xref: /linux/arch/riscv/include/asm/pgtable.h (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 #else
59 #define MODULES_VADDR		VMALLOC_START
60 #define MODULES_END		VMALLOC_END
61 #endif
62 
63 /*
64  * Roughly size the vmemmap space to be large enough to fit enough
65  * struct pages to map half the virtual address space. Then
66  * position vmemmap directly below the VMALLOC region.
67  */
68 #define VA_BITS_SV32 32
69 #ifdef CONFIG_64BIT
70 #define VA_BITS_SV39 39
71 #define VA_BITS_SV48 48
72 #define VA_BITS_SV57 57
73 
74 #define VA_BITS		(pgtable_l5_enabled ? \
75 				VA_BITS_SV57 : (pgtable_l4_enabled ? VA_BITS_SV48 : VA_BITS_SV39))
76 #else
77 #define VA_BITS		VA_BITS_SV32
78 #endif
79 
80 #define VMEMMAP_SHIFT \
81 	(VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
82 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
83 #define VMEMMAP_END	VMALLOC_START
84 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
85 
86 /*
87  * Define vmemmap for pfn_to_page & page_to_pfn calls. Needed if kernel
88  * is configured with CONFIG_SPARSEMEM_VMEMMAP enabled.
89  */
90 #define vmemmap		((struct page *)VMEMMAP_START - (phys_ram_base >> PAGE_SHIFT))
91 
92 #define PCI_IO_SIZE      SZ_16M
93 #define PCI_IO_END       VMEMMAP_START
94 #define PCI_IO_START     (PCI_IO_END - PCI_IO_SIZE)
95 
96 #define FIXADDR_TOP      PCI_IO_START
97 #ifdef CONFIG_64BIT
98 #define MAX_FDT_SIZE	 PMD_SIZE
99 #define FIX_FDT_SIZE	 (MAX_FDT_SIZE + SZ_2M)
100 #define FIXADDR_SIZE     (PMD_SIZE + FIX_FDT_SIZE)
101 #else
102 #define MAX_FDT_SIZE	 PGDIR_SIZE
103 #define FIX_FDT_SIZE	 MAX_FDT_SIZE
104 #define FIXADDR_SIZE     (PGDIR_SIZE + FIX_FDT_SIZE)
105 #endif
106 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
107 
108 #endif
109 
110 #ifndef __ASSEMBLY__
111 
112 #include <asm/page.h>
113 #include <asm/tlbflush.h>
114 #include <linux/mm_types.h>
115 #include <asm/compat.h>
116 #include <asm/cpufeature.h>
117 
118 #define __page_val_to_pfn(_val)  (((_val) & _PAGE_PFN_MASK) >> _PAGE_PFN_SHIFT)
119 
120 #ifdef CONFIG_64BIT
121 #include <asm/pgtable-64.h>
122 
123 #define VA_USER_SV39 (UL(1) << (VA_BITS_SV39 - 1))
124 #define VA_USER_SV48 (UL(1) << (VA_BITS_SV48 - 1))
125 #define VA_USER_SV57 (UL(1) << (VA_BITS_SV57 - 1))
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_IOREMAP	((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO)
203 #define PAGE_KERNEL_IO		__pgprot(_PAGE_IOREMAP)
204 
205 extern pgd_t swapper_pg_dir[];
206 extern pgd_t trampoline_pg_dir[];
207 extern pgd_t early_pg_dir[];
208 
209 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
210 static inline int pmd_present(pmd_t pmd)
211 {
212 	/*
213 	 * Checking for _PAGE_LEAF is needed too because:
214 	 * When splitting a THP, split_huge_page() will temporarily clear
215 	 * the present bit, in this situation, pmd_present() and
216 	 * pmd_trans_huge() still needs to return true.
