1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * Copyright (C) 2012 ARM Ltd. 4 */ 5 #ifndef __ASM_PGTABLE_H 6 #define __ASM_PGTABLE_H 7 8 #include <asm/bug.h> 9 #include <asm/proc-fns.h> 10 11 #include <asm/memory.h> 12 #include <asm/pgtable-hwdef.h> 13 #include <asm/pgtable-prot.h> 14 #include <asm/tlbflush.h> 15 16 /* 17 * VMALLOC range. 18 * 19 * VMALLOC_START: beginning of the kernel vmalloc space 20 * VMALLOC_END: extends to the available space below vmmemmap, PCI I/O space 21 * and fixed mappings 22 */ 23 #define VMALLOC_START (MODULES_END) 24 #define VMALLOC_END (- PUD_SIZE - VMEMMAP_SIZE - SZ_64K) 25 26 #define FIRST_USER_ADDRESS 0UL 27 28 #ifndef __ASSEMBLY__ 29 30 #include <asm/cmpxchg.h> 31 #include <asm/fixmap.h> 32 #include <linux/mmdebug.h> 33 #include <linux/mm_types.h> 34 #include <linux/sched.h> 35 36 extern struct page *vmemmap; 37 38 extern void __pte_error(const char *file, int line, unsigned long val); 39 extern void __pmd_error(const char *file, int line, unsigned long val); 40 extern void __pud_error(const char *file, int line, unsigned long val); 41 extern void __pgd_error(const char *file, int line, unsigned long val); 42 43 /* 44 * ZERO_PAGE is a global shared page that is always zero: used 45 * for zero-mapped memory areas etc.. 46 */ 47 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]; 48 #define ZERO_PAGE(vaddr) phys_to_page(__pa_symbol(empty_zero_page)) 49 50 #define pte_ERROR(pte) __pte_error(__FILE__, __LINE__, pte_val(pte)) 51 52 /* 53 * Macros to convert between a physical address and its placement in a 54 * page table entry, taking care of 52-bit addresses. 55 */ 56 #ifdef CONFIG_ARM64_PA_BITS_52 57 #define __pte_to_phys(pte) \ 58 ((pte_val(pte) & PTE_ADDR_LOW) | ((pte_val(pte) & PTE_ADDR_HIGH) << 36)) 59 #define __phys_to_pte_val(phys) (((phys) | ((phys) >> 36)) & PTE_ADDR_MASK) 60 #else 61 #define __pte_to_phys(pte) (pte_val(pte) & PTE_ADDR_MASK) 62 #define __phys_to_pte_val(phys) (phys) 63 #endif 64 65 #define pte_pfn(pte) (__pte_to_phys(pte) >> PAGE_SHIFT) 66 #define pfn_pte(pfn,prot) \ 67 __pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)) 68 69 #define pte_none(pte) (!pte_val(pte)) 70 #define pte_clear(mm,addr,ptep) set_pte(ptep, __pte(0)) 71 #define pte_page(pte) (pfn_to_page(pte_pfn(pte))) 72 73 /* 74 * The following only work if pte_present(). Undefined behaviour otherwise. 75 */ 76 #define pte_present(pte) (!!(pte_val(pte) & (PTE_VALID | PTE_PROT_NONE))) 77 #define pte_young(pte) (!!(pte_val(pte) & PTE_AF)) 78 #define pte_special(pte) (!!(pte_val(pte) & PTE_SPECIAL)) 79 #define pte_write(pte) (!!(pte_val(pte) & PTE_WRITE)) 80 #define pte_user_exec(pte) (!(pte_val(pte) & PTE_UXN)) 81 #define pte_cont(pte) (!!(pte_val(pte) & PTE_CONT)) 82 #define pte_devmap(pte) (!!(pte_val(pte) & PTE_DEVMAP)) 83 84 #define pte_cont_addr_end(addr, end) \ 85 ({ unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK; \ 86 (__boundary - 1 < (end) - 1) ? __boundary : (end); \ 87 }) 88 89 #define pmd_cont_addr_end(addr, end) \ 90 ({ unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK; \ 91 (__boundary - 1 < (end) - 1) ? __boundary : (end); \ 92 }) 93 94 #define pte_hw_dirty(pte) (pte_write(pte) && !(pte_val(pte) & PTE_RDONLY)) 95 #define pte_sw_dirty(pte) (!!(pte_val(pte) & PTE_DIRTY)) 96 #define pte_dirty(pte) (pte_sw_dirty(pte) || pte_hw_dirty(pte)) 97 98 #define pte_valid(pte) (!!