xref: /openbmc/linux/arch/x86/kernel/head_64.S (revision 2791f47d)
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#define GET_CR2_INTO(reg) GET_CR2_INTO_AX ; _ASM_MOV %_ASM_AX, reg
33#else
34#define INTERRUPT_RETURN iretq
35#define GET_CR2_INTO(reg) _ASM_MOV %cr2, reg
36#endif
37
38/*
39 * We are not able to switch in one step to the final KERNEL ADDRESS SPACE
40 * because we need identity-mapped pages.
41 */
42#define l4_index(x)	(((x) >> 39) & 511)
43#define pud_index(x)	(((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
44
45L4_PAGE_OFFSET = l4_index(__PAGE_OFFSET_BASE_L4)
46L4_START_KERNEL = l4_index(__START_KERNEL_map)
47
48L3_START_KERNEL = pud_index(__START_KERNEL_map)
49
50	.text
51	__HEAD
52	.code64
53SYM_CODE_START_NOALIGN(startup_64)
54	UNWIND_HINT_EMPTY
55	/*
56	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
57	 * and someone has loaded an identity mapped page table
58	 * for us.  These identity mapped page tables map all of the
59	 * kernel pages and possibly all of memory.
60	 *
61	 * %rsi holds a physical pointer to real_mode_data.
62	 *
63	 * We come here either directly from a 64bit bootloader, or from
64	 * arch/x86/boot/compressed/head_64.S.
65	 *
66	 * We only come here initially at boot nothing else comes here.
67	 *
68	 * Since we may be loaded at an address different from what we were
69	 * compiled to run at we first fixup the physical addresses in our page
70	 * tables and then reload them.
71	 */
72
73	/* Set up the stack for verify_cpu(), similar to initial_stack below */
74	leaq	(__end_init_task - SIZEOF_PTREGS)(%rip), %rsp
75
76	leaq	_text(%rip), %rdi
77	pushq	%rsi
78	call	startup_64_setup_env
79	popq	%rsi
80
81	/* Now switch to __KERNEL_CS so IRET works reliably */
82	pushq	$__KERNEL_CS
83	leaq	.Lon_kernel_cs(%rip), %rax
84	pushq	%rax
85	lretq
86
87.Lon_kernel_cs:
88	UNWIND_HINT_EMPTY
89
90	/* Sanitize CPU configuration */
91	call verify_cpu
92
93	/*
94	 * Perform pagetable fixups. Additionally, if SME is active, encrypt
95	 * the kernel and retrieve the modifier (SME encryption mask if SME
96	 * is active) to be added to the initial pgdir entry that will be
97	 * programmed into CR3.
98	 */
99	leaq	_text(%rip), %rdi
100	pushq	%rsi
101	call	__startup_64
102	popq	%rsi
103
104	/* Form the CR3 value being sure to include the CR3 modifier */
105	addq	$(early_top_pgt - __START_KERNEL_map), %rax
106	jmp 1f
107SYM_CODE_END(startup_64)
108
109SYM_CODE_START(secondary_startup_64)
110	UNWIND_HINT_EMPTY
111	/*
112	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
113	 * and someone has loaded a mapped page table.
114	 *
115	 * %rsi holds a physical pointer to real_mode_data.
116	 *
117	 * We come here either from startup_64 (using physical addresses)
118	 * or from trampoline.S (using virtual addresses).
119	 *
120	 * Using virtual addresses from trampoline.S removes the need
121	 * to have any identity mapped pages in the kernel page table
122	 * after the boot processor executes this code.
123	 */
124
125	/* Sanitize CPU configuration */
126	call verify_cpu
127
128	/*
129	 * The secondary_startup_64_no_verify entry point is only used by
130	 * SEV-ES guests. In those guests the call to verify_cpu() would cause
131	 * #VC exceptions which can not be handled at this stage of secondary
132	 * CPU bringup.
133	 *
134	 * All non SEV-ES systems, especially Intel systems, need to execute
135	 * verify_cpu() above to make sure NX is enabled.
