xref: /linux/arch/riscv/include/asm/pgtable.h (revision c894ec016c9d0418dd832202225a8c64f450d71e)
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 #ifdef CONFIG_64BIT
66 #define VA_BITS		(pgtable_l5_enabled ? \
67 				57 : (pgtable_l4_enabled ? 48 : 39))
68 #else
69 #define VA_BITS		32
70 #endif
71 
72 #define VMEMMAP_SHIFT \
73 	(VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
74 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
75 #define VMEMMAP_END	VMALLOC_START
76 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
77 
78 /*
79  * Define vmemmap for pfn_to_page & page_to_pfn calls. Needed if kernel
80  * is configured with CONFIG_SPARSEMEM_VMEMMAP enabled.
81  */
82 #define vmemmap		((struct page *)VMEMMAP_START)
83 
84 #define PCI_IO_SIZE      SZ_16M
85 #define PCI_IO_END       VMEMMAP_START
86 #define PCI_IO_START     (PCI_IO_END - PCI_IO_SIZE)
87 
88 #define FIXADDR_TOP      PCI_IO_START
89 #ifdef CONFIG_64BIT
90 #define MAX_FDT_SIZE	 PMD_SIZE
91 #define FIX_FDT_SIZE	 (MAX_FDT_SIZE + SZ_2M)
92 #define FIXADDR_SIZE     (PMD_SIZE + FIX_FDT_SIZE)
93 #else
94 #define MAX_FDT_SIZE	 PGDIR_SIZE
95 #define FIX_FDT_SIZE	 MAX_FDT_SIZE
96 #define FIXADDR_SIZE     (PGDIR_SIZE + FIX_FDT_SIZE)
97 #endif
98 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
99 
100 #endif
101 
102 #ifdef CONFIG_XIP_KERNEL
103 #define XIP_OFFSET		SZ_32M
104 #define XIP_OFFSET_MASK		(SZ_32M - 1)
105 #else
106 #define XIP_OFFSET		0
107 #endif
108 
109 #ifndef __ASSEMBLY__
110 
111 #include <asm/page.h>
112 #include <asm/tlbflush.h>
113 #include <linux/mm_types.h>
114 
115 #define __page_val_to_pfn(_val)  (((_val) & _PAGE_PFN_MASK) >> _PAGE_PFN_SHIFT)
116 
117 #ifdef CONFIG_64BIT
118 #include <asm/pgtable-64.h>
119 #else
120 #include <asm/pgtable-32.h>
121 #endif /* CONFIG_64BIT */
122 
123 #include <linux/page_table_check.h>
124 
125 #ifdef CONFIG_XIP_KERNEL
126 #define XIP_FIXUP(addr) ({							\
127 	uintptr_t __a = (uintptr_t)(addr);					\
128 	(__a >= CONFIG_XIP_PHYS_ADDR && \
129 	 __a < CONFIG_XIP_PHYS_ADDR + XIP_OFFSET * 2) ?	\
130 		__a - CONFIG_XIP_PHYS_ADDR + CONFIG_PHYS_RAM_BASE - XIP_OFFSET :\
131 		__a;								\
132 	})
133 #else
134 #define XIP_FIXUP(addr)		(addr)
135 #endif /* CONFIG_XIP_KERNEL */
136 
137 struct pt_alloc_ops {
138 	pte_t *(*get_pte_virt)(phys_addr_t pa);
139 	phys_addr_t (*alloc_pte)(uintptr_t va);
140 #ifndef __PAGETABLE_PMD_FOLDED
141 	pmd_t *(*get_pmd_virt)(phys_addr_t pa);
142 	phys_addr_t (*alloc_pmd)(uintptr_t va);
143 	pud_t *(*get_pud_virt)(phys_addr_t pa);
144 	phys_addr_t (*alloc_pud)(uintptr_t va);
145 	p4d_t *(*get_p4d_virt)(phys_addr_t pa);
146 	phys_addr_t (*alloc_p4d)(uintptr_t va);
147 #endif
148 };
149 
150 extern struct pt_alloc_ops pt_ops __initdata;
151 
152 #ifdef CONFIG_MMU
153 /* Number of PGD entries that a user-mode program can use */
154 #define USER_PTRS_PER_PGD   (TASK_SIZE / PGDIR_SIZE)
155 
156 /* Page protection bits */
157 #define _PAGE_BASE	(_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
158 
159 #define PAGE_NONE		__pgprot(_PAGE_PROT_NONE | _PAGE_READ)
160 #define PAGE_READ		__pgprot(_PAGE_BASE | _PAGE_READ)
161 #define PAGE_WRITE		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE)
