xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision e1cd7b80)
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 /*
66  * The RISC-V ISA doesn't yet specify how to query or modify PMAs, so we can't
67  * change the properties of memory regions.
68  */
69 #define _PAGE_IOREMAP _PAGE_KERNEL
70 
71 extern pgd_t swapper_pg_dir[];
72 
73 /* MAP_PRIVATE permissions: xwr (copy-on-write) */
74 #define __P000	PAGE_NONE
75 #define __P001	PAGE_READ
76 #define __P010	PAGE_COPY
77 #define __P011	PAGE_COPY
78 #define __P100	PAGE_EXEC
79 #define __P101	PAGE_READ_EXEC
80 #define __P110	PAGE_COPY_EXEC
81 #define __P111	PAGE_COPY_READ_EXEC
82 
83 /* MAP_SHARED permissions: xwr */
84 #define __S000	PAGE_NONE
85 #define __S001	PAGE_READ
86 #define __S010	PAGE_SHARED
87 #define __S011	PAGE_SHARED
88 #define __S100	PAGE_EXEC
89 #define __S101	PAGE_READ_EXEC
90 #define __S110	PAGE_SHARED_EXEC
91 #define __S111	PAGE_SHARED_EXEC
92 
93 static inline int pmd_present(pmd_t pmd)
94 {
95 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
96 }
97 
98 static inline int pmd_none(pmd_t pmd)
99 {
100 	return (pmd_val(pmd) == 0);
101 }
102 
103 static inline int pmd_bad(pmd_t pmd)
104 {
105 	return !pmd_present(pmd);
106 }
107 
108 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
109 {
110 	*pmdp = pmd;
111 }
112 
113 static inline void pmd_clear(pmd_t *pmdp)
114 {
115 	set_pmd(pmdp, __pmd(0));
116 }
117 
118 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
119 {
120 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
121 }
122 
123 static inline unsigned long _pgd_pfn(pgd_t pgd)
124 {
125 	return pgd_val(pgd) >> _PAGE_PFN_SHIFT;
126 }
127 
128 #define pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
129 
130 /* Locate an entry in the page global directory */
131 static inline pgd_t *pgd_offset(const struct mm_struct *mm, unsigned long addr)
132 {
133 	return mm->pgd + pgd_index(addr);
134 }
135 /* Locate an entry in the kernel page global directory */
136 #define pgd_offset_k(addr)      pgd_offset(&init_mm, (addr))
137 
138 static inline struct page *pmd_page(pmd_t pmd)
139 {
140 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
141 }
142 
143 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
144 {
145 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
146 }
147 
148 /* Yields the page frame number (PFN) of a page table entry */
149 static inline unsigned long pte_pfn(pte_t pte)
150 {
151 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
152 }
153 
154 #define pte_page(x)     pfn_to_page(pte_pfn(x))
155 
156 /* Constructs a page table entry */
157 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
158 {
159 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
160 }
161 
162 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
163 
164 #define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
165 
166 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long addr)
167 {
168 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(addr);
169 }
170 
171 #define pte_offset_map(dir, addr)	pte_offset_kernel((dir), (addr))
172 #define pte_unmap(pte)			((void)(pte))
173 
174 static inline int pte_present(pte_t pte)
175 {
176 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
177 }
178 
179 static inline int pte_none(pte_t pte)
180 {
181 	return (pte_val(pte) == 0);
182 }
183 
184 static inline int pte_write(pte_t pte)
185 {
186 	return pte_val(pte) & _PAGE_WRITE;
187 }
188 
189 static inline int pte_exec(pte_t pte)
190 {
191 	return pte_val(pte) & _PAGE_EXEC;
192 }
193 
194 static inline int pte_huge(pte_t pte)
195 {