136	 */
137SYM_INNER_LABEL(secondary_startup_64_no_verify, SYM_L_GLOBAL)
138	UNWIND_HINT_EMPTY
139
140	/*
141	 * Retrieve the modifier (SME encryption mask if SME is active) to be
142	 * added to the initial pgdir entry that will be programmed into CR3.
143	 */
144	pushq	%rsi
145	call	__startup_secondary_64
146	popq	%rsi
147
148	/* Form the CR3 value being sure to include the CR3 modifier */
149	addq	$(init_top_pgt - __START_KERNEL_map), %rax
1501:
151
152	/* Enable PAE mode, PGE and LA57 */
153	movl	$(X86_CR4_PAE | X86_CR4_PGE), %ecx
154#ifdef CONFIG_X86_5LEVEL
155	testl	$1, __pgtable_l5_enabled(%rip)
156	jz	1f
157	orl	$X86_CR4_LA57, %ecx
1581:
159#endif
160	movq	%rcx, %cr4
161
162	/* Setup early boot stage 4-/5-level pagetables. */
163	addq	phys_base(%rip), %rax
164	movq	%rax, %cr3
165
166	/* Ensure I am executing from virtual addresses */
167	movq	$1f, %rax
168	ANNOTATE_RETPOLINE_SAFE
169	jmp	*%rax
1701:
171	UNWIND_HINT_EMPTY
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	/*
208	 * Setup a boot time stack - Any secondary CPU will have lost its stack
209	 * by now because the cr3-switch above unmaps the real-mode stack
210	 */
211	movq initial_stack(%rip), %rsp
212
213	/* Setup and Load IDT */
214	pushq	%rsi
215	call	early_setup_idt
216	popq	%rsi
217
218	/* Check if nx is implemented */
219	movl	$0x80000001, %eax
220	cpuid
221	movl	%edx,%edi
222
223	/* Setup EFER (Extended Feature Enable Register) */
224	movl	$MSR_EFER, %ecx
225	rdmsr
226	btsl	$_EFER_SCE, %eax	/* Enable System Call */
227	btl	$20,%edi		/* No Execute supported? */
228	jnc     1f
229	btsl	$_EFER_NX, %eax
230	btsq	$_PAGE_BIT_NX,early_pmd_flags(%rip)
2311:	wrmsr				/* Make changes effective */
232
233	/* Setup cr0 */
234	movl	$CR0_STATE, %eax
235	/* Make changes effective */
236	movq	%rax, %cr0
237
238	/* zero EFLAGS after setting rsp */
239	pushq $0
240	popfq
241
242	/* rsi is pointer to real mode structure with interesting info.
243	   pass it to C */
244	movq	%rsi, %rdi
245
246.Ljump_to_C_code:
247	/*
248	 * Jump to run C code and to be on a real kernel address.
249	 * Since we are running on identity-mapped space we have to jump
250	 * to the full 64bit address, this is only possible as indirect
251	 * jump.  In addition we need to ensure %cs is set so we make this
252	 * a far return.
253	 *
254	 * Note: do not change to far jump indirect with 64bit offset.
255	 *
256	 * AMD does not support far jump indirect with 64bit offset.
257	 * AMD64 Architecture Programmer's Manual, Volume 3: states only
258	 *	JMP FAR mem16:16 FF /5 Far jump indirect,
259	 *		with the target specified by a far pointer in memory.
260	 *	JMP FAR mem16:32 FF /5 Far jump indirect,
261	 *		with the target specified by a far pointer in memory.
262	 *
263	 * Intel64 does support 64bit offset.
264	 * Software Developer Manual Vol 2: states:
265	 *	FF /5 JMP m16:16 Jump far, absolute indirect,
266	 *		address given in m16:16
267	 *	FF /5 JMP m16:32 Jump far, absolute indirect,
268	 *		address given in m16:32.
269	 *	REX.W + FF /5 JMP m16:64 Jump far, absolute indirect,
270	 *		address given in m16:64.
