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 #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/api.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_top_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;
  39 pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
  40
  41 #define __head  __section(.head.text)
  42
  43 static void __head *fixup_pointer(void *ptr, unsigned long physaddr)
  44 {
  45         return ptr - (void *)_text + (void *)physaddr;
  46 }
  47
  48 void __head __startup_64(unsigned long physaddr)
  49 {
  50         unsigned long load_delta, *p;
  51         pgdval_t *pgd;
  52         p4dval_t *p4d;
  53         pudval_t *pud;
  54         pmdval_t *pmd, pmd_entry;
  55         int i;
  56         unsigned int *next_pgt_ptr;
  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
  74         pgd = fixup_pointer(&early_top_pgt, physaddr);
  75         pgd[pgd_index(__START_KERNEL_map)] += load_delta;
  76
  77         if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
  78                 p4d = fixup_pointer(&level4_kernel_pgt, physaddr);
  79                 p4d[511] += load_delta;
  80         }
  81
  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          */
  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);
  98
  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         }
 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
 145 /* Wipe all early page tables except for the kernel symbol map */
 146 static void __init reset_early_page_tables(void)
 147 {
 148         memset(early_top_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1));
 149         next_early_pgt = 0;
 150         write_cr3(__pa_nodebug(early_top_pgt));
 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;
 157         pgdval_t pgd, *pgd_p;
 158         p4dval_t p4d, *p4d_p;
 159         pudval_t pud, *pud_p;
 160         pmdval_t pmd, *pmd_p;
 161
 162         /* Invalid address or early pgt is done ?  */
 163         if (physaddr >= MAXMEM || read_cr3_pa() != __pa_nodebug(early_top_pgt))
 164                 return -1;
 165
 166 again:
 167         pgd_p = &early_top_pgt[pgd_index(address)].pgd;
 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          */
 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);
 194         else {
 195                 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
 196                         reset_early_page_tables();
 197                         goto again;
 198                 }
 199
 200                 pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
 201                 memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);
 202                 *p4d_p = (p4dval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 203         }
 204         pud_p += pud_index(address);
 205         pud = *pud_p;
 206
 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++];
 216                 memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);
 217                 *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 218         }
 219         pmd = (physaddr & PMD_MASK) + early_pmd_flags;
 220         pmd_p[pmd_index(address)] = pmd;
 221
 222         return 0;
 223 }
 224
 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 {
 229         memset(__bss_start, 0,
 230                (unsigned long) __bss_stop - (unsigned long) __bss_start);
 231 }
 232
 233 static unsigned long get_cmd_line_ptr(void)
 234 {
 235         unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr;
 236
 237         cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32;
 238
 239         return cmd_line_ptr;
 240 }
 241
 242 static void __init copy_bootdata(char *real_mode_data)
 243 {
 244         char * command_line;
 245         unsigned long cmd_line_ptr;
 246
 247         memcpy(&boot_params, real_mode_data, sizeof boot_params);
 248         sanitize_boot_params(&boot_params);
 249         cmd_line_ptr = get_cmd_line_ptr();
 250         if (cmd_line_ptr) {
 251                 command_line = __va(cmd_line_ptr);
 252                 memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
 253         }
 254 }
 255
 256 asmlinkage __visible void __init x86_64_start_kernel(char * real_mode_data)
 257 {
 258         int i;
 259
 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          */
 264         BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map);
 265         BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE);
 266         BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE);
 267         BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0);
 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)));
 272         BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END);
 273
 274         cr4_init_shadow();
 275
 276         /* Kill off the identity-map trampoline */
 277         reset_early_page_tables();
 278
 279         clear_bss();
 280
 281         clear_page(init_top_pgt);
 282
 283         kasan_early_init();
 284
 285         for (i = 0; i < NUM_EXCEPTION_VECTORS; i++)
 286                 set_intr_gate(i, early_idt_handler_array[i]);
 287         load_idt((const struct desc_ptr *)&idt_descr);
 288
 289         copy_bootdata(__va(real_mode_data));
 290
 291         /*
 292          * Load microcode early on BSP.
 293          */
 294         load_ucode_bsp();
 295
 296         /* set init_top_pgt kernel high mapping*/
 297         init_top_pgt[511] = early_top_pgt[511];
 298
 299         x86_64_start_reservations(real_mode_data);
 300 }
 301
 302 void __init x86_64_start_reservations(char *real_mode_data)
 303 {
 304         /* version is always not zero if it is copied */
 305         if (!boot_params.hdr.version)
 306                 copy_bootdata(__va(real_mode_data));
 307
 308         x86_early_init_platform_quirks();
 309
 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
 318         start_kernel();
 319 }