xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision 88a6f899)
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_READ_EXEC
169 #define PAGE_SHARED		PAGE_WRITE
170 #define PAGE_SHARED_EXEC	PAGE_WRITE_EXEC
171 
172 #define _PAGE_KERNEL		(_PAGE_READ \
173 				| _PAGE_WRITE \
174 				| _PAGE_PRESENT \
175 				| _PAGE_ACCESSED \
176 				| _PAGE_DIRTY \
177 				| _PAGE_GLOBAL)
178 
179 #define PAGE_KERNEL		__pgprot(_PAGE_KERNEL)
180 #define PAGE_KERNEL_READ	__pgprot(_PAGE_KERNEL & ~_PAGE_WRITE)
181 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
182 #define PAGE_KERNEL_READ_EXEC	__pgprot((_PAGE_KERNEL & ~_PAGE_WRITE) \
183 					 | _PAGE_EXEC)
184 
185 #define PAGE_TABLE		__pgprot(_PAGE_TABLE)
186 
187 #define _PAGE_IOREMAP	((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO)
188 #define PAGE_KERNEL_IO		__pgprot(_PAGE_IOREMAP)
189 
190 extern pgd_t swapper_pg_dir[];
191 
192 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
193 static inline int pmd_present(pmd_t pmd)
194 {
195 	/*
196 	 * Checking for _PAGE_LEAF is needed too because:
197 	 * When splitting a THP, split_huge_page() will temporarily clear
198 	 * the present bit, in this situation, pmd_present() and
199 	 * pmd_trans_huge() still needs to return true.
200 	 */
201 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE | _PAGE_LEAF));
202 }
203 #else
204 static inline int pmd_present(pmd_t pmd)
205 {
206 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
207 }
208 #endif
209 
210 static inline int pmd_none(pmd_t pmd)
211 {
212 	return (pmd_val(pmd) == 0);
213 }
214 
215 static inline int pmd_bad(pmd_t pmd)
216 {
217 	return !pmd_present(pmd) || (pmd_val(pmd) & _PAGE_LEAF);
218 }
219 
220 #define pmd_leaf	pmd_leaf
221 static inline int pmd_leaf(pmd_t pmd)
222 {
223 	return pmd_present(pmd) && (pmd_val(pmd) & _PAGE_LEAF);
224 }
225 
226 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
227 {
228 	*pmdp = pmd;
229 }
230 
231 static inline void pmd_clear(pmd_t *pmdp)
232 {
233 	set_pmd(pmdp, __pmd(0));
234 }
235 
236 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
237 {
238 	unsigned long prot_val = pgprot_val(prot);
239 
240 	ALT_THEAD_PMA(prot_val);
241 
242 	return __pgd((pfn << _PAGE_PFN_SHIFT) | prot_val);
243 }
244 
245 static inline unsigned long _pgd_pfn(pgd_t pgd)
246 {
247 	return __page_val_to_pfn(pgd_val(pgd));
248 }
249 
250 static inline struct page *pmd_page(pmd_t pmd)
251 {
252 	return pfn_to_page(__page_val_to_pfn(pmd_val(pmd)));
253 }
254 
255 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
256 {
257 	return (unsigned long)pfn_to_virt(__page_val_to_pfn(pmd_val(pmd)));
258 }
259 
260 static inline pte_t pmd_pte(pmd_t pmd)
261 {
262 	return __pte(pmd_val(pmd));
263 }
264 
265 static inline pte_t pud_pte(pud_t pud)
266 {
267 	return __pte(pud_val(pud));
268 }
269 
270 #ifdef CONFIG_RISCV_ISA_SVNAPOT
271 
272 static __always_inline bool has_svnapot(void)
273 {
274 	return riscv_has_extension_likely(RISCV_ISA_EXT_SVNAPOT);
275 }
276 
277 static inline unsigned long pte_napot(pte_t pte)
278 {
279 	return pte_val(pte) & _PAGE_NAPOT;
280 }
281 
282 static inline pte_t pte_mknapot(pte_t pte, unsigned int order)
283 {
284 	int pos = order - 1 + _PAGE_PFN_SHIFT;
285 	unsigned long napot_bit = BIT(pos);
286 	unsigned long napot_mask = ~GENMASK(pos, _PAGE_PFN_SHIFT);
287 
288 	return __pte((pte_val(pte) & napot_mask) | napot_bit | _PAGE_NAPOT);
289 }
290 
291 #else
292 
293 static __always_inline bool has_svnapot(void) { return false; }
294 
295 static inline unsigned long pte_napot(pte_t pte)
296 {
297 	return 0;
298 }
299 
300 #endif /* CONFIG_RISCV_ISA_SVNAPOT */
301 
302 /* Yields the page frame number (PFN) of a page table entry */
