1/*
  2 * relocate_kernel.S - put the kernel image in place to boot
  3 * Copyright (C) 2002-2005 Eric Biederman  <ebiederm@xmission.com>
  4 *
  5 * This source code is licensed under the GNU General Public License,
  6 * Version 2.  See the file COPYING for more details.
  7 */
  8
  9#include <linux/linkage.h>
 
 
 10#include <asm/page_types.h>
 11#include <asm/kexec.h>
 12#include <asm/processor-flags.h>
 13#include <asm/pgtable_types.h>
 
 
 
 14
 15/*
 16 * Must be relocatable PIC code callable as a C function
 
 17 */
 18
 19#define PTR(x) (x << 3)
 20#define PAGE_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY)
 21
 22/*
 23 * control_page + KEXEC_CONTROL_CODE_MAX_SIZE
 24 * ~ control_page + PAGE_SIZE are used as data storage and stack for
 25 * jumping back
 26 */
 27#define DATA(offset)		(KEXEC_CONTROL_CODE_MAX_SIZE+(offset))
 28
 29/* Minimal CPU state */
 30#define RSP			DATA(0x0)
 31#define CR0			DATA(0x8)
 32#define CR3			DATA(0x10)
 33#define CR4			DATA(0x18)
 34
 35/* other data */
 36#define CP_PA_TABLE_PAGE	DATA(0x20)
 37#define CP_PA_SWAP_PAGE		DATA(0x28)
 38#define CP_PA_BACKUP_PAGES_MAP	DATA(0x30)
 39
 40	.text
 41	.align PAGE_SIZE
 42	.code64
 43	.globl relocate_kernel
 44relocate_kernel:
 
 
 45	/*
 46	 * %rdi indirection_page
 47	 * %rsi page_list
 48	 * %rdx start address
 49	 * %rcx preserve_context
 
 50	 */
 51
 52	/* Save the CPU context, used for jumping back */
 53	pushq %rbx
 54	pushq %rbp
 55	pushq %r12
 56	pushq %r13
 57	pushq %r14
 58	pushq %r15
 59	pushf
 60
 61	movq	PTR(VA_CONTROL_PAGE)(%rsi), %r11
 62	movq	%rsp, RSP(%r11)
 63	movq	%cr0, %rax
 64	movq	%rax, CR0(%r11)
 65	movq	%cr3, %rax
 66	movq	%rax, CR3(%r11)
 67	movq	%cr4, %rax
 68	movq	%rax, CR4(%r11)
 69
 
 
 
 70	/* zero out flags, and disable interrupts */
 71	pushq $0
 72	popfq
 73
 
 
 
 74	/*
 75	 * get physical address of control page now
 76	 * this is impossible after page table switch
 77	 */
 78	movq	PTR(PA_CONTROL_PAGE)(%rsi), %r8
 79
 80	/* get physical address of page table now too */
 81	movq	PTR(PA_TABLE_PAGE)(%rsi), %r9
 82
 83	/* get physical address of swap page now */
 84	movq	PTR(PA_SWAP_PAGE)(%rsi), %r10
 85
 86	/* save some information for jumping back */
 87	movq	%r9, CP_PA_TABLE_PAGE(%r11)
 88	movq	%r10, CP_PA_SWAP_PAGE(%r11)
 89	movq	%rdi, CP_PA_BACKUP_PAGES_MAP(%r11)
 90
 91	/* Switch to the identity mapped page tables */
 92	movq	%r9, %cr3
 93
 94	/* setup a new stack at the end of the physical control page */
 95	lea	PAGE_SIZE(%r8), %rsp
 96
 97	/* jump to identity mapped page */
 98	addq	$(identity_mapped - relocate_kernel), %r8
 99	pushq	%r8
 
100	ret
 
 
101
102identity_mapped:
 
103	/* set return address to 0 if not preserving context */
104	pushq	$0
105	/* store the start address on the stack */
106	pushq   %rdx
107
108	/*
 
 
 
 
 
 
 
