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>
21 #include <asm/processor-flags.h>
23 #ifdef CONFIG_PARAVIRT
24 #include <asm/asm-offsets.h>
25 #include <asm/paravirt.h>
27 #define GET_CR2_INTO_RCX movq %cr2, %rcx
30 /* we are not able to switch in one step to the final KERNEL ADRESS SPACE
31 * because we need identity-mapped pages.
42 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
43 * and someone has loaded an identity mapped page table
44 * for us. These identity mapped page tables map all of the
45 * kernel pages and possibly all of memory.
47 * %esi holds a physical pointer to real_mode_data.
49 * We come here either directly from a 64bit bootloader, or from
50 * arch/x86_64/boot/compressed/head.S.
52 * We only come here initially at boot nothing else comes here.
54 * Since we may be loaded at an address different from what we were
55 * compiled to run at we first fixup the physical addresses in our page
56 * tables and then reload them.
59 /* Compute the delta between the address I am compiled to run at and the
60 * address I am actually running at.
62 leaq _text(%rip), %rbp
63 subq $_text - __START_KERNEL_map, %rbp
65 /* Is the address not 2M aligned? */
67 andl $~PMD_PAGE_MASK, %eax
71 /* Is the address too large? */
72 leaq _text(%rip), %rdx
73 movq $PGDIR_SIZE, %rax
77 /* Fixup the physical addresses in the page table
79 addq %rbp, init_level4_pgt + 0(%rip)
80 addq %rbp, init_level4_pgt + (258*8)(%rip)
81 addq %rbp, init_level4_pgt + (511*8)(%rip)
83 addq %rbp, level3_ident_pgt + 0(%rip)
85 addq %rbp, level3_kernel_pgt + (510*8)(%rip)
86 addq %rbp, level3_kernel_pgt + (511*8)(%rip)
88 addq %rbp, level2_fixmap_pgt + (506*8)(%rip)
90 /* Add an Identity mapping if I am above 1G */
91 leaq _text(%rip), %rdi
92 andq $PMD_PAGE_MASK, %rdi
96 andq $(PTRS_PER_PUD - 1), %rax
99 leaq (level2_spare_pgt - __START_KERNEL_map + _KERNPG_TABLE)(%rbp), %rdx
100 leaq level3_ident_pgt(%rip), %rbx
101 movq %rdx, 0(%rbx, %rax, 8)
104 shrq $PMD_SHIFT, %rax
105 andq $(PTRS_PER_PMD - 1), %rax
106 leaq __PAGE_KERNEL_LARGE_EXEC(%rdi), %rdx
107 leaq level2_spare_pgt(%rip), %rbx
108 movq %rdx, 0(%rbx, %rax, 8)
112 * Fixup the kernel text+data virtual addresses. Note that
113 * we might write invalid pmds, when the kernel is relocated
114 * cleanup_highmap() fixes this up along with the mappings
118 leaq level2_kernel_pgt(%rip), %rdi
120 /* See if it is a valid page table entry */
124 /* Go to the next page */
129 /* Fixup phys_base */
130 addq %rbp, phys_base(%rip)
133 addq %rbp, trampoline_level4_pgt + 0(%rip)
134 addq %rbp, trampoline_level4_pgt + (511*8)(%rip)
137 /* Due to ENTRY(), sometimes the empty space gets filled with
138 * zeros. Better take a jmp than relying on empty space being
139 * filled with 0x90 (nop)
141 jmp secondary_startup_64
142 ENTRY(secondary_startup_64)
144 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
145 * and someone has loaded a mapped page table.
147 * %esi holds a physical pointer to real_mode_data.
149 * We come here either from startup_64 (using physical addresses)
150 * or from trampoline.S (using virtual addresses).
152 * Using virtual addresses from trampoline.S removes the need
153 * to have any identity mapped pages in the kernel page table
154 * after the boot processor executes this code.
157 /* Enable PAE mode and PGE */
158 movl $(X86_CR4_PAE | X86_CR4_PGE), %eax
161 /* Setup early boot stage 4 level pagetables. */
162 movq $(init_level4_pgt - __START_KERNEL_map), %rax
163 addq phys_base(%rip), %rax
166 /* Ensure I am executing from virtual addresses */
171 /* Check if nx is implemented */
172 movl $0x80000001, %eax
176 /* Setup EFER (Extended Feature Enable Register) */
179 btsl $_EFER_SCE, %eax /* Enable System Call */
180 btl $20,%edi /* No Execute supported? */
183 1: wrmsr /* Make changes effective */
186 #define CR0_STATE (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
187 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
189 movl $CR0_STATE, %eax
190 /* Make changes effective */
193 /* Setup a boot time stack */
194 movq init_rsp(%rip),%rsp
196 /* zero EFLAGS after setting rsp */
201 * We must switch to a new descriptor in kernel space for the GDT
202 * because soon the kernel won't have access anymore to the userspace
203 * addresses where we're currently running on. We have to do that here
204 * because in 32bit we couldn't load a 64bit linear address.
206 lgdt cpu_gdt_descr(%rip)
208 /* set up data segments. actually 0 would do too */
209 movl $__KERNEL_DS,%eax
215 * We don't really need to load %fs or %gs, but load them anyway
216 * to kill any stale realmode selectors. This allows execution
223 * Setup up a dummy PDA. this is just for some early bootup code
224 * that does in_interrupt()
226 movl $MSR_GS_BASE,%ecx
227 movq $empty_zero_page,%rax
232 /* esi is pointer to real mode structure with interesting info.
