Loading...
1/*
2 * linux/arch/x86_64/kernel/head.S -- start in 32bit and switch to 64bit
3 *
4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
6 * Copyright (C) 2000 Karsten Keil <kkeil@suse.de>
7 * Copyright (C) 2001,2002 Andi Kleen <ak@suse.de>
8 * Copyright (C) 2005 Eric Biederman <ebiederm@xmission.com>
9 */
10
11
12#include <linux/linkage.h>
13#include <linux/threads.h>
14#include <linux/init.h>
15#include <asm/segment.h>
16#include <asm/pgtable.h>
17#include <asm/page.h>
18#include <asm/msr.h>
19#include <asm/cache.h>
20#include <asm/processor-flags.h>
21#include <asm/percpu.h>
22#include <asm/nops.h>
23
24#ifdef CONFIG_PARAVIRT
25#include <asm/asm-offsets.h>
26#include <asm/paravirt.h>
27#define GET_CR2_INTO(reg) GET_CR2_INTO_RAX ; movq %rax, reg
28#else
29#define GET_CR2_INTO(reg) movq %cr2, reg
30#define INTERRUPT_RETURN iretq
31#endif
32
33/* we are not able to switch in one step to the final KERNEL ADDRESS SPACE
34 * because we need identity-mapped pages.
35 *
36 */
37
38#define pud_index(x) (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
39
40L4_PAGE_OFFSET = pgd_index(__PAGE_OFFSET)
41L3_PAGE_OFFSET = pud_index(__PAGE_OFFSET)
42L4_START_KERNEL = pgd_index(__START_KERNEL_map)
43L3_START_KERNEL = pud_index(__START_KERNEL_map)
44
45 .text
46 __HEAD
47 .code64
48 .globl startup_64
49startup_64:
50
51 /*
52 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
53 * and someone has loaded an identity mapped page table
54 * for us. These identity mapped page tables map all of the
55 * kernel pages and possibly all of memory.
56 *
57 * %esi holds a physical pointer to real_mode_data.
58 *
59 * We come here either directly from a 64bit bootloader, or from
60 * arch/x86_64/boot/compressed/head.S.
61 *
62 * We only come here initially at boot nothing else comes here.
63 *
64 * Since we may be loaded at an address different from what we were
65 * compiled to run at we first fixup the physical addresses in our page
66 * tables and then reload them.
67 */
68
69 /* Compute the delta between the address I am compiled to run at and the
70 * address I am actually running at.
71 */
72 leaq _text(%rip), %rbp
73 subq $_text - __START_KERNEL_map, %rbp
74
75 /* Is the address not 2M aligned? */
76 movq %rbp, %rax
77 andl $~PMD_PAGE_MASK, %eax
78 testl %eax, %eax
79 jnz bad_address
80
81 /* Is the address too large? */
82 leaq _text(%rip), %rdx
83 movq $PGDIR_SIZE, %rax
84 cmpq %rax, %rdx
85 jae bad_address
86
87 /* Fixup the physical addresses in the page table
88 */
89 addq %rbp, init_level4_pgt + 0(%rip)
90 addq %rbp, init_level4_pgt + (L4_PAGE_OFFSET*8)(%rip)
91 addq %rbp, init_level4_pgt + (L4_START_KERNEL*8)(%rip)
92
93 addq %rbp, level3_ident_pgt + 0(%rip)
94
95 addq %rbp, level3_kernel_pgt + (510*8)(%rip)
96 addq %rbp, level3_kernel_pgt + (511*8)(%rip)
97
98 addq %rbp, level2_fixmap_pgt + (506*8)(%rip)
99
100 /* Add an Identity mapping if I am above 1G */
101 leaq _text(%rip), %rdi
102 andq $PMD_PAGE_MASK, %rdi
103
104 movq %rdi, %rax
105 shrq $PUD_SHIFT, %rax
106 andq $(PTRS_PER_PUD - 1), %rax
107 jz ident_complete
108
109 leaq (level2_spare_pgt - __START_KERNEL_map + _KERNPG_TABLE)(%rbp), %rdx
110 leaq level3_ident_pgt(%rip), %rbx
111 movq %rdx, 0(%rbx, %rax, 8)
112
113 movq %rdi, %rax
114 shrq $PMD_SHIFT, %rax
115 andq $(PTRS_PER_PMD - 1), %rax
116 leaq __PAGE_KERNEL_IDENT_LARGE_EXEC(%rdi), %rdx
117 leaq level2_spare_pgt(%rip), %rbx
118 movq %rdx, 0(%rbx, %rax, 8)
119ident_complete:
120
121 /*
122 * Fixup the kernel text+data virtual addresses. Note that
123 * we might write invalid pmds, when the kernel is relocated
124 * cleanup_highmap() fixes this up along with the mappings
125 * beyond _end.
