Loading...
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
35 __HEAD
36 .code32
37ENTRY(startup_32)
38 /*
39 * 32bit entry is 0 and it is ABI so immutable!
40 * If we come here directly from a bootloader,
41 * kernel(text+data+bss+brk) ramdisk, zero_page, command line
42 * all need to be under the 4G limit.
43 */
44 cld
45 /*
46 * Test KEEP_SEGMENTS flag to see if the bootloader is asking
47 * us to not reload segments
48 */
49 testb $(1<<6), BP_loadflags(%esi)
50 jnz 1f
51
52 cli
53 movl $(__BOOT_DS), %eax
54 movl %eax, %ds
55 movl %eax, %es
56 movl %eax, %ss
571:
58
59/*
60 * Calculate the delta between where we were compiled to run
61 * at and where we were actually loaded at. This can only be done
62 * with a short local call on x86. Nothing else will tell us what
63 * address we are running at. The reserved chunk of the real-mode
64 * data at 0x1e4 (defined as a scratch field) are used as the stack
65 * for this calculation. Only 4 bytes are needed.
66 */
67 leal (BP_scratch+4)(%esi), %esp
68 call 1f
691: popl %ebp
70 subl $1b, %ebp
71
72/* setup a stack and make sure cpu supports long mode. */
73 movl $boot_stack_end, %eax
74 addl %ebp, %eax
75 movl %eax, %esp
76
77 call verify_cpu
78 testl %eax, %eax
79 jnz no_longmode
80
81/*
82 * Compute the delta between where we were compiled to run at
83 * and where the code will actually run at.
84 *
85 * %ebp contains the address we are loaded at by the boot loader and %ebx
86 * contains the address where we should move the kernel image temporarily
87 * for safe in-place decompression.
88 */
89
90#ifdef CONFIG_RELOCATABLE
91 movl %ebp, %ebx
92 movl BP_kernel_alignment(%esi), %eax
93 decl %eax
94 addl %eax, %ebx
95 notl %eax
96 andl %eax, %ebx
97 cmpl $LOAD_PHYSICAL_ADDR, %ebx
98 jge 1f
99#endif
100 movl $LOAD_PHYSICAL_ADDR, %ebx
1011:
102
103 /* Target address to relocate to for decompression */
104 addl $z_extract_offset, %ebx
105
106/*
107 * Prepare for entering 64 bit mode
108 */
109
110 /* Load new GDT with the 64bit segments using 32bit descriptor */
111 leal gdt(%ebp), %eax
112 movl %eax, gdt+2(%ebp)
113 lgdt gdt(%ebp)
114
115 /* Enable PAE mode */
116 movl %cr4, %eax
117 orl $X86_CR4_PAE, %eax
118 movl %eax, %cr4
119
120 /*
121 * Build early 4G boot pagetable
122 */
123 /* Initialize Page tables to 0 */
124 leal pgtable(%ebx), %edi
125 xorl %eax, %eax
126 movl $((4096*6)/4), %ecx
127 rep stosl
128
129 /* Build Level 4 */
130 leal pgtable + 0(%ebx), %edi
131 leal 0x1007 (%edi), %eax
132 movl %eax, 0(%edi)
133
134 /* Build Level 3 */
135 leal pgtable + 0x1000(%ebx), %edi
136 leal 0x1007(%edi), %eax
137 movl $4, %ecx
1381: movl %eax, 0x00(%edi)
139 addl $0x00001000, %eax
140 addl $8, %edi
141 decl %ecx
142 jnz 1b
143
144 /* Build Level 2 */
145 leal pgtable + 0x2000(%ebx), %edi
146 movl $0x00000183, %eax
147 movl $2048, %ecx
1481: movl %eax, 0(%edi)
149 addl $0x00200000, %eax
150 addl $8, %edi
151 decl %ecx
152 jnz 1b
153
154 /* Enable the boot page tables */
155 leal pgtable(%ebx), %eax
156 movl %eax, %cr3
157
158 /* Enable Long mode in EFER (Extended Feature Enable Register) */
159 movl $MSR_EFER, %ecx
160 rdmsr
161 btsl $_EFER_LME, %eax
162 wrmsr
163
164 /* After gdt is loaded */
165 xorl %eax, %eax
166 lldt %ax
167 movl $0x20, %eax
168 ltr %ax
169
170 /*
171 * Setup for the jump to 64bit mode
172 *
173 * When the jump is performend we will be in long mode but
174 * in 32bit compatibility mode with EFER.LME = 1, CS.L = 0, CS.D = 1
175 * (and in turn EFER.LMA = 1). To jump into 64bit mode we use
176 * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
177 * We place all of the values on our mini stack so lret can
178 * used to perform that far jump.
