1 /* SPDX-License-Identifier: GPL-2.0 */
5 * Copyright (C) 1991, 1992, 1993 Linus Torvalds
9 * head.S contains the 32-bit startup code.
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]
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
23 * High loaded stuff by Hans Lermen & Werner Almesberger, Feb. 1996
28 #include <linux/init.h>
29 #include <linux/linkage.h>
30 #include <asm/segment.h>
33 #include <asm/processor-flags.h>
34 #include <asm/asm-offsets.h>
35 #include <asm/bootparam.h>
39 * Locally defined symbols should be marked hidden:
48 SYM_FUNC_START(startup_32)
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.
59 * Calculate the delta between where we were compiled to run
60 * at and where we were actually loaded at. This can only be done
61 * with a short local call on x86. Nothing else will tell us what
62 * address we are running at. The reserved chunk of the real-mode
63 * data at 0x1e4 (defined as a scratch field) are used as the stack
64 * for this calculation. Only 4 bytes are needed.
66 leal (BP_scratch+4)(%esi), %esp
71 /* Load new GDT with the 64bit segments using 32bit descriptor */
76 /* Load segment registers with our descriptors */
84 /* setup a stack and make sure cpu supports long mode. */
85 leal boot_stack_end(%ebp), %esp
92 * Compute the delta between where we were compiled to run at
93 * and where the code will actually run at.
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.
100 #ifdef CONFIG_RELOCATABLE
102 movl BP_kernel_alignment(%esi), %eax
107 cmpl $LOAD_PHYSICAL_ADDR, %ebx
110 movl $LOAD_PHYSICAL_ADDR, %ebx
113 /* Target address to relocate to for decompression */
114 movl BP_init_size(%esi), %eax
119 * Prepare for entering 64 bit mode
122 /* Enable PAE mode */
124 orl $X86_CR4_PAE, %eax
128 * Build early 4G boot pagetable
131 * If SEV is active then set the encryption mask in the page tables.
132 * This will insure that when the kernel is copied and decompressed
133 * it will be done so encrypted.
135 call get_sev_encryption_bit
139 subl $32, %eax /* Encryption bit is always above bit 31 */
140 bts %eax, %edx /* Set encryption mask for page tables */
143 /* Initialize Page tables to 0 */
144 leal pgtable(%ebx), %edi
146 movl $(BOOT_INIT_PGT_SIZE/4), %ecx
150 leal pgtable + 0(%ebx), %edi
151 leal 0x1007 (%edi), %eax
156 leal pgtable + 0x1000(%ebx), %edi
157 leal 0x1007(%edi), %eax
159 1: movl %eax, 0x00(%edi)
160 addl %edx, 0x04(%edi)
161 addl $0x00001000, %eax
167 leal pgtable + 0x2000(%ebx), %edi
168 movl $0x00000183, %eax
170 1: movl %eax, 0(%edi)
172 addl $0x00200000, %eax
177 /* Enable the boot page tables */
178 leal pgtable(%ebx), %eax
181 /* Enable Long mode in EFER (Extended Feature Enable Register) */
184 btsl $_EFER_LME, %eax
187 /* After gdt is loaded */
190 movl $__BOOT_TSS, %eax
194 * Setup for the jump to 64bit mode
196 * When the jump is performend we will be in long mode but
197 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1
198 * (and in turn EFER.LMA = 1). To jump into 64bit mode we use
199 * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
200 * We place all of the values on our mini stack so lret can
201 * used to perform that far jump.
