1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * This code is used on x86_64 to create page table identity mappings on 4 * demand by building up a new set of page tables (or appending to the 5 * existing ones), and then switching over to them when ready. 6 * 7 * Copyright (C) 2015-2016 Yinghai Lu 8 * Copyright (C) 2016 Kees Cook 9 */ 10 11 /* 12 * Since we're dealing with identity mappings, physical and virtual 13 * addresses are the same, so override these defines which are ultimately 14 * used by the headers in misc.h. 15 */ 16 #define __pa(x) ((unsigned long)(x)) 17 #define __va(x) ((void *)((unsigned long)(x))) 18 19 /* No PAGE_TABLE_ISOLATION support needed either: */ 20 #undef CONFIG_PAGE_TABLE_ISOLATION 21 22 #include "error.h" 23 #include "misc.h" 24 25 /* These actually do the work of building the kernel identity maps. */ 26 #include <linux/pgtable.h> 27 #include <asm/cmpxchg.h> 28 #include <asm/trap_pf.h> 29 #include <asm/trapnr.h> 30 #include <asm/init.h> 31 /* Use the static base for this part of the boot process */ 32 #undef __PAGE_OFFSET 33 #define __PAGE_OFFSET __PAGE_OFFSET_BASE 34 #include "../../mm/ident_map.c" 35 36 #define _SETUP 37 #include <asm/setup.h> /* For COMMAND_LINE_SIZE */ 38 #undef _SETUP 39 40 extern unsigned long get_cmd_line_ptr(void); 41 42 /* Used by PAGE_KERN* macros: */ 43 pteval_t __default_kernel_pte_mask __read_mostly = ~0; 44 45 /* Used to track our page table allocation area. */ 46 struct alloc_pgt_data { 47 unsigned char *pgt_buf; 48 unsigned long pgt_buf_size; 49 unsigned long pgt_buf_offset; 50 }; 51 52 /* 53 * Allocates space for a page table entry, using struct alloc_pgt_data 54 * above. Besides the local callers, this is used as the allocation 55 * callback in mapping_info below. 56 */ 57 static void *alloc_pgt_page(void *context) 58 { 59 struct alloc_pgt_data *pages = (struct alloc_pgt_data *)context; 60 unsigned char *entry; 61 62 /* Validate there is space available for a new page. */ 63 if (pages->pgt_buf_offset >= pages->pgt_buf_size) { 64 debug_putstr("out of pgt_buf in " __FILE__ "!?\n"); 65 debug_putaddr(pages->pgt_buf_offset); 66 debug_putaddr(pages->pgt_buf_size); 67 return NULL; 68 } 69 70 entry = pages->pgt_buf + pages->pgt_buf_offset; 71 pages->pgt_buf_offset += PAGE_SIZE; 72 73 return entry; 74 } 75 76 /* Used to track our allocated page tables. */ 77 static struct alloc_pgt_data pgt_data; 78 79 /* The top level page table entry pointer. */ 80 static unsigned long top_level_pgt; 81 82 phys_addr_t physical_mask = (1ULL << __PHYSICAL_MASK_SHIFT) - 1; 83 84 /* 85 * Mapping information structure passed to kernel_ident_mapping_init(). 86 * Due to relocation, pointers must be assigned at run time not build time. 87 */ 88 static struct x86_mapping_info mapping_info; 89 90 /* 91 * Adds the specified range to the identity mappings. 92 */ 93 void kernel_add_identity_map(unsigned long start, unsigned long end) 94 { 95 int ret; 96 97 /* Align boundary to 2M. */ 98 start = round_down(start, PMD_SIZE); 99 end = round_up(end, PMD_SIZE); 100 if (start >= end) 101 return; 102 103 /* Build the mapping. */ 104 ret = kernel_ident_mapping_init(&mapping_info, (pgd_t *)top_level_pgt, start, end); 105 if (ret) 106 error("Error: kernel_ident_mapping_init() failed\n"); 107 } 108 109 /* Locates and clears a region for a new top level page table. */ 110 void initialize_identity_maps(void *rmode) 111 { 112 unsigned long cmdline; 113 114 /* Exclude the encryption mask from __PHYSICAL_MASK */ 115 physical_mask &= ~sme_me_mask; 116 117 /* Init mapping_info with run-time function/buffer pointers. */ 118 mapping_info.alloc_pgt_page = alloc_pgt_page; 119 mapping_info.context = &pgt_data; 120 mapping_info.page_flag = __PAGE_KERNEL_LARGE_EXEC | sme_me_mask; 121 mapping_info.kernpg_flag = _KERNPG_TABLE; 122 123 /* 124 * It should be impossible for this not to already be true, 125 * but since calling this a second time would rewind the other 126 * counters, let's just make sure this is reset too. 127 */ 128 pgt_data.pgt_buf_offset = 0; 129 130 /* 131 * If we came here via startup_32(), cr3 will be _pgtable already 132 * and we must append to the existing area instead of entirely 133 * overwriting it. 134 * 135 * With 5-level paging, we use '_pgtable' to allocate the p4d page table, 136 * the top-level page table is allocated separately. 