179 */
180 pushl $__KERNEL_CS
181 leal startup_64(%ebp), %eax
182#ifdef CONFIG_EFI_MIXED
183 movl efi32_config(%ebp), %ebx
184 cmp $0, %ebx
185 jz 1f
186 leal handover_entry(%ebp), %eax
1871:
188#endif
189 pushl %eax
190
191 /* Enter paged protected Mode, activating Long Mode */
192 movl $(X86_CR0_PG | X86_CR0_PE), %eax /* Enable Paging and Protected mode */
193 movl %eax, %cr0
194
195 /* Jump from 32bit compatibility mode into 64bit mode. */
196 lret
197ENDPROC(startup_32)
198
199#ifdef CONFIG_EFI_MIXED
200 .org 0x190
201ENTRY(efi32_stub_entry)
202 add $0x4, %esp /* Discard return address */
203 popl %ecx
204 popl %edx
205 popl %esi
206
207 leal (BP_scratch+4)(%esi), %esp
208 call 1f
2091: pop %ebp
210 subl $1b, %ebp
211
212 movl %ecx, efi32_config(%ebp)
213 movl %edx, efi32_config+8(%ebp)
214 sgdtl efi32_boot_gdt(%ebp)
215
216 leal efi32_config(%ebp), %eax
217 movl %eax, efi_config(%ebp)
218
219 jmp startup_32
220ENDPROC(efi32_stub_entry)
221#endif
222
223 .code64
224 .org 0x200
225ENTRY(startup_64)
226 /*
227 * 64bit entry is 0x200 and it is ABI so immutable!
228 * We come here either from startup_32 or directly from a
229 * 64bit bootloader.
230 * If we come here from a bootloader, kernel(text+data+bss+brk),
231 * ramdisk, zero_page, command line could be above 4G.
232 * We depend on an identity mapped page table being provided
233 * that maps our entire kernel(text+data+bss+brk), zero page
234 * and command line.
235 */
236#ifdef CONFIG_EFI_STUB
237 /*
238 * The entry point for the PE/COFF executable is efi_pe_entry, so
239 * only legacy boot loaders will execute this jmp.
240 */
241 jmp preferred_addr
242
243ENTRY(efi_pe_entry)
244 movq %rcx, efi64_config(%rip) /* Handle */
245 movq %rdx, efi64_config+8(%rip) /* EFI System table pointer */
246
247 leaq efi64_config(%rip), %rax
248 movq %rax, efi_config(%rip)
249
250 call 1f
2511: popq %rbp
252 subq $1b, %rbp
253
254 /*
255 * Relocate efi_config->call().
256 */
257 addq %rbp, efi64_config+88(%rip)
258
259 movq %rax, %rdi
260 call make_boot_params
261 cmpq $0,%rax
262 je fail
263 mov %rax, %rsi
264 leaq startup_32(%rip), %rax
265 movl %eax, BP_code32_start(%rsi)
266 jmp 2f /* Skip the relocation */
267
268handover_entry:
269 call 1f
2701: popq %rbp
271 subq $1b, %rbp
272
273 /*
274 * Relocate efi_config->call().
275 */
276 movq efi_config(%rip), %rax
277 addq %rbp, 88(%rax)
2782:
279 movq efi_config(%rip), %rdi
280 call efi_main
281 movq %rax,%rsi
282 cmpq $0,%rax
283 jne 2f
284fail:
285 /* EFI init failed, so hang. */
286 hlt
287 jmp fail
2882:
289 movl BP_code32_start(%esi), %eax
290 leaq preferred_addr(%rax), %rax
291 jmp *%rax
292
293preferred_addr:
294#endif
295
296 /* Setup data segments. */
297 xorl %eax, %eax
298 movl %eax, %ds
299 movl %eax, %es
300 movl %eax, %ss
301 movl %eax, %fs
302 movl %eax, %gs
303
304 /*
305 * Compute the decompressed kernel start address. It is where
306 * we were loaded at aligned to a 2M boundary. %rbp contains the
307 * decompressed kernel start address.
