xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision 1b39eacd)
1 /*
2  * Copyright (C) 2012 Regents of the University of California
3  *
4  *   This program is free software; you can redistribute it and/or
5  *   modify it under the terms of the GNU General Public License
6  *   as published by the Free Software Foundation, version 2.
7  *
8  *   This program is distributed in the hope that it will be useful,
9  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
10  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  *   GNU General Public License for more details.
12  */
13 
14 #ifndef _ASM_RISCV_PGTABLE_H
15 #define _ASM_RISCV_PGTABLE_H
16 
17 #include <linux/mmzone.h>
18 
19 #include <asm/pgtable-bits.h>
20 
21 #ifndef __ASSEMBLY__
22 
23 #ifdef CONFIG_MMU
24 
25 /* Page Upper Directory not used in RISC-V */
26 #include <asm-generic/pgtable-nopud.h>
27 #include <asm/page.h>
28 #include <asm/tlbflush.h>
29 #include <linux/mm_types.h>
30 
31 #ifdef CONFIG_64BIT
32 #include <asm/pgtable-64.h>
33 #else
34 #include <asm/pgtable-32.h>
35 #endif /* CONFIG_64BIT */
36 
37 /* Number of entries in the page global directory */
38 #define PTRS_PER_PGD    (PAGE_SIZE / sizeof(pgd_t))
39 /* Number of entries in the page table */
40 #define PTRS_PER_PTE    (PAGE_SIZE / sizeof(pte_t))
41 
42 /* Number of PGD entries that a user-mode program can use */
43 #define USER_PTRS_PER_PGD   (TASK_SIZE / PGDIR_SIZE)
44 #define FIRST_USER_ADDRESS  0
45 
46 /* Page protection bits */
47 #define _PAGE_BASE	(_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
48 
49 #define PAGE_NONE		__pgprot(0)
50 #define PAGE_READ		__pgprot(_PAGE_BASE | _PAGE_READ)
51 #define PAGE_WRITE		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE)
52 #define PAGE_EXEC		__pgprot(_PAGE_BASE | _PAGE_EXEC)
53 #define PAGE_READ_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
54 #define PAGE_WRITE_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ |	\
55 					 _PAGE_EXEC | _PAGE_WRITE)
56 
57 #define PAGE_COPY		PAGE_READ
58 #define PAGE_COPY_EXEC		PAGE_EXEC
59 #define PAGE_COPY_READ_EXEC	PAGE_READ_EXEC
60 #define PAGE_SHARED		PAGE_WRITE
61 #define PAGE_SHARED_EXEC	PAGE_WRITE_EXEC
62 
63 #define _PAGE_KERNEL		(_PAGE_READ \
64 				| _PAGE_WRITE \
65 				| _PAGE_PRESENT \
66 				| _PAGE_ACCESSED \
67 				| _PAGE_DIRTY)
68 
69 #define PAGE_KERNEL		__pgprot(_PAGE_KERNEL)
70 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
71 
72 extern pgd_t swapper_pg_dir[];
73 
74 /* MAP_PRIVATE permissions: xwr (copy-on-write) */
75 #define __P000	PAGE_NONE
76 #define __P001	PAGE_READ
77 #define __P010	PAGE_COPY
78 #define __P011	PAGE_COPY
79 #define __P100	PAGE_EXEC
80 #define __P101	PAGE_READ_EXEC
81 #define __P110	PAGE_COPY_EXEC
82 #define __P111	PAGE_COPY_READ_EXEC
83 
84 /* MAP_SHARED permissions: xwr */
85 #define __S000	PAGE_NONE
86 #define __S001	PAGE_READ
87 #define __S010	PAGE_SHARED
88 #define __S011	PAGE_SHARED
89 #define __S100	PAGE_EXEC
90 #define __S101	PAGE_READ_EXEC
91 #define __S110	PAGE_SHARED_EXEC
92 #define __S111	PAGE_SHARED_EXEC
93 
94 /*
95  * ZERO_PAGE is a global shared page that is always zero,
96  * used for zero-mapped memory areas, etc.
