1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * prepare to run common code 4 * 5 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE 6 */ 7 8 /* cpu_feature_enabled() cannot be used this early */ 9 #define USE_EARLY_PGTABLE_L5 10 11 #include <linux/init.h> 12 #include <linux/linkage.h> 13 #include <linux/types.h> 14 #include <linux/kernel.h> 15 #include <linux/string.h> 16 #include <linux/percpu.h> 17 #include <linux/start_kernel.h> 18 #include <linux/io.h> 19 #include <linux/memblock.h> 20 #include <linux/cc_platform.h> 21 #include <linux/pgtable.h> 22 23 #include <asm/asm.h> 24 #include <asm/page_64.h> 25 #include <asm/processor.h> 26 #include <asm/proto.h> 27 #include <asm/smp.h> 28 #include <asm/setup.h> 29 #include <asm/desc.h> 30 #include <asm/tlbflush.h> 31 #include <asm/sections.h> 32 #include <asm/kdebug.h> 33 #include <asm/e820/api.h> 34 #include <asm/bios_ebda.h> 35 #include <asm/bootparam_utils.h> 36 #include <asm/microcode.h> 37 #include <asm/kasan.h> 38 #include <asm/fixmap.h> 39 #include <asm/realmode.h> 40 #include <asm/extable.h> 41 #include <asm/trapnr.h> 42 #include <asm/sev.h> 43 #include <asm/tdx.h> 44 #include <asm/init.h> 45 46 /* 47 * Manage page tables very early on. 48 */ 49 extern pmd_t early_dynamic_pgts[EARLY_DYNAMIC_PAGE_TABLES][PTRS_PER_PMD]; 50 unsigned int __initdata next_early_pgt; 51 SYM_PIC_ALIAS(next_early_pgt); 52 pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX); 53 54 #ifdef CONFIG_X86_5LEVEL 55 unsigned int __pgtable_l5_enabled __ro_after_init; 56 unsigned int pgdir_shift __ro_after_init = 39; 57 EXPORT_SYMBOL(pgdir_shift); 58 unsigned int ptrs_per_p4d __ro_after_init = 1; 59 EXPORT_SYMBOL(ptrs_per_p4d); 60 #endif 61 62 #ifdef CONFIG_DYNAMIC_MEMORY_LAYOUT 63 unsigned long page_offset_base __ro_after_init = __PAGE_OFFSET_BASE_L4; 64 EXPORT_SYMBOL(page_offset_base); 65 unsigned long vmalloc_base __ro_after_init = __VMALLOC_BASE_L4; 66 EXPORT_SYMBOL(vmalloc_base); 67 unsigned long vmemmap_base __ro_after_init = __VMEMMAP_BASE_L4; 68 EXPORT_SYMBOL(vmemmap_base); 69 #endif 70 71 /* Wipe all early page tables except for the kernel symbol map */ 72 static void __init reset_early_page_tables(void) 73 { 74 memset(early_top_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1)); 75 next_early_pgt = 0; 76 write_cr3(__sme_pa_nodebug(early_top_pgt)); 77 } 78 79 /* Create a new PMD entry */ 80 bool __init __early_make_pgtable(unsigned long address, pmdval_t pmd) 81 { 82 unsigned long physaddr = address - __PAGE_OFFSET; 83 pgdval_t pgd, *pgd_p; 84 p4dval_t p4d, *p4d_p; 85 pudval_t pud, *pud_p; 86 pmdval_t *pmd_p; 87 88 /* Invalid address or early pgt is done ? */ 89 if (physaddr >= MAXMEM || read_cr3_pa() != __pa_nodebug(early_top_pgt)) 90 return false; 91 92 again: 93 pgd_p = &early_top_pgt[pgd_index(address)].pgd; 94 pgd = *pgd_p; 95 96 /* 97 * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is 98 * critical -- __PAGE_OFFSET would point us back into the dynamic 99 * range and we might end up looping forever... 100 */ 101 if (!pgtable_l5_enabled()) 102 p4d_p = pgd_p; 103 else if (pgd) 104 p4d_p = (p4dval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base); 105 else { 106 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) { 107 reset_early_page_tables(); 108 goto again; 109 } 110 111 p4d_p = (p4dval_t *)early_dynamic_pgts[next_early_pgt++]; 112 memset(p4d_p, 0, sizeof(*p4d_p) * PTRS_PER_P4D); 113 *pgd_p = (pgdval_t)p4d_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE; 114 } 115 p4d_p += p4d_index(address); 116 p4d = *p4d_p; 117 118 if (p4d) 119 pud_p = (pudval_t *)((p4d & PTE_PFN_MASK) + __START_KERNEL_map - phys_base); 120 else { 121 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) { 122 reset_early_page_tables(); 123 goto again; 124 } 125 126 pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++]; 127 memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD); 128 *p4d_p = (p4dval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE; 129 } 130 pud_p += pud_index(address); 131 pud = *pud_p; 132 133 if (pud) 134 pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base); 135 else { 136 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) { 137 reset_early_page_tables(); 138 goto again; 139 } 140 141 pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++]; 142 memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD); 143 *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE; 144 } 145 pmd_p[pmd_index(address)] = pmd; 146 147 return true; 148 } 149 150 static bool __init early_make_pgtable(unsigned long address) 151 { 152 unsigned long physaddr = address - __PAGE_OFFSET; 153 pmdval_t pmd; 154 155 pmd = (physaddr & PMD_MASK) + early_pmd_flags; 156 157 return __early_make_pgtable(address, pmd); 158 } 159 160 