162 #define PAGE_EXEC		__pgprot(_PAGE_BASE | _PAGE_EXEC)
163 #define PAGE_READ_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
164 #define PAGE_WRITE_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ |	\
165 					 _PAGE_EXEC | _PAGE_WRITE)
166 
167 #define PAGE_COPY		PAGE_READ
168 #define PAGE_COPY_EXEC		PAGE_EXEC
169 #define PAGE_COPY_READ_EXEC	PAGE_READ_EXEC
170 #define PAGE_SHARED		PAGE_WRITE
171 #define PAGE_SHARED_EXEC	PAGE_WRITE_EXEC
172 
173 #define _PAGE_KERNEL		(_PAGE_READ \
174 				| _PAGE_WRITE \
175 				| _PAGE_PRESENT \
176 				| _PAGE_ACCESSED \
177 				| _PAGE_DIRTY \
178 				| _PAGE_GLOBAL)
179 
180 #define PAGE_KERNEL		__pgprot(_PAGE_KERNEL)
181 #define PAGE_KERNEL_READ	__pgprot(_PAGE_KERNEL & ~_PAGE_WRITE)
182 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
183 #define PAGE_KERNEL_READ_EXEC	__pgprot((_PAGE_KERNEL & ~_PAGE_WRITE) \
184 					 | _PAGE_EXEC)
185 
186 #define PAGE_TABLE		__pgprot(_PAGE_TABLE)
187 
188 #define _PAGE_IOREMAP	((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO)
189 #define PAGE_KERNEL_IO		__pgprot(_PAGE_IOREMAP)
190 
191 extern pgd_t swapper_pg_dir[];
192 
193 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
194 static inline int pmd_present(pmd_t pmd)
195 {
196 	/*
197 	 * Checking for _PAGE_LEAF is needed too because:
198 	 * When splitting a THP, split_huge_page() will temporarily clear
199 	 * the present bit, in this situation, pmd_present() and
200 	 * pmd_trans_huge() still needs to return true.
201 	 */
202 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE | _PAGE_LEAF));
203 }
204 #else
205 static inline int pmd_present(pmd_t pmd)
206 {
207 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
208 }
209 #endif
210 
211 static inline int pmd_none(pmd_t pmd)
212 {
213 	return (pmd_val(pmd) == 0);
214 }
215 
216 static inline int pmd_bad(pmd_t pmd)
217 {
218 	return !pmd_present(pmd) || (pmd_val(pmd) & _PAGE_LEAF);
219 }
220 
221 #define pmd_leaf	pmd_leaf
222 static inline int pmd_leaf(pmd_t pmd)
223 {
224 	return pmd_present(pmd) && (pmd_val(pmd) & _PAGE_LEAF);
225 }
226 
227 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
228 {
229 	*pmdp = pmd;
230 }
231 
232 static inline void pmd_clear(pmd_t *pmdp)
233 {
234 	set_pmd(pmdp, __pmd(0));
235 }
236 
237 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
238 {
239 	unsigned long prot_val = pgprot_val(prot);
240 
241 	ALT_THEAD_PMA(prot_val);
242 
243 	return __pgd((pfn << _PAGE_PFN_SHIFT) | prot_val);
244 }
245 
246 static inline unsigned long _pgd_pfn(pgd_t pgd)
247 {
248 	return __page_val_to_pfn(pgd_val(pgd));
249 }
250 
251 static inline struct page *pmd_page(pmd_t pmd)
252 {
253 	return pfn_to_page(__page_val_to_pfn(pmd_val(pmd)));
254 }
255 
256 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
257 {
258 	return (unsigned long)pfn_to_virt(__page_val_to_pfn(pmd_val(pmd)));
259 }
260 
261 static inline pte_t pmd_pte(pmd_t pmd)
262 {
263 	return __pte(pmd_val(pmd));
264 }
265 
266 static inline pte_t pud_pte(pud_t pud)
267 {
268 	return __pte(pud_val(pud));
269 }
270 
271 #ifdef CONFIG_RISCV_ISA_SVNAPOT
272 
273 static __always_inline bool has_svnapot(void)
274 {
275 	return riscv_has_extension_likely(RISCV_ISA_EXT_SVNAPOT);
