2 * linux/arch/x86_64/kernel/head.S -- start in 32bit and switch to 64bit
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>
12 #include <linux/linkage.h>
13 #include <linux/threads.h>
14 #include <linux/init.h>
16 #include <asm/segment.h>
17 #include <asm/pgtable.h>
20 #include <asm/cache.h>
22 /* we are not able to switch in one step to the final KERNEL ADRESS SPACE
23 * because we need identity-mapped pages.
34 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
35 * and someone has loaded an identity mapped page table
36 * for us. These identity mapped page tables map all of the
37 * kernel pages and possibly all of memory.
39 * %esi holds a physical pointer to real_mode_data.
41 * We come here either directly from a 64bit bootloader, or from
42 * arch/x86_64/boot/compressed/head.S.
44 * We only come here initially at boot nothing else comes here.
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.
51 /* Compute the delta between the address I am compiled to run at and the
52 * address I am actually running at.
54 leaq _text(%rip), %rbp
55 subq $_text - __START_KERNEL_map, %rbp
57 /* Is the address not 2M aligned? */
59 andl $~LARGE_PAGE_MASK, %eax
63 /* Is the address too large? */
64 leaq _text(%rip), %rdx
65 movq $PGDIR_SIZE, %rax
69 /* Fixup the physical addresses in the page table
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)
75 addq %rbp, level3_ident_pgt + 0(%rip)
77 addq %rbp, level3_kernel_pgt + (510*8)(%rip)
78 addq %rbp, level3_kernel_pgt + (511*8)(%rip)
80 addq %rbp, level2_fixmap_pgt + (506*8)(%rip)
82 /* Add an Identity mapping if I am above 1G */
83 leaq _text(%rip), %rdi
84 andq $LARGE_PAGE_MASK, %rdi
88 andq $(PTRS_PER_PUD - 1), %rax
91 leaq (level2_spare_pgt - __START_KERNEL_map + _KERNPG_TABLE)(%rbp), %rdx
92 leaq level3_ident_pgt(%rip), %rbx
93 movq %rdx, 0(%rbx, %rax, 8)
97 andq $(PTRS_PER_PMD - 1), %rax
98 leaq __PAGE_KERNEL_LARGE_EXEC(%rdi), %rdx
99 leaq level2_spare_pgt(%rip), %rbx
100 movq %rdx, 0(%rbx, %rax, 8)
103 /* Fixup the kernel text+data virtual addresses
105 leaq level2_kernel_pgt(%rip), %rdi
107 /* See if it is a valid page table entry */
111 /* Go to the next page */
116 /* Fixup phys_base */
117 addq %rbp, phys_base(%rip)
120 addq %rbp, trampoline_level4_pgt + 0(%rip)
121 addq %rbp, trampoline_level4_pgt + (511*8)(%rip)
123 #ifdef CONFIG_ACPI_SLEEP
124 addq %rbp, wakeup_level4_pgt + 0(%rip)
125 addq %rbp, wakeup_level4_pgt + (511*8)(%rip)
128 /* Due to ENTRY(), sometimes the empty space gets filled with
129 * zeros. Better take a jmp than relying on empty space being
130 * filled with 0x90 (nop)
132 jmp secondary_startup_64
133 ENTRY(secondary_startup_64)
135 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
136 * and someone has loaded a mapped page table.
138 * %esi holds a physical pointer to real_mode_data.
140 * We come here either from startup_64 (using physical addresses)
141 * or from trampoline.S (using virtual addresses).
143 * Using virtual addresses from trampoline.S removes the need
144 * to have any identity mapped pages in the kernel page table
145 * after the boot processor executes this code.
148 /* Enable PAE mode and PGE */
154 /* Setup early boot stage 4 level pagetables. */
155 movq $(init_level4_pgt - __START_KERNEL_map), %rax
156 addq phys_base(%rip), %rax
159 /* Ensure I am executing from virtual addresses */
164 /* Check if nx is implemented */
165 movl $0x80000001, %eax
169 /* Setup EFER (Extended Feature Enable Register) */
172 btsl $_EFER_SCE, %eax /* Enable System Call */
173 btl $20,%edi /* No Execute supported? */
176 1: wrmsr /* Make changes effective */
179 #define CR0_PM 1 /* protected mode */
180 #define CR0_MP (1<<1)
181 #define CR0_ET (1<<4)
182 #define CR0_NE (1<<5)
183 #define CR0_WP (1<<16)
184 #define CR0_AM (1<<18)
185 #define CR0_PAGING (1<<31)
186 movl $CR0_PM|CR0_MP|CR0_ET|CR0_NE|CR0_WP|CR0_AM|CR0_PAGING,%eax
187 /* Make changes effective */
190 /* Setup a boot time stack */
191 movq init_rsp(%rip),%rsp
193 /* zero EFLAGS after setting rsp */
198 * We must switch to a new descriptor in kernel space for the GDT
199 * because soon the kernel won't have access anymore to the userspace
200 * addresses where we're currently running on. We have to do that here
201 * because in 32bit we couldn't load a 64bit linear address.
