개념 설명 전체
arch/riscv/include/asm/pgtable.h · v6.18.37
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 __ASSEMBLER__
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 MMAP_VA_BITS_64 ((VA_BITS >= VA_BITS_SV48) ? VA_BITS_SV48 : VA_BITS)
128 #define MMAP_MIN_VA_BITS_64 (VA_BITS_SV39)
129 #define MMAP_VA_BITS (is_compat_task() ? VA_BITS_SV32 : MMAP_VA_BITS_64)
130 #define MMAP_MIN_VA_BITS (is_compat_task() ? VA_BITS_SV32 : MMAP_MIN_VA_BITS_64)
131 #else
132 #include <asm/pgtable-32.h>
133 #endif /* CONFIG_64BIT */
134
135 #include <linux/page_table_check.h>
136
137 #ifdef CONFIG_XIP_KERNEL
138 #define XIP_FIXUP(addr) ({ \
139 extern char _sdata[], _start[], _end[]; \
140 uintptr_t __rom_start_data = CONFIG_XIP_PHYS_ADDR \
141 + (uintptr_t)&_sdata - (uintptr_t)&_start; \
142 uintptr_t __rom_end_data = CONFIG_XIP_PHYS_ADDR \
143 + (uintptr_t)&_end - (uintptr_t)&_start; \
144 uintptr_t __a = (uintptr_t)(addr); \
145 (__a >= __rom_start_data && __a < __rom_end_data) ? \
146 __a - __rom_start_data + CONFIG_PHYS_RAM_BASE : __a; \
147 })
148 #else
149 #define XIP_FIXUP(addr) (addr)
150 #endif /* CONFIG_XIP_KERNEL */
151
152 struct pt_alloc_ops {
153 pte_t *(*get_pte_virt)(phys_addr_t pa);
154 phys_addr_t (*alloc_pte)(uintptr_t va);
155 #ifndef __PAGETABLE_PMD_FOLDED
156 pmd_t *(*get_pmd_virt)(phys_addr_t pa);
157 phys_addr_t (*alloc_pmd)(uintptr_t va);
158 pud_t *(*get_pud_virt)(phys_addr_t pa);
159 phys_addr_t (*alloc_pud)(uintptr_t va);
160 p4d_t *(*get_p4d_virt)(phys_addr_t pa);
161 phys_addr_t (*alloc_p4d)(uintptr_t va);
162 #endif
163 };
164
165 extern struct pt_alloc_ops pt_ops __meminitdata;
166
167 #ifdef CONFIG_MMU
168 /* Number of PGD entries that a user-mode program can use */
169 #define USER_PTRS_PER_PGD (TASK_SIZE / PGDIR_SIZE)
170
171 /* Page protection bits */
172 #define _PAGE_BASE (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
173
174 #define PAGE_NONE __pgprot(_PAGE_PROT_NONE | _PAGE_READ)
175 #define PAGE_READ __pgprot(_PAGE_BASE | _PAGE_READ)
176 #define PAGE_WRITE __pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE)
177 #define PAGE_EXEC __pgprot(_PAGE_BASE | _PAGE_EXEC)
178 #define PAGE_READ_EXEC __pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
179 #define PAGE_WRITE_EXEC __pgprot(_PAGE_BASE | _PAGE_READ | \
180 _PAGE_EXEC | _PAGE_WRITE)
181
182 #define PAGE_COPY PAGE_READ
183 #define PAGE_COPY_EXEC PAGE_READ_EXEC
184 #define PAGE_SHARED PAGE_WRITE
185 #define PAGE_SHARED_EXEC PAGE_WRITE_EXEC
186
187 #define _PAGE_KERNEL (_PAGE_READ \
188 | _PAGE_WRITE \
189 | _PAGE_PRESENT \
190 | _PAGE_ACCESSED \
191 | _PAGE_DIRTY \
192 | _PAGE_GLOBAL)
193
194 #define PAGE_KERNEL __pgprot(_PAGE_KERNEL)
195 #define PAGE_KERNEL_READ __pgprot(_PAGE_KERNEL & ~_PAGE_WRITE)
196 #define PAGE_KERNEL_EXEC __pgprot(_PAGE_KERNEL | _PAGE_EXEC)
197 #define PAGE_KERNEL_READ_EXEC __pgprot((_PAGE_KERNEL & ~_PAGE_WRITE) \
198 | _PAGE_EXEC)
199
200 #define PAGE_TABLE __pgprot(_PAGE_TABLE)
201
202 #define _PAGE_KERNEL_NC ((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_NOCACHE)
203 #define _PAGE_IOREMAP ((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO)
204 #define PAGE_KERNEL_IO __pgprot(_PAGE_IOREMAP)
205
206 extern pgd_t swapper_pg_dir[];
207 extern pgd_t trampoline_pg_dir[];
208 extern pgd_t early_pg_dir[];
209
210 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
211 static inline int pmd_present(pmd_t pmd)
212 {
213 /*
214 * Checking for _PAGE_LEAF is needed too because:
215 * When splitting a THP, split_huge_page() will temporarily clear
216 * the present bit, in this situation, pmd_present() and
217 * pmd_trans_huge() still needs to return true.
