xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision ecd25094)
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 #define PAGE_TABLE		__pgprot(_PAGE_TABLE)
63 
64 extern pgd_t swapper_pg_dir[];
65 
66 /* MAP_PRIVATE permissions: xwr (copy-on-write) */
67 #define __P000	PAGE_NONE
68 #define __P001	PAGE_READ
69 #define __P010	PAGE_COPY
70 #define __P011	PAGE_COPY
71 #define __P100	PAGE_EXEC
72 #define __P101	PAGE_READ_EXEC
73 #define __P110	PAGE_COPY_EXEC
74 #define __P111	PAGE_COPY_READ_EXEC
75 
76 /* MAP_SHARED permissions: xwr */
77 #define __S000	PAGE_NONE
78 #define __S001	PAGE_READ
79 #define __S010	PAGE_SHARED
80 #define __S011	PAGE_SHARED
81 #define __S100	PAGE_EXEC
82 #define __S101	PAGE_READ_EXEC
83 #define __S110	PAGE_SHARED_EXEC
84 #define __S111	PAGE_SHARED_EXEC
85 
86 /*
87  * Roughly size the vmemmap space to be large enough to fit enough
88  * struct pages to map half the virtual address space. Then
89  * position vmemmap directly below the VMALLOC region.
90  */
91 #define VMEMMAP_SHIFT \
92 	(CONFIG_VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
93 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
94 #define VMEMMAP_END	(VMALLOC_START - 1)
95 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
96 
97 #define vmemmap		((struct page *)VMEMMAP_START)
98 
99 /*
100  * ZERO_PAGE is a global shared page that is always zero,
101  * used for zero-mapped memory areas, etc.
102  */
103 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
104 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
105 
106 static inline int pmd_present(pmd_t pmd)
107 {
108 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
109 }
110 
111 static inline int pmd_none(pmd_t pmd)
112 {
113 	return (pmd_val(pmd) == 0);
114 }
115 
116 static inline int pmd_bad(pmd_t pmd)
117 {
118 	return !pmd_present(pmd);
119 }
120 
121 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
122 {
123 	*pmdp = pmd;
124 }
125 
126 static inline void pmd_clear(pmd_t *pmdp)
127 {
128 	set_pmd(pmdp, __pmd(0));
129 }
130 
131 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
132 {
133 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
134 }
135 
136 static inline unsigned long _pgd_pfn(pgd_t pgd)
137 {
138 	return pgd_val(pgd) >> _PAGE_PFN_SHIFT;
139 }
140 
141 #define pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
142 
143 /* Locate an entry in the page global directory */
144 static inline pgd_t *pgd_offset(const struct mm_struct *mm, unsigned long addr)
145 {
146 	return mm->pgd + pgd_index(addr);
147 }
148 /* Locate an entry in the kernel page global directory */
149 #define pgd_offset_k(addr)      pgd_offset(&init_mm, (addr))
150 
151 static inline struct page *pmd_page(pmd_t pmd)
152 {
153 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
154 }
155 
156 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
157 {
158 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
159 }
160 
161 /* Yields the page frame number (PFN) of a page table entry */
162 static inline unsigned long pte_pfn(pte_t pte)
163 {
164 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
165 }
166 
167 #define pte_page(x)     pfn_to_page(pte_pfn(x))
168 
169 /* Constructs a page table entry */
170 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
171 {
172 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
173 }
174 
175 static inline pte_t mk_pte(struct page *page, pgprot_t prot)
176 {
177 	return pfn_pte(page_to_pfn(page), prot);
178 }
179 
180 #define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
181 
182 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long addr)
183 {
184 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(addr);
185 }
186 
187 #define pte_offset_map(dir, addr)	pte_offset_kernel((dir), (addr))
188 #define pte_unmap(pte)			((void)(pte))
189 
190 static inline int pte_present(pte_t pte)
191 {
192 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
193 }
194 
195 static inline int pte_none(pte_t pte)
196 {
197 	return (pte_val(pte) == 0);
198 }
199 
200 static inline int pte_write(pte_t pte)
201 {
202 	return pte_val(pte) & _PAGE_WRITE;
203 }
204 
205 static inline int pte_exec(pte_t pte)
206 {
207 	return pte_val(pte) & _PAGE_EXEC;
208 }
209 
210 static inline int pte_huge(pte_t pte)
211 {
212 	return pte_present(pte)
213 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
214 }
215 
216 static inline int pte_dirty(pte_t pte)
217 {
218 	return pte_val(pte) & _PAGE_DIRTY;
219 }
220 
221 static inline int pte_young(pte_t pte)
222 {
223 	return pte_val(pte) & _PAGE_ACCESSED;
224 }
225 
226 static inline int pte_special(pte_t pte)
227 {
228 	return pte_val(pte) & _PAGE_SPECIAL;
229 }
230 
231 /* static inline pte_t pte_rdprotect(pte_t pte) */
232 
233 static inline pte_t pte_wrprotect(pte_t pte)
234 {
235 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
236 }
237 
238 /* static inline pte_t pte_mkread(pte_t pte) */
239 
240 static inline pte_t pte_mkwrite(pte_t pte)
241 {
242 	return __pte(pte_val(pte) | _PAGE_WRITE);
243 }
244 
245 /* static inline pte_t pte_mkexec(pte_t pte) */
246 
247 static inline pte_t pte_mkdirty(pte_t pte)
248 {
249 	return __pte(pte_val(pte) | _PAGE_DIRTY);
250 }
251 
252 static inline pte_t pte_mkclean(pte_t pte)
253 {
254 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
255 }
256 
257 static inline pte_t pte_mkyoung(pte_t pte)
258 {
259 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
260 }
261 
262 static inline pte_t pte_mkold(pte_t pte)
263 {
264 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
265 }
266 
267 static inline pte_t pte_mkspecial(pte_t pte)
268 {
269 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
270 }
271 
272 static inline pte_t pte_mkhuge(pte_t pte)
273 {
274 	return pte;
275 }
276 
277 /* Modify page protection bits */
278 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
279 {
280 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
281 }
282 
283 #define pgd_ERROR(e) \
284 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
285 
286 
287 /* Commit new configuration to MMU hardware */
288 static inline void update_mmu_cache(struct vm_area_struct *vma,
289 	unsigned long address, pte_t *ptep)
290 {
291 	/*
292 	 * The kernel assumes that TLBs don't cache invalid entries, but
293 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
294 	 * cache flush; it is necessary even after writing invalid entries.
