]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
835c34a1 | 2 | * prepare to run common code |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
1da177e4 LT |
5 | */ |
6 | ||
be3606ff | 7 | #define DISABLE_BRANCH_PROFILING |
1da177e4 LT |
8 | #include <linux/init.h> |
9 | #include <linux/linkage.h> | |
10 | #include <linux/types.h> | |
11 | #include <linux/kernel.h> | |
12 | #include <linux/string.h> | |
13 | #include <linux/percpu.h> | |
eaf76e8b | 14 | #include <linux/start_kernel.h> |
8b664aa6 | 15 | #include <linux/io.h> |
72d7c3b3 | 16 | #include <linux/memblock.h> |
1da177e4 LT |
17 | |
18 | #include <asm/processor.h> | |
19 | #include <asm/proto.h> | |
20 | #include <asm/smp.h> | |
1da177e4 LT |
21 | #include <asm/setup.h> |
22 | #include <asm/desc.h> | |
f6c2e333 | 23 | #include <asm/pgtable.h> |
cfd243d4 | 24 | #include <asm/tlbflush.h> |
2bc0414e | 25 | #include <asm/sections.h> |
718fc13b | 26 | #include <asm/kdebug.h> |
66441bd3 | 27 | #include <asm/e820/api.h> |
47a3d5da | 28 | #include <asm/bios_ebda.h> |
5dcd14ec | 29 | #include <asm/bootparam_utils.h> |
feddc9de | 30 | #include <asm/microcode.h> |
ef7f0d6a | 31 | #include <asm/kasan.h> |
1da177e4 | 32 | |
8170e6be PA |
33 | /* |
34 | * Manage page tables very early on. | |
35 | */ | |
65ade2f8 | 36 | extern pgd_t early_top_pgt[PTRS_PER_PGD]; |
8170e6be | 37 | extern pmd_t early_dynamic_pgts[EARLY_DYNAMIC_PAGE_TABLES][PTRS_PER_PMD]; |
c88d7150 | 38 | static unsigned int __initdata next_early_pgt; |
5e427ec2 | 39 | pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX); |
8170e6be | 40 | |
26179670 KS |
41 | #define __head __section(.head.text) |
42 | ||
43 | static void __head *fixup_pointer(void *ptr, unsigned long physaddr) | |
c88d7150 KS |
44 | { |
45 | return ptr - (void *)_text + (void *)physaddr; | |
46 | } | |
47 | ||
26179670 | 48 | void __head __startup_64(unsigned long physaddr) |
c88d7150 KS |
49 | { |
50 | unsigned long load_delta, *p; | |
51 | pgdval_t *pgd; | |
032370b9 | 52 | p4dval_t *p4d; |
c88d7150 KS |
53 | pudval_t *pud; |
54 | pmdval_t *pmd, pmd_entry; | |
55 | int i; | |
187e91fe | 56 | unsigned int *next_pgt_ptr; |
c88d7150 KS |
57 | |
58 | /* Is the address too large? */ | |
59 | if (physaddr >> MAX_PHYSMEM_BITS) | |
60 | for (;;); | |
61 | ||
62 | /* | |
63 | * Compute the delta between the address I am compiled to run at | |
64 | * and the address I am actually running at. | |
65 | */ | |
66 | load_delta = physaddr - (unsigned long)(_text - __START_KERNEL_map); | |
67 | ||
68 | /* Is the address not 2M aligned? */ | |
69 | if (load_delta & ~PMD_PAGE_MASK) | |
70 | for (;;); | |
71 | ||
72 | /* Fixup the physical addresses in the page table */ | |
73 | ||
65ade2f8 | 74 | pgd = fixup_pointer(&early_top_pgt, physaddr); |
c88d7150 KS |
75 | pgd[pgd_index(__START_KERNEL_map)] += load_delta; |
76 | ||
032370b9 KS |
77 | if (IS_ENABLED(CONFIG_X86_5LEVEL)) { |
78 | p4d = fixup_pointer(&level4_kernel_pgt, physaddr); | |
79 | p4d[511] += load_delta; | |
80 | } | |
81 | ||
c88d7150 KS |
82 | pud = fixup_pointer(&level3_kernel_pgt, physaddr); |
83 | pud[510] += load_delta; | |
84 | pud[511] += load_delta; | |
85 | ||