218 */
219 return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE | _PAGE_LEAF));
220 }
221 #else
222 static inline int pmd_present(pmd_t pmd)
223 {
224 return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
225 }
226 #endif
227
228 static inline int pmd_none(pmd_t pmd)
229 {
230 return (pmd_val(pmd) == 0);
231 }
232
233 static inline int pmd_bad(pmd_t pmd)
234 {
235 return !pmd_present(pmd) || (pmd_val(pmd) & _PAGE_LEAF);
236 }
237
238 #define pmd_leaf pmd_leaf
239 static inline bool pmd_leaf(pmd_t pmd)
240 {
241 return pmd_present(pmd) && (pmd_val(pmd) & _PAGE_LEAF);
242 }
243
244 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
245 {
246 WRITE_ONCE(*pmdp, pmd);
247 }
248
249 static inline void pmd_clear(pmd_t *pmdp)
250 {
251 set_pmd(pmdp, __pmd(0));
252 }
253
254 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
255 {
256 unsigned long prot_val = pgprot_val(prot);
257
258 ALT_THEAD_PMA(prot_val);
259
260 return __pgd((pfn << _PAGE_PFN_SHIFT) | prot_val);
261 }
262
263 static inline unsigned long _pgd_pfn(pgd_t pgd)
264 {
265 return __page_val_to_pfn(pgd_val(pgd));
266 }
267
268 static inline struct page *pmd_page(pmd_t pmd)
269 {
270 return pfn_to_page(__page_val_to_pfn(pmd_val(pmd)));
271 }
272
273 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
274 {
275 return (unsigned long)pfn_to_virt(__page_val_to_pfn(pmd_val(pmd)));
276 }
277
278 static inline pte_t pmd_pte(pmd_t pmd)
279 {
280 return __pte(pmd_val(pmd));
281 }
282
283 static inline pte_t pud_pte(pud_t pud)
284 {
285 return __pte(pud_val(pud));
286 }
287
288 #ifdef CONFIG_RISCV_ISA_SVNAPOT
289
290 static __always_inline bool has_svnapot(void)
291 {
292 return riscv_has_extension_likely(RISCV_ISA_EXT_SVNAPOT);
293 }
294
295 static inline unsigned long pte_napot(pte_t pte)
296 {
297 return pte_val(pte) & _PAGE_NAPOT;
298 }
299
300 static inline pte_t pte_mknapot(pte_t pte, unsigned int order)
301 {
302 int pos = order - 1 + _PAGE_PFN_SHIFT;
303 unsigned long napot_bit = BIT(pos);
304 unsigned long napot_mask = ~GENMASK(pos, _PAGE_PFN_SHIFT);
305
306 return __pte((pte_val(pte) & napot_mask) | napot_bit | _PAGE_NAPOT);
307 }
308
309 #else
310
311 static __always_inline bool has_svnapot(void) { return false; }
312
313 static inline unsigned long pte_napot(pte_t pte)
314 {
315 return 0;
316 }
317
318 #endif /* CONFIG_RISCV_ISA_SVNAPOT */
319
320 /* Yields the page frame number (PFN) of a page table entry */
321 static inline unsigned long pte_pfn(pte_t pte)
322 {
323 unsigned long res = __page_val_to_pfn(pte_val(pte));
324
325 if (has_svnapot() && pte_napot(pte))
326 res = res & (res - 1UL);
327
328 return res;
329 }
330
331 #define pte_page(x) pfn_to_page(pte_pfn(x))
332
333 /* Constructs a page table entry */
334 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
335 {
336 unsigned long prot_val = pgprot_val(prot);
337
338 ALT_THEAD_PMA(prot_val);
339
340 return __pte((pfn << _PAGE_PFN_SHIFT) | prot_val);
341 }
342
343 #define pte_pgprot pte_pgprot
344 static inline pgprot_t pte_pgprot(pte_t pte)
345 {
346 unsigned long pfn = pte_pfn(pte);
347
348 return __pgprot(pte_val(pfn_pte(pfn, __pgprot(0))) ^ pte_val(pte));
349 }
350
351 static inline int pte_present(pte_t pte)
352 {
353 return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
354 }
355
356 #define pte_accessible pte_accessible
357 static inline unsigned long pte_accessible(struct mm_struct *mm, pte_t a)
358 {
359 if (pte_val(a) & _PAGE_PRESENT)