137 * 138 * p4d_offset(top_level_pgt, 0) would cover both the 4- and 5-level 139 * cases. On 4-level paging it's equal to 'top_level_pgt'. 140 */ 141 top_level_pgt = read_cr3_pa(); 142 if (p4d_offset((pgd_t *)top_level_pgt, 0) == (p4d_t *)_pgtable) { 143 pgt_data.pgt_buf = _pgtable + BOOT_INIT_PGT_SIZE; 144 pgt_data.pgt_buf_size = BOOT_PGT_SIZE - BOOT_INIT_PGT_SIZE; 145 memset(pgt_data.pgt_buf, 0, pgt_data.pgt_buf_size); 146 } else { 147 pgt_data.pgt_buf = _pgtable; 148 pgt_data.pgt_buf_size = BOOT_PGT_SIZE; 149 memset(pgt_data.pgt_buf, 0, pgt_data.pgt_buf_size); 150 top_level_pgt = (unsigned long)alloc_pgt_page(&pgt_data); 151 } 152 153 /* 154 * New page-table is set up - map the kernel image, boot_params and the 155 * command line. The uncompressed kernel requires boot_params and the 156 * command line to be mapped in the identity mapping. Map them 157 * explicitly here in case the compressed kernel does not touch them, 158 * or does not touch all the pages covering them. 159 */ 160 kernel_add_identity_map((unsigned long)_head, (unsigned long)_end); 161 boot_params = rmode; 162 kernel_add_identity_map((unsigned long)boot_params, (unsigned long)(boot_params + 1)); 163 cmdline = get_cmd_line_ptr(); 164 kernel_add_identity_map(cmdline, cmdline + COMMAND_LINE_SIZE); 165 166 sev_prep_identity_maps(top_level_pgt); 167 168 /* Load the new page-table. */ 169 write_cr3(top_level_pgt); 170 } 171 172 static pte_t *split_large_pmd(struct x86_mapping_info *info, 173 pmd_t *pmdp, unsigned long __address) 174 { 175 unsigned long page_flags; 176 unsigned long address; 177 pte_t *pte; 178 pmd_t pmd; 179 int i; 180 181 pte = (pte_t *)info->alloc_pgt_page(info->context); 182 if (!pte) 183 return NULL; 184 185 address = __address & PMD_MASK; 186 /* No large page - clear PSE flag */ 187 page_flags = info->page_flag & ~_PAGE_PSE; 188 189 /* Populate the PTEs */ 190 for (i = 0; i < PTRS_PER_PMD; i++) { 191 set_pte(&pte[i], __pte(address | page_flags)); 192 address += PAGE_SIZE; 193 } 194 195 /* 196 * Ideally we need to clear the large PMD first and do a TLB 197 * flush before we write the new PMD. But the 2M range of the 198 * PMD might contain the code we execute and/or the stack 199 * we are on, so we can't do that. But that should be safe here 200 * because we are going from large to small mappings and we are 201 * also the only user of the page-table, so there is no chance 202 * of a TLB multihit. 203 */ 204 pmd = __pmd((unsigned long)pte | info->kernpg_flag); 205 set_pmd(pmdp, pmd); 206 /* Flush TLB to establish the new PMD */ 207 write_cr3(top_level_pgt); 208 209 return pte + pte_index(__address); 210 } 211 212 static void clflush_page(unsigned long address) 213 { 214 unsigned int flush_size; 215 char *cl, *start, *end; 216 217 /* 218 * Hardcode cl-size to 64 - CPUID can't be used here because that might 219 * cause another #VC exception and the GHCB is not ready to use yet. 220 */ 221 flush_size = 64; 222 start = (char *)(address & PAGE_MASK); 223 end = start + PAGE_SIZE; 224 225 /* 226 * First make sure there are no pending writes on the cache-lines to 227 * flush. 