97  */
98 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
99 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
100 
101 static inline int pmd_present(pmd_t pmd)
102 {
103 	return (pmd_val(pmd) & _PAGE_PRESENT);
104 }
105 
106 static inline int pmd_none(pmd_t pmd)
107 {
108 	return (pmd_val(pmd) == 0);
109 }
110 
111 static inline int pmd_bad(pmd_t pmd)
112 {
113 	return !pmd_present(pmd);
114 }
115 
116 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
117 {
118 	*pmdp = pmd;
119 }
120 
121 static inline void pmd_clear(pmd_t *pmdp)
122 {
123 	set_pmd(pmdp, __pmd(0));
124 }
125 
126 
127 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
128 {
129 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
130 }
131 
132 #define pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
133 
134 /* Locate an entry in the page global directory */
135 static inline pgd_t *pgd_offset(const struct mm_struct *mm, unsigned long addr)
136 {
137 	return mm->pgd + pgd_index(addr);
138 }
139 /* Locate an entry in the kernel page global directory */
140 #define pgd_offset_k(addr)      pgd_offset(&init_mm, (addr))
141 
142 static inline struct page *pmd_page(pmd_t pmd)
143 {
144 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
145 }
146 
147 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
148 {
149 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
150 }
151 
152 /* Yields the page frame number (PFN) of a page table entry */
153 static inline unsigned long pte_pfn(pte_t pte)
154 {
155 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
156 }
157 
158 #define pte_page(x)     pfn_to_page(pte_pfn(x))
159 
160 /* Constructs a page table entry */
161 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
162 {
163 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
164 }
165 
166 static inline pte_t mk_pte(struct page *page, pgprot_t prot)
167 {
168 	return pfn_pte(page_to_pfn(page), prot);
169 }
170 
171 #define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
172 
173 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long addr)
174 {
175 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(addr);
176 }
177 
178 #define pte_offset_map(dir, addr)	pte_offset_kernel((dir), (addr))
179 #define pte_unmap(pte)			((void)(pte))
180 
181 static inline int pte_present(pte_t pte)
182 {
183 	return (pte_val(pte) & _PAGE_PRESENT);
184 }
185 
186 static inline int pte_none(pte_t pte)
187 {
188 	return (pte_val(pte) == 0);
189 }
190 
191 static inline int pte_write(pte_t pte)
192 {
193 	return pte_val(pte) & _PAGE_WRITE;
194 }
195 
196 static inline int pte_exec(pte_t pte)
197 {
198 	return pte_val(pte) & _PAGE_EXEC;
199 }
200 
201 static inline int pte_huge(pte_t pte)
202 {
203 	return pte_present(pte)
204 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
205 }
206 
207 static inline int pte_dirty(pte_t pte)
208 {
209 	return pte_val(pte) & _PAGE_DIRTY;
210 }
211 
212 static inline int pte_young(pte_t pte)
213 {
214 	return pte_val(pte) & _PAGE_ACCESSED;
215 }
216 
217 static inline int pte_special(pte_t pte)
218 {
219 	return pte_val(pte) & _PAGE_SPECIAL;
220 }
221 
222 /* static inline pte_t pte_rdprotect(pte_t pte) */
223 
224 static inline pte_t pte_wrprotect(pte_t pte)
225 {
226 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
227 }
228 
229 /* static inline pte_t pte_mkread(pte_t pte) */
230 
231 static inline pte_t pte_mkwrite(pte_t pte)
232 {
233 	return __pte(pte_val(pte) | _PAGE_WRITE);
234 }
235 
236 /* static inline pte_t pte_mkexec(pte_t pte) */
237 
238 static inline pte_t pte_mkdirty(pte_t pte)
239 {
240 	return __pte(pte_val(pte) | _PAGE_DIRTY);
241 }
242 
243 static inline pte_t pte_mkclean(pte_t pte)
244 {
245 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
246 }
247 
248 static inline pte_t pte_mkyoung(pte_t pte)
249 {
250 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
251 }
252 
253 static inline pte_t pte_mkold(pte_t pte)
254 {
255 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
256 }
257 
258 static inline pte_t pte_mkspecial(pte_t pte)
259 {
260 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
261 }
262 
263 /* Modify page protection bits */
264 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
265 {
266 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
267 }
268 
269 #define pgd_ERROR(e) \
270 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
271 
272 
273 /* Commit new configuration to MMU hardware */
274 static inline void update_mmu_cache(struct vm_area_struct *vma,
275 	unsigned long address, pte_t *ptep)
276 {
277 	/*
278 	 * The kernel assumes that TLBs don't cache invalid entries, but
279 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
280 	 * cache flush; it is necessary even after writing invalid entries.
