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>
15 #include <asm/segment.h>
16 #include <asm/pgtable.h>
19 #include <asm/cache.h>
20 #include <asm/processor-flags.h>
21 #include <asm/percpu.h>
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
29 #define GET_CR2_INTO(reg) movq %cr2, reg
30 #define INTERRUPT_RETURN iretq
33 /* we are not able to switch in one step to the final KERNEL ADDRESS SPACE
34 * because we need identity-mapped pages.
38 #define pud_index(x) (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
40 L4_PAGE_OFFSET = pgd_index(__PAGE_OFFSET)
41 L3_PAGE_OFFSET = pud_index(__PAGE_OFFSET)
42 L4_START_KERNEL = pgd_index(__START_KERNEL_map)
43 L3_START_KERNEL = pud_index(__START_KERNEL_map)
51 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
52 * and someone has loaded an identity mapped page table
53 * for us. These identity mapped page tables map all of the
54 * kernel pages and possibly all of memory.
56 * %rsi holds a physical pointer to real_mode_data.
58 * We come here either directly from a 64bit bootloader, or from
59 * arch/x86_64/boot/compressed/head.S.
61 * We only come here initially at boot nothing else comes here.
63 * Since we may be loaded at an address different from what we were
64 * compiled to run at we first fixup the physical addresses in our page
65 * tables and then reload them.
69 * Compute the delta between the address I am compiled to run at and the
70 * address I am actually running at.
72 leaq _text(%rip), %rbp
73 subq $_text - __START_KERNEL_map, %rbp
75 /* Is the address not 2M aligned? */
77 andl $~PMD_PAGE_MASK, %eax
82 * Is the address too large?
84 leaq _text(%rip), %rax
85 shrq $MAX_PHYSMEM_BITS, %rax
89 * Fixup the physical addresses in the page table
91 addq %rbp, early_level4_pgt + (L4_START_KERNEL*8)(%rip)
93 addq %rbp, level3_kernel_pgt + (510*8)(%rip)
94 addq %rbp, level3_kernel_pgt + (511*8)(%rip)
96 addq %rbp, level2_fixmap_pgt + (506*8)(%rip)
99 * Set up the identity mapping for the switchover. These
100 * entries should *NOT* have the global bit set! This also
101 * creates a bunch of nonsense entries but that is fine --
102 * it avoids problems around wraparound.
104 leaq _text(%rip), %rdi
105 leaq early_level4_pgt(%rip), %rbx
108 shrq $PGDIR_SHIFT, %rax
110 leaq (4096 + _KERNPG_TABLE)(%rbx), %rdx
111 movq %rdx, 0(%rbx,%rax,8)
112 movq %rdx, 8(%rbx,%rax,8)
116 shrq $PUD_SHIFT, %rax
117 andl $(PTRS_PER_PUD-1), %eax
118 movq %rdx, (4096+0)(%rbx,%rax,8)
119 movq %rdx, (4096+8)(%rbx,%rax,8)
123 shrq $PMD_SHIFT, %rdi
124 addq $(__PAGE_KERNEL_LARGE_EXEC & ~_PAGE_GLOBAL), %rax
125 leaq (_end - 1)(%rip), %rcx
126 shrq $PMD_SHIFT, %rcx
131 andq $(PTRS_PER_PMD - 1), %rdi
132 movq %rax, (%rbx,%rdi,8)
139 * Fixup the kernel text+data virtual addresses. Note that
140 * we might write invalid pmds, when the kernel is relocated
141 * cleanup_highmap() fixes this up along with the mappings
144 leaq level2_kernel_pgt(%rip), %rdi
146 /* See if it is a valid page table entry */
150 /* Go to the next page */
155 /* Fixup phys_base */
156 addq %rbp, phys_base(%rip)
158 movq $(early_level4_pgt - __START_KERNEL_map), %rax
160 ENTRY(secondary_startup_64)
162 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
163 * and someone has loaded a mapped page table.
165 * %rsi holds a physical pointer to real_mode_data.
167 * We come here either from startup_64 (using physical addresses)
168 * or from trampoline.S (using virtual addresses).
