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 static void __init *fixup_pointer(void *ptr, unsigned long physaddr)
  42 {
  43         return ptr - (void *)_text + (void *)physaddr;
  44 }
  45
  46 void __init __startup_64(unsigned long physaddr)
  47 {
  48         unsigned long load_delta, *p;
  49         pgdval_t *pgd;
  50         p4dval_t *p4d;
  51         pudval_t *pud;
  52         pmdval_t *pmd, pmd_entry;
  53         int i;
  54
  55         /* Is the address too large? */
  56         if (physaddr >> MAX_PHYSMEM_BITS)
  57                 for (;;);
  58
  59         /*
  60          * Compute the delta between the address I am compiled to run at
  61          * and the address I am actually running at.
  62          */
  63         load_delta = physaddr - (unsigned long)(_text - __START_KERNEL_map);
  64
  65         /* Is the address not 2M aligned? */
  66         if (load_delta & ~PMD_PAGE_MASK)
  67                 for (;;);
  68
  69         /* Fixup the physical addresses in the page table */
  70
  71         pgd = fixup_pointer(&early_top_pgt, physaddr);
  72         pgd[pgd_index(__START_KERNEL_map)] += load_delta;
  73
  74         if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
  75                 p4d = fixup_pointer(&level4_kernel_pgt, physaddr);
  76                 p4d[511] += load_delta;
  77         }
  78
  79         pud = fixup_pointer(&level3_kernel_pgt, physaddr);
  80         pud[510] += load_delta;
  81         pud[511] += load_delta;
  82
  83         pmd = fixup_pointer(level2_fixmap_pgt, physaddr);
  84         pmd[506] += load_delta;
  85
  86         /*
  87          * Set up the identity mapping for the switchover.  These
  88          * entries should *NOT* have the global bit set!  This also
  89          * creates a bunch of nonsense entries but that is fine --
  90          * it avoids problems around wraparound.
  91          */
  92
  93         pud = fixup_pointer(early_dynamic_pgts[next_early_pgt++], physaddr);
  94         pmd = fixup_pointer(early_dynamic_pgts[next_early_pgt++], physaddr);
  95
  96         if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
  97                 p4d = fixup_pointer(early_dynamic_pgts[next_early_pgt++], physaddr);
  98
  99                 i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
 100                 pgd[i + 0] = (pgdval_t)p4d + _KERNPG_TABLE;
 101                 pgd[i + 1] = (pgdval_t)p4d + _KERNPG_TABLE;
 102
 103                 i = (physaddr >> P4D_SHIFT) % PTRS_PER_P4D;
 104                 p4d[i + 0] = (pgdval_t)pud + _KERNPG_TABLE;
 105                 p4d[i + 1] = (pgdval_t)pud + _KERNPG_TABLE;
 106         } else {
 107                 i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
 108                 pgd[i + 0] = (pgdval_t)pud + _KERNPG_TABLE;
 109                 pgd[i + 1] = (pgdval_t)pud + _KERNPG_TABLE;
 110         }
 111
 112         i = (physaddr >> PUD_SHIFT) % PTRS_PER_PUD;
 113         pud[i + 0] = (pudval_t)pmd + _KERNPG_TABLE;
 114         pud[i + 1] = (pudval_t)pmd + _KERNPG_TABLE;
 115
 116         pmd_entry = __PAGE_KERNEL_LARGE_EXEC & ~_PAGE_GLOBAL;
 117         pmd_entry +=  physaddr;
 118
 119         for (i = 0; i < DIV_ROUND_UP(_end - _text, PMD_SIZE); i++) {
 120                 int idx = i + (physaddr >> PMD_SHIFT) % PTRS_PER_PMD;
 121                 pmd[idx] = pmd_entry + i * PMD_SIZE;
 122         }
 123
 124         /*
 125          * Fixup the kernel text+data virtual addresses. Note that
 126          * we might write invalid pmds, when the kernel is relocated
 127          * cleanup_highmap() fixes this up along with the mappings
 128          * beyond _end.
 129          */
 130
 131         pmd = fixup_pointer(level2_kernel_pgt, physaddr);
 132         for (i = 0; i < PTRS_PER_PMD; i++) {
 133                 if (pmd[i] & _PAGE_PRESENT)
 134                         pmd[i] += load_delta;
 135         }
 136
 137         /* Fixup phys_base */
 138         p = fixup_pointer(&phys_base, physaddr);
 139         *p += load_delta;
 140 }
 141
 142 /* Wipe all early page tables except for the kernel symbol map */
 143 static void __init reset_early_page_tables(void)
 144 {
 145         memset(early_top_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1));
 146         next_early_pgt = 0;
 147         write_cr3(__pa_nodebug(early_top_pgt));
 148 }
 149
 150 /* Create a new PMD entry */
 151 int __init early_make_pgtable(unsigned long address)
 152 {
 153         unsigned long physaddr = address - __PAGE_OFFSET;
 154         pgdval_t pgd, *pgd_p;
 155         p4dval_t p4d, *p4d_p;
 156         pudval_t pud, *pud_p;
 157         pmdval_t pmd, *pmd_p;
 158
 159         /* Invalid address or early pgt is done ?  */
 160         if (physaddr >= MAXMEM || read_cr3_pa() != __pa_nodebug(early_top_pgt))
 161                 return -1;
 162
 163 again:
 164         pgd_p = &early_top_pgt[pgd_index(address)].pgd;
 165         pgd = *pgd_p;
 166
 167         /*
 168          * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is
 169          * critical -- __PAGE_OFFSET would point us back into the dynamic
 170          * range and we might end up looping forever...