(pte_val(pte) & PTE_VALID)) 99 /* 100 * Execute-only user mappings do not have the PTE_USER bit set. All valid 101 * kernel mappings have the PTE_UXN bit set. 102 */ 103 #define pte_valid_not_user(pte) \ 104 ((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN)) 105 #define pte_valid_young(pte) \ 106 ((pte_val(pte) & (PTE_VALID | PTE_AF)) == (PTE_VALID | PTE_AF)) 107 #define pte_valid_user(pte) \ 108 ((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER)) 109 110 /* 111 * Could the pte be present in the TLB? We must check mm_tlb_flush_pending 112 * so that we don't erroneously return false for pages that have been 113 * remapped as PROT_NONE but are yet to be flushed from the TLB. 114 */ 115 #define pte_accessible(mm, pte) \ 116 (mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid_young(pte)) 117 118 /* 119 * p??_access_permitted() is true for valid user mappings (subject to the 120 * write permission check) other than user execute-only which do not have the 121 * PTE_USER bit set. PROT_NONE mappings do not have the PTE_VALID bit set. 122 */ 123 #define pte_access_permitted(pte, write) \ 124 (pte_valid_user(pte) && (!(write) || pte_write(pte))) 125 #define pmd_access_permitted(pmd, write) \ 126 (pte_access_permitted(pmd_pte(pmd), (write))) 127 #define pud_access_permitted(pud, write) \ 128 (pte_access_permitted(pud_pte(pud), (write))) 129 130 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot) 131 { 132 pte_val(pte) &= ~pgprot_val(prot); 133 return pte; 134 } 135 136 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot) 137 { 138 pte_val(pte) |= pgprot_val(prot); 139 return pte; 140 } 141 142 static inline pte_t pte_wrprotect(pte_t pte) 143 { 144 pte = clear_pte_bit(pte, __pgprot(PTE_WRITE)); 145 pte = set_pte_bit(pte, __pgprot(PTE_RDONLY)); 146 return pte; 147 } 148 149 static inline pte_t pte_mkwrite(pte_t pte) 150 { 151 pte = set_pte_bit(pte, __pgprot(PTE_WRITE)); 152 pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY)); 153 return pte; 154 } 155 156 static inline pte_t pte_mkclean(pte_t pte) 157 { 158 pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY)); 159 pte = set_pte_bit(pte, __pgprot(PTE_RDONLY)); 160 161 return pte; 162 } 163 164 static inline pte_t pte_mkdirty(pte_t pte) 165 { 166 pte = set_pte_bit(pte, __pgprot(PTE_DIRTY)); 167 168 if (pte_write(pte)) 169 pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY)); 170 171 return pte; 172 } 173 174 static inline pte_t pte_mkold(pte_t pte) 175 { 176 return clear_pte_bit(pte, __pgprot(PTE_AF)); 177 } 178 179 static inline pte_t pte_mkyoung(pte_t pte) 180 { 181 return set_pte_bit(pte, __pgprot(PTE_AF)); 182 } 183 184 static inline pte_t pte_mkspecial(pte_t pte) 185 { 186 return set_pte_bit(pte, __pgprot(PTE_SPECIAL)); 187 } 188 189 static inline pte_t pte_mkcont(pte_t pte) 190 { 191 pte = set_pte_bit(pte, __pgprot(PTE_CONT)); 192 return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE)); 193 } 194 195 static inline pte_t pte_mknoncont(pte_t pte) 196 { 197 return clear_pte_bit(pte, __pgprot(PTE_CONT)); 198 } 199 200 static inline pte_t pte_mkpresent(pte_t pte) 201 { 202 return set_pte_bit(pte, __pgprot(PTE_VALID)); 203 } 204 205 static inline pmd_t pmd_mkcont(pmd_t pmd) 206 { 207 return __pmd(pmd_val(pmd) | PMD_SECT_CONT); 208 } 209 210 static inline pte_t pte_mkdevmap(pte_t pte) 211 { 212 return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL)); 213 } 214 215 static inline void set_pte(pte_t *ptep, pte_t pte) 216 { 217 WRITE_ONCE(*ptep, pte); 218 219 /* 220 * Only if the new pte is valid and kernel, otherwise TLB maintenance 221 * or update_mmu_cache() have the necessary barriers. 