xref: /openbmc/linux/arch/x86/kernel/head_64.S (revision d87c25e8)
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	/*
170	 * Switch to new page-table
171	 *
172	 * For the boot CPU this switches to early_top_pgt which still has the
173	 * indentity mappings present. The secondary CPUs will switch to the
174	 * init_top_pgt here, away from the trampoline_pgd and unmap the
175	 * indentity mapped ranges.
176	 */
177	movq	%rax, %cr3
178
179	/*
180	 * Do a global TLB flush after the CR3 switch to make sure the TLB
181	 * entries from the identity mapping are flushed.
182	 */
183	movq	%cr4, %rcx
184	movq	%rcx, %rax
185	xorq	$X86_CR4_PGE, %rcx
186	movq	%rcx, %cr4
187	movq	%rax, %cr4
188
189	/* Ensure I am executing from virtual addresses */
190	movq	$1f, %rax
191	ANNOTATE_RETPOLINE_SAFE
192	jmp	*%rax
1931:
194	UNWIND_HINT_EMPTY
195
196	/*
197	 * We must switch to a new descriptor in kernel space for the GDT
198	 * because soon the kernel won't have access anymore to the userspace
199	 * addresses where we're currently running on. We have to do that here
200	 * because in 32bit we couldn't load a 64bit linear address.
201	 */
202	lgdt	early_gdt_descr(%rip)
203
204	/* set up data segments */
205	xorl %eax,%eax
206	movl %eax,%ds
207	movl %eax,%ss
208	movl %eax,%es
209
210	/*
211	 * We don't really need to load %fs or %gs, but load them anyway
212	 * to kill any stale realmode selectors.  This allows execution
213	 * under VT hardware.
214	 */
215	movl %eax,%fs
216	movl %eax,%gs
217
218	/* Set up %gs.
219	 *
220	 * The base of %gs always points to fixed_percpu_data. If the
221	 * stack protector canary is enabled, it is located at %gs:40.
222	 * Note that, on SMP, the boot cpu uses init data section until
223	 * the per cpu areas are set up.
224	 */
225	movl	$MSR_GS_BASE,%ecx
226	movl	initial_gs(%rip),%eax
227	movl	initial_gs+4(%rip),%edx
228	wrmsr
229
230	/*
231	 * Setup a boot time stack - Any secondary CPU will have lost its stack
232	 * by now because the cr3-switch above unmaps the real-mode stack
233	 */
234	movq initial_stack(%rip), %rsp
235
236	/* Setup and Load IDT */
237	pushq	%rsi
238	call	early_setup_idt
239	popq	%rsi
240
241	/* Check if nx is implemented */
242	movl	$0x80000001, %eax
243	cpuid
244	movl	%edx,%edi
245
246	/* Setup EFER (Extended Feature Enable Register) */
247	movl	$MSR_EFER, %ecx
248	rdmsr
249	btsl	$_EFER_SCE, %eax	/* Enable System Call */
250	btl	$20,%edi		/* No Execute supported? */
251	jnc     1f
252	btsl	$_EFER_NX, %eax
253	btsq	$_PAGE_BIT_NX,early_pmd_flags(%rip)
2541:	wrmsr				/* Make changes effective */
255
256	/* Setup cr0 */
257	movl	$CR0_STATE, %eax
258	/* Make changes effective */
259	movq	%rax, %cr0
260
261	/* zero EFLAGS after setting rsp */
262	pushq $0
263	popfq
264
265	/* rsi is pointer to real mode structure with interesting info.
266	   pass it to C */
267	movq	%rsi, %rdi
268
269.Ljump_to_C_code:
270	/*
271	 * Jump to run C code and to be on a real kernel address.
272	 * Since we are running on identity-mapped space we have to jump
273	 * to the full 64bit address, this is only possible as indirect
274	 * jump.  In addition we need to ensure %cs is set so we make this
275	 * a far return.
276	 *
277	 * Note: do not change to far jump indirect with 64bit offset.
278	 *
279	 * AMD does not support far jump indirect with 64bit offset.
280	 * AMD64 Architecture Programmer's Manual, Volume 3: states only
281	 *	JMP FAR mem16:16 FF /5 Far jump indirect,
282	 *		with the target specified by a far pointer in memory.
283	 *	JMP FAR mem16:32 FF /5 Far jump indirect,
284	 *		with the target specified by a far pointer in memory.
