1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * Copyright (C) 2012 Regents of the University of California 4 */ 5 6 #ifndef _ASM_RISCV_PGTABLE_H 7 #define _ASM_RISCV_PGTABLE_H 8 9 #include <linux/mmzone.h> 10 #include <linux/sizes.h> 11 12 #include <asm/pgtable-bits.h> 13 14 #ifndef CONFIG_MMU 15 #ifdef CONFIG_RELOCATABLE 16 #define KERNEL_LINK_ADDR UL(0) 17 #else 18 #define KERNEL_LINK_ADDR _AC(CONFIG_PHYS_RAM_BASE, UL) 19 #endif 20 #define KERN_VIRT_SIZE (UL(-1)) 21 #else 22 23 #define ADDRESS_SPACE_END (UL(-1)) 24 25 #ifdef CONFIG_64BIT 26 /* Leave 2GB for kernel and BPF at the end of the address space */ 27 #define KERNEL_LINK_ADDR (ADDRESS_SPACE_END - SZ_2G + 1) 28 #else 29 #define KERNEL_LINK_ADDR PAGE_OFFSET 30 #endif 31 32 /* Number of entries in the page global directory */ 33 #define PTRS_PER_PGD (PAGE_SIZE / sizeof(pgd_t)) 34 /* Number of entries in the page table */ 35 #define PTRS_PER_PTE (PAGE_SIZE / sizeof(pte_t)) 36 37 /* 38 * Half of the kernel address space (1/4 of the entries of the page global 39 * directory) is for the direct mapping. 40 */ 41 #define KERN_VIRT_SIZE ((PTRS_PER_PGD / 2 * PGDIR_SIZE) / 2) 42 43 #define VMALLOC_SIZE (KERN_VIRT_SIZE >> 1) 44 #define VMALLOC_END PAGE_OFFSET 45 #define VMALLOC_START (PAGE_OFFSET - VMALLOC_SIZE) 46 47 #define BPF_JIT_REGION_SIZE (SZ_128M) 48 #ifdef CONFIG_64BIT 49 #define BPF_JIT_REGION_START (BPF_JIT_REGION_END - BPF_JIT_REGION_SIZE) 50 #define BPF_JIT_REGION_END (MODULES_END) 51 #else 52 #define BPF_JIT_REGION_START (PAGE_OFFSET - BPF_JIT_REGION_SIZE) 53 #define BPF_JIT_REGION_END (VMALLOC_END) 54 #endif 55 56 /* Modules always live before the kernel */ 57 #ifdef CONFIG_64BIT 58 /* This is used to define the end of the KASAN shadow region */ 59 #define MODULES_LOWEST_VADDR (KERNEL_LINK_ADDR - SZ_2G) 60 #define MODULES_VADDR (PFN_ALIGN((unsigned long)&_end) - SZ_2G) 61 #define MODULES_END (PFN_ALIGN((unsigned long)&_start)) 62 #else 63 #define MODULES_VADDR VMALLOC_START 64 #define MODULES_END VMALLOC_END 65 #endif 66 67 /* 68 * Roughly size the vmemmap space to be large enough to fit enough 69 * struct pages to map half the virtual address space. Then 70 * position vmemmap directly below the VMALLOC region. 71 */ 72 #define VA_BITS_SV32 32 73 #ifdef CONFIG_64BIT 74 #define VA_BITS_SV39 39 75 #define VA_BITS_SV48 48 76 #define VA_BITS_SV57 57 77 78 #define VA_BITS (pgtable_l5_enabled ? \ 79 VA_BITS_SV57 : (pgtable_l4_enabled ? VA_BITS_SV48 : VA_BITS_SV39)) 80 #else 81 #define VA_BITS VA_BITS_SV32 82 #endif 83 84 #define VMEMMAP_SHIFT \ 85 (VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT) 86 #define VMEMMAP_SIZE BIT(VMEMMAP_SHIFT) 87 #define VMEMMAP_END VMALLOC_START 88 #define VMEMMAP_START (VMALLOC_START - VMEMMAP_SIZE) 89 90 /* 91 * Define vmemmap for pfn_to_page & page_to_pfn calls. Needed if kernel 92 * is configured with CONFIG_SPARSEMEM_VMEMMAP enabled. 93 */ 94 #define vmemmap ((struct page *)VMEMMAP_START - vmemmap_start_pfn) 95 96 #define PCI_IO_SIZE SZ_16M 97 #define PCI_IO_END VMEMMAP_START 98 #define PCI_IO_START (PCI_IO_END - PCI_IO_SIZE) 99 100 #define FIXADDR_TOP PCI_IO_START 101 #ifdef CONFIG_64BIT 102 #define MAX_FDT_SIZE PMD_SIZE 103 #define FIX_FDT_SIZE (MAX_FDT_SIZE + SZ_2M) 104 #define FIXADDR_SIZE (PMD_SIZE + FIX_FDT_SIZE) 105 #else 106 #define MAX_FDT_SIZE PGDIR_SIZE 107 #define FIX_FDT_SIZE MAX_FDT_SIZE 108 #define FIXADDR_SIZE (PGDIR_SIZE + FIX_FDT_SIZE) 109 #endif 110 #define FIXADDR_START (FIXADDR_TOP - FIXADDR_SIZE) 111 112 #endif 113 114 #ifndef __ASSEMBLY__ 115 116 #include <asm/page.h> 117 #include <asm/tlbflush.h> 118 #include <linux/mm_types.h> 119 #include <asm/compat.h> 120 #include <asm/cpufeature.h> 121 122 #define __page_val_to_pfn(_val) (((_val) & _PAGE_PFN_MASK) >> _PAGE_PFN_SHIFT) 123 124 #ifdef CONFIG_64BIT 125 #include <asm/pgtable-64.h> 126 127 #define VA_USER_SV39 (UL(1) << (VA_BITS_SV39 - 1)) 128 #define VA_USER_SV48 (UL(1) << (VA_BITS_SV48 - 1)) 129 #define VA_USER_SV57 (UL(1) << (VA_BITS_SV57 - 1)) 130 131 #define MMAP_VA_BITS_64 ((VA_BITS >= VA_BITS_SV48) ? VA_BITS_SV48 : VA_BITS) 132 #define MMAP_MIN_VA_BITS_64 (VA_BITS_SV39) 133 #define MMAP_VA_BITS (is_compat_task() ? VA_BITS_SV32 : MMAP_VA_BITS_64) 134 #define MMAP_MIN_VA_BITS (is_compat_task() ? VA_BITS_SV32 : MMAP_MIN_VA_BITS_64) 135 #else 136 #include <asm/pgtable-32.h> 137 #endif /* CONFIG_64BIT */ 138 139 #include <linux/page_table_check.h> 140 141 #ifdef CONFIG_XIP_KERNEL 142 #define XIP_FIXUP(addr) ({ \ 143 extern char _sdata[], _start[], _end[]; \ 144 uintptr_t __rom_start_data = CONFIG_XIP_PHYS_ADDR \ 145 + (uintptr_t)&_sdata - (uintptr_t)&_start; \ 146 uintptr_t __rom_end_data = CONFIG_XIP_PHYS_ADDR \ 147 + (uintptr_t)&_end - (uintptr_t)&_start; \ 148 uintptr_t __a = (uintptr_t)(addr); \ 149 (__a >= __rom_start_data && __a < __rom_end_data) ? \ 150 __a - __rom_start_data + CONFIG_PHYS_RAM_BASE : __a; \ 151 }) 152 #else 153 #define XIP_FIXUP(addr) (addr) 154 #endif /* CONFIG_XIP_KERNEL */ 155 156 struct pt_alloc_ops { 157 pte_t *(*get_pte_virt)(phys_addr_t pa); 158 phys_addr_t (*alloc_pte)(uintptr_t va); 159 #ifndef __PAGETABLE_PMD_FOLDED 160 pmd_t *(*get_pmd_virt)(phys_addr_t pa); 161 phys_addr_t (*alloc_pmd)(uintptr_t va); 162 pud_t *(*get_pud_virt)(phys_addr_t pa); 163 phys_addr_t (*alloc_pud)(uintptr_t va); 164 p4d_t *(*get_p4d_virt)(phys_addr_t pa); 165 phys_addr_t (*alloc_p4d)(uintptr_t va); 166 #endif 167 }; 168 169 extern struct pt_alloc_ops pt_ops __meminitdata; 170 171 #ifdef CONFIG_MMU 172 /* Number of PGD entries that a user-mode program can use */ 173 #define USER_PTRS_PER_PGD (TASK_SIZE / PGDIR_SIZE) 174 175 /* Page protection bits */ 176 #define _PAGE_BASE (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER) 177 178 #define PAGE_NONE __pgprot(_PAGE_PROT_NONE | _PAGE_READ) 179 #define PAGE_READ __pgprot(_PAGE_BASE | _PAGE_READ) 180 #define PAGE_WRITE __pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE) 181 #define PAGE_EXEC __pgprot(_PAGE_BASE | _PAGE_EXEC) 182 #define PAGE_READ_EXEC __pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC) 183 #define PAGE_WRITE_EXEC __pgprot(_PAGE_BASE | _PAGE_READ | \ 184 _PAGE_EXEC | _PAGE_WRITE) 185 186 #define PAGE_COPY PAGE_READ 187 #define PAGE_COPY_EXEC PAGE_READ_EXEC 188 #define PAGE_SHARED PAGE_WRITE 189 #define PAGE_SHARED_EXEC PAGE_WRITE_EXEC 190 191 #define _PAGE_KERNEL (_PAGE_READ \ 192 | _PAGE_WRITE \ 193 | _PAGE_PRESENT \ 194 | _PAGE_ACCESSED \ 195 | _PAGE_DIRTY \ 196 | _PAGE_GLOBAL) 197 198 #define PAGE_KERNEL __pgprot(_PAGE_KERNEL) 199 #define PAGE_KERNEL_READ __pgprot(_PAGE_KERNEL & ~_PAGE_WRITE) 200 #define PAGE_KERNEL_EXEC __pgprot(_PAGE_KERNEL | _PAGE_EXEC) 201 #define PAGE_KERNEL_READ_EXEC __pgprot((_PAGE_KERNEL & ~_PAGE_WRITE) \ 202 | _PAGE_EXEC) 203 204 #define PAGE_TABLE __pgprot(_PAGE_TABLE) 205 206 #define _PAGE_IOREMAP ((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO) 207 #define PAGE_KERNEL_IO __pgprot(_PAGE_IOREMAP) 208 209 extern pgd_t swapper_pg_dir[]; 210 extern pgd_t trampoline_pg_dir[]; 211 extern pgd_t early_pg_dir[]; 212 213 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 214 static inline int pmd_present(pmd_t pmd) 215 { 216 /* 217 * Checking for _PAGE_LEAF is needed too because: 218 * When splitting a THP, split_huge_page() will temporarily clear 219 * the present bit, in this situation, pmd_present() and 220 * pmd_trans_huge() still needs to return true. 221 */ 222 return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE | _PAGE_LEAF)); 223 } 224 #else 225 static inline int pmd_present(pmd_t pmd) 226 { 227 return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE)); 228 } 229 #endif 230 231 static inline int pmd_none(pmd_t pmd) 232 { 233 return (pmd_val(pmd) == 0); 234 } 235 236 static inline int pmd_bad(pmd_t pmd) 237 { 238 return !pmd_present(pmd) || (pmd_val(pmd) & _PAGE_LEAF); 239 } 240 241 #define pmd_leaf pmd_leaf 242 static inline bool pmd_leaf(pmd_t pmd) 243 { 244 return pmd_present(pmd) && (pmd_val(pmd) & _PAGE_LEAF); 245 } 246 247 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd) 248 { 249 