222 */ 223 if (pte_valid_not_user(pte)) { 224 dsb(ishst); 225 isb(); 226 } 227 } 228 229 extern void __sync_icache_dcache(pte_t pteval); 230 231 /* 232 * PTE bits configuration in the presence of hardware Dirty Bit Management 233 * (PTE_WRITE == PTE_DBM): 234 * 235 * Dirty Writable | PTE_RDONLY PTE_WRITE PTE_DIRTY (sw) 236 * 0 0 | 1 0 0 237 * 0 1 | 1 1 0 238 * 1 0 | 1 0 1 239 * 1 1 | 0 1 x 240 * 241 * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via 242 * the page fault mechanism. Checking the dirty status of a pte becomes: 243 * 244 * PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY) 245 */ 246 247 static inline void __check_racy_pte_update(struct mm_struct *mm, pte_t *ptep, 248 pte_t pte) 249 { 250 pte_t old_pte; 251 252 if (!IS_ENABLED(CONFIG_DEBUG_VM)) 253 return; 254 255 old_pte = READ_ONCE(*ptep); 256 257 if (!pte_valid(old_pte) || !pte_valid(pte)) 258 return; 259 if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1) 260 return; 261 262 /* 263 * Check for potential race with hardware updates of the pte 264 * (ptep_set_access_flags safely changes valid ptes without going 265 * through an invalid entry). 266 */ 267 VM_WARN_ONCE(!pte_young(pte), 268 "%s: racy access flag clearing: 0x%016llx -> 0x%016llx", 269 __func__, pte_val(old_pte), pte_val(pte)); 270 VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte), 271 "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx", 272 __func__, pte_val(old_pte), pte_val(pte)); 273 } 274 275 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr, 276 pte_t *ptep, pte_t pte) 277 { 278 if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte)) 279 __sync_icache_dcache(pte); 280 281 __check_racy_pte_update(mm, ptep, pte); 282 283 set_pte(ptep, pte); 284 } 285 286 #define __HAVE_ARCH_PTE_SAME 287 static inline int pte_same(pte_t pte_a, pte_t pte_b) 288 { 289 pteval_t lhs, rhs; 290 291 lhs = pte_val(pte_a); 292 rhs = pte_val(pte_b); 293 294 if (pte_present(pte_a)) 295 lhs &= ~PTE_RDONLY; 296 297 if (pte_present(pte_b)) 298 rhs &= ~PTE_RDONLY; 299 300 return (lhs == rhs); 301 } 302 303 /* 304 * Huge pte definitions. 305 */ 306 #define pte_mkhuge(pte) (__pte(pte_val(pte) & ~PTE_TABLE_BIT)) 307 308 /* 309 * Hugetlb definitions. 310 */ 311 #define HUGE_MAX_HSTATE 4 312 #define HPAGE_SHIFT PMD_SHIFT 313 #define HPAGE_SIZE (_AC(1, UL) << HPAGE_SHIFT) 314 #define HPAGE_MASK (~(HPAGE_SIZE - 1)) 315 #define HUGETLB_PAGE_ORDER (HPAGE_SHIFT - PAGE_SHIFT) 316 317 static inline pte_t pgd_pte(pgd_t pgd) 318 { 319 return __pte(pgd_val(pgd)); 320 } 321 322 static inline pte_t pud_pte(pud_t pud) 323 { 324 return __pte(pud_val(pud)); 325 } 326 327 static inline pud_t pte_pud(pte_t pte) 328 { 329 return __pud(pte_val(pte)); 330 } 331 332 static inline pmd_t pud_pmd(pud_t pud) 333 { 334 return __pmd(pud_val(pud)); 335 } 336 337 static inline pte_t pmd_pte(pmd_t pmd) 338 { 339 return __pte(pmd_val(pmd)); 340 } 341 342 static inline pmd_t pte_pmd(pte_t pte) 343 { 344 return __pmd(pte_val(pte)); 345 } 346 347 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot) 348 { 349 return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT); 350 } 351 352 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot) 353 { 354 return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT); 355 } 356 357 #ifdef CONFIG_NUMA_BALANCING 358 /* 359 * See the comment in include/asm-generic/pgtable.h 360 */ 361 static inline int pte_protnone(pte_t pte) 362 { 363 return (pte_val(pte) & (PTE_VALID | PTE_PROT_NONE)) == PTE_PROT_NONE; 364 } 365 366 static inline int pmd_protnone(pmd_t pmd) 367 { 368 return pte_protnone(pmd_pte(pmd)); 369 } 370 #endif 371 372 /* 373 * THP definitions. 