285	 *
286	 * Intel64 does support 64bit offset.
287	 * Software Developer Manual Vol 2: states:
288	 *	FF /5 JMP m16:16 Jump far, absolute indirect,
289	 *		address given in m16:16
290	 *	FF /5 JMP m16:32 Jump far, absolute indirect,
291	 *		address given in m16:32.
292	 *	REX.W + FF /5 JMP m16:64 Jump far, absolute indirect,
293	 *		address given in m16:64.
294	 */
295	pushq	$.Lafter_lret	# put return address on stack for unwinder
296	xorl	%ebp, %ebp	# clear frame pointer
297	movq	initial_code(%rip), %rax
298	pushq	$__KERNEL_CS	# set correct cs
299	pushq	%rax		# target address in negative space
300	lretq
301.Lafter_lret:
302SYM_CODE_END(secondary_startup_64)
303
304#include "verify_cpu.S"
305#include "sev_verify_cbit.S"
306
307#ifdef CONFIG_HOTPLUG_CPU
308/*
309 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
310 * up already except stack. We just set up stack here. Then call
311 * start_secondary() via .Ljump_to_C_code.
312 */
313SYM_CODE_START(start_cpu0)
314	UNWIND_HINT_EMPTY
315	movq	initial_stack(%rip), %rsp
316	jmp	.Ljump_to_C_code
317SYM_CODE_END(start_cpu0)
318#endif
319
320#ifdef CONFIG_AMD_MEM_ENCRYPT
321/*
322 * VC Exception handler used during early boot when running on kernel
323 * addresses, but before the switch to the idt_table can be made.
324 * The early_idt_handler_array can't be used here because it calls into a lot
325 * of __init code and this handler is also used during CPU offlining/onlining.
326 * Therefore this handler ends up in the .text section so that it stays around
327 * when .init.text is freed.
328 */
329SYM_CODE_START_NOALIGN(vc_boot_ghcb)
330	UNWIND_HINT_IRET_REGS offset=8
331
332	/* Build pt_regs */
333	PUSH_AND_CLEAR_REGS
334
335	/* Call C handler */
336	movq    %rsp, %rdi
337	movq	ORIG_RAX(%rsp), %rsi
338	movq	initial_vc_handler(%rip), %rax
339	ANNOTATE_RETPOLINE_SAFE
340	call	*%rax
341
342	/* Unwind pt_regs */
343	POP_REGS
344
345	/* Remove Error Code */
346	addq    $8, %rsp
347
348	/* Pure iret required here - don't use INTERRUPT_RETURN */
349	iretq
350SYM_CODE_END(vc_boot_ghcb)
351#endif
352
353	/* Both SMP bootup and ACPI suspend change these variables */
354	__REFDATA
355	.balign	8
356SYM_DATA(initial_code,	.quad x86_64_start_kernel)
357SYM_DATA(initial_gs,	.quad INIT_PER_CPU_VAR(fixed_percpu_data))
358#ifdef CONFIG_AMD_MEM_ENCRYPT
359SYM_DATA(initial_vc_handler,	.quad handle_vc_boot_ghcb)
360#endif
361
362/*
363 * The FRAME_SIZE gap is a convention which helps the in-kernel unwinder
364 * reliably detect the end of the stack.
365 */
366SYM_DATA(initial_stack, .quad init_thread_union + THREAD_SIZE - FRAME_SIZE)
367	__FINITDATA
368
369	__INIT
370SYM_CODE_START(early_idt_handler_array)
371	i = 0
372	.rept NUM_EXCEPTION_VECTORS
373	.if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
374		UNWIND_HINT_IRET_REGS
375		pushq $0	# Dummy error code, to make stack frame uniform
376	.else
377		UNWIND_HINT_IRET_REGS offset=8
378	.endif
379	pushq $i		# 72(%rsp) Vector number
380	jmp early_idt_handler_common
381	UNWIND_HINT_IRET_REGS
382	i = i + 1
383	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
384	.endr
385	UNWIND_HINT_IRET_REGS offset=16
386SYM_CODE_END(early_idt_handler_array)
387
388SYM_CODE_START_LOCAL(early_idt_handler_common)
389	/*
390	 * The stack is the hardware frame, an error code or zero, and the
391	 * vector number.