126 */
127
128 leaq level2_kernel_pgt(%rip), %rdi
129 leaq 4096(%rdi), %r8
130 /* See if it is a valid page table entry */
1311: testq $1, 0(%rdi)
132 jz 2f
133 addq %rbp, 0(%rdi)
134 /* Go to the next page */
1352: addq $8, %rdi
136 cmp %r8, %rdi
137 jne 1b
138
139 /* Fixup phys_base */
140 addq %rbp, phys_base(%rip)
141
142 /* Due to ENTRY(), sometimes the empty space gets filled with
143 * zeros. Better take a jmp than relying on empty space being
144 * filled with 0x90 (nop)
145 */
146 jmp secondary_startup_64
147ENTRY(secondary_startup_64)
148 /*
149 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
150 * and someone has loaded a mapped page table.
151 *
152 * %esi holds a physical pointer to real_mode_data.
153 *
154 * We come here either from startup_64 (using physical addresses)
155 * or from trampoline.S (using virtual addresses).
156 *
157 * Using virtual addresses from trampoline.S removes the need
158 * to have any identity mapped pages in the kernel page table
159 * after the boot processor executes this code.
160 */
161
162 /* Enable PAE mode and PGE */
163 movl $(X86_CR4_PAE | X86_CR4_PGE), %eax
164 movq %rax, %cr4
165
166 /* Setup early boot stage 4 level pagetables. */
167 movq $(init_level4_pgt - __START_KERNEL_map), %rax
168 addq phys_base(%rip), %rax
169 movq %rax, %cr3
170
171 /* Ensure I am executing from virtual addresses */
172 movq $1f, %rax
173 jmp *%rax
1741:
175
176 /* Check if nx is implemented */
177 movl $0x80000001, %eax
178 cpuid
179 movl %edx,%edi
180
181 /* Setup EFER (Extended Feature Enable Register) */
182 movl $MSR_EFER, %ecx
183 rdmsr
184 btsl $_EFER_SCE, %eax /* Enable System Call */
185 btl $20,%edi /* No Execute supported? */
186 jnc 1f
187 btsl $_EFER_NX, %eax
1881: wrmsr /* Make changes effective */
189
190 /* Setup cr0 */
191#define CR0_STATE (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
192 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
193 X86_CR0_PG)
194 movl $CR0_STATE, %eax
195 /* Make changes effective */
196 movq %rax, %cr0
197
198 /* Setup a boot time stack */
199 movq stack_start(%rip),%rsp
200
201 /* zero EFLAGS after setting rsp */
202 pushq $0
203 popfq
204
205 /*
206 * We must switch to a new descriptor in kernel space for the GDT
207 * because soon the kernel won't have access anymore to the userspace
208 * addresses where we're currently running on. We have to do that here
209 * because in 32bit we couldn't load a 64bit linear address.
210 */
211 lgdt early_gdt_descr(%rip)
212
213 /* set up data segments */
214 xorl %eax,%eax
215 movl %eax,%ds
216 movl %eax,%ss
217 movl %eax,%es
218
219 /*
220 * We don't really need to load %fs or %gs, but load them anyway
221 * to kill any stale realmode selectors. This allows execution
222 * under VT hardware.
223 */
224 movl %eax,%fs
225 movl %eax,%gs
226
227 /* Set up %gs.
228 *
229 * The base of %gs always points to the bottom of the irqstack
230 * union. If the stack protector canary is enabled, it is
231 * located at %gs:40. Note that, on SMP, the boot cpu uses
232 * init data section till per cpu areas are set up.
233 */
234 movl $MSR_GS_BASE,%ecx
235 movl initial_gs(%rip),%eax
236 movl initial_gs+4(%rip),%edx
237 wrmsr
238
239 /* esi is pointer to real mode structure with interesting info.
