xref: /linux/arch/riscv/include/asm/pgtable.h (revision b91540d52a08b65eb6a2b09132e1bd54fa82754c)
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_MMU
23 
24 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
25 #define VMALLOC_END      (PAGE_OFFSET - 1)
26 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
27 
28 #define BPF_JIT_REGION_SIZE	(SZ_128M)
29 #define BPF_JIT_REGION_START	(PAGE_OFFSET - BPF_JIT_REGION_SIZE)
30 #define BPF_JIT_REGION_END	(VMALLOC_END)
31 
32 /*
33  * Roughly size the vmemmap space to be large enough to fit enough
34  * struct pages to map half the virtual address space. Then
35  * position vmemmap directly below the VMALLOC region.
36  */
37 #define VMEMMAP_SHIFT \
38 	(CONFIG_VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
39 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
40 #define VMEMMAP_END	(VMALLOC_START - 1)
41 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
42 
43 /*
44  * Define vmemmap for pfn_to_page & page_to_pfn calls. Needed if kernel
45  * is configured with CONFIG_SPARSEMEM_VMEMMAP enabled.
46  */
47 #define vmemmap		((struct page *)VMEMMAP_START)
48 
49 #define PCI_IO_SIZE      SZ_16M
50 #define PCI_IO_END       VMEMMAP_START
51 #define PCI_IO_START     (PCI_IO_END - PCI_IO_SIZE)
52 
53 #define FIXADDR_TOP      PCI_IO_START
54 #ifdef CONFIG_64BIT
55 #define FIXADDR_SIZE     PMD_SIZE
56 #else
57 #define FIXADDR_SIZE     PGDIR_SIZE
58 #endif
59 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
60 
61 #endif
62 
63 #ifdef CONFIG_64BIT
64 #include <asm/pgtable-64.h>
65 #else
66 #include <asm/pgtable-32.h>
67 #endif /* CONFIG_64BIT */
68 
69 #ifdef CONFIG_MMU
70 /* Number of entries in the page global directory */
71 #define PTRS_PER_PGD    (PAGE_SIZE / sizeof(pgd_t))
72 /* Number of entries in the page table */
73 #define PTRS_PER_PTE    (PAGE_SIZE / sizeof(pte_t))
74 
75 /* Number of PGD entries that a user-mode program can use */
76 #define USER_PTRS_PER_PGD   (TASK_SIZE / PGDIR_SIZE)
77 
78 /* Page protection bits */
79 #define _PAGE_BASE	(_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
80 
81 #define PAGE_NONE		__pgprot(_PAGE_PROT_NONE)
82 #define PAGE_READ		__pgprot(_PAGE_BASE | _PAGE_READ)
83 #define PAGE_WRITE		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE)
84 #define PAGE_EXEC		__pgprot(_PAGE_BASE | _PAGE_EXEC)
85 #define PAGE_READ_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
86 #define PAGE_WRITE_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ |	\
87 					 _PAGE_EXEC | _PAGE_WRITE)
88 
89 #define PAGE_COPY		PAGE_READ
90 #define PAGE_COPY_EXEC		PAGE_EXEC
91 #define PAGE_COPY_READ_EXEC	PAGE_READ_EXEC
92 #define PAGE_SHARED		PAGE_WRITE
93 #define PAGE_SHARED_EXEC	PAGE_WRITE_EXEC
94 
95 #define _PAGE_KERNEL		(_PAGE_READ \
96 				| _PAGE_WRITE \
97 				| _PAGE_PRESENT \
98 				| _PAGE_ACCESSED \
99 				| _PAGE_DIRTY)
100 
101 #define PAGE_KERNEL		__pgprot(_PAGE_KERNEL)
102 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
103 #define PAGE_KERNEL_READ	__pgprot(_PAGE_KERNEL & ~_PAGE_WRITE)
104 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
105 #define PAGE_KERNEL_READ_EXEC	__pgprot((_PAGE_KERNEL & ~_PAGE_WRITE) \
106 					 | _PAGE_EXEC)
107 
108 #define PAGE_TABLE		__pgprot(_PAGE_TABLE)
109 
110 /*
111  * The RISC-V ISA doesn't yet specify how to query or modify PMAs, so we can't
112  * change the properties of memory regions.
