xref: /linux/arch/riscv/include/asm/pgtable.h (revision eb01fe7abbe2d0b38824d2a93fdb4cc3eaf2ccc1)
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 #define KERNEL_LINK_ADDR	PAGE_OFFSET
16 #define KERN_VIRT_SIZE		(UL(-1))
17 #else
18 
19 #define ADDRESS_SPACE_END	(UL(-1))
20 
21 #ifdef CONFIG_64BIT
22 /* Leave 2GB for kernel and BPF at the end of the address space */
23 #define KERNEL_LINK_ADDR	(ADDRESS_SPACE_END - SZ_2G + 1)
24 #else
25 #define KERNEL_LINK_ADDR	PAGE_OFFSET
26 #endif
27 
28 /* Number of entries in the page global directory */
29 #define PTRS_PER_PGD    (PAGE_SIZE / sizeof(pgd_t))
30 /* Number of entries in the page table */
31 #define PTRS_PER_PTE    (PAGE_SIZE / sizeof(pte_t))
32 
33 /*
34  * Half of the kernel address space (1/4 of the entries of the page global
35  * directory) is for the direct mapping.
36  */
37 #define KERN_VIRT_SIZE          ((PTRS_PER_PGD / 2 * PGDIR_SIZE) / 2)
38 
39 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
40 #define VMALLOC_END      PAGE_OFFSET
41 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
42 
43 #define BPF_JIT_REGION_SIZE	(SZ_128M)
44 #ifdef CONFIG_64BIT
45 #define BPF_JIT_REGION_START	(BPF_JIT_REGION_END - BPF_JIT_REGION_SIZE)
46 #define BPF_JIT_REGION_END	(MODULES_END)
47 #else
48 #define BPF_JIT_REGION_START	(PAGE_OFFSET - BPF_JIT_REGION_SIZE)
49 #define BPF_JIT_REGION_END	(VMALLOC_END)
50 #endif
51 
52 /* Modules always live before the kernel */
53 #ifdef CONFIG_64BIT
54 /* This is used to define the end of the KASAN shadow region */
55 #define MODULES_LOWEST_VADDR	(KERNEL_LINK_ADDR - SZ_2G)
56 #define MODULES_VADDR		(PFN_ALIGN((unsigned long)&_end) - SZ_2G)
57 #define MODULES_END		(PFN_ALIGN((unsigned long)&_start))
58 #endif
59 
60 /*
61  * Roughly size the vmemmap space to be large enough to fit enough
62  * struct pages to map half the virtual address space. Then
63  * position vmemmap directly below the VMALLOC region.
64  */
65 #define VA_BITS_SV32 32
66 #ifdef CONFIG_64BIT
67 #define VA_BITS_SV39 39
68 #define VA_BITS_SV48 48
69 #define VA_BITS_SV57 57
70 
71 #define VA_BITS		(pgtable_l5_enabled ? \
72 				VA_BITS_SV57 : (pgtable_l4_enabled ? VA_BITS_SV48 : VA_BITS_SV39))
73 #else
74 #define VA_BITS		VA_BITS_SV32
75 #endif
76 
77 #define VMEMMAP_SHIFT \
78 	(VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
79 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
80 #define VMEMMAP_END	VMALLOC_START
81 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
82 
83 /*
84  * Define vmemmap for pfn_to_page & page_to_pfn calls. Needed if kernel
85  * is configured with CONFIG_SPARSEMEM_VMEMMAP enabled.