240 pass it to C */
241 movl %esi, %edi
242
243 /* Finally jump to run C code and to be on real kernel address
244 * Since we are running on identity-mapped space we have to jump
245 * to the full 64bit address, this is only possible as indirect
246 * jump. In addition we need to ensure %cs is set so we make this
247 * a far return.
248 */
249 movq initial_code(%rip),%rax
250 pushq $0 # fake return address to stop unwinder
251 pushq $__KERNEL_CS # set correct cs
252 pushq %rax # target address in negative space
253 lretq
254
255 /* SMP bootup changes these two */
256 __REFDATA
257 .align 8
258 ENTRY(initial_code)
259 .quad x86_64_start_kernel
260 ENTRY(initial_gs)
261 .quad INIT_PER_CPU_VAR(irq_stack_union)
262
263 ENTRY(stack_start)
264 .quad init_thread_union+THREAD_SIZE-8
265 .word 0
266 __FINITDATA
267
268bad_address:
269 jmp bad_address
270
271 .section ".init.text","ax"
272 .globl early_idt_handlers
273early_idt_handlers:
274 # 104(%rsp) %rflags
275 # 96(%rsp) %cs
276 # 88(%rsp) %rip
277 # 80(%rsp) error code
278 i = 0
279 .rept NUM_EXCEPTION_VECTORS
280 .if (EXCEPTION_ERRCODE_MASK >> i) & 1
281 ASM_NOP2
282 .else
283 pushq $0 # Dummy error code, to make stack frame uniform
284 .endif
285 pushq $i # 72(%rsp) Vector number
286 jmp early_idt_handler
287 i = i + 1
288 .endr
289
290ENTRY(early_idt_handler)
291 cld
292
293 cmpl $2,early_recursion_flag(%rip)
294 jz 1f
295 incl early_recursion_flag(%rip)
296
297 pushq %rax # 64(%rsp)
298 pushq %rcx # 56(%rsp)
299 pushq %rdx # 48(%rsp)
300 pushq %rsi # 40(%rsp)
301 pushq %rdi # 32(%rsp)
302 pushq %r8 # 24(%rsp)
303 pushq %r9 # 16(%rsp)
304 pushq %r10 # 8(%rsp)
305 pushq %r11 # 0(%rsp)
306
307 cmpl $__KERNEL_CS,96(%rsp)
308 jne 10f
309
310 leaq 88(%rsp),%rdi # Pointer to %rip
311 call early_fixup_exception
312 andl %eax,%eax
313 jnz 20f # Found an exception entry
314
31510:
316#ifdef CONFIG_EARLY_PRINTK
317 GET_CR2_INTO(%r9) # can clobber any volatile register if pv
318 movl 80(%rsp),%r8d # error code
319 movl 72(%rsp),%esi # vector number
320 movl 96(%rsp),%edx # %cs
321 movq 88(%rsp),%rcx # %rip
322 xorl %eax,%eax
323 leaq early_idt_msg(%rip),%rdi
324 call early_printk
325 cmpl $2,early_recursion_flag(%rip)
326 jz 1f
327 call dump_stack
328#ifdef CONFIG_KALLSYMS
329 leaq early_idt_ripmsg(%rip),%rdi
330 movq 40(%rsp),%rsi # %rip again
331 call __print_symbol
332#endif
333#endif /* EARLY_PRINTK */
3341: hlt
335 jmp 1b
336
33720: # Exception table entry found
338 popq %r11
339 popq %r10
340 popq %r9
341 popq %r8
342 popq %rdi
343 popq %rsi
344 popq %rdx
345 popq %rcx
346 popq %rax
347 addq $16,%rsp # drop vector number and error code
348 decl early_recursion_flag(%rip)
349 INTERRUPT_RETURN
350
351 .balign 4
352early_recursion_flag:
353 .long 0
354
355#ifdef CONFIG_EARLY_PRINTK
356early_idt_msg:
357 .asciz "PANIC: early exception %02lx rip %lx:%lx error %lx cr2 %lx\n"
358early_idt_ripmsg:
359 .asciz "RIP %s\n"
360#endif /* CONFIG_EARLY_PRINTK */
361 .previous
362
363#define NEXT_PAGE(name) \
364 .balign PAGE_SIZE; \
365ENTRY(name)
366
367/* Automate the creation of 1 to 1 mapping pmd entries */
368#define PMDS(START, PERM, COUNT) \
369 i = 0 ; \
370 .rept (COUNT) ; \
371 .quad (START) + (i << PMD_SHIFT) + (PERM) ; \
372 i = i + 1 ; \
373 .endr
374
375 .data
376 /*
377 * This default setting generates an ident mapping at address 0x100000
378 * and a mapping for the kernel that precisely maps virtual address
379 * 0xffffffff80000000 to physical address 0x000000. (always using
