Commit | Line | Data |
---|---|---|
5234f5eb EB |
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> | |
0341c14d | 10 | #include <asm/page_types.h> |
4bfaaef0 | 11 | #include <asm/kexec.h> |
fd3af531 | 12 | #include <asm/processor-flags.h> |
0341c14d | 13 | #include <asm/pgtable_types.h> |
5234f5eb | 14 | |
4bfaaef0 MD |
15 | /* |
16 | * Must be relocatable PIC code callable as a C function | |
17 | */ | |
18 | ||
19 | #define PTR(x) (x << 3) | |
366932de | 20 | #define PAGE_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY) |
4bfaaef0 | 21 | |
fee7b0d8 HY |
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 | ||
4bfaaef0 | 40 | .text |
288621e3 | 41 | .align PAGE_SIZE |
5234f5eb | 42 | .code64 |
4bfaaef0 MD |
43 | .globl relocate_kernel |
44 | relocate_kernel: | |
fef3a7a1 HY |
45 | /* |
46 | * %rdi indirection_page | |
4bfaaef0 MD |
47 | * %rsi page_list |
48 | * %rdx start address | |
fee7b0d8 | 49 | * %rcx preserve_context |
4e237903 | 50 | * %r8 sme_active |
4bfaaef0 MD |
51 | */ |
52 | ||
fee7b0d8 HY |
53 | /* Save the CPU context, used for jumping back */ |
54 | pushq %rbx | |
55 | pushq %rbp | |
56 | pushq %r12 | |
57 | pushq %r13 | |
58 | pushq %r14 | |
59 | pushq %r15 | |
60 | pushf | |
61 | ||
62 | movq PTR(VA_CONTROL_PAGE)(%rsi), %r11 | |
63 | movq %rsp, RSP(%r11) | |
64 | movq %cr0, %rax | |
65 | movq %rax, CR0(%r11) | |
66 | movq %cr3, %rax | |
67 | movq %rax, CR3(%r11) | |
68 | movq %cr4, %rax | |
69 | movq %rax, CR4(%r11) | |
70 | ||
5bf30316 KS |
71 | /* Save CR4. Required to enable the right paging mode later. */ |
72 | movq %rax, %r13 | |
73 | ||
5234f5eb EB |
74 | /* zero out flags, and disable interrupts */ |
75 | pushq $0 | |
76 | popfq | |
77 | ||
4e237903 TL |
78 | /* Save SME active flag */ |
79 | movq %r8, %r12 | |
80 | ||
fef3a7a1 HY |
81 | /* |
82 | * get physical address of control page now | |
83 | * this is impossible after page table switch | |
84 | */ | |
4bfaaef0 MD |
85 | movq PTR(PA_CONTROL_PAGE)(%rsi), %r8 |
86 | ||
87 | /* get physical address of page table now too */ | |
fee7b0d8 HY |
88 | movq PTR(PA_TABLE_PAGE)(%rsi), %r9 |
89 | ||
90 | /* get physical address of swap page now */ | |
91 | movq PTR(PA_SWAP_PAGE)(%rsi), %r10 | |
92 | ||
93 | /* save some information for jumping back */ | |
94 | movq %r9, CP_PA_TABLE_PAGE(%r11) | |
95 | movq %r10, CP_PA_SWAP_PAGE(%r11) | |
96 | movq %rdi, CP_PA_BACKUP_PAGES_MAP(%r11) | |
5234f5eb | 97 | |
f5deb796 | 98 | /* Switch to the identity mapped page tables */ |
fee7b0d8 | 99 | movq %r9, %cr3 |
4bfaaef0 MD |
100 | |
101 | /* setup a new stack at the end of the physical control page */ | |
a7bba17b | 102 | lea PAGE_SIZE(%r8), %rsp |
4bfaaef0 MD |
103 | |
104 | /* jump to identity mapped page */ | |
105 | addq $(identity_mapped - relocate_kernel), %r8 | |
