xref: /linux/arch/x86/kernel/head_64.S (revision 68550cbc6129159b7a6434796b721e8b66ee12f6)
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/*
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#define l4_index(x)	(((x) >> 39) & 511)
34#define pud_index(x)	(((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
35
36L4_PAGE_OFFSET = l4_index(__PAGE_OFFSET_BASE_L4)
37L4_START_KERNEL = l4_index(__START_KERNEL_map)
38
39L3_START_KERNEL = pud_index(__START_KERNEL_map)
40
41	.text
42	__HEAD
43	.code64
44SYM_CODE_START_NOALIGN(startup_64)
45	UNWIND_HINT_EMPTY
46	/*
47	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
48	 * and someone has loaded an identity mapped page table
49	 * for us.  These identity mapped page tables map all of the
50	 * kernel pages and possibly all of memory.
51	 *
52	 * %rsi holds a physical pointer to real_mode_data.
53	 *
54	 * We come here either directly from a 64bit bootloader, or from
55	 * arch/x86/boot/compressed/head_64.S.
56	 *
57	 * We only come here initially at boot nothing else comes here.
58	 *
59	 * Since we may be loaded at an address different from what we were
60	 * compiled to run at we first fixup the physical addresses in our page
61	 * tables and then reload them.
62	 */
63
64	/* Set up the stack for verify_cpu(), similar to initial_stack below */
65	leaq	(__end_init_task - FRAME_SIZE)(%rip), %rsp
66
67	leaq	_text(%rip), %rdi
68	pushq	%rsi
69	call	startup_64_setup_env
70	popq	%rsi
71
72	/* Now switch to __KERNEL_CS so IRET works reliably */
73	pushq	$__KERNEL_CS
74	leaq	.Lon_kernel_cs(%rip), %rax
75	pushq	%rax
76	lretq
77
78.Lon_kernel_cs:
79	UNWIND_HINT_EMPTY
80
81	/* Sanitize CPU configuration */
82	call verify_cpu
83
84	/*
85	 * Perform pagetable fixups. Additionally, if SME is active, encrypt
86	 * the kernel and retrieve the modifier (SME encryption mask if SME
87	 * is active) to be added to the initial pgdir entry that will be
88	 * programmed into CR3.
89	 */
90	leaq	_text(%rip), %rdi
91	pushq	%rsi
92	call	__startup_64
93	popq	%rsi
94
95	/* Form the CR3 value being sure to include the CR3 modifier */
96	addq	$(early_top_pgt - __START_KERNEL_map), %rax
97	jmp 1f
98SYM_CODE_END(startup_64)
99
100SYM_CODE_START(secondary_startup_64)
101	UNWIND_HINT_EMPTY
102	/*
103	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
104	 * and someone has loaded a mapped page table.
105	 *
106	 * %rsi holds a physical pointer to real_mode_data.
107	 *
108	 * We come here either from startup_64 (using physical addresses)
109	 * or from trampoline.S (using virtual addresses).
110	 *
111	 * Using virtual addresses from trampoline.S removes the need
112	 * to have any identity mapped pages in the kernel page table
113	 * after the boot processor executes this code.
114	 */
115
116	/* Sanitize CPU configuration */
117	call verify_cpu
118
119	/*
120	 * The secondary_startup_64_no_verify entry point is only used by
121	 * SEV-ES guests. In those guests the call to verify_cpu() would cause
122	 * #VC exceptions which can not be handled at this stage of secondary
123	 * CPU bringup.
124	 *
125	 * All non SEV-ES systems, especially Intel systems, need to execute
126	 * verify_cpu() above to make sure NX is enabled.
127	 */
128SYM_INNER_LABEL(secondary_startup_64_no_verify, SYM_L_GLOBAL)
129	UNWIND_HINT_EMPTY
130
131	/*
132	 * Retrieve the modifier (SME encryption mask if SME is active) to be
133	 * added to the initial pgdir entry that will be programmed into CR3.