308 *
309 * If it is a relocatable kernel then decompress and run the kernel
310 * from load address aligned to 2MB addr, otherwise decompress and
311 * run the kernel from LOAD_PHYSICAL_ADDR
312 *
313 * We cannot rely on the calculation done in 32-bit mode, since we
314 * may have been invoked via the 64-bit entry point.
315 */
316
317 /* Start with the delta to where the kernel will run at. */
318#ifdef CONFIG_RELOCATABLE
319 leaq startup_32(%rip) /* - $startup_32 */, %rbp
320 movl BP_kernel_alignment(%rsi), %eax
321 decl %eax
322 addq %rax, %rbp
323 notq %rax
324 andq %rax, %rbp
325 cmpq $LOAD_PHYSICAL_ADDR, %rbp
326 jge 1f
327#endif
328 movq $LOAD_PHYSICAL_ADDR, %rbp
3291:
330
331 /* Target address to relocate to for decompression */
332 leaq z_extract_offset(%rbp), %rbx
333
334 /* Set up the stack */
335 leaq boot_stack_end(%rbx), %rsp
336
337 /* Zero EFLAGS */
338 pushq $0
339 popfq
340
341/*
342 * Copy the compressed kernel to the end of our buffer
343 * where decompression in place becomes safe.
344 */
345 pushq %rsi
346 leaq (_bss-8)(%rip), %rsi
347 leaq (_bss-8)(%rbx), %rdi
348 movq $_bss /* - $startup_32 */, %rcx
349 shrq $3, %rcx
350 std
351 rep movsq
352 cld
353 popq %rsi
354
355/*
356 * Jump to the relocated address.
357 */
358 leaq relocated(%rbx), %rax
359 jmp *%rax
360
361#ifdef CONFIG_EFI_STUB
362 .org 0x390
363ENTRY(efi64_stub_entry)
364 movq %rdi, efi64_config(%rip) /* Handle */
365 movq %rsi, efi64_config+8(%rip) /* EFI System table pointer */
366
367 leaq efi64_config(%rip), %rax
368 movq %rax, efi_config(%rip)
369
370 movq %rdx, %rsi
371 jmp handover_entry
372ENDPROC(efi64_stub_entry)
373#endif
374
375 .text
376relocated:
377
378/*
379 * Clear BSS (stack is currently empty)
380 */
381 xorl %eax, %eax
382 leaq _bss(%rip), %rdi
383 leaq _ebss(%rip), %rcx
384 subq %rdi, %rcx
385 shrq $3, %rcx
386 rep stosq
387
388/*
389 * Adjust our own GOT
390 */
391 leaq _got(%rip), %rdx
392 leaq _egot(%rip), %rcx
3931:
394 cmpq %rcx, %rdx
395 jae 2f
396 addq %rbx, (%rdx)
397 addq $8, %rdx
398 jmp 1b
3992:
400
401/*
402 * Do the decompression, and jump to the new kernel..
403 */
404 pushq %rsi /* Save the real mode argument */
405 movq %rsi, %rdi /* real mode address */
406 leaq boot_heap(%rip), %rsi /* malloc area for uncompression */
407 leaq input_data(%rip), %rdx /* input_data */
408 movl $z_input_len, %ecx /* input_len */
409 movq %rbp, %r8 /* output target address */
410 movq $z_output_len, %r9 /* decompressed length */
411 call decompress_kernel /* returns kernel location in %rax */
412 popq %rsi
413
414/*
415 * Jump to the decompressed kernel.