217 	 */
218 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE | _PAGE_LEAF));
219 }
220 #else
221 static inline int pmd_present(pmd_t pmd)
222 {
223 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
224 }
225 #endif
226 
227 static inline int pmd_none(pmd_t pmd)
228 {
229 	return (pmd_val(pmd) == 0);
230 }
231 
232 static inline int pmd_bad(pmd_t pmd)
233 {
234 	return !pmd_present(pmd) || (pmd_val(pmd) & _PAGE_LEAF);
235 }
236 
237 #define pmd_leaf	pmd_leaf
238 static inline bool pmd_leaf(pmd_t pmd)
239 {
240 	return pmd_present(pmd) && (pmd_val(pmd) & _PAGE_LEAF);
241 }
242 
243 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
244 {
245 	WRITE_ONCE(*pmdp, pmd);
246 }
247 
248 static inline void pmd_clear(pmd_t *pmdp)
249 {
250 	set_pmd(pmdp, __pmd(0));
251 }
252 
253 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
254 {
255 	unsigned long prot_val = pgprot_val(prot);
256 
257 	ALT_THEAD_PMA(prot_val);
258 
259 	return __pgd((pfn << _PAGE_PFN_SHIFT) | prot_val);
260 }
261 
262 static inline unsigned long _pgd_pfn(pgd_t pgd)
263 {
264 	return __page_val_to_pfn(pgd_val(pgd));
265 }
266 
267 static inline struct page *pmd_page(pmd_t pmd)
268 {
269 	return pfn_to_page(__page_val_to_pfn(pmd_val(pmd)));
270 }
271 
272 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
273 {
274 	return (unsigned long)pfn_to_virt(__page_val_to_pfn(pmd_val(pmd)));
275 }
276 
277 static inline pte_t pmd_pte(pmd_t pmd)
278 {
279 	return __pte(pmd_val(pmd));
280 }
281 
282 static inline pte_t pud_pte(pud_t pud)
283 {
284 	return __pte(pud_val(pud));
285 }
286 
287 #ifdef CONFIG_RISCV_ISA_SVNAPOT
288 
289 static __always_inline bool has_svnapot(void)
290 {
291 	return riscv_has_extension_likely(RISCV_ISA_EXT_SVNAPOT);
292 }
293 
294 static inline unsigned long pte_napot(pte_t pte)
295 {
296 	return pte_val(pte) & _PAGE_NAPOT;
297 }
298 
299 static inline pte_t pte_mknapot(pte_t pte, unsigned int order)
300 {
301 	int pos = order - 1 + _PAGE_PFN_SHIFT;
302 	unsigned long napot_bit = BIT(pos);
303 	unsigned long napot_mask = ~GENMASK(pos, _PAGE_PFN_SHIFT);
304 
305 	return __pte((pte_val(pte) & napot_mask) | napot_bit | _PAGE_NAPOT);
306 }
307 
308 #else
309 
310 static __always_inline bool has_svnapot(void) { return false; }
311 
312 static inline unsigned long pte_napot(pte_t pte)
313 {
314 	return 0;
315 }
316 
317 #endif /* CONFIG_RISCV_ISA_SVNAPOT */
318 
319 /* Yields the page frame number (PFN) of a page table entry */
320 static inline unsigned long pte_pfn(pte_t pte)
321 {
322 	unsigned long res  = __page_val_to_pfn(pte_val(pte));
323 
324 	if (has_svnapot() && pte_napot(pte))
325 		res = res & (res - 1UL);
326 
327 	return res;
328 }
329 
330 #define pte_page(x)     pfn_to_page(pte_pfn(x))
331 
332 /* Constructs a page table entry */
333 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
334 {
335 	unsigned long prot_val = pgprot_val(prot);
336 
337 	ALT_THEAD_PMA(prot_val);
338 
339 	return __pte((pfn << _PAGE_PFN_SHIFT) | prot_val);
340 }
341 
342 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
343 
344 static inline int pte_present(pte_t pte)
345 {
346 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
347 }
348 
349 #define pte_accessible pte_accessible
350 static inline unsigned long pte_accessible(struct mm_struct *mm, pte_t a)
