xref: /openbmc/linux/arch/x86/kernel/head_64.S (revision 48927bbb)
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/segment.h>
16#include <asm/pgtable.h>
17#include <asm/page.h>
18#include <asm/msr.h>
19#include <asm/cache.h>
20#include <asm/processor-flags.h>
21#include <asm/percpu.h>
22
23#ifdef CONFIG_PARAVIRT
24#include <asm/asm-offsets.h>
25#include <asm/paravirt.h>
26#else
27#define GET_CR2_INTO_RCX movq %cr2, %rcx
28#endif
29
30/* we are not able to switch in one step to the final KERNEL ADDRESS SPACE
31 * because we need identity-mapped pages.
32 *
33 */
34
35#define pud_index(x)	(((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
36
37L4_PAGE_OFFSET = pgd_index(__PAGE_OFFSET)
38L3_PAGE_OFFSET = pud_index(__PAGE_OFFSET)
39L4_START_KERNEL = pgd_index(__START_KERNEL_map)
40L3_START_KERNEL = pud_index(__START_KERNEL_map)
41
42	.text
43	__HEAD
44	.code64
45	.globl startup_64
46startup_64:
47
48	/*
49	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
50	 * and someone has loaded an identity mapped page table
51	 * for us.  These identity mapped page tables map all of the
52	 * kernel pages and possibly all of memory.
53	 *
54	 * %esi holds a physical pointer to real_mode_data.
55	 *
56	 * We come here either directly from a 64bit bootloader, or from
57	 * arch/x86_64/boot/compressed/head.S.
58	 *
59	 * We only come here initially at boot nothing else comes here.
60	 *
61	 * Since we may be loaded at an address different from what we were
62	 * compiled to run at we first fixup the physical addresses in our page
63	 * tables and then reload them.
64	 */
65
66	/* Compute the delta between the address I am compiled to run at and the
67	 * address I am actually running at.
68	 */
69	leaq	_text(%rip), %rbp
70	subq	$_text - __START_KERNEL_map, %rbp
71
72	/* Is the address not 2M aligned? */
73	movq	%rbp, %rax
74	andl	$~PMD_PAGE_MASK, %eax
75	testl	%eax, %eax
76	jnz	bad_address
77
78	/* Is the address too large? */
79	leaq	_text(%rip), %rdx
80	movq	$PGDIR_SIZE, %rax
81	cmpq	%rax, %rdx
82	jae	bad_address
83
84	/* Fixup the physical addresses in the page table
85	 */
86	addq	%rbp, init_level4_pgt + 0(%rip)
87	addq	%rbp, init_level4_pgt + (L4_PAGE_OFFSET*8)(%rip)
88	addq	%rbp, init_level4_pgt + (L4_START_KERNEL*8)(%rip)
89
90	addq	%rbp, level3_ident_pgt + 0(%rip)
91
92	addq	%rbp, level3_kernel_pgt + (510*8)(%rip)
93	addq	%rbp, level3_kernel_pgt + (511*8)(%rip)
94
95	addq	%rbp, level2_fixmap_pgt + (506*8)(%rip)
96
97	/* Add an Identity mapping if I am above 1G */
98	leaq	_text(%rip), %rdi
99	andq	$PMD_PAGE_MASK, %rdi
100
101	movq	%rdi, %rax
102	shrq	$PUD_SHIFT, %rax
103	andq	$(PTRS_PER_PUD - 1), %rax
104	jz	ident_complete
105
106	leaq	(level2_spare_pgt - __START_KERNEL_map + _KERNPG_TABLE)(%rbp), %rdx
107	leaq	level3_ident_pgt(%rip), %rbx
108	movq	%rdx, 0(%rbx, %rax, 8)
109
110	movq	%rdi, %rax
111	shrq	$PMD_SHIFT, %rax
112	andq	$(PTRS_PER_PMD - 1), %rax
113	leaq	__PAGE_KERNEL_IDENT_LARGE_EXEC(%rdi), %rdx
114	leaq	level2_spare_pgt(%rip), %rbx
115	movq	%rdx, 0(%rbx, %rax, 8)
116ident_complete:
117
118	/*
119	 * Fixup the kernel text+data virtual addresses. Note that
120	 * we might write invalid pmds, when the kernel is relocated
121	 * cleanup_highmap() fixes this up along with the mappings
122	 * beyond _end.
123	 */
124
125	leaq	level2_kernel_pgt(%rip), %rdi
126	leaq	4096(%rdi), %r8
127	/* See if it is a valid page table entry */
1281:	testq	$1, 0(%rdi)
129	jz	2f
130	addq	%rbp, 0(%rdi)
131	/* Go to the next page */
1322:	addq	$8, %rdi
133	cmp	%r8, %rdi
134	jne	1b
135
136	/* Fixup phys_base */
137	addq	%rbp, phys_base(%rip)
138
139	/* Due to ENTRY(), sometimes the empty space gets filled with
140	 * zeros. Better take a jmp than relying on empty space being
141	 * filled with 0x90 (nop)
142	 */
143	jmp secondary_startup_64
144ENTRY(secondary_startup_64)
145	/*
146	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
147	 * and someone has loaded a mapped page table.