276 }
277 
278 static inline unsigned long pte_napot(pte_t pte)
279 {
280 	return pte_val(pte) & _PAGE_NAPOT;
281 }
282 
283 static inline pte_t pte_mknapot(pte_t pte, unsigned int order)
284 {
285 	int pos = order - 1 + _PAGE_PFN_SHIFT;
286 	unsigned long napot_bit = BIT(pos);
287 	unsigned long napot_mask = ~GENMASK(pos, _PAGE_PFN_SHIFT);
288 
289 	return __pte((pte_val(pte) & napot_mask) | napot_bit | _PAGE_NAPOT);
290 }
291 
292 #else
293 
294 static __always_inline bool has_svnapot(void) { return false; }
295 
296 static inline unsigned long pte_napot(pte_t pte)
297 {
298 	return 0;
299 }
300 
301 #endif /* CONFIG_RISCV_ISA_SVNAPOT */
302 
303 /* Yields the page frame number (PFN) of a page table entry */
304 static inline unsigned long pte_pfn(pte_t pte)
305 {
306 	unsigned long res  = __page_val_to_pfn(pte_val(pte));
307 
308 	if (has_svnapot() && pte_napot(pte))
309 		res = res & (res - 1UL);
310 
311 	return res;
312 }
313 
314 #define pte_page(x)     pfn_to_page(pte_pfn(x))
315 
316 /* Constructs a page table entry */
317 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
318 {
319 	unsigned long prot_val = pgprot_val(prot);
320 
321 	ALT_THEAD_PMA(prot_val);
322 
323 	return __pte((pfn << _PAGE_PFN_SHIFT) | prot_val);
324 }
325 
326 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
327 
328 static inline int pte_present(pte_t pte)
329 {
330 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
331 }
332 
333 static inline int pte_none(pte_t pte)
334 {
335 	return (pte_val(pte) == 0);
336 }
337 
338 static inline int pte_write(pte_t pte)
339 {
340 	return pte_val(pte) & _PAGE_WRITE;
341 }
342 
343 static inline int pte_exec(pte_t pte)
344 {
345 	return pte_val(pte) & _PAGE_EXEC;
346 }
347 
348 static inline int pte_user(pte_t pte)
349 {
350 	return pte_val(pte) & _PAGE_USER;
351 }
352 
353 static inline int pte_huge(pte_t pte)
354 {
355 	return pte_present(pte) && (pte_val(pte) & _PAGE_LEAF);
356 }
357 
358 static inline int pte_dirty(pte_t pte)
359 {
360 	return pte_val(pte) & _PAGE_DIRTY;
361 }
362 
363 static inline int pte_young(pte_t pte)
364 {
365 	return pte_val(pte) & _PAGE_ACCESSED;
366 }
367 
368 static inline int pte_special(pte_t pte)
369 {
370 	return pte_val(pte) & _PAGE_SPECIAL;
371 }
372 
373 /* static inline pte_t pte_rdprotect(pte_t pte) */
374 
375 static inline pte_t pte_wrprotect(pte_t pte)
376 {
377 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
378 }
379 
380 /* static inline pte_t pte_mkread(pte_t pte) */
381 
382 static inline pte_t pte_mkwrite(pte_t pte)
383 {
384 	return __pte(pte_val(pte) | _PAGE_WRITE);
385 }
386 
387 /* static inline pte_t pte_mkexec(pte_t pte) */
388 
389 static inline pte_t pte_mkdirty(pte_t pte)
390 {
391 	return __pte(pte_val(pte) | _PAGE_DIRTY);
392 }
393 
394 static inline pte_t pte_mkclean(pte_t pte)
395 {
396 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
397 }
398 
399 static inline pte_t pte_mkyoung(pte_t pte)
400 {
401 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
402 }
403 
404 static inline pte_t pte_mkold(pte_t pte)
405 {
406 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
407 }
408 
409 static inline pte_t pte_mkspecial(pte_t pte)
410 {
