xref: /openbmc/linux/arch/x86/kernel/head_64.S (revision 3db55767)
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_END_OF_STACK
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() */
65	leaq	(__end_init_task - PTREGS_SIZE)(%rip), %rsp
66
67	leaq	_text(%rip), %rdi
68
69	/* Setup GSBASE to allow stack canary access for C code */
70	movl	$MSR_GS_BASE, %ecx
71	leaq	INIT_PER_CPU_VAR(fixed_percpu_data)(%rip), %rdx
72	movl	%edx, %eax
73	shrq	$32,  %rdx
74	wrmsr
75
76	pushq	%rsi
77	call	startup_64_setup_env
78	popq	%rsi
79
80#ifdef CONFIG_AMD_MEM_ENCRYPT
81	/*
82	 * Activate SEV/SME memory encryption if supported/enabled. This needs to
83	 * be done now, since this also includes setup of the SEV-SNP CPUID table,
84	 * which needs to be done before any CPUID instructions are executed in
85	 * subsequent code.
86	 */
87	movq	%rsi, %rdi
88	pushq	%rsi
89	call	sme_enable
90	popq	%rsi
91#endif
92
93	/* Now switch to __KERNEL_CS so IRET works reliably */
94	pushq	$__KERNEL_CS
95	leaq	.Lon_kernel_cs(%rip), %rax
96	pushq	%rax
97	lretq
98
99.Lon_kernel_cs:
100	UNWIND_HINT_END_OF_STACK
101
102	/* Sanitize CPU configuration */
103	call verify_cpu
104
105	/*
106	 * Perform pagetable fixups. Additionally, if SME is active, encrypt
107	 * the kernel and retrieve the modifier (SME encryption mask if SME
108	 * is active) to be added to the initial pgdir entry that will be
109	 * programmed into CR3.
110	 */
111	leaq	_text(%rip), %rdi
112	pushq	%rsi
113	call	__startup_64
114	popq	%rsi
115
116	/* Form the CR3 value being sure to include the CR3 modifier */
117	addq	$(early_top_pgt - __START_KERNEL_map), %rax
118	jmp 1f
119SYM_CODE_END(startup_64)
120
121SYM_CODE_START(secondary_startup_64)
122	UNWIND_HINT_END_OF_STACK
123	ANNOTATE_NOENDBR
124	/*
125	 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
126	 * and someone has loaded a mapped page table.
127	 *
128	 * %rsi holds a physical pointer to real_mode_data.
129	 *
130	 * We come here either from startup_64 (using physical addresses)
131	 * or from trampoline.S (using virtual addresses).
132	 *
133	 * Using virtual addresses from trampoline.S removes the need
134	 * to have any identity mapped pages in the kernel page table
135	 * after the boot processor executes this code.
136	 */
137
138	/* Sanitize CPU configuration */
139	call verify_cpu
140
141	/*
142	 * The secondary_startup_64_no_verify entry point is only used by
143	 * SEV-ES guests. In those guests the call to verify_cpu() would cause
144	 * #VC exceptions which can not be handled at this stage of secondary
145	 * CPU bringup.
146	 *
147	 * All non SEV-ES systems, especially Intel systems, need to execute
148	 * verify_cpu() above to make sure NX is enabled.
149	 */
150SYM_INNER_LABEL(secondary_startup_64_no_verify, SYM_L_GLOBAL)
151	UNWIND_HINT_END_OF_STACK
152	ANNOTATE_NOENDBR
153
154	/*
155	 * Retrieve the modifier (SME encryption mask if SME is active) to be
156	 * added to the initial pgdir entry that will be programmed into CR3.
157	 */
158#ifdef CONFIG_AMD_MEM_ENCRYPT
159	movq	sme_me_mask, %rax
160#else
161	xorq	%rax, %rax
162#endif
163
164	/* Form the CR3 value being sure to include the CR3 modifier */
165	addq	$(init_top_pgt - __START_KERNEL_map), %rax
1661:
167
168#ifdef CONFIG_X86_MCE
169	/*
170	 * Preserve CR4.MCE if the kernel will enable #MC support.