86  */
87 #define vmemmap		((struct page *)VMEMMAP_START - (phys_ram_base >> PAGE_SHIFT))
88 
89 #define PCI_IO_SIZE      SZ_16M
90 #define PCI_IO_END       VMEMMAP_START
91 #define PCI_IO_START     (PCI_IO_END - PCI_IO_SIZE)
92 
93 #define FIXADDR_TOP      PCI_IO_START
94 #ifdef CONFIG_64BIT
95 #define MAX_FDT_SIZE	 PMD_SIZE
96 #define FIX_FDT_SIZE	 (MAX_FDT_SIZE + SZ_2M)
97 #define FIXADDR_SIZE     (PMD_SIZE + FIX_FDT_SIZE)
98 #else
99 #define MAX_FDT_SIZE	 PGDIR_SIZE
100 #define FIX_FDT_SIZE	 MAX_FDT_SIZE
101 #define FIXADDR_SIZE     (PGDIR_SIZE + FIX_FDT_SIZE)
102 #endif
103 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
104 
105 #endif
106 
107 #ifdef CONFIG_XIP_KERNEL
108 #define XIP_OFFSET		SZ_32M
109 #define XIP_OFFSET_MASK		(SZ_32M - 1)
110 #else
111 #define XIP_OFFSET		0
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 
121 #define __page_val_to_pfn(_val)  (((_val) & _PAGE_PFN_MASK) >> _PAGE_PFN_SHIFT)
122 
123 #ifdef CONFIG_64BIT
124 #include <asm/pgtable-64.h>
125 
126 #define VA_USER_SV39 (UL(1) << (VA_BITS_SV39 - 1))
127 #define VA_USER_SV48 (UL(1) << (VA_BITS_SV48 - 1))
128 #define VA_USER_SV57 (UL(1) << (VA_BITS_SV57 - 1))
129 
130 #ifdef CONFIG_COMPAT
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 #define MMAP_VA_BITS ((VA_BITS >= VA_BITS_SV48) ? VA_BITS_SV48 : VA_BITS)
137 #define MMAP_MIN_VA_BITS (VA_BITS_SV39)
138 #endif /* CONFIG_COMPAT */
139 
140 #else
141 #include <asm/pgtable-32.h>
142 #endif /* CONFIG_64BIT */
143 
144 #include <linux/page_table_check.h>
145 
146 #ifdef CONFIG_XIP_KERNEL
147 #define XIP_FIXUP(addr) ({							\
148 	uintptr_t __a = (uintptr_t)(addr);					\
149 	(__a >= CONFIG_XIP_PHYS_ADDR && \
150 	 __a < CONFIG_XIP_PHYS_ADDR + XIP_OFFSET * 2) ?	\
151 		__a - CONFIG_XIP_PHYS_ADDR + CONFIG_PHYS_RAM_BASE - XIP_OFFSET :\
152 		__a;								\
153 	})
154 #else
155 #define XIP_FIXUP(addr)		(addr)
156 #endif /* CONFIG_XIP_KERNEL */
157 
158 struct pt_alloc_ops {
159 	pte_t *(*get_pte_virt)(phys_addr_t pa);
160 	phys_addr_t (*alloc_pte)(uintptr_t va);
161 #ifndef __PAGETABLE_PMD_FOLDED
162 	pmd_t *(*get_pmd_virt)(phys_addr_t pa);
163 	phys_addr_t (*alloc_pmd)(uintptr_t va);
164 	pud_t *(*get_pud_virt)(phys_addr_t pa);
165 	phys_addr_t (*alloc_pud)(uintptr_t va);
166 	p4d_t *(*get_p4d_virt)(phys_addr_t pa);
167 	phys_addr_t (*alloc_p4d)(uintptr_t va);
168 #endif
169 };
170 
171 extern struct pt_alloc_ops pt_ops __initdata;
172 
173 #ifdef CONFIG_MMU
174 /* Number of PGD entries that a user-mode program can use */
175 #define USER_PTRS_PER_PGD   (TASK_SIZE / PGDIR_SIZE)
176 
177 /* Page protection bits */
178 #define _PAGE_BASE	(_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
179 
180 #define PAGE_NONE		__pgprot(_PAGE_PROT_NONE | _PAGE_READ)
181 #define PAGE_READ		__pgprot(_PAGE_BASE | _PAGE_READ)
182 #define PAGE_WRITE		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE)
183 #define PAGE_EXEC		__pgprot(_PAGE_BASE | _PAGE_EXEC)
184 #define PAGE_READ_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
185 #define PAGE_WRITE_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ |	\
186 					 _PAGE_EXEC | _PAGE_WRITE)
187 
188 #define PAGE_COPY		PAGE_READ
189 #define PAGE_COPY_EXEC		PAGE_READ_EXEC
190 #define PAGE_SHARED		PAGE_WRITE
191 #define PAGE_SHARED_EXEC	PAGE_WRITE_EXEC
192 
193 #define _PAGE_KERNEL		(_PAGE_READ \
194 				| _PAGE_WRITE \
195 				| _PAGE_PRESENT \
196 				| _PAGE_ACCESSED \
197 				| _PAGE_DIRTY \
198 				| _PAGE_GLOBAL)
199 
200 #define PAGE_KERNEL		__pgprot(_PAGE_KERNEL)
201 #define PAGE_KERNEL_READ	__pgprot(_PAGE_KERNEL & ~_PAGE_WRITE)