411 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
412 }
413 
414 static inline pte_t pte_mkhuge(pte_t pte)
415 {
416 	return pte;
417 }
418 
419 #ifdef CONFIG_NUMA_BALANCING
420 /*
421  * See the comment in include/asm-generic/pgtable.h
422  */
423 static inline int pte_protnone(pte_t pte)
424 {
425 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)) == _PAGE_PROT_NONE;
426 }
427 
428 static inline int pmd_protnone(pmd_t pmd)
429 {
430 	return pte_protnone(pmd_pte(pmd));
431 }
432 #endif
433 
434 /* Modify page protection bits */
435 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
436 {
437 	unsigned long newprot_val = pgprot_val(newprot);
438 
439 	ALT_THEAD_PMA(newprot_val);
440 
441 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | newprot_val);
442 }
443 
444 #define pgd_ERROR(e) \
445 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
446 
447 
448 /* Commit new configuration to MMU hardware */
449 static inline void update_mmu_cache(struct vm_area_struct *vma,
450 	unsigned long address, pte_t *ptep)
451 {
452 	/*
453 	 * The kernel assumes that TLBs don't cache invalid entries, but
454 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
455 	 * cache flush; it is necessary even after writing invalid entries.
456 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
457 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
458 	 */
459 	local_flush_tlb_page(address);
460 }
461 
462 #define __HAVE_ARCH_UPDATE_MMU_TLB
463 #define update_mmu_tlb update_mmu_cache
464 
465 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
466 		unsigned long address, pmd_t *pmdp)
467 {
468 	pte_t *ptep = (pte_t *)pmdp;
469 
470 	update_mmu_cache(vma, address, ptep);
471 }
472 
473 #define __HAVE_ARCH_PTE_SAME
474 static inline int pte_same(pte_t pte_a, pte_t pte_b)
475 {
476 	return pte_val(pte_a) == pte_val(pte_b);
477 }
478 
479 /*
480  * Certain architectures need to do special things when PTEs within
481  * a page table are directly modified.  Thus, the following hook is
482  * made available.
483  */
484 static inline void set_pte(pte_t *ptep, pte_t pteval)
485 {
486 	*ptep = pteval;
487 }
488 
489 void flush_icache_pte(pte_t pte);
490 
491 static inline void __set_pte_at(struct mm_struct *mm,
492 	unsigned long addr, pte_t *ptep, pte_t pteval)
493 {
494 	if (pte_present(pteval) && pte_exec(pteval))
495 		flush_icache_pte(pteval);
496 
497 	set_pte(ptep, pteval);
498 }
499 
500 static inline void set_pte_at(struct mm_struct *mm,
501 	unsigned long addr, pte_t *ptep, pte_t pteval)
502 {
503 	page_table_check_pte_set(mm, addr, ptep, pteval);
504 	__set_pte_at(mm, addr, ptep, pteval);
505 }
506 
507 static inline void pte_clear(struct mm_struct *mm,
508 	unsigned long addr, pte_t *ptep)
509 {
510 	__set_pte_at(mm, addr, ptep, __pte(0));
511 }
512 
513 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
514 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
515 					unsigned long address, pte_t *ptep,
516 					pte_t entry, int dirty)
517 {
518 	if (!pte_same(*ptep, entry))
519 		set_pte_at(vma->vm_mm, address, ptep, entry);
520 	/*
521 	 * update_mmu_cache will unconditionally execute, handling both
522 	 * the case that the PTE changed and the spurious fault case.
