xref: /openbmc/linux/arch/x86/kernel/head_64.S (revision b12d8db8) 
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 *  Copyright (C) 2005 Eric Biederman <ebiederm@xmission.com>
9 */
10
11
12#include <linux/linkage.h>
13#include <linux/threads.h>
14#include <linux/init.h>
15#include <asm/desc.h>
16#include <asm/segment.h>
17#include <asm/pgtable.h>
18#include <asm/page.h>
19#include <asm/msr.h>
20#include <asm/cache.h>
21#include <asm/processor-flags.h>
22#include <asm/percpu.h>
23
24#ifdef CONFIG_PARAVIRT
25#include <asm/asm-offsets.h>
26#include <asm/paravirt.h>
27#else
28#define GET_CR2_INTO_RCX movq %cr2, %rcx
29#endif
30
31/* we are not able to switch in one step to the final KERNEL ADRESS SPACE
32 * because we need identity-mapped pages.
33 *
34 */
35
36#define pud_index(x)	(((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
37
38L4_PAGE_OFFSET = pgd_index(__PAGE_OFFSET)
39L3_PAGE_OFFSET = pud_index(__PAGE_OFFSET)
40L4_START_KERNEL = pgd_index(__START_KERNEL_map)
41L3_START_KERNEL = pud_index(__START_KERNEL_map)
42
43	.text
44	.section .text.head
45	.code64
46	.globl startup_64
47startup_64:
48
49	/*
50	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
51	 * and someone has loaded an identity mapped page table
52	 * for us.  These identity mapped page tables map all of the
53	 * kernel pages and possibly all of memory.
54	 *
55	 * %esi holds a physical pointer to real_mode_data.
56	 *
57	 * We come here either directly from a 64bit bootloader, or from
58	 * arch/x86_64/boot/compressed/head.S.
59	 *
60	 * We only come here initially at boot nothing else comes here.
61	 *
62	 * Since we may be loaded at an address different from what we were
63	 * compiled to run at we first fixup the physical addresses in our page
64	 * tables and then reload them.
65	 */
66
67	/* Compute the delta between the address I am compiled to run at and the
68	 * address I am actually running at.
69	 */
70	leaq	_text(%rip), %rbp
71	subq	$_text - __START_KERNEL_map, %rbp
72
73	/* Is the address not 2M aligned? */
74	movq	%rbp, %rax
75	andl	$~PMD_PAGE_MASK, %eax
76	testl	%eax, %eax
77	jnz	bad_address
78
79	/* Is the address too large? */
80	leaq	_text(%rip), %rdx
81	movq	$PGDIR_SIZE, %rax
82	cmpq	%rax, %rdx
83	jae	bad_address
84
85	/* Fixup the physical addresses in the page table
86	 */
87	addq	%rbp, init_level4_pgt + 0(%rip)
88	addq	%rbp, init_level4_pgt + (L4_PAGE_OFFSET*8)(%rip)
89	addq	%rbp, init_level4_pgt + (L4_START_KERNEL*8)(%rip)
90
91	addq	%rbp, level3_ident_pgt + 0(%rip)
92
93	addq	%rbp, level3_kernel_pgt + (510*8)(%rip)
94	addq	%rbp, level3_kernel_pgt + (511*8)(%rip)
95
96	addq	%rbp, level2_fixmap_pgt + (506*8)(%rip)
97
98	/* Add an Identity mapping if I am above 1G */
99	leaq	_text(%rip), %rdi
100	andq	$PMD_PAGE_MASK, %rdi
101
102	movq	%rdi, %rax
103	shrq	$PUD_SHIFT, %rax
104	andq	$(PTRS_PER_PUD - 1), %rax
105	jz	ident_complete
106
107	leaq	(level2_spare_pgt - __START_KERNEL_map + _KERNPG_TABLE)(%rbp), %rdx
108	leaq	level3_ident_pgt(%rip), %rbx
109	movq	%rdx, 0(%rbx, %rax, 8)
110
111	movq	%rdi, %rax
112	shrq	$PMD_SHIFT, %rax
113	andq	$(PTRS_PER_PMD - 1), %rax
114	leaq	__PAGE_KERNEL_IDENT_LARGE_EXEC(%rdi), %rdx
115	leaq	level2_spare_pgt(%rip), %rbx
116	movq	%rdx, 0(%rbx, %rax, 8)
117ident_complete:
118
119	/*
120	 * Fixup the kernel text+data virtual addresses. Note that
121	 * we might write invalid pmds, when the kernel is relocated
122	 * cleanup_highmap() fixes this up along with the mappings
123	 * beyond _end.