134	 */
135	pushq	%rsi
136	call	__startup_secondary_64
137	popq	%rsi
138
139	/* Form the CR3 value being sure to include the CR3 modifier */
140	addq	$(init_top_pgt - __START_KERNEL_map), %rax
1411:
142
143	/* Enable PAE mode, PGE and LA57 */
144	movl	$(X86_CR4_PAE | X86_CR4_PGE), %ecx
145#ifdef CONFIG_X86_5LEVEL
146	testl	$1, __pgtable_l5_enabled(%rip)
147	jz	1f
148	orl	$X86_CR4_LA57, %ecx
1491:
150#endif
151	movq	%rcx, %cr4
152
153	/* Setup early boot stage 4-/5-level pagetables. */
154	addq	phys_base(%rip), %rax
155
156	/*
157	 * For SEV guests: Verify that the C-bit is correct. A malicious
158	 * hypervisor could lie about the C-bit position to perform a ROP
159	 * attack on the guest by writing to the unencrypted stack and wait for
160	 * the next RET instruction.
161	 * %rsi carries pointer to realmode data and is callee-clobbered. Save
162	 * and restore it.
163	 */
164	pushq	%rsi
165	movq	%rax, %rdi
166	call	sev_verify_cbit
167	popq	%rsi
168
169	/* Switch to new page-table */
170	movq	%rax, %cr3
171
172	/* Ensure I am executing from virtual addresses */
173	movq	$1f, %rax
174	ANNOTATE_RETPOLINE_SAFE
175	jmp	*%rax
1761:
177	UNWIND_HINT_EMPTY
178
179	/*
180	 * We must switch to a new descriptor in kernel space for the GDT
181	 * because soon the kernel won't have access anymore to the userspace
182	 * addresses where we're currently running on. We have to do that here
183	 * because in 32bit we couldn't load a 64bit linear address.
184	 */
185	lgdt	early_gdt_descr(%rip)
186
187	/* set up data segments */
188	xorl %eax,%eax
189	movl %eax,%ds
190	movl %eax,%ss
191	movl %eax,%es
192
193	/*
194	 * We don't really need to load %fs or %gs, but load them anyway
195	 * to kill any stale realmode selectors.  This allows execution
196	 * under VT hardware.
197	 */
198	movl %eax,%fs
199	movl %eax,%gs
200
201	/* Set up %gs.
202	 *
203	 * The base of %gs always points to fixed_percpu_data. If the
204	 * stack protector canary is enabled, it is located at %gs:40.
205	 * Note that, on SMP, the boot cpu uses init data section until
206	 * the per cpu areas are set up.
207	 */
208	movl	$MSR_GS_BASE,%ecx
209	movl	initial_gs(%rip),%eax
210	movl	initial_gs+4(%rip),%edx
211	wrmsr
212
213	/*
214	 * Setup a boot time stack - Any secondary CPU will have lost its stack
215	 * by now because the cr3-switch above unmaps the real-mode stack
216	 */
217	movq initial_stack(%rip), %rsp
218
219	/* Setup and Load IDT */
220	pushq	%rsi
221	call	early_setup_idt
222	popq	%rsi
223
224	/* Check if nx is implemented */
225	movl	$0x80000001, %eax
226	cpuid
227	movl	%edx,%edi
228
229	/* Setup EFER (Extended Feature Enable Register) */
230	movl	$MSR_EFER, %ecx
231	rdmsr
232	btsl	$_EFER_SCE, %eax	/* Enable System Call */
233	btl	$20,%edi		/* No Execute supported? */
234	jnc     1f
235	btsl	$_EFER_NX, %eax
236	btsq	$_PAGE_BIT_NX,early_pmd_flags(%rip)
2371:	wrmsr				/* Make changes effective */
238
239	/* Setup cr0 */
240	movl	$CR0_STATE, %eax
241	/* Make changes effective */
242	movq	%rax, %cr0
243
244	/* zero EFLAGS after setting rsp */
245	pushq $0
246	popfq
247
248	/* rsi is pointer to real mode structure with interesting info.
249	   pass it to C */
250	movq	%rsi, %rdi
251
252.Ljump_to_C_code:
253	/*
254	 * Jump to run C code and to be on a real kernel address.
255	 * Since we are running on identity-mapped space we have to jump
256	 * to the full 64bit address, this is only possible as indirect
257	 * jump.  In addition we need to ensure %cs is set so we make this
258	 * a far return.
259	 *
260	 * Note: do not change to far jump indirect with 64bit offset.
261	 *
262	 * AMD does not support far jump indirect with 64bit offset.
263	 * AMD64 Architecture Programmer's Manual, Volume 3: states only
264	 *	JMP FAR mem16:16 FF /5 Far jump indirect,
265	 *		with the target specified by a far pointer in memory.
