xref: /openbmc/linux/arch/x86/kernel/head_64.S (revision a16be368)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 *  linux/arch/x86/kernel/head_64.S -- start in 32bit and switch to 64bit
4 *
5 *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 *  Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
7 *  Copyright (C) 2000 Karsten Keil <kkeil@suse.de>
8 *  Copyright (C) 2001,2002 Andi Kleen <ak@suse.de>
9 *  Copyright (C) 2005 Eric Biederman <ebiederm@xmission.com>
10 */
11
12
13#include <linux/linkage.h>
14#include <linux/threads.h>
15#include <linux/init.h>
16#include <linux/pgtable.h>
17#include <asm/segment.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#include <asm/nops.h>
24#include "../entry/calling.h"
25#include <asm/export.h>
26#include <asm/nospec-branch.h>
27#include <asm/fixmap.h>
28
29#ifdef CONFIG_PARAVIRT_XXL
30#include <asm/asm-offsets.h>
31#include <asm/paravirt.h>
32#else
33#define INTERRUPT_RETURN iretq
34#endif
35
36/* we are not able to switch in one step to the final KERNEL ADDRESS SPACE
37 * because we need identity-mapped pages.
38 *
39 */
40
41#define l4_index(x)	(((x) >> 39) & 511)
42#define pud_index(x)	(((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
43
44L4_PAGE_OFFSET = l4_index(__PAGE_OFFSET_BASE_L4)
45L4_START_KERNEL = l4_index(__START_KERNEL_map)
46
47L3_START_KERNEL = pud_index(__START_KERNEL_map)
48
49	.text
50	__HEAD
51	.code64
52SYM_CODE_START_NOALIGN(startup_64)
53	UNWIND_HINT_EMPTY
54	/*
55	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
56	 * and someone has loaded an identity mapped page table
57	 * for us.  These identity mapped page tables map all of the
58	 * kernel pages and possibly all of memory.
59	 *
60	 * %rsi holds a physical pointer to real_mode_data.
61	 *
62	 * We come here either directly from a 64bit bootloader, or from
63	 * arch/x86/boot/compressed/head_64.S.
64	 *
65	 * We only come here initially at boot nothing else comes here.
66	 *
67	 * Since we may be loaded at an address different from what we were
68	 * compiled to run at we first fixup the physical addresses in our page
69	 * tables and then reload them.
70	 */
71
72	/* Set up the stack for verify_cpu(), similar to initial_stack below */
73	leaq	(__end_init_task - SIZEOF_PTREGS)(%rip), %rsp
74
75	/* Sanitize CPU configuration */
76	call verify_cpu
77
78	/*
79	 * Perform pagetable fixups. Additionally, if SME is active, encrypt
80	 * the kernel and retrieve the modifier (SME encryption mask if SME
81	 * is active) to be added to the initial pgdir entry that will be
82	 * programmed into CR3.
83	 */
84	leaq	_text(%rip), %rdi
85	pushq	%rsi
86	call	__startup_64
87	popq	%rsi
88
89	/* Form the CR3 value being sure to include the CR3 modifier */
90	addq	$(early_top_pgt - __START_KERNEL_map), %rax
91	jmp 1f
92SYM_CODE_END(startup_64)
93
94SYM_CODE_START(secondary_startup_64)
95	UNWIND_HINT_EMPTY
96	/*
97	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
98	 * and someone has loaded a mapped page table.
99	 *
100	 * %rsi holds a physical pointer to real_mode_data.
101	 *
102	 * We come here either from startup_64 (using physical addresses)
103	 * or from trampoline.S (using virtual addresses).
104	 *
105	 * Using virtual addresses from trampoline.S removes the need
106	 * to have any identity mapped pages in the kernel page table
107	 * after the boot processor executes this code.
108	 */
109
110	/* Sanitize CPU configuration */
111	call verify_cpu
112
113	/*
114	 * Retrieve the modifier (SME encryption mask if SME is active) to be
115	 * added to the initial pgdir entry that will be programmed into CR3.