void __init do_early_exception(struct pt_regs *regs, int trapnr) 161 { 162 if (trapnr == X86_TRAP_PF && 163 early_make_pgtable(native_read_cr2())) 164 return; 165 166 if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT) && 167 trapnr == X86_TRAP_VC && handle_vc_boot_ghcb(regs)) 168 return; 169 170 if (trapnr == X86_TRAP_VE && tdx_early_handle_ve(regs)) 171 return; 172 173 early_fixup_exception(regs, trapnr); 174 } 175 176 /* Don't add a printk in there. printk relies on the PDA which is not initialized 177 yet. */ 178 void __init clear_bss(void) 179 { 180 memset(__bss_start, 0, 181 (unsigned long) __bss_stop - (unsigned long) __bss_start); 182 memset(__brk_base, 0, 183 (unsigned long) __brk_limit - (unsigned long) __brk_base); 184 } 185 186 static unsigned long get_cmd_line_ptr(void) 187 { 188 unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr; 189 190 cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32; 191 192 return cmd_line_ptr; 193 } 194 195 static void __init copy_bootdata(char *real_mode_data) 196 { 197 char * command_line; 198 unsigned long cmd_line_ptr; 199 200 /* 201 * If SME is active, this will create decrypted mappings of the 202 * boot data in advance of the copy operations. 203 */ 204 sme_map_bootdata(real_mode_data); 205 206 memcpy(&boot_params, real_mode_data, sizeof(boot_params)); 207 sanitize_boot_params(&boot_params); 208 cmd_line_ptr = get_cmd_line_ptr(); 209 if (cmd_line_ptr) { 210 command_line = __va(cmd_line_ptr); 211 memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE); 212 } 213 214 /* 215 * The old boot data is no longer needed and won't be reserved, 216 * freeing up that memory for use by the system. If SME is active, 217 * we need to remove the mappings that were created so that the 218 * memory doesn't remain mapped as decrypted. 219 */ 220 sme_unmap_bootdata(real_mode_data); 221 } 222 223 asmlinkage __visible void __init __noreturn x86_64_start_kernel(char * real_mode_data) 224 { 225 /* 226 * Build-time sanity checks on the kernel image and module 227 * area mappings. (these are purely build-time and produce no code) 228 */ 229 BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map); 230 BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE); 231 BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE); 232 BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0); 233 BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0); 234 BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL)); 235 MAYBE_BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) == 236 (__START_KERNEL & PGDIR_MASK))); 237 BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END); 238 239 cr4_init_shadow(); 240 241 /* Kill off the identity-map trampoline */ 242 reset_early_page_tables(); 243 244 if (pgtable_l5_enabled()) { 245 page_offset_base = __PAGE_OFFSET_BASE_L5; 246 vmalloc_base = __VMALLOC_BASE_L5; 247 vmemmap_base = __VMEMMAP_BASE_L5; 248 } 249 250 clear_bss(); 251 252 /* 253 * This needs to happen *before* kasan_early_init() because latter maps stuff 254 * into that page. 255 */ 256 clear_page(init_top_pgt); 257 258 /* 259 * SME support may update early_pmd_flags to include the memory 260 * encryption mask, so it needs to be called before anything 261 * that may generate a page fault. 262 */ 263 sme_early_init(); 264 265 kasan_early_init(); 266 267 /* 268 * Flush global TLB entries which could be left over from the trampoline page 269 * table. 270 * 271 * This needs to happen *after* kasan_early_init() as KASAN-enabled .configs 272 * instrument native_write_cr4() so KASAN must be initialized for that 273 * instrumentation to work. 274 */ 275 __native_tlb_flush_global(this_cpu_read(cpu_tlbstate.cr4)); 276 277 idt_setup_early_handler(); 278 279 /* Needed before cc_platform_has() can be used for TDX */ 280 tdx_early_init(); 281 282 copy_bootdata(__va(real_mode_data)); 283 284 /* 285 * Load microcode early on BSP. 286 */ 287 load_ucode_bsp(); 288 289 /* set init_top_pgt kernel high mapping*/ 290 init_top_pgt[511] = early_top_pgt[511]; 291 292 x86_64_start_reservations(real_mode_data); 293 } 294 295 void __init __noreturn x86_64_start_reservations(char *real_mode_data) 296 { 297 /* version is always not zero if it is copied */ 298 if (!boot_params.hdr.version) 299 copy_bootdata(__va(real_mode_data)); 300 301 x86_early_init_platform_quirks(); 302 303 switch (boot_params.hdr.hardware_subarch) { 304 case X86_SUBARCH_INTEL_MID: 305 x86_intel_mid_early_setup(); 306 break; 307 default: 308 break; 309 } 310 311 start_kernel(); 312 } 313 314 void early_setup_idt(void) 315 { 316 void *handler = NULL; 317 318 if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT)) { 319 setup_ghcb(); 320 handler = vc_boot_ghcb; 321 } 322 323 startup_64_load_idt(handler); 324 } 325