xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision ef97e774)
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 /* Yields the page frame number (PFN) of a page table entry */
272 static inline unsigned long pte_pfn(pte_t pte)
273 {
274 	return __page_val_to_pfn(pte_val(pte));
275 }
276 
277 #define pte_page(x)     pfn_to_page(pte_pfn(x))
278 
279 /* Constructs a page table entry */
280 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
281 {
282 	unsigned long prot_val = pgprot_val(prot);
283 
284 	ALT_THEAD_PMA(prot_val);
285 
286 	return __pte((pfn << _PAGE_PFN_SHIFT) | prot_val);
287 }
288 
289 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
290 
291 static inline int pte_present(pte_t pte)
292 {
293 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
294 }
295 
296 static inline int pte_none(pte_t pte)
297 {
298 	return (pte_val(pte) == 0);
299 }
300 
301 static inline int pte_write(pte_t pte)
302 {
303 	return pte_val(pte) & _PAGE_WRITE;
304 }
305 
306 static inline int pte_exec(pte_t pte)
307 {
308 	return pte_val(pte) & _PAGE_EXEC;
309 }
310 
311 static inline int pte_user(pte_t pte)
312 {
313 	return pte_val(pte) & _PAGE_USER;
314 }
315 
316 static inline int pte_huge(pte_t pte)
317 {
318 	return pte_present(pte) && (pte_val(pte) & _PAGE_LEAF);
319 }
320 
321 static inline int pte_dirty(pte_t pte)
322 {
323 	return pte_val(pte) & _PAGE_DIRTY;
324 }
325 
326 static inline int pte_young(pte_t pte)
327 {
328 	return pte_val(pte) & _PAGE_ACCESSED;
329 }
330 
331 static inline int pte_special(pte_t pte)
332 {
333 	return pte_val(pte) & _PAGE_SPECIAL;
334 }
335 
336 /* static inline pte_t pte_rdprotect(pte_t pte) */
337 
338 static inline pte_t pte_wrprotect(pte_t pte)
339 {
340 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
341 }
342 
343 /* static inline pte_t pte_mkread(pte_t pte) */
344 
345 static inline pte_t pte_mkwrite(pte_t pte)
346 {
347 	return __pte(pte_val(pte) | _PAGE_WRITE);
348 }
349 
350 /* static inline pte_t pte_mkexec(pte_t pte) */
351 
352 static inline pte_t pte_mkdirty(pte_t pte)
353 {
354 	return __pte(pte_val(pte) | _PAGE_DIRTY);
355 }
356 
357 static inline pte_t pte_mkclean(pte_t pte)
358 {
359 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
360 }
361 
362 static inline pte_t pte_mkyoung(pte_t pte)
363 {
364 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
365 }
366 
367 static inline pte_t pte_mkold(pte_t pte)
368 {
369 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
370 }
371 
372 static inline pte_t pte_mkspecial(pte_t pte)
373 {
374 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
375 }
376 
377 static inline pte_t pte_mkhuge(pte_t pte)
378 {
379 	return pte;
380 }
381 
382 #ifdef CONFIG_NUMA_BALANCING
383 /*
384  * See the comment in include/asm-generic/pgtable.h
385  */
386 static inline int pte_protnone(pte_t pte)
387 {
388 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)) == _PAGE_PROT_NONE;
389 }
390 
391 static inline int pmd_protnone(pmd_t pmd)
392 {
393 	return pte_protnone(pmd_pte(pmd));
394 }
395 #endif
396 
397 /* Modify page protection bits */
398 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
399 {
400 	unsigned long newprot_val = pgprot_val(newprot);
401 
402 	ALT_THEAD_PMA(newprot_val);
403 
404 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | newprot_val);
405 }
406 
407 #define pgd_ERROR(e) \
408 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
409 
410 
411 /* Commit new configuration to MMU hardware */
412 static inline void update_mmu_cache(struct vm_area_struct *vma,
413 	unsigned long address, pte_t *ptep)
414 {
415 	/*
416 	 * The kernel assumes that TLBs don't cache invalid entries, but
417 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
418 	 * cache flush; it is necessary even after writing invalid entries.
419 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
420 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
421 	 */
422 	local_flush_tlb_page(address);
423 }
424 
425 #define __HAVE_ARCH_UPDATE_MMU_TLB
426 #define update_mmu_tlb update_mmu_cache
427 
428 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
429 		unsigned long address, pmd_t *pmdp)
430 {
431 	pte_t *ptep = (pte_t *)pmdp;
432 
433 	update_mmu_cache(vma, address, ptep);
434 }
435 
436 #define __HAVE_ARCH_PTE_SAME
437 static inline int pte_same(pte_t pte_a, pte_t pte_b)
438 {
439 	return pte_val(pte_a) == pte_val(pte_b);
440 }
441 
442 /*
443  * Certain architectures need to do special things when PTEs within
444  * a page table are directly modified.  Thus, the following hook is
445  * made available.