360 return true;
361
362 if ((pte_val(a) & _PAGE_PROT_NONE) &&
363 atomic_read(&mm->tlb_flush_pending))
364 return true;
365
366 return false;
367 }
368
369 static inline int pte_none(pte_t pte)
370 {
371 return (pte_val(pte) == 0);
372 }
373
374 static inline int pte_write(pte_t pte)
375 {
376 return pte_val(pte) & _PAGE_WRITE;
377 }
378
379 static inline int pte_exec(pte_t pte)
380 {
381 return pte_val(pte) & _PAGE_EXEC;
382 }
383
384 static inline int pte_user(pte_t pte)
385 {
386 return pte_val(pte) & _PAGE_USER;
387 }
388
389 static inline int pte_huge(pte_t pte)
390 {
391 return pte_present(pte) && (pte_val(pte) & _PAGE_LEAF);
392 }
393
394 static inline int pte_dirty(pte_t pte)
395 {
396 return pte_val(pte) & _PAGE_DIRTY;
397 }
398
399 static inline int pte_young(pte_t pte)
400 {
401 return pte_val(pte) & _PAGE_ACCESSED;
402 }
403
404 static inline int pte_special(pte_t pte)
405 {
406 return pte_val(pte) & _PAGE_SPECIAL;
407 }
408
409 /* static inline pte_t pte_rdprotect(pte_t pte) */
410
411 static inline pte_t pte_wrprotect(pte_t pte)
412 {
413 return __pte(pte_val(pte) & ~(_PAGE_WRITE));
414 }
415
416 /* static inline pte_t pte_mkread(pte_t pte) */
417
418 static inline pte_t pte_mkwrite_novma(pte_t pte)
419 {
420 return __pte(pte_val(pte) | _PAGE_WRITE);
421 }
422
423 /* static inline pte_t pte_mkexec(pte_t pte) */
424
425 static inline pte_t pte_mkdirty(pte_t pte)
426 {
427 return __pte(pte_val(pte) | _PAGE_DIRTY);
428 }
429
430 static inline pte_t pte_mkclean(pte_t pte)
431 {
432 return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
433 }
434
435 static inline pte_t pte_mkyoung(pte_t pte)
436 {
437 return __pte(pte_val(pte) | _PAGE_ACCESSED);
438 }
439
440 static inline pte_t pte_mkold(pte_t pte)
441 {
442 return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
443 }
444
445 static inline pte_t pte_mkspecial(pte_t pte)
446 {
447 return __pte(pte_val(pte) | _PAGE_SPECIAL);
448 }
449
450 static inline pte_t pte_mkhuge(pte_t pte)
451 {
452 return pte;
453 }
454
455 #ifdef CONFIG_RISCV_ISA_SVNAPOT
456 #define pte_leaf_size(pte) (pte_napot(pte) ? \
457 napot_cont_size(napot_cont_order(pte)) :\
458 PAGE_SIZE)
459 #endif
460
461 #ifdef CONFIG_NUMA_BALANCING
462 /*
463 * See the comment in include/asm-generic/pgtable.h
464 */
465 static inline int pte_protnone(pte_t pte)
466 {
467 return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)) == _PAGE_PROT_NONE;
468 }
469
470 static inline int pmd_protnone(pmd_t pmd)
471 {
472 return pte_protnone(pmd_pte(pmd));
473 }
474 #endif
475
476 /* Modify page protection bits */
477 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
478 {
479 unsigned long newprot_val = pgprot_val(newprot);
480
481 ALT_THEAD_PMA(newprot_val);
482
483 return __pte((pte_val(pte) & _PAGE_CHG_MASK) | newprot_val);
484 }
485
486 #define pgd_ERROR(e) \
487 pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
488
489
490 /* Commit new configuration to MMU hardware */
491 static inline void update_mmu_cache_range(struct vm_fault *vmf,
492 struct vm_area_struct *vma, unsigned long address,
493 pte_t *ptep, unsigned int nr)
494 {
495 asm goto(ALTERNATIVE("nop", "j %l[svvptc]", 0, RISCV_ISA_EXT_SVVPTC, 1)
496 : : : : svvptc);
497
498 /*
499 * The kernel assumes that TLBs don't cache invalid entries, but
500 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
501 * cache flush; it is necessary even after writing invalid entries.
