xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision 9e953cda5cdf1c230a3c9b7fc4d5e94f15885a9b)
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 
11 #include <asm/pgtable-bits.h>
12 
13 #ifndef __ASSEMBLY__
14 
15 /* Page Upper Directory not used in RISC-V */
16 #include <asm-generic/pgtable-nopud.h>
17 #include <asm/page.h>
18 #include <asm/tlbflush.h>
19 #include <linux/mm_types.h>
20 
21 #ifdef CONFIG_64BIT
22 #include <asm/pgtable-64.h>
23 #else
24 #include <asm/pgtable-32.h>
25 #endif /* CONFIG_64BIT */
26 
27 /* Number of entries in the page global directory */
28 #define PTRS_PER_PGD    (PAGE_SIZE / sizeof(pgd_t))
29 /* Number of entries in the page table */
30 #define PTRS_PER_PTE    (PAGE_SIZE / sizeof(pte_t))
31 
32 /* Number of PGD entries that a user-mode program can use */
33 #define USER_PTRS_PER_PGD   (TASK_SIZE / PGDIR_SIZE)
34 #define FIRST_USER_ADDRESS  0
35 
36 /* Page protection bits */
37 #define _PAGE_BASE	(_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
38 
39 #define PAGE_NONE		__pgprot(_PAGE_PROT_NONE)
40 #define PAGE_READ		__pgprot(_PAGE_BASE | _PAGE_READ)
41 #define PAGE_WRITE		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE)
42 #define PAGE_EXEC		__pgprot(_PAGE_BASE | _PAGE_EXEC)
43 #define PAGE_READ_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
44 #define PAGE_WRITE_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ |	\
45 					 _PAGE_EXEC | _PAGE_WRITE)
46 
47 #define PAGE_COPY		PAGE_READ
48 #define PAGE_COPY_EXEC		PAGE_EXEC
49 #define PAGE_COPY_READ_EXEC	PAGE_READ_EXEC
50 #define PAGE_SHARED		PAGE_WRITE
51 #define PAGE_SHARED_EXEC	PAGE_WRITE_EXEC
52 
53 #define _PAGE_KERNEL		(_PAGE_READ \
54 				| _PAGE_WRITE \
55 				| _PAGE_PRESENT \
56 				| _PAGE_ACCESSED \
57 				| _PAGE_DIRTY)
58 
59 #define PAGE_KERNEL		__pgprot(_PAGE_KERNEL)
60 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
61 
62 extern pgd_t swapper_pg_dir[];
63 
64 /* MAP_PRIVATE permissions: xwr (copy-on-write) */
65 #define __P000	PAGE_NONE
66 #define __P001	PAGE_READ
67 #define __P010	PAGE_COPY
68 #define __P011	PAGE_COPY
69 #define __P100	PAGE_EXEC
70 #define __P101	PAGE_READ_EXEC
71 #define __P110	PAGE_COPY_EXEC
72 #define __P111	PAGE_COPY_READ_EXEC
73 
74 /* MAP_SHARED permissions: xwr */
75 #define __S000	PAGE_NONE
76 #define __S001	PAGE_READ
77 #define __S010	PAGE_SHARED
78 #define __S011	PAGE_SHARED
79 #define __S100	PAGE_EXEC
80 #define __S101	PAGE_READ_EXEC
81 #define __S110	PAGE_SHARED_EXEC
82 #define __S111	PAGE_SHARED_EXEC
83 
84 /*
85  * ZERO_PAGE is a global shared page that is always zero,
86  * used for zero-mapped memory areas, etc.
