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
  57         /* Is the address too large? */
  58         if (physaddr >> MAX_PHYSMEM_BITS)
  59                 for (;;);
  60
  61         /*
  62          * Compute the delta between the address I am compiled to run at
  63          * and the address I am actually running at.
  64          */
  65         load_delta = physaddr - (unsigned long)(_text - __START_KERNEL_map);
  66
  67         /* Is the address not 2M aligned? */
  68         if (load_delta & ~PMD_PAGE_MASK)
  69                 for (;;);
  70
  71         /* Fixup the physical addresses in the page table */
  72
  73         pgd = fixup_pointer(&early_top_pgt, physaddr);
  74         pgd[pgd_index(__START_KERNEL_map)] += load_delta;
  75
  76         if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
  77                 p4d = fixup_pointer(&level4_kernel_pgt, physaddr);
  78                 p4d[511] += load_delta;
  79         }
  80
  81         pud = fixup_pointer(&level3_kernel_pgt, physaddr);
  82         pud[510] += load_delta;
  83         pud[511] += load_delta;
  84
  85         pmd = fixup_pointer(level2_fixmap_pgt, physaddr);
  86         pmd[506] += load_delta;
  87
  88         /*
  89          * Set up the identity mapping for the switchover.  These
  90          * entries should *NOT* have the global bit set!  This also
  91          * creates a bunch of nonsense entries but that is fine --
  92          * it avoids problems around wraparound.
  93          */
  94
  95         pud = fixup_pointer(early_dynamic_pgts[next_early_pgt++], physaddr);
  96         pmd = fixup_pointer(early_dynamic_pgts[next_early_pgt++], physaddr);
  97
  98         if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
  99                 p4d = fixup_pointer(early_dynamic_pgts[next_early_pgt++], physaddr);
 100
 101                 i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
 102                 pgd[i + 0] = (pgdval_t)p4d + _KERNPG_TABLE;
 103                 pgd[i + 1] = (pgdval_t)p4d + _KERNPG_TABLE;
 104
 105                 i = (physaddr >> P4D_SHIFT) % PTRS_PER_P4D;
 106                 p4d[i + 0] = (pgdval_t)pud + _KERNPG_TABLE;
 107                 p4d[i + 1] = (pgdval_t)pud + _KERNPG_TABLE;
 108         } else {
 109                 i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
 110                 pgd[i + 0] = (pgdval_t)pud + _KERNPG_TABLE;
 111                 pgd[i + 1] = (pgdval_t)pud + _KERNPG_TABLE;
 112         }
 113
 114         i = (physaddr >> PUD_SHIFT) % PTRS_PER_PUD;
 115         pud[i + 0] = (pudval_t)pmd + _KERNPG_TABLE;
 116         pud[i + 1] = (pudval_t)pmd + _KERNPG_TABLE;
 117
 118         pmd_entry = __PAGE_KERNEL_LARGE_EXEC & ~_PAGE_GLOBAL;
 119         pmd_entry +=  physaddr;
 120
 121         for (i = 0; i < DIV_ROUND_UP(_end - _text, PMD_SIZE); i++) {
 122                 int idx = i + (physaddr >> PMD_SHIFT) % PTRS_PER_PMD;
 123                 pmd[idx] = pmd_entry + i * PMD_SIZE;
 124         }
 125
 126         /*
 127          * Fixup the kernel text+data virtual addresses. Note that
 128          * we might write invalid pmds, when the kernel is relocated
 129          * cleanup_highmap() fixes this up along with the mappings
 130          * beyond _end.
 131          */
 132
 133         pmd = fixup_pointer(level2_kernel_pgt, physaddr);
 134         for (i = 0; i < PTRS_PER_PMD; i++) {
 135                 if (pmd[i] & _PAGE_PRESENT)
 136                         pmd[i] += load_delta;
 137         }
 138
 139         /* Fixup phys_base */
 140         p = fixup_pointer(&phys_base, physaddr);
 141         *p += load_delta;
 142 }
 143
 144 /* Wipe all early page tables except for the kernel symbol map */
 145 static void __init reset_early_page_tables(void)
 146 {
 147         memset(early_top_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1));
 148         next_early_pgt = 0;
 149         write_cr3(__pa_nodebug(early_top_pgt));
 150 }
 151
 152 /* Create a new PMD entry */
 153 int __init early_make_pgtable(unsigned long address)
 154 {
 155         unsigned long physaddr = address - __PAGE_OFFSET;
 156         pgdval_t pgd, *pgd_p;
 157         p4dval_t p4d, *p4d_p;
 158         pudval_t pud, *pud_p;
 159         pmdval_t pmd, *pmd_p;
 160
 161         /* Invalid address or early pgt is done ?  */
 162         if (physaddr >= MAXMEM || read_cr3_pa() != __pa_nodebug(early_top_pgt))
 163                 return -1;
 164
 165 again:
 166         pgd_p = &early_top_pgt[pgd_index(address)].pgd;
 167         pgd = *pgd_p;
 168
 169         /*
 170          * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is
 171          * critical -- __PAGE_OFFSET would point us back into the dynamic
 172          * range and we might end up looping forever...
