xref: /linux/arch/riscv/include/asm/pgtable.h (revision 6bd33e1ece528f67646db33bf97406b747dafda0)
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 __ASSEMBLY__
15 
16 /* Page Upper Directory not used in RISC-V */
17 #include <asm-generic/pgtable-nopud.h>
18 #include <asm/page.h>
19 #include <asm/tlbflush.h>
20 #include <linux/mm_types.h>
21 
22 #ifdef CONFIG_64BIT
23 #include <asm/pgtable-64.h>
24 #else
25 #include <asm/pgtable-32.h>
26 #endif /* CONFIG_64BIT */
27 
28 #ifdef CONFIG_MMU
29 /* Number of entries in the page global directory */
30 #define PTRS_PER_PGD    (PAGE_SIZE / sizeof(pgd_t))
31 /* Number of entries in the page table */
32 #define PTRS_PER_PTE    (PAGE_SIZE / sizeof(pte_t))
33 
34 /* Number of PGD entries that a user-mode program can use */
35 #define USER_PTRS_PER_PGD   (TASK_SIZE / PGDIR_SIZE)
36 
37 /* Page protection bits */
38 #define _PAGE_BASE	(_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
39 
40 #define PAGE_NONE		__pgprot(_PAGE_PROT_NONE)
41 #define PAGE_READ		__pgprot(_PAGE_BASE | _PAGE_READ)
42 #define PAGE_WRITE		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE)
43 #define PAGE_EXEC		__pgprot(_PAGE_BASE | _PAGE_EXEC)
44 #define PAGE_READ_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
45 #define PAGE_WRITE_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ |	\
46 					 _PAGE_EXEC | _PAGE_WRITE)
47 
48 #define PAGE_COPY		PAGE_READ
49 #define PAGE_COPY_EXEC		PAGE_EXEC
50 #define PAGE_COPY_READ_EXEC	PAGE_READ_EXEC
51 #define PAGE_SHARED		PAGE_WRITE
52 #define PAGE_SHARED_EXEC	PAGE_WRITE_EXEC
53 
54 #define _PAGE_KERNEL		(_PAGE_READ \
55 				| _PAGE_WRITE \
56 				| _PAGE_PRESENT \
57 				| _PAGE_ACCESSED \
58 				| _PAGE_DIRTY)
59 
60 #define PAGE_KERNEL		__pgprot(_PAGE_KERNEL)
61 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
62 
63 #define PAGE_TABLE		__pgprot(_PAGE_TABLE)
64 
65 extern pgd_t swapper_pg_dir[];
66 
67 /* MAP_PRIVATE permissions: xwr (copy-on-write) */
68 #define __P000	PAGE_NONE
69 #define __P001	PAGE_READ
70 #define __P010	PAGE_COPY
71 #define __P011	PAGE_COPY
72 #define __P100	PAGE_EXEC
73 #define __P101	PAGE_READ_EXEC
74 #define __P110	PAGE_COPY_EXEC
75 #define __P111	PAGE_COPY_READ_EXEC
76 
77 /* MAP_SHARED permissions: xwr */
78 #define __S000	PAGE_NONE
79 #define __S001	PAGE_READ
80 #define __S010	PAGE_SHARED
81 #define __S011	PAGE_SHARED
82 #define __S100	PAGE_EXEC
83 #define __S101	PAGE_READ_EXEC
84 #define __S110	PAGE_SHARED_EXEC
85 #define __S111	PAGE_SHARED_EXEC
86 
87 static inline int pmd_present(pmd_t pmd)
88 {
89 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
90 }
91 
92 static inline int pmd_none(pmd_t pmd)
93 {
94 	return (pmd_val(pmd) == 0);
95 }
96 
97 static inline int pmd_bad(pmd_t pmd)
98 {
99 	return !pmd_present(pmd);
100 }
101 
102 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
103 {
104 	*pmdp = pmd;
105 }
106 
107 static inline void pmd_clear(pmd_t *pmdp)
108 {
109 	set_pmd(pmdp, __pmd(0));
110 }
111 
112 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
113 {
114 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
115 }
116 
117 static inline unsigned long _pgd_pfn(pgd_t pgd)
118 {
119 	return pgd_val(pgd) >> _PAGE_PFN_SHIFT;
120 }
121 
122 #define pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
123 
124 /* Locate an entry in the page global directory */
125 static inline pgd_t *pgd_offset(const struct mm_struct *mm, unsigned long addr)
126 {
127 	return mm->pgd + pgd_index(addr);
128 }
129 /* Locate an entry in the kernel page global directory */
130 #define pgd_offset_k(addr)      pgd_offset(&init_mm, (addr))
131 
132 static inline struct page *pmd_page(pmd_t pmd)
133 {
134 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
135 }
136 
137 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
138 {
139 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
140 }
141 
142 /* Yields the page frame number (PFN) of a page table entry */
143 static inline unsigned long pte_pfn(pte_t pte)
144 {
145 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
146 }
147 
148 #define pte_page(x)     pfn_to_page(pte_pfn(x))
149 
150 /* Constructs a page table entry */
151 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
152 {
153 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
154 }
155 
156 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
157 