204 leal startup_64(%ebp), %eax
205 #ifdef CONFIG_EFI_MIXED
206 movl efi32_boot_args(%ebp), %edi
209 leal efi64_stub_entry(%ebp), %eax
210 movl efi32_boot_args+4(%ebp), %esi
211 movl efi32_boot_args+8(%ebp), %edx // saved bootparams pointer
214 leal efi_pe_entry(%ebp), %eax
215 movl %edi, %ecx // MS calling convention
221 /* Enter paged protected Mode, activating Long Mode */
222 movl $(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */
225 /* Jump from 32bit compatibility mode into 64bit mode. */
227 SYM_FUNC_END(startup_32)
229 #ifdef CONFIG_EFI_MIXED
231 SYM_FUNC_START(efi32_stub_entry)
232 add $0x4, %esp /* Discard return address */
241 movl %esi, efi32_boot_args+8(%ebp)
242 SYM_INNER_LABEL(efi32_pe_stub_entry, SYM_L_LOCAL)
243 movl %ecx, efi32_boot_args(%ebp)
244 movl %edx, efi32_boot_args+4(%ebp)
245 movb $0, efi_is64(%ebp)
247 /* Save firmware GDTR and code/data selectors */
248 sgdtl efi32_boot_gdt(%ebp)
249 movw %cs, efi32_boot_cs(%ebp)
250 movw %ds, efi32_boot_ds(%ebp)
254 btrl $X86_CR0_PG_BIT, %eax
258 SYM_FUNC_END(efi32_stub_entry)
263 SYM_CODE_START(startup_64)
265 * 64bit entry is 0x200 and it is ABI so immutable!
266 * We come here either from startup_32 or directly from a
268 * If we come here from a bootloader, kernel(text+data+bss+brk),
269 * ramdisk, zero_page, command line could be above 4G.
270 * We depend on an identity mapped page table being provided
271 * that maps our entire kernel(text+data+bss+brk), zero page
278 /* Setup data segments. */
287 * Compute the decompressed kernel start address. It is where
288 * we were loaded at aligned to a 2M boundary. %rbp contains the
289 * decompressed kernel start address.
291 * If it is a relocatable kernel then decompress and run the kernel
292 * from load address aligned to 2MB addr, otherwise decompress and
293 * run the kernel from LOAD_PHYSICAL_ADDR
295 * We cannot rely on the calculation done in 32-bit mode, since we
296 * may have been invoked via the 64-bit entry point.
299 /* Start with the delta to where the kernel will run at. */
300 #ifdef CONFIG_RELOCATABLE
301 leaq startup_32(%rip) /* - $startup_32 */, %rbp
302 movl BP_kernel_alignment(%rsi), %eax
307 cmpq $LOAD_PHYSICAL_ADDR, %rbp
310 movq $LOAD_PHYSICAL_ADDR, %rbp
313 /* Target address to relocate to for decompression */
314 movl BP_init_size(%rsi), %ebx
318 /* Set up the stack */
319 leaq boot_stack_end(%rbx), %rsp
322 * paging_prepare() and cleanup_trampoline() below can have GOT
323 * references. Adjust the table with address we are running at.
325 * Zero RAX for adjust_got: the GOT was not adjusted before;
326 * there's no adjustment to undo.
331 * Calculate the address the binary is loaded at and use it as
341 * At this point we are in long mode with 4-level paging enabled,
342 * but we might want to enable 5-level paging or vice versa.
344 * The problem is that we cannot do it directly. Setting or clearing
345 * CR4.LA57 in long mode would trigger #GP. So we need to switch off
346 * long mode and paging first.
348 * We also need a trampoline in lower memory to switch over from
349 * 4- to 5-level paging for cases when the bootloader puts the kernel
350 * above 4G, but didn't enable 5-level paging for us.
352 * The same trampoline can be used to switch from 5- to 4-level paging
353 * mode, like when starting 4-level paging kernel via kexec() when
354 * original kernel worked in 5-level paging mode.
356 * For the trampoline, we need the top page table to reside in lower
357 * memory as we don't have a way to load 64-bit values into CR3 in
360 * We go though the trampoline even if we don't have to: if we're
361 * already in a desired paging mode. This way the trampoline code gets
362 * tested on every boot.
365 /* Make sure we have GDT with 32-bit code segment */
366 leaq gdt64(%rip), %rax
371 * paging_prepare() sets up the trampoline and checks if we need to
372 * enable 5-level paging.
374 * paging_prepare() returns a two-quadword structure which lands
376 * - Address of the trampoline is returned in RAX.
377 * - Non zero RDX means trampoline needs to enable 5-level
380 * RSI holds real mode data and needs to be preserved across
381 * this function call.
384 movq %rsi, %rdi /* real mode address */
388 /* Save the trampoline address in RCX */
392 * Load the address of trampoline_return() into RDI.
393 * It will be used by the trampoline to return to the main code.
395 leaq trampoline_return(%rip), %rdi
397 /* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */
399 leaq TRAMPOLINE_32BIT_CODE_OFFSET(%rax), %rax
403 /* Restore the stack, the 32-bit trampoline uses its own stack */
404 leaq boot_stack_end(%rbx), %rsp
407 * cleanup_trampoline() would restore trampoline memory.