196 	return pte_present(pte)
197 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
198 }
199 
200 static inline int pte_dirty(pte_t pte)
201 {
202 	return pte_val(pte) & _PAGE_DIRTY;
203 }
204 
205 static inline int pte_young(pte_t pte)
206 {
207 	return pte_val(pte) & _PAGE_ACCESSED;
208 }
209 
210 static inline int pte_special(pte_t pte)
211 {
212 	return pte_val(pte) & _PAGE_SPECIAL;
213 }
214 
215 /* static inline pte_t pte_rdprotect(pte_t pte) */
216 
217 static inline pte_t pte_wrprotect(pte_t pte)
218 {
219 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
220 }
221 
222 /* static inline pte_t pte_mkread(pte_t pte) */
223 
224 static inline pte_t pte_mkwrite(pte_t pte)
225 {
226 	return __pte(pte_val(pte) | _PAGE_WRITE);
227 }
228 
229 /* static inline pte_t pte_mkexec(pte_t pte) */
230 
231 static inline pte_t pte_mkdirty(pte_t pte)
232 {
233 	return __pte(pte_val(pte) | _PAGE_DIRTY);
234 }
235 
236 static inline pte_t pte_mkclean(pte_t pte)
237 {
238 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
239 }
240 
241 static inline pte_t pte_mkyoung(pte_t pte)
242 {
243 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
244 }
245 
246 static inline pte_t pte_mkold(pte_t pte)
247 {
248 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
249 }
250 
251 static inline pte_t pte_mkspecial(pte_t pte)
252 {
253 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
254 }
255 
256 static inline pte_t pte_mkhuge(pte_t pte)
257 {
258 	return pte;
259 }
260 
261 /* Modify page protection bits */
262 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
263 {
264 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
265 }
266 
267 #define pgd_ERROR(e) \
268 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
269 
270 
271 /* Commit new configuration to MMU hardware */
272 static inline void update_mmu_cache(struct vm_area_struct *vma,
273 	unsigned long address, pte_t *ptep)
274 {
275 	/*
276 	 * The kernel assumes that TLBs don't cache invalid entries, but
277 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
278 	 * cache flush; it is necessary even after writing invalid entries.
279 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
280 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
281 	 */
282 	local_flush_tlb_page(address);
283 }
284 
285 #define __HAVE_ARCH_PTE_SAME
286 static inline int pte_same(pte_t pte_a, pte_t pte_b)
287 {
288 	return pte_val(pte_a) == pte_val(pte_b);
289 }
290 
291 /*
292  * Certain architectures need to do special things when PTEs within
293  * a page table are directly modified.  Thus, the following hook is
294  * made available.
295  */
296 static inline void set_pte(pte_t *ptep, pte_t pteval)
297 {
298 	*ptep = pteval;
299 }
300 
301 void flush_icache_pte(pte_t pte);
302 
303 static inline void set_pte_at(struct mm_struct *mm,
304 	unsigned long addr, pte_t *ptep, pte_t pteval)
305 {
306 	if (pte_present(pteval) && pte_exec(pteval))
307 		flush_icache_pte(pteval);
308 
309 	set_pte(ptep, pteval);
310 }
311 
312 static inline void pte_clear(struct mm_struct *mm,
313 	unsigned long addr, pte_t *ptep)
314 {
315 	set_pte_at(mm, addr, ptep, __pte(0));
316 }
317 
318 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
319 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
320 					unsigned long address, pte_t *ptep,
321 					pte_t entry, int dirty)
322 {
323 	if (!pte_same(*ptep, entry))
324 		set_pte_at(vma->vm_mm, address, ptep, entry);
325 	/*
326 	 * update_mmu_cache will unconditionally execute, handling both
327 	 * the case that the PTE changed and the spurious fault case.