171	 * Clearing MCE may fault in some environments (that also force #MC
172	 * support). Any machine check that occurs before #MC support is fully
173	 * configured will crash the system regardless of the CR4.MCE value set
174	 * here.
175	 */
176	movq	%cr4, %rcx
177	andl	$X86_CR4_MCE, %ecx
178#else
179	movl	$0, %ecx
180#endif
181
182	/* Enable PAE mode, PGE and LA57 */
183	orl	$(X86_CR4_PAE | X86_CR4_PGE), %ecx
184#ifdef CONFIG_X86_5LEVEL
185	testl	$1, __pgtable_l5_enabled(%rip)
186	jz	1f
187	orl	$X86_CR4_LA57, %ecx
1881:
189#endif
190	movq	%rcx, %cr4
191
192	/* Setup early boot stage 4-/5-level pagetables. */
193	addq	phys_base(%rip), %rax
194
195	/*
196	 * For SEV guests: Verify that the C-bit is correct. A malicious
197	 * hypervisor could lie about the C-bit position to perform a ROP
198	 * attack on the guest by writing to the unencrypted stack and wait for
199	 * the next RET instruction.
200	 * %rsi carries pointer to realmode data and is callee-clobbered. Save
201	 * and restore it.
202	 */
203	pushq	%rsi
204	movq	%rax, %rdi
205	call	sev_verify_cbit
206	popq	%rsi
207
208	/*
209	 * Switch to new page-table
210	 *
211	 * For the boot CPU this switches to early_top_pgt which still has the
212	 * indentity mappings present. The secondary CPUs will switch to the
213	 * init_top_pgt here, away from the trampoline_pgd and unmap the
214	 * indentity mapped ranges.
215	 */
216	movq	%rax, %cr3
217
218	/*
219	 * Do a global TLB flush after the CR3 switch to make sure the TLB
220	 * entries from the identity mapping are flushed.
221	 */
222	movq	%cr4, %rcx
223	movq	%rcx, %rax
224	xorq	$X86_CR4_PGE, %rcx
225	movq	%rcx, %cr4
226	movq	%rax, %cr4
227
228	/* Ensure I am executing from virtual addresses */
229	movq	$1f, %rax
230	ANNOTATE_RETPOLINE_SAFE
231	jmp	*%rax
2321:
233	UNWIND_HINT_END_OF_STACK
234	ANNOTATE_NOENDBR // above
235
236#ifdef CONFIG_SMP
237	movl	smpboot_control(%rip), %ecx
238
239	/* Get the per cpu offset for the given CPU# which is in ECX */
240	movq	__per_cpu_offset(,%rcx,8), %rdx
241#else
242	xorl	%edx, %edx /* zero-extended to clear all of RDX */
243#endif /* CONFIG_SMP */
244
245	/*
246	 * Setup a boot time stack - Any secondary CPU will have lost its stack
247	 * by now because the cr3-switch above unmaps the real-mode stack.
248	 *
249	 * RDX contains the per-cpu offset
250	 */
251	movq	pcpu_hot + X86_current_task(%rdx), %rax
252	movq	TASK_threadsp(%rax), %rsp
253
254	/*
255	 * We must switch to a new descriptor in kernel space for the GDT
256	 * because soon the kernel won't have access anymore to the userspace
257	 * addresses where we're currently running on. We have to do that here
258	 * because in 32bit we couldn't load a 64bit linear address.
259	 */
260	subq	$16, %rsp
261	movw	$(GDT_SIZE-1), (%rsp)
262	leaq	gdt_page(%rdx), %rax
263	movq	%rax, 2(%rsp)
264	lgdt	(%rsp)
265	addq	$16, %rsp
266
267	/* set up data segments */
268	xorl %eax,%eax
269	movl %eax,%ds
270	movl %eax,%ss
271	movl %eax,%es
272
273	/*
274	 * We don't really need to load %fs or %gs, but load them anyway
275	 * to kill any stale realmode selectors.  This allows execution
276	 * under VT hardware.