228 */ 229 asm volatile("mfence" : : : "memory"); 230 231 for (cl = start; cl != end; cl += flush_size) 232 clflush(cl); 233 } 234 235 static int set_clr_page_flags(struct x86_mapping_info *info, 236 unsigned long address, 237 pteval_t set, pteval_t clr) 238 { 239 pgd_t *pgdp = (pgd_t *)top_level_pgt; 240 p4d_t *p4dp; 241 pud_t *pudp; 242 pmd_t *pmdp; 243 pte_t *ptep, pte; 244 245 /* 246 * First make sure there is a PMD mapping for 'address'. 247 * It should already exist, but keep things generic. 248 * 249 * To map the page just read from it and fault it in if there is no 250 * mapping yet. kernel_add_identity_map() can't be called here because 251 * that would unconditionally map the address on PMD level, destroying 252 * any PTE-level mappings that might already exist. Use assembly here 253 * so the access won't be optimized away. 254 */ 255 asm volatile("mov %[address], %%r9" 256 :: [address] "g" (*(unsigned long *)address) 257 : "r9", "memory"); 258 259 /* 260 * The page is mapped at least with PMD size - so skip checks and walk 261 * directly to the PMD. 262 */ 263 p4dp = p4d_offset(pgdp, address); 264 pudp = pud_offset(p4dp, address); 265 pmdp = pmd_offset(pudp, address); 266 267 if (pmd_large(*pmdp)) 268 ptep = split_large_pmd(info, pmdp, address); 269 else 270 ptep = pte_offset_kernel(pmdp, address); 271 272 if (!ptep) 273 return -ENOMEM; 274 275 /* 276 * Changing encryption attributes of a page requires to flush it from 277 * the caches. 278 */ 279 if ((set | clr) & _PAGE_ENC) { 280 clflush_page(address); 281 282 /* 283 * If the encryption attribute is being cleared, change the page state 284 * to shared in the RMP table. 285 */ 286 if (clr) 287 snp_set_page_shared(__pa(address & PAGE_MASK)); 288 } 289 290 /* Update PTE */ 291 pte = *ptep; 292 pte = pte_set_flags(pte, set); 293 pte = pte_clear_flags(pte, clr); 294 set_pte(ptep, pte); 295 296 /* 297 * If the encryption attribute is being set, then change the page state to 298 * private in the RMP entry. The page state change must be done after the PTE 299 * is updated. 300 */ 301 if (set & _PAGE_ENC) 302 snp_set_page_private(__pa(address & PAGE_MASK)); 303 304 /* Flush TLB after changing encryption attribute */ 305 write_cr3(top_level_pgt); 306 307 return 0; 308 } 309 310 int set_page_decrypted(unsigned long address) 311 { 312 return set_clr_page_flags(&mapping_info, address, 0, _PAGE_ENC); 313 } 314 315 int set_page_encrypted(unsigned long address) 316 { 317 return set_clr_page_flags(&mapping_info, address, _PAGE_ENC, 0); 318 } 319 320 int set_page_non_present(unsigned long address) 321 { 322 return set_clr_page_flags(&mapping_info, address, 0, _PAGE_PRESENT); 323 } 324 325 static void do_pf_error(const char *msg, unsigned long error_code, 326 unsigned long address, unsigned long ip) 327 { 328 error_putstr(msg); 329 330 error_putstr("\nError Code: "); 331 error_puthex(error_code); 332 error_putstr("\nCR2: 0x"); 333 error_puthex(address); 334 error_putstr("\nRIP relative to _head: 0x"); 335 error_puthex(ip - (unsigned long)_head); 336 error_putstr("\n"); 337 338 error("Stopping.\n"); 339 } 340 341 void do_boot_page_fault(struct pt_regs *regs, unsigned long error_code) 342 { 343 unsigned long address = native_read_cr2(); 344 unsigned long end; 345 bool ghcb_fault; 346 347 ghcb_fault = sev_es_check_ghcb_fault(address); 348 349 address &= PMD_MASK; 350 end = address + PMD_SIZE; 351 352 /* 353 * Check for unexpected error codes. Unexpected are: 354 * - Faults on present pages 355 * - User faults 356 * - Reserved bits set 357 */ 358 if (error_code & (X86_PF_PROT | X86_PF_USER | X86_PF_RSVD)) 359 do_pf_error("Unexpected page-fault:", error_code, address, regs->ip); 360 else if (ghcb_fault) 361 do_pf_error("Page-fault on GHCB page:", error_code, address, regs->ip); 362 363 /* 364 * Error code is sane - now identity map the 2M region around 365 * the faulting address. 366 */ 367 kernel_add_identity_map(address, end); 368 } 369