380 * 2Mbyte large pages provided by PAE mode)
381 */
382NEXT_PAGE(init_level4_pgt)
383 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
384 .org init_level4_pgt + L4_PAGE_OFFSET*8, 0
385 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
386 .org init_level4_pgt + L4_START_KERNEL*8, 0
387 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
388 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
389
390NEXT_PAGE(level3_ident_pgt)
391 .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
392 .fill 511,8,0
393
394NEXT_PAGE(level3_kernel_pgt)
395 .fill L3_START_KERNEL,8,0
396 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
397 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE
398 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
399
400NEXT_PAGE(level2_fixmap_pgt)
401 .fill 506,8,0
402 .quad level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
403 /* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */
404 .fill 5,8,0
405
406NEXT_PAGE(level1_fixmap_pgt)
407 .fill 512,8,0
408
409NEXT_PAGE(level2_ident_pgt)
410 /* Since I easily can, map the first 1G.
411 * Don't set NX because code runs from these pages.
412 */
413 PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
414
415NEXT_PAGE(level2_kernel_pgt)
416 /*
417 * 512 MB kernel mapping. We spend a full page on this pagetable
418 * anyway.
419 *
420 * The kernel code+data+bss must not be bigger than that.
421 *
422 * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
423 * If you want to increase this then increase MODULES_VADDR
424 * too.)
425 */
426 PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
427 KERNEL_IMAGE_SIZE/PMD_SIZE)
428
429NEXT_PAGE(level2_spare_pgt)
430 .fill 512, 8, 0
431
432#undef PMDS
433#undef NEXT_PAGE
434
435 .data
436 .align 16
437 .globl early_gdt_descr
438early_gdt_descr:
439 .word GDT_ENTRIES*8-1
440early_gdt_descr_base:
441 .quad INIT_PER_CPU_VAR(gdt_page)
442
443ENTRY(phys_base)
444 /* This must match the first entry in level2_kernel_pgt */
445 .quad 0x0000000000000000
446
447#include "../../x86/xen/xen-head.S"
448
449 .section .bss, "aw", @nobits
450 .align L1_CACHE_BYTES
451ENTRY(idt_table)
452 .skip IDT_ENTRIES * 16
453
454 .align L1_CACHE_BYTES
455ENTRY(nmi_idt_table)
456 .skip IDT_ENTRIES * 16
457
458 __PAGE_ALIGNED_BSS
459 .align PAGE_SIZE
460ENTRY(empty_zero_page)
461 .skip PAGE_SIZE
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * linux/arch/x86/kernel/head_64.S -- start in 32bit and switch to 64bit
4 *
5 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
7 * Copyright (C) 2000 Karsten Keil <kkeil@suse.de>
8 * Copyright (C) 2001,2002 Andi Kleen <ak@suse.de>
9 * Copyright (C) 2005 Eric Biederman <ebiederm@xmission.com>
10 */
11
12
13#include <linux/linkage.h>
14#include <linux/threads.h>
15#include <linux/init.h>
16#include <linux/pgtable.h>
17#include <asm/segment.h>
18#include <asm/page.h>
19#include <asm/msr.h>
20#include <asm/cache.h>
21#include <asm/processor-flags.h>
22#include <asm/percpu.h>
23#include <asm/nops.h>
24#include "../entry/calling.h"
25#include <asm/export.h>
26#include <asm/nospec-branch.h>
27#include <asm/fixmap.h>
28
29#ifdef CONFIG_PARAVIRT_XXL
30#include <asm/asm-offsets.h>
31#include <asm/paravirt.h>
32#define GET_CR2_INTO(reg) GET_CR2_INTO_AX ; _ASM_MOV %_ASM_AX, reg
33#else
34#define INTERRUPT_RETURN iretq
35#define GET_CR2_INTO(reg) _ASM_MOV %cr2, reg
36#endif
37
38/*
39 * We are not able to switch in one step to the final KERNEL ADDRESS SPACE
40 * because we need identity-mapped pages.