158 #define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
159 
160 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long addr)
161 {
162 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(addr);
163 }
164 
165 #define pte_offset_map(dir, addr)	pte_offset_kernel((dir), (addr))
166 #define pte_unmap(pte)			((void)(pte))
167 
168 static inline int pte_present(pte_t pte)
169 {
170 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
171 }
172 
173 static inline int pte_none(pte_t pte)
174 {
175 	return (pte_val(pte) == 0);
176 }
177 
178 static inline int pte_write(pte_t pte)
179 {
180 	return pte_val(pte) & _PAGE_WRITE;
181 }
182 
183 static inline int pte_exec(pte_t pte)
184 {
185 	return pte_val(pte) & _PAGE_EXEC;
186 }
187 
188 static inline int pte_huge(pte_t pte)
189 {
190 	return pte_present(pte)
191 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
192 }
193 
194 static inline int pte_dirty(pte_t pte)
195 {
196 	return pte_val(pte) & _PAGE_DIRTY;
197 }
198 
199 static inline int pte_young(pte_t pte)
200 {
201 	return pte_val(pte) & _PAGE_ACCESSED;
202 }
203 
204 static inline int pte_special(pte_t pte)
205 {
206 	return pte_val(pte) & _PAGE_SPECIAL;
207 }
208 
209 /* static inline pte_t pte_rdprotect(pte_t pte) */
210 
211 static inline pte_t pte_wrprotect(pte_t pte)
212 {
213 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
214 }
215 
216 /* static inline pte_t pte_mkread(pte_t pte) */
217 
218 static inline pte_t pte_mkwrite(pte_t pte)
219 {
220 	return __pte(pte_val(pte) | _PAGE_WRITE);
221 }
222 
223 /* static inline pte_t pte_mkexec(pte_t pte) */
224 
225 static inline pte_t pte_mkdirty(pte_t pte)
226 {
227 	return __pte(pte_val(pte) | _PAGE_DIRTY);
228 }
229 
230 static inline pte_t pte_mkclean(pte_t pte)
231 {
232 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
233 }
234 
235 static inline pte_t pte_mkyoung(pte_t pte)
236 {
237 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
238 }
239 
240 static inline pte_t pte_mkold(pte_t pte)
241 {
242 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
243 }
244 
245 static inline pte_t pte_mkspecial(pte_t pte)
246 {
247 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
248 }
249 
250 static inline pte_t pte_mkhuge(pte_t pte)
251 {
252 	return pte;
253 }
254 
255 /* Modify page protection bits */
256 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
257 {
258 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
259 }
260 
261 #define pgd_ERROR(e) \
262 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
263 
264 
265 /* Commit new configuration to MMU hardware */
266 static inline void update_mmu_cache(struct vm_area_struct *vma,
267 	unsigned long address, pte_t *ptep)
268 {
269 	/*
270 	 * The kernel assumes that TLBs don't cache invalid entries, but
271 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
272 	 * cache flush; it is necessary even after writing invalid entries.
273 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
274 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
275 	 */
276 	local_flush_tlb_page(address);
277 }
278 
279 #define __HAVE_ARCH_PTE_SAME
280 static inline int pte_same(pte_t pte_a, pte_t pte_b)
281 {
282 	return pte_val(pte_a) == pte_val(pte_b);
283 }
284 
285 /*
286  * Certain architectures need to do special things when PTEs within
287  * a page table are directly modified.  Thus, the following hook is
288  * made available.
289  */
290 static inline void set_pte(pte_t *ptep, pte_t pteval)
291 {
292 	*ptep = pteval;
293 }
294 
295 void flush_icache_pte(pte_t pte);
296 
297 static inline void set_pte_at(struct mm_struct *mm,
298 	unsigned long addr, pte_t *ptep, pte_t pteval)
299 {
300 	if (pte_present(pteval) && pte_exec(pteval))
301 		flush_icache_pte(pteval);
302 
303 	set_pte(ptep, pteval);
304 }
305 
306 static inline void pte_clear(struct mm_struct *mm,
307 	unsigned long addr, pte_t *ptep)
308 {
309 	set_pte_at(mm, addr, ptep, __pte(0));
310 }
311 
312 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
313 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
314 					unsigned long address, pte_t *ptep,
315 					pte_t entry, int dirty)
316 {
317 	if (!pte_same(*ptep, entry))
318 		set_pte_at(vma->vm_mm, address, ptep, entry);
319 	/*
320 	 * update_mmu_cache will unconditionally execute, handling both
321 	 * the case that the PTE changed and the spurious fault case.