446  */
447 static inline void set_pte(pte_t *ptep, pte_t pteval)
448 {
449 	*ptep = pteval;
450 }
451 
452 void flush_icache_pte(pte_t pte);
453 
454 static inline void __set_pte_at(struct mm_struct *mm,
455 	unsigned long addr, pte_t *ptep, pte_t pteval)
456 {
457 	if (pte_present(pteval) && pte_exec(pteval))
458 		flush_icache_pte(pteval);
459 
460 	set_pte(ptep, pteval);
461 }
462 
463 static inline void set_pte_at(struct mm_struct *mm,
464 	unsigned long addr, pte_t *ptep, pte_t pteval)
465 {
466 	page_table_check_pte_set(mm, addr, ptep, pteval);
467 	__set_pte_at(mm, addr, ptep, pteval);
468 }
469 
470 static inline void pte_clear(struct mm_struct *mm,
471 	unsigned long addr, pte_t *ptep)
472 {
473 	__set_pte_at(mm, addr, ptep, __pte(0));
474 }
475 
476 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
477 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
478 					unsigned long address, pte_t *ptep,
479 					pte_t entry, int dirty)
480 {
481 	if (!pte_same(*ptep, entry))
482 		set_pte_at(vma->vm_mm, address, ptep, entry);
483 	/*
484 	 * update_mmu_cache will unconditionally execute, handling both
485 	 * the case that the PTE changed and the spurious fault case.
486 	 */
487 	return true;
488 }
489 
490 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
491 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
492 				       unsigned long address, pte_t *ptep)
493 {
494 	pte_t pte = __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
495 
496 	page_table_check_pte_clear(mm, address, pte);
497 
498 	return pte;
499 }
500 
501 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
502 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
503 					    unsigned long address,
504 					    pte_t *ptep)
505 {
506 	if (!pte_young(*ptep))
507 		return 0;
508 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
509 }
510 
511 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
512 static inline void ptep_set_wrprotect(struct mm_struct *mm,
513 				      unsigned long address, pte_t *ptep)
514 {
515 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
516 }
517 
518 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
519 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
520 					 unsigned long address, pte_t *ptep)
521 {
522 	/*
523 	 * This comment is borrowed from x86, but applies equally to RISC-V:
524 	 *
525 	 * Clearing the accessed bit without a TLB flush
526 	 * doesn't cause data corruption. [ It could cause incorrect
527 	 * page aging and the (mistaken) reclaim of hot pages, but the
528 	 * chance of that should be relatively low. ]
529 	 *
530 	 * So as a performance optimization don't flush the TLB when
531 	 * clearing the accessed bit, it will eventually be flushed by
532 	 * a context switch or a VM operation anyway. [ In the rare
533 	 * event of it not getting flushed for a long time the delay
534 	 * shouldn't really matter because there's no real memory
535 	 * pressure for swapout to react to. ]
536 	 */
537 	return ptep_test_and_clear_young(vma, address, ptep);
538 }
539 
540 #define pgprot_noncached pgprot_noncached
541 static inline pgprot_t pgprot_noncached(pgprot_t _prot)
542 {
543 	unsigned long prot = pgprot_val(_prot);
544 
545 	prot &= ~_PAGE_MTMASK;
546 	prot |= _PAGE_IO;
547 
548 	return __pgprot(prot);
549 }
550 
551 #define pgprot_writecombine pgprot_writecombine
552 static inline pgprot_t pgprot_writecombine(pgprot_t _prot)
553 {
554 	unsigned long prot = pgprot_val(_prot);
555 
556 	prot &= ~_PAGE_MTMASK;
557 	prot |= _PAGE_NOCACHE;
558 
559 	return __pgprot(prot);
560 }
561 
562 /*
563  * THP functions
564  */
565 static inline pmd_t pte_pmd(pte_t pte)
566 {
567 	return __pmd(pte_val(pte));
568 }
569 
570 static inline pmd_t pmd_mkhuge(pmd_t pmd)
571 {
572 	return pmd;
573 }
574 
575 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
576 {
577 	return __pmd(pmd_val(pmd) & ~(_PAGE_PRESENT|_PAGE_PROT_NONE));
578 }
579 
580 #define __pmd_to_phys(pmd)  (__page_val_to_pfn(pmd_val(pmd)) << PAGE_SHIFT)
581 
582 static inline unsigned long pmd_pfn(pmd_t pmd)
583 {
584 	return ((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT);
585 }
586 