416 */
417 jmp *%rax
418
419 .code32
420no_longmode:
421 /* This isn't an x86-64 CPU so hang */
4221:
423 hlt
424 jmp 1b
425
426#include "../../kernel/verify_cpu.S"
427
428 .data
429gdt:
430 .word gdt_end - gdt
431 .long gdt
432 .word 0
433 .quad 0x0000000000000000 /* NULL descriptor */
434 .quad 0x00af9a000000ffff /* __KERNEL_CS */
435 .quad 0x00cf92000000ffff /* __KERNEL_DS */
436 .quad 0x0080890000000000 /* TS descriptor */
437 .quad 0x0000000000000000 /* TS continued */
438gdt_end:
439
440#ifdef CONFIG_EFI_STUB
441efi_config:
442 .quad 0
443
444#ifdef CONFIG_EFI_MIXED
445 .global efi32_config
446efi32_config:
447 .fill 11,8,0
448 .quad efi64_thunk
449 .byte 0
450#endif
451
452 .global efi64_config
453efi64_config:
454 .fill 11,8,0
455 .quad efi_call6
456 .byte 1
457#endif /* CONFIG_EFI_STUB */
458
459/*
460 * Stack and heap for uncompression
461 */
462 .bss
463 .balign 4
464boot_heap:
465 .fill BOOT_HEAP_SIZE, 1, 0
466boot_stack:
467 .fill BOOT_STACK_SIZE, 1, 0
468boot_stack_end:
469
470/*
471 * Space for page tables (not in .bss so not zeroed)
472 */
473 .section ".pgtable","a",@nobits
474 .balign 4096
475pgtable:
476 .fill 6*4096, 1, 0
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 .Lno_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 .Ladjust_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 * paging_prepare() returns a two-quadword structure which lands
362 * into RDX:RAX:
363 * - Address of the trampoline is returned in RAX.
364 * - Non zero RDX means trampoline needs to enable 5-level
365 * paging.
366 *
367 * RSI holds real mode data and needs to be preserved across
368 * this function call.
369 */
370 pushq %rsi
371 movq %rsi, %rdi /* real mode address */
372 call paging_prepare
373 popq %rsi
374
375 /* Save the trampoline address in RCX */
376 movq %rax, %rcx
377
378 /*
379 * Load the address of trampoline_return() into RDI.
380 * It will be used by the trampoline to return to the main code.
381 */
382 leaq trampoline_return(%rip), %rdi
383
384 /* Switch to compatibility mode (CS.L = 0 CS.D = 1) via far return */
385 pushq $__KERNEL32_CS
386 leaq TRAMPOLINE_32BIT_CODE_OFFSET(%rax), %rax
387 pushq %rax
388 lretq
389trampoline_return:
390 /* Restore the stack, the 32-bit trampoline uses its own stack */
391 leaq boot_stack_end(%rbx), %rsp
392
393 /*
394 * cleanup_trampoline() would restore trampoline memory.
395 *
396 * RDI is address of the page table to use instead of page table
397 * in trampoline memory (if required).
398 *
399 * RSI holds real mode data and needs to be preserved across
400 * this function call.
401 */
402 pushq %rsi
403 leaq top_pgtable(%rbx), %rdi
404 call cleanup_trampoline
405 popq %rsi
406
407 /* Zero EFLAGS */
408 pushq $0
409 popfq
410
411 /*
412 * Previously we've adjusted the GOT with address the binary was
413 * loaded at. Now we need to re-adjust for relocation address.
414 *
415 * Calculate the address the binary is loaded at, so that we can
416 * undo the previous GOT adjustment.
417 */
418 call 1f
4191: popq %rax
420 subq $1b, %rax
421
422 /* The new adjustment is the relocation address */
423 movq %rbx, %rdi
424 call .Ladjust_got
425
426/*
427 * Copy the compressed kernel to the end of our buffer
428 * where decompression in place becomes safe.
429 */
430 pushq %rsi
431 leaq (_bss-8)(%rip), %rsi
432 leaq (_bss-8)(%rbx), %rdi
433 movq $_bss /* - $startup_32 */, %rcx
434 shrq $3, %rcx
435 std
436 rep movsq
437 cld
438 popq %rsi
439
440/*
441 * Jump to the relocated address.
442 */
443 leaq .Lrelocated(%rbx), %rax
444 jmp *%rax
445
446#ifdef CONFIG_EFI_STUB
447
448/* The entry point for the PE/COFF executable is efi_pe_entry. */
449ENTRY(efi_pe_entry)
450 movq %rcx, efi64_config(%rip) /* Handle */
451 movq %rdx, efi64_config+8(%rip) /* EFI System table pointer */
452
453 leaq efi64_config(%rip), %rax
454 movq %rax, efi_config(%rip)
455
456 call 1f
4571: popq %rbp
458 subq $1b, %rbp
459
460 /*
461 * Relocate efi_config->call().