41 */
42#define l4_index(x) (((x) >> 39) & 511)
43#define pud_index(x) (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
44
45L4_PAGE_OFFSET = l4_index(__PAGE_OFFSET_BASE_L4)
46L4_START_KERNEL = l4_index(__START_KERNEL_map)
47
48L3_START_KERNEL = pud_index(__START_KERNEL_map)
49
50 .text
51 __HEAD
52 .code64
53SYM_CODE_START_NOALIGN(startup_64)
54 UNWIND_HINT_EMPTY
55 /*
56 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
57 * and someone has loaded an identity mapped page table
58 * for us. These identity mapped page tables map all of the
59 * kernel pages and possibly all of memory.
60 *
61 * %rsi holds a physical pointer to real_mode_data.
62 *
63 * We come here either directly from a 64bit bootloader, or from
64 * arch/x86/boot/compressed/head_64.S.
65 *
66 * We only come here initially at boot nothing else comes here.
67 *
68 * Since we may be loaded at an address different from what we were
69 * compiled to run at we first fixup the physical addresses in our page
70 * tables and then reload them.
71 */
72
73 /* Set up the stack for verify_cpu(), similar to initial_stack below */
74 leaq (__end_init_task - SIZEOF_PTREGS)(%rip), %rsp
75
76 /* Sanitize CPU configuration */
77 call verify_cpu
78
79 /*
80 * Perform pagetable fixups. Additionally, if SME is active, encrypt
81 * the kernel and retrieve the modifier (SME encryption mask if SME
82 * is active) to be added to the initial pgdir entry that will be
83 * programmed into CR3.
84 */
85 leaq _text(%rip), %rdi
86 pushq %rsi
87 call __startup_64
88 popq %rsi
89
90 /* Form the CR3 value being sure to include the CR3 modifier */
91 addq $(early_top_pgt - __START_KERNEL_map), %rax
92 jmp 1f
93SYM_CODE_END(startup_64)
94
95SYM_CODE_START(secondary_startup_64)
96 UNWIND_HINT_EMPTY
97 /*
98 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
99 * and someone has loaded a mapped page table.
100 *
101 * %rsi holds a physical pointer to real_mode_data.
102 *
103 * We come here either from startup_64 (using physical addresses)
104 * or from trampoline.S (using virtual addresses).
105 *
106 * Using virtual addresses from trampoline.S removes the need
107 * to have any identity mapped pages in the kernel page table
108 * after the boot processor executes this code.
109 */
110
111 /* Sanitize CPU configuration */
112 call verify_cpu
113
114 /*
115 * Retrieve the modifier (SME encryption mask if SME is active) to be
116 * added to the initial pgdir entry that will be programmed into CR3.
117 */
118 pushq %rsi
119 call __startup_secondary_64
120 popq %rsi
121
122 /* Form the CR3 value being sure to include the CR3 modifier */
123 addq $(init_top_pgt - __START_KERNEL_map), %rax
1241:
125
126 /* Enable PAE mode, PGE and LA57 */
127 movl $(X86_CR4_PAE | X86_CR4_PGE), %ecx
128#ifdef CONFIG_X86_5LEVEL
129 testl $1, __pgtable_l5_enabled(%rip)
130 jz 1f
131 orl $X86_CR4_LA57, %ecx
1321:
133#endif
134 movq %rcx, %cr4
135
136 /* Setup early boot stage 4-/5-level pagetables. */
137 addq phys_base(%rip), %rax
138 movq %rax, %cr3
139
140 /* Ensure I am executing from virtual addresses */
141 movq $1f, %rax
142 ANNOTATE_RETPOLINE_SAFE
143 jmp *%rax
1441:
145 UNWIND_HINT_EMPTY
146
147 /* Check if nx is implemented */
148 movl $0x80000001, %eax
149 cpuid
150 movl %edx,%edi
151
152 /* Setup EFER (Extended Feature Enable Register) */
153 movl $MSR_EFER, %ecx
154 rdmsr
155 btsl $_EFER_SCE, %eax /* Enable System Call */
156 btl $20,%edi /* No Execute supported? */
157 jnc 1f
158 btsl $_EFER_NX, %eax
159 btsq $_PAGE_BIT_NX,early_pmd_flags(%rip)
1601: wrmsr /* Make changes effective */
161
162 /* Setup cr0 */
163 movl $CR0_STATE, %eax
164 /* Make changes effective */
165 movq %rax, %cr0
166
167 /* Setup a boot time stack */
168 movq initial_stack(%rip), %rsp
169
170 /* zero EFLAGS after setting rsp */
171 pushq $0
172 popfq
173
174 /*
175 * We must switch to a new descriptor in kernel space for the GDT
176 * because soon the kernel won't have access anymore to the userspace
177 * addresses where we're currently running on. We have to do that here
178 * because in 32bit we couldn't load a 64bit linear address.