523 	 */
524 	return true;
525 }
526 
527 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
528 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
529 				       unsigned long address, pte_t *ptep)
530 {
531 	pte_t pte = __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
532 
533 	page_table_check_pte_clear(mm, address, pte);
534 
535 	return pte;
536 }
537 
538 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
539 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
540 					    unsigned long address,
541 					    pte_t *ptep)
542 {
543 	if (!pte_young(*ptep))
544 		return 0;
545 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
546 }
547 
548 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
549 static inline void ptep_set_wrprotect(struct mm_struct *mm,
550 				      unsigned long address, pte_t *ptep)
551 {
552 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
553 }
554 
555 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
556 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
557 					 unsigned long address, pte_t *ptep)
558 {
559 	/*
560 	 * This comment is borrowed from x86, but applies equally to RISC-V:
561 	 *
562 	 * Clearing the accessed bit without a TLB flush
563 	 * doesn't cause data corruption. [ It could cause incorrect
564 	 * page aging and the (mistaken) reclaim of hot pages, but the
565 	 * chance of that should be relatively low. ]
566 	 *
567 	 * So as a performance optimization don't flush the TLB when
568 	 * clearing the accessed bit, it will eventually be flushed by
569 	 * a context switch or a VM operation anyway. [ In the rare
570 	 * event of it not getting flushed for a long time the delay
571 	 * shouldn't really matter because there's no real memory
572 	 * pressure for swapout to react to. ]
573 	 */
574 	return ptep_test_and_clear_young(vma, address, ptep);
575 }
576 
577 #define pgprot_noncached pgprot_noncached
578 static inline pgprot_t pgprot_noncached(pgprot_t _prot)
579 {
580 	unsigned long prot = pgprot_val(_prot);
581 
582 	prot &= ~_PAGE_MTMASK;
583 	prot |= _PAGE_IO;
584 
585 	return __pgprot(prot);
586 }
587 
588 #define pgprot_writecombine pgprot_writecombine
589 static inline pgprot_t pgprot_writecombine(pgprot_t _prot)
590 {
591 	unsigned long prot = pgprot_val(_prot);
592 
593 	prot &= ~_PAGE_MTMASK;
594 	prot |= _PAGE_NOCACHE;
595 
596 	return __pgprot(prot);
597 }
598 
599 /*
600  * THP functions
601  */
602 static inline pmd_t pte_pmd(pte_t pte)
603 {
604 	return __pmd(pte_val(pte));
605 }
606 
607 static inline pmd_t pmd_mkhuge(pmd_t pmd)
608 {
609 	return pmd;
610 }
611 
612 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
613 {
614 	return __pmd(pmd_val(pmd) & ~(_PAGE_PRESENT|_PAGE_PROT_NONE));
615 }
616 
617 #define __pmd_to_phys(pmd)  (__page_val_to_pfn(pmd_val(pmd)) << PAGE_SHIFT)
618 
619 static inline unsigned long pmd_pfn(pmd_t pmd)
620 {
621 	return ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT);
622 }
623 
624 #define __pud_to_phys(pud)  (__page_val_to_pfn(pud_val(pud)) << PAGE_SHIFT)
625 
626 static inline unsigned long pud_pfn(pud_t pud)
627 {
628 	return ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT);
629 }
630 
631 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
632 {
633 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
634 }
635 
636 #define pmd_write pmd_write
637 static inline int pmd_write(pmd_t pmd)
638 {
639 	return pte_write(pmd_pte(pmd));
640 }
641 
642 static inline int pmd_dirty(pmd_t pmd)
643 {
644 	return pte_dirty(pmd_pte(pmd));
645 }
646 
647 #define pmd_young pmd_young