202 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
203 #define PAGE_KERNEL_READ_EXEC	__pgprot((_PAGE_KERNEL & ~_PAGE_WRITE) \
204 					 | _PAGE_EXEC)
205 
206 #define PAGE_TABLE		__pgprot(_PAGE_TABLE)
207 
208 #define _PAGE_IOREMAP	((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO)
209 #define PAGE_KERNEL_IO		__pgprot(_PAGE_IOREMAP)
210 
211 extern pgd_t swapper_pg_dir[];
212 extern pgd_t trampoline_pg_dir[];
213 extern pgd_t early_pg_dir[];
214 
215 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
216 static inline int pmd_present(pmd_t pmd)
217 {
218 	/*
219 	 * Checking for _PAGE_LEAF is needed too because:
220 	 * When splitting a THP, split_huge_page() will temporarily clear
221 	 * the present bit, in this situation, pmd_present() and
222 	 * pmd_trans_huge() still needs to return true.
223 	 */
224 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE | _PAGE_LEAF));
225 }
226 #else
227 static inline int pmd_present(pmd_t pmd)
228 {
229 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
230 }
231 #endif
232 
233 static inline int pmd_none(pmd_t pmd)
234 {
235 	return (pmd_val(pmd) == 0);
236 }
237 
238 static inline int pmd_bad(pmd_t pmd)
239 {
240 	return !pmd_present(pmd) || (pmd_val(pmd) & _PAGE_LEAF);
241 }
242 
243 #define pmd_leaf	pmd_leaf
244 static inline bool pmd_leaf(pmd_t pmd)
245 {
246 	return pmd_present(pmd) && (pmd_val(pmd) & _PAGE_LEAF);
247 }
248 
249 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
250 {
251 	WRITE_ONCE(*pmdp, pmd);
252 }
253 
254 static inline void pmd_clear(pmd_t *pmdp)
255 {
256 	set_pmd(pmdp, __pmd(0));
257 }
258 
259 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
260 {
261 	unsigned long prot_val = pgprot_val(prot);
262 
263 	ALT_THEAD_PMA(prot_val);
264 
265 	return __pgd((pfn << _PAGE_PFN_SHIFT) | prot_val);
266 }
267 
268 static inline unsigned long _pgd_pfn(pgd_t pgd)
269 {
270 	return __page_val_to_pfn(pgd_val(pgd));
271 }
272 
273 static inline struct page *pmd_page(pmd_t pmd)
274 {
275 	return pfn_to_page(__page_val_to_pfn(pmd_val(pmd)));
276 }
277 
278 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
279 {
280 	return (unsigned long)pfn_to_virt(__page_val_to_pfn(pmd_val(pmd)));
281 }
282 
283 static inline pte_t pmd_pte(pmd_t pmd)
284 {
285 	return __pte(pmd_val(pmd));
286 }
287 
288 static inline pte_t pud_pte(pud_t pud)
289 {
290 	return __pte(pud_val(pud));
291 }
292 
293 #ifdef CONFIG_RISCV_ISA_SVNAPOT
294 #include <asm/cpufeature.h>
295 
296 static __always_inline bool has_svnapot(void)
297 {
298 	return riscv_has_extension_likely(RISCV_ISA_EXT_SVNAPOT);
299 }
300 
301 static inline unsigned long pte_napot(pte_t pte)
302 {
303 	return pte_val(pte) & _PAGE_NAPOT;
304 }
305 
306 static inline pte_t pte_mknapot(pte_t pte, unsigned int order)
307 {
308 	int pos = order - 1 + _PAGE_PFN_SHIFT;
309 	unsigned long napot_bit = BIT(pos);
310 	unsigned long napot_mask = ~GENMASK(pos, _PAGE_PFN_SHIFT);
311 
312 	return __pte((pte_val(pte) & napot_mask) | napot_bit | _PAGE_NAPOT);
313 }
314 
315 #else
316 
317 static __always_inline bool has_svnapot(void) { return false; }
318 
319 static inline unsigned long pte_napot(pte_t pte)
320 {
321 	return 0;
322 }
323 
324 #endif /* CONFIG_RISCV_ISA_SVNAPOT */
325 
326 /* Yields the page frame number (PFN) of a page table entry */
327 static inline unsigned long pte_pfn(pte_t pte)
328 {
329 	unsigned long res  = __page_val_to_pfn(pte_val(pte));
330 
331 	if (has_svnapot() && pte_napot(pte))
332 		res = res & (res - 1UL);
333 
334 	return res;
335 }
336 
337 #define pte_page(x)     pfn_to_page(pte_pfn(x))
338 
339 /* Constructs a page table entry */
