xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision 20e2fc42)
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 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
87 #define VMALLOC_END      (PAGE_OFFSET - 1)
88 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
89 
90 /*
91  * Roughly size the vmemmap space to be large enough to fit enough
92  * struct pages to map half the virtual address space. Then
93  * position vmemmap directly below the VMALLOC region.
94  */
95 #define VMEMMAP_SHIFT \
96 	(CONFIG_VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
97 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
98 #define VMEMMAP_END	(VMALLOC_START - 1)
99 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
100 
101 #define vmemmap		((struct page *)VMEMMAP_START)
102 
103 #define FIXADDR_TOP      (VMEMMAP_START)
104 #ifdef CONFIG_64BIT
105 #define FIXADDR_SIZE     PMD_SIZE
106 #else
107 #define FIXADDR_SIZE     PGDIR_SIZE
108 #endif
109 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
110 
111 /*
112  * ZERO_PAGE is a global shared page that is always zero,
113  * used for zero-mapped memory areas, etc.
114  */
115 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
116 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
117 
118 static inline int pmd_present(pmd_t pmd)
119 {
120 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
121 }
122 
123 static inline int pmd_none(pmd_t pmd)
124 {
125 	return (pmd_val(pmd) == 0);
126 }
127 
128 static inline int pmd_bad(pmd_t pmd)
129 {
130 	return !pmd_present(pmd);
131 }
132 
133 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
134 {
135 	*pmdp = pmd;
136 }
137 
138 static inline void pmd_clear(pmd_t *pmdp)
139 {
140 	set_pmd(pmdp, __pmd(0));
141 }
142 
143 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
144 {
145 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
146 }
147 
148 static inline unsigned long _pgd_pfn(pgd_t pgd)
149 {
150 	return pgd_val(pgd) >> _PAGE_PFN_SHIFT;
151 }
152 
153 #define pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
154 
155 /* Locate an entry in the page global directory */
156 static inline pgd_t *pgd_offset(const struct mm_struct *mm, unsigned long addr)
157 {
158 	return mm->pgd + pgd_index(addr);
159 }
160 /* Locate an entry in the kernel page global directory */
161 #define pgd_offset_k(addr)      pgd_offset(&init_mm, (addr))
162 
163 static inline struct page *pmd_page(pmd_t pmd)
164 {
165 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
166 }
167 
168 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
169 {
170 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
171 }
172 
173 /* Yields the page frame number (PFN) of a page table entry */
174 static inline unsigned long pte_pfn(pte_t pte)
175 {
176 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
177 }
178 
179 #define pte_page(x)     pfn_to_page(pte_pfn(x))
180 
181 /* Constructs a page table entry */
182 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
183 {
184 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
185 }
186 
187 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
188 
189 #define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
190 
191 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long addr)
192 {
193 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(addr);
194 }
195 
196 #define pte_offset_map(dir, addr)	pte_offset_kernel((dir), (addr))
197 #define pte_unmap(pte)			((void)(pte))
198 
199 static inline int pte_present(pte_t pte)
200 {
201 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
202 }
203 
204 static inline int pte_none(pte_t pte)
205 {
206 	return (pte_val(pte) == 0);
207 }
208 
209 static inline int pte_write(pte_t pte)
210 {
211 	return pte_val(pte) & _PAGE_WRITE;
212 }
213 
214 static inline int pte_exec(pte_t pte)
215 {
216 	return pte_val(pte) & _PAGE_EXEC;
217 }
218 
219 static inline int pte_huge(pte_t pte)
220 {
221 	return pte_present(pte)
222 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
223 }
224 
225 static inline int pte_dirty(pte_t pte)
226 {
227 	return pte_val(pte) & _PAGE_DIRTY;
228 }
229 
230 static inline int pte_young(pte_t pte)
231 {
232 	return pte_val(pte) & _PAGE_ACCESSED;
233 }
234 
235 static inline int pte_special(pte_t pte)
236 {
237 	return pte_val(pte) & _PAGE_SPECIAL;
238 }
239 
240 /* static inline pte_t pte_rdprotect(pte_t pte) */
241 
242 static inline pte_t pte_wrprotect(pte_t pte)
243 {
244 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
245 }
246 
247 /* static inline pte_t pte_mkread(pte_t pte) */
248 
249 static inline pte_t pte_mkwrite(pte_t pte)
250 {
251 	return __pte(pte_val(pte) | _PAGE_WRITE);
252 }
253 
254 /* static inline pte_t pte_mkexec(pte_t pte) */
255 
256 static inline pte_t pte_mkdirty(pte_t pte)
257 {
258 	return __pte(pte_val(pte) | _PAGE_DIRTY);
259 }
260 
261 static inline pte_t pte_mkclean(pte_t pte)
262 {
263 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
264 }
265 
266 static inline pte_t pte_mkyoung(pte_t pte)
267 {
268 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
269 }
270 
271 static inline pte_t pte_mkold(pte_t pte)
272 {
273 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
274 }
275 
276 static inline pte_t pte_mkspecial(pte_t pte)
277 {
278 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
279 }
280 
281 static inline pte_t pte_mkhuge(pte_t pte)
282 {
283 	return pte;
284 }
285 
286 /* Modify page protection bits */
287 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
288 {
289 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
290 }
291 
292 #define pgd_ERROR(e) \
293 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
294 
295 
296 /* Commit new configuration to MMU hardware */
297 static inline void update_mmu_cache(struct vm_area_struct *vma,
298 	unsigned long address, pte_t *ptep)
299 {
300 	/*
301 	 * The kernel assumes that TLBs don't cache invalid entries, but
302 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
303 	 * cache flush; it is necessary even after writing invalid entries.