148	 *
149	 * %esi holds a physical pointer to real_mode_data.
150	 *
151	 * We come here either from startup_64 (using physical addresses)
152	 * or from trampoline.S (using virtual addresses).
153	 *
154	 * Using virtual addresses from trampoline.S removes the need
155	 * to have any identity mapped pages in the kernel page table
156	 * after the boot processor executes this code.
157	 */
158
159	/* Enable PAE mode and PGE */
160	movl	$(X86_CR4_PAE | X86_CR4_PGE), %eax
161	movq	%rax, %cr4
162
163	/* Setup early boot stage 4 level pagetables. */
164	movq	$(init_level4_pgt - __START_KERNEL_map), %rax
165	addq	phys_base(%rip), %rax
166	movq	%rax, %cr3
167
168	/* Ensure I am executing from virtual addresses */
169	movq	$1f, %rax
170	jmp	*%rax
1711:
172
173	/* Check if nx is implemented */
174	movl	$0x80000001, %eax
175	cpuid
176	movl	%edx,%edi
177
178	/* Setup EFER (Extended Feature Enable Register) */
179	movl	$MSR_EFER, %ecx
180	rdmsr
181	btsl	$_EFER_SCE, %eax	/* Enable System Call */
182	btl	$20,%edi		/* No Execute supported? */
183	jnc     1f
184	btsl	$_EFER_NX, %eax
1851:	wrmsr				/* Make changes effective */
186
187	/* Setup cr0 */
188#define CR0_STATE	(X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
189			 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
190			 X86_CR0_PG)
191	movl	$CR0_STATE, %eax
192	/* Make changes effective */
193	movq	%rax, %cr0
194
195	/* Setup a boot time stack */
196	movq stack_start(%rip),%rsp
197
198	/* zero EFLAGS after setting rsp */
199	pushq $0
200	popfq
201
202	/*
203	 * We must switch to a new descriptor in kernel space for the GDT
204	 * because soon the kernel won't have access anymore to the userspace
205	 * addresses where we're currently running on. We have to do that here
206	 * because in 32bit we couldn't load a 64bit linear address.
207	 */
208	lgdt	early_gdt_descr(%rip)
209
210	/* set up data segments */
211	xorl %eax,%eax
212	movl %eax,%ds
213	movl %eax,%ss
214	movl %eax,%es
215
216	/*
217	 * We don't really need to load %fs or %gs, but load them anyway
218	 * to kill any stale realmode selectors.  This allows execution
219	 * under VT hardware.
220	 */
221	movl %eax,%fs
222	movl %eax,%gs
223
224	/* Set up %gs.
225	 *
226	 * The base of %gs always points to the bottom of the irqstack
227	 * union.  If the stack protector canary is enabled, it is
228	 * located at %gs:40.  Note that, on SMP, the boot cpu uses
229	 * init data section till per cpu areas are set up.
230	 */
231	movl	$MSR_GS_BASE,%ecx
232	movl	initial_gs(%rip),%eax
233	movl	initial_gs+4(%rip),%edx
234	wrmsr
235
236	/* esi is pointer to real mode structure with interesting info.
237	   pass it to C */
238	movl	%esi, %edi
239
240	/* Finally jump to run C code and to be on real kernel address
241	 * Since we are running on identity-mapped space we have to jump
242	 * to the full 64bit address, this is only possible as indirect
243	 * jump.  In addition we need to ensure %cs is set so we make this
244	 * a far return.
245	 */
246	movq	initial_code(%rip),%rax
247	pushq	$0		# fake return address to stop unwinder
248	pushq	$__KERNEL_CS	# set correct cs
249	pushq	%rax		# target address in negative space
250	lretq
251
252	/* SMP bootup changes these two */
253	__REFDATA
254	.align	8
255	ENTRY(initial_code)
256	.quad	x86_64_start_kernel
257	ENTRY(initial_gs)
258	.quad	INIT_PER_CPU_VAR(irq_stack_union)
259
260	ENTRY(stack_start)
261	.quad  init_thread_union+THREAD_SIZE-8
262	.word  0
263	__FINITDATA
264
265bad_address:
266	jmp bad_address
267
268	.section ".init.text","ax"
269#ifdef CONFIG_EARLY_PRINTK
270	.globl early_idt_handlers
271early_idt_handlers:
272	i = 0
273	.rept NUM_EXCEPTION_VECTORS
274	movl $i, %esi
275	jmp early_idt_handler
276	i = i + 1
277	.endr
278#endif
279
280ENTRY(early_idt_handler)
281#ifdef CONFIG_EARLY_PRINTK
282	cmpl $2,early_recursion_flag(%rip)
283	jz  1f
284	incl early_recursion_flag(%rip)
285	GET_CR2_INTO_RCX
286	movq %rcx,%r9
287	xorl %r8d,%r8d		# zero for error code
288	movl %esi,%ecx		# get vector number
289	# Test %ecx against mask of vectors that push error code.