351 {
352 	if (pte_val(a) & _PAGE_PRESENT)
353 		return true;
354 
355 	if ((pte_val(a) & _PAGE_PROT_NONE) &&
356 	    atomic_read(&mm->tlb_flush_pending))
357 		return true;
358 
359 	return false;
360 }
361 
362 static inline int pte_none(pte_t pte)
363 {
364 	return (pte_val(pte) == 0);
365 }
366 
367 static inline int pte_write(pte_t pte)
368 {
369 	return pte_val(pte) & _PAGE_WRITE;
370 }
371 
372 static inline int pte_exec(pte_t pte)
373 {
374 	return pte_val(pte) & _PAGE_EXEC;
375 }
376 
377 static inline int pte_user(pte_t pte)
378 {
379 	return pte_val(pte) & _PAGE_USER;
380 }
381 
382 static inline int pte_huge(pte_t pte)
383 {
384 	return pte_present(pte) && (pte_val(pte) & _PAGE_LEAF);
385 }
386 
387 static inline int pte_dirty(pte_t pte)
388 {
389 	return pte_val(pte) & _PAGE_DIRTY;
390 }
391 
392 static inline int pte_young(pte_t pte)
393 {
394 	return pte_val(pte) & _PAGE_ACCESSED;
395 }
396 
397 static inline int pte_special(pte_t pte)
398 {
399 	return pte_val(pte) & _PAGE_SPECIAL;
400 }
401 
402 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
403 static inline int pte_devmap(pte_t pte)
404 {
405 	return pte_val(pte) & _PAGE_DEVMAP;
406 }
407 #endif
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_mkdevmap(pte_t pte)
451 {
452 	return __pte(pte_val(pte) | _PAGE_DEVMAP);
453 }
454 
455 static inline pte_t pte_mkhuge(pte_t pte)
456 {
457 	return pte;
458 }
459 
460 #ifdef CONFIG_RISCV_ISA_SVNAPOT
461 #define pte_leaf_size(pte)	(pte_napot(pte) ?				\
462 					napot_cont_size(napot_cont_order(pte)) :\
463 					PAGE_SIZE)
464 #endif
465 
466 #ifdef CONFIG_NUMA_BALANCING
467 /*
468  * See the comment in include/asm-generic/pgtable.h
469  */
470 static inline int pte_protnone(pte_t pte)
471 {
472 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)) == _PAGE_PROT_NONE;
473 }
474 
475 static inline int pmd_protnone(pmd_t pmd)
476 {
477 	return pte_protnone(pmd_pte(pmd));
478 }
479 #endif
480 
481 /* Modify page protection bits */
482 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
483 {
484 	unsigned long newprot_val = pgprot_val(newprot);
485 
486 	ALT_THEAD_PMA(newprot_val);
487 
488 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | newprot_val);
489 }
490 
491 #define pgd_ERROR(e) \
492 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
493 
494 
495 /* Commit new configuration to MMU hardware */
496 static inline void update_mmu_cache_range(struct vm_fault *vmf,
497 		struct vm_area_struct *vma, unsigned long address,
498 		pte_t *ptep, unsigned int nr)
499 {
500 	asm goto(ALTERNATIVE("nop", "j %l[svvptc]", 0, RISCV_ISA_EXT_SVVPTC, 1)
501 		 : : : : svvptc);
502 
503 	/*
504 	 * The kernel assumes that TLBs don't cache invalid entries, but
505 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
506 	 * cache flush; it is necessary even after writing invalid entries.
507 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
508 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
509 	 */
510 	while (nr--)
511 		local_flush_tlb_page(address + nr * PAGE_SIZE);
512 
513 svvptc:;
514 	/*
515 	 * Svvptc guarantees that the new valid pte will be visible within
516 	 * a bounded timeframe, so when the uarch does not cache invalid
517 	 * entries, we don't have to do anything.