113  */
114 #define _PAGE_IOREMAP _PAGE_KERNEL
115 
116 extern pgd_t swapper_pg_dir[];
117 
118 /* MAP_PRIVATE permissions: xwr (copy-on-write) */
119 #define __P000	PAGE_NONE
120 #define __P001	PAGE_READ
121 #define __P010	PAGE_COPY
122 #define __P011	PAGE_COPY
123 #define __P100	PAGE_EXEC
124 #define __P101	PAGE_READ_EXEC
125 #define __P110	PAGE_COPY_EXEC
126 #define __P111	PAGE_COPY_READ_EXEC
127 
128 /* MAP_SHARED permissions: xwr */
129 #define __S000	PAGE_NONE
130 #define __S001	PAGE_READ
131 #define __S010	PAGE_SHARED
132 #define __S011	PAGE_SHARED
133 #define __S100	PAGE_EXEC
134 #define __S101	PAGE_READ_EXEC
135 #define __S110	PAGE_SHARED_EXEC
136 #define __S111	PAGE_SHARED_EXEC
137 
138 static inline int pmd_present(pmd_t pmd)
139 {
140 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
141 }
142 
143 static inline int pmd_none(pmd_t pmd)
144 {
145 	return (pmd_val(pmd) == 0);
146 }
147 
148 static inline int pmd_bad(pmd_t pmd)
149 {
150 	return !pmd_present(pmd);
151 }
152 
153 #define pmd_leaf	pmd_leaf
154 static inline int pmd_leaf(pmd_t pmd)
155 {
156 	return pmd_present(pmd) &&
157 	       (pmd_val(pmd) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
158 }
159 
160 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
161 {
162 	*pmdp = pmd;
163 }
164 
165 static inline void pmd_clear(pmd_t *pmdp)
166 {
167 	set_pmd(pmdp, __pmd(0));
168 }
169 
170 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
171 {
172 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
173 }
174 
175 static inline unsigned long _pgd_pfn(pgd_t pgd)
176 {
177 	return pgd_val(pgd) >> _PAGE_PFN_SHIFT;
178 }
179 
180 static inline struct page *pmd_page(pmd_t pmd)
181 {
182 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
183 }
184 
185 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
186 {
187 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
188 }
189 
190 /* Yields the page frame number (PFN) of a page table entry */
191 static inline unsigned long pte_pfn(pte_t pte)
192 {
193 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
194 }
195 
196 #define pte_page(x)     pfn_to_page(pte_pfn(x))
197 
198 /* Constructs a page table entry */
199 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
200 {
201 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
202 }
203 
204 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
205 
206 static inline int pte_present(pte_t pte)
207 {
208 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
209 }
210 
211 static inline int pte_none(pte_t pte)
212 {
213 	return (pte_val(pte) == 0);
214 }
215 
216 static inline int pte_write(pte_t pte)
217 {
218 	return pte_val(pte) & _PAGE_WRITE;
219 }
220 
221 static inline int pte_exec(pte_t pte)
222 {
223 	return pte_val(pte) & _PAGE_EXEC;
224 }
225 
226 static inline int pte_huge(pte_t pte)
227 {
228 	return pte_present(pte)
229 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
230 }
231 
232 static inline int pte_dirty(pte_t pte)
233 {
234 	return pte_val(pte) & _PAGE_DIRTY;
235 }
236 
237 static inline int pte_young(pte_t pte)
238 {
239 	return pte_val(pte) & _PAGE_ACCESSED;
240 }
241 
242 static inline int pte_special(pte_t pte)
243 {
244 	return pte_val(pte) & _PAGE_SPECIAL;
245 }
246 
247 /* static inline pte_t pte_rdprotect(pte_t pte) */
248 
249 static inline pte_t pte_wrprotect(pte_t pte)
250 {
251 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
252 }
253 
254 /* static inline pte_t pte_mkread(pte_t pte) */
255 
256 static inline pte_t pte_mkwrite(pte_t pte)
257 {
258 	return __pte(pte_val(pte) | _PAGE_WRITE);
259 }
260 
261 /* static inline pte_t pte_mkexec(pte_t pte) */
262 
263 static inline pte_t pte_mkdirty(pte_t pte)
264 {
265 	return __pte(pte_val(pte) | _PAGE_DIRTY);
266 }
267 
268 static inline pte_t pte_mkclean(pte_t pte)
269 {
270 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
271 }
272 
273 static inline pte_t pte_mkyoung(pte_t pte)
274 {
275 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
276 }
277 
278 static inline pte_t pte_mkold(pte_t pte)
279 {
280 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
281 }
282 
283 static inline pte_t pte_mkspecial(pte_t pte)
284 {
285 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
286 }
287 
288 static inline pte_t pte_mkhuge(pte_t pte)
289 {
290 	return pte;
291 }
292 
293 /* Modify page protection bits */
294 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
295 {
296 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
297 }
298 
299 #define pgd_ERROR(e) \
300 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
301 
302 
303 /* Commit new configuration to MMU hardware */
304 static inline void update_mmu_cache(struct vm_area_struct *vma,
305 	unsigned long address, pte_t *ptep)
306 {
307 	/*
308 	 * The kernel assumes that TLBs don't cache invalid entries, but
309 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
310 	 * cache flush; it is necessary even after writing invalid entries.
311 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
312 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
313 	 */
314 	local_flush_tlb_page(address);
315 }
316 
317 #define __HAVE_ARCH_PTE_SAME
318 static inline int pte_same(pte_t pte_a, pte_t pte_b)
319 {
320 	return pte_val(pte_a) == pte_val(pte_b);
321 }
322 
323 /*
324  * Certain architectures need to do special things when PTEs within
325  * a page table are directly modified.  Thus, the following hook is
326  * made available.