WRITE_ONCE(*pmdp, pmd); 250 } 251 252 static inline void pmd_clear(pmd_t *pmdp) 253 { 254 set_pmd(pmdp, __pmd(0)); 255 } 256 257 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot) 258 { 259 unsigned long prot_val = pgprot_val(prot); 260 261 ALT_THEAD_PMA(prot_val); 262 263 return __pgd((pfn << _PAGE_PFN_SHIFT) | prot_val); 264 } 265 266 static inline unsigned long _pgd_pfn(pgd_t pgd) 267 { 268 return __page_val_to_pfn(pgd_val(pgd)); 269 } 270 271 static inline struct page *pmd_page(pmd_t pmd) 272 { 273 return pfn_to_page(__page_val_to_pfn(pmd_val(pmd))); 274 } 275 276 static inline unsigned long pmd_page_vaddr(pmd_t pmd) 277 { 278 return (unsigned long)pfn_to_virt(__page_val_to_pfn(pmd_val(pmd))); 279 } 280 281 static inline pte_t pmd_pte(pmd_t pmd) 282 { 283 return __pte(pmd_val(pmd)); 284 } 285 286 static inline pte_t pud_pte(pud_t pud) 287 { 288 return __pte(pud_val(pud)); 289 } 290 291 #ifdef CONFIG_RISCV_ISA_SVNAPOT 292 293 static __always_inline bool has_svnapot(void) 294 { 295 return riscv_has_extension_likely(RISCV_ISA_EXT_SVNAPOT); 296 } 297 298 static inline unsigned long pte_napot(pte_t pte) 299 { 300 return pte_val(pte) & _PAGE_NAPOT; 301 } 302 303 static inline pte_t pte_mknapot(pte_t pte, unsigned int order) 304 { 305 int pos = order - 1 + _PAGE_PFN_SHIFT; 306 unsigned long napot_bit = BIT(pos); 307 unsigned long napot_mask = ~GENMASK(pos, _PAGE_PFN_SHIFT); 308 309 return __pte((pte_val(pte) & napot_mask) | napot_bit | _PAGE_NAPOT); 310 } 311 312 #else 313 314 static __always_inline bool has_svnapot(void) { return false; } 315 316 static inline unsigned long pte_napot(pte_t pte) 317 { 318 return 0; 319 } 320 321 #endif /* CONFIG_RISCV_ISA_SVNAPOT */ 322 323 /* Yields the page frame number (PFN) of a page table entry */ 324 static inline unsigned long pte_pfn(pte_t pte) 325 { 326 unsigned long res = __page_val_to_pfn(pte_val(pte)); 327 328 if (has_svnapot() && pte_napot(pte)) 329 res = res & (res - 1UL); 330 331 return res; 332 } 333 334 #define pte_page(x) pfn_to_page(pte_pfn(x)) 335 336 /* Constructs a page table entry */ 337 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot) 338 { 339 unsigned long prot_val = pgprot_val(prot); 340 341 ALT_THEAD_PMA(prot_val); 342 343 return __pte((pfn << _PAGE_PFN_SHIFT) | prot_val); 344 } 345 346 #define pte_pgprot pte_pgprot 347 static inline pgprot_t pte_pgprot(pte_t pte) 348 { 349 unsigned long pfn = pte_pfn(pte); 350 351 return __pgprot(pte_val(pfn_pte(pfn, __pgprot(0))) ^ pte_val(pte)); 352 } 353 354 static inline int pte_present(pte_t pte) 355 { 356 return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)); 357 } 358 359 #define pte_accessible pte_accessible 360 static inline unsigned long pte_accessible(struct mm_struct *mm, pte_t a) 361 { 362 if (pte_val(a) & _PAGE_PRESENT) 363 return true; 364 365 if ((pte_val(a) & _PAGE_PROT_NONE) && 366 atomic_read(&mm->tlb_flush_pending)) 367 return true; 368 369 return false; 370 } 371 372 static inline int pte_none(pte_t pte) 373 { 374 return (pte_val(pte) == 0); 375 } 376 377 static inline int pte_write(pte_t pte) 378 { 379 return pte_val(pte) & _PAGE_WRITE; 380 } 381 382 static inline int pte_exec(pte_t pte) 383 { 384 return pte_val(pte) & _PAGE_EXEC; 385 } 386 387 static inline int pte_user(pte_t pte) 388 { 389 return pte_val(pte) & _PAGE_USER; 390 } 391 392 static inline int pte_huge(pte_t pte) 393 { 394 return pte_present(pte) && (pte_val(pte) & _PAGE_LEAF); 395 } 396 397 static inline int pte_dirty(pte_t pte) 398 { 399 return pte_val(pte) & _PAGE_DIRTY; 400 } 401 402 static inline int pte_young(pte_t pte) 403 { 404 return pte_val(pte) & _PAGE_ACCESSED; 405 } 406 407 static inline int pte_special(pte_t pte) 408 { 409 return pte_val(pte) & _PAGE_SPECIAL; 410 } 411 412 /* static inline pte_t pte_rdprotect(pte_t pte) */ 413 414 static inline pte_t pte_wrprotect(pte_t pte) 415 { 416 return __pte(pte_val(pte) & ~(_PAGE_WRITE)); 417 } 418 419 /* static inline pte_t pte_mkread(pte_t pte) */ 420 421 static inline pte_t pte_mkwrite_novma(pte_t pte) 422 { 423 return __pte(pte_val(pte) | _PAGE_WRITE); 424 } 425 426 /* static inline pte_t pte_mkexec(pte_t pte) */ 427 428 static inline pte_t pte_mkdirty(pte_t pte) 429 { 430 return __pte(pte_val(pte) | _PAGE_DIRTY); 431 } 432 433 static inline pte_t pte_mkclean(pte_t pte) 434 { 435 return __pte(pte_val(pte) & ~(_PAGE_DIRTY)); 436 } 437 438 static inline pte_t pte_mkyoung(pte_t pte) 439 { 440 return __pte(pte_val(pte) | _PAGE_ACCESSED); 441 } 442 443 static inline pte_t pte_mkold(pte_t pte) 444 { 445 return __pte(pte_val(pte) & ~(_PAGE_ACCESSED)); 446 } 447 448 static inline pte_t pte_mkspecial(pte_t pte) 449 { 450 return __pte(pte_val(pte) | _PAGE_SPECIAL); 451 } 452 453 static inline pte_t pte_mkhuge(pte_t pte) 454 { 455 return pte; 456 } 457 458 #ifdef CONFIG_RISCV_ISA_SVNAPOT 459 #define pte_leaf_size(pte) (pte_napot(pte) ? \ 460 napot_cont_size(napot_cont_order(pte)) :\ 461 PAGE_SIZE) 462 #endif 463 464 #ifdef CONFIG_NUMA_BALANCING 465 /* 466 * See the comment in include/asm-generic/pgtable.h 467 */ 468 static inline int pte_protnone(pte_t pte) 469 { 470 return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)) == _PAGE_PROT_NONE; 471 } 472 473 static inline int pmd_protnone(pmd_t pmd) 474 { 475 return pte_protnone(pmd_pte(pmd)); 476 } 477 #endif 478 479 /* Modify page protection bits */ 480 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot) 481 { 482 unsigned long newprot_val = pgprot_val(newprot); 483 484 ALT_THEAD_PMA(newprot_val); 485 486 return __pte((pte_val(pte) & _PAGE_CHG_MASK) | newprot_val); 487 } 488 489 #define pgd_ERROR(e) \ 490 pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e)) 491 492 493 /* Commit new configuration to MMU hardware */ 494 static inline void update_mmu_cache_range(struct vm_fault *vmf, 495 struct vm_area_struct *vma, unsigned long address, 496 pte_t *ptep, unsigned int nr) 497 { 498 asm goto(ALTERNATIVE("nop", "j %l[svvptc]", 0, RISCV_ISA_EXT_SVVPTC, 1) 499 : : : : svvptc); 500 501 /* 502 * The kernel assumes that TLBs don't cache invalid entries, but 503 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a 504 * cache flush; it is necessary even after writing invalid entries. 505 * Relying on flush_tlb_fix_spurious_fault would suffice, but 506 * the extra traps reduce performance. So, eagerly SFENCE.VMA. 507 */ 508 while (nr--) 509 local_flush_tlb_page(address + nr * PAGE_SIZE); 510 511 svvptc:; 512 /* 513 * Svvptc guarantees that the new valid pte will be visible within 514 * a bounded timeframe, so when the uarch does not cache invalid 515 * entries, we don't have to do anything. 516 */ 517 } 518 #define update_mmu_cache(vma, addr, ptep) \ 519 update_mmu_cache_range(NULL, vma, addr, ptep, 1) 520 521 #define update_mmu_tlb_range(vma, addr, ptep, nr) \ 522 update_mmu_cache_range(NULL, vma, addr, ptep, nr) 523 524 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma, 525 unsigned long address, pmd_t *pmdp) 526 { 527 pte_t *ptep = (pte_t *)pmdp; 528 529 update_mmu_cache(vma, address, ptep); 530 } 531 532 #define __HAVE_ARCH_PTE_SAME 533 static inline int pte_same(pte_t pte_a, pte_t pte_b) 534 { 535 return pte_val(pte_a) == pte_val(pte_b); 536 } 537 538 /* 539 * Certain architectures need to do special things when PTEs within 540 * a page table are directly modified. Thus, the following hook is 541 * made available. 542 */ 543 static inline void set_pte(pte_t *ptep, pte_t pteval) 544 { 545 WRITE_ONCE(*ptep, pteval); 546 } 547 548 void flush_icache_pte(struct mm_struct *mm, pte_t pte); 549 550 static inline void __set_pte_at(struct mm_struct *mm, pte_t *ptep, pte_t pteval) 551 { 552 if (pte_present(pteval) && pte_exec(pteval)) 553 flush_icache_pte(mm, pteval); 554 555 set_pte(ptep, pteval); 556 } 557 558 #define PFN_PTE_SHIFT _PAGE_PFN_SHIFT 559 560 static inline void set_ptes(struct mm_struct *mm, unsigned long addr, 561 pte_t *ptep, pte_t pteval, unsigned int nr) 562 { 563 page_table_check_ptes_set(mm, ptep, pteval, nr); 564 565 for (;;) { 566 __set_pte_at(mm, ptep, pteval); 567 if (--nr == 0) 568 break; 