462 */
463 addq %rbp, efi64_config+40(%rip)
464
465 movq %rax, %rdi
466 call make_boot_params
467 cmpq $0,%rax
468 je fail
469 mov %rax, %rsi
470 leaq startup_32(%rip), %rax
471 movl %eax, BP_code32_start(%rsi)
472 jmp 2f /* Skip the relocation */
473
474handover_entry:
475 call 1f
4761: popq %rbp
477 subq $1b, %rbp
478
479 /*
480 * Relocate efi_config->call().
481 */
482 movq efi_config(%rip), %rax
483 addq %rbp, 40(%rax)
4842:
485 movq efi_config(%rip), %rdi
486 call efi_main
487 movq %rax,%rsi
488 cmpq $0,%rax
489 jne 2f
490fail:
491 /* EFI init failed, so hang. */
492 hlt
493 jmp fail
4942:
495 movl BP_code32_start(%esi), %eax
496 leaq startup_64(%rax), %rax
497 jmp *%rax
498ENDPROC(efi_pe_entry)
499
500 .org 0x390
501ENTRY(efi64_stub_entry)
502 movq %rdi, efi64_config(%rip) /* Handle */
503 movq %rsi, efi64_config+8(%rip) /* EFI System table pointer */
504
505 leaq efi64_config(%rip), %rax
506 movq %rax, efi_config(%rip)
507
508 movq %rdx, %rsi
509 jmp handover_entry
510ENDPROC(efi64_stub_entry)
511#endif
512
513 .text
514.Lrelocated:
515
516/*
517 * Clear BSS (stack is currently empty)
518 */
519 xorl %eax, %eax
520 leaq _bss(%rip), %rdi
521 leaq _ebss(%rip), %rcx
522 subq %rdi, %rcx
523 shrq $3, %rcx
524 rep stosq
525
526/*
527 * Do the extraction, and jump to the new kernel..
528 */
529 pushq %rsi /* Save the real mode argument */
530 movq %rsi, %rdi /* real mode address */
531 leaq boot_heap(%rip), %rsi /* malloc area for uncompression */
532 leaq input_data(%rip), %rdx /* input_data */
533 movl $z_input_len, %ecx /* input_len */
534 movq %rbp, %r8 /* output target address */
535 movq $z_output_len, %r9 /* decompressed length, end of relocs */
536 call extract_kernel /* returns kernel location in %rax */
537 popq %rsi
538
539/*
540 * Jump to the decompressed kernel.
541 */
542 jmp *%rax
543
544/*
545 * Adjust the global offset table
546 *
547 * RAX is the previous adjustment of the table to undo (use 0 if it's the
548 * first time we touch GOT).
549 * RDI is the new adjustment to apply.
550 */
551.Ladjust_got:
552 /* Walk through the GOT adding the address to the entries */
553 leaq _got(%rip), %rdx
554 leaq _egot(%rip), %rcx
5551:
556 cmpq %rcx, %rdx
557 jae 2f
558 subq %rax, (%rdx) /* Undo previous adjustment */
559 addq %rdi, (%rdx) /* Apply the new adjustment */
560 addq $8, %rdx
561 jmp 1b
5622:
563 ret
564
565 .code32
566/*
567 * This is the 32-bit trampoline that will be copied over to low memory.
568 *
569 * RDI contains the return address (might be above 4G).
570 * ECX contains the base address of the trampoline memory.
571 * Non zero RDX means trampoline needs to enable 5-level paging.