179 */
180 lgdt early_gdt_descr(%rip)
181
182 /* set up data segments */
183 xorl %eax,%eax
184 movl %eax,%ds
185 movl %eax,%ss
186 movl %eax,%es
187
188 /*
189 * We don't really need to load %fs or %gs, but load them anyway
190 * to kill any stale realmode selectors. This allows execution
191 * under VT hardware.
192 */
193 movl %eax,%fs
194 movl %eax,%gs
195
196 /* Set up %gs.
197 *
198 * The base of %gs always points to fixed_percpu_data. If the
199 * stack protector canary is enabled, it is located at %gs:40.
200 * Note that, on SMP, the boot cpu uses init data section until
201 * the per cpu areas are set up.
202 */
203 movl $MSR_GS_BASE,%ecx
204 movl initial_gs(%rip),%eax
205 movl initial_gs+4(%rip),%edx
206 wrmsr
207
208 /* rsi is pointer to real mode structure with interesting info.
209 pass it to C */
210 movq %rsi, %rdi
211
212.Ljump_to_C_code:
213 /*
214 * Jump to run C code and to be on a real kernel address.
215 * Since we are running on identity-mapped space we have to jump
216 * to the full 64bit address, this is only possible as indirect
217 * jump. In addition we need to ensure %cs is set so we make this
218 * a far return.
219 *
220 * Note: do not change to far jump indirect with 64bit offset.
221 *
222 * AMD does not support far jump indirect with 64bit offset.
223 * AMD64 Architecture Programmer's Manual, Volume 3: states only
224 * JMP FAR mem16:16 FF /5 Far jump indirect,
225 * with the target specified by a far pointer in memory.
226 * JMP FAR mem16:32 FF /5 Far jump indirect,
227 * with the target specified by a far pointer in memory.
228 *
229 * Intel64 does support 64bit offset.
230 * Software Developer Manual Vol 2: states:
231 * FF /5 JMP m16:16 Jump far, absolute indirect,
232 * address given in m16:16
233 * FF /5 JMP m16:32 Jump far, absolute indirect,
234 * address given in m16:32.
235 * REX.W + FF /5 JMP m16:64 Jump far, absolute indirect,
236 * address given in m16:64.
237 */
238 pushq $.Lafter_lret # put return address on stack for unwinder
239 xorl %ebp, %ebp # clear frame pointer
240 movq initial_code(%rip), %rax
241 pushq $__KERNEL_CS # set correct cs
242 pushq %rax # target address in negative space
243 lretq
244.Lafter_lret:
245SYM_CODE_END(secondary_startup_64)
246
247#include "verify_cpu.S"
248
249#ifdef CONFIG_HOTPLUG_CPU
250/*
251 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
252 * up already except stack. We just set up stack here. Then call
253 * start_secondary() via .Ljump_to_C_code.
254 */
255SYM_CODE_START(start_cpu0)
256 UNWIND_HINT_EMPTY
257 movq initial_stack(%rip), %rsp
258 jmp .Ljump_to_C_code
259SYM_CODE_END(start_cpu0)
260#endif
261
262 /* Both SMP bootup and ACPI suspend change these variables */
263 __REFDATA
264 .balign 8
265SYM_DATA(initial_code, .quad x86_64_start_kernel)
266SYM_DATA(initial_gs, .quad INIT_PER_CPU_VAR(fixed_percpu_data))
267
268/*
269 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel unwinder
270 * reliably detect the end of the stack.
