret
225ENDPROC(startup_32)
226
227#ifdef CONFIG_EFI_MIXED
228 .org 0x190
229ENTRY(efi32_stub_entry)
230 add $0x4, %esp /* Discard return address */
231 popl %ecx
232 popl %edx
233 popl %esi
234
235 leal (BP_scratch+4)(%esi), %esp
236 call 1f
2371: pop %ebp
238 subl $1b, %ebp
239
240 movl %ecx, efi32_config(%ebp)
241 movl %edx, efi32_config+8(%ebp)
242 sgdtl efi32_boot_gdt(%ebp)
243
244 leal efi32_config(%ebp), %eax
245 movl %eax, efi_config(%ebp)
246
247 jmp startup_32
248ENDPROC(efi32_stub_entry)
249#endif
250
251 .code64
252 .org 0x200
253ENTRY(startup_64)
254 /*
255 * 64bit entry is 0x200 and it is ABI so immutable!
256 * We come here either from startup_32 or directly from a
257 * 64bit bootloader.
258 * If we come here from a bootloader, kernel(text+data+bss+brk),
259 * ramdisk, zero_page, command line could be above 4G.
260 * We depend on an identity mapped page table being provided
261 * that maps our entire kernel(text+data+bss+brk), zero page
262 * and command line.
263 */
264
265 /* Setup data segments. */
266 xorl %eax, %eax
267 movl %eax, %ds
268 movl %eax, %es
269 movl %eax, %ss
270 movl %eax, %fs
271 movl %eax, %gs
272
273 /*
274 * Compute the decompressed kernel start address. It is where
275 * we were loaded at aligned to a 2M boundary. %rbp contains the
276 * decompressed kernel start address.
277 *
278 * If it is a relocatable kernel then decompress and run the kernel
279 * from load address aligned to 2MB addr, otherwise decompress and
280 * run the kernel from LOAD_PHYSICAL_ADDR
281 *
282 * We cannot rely on the calculation done in 32-bit mode, since we
283 * may have been invoked via the 64-bit entry point.
284 */
285
286 /* Start with the delta to where the kernel will run at. */
287#ifdef CONFIG_RELOCATABLE
288 leaq startup_32(%rip) /* - $startup_32 */, %rbp
289 movl BP_kernel_alignment(%rsi), %eax
290 decl %eax
291 addq %rax, %rbp
292 notq %rax
293 andq %rax, %rbp
294 cmpq $LOAD_PHYSICAL_ADDR, %rbp
295 jge 1f
296#endif
297 movq $LOAD_PHYSICAL_ADDR, %rbp
2981:
299
300 /* Target address to relocate to for decompression */
301 movl BP_init_size(%rsi), %ebx
302 subl $_end, %ebx
303 addq %rbp, %rbx
304
305 /* Set up the stack */
306 leaq boot_stack_end(%rbx), %rsp
307
308 /*
309 * paging_prepare() and cleanup_trampoline() below can have GOT
310 * references. Adjust the table with address we are running at.
311 *
312 * Zero RAX for adjust_got: the GOT was not adjusted before;
313 * there's no adjustment to undo.
314 */
315 xorq %rax, %rax
316
317 /*
318 * Calculate the address the binary is loaded at and use it as
319 * a GOT adjustment.
320 */
321 call 1f
3221: popq %rdi
323 subq $1b, %rdi
324
325 call adjust_got
326
327 /*
328 * At this point we are in long mode with 4-level paging enabled,
329 * but we might want to enable 5-level paging or vice versa.
330 *
331 * The problem is that we cannot do it directly. Setting or clearing
332 * CR4.LA57 in long mode would trigger #GP. So we need to switch off
333 * long mode and paging first.
334 *
335 * We also need a trampoline in lower memory to switch over from
336 * 4- to 5-level paging for cases when the bootloader puts the kernel
337 * above 4G, but didn't enable 5-level paging for us.
338 *
339 * The same trampoline can be used to switch from 5- to 4-level paging
340 * mode, like when starting 4-level paging kernel via kexec() when
341 * original kernel worked in 5-level paging mode.
342 *
343 * For the trampoline, we need the top page table to reside in lower
344 * memory as we don't have a way to load 64-bit values into CR3 in
345 * 32-bit mode.
346 *
347 * We go though the trampoline even if we don't have to: if we're
348 * already in a desired paging mode. This way the trampoline code gets
349 * tested on every boot.
350 */
351
352 /* Make sure we have GDT with 32-bit code segment */
353 leaq gdt(%rip), %rax
354 movq %rax, gdt64+2(%rip)
355 lgdt gdt64(%rip)
356
357 /*
358 * paging_prepare() sets up the trampoline and checks if we need to
359 * enable 5-level paging.
360 *
361 * Address of the trampoline is returned in RAX.
362 * Non zero RDX on return means we need to enable 5-level paging.
363 *
364 * RSI holds real mode data and needs to be preserved across
365 * this function call.
366 */
367 pushq %rsi
368 movq %rsi, %rdi /* real mode address */
369 call paging_prepare
370 popq %rsi
371
372 /* Save the trampoline address in RCX */
373 movq %rax, %rcx
374
375 /*
376 * Load the address of trampoline_return() into RDI.