587 #define __pud_to_phys(pud)  (__page_val_to_pfn(pud_val(pud)) << PAGE_SHIFT)
588 
589 static inline unsigned long pud_pfn(pud_t pud)
590 {
591 	return ((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT);
592 }
593 
594 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
595 {
596 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
597 }
598 
599 #define pmd_write pmd_write
600 static inline int pmd_write(pmd_t pmd)
601 {
602 	return pte_write(pmd_pte(pmd));
603 }
604 
605 static inline int pmd_dirty(pmd_t pmd)
606 {
607 	return pte_dirty(pmd_pte(pmd));
608 }
609 
610 #define pmd_young pmd_young
611 static inline int pmd_young(pmd_t pmd)
612 {
613 	return pte_young(pmd_pte(pmd));
614 }
615 
616 static inline int pmd_user(pmd_t pmd)
617 {
618 	return pte_user(pmd_pte(pmd));
619 }
620 
621 static inline pmd_t pmd_mkold(pmd_t pmd)
622 {
623 	return pte_pmd(pte_mkold(pmd_pte(pmd)));
624 }
625 
626 static inline pmd_t pmd_mkyoung(pmd_t pmd)
627 {
628 	return pte_pmd(pte_mkyoung(pmd_pte(pmd)));
629 }
630 
631 static inline pmd_t pmd_mkwrite(pmd_t pmd)
632 {
633 	return pte_pmd(pte_mkwrite(pmd_pte(pmd)));
634 }
635 
636 static inline pmd_t pmd_wrprotect(pmd_t pmd)
637 {
638 	return pte_pmd(pte_wrprotect(pmd_pte(pmd)));
639 }
640 
641 static inline pmd_t pmd_mkclean(pmd_t pmd)
642 {
643 	return pte_pmd(pte_mkclean(pmd_pte(pmd)));
644 }
645 
646 static inline pmd_t pmd_mkdirty(pmd_t pmd)
647 {
648 	return pte_pmd(pte_mkdirty(pmd_pte(pmd)));
649 }
650 
651 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
652 				pmd_t *pmdp, pmd_t pmd)
653 {
654 	page_table_check_pmd_set(mm, addr, pmdp, pmd);
655 	return __set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd));
656 }
657 
658 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
659 				pud_t *pudp, pud_t pud)
660 {
661 	page_table_check_pud_set(mm, addr, pudp, pud);
662 	return __set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud));
663 }
664 
665 #ifdef CONFIG_PAGE_TABLE_CHECK
666 static inline bool pte_user_accessible_page(pte_t pte)
667 {
668 	return pte_present(pte) && pte_user(pte);
669 }
670 
671 static inline bool pmd_user_accessible_page(pmd_t pmd)
672 {
673 	return pmd_leaf(pmd) && pmd_user(pmd);
674 }
675 
676 static inline bool pud_user_accessible_page(pud_t pud)
677 {
678 	return pud_leaf(pud) && pud_user(pud);
679 }
680 #endif
681 
682 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
683 static inline int pmd_trans_huge(pmd_t pmd)
684 {
685 	return pmd_leaf(pmd);
686 }
687 
688 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
689 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
690 					unsigned long address, pmd_t *pmdp,
691 					pmd_t entry, int dirty)
692 {
693 	return ptep_set_access_flags(vma, address, (pte_t *)pmdp, pmd_pte(entry), dirty);
694 }
695 
696 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
697 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
698 					unsigned long address, pmd_t *pmdp)
699 {
700 	return ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
701 }
702 
703 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
704 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
705 					unsigned long address, pmd_t *pmdp)
706 {
707 	pmd_t pmd = __pmd(atomic_long_xchg((atomic_long_t *)pmdp, 0));
708 
709 	page_table_check_pmd_clear(mm, address, pmd);
710 
711 	return pmd;
712 }
713 
714 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
715 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
716 					unsigned long address, pmd_t *pmdp)
717 {
718 	ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
719 }
720 
721 #define pmdp_establish pmdp_establish
722 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
723 				unsigned long address, pmd_t *pmdp, pmd_t pmd)
724 {
725 	page_table_check_pmd_set(vma->vm_mm, address, pmdp, pmd);
726 	return __pmd(atomic_long_xchg((atomic_long_t *)pmdp, pmd_val(pmd)));
727 }
728 
729 #define pmdp_collapse_flush pmdp_collapse_flush
730 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
731 				 unsigned long address, pmd_t *pmdp);
732 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
733 
734 /*
735  * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
736  * are !pte_none() && !pte_present().