648 static inline int pmd_young(pmd_t pmd)
649 {
650 	return pte_young(pmd_pte(pmd));
651 }
652 
653 static inline int pmd_user(pmd_t pmd)
654 {
655 	return pte_user(pmd_pte(pmd));
656 }
657 
658 static inline pmd_t pmd_mkold(pmd_t pmd)
659 {
660 	return pte_pmd(pte_mkold(pmd_pte(pmd)));
661 }
662 
663 static inline pmd_t pmd_mkyoung(pmd_t pmd)
664 {
665 	return pte_pmd(pte_mkyoung(pmd_pte(pmd)));
666 }
667 
668 static inline pmd_t pmd_mkwrite(pmd_t pmd)
669 {
670 	return pte_pmd(pte_mkwrite(pmd_pte(pmd)));
671 }
672 
673 static inline pmd_t pmd_wrprotect(pmd_t pmd)
674 {
675 	return pte_pmd(pte_wrprotect(pmd_pte(pmd)));
676 }
677 
678 static inline pmd_t pmd_mkclean(pmd_t pmd)
679 {
680 	return pte_pmd(pte_mkclean(pmd_pte(pmd)));
681 }
682 
683 static inline pmd_t pmd_mkdirty(pmd_t pmd)
684 {
685 	return pte_pmd(pte_mkdirty(pmd_pte(pmd)));
686 }
687 
688 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
689 				pmd_t *pmdp, pmd_t pmd)
690 {
691 	page_table_check_pmd_set(mm, addr, pmdp, pmd);
692 	return __set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd));
693 }
694 
695 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
696 				pud_t *pudp, pud_t pud)
697 {
698 	page_table_check_pud_set(mm, addr, pudp, pud);
699 	return __set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud));
700 }
701 
702 #ifdef CONFIG_PAGE_TABLE_CHECK
703 static inline bool pte_user_accessible_page(pte_t pte)
704 {
705 	return pte_present(pte) && pte_user(pte);
706 }
707 
708 static inline bool pmd_user_accessible_page(pmd_t pmd)
709 {
710 	return pmd_leaf(pmd) && pmd_user(pmd);
711 }
712 
713 static inline bool pud_user_accessible_page(pud_t pud)
714 {
715 	return pud_leaf(pud) && pud_user(pud);
716 }
717 #endif
718 
719 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
720 static inline int pmd_trans_huge(pmd_t pmd)
721 {
722 	return pmd_leaf(pmd);
723 }
724 
725 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
726 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
727 					unsigned long address, pmd_t *pmdp,
728 					pmd_t entry, int dirty)
729 {
730 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
731 }
732 
733 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
734 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
735 					unsigned long address, pmd_t *pmdp)
736 {
737 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
738 }
739 
740 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
741 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
742 					unsigned long address, pmd_t *pmdp)
743 {
744 	pmd_t pmd = __pmd(atomic_long_xchg((atomic_long_t *)pmdp, 0));
745 
746 	page_table_check_pmd_clear(mm, address, pmd);
747 
748 	return pmd;
749 }
750 
751 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
752 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
753 					unsigned long address, pmd_t *pmdp)
754 {
755 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
756 }
757 
758 #define pmdp_establish pmdp_establish
759 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
760 				unsigned long address, pmd_t *pmdp, pmd_t pmd)
761 {
762 	page_table_check_pmd_set(vma->vm_mm, address, pmdp, pmd);
763 	return __pmd(atomic_long_xchg((atomic_long_t *)pmdp, pmd_val(pmd)));
764 }
765 
766 #define pmdp_collapse_flush pmdp_collapse_flush
767 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
768 				 unsigned long address, pmd_t *pmdp);
769 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
770 
771 /*
772  * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
773  * are !pte_none() && !pte_present().