266	 *	JMP FAR mem16:32 FF /5 Far jump indirect,
267	 *		with the target specified by a far pointer in memory.
268	 *
269	 * Intel64 does support 64bit offset.
270	 * Software Developer Manual Vol 2: states:
271	 *	FF /5 JMP m16:16 Jump far, absolute indirect,
272	 *		address given in m16:16
273	 *	FF /5 JMP m16:32 Jump far, absolute indirect,
274	 *		address given in m16:32.
275	 *	REX.W + FF /5 JMP m16:64 Jump far, absolute indirect,
276	 *		address given in m16:64.
277	 */
278	pushq	$.Lafter_lret	# put return address on stack for unwinder
279	xorl	%ebp, %ebp	# clear frame pointer
280	movq	initial_code(%rip), %rax
281	pushq	$__KERNEL_CS	# set correct cs
282	pushq	%rax		# target address in negative space
283	lretq
284.Lafter_lret:
285SYM_CODE_END(secondary_startup_64)
286
287#include "verify_cpu.S"
288#include "sev_verify_cbit.S"
289
290#ifdef CONFIG_HOTPLUG_CPU
291/*
292 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
293 * up already except stack. We just set up stack here. Then call
294 * start_secondary() via .Ljump_to_C_code.
295 */
296SYM_CODE_START(start_cpu0)
297	UNWIND_HINT_EMPTY
298	movq	initial_stack(%rip), %rsp
299	jmp	.Ljump_to_C_code
300SYM_CODE_END(start_cpu0)
301#endif
302
303#ifdef CONFIG_AMD_MEM_ENCRYPT
304/*
305 * VC Exception handler used during early boot when running on kernel
306 * addresses, but before the switch to the idt_table can be made.
307 * The early_idt_handler_array can't be used here because it calls into a lot
308 * of __init code and this handler is also used during CPU offlining/onlining.
309 * Therefore this handler ends up in the .text section so that it stays around
310 * when .init.text is freed.
311 */
312SYM_CODE_START_NOALIGN(vc_boot_ghcb)
313	UNWIND_HINT_IRET_REGS offset=8
314
315	/* Build pt_regs */
316	PUSH_AND_CLEAR_REGS
317
318	/* Call C handler */
319	movq    %rsp, %rdi
320	movq	ORIG_RAX(%rsp), %rsi
321	movq	initial_vc_handler(%rip), %rax
322	ANNOTATE_RETPOLINE_SAFE
323	call	*%rax
324
325	/* Unwind pt_regs */
326	POP_REGS
327
328	/* Remove Error Code */
329	addq    $8, %rsp
330
331	/* Pure iret required here - don't use INTERRUPT_RETURN */
332	iretq
333SYM_CODE_END(vc_boot_ghcb)
334#endif
335
336	/* Both SMP bootup and ACPI suspend change these variables */
337	__REFDATA
338	.balign	8
339SYM_DATA(initial_code,	.quad x86_64_start_kernel)
340SYM_DATA(initial_gs,	.quad INIT_PER_CPU_VAR(fixed_percpu_data))
341#ifdef CONFIG_AMD_MEM_ENCRYPT
342SYM_DATA(initial_vc_handler,	.quad handle_vc_boot_ghcb)
343#endif
344
345/*
346 * The FRAME_SIZE gap is a convention which helps the in-kernel unwinder
347 * reliably detect the end of the stack.
348 */
349SYM_DATA(initial_stack, .quad init_thread_union + THREAD_SIZE - FRAME_SIZE)
350	__FINITDATA
351
352	__INIT
353SYM_CODE_START(early_idt_handler_array)
354	i = 0
355	.rept NUM_EXCEPTION_VECTORS
356	.if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
357		UNWIND_HINT_IRET_REGS
358		pushq $0	# Dummy error code, to make stack frame uniform
359	.else
360		UNWIND_HINT_IRET_REGS offset=8
361	.endif
362	pushq $i		# 72(%rsp) Vector number
363	jmp early_idt_handler_common
364	UNWIND_HINT_IRET_REGS
365	i = i + 1
366	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
367	.endr
368	UNWIND_HINT_IRET_REGS offset=16
369SYM_CODE_END(early_idt_handler_array)
370
371SYM_CODE_START_LOCAL(early_idt_handler_common)
372	/*
373	 * The stack is the hardware frame, an error code or zero, and the
374	 * vector number.