271	 */
272	pushq	$.Lafter_lret	# put return address on stack for unwinder
273	xorl	%ebp, %ebp	# clear frame pointer
274	movq	initial_code(%rip), %rax
275	pushq	$__KERNEL_CS	# set correct cs
276	pushq	%rax		# target address in negative space
277	lretq
278.Lafter_lret:
279SYM_CODE_END(secondary_startup_64)
280
281#include "verify_cpu.S"
282
283#ifdef CONFIG_HOTPLUG_CPU
284/*
285 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
286 * up already except stack. We just set up stack here. Then call
287 * start_secondary() via .Ljump_to_C_code.
288 */
289SYM_CODE_START(start_cpu0)
290	UNWIND_HINT_EMPTY
291	movq	initial_stack(%rip), %rsp
292	jmp	.Ljump_to_C_code
293SYM_CODE_END(start_cpu0)
294#endif
295
296#ifdef CONFIG_AMD_MEM_ENCRYPT
297/*
298 * VC Exception handler used during early boot when running on kernel
299 * addresses, but before the switch to the idt_table can be made.
300 * The early_idt_handler_array can't be used here because it calls into a lot
301 * of __init code and this handler is also used during CPU offlining/onlining.
302 * Therefore this handler ends up in the .text section so that it stays around
303 * when .init.text is freed.
304 */
305SYM_CODE_START_NOALIGN(vc_boot_ghcb)
306	UNWIND_HINT_IRET_REGS offset=8
307
308	/* Build pt_regs */
309	PUSH_AND_CLEAR_REGS
310
311	/* Call C handler */
312	movq    %rsp, %rdi
313	movq	ORIG_RAX(%rsp), %rsi
314	movq	initial_vc_handler(%rip), %rax
315	ANNOTATE_RETPOLINE_SAFE
316	call	*%rax
317
318	/* Unwind pt_regs */
319	POP_REGS
320
321	/* Remove Error Code */
322	addq    $8, %rsp
323
324	/* Pure iret required here - don't use INTERRUPT_RETURN */
325	iretq
326SYM_CODE_END(vc_boot_ghcb)
327#endif
328
329	/* Both SMP bootup and ACPI suspend change these variables */
330	__REFDATA
331	.balign	8
332SYM_DATA(initial_code,	.quad x86_64_start_kernel)
333SYM_DATA(initial_gs,	.quad INIT_PER_CPU_VAR(fixed_percpu_data))
334#ifdef CONFIG_AMD_MEM_ENCRYPT
335SYM_DATA(initial_vc_handler,	.quad handle_vc_boot_ghcb)
336#endif
337
338/*
339 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel unwinder
340 * reliably detect the end of the stack.
341 */
342SYM_DATA(initial_stack, .quad init_thread_union + THREAD_SIZE - SIZEOF_PTREGS)
343	__FINITDATA
344
345	__INIT
346SYM_CODE_START(early_idt_handler_array)
347	i = 0
348	.rept NUM_EXCEPTION_VECTORS
349	.if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
350		UNWIND_HINT_IRET_REGS
351		pushq $0	# Dummy error code, to make stack frame uniform
352	.else
353		UNWIND_HINT_IRET_REGS offset=8
354	.endif
355	pushq $i		# 72(%rsp) Vector number
356	jmp early_idt_handler_common
357	UNWIND_HINT_IRET_REGS
358	i = i + 1
359	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
360	.endr
361	UNWIND_HINT_IRET_REGS offset=16
362SYM_CODE_END(early_idt_handler_array)
363
364SYM_CODE_START_LOCAL(early_idt_handler_common)
365	/*
366	 * The stack is the hardware frame, an error code or zero, and the
367	 * vector number.
