]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
9 | * $Id: head.S,v 1.49 2002/03/19 17:39:25 ak Exp $ | |
10 | */ | |
11 | ||
12 | ||
13 | #include <linux/linkage.h> | |
14 | #include <linux/threads.h> | |
f6c2e333 | 15 | #include <linux/init.h> |
1da177e4 LT |
16 | #include <asm/desc.h> |
17 | #include <asm/segment.h> | |
18 | #include <asm/page.h> | |
19 | #include <asm/msr.h> | |
20 | #include <asm/cache.h> | |
21 | ||
22 | /* we are not able to switch in one step to the final KERNEL ADRESS SPACE | |
23 | * because we need identity-mapped pages on setup so define __START_KERNEL to | |
24 | * 0x100000 for this stage | |
25 | * | |
26 | */ | |
27 | ||
28 | .text | |
eaeae0cc | 29 | .section .bootstrap.text |
1da177e4 LT |
30 | .code32 |
31 | .globl startup_32 | |
32 | /* %bx: 1 if coming from smp trampoline on secondary cpu */ | |
33 | startup_32: | |
34 | ||
35 | /* | |
36 | * At this point the CPU runs in 32bit protected mode (CS.D = 1) with | |
37 | * paging disabled and the point of this file is to switch to 64bit | |
38 | * long mode with a kernel mapping for kerneland to jump into the | |
39 | * kernel virtual addresses. | |
40 | * There is no stack until we set one up. | |
41 | */ | |
42 | ||
43 | /* Initialize the %ds segment register */ | |
44 | movl $__KERNEL_DS,%eax | |
45 | movl %eax,%ds | |
46 | ||
47 | /* Load new GDT with the 64bit segments using 32bit descriptor */ | |
48 | lgdt pGDT32 - __START_KERNEL_map | |
49 | ||
50 | /* If the CPU doesn't support CPUID this will double fault. | |
51 | * Unfortunately it is hard to check for CPUID without a stack. | |
52 | */ | |
53 | ||
54 | /* Check if extended functions are implemented */ | |
55 | movl $0x80000000, %eax | |
56 | cpuid | |
57 | cmpl $0x80000000, %eax | |
58 | jbe no_long_mode | |
59 | /* Check if long mode is implemented */ | |
60 | mov $0x80000001, %eax | |
61 | cpuid | |
62 | btl $29, %edx | |
63 | jnc no_long_mode | |
64 | ||
65 | /* | |
66 | * Prepare for entering 64bits mode | |
67 | */ | |
68 | ||
69 | /* Enable PAE mode */ | |
70 | xorl %eax, %eax | |
71 | btsl $5, %eax | |
72 | movl %eax, %cr4 | |
73 | ||
74 | /* Setup early boot stage 4 level pagetables */ | |
f6c2e333 | 75 | movl $(boot_level4_pgt - __START_KERNEL_map), %eax |
1da177e4 LT |
76 | movl %eax, %cr3 |
77 | ||
78 | /* Setup EFER (Extended Feature Enable Register) */ | |
79 | movl $MSR_EFER, %ecx | |
80 | rdmsr | |
81 | ||
82 | /* Enable Long Mode */ | |
83 | btsl $_EFER_LME, %eax | |
84 | ||
85 | /* Make changes effective */ | |
86 | wrmsr | |
87 | ||
88 | xorl %eax, %eax | |
89 | btsl $31, %eax /* Enable paging and in turn activate Long Mode */ | |
90 | btsl $0, %eax /* Enable protected mode */ | |
91 | /* Make changes effective */ | |
92 | movl %eax, %cr0 | |
93 | /* | |
94 | * At this point we're in long mode but in 32bit compatibility mode | |
95 | * with EFER.LME = 1, CS.L = 0, CS.D = 1 (and in turn | |
96 | * EFER.LMA = 1). Now we want to jump in 64bit mode, to do that we use | |
97 | * the new gdt/idt that has __KERNEL_CS with CS.L = 1. | |
98 | */ | |
99 | ljmp $__KERNEL_CS, $(startup_64 - __START_KERNEL_map) | |
100 | ||
101 | .code64 | |
102 | .org 0x100 | |
103 | .globl startup_64 | |
104 | startup_64: | |
105 | /* We come here either from startup_32 | |