86 | pmd = fixup_pointer(level2_fixmap_pgt, physaddr); | |
87 | pmd[506] += load_delta; | |
88 | ||
89 | /* | |
90 | * Set up the identity mapping for the switchover. These | |
91 | * entries should *NOT* have the global bit set! This also | |
92 | * creates a bunch of nonsense entries but that is fine -- | |
93 | * it avoids problems around wraparound. | |
94 | */ | |
187e91fe AP |
95 | next_pgt_ptr = fixup_pointer(&next_early_pgt, physaddr); |
96 | pud = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], physaddr); | |
97 | pmd = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], physaddr); | |
c88d7150 | 98 | |
032370b9 KS |
99 | if (IS_ENABLED(CONFIG_X86_5LEVEL)) { |
100 | p4d = fixup_pointer(early_dynamic_pgts[next_early_pgt++], physaddr); | |
101 | ||
102 | i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD; | |
103 | pgd[i + 0] = (pgdval_t)p4d + _KERNPG_TABLE; | |
104 | pgd[i + 1] = (pgdval_t)p4d + _KERNPG_TABLE; | |
105 | ||
106 | i = (physaddr >> P4D_SHIFT) % PTRS_PER_P4D; | |
107 | p4d[i + 0] = (pgdval_t)pud + _KERNPG_TABLE; | |
108 | p4d[i + 1] = (pgdval_t)pud + _KERNPG_TABLE; | |
109 | } else { | |
110 | i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD; | |
111 | pgd[i + 0] = (pgdval_t)pud + _KERNPG_TABLE; | |
112 | pgd[i + 1] = (pgdval_t)pud + _KERNPG_TABLE; | |
113 | } | |
c88d7150 KS |
114 | |
115 | i = (physaddr >> PUD_SHIFT) % PTRS_PER_PUD; | |
116 | pud[i + 0] = (pudval_t)pmd + _KERNPG_TABLE; | |
117 | pud[i + 1] = (pudval_t)pmd + _KERNPG_TABLE; | |
118 | ||
119 | pmd_entry = __PAGE_KERNEL_LARGE_EXEC & ~_PAGE_GLOBAL; | |
120 | pmd_entry += physaddr; | |
121 | ||
122 | for (i = 0; i < DIV_ROUND_UP(_end - _text, PMD_SIZE); i++) { | |
123 | int idx = i + (physaddr >> PMD_SHIFT) % PTRS_PER_PMD; | |
124 | pmd[idx] = pmd_entry + i * PMD_SIZE; | |
125 | } | |
126 | ||
127 | /* | |
128 | * Fixup the kernel text+data virtual addresses. Note that | |
129 | * we might write invalid pmds, when the kernel is relocated | |
130 | * cleanup_highmap() fixes this up along with the mappings | |
131 | * beyond _end. | |
132 | */ | |
133 | ||
134 | pmd = fixup_pointer(level2_kernel_pgt, physaddr); | |
135 | for (i = 0; i < PTRS_PER_PMD; i++) { | |
136 | if (pmd[i] & _PAGE_PRESENT) | |
137 | pmd[i] += load_delta; | |
138 | } | |
139 | ||
140 | /* Fixup phys_base */ | |
141 | p = fixup_pointer(&phys_base, physaddr); | |
142 | *p += load_delta; | |
143 | } | |
144 | ||
8170e6be PA |
145 | /* Wipe all early page tables except for the kernel symbol map */ |
146 | static void __init reset_early_page_tables(void) | |
cfd243d4 | 147 | { |
65ade2f8 | 148 | memset(early_top_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1)); |
8170e6be | 149 | next_early_pgt = 0; |
65ade2f8 | 150 | write_cr3(__pa_nodebug(early_top_pgt)); |
8170e6be PA |
151 | } |
152 | ||
153 | /* Create a new PMD entry */ | |
154 | int __init early_make_pgtable(unsigned long address) | |
155 | { | |
156 | unsigned long physaddr = address - __PAGE_OFFSET; | |
8170e6be | 157 | pgdval_t pgd, *pgd_p; |
032370b9 | 158 | p4dval_t p4d, *p4d_p; |
6b9c75ac | 159 | pudval_t pud, *pud_p; |
8170e6be PA |
160 | pmdval_t pmd, *pmd_p; |
161 | ||