340 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
341 {
342 	unsigned long prot_val = pgprot_val(prot);
343 
344 	ALT_THEAD_PMA(prot_val);
345 
346 	return __pte((pfn << _PAGE_PFN_SHIFT) | prot_val);
347 }
348 
349 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
350 
351 static inline int pte_present(pte_t pte)
352 {
353 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
354 }
355 
356 static inline int pte_none(pte_t pte)
357 {
358 	return (pte_val(pte) == 0);
359 }
360 
361 static inline int pte_write(pte_t pte)
362 {
363 	return pte_val(pte) & _PAGE_WRITE;
364 }
365 
366 static inline int pte_exec(pte_t pte)
367 {
368 	return pte_val(pte) & _PAGE_EXEC;
369 }
370 
371 static inline int pte_user(pte_t pte)
372 {
373 	return pte_val(pte) & _PAGE_USER;
374 }
375 
376 static inline int pte_huge(pte_t pte)
377 {
378 	return pte_present(pte) && (pte_val(pte) & _PAGE_LEAF);
379 }
380 
381 static inline int pte_dirty(pte_t pte)
382 {
383 	return pte_val(pte) & _PAGE_DIRTY;
384 }
385 
386 static inline int pte_young(pte_t pte)
387 {
388 	return pte_val(pte) & _PAGE_ACCESSED;
389 }
390 
391 static inline int pte_special(pte_t pte)
392 {
393 	return pte_val(pte) & _PAGE_SPECIAL;
394 }
395 
396 /* static inline pte_t pte_rdprotect(pte_t pte) */
397 
398 static inline pte_t pte_wrprotect(pte_t pte)
399 {
400 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
401 }
402 
403 /* static inline pte_t pte_mkread(pte_t pte) */
404 
405 static inline pte_t pte_mkwrite_novma(pte_t pte)
406 {
407 	return __pte(pte_val(pte) | _PAGE_WRITE);
408 }
409 
410 /* static inline pte_t pte_mkexec(pte_t pte) */
411 
412 static inline pte_t pte_mkdirty(pte_t pte)
413 {
414 	return __pte(pte_val(pte) | _PAGE_DIRTY);
415 }
416 
417 static inline pte_t pte_mkclean(pte_t pte)
418 {
419 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
420 }
421 
422 static inline pte_t pte_mkyoung(pte_t pte)
423 {
424 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
425 }
426 
427 static inline pte_t pte_mkold(pte_t pte)
428 {
429 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
430 }
431 
432 static inline pte_t pte_mkspecial(pte_t pte)
433 {
434 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
435 }
436 
437 static inline pte_t pte_mkhuge(pte_t pte)
438 {
439 	return pte;
440 }
441 
442 #define pte_leaf_size(pte)	(pte_napot(pte) ?				\
443 					napot_cont_size(napot_cont_order(pte)) :\
444 					PAGE_SIZE)
445 
446 #ifdef CONFIG_NUMA_BALANCING
447 /*
448  * See the comment in include/asm-generic/pgtable.h
449  */
450 static inline int pte_protnone(pte_t pte)
451 {
452 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)) == _PAGE_PROT_NONE;
453 }
454 
455 static inline int pmd_protnone(pmd_t pmd)
456 {
457 	return pte_protnone(pmd_pte(pmd));
458 }
459 #endif
460 
461 /* Modify page protection bits */
462 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
463 {
464 	unsigned long newprot_val = pgprot_val(newprot);
465 
466 	ALT_THEAD_PMA(newprot_val);
467 
468 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | newprot_val);
469 }
470 
471 #define pgd_ERROR(e) \
472 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
473 
474 
475 /* Commit new configuration to MMU hardware */
476 static inline void update_mmu_cache_range(struct vm_fault *vmf,
477 		struct vm_area_struct *vma, unsigned long address,
478 		pte_t *ptep, unsigned int nr)
479 {
480 	/*
481 	 * The kernel assumes that TLBs don't cache invalid entries, but
482 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
483 	 * cache flush; it is necessary even after writing invalid entries.