502 * Relying on flush_tlb_fix_spurious_fault would suffice, but
503 * the extra traps reduce performance. So, eagerly SFENCE.VMA.
504 */
505 while (nr--)
506 local_flush_tlb_page(address + nr * PAGE_SIZE);
507
508 svvptc:;
509 /*
510 * Svvptc guarantees that the new valid pte will be visible within
511 * a bounded timeframe, so when the uarch does not cache invalid
512 * entries, we don't have to do anything.
513 */
514 }
515 #define update_mmu_cache(vma, addr, ptep) \
516 update_mmu_cache_range(NULL, vma, addr, ptep, 1)
517
518 #define update_mmu_tlb_range(vma, addr, ptep, nr) \
519 update_mmu_cache_range(NULL, vma, addr, ptep, nr)
520
521 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
522 unsigned long address, pmd_t *pmdp)
523 {
524 pte_t *ptep = (pte_t *)pmdp;
525
526 update_mmu_cache(vma, address, ptep);
527 }
528
529 #define __HAVE_ARCH_PTE_SAME
530 static inline int pte_same(pte_t pte_a, pte_t pte_b)
531 {
532 return pte_val(pte_a) == pte_val(pte_b);
533 }
534
535 /*
536 * Certain architectures need to do special things when PTEs within
537 * a page table are directly modified. Thus, the following hook is
538 * made available.
539 */
540 static inline void set_pte(pte_t *ptep, pte_t pteval)
541 {
542 WRITE_ONCE(*ptep, pteval);
543 }
544
545 void flush_icache_pte(struct mm_struct *mm, pte_t pte);
546
547 static inline void __set_pte_at(struct mm_struct *mm, pte_t *ptep, pte_t pteval)
548 {
549 if (pte_present(pteval) && pte_exec(pteval))
550 flush_icache_pte(mm, pteval);
551
552 set_pte(ptep, pteval);
553 }
554
555 #define PFN_PTE_SHIFT _PAGE_PFN_SHIFT
556
557 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
558 pte_t *ptep, pte_t pteval, unsigned int nr)
559 {
560 page_table_check_ptes_set(mm, ptep, pteval, nr);
561
562 for (;;) {
563 __set_pte_at(mm, ptep, pteval);
564 if (--nr == 0)
565 break;
566 ptep++;
567 pte_val(pteval) += 1 << _PAGE_PFN_SHIFT;
568 }
569 }
570 #define set_ptes set_ptes
571
572 static inline void pte_clear(struct mm_struct *mm,
573 unsigned long addr, pte_t *ptep)
574 {
575 __set_pte_at(mm, ptep, __pte(0));
576 }
577
578 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS /* defined in mm/pgtable.c */
579 extern int ptep_set_access_flags(struct vm_area_struct *vma, unsigned long address,
580 pte_t *ptep, pte_t entry, int dirty);
581 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG /* defined in mm/pgtable.c */
582 extern int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long address,
583 pte_t *ptep);
584
585 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
586 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
587 unsigned long address, pte_t *ptep)
588 {
589 pte_t pte = __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
590
591 page_table_check_pte_clear(mm, pte);
592
593 return pte;
594 }
595
596 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
597 static inline void ptep_set_wrprotect(struct mm_struct *mm,
598 unsigned long address, pte_t *ptep)
599 {
600 atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
601 }
602
603 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
604 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
605 unsigned long address, pte_t *ptep)
606 {
607 /*
608 * This comment is borrowed from x86, but applies equally to RISC-V:
609 *
610 * Clearing the accessed bit without a TLB flush
611 * doesn't cause data corruption. [ It could cause incorrect
612 * page aging and the (mistaken) reclaim of hot pages, but the
613 * chance of that should be relatively low. ]
614 *
615 * So as a performance optimization don't flush the TLB when
616 * clearing the accessed bit, it will eventually be flushed by
617 * a context switch or a VM operation anyway. [ In the rare
618 * event of it not getting flushed for a long time the delay