277	 */
278	movl %eax,%fs
279	movl %eax,%gs
280
281	/* Set up %gs.
282	 *
283	 * The base of %gs always points to fixed_percpu_data. If the
284	 * stack protector canary is enabled, it is located at %gs:40.
285	 * Note that, on SMP, the boot cpu uses init data section until
286	 * the per cpu areas are set up.
287	 */
288	movl	$MSR_GS_BASE,%ecx
289#ifndef CONFIG_SMP
290	leaq	INIT_PER_CPU_VAR(fixed_percpu_data)(%rip), %rdx
291#endif
292	movl	%edx, %eax
293	shrq	$32, %rdx
294	wrmsr
295
296	/* Setup and Load IDT */
297	pushq	%rsi
298	call	early_setup_idt
299	popq	%rsi
300
301	/* Check if nx is implemented */
302	movl	$0x80000001, %eax
303	cpuid
304	movl	%edx,%edi
305
306	/* Setup EFER (Extended Feature Enable Register) */
307	movl	$MSR_EFER, %ecx
308	rdmsr
309	/*
310	 * Preserve current value of EFER for comparison and to skip
311	 * EFER writes if no change was made (for TDX guest)
312	 */
313	movl    %eax, %edx
314	btsl	$_EFER_SCE, %eax	/* Enable System Call */
315	btl	$20,%edi		/* No Execute supported? */
316	jnc     1f
317	btsl	$_EFER_NX, %eax
318	btsq	$_PAGE_BIT_NX,early_pmd_flags(%rip)
319
320	/* Avoid writing EFER if no change was made (for TDX guest) */
3211:	cmpl	%edx, %eax
322	je	1f
323	xor	%edx, %edx
324	wrmsr				/* Make changes effective */
3251:
326	/* Setup cr0 */
327	movl	$CR0_STATE, %eax
328	/* Make changes effective */
329	movq	%rax, %cr0
330
331	/* zero EFLAGS after setting rsp */
332	pushq $0
333	popfq
334
335	/* rsi is pointer to real mode structure with interesting info.
336	   pass it to C */
337	movq	%rsi, %rdi
338
339.Ljump_to_C_code:
340	/*
341	 * Jump to run C code and to be on a real kernel address.
342	 * Since we are running on identity-mapped space we have to jump
343	 * to the full 64bit address, this is only possible as indirect
344	 * jump.  In addition we need to ensure %cs is set so we make this
345	 * a far return.
346	 *
347	 * Note: do not change to far jump indirect with 64bit offset.
348	 *
349	 * AMD does not support far jump indirect with 64bit offset.
350	 * AMD64 Architecture Programmer's Manual, Volume 3: states only
351	 *	JMP FAR mem16:16 FF /5 Far jump indirect,
352	 *		with the target specified by a far pointer in memory.
353	 *	JMP FAR mem16:32 FF /5 Far jump indirect,
354	 *		with the target specified by a far pointer in memory.
355	 *
356	 * Intel64 does support 64bit offset.
357	 * Software Developer Manual Vol 2: states:
358	 *	FF /5 JMP m16:16 Jump far, absolute indirect,
359	 *		address given in m16:16
360	 *	FF /5 JMP m16:32 Jump far, absolute indirect,
361	 *		address given in m16:32.
362	 *	REX.W + FF /5 JMP m16:64 Jump far, absolute indirect,
363	 *		address given in m16:64.
364	 */
365	pushq	$.Lafter_lret	# put return address on stack for unwinder
366	xorl	%ebp, %ebp	# clear frame pointer
367	movq	initial_code(%rip), %rax
368	pushq	$__KERNEL_CS	# set correct cs
369	pushq	%rax		# target address in negative space
370	lretq
371.Lafter_lret:
372	ANNOTATE_NOENDBR
373SYM_CODE_END(secondary_startup_64)
374
375#include "verify_cpu.S"
376#include "sev_verify_cbit.S"
377
378#ifdef CONFIG_HOTPLUG_CPU
379/*
380 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
381 * up already except stack. We just set up stack here. Then call
382 * start_secondary() via .Ljump_to_C_code.