484 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
485 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
486 	 */
487 	while (nr--)
488 		local_flush_tlb_page(address + nr * PAGE_SIZE);
489 }
490 #define update_mmu_cache(vma, addr, ptep) \
491 	update_mmu_cache_range(NULL, vma, addr, ptep, 1)
492 
493 #define __HAVE_ARCH_UPDATE_MMU_TLB
494 #define update_mmu_tlb update_mmu_cache
495 
496 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
497 		unsigned long address, pmd_t *pmdp)
498 {
499 	pte_t *ptep = (pte_t *)pmdp;
500 
501 	update_mmu_cache(vma, address, ptep);
502 }
503 
504 #define __HAVE_ARCH_PTE_SAME
505 static inline int pte_same(pte_t pte_a, pte_t pte_b)
506 {
507 	return pte_val(pte_a) == pte_val(pte_b);
508 }
509 
510 /*
511  * Certain architectures need to do special things when PTEs within
512  * a page table are directly modified.  Thus, the following hook is
513  * made available.
514  */
515 static inline void set_pte(pte_t *ptep, pte_t pteval)
516 {
517 	WRITE_ONCE(*ptep, pteval);
518 }
519 
520 void flush_icache_pte(pte_t pte);
521 
522 static inline void __set_pte_at(pte_t *ptep, pte_t pteval)
523 {
524 	if (pte_present(pteval) && pte_exec(pteval))
525 		flush_icache_pte(pteval);
526 
527 	set_pte(ptep, pteval);
528 }
529 
530 #define PFN_PTE_SHIFT		_PAGE_PFN_SHIFT
531 
532 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
533 		pte_t *ptep, pte_t pteval, unsigned int nr)
534 {
535 	page_table_check_ptes_set(mm, ptep, pteval, nr);
536 
537 	for (;;) {
538 		__set_pte_at(ptep, pteval);
539 		if (--nr == 0)
540 			break;
541 		ptep++;
542 		pte_val(pteval) += 1 << _PAGE_PFN_SHIFT;
543 	}
544 }
545 #define set_ptes set_ptes
546 
547 static inline void pte_clear(struct mm_struct *mm,
548 	unsigned long addr, pte_t *ptep)
549 {
550 	__set_pte_at(ptep, __pte(0));
551 }
552 
553 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS	/* defined in mm/pgtable.c */
554 extern int ptep_set_access_flags(struct vm_area_struct *vma, unsigned long address,
555 				 pte_t *ptep, pte_t entry, int dirty);
556 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG	/* defined in mm/pgtable.c */
557 extern int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long address,
558 				     pte_t *ptep);
559 
560 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
561 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
562 				       unsigned long address, pte_t *ptep)
563 {
564 	pte_t pte = __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
565 
566 	page_table_check_pte_clear(mm, pte);
567 
568 	return pte;
569 }
570 
571 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
572 static inline void ptep_set_wrprotect(struct mm_struct *mm,
573 				      unsigned long address, pte_t *ptep)
574 {
575 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
576 }
577 
578 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
579 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
580 					 unsigned long address, pte_t *ptep)
581 {
582 	/*
583 	 * This comment is borrowed from x86, but applies equally to RISC-V:
584 	 *
585 	 * Clearing the accessed bit without a TLB flush
586 	 * doesn't cause data corruption. [ It could cause incorrect
587 	 * page aging and the (mistaken) reclaim of hot pages, but the
588 	 * chance of that should be relatively low. ]
589 	 *
590 	 * So as a performance optimization don't flush the TLB when
591 	 * clearing the accessed bit, it will eventually be flushed by
592 	 * a context switch or a VM operation anyway. [ In the rare
593 	 * event of it not getting flushed for a long time the delay