328 	 */
329 	return true;
330 }
331 
332 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
333 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
334 				       unsigned long address, pte_t *ptep)
335 {
336 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
337 }
338 
339 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
340 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
341 					    unsigned long address,
342 					    pte_t *ptep)
343 {
344 	if (!pte_young(*ptep))
345 		return 0;
346 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
347 }
348 
349 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
350 static inline void ptep_set_wrprotect(struct mm_struct *mm,
351 				      unsigned long address, pte_t *ptep)
352 {
353 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
354 }
355 
356 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
357 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
358 					 unsigned long address, pte_t *ptep)
359 {
360 	/*
361 	 * This comment is borrowed from x86, but applies equally to RISC-V:
362 	 *
363 	 * Clearing the accessed bit without a TLB flush
364 	 * doesn't cause data corruption. [ It could cause incorrect
365 	 * page aging and the (mistaken) reclaim of hot pages, but the
366 	 * chance of that should be relatively low. ]
367 	 *
368 	 * So as a performance optimization don't flush the TLB when
369 	 * clearing the accessed bit, it will eventually be flushed by
370 	 * a context switch or a VM operation anyway. [ In the rare
371 	 * event of it not getting flushed for a long time the delay
372 	 * shouldn't really matter because there's no real memory
373 	 * pressure for swapout to react to. ]
374 	 */
375 	return ptep_test_and_clear_young(vma, address, ptep);
376 }
377 
378 /*
379  * Encode and decode a swap entry
380  *
381  * Format of swap PTE:
382  *	bit            0:	_PAGE_PRESENT (zero)
383  *	bit            1:	_PAGE_PROT_NONE (zero)
384  *	bits      2 to 6:	swap type
385  *	bits 7 to XLEN-1:	swap offset
386  */
387 #define __SWP_TYPE_SHIFT	2
388 #define __SWP_TYPE_BITS		5
389 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
390 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
391 
392 #define MAX_SWAPFILES_CHECK()	\
393 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
394 
395 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
396 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
397 #define __swp_entry(type, offset) ((swp_entry_t) \
398 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
399 
400 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
401 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
402 
403 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
404 #define VMALLOC_END      (PAGE_OFFSET - 1)
405 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
406 
407 /*
408  * Roughly size the vmemmap space to be large enough to fit enough
409  * struct pages to map half the virtual address space. Then
410  * position vmemmap directly below the VMALLOC region.
411  */
412 #define VMEMMAP_SHIFT \
413 	(CONFIG_VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
414 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
415 #define VMEMMAP_END	(VMALLOC_START - 1)
416 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
417 
418 #define vmemmap		((struct page *)VMEMMAP_START)
419 
420 #define PCI_IO_SIZE      SZ_16M
421 #define PCI_IO_END       VMEMMAP_START
422 #define PCI_IO_START     (PCI_IO_END - PCI_IO_SIZE)
423 
424 #define FIXADDR_TOP      PCI_IO_START
425 #ifdef CONFIG_64BIT
426 #define FIXADDR_SIZE     PMD_SIZE
427 #else
428 #define FIXADDR_SIZE     PGDIR_SIZE
429 #endif
430 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
431 
432 /*
433  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
434  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
435  */
436 #ifdef CONFIG_64BIT
437 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
438 #else
439 #define TASK_SIZE FIXADDR_START
440 #endif
441 
442 #else /* CONFIG_MMU */
443 
444 #define PAGE_KERNEL		__pgprot(0)
445 #define swapper_pg_dir		NULL
446 #define VMALLOC_START		0
447 
448 #define TASK_SIZE 0xffffffffUL
449 
450 #endif /* !CONFIG_MMU */
451 
452 #define kern_addr_valid(addr)   (1) /* FIXME */
453 
454 extern void *dtb_early_va;
455 void setup_bootmem(void);
456 void paging_init(void);
457 
458 #define FIRST_USER_ADDRESS  0
459 
460 /*
461  * ZERO_PAGE is a global shared page that is always zero,
462  * used for zero-mapped memory areas, etc.
463  */
464 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
465 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
466 
467 #include <asm-generic/pgtable.h>
468 
469 #endif /* !__ASSEMBLY__ */
470 
471 #endif /* _ASM_RISCV_PGTABLE_H */
472