116	 */
117	pushq	%rsi
118	call	__startup_secondary_64
119	popq	%rsi
120
121	/* Form the CR3 value being sure to include the CR3 modifier */
122	addq	$(init_top_pgt - __START_KERNEL_map), %rax
1231:
124
125	/* Enable PAE mode, PGE and LA57 */
126	movl	$(X86_CR4_PAE | X86_CR4_PGE), %ecx
127#ifdef CONFIG_X86_5LEVEL
128	testl	$1, __pgtable_l5_enabled(%rip)
129	jz	1f
130	orl	$X86_CR4_LA57, %ecx
1311:
132#endif
133	movq	%rcx, %cr4
134
135	/* Setup early boot stage 4-/5-level pagetables. */
136	addq	phys_base(%rip), %rax
137	movq	%rax, %cr3
138
139	/* Ensure I am executing from virtual addresses */
140	movq	$1f, %rax
141	ANNOTATE_RETPOLINE_SAFE
142	jmp	*%rax
1431:
144	UNWIND_HINT_EMPTY
145
146	/* Check if nx is implemented */
147	movl	$0x80000001, %eax
148	cpuid
149	movl	%edx,%edi
150
151	/* Setup EFER (Extended Feature Enable Register) */
152	movl	$MSR_EFER, %ecx
153	rdmsr
154	btsl	$_EFER_SCE, %eax	/* Enable System Call */
155	btl	$20,%edi		/* No Execute supported? */
156	jnc     1f
157	btsl	$_EFER_NX, %eax
158	btsq	$_PAGE_BIT_NX,early_pmd_flags(%rip)
1591:	wrmsr				/* Make changes effective */
160
161	/* Setup cr0 */
162	movl	$CR0_STATE, %eax
163	/* Make changes effective */
164	movq	%rax, %cr0
165
166	/* Setup a boot time stack */
167	movq initial_stack(%rip), %rsp
168
169	/* zero EFLAGS after setting rsp */
170	pushq $0
171	popfq
172
173	/*
174	 * We must switch to a new descriptor in kernel space for the GDT
175	 * because soon the kernel won't have access anymore to the userspace
176	 * addresses where we're currently running on. We have to do that here
177	 * because in 32bit we couldn't load a 64bit linear address.
178	 */
179	lgdt	early_gdt_descr(%rip)
180
181	/* set up data segments */
182	xorl %eax,%eax
183	movl %eax,%ds
184	movl %eax,%ss
185	movl %eax,%es
186
187	/*
188	 * We don't really need to load %fs or %gs, but load them anyway
189	 * to kill any stale realmode selectors.  This allows execution
190	 * under VT hardware.
191	 */
192	movl %eax,%fs
193	movl %eax,%gs
194
195	/* Set up %gs.
196	 *
197	 * The base of %gs always points to fixed_percpu_data. If the
198	 * stack protector canary is enabled, it is located at %gs:40.
199	 * Note that, on SMP, the boot cpu uses init data section until
200	 * the per cpu areas are set up.
201	 */
202	movl	$MSR_GS_BASE,%ecx
203	movl	initial_gs(%rip),%eax
204	movl	initial_gs+4(%rip),%edx
205	wrmsr
206
207	/* rsi is pointer to real mode structure with interesting info.
208	   pass it to C */
209	movq	%rsi, %rdi
210
211.Ljump_to_C_code:
212	/*
213	 * Jump to run C code and to be on a real kernel address.
214	 * Since we are running on identity-mapped space we have to jump
215	 * to the full 64bit address, this is only possible as indirect
216	 * jump.  In addition we need to ensure %cs is set so we make this
217	 * a far return.
218	 *
219	 * Note: do not change to far jump indirect with 64bit offset.
220	 *
221	 * AMD does not support far jump indirect with 64bit offset.
222	 * AMD64 Architecture Programmer's Manual, Volume 3: states only
223	 *	JMP FAR mem16:16 FF /5 Far jump indirect,
224	 *		with the target specified by a far pointer in memory.
225	 *	JMP FAR mem16:32 FF /5 Far jump indirect,
226	 *		with the target specified by a far pointer in memory.
227	 *
228	 * Intel64 does support 64bit offset.
229	 * Software Developer Manual Vol 2: states:
230	 *	FF /5 JMP m16:16 Jump far, absolute indirect,
231	 *		address given in m16:16
232	 *	FF /5 JMP m16:32 Jump far, absolute indirect,
233	 *		address given in m16:32.
234	 *	REX.W + FF /5 JMP m16:64 Jump far, absolute indirect,
235	 *		address given in m16:64.
236	 */
237	pushq	$.Lafter_lret	# put return address on stack for unwinder
238	xorl	%ebp, %ebp	# clear frame pointer
239	movq	initial_code(%rip), %rax
240	pushq	$__KERNEL_CS	# set correct cs
241	pushq	%rax		# target address in negative space
242	lretq
243.Lafter_lret:
244SYM_CODE_END(secondary_startup_64)
245
246#include "verify_cpu.S"
247
248#ifdef CONFIG_HOTPLUG_CPU
249/*
250 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
251 * up already except stack. We just set up stack here. Then call
252 * start_secondary() via .Ljump_to_C_code.