271 */
272SYM_DATA(initial_stack, .quad init_thread_union + THREAD_SIZE - SIZEOF_PTREGS)
273 __FINITDATA
274
275 __INIT
276SYM_CODE_START(early_idt_handler_array)
277 i = 0
278 .rept NUM_EXCEPTION_VECTORS
279 .if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
280 UNWIND_HINT_IRET_REGS
281 pushq $0 # Dummy error code, to make stack frame uniform
282 .else
283 UNWIND_HINT_IRET_REGS offset=8
284 .endif
285 pushq $i # 72(%rsp) Vector number
286 jmp early_idt_handler_common
287 UNWIND_HINT_IRET_REGS
288 i = i + 1
289 .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
290 .endr
291 UNWIND_HINT_IRET_REGS offset=16
292SYM_CODE_END(early_idt_handler_array)
293
294SYM_CODE_START_LOCAL(early_idt_handler_common)
295 /*
296 * The stack is the hardware frame, an error code or zero, and the
297 * vector number.
298 */
299 cld
300
301 incl early_recursion_flag(%rip)
302
303 /* The vector number is currently in the pt_regs->di slot. */
304 pushq %rsi /* pt_regs->si */
305 movq 8(%rsp), %rsi /* RSI = vector number */
306 movq %rdi, 8(%rsp) /* pt_regs->di = RDI */
307 pushq %rdx /* pt_regs->dx */
308 pushq %rcx /* pt_regs->cx */
309 pushq %rax /* pt_regs->ax */
310 pushq %r8 /* pt_regs->r8 */
311 pushq %r9 /* pt_regs->r9 */
312 pushq %r10 /* pt_regs->r10 */
313 pushq %r11 /* pt_regs->r11 */
314 pushq %rbx /* pt_regs->bx */
315 pushq %rbp /* pt_regs->bp */
316 pushq %r12 /* pt_regs->r12 */
317 pushq %r13 /* pt_regs->r13 */
318 pushq %r14 /* pt_regs->r14 */
319 pushq %r15 /* pt_regs->r15 */
320 UNWIND_HINT_REGS
321
322 cmpq $14,%rsi /* Page fault? */
323 jnz 10f
324 GET_CR2_INTO(%rdi) /* can clobber %rax if pv */
325 call early_make_pgtable
326 andl %eax,%eax
327 jz 20f /* All good */
328
32910:
330 movq %rsp,%rdi /* RDI = pt_regs; RSI is already trapnr */
331 call early_fixup_exception
332
33320:
334 decl early_recursion_flag(%rip)
335 jmp restore_regs_and_return_to_kernel
336SYM_CODE_END(early_idt_handler_common)
337
338
339#define SYM_DATA_START_PAGE_ALIGNED(name) \
340 SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE)
341
342#ifdef CONFIG_PAGE_TABLE_ISOLATION
343/*
344 * Each PGD needs to be 8k long and 8k aligned. We do not
345 * ever go out to userspace with these, so we do not
346 * strictly *need* the second page, but this allows us to
347 * have a single set_pgd() implementation that does not
348 * need to worry about whether it has 4k or 8k to work
349 * with.
350 *
351 * This ensures PGDs are 8k long:
352 */
353#define PTI_USER_PGD_FILL 512
354/* This ensures they are 8k-aligned: */
355#define SYM_DATA_START_PTI_ALIGNED(name) \
356 SYM_START(name, SYM_L_GLOBAL, .balign 2 * PAGE_SIZE)
357#else
358#define SYM_DATA_START_PTI_ALIGNED(name) \
359 SYM_DATA_START_PAGE_ALIGNED(name)
360#define PTI_USER_PGD_FILL 0
361#endif
362
363/* Automate the creation of 1 to 1 mapping pmd entries */
364#define PMDS(START, PERM, COUNT) \
365 i = 0 ; \
366 .rept (COUNT) ; \
367 .quad (START) + (i << PMD_SHIFT) + (PERM) ; \
368 i = i + 1 ; \
369 .endr
370
371 __INITDATA
372 .balign 4
373
374SYM_DATA_START_PTI_ALIGNED(early_top_pgt)
375 .fill 512,8,0
376 .fill PTI_USER_PGD_FILL,8,0
377SYM_DATA_END(early_top_pgt)
378
379SYM_DATA_START_PAGE_ALIGNED(early_dynamic_pgts)
380 .fill 512*EARLY_DYNAMIC_PAGE_TABLES,8,0
381SYM_DATA_END(early_dynamic_pgts)
382
383SYM_DATA(early_recursion_flag, .long 0)
384
385 .data
386
387#if defined(CONFIG_XEN_PV) || defined(CONFIG_PVH)
388SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
389 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
390 .org init_top_pgt + L4_PAGE_OFFSET*8, 0
391 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
392 .org init_top_pgt + L4_START_KERNEL*8, 0
393 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
394 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
395 .fill PTI_USER_PGD_FILL,8,0
396SYM_DATA_END(init_top_pgt)
397
398SYM_DATA_START_PAGE_ALIGNED(level3_ident_pgt)
399 .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
400 .fill 511, 8, 0
401SYM_DATA_END(level3_ident_pgt)
402SYM_DATA_START_PAGE_ALIGNED(level2_ident_pgt)
403 /*
404 * Since I easily can, map the first 1G.
