arch/x86/kernel/head64.c

   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
  30 /*
  31  * Manage page tables very early on.
  32  */
  33 extern pgd_t early_level4_pgt[PTRS_PER_PGD];
  34 extern pmd_t early_dynamic_pgts[EARLY_DYNAMIC_PAGE_TABLES][PTRS_PER_PMD];
  35 static unsigned int __initdata next_early_pgt = 2;
  36
  37 /* Wipe all early page tables except for the kernel symbol map */
  38 static void __init reset_early_page_tables(void)
  39 {
  40         unsigned long i;
  41
  42         for (i = 0; i < PTRS_PER_PGD-1; i++)
  43                 early_level4_pgt[i].pgd = 0;
  44
  45         next_early_pgt = 0;
  46
  47         write_cr3(__pa(early_level4_pgt));
  48 }
  49
  50 /* Create a new PMD entry */
  51 int __init early_make_pgtable(unsigned long address)
  52 {
  53         unsigned long physaddr = address - __PAGE_OFFSET;
  54         unsigned long i;
  55         pgdval_t pgd, *pgd_p;
  56         pudval_t pud, *pud_p;
  57         pmdval_t pmd, *pmd_p;
  58
  59         /* Invalid address or early pgt is done ?  */
  60         if (physaddr >= MAXMEM || read_cr3() != __pa(early_level4_pgt))
  61                 return -1;
  62
  63 again:
  64         pgd_p = &early_level4_pgt[pgd_index(address)].pgd;
  65         pgd = *pgd_p;
  66
  67         /*
  68          * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is
  69          * critical -- __PAGE_OFFSET would point us back into the dynamic
  70          * range and we might end up looping forever...
  71          */
  72         if (pgd)
  73                 pud_p = (pudval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
  74         else {
  75                 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
  76                         reset_early_page_tables();
  77                         goto again;
  78                 }
  79
  80                 pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
  81                 for (i = 0; i < PTRS_PER_PUD; i++)
  82                         pud_p[i] = 0;
  83                 *pgd_p = (pgdval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
  84         }
  85         pud_p += pud_index(address);
  86         pud = *pud_p;
  87
  88         if (pud)
  89                 pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
  90         else {
  91                 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
  92                         reset_early_page_tables();
  93                         goto again;
  94                 }
  95
  96                 pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
  97                 for (i = 0; i < PTRS_PER_PMD; i++)
  98                         pmd_p[i] = 0;
  99                 *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 100         }
 101         pmd = (physaddr & PMD_MASK) + (__PAGE_KERNEL_LARGE & ~_PAGE_GLOBAL);
 102         pmd_p[pmd_index(address)] = pmd;
 103
 104         return 0;
 105 }
 106
 107 /* Don't add a printk in there. printk relies on the PDA which is not initialized
 108    yet. */
 109 static void __init clear_bss(void)
 110 {
 111         memset(__bss_start, 0,
 112                (unsigned long) __bss_stop - (unsigned long) __bss_start);
 113 }
 114
 115 static unsigned long get_cmd_line_ptr(void)
 116 {
 117         unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr;
 118
 119         cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32;
 120
 121         return cmd_line_ptr;
 122 }
 123
 124 static void __init copy_bootdata(char *real_mode_data)
 125 {
 126         char * command_line;
 127         unsigned long cmd_line_ptr;
 128
 129         memcpy(&boot_params, real_mode_data, sizeof boot_params);
 130         sanitize_boot_params(&boot_params);
 131         cmd_line_ptr = get_cmd_line_ptr();
 132         if (cmd_line_ptr) {
 133                 command_line = __va(cmd_line_ptr);
 134                 memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
 135         }
 136 }
 137
 138 void __init x86_64_start_kernel(char * real_mode_data)
 139 {
 140         int i;
 141
 142         /*
 143          * Build-time sanity checks on the kernel image and module
 144          * area mappings. (these are purely build-time and produce no code)
 145          */
 146         BUILD_BUG_ON(MODULES_VADDR < KERNEL_IMAGE_START);
 147         BUILD_BUG_ON(MODULES_VADDR-KERNEL_IMAGE_START < KERNEL_IMAGE_SIZE);
 148         BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE);
 149         BUILD_BUG_ON((KERNEL_IMAGE_START & ~PMD_MASK) != 0);
 150         BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0);
 151         BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
 152         BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
 153                                 (__START_KERNEL & PGDIR_MASK)));
 154         BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END);
 155
 156         /* Kill off the identity-map trampoline */
 157         reset_early_page_tables();
 158
 159         /* clear bss before set_intr_gate with early_idt_handler */
 160         clear_bss();
 161
 162         for (i = 0; i < NUM_EXCEPTION_VECTORS; i++) {
 163 #ifdef CONFIG_EARLY_PRINTK
 164                 set_intr_gate(i, &early_idt_handlers[i]);
 165 #else
 166                 set_intr_gate(i, early_idt_handler);
 167 #endif
 168         }
 169         load_idt((const struct desc_ptr *)&idt_descr);
 170
 171         copy_bootdata(__va(real_mode_data));
 172
 173         if (console_loglevel == 10)
 174                 early_printk("Kernel alive\n");
 175
 176         clear_page(init_level4_pgt);
 177         /* set init_level4_pgt kernel high mapping*/
 178         init_level4_pgt[511] = early_level4_pgt[511];
 179
 180         x86_64_start_reservations(real_mode_data);
 181 }
 182
 183 void __init x86_64_start_reservations(char *real_mode_data)
 184 {
 185         /* version is always not zero if it is copied */
 186         if (!boot_params.hdr.version)
 187                 copy_bootdata(__va(real_mode_data));
 188
 189         reserve_ebda_region();
 190
 191         start_kernel();
 192 }