162 | /* Invalid address or early pgt is done ? */ | |
65ade2f8 | 163 | if (physaddr >= MAXMEM || read_cr3_pa() != __pa_nodebug(early_top_pgt)) |
8170e6be PA |
164 | return -1; |
165 | ||
6b9c75ac | 166 | again: |
65ade2f8 | 167 | pgd_p = &early_top_pgt[pgd_index(address)].pgd; |
8170e6be PA |
168 | pgd = *pgd_p; |
169 | ||
170 | /* | |
171 | * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is | |
172 | * critical -- __PAGE_OFFSET would point us back into the dynamic | |
173 | * range and we might end up looping forever... | |
174 | */ | |
032370b9 KS |
175 | if (!IS_ENABLED(CONFIG_X86_5LEVEL)) |
176 | p4d_p = pgd_p; | |
177 | else if (pgd) | |
178 | p4d_p = (p4dval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base); | |
179 | else { | |
180 | if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) { | |
181 | reset_early_page_tables(); | |
182 | goto again; | |
183 | } | |
184 | ||
185 | p4d_p = (p4dval_t *)early_dynamic_pgts[next_early_pgt++]; | |
186 | memset(p4d_p, 0, sizeof(*p4d_p) * PTRS_PER_P4D); | |
187 | *pgd_p = (pgdval_t)p4d_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE; | |
188 | } | |
189 | p4d_p += p4d_index(address); | |
190 | p4d = *p4d_p; | |
191 | ||
192 | if (p4d) | |
193 | pud_p = (pudval_t *)((p4d & PTE_PFN_MASK) + __START_KERNEL_map - phys_base); | |
6b9c75ac YL |
194 | else { |
195 | if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) { | |
8170e6be | 196 | reset_early_page_tables(); |
6b9c75ac YL |
197 | goto again; |
198 | } | |
8170e6be PA |
199 | |
200 | pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++]; | |
a91bbe01 | 201 | memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD); |
032370b9 | 202 | *p4d_p = (p4dval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE; |
8170e6be | 203 | } |
6b9c75ac YL |
204 | pud_p += pud_index(address); |
205 | pud = *pud_p; | |
8170e6be | 206 | |
6b9c75ac YL |
207 | if (pud) |
208 | pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base); | |
209 | else { | |
210 | if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) { | |
211 | reset_early_page_tables(); | |
212 | goto again; | |
213 | } | |
214 | ||
215 | pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++]; | |
a91bbe01 | 216 | memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD); |
6b9c75ac YL |
217 | *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE; |
218 | } | |
78d77df7 | 219 | pmd = (physaddr & PMD_MASK) + early_pmd_flags; |
6b9c75ac | 220 | pmd_p[pmd_index(address)] = pmd; |
8170e6be PA |
221 | |
222 | return 0; | |
cfd243d4 VG |
223 | } |
224 | ||
1da177e4 LT |
225 | /* Don't add a printk in there. printk relies on the PDA which is not initialized |
226 | yet. */ | |
227 | static void __init clear_bss(void) | |
228 | { | |
1da177e4 | 229 | memset(__bss_start, 0, |
2bc0414e | 230 | (unsigned long) __bss_stop - (unsigned long) __bss_start); |
1da177e4 LT |
231 | } |
232 | ||
f1da834c YL |
233 | static unsigned long get_cmd_line_ptr(void) |
234 | { | |
235 | unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr; | |
236 | ||
ee92d815 YL |
237 | cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32; |
238 | ||
f1da834c YL |