106 | * or directly from a 64bit bootloader. | |
107 | * Since we may have come directly from a bootloader we | |
108 | * reload the page tables here. | |
109 | */ | |
110 | ||
111 | /* Enable PAE mode and PGE */ | |
112 | xorq %rax, %rax | |
113 | btsq $5, %rax | |
114 | btsq $7, %rax | |
115 | movq %rax, %cr4 | |
116 | ||
117 | /* Setup early boot stage 4 level pagetables. */ | |
f6c2e333 | 118 | movq $(boot_level4_pgt - __START_KERNEL_map), %rax |
1da177e4 LT |
119 | movq %rax, %cr3 |
120 | ||
121 | /* Check if nx is implemented */ | |
122 | movl $0x80000001, %eax | |
123 | cpuid | |
124 | movl %edx,%edi | |
125 | ||
126 | /* Setup EFER (Extended Feature Enable Register) */ | |
127 | movl $MSR_EFER, %ecx | |
128 | rdmsr | |
129 | ||
130 | /* Enable System Call */ | |
131 | btsl $_EFER_SCE, %eax | |
132 | ||
133 | /* No Execute supported? */ | |
134 | btl $20,%edi | |
135 | jnc 1f | |
136 | btsl $_EFER_NX, %eax | |
137 | 1: | |
138 | /* Make changes effective */ | |
139 | wrmsr | |
140 | ||
141 | /* Setup cr0 */ | |
3829ee6b AK |
142 | #define CR0_PM 1 /* protected mode */ |
143 | #define CR0_MP (1<<1) | |
144 | #define CR0_ET (1<<4) | |
145 | #define CR0_NE (1<<5) | |
146 | #define CR0_WP (1<<16) | |
147 | #define CR0_AM (1<<18) | |
148 | #define CR0_PAGING (1<<31) | |
149 | movl $CR0_PM|CR0_MP|CR0_ET|CR0_NE|CR0_WP|CR0_AM|CR0_PAGING,%eax | |
1da177e4 LT |
150 | /* Make changes effective */ |
151 | movq %rax, %cr0 | |
152 | ||
153 | /* Setup a boot time stack */ | |
154 | movq init_rsp(%rip),%rsp | |
155 | ||
156 | /* zero EFLAGS after setting rsp */ | |
157 | pushq $0 | |
158 | popfq | |
159 | ||
160 | /* | |
161 | * We must switch to a new descriptor in kernel space for the GDT | |
162 | * because soon the kernel won't have access anymore to the userspace | |
163 | * 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. | |
165 | */ | |
166 | lgdt cpu_gdt_descr | |
167 | ||
168 | /* | |
169 | * Setup up a dummy PDA. this is just for some early bootup code | |
170 | * that does in_interrupt() | |
171 | */ | |
172 | movl $MSR_GS_BASE,%ecx | |
173 | movq $empty_zero_page,%rax | |
174 | movq %rax,%rdx | |
175 | shrq $32,%rdx | |
176 | wrmsr | |
177 | ||
178 | /* set up data segments. actually 0 would do too */ | |
179 | movl $__KERNEL_DS,%eax | |
180 | movl %eax,%ds | |
181 | movl %eax,%ss | |
182 | movl %eax,%es | |
183 | ||
184 | /* esi is pointer to real mode structure with interesting info. | |
185 | pass it to C */ | |
186 | movl %esi, %edi | |
187 | ||
188 | /* Finally jump to run C code and to be on real kernel address | |
189 | * Since we are running on identity-mapped space we have to jump | |
190 | * to the full 64bit address , this is only possible as indirect | |
191 | * jump | |
192 | */ | |
193 | movq initial_code(%rip),%rax | |
194 | jmp *%rax | |
195 | ||
e57113bc JB |
196 | /* SMP bootup changes these two */ |
197 | .align 8 | |
1da177e4 LT |
198 | .globl initial_code |
199 | initial_code: | |
200 | .quad x86_64_start_kernel | |
201 | .globl init_rsp | |
202 | init_rsp: | |
203 | .quad init_thread_union+THREAD_SIZE-8 | |
204 | ||
205 | ENTRY(early_idt_handler) | |
b957591f AK |
206 | cmpl $2,early_recursion_flag(%rip) |
207 | jz 1f | |
208 | incl early_recursion_flag(%rip) | |
1da177e4 LT |
209 | xorl %eax,%eax |