303 static inline unsigned long pte_pfn(pte_t pte)
304 {
305 	unsigned long res  = __page_val_to_pfn(pte_val(pte));
306 
307 	if (has_svnapot() && pte_napot(pte))
308 		res = res & (res - 1UL);
309 
310 	return res;
311 }
312 
313 #define pte_page(x)     pfn_to_page(pte_pfn(x))
314 
315 /* Constructs a page table entry */
316 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
317 {
318 	unsigned long prot_val = pgprot_val(prot);
319 
320 	ALT_THEAD_PMA(prot_val);
321 
322 	return __pte((pfn << _PAGE_PFN_SHIFT) | prot_val);
323 }
324 
325 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
326 
327 static inline int pte_present(pte_t pte)
328 {
329 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
330 }
331 
332 static inline int pte_none(pte_t pte)
333 {
334 	return (pte_val(pte) == 0);
335 }
336 
337 static inline int pte_write(pte_t pte)
338 {
339 	return pte_val(pte) & _PAGE_WRITE;
340 }
341 
342 static inline int pte_exec(pte_t pte)
343 {
344 	return pte_val(pte) & _PAGE_EXEC;
345 }
346 
347 static inline int pte_user(pte_t pte)
348 {
349 	return pte_val(pte) & _PAGE_USER;
350 }
351 
352 static inline int pte_huge(pte_t pte)
353 {
354 	return pte_present(pte) && (pte_val(pte) & _PAGE_LEAF);
355 }
356 
357 static inline int pte_dirty(pte_t pte)
358 {
359 	return pte_val(pte) & _PAGE_DIRTY;
360 }
361 
362 static inline int pte_young(pte_t pte)
363 {
364 	return pte_val(pte) & _PAGE_ACCESSED;
365 }
366 
367 static inline int pte_special(pte_t pte)
368 {
369 	return pte_val(pte) & _PAGE_SPECIAL;
370 }
371 
372 /* static inline pte_t pte_rdprotect(pte_t pte) */
373 
374 static inline pte_t pte_wrprotect(pte_t pte)
375 {
376 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
377 }
378 
379 /* static inline pte_t pte_mkread(pte_t pte) */
380 
381 static inline pte_t pte_mkwrite(pte_t pte)
382 {
383 	return __pte(pte_val(pte) | _PAGE_WRITE);
384 }
385 
386 /* static inline pte_t pte_mkexec(pte_t pte) */
387 
388 static inline pte_t pte_mkdirty(pte_t pte)
389 {
390 	return __pte(pte_val(pte) | _PAGE_DIRTY);
391 }
392 
393 static inline pte_t pte_mkclean(pte_t pte)
394 {
395 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
396 }
397 
398 static inline pte_t pte_mkyoung(pte_t pte)
399 {
400 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
401 }
402 
403 static inline pte_t pte_mkold(pte_t pte)
404 {
405 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
406 }
407 
408 static inline pte_t pte_mkspecial(pte_t pte)
409 {
410 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
411 }
412 
413 static inline pte_t pte_mkhuge(pte_t pte)
414 {
415 	return pte;
416 }
417 
418 #ifdef CONFIG_NUMA_BALANCING
419 /*
420  * See the comment in include/asm-generic/pgtable.h
421  */
422 static inline int pte_protnone(pte_t pte)
423 {
424 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)) == _PAGE_PROT_NONE;
425 }
426 
427 static inline int pmd_protnone(pmd_t pmd)
428 {
429 	return pte_protnone(pmd_pte(pmd));
430 }
431 #endif
432 
433 /* Modify page protection bits */
434 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
435 {
436 	unsigned long newprot_val = pgprot_val(newprot);
437 
438 	ALT_THEAD_PMA(newprot_val);
439 
440 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | newprot_val);
441 }
442 
443 #define pgd_ERROR(e) \
444 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
445 
446 
447 /* Commit new configuration to MMU hardware */
448 static inline void update_mmu_cache(struct vm_area_struct *vma,
449 	unsigned long address, pte_t *ptep)
450 {
451 	/*
452 	 * The kernel assumes that TLBs don't cache invalid entries, but
453 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
454 	 * cache flush; it is necessary even after writing invalid entries.