383 */
384SYM_CODE_START(start_cpu0)
385	ANNOTATE_NOENDBR
386	UNWIND_HINT_END_OF_STACK
387
388	/* Find the idle task stack */
389	movq	PER_CPU_VAR(pcpu_hot) + X86_current_task, %rcx
390	movq	TASK_threadsp(%rcx), %rsp
391
392	jmp	.Ljump_to_C_code
393SYM_CODE_END(start_cpu0)
394#endif
395
396#ifdef CONFIG_AMD_MEM_ENCRYPT
397/*
398 * VC Exception handler used during early boot when running on kernel
399 * addresses, but before the switch to the idt_table can be made.
400 * The early_idt_handler_array can't be used here because it calls into a lot
401 * of __init code and this handler is also used during CPU offlining/onlining.
402 * Therefore this handler ends up in the .text section so that it stays around
403 * when .init.text is freed.
404 */
405SYM_CODE_START_NOALIGN(vc_boot_ghcb)
406	UNWIND_HINT_IRET_REGS offset=8
407	ENDBR
408
409	/* Build pt_regs */
410	PUSH_AND_CLEAR_REGS
411
412	/* Call C handler */
413	movq    %rsp, %rdi
414	movq	ORIG_RAX(%rsp), %rsi
415	movq	initial_vc_handler(%rip), %rax
416	ANNOTATE_RETPOLINE_SAFE
417	call	*%rax
418
419	/* Unwind pt_regs */
420	POP_REGS
421
422	/* Remove Error Code */
423	addq    $8, %rsp
424
425	iretq
426SYM_CODE_END(vc_boot_ghcb)
427#endif
428
429	/* Both SMP bootup and ACPI suspend change these variables */
430	__REFDATA
431	.balign	8
432SYM_DATA(initial_code,	.quad x86_64_start_kernel)
433#ifdef CONFIG_AMD_MEM_ENCRYPT
434SYM_DATA(initial_vc_handler,	.quad handle_vc_boot_ghcb)
435#endif
436	__FINITDATA
437
438	__INIT
439SYM_CODE_START(early_idt_handler_array)
440	i = 0
441	.rept NUM_EXCEPTION_VECTORS
442	.if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
443		UNWIND_HINT_IRET_REGS
444		ENDBR
445		pushq $0	# Dummy error code, to make stack frame uniform
446	.else
447		UNWIND_HINT_IRET_REGS offset=8
448		ENDBR
449	.endif
450	pushq $i		# 72(%rsp) Vector number
451	jmp early_idt_handler_common
452	UNWIND_HINT_IRET_REGS
453	i = i + 1
454	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
455	.endr
456SYM_CODE_END(early_idt_handler_array)
457	ANNOTATE_NOENDBR // early_idt_handler_array[NUM_EXCEPTION_VECTORS]
458
459SYM_CODE_START_LOCAL(early_idt_handler_common)
460	UNWIND_HINT_IRET_REGS offset=16
461	/*
462	 * The stack is the hardware frame, an error code or zero, and the
463	 * vector number.