569 ptep++; 570 pte_val(pteval) += 1 << _PAGE_PFN_SHIFT; 571 } 572 } 573 #define set_ptes set_ptes 574 575 static inline void pte_clear(struct mm_struct *mm, 576 unsigned long addr, pte_t *ptep) 577 { 578 __set_pte_at(mm, ptep, __pte(0)); 579 } 580 581 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS /* defined in mm/pgtable.c */ 582 extern int ptep_set_access_flags(struct vm_area_struct *vma, unsigned long address, 583 pte_t *ptep, pte_t entry, int dirty); 584 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG /* defined in mm/pgtable.c */ 585 extern int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long address, 586 pte_t *ptep); 587 588 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR 589 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, 590 unsigned long address, pte_t *ptep) 591 { 592 pte_t pte = __pte(atomic_long_xchg((atomic_long_t *)ptep, 0)); 593 594 page_table_check_pte_clear(mm, pte); 595 596 return pte; 597 } 598 599 #define __HAVE_ARCH_PTEP_SET_WRPROTECT 600 static inline void ptep_set_wrprotect(struct mm_struct *mm, 601 unsigned long address, pte_t *ptep) 602 { 603 atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep); 604 } 605 606 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH 607 static inline int ptep_clear_flush_young(struct vm_area_struct *vma, 608 unsigned long address, pte_t *ptep) 609 { 610 /* 611 * This comment is borrowed from x86, but applies equally to RISC-V: 612 * 613 * Clearing the accessed bit without a TLB flush 614 * doesn't cause data corruption. [ It could cause incorrect 615 * page aging and the (mistaken) reclaim of hot pages, but the 616 * chance of that should be relatively low. ] 617 * 618 * So as a performance optimization don't flush the TLB when 619 * clearing the accessed bit, it will eventually be flushed by 620 * a context switch or a VM operation anyway. [ In the rare 621 * event of it not getting flushed for a long time the delay 622 * shouldn't really matter because there's no real memory 623 * pressure for swapout to react to. ] 624 */ 625 return ptep_test_and_clear_young(vma, address, ptep); 626 } 627 628 #define pgprot_nx pgprot_nx 629 static inline pgprot_t pgprot_nx(pgprot_t _prot) 630 { 631 return __pgprot(pgprot_val(_prot) & ~_PAGE_EXEC); 632 } 633 634 #define pgprot_noncached pgprot_noncached 635 static inline pgprot_t pgprot_noncached(pgprot_t _prot) 636 { 637 unsigned long prot = pgprot_val(_prot); 638 639 prot &= ~_PAGE_MTMASK; 640 prot |= _PAGE_IO; 641 642 return __pgprot(prot); 643 } 644 645 #define pgprot_writecombine pgprot_writecombine 646 static inline pgprot_t pgprot_writecombine(pgprot_t _prot) 647 { 648 unsigned long prot = pgprot_val(_prot); 649 650 prot &= ~_PAGE_MTMASK; 651 prot |= _PAGE_NOCACHE; 652 653 return __pgprot(prot); 654 } 655 656 /* 657 * Both Svade and Svadu control the hardware behavior when the PTE A/D bits need to be set. By 658 * default the M-mode firmware enables the hardware updating scheme when only Svadu is present in 659 * DT. 660 */ 661 #define arch_has_hw_pte_young arch_has_hw_pte_young 662 static inline bool arch_has_hw_pte_young(void) 663 { 664 return riscv_has_extension_unlikely(RISCV_ISA_EXT_SVADU); 665 } 666 667 /* 668 * THP functions 669 */ 670 static inline pmd_t pte_pmd(pte_t pte) 671 { 672 return __pmd(pte_val(pte)); 673 } 674 675 static inline pud_t pte_pud(pte_t pte) 676 { 677 return __pud(pte_val(pte)); 678 } 679 680 static inline pmd_t pmd_mkhuge(pmd_t pmd) 681 { 682 return pmd; 683 } 684 685 static inline pmd_t pmd_mkinvalid(pmd_t pmd) 686 { 687 return __pmd(pmd_val(pmd) & ~(_PAGE_PRESENT|_PAGE_PROT_NONE)); 688 } 689 690 #define __pmd_to_phys(pmd) (__page_val_to_pfn(pmd_val(pmd)) << PAGE_SHIFT) 691 692 static inline unsigned long pmd_pfn(pmd_t pmd) 693 { 694 return ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT); 695 } 696 697 #define __pud_to_phys(pud) (__page_val_to_pfn(pud_val(pud)) << PAGE_SHIFT) 698 699 #define pud_pfn pud_pfn 700 static inline unsigned long pud_pfn(pud_t pud) 701 { 702 return ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT); 703 } 704 705 #define pmd_pgprot pmd_pgprot 706 static inline pgprot_t pmd_pgprot(pmd_t pmd) 707 { 708 return pte_pgprot(pmd_pte(pmd)); 709 } 710 711 #define pud_pgprot pud_pgprot 712 static inline pgprot_t pud_pgprot(pud_t pud) 713 { 714 return pte_pgprot(pud_pte(pud)); 715 } 716 717 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot) 718 { 719 return pte_pmd(pte_modify(pmd_pte(pmd), newprot)); 720 } 721 722 #define pmd_write pmd_write 723 static inline int pmd_write(pmd_t pmd) 724 { 725 return pte_write(pmd_pte(pmd)); 726 } 727 728 #define pud_write pud_write 729 static inline int pud_write(pud_t pud) 730 { 731 return pte_write(pud_pte(pud)); 732 } 733 734 #define pmd_dirty pmd_dirty 735 static inline int pmd_dirty(pmd_t pmd) 736 { 737 return pte_dirty(pmd_pte(pmd)); 738 } 739 740 #define pmd_young pmd_young 741 static inline int pmd_young(pmd_t pmd) 742 { 743 return pte_young(pmd_pte(pmd)); 744 } 745 746 static inline int pmd_user(pmd_t pmd) 747 { 748 return pte_user(pmd_pte(pmd)); 749 } 750 751 static inline pmd_t pmd_mkold(pmd_t pmd) 752 { 753 return pte_pmd(pte_mkold(pmd_pte(pmd))); 754 } 755 756 static inline pmd_t pmd_mkyoung(pmd_t pmd) 757 { 758 return pte_pmd(pte_mkyoung(pmd_pte(pmd))); 759 } 760 761 static inline pmd_t pmd_mkwrite_novma(pmd_t pmd) 762 { 763 return pte_pmd(pte_mkwrite_novma(pmd_pte(pmd))); 764 } 765 766 static inline pmd_t pmd_wrprotect(pmd_t pmd) 767 { 768 return pte_pmd(pte_wrprotect(pmd_pte(pmd))); 769 } 770 771 static inline pmd_t pmd_mkclean(pmd_t pmd) 772 { 773 return pte_pmd(pte_mkclean(pmd_pte(pmd))); 774 } 775 776 static inline pmd_t pmd_mkdirty(pmd_t pmd) 777 { 778 return pte_pmd(pte_mkdirty(pmd_pte(pmd))); 779 } 780 781 #ifdef CONFIG_ARCH_SUPPORTS_PMD_PFNMAP 782 static inline bool pmd_special(pmd_t pmd) 783 { 784 return pte_special(pmd_pte(pmd)); 785 } 786 787 static inline pmd_t pmd_mkspecial(pmd_t pmd) 788 { 789 return pte_pmd(pte_mkspecial(pmd_pte(pmd))); 790 } 791 #endif 792 793 #ifdef CONFIG_ARCH_SUPPORTS_PUD_PFNMAP 794 static inline bool pud_special(pud_t pud) 795 { 796 return pte_special(pud_pte(pud)); 797 } 798 799 static inline pud_t pud_mkspecial(pud_t pud) 800 { 801 return pte_pud(pte_mkspecial(pud_pte(pud))); 802 } 803 #endif 804 805 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr, 806 pmd_t *pmdp, pmd_t pmd) 807 { 808 page_table_check_pmd_set(mm, pmdp, pmd); 809 return __set_pte_at(mm, (pte_t *)pmdp, pmd_pte(pmd)); 810 } 811 812 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr, 813 pud_t *pudp, pud_t pud) 814 { 815 page_table_check_pud_set(mm, pudp, pud); 816 return __set_pte_at(mm, (pte_t *)pudp, pud_pte(pud)); 817 } 818 819 #ifdef CONFIG_PAGE_TABLE_CHECK 820 static inline bool pte_user_accessible_page(pte_t pte) 821 { 822 return pte_present(pte) && pte_user(pte); 823 } 824 825 static inline bool pmd_user_accessible_page(pmd_t pmd) 826 { 827 return pmd_leaf(pmd) && pmd_user(pmd); 828 } 829 830 static inline bool pud_user_accessible_page(pud_t pud) 831 { 832 return pud_leaf(pud) && pud_user(pud); 833 } 834 #endif 835 836 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 837 static inline int pmd_trans_huge(pmd_t pmd) 838 { 839 return pmd_leaf(pmd); 840 } 841 842 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS 843 static inline int pmdp_set_access_flags(struct vm_area_struct *vma, 844 unsigned long address, pmd_t *pmdp, 845 pmd_t entry, int dirty) 846 { 847 return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty); 848 } 849 850 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG 851 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma, 852 unsigned long address, pmd_t *pmdp) 853 { 854 return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp); 855 } 856 857 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR 858 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, 859 unsigned long address, pmd_t *pmdp) 860 { 861 pmd_t pmd = __pmd(atomic_long_xchg((atomic_long_t *)pmdp, 0)); 862 863 page_table_check_pmd_clear(mm, pmd); 