aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86_64/kernel/head.S
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86_64/kernel/head.S')
-rw-r--r--arch/x86_64/kernel/head.S233
1 files changed, 130 insertions, 103 deletions
diff --git a/arch/x86_64/kernel/head.S b/arch/x86_64/kernel/head.S
index c211e52f1333..36aa98a6d15c 100644
--- a/arch/x86_64/kernel/head.S
+++ b/arch/x86_64/kernel/head.S
@@ -5,6 +5,7 @@
5 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz> 5 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
6 * Copyright (C) 2000 Karsten Keil <kkeil@suse.de> 6 * Copyright (C) 2000 Karsten Keil <kkeil@suse.de>
7 * Copyright (C) 2001,2002 Andi Kleen <ak@suse.de> 7 * Copyright (C) 2001,2002 Andi Kleen <ak@suse.de>
8 * Copyright (C) 2005 Eric Biederman <ebiederm@xmission.com>
8 */ 9 */
9 10
10 11
@@ -17,95 +18,127 @@
17#include <asm/page.h> 18#include <asm/page.h>
18#include <asm/msr.h> 19#include <asm/msr.h>
19#include <asm/cache.h> 20#include <asm/cache.h>
20 21
21/* we are not able to switch in one step to the final KERNEL ADRESS SPACE 22/* we are not able to switch in one step to the final KERNEL ADRESS SPACE
22 * because we need identity-mapped pages on setup so define __START_KERNEL to 23 * because we need identity-mapped pages.
23 * 0x100000 for this stage 24 *
24 *
25 */ 25 */
26 26
27 .text 27 .text
28 .section .bootstrap.text 28 .section .bootstrap.text
29 .code32 29 .code64
30 .globl startup_32 30 .globl startup_64
31/* %bx: 1 if coming from smp trampoline on secondary cpu */ 31startup_64:
32startup_32: 32
33
34 /* 33 /*
35 * At this point the CPU runs in 32bit protected mode (CS.D = 1) with 34 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
36 * paging disabled and the point of this file is to switch to 64bit 35 * and someone has loaded an identity mapped page table
37 * long mode with a kernel mapping for kerneland to jump into the 36 * for us. These identity mapped page tables map all of the
38 * kernel virtual addresses. 37 * kernel pages and possibly all of memory.
39 * There is no stack until we set one up. 38 *
39 * %esi holds a physical pointer to real_mode_data.
40 *
41 * We come here either directly from a 64bit bootloader, or from
42 * arch/x86_64/boot/compressed/head.S.
43 *
44 * We only come here initially at boot nothing else comes here.
45 *
46 * Since we may be loaded at an address different from what we were
47 * compiled to run at we first fixup the physical addresses in our page
48 * tables and then reload them.
40 */ 49 */
41 50
42 /* Initialize the %ds segment register */ 51 /* Compute the delta between the address I am compiled to run at and the
43 movl $__KERNEL_DS,%eax 52 * address I am actually running at.
44 movl %eax,%ds
45
46 /* Load new GDT with the 64bit segments using 32bit descriptor */
47 lgdt pGDT32 - __START_KERNEL_map
48
49 /* If the CPU doesn't support CPUID this will double fault.
50 * Unfortunately it is hard to check for CPUID without a stack.
51 */ 53 */
52 54 leaq _text(%rip), %rbp
53 /* Check if extended functions are implemented */ 55 subq $_text - __START_KERNEL_map, %rbp
54 movl $0x80000000, %eax 56
55 cpuid 57 /* Is the address not 2M aligned? */
56 cmpl $0x80000000, %eax 58 movq %rbp, %rax
57 jbe no_long_mode 59 andl $~LARGE_PAGE_MASK, %eax
58 /* Check if long mode is implemented */ 60 testl %eax, %eax
59 mov $0x80000001, %eax 61 jnz bad_address
60 cpuid 62
61 btl $29, %edx 63 /* Is the address too large? */
62 jnc no_long_mode 64 leaq _text(%rip), %rdx
63 65 movq $PGDIR_SIZE, %rax
64 /* 66 cmpq %rax, %rdx
65 * Prepare for entering 64bits mode 67 jae bad_address
68
69 /* Fixup the physical addresses in the page table
66 */ 70 */
71 addq %rbp, init_level4_pgt + 0(%rip)
72 addq %rbp, init_level4_pgt + (258*8)(%rip)
73 addq %rbp, init_level4_pgt + (511*8)(%rip)
74
75 addq %rbp, level3_ident_pgt + 0(%rip)
76 addq %rbp, level3_kernel_pgt + (510*8)(%rip)
77
78 /* Add an Identity mapping if I am above 1G */
79 leaq _text(%rip), %rdi
80 andq $LARGE_PAGE_MASK, %rdi
81
82 movq %rdi, %rax
83 shrq $PUD_SHIFT, %rax
84 andq $(PTRS_PER_PUD - 1), %rax
85 jz ident_complete
86
87 leaq (level2_spare_pgt - __START_KERNEL_map + _KERNPG_TABLE)(%rbp), %rdx
88 leaq level3_ident_pgt(%rip), %rbx
89 movq %rdx, 0(%rbx, %rax, 8)
90
91 movq %rdi, %rax
92 shrq $PMD_SHIFT, %rax
93 andq $(PTRS_PER_PMD - 1), %rax
94 leaq __PAGE_KERNEL_LARGE_EXEC(%rdi), %rdx