464	 */
465	cld
466
467	incl early_recursion_flag(%rip)
468
469	/* The vector number is currently in the pt_regs->di slot. */
470	pushq %rsi				/* pt_regs->si */
471	movq 8(%rsp), %rsi			/* RSI = vector number */
472	movq %rdi, 8(%rsp)			/* pt_regs->di = RDI */
473	pushq %rdx				/* pt_regs->dx */
474	pushq %rcx				/* pt_regs->cx */
475	pushq %rax				/* pt_regs->ax */
476	pushq %r8				/* pt_regs->r8 */
477	pushq %r9				/* pt_regs->r9 */
478	pushq %r10				/* pt_regs->r10 */
479	pushq %r11				/* pt_regs->r11 */
480	pushq %rbx				/* pt_regs->bx */
481	pushq %rbp				/* pt_regs->bp */
482	pushq %r12				/* pt_regs->r12 */
483	pushq %r13				/* pt_regs->r13 */
484	pushq %r14				/* pt_regs->r14 */
485	pushq %r15				/* pt_regs->r15 */
486	UNWIND_HINT_REGS
487
488	movq %rsp,%rdi		/* RDI = pt_regs; RSI is already trapnr */
489	call do_early_exception
490
491	decl early_recursion_flag(%rip)
492	jmp restore_regs_and_return_to_kernel
493SYM_CODE_END(early_idt_handler_common)
494
495#ifdef CONFIG_AMD_MEM_ENCRYPT
496/*
497 * VC Exception handler used during very early boot. The
498 * early_idt_handler_array can't be used because it returns via the
499 * paravirtualized INTERRUPT_RETURN and pv-ops don't work that early.
500 *
501 * XXX it does, fix this.
502 *
503 * This handler will end up in the .init.text section and not be
504 * available to boot secondary CPUs.
505 */
506SYM_CODE_START_NOALIGN(vc_no_ghcb)
507	UNWIND_HINT_IRET_REGS offset=8
508	ENDBR
509
510	/* Build pt_regs */
511	PUSH_AND_CLEAR_REGS
512
513	/* Call C handler */
514	movq    %rsp, %rdi
515	movq	ORIG_RAX(%rsp), %rsi
516	call    do_vc_no_ghcb
517
518	/* Unwind pt_regs */
519	POP_REGS
520
521	/* Remove Error Code */
522	addq    $8, %rsp
523
524	/* Pure iret required here - don't use INTERRUPT_RETURN */
525	iretq
526SYM_CODE_END(vc_no_ghcb)
527#endif
528
529#define SYM_DATA_START_PAGE_ALIGNED(name)			\
530	SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE)
531
532#ifdef CONFIG_PAGE_TABLE_ISOLATION
533/*
534 * Each PGD needs to be 8k long and 8k aligned.  We do not
535 * ever go out to userspace with these, so we do not
536 * strictly *need* the second page, but this allows us to
537 * have a single set_pgd() implementation that does not
538 * need to worry about whether it has 4k or 8k to work
539 * with.
540 *
541 * This ensures PGDs are 8k long:
542 */
543#define PTI_USER_PGD_FILL	512
544/* This ensures they are 8k-aligned: */
545#define SYM_DATA_START_PTI_ALIGNED(name) \
546	SYM_START(name, SYM_L_GLOBAL, .balign 2 * PAGE_SIZE)
547#else
548#define SYM_DATA_START_PTI_ALIGNED(name) \
549	SYM_DATA_START_PAGE_ALIGNED(name)
550#define PTI_USER_PGD_FILL	0
551#endif
552
553/* Automate the creation of 1 to 1 mapping pmd entries */
554#define PMDS(START, PERM, COUNT)			\
555	i = 0 ;						\
556	.rept (COUNT) ;					\
557	.quad	(START) + (i << PMD_SHIFT) + (PERM) ;	\
558	i = i + 1 ;					\
559	.endr
560
561	__INITDATA
562	.balign 4
563
564SYM_DATA_START_PTI_ALIGNED(early_top_pgt)
565	.fill	512,8,0
566	.fill	PTI_USER_PGD_FILL,8,0
567SYM_DATA_END(early_top_pgt)
568
569SYM_DATA_START_PAGE_ALIGNED(early_dynamic_pgts)
570	.fill	512*EARLY_DYNAMIC_PAGE_TABLES,8,0
571SYM_DATA_END(early_dynamic_pgts)
572
573SYM_DATA(early_recursion_flag, .long 0)
574
575	.data
576
577#if defined(CONFIG_XEN_PV) || defined(CONFIG_PVH)
578SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
579	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
580	.org    init_top_pgt + L4_PAGE_OFFSET*8, 0
581	.quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
582	.org    init_top_pgt + L4_START_KERNEL*8, 0
583	/* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
584	.quad   level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
585	.fill	PTI_USER_PGD_FILL,8,0
586SYM_DATA_END(init_top_pgt)
587
588SYM_DATA_START_PAGE_ALIGNED(level3_ident_pgt)
589	.quad	level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
590	.fill	511, 8, 0
591SYM_DATA_END(level3_ident_pgt)
592SYM_DATA_START_PAGE_ALIGNED(level2_ident_pgt)
593	/*
594	 * Since I easily can, map the first 1G.
