xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision e7253313)
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 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
94 #define VMALLOC_END      (PAGE_OFFSET - 1)
95 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
96 
97 /*
98  * Roughly size the vmemmap space to be large enough to fit enough
99  * struct pages to map half the virtual address space. Then
100  * position vmemmap directly below the VMALLOC region.
101  */
102 #define VMEMMAP_SHIFT \
103 	(CONFIG_VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
104 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
105 #define VMEMMAP_END	(VMALLOC_START - 1)
106 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
107 
108 /*
109  * Define vmemmap for pfn_to_page & page_to_pfn calls. Needed if kernel
110  * is configured with CONFIG_SPARSEMEM_VMEMMAP enabled.
111  */
112 #define vmemmap		((struct page *)VMEMMAP_START)
113 
114 static inline int pmd_present(pmd_t pmd)
115 {
116 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
117 }
118 
119 static inline int pmd_none(pmd_t pmd)
120 {
121 	return (pmd_val(pmd) == 0);
122 }
123 
124 static inline int pmd_bad(pmd_t pmd)
125 {
126 	return !pmd_present(pmd);
127 }
128 
129 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
130 {
131 	*pmdp = pmd;
132 }
133 
134 static inline void pmd_clear(pmd_t *pmdp)
135 {
136 	set_pmd(pmdp, __pmd(0));
137 }
138 
139 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
140 {
141 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
142 }
143 
144 static inline unsigned long _pgd_pfn(pgd_t pgd)
145 {
146 	return pgd_val(pgd) >> _PAGE_PFN_SHIFT;
147 }
148 
149 #define pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
150 
151 /* Locate an entry in the page global directory */
152 static inline pgd_t *pgd_offset(const struct mm_struct *mm, unsigned long addr)
153 {
154 	return mm->pgd + pgd_index(addr);
155 }
156 /* Locate an entry in the kernel page global directory */
157 #define pgd_offset_k(addr)      pgd_offset(&init_mm, (addr))
158 
159 static inline struct page *pmd_page(pmd_t pmd)
160 {
161 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
162 }
163 
164 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
165 {
166 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
167 }
168 
169 /* Yields the page frame number (PFN) of a page table entry */
170 static inline unsigned long pte_pfn(pte_t pte)
171 {
172 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
173 }
174 
175 #define pte_page(x)     pfn_to_page(pte_pfn(x))
176 
177 /* Constructs a page table entry */
178 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
179 {
180 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
181 }
182 
183 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
184 
185 #define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
186 
187 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long addr)
188 {
189 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(addr);
190 }
191 
192 #define pte_offset_map(dir, addr)	pte_offset_kernel((dir), (addr))
193 #define pte_unmap(pte)			((void)(pte))
194 
195 static inline int pte_present(pte_t pte)
196 {
197 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
198 }
199 
200 static inline int pte_none(pte_t pte)
201 {
202 	return (pte_val(pte) == 0);
203 }
204 
205 static inline int pte_write(pte_t pte)
206 {
207 	return pte_val(pte) & _PAGE_WRITE;
208 }
209 
210 static inline int pte_exec(pte_t pte)
211 {
212 	return pte_val(pte) & _PAGE_EXEC;
213 }
214 
215 static inline int pte_huge(pte_t pte)
216 {
217 	return pte_present(pte)
218 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
219 }
220 
221 static inline int pte_dirty(pte_t pte)
222 {
223 	return pte_val(pte) & _PAGE_DIRTY;
224 }
225 
226 static inline int pte_young(pte_t pte)
227 {
228 	return pte_val(pte) & _PAGE_ACCESSED;
229 }
230 
231 static inline int pte_special(pte_t pte)
232 {
233 	return pte_val(pte) & _PAGE_SPECIAL;
234 }
235 
236 /* static inline pte_t pte_rdprotect(pte_t pte) */
237 
238 static inline pte_t pte_wrprotect(pte_t pte)
239 {
240 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
241 }
242 
243 /* static inline pte_t pte_mkread(pte_t pte) */
244 
245 static inline pte_t pte_mkwrite(pte_t pte)
246 {
247 	return __pte(pte_val(pte) | _PAGE_WRITE);
248 }
249 
250 /* static inline pte_t pte_mkexec(pte_t pte) */
251 
252 static inline pte_t pte_mkdirty(pte_t pte)
253 {
254 	return __pte(pte_val(pte) | _PAGE_DIRTY);
255 }
256 
257 static inline pte_t pte_mkclean(pte_t pte)
258 {
259 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
260 }
261 
262 static inline pte_t pte_mkyoung(pte_t pte)
263 {
264 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
265 }
266 
267 static inline pte_t pte_mkold(pte_t pte)
268 {
269 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
270 }
271 
272 static inline pte_t pte_mkspecial(pte_t pte)
273 {
274 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
275 }
276 
277 static inline pte_t pte_mkhuge(pte_t pte)
278 {
279 	return pte;
280 }
281 
282 /* Modify page protection bits */
283 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
284 {
285 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
286 }
287 
288 #define pgd_ERROR(e) \
289 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
290 
291 
292 /* Commit new configuration to MMU hardware */
293 static inline void update_mmu_cache(struct vm_area_struct *vma,
294 	unsigned long address, pte_t *ptep)
295 {
296 	/*
297 	 * The kernel assumes that TLBs don't cache invalid entries, but
298 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
299 	 * cache flush; it is necessary even after writing invalid entries.