368	 */
369	cld
370
371	incl early_recursion_flag(%rip)
372
373	/* The vector number is currently in the pt_regs->di slot. */
374	pushq %rsi				/* pt_regs->si */
375	movq 8(%rsp), %rsi			/* RSI = vector number */
376	movq %rdi, 8(%rsp)			/* pt_regs->di = RDI */
377	pushq %rdx				/* pt_regs->dx */
378	pushq %rcx				/* pt_regs->cx */
379	pushq %rax				/* pt_regs->ax */
380	pushq %r8				/* pt_regs->r8 */
381	pushq %r9				/* pt_regs->r9 */
382	pushq %r10				/* pt_regs->r10 */
383	pushq %r11				/* pt_regs->r11 */
384	pushq %rbx				/* pt_regs->bx */
385	pushq %rbp				/* pt_regs->bp */
386	pushq %r12				/* pt_regs->r12 */
387	pushq %r13				/* pt_regs->r13 */
388	pushq %r14				/* pt_regs->r14 */
389	pushq %r15				/* pt_regs->r15 */
390	UNWIND_HINT_REGS
391
392	movq %rsp,%rdi		/* RDI = pt_regs; RSI is already trapnr */
393	call do_early_exception
394
395	decl early_recursion_flag(%rip)
396	jmp restore_regs_and_return_to_kernel
397SYM_CODE_END(early_idt_handler_common)
398
399#ifdef CONFIG_AMD_MEM_ENCRYPT
400/*
401 * VC Exception handler used during very early boot. The
402 * early_idt_handler_array can't be used because it returns via the
403 * paravirtualized INTERRUPT_RETURN and pv-ops don't work that early.
404 *
405 * This handler will end up in the .init.text section and not be
406 * available to boot secondary CPUs.
407 */
408SYM_CODE_START_NOALIGN(vc_no_ghcb)
409	UNWIND_HINT_IRET_REGS offset=8
410
411	/* Build pt_regs */
412	PUSH_AND_CLEAR_REGS
413
414	/* Call C handler */
415	movq    %rsp, %rdi
416	movq	ORIG_RAX(%rsp), %rsi
417	call    do_vc_no_ghcb
418
419	/* Unwind pt_regs */
420	POP_REGS
421
422	/* Remove Error Code */
423	addq    $8, %rsp
424
425	/* Pure iret required here - don't use INTERRUPT_RETURN */
426	iretq
427SYM_CODE_END(vc_no_ghcb)
428#endif
429
430#define SYM_DATA_START_PAGE_ALIGNED(name)			\
431	SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE)
432
433#ifdef CONFIG_PAGE_TABLE_ISOLATION
434/*
435 * Each PGD needs to be 8k long and 8k aligned.  We do not
436 * ever go out to userspace with these, so we do not
437 * strictly *need* the second page, but this allows us to
438 * have a single set_pgd() implementation that does not
439 * need to worry about whether it has 4k or 8k to work
440 * with.
441 *
442 * This ensures PGDs are 8k long:
443 */
444#define PTI_USER_PGD_FILL	512
445/* This ensures they are 8k-aligned: */
446#define SYM_DATA_START_PTI_ALIGNED(name) \
447	SYM_START(name, SYM_L_GLOBAL, .balign 2 * PAGE_SIZE)
448#else
449#define SYM_DATA_START_PTI_ALIGNED(name) \
450	SYM_DATA_START_PAGE_ALIGNED(name)
451#define PTI_USER_PGD_FILL	0
452#endif
453
454/* Automate the creation of 1 to 1 mapping pmd entries */
455#define PMDS(START, PERM, COUNT)			\
456	i = 0 ;						\
457	.rept (COUNT) ;					\
458	.quad	(START) + (i << PMD_SHIFT) + (PERM) ;	\
459	i = i + 1 ;					\
460	.endr
461
462	__INITDATA
463	.balign 4
464
465SYM_DATA_START_PTI_ALIGNED(early_top_pgt)
466	.fill	512,8,0
467	.fill	PTI_USER_PGD_FILL,8,0
468SYM_DATA_END(early_top_pgt)
469
470SYM_DATA_START_PAGE_ALIGNED(early_dynamic_pgts)
471	.fill	512*EARLY_DYNAMIC_PAGE_TABLES,8,0
472SYM_DATA_END(early_dynamic_pgts)
473
474SYM_DATA(early_recursion_flag, .long 0)
475
476	.data
477
478#if defined(CONFIG_XEN_PV) || defined(CONFIG_PVH)
479SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
480	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
481	.org    init_top_pgt + L4_PAGE_OFFSET*8, 0
482	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
483	.org    init_top_pgt + L4_START_KERNEL*8, 0
484	/* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
485	.quad   level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
486	.fill	PTI_USER_PGD_FILL,8,0
487SYM_DATA_END(init_top_pgt)
488
489SYM_DATA_START_PAGE_ALIGNED(level3_ident_pgt)
490	.quad	level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
491	.fill	511, 8, 0
492SYM_DATA_END(level3_ident_pgt)
493SYM_DATA_START_PAGE_ALIGNED(level2_ident_pgt)
494	/*
495	 * Since I easily can, map the first 1G.