87  */
88 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
89 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
90 
91 static inline int pmd_present(pmd_t pmd)
92 {
93 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
94 }
95 
96 static inline int pmd_none(pmd_t pmd)
97 {
98 	return (pmd_val(pmd) == 0);
99 }
100 
101 static inline int pmd_bad(pmd_t pmd)
102 {
103 	return !pmd_present(pmd);
104 }
105 
106 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
107 {
108 	*pmdp = pmd;
109 }
110 
111 static inline void pmd_clear(pmd_t *pmdp)
112 {
113 	set_pmd(pmdp, __pmd(0));
114 }
115 
116 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
117 {
118 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
119 }
120 
121 #define pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
122 
123 /* Locate an entry in the page global directory */
124 static inline pgd_t *pgd_offset(const struct mm_struct *mm, unsigned long addr)
125 {
126 	return mm->pgd + pgd_index(addr);
127 }
128 /* Locate an entry in the kernel page global directory */
129 #define pgd_offset_k(addr)      pgd_offset(&init_mm, (addr))
130 
131 static inline struct page *pmd_page(pmd_t pmd)
132 {
133 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
134 }
135 
136 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
137 {
138 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
139 }
140 
141 /* Yields the page frame number (PFN) of a page table entry */
142 static inline unsigned long pte_pfn(pte_t pte)
143 {
144 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
145 }
146 
147 #define pte_page(x)     pfn_to_page(pte_pfn(x))
148 
149 /* Constructs a page table entry */
150 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
151 {
152 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
153 }
154 
155 static inline pte_t mk_pte(struct page *page, pgprot_t prot)
156 {
157 	return pfn_pte(page_to_pfn(page), prot);
158 }
159 
160 #define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
161 
162 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long addr)
163 {
164 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(addr);
165 }
166 
167 #define pte_offset_map(dir, addr)	pte_offset_kernel((dir), (addr))
168 #define pte_unmap(pte)			((void)(pte))
169 
170 static inline int pte_present(pte_t pte)
171 {
172 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
173 }
174 
175 static inline int pte_none(pte_t pte)
176 {
177 	return (pte_val(pte) == 0);
178 }
179 
180 static inline int pte_write(pte_t pte)
181 {
182 	return pte_val(pte) & _PAGE_WRITE;
183 }
184 
185 static inline int pte_exec(pte_t pte)
186 {
187 	return pte_val(pte) & _PAGE_EXEC;
188 }
189 
190 static inline int pte_huge(pte_t pte)
191 {
192 	return pte_present(pte)
193 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
194 }
195 
196 static inline int pte_dirty(pte_t pte)
197 {
198 	return pte_val(pte) & _PAGE_DIRTY;
199 }
200 
201 static inline int pte_young(pte_t pte)
202 {
203 	return pte_val(pte) & _PAGE_ACCESSED;
204 }
205 
206 static inline int pte_special(pte_t pte)
207 {
208 	return pte_val(pte) & _PAGE_SPECIAL;
209 }
210 
211 /* static inline pte_t pte_rdprotect(pte_t pte) */
212 
213 static inline pte_t pte_wrprotect(pte_t pte)
214 {
215 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
216 }
217 
218 /* static inline pte_t pte_mkread(pte_t pte) */
219 
220 static inline pte_t pte_mkwrite(pte_t pte)
221 {
222 	return __pte(pte_val(pte) | _PAGE_WRITE);
223 }
224 
225 /* static inline pte_t pte_mkexec(pte_t pte) */
226 
227 static inline pte_t pte_mkdirty(pte_t pte)
228 {
229 	return __pte(pte_val(pte) | _PAGE_DIRTY);
230 }
231 
232 static inline pte_t pte_mkclean(pte_t pte)
233 {
234 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
235 }
236 
237 static inline pte_t pte_mkyoung(pte_t pte)
238 {
239 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
240 }
241 
242 static inline pte_t pte_mkold(pte_t pte)
243 {
244 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
245 }
246 
247 static inline pte_t pte_mkspecial(pte_t pte)
248 {
249 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
250 }
251 
252 static inline pte_t pte_mkhuge(pte_t pte)
253 {
254 	return pte;
255 }
256 
257 /* Modify page protection bits */
258 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
259 {
260 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
261 }
262 
263 #define pgd_ERROR(e) \
264 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
265 
266 
267 /* Commit new configuration to MMU hardware */
268 static inline void update_mmu_cache(struct vm_area_struct *vma,
269 	unsigned long address, pte_t *ptep)
270 {
271 	/*
272 	 * The kernel assumes that TLBs don't cache invalid entries, but
273 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
274 	 * cache flush; it is necessary even after writing invalid entries.
275 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
276 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
277 	 */
278 	local_flush_tlb_page(address);
279 }
280 
281 #define __HAVE_ARCH_PTE_SAME
282 static inline int pte_same(pte_t pte_a, pte_t pte_b)
283 {
284 	return pte_val(pte_a) == pte_val(pte_b);
285 }
286 
287 /*
288  * Certain architectures need to do special things when PTEs within
289  * a page table are directly modified.  Thus, the following hook is
290  * made available.