594 	 * shouldn't really matter because there's no real memory
595 	 * pressure for swapout to react to. ]
596 	 */
597 	return ptep_test_and_clear_young(vma, address, ptep);
598 }
599 
600 #define pgprot_noncached pgprot_noncached
601 static inline pgprot_t pgprot_noncached(pgprot_t _prot)
602 {
603 	unsigned long prot = pgprot_val(_prot);
604 
605 	prot &= ~_PAGE_MTMASK;
606 	prot |= _PAGE_IO;
607 
608 	return __pgprot(prot);
609 }
610 
611 #define pgprot_writecombine pgprot_writecombine
612 static inline pgprot_t pgprot_writecombine(pgprot_t _prot)
613 {
614 	unsigned long prot = pgprot_val(_prot);
615 
616 	prot &= ~_PAGE_MTMASK;
617 	prot |= _PAGE_NOCACHE;
618 
619 	return __pgprot(prot);
620 }
621 
622 /*
623  * THP functions
624  */
625 static inline pmd_t pte_pmd(pte_t pte)
626 {
627 	return __pmd(pte_val(pte));
628 }
629 
630 static inline pmd_t pmd_mkhuge(pmd_t pmd)
631 {
632 	return pmd;
633 }
634 
635 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
636 {
637 	return __pmd(pmd_val(pmd) & ~(_PAGE_PRESENT|_PAGE_PROT_NONE));
638 }
639 
640 #define __pmd_to_phys(pmd)  (__page_val_to_pfn(pmd_val(pmd)) << PAGE_SHIFT)
641 
642 static inline unsigned long pmd_pfn(pmd_t pmd)
643 {
644 	return ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT);
645 }
646 
647 #define __pud_to_phys(pud)  (__page_val_to_pfn(pud_val(pud)) << PAGE_SHIFT)
648 
649 static inline unsigned long pud_pfn(pud_t pud)
650 {
651 	return ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT);
652 }
653 
654 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
655 {
656 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
657 }
658 
659 #define pmd_write pmd_write
660 static inline int pmd_write(pmd_t pmd)
661 {
662 	return pte_write(pmd_pte(pmd));
663 }
664 
665 #define pmd_dirty pmd_dirty
666 static inline int pmd_dirty(pmd_t pmd)
667 {
668 	return pte_dirty(pmd_pte(pmd));
669 }
670 
671 #define pmd_young pmd_young
672 static inline int pmd_young(pmd_t pmd)
673 {
674 	return pte_young(pmd_pte(pmd));
675 }
676 
677 static inline int pmd_user(pmd_t pmd)
678 {
679 	return pte_user(pmd_pte(pmd));
680 }
681 
682 static inline pmd_t pmd_mkold(pmd_t pmd)
683 {
684 	return pte_pmd(pte_mkold(pmd_pte(pmd)));
685 }
686 
687 static inline pmd_t pmd_mkyoung(pmd_t pmd)
688 {
689 	return pte_pmd(pte_mkyoung(pmd_pte(pmd)));
690 }
691 
692 static inline pmd_t pmd_mkwrite_novma(pmd_t pmd)
693 {
694 	return pte_pmd(pte_mkwrite_novma(pmd_pte(pmd)));
695 }
696 
697 static inline pmd_t pmd_wrprotect(pmd_t pmd)
698 {
699 	return pte_pmd(pte_wrprotect(pmd_pte(pmd)));
700 }
701 
702 static inline pmd_t pmd_mkclean(pmd_t pmd)
703 {
704 	return pte_pmd(pte_mkclean(pmd_pte(pmd)));
705 }
706 
707 static inline pmd_t pmd_mkdirty(pmd_t pmd)
708 {
709 	return pte_pmd(pte_mkdirty(pmd_pte(pmd)));
710 }
711 
712 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
713 				pmd_t *pmdp, pmd_t pmd)
714 {
715 	page_table_check_pmd_set(mm, pmdp, pmd);
716 	return __set_pte_at((pte_t *)pmdp, pmd_pte(pmd));
717 }
718 
719 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
720 				pud_t *pudp, pud_t pud)
721 {
722 	page_table_check_pud_set(mm, pudp, pud);
723 	return __set_pte_at((pte_t *)pudp, pud_pte(pud));
724 }
725 
726 #ifdef CONFIG_PAGE_TABLE_CHECK
727 static inline bool pte_user_accessible_page(pte_t pte)
728 {
729 	return pte_present(pte) && pte_user(pte);
730 }
731 
732 static inline bool pmd_user_accessible_page(pmd_t pmd)
733 {
734 	return pmd_leaf(pmd) && pmd_user(pmd);
735 }
736 
737 static inline bool pud_user_accessible_page(pud_t pud)
738 {
739 	return pud_leaf(pud) && pud_user(pud);
740 }