375	 */
376	cld
377
378	incl early_recursion_flag(%rip)
379
380	/* The vector number is currently in the pt_regs->di slot. */
381	pushq %rsi				/* pt_regs->si */
382	movq 8(%rsp), %rsi			/* RSI = vector number */
383	movq %rdi, 8(%rsp)			/* pt_regs->di = RDI */
384	pushq %rdx				/* pt_regs->dx */
385	pushq %rcx				/* pt_regs->cx */
386	pushq %rax				/* pt_regs->ax */
387	pushq %r8				/* pt_regs->r8 */
388	pushq %r9				/* pt_regs->r9 */
389	pushq %r10				/* pt_regs->r10 */
390	pushq %r11				/* pt_regs->r11 */
391	pushq %rbx				/* pt_regs->bx */
392	pushq %rbp				/* pt_regs->bp */
393	pushq %r12				/* pt_regs->r12 */
394	pushq %r13				/* pt_regs->r13 */
395	pushq %r14				/* pt_regs->r14 */
396	pushq %r15				/* pt_regs->r15 */
397	UNWIND_HINT_REGS
398
399	movq %rsp,%rdi		/* RDI = pt_regs; RSI is already trapnr */
400	call do_early_exception
401
402	decl early_recursion_flag(%rip)
403	jmp restore_regs_and_return_to_kernel
404SYM_CODE_END(early_idt_handler_common)
405
406#ifdef CONFIG_AMD_MEM_ENCRYPT
407/*
408 * VC Exception handler used during very early boot. The
409 * early_idt_handler_array can't be used because it returns via the
410 * paravirtualized INTERRUPT_RETURN and pv-ops don't work that early.
411 *
412 * This handler will end up in the .init.text section and not be
413 * available to boot secondary CPUs.
414 */
415SYM_CODE_START_NOALIGN(vc_no_ghcb)
416	UNWIND_HINT_IRET_REGS offset=8
417
418	/* Build pt_regs */
419	PUSH_AND_CLEAR_REGS
420
421	/* Call C handler */
422	movq    %rsp, %rdi
423	movq	ORIG_RAX(%rsp), %rsi
424	call    do_vc_no_ghcb
425
426	/* Unwind pt_regs */
427	POP_REGS
428
429	/* Remove Error Code */
430	addq    $8, %rsp
431
432	/* Pure iret required here - don't use INTERRUPT_RETURN */
433	iretq
434SYM_CODE_END(vc_no_ghcb)
435#endif
436
437#define SYM_DATA_START_PAGE_ALIGNED(name)			\
438	SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE)
439
440#ifdef CONFIG_PAGE_TABLE_ISOLATION
441/*
442 * Each PGD needs to be 8k long and 8k aligned.  We do not
443 * ever go out to userspace with these, so we do not
444 * strictly *need* the second page, but this allows us to
445 * have a single set_pgd() implementation that does not
446 * need to worry about whether it has 4k or 8k to work
447 * with.
448 *
449 * This ensures PGDs are 8k long:
450 */
451#define PTI_USER_PGD_FILL	512
452/* This ensures they are 8k-aligned: */
453#define SYM_DATA_START_PTI_ALIGNED(name) \
454	SYM_START(name, SYM_L_GLOBAL, .balign 2 * PAGE_SIZE)
455#else
456#define SYM_DATA_START_PTI_ALIGNED(name) \
457	SYM_DATA_START_PAGE_ALIGNED(name)
458#define PTI_USER_PGD_FILL	0
459#endif
460
461/* Automate the creation of 1 to 1 mapping pmd entries */
462#define PMDS(START, PERM, COUNT)			\
463	i = 0 ;						\
464	.rept (COUNT) ;					\
465	.quad	(START) + (i << PMD_SHIFT) + (PERM) ;	\
466	i = i + 1 ;					\
467	.endr
468
469	__INITDATA
470	.balign 4
471
472SYM_DATA_START_PTI_ALIGNED(early_top_pgt)
473	.fill	512,8,0
474	.fill	PTI_USER_PGD_FILL,8,0
475SYM_DATA_END(early_top_pgt)
476
477SYM_DATA_START_PAGE_ALIGNED(early_dynamic_pgts)
478	.fill	512*EARLY_DYNAMIC_PAGE_TABLES,8,0
479SYM_DATA_END(early_dynamic_pgts)
480
481SYM_DATA(early_recursion_flag, .long 0)
482
483	.data
484
485#if defined(CONFIG_XEN_PV) || defined(CONFIG_PVH)
486SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
487	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
488	.org    init_top_pgt + L4_PAGE_OFFSET*8, 0
489	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
490	.org    init_top_pgt + L4_START_KERNEL*8, 0
491	/* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
492	.quad   level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
493	.fill	PTI_USER_PGD_FILL,8,0
494SYM_DATA_END(init_top_pgt)
495
496SYM_DATA_START_PAGE_ALIGNED(level3_ident_pgt)
497	.quad	level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
498	.fill	511, 8, 0
499SYM_DATA_END(level3_ident_pgt)
500SYM_DATA_START_PAGE_ALIGNED(level2_ident_pgt)
501	/*
502	 * Since I easily can, map the first 1G.