106 | pushq %r8 | |
107 | ret | |
108 | ||
109 | identity_mapped: | |
050438ed HY |
110 | /* set return address to 0 if not preserving context */ |
111 | pushq $0 | |
4bfaaef0 MD |
112 | /* store the start address on the stack */ |
113 | pushq %rdx | |
5234f5eb | 114 | |
fef3a7a1 HY |
115 | /* |
116 | * Set cr0 to a known state: | |
fd3af531 | 117 | * - Paging enabled |
118 | * - Alignment check disabled | |
119 | * - Write protect disabled | |
120 | * - No task switch | |
121 | * - Don't do FP software emulation. | |
122 | * - Proctected mode enabled | |
5234f5eb EB |
123 | */ |
124 | movq %cr0, %rax | |
fd3af531 | 125 | andq $~(X86_CR0_AM | X86_CR0_WP | X86_CR0_TS | X86_CR0_EM), %rax |
126 | orl $(X86_CR0_PG | X86_CR0_PE), %eax | |
5234f5eb EB |
127 | movq %rax, %cr0 |
128 | ||
fef3a7a1 HY |
129 | /* |
130 | * Set cr4 to a known state: | |
fd3af531 | 131 | * - physical address extension enabled |
5bf30316 | 132 | * - 5-level paging, if it was enabled before |
5234f5eb | 133 | */ |
a734b4a2 | 134 | movl $X86_CR4_PAE, %eax |
5bf30316 KS |
135 | testq $X86_CR4_LA57, %r13 |
136 | jz 1f | |
137 | orl $X86_CR4_LA57, %eax | |
138 | 1: | |
5234f5eb EB |
139 | movq %rax, %cr4 |
140 | ||
141 | jmp 1f | |
142 | 1: | |
143 | ||
f5deb796 | 144 | /* Flush the TLB (needed?) */ |
fee7b0d8 HY |
145 | movq %r9, %cr3 |
146 | ||
4e237903 TL |
147 | /* |
148 | * If SME is active, there could be old encrypted cache line | |
149 | * entries that will conflict with the now unencrypted memory | |
150 | * used by kexec. Flush the caches before copying the kernel. | |
151 | */ | |
152 | testq %r12, %r12 | |
153 | jz 1f | |
154 | wbinvd | |
155 | 1: | |
156 | ||
fee7b0d8 HY |
157 | movq %rcx, %r11 |
158 | call swap_pages | |
159 | ||
160 | /* | |
161 | * To be certain of avoiding problems with self-modifying code | |
162 | * I need to execute a serializing instruction here. | |
163 | * So I flush the TLB by reloading %cr3 here, it's handy, | |
164 | * and not processor dependent. | |
165 | */ | |
166 | movq %cr3, %rax | |
167 | movq %rax, %cr3 | |
168 | ||
169 | /* | |
170 | * set all of the registers to known values | |
171 | * leave %rsp alone | |
172 | */ | |
173 | ||
174 | testq %r11, %r11 | |
175 | jnz 1f | |
f037e416 PA |
176 | xorl %eax, %eax |
177 | xorl %ebx, %ebx | |
178 | xorl %ecx, %ecx | |
179 | xorl %edx, %edx | |
180 | xorl %esi, %esi | |
181 | xorl %edi, %edi | |
182 | xorl %ebp, %ebp | |
183 | xorl %r8d, %r8d | |
184 | xorl %r9d, %r9d | |
185 | xorl %r10d, %r10d | |
186 | xorl %r11d, %r11d | |
187 | xorl %r12d, %r12d | |
188 | xorl %r13d, %r13d | |
189 | xorl %r14d, %r14d | |
190 | xorl %r15d, %r15d | |
fee7b0d8 HY |
191 | |
192 | ret | |
193 | ||
194 | 1: | |
195 | popq %rdx | |
196 | leaq PAGE_SIZE(%r10), %rsp | |