374 */ 375 376 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 377 #define pmd_trans_huge(pmd) (pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT)) 378 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 379 380 #define pmd_present(pmd) pte_present(pmd_pte(pmd)) 381 #define pmd_dirty(pmd) pte_dirty(pmd_pte(pmd)) 382 #define pmd_young(pmd) pte_young(pmd_pte(pmd)) 383 #define pmd_valid(pmd) pte_valid(pmd_pte(pmd)) 384 #define pmd_wrprotect(pmd) pte_pmd(pte_wrprotect(pmd_pte(pmd))) 385 #define pmd_mkold(pmd) pte_pmd(pte_mkold(pmd_pte(pmd))) 386 #define pmd_mkwrite(pmd) pte_pmd(pte_mkwrite(pmd_pte(pmd))) 387 #define pmd_mkclean(pmd) pte_pmd(pte_mkclean(pmd_pte(pmd))) 388 #define pmd_mkdirty(pmd) pte_pmd(pte_mkdirty(pmd_pte(pmd))) 389 #define pmd_mkyoung(pmd) pte_pmd(pte_mkyoung(pmd_pte(pmd))) 390 #define pmd_mknotpresent(pmd) (__pmd(pmd_val(pmd) & ~PMD_SECT_VALID)) 391 392 #define pmd_thp_or_huge(pmd) (pmd_huge(pmd) || pmd_trans_huge(pmd)) 393 394 #define pmd_write(pmd) pte_write(pmd_pte(pmd)) 395 396 #define pmd_mkhuge(pmd) (__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT)) 397 398 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 399 #define pmd_devmap(pmd) pte_devmap(pmd_pte(pmd)) 400 #endif 401 static inline pmd_t pmd_mkdevmap(pmd_t pmd) 402 { 403 return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP))); 404 } 405 406 #define __pmd_to_phys(pmd) __pte_to_phys(pmd_pte(pmd)) 407 #define __phys_to_pmd_val(phys) __phys_to_pte_val(phys) 408 #define pmd_pfn(pmd) ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT) 409 #define pfn_pmd(pfn,prot) __pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)) 410 #define mk_pmd(page,prot) pfn_pmd(page_to_pfn(page),prot) 411 412 #define pud_young(pud) pte_young(pud_pte(pud)) 413 #define pud_mkyoung(pud) pte_pud(pte_mkyoung(pud_pte(pud))) 414 #define pud_write(pud) pte_write(pud_pte(pud)) 415 416 #define pud_mkhuge(pud) (__pud(pud_val(pud) & ~PUD_TABLE_BIT)) 417 418 #define __pud_to_phys(pud) __pte_to_phys(pud_pte(pud)) 419 #define __phys_to_pud_val(phys) __phys_to_pte_val(phys) 420 #define pud_pfn(pud) ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT) 421 #define pfn_pud(pfn,prot) __pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot)) 422 423 #define set_pmd_at(mm, addr, pmdp, pmd) set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd)) 424 425 #define __pgd_to_phys(pgd) __pte_to_phys(pgd_pte(pgd)) 426 #define __phys_to_pgd_val(phys) __phys_to_pte_val(phys) 427 428 #define __pgprot_modify(prot,mask,bits) \ 429 __pgprot((pgprot_val(prot) & ~(mask)) | (bits)) 430 431 /* 432 * Mark the prot value as uncacheable and unbufferable. 433 */ 434 #define pgprot_noncached(prot) \ 435 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN) 436 #define pgprot_writecombine(prot) \ 437 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN) 438 #define pgprot_device(prot) \ 439 __pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN) 440 /* 441 * DMA allocations for non-coherent devices use what the Arm architecture calls 442 * "Normal non-cacheable" memory, which permits speculation, unaligned accesses 443 * and merging of writes. This is different from "Device-nGnR[nE]" memory which 444 * is intended for MMIO and thus forbids speculation, preserves access size, 445 * requires strict alignment and can also force write responses to come from the 446 * endpoint. 