377 * It will be used by the trampoline to return to the main code.
378 */
379 leaq trampoline_return(%rip), %rdi
380
381 /* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */
382 pushq $__KERNEL32_CS
383 leaq TRAMPOLINE_32BIT_CODE_OFFSET(%rax), %rax
384 pushq %rax
385 lretq
386trampoline_return:
387 /* Restore the stack, the 32-bit trampoline uses its own stack */
388 leaq boot_stack_end(%rbx), %rsp
389
390 /*
391 * cleanup_trampoline() would restore trampoline memory.
392 *
393 * RDI is address of the page table to use instead of page table
394 * in trampoline memory (if required).
395 *
396 * RSI holds real mode data and needs to be preserved across
397 * this function call.
398 */
399 pushq %rsi
400 leaq top_pgtable(%rbx), %rdi
401 call cleanup_trampoline
402 popq %rsi
403
404 /* Zero EFLAGS */
405 pushq $0
406 popfq
407
408 /*
409 * Previously we've adjusted the GOT with address the binary was
410 * loaded at. Now we need to re-adjust for relocation address.
411 *
412 * Calculate the address the binary is loaded at, so that we can
413 * undo the previous GOT adjustment.
414 */
415 call 1f
4161: popq %rax
417 subq $1b, %rax
418
419 /* The new adjustment is the relocation address */
420 movq %rbx, %rdi
421 call adjust_got
422
423/*
424 * Copy the compressed kernel to the end of our buffer
425 * where decompression in place becomes safe.
426 */
427 pushq %rsi
428 leaq (_bss-8)(%rip), %rsi
429 leaq (_bss-8)(%rbx), %rdi
430 movq $_bss /* - $startup_32 */, %rcx
431 shrq $3, %rcx
432 std
433 rep movsq
434 cld
435 popq %rsi
436
437/*
438 * Jump to the relocated address.
439 */
440 leaq relocated(%rbx), %rax
441 jmp *%rax
442
443#ifdef CONFIG_EFI_STUB
444
445/* The entry point for the PE/COFF executable is efi_pe_entry. */
446ENTRY(efi_pe_entry)
447 movq %rcx, efi64_config(%rip) /* Handle */
448 movq %rdx, efi64_config+8(%rip) /* EFI System table pointer */
449
450 leaq efi64_config(%rip), %rax
451 movq %rax, efi_config(%rip)
452
453 call 1f
4541: popq %rbp
455 subq $1b, %rbp
456
457 /*
458 * Relocate efi_config->call().
459 */
460 addq %rbp, efi64_config+40(%rip)
461
462 movq %rax, %rdi
463 call make_boot_params
464 cmpq $0,%rax
465 je fail
466 mov %rax, %rsi
467 leaq startup_32(%rip), %rax
468 movl %eax, BP_code32_start(%rsi)
469 jmp 2f /* Skip the relocation */
470
471handover_entry:
472 call 1f
4731: popq %rbp
474 subq $1b, %rbp
475
476 /*
477 * Relocate efi_config->call().
478 */
479 movq efi_config(%rip), %rax
480 addq %rbp, 40(%rax)
4812:
482 movq efi_config(%rip), %rdi
483 call efi_main
484 movq %rax,%rsi
485 cmpq $0,%rax
486 jne 2f
487fail:
488 /* EFI init failed, so hang. */
489 hlt
490 jmp fail
4912:
492 movl BP_code32_start(%esi), %eax
493 leaq startup_64(%rax), %rax
494 jmp *%rax
495ENDPROC(efi_pe_entry)
496
497 .org 0x390
498ENTRY(efi64_stub_entry)
499 movq %rdi, efi64_config(%rip) /* Handle */
500 movq %rsi, efi64_config+8(%rip) /* EFI System table pointer */
501
502 leaq efi64_config(%rip), %rax
503 movq %rax, efi_config(%rip)
504
505 movq %rdx, %rsi
506 jmp handover_entry
507ENDPROC(efi64_stub_entry)
508#endif
509
510 .text
511relocated:
512
513/*
514 * Clear BSS (stack is currently empty)
515 */
516 xorl %eax, %eax
517 leaq _bss(%rip), %rdi
518 leaq _ebss(%rip), %rcx
519 subq %rdi, %rcx
520 shrq $3, %rcx
521 rep stosq
522
523/*
524 * Do the extraction, and jump to the new kernel..
525 */
526 pushq %rsi /* Save the real mode argument */
527 movq %rsi, %rdi /* real mode address */
528 leaq boot_heap(%rip), %rsi /* malloc area for uncompression */
529 leaq input_data(%rip), %rdx /* input_data */
530 movl $z_input_len, %ecx /* input_len */
531 movq %rbp, %r8 /* output target address */
532 movq $z_output_len, %r9 /* decompressed length, end of relocs */
533 call extract_kernel /* returns kernel location in %rax */
534 popq %rsi
535
536/*
537 * Jump to the decompressed kernel.
538 */
539 jmp *%rax
540
541/*
542 * Adjust the global offset table
543 *
544 * RAX is the previous adjustment of the table to undo (use 0 if it's the
545 * first time we touch GOT).