290	cmpl $31,%ecx
291	ja 0f
292	movl $1,%eax
293	salq %cl,%rax
294	testl $0x27d00,%eax
295	je 0f
296	popq %r8		# get error code
2970:	movq 0(%rsp),%rcx	# get ip
298	movq 8(%rsp),%rdx	# get cs
299	xorl %eax,%eax
300	leaq early_idt_msg(%rip),%rdi
301	call early_printk
302	cmpl $2,early_recursion_flag(%rip)
303	jz  1f
304	call dump_stack
305#ifdef CONFIG_KALLSYMS
306	leaq early_idt_ripmsg(%rip),%rdi
307	movq 0(%rsp),%rsi	# get rip again
308	call __print_symbol
309#endif
310#endif /* EARLY_PRINTK */
3111:	hlt
312	jmp 1b
313
314#ifdef CONFIG_EARLY_PRINTK
315early_recursion_flag:
316	.long 0
317
318early_idt_msg:
319	.asciz "PANIC: early exception %02lx rip %lx:%lx error %lx cr2 %lx\n"
320early_idt_ripmsg:
321	.asciz "RIP %s\n"
322#endif /* CONFIG_EARLY_PRINTK */
323	.previous
324
325#define NEXT_PAGE(name) \
326	.balign	PAGE_SIZE; \
327ENTRY(name)
328
329/* Automate the creation of 1 to 1 mapping pmd entries */
330#define PMDS(START, PERM, COUNT)			\
331	i = 0 ;						\
332	.rept (COUNT) ;					\
333	.quad	(START) + (i << PMD_SHIFT) + (PERM) ;	\
334	i = i + 1 ;					\
335	.endr
336
337	.data
338	/*
339	 * This default setting generates an ident mapping at address 0x100000
340	 * and a mapping for the kernel that precisely maps virtual address
341	 * 0xffffffff80000000 to physical address 0x000000. (always using
342	 * 2Mbyte large pages provided by PAE mode)
343	 */
344NEXT_PAGE(init_level4_pgt)
345	.quad	level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
346	.org	init_level4_pgt + L4_PAGE_OFFSET*8, 0
347	.quad	level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
348	.org	init_level4_pgt + L4_START_KERNEL*8, 0
349	/* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
350	.quad	level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
351
352NEXT_PAGE(level3_ident_pgt)
353	.quad	level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
354	.fill	511,8,0
355
356NEXT_PAGE(level3_kernel_pgt)
357	.fill	L3_START_KERNEL,8,0
358	/* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
359	.quad	level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE
360	.quad	level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
361
362NEXT_PAGE(level2_fixmap_pgt)
363	.fill	506,8,0
364	.quad	level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
365	/* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */
366	.fill	5,8,0
367
368NEXT_PAGE(level1_fixmap_pgt)
369	.fill	512,8,0
370
371NEXT_PAGE(level2_ident_pgt)
372	/* Since I easily can, map the first 1G.
373	 * Don't set NX because code runs from these pages.
374	 */
375	PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
376
377NEXT_PAGE(level2_kernel_pgt)
378	/*
379	 * 512 MB kernel mapping. We spend a full page on this pagetable
380	 * anyway.
381	 *
382	 * The kernel code+data+bss must not be bigger than that.
383	 *
384	 * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
385	 *  If you want to increase this then increase MODULES_VADDR
386	 *  too.)
387	 */
388	PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
389		KERNEL_IMAGE_SIZE/PMD_SIZE)
390
391NEXT_PAGE(level2_spare_pgt)
392	.fill   512, 8, 0
393
394#undef PMDS
395#undef NEXT_PAGE
396
397	.data
398	.align 16
399	.globl early_gdt_descr
400early_gdt_descr:
401	.word	GDT_ENTRIES*8-1
402early_gdt_descr_base:
403	.quad	INIT_PER_CPU_VAR(gdt_page)
404
405ENTRY(phys_base)
406	/* This must match the first entry in level2_kernel_pgt */
407	.quad   0x0000000000000000
408
409#include "../../x86/xen/xen-head.S"
410
411	.section .bss, "aw", @nobits
412	.align L1_CACHE_BYTES
413ENTRY(idt_table)
414	.skip IDT_ENTRIES * 16
415
416	.align L1_CACHE_BYTES
417ENTRY(nmi_idt_table)
418	.skip IDT_ENTRIES * 16
419
420	__PAGE_ALIGNED_BSS
421	.align PAGE_SIZE
422ENTRY(empty_zero_page)
423	.skip PAGE_SIZE
424