253 */
254SYM_CODE_START(start_cpu0)
255	UNWIND_HINT_EMPTY
256	movq	initial_stack(%rip), %rsp
257	jmp	.Ljump_to_C_code
258SYM_CODE_END(start_cpu0)
259#endif
260
261	/* Both SMP bootup and ACPI suspend change these variables */
262	__REFDATA
263	.balign	8
264SYM_DATA(initial_code,	.quad x86_64_start_kernel)
265SYM_DATA(initial_gs,	.quad INIT_PER_CPU_VAR(fixed_percpu_data))
266
267/*
268 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel unwinder
269 * reliably detect the end of the stack.
270 */
271SYM_DATA(initial_stack, .quad init_thread_union + THREAD_SIZE - SIZEOF_PTREGS)
272	__FINITDATA
273
274	__INIT
275SYM_CODE_START(early_idt_handler_array)
276	i = 0
277	.rept NUM_EXCEPTION_VECTORS
278	.if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
279		UNWIND_HINT_IRET_REGS
280		pushq $0	# Dummy error code, to make stack frame uniform
281	.else
282		UNWIND_HINT_IRET_REGS offset=8
283	.endif
284	pushq $i		# 72(%rsp) Vector number
285	jmp early_idt_handler_common
286	UNWIND_HINT_IRET_REGS
287	i = i + 1
288	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
289	.endr
290	UNWIND_HINT_IRET_REGS offset=16
291SYM_CODE_END(early_idt_handler_array)
292
293SYM_CODE_START_LOCAL(early_idt_handler_common)
294	/*
295	 * The stack is the hardware frame, an error code or zero, and the
296	 * vector number.
297	 */
298	cld
299
300	incl early_recursion_flag(%rip)
301
302	/* The vector number is currently in the pt_regs->di slot. */
303	pushq %rsi				/* pt_regs->si */
304	movq 8(%rsp), %rsi			/* RSI = vector number */
305	movq %rdi, 8(%rsp)			/* pt_regs->di = RDI */
306	pushq %rdx				/* pt_regs->dx */
307	pushq %rcx				/* pt_regs->cx */
308	pushq %rax				/* pt_regs->ax */
309	pushq %r8				/* pt_regs->r8 */
310	pushq %r9				/* pt_regs->r9 */
311	pushq %r10				/* pt_regs->r10 */
312	pushq %r11				/* pt_regs->r11 */
313	pushq %rbx				/* pt_regs->bx */
314	pushq %rbp				/* pt_regs->bp */
315	pushq %r12				/* pt_regs->r12 */
316	pushq %r13				/* pt_regs->r13 */
317	pushq %r14				/* pt_regs->r14 */
318	pushq %r15				/* pt_regs->r15 */
319	UNWIND_HINT_REGS
320
321	cmpq $14,%rsi		/* Page fault? */
322	jnz 10f
323	GET_CR2_INTO(%rdi)	/* can clobber %rax if pv */
324	call early_make_pgtable
325	andl %eax,%eax
326	jz 20f			/* All good */
327
32810:
329	movq %rsp,%rdi		/* RDI = pt_regs; RSI is already trapnr */
330	call early_fixup_exception
331
33220:
333	decl early_recursion_flag(%rip)
334	jmp restore_regs_and_return_to_kernel
335SYM_CODE_END(early_idt_handler_common)
336
337
338#define SYM_DATA_START_PAGE_ALIGNED(name)			\
339	SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE)
340
341#ifdef CONFIG_PAGE_TABLE_ISOLATION
342/*
343 * Each PGD needs to be 8k long and 8k aligned.  We do not
344 * ever go out to userspace with these, so we do not
345 * strictly *need* the second page, but this allows us to
346 * have a single set_pgd() implementation that does not
347 * need to worry about whether it has 4k or 8k to work
348 * with.