496	 * Don't set NX because code runs from these pages.
497	 *
498	 * Note: This sets _PAGE_GLOBAL despite whether
499	 * the CPU supports it or it is enabled.  But,
500	 * the CPU should ignore the bit.
501	 */
502	PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
503SYM_DATA_END(level2_ident_pgt)
504#else
505SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
506	.fill	512,8,0
507	.fill	PTI_USER_PGD_FILL,8,0
508SYM_DATA_END(init_top_pgt)
509#endif
510
511#ifdef CONFIG_X86_5LEVEL
512SYM_DATA_START_PAGE_ALIGNED(level4_kernel_pgt)
513	.fill	511,8,0
514	.quad	level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
515SYM_DATA_END(level4_kernel_pgt)
516#endif
517
518SYM_DATA_START_PAGE_ALIGNED(level3_kernel_pgt)
519	.fill	L3_START_KERNEL,8,0
520	/* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
521	.quad	level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
522	.quad	level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
523SYM_DATA_END(level3_kernel_pgt)
524
525SYM_DATA_START_PAGE_ALIGNED(level2_kernel_pgt)
526	/*
527	 * 512 MB kernel mapping. We spend a full page on this pagetable
528	 * anyway.
529	 *
530	 * The kernel code+data+bss must not be bigger than that.
531	 *
532	 * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
533	 *  If you want to increase this then increase MODULES_VADDR
534	 *  too.)
535	 *
536	 *  This table is eventually used by the kernel during normal
537	 *  runtime.  Care must be taken to clear out undesired bits
538	 *  later, like _PAGE_RW or _PAGE_GLOBAL in some cases.
539	 */
540	PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
541		KERNEL_IMAGE_SIZE/PMD_SIZE)
542SYM_DATA_END(level2_kernel_pgt)
543
544SYM_DATA_START_PAGE_ALIGNED(level2_fixmap_pgt)
545	.fill	(512 - 4 - FIXMAP_PMD_NUM),8,0
546	pgtno = 0
547	.rept (FIXMAP_PMD_NUM)
548	.quad level1_fixmap_pgt + (pgtno << PAGE_SHIFT) - __START_KERNEL_map \
549		+ _PAGE_TABLE_NOENC;
550	pgtno = pgtno + 1
551	.endr
552	/* 6 MB reserved space + a 2MB hole */
553	.fill	4,8,0
554SYM_DATA_END(level2_fixmap_pgt)
555
556SYM_DATA_START_PAGE_ALIGNED(level1_fixmap_pgt)
557	.rept (FIXMAP_PMD_NUM)
558	.fill	512,8,0
559	.endr
560SYM_DATA_END(level1_fixmap_pgt)
561
562#undef PMDS
563
564	.data
565	.align 16
566
567SYM_DATA(early_gdt_descr,		.word GDT_ENTRIES*8-1)
568SYM_DATA_LOCAL(early_gdt_descr_base,	.quad INIT_PER_CPU_VAR(gdt_page))
569
570	.align 16
571/* This must match the first entry in level2_kernel_pgt */
572SYM_DATA(phys_base, .quad 0x0)
573EXPORT_SYMBOL(phys_base)
574
575#include "../../x86/xen/xen-head.S"
576
577	__PAGE_ALIGNED_BSS
578SYM_DATA_START_PAGE_ALIGNED(empty_zero_page)
579	.skip PAGE_SIZE
580SYM_DATA_END(empty_zero_page)
581EXPORT_SYMBOL(empty_zero_page)
582
583