619 * shouldn't really matter because there's no real memory
620 * pressure for swapout to react to. ]
621 */
622 return ptep_test_and_clear_young(vma, address, ptep);
623 }
624
625 #define pgprot_nx pgprot_nx
626 static inline pgprot_t pgprot_nx(pgprot_t _prot)
627 {
628 return __pgprot(pgprot_val(_prot) & ~_PAGE_EXEC);
629 }
630
631 #define pgprot_noncached pgprot_noncached
632 static inline pgprot_t pgprot_noncached(pgprot_t _prot)
633 {
634 unsigned long prot = pgprot_val(_prot);
635
636 prot &= ~_PAGE_MTMASK;
637 prot |= _PAGE_IO;
638
639 return __pgprot(prot);
640 }
641
642 #define pgprot_writecombine pgprot_writecombine
643 static inline pgprot_t pgprot_writecombine(pgprot_t _prot)
644 {
645 unsigned long prot = pgprot_val(_prot);
646
647 prot &= ~_PAGE_MTMASK;
648 prot |= _PAGE_NOCACHE;
649
650 return __pgprot(prot);
651 }
652
653 #define pgprot_dmacoherent pgprot_writecombine
654
655 /*
656 * Both Svade and Svadu control the hardware behavior when the PTE A/D bits need to be set. By
657 * default the M-mode firmware enables the hardware updating scheme when only Svadu is present in
658 * DT.
659 */
660 #define arch_has_hw_pte_young arch_has_hw_pte_young
661 static inline bool arch_has_hw_pte_young(void)
662 {
663 return riscv_has_extension_unlikely(RISCV_ISA_EXT_SVADU);
664 }
665
666 /*
667 * THP functions
668 */
669 static inline pmd_t pte_pmd(pte_t pte)
670 {
671 return __pmd(pte_val(pte));
672 }
673
674 static inline pud_t pte_pud(pte_t pte)
675 {
676 return __pud(pte_val(pte));
677 }
678
679 static inline pmd_t pmd_mkhuge(pmd_t pmd)
680 {
681 return pmd;
682 }
683
684 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
685 {
686 return __pmd(pmd_val(pmd) & ~(_PAGE_PRESENT|_PAGE_PROT_NONE));
687 }
688
689 #define __pmd_to_phys(pmd) (__page_val_to_pfn(pmd_val(pmd)) << PAGE_SHIFT)
690
691 static inline unsigned long pmd_pfn(pmd_t pmd)
692 {
693 return ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT);
694 }
695
696 #define __pud_to_phys(pud) (__page_val_to_pfn(pud_val(pud)) << PAGE_SHIFT)
697
698 #define pud_pfn pud_pfn
699 static inline unsigned long pud_pfn(pud_t pud)
700 {
701 return ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT);
702 }
703
704 #define pmd_pgprot pmd_pgprot
705 static inline pgprot_t pmd_pgprot(pmd_t pmd)
706 {
707 return pte_pgprot(pmd_pte(pmd));
708 }
709
710 #define pud_pgprot pud_pgprot
711 static inline pgprot_t pud_pgprot(pud_t pud)
712 {
713 return pte_pgprot(pud_pte(pud));
714 }
715
716 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
717 {
718 return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
719 }
720
721 #define pmd_write pmd_write
722 static inline int pmd_write(pmd_t pmd)
723 {
724 return pte_write(pmd_pte(pmd));
725 }
726
727 #define pud_write pud_write
728 static inline int pud_write(pud_t pud)
729 {
730 return pte_write(pud_pte(pud));
731 }
732
733 #define pmd_dirty pmd_dirty
734 static inline int pmd_dirty(pmd_t pmd)
735 {
736 return pte_dirty(pmd_pte(pmd));
737 }
738
739 #define pmd_young pmd_young
740 static inline int pmd_young(pmd_t pmd)
741 {
742 return pte_young(pmd_pte(pmd));
743 }
744
745 static inline int pmd_user(pmd_t pmd)
746 {
747 return pte_user(pmd_pte(pmd));
748 }
749
750 static inline pmd_t pmd_mkold(pmd_t pmd)
751 {
752 return pte_pmd(pte_mkold(pmd_pte(pmd)));
753 }
754
755 static inline pmd_t pmd_mkyoung(pmd_t pmd)
756 {
757 return pte_pmd(pte_mkyoung(pmd_pte(pmd)));
758 }
759
760 static inline pmd_t pmd_mkwrite_novma(pmd_t pmd)
761 {
762 return pte_pmd(pte_mkwrite_novma(pmd_pte(pmd)));
763 }
764
765 static inline pmd_t pmd_wrprotect(pmd_t pmd)
766 {
767 return pte_pmd(pte_wrprotect(pmd_pte(pmd)));
768 }
769
770 static inline pmd_t pmd_mkclean(pmd_t pmd)
771 {
772 return pte_pmd(pte_mkclean(pmd_pte(pmd)));
773 }
774
775 static inline pmd_t pmd_mkdirty(pmd_t pmd)
776 {