447 */ 448 #define pgprot_dmacoherent(prot) \ 449 __pgprot_modify(prot, PTE_ATTRINDX_MASK, \ 450 PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN) 451 452 #define __HAVE_PHYS_MEM_ACCESS_PROT 453 struct file; 454 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, 455 unsigned long size, pgprot_t vma_prot); 456 457 #define pmd_none(pmd) (!pmd_val(pmd)) 458 459 #define pmd_bad(pmd) (!(pmd_val(pmd) & PMD_TABLE_BIT)) 460 461 #define pmd_table(pmd) ((pmd_val(pmd) & PMD_TYPE_MASK) == \ 462 PMD_TYPE_TABLE) 463 #define pmd_sect(pmd) ((pmd_val(pmd) & PMD_TYPE_MASK) == \ 464 PMD_TYPE_SECT) 465 466 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3 467 static inline bool pud_sect(pud_t pud) { return false; } 468 static inline bool pud_table(pud_t pud) { return true; } 469 #else 470 #define pud_sect(pud) ((pud_val(pud) & PUD_TYPE_MASK) == \ 471 PUD_TYPE_SECT) 472 #define pud_table(pud) ((pud_val(pud) & PUD_TYPE_MASK) == \ 473 PUD_TYPE_TABLE) 474 #endif 475 476 extern pgd_t init_pg_dir[PTRS_PER_PGD]; 477 extern pgd_t init_pg_end[]; 478 extern pgd_t swapper_pg_dir[PTRS_PER_PGD]; 479 extern pgd_t idmap_pg_dir[PTRS_PER_PGD]; 480 extern pgd_t tramp_pg_dir[PTRS_PER_PGD]; 481 482 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd); 483 484 static inline bool in_swapper_pgdir(void *addr) 485 { 486 return ((unsigned long)addr & PAGE_MASK) == 487 ((unsigned long)swapper_pg_dir & PAGE_MASK); 488 } 489 490 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd) 491 { 492 #ifdef __PAGETABLE_PMD_FOLDED 493 if (in_swapper_pgdir(pmdp)) { 494 set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd))); 495 return; 496 } 497 #endif /* __PAGETABLE_PMD_FOLDED */ 498 499 WRITE_ONCE(*pmdp, pmd); 500 501 if (pmd_valid(pmd)) { 502 dsb(ishst); 503 isb(); 504 } 505 } 506 507 static inline void pmd_clear(pmd_t *pmdp) 508 { 509 set_pmd(pmdp, __pmd(0)); 510 } 511 512 static inline phys_addr_t pmd_page_paddr(pmd_t pmd) 513 { 514 return __pmd_to_phys(pmd); 515 } 516 517 static inline void pte_unmap(pte_t *pte) { } 518 519 /* Find an entry in the third-level page table. */ 520 #define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) 521 522 #define pte_offset_phys(dir,addr) (pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t)) 523 #define pte_offset_kernel(dir,addr) ((pte_t *)__va(pte_offset_phys((dir), (addr)))) 524 525 #define pte_offset_map(dir,addr) pte_offset_kernel((dir), (addr)) 526 527 #define pte_set_fixmap(addr) ((pte_t *)set_fixmap_offset(FIX_PTE, addr)) 528 #define pte_set_fixmap_offset(pmd, addr) pte_set_fixmap(pte_offset_phys(pmd, addr)) 529 #define pte_clear_fixmap() clear_fixmap(FIX_PTE) 530 531 #define pmd_page(pmd) pfn_to_page(__phys_to_pfn(__pmd_to_phys(pmd))) 532 533 /* use ONLY for statically allocated translation tables */ 534 #define pte_offset_kimg(dir,addr) ((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr)))) 535 536 /* 537 * Conversion functions: convert a page and protection to a page entry, 538 * and a page entry and page directory to the page they refer to. 539 */ 540 #define mk_pte(page,prot) pfn_pte(page_to_pfn(page),prot) 541 542 #if CONFIG_PGTABLE_LEVELS > 2 543 544 #define pmd_ERROR(pmd) __pmd_error(__FILE__, __LINE__, pmd_val(pmd)) 545 546 #define pud_none(pud) (!pud_val(pud)) 547 #define pud_bad(pud) (!