 
109	 * Set cr0 to a known state:
110	 *  - Paging enabled
111	 *  - Alignment check disabled
112	 *  - Write protect disabled
113	 *  - No task switch
114	 *  - Don't do FP software emulation.
115	 *  - Proctected mode enabled
116	 */
117	movq	%cr0, %rax
118	andq	$~(X86_CR0_AM | X86_CR0_WP | X86_CR0_TS | X86_CR0_EM), %rax
119	orl	$(X86_CR0_PG | X86_CR0_PE), %eax
120	movq	%rax, %cr0
121
122	/*
123	 * Set cr4 to a known state:
124	 *  - physical address extension enabled
 
 
 
 
 
 
125	 */
126	movq	$X86_CR4_PAE, %rax
127	movq	%rax, %cr4
128
129	jmp 1f
1301:
131
132	/* Flush the TLB (needed?) */
133	movq	%r9, %cr3
134
 
 
 
 
 
 
 
 
 
 
 
135	movq	%rcx, %r11
136	call	swap_pages
137
138	/*
139	 * To be certain of avoiding problems with self-modifying code
140	 * I need to execute a serializing instruction here.
141	 * So I flush the TLB by reloading %cr3 here, it's handy,
142	 * and not processor dependent.
143	 */
144	movq	%cr3, %rax
145	movq	%rax, %cr3
146
147	/*
148	 * set all of the registers to known values
149	 * leave %rsp alone
150	 */
151
152	testq	%r11, %r11
153	jnz 1f
154	xorq	%rax, %rax
155	xorq	%rbx, %rbx
156	xorq    %rcx, %rcx
157	xorq    %rdx, %rdx
158	xorq    %rsi, %rsi
159	xorq    %rdi, %rdi
160	xorq    %rbp, %rbp
161	xorq	%r8,  %r8
162	xorq	%r9,  %r9
163	xorq	%r10, %r9
164	xorq	%r11, %r11
165	xorq	%r12, %r12
166	xorq	%r13, %r13
167	xorq	%r14, %r14
168	xorq	%r15, %r15
169
 
170	ret
 
171
1721:
173	popq	%rdx
174	leaq	PAGE_SIZE(%r10), %rsp
 
175	call	*%rdx
176
177	/* get the re-entry point of the peer system */
178	movq	0(%rsp), %rbp
179	call	1f
1801:
181	popq	%r8
182	subq	$(1b - relocate_kernel), %r8
183	movq	CP_PA_SWAP_PAGE(%r8), %r10
184	movq	CP_PA_BACKUP_PAGES_MAP(%r8), %rdi
185	movq	CP_PA_TABLE_PAGE(%r8), %rax
186	movq	%rax, %cr3
187	lea	PAGE_SIZE(%r8), %rsp
188	call	swap_pages
189	movq	$virtual_mapped, %rax
190	pushq	%rax
 
191	ret
 
 
192
193virtual_mapped:
 
 
194	movq	RSP(%r8), %rsp
195	movq	CR4(%r8), %rax
196	movq	%rax, %cr4
197	movq	CR3(%r8), %rax
198	movq	CR0(%r8), %r8
199	movq	%rax, %cr3
200	movq	%r8, %cr0
 
 
 
 
 
 
 
201	movq	%rbp, %rax
202
203	popf
204	popq	%r15
205	popq	%r14
206	popq	%r13
207	popq	%r12
208	popq	%rbp
209	popq	%rbx
 
210	ret
 
 
211
212	/* Do the copies */
213swap_pages:
214	movq	%rdi, %rcx 	/* Put the page_list in %rcx */
215	xorq	%rdi, %rdi
216	xorq	%rsi, %rsi
 
217	jmp	1f
218
2190:	/* top, read another word for the indirection page */
220
221	movq	(%rbx), %rcx
222	addq	$8,	%rbx
2231:
224	testq	$0x1,	%rcx  /* is it a destination page? */
225	jz	2f
226	movq	%rcx,	%rdi
227	andq	$0xfffffffffffff000, %rdi
228	jmp	0b
2292:
230	testq	$0x2,	%rcx  /* is it an indirection page? */
231	jz	2f
232	movq	%rcx,   %rbx
233	andq	$0xfffffffffffff000, %rbx
234	jmp	0b
2352:
236	testq	$0x4,	%rcx  /* is it the done indicator? */
237	jz	2f
238	jmp	3f
2392:
240	testq	$0x8,	%rcx  /* is it the source indicator? */
241	jz	0b	      /* Ignore it otherwise */
242	movq	%rcx,   %rsi  /* For ever source page do a copy */
243	andq	$0xfffffffffffff000, %rsi
244
245	movq	%rdi, %rdx
246	movq	%rsi, %rax
247
 