170 * Using virtual addresses from trampoline.S removes the need
171 * to have any identity mapped pages in the kernel page table
172 * after the boot processor executes this code.
175 movq $(init_level4_pgt - __START_KERNEL_map), %rax
178 /* Enable PAE mode and PGE */
179 movl $(X86_CR4_PAE | X86_CR4_PGE), %ecx
182 /* Setup early boot stage 4 level pagetables. */
183 addq phys_base(%rip), %rax
186 /* Ensure I am executing from virtual addresses */
191 /* Check if nx is implemented */
192 movl $0x80000001, %eax
196 /* Setup EFER (Extended Feature Enable Register) */
199 btsl $_EFER_SCE, %eax /* Enable System Call */
200 btl $20,%edi /* No Execute supported? */
203 1: wrmsr /* Make changes effective */
206 #define CR0_STATE (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
207 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
209 movl $CR0_STATE, %eax
210 /* Make changes effective */
213 /* Setup a boot time stack */
214 movq stack_start(%rip), %rsp
216 /* zero EFLAGS after setting rsp */
221 * We must switch to a new descriptor in kernel space for the GDT
222 * because soon the kernel won't have access anymore to the userspace
223 * addresses where we're currently running on. We have to do that here
224 * because in 32bit we couldn't load a 64bit linear address.
226 lgdt early_gdt_descr(%rip)
228 /* set up data segments */
235 * We don't really need to load %fs or %gs, but load them anyway
236 * to kill any stale realmode selectors. This allows execution
244 * The base of %gs always points to the bottom of the irqstack
245 * union. If the stack protector canary is enabled, it is
246 * located at %gs:40. Note that, on SMP, the boot cpu uses
247 * init data section till per cpu areas are set up.
249 movl $MSR_GS_BASE,%ecx
250 movl initial_gs(%rip),%eax
251 movl initial_gs+4(%rip),%edx
254 /* rsi is pointer to real mode structure with interesting info.
258 /* Finally jump to run C code and to be on real kernel address
259 * Since we are running on identity-mapped space we have to jump
260 * to the full 64bit address, this is only possible as indirect
261 * jump. In addition we need to ensure %cs is set so we make this
264 * Note: do not change to far jump indirect with 64bit offset.
266 * AMD does not support far jump indirect with 64bit offset.
267 * AMD64 Architecture Programmer's Manual, Volume 3: states only
268 * JMP FAR mem16:16 FF /5 Far jump indirect,
269 * with the target specified by a far pointer in memory.
270 * JMP FAR mem16:32 FF /5 Far jump indirect,
271 * with the target specified by a far pointer in memory.
273 * Intel64 does support 64bit offset.
274 * Software Developer Manual Vol 2: states:
275 * FF /5 JMP m16:16 Jump far, absolute indirect,
276 * address given in m16:16
277 * FF /5 JMP m16:32 Jump far, absolute indirect,
278 * address given in m16:32.
279 * REX.W + FF /5 JMP m16:64 Jump far, absolute indirect,
280 * address given in m16:64.
282 movq initial_code(%rip),%rax
283 pushq $0 # fake return address to stop unwinder
284 pushq $__KERNEL_CS # set correct cs
285 pushq %rax # target address in negative space
288 #ifdef CONFIG_HOTPLUG_CPU
290 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
291 * up already except stack. We just set up stack here. Then call
295 movq stack_start(%rip),%rsp
296 movq initial_code(%rip),%rax
297 pushq $0 # fake return address to stop unwinder
298 pushq $__KERNEL_CS # set correct cs
299 pushq %rax # target address in negative space
304 /* SMP bootup changes these two */
308 .quad x86_64_start_kernel
310 .quad INIT_PER_CPU_VAR(irq_stack_union)
313 .quad init_thread_union+THREAD_SIZE-8
321 .globl early_idt_handlers
326 # 80(%rsp) error code
328 .rept NUM_EXCEPTION_VECTORS
329 .if (EXCEPTION_ERRCODE_MASK >> i) & 1
332 pushq $0 # Dummy error code, to make stack frame uniform
334 pushq $i # 72(%rsp) Vector number
335 jmp early_idt_handler
339 /* This is global to keep gas from relaxing the jumps */
340 ENTRY(early_idt_handler)
343 cmpl $2,early_recursion_flag(%rip)
345 incl early_recursion_flag(%rip)
347 pushq %rax # 64(%rsp)
348 pushq %rcx # 56(%rsp)
349 pushq %rdx # 48(%rsp)
350 pushq %rsi # 40(%rsp)
351 pushq %rdi # 32(%rsp)
357 cmpl $__KERNEL_CS,96(%rsp)
360 cmpl $14,72(%rsp) # Page fault?