409 * RDI is address of the page table to use instead of page table
410 * in trampoline memory (if required).
412 * RSI holds real mode data and needs to be preserved across
413 * this function call.
416 leaq top_pgtable(%rbx), %rdi
417 call cleanup_trampoline
425 * Previously we've adjusted the GOT with address the binary was
426 * loaded at. Now we need to re-adjust for relocation address.
428 * Calculate the address the binary is loaded at, so that we can
429 * undo the previous GOT adjustment.
435 /* The new adjustment is the relocation address */
440 * Copy the compressed kernel to the end of our buffer
441 * where decompression in place becomes safe.
444 leaq (_bss-8)(%rip), %rsi
445 leaq (_bss-8)(%rbx), %rdi
446 movq $_bss /* - $startup_32 */, %rcx
454 * The GDT may get overwritten either during the copy we just did or
455 * during extract_kernel below. To avoid any issues, repoint the GDTR
456 * to the new copy of the GDT.
458 leaq gdt64(%rbx), %rax
464 * Jump to the relocated address.
466 leaq .Lrelocated(%rbx), %rax
468 SYM_CODE_END(startup_64)
470 #ifdef CONFIG_EFI_STUB
472 SYM_FUNC_START(efi64_stub_entry)
473 SYM_FUNC_START_ALIAS(efi_stub_entry)
474 and $~0xf, %rsp /* realign the stack */
477 movl BP_code32_start(%esi), %eax
478 leaq startup_64(%rax), %rax
480 SYM_FUNC_END(efi64_stub_entry)
481 SYM_FUNC_END_ALIAS(efi_stub_entry)
485 SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated)
488 * Clear BSS (stack is currently empty)
491 leaq _bss(%rip), %rdi
492 leaq _ebss(%rip), %rcx
498 * Do the extraction, and jump to the new kernel..
500 pushq %rsi /* Save the real mode argument */
501 movq %rsi, %rdi /* real mode address */
502 leaq boot_heap(%rip), %rsi /* malloc area for uncompression */
503 leaq input_data(%rip), %rdx /* input_data */
504 movl $z_input_len, %ecx /* input_len */
505 movq %rbp, %r8 /* output target address */
506 movq $z_output_len, %r9 /* decompressed length, end of relocs */
507 call extract_kernel /* returns kernel location in %rax */
511 * Jump to the decompressed kernel.
514 SYM_FUNC_END(.Lrelocated)
517 * Adjust the global offset table
519 * RAX is the previous adjustment of the table to undo (use 0 if it's the
520 * first time we touch GOT).
521 * RDI is the new adjustment to apply.
524 /* Walk through the GOT adding the address to the entries */
525 leaq _got(%rip), %rdx
526 leaq _egot(%rip), %rcx
530 subq %rax, (%rdx) /* Undo previous adjustment */
531 addq %rdi, (%rdx) /* Apply the new adjustment */
539 * This is the 32-bit trampoline that will be copied over to low memory.
541 * RDI contains the return address (might be above 4G).
542 * ECX contains the base address of the trampoline memory.
543 * Non zero RDX means trampoline needs to enable 5-level paging.
545 SYM_CODE_START(trampoline_32bit_src)
546 /* Set up data and stack segments */
547 movl $__KERNEL_DS, %eax
551 /* Set up new stack */
552 leal TRAMPOLINE_32BIT_STACK_END(%ecx), %esp
556 btrl $X86_CR0_PG_BIT, %eax
559 /* Check what paging mode we want to be in after the trampoline */
563 /* We want 5-level paging: don't touch CR3 if it already points to 5-level page tables */
565 testl $X86_CR4_LA57, %eax
569 /* We want 4-level paging: don't touch CR3 if it already points to 4-level page tables */
571 testl $X86_CR4_LA57, %eax
574 /* Point CR3 to the trampoline's new top level page table */
575 leal TRAMPOLINE_32BIT_PGTABLE_OFFSET(%ecx), %eax
578 /* Set EFER.LME=1 as a precaution in case hypervsior pulls the rug */
583 btsl $_EFER_LME, %eax
588 /* Enable PAE and LA57 (if required) paging modes */
589 movl $X86_CR4_PAE, %eax
592 orl $X86_CR4_LA57, %eax
596 /* Calculate address of paging_enabled() once we are executing in the trampoline */
597 leal .Lpaging_enabled - trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_OFFSET(%ecx), %eax
599 /* Prepare the stack for far return to Long Mode */
603 /* Enable paging again */
604 movl $(X86_CR0_PG | X86_CR0_PE), %eax
608 SYM_CODE_END(trampoline_32bit_src)
611 SYM_FUNC_START_LOCAL_NOALIGN(.Lpaging_enabled)
612 /* Return from the trampoline */
614 SYM_FUNC_END(.Lpaging_enabled)
617 * The trampoline code has a size limit.