503	 * Don't set NX because code runs from these pages.
504	 *
505	 * Note: This sets _PAGE_GLOBAL despite whether
506	 * the CPU supports it or it is enabled.  But,
507	 * the CPU should ignore the bit.
508	 */
509	PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
510SYM_DATA_END(level2_ident_pgt)
511#else
512SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
513	.fill	512,8,0
514	.fill	PTI_USER_PGD_FILL,8,0
515SYM_DATA_END(init_top_pgt)
516#endif
517
518#ifdef CONFIG_X86_5LEVEL
519SYM_DATA_START_PAGE_ALIGNED(level4_kernel_pgt)
520	.fill	511,8,0
521	.quad	level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
522SYM_DATA_END(level4_kernel_pgt)
523#endif
524
525SYM_DATA_START_PAGE_ALIGNED(level3_kernel_pgt)
526	.fill	L3_START_KERNEL,8,0
527	/* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
528	.quad	level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
529	.quad	level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
530SYM_DATA_END(level3_kernel_pgt)
531
532SYM_DATA_START_PAGE_ALIGNED(level2_kernel_pgt)
533	/*
534	 * Kernel high mapping.
535	 *
536	 * The kernel code+data+bss must be located below KERNEL_IMAGE_SIZE in
537	 * virtual address space, which is 1 GiB if RANDOMIZE_BASE is enabled,
538	 * 512 MiB otherwise.
539	 *
540	 * (NOTE: after that starts the module area, see MODULES_VADDR.)
541	 *
542	 * This table is eventually used by the kernel during normal runtime.
543	 * Care must be taken to clear out undesired bits later, like _PAGE_RW
544	 * or _PAGE_GLOBAL in some cases.
545	 */
546	PMDS(0, __PAGE_KERNEL_LARGE_EXEC, KERNEL_IMAGE_SIZE/PMD_SIZE)
547SYM_DATA_END(level2_kernel_pgt)
548
549SYM_DATA_START_PAGE_ALIGNED(level2_fixmap_pgt)
550	.fill	(512 - 4 - FIXMAP_PMD_NUM),8,0
551	pgtno = 0
552	.rept (FIXMAP_PMD_NUM)
553	.quad level1_fixmap_pgt + (pgtno << PAGE_SHIFT) - __START_KERNEL_map \
554		+ _PAGE_TABLE_NOENC;
555	pgtno = pgtno + 1
556	.endr
557	/* 6 MB reserved space + a 2MB hole */
558	.fill	4,8,0
559SYM_DATA_END(level2_fixmap_pgt)
560
561SYM_DATA_START_PAGE_ALIGNED(level1_fixmap_pgt)
562	.rept (FIXMAP_PMD_NUM)
563	.fill	512,8,0
564	.endr
565SYM_DATA_END(level1_fixmap_pgt)
566
567#undef PMDS
568
569	.data
570	.align 16
571
572SYM_DATA(early_gdt_descr,		.word GDT_ENTRIES*8-1)
573SYM_DATA_LOCAL(early_gdt_descr_base,	.quad INIT_PER_CPU_VAR(gdt_page))
574
575	.align 16
576/* This must match the first entry in level2_kernel_pgt */
577SYM_DATA(phys_base, .quad 0x0)
578EXPORT_SYMBOL(phys_base)
579
580#include "../../x86/xen/xen-head.S"
581
582	__PAGE_ALIGNED_BSS
583SYM_DATA_START_PAGE_ALIGNED(empty_zero_page)
584	.skip PAGE_SIZE
585SYM_DATA_END(empty_zero_page)
586EXPORT_SYMBOL(empty_zero_page)
587
588