(pud_val(pud) & PUD_TABLE_BIT)) 548 #define pud_present(pud) pte_present(pud_pte(pud)) 549 #define pud_valid(pud) pte_valid(pud_pte(pud)) 550 551 static inline void set_pud(pud_t *pudp, pud_t pud) 552 { 553 #ifdef __PAGETABLE_PUD_FOLDED 554 if (in_swapper_pgdir(pudp)) { 555 set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud))); 556 return; 557 } 558 #endif /* __PAGETABLE_PUD_FOLDED */ 559 560 WRITE_ONCE(*pudp, pud); 561 562 if (pud_valid(pud)) { 563 dsb(ishst); 564 isb(); 565 } 566 } 567 568 static inline void pud_clear(pud_t *pudp) 569 { 570 set_pud(pudp, __pud(0)); 571 } 572 573 static inline phys_addr_t pud_page_paddr(pud_t pud) 574 { 575 return __pud_to_phys(pud); 576 } 577 578 /* Find an entry in the second-level page table. */ 579 #define pmd_index(addr) (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)) 580 581 #define pmd_offset_phys(dir, addr) (pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t)) 582 #define pmd_offset(dir, addr) ((pmd_t *)__va(pmd_offset_phys((dir), (addr)))) 583 584 #define pmd_set_fixmap(addr) ((pmd_t *)set_fixmap_offset(FIX_PMD, addr)) 585 #define pmd_set_fixmap_offset(pud, addr) pmd_set_fixmap(pmd_offset_phys(pud, addr)) 586 #define pmd_clear_fixmap() clear_fixmap(FIX_PMD) 587 588 #define pud_page(pud) pfn_to_page(__phys_to_pfn(__pud_to_phys(pud))) 589 590 /* use ONLY for statically allocated translation tables */ 591 #define pmd_offset_kimg(dir,addr) ((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr)))) 592 593 #else 594 595 #define pud_page_paddr(pud) ({ BUILD_BUG(); 0; }) 596 597 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */ 598 #define pmd_set_fixmap(addr) NULL 599 #define pmd_set_fixmap_offset(pudp, addr) ((pmd_t *)pudp) 600 #define pmd_clear_fixmap() 601 602 #define pmd_offset_kimg(dir,addr) ((pmd_t *)dir) 603 604 #endif /* CONFIG_PGTABLE_LEVELS > 2 */ 605 606 #if CONFIG_PGTABLE_LEVELS > 3 607 608 #define pud_ERROR(pud) __pud_error(__FILE__, __LINE__, pud_val(pud)) 609 610 #define pgd_none(pgd) (!pgd_val(pgd)) 611 #define pgd_bad(pgd) (!(pgd_val(pgd) & 2)) 612 #define pgd_present(pgd) (pgd_val(pgd)) 613 614 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd) 615 { 616 if (in_swapper_pgdir(pgdp)) { 617 set_swapper_pgd(pgdp, pgd); 618 return; 619 } 620 621 WRITE_ONCE(*pgdp, pgd); 622 dsb(ishst); 623 isb(); 624 } 625 626 static inline void pgd_clear(pgd_t *pgdp) 627 { 628 set_pgd(pgdp, __pgd(0)); 629 } 630 631 static inline phys_addr_t pgd_page_paddr(pgd_t pgd) 632 { 633 return __pgd_to_phys(pgd); 634 } 635 636 /* Find an entry in the frst-level page table. */ 637 #define pud_index(addr) (((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1)) 638 639 #define pud_offset_phys(dir, addr) (pgd_page_paddr(READ_ONCE(*(dir))) + pud_index(addr) * sizeof(pud_t)) 640 #define pud_offset(dir, addr) ((pud_t *)__va(pud_offset_phys((dir), (addr)))) 641 642 #define pud_set_fixmap(addr) ((pud_t *)set_fixmap_offset(FIX_PUD, addr)) 643 #define pud_set_fixmap_offset(pgd, addr) pud_set_fixmap(pud_offset_phys(pgd, addr)) 644 #define pud_clear_fixmap() clear_fixmap(FIX_PUD) 645 646 #define pgd_page(pgd) pfn_to_page(__phys_to_pfn(__pgd_to_phys(pgd))) 647 648 /* use ONLY for statically allocated translation tables */ 649 #define pud_offset_kimg(dir,addr) ((pud_t *)__phys_to_kimg(pud_offset_phys((dir), (addr)))) 650 651 #else 652 653 #define pgd_page_paddr(pgd) ({ BUILD_BUG(); 0;}) 654 655 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */ 656 #define pud_set_fixmap(addr) NULL 657 #define pud_set_fixmap_offset(pgdp, addr) ((pud_t *)pgdp) 658 #define pud_clear_fixmap() 659 660 #define pud_offset_kimg(dir,addr) ((pud_t *)dir) 661 662 #endif /* CONFIG_PGTABLE_LEVELS > 3 */ 663 664 #define pgd_ERROR(pgd) __pgd_error(__FILE__, __LINE__, pgd_val(pgd)) 665 666 /* to find an entry in a