777 return pte_pmd(pte_mkdirty(pmd_pte(pmd)));
778 }
779
780 #ifdef CONFIG_ARCH_SUPPORTS_PMD_PFNMAP
781 static inline bool pmd_special(pmd_t pmd)
782 {
783 return pte_special(pmd_pte(pmd));
784 }
785
786 static inline pmd_t pmd_mkspecial(pmd_t pmd)
787 {
788 return pte_pmd(pte_mkspecial(pmd_pte(pmd)));
789 }
790 #endif
791
792 #ifdef CONFIG_ARCH_SUPPORTS_PUD_PFNMAP
793 static inline bool pud_special(pud_t pud)
794 {
795 return pte_special(pud_pte(pud));
796 }
797
798 static inline pud_t pud_mkspecial(pud_t pud)
799 {
800 return pte_pud(pte_mkspecial(pud_pte(pud)));
801 }
802 #endif
803
804 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
805 pmd_t *pmdp, pmd_t pmd)
806 {
807 page_table_check_pmd_set(mm, pmdp, pmd);
808 return __set_pte_at(mm, (pte_t *)pmdp, pmd_pte(pmd));
809 }
810
811 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
812 pud_t *pudp, pud_t pud)
813 {
814 page_table_check_pud_set(mm, pudp, pud);
815 return __set_pte_at(mm, (pte_t *)pudp, pud_pte(pud));
816 }
817
818 #ifdef CONFIG_PAGE_TABLE_CHECK
819 static inline bool pte_user_accessible_page(pte_t pte)
820 {
821 return pte_present(pte) && pte_user(pte);
822 }
823
824 static inline bool pmd_user_accessible_page(pmd_t pmd)
825 {
826 return pmd_leaf(pmd) && pmd_user(pmd);
827 }
828
829 static inline bool pud_user_accessible_page(pud_t pud)
830 {
831 return pud_leaf(pud) && pud_user(pud);
832 }
833 #endif
834
835 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
836 static inline int pmd_trans_huge(pmd_t pmd)
837 {
838 return pmd_leaf(pmd);
839 }
840
841 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
842 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
843 unsigned long address, pmd_t *pmdp,
844 pmd_t entry, int dirty)
845 {
846 return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
847 }
848
849 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
850 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
851 unsigned long address, pmd_t *pmdp)
852 {
853 return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
854 }
855
856 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
857 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
858 unsigned long address, pmd_t *pmdp)
859 {
860 pmd_t pmd = __pmd(atomic_long_xchg((atomic_long_t *)pmdp, 0));
861
862 page_table_check_pmd_clear(mm, pmd);
863
864 return pmd;
865 }
866
867 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
868 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
869 unsigned long address, pmd_t *pmdp)
870 {
871 ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
872 }
873
874 #define pmdp_establish pmdp_establish
875 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
876 unsigned long address, pmd_t *pmdp, pmd_t pmd)
877 {
878 page_table_check_pmd_set(vma->vm_mm, pmdp, pmd);
879 return __pmd(atomic_long_xchg((atomic_long_t *)pmdp, pmd_val(pmd)));
880 }
881
882 #define pmdp_collapse_flush pmdp_collapse_flush
883 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
884 unsigned long address, pmd_t *pmdp);
885
886 static inline pud_t pud_wrprotect(pud_t pud)
887 {
888 return pte_pud(pte_wrprotect(pud_pte(pud)));
889 }
890
891 static inline int pud_trans_huge(pud_t pud)
892 {
893 return pud_leaf(pud);
894 }
895
896 static inline int pud_dirty(pud_t pud)
897 {
898 return pte_dirty(pud_pte(pud));
899 }
900
901 static inline pud_t pud_mkyoung(pud_t pud)
902 {
903 return pte_pud(pte_mkyoung(pud_pte(pud)));
904 }
905
906 static inline pud_t pud_mkold(pud_t pud)
907 {
908 return pte_pud(pte_mkold(pud_pte(pud)));
909 }
910
911 static inline pud_t pud_mkdirty(pud_t pud)
912 {
913 return pte_pud(pte_mkdirty(pud_pte(pud)));
914 }
915
916 static inline pud_t pud_mkclean(pud_t pud)
917 {
918 return pte_pud(pte_mkclean(pud_pte(pud)));
919 }
920
921 static inline pud_t pud_mkwrite(pud_t pud)
922 {