197 | call *%rdx | |
198 | ||
199 | /* get the re-entry point of the peer system */ | |
200 | movq 0(%rsp), %rbp | |
201 | call 1f | |
202 | 1: | |
203 | popq %r8 | |
204 | subq $(1b - relocate_kernel), %r8 | |
205 | movq CP_PA_SWAP_PAGE(%r8), %r10 | |
206 | movq CP_PA_BACKUP_PAGES_MAP(%r8), %rdi | |
207 | movq CP_PA_TABLE_PAGE(%r8), %rax | |
208 | movq %rax, %cr3 | |
209 | lea PAGE_SIZE(%r8), %rsp | |
210 | call swap_pages | |
211 | movq $virtual_mapped, %rax | |
212 | pushq %rax | |
213 | ret | |
214 | ||
215 | virtual_mapped: | |
216 | movq RSP(%r8), %rsp | |
217 | movq CR4(%r8), %rax | |
218 | movq %rax, %cr4 | |
219 | movq CR3(%r8), %rax | |
220 | movq CR0(%r8), %r8 | |
221 | movq %rax, %cr3 | |
222 | movq %r8, %cr0 | |
223 | movq %rbp, %rax | |
224 | ||
225 | popf | |
226 | popq %r15 | |
227 | popq %r14 | |
228 | popq %r13 | |
229 | popq %r12 | |
230 | popq %rbp | |
231 | popq %rbx | |
232 | ret | |
5234f5eb EB |
233 | |
234 | /* Do the copies */ | |
fee7b0d8 | 235 | swap_pages: |
5234f5eb | 236 | movq %rdi, %rcx /* Put the page_list in %rcx */ |
f037e416 PA |
237 | xorl %edi, %edi |
238 | xorl %esi, %esi | |
5234f5eb EB |
239 | jmp 1f |
240 | ||
241 | 0: /* top, read another word for the indirection page */ | |
242 | ||
243 | movq (%rbx), %rcx | |
244 | addq $8, %rbx | |
245 | 1: | |
3e1aa7cb | 246 | testb $0x1, %cl /* is it a destination page? */ |
5234f5eb EB |
247 | jz 2f |
248 | movq %rcx, %rdi | |
249 | andq $0xfffffffffffff000, %rdi | |
250 | jmp 0b | |
251 | 2: | |
3e1aa7cb | 252 | testb $0x2, %cl /* is it an indirection page? */ |
5234f5eb EB |
253 | jz 2f |
254 | movq %rcx, %rbx | |
255 | andq $0xfffffffffffff000, %rbx | |
256 | jmp 0b | |
257 | 2: | |
3e1aa7cb | 258 | testb $0x4, %cl /* is it the done indicator? */ |
5234f5eb EB |
259 | jz 2f |
260 | jmp 3f | |
261 | 2: | |
3e1aa7cb | 262 | testb $0x8, %cl /* is it the source indicator? */ |
5234f5eb EB |
263 | jz 0b /* Ignore it otherwise */ |
264 | movq %rcx, %rsi /* For ever source page do a copy */ | |
265 | andq $0xfffffffffffff000, %rsi | |
266 | ||
fee7b0d8 HY |
267 | movq %rdi, %rdx |
268 | movq %rsi, %rax | |
269 | ||
270 | movq %r10, %rdi | |
a734b4a2 | 271 | movl $512, %ecx |
5234f5eb | 272 | rep ; movsq |
5234f5eb | 273 | |
fee7b0d8 HY |
274 | movq %rax, %rdi |
275 | movq %rdx, %rsi | |
a734b4a2 | 276 | movl $512, %ecx |
fee7b0d8 | 277 | rep ; movsq |
5234f5eb | 278 | |
fee7b0d8 HY |
279 | movq %rdx, %rdi |
280 | movq %r10, %rsi | |
a734b4a2 | 281 | movl $512, %ecx |
fee7b0d8 | 282 | rep ; movsq |
5234f5eb | 283 | |
fee7b0d8 HY |
284 | lea PAGE_SIZE(%rax), %rsi |
285 | jmp 0b | |
286 | 3: | |
5234f5eb | 287 | ret |
fee7b0d8 HY |
288 | |
289 | .globl kexec_control_code_size | |
290 | .set kexec_control_code_size, . - relocate_kernel |