95 leaq level2_spare_pgt(%rip), %rbx
96 movq %rdx, 0(%rbx, %rax, 8)
97ident_complete:
98
99 /* Fixup the kernel text+data virtual addresses
100 */
101 leaq level2_kernel_pgt(%rip), %rdi
102 leaq 4096(%rdi), %r8
103 /* See if it is a valid page table entry */
1041: testq $1, 0(%rdi)
105 jz 2f
106 addq %rbp, 0(%rdi)
107 /* Go to the next page */
1082: addq $8, %rdi
109 cmp %r8, %rdi
110 jne 1b
111
112 /* Fixup phys_base */
113 addq %rbp, phys_base(%rip)
67 114
68 /* Enable PAE mode */ 115#ifdef CONFIG_SMP
69 xorl %eax, %eax 116 addq %rbp, trampoline_level4_pgt + 0(%rip)
70 btsl $5, %eax 117 addq %rbp, trampoline_level4_pgt + (511*8)(%rip)
71 movl %eax, %cr4 118#endif
72 119#ifdef CONFIG_ACPI_SLEEP
73 /* Setup early boot stage 4 level pagetables */ 120 addq %rbp, wakeup_level4_pgt + 0(%rip)
74 movl $(init_level4_pgt - __START_KERNEL_map), %eax 121 addq %rbp, wakeup_level4_pgt + (511*8)(%rip)
75 movl %eax, %cr3 122#endif
76
77 /* Setup EFER (Extended Feature Enable Register) */
78 movl $MSR_EFER, %ecx
79 rdmsr
80
81 /* Enable Long Mode */
82 btsl $_EFER_LME, %eax
83
84 /* Make changes effective */
85 wrmsr
86 123
87 xorl %eax, %eax 124 /* Due to ENTRY(), sometimes the empty space gets filled with
88 btsl $31, %eax /* Enable paging and in turn activate Long Mode */ 125 * zeros. Better take a jmp than relying on empty space being
89 btsl $0, %eax /* Enable protected mode */ 126 * filled with 0x90 (nop)
90 /* Make changes effective */
91 movl %eax, %cr0
92 /*
93 * At this point we're in long mode but in 32bit compatibility mode
94 * with EFER.LME = 1, CS.L = 0, CS.D = 1 (and in turn
95 * EFER.LMA = 1). Now we want to jump in 64bit mode, to do that we use
96 * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
97 */ 127 */
98 ljmp $__KERNEL_CS, $(startup_64 - __START_KERNEL_map) 128 jmp secondary_startup_64
99
100 .code64
101 .org 0x100
102 .globl startup_64
103startup_64:
104ENTRY(secondary_startup_64) 129ENTRY(secondary_startup_64)
105 /* We come here either from startup_32 130 /*
106 * or directly from a 64bit bootloader. 131 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
107 * Since we may have come directly from a bootloader we 132 * and someone has loaded a mapped page table.
108 * reload the page tables here. 133 *
134 * %esi holds a physical pointer to real_mode_data.
135 *
136 * We come here either from startup_64 (using physical addresses)
137 * or from trampoline.S (using virtual addresses).
138 *
139 * Using virtual addresses from trampoline.S removes the need
140 * to have any identity mapped pages in the kernel page table
141 * after the boot processor executes this code.
109 */ 142 */
110 143
111 /* Enable PAE mode and PGE */ 144 /* Enable PAE mode and PGE */
@@ -116,8 +149,14 @@ ENTRY(secondary_startup_64)
116 149
117 /* Setup early boot stage 4 level pagetables. */ 150 /* Setup early boot stage 4 level pagetables. */
118 movq $(init_level4_pgt - __START_KERNEL_map), %rax 151 movq $(init_level4_pgt - __START_KERNEL_map), %rax
152 addq phys_base(%rip), %rax
119 movq %rax, %cr3 153 movq %rax, %cr3
120 154
155 /* Ensure I am executing from virtual addresses */
156 movq $1f, %rax
157 jmp *%rax
1581:
159
121 /* Check if nx is implemented */ 160 /* Check if nx is implemented */
122 movl $0x80000001, %eax 161 movl $0x80000001, %eax
123 cpuid 162 cpuid
@@ -126,17 +165,11 @@ ENTRY(secondary_startup_64)
126 /* Setup EFER (Extended Feature Enable Register) */ 165 /* Setup EFER (Extended Feature Enable Register) */
127 movl $MSR_EFER, %ecx 166 movl $MSR_EFER, %ecx
128 rdmsr 167 rdmsr
129 168 btsl $_EFER_SCE, %eax /* Enable System Call */
130 /* Enable System Call */ 169 btl $20,%edi /* No Execute supported? */
131 btsl $_EFER_SCE, %eax
132
133 /* No Execute supported? */
134 btl $20,%edi
135 jnc 1f 170 jnc 1f
136 btsl $_EFER_NX, %eax 171 btsl $_EFER_NX, %eax
1371: 1721: wrmsr /* Make changes effective */
138 /* Make changes effective */
139 wrmsr
140 173
141 /* Setup cr0 */ 174 /* Setup cr0 */
142#define CR0_PM 1 /* protected mode */ 175#define CR0_PM 1 /* protected mode */
@@ -163,7 +196,7 @@ ENTRY(secondary_startup_64)
163 * addresses where we're currently running on. We have to do that here 196 * addresses where we're currently running on. We have to do that here
164 * because in 32bit we couldn't load a 64bit linear address. 197 * because in 32bit we couldn't load a 64bit linear address.