923 return pte_pud(pte_mkwrite_novma(pud_pte(pud)));
924 }
925
926 static inline pud_t pud_mkhuge(pud_t pud)
927 {
928 return pud;
929 }
930
931 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
932 unsigned long address, pud_t *pudp,
933 pud_t entry, int dirty)
934 {
935 return ptep_set_access_flags(vma, address, (pte_t *)pudp, pud_pte(entry), dirty);
936 }
937
938 static inline int pudp_test_and_clear_young(struct vm_area_struct *vma,
939 unsigned long address, pud_t *pudp)
940 {
941 return ptep_test_and_clear_young(vma, address, (pte_t *)pudp);
942 }
943
944 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
945 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
946 unsigned long address, pud_t *pudp)
947 {
948 #ifdef CONFIG_SMP
949 pud_t pud = __pud(xchg(&pudp->pud, 0));
950 #else
951 pud_t pud = *pudp;
952
953 pud_clear(pudp);
954 #endif
955
956 page_table_check_pud_clear(mm, pud);
957
958 return pud;
959 }
960
961 static inline int pud_young(pud_t pud)
962 {
963 return pte_young(pud_pte(pud));
964 }
965
966 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
967 unsigned long address, pud_t *pudp)
968 {
969 pte_t *ptep = (pte_t *)pudp;
970
971 update_mmu_cache(vma, address, ptep);
972 }
973
974 static inline pud_t pudp_establish(struct vm_area_struct *vma,
975 unsigned long address, pud_t *pudp, pud_t pud)
976 {
977 page_table_check_pud_set(vma->vm_mm, pudp, pud);
978 return __pud(atomic_long_xchg((atomic_long_t *)pudp, pud_val(pud)));
979 }
980
981 static inline pud_t pud_mkinvalid(pud_t pud)
982 {
983 return __pud(pud_val(pud) & ~(_PAGE_PRESENT | _PAGE_PROT_NONE));
984 }
985
986 extern pud_t pudp_invalidate(struct vm_area_struct *vma, unsigned long address,
987 pud_t *pudp);
988
989 static inline pud_t pud_modify(pud_t pud, pgprot_t newprot)
990 {
991 return pte_pud(pte_modify(pud_pte(pud), newprot));
992 }
993
994 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
995
996 /*
997 * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
998 * are !pte_none() && !pte_present().
999 *
1000 * Format of swap PTE:
1001 * bit 0: _PAGE_PRESENT (zero)
1002 * bit 1 to 3: _PAGE_LEAF (zero)
1003 * bit 5: _PAGE_PROT_NONE (zero)
1004 * bit 6: exclusive marker
1005 * bits 7 to 11: swap type
1006 * bits 12 to XLEN-1: swap offset
1007 */
1008 #define __SWP_TYPE_SHIFT 7
1009 #define __SWP_TYPE_BITS 5
1010 #define __SWP_TYPE_MASK ((1UL << __SWP_TYPE_BITS) - 1)
1011 #define __SWP_OFFSET_SHIFT (__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
1012
1013 #define MAX_SWAPFILES_CHECK() \
1014 BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
1015
1016 #define __swp_type(x) (((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
1017 #define __swp_offset(x) ((x).val >> __SWP_OFFSET_SHIFT)
1018 #define __swp_entry(type, offset) ((swp_entry_t) \
1019 { (((type) & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT) | \
1020 ((offset) << __SWP_OFFSET_SHIFT) })
1021
1022 #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) })
1023 #define __swp_entry_to_pte(x) ((pte_t) { (x).val })
1024
1025 static inline bool pte_swp_exclusive(pte_t pte)
1026 {
1027 return pte_val(pte) & _PAGE_SWP_EXCLUSIVE;
1028 }
1029
1030 static inline pte_t pte_swp_mkexclusive(pte_t pte)
1031 {
1032 return __pte(pte_val(pte) | _PAGE_SWP_EXCLUSIVE);
1033 }
1034
1035 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
1036 {
1037 return __pte(pte_val(pte) & ~_PAGE_SWP_EXCLUSIVE);
1038 }
1039
1040 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1041 #define __pmd_to_swp_entry(pmd) ((swp_entry_t) { pmd_val(pmd) })
1042 #define __swp_entry_to_pmd(swp) __pmd((swp).val)
1043 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
1044
1045 /*
1046 * In the RV64 Linux scheme, we give the user half of the virtual-address space
1047 * and give the kernel the other (upper) half.