248	movq	%r10, %rdi
249	movq	$512,   %rcx
250	rep ; movsq
251
 
252	movq	%rax, %rdi
253	movq	%rdx, %rsi
254	movq	$512,   %rcx
255	rep ; movsq
256
 
257	movq	%rdx, %rdi
258	movq	%r10, %rsi
259	movq	$512,   %rcx
260	rep ; movsq
261
262	lea	PAGE_SIZE(%rax), %rsi
263	jmp	0b
2643:
 
265	ret
 
 
266
267	.globl kexec_control_code_size
268.set kexec_control_code_size, . - relocate_kernel

  1/* SPDX-License-Identifier: GPL-2.0-only */
  2/*
  3 * relocate_kernel.S - put the kernel image in place to boot
  4 * Copyright (C) 2002-2005 Eric Biederman  <ebiederm@xmission.com>
 
 
 
  5 */
  6
  7#include <linux/linkage.h>
  8#include <linux/stringify.h>
  9#include <asm/alternative.h>
 10#include <asm/page_types.h>
 11#include <asm/kexec.h>
 12#include <asm/processor-flags.h>
 13#include <asm/pgtable_types.h>
 14#include <asm/nospec-branch.h>
 15#include <asm/unwind_hints.h>
 16#include <asm/asm-offsets.h>
 17
 18/*
 19 * Must be relocatable PIC code callable as a C function, in particular
 20 * there must be a plain RET and not jump to return thunk.
 21 */
 22
 23#define PTR(x) (x << 3)
 24#define PAGE_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY)
 25
 26/*
 27 * control_page + KEXEC_CONTROL_CODE_MAX_SIZE
 28 * ~ control_page + PAGE_SIZE are used as data storage and stack for
 29 * jumping back
 30 */
 31#define DATA(offset)		(KEXEC_CONTROL_CODE_MAX_SIZE+(offset))
 32
 33/* Minimal CPU state */
 34#define RSP			DATA(0x0)
 35#define CR0			DATA(0x8)
 36#define CR3			DATA(0x10)
 37#define CR4			DATA(0x18)
 38
 39/* other data */
 40#define CP_PA_TABLE_PAGE	DATA(0x20)
 41#define CP_PA_SWAP_PAGE		DATA(0x28)
 42#define CP_PA_BACKUP_PAGES_MAP	DATA(0x30)
 43
 44	.text
 45	.align PAGE_SIZE
 46	.code64
 47SYM_CODE_START_NOALIGN(relocate_range)
 48SYM_CODE_START_NOALIGN(relocate_kernel)
 49	UNWIND_HINT_END_OF_STACK
 50	ANNOTATE_NOENDBR
 51	/*
 52	 * %rdi indirection_page
 53	 * %rsi page_list
 54	 * %rdx start address
 55	 * %rcx preserve_context
 56	 * %r8  host_mem_enc_active
 57	 */
 58
 59	/* Save the CPU context, used for jumping back */
 60	pushq %rbx
 61	pushq %rbp
 62	pushq %r12
 63	pushq %r13
 64	pushq %r14
 65	pushq %r15
 66	pushf
 67
 68	movq	PTR(VA_CONTROL_PAGE)(%rsi), %r11
 69	movq	%rsp, RSP(%r11)
 70	movq	%cr0, %rax
 71	movq	%rax, CR0(%r11)
 72	movq	%cr3, %rax
 73	movq	%rax, CR3(%r11)
 74	movq	%cr4, %rax
 75	movq	%rax, CR4(%r11)
 76
 77	/* Save CR4. Required to enable the right paging mode later. */
 78	movq	%rax, %r13
 79
 80	/* zero out flags, and disable interrupts */
 81	pushq $0
 82	popfq
 83
 84	/* Save SME active flag */
 85	movq	%r8, %r12
 86
 87	/*
 88	 * get physical address of control page now
 89	 * this is impossible after page table switch
 90	 */
 91	movq	PTR(PA_CONTROL_PAGE)(%rsi), %r8
 92
 93	/* get physical address of page table now too */
 94	movq	PTR(PA_TABLE_PAGE)(%rsi), %r9
 95
 96	/* get physical address of swap page now */
 97	movq	PTR(PA_SWAP_PAGE)(%rsi), %r10
 98
 99	/* save some information for jumping back */
100	movq	%r9, CP_PA_TABLE_PAGE(%r11)
101	movq	%r10, CP_PA_SWAP_PAGE(%r11)
102	movq	%rdi, CP_PA_BACKUP_PAGES_MAP(%r11)
103
104	/* Switch to the identity mapped page tables */
105	movq	%r9, %cr3
106
107	/* setup a new stack at the end of the physical control page */
108	lea	PAGE_SIZE(%r8), %rsp
109
110	/* jump to identity mapped page */
111	addq	$(identity_mapped - relocate_kernel), %r8
112	pushq	%r8
113	ANNOTATE_UNRET_SAFE
114	ret
115	int3
116SYM_CODE_END(relocate_kernel)
117
118SYM_CODE_START_LOCAL_NOALIGN(identity_mapped)
119	UNWIND_HINT_END_OF_STACK
120	/* set return address to 0 if not preserving context */
121	pushq	$0
122	/* store the start address on the stack */
123	pushq   %rdx
124
125	/*
126	 * Clear X86_CR4_CET (if it was set) such that we can clear CR0_WP
127	 * below.
128	 */
129	movq	%cr4, %rax
130	andq	$~(X86_CR4_CET), %rax
131	movq	%rax, %cr4
132
133	/*
134	 * Set cr0 to a known state:
135	 *  - Paging enabled
136	 *  - Alignment check disabled
137	 *  - Write protect disabled
138	 *  - No task switch
139	 *  - Don't do FP software emulation.
140	 *  - Protected mode enabled
141	 */
142	movq	%cr0, %rax
143	andq	$~(X86_CR0_AM | X86_CR0_WP | X86_CR0_TS | X86_CR0_EM), %rax
144	orl	$(X86_CR0_PG | X86_CR0_PE), %eax
145	movq	%rax, %cr0
146
147	/*
148	 * Set cr4 to a known state:
149	 *  - physical address extension enabled
150	 *  - 5-level paging, if it was enabled before
151	 *  - Machine check exception on TDX guest, if it was enabled before.
152	 *    Clearing MCE might not be allowed in TDX guests, depending on setup.
153	 *
154	 * Use R13 that contains the original CR4 value, read in relocate_kernel().
155	 * PAE is always set in the original CR4.
156	 */
157	andl	$(X86_CR4_PAE | X86_CR4_LA57), %r13d
158	ALTERNATIVE "", __stringify(orl $X86_CR4_MCE, %r13d), X86_FEATURE_TDX_GUEST
159	movq	%r13, %cr4
 