210 | movq 8(%rsp),%rsi # get rip | |
211 | movq (%rsp),%rdx | |
212 | movq %cr2,%rcx | |
213 | leaq early_idt_msg(%rip),%rdi | |
214 | call early_printk | |
b957591f AK |
215 | cmpl $2,early_recursion_flag(%rip) |
216 | jz 1f | |
217 | call dump_stack | |
6574ffd7 AK |
218 | #ifdef CONFIG_KALLSYMS |
219 | leaq early_idt_ripmsg(%rip),%rdi | |
220 | movq 8(%rsp),%rsi # get rip again | |
221 | call __print_symbol | |
222 | #endif | |
1da177e4 LT |
223 | 1: hlt |
224 | jmp 1b | |
b957591f AK |
225 | early_recursion_flag: |
226 | .long 0 | |
1da177e4 LT |
227 | |
228 | early_idt_msg: | |
229 | .asciz "PANIC: early exception rip %lx error %lx cr2 %lx\n" | |
6574ffd7 AK |
230 | early_idt_ripmsg: |
231 | .asciz "RIP %s\n" | |
1da177e4 LT |
232 | |
233 | .code32 | |
234 | ENTRY(no_long_mode) | |
235 | /* This isn't an x86-64 CPU so hang */ | |
236 | 1: | |
237 | jmp 1b | |
238 | ||
239 | .org 0xf00 | |
240 | .globl pGDT32 | |
241 | pGDT32: | |
e57113bc | 242 | .word gdt_end-cpu_gdt_table-1 |
1da177e4 LT |
243 | .long cpu_gdt_table-__START_KERNEL_map |
244 | ||
245 | .org 0xf10 | |
246 | ljumpvector: | |
247 | .long startup_64-__START_KERNEL_map | |
248 | .word __KERNEL_CS | |
249 | ||
250 | ENTRY(stext) | |
251 | ENTRY(_stext) | |
252 | ||
f0cf5d1a JB |
253 | $page = 0 |
254 | #define NEXT_PAGE(name) \ | |
255 | $page = $page + 1; \ | |
256 | .org $page * 0x1000; \ | |
257 | phys_/**/name = $page * 0x1000 + __PHYSICAL_START; \ | |
258 | ENTRY(name) | |
259 | ||
260 | NEXT_PAGE(init_level4_pgt) | |
f6c2e333 SS |
261 | /* This gets initialized in x86_64_start_kernel */ |
262 | .fill 512,8,0 | |
1da177e4 | 263 | |
f0cf5d1a JB |
264 | NEXT_PAGE(level3_ident_pgt) |
265 | .quad phys_level2_ident_pgt | 0x007 | |
1da177e4 LT |
266 | .fill 511,8,0 |
267 | ||
f0cf5d1a | 268 | NEXT_PAGE(level3_kernel_pgt) |
1da177e4 LT |
269 | .fill 510,8,0 |
270 | /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */ | |
f0cf5d1a | 271 | .quad phys_level2_kernel_pgt | 0x007 |
1da177e4 LT |
272 | .fill 1,8,0 |
273 | ||
f0cf5d1a | 274 | NEXT_PAGE(level2_ident_pgt) |
1da177e4 | 275 | /* 40MB for bootup. */ |
f0cf5d1a JB |
276 | i = 0 |
277 | .rept 20 | |
278 | .quad i << 21 | 0x083 | |
279 | i = i + 1 | |
280 | .endr | |
1da177e4 LT |
281 | /* Temporary mappings for the super early allocator in arch/x86_64/mm/init.c */ |
282 | .globl temp_boot_pmds | |
283 | temp_boot_pmds: | |
284 | .fill 492,8,0 | |
285 | ||
f0cf5d1a | 286 | NEXT_PAGE(level2_kernel_pgt) |
1da177e4 LT |
287 | /* 40MB kernel mapping. The kernel code cannot be bigger than that. |
288 | When you change this change KERNEL_TEXT_SIZE in page.h too. */ | |
289 | /* (2^48-(2*1024*1024*1024)-((2^39)*511)-((2^30)*510)) = 0 */ | |
f0cf5d1a JB |
290 | i = 0 |
291 | .rept 20 | |
292 | .quad i << 21 | 0x183 | |
293 | i = i + 1 | |
294 | .endr | |
1da177e4 LT |
295 | /* Module mapping starts here */ |
296 | .fill 492,8,0 | |
297 | ||
f0cf5d1a JB |
298 | NEXT_PAGE(level3_physmem_pgt) |
299 | .quad phys_level2_kernel_pgt | 0x007 /* so that __va works even before pagetable_init */ | |
300 | .fill 511,8,0 | |
1da177e4 | 301 | |
f0cf5d1a | 302 | #undef NEXT_PAGE |
1da177e4 | 303 | |
f0cf5d1a | 304 | .data |
1da177e4 | 305 | |
1da177e4 | 306 | #ifdef CONFIG_ACPI_SLEEP |
f0cf5d1a | 307 | .align PAGE_SIZE |