405 * Don't set NX because code runs from these pages.
406 *
407 * Note: This sets _PAGE_GLOBAL despite whether
408 * the CPU supports it or it is enabled. But,
409 * the CPU should ignore the bit.
410 */
411 PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
412SYM_DATA_END(level2_ident_pgt)
413#else
414SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
415 .fill 512,8,0
416 .fill PTI_USER_PGD_FILL,8,0
417SYM_DATA_END(init_top_pgt)
418#endif
419
420#ifdef CONFIG_X86_5LEVEL
421SYM_DATA_START_PAGE_ALIGNED(level4_kernel_pgt)
422 .fill 511,8,0
423 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
424SYM_DATA_END(level4_kernel_pgt)
425#endif
426
427SYM_DATA_START_PAGE_ALIGNED(level3_kernel_pgt)
428 .fill L3_START_KERNEL,8,0
429 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
430 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
431 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
432SYM_DATA_END(level3_kernel_pgt)
433
434SYM_DATA_START_PAGE_ALIGNED(level2_kernel_pgt)
435 /*
436 * 512 MB kernel mapping. We spend a full page on this pagetable
437 * anyway.
438 *
439 * The kernel code+data+bss must not be bigger than that.
440 *
441 * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
442 * If you want to increase this then increase MODULES_VADDR
443 * too.)
444 *
445 * This table is eventually used by the kernel during normal
446 * runtime. Care must be taken to clear out undesired bits
447 * later, like _PAGE_RW or _PAGE_GLOBAL in some cases.
448 */
449 PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
450 KERNEL_IMAGE_SIZE/PMD_SIZE)
451SYM_DATA_END(level2_kernel_pgt)
452
453SYM_DATA_START_PAGE_ALIGNED(level2_fixmap_pgt)
454 .fill (512 - 4 - FIXMAP_PMD_NUM),8,0
455 pgtno = 0
456 .rept (FIXMAP_PMD_NUM)
457 .quad level1_fixmap_pgt + (pgtno << PAGE_SHIFT) - __START_KERNEL_map \
458 + _PAGE_TABLE_NOENC;
459 pgtno = pgtno + 1
460 .endr
461 /* 6 MB reserved space + a 2MB hole */
462 .fill 4,8,0
463SYM_DATA_END(level2_fixmap_pgt)
464
465SYM_DATA_START_PAGE_ALIGNED(level1_fixmap_pgt)
466 .rept (FIXMAP_PMD_NUM)
467 .fill 512,8,0
468 .endr
469SYM_DATA_END(level1_fixmap_pgt)
470
471#undef PMDS
472
473 .data
474 .align 16
475
476SYM_DATA(early_gdt_descr, .word GDT_ENTRIES*8-1)
477SYM_DATA_LOCAL(early_gdt_descr_base, .quad INIT_PER_CPU_VAR(gdt_page))
478
479 .align 16
480/* This must match the first entry in level2_kernel_pgt */
481SYM_DATA(phys_base, .quad 0x0)
482EXPORT_SYMBOL(phys_base)
483
484#include "../../x86/xen/xen-head.S"
485
486 __PAGE_ALIGNED_BSS
487SYM_DATA_START_PAGE_ALIGNED(empty_zero_page)
488 .skip PAGE_SIZE
489SYM_DATA_END(empty_zero_page)
490EXPORT_SYMBOL(empty_zero_page)
491