165 */ 198 */
166 lgdt cpu_gdt_descr 199 lgdt cpu_gdt_descr(%rip)
167 200
168 /* set up data segments. actually 0 would do too */ 201 /* set up data segments. actually 0 would do too */
169 movl $__KERNEL_DS,%eax 202 movl $__KERNEL_DS,%eax
@@ -214,6 +247,9 @@ initial_code:
214init_rsp: 247init_rsp:
215 .quad init_thread_union+THREAD_SIZE-8 248 .quad init_thread_union+THREAD_SIZE-8
216 249
250bad_address:
251 jmp bad_address
252
217ENTRY(early_idt_handler) 253ENTRY(early_idt_handler)
218 cmpl $2,early_recursion_flag(%rip) 254 cmpl $2,early_recursion_flag(%rip)
219 jz 1f 255 jz 1f
@@ -242,23 +278,7 @@ early_idt_msg:
242early_idt_ripmsg: 278early_idt_ripmsg:
243 .asciz "RIP %s\n" 279 .asciz "RIP %s\n"
244 280
245.code32 281.balign PAGE_SIZE
246ENTRY(no_long_mode)
247 /* This isn't an x86-64 CPU so hang */
2481:
249 jmp 1b
250
251.org 0xf00
252 .globl pGDT32
253pGDT32:
254 .word gdt_end-cpu_gdt_table-1
255 .long cpu_gdt_table-__START_KERNEL_map
256
257.org 0xf10
258ljumpvector:
259 .long startup_64-__START_KERNEL_map
260 .word __KERNEL_CS
261
262ENTRY(stext) 282ENTRY(stext)
263ENTRY(_stext) 283ENTRY(_stext)
264 284
@@ -303,7 +323,7 @@ NEXT_PAGE(level2_ident_pgt)
303 * Don't set NX because code runs from these pages. 323 * Don't set NX because code runs from these pages.
304 */ 324 */
305 PMDS(0x0000000000000000, __PAGE_KERNEL_LARGE_EXEC, PTRS_PER_PMD) 325 PMDS(0x0000000000000000, __PAGE_KERNEL_LARGE_EXEC, PTRS_PER_PMD)
306 326
307NEXT_PAGE(level2_kernel_pgt) 327NEXT_PAGE(level2_kernel_pgt)
308 /* 40MB kernel mapping. The kernel code cannot be bigger than that. 328 /* 40MB kernel mapping. The kernel code cannot be bigger than that.
309 When you change this change KERNEL_TEXT_SIZE in page.h too. */ 329 When you change this change KERNEL_TEXT_SIZE in page.h too. */
@@ -313,6 +333,9 @@ NEXT_PAGE(level2_kernel_pgt)
313 /* Module mapping starts here */ 333 /* Module mapping starts here */
314 .fill (PTRS_PER_PMD - (KERNEL_TEXT_SIZE/PMD_SIZE)),8,0 334 .fill (PTRS_PER_PMD - (KERNEL_TEXT_SIZE/PMD_SIZE)),8,0
315 335
336NEXT_PAGE(level2_spare_pgt)
337 .fill 512,8,0
338
316#undef PMDS 339#undef PMDS
317#undef NEXT_PAGE 340#undef NEXT_PAGE
318 341
@@ -330,6 +353,10 @@ gdt:
330 .endr 353 .endr
331#endif 354#endif
332 355
356ENTRY(phys_base)
357 /* This must match the first entry in level2_kernel_pgt */
358 .quad 0x0000000000000000
359
333/* We need valid kernel segments for data and code in long mode too 360/* We need valid kernel segments for data and code in long mode too
334 * IRET will check the segment types kkeil 2000/10/28 361 * IRET will check the segment types kkeil 2000/10/28
335 * Also sysret mandates a special GDT layout 362 * Also sysret mandates a special GDT layout