1048 */
1049 #ifdef CONFIG_64BIT
1050 #define KERN_VIRT_START (-(BIT(VA_BITS)) + TASK_SIZE)
1051 #else
1052 #define KERN_VIRT_START FIXADDR_START
1053 #endif
1054
1055 /*
1056 * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
1057 * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
1058 * Task size is:
1059 * - 0x9fc00000 (~2.5GB) for RV32.
1060 * - 0x4000000000 ( 256GB) for RV64 using SV39 mmu
1061 * - 0x800000000000 ( 128TB) for RV64 using SV48 mmu
1062 * - 0x100000000000000 ( 64PB) for RV64 using SV57 mmu
1063 *
1064 * Note that PGDIR_SIZE must evenly divide TASK_SIZE since "RISC-V
1065 * Instruction Set Manual Volume II: Privileged Architecture" states that
1066 * "load and store effective addresses, which are 64bits, must have bits
1067 * 63–48 all equal to bit 47, or else a page-fault exception will occur."
1068 * Similarly for SV57, bits 63–57 must be equal to bit 56.
1069 */
1070 #ifdef CONFIG_64BIT
1071 #define TASK_SIZE_64 (PGDIR_SIZE * PTRS_PER_PGD / 2)
1072
1073 #ifdef CONFIG_COMPAT
1074 #define TASK_SIZE_32 (_AC(0x80000000, UL) - PAGE_SIZE)
1075 #define TASK_SIZE (is_compat_task() ? \
1076 TASK_SIZE_32 : TASK_SIZE_64)
1077 #else
1078 #define TASK_SIZE TASK_SIZE_64
1079 #endif
1080
1081 #else
1082 #define TASK_SIZE FIXADDR_START
1083 #endif
1084
1085 #else /* CONFIG_MMU */
1086
1087 #define PAGE_SHARED __pgprot(0)
1088 #define PAGE_KERNEL __pgprot(0)
1089 #define swapper_pg_dir NULL
1090 #define TASK_SIZE _AC(-1, UL)
1091 #define VMALLOC_START _AC(0, UL)
1092 #define VMALLOC_END TASK_SIZE
1093
1094 #endif /* !CONFIG_MMU */
1095
1096 extern char _start[];
1097 extern void *_dtb_early_va;
1098 extern uintptr_t _dtb_early_pa;
1099 #if defined(CONFIG_XIP_KERNEL) && defined(CONFIG_MMU)
1100 #define dtb_early_va (*(void **)XIP_FIXUP(&_dtb_early_va))
1101 #define dtb_early_pa (*(uintptr_t *)XIP_FIXUP(&_dtb_early_pa))
1102 #else
1103 #define dtb_early_va _dtb_early_va
1104 #define dtb_early_pa _dtb_early_pa
1105 #endif /* CONFIG_XIP_KERNEL */
1106 extern u64 satp_mode;
1107
1108 void paging_init(void);
1109 void misc_mem_init(void);
1110
1111 /*
1112 * ZERO_PAGE is a global shared page that is always zero,
1113 * used for zero-mapped memory areas, etc.
1114 */
1115 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
1116 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
1117
1118 /*
1119 * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
1120 * TLB flush will be required as a result of the "set". For example, use
1121 * in scenarios where it is known ahead of time that the routine is
1122 * setting non-present entries, or re-setting an existing entry to the
1123 * same value. Otherwise, use the typical "set" helpers and flush the
1124 * TLB.
1125 */
1126 #define set_p4d_safe(p4dp, p4d) \
1127 ({ \
1128 WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
1129 set_p4d(p4dp, p4d); \
1130 })
1131
1132 #define set_pgd_safe(pgdp, pgd) \
1133 ({ \
1134 WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
1135 set_pgd(pgdp, pgd); \
1136 })
1137 #endif /* !__ASSEMBLER__ */
1138
1139 #endif /* _ASM_RISCV_PGTABLE_H */