236 /* Finally jump to run C code and to be on real kernel address
237 * Since we are running on identity-mapped space we have to jump
238 * to the full 64bit address, this is only possible as indirect
239 * jump. In addition we need to ensure %cs is set so we make this
242 movq initial_code(%rip),%rax
243 pushq $0 # fake return address to stop unwinder
244 pushq $__KERNEL_CS # set correct cs
245 pushq %rax # target address in negative space
248 /* SMP bootup changes these two */
252 .quad x86_64_start_kernel
256 .quad init_thread_union+THREAD_SIZE-8
261 .section ".init.text","ax"
262 #ifdef CONFIG_EARLY_PRINTK
263 .globl early_idt_handlers
266 .rept NUM_EXCEPTION_VECTORS
268 jmp early_idt_handler
273 ENTRY(early_idt_handler)
274 #ifdef CONFIG_EARLY_PRINTK
275 cmpl $2,early_recursion_flag(%rip)
277 incl early_recursion_flag(%rip)
280 xorl %r8d,%r8d # zero for error code
281 movl %esi,%ecx # get vector number
282 # Test %ecx against mask of vectors that push error code.
289 popq %r8 # get error code
290 0: movq 0(%rsp),%rcx # get ip
291 movq 8(%rsp),%rdx # get cs
293 leaq early_idt_msg(%rip),%rdi
295 cmpl $2,early_recursion_flag(%rip)
298 #ifdef CONFIG_KALLSYMS
299 leaq early_idt_ripmsg(%rip),%rdi
300 movq 8(%rsp),%rsi # get rip again
303 #endif /* EARLY_PRINTK */
307 #ifdef CONFIG_EARLY_PRINTK
308 early_recursion_flag:
312 .asciz "PANIC: early exception %02lx rip %lx:%lx error %lx cr2 %lx\n"
315 #endif /* CONFIG_EARLY_PRINTK */
320 #define NEXT_PAGE(name) \
324 /* Automate the creation of 1 to 1 mapping pmd entries */
325 #define PMDS(START, PERM, COUNT) \
328 .quad (START) + (i << PMD_SHIFT) + (PERM) ; \
333 * This default setting generates an ident mapping at address 0x100000
334 * and a mapping for the kernel that precisely maps virtual address
335 * 0xffffffff80000000 to physical address 0x000000. (always using
336 * 2Mbyte large pages provided by PAE mode)
338 NEXT_PAGE(init_level4_pgt)
339 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
341 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
343 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
344 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
346 NEXT_PAGE(level3_ident_pgt)
347 .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
350 NEXT_PAGE(level3_kernel_pgt)
352 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
353 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE
354 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
356 NEXT_PAGE(level2_fixmap_pgt)
358 .quad level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
359 /* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */
362 NEXT_PAGE(level1_fixmap_pgt)
365 NEXT_PAGE(level2_ident_pgt)
366 /* Since I easily can, map the first 1G.
367 * Don't set NX because code runs from these pages.
369 PMDS(0, __PAGE_KERNEL_LARGE_EXEC, PTRS_PER_PMD)
371 NEXT_PAGE(level2_kernel_pgt)
373 * 512 MB kernel mapping. We spend a full page on this pagetable
376 * The kernel code+data+bss must not be bigger than that.
378 * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
379 * If you want to increase this then increase MODULES_VADDR
382 PMDS(0, __PAGE_KERNEL_LARGE_EXEC|_PAGE_GLOBAL,
383 KERNEL_IMAGE_SIZE/PMD_SIZE)
385 NEXT_PAGE(level2_spare_pgt)
395 .word gdt_end-cpu_gdt_table-1
406 /* This must match the first entry in level2_kernel_pgt */
407 .quad 0x0000000000000000
409 /* We need valid kernel segments for data and code in long mode too
410 * IRET will check the segment types kkeil 2000/10/28
411 * Also sysret mandates a special GDT layout
414 .section .data.page_aligned, "aw"
417 /* The TLS descriptors are currently at a different place compared to i386.
418 Hopefully nobody expects them at a fixed place (Wine?) */
421 .quad 0x0000000000000000 /* NULL descriptor */
422 .quad 0x00cf9b000000ffff /* __KERNEL32_CS */
423 .quad 0x00af9b000000ffff /* __KERNEL_CS */
424 .quad 0x00cf93000000ffff /* __KERNEL_DS */
425 .quad 0x00cffb000000ffff /* __USER32_CS */
426 .quad 0x00cff3000000ffff /* __USER_DS, __USER32_DS */
427 .quad 0x00affb000000ffff /* __USER_CS */
428 .quad 0x0 /* unused */
431 .quad 0,0,0 /* three TLS descriptors */
432 .quad 0x0000f40000000000 /* node/CPU stored in limit */
434 /* asm/segment.h:GDT_ENTRIES must match this */
435 /* This should be a multiple of the cache line size */
436 /* GDTs of other CPUs are now dynamically allocated */
438 /* zero the remaining page */
439 .fill PAGE_SIZE / 8 - GDT_ENTRIES,8,0
441 .section .bss, "aw", @nobits
442 .align L1_CACHE_BYTES
446 .section .bss.page_aligned, "aw", @nobits
448 ENTRY(empty_zero_page)