741 #endif
742 
743 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
744 static inline int pmd_trans_huge(pmd_t pmd)
745 {
746 	return pmd_leaf(pmd);
747 }
748 
749 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
750 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
751 					unsigned long address, pmd_t *pmdp,
752 					pmd_t entry, int dirty)
753 {
754 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
755 }
756 
757 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
758 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
759 					unsigned long address, pmd_t *pmdp)
760 {
761 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
762 }
763 
764 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
765 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
766 					unsigned long address, pmd_t *pmdp)
767 {
768 	pmd_t pmd = __pmd(atomic_long_xchg((atomic_long_t *)pmdp, 0));
769 
770 	page_table_check_pmd_clear(mm, pmd);
771 
772 	return pmd;
773 }
774 
775 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
776 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
777 					unsigned long address, pmd_t *pmdp)
778 {
779 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
780 }
781 
782 #define pmdp_establish pmdp_establish
783 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
784 				unsigned long address, pmd_t *pmdp, pmd_t pmd)
785 {
786 	page_table_check_pmd_set(vma->vm_mm, pmdp, pmd);
787 	return __pmd(atomic_long_xchg((atomic_long_t *)pmdp, pmd_val(pmd)));
788 }
789 
790 #define pmdp_collapse_flush pmdp_collapse_flush
791 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
792 				 unsigned long address, pmd_t *pmdp);
793 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
794 
795 /*
796  * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
797  * are !pte_none() && !pte_present().
798  *
799  * Format of swap PTE:
800  *	bit            0:	_PAGE_PRESENT (zero)
801  *	bit       1 to 3:       _PAGE_LEAF (zero)
802  *	bit            5:	_PAGE_PROT_NONE (zero)
803  *	bit            6:	exclusive marker
804  *	bits      7 to 11:	swap type
805  *	bits 12 to XLEN-1:	swap offset
806  */
807 #define __SWP_TYPE_SHIFT	7
808 #define __SWP_TYPE_BITS		5
809 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
810 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
811 
812 #define MAX_SWAPFILES_CHECK()	\
813 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
814 
815 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
816 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
817 #define __swp_entry(type, offset) ((swp_entry_t) \
818 	{ (((type) & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT) | \
819 	  ((offset) << __SWP_OFFSET_SHIFT) })
820 
821 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
822 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
823 
824 static inline int pte_swp_exclusive(pte_t pte)
825 {
826 	return pte_val(pte) & _PAGE_SWP_EXCLUSIVE;
827 }
828 
829 static inline pte_t pte_swp_mkexclusive(pte_t pte)
830 {
831 	return __pte(pte_val(pte) | _PAGE_SWP_EXCLUSIVE);
832 }
833 
834 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
835 {
836 	return __pte(pte_val(pte) & ~_PAGE_SWP_EXCLUSIVE);
837 }
838 
839 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
840 #define __pmd_to_swp_entry(pmd) ((swp_entry_t) { pmd_val(pmd) })
841 #define __swp_entry_to_pmd(swp) __pmd((swp).val)
842 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
843 
844 /*
845  * In the RV64 Linux scheme, we give the user half of the virtual-address space
846  * and give the kernel the other (upper) half.