774  *
775  * Format of swap PTE:
776  *	bit            0:	_PAGE_PRESENT (zero)
777  *	bit       1 to 3:       _PAGE_LEAF (zero)
778  *	bit            5:	_PAGE_PROT_NONE (zero)
779  *	bit            6:	exclusive marker
780  *	bits      7 to 11:	swap type
781  *	bits 11 to XLEN-1:	swap offset
782  */
783 #define __SWP_TYPE_SHIFT	7
784 #define __SWP_TYPE_BITS		5
785 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
786 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
787 
788 #define MAX_SWAPFILES_CHECK()	\
789 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
790 
791 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
792 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
793 #define __swp_entry(type, offset) ((swp_entry_t) \
794 	{ (((type) & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT) | \
795 	  ((offset) << __SWP_OFFSET_SHIFT) })
796 
797 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
798 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
799 
800 static inline int pte_swp_exclusive(pte_t pte)
801 {
802 	return pte_val(pte) & _PAGE_SWP_EXCLUSIVE;
803 }
804 
805 static inline pte_t pte_swp_mkexclusive(pte_t pte)
806 {
807 	return __pte(pte_val(pte) | _PAGE_SWP_EXCLUSIVE);
808 }
809 
810 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
811 {
812 	return __pte(pte_val(pte) & ~_PAGE_SWP_EXCLUSIVE);
813 }
814 
815 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
816 #define __pmd_to_swp_entry(pmd) ((swp_entry_t) { pmd_val(pmd) })
817 #define __swp_entry_to_pmd(swp) __pmd((swp).val)
818 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
819 
820 /*
821  * In the RV64 Linux scheme, we give the user half of the virtual-address space
822  * and give the kernel the other (upper) half.
823  */
824 #ifdef CONFIG_64BIT
825 #define KERN_VIRT_START	(-(BIT(VA_BITS)) + TASK_SIZE)
826 #else
827 #define KERN_VIRT_START	FIXADDR_START
828 #endif
829 
830 /*
831  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
832  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
833  * Task size is:
834  * -     0x9fc00000 (~2.5GB) for RV32.
835  * -   0x4000000000 ( 256GB) for RV64 using SV39 mmu
836  * - 0x800000000000 ( 128TB) for RV64 using SV48 mmu
837  *
838  * Note that PGDIR_SIZE must evenly divide TASK_SIZE since "RISC-V
839  * Instruction Set Manual Volume II: Privileged Architecture" states that
840  * "load and store effective addresses, which are 64bits, must have bits
841  * 63–48 all equal to bit 47, or else a page-fault exception will occur."
842  */
843 #ifdef CONFIG_64BIT
844 #define TASK_SIZE_64	(PGDIR_SIZE * PTRS_PER_PGD / 2)
845 #define TASK_SIZE_MIN	(PGDIR_SIZE_L3 * PTRS_PER_PGD / 2)
846 
847 #ifdef CONFIG_COMPAT
848 #define TASK_SIZE_32	(_AC(0x80000000, UL) - PAGE_SIZE)
849 #define TASK_SIZE	(test_thread_flag(TIF_32BIT) ? \
850 			 TASK_SIZE_32 : TASK_SIZE_64)
851 #else
852 #define TASK_SIZE	TASK_SIZE_64
853 #endif
854 
855 #else
856 #define TASK_SIZE	FIXADDR_START
857 #define TASK_SIZE_MIN	TASK_SIZE
858 #endif
859 
860 #else /* CONFIG_MMU */
861 
862 #define PAGE_SHARED		__pgprot(0)
863 #define PAGE_KERNEL		__pgprot(0)
864 #define swapper_pg_dir		NULL
865 #define TASK_SIZE		0xffffffffUL
866 #define VMALLOC_START		0
867 #define VMALLOC_END		TASK_SIZE
868 
869 #endif /* !CONFIG_MMU */
870 
871 extern char _start[];
872 extern void *_dtb_early_va;
873 extern uintptr_t _dtb_early_pa;
874 #if defined(CONFIG_XIP_KERNEL) && defined(CONFIG_MMU)
875 #define dtb_early_va	(*(void **)XIP_FIXUP(&_dtb_early_va))
876 #define dtb_early_pa	(*(uintptr_t *)XIP_FIXUP(&_dtb_early_pa))
877 #else
878 #define dtb_early_va	_dtb_early_va
879 #define dtb_early_pa	_dtb_early_pa
880 #endif /* CONFIG_XIP_KERNEL */
881 extern u64 satp_mode;
882 extern bool pgtable_l4_enabled;
883 
884 void paging_init(void);
885 void misc_mem_init(void);
886 
887 /*
888  * ZERO_PAGE is a global shared page that is always zero,
889  * used for zero-mapped memory areas, etc.
890  */
891 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
892 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
893 
894 #endif /* !__ASSEMBLY__ */
895 
896 #endif /* _ASM_RISCV_PGTABLE_H */
897