124	 */
125
126	leaq	level2_kernel_pgt(%rip), %rdi
127	leaq	4096(%rdi), %r8
128	/* See if it is a valid page table entry */
1291:	testq	$1, 0(%rdi)
130	jz	2f
131	addq	%rbp, 0(%rdi)
132	/* Go to the next page */
1332:	addq	$8, %rdi
134	cmp	%r8, %rdi
135	jne	1b
136
137	/* Fixup phys_base */
138	addq	%rbp, phys_base(%rip)
139
140#ifdef CONFIG_X86_TRAMPOLINE
141	addq	%rbp, trampoline_level4_pgt + 0(%rip)
142	addq	%rbp, trampoline_level4_pgt + (511*8)(%rip)
143#endif
144
145	/* Due to ENTRY(), sometimes the empty space gets filled with
146	 * zeros. Better take a jmp than relying on empty space being
147	 * filled with 0x90 (nop)
148	 */
149	jmp secondary_startup_64
150ENTRY(secondary_startup_64)
151	/*
152	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
153	 * and someone has loaded a mapped page table.
154	 *
155	 * %esi holds a physical pointer to real_mode_data.
156	 *
157	 * We come here either from startup_64 (using physical addresses)
158	 * or from trampoline.S (using virtual addresses).
159	 *
160	 * Using virtual addresses from trampoline.S removes the need
161	 * to have any identity mapped pages in the kernel page table
162	 * after the boot processor executes this code.
163	 */
164
165	/* Enable PAE mode and PGE */
166	movl	$(X86_CR4_PAE | X86_CR4_PGE), %eax
167	movq	%rax, %cr4
168
169	/* Setup early boot stage 4 level pagetables. */
170	movq	$(init_level4_pgt - __START_KERNEL_map), %rax
171	addq	phys_base(%rip), %rax
172	movq	%rax, %cr3
173
174	/* Ensure I am executing from virtual addresses */
175	movq	$1f, %rax
176	jmp	*%rax
1771:
178
179	/* Check if nx is implemented */
180	movl	$0x80000001, %eax
181	cpuid
182	movl	%edx,%edi
183
184	/* Setup EFER (Extended Feature Enable Register) */
185	movl	$MSR_EFER, %ecx
186	rdmsr
187	btsl	$_EFER_SCE, %eax	/* Enable System Call */
188	btl	$20,%edi		/* No Execute supported? */
189	jnc     1f
190	btsl	$_EFER_NX, %eax
1911:	wrmsr				/* Make changes effective */
192
193	/* Setup cr0 */
194#define CR0_STATE	(X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
195			 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
196			 X86_CR0_PG)
197	movl	$CR0_STATE, %eax
198	/* Make changes effective */
199	movq	%rax, %cr0
200
201	/* Setup a boot time stack */
202	movq stack_start(%rip),%rsp
203
204	/* zero EFLAGS after setting rsp */
205	pushq $0
206	popfq
207
208#ifdef CONFIG_SMP
209	/*
210	 * early_gdt_base should point to the gdt_page in static percpu init
211	 * data area.  Computing this requires two symbols - __per_cpu_load
212	 * and per_cpu__gdt_page.  As linker can't do no such relocation, do
213	 * it by hand.  As early_gdt_descr is manipulated by C code for
214	 * secondary CPUs, this should be done only once for the boot CPU
215	 * when early_gdt_descr_base contains zero.