1da177e4 | 308 | ENTRY(wakeup_level4_pgt) |
f0cf5d1a | 309 | .quad phys_level3_ident_pgt | 0x007 |
1da177e4 | 310 | .fill 255,8,0 |
f0cf5d1a | 311 | .quad phys_level3_physmem_pgt | 0x007 |
1da177e4 LT |
312 | .fill 254,8,0 |
313 | /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */ | |
f0cf5d1a | 314 | .quad phys_level3_kernel_pgt | 0x007 |
1da177e4 LT |
315 | #endif |
316 | ||
f6c2e333 SS |
317 | #ifndef CONFIG_HOTPLUG_CPU |
318 | __INITDATA | |
319 | #endif | |
320 | /* | |
321 | * This default setting generates an ident mapping at address 0x100000 | |
322 | * and a mapping for the kernel that precisely maps virtual address | |
323 | * 0xffffffff80000000 to physical address 0x000000. (always using | |
324 | * 2Mbyte large pages provided by PAE mode) | |
325 | */ | |
326 | .align PAGE_SIZE | |
327 | ENTRY(boot_level4_pgt) | |
f0cf5d1a | 328 | .quad phys_level3_ident_pgt | 0x007 |
f6c2e333 | 329 | .fill 255,8,0 |
f0cf5d1a | 330 | .quad phys_level3_physmem_pgt | 0x007 |
f6c2e333 SS |
331 | .fill 254,8,0 |
332 | /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */ | |
f0cf5d1a | 333 | .quad phys_level3_kernel_pgt | 0x007 |
f6c2e333 | 334 | |
1da177e4 LT |
335 | .data |
336 | ||
337 | .align 16 | |
338 | .globl cpu_gdt_descr | |
339 | cpu_gdt_descr: | |
e57113bc | 340 | .word gdt_end-cpu_gdt_table-1 |
1da177e4 LT |
341 | gdt: |
342 | .quad cpu_gdt_table | |
343 | #ifdef CONFIG_SMP | |
344 | .rept NR_CPUS-1 | |
345 | .word 0 | |
346 | .quad 0 | |
347 | .endr | |
348 | #endif | |
349 | ||
350 | /* We need valid kernel segments for data and code in long mode too | |
351 | * IRET will check the segment types kkeil 2000/10/28 | |
352 | * Also sysret mandates a special GDT layout | |
353 | */ | |
354 | ||
e57113bc JB |
355 | .section .data.page_aligned, "aw" |
356 | .align PAGE_SIZE | |
1da177e4 LT |
357 | |
358 | /* The TLS descriptors are currently at a different place compared to i386. | |
359 | Hopefully nobody expects them at a fixed place (Wine?) */ | |
360 | ||
361 | ENTRY(cpu_gdt_table) | |
362 | .quad 0x0000000000000000 /* NULL descriptor */ | |
cdc4b9c0 | 363 | .quad 0x0 /* unused */ |
1da177e4 LT |
364 | .quad 0x00af9a000000ffff /* __KERNEL_CS */ |
365 | .quad 0x00cf92000000ffff /* __KERNEL_DS */ | |
366 | .quad 0x00cffa000000ffff /* __USER32_CS */ | |
367 | .quad 0x00cff2000000ffff /* __USER_DS, __USER32_DS */ | |
368 | .quad 0x00affa000000ffff /* __USER_CS */ | |
369 | .quad 0x00cf9a000000ffff /* __KERNEL32_CS */ | |
370 | .quad 0,0 /* TSS */ | |
371 | .quad 0,0 /* LDT */ | |
372 | .quad 0,0,0 /* three TLS descriptors */ | |
cdc4b9c0 | 373 | .quad 0 /* unused */ |
1da177e4 LT |
374 | gdt_end: |
375 | /* asm/segment.h:GDT_ENTRIES must match this */ | |
376 | /* This should be a multiple of the cache line size */ | |
c11efdf9 RT |
377 | /* GDTs of other CPUs are now dynamically allocated */ |
378 | ||
379 | /* zero the remaining page */ | |
380 | .fill PAGE_SIZE / 8 - GDT_ENTRIES,8,0 | |
1da177e4 | 381 | |
e57113bc JB |
382 | .section .bss, "aw", @nobits |
383 | .align L1_CACHE_BYTES | |
384 | ENTRY(idt_table) | |
385 | .skip 256 * 16 | |
1da177e4 | 386 | |
e57113bc JB |
387 | .section .bss.page_aligned, "aw", @nobits |
388 | .align PAGE_SIZE | |
389 | ENTRY(empty_zero_page) | |
390 | .skip PAGE_SIZE |