546 * RDI is the new adjustment to apply.
547 */
548adjust_got:
549 /* Walk through the GOT adding the address to the entries */
550 leaq _got(%rip), %rdx
551 leaq _egot(%rip), %rcx
5521:
553 cmpq %rcx, %rdx
554 jae 2f
555 subq %rax, (%rdx) /* Undo previous adjustment */
556 addq %rdi, (%rdx) /* Apply the new adjustment */
557 addq $8, %rdx
558 jmp 1b
5592:
560 ret
561
562 .code32
563/*
564 * This is the 32-bit trampoline that will be copied over to low memory.
565 *
566 * RDI contains the return address (might be above 4G).
567 * ECX contains the base address of the trampoline memory.
568 * Non zero RDX on return means we need to enable 5-level paging.
569 */
570ENTRY(trampoline_32bit_src)
571 /* Set up data and stack segments */
572 movl $__KERNEL_DS, %eax
573 movl %eax, %ds
574 movl %eax, %ss
575
576 /* Set up new stack */
577 leal TRAMPOLINE_32BIT_STACK_END(%ecx), %esp
578
579 /* Disable paging */
580 movl %cr0, %eax
581 btrl $X86_CR0_PG_BIT, %eax
582 movl %eax, %cr0
583
584 /* Check what paging mode we want to be in after the trampoline */
585 cmpl $0, %edx
586 jz 1f
587
588 /* We want 5-level paging: don't touch CR3 if it already points to 5-level page tables */
589 movl %cr4, %eax
590 testl $X86_CR4_LA57, %eax
591 jnz 3f
592 jmp 2f
5931:
594 /* We want 4-level paging: don't touch CR3 if it already points to 4-level page tables */
595 movl %cr4, %eax
596 testl $X86_CR4_LA57, %eax
597 jz 3f
5982:
599 /* Point CR3 to the trampoline's new top level page table */
600 leal TRAMPOLINE_32BIT_PGTABLE_OFFSET(%ecx), %eax
601 movl %eax, %cr3
6023:
603 /* Enable PAE and LA57 (if required) paging modes */
604 movl $X86_CR4_PAE, %eax
605 cmpl $0, %edx
606 jz 1f
607 orl $X86_CR4_LA57, %eax
6081:
609 movl %eax, %cr4
610
611 /* Calculate address of paging_enabled() once we are executing in the trampoline */
612 leal paging_enabled - trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_OFFSET(%ecx), %eax
613
614 /* Prepare the stack for far return to Long Mode */
615 pushl $__KERNEL_CS
616 pushl %eax
617
618 /* Enable paging again */
619 movl $(X86_CR0_PG | X86_CR0_PE), %eax
620 movl %eax, %cr0
621
622 lret
623
624 .code64
625paging_enabled:
626 /* Return from the trampoline */
627 jmp *%rdi
628
629 /*
630 * The trampoline code has a size limit.
631 * Make sure we fail to compile if the trampoline code grows
632 * beyond TRAMPOLINE_32BIT_CODE_SIZE bytes.
633 */
634 .org trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_SIZE
635
636 .code32
637no_longmode:
638 /* This isn't an x86-64 CPU, so hang intentionally, we cannot continue */
6391:
640 hlt
641 jmp 1b
642
643#include "../../kernel/verify_cpu.S"
644
645 .data
646gdt64:
647 .word gdt_end - gdt
648 .long 0
649 .word 0
650 .quad 0
651gdt:
652 .word gdt_end - gdt
653 .long gdt
654 .word 0
655 .quad 0x00cf9a000000ffff /* __KERNEL32_CS */
656 .quad 0x00af9a000000ffff /* __KERNEL_CS */
657 .quad 0x00cf92000000ffff /* __KERNEL_DS */
658 .quad 0x0080890000000000 /* TS descriptor */
659 .quad 0x0000000000000000 /* TS continued */
660gdt_end:
661
662#ifdef CONFIG_EFI_STUB
663efi_config:
664 .quad 0
665
666#ifdef CONFIG_EFI_MIXED
667 .global efi32_config
668efi32_config:
669 .fill 5,8,0
670 .quad efi64_thunk
671 .byte 0
672#endif
673
674 .global efi64_config
675efi64_config:
676 .fill 5,8,0
677 .quad efi_call
678 .byte 1
679#endif /* CONFIG_EFI_STUB */
680
681/*
682 * Stack and heap for uncompression
683 */
684 .bss
685 .balign 4
686boot_heap:
687 .fill BOOT_HEAP_SIZE, 1, 0
688boot_stack:
689 .fill BOOT_STACK_SIZE, 1, 0
690boot_stack_end:
691
692/*
693 * Space for page tables (not in .bss so not zeroed)
694 */
695 .section ".pgtable","a",@nobits
696 .balign 4096
697pgtable:
698 .fill BOOT_PGT_SIZE, 1, 0
699
700/*
701 * The page table is going to be used instead of page table in the trampoline
702 * memory.
703 */
704top_pgtable:
705 .fill PAGE_SIZE, 1, 0