392	 */
393	cld
394
395	incl early_recursion_flag(%rip)
396
397	/* The vector number is currently in the pt_regs->di slot. */
398	pushq %rsi				/* pt_regs->si */
399	movq 8(%rsp), %rsi			/* RSI = vector number */
400	movq %rdi, 8(%rsp)			/* pt_regs->di = RDI */
401	pushq %rdx				/* pt_regs->dx */
402	pushq %rcx				/* pt_regs->cx */
403	pushq %rax				/* pt_regs->ax */
404	pushq %r8				/* pt_regs->r8 */
405	pushq %r9				/* pt_regs->r9 */
406	pushq %r10				/* pt_regs->r10 */
407	pushq %r11				/* pt_regs->r11 */
408	pushq %rbx				/* pt_regs->bx */
409	pushq %rbp				/* pt_regs->bp */
410	pushq %r12				/* pt_regs->r12 */
411	pushq %r13				/* pt_regs->r13 */
412	pushq %r14				/* pt_regs->r14 */
413	pushq %r15				/* pt_regs->r15 */
414	UNWIND_HINT_REGS
415
416	movq %rsp,%rdi		/* RDI = pt_regs; RSI is already trapnr */
417	call do_early_exception
418
419	decl early_recursion_flag(%rip)
420	jmp restore_regs_and_return_to_kernel
421SYM_CODE_END(early_idt_handler_common)
422
423#ifdef CONFIG_AMD_MEM_ENCRYPT
424/*
425 * VC Exception handler used during very early boot. The
426 * early_idt_handler_array can't be used because it returns via the
427 * paravirtualized INTERRUPT_RETURN and pv-ops don't work that early.
428 *
429 * This handler will end up in the .init.text section and not be
430 * available to boot secondary CPUs.
431 */
432SYM_CODE_START_NOALIGN(vc_no_ghcb)
433	UNWIND_HINT_IRET_REGS offset=8
434
435	/* Build pt_regs */
436	PUSH_AND_CLEAR_REGS
437
438	/* Call C handler */
439	movq    %rsp, %rdi
440	movq	ORIG_RAX(%rsp), %rsi
441	call    do_vc_no_ghcb
442
443	/* Unwind pt_regs */
444	POP_REGS
445
446	/* Remove Error Code */
447	addq    $8, %rsp
448
449	/* Pure iret required here - don't use INTERRUPT_RETURN */
450	iretq
451SYM_CODE_END(vc_no_ghcb)
452#endif
453
454#define SYM_DATA_START_PAGE_ALIGNED(name)			\
455	SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE)
456
457#ifdef CONFIG_PAGE_TABLE_ISOLATION
458/*
459 * Each PGD needs to be 8k long and 8k aligned.  We do not
460 * ever go out to userspace with these, so we do not
461 * strictly *need* the second page, but this allows us to
462 * have a single set_pgd() implementation that does not
463 * need to worry about whether it has 4k or 8k to work
464 * with.
465 *
466 * This ensures PGDs are 8k long:
467 */
468#define PTI_USER_PGD_FILL	512
469/* This ensures they are 8k-aligned: */
470#define SYM_DATA_START_PTI_ALIGNED(name) \
471	SYM_START(name, SYM_L_GLOBAL, .balign 2 * PAGE_SIZE)
472#else
473#define SYM_DATA_START_PTI_ALIGNED(name) \
474	SYM_DATA_START_PAGE_ALIGNED(name)
475#define PTI_USER_PGD_FILL	0
476#endif
477
478/* Automate the creation of 1 to 1 mapping pmd entries */
479#define PMDS(START, PERM, COUNT)			\
480	i = 0 ;						\
481	.rept (COUNT) ;					\
482	.quad	(START) + (i << PMD_SHIFT) + (PERM) ;	\
483	i = i + 1 ;					\
484	.endr
485
486	__INITDATA
487	.balign 4
488
489SYM_DATA_START_PTI_ALIGNED(early_top_pgt)
490	.fill	512,8,0
491	.fill	PTI_USER_PGD_FILL,8,0
492SYM_DATA_END(early_top_pgt)
493
494SYM_DATA_START_PAGE_ALIGNED(early_dynamic_pgts)
495	.fill	512*EARLY_DYNAMIC_PAGE_TABLES,8,0
496SYM_DATA_END(early_dynamic_pgts)
497
498SYM_DATA(early_recursion_flag, .long 0)
499
500	.data
501
502#if defined(CONFIG_XEN_PV) || defined(CONFIG_PVH)
503SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
504	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
505	.org    init_top_pgt + L4_PAGE_OFFSET*8, 0
506	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
507	.org    init_top_pgt + L4_START_KERNEL*8, 0
508	/* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
509	.quad   level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
510	.fill	PTI_USER_PGD_FILL,8,0
511SYM_DATA_END(init_top_pgt)
512
513SYM_DATA_START_PAGE_ALIGNED(level3_ident_pgt)
514	.quad	level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
515	.fill	511, 8, 0
516SYM_DATA_END(level3_ident_pgt)
517SYM_DATA_START_PAGE_ALIGNED(level2_ident_pgt)
518	/*
519	 * Since I easily can, map the first 1G.