572 */
573ENTRY(trampoline_32bit_src)
574 /* Set up data and stack segments */
575 movl $__KERNEL_DS, %eax
576 movl %eax, %ds
577 movl %eax, %ss
578
579 /* Set up new stack */
580 leal TRAMPOLINE_32BIT_STACK_END(%ecx), %esp
581
582 /* Disable paging */
583 movl %cr0, %eax
584 btrl $X86_CR0_PG_BIT, %eax
585 movl %eax, %cr0
586
587 /* Check what paging mode we want to be in after the trampoline */
588 cmpl $0, %edx
589 jz 1f
590
591 /* We want 5-level paging: don't touch CR3 if it already points to 5-level page tables */
592 movl %cr4, %eax
593 testl $X86_CR4_LA57, %eax
594 jnz 3f
595 jmp 2f
5961:
597 /* We want 4-level paging: don't touch CR3 if it already points to 4-level page tables */
598 movl %cr4, %eax
599 testl $X86_CR4_LA57, %eax
600 jz 3f
6012:
602 /* Point CR3 to the trampoline's new top level page table */
603 leal TRAMPOLINE_32BIT_PGTABLE_OFFSET(%ecx), %eax
604 movl %eax, %cr3
6053:
606 /* Set EFER.LME=1 as a precaution in case hypervsior pulls the rug */
607 pushl %ecx
608 pushl %edx
609 movl $MSR_EFER, %ecx
610 rdmsr
611 btsl $_EFER_LME, %eax
612 wrmsr
613 popl %edx
614 popl %ecx
615
616 /* Enable PAE and LA57 (if required) paging modes */
617 movl $X86_CR4_PAE, %eax
618 cmpl $0, %edx
619 jz 1f
620 orl $X86_CR4_LA57, %eax
6211:
622 movl %eax, %cr4
623
624 /* Calculate address of paging_enabled() once we are executing in the trampoline */
625 leal .Lpaging_enabled - trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_OFFSET(%ecx), %eax
626
627 /* Prepare the stack for far return to Long Mode */
628 pushl $__KERNEL_CS
629 pushl %eax
630
631 /* Enable paging again */
632 movl $(X86_CR0_PG | X86_CR0_PE), %eax
633 movl %eax, %cr0
634
635 lret
636
637 .code64
638.Lpaging_enabled:
639 /* Return from the trampoline */
640 jmp *%rdi
641
642 /*
643 * The trampoline code has a size limit.
644 * Make sure we fail to compile if the trampoline code grows
645 * beyond TRAMPOLINE_32BIT_CODE_SIZE bytes.
646 */
647 .org trampoline_32bit_src + TRAMPOLINE_32BIT_CODE_SIZE
648
649 .code32
650.Lno_longmode:
651 /* This isn't an x86-64 CPU, so hang intentionally, we cannot continue */
6521:
653 hlt
654 jmp 1b
655
656#include "../../kernel/verify_cpu.S"
657
658 .data
659gdt64:
660 .word gdt_end - gdt
661 .quad 0
662 .balign 8
663gdt:
664 .word gdt_end - gdt
665 .long gdt
666 .word 0
667 .quad 0x00cf9a000000ffff /* __KERNEL32_CS */
668 .quad 0x00af9a000000ffff /* __KERNEL_CS */
669 .quad 0x00cf92000000ffff /* __KERNEL_DS */
670 .quad 0x0080890000000000 /* TS descriptor */
671 .quad 0x0000000000000000 /* TS continued */
672gdt_end:
673
674#ifdef CONFIG_EFI_STUB
675efi_config:
676 .quad 0
677
678#ifdef CONFIG_EFI_MIXED
679 .global efi32_config
680efi32_config:
681 .fill 5,8,0
682 .quad efi64_thunk
683 .byte 0
684#endif
685
686 .global efi64_config
687efi64_config:
688 .fill 5,8,0
689 .quad efi_call
690 .byte 1
691#endif /* CONFIG_EFI_STUB */
692
693/*
694 * Stack and heap for uncompression
695 */
696 .bss
697 .balign 4
698boot_heap:
699 .fill BOOT_HEAP_SIZE, 1, 0
700boot_stack:
701 .fill BOOT_STACK_SIZE, 1, 0
702boot_stack_end:
703
704/*
705 * Space for page tables (not in .bss so not zeroed)
706 */
707 .section ".pgtable","a",@nobits
708 .balign 4096
709pgtable:
710 .fill BOOT_PGT_SIZE, 1, 0
711
712/*
713 * The page table is going to be used instead of page table in the trampoline
714 * memory.
715 */
716top_pgtable:
717 .fill PAGE_SIZE, 1, 0