 173          */
 174         if (!IS_ENABLED(CONFIG_X86_5LEVEL))
 175                 p4d_p = pgd_p;
 176         else if (pgd)
 177                 p4d_p = (p4dval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
 178         else {
 179                 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
 180                         reset_early_page_tables();
 181                         goto again;
 182                 }
 183
 184                 p4d_p = (p4dval_t *)early_dynamic_pgts[next_early_pgt++];
 185                 memset(p4d_p, 0, sizeof(*p4d_p) * PTRS_PER_P4D);
 186                 *pgd_p = (pgdval_t)p4d_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 187         }
 188         p4d_p += p4d_index(address);
 189         p4d = *p4d_p;
 190
 191         if (p4d)
 192                 pud_p = (pudval_t *)((p4d & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
 193         else {
 194                 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
 195                         reset_early_page_tables();
 196                         goto again;
 197                 }
 198
 199                 pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
 200                 memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);
 201                 *p4d_p = (p4dval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 202         }
 203         pud_p += pud_index(address);
 204         pud = *pud_p;
 205
 206         if (pud)
 207                 pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
 208         else {
 209                 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
 210                         reset_early_page_tables();
 211                         goto again;
 212                 }
 213
 214                 pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
 215                 memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);
 216                 *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 217         }
 218         pmd = (physaddr & PMD_MASK) + early_pmd_flags;
 219         pmd_p[pmd_index(address)] = pmd;
 220
 221         return 0;
 222 }
 223
 224 /* Don't add a printk in there. printk relies on the PDA which is not initialized
 225    yet. */
 226 static void __init clear_bss(void)
 227 {
 228         memset(__bss_start, 0,
 229                (unsigned long) __bss_stop - (unsigned long) __bss_start);
 230 }
 231
 232 static unsigned long get_cmd_line_ptr(void)
 233 {
 234         unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr;
 235
 236         cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32;
 237
 238         return cmd_line_ptr;
 239 }
 240
 241 static void __init copy_bootdata(char *real_mode_data)
 242 {
 243         char * command_line;
 244         unsigned long cmd_line_ptr;
 245
 246         memcpy(&boot_params, real_mode_data, sizeof boot_params);
 247         sanitize_boot_params(&boot_params);
 248         cmd_line_ptr = get_cmd_line_ptr();
 249         if (cmd_line_ptr) {
 250                 command_line = __va(cmd_line_ptr);
 251                 memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
 252         }
 253 }
 254
 255 asmlinkage __visible void __init x86_64_start_kernel(char * real_mode_data)
 256 {
 257         int i;
 258
 259         /*
 260          * Build-time sanity checks on the kernel image and module
 261          * area mappings. (these are purely build-time and produce no code)
 262          */
 263         BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map);
 264         BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE);
 265         BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE);
 266         BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0);
 267         BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0);
 268         BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
 269         BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
 270                                 (__START_KERNEL & PGDIR_MASK)));
 271         BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END);
 272
 273         cr4_init_shadow();
 274
 275         /* Kill off the identity-map trampoline */
 276         reset_early_page_tables();
 277
 278         clear_bss();
 279
 280         clear_page(init_top_pgt);
 281
 282         kasan_early_init();
 283
 284         for (i = 0; i < NUM_EXCEPTION_VECTORS; i++)
 285                 set_intr_gate(i, early_idt_handler_array[i]);
 286         load_idt((const struct desc_ptr *)&idt_descr);
 287
 288         copy_bootdata(__va(real_mode_data));
 289
 290         /*
 291          * Load microcode early on BSP.
 292          */
 293         load_ucode_bsp();
 294
 295         /* set init_top_pgt kernel high mapping*/
 296         init_top_pgt[511] = early_top_pgt[511];
 297
 298         x86_64_start_reservations(real_mode_data);
 299 }
 300
 301 void __init x86_64_start_reservations(char *real_mode_data)
 302 {
 303         /* version is always not zero if it is copied */
 304         if (!boot_params.hdr.version)
 305                 copy_bootdata(__va(real_mode_data));
 306
 307         x86_early_init_platform_quirks();
 308
 309         switch (boot_params.hdr.hardware_subarch) {
 310         case X86_SUBARCH_INTEL_MID:
 311                 x86_intel_mid_early_setup();
 312                 break;
 313         default:
 314                 break;
 315         }
 316
 317         start_kernel();
 318 }