518 	 */
519 }
520 #define update_mmu_cache(vma, addr, ptep) \
521 	update_mmu_cache_range(NULL, vma, addr, ptep, 1)
522 
523 #define update_mmu_tlb_range(vma, addr, ptep, nr) \
524 	update_mmu_cache_range(NULL, vma, addr, ptep, nr)
525 
526 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
527 		unsigned long address, pmd_t *pmdp)
528 {
529 	pte_t *ptep = (pte_t *)pmdp;
530 
531 	update_mmu_cache(vma, address, ptep);
532 }
533 
534 #define __HAVE_ARCH_PTE_SAME
535 static inline int pte_same(pte_t pte_a, pte_t pte_b)
536 {
537 	return pte_val(pte_a) == pte_val(pte_b);
538 }
539 
540 /*
541  * Certain architectures need to do special things when PTEs within
542  * a page table are directly modified.  Thus, the following hook is
543  * made available.
544  */
545 static inline void set_pte(pte_t *ptep, pte_t pteval)
546 {
547 	WRITE_ONCE(*ptep, pteval);
548 }
549 
550 void flush_icache_pte(struct mm_struct *mm, pte_t pte);
551 
552 static inline void __set_pte_at(struct mm_struct *mm, pte_t *ptep, pte_t pteval)
553 {
554 	if (pte_present(pteval) && pte_exec(pteval))
555 		flush_icache_pte(mm, pteval);
556 
557 	set_pte(ptep, pteval);
558 }
559 
560 #define PFN_PTE_SHIFT		_PAGE_PFN_SHIFT
561 
562 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
563 		pte_t *ptep, pte_t pteval, unsigned int nr)
564 {
565 	page_table_check_ptes_set(mm, ptep, pteval, nr);
566 
567 	for (;;) {
568 		__set_pte_at(mm, ptep, pteval);
569 		if (--nr == 0)
570 			break;
571 		ptep++;
572 		pte_val(pteval) += 1 << _PAGE_PFN_SHIFT;
573 	}
574 }
575 #define set_ptes set_ptes
576 
577 static inline void pte_clear(struct mm_struct *mm,
578 	unsigned long addr, pte_t *ptep)
579 {
580 	__set_pte_at(mm, ptep, __pte(0));
581 }
582 
583 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS	/* defined in mm/pgtable.c */
584 extern int ptep_set_access_flags(struct vm_area_struct *vma, unsigned long address,
585 				 pte_t *ptep, pte_t entry, int dirty);
586 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG	/* defined in mm/pgtable.c */
587 extern int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long address,
588 				     pte_t *ptep);
589 
590 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
591 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
592 				       unsigned long address, pte_t *ptep)
593 {
594 	pte_t pte = __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
595 
596 	page_table_check_pte_clear(mm, pte);
597 
598 	return pte;
599 }
600 
601 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
602 static inline void ptep_set_wrprotect(struct mm_struct *mm,
603 				      unsigned long address, pte_t *ptep)
604 {
605 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
606 }
607 
608 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
609 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
610 					 unsigned long address, pte_t *ptep)
611 {
612 	/*
613 	 * This comment is borrowed from x86, but applies equally to RISC-V:
614 	 *
615 	 * Clearing the accessed bit without a TLB flush
616 	 * doesn't cause data corruption. [ It could cause incorrect
617 	 * page aging and the (mistaken) reclaim of hot pages, but the
618 	 * chance of that should be relatively low. ]
619 	 *
620 	 * So as a performance optimization don't flush the TLB when
621 	 * clearing the accessed bit, it will eventually be flushed by
622 	 * a context switch or a VM operation anyway. [ In the rare
623 	 * event of it not getting flushed for a long time the delay
624 	 * shouldn't really matter because there's no real memory
625 	 * pressure for swapout to react to. ]
626 	 */
627 	return ptep_test_and_clear_young(vma, address, ptep);
628 }
629 
630 #define pgprot_nx pgprot_nx
631 static inline pgprot_t pgprot_nx(pgprot_t _prot)
632 {
633 	return __pgprot(pgprot_val(_prot) & ~_PAGE_EXEC);
634 }
635 
636 #define pgprot_noncached pgprot_noncached
637 static inline pgprot_t pgprot_noncached(pgprot_t _prot)
638 {
639 	unsigned long prot = pgprot_val(_prot);
640 
641 	prot &= ~_PAGE_MTMASK;
642 	prot |= _PAGE_IO;
643 
644 	return __pgprot(prot);
645 }
646 
647 #define pgprot_writecombine pgprot_writecombine
648 static inline pgprot_t pgprot_writecombine(pgprot_t _prot)
649 {
650 	unsigned long prot = pgprot_val(_prot);
651 
652 	prot &= ~_PAGE_MTMASK;
653 	prot |= _PAGE_NOCACHE;
654 
655 	return __pgprot(prot);
656 }
657 
658 /*
659  * Both Svade and Svadu control the hardware behavior when the PTE A/D bits need to be set. By
660  * default the M-mode firmware enables the hardware updating scheme when only Svadu is present in
661  * DT.