455 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
456 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
457 	 */
458 	local_flush_tlb_page(address);
459 }
460 
461 #define __HAVE_ARCH_UPDATE_MMU_TLB
462 #define update_mmu_tlb update_mmu_cache
463 
464 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
465 		unsigned long address, pmd_t *pmdp)
466 {
467 	pte_t *ptep = (pte_t *)pmdp;
468 
469 	update_mmu_cache(vma, address, ptep);
470 }
471 
472 #define __HAVE_ARCH_PTE_SAME
473 static inline int pte_same(pte_t pte_a, pte_t pte_b)
474 {
475 	return pte_val(pte_a) == pte_val(pte_b);
476 }
477 
478 /*
479  * Certain architectures need to do special things when PTEs within
480  * a page table are directly modified.  Thus, the following hook is
481  * made available.
482  */
483 static inline void set_pte(pte_t *ptep, pte_t pteval)
484 {
485 	*ptep = pteval;
486 }
487 
488 void flush_icache_pte(pte_t pte);
489 
490 static inline void __set_pte_at(struct mm_struct *mm,
491 	unsigned long addr, pte_t *ptep, pte_t pteval)
492 {
493 	if (pte_present(pteval) && pte_exec(pteval))
494 		flush_icache_pte(pteval);
495 
496 	set_pte(ptep, pteval);
497 }
498 
499 static inline void set_pte_at(struct mm_struct *mm,
500 	unsigned long addr, pte_t *ptep, pte_t pteval)
501 {
502 	page_table_check_pte_set(mm, addr, ptep, pteval);
503 	__set_pte_at(mm, addr, ptep, pteval);
504 }
505 
506 static inline void pte_clear(struct mm_struct *mm,
507 	unsigned long addr, pte_t *ptep)
508 {
509 	__set_pte_at(mm, addr, ptep, __pte(0));
510 }
511 
512 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
513 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
514 					unsigned long address, pte_t *ptep,
515 					pte_t entry, int dirty)
516 {
517 	if (!pte_same(*ptep, entry))
518 		set_pte_at(vma->vm_mm, address, ptep, entry);
519 	/*
520 	 * update_mmu_cache will unconditionally execute, handling both
521 	 * the case that the PTE changed and the spurious fault case.