216	 */
217	movq	early_gdt_descr_base(%rip), %rax
218	testq	%rax, %rax
219	jnz	1f
220	movq	$__per_cpu_load, %rax
221	addq	$per_cpu__gdt_page, %rax
222	movq	%rax, early_gdt_descr_base(%rip)
2231:
224#endif
225	/*
226	 * We must switch to a new descriptor in kernel space for the GDT
227	 * because soon the kernel won't have access anymore to the userspace
228	 * addresses where we're currently running on. We have to do that here
229	 * because in 32bit we couldn't load a 64bit linear address.
230	 */
231	lgdt	early_gdt_descr(%rip)
232
233	/* set up data segments. actually 0 would do too */
234	movl $__KERNEL_DS,%eax
235	movl %eax,%ds
236	movl %eax,%ss
237	movl %eax,%es
238
239	/*
240	 * We don't really need to load %fs or %gs, but load them anyway
241	 * to kill any stale realmode selectors.  This allows execution
242	 * under VT hardware.
243	 */
244	movl %eax,%fs
245	movl %eax,%gs
246
247	/* Set up %gs.
248	 *
249	 * On SMP, %gs should point to the per-cpu area.  For initial
250	 * boot, make %gs point to the init data section.  For a
251	 * secondary CPU,initial_gs should be set to its pda address
252	 * before the CPU runs this code.
253	 *
254	 * On UP, initial_gs points to PER_CPU_VAR(__pda) and doesn't
255	 * change.
256	 */
257	movl	$MSR_GS_BASE,%ecx
258	movq	initial_gs(%rip),%rax
259	movq    %rax,%rdx
260	shrq	$32,%rdx
261	wrmsr
262
263	/* esi is pointer to real mode structure with interesting info.
264	   pass it to C */
265	movl	%esi, %edi
266
267	/* Finally jump to run C code and to be on real kernel address
268	 * Since we are running on identity-mapped space we have to jump
269	 * to the full 64bit address, this is only possible as indirect
270	 * jump.  In addition we need to ensure %cs is set so we make this
271	 * a far return.
272	 */
273	movq	initial_code(%rip),%rax
274	pushq	$0		# fake return address to stop unwinder
275	pushq	$__KERNEL_CS	# set correct cs
276	pushq	%rax		# target address in negative space
277	lretq
278
279	/* SMP bootup changes these two */
280	__REFDATA
281	.align	8
282	ENTRY(initial_code)
283	.quad	x86_64_start_kernel
284	ENTRY(initial_gs)
285#ifdef CONFIG_SMP
286	.quad	__per_cpu_load
287#else
288	.quad	PER_CPU_VAR(__pda)
289#endif
290	__FINITDATA
291
292	ENTRY(stack_start)
293	.quad  init_thread_union+THREAD_SIZE-8
294	.word  0
295
296bad_address:
297	jmp bad_address
298
299	.section ".init.text","ax"
300#ifdef CONFIG_EARLY_PRINTK
301	.globl early_idt_handlers
302early_idt_handlers:
303	i = 0
304	.rept NUM_EXCEPTION_VECTORS
305	movl $i, %esi
306	jmp early_idt_handler
307	i = i + 1
308	.endr
309#endif
310
311ENTRY(early_idt_handler)
312#ifdef CONFIG_EARLY_PRINTK
313	cmpl $2,early_recursion_flag(%rip)
314	jz  1f
315	incl early_recursion_flag(%rip)
316	GET_CR2_INTO_RCX
317	movq %rcx,%r9
318	xorl %r8d,%r8d		# zero for error code
319	movl %esi,%ecx		# get vector number
320	# Test %ecx against mask of vectors that push error code.