662  */
663 #define arch_has_hw_pte_young arch_has_hw_pte_young
664 static inline bool arch_has_hw_pte_young(void)
665 {
666 	return riscv_has_extension_unlikely(RISCV_ISA_EXT_SVADU);
667 }
668 
669 /*
670  * THP functions
671  */
672 static inline pmd_t pte_pmd(pte_t pte)
673 {
674 	return __pmd(pte_val(pte));
675 }
676 
677 static inline pmd_t pmd_mkhuge(pmd_t pmd)
678 {
679 	return pmd;
680 }
681 
682 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
683 {
684 	return __pmd(pmd_val(pmd) & ~(_PAGE_PRESENT|_PAGE_PROT_NONE));
685 }
686 
687 #define __pmd_to_phys(pmd)  (__page_val_to_pfn(pmd_val(pmd)) << PAGE_SHIFT)
688 
689 static inline unsigned long pmd_pfn(pmd_t pmd)
690 {
691 	return ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT);
692 }
693 
694 #define __pud_to_phys(pud)  (__page_val_to_pfn(pud_val(pud)) << PAGE_SHIFT)
695 
696 #define pud_pfn pud_pfn
697 static inline unsigned long pud_pfn(pud_t pud)
698 {
699 	return ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT);
700 }
701 
702 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
703 {
704 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
705 }
706 
707 #define pmd_write pmd_write
708 static inline int pmd_write(pmd_t pmd)
709 {
710 	return pte_write(pmd_pte(pmd));
711 }
712 
713 #define pud_write pud_write
714 static inline int pud_write(pud_t pud)
715 {
716 	return pte_write(pud_pte(pud));
717 }
718 
719 #define pmd_dirty pmd_dirty
720 static inline int pmd_dirty(pmd_t pmd)
721 {
722 	return pte_dirty(pmd_pte(pmd));
723 }
724 
725 #define pmd_young pmd_young
726 static inline int pmd_young(pmd_t pmd)
727 {
728 	return pte_young(pmd_pte(pmd));
729 }
730 
731 static inline int pmd_user(pmd_t pmd)
732 {
733 	return pte_user(pmd_pte(pmd));
734 }
735 
736 static inline pmd_t pmd_mkold(pmd_t pmd)
737 {
738 	return pte_pmd(pte_mkold(pmd_pte(pmd)));
739 }
740 
741 static inline pmd_t pmd_mkyoung(pmd_t pmd)
742 {
743 	return pte_pmd(pte_mkyoung(pmd_pte(pmd)));
744 }
745 
746 static inline pmd_t pmd_mkwrite_novma(pmd_t pmd)
747 {
748 	return pte_pmd(pte_mkwrite_novma(pmd_pte(pmd)));
749 }
750 
751 static inline pmd_t pmd_wrprotect(pmd_t pmd)
752 {
753 	return pte_pmd(pte_wrprotect(pmd_pte(pmd)));
754 }
755 
756 static inline pmd_t pmd_mkclean(pmd_t pmd)
757 {
758 	return pte_pmd(pte_mkclean(pmd_pte(pmd)));
759 }
760 
761 static inline pmd_t pmd_mkdirty(pmd_t pmd)
762 {
763 	return pte_pmd(pte_mkdirty(pmd_pte(pmd)));
764 }
765 
766 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
767 {
768 	return pte_pmd(pte_mkdevmap(pmd_pte(pmd)));
769 }
770 
771 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
772 				pmd_t *pmdp, pmd_t pmd)
773 {
774 	page_table_check_pmd_set(mm, pmdp, pmd);
775 	return __set_pte_at(mm, (pte_t *)pmdp, pmd_pte(pmd));
776 }
777 
778 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
779 				pud_t *pudp, pud_t pud)
780 {
781 	page_table_check_pud_set(mm, pudp, pud);
782 	return __set_pte_at(mm, (pte_t *)pudp, pud_pte(pud));
783 }
784 
785 #ifdef CONFIG_PAGE_TABLE_CHECK
786 static inline bool pte_user_accessible_page(pte_t pte)
787 {
788 	return pte_present(pte) && pte_user(pte);
789 }
790 
791 static inline bool pmd_user_accessible_page(pmd_t pmd)
792 {
793 	return pmd_leaf(pmd) && pmd_user(pmd);
794 }
795 
796 static inline bool pud_user_accessible_page(pud_t pud)
797 {
798 	return pud_leaf(pud) && pud_user(pud);
799 }
800 #endif
801 
802 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
803 static inline int pmd_trans_huge(pmd_t pmd)