281 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
282 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
283 	 */
284 	local_flush_tlb_page(address);
285 }
286 
287 #define __HAVE_ARCH_PTE_SAME
288 static inline int pte_same(pte_t pte_a, pte_t pte_b)
289 {
290 	return pte_val(pte_a) == pte_val(pte_b);
291 }
292 
293 /*
294  * Certain architectures need to do special things when PTEs within
295  * a page table are directly modified.  Thus, the following hook is
296  * made available.
297  */
298 static inline void set_pte(pte_t *ptep, pte_t pteval)
299 {
300 	*ptep = pteval;
301 }
302 
303 void flush_icache_pte(pte_t pte);
304 
305 static inline void set_pte_at(struct mm_struct *mm,
306 	unsigned long addr, pte_t *ptep, pte_t pteval)
307 {
308 	if (pte_present(pteval) && pte_exec(pteval))
309 		flush_icache_pte(pteval);
310 
311 	set_pte(ptep, pteval);
312 }
313 
314 static inline void pte_clear(struct mm_struct *mm,
315 	unsigned long addr, pte_t *ptep)
316 {
317 	set_pte_at(mm, addr, ptep, __pte(0));
318 }
319 
320 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
321 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
322 					unsigned long address, pte_t *ptep,
323 					pte_t entry, int dirty)
324 {
325 	if (!pte_same(*ptep, entry))
326 		set_pte_at(vma->vm_mm, address, ptep, entry);
327 	/*
328 	 * update_mmu_cache will unconditionally execute, handling both
329 	 * the case that the PTE changed and the spurious fault case.
330 	 */
331 	return true;
332 }
333 
334 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
335 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
336 				       unsigned long address, pte_t *ptep)
337 {
338 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
339 }
340 
341 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
342 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
343 					    unsigned long address,
344 					    pte_t *ptep)
345 {
346 	if (!pte_young(*ptep))
347 		return 0;
348 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
349 }
350 
351 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
352 static inline void ptep_set_wrprotect(struct mm_struct *mm,
353 				      unsigned long address, pte_t *ptep)
354 {
355 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
356 }
357 
358 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
359 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
360 					 unsigned long address, pte_t *ptep)
361 {
362 	/*
363 	 * This comment is borrowed from x86, but applies equally to RISC-V:
364 	 *
365 	 * Clearing the accessed bit without a TLB flush
366 	 * doesn't cause data corruption. [ It could cause incorrect
367 	 * page aging and the (mistaken) reclaim of hot pages, but the
368 	 * chance of that should be relatively low. ]
369 	 *
370 	 * So as a performance optimization don't flush the TLB when
371 	 * clearing the accessed bit, it will eventually be flushed by
372 	 * a context switch or a VM operation anyway. [ In the rare
373 	 * event of it not getting flushed for a long time the delay
374 	 * shouldn't really matter because there's no real memory
375 	 * pressure for swapout to react to. ]
376 	 */
377 	return ptep_test_and_clear_young(vma, address, ptep);
378 }
379 
380 /*
381  * Encode and decode a swap entry
382  *
383  * Format of swap PTE:
384  *	bit            0:	_PAGE_PRESENT (zero)
385  *	bit            1:	reserved for future use (zero)
386  *	bits      2 to 6:	swap type
387  *	bits 7 to XLEN-1:	swap offset
388  */
389 #define __SWP_TYPE_SHIFT	2
390 #define __SWP_TYPE_BITS		5
391 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
392 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
393 
394 #define MAX_SWAPFILES_CHECK()	\
395 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
396 
397 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
398 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
399 #define __swp_entry(type, offset) ((swp_entry_t) \
400 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
401 
402 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
403 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
404 
405 #ifdef CONFIG_FLATMEM
406 #define kern_addr_valid(addr)   (1) /* FIXME */
407 #endif
408 
409 extern void paging_init(void);
410 
411 static inline void pgtable_cache_init(void)
412 {
413 	/* No page table caches to initialize */
414 }
415 
416 #endif /* CONFIG_MMU */
417 
418 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
419 #define VMALLOC_END      (PAGE_OFFSET - 1)
420 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
421 
422 /*
423  * Task size is 0x40000000000 for RV64 or 0xb800000 for RV32.
424  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
425  */
426 #ifdef CONFIG_64BIT
427 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
428 #else
429 #define TASK_SIZE VMALLOC_START
430 #endif
431 
432 #include <asm-generic/pgtable.h>
433 
434 #endif /* !__ASSEMBLY__ */
435 
436 #endif /* _ASM_RISCV_PGTABLE_H */
437