522 	 */
523 	return true;
524 }
525 
526 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
527 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
528 				       unsigned long address, pte_t *ptep)
529 {
530 	pte_t pte = __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
531 
532 	page_table_check_pte_clear(mm, address, pte);
533 
534 	return pte;
535 }
536 
537 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
538 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
539 					    unsigned long address,
540 					    pte_t *ptep)
541 {
542 	if (!pte_young(*ptep))
543 		return 0;
544 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
545 }
546 
547 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
548 static inline void ptep_set_wrprotect(struct mm_struct *mm,
549 				      unsigned long address, pte_t *ptep)
550 {
551 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
552 }
553 
554 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
555 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
556 					 unsigned long address, pte_t *ptep)
557 {
558 	/*
559 	 * This comment is borrowed from x86, but applies equally to RISC-V:
560 	 *
561 	 * Clearing the accessed bit without a TLB flush
562 	 * doesn't cause data corruption. [ It could cause incorrect
563 	 * page aging and the (mistaken) reclaim of hot pages, but the
564 	 * chance of that should be relatively low. ]
565 	 *
566 	 * So as a performance optimization don't flush the TLB when
567 	 * clearing the accessed bit, it will eventually be flushed by
568 	 * a context switch or a VM operation anyway. [ In the rare
569 	 * event of it not getting flushed for a long time the delay
570 	 * shouldn't really matter because there's no real memory
571 	 * pressure for swapout to react to. ]
572 	 */
573 	return ptep_test_and_clear_young(vma, address, ptep);
574 }
575 
576 #define pgprot_noncached pgprot_noncached
577 static inline pgprot_t pgprot_noncached(pgprot_t _prot)
578 {
579 	unsigned long prot = pgprot_val(_prot);
580 
581 	prot &= ~_PAGE_MTMASK;
582 	prot |= _PAGE_IO;
583 
584 	return __pgprot(prot);
585 }
586 
587 #define pgprot_writecombine pgprot_writecombine
588 static inline pgprot_t pgprot_writecombine(pgprot_t _prot)
589 {
590 	unsigned long prot = pgprot_val(_prot);
591 
592 	prot &= ~_PAGE_MTMASK;
593 	prot |= _PAGE_NOCACHE;
594 
595 	return __pgprot(prot);
596 }
597 
598 /*
599  * THP functions
600  */
601 static inline pmd_t pte_pmd(pte_t pte)
602 {
603 	return __pmd(pte_val(pte));
604 }
605 
606 static inline pmd_t pmd_mkhuge(pmd_t pmd)
607 {
608 	return pmd;
609 }
610 
611 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
612 {
613 	return __pmd(pmd_val(pmd) & ~(_PAGE_PRESENT|_PAGE_PROT_NONE));
614 }
615 
616 #define __pmd_to_phys(pmd)  (__page_val_to_pfn(pmd_val(pmd)) << PAGE_SHIFT)
617 
618 static inline unsigned long pmd_pfn(pmd_t pmd)
619 {
620 	return ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT);
621 }
622 
623 #define __pud_to_phys(pud)  (__page_val_to_pfn(pud_val(pud)) << PAGE_SHIFT)
624 
625 static inline unsigned long pud_pfn(pud_t pud)
626 {
627 	return ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT);
628 }
629 
630 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
631 {
632 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
633 }
634 
635 #define pmd_write pmd_write
636 static inline int pmd_write(pmd_t pmd)
637 {
638 	return pte_write(pmd_pte(pmd));
639 }
640 
641 static inline int pmd_dirty(pmd_t pmd)
642 {
643 	return pte_dirty(pmd_pte(pmd));
644 }
645 
646 #define pmd_young pmd_young
647 static inline int pmd_young(pmd_t pmd)
648 {
649 	return pte_young(pmd_pte(pmd));
650 }
651 
652 static inline int pmd_user(pmd_t pmd)
653 {
654 	return pte_user(pmd_pte(pmd));
655 }
656 
657 static inline pmd_t pmd_mkold(pmd_t pmd)
658 {
659 	return pte_pmd(pte_mkold(pmd_pte(pmd)));
660 }
661 
662 static inline pmd_t pmd_mkyoung(pmd_t pmd)
663 {
664 	return pte_pmd(pte_mkyoung(pmd_pte(pmd)));
665 }
666 
667 static inline pmd_t pmd_mkwrite(pmd_t pmd)
668 {
669 	return pte_pmd(pte_mkwrite(pmd_pte(pmd)));
670 }
671 
672 static inline pmd_t pmd_wrprotect(pmd_t pmd)
673 {
674 	return pte_pmd(pte_wrprotect(pmd_pte(pmd)));
675 }
676 
677 static inline pmd_t pmd_mkclean(pmd_t pmd)
678 {
679 	return pte_pmd(pte_mkclean(pmd_pte(pmd)));
680 }
681 
682 static inline pmd_t pmd_mkdirty(pmd_t pmd)
683 {
684 	return pte_pmd(pte_mkdirty(pmd_pte(pmd)));
685 }
686 