349 *
350 * This ensures PGDs are 8k long:
351 */
352#define PTI_USER_PGD_FILL	512
353/* This ensures they are 8k-aligned: */
354#define SYM_DATA_START_PTI_ALIGNED(name) \
355	SYM_START(name, SYM_L_GLOBAL, .balign 2 * PAGE_SIZE)
356#else
357#define SYM_DATA_START_PTI_ALIGNED(name) \
358	SYM_DATA_START_PAGE_ALIGNED(name)
359#define PTI_USER_PGD_FILL	0
360#endif
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	__INITDATA
371	.balign 4
372
373SYM_DATA_START_PTI_ALIGNED(early_top_pgt)
374	.fill	512,8,0
375	.fill	PTI_USER_PGD_FILL,8,0
376SYM_DATA_END(early_top_pgt)
377
378SYM_DATA_START_PAGE_ALIGNED(early_dynamic_pgts)
379	.fill	512*EARLY_DYNAMIC_PAGE_TABLES,8,0
380SYM_DATA_END(early_dynamic_pgts)
381
382SYM_DATA(early_recursion_flag, .long 0)
383
384	.data
385
386#if defined(CONFIG_XEN_PV) || defined(CONFIG_PVH)
387SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
388	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
389	.org    init_top_pgt + L4_PAGE_OFFSET*8, 0
390	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
391	.org    init_top_pgt + L4_START_KERNEL*8, 0
392	/* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
393	.quad   level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
394	.fill	PTI_USER_PGD_FILL,8,0
395SYM_DATA_END(init_top_pgt)
396
397SYM_DATA_START_PAGE_ALIGNED(level3_ident_pgt)
398	.quad	level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
399	.fill	511, 8, 0
400SYM_DATA_END(level3_ident_pgt)
401SYM_DATA_START_PAGE_ALIGNED(level2_ident_pgt)
402	/*
403	 * Since I easily can, map the first 1G.
404	 * Don't set NX because code runs from these pages.
405	 *
406	 * Note: This sets _PAGE_GLOBAL despite whether
407	 * the CPU supports it or it is enabled.  But,
408	 * the CPU should ignore the bit.
409	 */
410	PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
411SYM_DATA_END(level2_ident_pgt)
412#else
413SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
414	.fill	512,8,0
415	.fill	PTI_USER_PGD_FILL,8,0
416SYM_DATA_END(init_top_pgt)
417#endif
418
419#ifdef CONFIG_X86_5LEVEL
420SYM_DATA_START_PAGE_ALIGNED(level4_kernel_pgt)
421	.fill	511,8,0
422	.quad	level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
423SYM_DATA_END(level4_kernel_pgt)
424#endif
425
426SYM_DATA_START_PAGE_ALIGNED(level3_kernel_pgt)
427	.fill	L3_START_KERNEL,8,0
428	/* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
429	.quad	level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
430	.quad	level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
431SYM_DATA_END(level3_kernel_pgt)
432
433SYM_DATA_START_PAGE_ALIGNED(level2_kernel_pgt)
434	/*
435	 * 512 MB kernel mapping. We spend a full page on this pagetable
436	 * anyway.
437	 *
438	 * The kernel code+data+bss must not be bigger than that.
439	 *
440	 * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
441	 *  If you want to increase this then increase MODULES_VADDR
442	 *  too.)
443	 *
444	 *  This table is eventually used by the kernel during normal
445	 *  runtime.  Care must be taken to clear out undesired bits
446	 *  later, like _PAGE_RW or _PAGE_GLOBAL in some cases.
447	 */
448	PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
449		KERNEL_IMAGE_SIZE/PMD_SIZE)
450SYM_DATA_END(level2_kernel_pgt)
451
452SYM_DATA_START_PAGE_ALIGNED(level2_fixmap_pgt)
453	.fill	(512 - 4 - FIXMAP_PMD_NUM),8,0
454	pgtno = 0
455	.rept (FIXMAP_PMD_NUM)
456	.quad level1_fixmap_pgt + (pgtno << PAGE_SHIFT) - __START_KERNEL_map \
457		+ _PAGE_TABLE_NOENC;
458	pgtno = pgtno + 1
459	.endr
460	/* 6 MB reserved space + a 2MB hole */
461	.fill	4,8,0
462SYM_DATA_END(level2_fixmap_pgt)
463
464SYM_DATA_START_PAGE_ALIGNED(level1_fixmap_pgt)
465	.rept (FIXMAP_PMD_NUM)
466	.fill	512,8,0
467	.endr
468SYM_DATA_END(level1_fixmap_pgt)
469
470#undef PMDS
471
472	.data
473	.align 16
474
475SYM_DATA(early_gdt_descr,		.word GDT_ENTRIES*8-1)
476SYM_DATA_LOCAL(early_gdt_descr_base,	.quad INIT_PER_CPU_VAR(gdt_page))
477
478	.align 16
479/* This must match the first entry in level2_kernel_pgt */
480SYM_DATA(phys_base, .quad 0x0)
481EXPORT_SYMBOL(phys_base)
482
483#include "../../x86/xen/xen-head.S"
484
485	__PAGE_ALIGNED_BSS
486SYM_DATA_START_PAGE_ALIGNED(empty_zero_page)
487	.skip PAGE_SIZE
488SYM_DATA_END(empty_zero_page)
489EXPORT_SYMBOL(empty_zero_page)
490
491