321	cmpl $31,%ecx
322	ja 0f
323	movl $1,%eax
324	salq %cl,%rax
325	testl $0x27d00,%eax
326	je 0f
327	popq %r8		# get error code
3280:	movq 0(%rsp),%rcx	# get ip
329	movq 8(%rsp),%rdx	# get cs
330	xorl %eax,%eax
331	leaq early_idt_msg(%rip),%rdi
332	call early_printk
333	cmpl $2,early_recursion_flag(%rip)
334	jz  1f
335	call dump_stack
336#ifdef CONFIG_KALLSYMS
337	leaq early_idt_ripmsg(%rip),%rdi
338	movq 0(%rsp),%rsi	# get rip again
339	call __print_symbol
340#endif
341#endif /* EARLY_PRINTK */
3421:	hlt
343	jmp 1b
344
345#ifdef CONFIG_EARLY_PRINTK
346early_recursion_flag:
347	.long 0
348
349early_idt_msg:
350	.asciz "PANIC: early exception %02lx rip %lx:%lx error %lx cr2 %lx\n"
351early_idt_ripmsg:
352	.asciz "RIP %s\n"
353#endif /* CONFIG_EARLY_PRINTK */
354	.previous
355
356.balign PAGE_SIZE
357
358#define NEXT_PAGE(name) \
359	.balign	PAGE_SIZE; \
360ENTRY(name)
361
362/* Automate the creation of 1 to 1 mapping pmd entries */
363#define PMDS(START, PERM, COUNT)			\
364	i = 0 ;						\
365	.rept (COUNT) ;					\
366	.quad	(START) + (i << PMD_SHIFT) + (PERM) ;	\
367	i = i + 1 ;					\
368	.endr
369
370	/*
371	 * This default setting generates an ident mapping at address 0x100000
372	 * and a mapping for the kernel that precisely maps virtual address
373	 * 0xffffffff80000000 to physical address 0x000000. (always using
374	 * 2Mbyte large pages provided by PAE mode)
375	 */
376NEXT_PAGE(init_level4_pgt)
377	.quad	level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
378	.org	init_level4_pgt + L4_PAGE_OFFSET*8, 0
379	.quad	level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
380	.org	init_level4_pgt + L4_START_KERNEL*8, 0
381	/* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
382	.quad	level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
383
384NEXT_PAGE(level3_ident_pgt)
385	.quad	level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
386	.fill	511,8,0
387
388NEXT_PAGE(level3_kernel_pgt)
389	.fill	L3_START_KERNEL,8,0
390	/* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
391	.quad	level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE
392	.quad	level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
393
394NEXT_PAGE(level2_fixmap_pgt)
395	.fill	506,8,0
396	.quad	level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
397	/* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */
398	.fill	5,8,0
399
400NEXT_PAGE(level1_fixmap_pgt)
401	.fill	512,8,0
402
403NEXT_PAGE(level2_ident_pgt)
404	/* Since I easily can, map the first 1G.
405	 * Don't set NX because code runs from these pages.
406	 */
407	PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
408
409NEXT_PAGE(level2_kernel_pgt)
410	/*
411	 * 512 MB kernel mapping. We spend a full page on this pagetable
412	 * anyway.
413	 *
414	 * The kernel code+data+bss must not be bigger than that.
415	 *
416	 * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
417	 *  If you want to increase this then increase MODULES_VADDR
418	 *  too.)
419	 */
420	PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
421		KERNEL_IMAGE_SIZE/PMD_SIZE)
422
423NEXT_PAGE(level2_spare_pgt)
424	.fill   512, 8, 0
425
426#undef PMDS
427#undef NEXT_PAGE
428
429	.data
430	.align 16
431	.globl early_gdt_descr
432early_gdt_descr:
433	.word	GDT_ENTRIES*8-1
434#ifdef CONFIG_SMP
435early_gdt_descr_base:
436	.quad   0x0000000000000000
437#else
438	.quad	per_cpu__gdt_page
439#endif
440
441ENTRY(phys_base)
442	/* This must match the first entry in level2_kernel_pgt */
443	.quad   0x0000000000000000
444
445#include "../../x86/xen/xen-head.S"
446
447	.section .bss, "aw", @nobits
448	.align L1_CACHE_BYTES
449ENTRY(idt_table)
450	.skip 256 * 16
451
452	.section .bss.page_aligned, "aw", @nobits
453	.align PAGE_SIZE
454ENTRY(empty_zero_page)
455	.skip PAGE_SIZE
456