239 | return cmd_line_ptr; |
240 | } | |
241 | ||
1da177e4 LT |
242 | static void __init copy_bootdata(char *real_mode_data) |
243 | { | |
1da177e4 | 244 | char * command_line; |
f1da834c | 245 | unsigned long cmd_line_ptr; |
1da177e4 | 246 | |
30c82645 | 247 | memcpy(&boot_params, real_mode_data, sizeof boot_params); |
5dcd14ec | 248 | sanitize_boot_params(&boot_params); |
f1da834c YL |
249 | cmd_line_ptr = get_cmd_line_ptr(); |
250 | if (cmd_line_ptr) { | |
251 | command_line = __va(cmd_line_ptr); | |
30c82645 | 252 | memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE); |
1da177e4 | 253 | } |
1da177e4 LT |
254 | } |
255 | ||
2605fc21 | 256 | asmlinkage __visible void __init x86_64_start_kernel(char * real_mode_data) |
1da177e4 | 257 | { |
1da177e4 LT |
258 | int i; |
259 | ||
b4e0409a IM |
260 | /* |
261 | * Build-time sanity checks on the kernel image and module | |
262 | * area mappings. (these are purely build-time and produce no code) | |
263 | */ | |
8e3c2a8c BP |
264 | BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map); |
265 | BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE); | |
b4e0409a | 266 | BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE); |
8e3c2a8c | 267 | BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0); |
b4e0409a IM |
268 | BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0); |
269 | BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL)); | |
270 | BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) == | |
271 | (__START_KERNEL & PGDIR_MASK))); | |
66d4bdf2 | 272 | BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END); |
b4e0409a | 273 | |
1e02ce4c AL |
274 | cr4_init_shadow(); |
275 | ||
8170e6be PA |
276 | /* Kill off the identity-map trampoline */ |
277 | reset_early_page_tables(); | |
278 | ||
3df0af0e YL |
279 | clear_bss(); |
280 | ||
65ade2f8 | 281 | clear_page(init_top_pgt); |
d0f77d4d | 282 | |
5d5aa3cf AP |
283 | kasan_early_init(); |
284 | ||
ac630dd9 | 285 | for (i = 0; i < NUM_EXCEPTION_VECTORS; i++) |
425be567 | 286 | set_intr_gate(i, early_idt_handler_array[i]); |
9d1c6e7c | 287 | load_idt((const struct desc_ptr *)&idt_descr); |
f6c2e333 | 288 | |
fa2bbce9 YL |
289 | copy_bootdata(__va(real_mode_data)); |
290 | ||
feddc9de FY |
291 | /* |
292 | * Load microcode early on BSP. | |
293 | */ | |
294 | load_ucode_bsp(); | |
295 | ||
65ade2f8 KS |
296 | /* set init_top_pgt kernel high mapping*/ |
297 | init_top_pgt[511] = early_top_pgt[511]; | |
8170e6be | 298 | |
f97013fd JF |
299 | x86_64_start_reservations(real_mode_data); |
300 | } | |
301 | ||
302 | void __init x86_64_start_reservations(char *real_mode_data) | |
303 | { | |
fa2bbce9 YL |
304 | /* version is always not zero if it is copied */ |
305 | if (!boot_params.hdr.version) | |
306 | copy_bootdata(__va(real_mode_data)); | |
9de819fe | 307 | |
8d152e7a | 308 | x86_early_init_platform_quirks(); |
75175278 | 309 | |
3fda5bb4 AS |
310 | switch (boot_params.hdr.hardware_subarch) { |
311 | case X86_SUBARCH_INTEL_MID: | |
312 | x86_intel_mid_early_setup(); | |
313 | break; | |
314 | default: | |
315 | break; | |
316 | } | |
317 | ||
1da177e4 LT |
318 | start_kernel(); |
319 | } |