327  */
328 static inline void set_pte(pte_t *ptep, pte_t pteval)
329 {
330 	*ptep = pteval;
331 }
332 
333 void flush_icache_pte(pte_t pte);
334 
335 static inline void set_pte_at(struct mm_struct *mm,
336 	unsigned long addr, pte_t *ptep, pte_t pteval)
337 {
338 	if (pte_present(pteval) && pte_exec(pteval))
339 		flush_icache_pte(pteval);
340 
341 	set_pte(ptep, pteval);
342 }
343 
344 static inline void pte_clear(struct mm_struct *mm,
345 	unsigned long addr, pte_t *ptep)
346 {
347 	set_pte_at(mm, addr, ptep, __pte(0));
348 }
349 
350 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
351 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
352 					unsigned long address, pte_t *ptep,
353 					pte_t entry, int dirty)
354 {
355 	if (!pte_same(*ptep, entry))
356 		set_pte_at(vma->vm_mm, address, ptep, entry);
357 	/*
358 	 * update_mmu_cache will unconditionally execute, handling both
359 	 * the case that the PTE changed and the spurious fault case.
360 	 */
361 	return true;
362 }
363 
364 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
365 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
366 				       unsigned long address, pte_t *ptep)
367 {
368 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
369 }
370 
371 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
372 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
373 					    unsigned long address,
374 					    pte_t *ptep)
375 {
376 	if (!pte_young(*ptep))
377 		return 0;
378 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
379 }
380 
381 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
382 static inline void ptep_set_wrprotect(struct mm_struct *mm,
383 				      unsigned long address, pte_t *ptep)
384 {
385 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
386 }
387 
388 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
389 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
390 					 unsigned long address, pte_t *ptep)
391 {
392 	/*
393 	 * This comment is borrowed from x86, but applies equally to RISC-V:
394 	 *
395 	 * Clearing the accessed bit without a TLB flush
396 	 * doesn't cause data corruption. [ It could cause incorrect
397 	 * page aging and the (mistaken) reclaim of hot pages, but the
398 	 * chance of that should be relatively low. ]
399 	 *
400 	 * So as a performance optimization don't flush the TLB when
401 	 * clearing the accessed bit, it will eventually be flushed by
402 	 * a context switch or a VM operation anyway. [ In the rare
403 	 * event of it not getting flushed for a long time the delay
404 	 * shouldn't really matter because there's no real memory
405 	 * pressure for swapout to react to. ]
406 	 */
407 	return ptep_test_and_clear_young(vma, address, ptep);
408 }
409 
410 /*
411  * Encode and decode a swap entry
412  *
413  * Format of swap PTE:
414  *	bit            0:	_PAGE_PRESENT (zero)
415  *	bit            1:	_PAGE_PROT_NONE (zero)
416  *	bits      2 to 6:	swap type
417  *	bits 7 to XLEN-1:	swap offset
418  */
419 #define __SWP_TYPE_SHIFT	2
420 #define __SWP_TYPE_BITS		5
421 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
422 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
423 
424 #define MAX_SWAPFILES_CHECK()	\
425 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
426 
427 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
428 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
429 #define __swp_entry(type, offset) ((swp_entry_t) \
430 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
431 
432 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
433 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
434 
435 /*
436  * In the RV64 Linux scheme, we give the user half of the virtual-address space
437  * and give the kernel the other (upper) half.
438  */
439 #ifdef CONFIG_64BIT
440 #define KERN_VIRT_START	(-(BIT(CONFIG_VA_BITS)) + TASK_SIZE)
441 #else
442 #define KERN_VIRT_START	FIXADDR_START
443 #endif
444 
445 /*
446  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
447  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
448  */
449 #ifdef CONFIG_64BIT
450 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
451 #else
452 #define TASK_SIZE FIXADDR_START
453 #endif
454 
455 #else /* CONFIG_MMU */
456 
457 #define PAGE_SHARED		__pgprot(0)
458 #define PAGE_KERNEL		__pgprot(0)
459 #define swapper_pg_dir		NULL
460 #define TASK_SIZE		0xffffffffUL
461 #define VMALLOC_START		0
462 #define VMALLOC_END		TASK_SIZE
463 
464 static inline void __kernel_map_pages(struct page *page, int numpages, int enable) {}
465 
466 #endif /* !CONFIG_MMU */
467 
468 #define kern_addr_valid(addr)   (1) /* FIXME */
469 
470 extern void *dtb_early_va;
471 extern uintptr_t dtb_early_pa;
472 void setup_bootmem(void);
473 void paging_init(void);
474 
475 #define FIRST_USER_ADDRESS  0
476 
477 /*
478  * ZERO_PAGE is a global shared page that is always zero,
479  * used for zero-mapped memory areas, etc.
480  */
481 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
482 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
483 
484 #endif /* !__ASSEMBLY__ */
485 
486 #endif /* _ASM_RISCV_PGTABLE_H */
487