 
160
161	/* Flush the TLB (needed?) */
162	movq	%r9, %cr3
163
164	/*
165	 * If SME is active, there could be old encrypted cache line
166	 * entries that will conflict with the now unencrypted memory
167	 * used by kexec. Flush the caches before copying the kernel.
168	 */
169	testq	%r12, %r12
170	jz .Lsme_off
171	wbinvd
172.Lsme_off:
173
174	/* Save the preserve_context to %r11 as swap_pages clobbers %rcx. */
175	movq	%rcx, %r11
176	call	swap_pages
177
178	/*
179	 * To be certain of avoiding problems with self-modifying code
180	 * I need to execute a serializing instruction here.
181	 * So I flush the TLB by reloading %cr3 here, it's handy,
182	 * and not processor dependent.
183	 */
184	movq	%cr3, %rax
185	movq	%rax, %cr3
186
187	/*
188	 * set all of the registers to known values
189	 * leave %rsp alone
190	 */
191
192	testq	%r11, %r11
193	jnz .Lrelocate
194	xorl	%eax, %eax
195	xorl	%ebx, %ebx
196	xorl    %ecx, %ecx
197	xorl    %edx, %edx
198	xorl    %esi, %esi
199	xorl    %edi, %edi
200	xorl    %ebp, %ebp
201	xorl	%r8d, %r8d
202	xorl	%r9d, %r9d
203	xorl	%r10d, %r10d
204	xorl	%r11d, %r11d
205	xorl	%r12d, %r12d
206	xorl	%r13d, %r13d
207	xorl	%r14d, %r14d
208	xorl	%r15d, %r15d
209
210	ANNOTATE_UNRET_SAFE
211	ret
212	int3
213
214.Lrelocate:
215	popq	%rdx
216	leaq	PAGE_SIZE(%r10), %rsp
217	ANNOTATE_RETPOLINE_SAFE
218	call	*%rdx
219
220	/* get the re-entry point of the peer system */
221	movq	0(%rsp), %rbp
222	leaq	relocate_kernel(%rip), %r8
 
