1 /*
2 * prepare to run common code
3 *
4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 */
6
7 #define DISABLE_BRANCH_PROFILING
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>
14 #include <linux/start_kernel.h>
15 #include <linux/io.h>
16 #include <linux/memblock.h>
17
18 #include <asm/processor.h>
19 #include <asm/proto.h>
20 #include <asm/smp.h>
21 #include <asm/setup.h>
22 #include <asm/desc.h>
23 #include <asm/pgtable.h>
24 #include <asm/tlbflush.h>
25 #include <asm/sections.h>
26 #include <asm/kdebug.h>
27 #include <asm/e820.h>
28 #include <asm/bios_ebda.h>
29 #include <asm/bootparam_utils.h>
30 #include <asm/microcode.h>
31 #include <asm/kasan.h>
32
33 /*
34 * Manage page tables very early on.
35 */
36 extern pgd_t early_level4_pgt[PTRS_PER_PGD];
37 extern pmd_t early_dynamic_pgts[EARLY_DYNAMIC_PAGE_TABLES][PTRS_PER_PMD];
38 static unsigned int __initdata next_early_pgt = 2;
39 pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
40
41 /* Wipe all early page tables except for the kernel symbol map */
reset_early_page_tables(void)42 static void __init reset_early_page_tables(void)
43 {
44 unsigned long i;
45
46 for (i = 0; i < PTRS_PER_PGD-1; i++)
47 early_level4_pgt[i].pgd = 0;
48
49 next_early_pgt = 0;
50
51 write_cr3(__pa_nodebug(early_level4_pgt));
52 }
53
54 /* Create a new PMD entry */
early_make_pgtable(unsigned long address)55 int __init early_make_pgtable(unsigned long address)
56 {
57 unsigned long physaddr = address - __PAGE_OFFSET;
58 unsigned long i;
59 pgdval_t pgd, *pgd_p;
60 pudval_t pud, *pud_p;
61 pmdval_t pmd, *pmd_p;
62
63 /* Invalid address or early pgt is done ? */
64 if (physaddr >= MAXMEM || read_cr3() != __pa_nodebug(early_level4_pgt))
65 return -1;
66
67 again:
68 pgd_p = &early_level4_pgt[pgd_index(address)].pgd;
69 pgd = *pgd_p;
70
71 /*
72 * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is
73 * critical -- __PAGE_OFFSET would point us back into the dynamic
74 * range and we might end up looping forever...
75 */
76 if (pgd)
77 pud_p = (pudval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
78 else {
79 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
80 reset_early_page_tables();
81 goto again;
82 }
83
84 pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
85 for (i = 0; i < PTRS_PER_PUD; i++)
86 pud_p[i] = 0;
87 *pgd_p = (pgdval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
88 }
89 pud_p += pud_index(address);
90 pud = *pud_p;
91
92 if (pud)
93 pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
94 else {
95 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
96 reset_early_page_tables();
97 goto again;
98 }
99
100 pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
101 for (i = 0; i < PTRS_PER_PMD; i++)
102 pmd_p[i] = 0;
103 *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
104 }
105 pmd = (physaddr & PMD_MASK) + early_pmd_flags;
106 pmd_p[pmd_index(address)] = pmd;
107
108 return 0;
109 }
110
111 /* Don't add a printk in there. printk relies on the PDA which is not initialized
112 yet. */
clear_bss(void)113 static void __init clear_bss(void)
114 {
115 memset(__bss_start, 0,
116 (unsigned long) __bss_stop - (unsigned long) __bss_start);
117 }
118
get_cmd_line_ptr(void)119 static unsigned long get_cmd_line_ptr(void)
120 {
121 unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr;
122
123 cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32;
124
125 return cmd_line_ptr;
126 }
127
copy_bootdata(char * real_mode_data)128 static void __init copy_bootdata(char *real_mode_data)
129 {
130 char * command_line;
131 unsigned long cmd_line_ptr;
132
133 memcpy(&boot_params, real_mode_data, sizeof boot_params);
134 sanitize_boot_params(&boot_params);
135 cmd_line_ptr = get_cmd_line_ptr();
136 if (cmd_line_ptr) {
137 command_line = __va(cmd_line_ptr);
138 memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
139 }
140 }
141
x86_64_start_kernel(char * real_mode_data)142 asmlinkage __visible void __init x86_64_start_kernel(char * real_mode_data)
143 {
144 int i;
145
146 /*
147 * Build-time sanity checks on the kernel image and module
148 * area mappings. (these are purely build-time and produce no code)
149 */
150 BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map);
151 BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE);
152 BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE);
153 BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0);
154 BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0);
155 BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
156 BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
157 (__START_KERNEL & PGDIR_MASK)));
158 BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END);
159
160 cr4_init_shadow();
161
162 /* Kill off the identity-map trampoline */
163 reset_early_page_tables();
164
165 clear_bss();
166
167 clear_page(init_level4_pgt);
168
169 kasan_early_init();
170
171 for (i = 0; i < NUM_EXCEPTION_VECTORS; i++)
172 set_intr_gate(i, early_idt_handler_array[i]);
173 load_idt((const struct desc_ptr *)&idt_descr);
174
175 copy_bootdata(__va(real_mode_data));
176
177 /*
178 * Load microcode early on BSP.
179 */
180 load_ucode_bsp();
181
182 /* set init_level4_pgt kernel high mapping*/
183 init_level4_pgt[511] = early_level4_pgt[511];
184
185 x86_64_start_reservations(real_mode_data);
186 }
187
x86_64_start_reservations(char * real_mode_data)188 void __init x86_64_start_reservations(char *real_mode_data)
189 {
190 /* version is always not zero if it is copied */
191 if (!boot_params.hdr.version)
192 copy_bootdata(__va(real_mode_data));
193
194 reserve_ebda_region();
195
196 start_kernel();
197 }
198