304 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
305 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
306 	 */
307 	local_flush_tlb_page(address);
308 }
309 
310 #define __HAVE_ARCH_PTE_SAME
311 static inline int pte_same(pte_t pte_a, pte_t pte_b)
312 {
313 	return pte_val(pte_a) == pte_val(pte_b);
314 }
315 
316 /*
317  * Certain architectures need to do special things when PTEs within
318  * a page table are directly modified.  Thus, the following hook is
319  * made available.
320  */
321 static inline void set_pte(pte_t *ptep, pte_t pteval)
322 {
323 	*ptep = pteval;
324 }
325 
326 void flush_icache_pte(pte_t pte);
327 
328 static inline void set_pte_at(struct mm_struct *mm,
329 	unsigned long addr, pte_t *ptep, pte_t pteval)
330 {
331 	if (pte_present(pteval) && pte_exec(pteval))
332 		flush_icache_pte(pteval);
333 
334 	set_pte(ptep, pteval);
335 }
336 
337 static inline void pte_clear(struct mm_struct *mm,
338 	unsigned long addr, pte_t *ptep)
339 {
340 	set_pte_at(mm, addr, ptep, __pte(0));
341 }
342 
343 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
344 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
345 					unsigned long address, pte_t *ptep,
346 					pte_t entry, int dirty)
347 {
348 	if (!pte_same(*ptep, entry))
349 		set_pte_at(vma->vm_mm, address, ptep, entry);
350 	/*
351 	 * update_mmu_cache will unconditionally execute, handling both
352 	 * the case that the PTE changed and the spurious fault case.
353 	 */
354 	return true;
355 }
356 
357 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
358 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
359 				       unsigned long address, pte_t *ptep)
360 {
361 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
362 }
363 
364 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
365 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
366 					    unsigned long address,
367 					    pte_t *ptep)
368 {
369 	if (!pte_young(*ptep))
370 		return 0;
371 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
372 }
373 
374 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
375 static inline void ptep_set_wrprotect(struct mm_struct *mm,
376 				      unsigned long address, pte_t *ptep)
377 {
378 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
379 }
380 
381 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
382 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
383 					 unsigned long address, pte_t *ptep)
384 {
385 	/*
386 	 * This comment is borrowed from x86, but applies equally to RISC-V:
387 	 *
388 	 * Clearing the accessed bit without a TLB flush
389 	 * doesn't cause data corruption. [ It could cause incorrect
390 	 * page aging and the (mistaken) reclaim of hot pages, but the
391 	 * chance of that should be relatively low. ]
392 	 *
393 	 * So as a performance optimization don't flush the TLB when
394 	 * clearing the accessed bit, it will eventually be flushed by
395 	 * a context switch or a VM operation anyway. [ In the rare
396 	 * event of it not getting flushed for a long time the delay
397 	 * shouldn't really matter because there's no real memory
398 	 * pressure for swapout to react to. ]
399 	 */
400 	return ptep_test_and_clear_young(vma, address, ptep);
401 }
402 
403 /*
404  * Encode and decode a swap entry
405  *
406  * Format of swap PTE:
407  *	bit            0:	_PAGE_PRESENT (zero)
408  *	bit            1:	_PAGE_PROT_NONE (zero)
409  *	bits      2 to 6:	swap type
410  *	bits 7 to XLEN-1:	swap offset
411  */
412 #define __SWP_TYPE_SHIFT	2
413 #define __SWP_TYPE_BITS		5
414 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
415 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
416 
417 #define MAX_SWAPFILES_CHECK()	\
418 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
419 
420 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
421 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
422 #define __swp_entry(type, offset) ((swp_entry_t) \
423 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
424 
425 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
426 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
427 
428 #define kern_addr_valid(addr)   (1) /* FIXME */
429 
430 extern void *dtb_early_va;
431 extern void setup_bootmem(void);
432 extern void paging_init(void);
433 
434 /*
435  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
436  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
437  */
438 #ifdef CONFIG_64BIT
439 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
440 #else
441 #define TASK_SIZE FIXADDR_START
442 #endif
443 
444 #include <asm-generic/pgtable.h>
445 
446 #endif /* !__ASSEMBLY__ */
447 
448 #endif /* _ASM_RISCV_PGTABLE_H */
449