362 GET_CR2_INTO(%rdi) # can clobber any volatile register if pv
363 call early_make_pgtable
368 leaq 88(%rsp),%rdi # Pointer to %rip
369 call early_fixup_exception
371 jnz 20f # Found an exception entry
374 #ifdef CONFIG_EARLY_PRINTK
375 GET_CR2_INTO(%r9) # can clobber any volatile register if pv
376 movl 80(%rsp),%r8d # error code
377 movl 72(%rsp),%esi # vector number
378 movl 96(%rsp),%edx # %cs
379 movq 88(%rsp),%rcx # %rip
381 leaq early_idt_msg(%rip),%rdi
383 cmpl $2,early_recursion_flag(%rip)
386 #ifdef CONFIG_KALLSYMS
387 leaq early_idt_ripmsg(%rip),%rdi
388 movq 40(%rsp),%rsi # %rip again
391 #endif /* EARLY_PRINTK */
395 20: # Exception table entry found or page table generated
405 addq $16,%rsp # drop vector number and error code
406 decl early_recursion_flag(%rip)
408 ENDPROC(early_idt_handler)
413 early_recursion_flag:
416 #ifdef CONFIG_EARLY_PRINTK
418 .asciz "PANIC: early exception %02lx rip %lx:%lx error %lx cr2 %lx\n"
421 #endif /* CONFIG_EARLY_PRINTK */
423 #define NEXT_PAGE(name) \
427 /* Automate the creation of 1 to 1 mapping pmd entries */
428 #define PMDS(START, PERM, COUNT) \
431 .quad (START) + (i << PMD_SHIFT) + (PERM) ; \
436 NEXT_PAGE(early_level4_pgt)
438 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
440 NEXT_PAGE(early_dynamic_pgts)
441 .fill 512*EARLY_DYNAMIC_PAGE_TABLES,8,0
446 NEXT_PAGE(init_level4_pgt)
449 NEXT_PAGE(init_level4_pgt)
450 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
451 .org init_level4_pgt + L4_PAGE_OFFSET*8, 0
452 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
453 .org init_level4_pgt + L4_START_KERNEL*8, 0
454 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
455 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
457 NEXT_PAGE(level3_ident_pgt)
458 .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
460 NEXT_PAGE(level2_ident_pgt)
461 /* Since I easily can, map the first 1G.
462 * Don't set NX because code runs from these pages.
464 PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
467 NEXT_PAGE(level3_kernel_pgt)
468 .fill L3_START_KERNEL,8,0
469 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
470 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE
471 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
473 NEXT_PAGE(level2_kernel_pgt)
475 * 512 MB kernel mapping. We spend a full page on this pagetable
478 * The kernel code+data+bss must not be bigger than that.
480 * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
481 * If you want to increase this then increase MODULES_VADDR
484 PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
485 KERNEL_IMAGE_SIZE/PMD_SIZE)
487 NEXT_PAGE(level2_fixmap_pgt)
489 .quad level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
490 /* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */
493 NEXT_PAGE(level1_fixmap_pgt)
500 .globl early_gdt_descr
502 .word GDT_ENTRIES*8-1
503 early_gdt_descr_base:
504 .quad INIT_PER_CPU_VAR(gdt_page)
507 /* This must match the first entry in level2_kernel_pgt */
508 .quad 0x0000000000000000
510 #include "../../x86/xen/xen-head.S"
512 .section .bss, "aw", @nobits
513 .align L1_CACHE_BYTES
515 .skip IDT_ENTRIES * 16
517 .align L1_CACHE_BYTES
519 .skip IDT_ENTRIES * 16
522 NEXT_PAGE(empty_zero_page)