864 865 return pmd; 866 } 867 868 #define __HAVE_ARCH_PMDP_SET_WRPROTECT 869 static inline void pmdp_set_wrprotect(struct mm_struct *mm, 870 unsigned long address, pmd_t *pmdp) 871 { 872 ptep_set_wrprotect(mm, address, (pte_t *)pmdp); 873 } 874 875 #define pmdp_establish pmdp_establish 876 static inline pmd_t pmdp_establish(struct vm_area_struct *vma, 877 unsigned long address, pmd_t *pmdp, pmd_t pmd) 878 { 879 page_table_check_pmd_set(vma->vm_mm, pmdp, pmd); 880 return __pmd(atomic_long_xchg((atomic_long_t *)pmdp, pmd_val(pmd))); 881 } 882 883 #define pmdp_collapse_flush pmdp_collapse_flush 884 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma, 885 unsigned long address, pmd_t *pmdp); 886 887 static inline pud_t pud_wrprotect(pud_t pud) 888 { 889 return pte_pud(pte_wrprotect(pud_pte(pud))); 890 } 891 892 static inline int pud_trans_huge(pud_t pud) 893 { 894 return pud_leaf(pud); 895 } 896 897 static inline int pud_dirty(pud_t pud) 898 { 899 return pte_dirty(pud_pte(pud)); 900 } 901 902 static inline pud_t pud_mkyoung(pud_t pud) 903 { 904 return pte_pud(pte_mkyoung(pud_pte(pud))); 905 } 906 907 static inline pud_t pud_mkold(pud_t pud) 908 { 909 return pte_pud(pte_mkold(pud_pte(pud))); 910 } 911 912 static inline pud_t pud_mkdirty(pud_t pud) 913 { 914 return pte_pud(pte_mkdirty(pud_pte(pud))); 915 } 916 917 static inline pud_t pud_mkclean(pud_t pud) 918 { 919 return pte_pud(pte_mkclean(pud_pte(pud))); 920 } 921 922 static inline pud_t pud_mkwrite(pud_t pud) 923 { 924 return pte_pud(pte_mkwrite_novma(pud_pte(pud))); 925 } 926 927 static inline pud_t pud_mkhuge(pud_t pud) 928 { 929 return pud; 930 } 931 932 static inline int pudp_set_access_flags(struct vm_area_struct *vma, 933 unsigned long address, pud_t *pudp, 934 pud_t entry, int dirty) 935 { 936 return ptep_set_access_flags(vma, address, (pte_t *)pudp, pud_pte(entry), dirty); 937 } 938 939 static inline int pudp_test_and_clear_young(struct vm_area_struct *vma, 940 unsigned long address, pud_t *pudp) 941 { 942 return ptep_test_and_clear_young(vma, address, (pte_t *)pudp); 943 } 944 945 static inline int pud_young(pud_t pud) 946 { 947 return pte_young(pud_pte(pud)); 948 } 949 950 static inline void update_mmu_cache_pud(struct vm_area_struct *vma, 951 unsigned long address, pud_t *pudp) 952 { 953 pte_t *ptep = (pte_t *)pudp; 954 955 update_mmu_cache(vma, address, ptep); 956 } 957 958 static inline pud_t pudp_establish(struct vm_area_struct *vma, 959 unsigned long address, pud_t *pudp, pud_t pud) 960 { 961 page_table_check_pud_set(vma->vm_mm, pudp, pud); 962 return __pud(atomic_long_xchg((atomic_long_t *)pudp, pud_val(pud))); 963 } 964 965 static inline pud_t pud_mkinvalid(pud_t pud) 966 { 967 return __pud(pud_val(pud) & ~(_PAGE_PRESENT | _PAGE_PROT_NONE)); 968 } 969 970 extern pud_t pudp_invalidate(struct vm_area_struct *vma, unsigned long address, 971 pud_t *pudp); 972 973 static inline pud_t pud_modify(pud_t pud, pgprot_t newprot) 974 { 975 return pte_pud(pte_modify(pud_pte(pud), newprot)); 976 } 977 978 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 979 980 /* 981 * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that 982 * are !pte_none() && !pte_present(). 983 * 984 * Format of swap PTE: 985 * bit 0: _PAGE_PRESENT (zero) 986 * bit 1 to 3: _PAGE_LEAF (zero) 987 * bit 5: _PAGE_PROT_NONE (zero) 988 * bit 6: exclusive marker 989 * bits 7 to 11: swap type 990 * bits 12 to XLEN-1: swap offset 991 */ 992 #define __SWP_TYPE_SHIFT 7 993 #define __SWP_TYPE_BITS 5 994 #define __SWP_TYPE_MASK ((1UL << __SWP_TYPE_BITS) - 1) 995 #define __SWP_OFFSET_SHIFT (__SWP_TYPE_BITS + __SWP_TYPE_SHIFT) 996 997 #define MAX_SWAPFILES_CHECK() \ 998 BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS) 999 1000 #define __swp_type(x) (((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK) 1001 #define __swp_offset(x) ((x).val >> __SWP_OFFSET_SHIFT) 1002 #define __swp_entry(type, offset) ((swp_entry_t) \ 1003 { (((type) & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT) | \ 1004 ((offset) << __SWP_OFFSET_SHIFT) }) 