737  *
738  * Format of swap PTE:
739  *	bit            0:	_PAGE_PRESENT (zero)
740  *	bit       1 to 3:       _PAGE_LEAF (zero)
741  *	bit            5:	_PAGE_PROT_NONE (zero)
742  *	bit            6:	exclusive marker
743  *	bits      7 to 11:	swap type
744  *	bits 11 to XLEN-1:	swap offset
745  */
746 #define __SWP_TYPE_SHIFT	7
747 #define __SWP_TYPE_BITS		5
748 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
749 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
750 
751 #define MAX_SWAPFILES_CHECK()	\
752 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
753 
754 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
755 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
756 #define __swp_entry(type, offset) ((swp_entry_t) \
757 	{ (((type) & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT) | \
758 	  ((offset) << __SWP_OFFSET_SHIFT) })
759 
760 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
761 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
762 
763 static inline int pte_swp_exclusive(pte_t pte)
764 {
765 	return pte_val(pte) & _PAGE_SWP_EXCLUSIVE;
766 }
767 
768 static inline pte_t pte_swp_mkexclusive(pte_t pte)
769 {
770 	return __pte(pte_val(pte) | _PAGE_SWP_EXCLUSIVE);
771 }
772 
773 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
774 {
775 	return __pte(pte_val(pte) & ~_PAGE_SWP_EXCLUSIVE);
776 }
777 
778 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
779 #define __pmd_to_swp_entry(pmd) ((swp_entry_t) { pmd_val(pmd) })
780 #define __swp_entry_to_pmd(swp) __pmd((swp).val)
781 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
782 
783 /*
784  * In the RV64 Linux scheme, we give the user half of the virtual-address space
785  * and give the kernel the other (upper) half.
786  */
787 #ifdef CONFIG_64BIT
788 #define KERN_VIRT_START	(-(BIT(VA_BITS)) + TASK_SIZE)
789 #else
790 #define KERN_VIRT_START	FIXADDR_START
791 #endif
792 
793 /*
794  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
795  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
796  * Task size is:
797  * -     0x9fc00000 (~2.5GB) for RV32.
798  * -   0x4000000000 ( 256GB) for RV64 using SV39 mmu
799  * - 0x800000000000 ( 128TB) for RV64 using SV48 mmu
800  *
801  * Note that PGDIR_SIZE must evenly divide TASK_SIZE since "RISC-V
802  * Instruction Set Manual Volume II: Privileged Architecture" states that
803  * "load and store effective addresses, which are 64bits, must have bits
804  * 63–48 all equal to bit 47, or else a page-fault exception will occur."
805  */
806 #ifdef CONFIG_64BIT
807 #define TASK_SIZE_64	(PGDIR_SIZE * PTRS_PER_PGD / 2)
808 #define TASK_SIZE_MIN	(PGDIR_SIZE_L3 * PTRS_PER_PGD / 2)
809 
810 #ifdef CONFIG_COMPAT
811 #define TASK_SIZE_32	(_AC(0x80000000, UL) - PAGE_SIZE)
812 #define TASK_SIZE	(test_thread_flag(TIF_32BIT) ? \
813 			 TASK_SIZE_32 : TASK_SIZE_64)
814 #else
815 #define TASK_SIZE	TASK_SIZE_64
816 #endif
817 
818 #else
819 #define TASK_SIZE	FIXADDR_START
820 #define TASK_SIZE_MIN	TASK_SIZE
821 #endif
822 
823 #else /* CONFIG_MMU */
824 
825 #define PAGE_SHARED		__pgprot(0)
826 #define PAGE_KERNEL		__pgprot(0)
827 #define swapper_pg_dir		NULL
828 #define TASK_SIZE		0xffffffffUL
829 #define VMALLOC_START		0
830 #define VMALLOC_END		TASK_SIZE
831 
832 #endif /* !CONFIG_MMU */
833 
834 extern char _start[];
835 extern void *_dtb_early_va;
836 extern uintptr_t _dtb_early_pa;
837 #if defined(CONFIG_XIP_KERNEL) && defined(CONFIG_MMU)
838 #define dtb_early_va	(*(void **)XIP_FIXUP(&_dtb_early_va))
839 #define dtb_early_pa	(*(uintptr_t *)XIP_FIXUP(&_dtb_early_pa))
840 #else
841 #define dtb_early_va	_dtb_early_va
842 #define dtb_early_pa	_dtb_early_pa
843 #endif /* CONFIG_XIP_KERNEL */
844 extern u64 satp_mode;
845 extern bool pgtable_l4_enabled;
846 
847 void paging_init(void);
848 void misc_mem_init(void);
849 
850 /*
851  * ZERO_PAGE is a global shared page that is always zero,
852  * used for zero-mapped memory areas, etc.
853  */
854 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
855 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
856 
857 #endif /* !__ASSEMBLY__ */
858 
859 #endif /* _ASM_RISCV_PGTABLE_H */
860