687 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
688 				pmd_t *pmdp, pmd_t pmd)
689 {
690 	page_table_check_pmd_set(mm, addr, pmdp, pmd);
691 	return __set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd));
692 }
693 
694 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
695 				pud_t *pudp, pud_t pud)
696 {
697 	page_table_check_pud_set(mm, addr, pudp, pud);
698 	return __set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud));
699 }
700 
701 #ifdef CONFIG_PAGE_TABLE_CHECK
702 static inline bool pte_user_accessible_page(pte_t pte)
703 {
704 	return pte_present(pte) && pte_user(pte);
705 }
706 
707 static inline bool pmd_user_accessible_page(pmd_t pmd)
708 {
709 	return pmd_leaf(pmd) && pmd_user(pmd);
710 }
711 
712 static inline bool pud_user_accessible_page(pud_t pud)
713 {
714 	return pud_leaf(pud) && pud_user(pud);
715 }
716 #endif
717 
718 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
719 static inline int pmd_trans_huge(pmd_t pmd)
720 {
721 	return pmd_leaf(pmd);
722 }
723 
724 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
725 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
726 					unsigned long address, pmd_t *pmdp,
727 					pmd_t entry, int dirty)
728 {
729 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
730 }
731 
732 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
733 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
734 					unsigned long address, pmd_t *pmdp)
735 {
736 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
737 }
738 
739 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
740 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
741 					unsigned long address, pmd_t *pmdp)
742 {
743 	pmd_t pmd = __pmd(atomic_long_xchg((atomic_long_t *)pmdp, 0));
744 
745 	page_table_check_pmd_clear(mm, address, pmd);
746 
747 	return pmd;
748 }
749 
750 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
751 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
752 					unsigned long address, pmd_t *pmdp)
753 {
754 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
755 }
756 
757 #define pmdp_establish pmdp_establish
758 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
759 				unsigned long address, pmd_t *pmdp, pmd_t pmd)
760 {
761 	page_table_check_pmd_set(vma->vm_mm, address, pmdp, pmd);
762 	return __pmd(atomic_long_xchg((atomic_long_t *)pmdp, pmd_val(pmd)));
763 }
764 
765 #define pmdp_collapse_flush pmdp_collapse_flush
766 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
767 				 unsigned long address, pmd_t *pmdp);
768 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
769 
770 /*
771  * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
772  * are !pte_none() && !pte_present().
773  *
774  * Format of swap PTE:
775  *	bit            0:	_PAGE_PRESENT (zero)
776  *	bit       1 to 3:       _PAGE_LEAF (zero)
777  *	bit            5:	_PAGE_PROT_NONE (zero)
778  *	bit            6:	exclusive marker
779  *	bits      7 to 11:	swap type
780  *	bits 11 to XLEN-1:	swap offset
781  */
782 #define __SWP_TYPE_SHIFT	7
783 #define __SWP_TYPE_BITS		5
784 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
785 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
786 
787 #define MAX_SWAPFILES_CHECK()	\
788 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
789 
790 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
791 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
792 #define __swp_entry(type, offset) ((swp_entry_t) \
793 	{ (((type) & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT) | \
794 	  ((offset) << __SWP_OFFSET_SHIFT) })
795 
796 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
797 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
798 
799 static inline int pte_swp_exclusive(pte_t pte)
800 {
801 	return pte_val(pte) & _PAGE_SWP_EXCLUSIVE;
802 }
803 
804 static inline pte_t pte_swp_mkexclusive(pte_t pte)
805 {
806 	return __pte(pte_val(pte) | _PAGE_SWP_EXCLUSIVE);
807 }
808 
809 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
810 {
811 	return __pte(pte_val(pte) & ~_PAGE_SWP_EXCLUSIVE);
812 }
813 
814 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
815 #define __pmd_to_swp_entry(pmd) ((swp_entry_t) { pmd_val(pmd) })
816 #define __swp_entry_to_pmd(swp) __pmd((swp).val)
817 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
818 
819 /*
820  * In the RV64 Linux scheme, we give the user half of the virtual-address space
821  * and give the kernel the other (upper) half.