322 	 */
323 	return true;
324 }
325 
326 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
327 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
328 				       unsigned long address, pte_t *ptep)
329 {
330 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
331 }
332 
333 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
334 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
335 					    unsigned long address,
336 					    pte_t *ptep)
337 {
338 	if (!pte_young(*ptep))
339 		return 0;
340 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
341 }
342 
343 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
344 static inline void ptep_set_wrprotect(struct mm_struct *mm,
345 				      unsigned long address, pte_t *ptep)
346 {
347 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
348 }
349 
350 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
351 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
352 					 unsigned long address, pte_t *ptep)
353 {
354 	/*
355 	 * This comment is borrowed from x86, but applies equally to RISC-V:
356 	 *
357 	 * Clearing the accessed bit without a TLB flush
358 	 * doesn't cause data corruption. [ It could cause incorrect
359 	 * page aging and the (mistaken) reclaim of hot pages, but the
360 	 * chance of that should be relatively low. ]
361 	 *
362 	 * So as a performance optimization don't flush the TLB when
363 	 * clearing the accessed bit, it will eventually be flushed by
364 	 * a context switch or a VM operation anyway. [ In the rare
365 	 * event of it not getting flushed for a long time the delay
366 	 * shouldn't really matter because there's no real memory
367 	 * pressure for swapout to react to. ]
368 	 */
369 	return ptep_test_and_clear_young(vma, address, ptep);
370 }
371 
372 /*
373  * Encode and decode a swap entry
374  *
375  * Format of swap PTE:
376  *	bit            0:	_PAGE_PRESENT (zero)
377  *	bit            1:	_PAGE_PROT_NONE (zero)
378  *	bits      2 to 6:	swap type
379  *	bits 7 to XLEN-1:	swap offset
380  */
381 #define __SWP_TYPE_SHIFT	2
382 #define __SWP_TYPE_BITS		5
383 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
384 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
385 
386 #define MAX_SWAPFILES_CHECK()	\
387 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
388 
389 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
390 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
391 #define __swp_entry(type, offset) ((swp_entry_t) \
392 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
393 
394 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
395 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
396 
397 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
398 #define VMALLOC_END      (PAGE_OFFSET - 1)
399 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
400 
401 /*
402  * Roughly size the vmemmap space to be large enough to fit enough
403  * struct pages to map half the virtual address space. Then
404  * position vmemmap directly below the VMALLOC region.
405  */
406 #define VMEMMAP_SHIFT \
407 	(CONFIG_VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
408 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
409 #define VMEMMAP_END	(VMALLOC_START - 1)
410 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
411 
412 #define vmemmap		((struct page *)VMEMMAP_START)
413 
414 #define PCI_IO_SIZE      SZ_16M
415 #define PCI_IO_END       VMEMMAP_START
416 #define PCI_IO_START     (PCI_IO_END - PCI_IO_SIZE)
417 
418 #define FIXADDR_TOP      PCI_IO_START
419 #ifdef CONFIG_64BIT
420 #define FIXADDR_SIZE     PMD_SIZE
421 #else
422 #define FIXADDR_SIZE     PGDIR_SIZE
423 #endif
424 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
425 
426 /*
427  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
428  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
429  */
430 #ifdef CONFIG_64BIT
431 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
432 #else
433 #define TASK_SIZE FIXADDR_START
434 #endif
435 
436 #else /* CONFIG_MMU */
437 
438 #define PAGE_KERNEL		__pgprot(0)
439 #define swapper_pg_dir		NULL
440 #define VMALLOC_START		0
441 
442 #define TASK_SIZE 0xffffffffUL
443 
444 #endif /* !CONFIG_MMU */
445 
446 #define kern_addr_valid(addr)   (1) /* FIXME */
447 
448 extern void *dtb_early_va;
449 void setup_bootmem(void);
450 void paging_init(void);
451 
452 #define FIRST_USER_ADDRESS  0
453 
454 /*
455  * ZERO_PAGE is a global shared page that is always zero,
456  * used for zero-mapped memory areas, etc.
457  */
458 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
459 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
460 
461 #include <asm-generic/pgtable.h>
462 
463 #endif /* !__ASSEMBLY__ */
464 
465 #endif /* _ASM_RISCV_PGTABLE_H */
466