804 {
805 	return pmd_leaf(pmd);
806 }
807 
808 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
809 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
810 					unsigned long address, pmd_t *pmdp,
811 					pmd_t entry, int dirty)
812 {
813 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
814 }
815 
816 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
817 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
818 					unsigned long address, pmd_t *pmdp)
819 {
820 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
821 }
822 
823 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
824 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
825 					unsigned long address, pmd_t *pmdp)
826 {
827 	pmd_t pmd = __pmd(atomic_long_xchg((atomic_long_t *)pmdp, 0));
828 
829 	page_table_check_pmd_clear(mm, pmd);
830 
831 	return pmd;
832 }
833 
834 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
835 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
836 					unsigned long address, pmd_t *pmdp)
837 {
838 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
839 }
840 
841 #define pmdp_establish pmdp_establish
842 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
843 				unsigned long address, pmd_t *pmdp, pmd_t pmd)
844 {
845 	page_table_check_pmd_set(vma->vm_mm, pmdp, pmd);
846 	return __pmd(atomic_long_xchg((atomic_long_t *)pmdp, pmd_val(pmd)));
847 }
848 
849 #define pmdp_collapse_flush pmdp_collapse_flush
850 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
851 				 unsigned long address, pmd_t *pmdp);
852 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
853 
854 /*
855  * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
856  * are !pte_none() && !pte_present().
857  *
858  * Format of swap PTE:
859  *	bit            0:	_PAGE_PRESENT (zero)
860  *	bit       1 to 3:       _PAGE_LEAF (zero)
861  *	bit            5:	_PAGE_PROT_NONE (zero)
862  *	bit            6:	exclusive marker
863  *	bits      7 to 11:	swap type
864  *	bits 12 to XLEN-1:	swap offset
865  */
866 #define __SWP_TYPE_SHIFT	7
867 #define __SWP_TYPE_BITS		5
868 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
869 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
870 
871 #define MAX_SWAPFILES_CHECK()	\
872 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
873 
874 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
875 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
876 #define __swp_entry(type, offset) ((swp_entry_t) \
877 	{ (((type) & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT) | \
878 	  ((offset) << __SWP_OFFSET_SHIFT) })
879 
880 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
881 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
882 
883 static inline int pte_swp_exclusive(pte_t pte)
884 {
885 	return pte_val(pte) & _PAGE_SWP_EXCLUSIVE;
886 }
887 
888 static inline pte_t pte_swp_mkexclusive(pte_t pte)
889 {
890 	return __pte(pte_val(pte) | _PAGE_SWP_EXCLUSIVE);
891 }
892 
893 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
894 {
895 	return __pte(pte_val(pte) & ~_PAGE_SWP_EXCLUSIVE);
896 }
897 
898 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
899 #define __pmd_to_swp_entry(pmd) ((swp_entry_t) { pmd_val(pmd) })
900 #define __swp_entry_to_pmd(swp) __pmd((swp).val)
901 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
902 
903 /*
904  * In the RV64 Linux scheme, we give the user half of the virtual-address space
905  * and give the kernel the other (upper) half.