1005 1006 #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) }) 1007 #define __swp_entry_to_pte(x) ((pte_t) { (x).val }) 1008 1009 static inline bool pte_swp_exclusive(pte_t pte) 1010 { 1011 return pte_val(pte) & _PAGE_SWP_EXCLUSIVE; 1012 } 1013 1014 static inline pte_t pte_swp_mkexclusive(pte_t pte) 1015 { 1016 return __pte(pte_val(pte) | _PAGE_SWP_EXCLUSIVE); 1017 } 1018 1019 static inline pte_t pte_swp_clear_exclusive(pte_t pte) 1020 { 1021 return __pte(pte_val(pte) & ~_PAGE_SWP_EXCLUSIVE); 1022 } 1023 1024 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1025 #define __pmd_to_swp_entry(pmd) ((swp_entry_t) { pmd_val(pmd) }) 1026 #define __swp_entry_to_pmd(swp) __pmd((swp).val) 1027 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */ 1028 1029 /* 1030 * In the RV64 Linux scheme, we give the user half of the virtual-address space 1031 * and give the kernel the other (upper) half. 1032 */ 1033 #ifdef CONFIG_64BIT 1034 #define KERN_VIRT_START (-(BIT(VA_BITS)) + TASK_SIZE) 1035 #else 1036 #define KERN_VIRT_START FIXADDR_START 1037 #endif 1038 1039 /* 1040 * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32. 1041 * Note that PGDIR_SIZE must evenly divide TASK_SIZE. 1042 * Task size is: 1043 * - 0x9fc00000 (~2.5GB) for RV32. 1044 * - 0x4000000000 ( 256GB) for RV64 using SV39 mmu 1045 * - 0x800000000000 ( 128TB) for RV64 using SV48 mmu 1046 * - 0x100000000000000 ( 64PB) for RV64 using SV57 mmu 1047 * 1048 * Note that PGDIR_SIZE must evenly divide TASK_SIZE since "RISC-V 1049 * Instruction Set Manual Volume II: Privileged Architecture" states that 1050 * "load and store effective addresses, which are 64bits, must have bits 1051 * 63–48 all equal to bit 47, or else a page-fault exception will occur." 1052 * Similarly for SV57, bits 63–57 must be equal to bit 56. 1053 */ 1054 #ifdef CONFIG_64BIT 1055 #define TASK_SIZE_64 (PGDIR_SIZE * PTRS_PER_PGD / 2) 1056 1057 #ifdef CONFIG_COMPAT 1058 #define TASK_SIZE_32 (_AC(0x80000000, UL) - PAGE_SIZE) 1059 #define TASK_SIZE (is_compat_task() ? \ 1060 TASK_SIZE_32 : TASK_SIZE_64) 1061 #else 1062 #define TASK_SIZE TASK_SIZE_64 1063 #endif 1064 1065 #else 1066 #define TASK_SIZE FIXADDR_START 1067 #endif 1068 1069 #else /* CONFIG_MMU */ 1070 1071 #define PAGE_SHARED __pgprot(0) 1072 #define PAGE_KERNEL __pgprot(0) 1073 #define swapper_pg_dir NULL 1074 #define TASK_SIZE _AC(-1, UL) 1075 #define VMALLOC_START _AC(0, UL) 1076 #define VMALLOC_END TASK_SIZE 1077 1078 #endif /* !CONFIG_MMU */ 1079 1080 extern char _start[]; 1081 extern void *_dtb_early_va; 1082 extern uintptr_t _dtb_early_pa; 1083 #if defined(CONFIG_XIP_KERNEL) && defined(CONFIG_MMU) 1084 #define dtb_early_va (*(void **)XIP_FIXUP(&_dtb_early_va)) 1085 #define dtb_early_pa (*(uintptr_t *)XIP_FIXUP(&_dtb_early_pa)) 1086 #else 1087 #define dtb_early_va _dtb_early_va 1088 #define dtb_early_pa _dtb_early_pa 1089 #endif /* CONFIG_XIP_KERNEL */ 1090 extern u64 satp_mode; 1091 1092 void paging_init(void); 1093 void misc_mem_init(void); 1094 1095 /* 1096 * ZERO_PAGE is a global shared page that is always zero, 1097 * used for zero-mapped memory areas, etc. 1098 */ 1099 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]; 1100 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page)) 1101 1102 /* 1103 * Use set_p*_safe(), and elide TLB flushing, when confident that *no* 1104 * TLB flush will be required as a result of the "set". For example, use 1105 * in scenarios where it is known ahead of time that the routine is 1106 * setting non-present entries, or re-setting an existing entry to the 1107 * same value. Otherwise, use the typical "set" helpers and flush the 1108 * TLB. 1109 */ 1110 #define set_p4d_safe(p4dp, p4d) \ 1111 ({ \ 1112 WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \ 1113 set_p4d(p4dp, p4d); \ 1114 }) 1115 1116 #define set_pgd_safe(pgdp, pgd) \ 1117 ({ \ 1118 WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \ 1119 set_pgd(pgdp, pgd); \ 1120 }) 1121 #endif /* !__ASSEMBLY__ */ 1122 1123 #endif /* _ASM_RISCV_PGTABLE_H */ 1124