300 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
301 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
302 	 */
303 	local_flush_tlb_page(address);
304 }
305 
306 #define __HAVE_ARCH_PTE_SAME
307 static inline int pte_same(pte_t pte_a, pte_t pte_b)
308 {
309 	return pte_val(pte_a) == pte_val(pte_b);
310 }
311 
312 /*
313  * Certain architectures need to do special things when PTEs within
314  * a page table are directly modified.  Thus, the following hook is
315  * made available.
316  */
317 static inline void set_pte(pte_t *ptep, pte_t pteval)
318 {
319 	*ptep = pteval;
320 }
321 
322 void flush_icache_pte(pte_t pte);
323 
324 static inline void set_pte_at(struct mm_struct *mm,
325 	unsigned long addr, pte_t *ptep, pte_t pteval)
326 {
327 	if (pte_present(pteval) && pte_exec(pteval))
328 		flush_icache_pte(pteval);
329 
330 	set_pte(ptep, pteval);
331 }
332 
333 static inline void pte_clear(struct mm_struct *mm,
334 	unsigned long addr, pte_t *ptep)
335 {
336 	set_pte_at(mm, addr, ptep, __pte(0));
337 }
338 
339 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
340 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
341 					unsigned long address, pte_t *ptep,
342 					pte_t entry, int dirty)
343 {
344 	if (!pte_same(*ptep, entry))
345 		set_pte_at(vma->vm_mm, address, ptep, entry);
346 	/*
347 	 * update_mmu_cache will unconditionally execute, handling both
348 	 * the case that the PTE changed and the spurious fault case.
349 	 */
350 	return true;
351 }
352 
353 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
354 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
355 				       unsigned long address, pte_t *ptep)
356 {
357 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
358 }
359 
360 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
361 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
362 					    unsigned long address,
363 					    pte_t *ptep)
364 {
365 	if (!pte_young(*ptep))
366 		return 0;
367 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
368 }
369 
370 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
371 static inline void ptep_set_wrprotect(struct mm_struct *mm,
372 				      unsigned long address, pte_t *ptep)
373 {
374 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
375 }
376 
377 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
378 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
379 					 unsigned long address, pte_t *ptep)
380 {
381 	/*
382 	 * This comment is borrowed from x86, but applies equally to RISC-V:
383 	 *
384 	 * Clearing the accessed bit without a TLB flush
385 	 * doesn't cause data corruption. [ It could cause incorrect
386 	 * page aging and the (mistaken) reclaim of hot pages, but the
387 	 * chance of that should be relatively low. ]
388 	 *
389 	 * So as a performance optimization don't flush the TLB when
390 	 * clearing the accessed bit, it will eventually be flushed by
391 	 * a context switch or a VM operation anyway. [ In the rare
392 	 * event of it not getting flushed for a long time the delay
393 	 * shouldn't really matter because there's no real memory
394 	 * pressure for swapout to react to. ]
395 	 */
396 	return ptep_test_and_clear_young(vma, address, ptep);
397 }
398 
399 /*
400  * Encode and decode a swap entry
401  *
402  * Format of swap PTE:
403  *	bit            0:	_PAGE_PRESENT (zero)
404  *	bit            1:	_PAGE_PROT_NONE (zero)
405  *	bits      2 to 6:	swap type
406  *	bits 7 to XLEN-1:	swap offset
407  */
408 #define __SWP_TYPE_SHIFT	2
409 #define __SWP_TYPE_BITS		5
410 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
411 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
412 
413 #define MAX_SWAPFILES_CHECK()	\
414 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
415 
416 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
417 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
418 #define __swp_entry(type, offset) ((swp_entry_t) \
419 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
420 
421 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
422 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
423 
424 #define PCI_IO_SIZE      SZ_16M
425 #define PCI_IO_END       VMEMMAP_START
426 #define PCI_IO_START     (PCI_IO_END - PCI_IO_SIZE)
427 
428 #define FIXADDR_TOP      PCI_IO_START
429 #ifdef CONFIG_64BIT
430 #define FIXADDR_SIZE     PMD_SIZE
431 #else
432 #define FIXADDR_SIZE     PGDIR_SIZE
433 #endif
434 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
435 
436 /*
437  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
438  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
439  */
440 #ifdef CONFIG_64BIT
441 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
442 #else
443 #define TASK_SIZE FIXADDR_START
444 #endif
445 
446 #else /* CONFIG_MMU */
447 
448 #define PAGE_KERNEL		__pgprot(0)
449 #define swapper_pg_dir		NULL
450 #define VMALLOC_START		0
451 
452 #define TASK_SIZE 0xffffffffUL
453 
454 #endif /* !CONFIG_MMU */
455 
456 #define kern_addr_valid(addr)   (1) /* FIXME */
457 
458 extern void *dtb_early_va;
459 void setup_bootmem(void);
460 void paging_init(void);
461 
462 #define FIRST_USER_ADDRESS  0
463 
464 /*
465  * ZERO_PAGE is a global shared page that is always zero,
466  * used for zero-mapped memory areas, etc.
467  */
468 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
469 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
470 
471 #include <asm-generic/pgtable.h>
472 
473 #endif /* !__ASSEMBLY__ */
474 
475 #endif /* _ASM_RISCV_PGTABLE_H */
476