822  */
823 #ifdef CONFIG_64BIT
824 #define KERN_VIRT_START	(-(BIT(VA_BITS)) + TASK_SIZE)
825 #else
826 #define KERN_VIRT_START	FIXADDR_START
827 #endif
828 
829 /*
830  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
831  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
832  * Task size is:
833  * -     0x9fc00000 (~2.5GB) for RV32.
834  * -   0x4000000000 ( 256GB) for RV64 using SV39 mmu
835  * - 0x800000000000 ( 128TB) for RV64 using SV48 mmu
836  *
837  * Note that PGDIR_SIZE must evenly divide TASK_SIZE since "RISC-V
838  * Instruction Set Manual Volume II: Privileged Architecture" states that
839  * "load and store effective addresses, which are 64bits, must have bits
840  * 63–48 all equal to bit 47, or else a page-fault exception will occur."
841  */
842 #ifdef CONFIG_64BIT
843 #define TASK_SIZE_64	(PGDIR_SIZE * PTRS_PER_PGD / 2)
844 #define TASK_SIZE_MIN	(PGDIR_SIZE_L3 * PTRS_PER_PGD / 2)
845 
846 #ifdef CONFIG_COMPAT
847 #define TASK_SIZE_32	(_AC(0x80000000, UL) - PAGE_SIZE)
848 #define TASK_SIZE	(test_thread_flag(TIF_32BIT) ? \
849 			 TASK_SIZE_32 : TASK_SIZE_64)
850 #else
851 #define TASK_SIZE	TASK_SIZE_64
852 #endif
853 
854 #else
855 #define TASK_SIZE	FIXADDR_START
856 #define TASK_SIZE_MIN	TASK_SIZE
857 #endif
858 
859 #else /* CONFIG_MMU */
860 
861 #define PAGE_SHARED		__pgprot(0)
862 #define PAGE_KERNEL		__pgprot(0)
863 #define swapper_pg_dir		NULL
864 #define TASK_SIZE		0xffffffffUL
865 #define VMALLOC_START		0
866 #define VMALLOC_END		TASK_SIZE
867 
868 #endif /* !CONFIG_MMU */
869 
870 extern char _start[];
871 extern void *_dtb_early_va;
872 extern uintptr_t _dtb_early_pa;
873 #if defined(CONFIG_XIP_KERNEL) && defined(CONFIG_MMU)
874 #define dtb_early_va	(*(void **)XIP_FIXUP(&_dtb_early_va))
875 #define dtb_early_pa	(*(uintptr_t *)XIP_FIXUP(&_dtb_early_pa))
876 #else
877 #define dtb_early_va	_dtb_early_va
878 #define dtb_early_pa	_dtb_early_pa
879 #endif /* CONFIG_XIP_KERNEL */
880 extern u64 satp_mode;
881 extern bool pgtable_l4_enabled;
882 
883 void paging_init(void);
884 void misc_mem_init(void);
885 
886 /*
887  * ZERO_PAGE is a global shared page that is always zero,
888  * used for zero-mapped memory areas, etc.
889  */
890 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
891 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
892 
893 #endif /* !__ASSEMBLY__ */
894 
895 #endif /* _ASM_RISCV_PGTABLE_H */
896