847  */
848 #ifdef CONFIG_64BIT
849 #define KERN_VIRT_START	(-(BIT(VA_BITS)) + TASK_SIZE)
850 #else
851 #define KERN_VIRT_START	FIXADDR_START
852 #endif
853 
854 /*
855  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
856  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
857  * Task size is:
858  * -        0x9fc00000	(~2.5GB) for RV32.
859  * -      0x4000000000	( 256GB) for RV64 using SV39 mmu
860  * -    0x800000000000	( 128TB) for RV64 using SV48 mmu
861  * - 0x100000000000000	(  64PB) for RV64 using SV57 mmu
862  *
863  * Note that PGDIR_SIZE must evenly divide TASK_SIZE since "RISC-V
864  * Instruction Set Manual Volume II: Privileged Architecture" states that
865  * "load and store effective addresses, which are 64bits, must have bits
866  * 63–48 all equal to bit 47, or else a page-fault exception will occur."
867  * Similarly for SV57, bits 63–57 must be equal to bit 56.
868  */
869 #ifdef CONFIG_64BIT
870 #define TASK_SIZE_64	(PGDIR_SIZE * PTRS_PER_PGD / 2)
871 #define TASK_SIZE_MIN	(PGDIR_SIZE_L3 * PTRS_PER_PGD / 2)
872 
873 #ifdef CONFIG_COMPAT
874 #define TASK_SIZE_32	(_AC(0x80000000, UL))
875 #define TASK_SIZE	(test_thread_flag(TIF_32BIT) ? \
876 			 TASK_SIZE_32 : TASK_SIZE_64)
877 #else
878 #define TASK_SIZE	TASK_SIZE_64
879 #endif
880 
881 #else
882 #define TASK_SIZE	FIXADDR_START
883 #define TASK_SIZE_MIN	TASK_SIZE
884 #endif
885 
886 #else /* CONFIG_MMU */
887 
888 #define PAGE_SHARED		__pgprot(0)
889 #define PAGE_KERNEL		__pgprot(0)
890 #define swapper_pg_dir		NULL
891 #define TASK_SIZE		0xffffffffUL
892 #define VMALLOC_START		_AC(0, UL)
893 #define VMALLOC_END		TASK_SIZE
894 
895 #endif /* !CONFIG_MMU */
896 
897 extern char _start[];
898 extern void *_dtb_early_va;
899 extern uintptr_t _dtb_early_pa;
900 #if defined(CONFIG_XIP_KERNEL) && defined(CONFIG_MMU)
901 #define dtb_early_va	(*(void **)XIP_FIXUP(&_dtb_early_va))
902 #define dtb_early_pa	(*(uintptr_t *)XIP_FIXUP(&_dtb_early_pa))
903 #else
904 #define dtb_early_va	_dtb_early_va
905 #define dtb_early_pa	_dtb_early_pa
906 #endif /* CONFIG_XIP_KERNEL */
907 extern u64 satp_mode;
908 
909 void paging_init(void);
910 void misc_mem_init(void);
911 
912 /*
913  * ZERO_PAGE is a global shared page that is always zero,
914  * used for zero-mapped memory areas, etc.
915  */
916 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
917 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
918 
919 #endif /* !__ASSEMBLY__ */
920 
921 #endif /* _ASM_RISCV_PGTABLE_H */
922