295 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
296 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
297 	 */
298 	local_flush_tlb_page(address);
299 }
300 
301 #define __HAVE_ARCH_PTE_SAME
302 static inline int pte_same(pte_t pte_a, pte_t pte_b)
303 {
304 	return pte_val(pte_a) == pte_val(pte_b);
305 }
306 
307 /*
308  * Certain architectures need to do special things when PTEs within
309  * a page table are directly modified.  Thus, the following hook is
310  * made available.
311  */
312 static inline void set_pte(pte_t *ptep, pte_t pteval)
313 {
314 	*ptep = pteval;
315 }
316 
317 void flush_icache_pte(pte_t pte);
318 
319 static inline void set_pte_at(struct mm_struct *mm,
320 	unsigned long addr, pte_t *ptep, pte_t pteval)
321 {
322 	if (pte_present(pteval) && pte_exec(pteval))
323 		flush_icache_pte(pteval);
324 
325 	set_pte(ptep, pteval);
326 }
327 
328 static inline void pte_clear(struct mm_struct *mm,
329 	unsigned long addr, pte_t *ptep)
330 {
331 	set_pte_at(mm, addr, ptep, __pte(0));
332 }
333 
334 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
335 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
336 					unsigned long address, pte_t *ptep,
337 					pte_t entry, int dirty)
338 {
339 	if (!pte_same(*ptep, entry))
340 		set_pte_at(vma->vm_mm, address, ptep, entry);
341 	/*
342 	 * update_mmu_cache will unconditionally execute, handling both
343 	 * the case that the PTE changed and the spurious fault case.
344 	 */
345 	return true;
346 }
347 
348 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
349 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
350 				       unsigned long address, pte_t *ptep)
351 {
352 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
353 }
354 
355 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
356 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
357 					    unsigned long address,
358 					    pte_t *ptep)
359 {
360 	if (!pte_young(*ptep))
361 		return 0;
362 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
363 }
364 
365 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
366 static inline void ptep_set_wrprotect(struct mm_struct *mm,
367 				      unsigned long address, pte_t *ptep)
368 {
369 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
370 }
371 
372 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
373 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
374 					 unsigned long address, pte_t *ptep)
375 {
376 	/*
377 	 * This comment is borrowed from x86, but applies equally to RISC-V:
378 	 *
379 	 * Clearing the accessed bit without a TLB flush
380 	 * doesn't cause data corruption. [ It could cause incorrect
381 	 * page aging and the (mistaken) reclaim of hot pages, but the
382 	 * chance of that should be relatively low. ]
383 	 *
384 	 * So as a performance optimization don't flush the TLB when
385 	 * clearing the accessed bit, it will eventually be flushed by
386 	 * a context switch or a VM operation anyway. [ In the rare
387 	 * event of it not getting flushed for a long time the delay
388 	 * shouldn't really matter because there's no real memory
389 	 * pressure for swapout to react to. ]
390 	 */
391 	return ptep_test_and_clear_young(vma, address, ptep);
392 }
393 
394 /*
395  * Encode and decode a swap entry
396  *
397  * Format of swap PTE:
398  *	bit            0:	_PAGE_PRESENT (zero)
399  *	bit            1:	_PAGE_PROT_NONE (zero)
400  *	bits      2 to 6:	swap type
401  *	bits 7 to XLEN-1:	swap offset
402  */
403 #define __SWP_TYPE_SHIFT	2
404 #define __SWP_TYPE_BITS		5
405 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
406 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
407 
408 #define MAX_SWAPFILES_CHECK()	\
409 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
410 
411 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
412 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
413 #define __swp_entry(type, offset) ((swp_entry_t) \
414 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
415 
416 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
417 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
418 
419 #ifdef CONFIG_FLATMEM
420 #define kern_addr_valid(addr)   (1) /* FIXME */
421 #endif
422 
423 extern void *dtb_early_va;
424 extern void setup_bootmem(void);
425 extern void paging_init(void);
426 
427 static inline void pgtable_cache_init(void)
428 {
429 	/* No page table caches to initialize */
430 }
431 
432 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
433 #define VMALLOC_END      (PAGE_OFFSET - 1)
434 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
435 
436 #define FIXADDR_TOP      VMALLOC_START
437 #ifdef CONFIG_64BIT
438 #define FIXADDR_SIZE     PMD_SIZE
439 #else
440 #define FIXADDR_SIZE     PGDIR_SIZE
441 #endif
442 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
443 
444 /*
445  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
446  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
447  */
448 #ifdef CONFIG_64BIT
449 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
450 #else
451 #define TASK_SIZE FIXADDR_START
452 #endif
453 
454 #include <asm-generic/pgtable.h>
455 
456 #endif /* !__ASSEMBLY__ */
457 
458 #endif /* _ASM_RISCV_PGTABLE_H */
459