618 * Make sure we fail to compile if the trampoline code grows
619 * beyond TRAMPOLINE_32BIT_CODE_SIZE bytes.
621 .org trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_SIZE
624 SYM_FUNC_START_LOCAL_NOALIGN(.Lno_longmode)
625 /* This isn't an x86-64 CPU, so hang intentionally, we cannot continue */
629 SYM_FUNC_END(.Lno_longmode)
631 #include "../../kernel/verify_cpu.S"
634 SYM_DATA_START_LOCAL(gdt64)
635 .word gdt_end - gdt - 1
639 SYM_DATA_START_LOCAL(gdt)
640 .word gdt_end - gdt - 1
643 .quad 0x00cf9a000000ffff /* __KERNEL32_CS */
644 .quad 0x00af9a000000ffff /* __KERNEL_CS */
645 .quad 0x00cf92000000ffff /* __KERNEL_DS */
646 .quad 0x0080890000000000 /* TS descriptor */
647 .quad 0x0000000000000000 /* TS continued */
648 SYM_DATA_END_LABEL(gdt, SYM_L_LOCAL, gdt_end)
650 #ifdef CONFIG_EFI_MIXED
651 SYM_DATA_LOCAL(efi32_boot_args, .long 0, 0, 0)
652 SYM_DATA(efi_is64, .byte 1)
654 #define ST32_boottime 60 // offsetof(efi_system_table_32_t, boottime)
655 #define BS32_handle_protocol 88 // offsetof(efi_boot_services_32_t, handle_protocol)
656 #define LI32_image_base 32 // offsetof(efi_loaded_image_32_t, image_base)
660 SYM_FUNC_START(efi32_pe_entry)
663 call verify_cpu // check for long mode support
665 movl $0x80000003, %eax // EFI_UNSUPPORTED
672 /* Get the loaded image protocol pointer from the image handle */
673 subl $12, %esp // space for the loaded image pointer
674 pushl %esp // pass its address
676 pushl %eax // pass the GUID address
677 pushl 28(%esp) // pass the image handle
679 movl 36(%esp), %eax // sys_table
680 movl ST32_boottime(%eax), %eax // sys_table->boottime
681 call *BS32_handle_protocol(%eax) // sys_table->boottime->handle_protocol
685 movl 32(%esp), %ecx // image_handle
686 movl 36(%esp), %edx // sys_table
687 movl 12(%esp), %esi // loaded_image
688 movl LI32_image_base(%esi), %esi // loaded_image->image_base
689 jmp efi32_pe_stub_entry
694 SYM_FUNC_END(efi32_pe_entry)
697 /* EFI loaded image protocol GUID */
700 .byte 0x8E, 0x3F, 0x00, 0xA0, 0xC9, 0x69, 0x72, 0x3B
704 * Stack and heap for uncompression
708 SYM_DATA_LOCAL(boot_heap, .fill BOOT_HEAP_SIZE, 1, 0)
710 SYM_DATA_START_LOCAL(boot_stack)
711 .fill BOOT_STACK_SIZE, 1, 0
712 SYM_DATA_END_LABEL(boot_stack, SYM_L_LOCAL, boot_stack_end)
715 * Space for page tables (not in .bss so not zeroed)
717 .section ".pgtable","a",@nobits
719 SYM_DATA_LOCAL(pgtable, .fill BOOT_PGT_SIZE, 1, 0)
722 * The page table is going to be used instead of page table in the trampoline
725 SYM_DATA_LOCAL(top_pgtable, .fill PAGE_SIZE, 1, 0)