291  */
292 static inline void set_pte(pte_t *ptep, pte_t pteval)
293 {
294 	*ptep = pteval;
295 }
296 
297 void flush_icache_pte(pte_t pte);
298 
299 static inline void set_pte_at(struct mm_struct *mm,
300 	unsigned long addr, pte_t *ptep, pte_t pteval)
301 {
302 	if (pte_present(pteval) && pte_exec(pteval))
303 		flush_icache_pte(pteval);
304 
305 	set_pte(ptep, pteval);
306 }
307 
308 static inline void pte_clear(struct mm_struct *mm,
309 	unsigned long addr, pte_t *ptep)
310 {
311 	set_pte_at(mm, addr, ptep, __pte(0));
312 }
313 
314 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
315 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
316 					unsigned long address, pte_t *ptep,
317 					pte_t entry, int dirty)
318 {
319 	if (!pte_same(*ptep, entry))
320 		set_pte_at(vma->vm_mm, address, ptep, entry);
321 	/*
322 	 * update_mmu_cache will unconditionally execute, handling both
323 	 * the case that the PTE changed and the spurious fault case.
324 	 */
325 	return true;
326 }
327 
328 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
329 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
330 				       unsigned long address, pte_t *ptep)
331 {
332 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
333 }
334 
335 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
336 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
337 					    unsigned long address,
338 					    pte_t *ptep)
339 {
340 	if (!pte_young(*ptep))
341 		return 0;
342 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
343 }
344 
345 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
346 static inline void ptep_set_wrprotect(struct mm_struct *mm,
347 				      unsigned long address, pte_t *ptep)
348 {
349 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
350 }
351 
352 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
353 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
354 					 unsigned long address, pte_t *ptep)
355 {
356 	/*
357 	 * This comment is borrowed from x86, but applies equally to RISC-V:
358 	 *
359 	 * Clearing the accessed bit without a TLB flush
360 	 * doesn't cause data corruption. [ It could cause incorrect
361 	 * page aging and the (mistaken) reclaim of hot pages, but the
362 	 * chance of that should be relatively low. ]
363 	 *
364 	 * So as a performance optimization don't flush the TLB when
365 	 * clearing the accessed bit, it will eventually be flushed by
366 	 * a context switch or a VM operation anyway. [ In the rare
367 	 * event of it not getting flushed for a long time the delay
368 	 * shouldn't really matter because there's no real memory
369 	 * pressure for swapout to react to. ]
370 	 */
371 	return ptep_test_and_clear_young(vma, address, ptep);
372 }
373 
374 /*
375  * Encode and decode a swap entry
376  *
377  * Format of swap PTE:
378  *	bit            0:	_PAGE_PRESENT (zero)
379  *	bit            1:	_PAGE_PROT_NONE (zero)
380  *	bits      2 to 6:	swap type
381  *	bits 7 to XLEN-1:	swap offset
382  */
383 #define __SWP_TYPE_SHIFT	2
384 #define __SWP_TYPE_BITS		5
385 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
386 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
387 
388 #define MAX_SWAPFILES_CHECK()	\
389 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
390 
391 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
392 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
393 #define __swp_entry(type, offset) ((swp_entry_t) \
394 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
395 
396 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
397 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
398 
399 #ifdef CONFIG_FLATMEM
400 #define kern_addr_valid(addr)   (1) /* FIXME */
401 #endif
402 
403 extern void setup_bootmem(void);
404 extern void paging_init(void);
405 
406 static inline void pgtable_cache_init(void)
407 {
408 	/* No page table caches to initialize */
409 }
410 
411 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
412 #define VMALLOC_END      (PAGE_OFFSET - 1)
413 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
414 
415 /*
416  * Task size is 0x4000000000 for RV64 or 0xb800000 for RV32.
417  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
418  */
419 #ifdef CONFIG_64BIT
420 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
421 #else
422 #define TASK_SIZE VMALLOC_START
423 #endif
424 
425 #include <asm-generic/pgtable.h>
426 
427 #endif /* !__ASSEMBLY__ */
428 
429 #endif /* _ASM_RISCV_PGTABLE_H */
430