595	 * Don't set NX because code runs from these pages.
596	 *
597	 * Note: This sets _PAGE_GLOBAL despite whether
598	 * the CPU supports it or it is enabled.  But,
599	 * the CPU should ignore the bit.
600	 */
601	PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
602SYM_DATA_END(level2_ident_pgt)
603#else
604SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
605	.fill	512,8,0
606	.fill	PTI_USER_PGD_FILL,8,0
607SYM_DATA_END(init_top_pgt)
608#endif
609
610#ifdef CONFIG_X86_5LEVEL
611SYM_DATA_START_PAGE_ALIGNED(level4_kernel_pgt)
612	.fill	511,8,0
613	.quad	level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
614SYM_DATA_END(level4_kernel_pgt)
615#endif
616
617SYM_DATA_START_PAGE_ALIGNED(level3_kernel_pgt)
618	.fill	L3_START_KERNEL,8,0
619	/* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
620	.quad	level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
621	.quad	level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
622SYM_DATA_END(level3_kernel_pgt)
623
624SYM_DATA_START_PAGE_ALIGNED(level2_kernel_pgt)
625	/*
626	 * Kernel high mapping.
627	 *
628	 * The kernel code+data+bss must be located below KERNEL_IMAGE_SIZE in
629	 * virtual address space, which is 1 GiB if RANDOMIZE_BASE is enabled,
630	 * 512 MiB otherwise.
631	 *
632	 * (NOTE: after that starts the module area, see MODULES_VADDR.)
633	 *
634	 * This table is eventually used by the kernel during normal runtime.
635	 * Care must be taken to clear out undesired bits later, like _PAGE_RW
636	 * or _PAGE_GLOBAL in some cases.
637	 */
638	PMDS(0, __PAGE_KERNEL_LARGE_EXEC, KERNEL_IMAGE_SIZE/PMD_SIZE)
639SYM_DATA_END(level2_kernel_pgt)
640
641SYM_DATA_START_PAGE_ALIGNED(level2_fixmap_pgt)
642	.fill	(512 - 4 - FIXMAP_PMD_NUM),8,0
643	pgtno = 0
644	.rept (FIXMAP_PMD_NUM)
645	.quad level1_fixmap_pgt + (pgtno << PAGE_SHIFT) - __START_KERNEL_map \
646		+ _PAGE_TABLE_NOENC;
647	pgtno = pgtno + 1
648	.endr
649	/* 6 MB reserved space + a 2MB hole */
650	.fill	4,8,0
651SYM_DATA_END(level2_fixmap_pgt)
652
653SYM_DATA_START_PAGE_ALIGNED(level1_fixmap_pgt)
654	.rept (FIXMAP_PMD_NUM)
655	.fill	512,8,0
656	.endr
657SYM_DATA_END(level1_fixmap_pgt)
658
659#undef PMDS
660
661	.data
662	.align 16
663
664SYM_DATA(smpboot_control,		.long 0)
665
666	.align 16
667/* This must match the first entry in level2_kernel_pgt */
668SYM_DATA(phys_base, .quad 0x0)
669EXPORT_SYMBOL(phys_base)
670
671#include "../../x86/xen/xen-head.S"
672
673	__PAGE_ALIGNED_BSS
674SYM_DATA_START_PAGE_ALIGNED(empty_zero_page)
675	.skip PAGE_SIZE
676SYM_DATA_END(empty_zero_page)
677EXPORT_SYMBOL(empty_zero_page)
678
679