520	 * Don't set NX because code runs from these pages.
521	 *
522	 * Note: This sets _PAGE_GLOBAL despite whether
523	 * the CPU supports it or it is enabled.  But,
524	 * the CPU should ignore the bit.
525	 */
526	PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
527SYM_DATA_END(level2_ident_pgt)
528#else
529SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
530	.fill	512,8,0
531	.fill	PTI_USER_PGD_FILL,8,0
532SYM_DATA_END(init_top_pgt)
533#endif
534
535#ifdef CONFIG_X86_5LEVEL
536SYM_DATA_START_PAGE_ALIGNED(level4_kernel_pgt)
537	.fill	511,8,0
538	.quad	level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
539SYM_DATA_END(level4_kernel_pgt)
540#endif
541
542SYM_DATA_START_PAGE_ALIGNED(level3_kernel_pgt)
543	.fill	L3_START_KERNEL,8,0
544	/* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
545	.quad	level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
546	.quad	level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
547SYM_DATA_END(level3_kernel_pgt)
548
549SYM_DATA_START_PAGE_ALIGNED(level2_kernel_pgt)
550	/*
551	 * Kernel high mapping.
552	 *
553	 * The kernel code+data+bss must be located below KERNEL_IMAGE_SIZE in
554	 * virtual address space, which is 1 GiB if RANDOMIZE_BASE is enabled,
555	 * 512 MiB otherwise.
556	 *
557	 * (NOTE: after that starts the module area, see MODULES_VADDR.)
558	 *
559	 * This table is eventually used by the kernel during normal runtime.
560	 * Care must be taken to clear out undesired bits later, like _PAGE_RW
561	 * or _PAGE_GLOBAL in some cases.
562	 */
563	PMDS(0, __PAGE_KERNEL_LARGE_EXEC, KERNEL_IMAGE_SIZE/PMD_SIZE)
564SYM_DATA_END(level2_kernel_pgt)
565
566SYM_DATA_START_PAGE_ALIGNED(level2_fixmap_pgt)
567	.fill	(512 - 4 - FIXMAP_PMD_NUM),8,0
568	pgtno = 0
569	.rept (FIXMAP_PMD_NUM)
570	.quad level1_fixmap_pgt + (pgtno << PAGE_SHIFT) - __START_KERNEL_map \
571		+ _PAGE_TABLE_NOENC;
572	pgtno = pgtno + 1
573	.endr
574	/* 6 MB reserved space + a 2MB hole */
575	.fill	4,8,0
576SYM_DATA_END(level2_fixmap_pgt)
577
578SYM_DATA_START_PAGE_ALIGNED(level1_fixmap_pgt)
579	.rept (FIXMAP_PMD_NUM)
580	.fill	512,8,0
581	.endr
582SYM_DATA_END(level1_fixmap_pgt)
583
584#undef PMDS
585
586	.data
587	.align 16
588
589SYM_DATA(early_gdt_descr,		.word GDT_ENTRIES*8-1)
590SYM_DATA_LOCAL(early_gdt_descr_base,	.quad INIT_PER_CPU_VAR(gdt_page))
591
592	.align 16
593/* This must match the first entry in level2_kernel_pgt */
594SYM_DATA(phys_base, .quad 0x0)
595EXPORT_SYMBOL(phys_base)
596
597#include "../../x86/xen/xen-head.S"
598
599	__PAGE_ALIGNED_BSS
600SYM_DATA_START_PAGE_ALIGNED(empty_zero_page)
601	.skip PAGE_SIZE
602SYM_DATA_END(empty_zero_page)
603EXPORT_SYMBOL(empty_zero_page)
604
605