 171          */
 172         if (!IS_ENABLED(CONFIG_X86_5LEVEL))
 173                 p4d_p = pgd_p;
 174         else if (pgd)
 175                 p4d_p = (p4dval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
 176         else {
 177                 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
 178                         reset_early_page_tables();
 179                         goto again;
 180                 }
 181
 182                 p4d_p = (p4dval_t *)early_dynamic_pgts[next_early_pgt++];
 183                 memset(p4d_p, 0, sizeof(*p4d_p) * PTRS_PER_P4D);
 184                 *pgd_p = (pgdval_t)p4d_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 185         }
 186         p4d_p += p4d_index(address);
 187         p4d = *p4d_p;
 188
 189         if (p4d)
 190                 pud_p = (pudval_t *)((p4d & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
 191         else {
 192                 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
 193                         reset_early_page_tables();
 194                         goto again;
 195                 }
 196
 197                 pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
 198                 memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);
 199                 *p4d_p = (p4dval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 200         }
 201         pud_p += pud_index(address);
 202         pud = *pud_p;
 203
 204         if (pud)
 205                 pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
 206         else {
 207                 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
 208                         reset_early_page_tables();
 209                         goto again;
 210                 }
 211
 212                 pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
 213                 memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);
 214                 *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 215         }
 216         pmd = (physaddr & PMD_MASK) + early_pmd_flags;
 217         pmd_p[pmd_index(address)] = pmd;
 218
 219         return 0;
 220 }
 221
 222 /* Don't add a printk in there. printk relies on the PDA which is not initialized
 223    yet. */
 224 static void __init clear_bss(void)
 225 {
 226         memset(__bss_start, 0,
 227                (unsigned long) __bss_stop - (unsigned long) __bss_start);
 228 }
 229
 230 static unsigned long get_cmd_line_ptr(void)
 231 {
 232         unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr;
 233
 234         cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32;
 235
 236         return cmd_line_ptr;
 237 }
 238
 239 static void __init copy_bootdata(char *real_mode_data)
 240 {
 241         char * command_line;
 242         unsigned long cmd_line_ptr;
 243
 244         memcpy(&boot_params, real_mode_data, sizeof boot_params);
 245         sanitize_boot_params(&boot_params);
 246         cmd_line_ptr = get_cmd_line_ptr();
 247         if (cmd_line_ptr) {
 248                 command_line = __va(cmd_line_ptr);
 249                 memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
 250         }
 251 }
 252
 253 asmlinkage __visible void __init x86_64_start_kernel(char * real_mode_data)
 254 {
 255         int i;
 256
 257         /*
 258          * Build-time sanity checks on the kernel image and module
 259          * area mappings. (these are purely build-time and produce no code)
 260          */
 261         BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map);
 262         BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE);
 263         BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE);
 264         BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0);
 265         BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0);
 266         BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
 267         BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
 268                                 (__START_KERNEL & PGDIR_MASK)));
 269         BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END);
 270
 271         cr4_init_shadow();
 272
 273         /* Kill off the identity-map trampoline */
 274         reset_early_page_tables();
 275
 276         clear_bss();
 277
 278         clear_page(init_top_pgt);
 279
 280         kasan_early_init();
 281
 282         for (i = 0; i < NUM_EXCEPTION_VECTORS; i++)
 283                 set_intr_gate(i, early_idt_handler_array[i]);
 284         load_idt((const struct desc_ptr *)&idt_descr);
 285
 286         copy_bootdata(__va(real_mode_data));
 287
 288         /*
 289          * Load microcode early on BSP.
 290          */
 291         load_ucode_bsp();
 292
 293         /* set init_top_pgt kernel high mapping*/
 294         init_top_pgt[511] = early_top_pgt[511];
 295
 296         x86_64_start_reservations(real_mode_data);
 297 }
 298
 299 void __init x86_64_start_reservations(char *real_mode_data)
 300 {
 301         /* version is always not zero if it is copied */
 302         if (!boot_params.hdr.version)
 303                 copy_bootdata(__va(real_mode_data));
 304
 305         x86_early_init_platform_quirks();
 306
 307         switch (boot_params.hdr.hardware_subarch) {
 308         case X86_SUBARCH_INTEL_MID:
 309                 x86_intel_mid_early_setup();
 310                 break;
 311         default:
 312                 break;
 313         }
 314
 315         start_kernel();
 316 }