 
 
223	movq	CP_PA_SWAP_PAGE(%r8), %r10
224	movq	CP_PA_BACKUP_PAGES_MAP(%r8), %rdi
225	movq	CP_PA_TABLE_PAGE(%r8), %rax
226	movq	%rax, %cr3
227	lea	PAGE_SIZE(%r8), %rsp
228	call	swap_pages
229	movq	$virtual_mapped, %rax
230	pushq	%rax
231	ANNOTATE_UNRET_SAFE
232	ret
233	int3
234SYM_CODE_END(identity_mapped)
235
236SYM_CODE_START_LOCAL_NOALIGN(virtual_mapped)
237	UNWIND_HINT_END_OF_STACK
238	ANNOTATE_NOENDBR // RET target, above
239	movq	RSP(%r8), %rsp
240	movq	CR4(%r8), %rax
241	movq	%rax, %cr4
242	movq	CR3(%r8), %rax
243	movq	CR0(%r8), %r8
244	movq	%rax, %cr3
245	movq	%r8, %cr0
246
247#ifdef CONFIG_KEXEC_JUMP
248	/* Saved in save_processor_state. */
249	movq    $saved_context, %rax
250	lgdt    saved_context_gdt_desc(%rax)
251#endif
252
253	movq	%rbp, %rax
254
255	popf
256	popq	%r15
257	popq	%r14
258	popq	%r13
259	popq	%r12
260	popq	%rbp
261	popq	%rbx
262	ANNOTATE_UNRET_SAFE
263	ret
264	int3
265SYM_CODE_END(virtual_mapped)
266
267	/* Do the copies */
268SYM_CODE_START_LOCAL_NOALIGN(swap_pages)
269	UNWIND_HINT_END_OF_STACK
270	movq	%rdi, %rcx	/* Put the indirection_page in %rcx */
271	xorl	%edi, %edi
272	xorl	%esi, %esi
273	jmp	1f
274
2750:	/* top, read another word for the indirection page */
276
277	movq	(%rbx), %rcx
278	addq	$8,	%rbx
2791:
280	testb	$0x1,	%cl   /* is it a destination page? */
281	jz	2f
282	movq	%rcx,	%rdi
283	andq	$0xfffffffffffff000, %rdi
284	jmp	0b
2852:
286	testb	$0x2,	%cl   /* is it an indirection page? */
287	jz	2f
288	movq	%rcx,   %rbx
289	andq	$0xfffffffffffff000, %rbx
290	jmp	0b
2912:
292	testb	$0x4,	%cl   /* is it the done indicator? */
293	jz	2f
294	jmp	3f
2952:
296	testb	$0x8,	%cl   /* is it the source indicator? */
297	jz	0b	      /* Ignore it otherwise */
298	movq	%rcx,   %rsi  /* For ever source page do a copy */
299	andq	$0xfffffffffffff000, %rsi
300
301	movq	%rdi, %rdx    /* Save destination page to %rdx */
302	movq	%rsi, %rax    /* Save source page to %rax */
303
304	/* copy source page to swap page */
305	movq	%r10, %rdi
306	movl	$512, %ecx
307	rep ; movsq
308
309	/* copy destination page to source page */
310	movq	%rax, %rdi
311	movq	%rdx, %rsi
312	movl	$512, %ecx
313	rep ; movsq
314
315	/* copy swap page to destination page */
316	movq	%rdx, %rdi
317	movq	%r10, %rsi
318	movl	$512, %ecx
319	rep ; movsq
320
321	lea	PAGE_SIZE(%rax), %rsi
322	jmp	0b
3233:
324	ANNOTATE_UNRET_SAFE
325	ret
326	int3
327SYM_CODE_END(swap_pages)
328
329	.skip KEXEC_CONTROL_CODE_MAX_SIZE - (. - relocate_kernel), 0xcc
330SYM_CODE_END(relocate_range);