xref: /openbmc/linux/arch/x86/kernel/head64.c (revision bc5aa3a0)
1 /*
2  *  prepare to run common code
3  *
4  *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5  */
6 
7 #include <linux/init.h>
8 #include <linux/linkage.h>
9 #include <linux/types.h>
10 #include <linux/kernel.h>
11 #include <linux/string.h>
12 #include <linux/percpu.h>
13 #include <linux/start_kernel.h>
14 #include <linux/io.h>
15 #include <linux/memblock.h>
16 
17 #include <asm/processor.h>
18 #include <asm/proto.h>
19 #include <asm/smp.h>
20 #include <asm/setup.h>
21 #include <asm/desc.h>
22 #include <asm/pgtable.h>
23 #include <asm/tlbflush.h>
24 #include <asm/sections.h>
25 #include <asm/kdebug.h>
26 #include <asm/e820.h>
27 #include <asm/bios_ebda.h>
28 #include <asm/bootparam_utils.h>
29 #include <asm/microcode.h>
30 #include <asm/kasan.h>
31 
32 /*
33  * Manage page tables very early on.
34  */
35 extern pgd_t early_level4_pgt[PTRS_PER_PGD];
36 extern pmd_t early_dynamic_pgts[EARLY_DYNAMIC_PAGE_TABLES][PTRS_PER_PMD];
37 static unsigned int __initdata next_early_pgt = 2;
38 pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
39 
40 /* Wipe all early page tables except for the kernel symbol map */
41 static void __init reset_early_page_tables(void)
42 {
43 	memset(early_level4_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1));
44 	next_early_pgt = 0;
45 	write_cr3(__pa_nodebug(early_level4_pgt));
46 }
47 
48 /* Create a new PMD entry */
49 int __init early_make_pgtable(unsigned long address)
50 {
51 	unsigned long physaddr = address - __PAGE_OFFSET;
52 	pgdval_t pgd, *pgd_p;
53 	pudval_t pud, *pud_p;
54 	pmdval_t pmd, *pmd_p;
55 
56 	/* Invalid address or early pgt is done ?  */
57 	if (physaddr >= MAXMEM || read_cr3() != __pa_nodebug(early_level4_pgt))
58 		return -1;
59 
60 again:
61 	pgd_p = &early_level4_pgt[pgd_index(address)].pgd;
62 	pgd = *pgd_p;
63 
64 	/*
65 	 * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is
66 	 * critical -- __PAGE_OFFSET would point us back into the dynamic
67 	 * range and we might end up looping forever...
68 	 */
69 	if (pgd)
70 		pud_p = (pudval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
71 	else {
72 		if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
73 			reset_early_page_tables();
74 			goto again;
75 		}
76 
77 		pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
78 		memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);
79 		*pgd_p = (pgdval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
80 	}
81 	pud_p += pud_index(address);
82 	pud = *pud_p;
83 
84 	if (pud)
85 		pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
86 	else {
87 		if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
88 			reset_early_page_tables();
89 			goto again;
90 		}
91 
92 		pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
93 		memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);
94 		*pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
95 	}
96 	pmd = (physaddr & PMD_MASK) + early_pmd_flags;
97 	pmd_p[pmd_index(address)] = pmd;
98 
99 	return 0;
100 }
101 
102 /* Don't add a printk in there. printk relies on the PDA which is not initialized
103    yet. */
104 static void __init clear_bss(void)
105 {
106 	memset(__bss_start, 0,
107 	       (unsigned long) __bss_stop - (unsigned long) __bss_start);
108 }
109 
110 static unsigned long get_cmd_line_ptr(void)
111 {
112 	unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr;
113 
114 	cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32;
115 
116 	return cmd_line_ptr;
117 }
118 
119 static void __init copy_bootdata(char *real_mode_data)
120 {
121 	char * command_line;
122 	unsigned long cmd_line_ptr;
123 
124 	memcpy(&boot_params, real_mode_data, sizeof boot_params);
125 	sanitize_boot_params(&boot_params);
126 	cmd_line_ptr = get_cmd_line_ptr();
127 	if (cmd_line_ptr) {
128 		command_line = __va(cmd_line_ptr);
129 		memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
130 	}
131 }
132 
133 asmlinkage __visible void __init x86_64_start_kernel(char * real_mode_data)
134 {
135 	int i;
136 
137 	/*
138 	 * Build-time sanity checks on the kernel image and module
139 	 * area mappings. (these are purely build-time and produce no code)
140 	 */
141 	BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map);
142 	BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE);
143 	BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE);
144 	BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0);
145 	BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0);
146 	BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
147 	BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
148 				(__START_KERNEL & PGDIR_MASK)));
149 	BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END);
150 
151 	cr4_init_shadow();
152 
153 	/* Kill off the identity-map trampoline */
154 	reset_early_page_tables();
155 
156 	clear_bss();
157 
158 	clear_page(init_level4_pgt);
159 
160 	kasan_early_init();
161 
162 	for (i = 0; i < NUM_EXCEPTION_VECTORS; i++)
163 		set_intr_gate(i, early_idt_handler_array[i]);
164 	load_idt((const struct desc_ptr *)&idt_descr);
165 
166 	copy_bootdata(__va(real_mode_data));
167 
168 	/*
169 	 * Load microcode early on BSP.
170 	 */
171 	load_ucode_bsp();
172 
173 	/* set init_level4_pgt kernel high mapping*/
174 	init_level4_pgt[511] = early_level4_pgt[511];
175 
176 	x86_64_start_reservations(real_mode_data);
177 }
178 
179 void __init x86_64_start_reservations(char *real_mode_data)
180 {
181 	/* version is always not zero if it is copied */
182 	if (!boot_params.hdr.version)
183 		copy_bootdata(__va(real_mode_data));
184 
185 	x86_early_init_platform_quirks();
186 
187 	switch (boot_params.hdr.hardware_subarch) {
188 	case X86_SUBARCH_INTEL_MID:
189 		x86_intel_mid_early_setup();
190 		break;
191 	default:
192 		break;
193 	}
194 
195 	start_kernel();
196 }
197