906  */
907 #ifdef CONFIG_64BIT
908 #define KERN_VIRT_START	(-(BIT(VA_BITS)) + TASK_SIZE)
909 #else
910 #define KERN_VIRT_START	FIXADDR_START
911 #endif
912 
913 /*
914  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
915  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
916  * Task size is:
917  * -        0x9fc00000	(~2.5GB) for RV32.
918  * -      0x4000000000	( 256GB) for RV64 using SV39 mmu
919  * -    0x800000000000	( 128TB) for RV64 using SV48 mmu
920  * - 0x100000000000000	(  64PB) for RV64 using SV57 mmu
921  *
922  * Note that PGDIR_SIZE must evenly divide TASK_SIZE since "RISC-V
923  * Instruction Set Manual Volume II: Privileged Architecture" states that
924  * "load and store effective addresses, which are 64bits, must have bits
925  * 63–48 all equal to bit 47, or else a page-fault exception will occur."
926  * Similarly for SV57, bits 63–57 must be equal to bit 56.
927  */
928 #ifdef CONFIG_64BIT
929 #define TASK_SIZE_64	(PGDIR_SIZE * PTRS_PER_PGD / 2)
930 #define TASK_SIZE_MAX	LONG_MAX
931 
932 #ifdef CONFIG_COMPAT
933 #define TASK_SIZE_32	(_AC(0x80000000, UL) - PAGE_SIZE)
934 #define TASK_SIZE	(is_compat_task() ? \
935 			 TASK_SIZE_32 : TASK_SIZE_64)
936 #else
937 #define TASK_SIZE	TASK_SIZE_64
938 #endif
939 
940 #else
941 #define TASK_SIZE	FIXADDR_START
942 #endif
943 
944 #else /* CONFIG_MMU */
945 
946 #define PAGE_SHARED		__pgprot(0)
947 #define PAGE_KERNEL		__pgprot(0)
948 #define swapper_pg_dir		NULL
949 #define TASK_SIZE		_AC(-1, UL)
950 #define VMALLOC_START		_AC(0, UL)
951 #define VMALLOC_END		TASK_SIZE
952 
953 #endif /* !CONFIG_MMU */
954 
955 extern char _start[];
956 extern void *_dtb_early_va;
957 extern uintptr_t _dtb_early_pa;
958 #if defined(CONFIG_XIP_KERNEL) && defined(CONFIG_MMU)
959 #define dtb_early_va	(*(void **)XIP_FIXUP(&_dtb_early_va))
960 #define dtb_early_pa	(*(uintptr_t *)XIP_FIXUP(&_dtb_early_pa))
961 #else
962 #define dtb_early_va	_dtb_early_va
963 #define dtb_early_pa	_dtb_early_pa
964 #endif /* CONFIG_XIP_KERNEL */
965 extern u64 satp_mode;
966 
967 void paging_init(void);
968 void misc_mem_init(void);
969 
970 /*
971  * ZERO_PAGE is a global shared page that is always zero,
972  * used for zero-mapped memory areas, etc.
973  */
974 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
975 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
976 
977 /*
978  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
979  * TLB flush will be required as a result of the "set". For example, use
980  * in scenarios where it is known ahead of time that the routine is
981  * setting non-present entries, or re-setting an existing entry to the
982  * same value. Otherwise, use the typical "set" helpers and flush the
983  * TLB.
984  */
985 #define set_p4d_safe(p4dp, p4d) \
986 ({ \
987 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
988 	set_p4d(p4dp, p4d); \
989 })
990 
991 #define set_pgd_safe(pgdp, pgd) \
992 ({ \
993 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
994 	set_pgd(pgdp, pgd); \
995 })
996 #endif /* !__ASSEMBLY__ */
997 
998 #endif /* _ASM_RISCV_PGTABLE_H */
999