page-table-directory */ 667 #define pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1)) 668 669 #define pgd_offset_raw(pgd, addr) ((pgd) + pgd_index(addr)) 670 671 #define pgd_offset(mm, addr) (pgd_offset_raw((mm)->pgd, (addr))) 672 673 /* to find an entry in a kernel page-table-directory */ 674 #define pgd_offset_k(addr) pgd_offset(&init_mm, addr) 675 676 #define pgd_set_fixmap(addr) ((pgd_t *)set_fixmap_offset(FIX_PGD, addr)) 677 #define pgd_clear_fixmap() clear_fixmap(FIX_PGD) 678 679 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot) 680 { 681 const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY | 682 PTE_PROT_NONE | PTE_VALID | PTE_WRITE; 683 /* preserve the hardware dirty information */ 684 if (pte_hw_dirty(pte)) 685 pte = pte_mkdirty(pte); 686 pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask); 687 return pte; 688 } 689 690 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot) 691 { 692 return pte_pmd(pte_modify(pmd_pte(pmd), newprot)); 693 } 694 695 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS 696 extern int ptep_set_access_flags(struct vm_area_struct *vma, 697 unsigned long address, pte_t *ptep, 698 pte_t entry, int dirty); 699 700 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 701 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS 702 static inline int pmdp_set_access_flags(struct vm_area_struct *vma, 703 unsigned long address, pmd_t *pmdp, 704 pmd_t entry, int dirty) 705 { 706 return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty); 707 } 708 709 static inline int pud_devmap(pud_t pud) 710 { 711 return 0; 712 } 713 714 static inline int pgd_devmap(pgd_t pgd) 715 { 716 return 0; 717 } 718 #endif 719 720 /* 721 * Atomic pte/pmd modifications. 722 */ 723 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG 724 static inline int __ptep_test_and_clear_young(pte_t *ptep) 725 { 726 pte_t old_pte, pte; 727 728 pte = READ_ONCE(*ptep); 729 do { 730 old_pte = pte; 731 pte = pte_mkold(pte); 732 pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep), 733 pte_val(old_pte), pte_val(pte)); 734 } while (pte_val(pte) != pte_val(old_pte)); 735 736 return pte_young(pte); 737 } 738 739 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma, 740 unsigned long address, 741 pte_t *ptep) 742 { 743 return __ptep_test_and_clear_young(ptep); 744 } 745 746 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH 747 static inline int ptep_clear_flush_young(struct vm_area_struct *vma, 748 unsigned long address, pte_t *ptep) 749 { 750 int young = ptep_test_and_clear_young(vma, address, ptep); 751 752 if (young) { 753 /* 754 * We can elide the trailing DSB here since the worst that can 755 * happen is that a CPU continues to use the young entry in its 756 * TLB and we mistakenly reclaim the associated page. The 757 * window for such an event is bounded by the next 758 * context-switch, which provides a DSB to complete the TLB 759 * invalidation. 760 */ 761 flush_tlb_page_nosync(vma, address); 762 } 763 764 return young; 765 } 766 767 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 768 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG 769 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma, 770 unsigned long address, 771 pmd_t *pmdp) 772 { 773 return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp); 774 } 775 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 776 777 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR 778 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, 779 unsigned long address, pte_t *ptep) 780 { 781 return __pte(xchg_relaxed(&pte_val(*ptep), 0)); 782 } 783 784 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 785 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR 786 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, 787 unsigned long address, pmd_t *pmdp) 788 { 789 return pte_pmd(ptep_get_and_clear(mm, address, (pte_t *)pmdp)); 790 } 791 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 792 793 /* 794 * ptep_set_wrprotect - mark read-only while trasferring potential hardware 795 * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit. 796 */ 797 #define __HAVE_ARCH_PTEP_SET_WRPROTECT 798 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep) 799 { 800 pte_t old_pte, pte; 801 802 pte = READ_ONCE(*ptep); 803 do { 804 old_pte = pte; 805 /* 806 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY 807 * clear), set the PTE_DIRTY bit. 808 */ 809 if (pte_hw_dirty(pte)) 810 pte = pte_mkdirty(pte); 811 pte = pte_wrprotect(pte); 812 pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep), 813 pte_val(old_pte), pte_val(pte)); 814 } while (pte_val(pte) != pte_val(old_pte)); 815 } 816 817 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 818 #define __HAVE_ARCH_PMDP_SET_WRPROTECT 819 static inline void pmdp_set_wrprotect(struct mm_struct *mm, 820 unsigned long address, pmd_t *pmdp) 821 { 822 ptep_set_wrprotect(mm, address, (pte_t *)pmdp); 823 } 824 825 #define pmdp_establish pmdp_establish 826 static inline pmd_t pmdp_establish(struct vm_area_struct *vma, 827 unsigned long address, pmd_t *pmdp, pmd_t pmd) 828 { 829 return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd))); 830 } 831 #endif 832 833 /* 834 * Encode and decode a swap entry: 835 * bits 0-1: present (must be zero) 836 * bits 2-7: swap type 837 * bits 8-57: swap offset 838 * bit 58: PTE_PROT_NONE (must be zero) 839 */ 840 #define __SWP_TYPE_SHIFT 2 841 #define __SWP_TYPE_BITS 6 842 #define __SWP_OFFSET_BITS 50 843 #define __SWP_TYPE_MASK ((1 << __SWP_TYPE_BITS) - 1) 844 #define __SWP_OFFSET_SHIFT (__SWP_TYPE_BITS + __SWP_TYPE_SHIFT) 845 #define __SWP_OFFSET_MASK ((1UL << __SWP_OFFSET_BITS) - 1) 846 847 #define __swp_type(x) (((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK) 848 #define __swp_offset(x) (((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK) 849 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) }) 850 851 #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) }) 852 #define __swp_entry_to_pte(swp) ((pte_t) { (swp).val }) 853 854 /* 855 * Ensure that there are not more swap files than can be encoded in the kernel 856 * PTEs. 857 */ 858 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS) 859 860 extern int kern_addr_valid(unsigned long addr); 861 862 #include <asm-generic/pgtable.h> 863 864 /* 865 * On AArch64, the cache coherency is handled via the set_pte_at() function. 866 */ 867 static inline void update_mmu_cache(struct vm_area_struct *vma, 868 unsigned long addr, pte_t *ptep) 869 { 870 /* 871 * We don't do anything here, so there's a very small chance of 872 * us retaking a user fault which we just fixed up. The alternative 873 * is doing a dsb(ishst), but that penalises the fastpath. 874 */ 875 } 876 877 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0) 878 879 #ifdef CONFIG_ARM64_PA_BITS_52 880 #define phys_to_ttbr(addr) (((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52) 881 #else 882 #define phys_to_ttbr(addr) (addr) 883 #endif 884 885 #endif /* !__ASSEMBLY__ */ 886 887 #endif /* __ASM_PGTABLE_H */ 888