203 lgdt cpu_gdt_descr(%rip)
205 /* set up data segments. actually 0 would do too */
206 movl $__KERNEL_DS,%eax
212 * We don't really need to load %fs or %gs, but load them anyway
213 * to kill any stale realmode selectors. This allows execution
220 * Setup up a dummy PDA. this is just for some early bootup code
221 * that does in_interrupt()
223 movl $MSR_GS_BASE,%ecx
224 movq $empty_zero_page,%rax
229 /* esi is pointer to real mode structure with interesting info.
233 /* Finally jump to run C code and to be on real kernel address
234 * Since we are running on identity-mapped space we have to jump
235 * to the full 64bit address, this is only possible as indirect
236 * jump. In addition we need to ensure %cs is set so we make this
239 movq initial_code(%rip),%rax
240 pushq $0 # fake return address to stop unwinder
241 pushq $__KERNEL_CS # set correct cs
242 pushq %rax # target address in negative space
245 /* SMP bootup changes these two */
246 #ifndef CONFIG_HOTPLUG_CPU
247 .pushsection .init.data
252 .quad x86_64_start_kernel
253 #ifndef CONFIG_HOTPLUG_CPU
258 .quad init_thread_union+THREAD_SIZE-8
263 ENTRY(early_idt_handler)
264 cmpl $2,early_recursion_flag(%rip)
266 incl early_recursion_flag(%rip)
268 movq 8(%rsp),%rsi # get rip
271 leaq early_idt_msg(%rip),%rdi
273 cmpl $2,early_recursion_flag(%rip)
276 #ifdef CONFIG_KALLSYMS
277 leaq early_idt_ripmsg(%rip),%rdi
278 movq 8(%rsp),%rsi # get rip again
283 early_recursion_flag:
287 .asciz "PANIC: early exception rip %lx error %lx cr2 %lx\n"
293 #define NEXT_PAGE(name) \
297 /* Automate the creation of 1 to 1 mapping pmd entries */
298 #define PMDS(START, PERM, COUNT) \
301 .quad (START) + (i << 21) + (PERM) ; \
306 * This default setting generates an ident mapping at address 0x100000
307 * and a mapping for the kernel that precisely maps virtual address
308 * 0xffffffff80000000 to physical address 0x000000. (always using
309 * 2Mbyte large pages provided by PAE mode)
311 NEXT_PAGE(init_level4_pgt)
312 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
314 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
316 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
317 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
319 NEXT_PAGE(level3_ident_pgt)
320 .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
323 NEXT_PAGE(level3_kernel_pgt)
325 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
326 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE
327 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
329 NEXT_PAGE(level2_fixmap_pgt)
331 .quad level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
332 /* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */
335 NEXT_PAGE(level1_fixmap_pgt)
338 NEXT_PAGE(level2_ident_pgt)
339 /* Since I easily can, map the first 1G.
340 * Don't set NX because code runs from these pages.
342 PMDS(0x0000000000000000, __PAGE_KERNEL_LARGE_EXEC, PTRS_PER_PMD)
344 NEXT_PAGE(level2_kernel_pgt)
345 /* 40MB kernel mapping. The kernel code cannot be bigger than that.
346 When you change this change KERNEL_TEXT_SIZE in page.h too. */
347 /* (2^48-(2*1024*1024*1024)-((2^39)*511)-((2^30)*510)) = 0 */
348 PMDS(0x0000000000000000, __PAGE_KERNEL_LARGE_EXEC|_PAGE_GLOBAL, KERNEL_TEXT_SIZE/PMD_SIZE)
349 /* Module mapping starts here */
350 .fill (PTRS_PER_PMD - (KERNEL_TEXT_SIZE/PMD_SIZE)),8,0
352 NEXT_PAGE(level2_spare_pgt)
362 .word gdt_end-cpu_gdt_table-1
373 /* This must match the first entry in level2_kernel_pgt */
374 .quad 0x0000000000000000
376 /* We need valid kernel segments for data and code in long mode too
377 * IRET will check the segment types kkeil 2000/10/28
378 * Also sysret mandates a special GDT layout
381 .section .data.page_aligned, "aw"
384 /* The TLS descriptors are currently at a different place compared to i386.
385 Hopefully nobody expects them at a fixed place (Wine?) */
388 .quad 0x0000000000000000 /* NULL descriptor */
389 .quad 0x00cf9b000000ffff /* __KERNEL32_CS */
390 .quad 0x00af9b000000ffff /* __KERNEL_CS */
391 .quad 0x00cf93000000ffff /* __KERNEL_DS */
392 .quad 0x00cffb000000ffff /* __USER32_CS */
393 .quad 0x00cff3000000ffff /* __USER_DS, __USER32_DS */
394 .quad 0x00affb000000ffff /* __USER_CS */
395 .quad 0x0 /* unused */
398 .quad 0,0,0 /* three TLS descriptors */
399 .quad 0x0000f40000000000 /* node/CPU stored in limit */
401 /* asm/segment.h:GDT_ENTRIES must match this */
402 /* This should be a multiple of the cache line size */
403 /* GDTs of other CPUs are now dynamically allocated */
405 /* zero the remaining page */
406 .fill PAGE_SIZE / 8 - GDT_ENTRIES,8,0
408 .section .bss, "aw", @nobits
409 .align L1_CACHE_BYTES
413 .section .bss.page_aligned, "aw", @nobits
415 ENTRY(empty_zero_page)