xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision 31e67366)
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_READ	__pgprot(_PAGE_KERNEL & ~_PAGE_WRITE)
103 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
104 #define PAGE_KERNEL_READ_EXEC	__pgprot((_PAGE_KERNEL & ~_PAGE_WRITE) \
105 					 | _PAGE_EXEC)
106 
107 #define PAGE_TABLE		__pgprot(_PAGE_TABLE)
108 
109 /*
110  * The RISC-V ISA doesn't yet specify how to query or modify PMAs, so we can't
111  * change the properties of memory regions.
112  */
113 #define _PAGE_IOREMAP _PAGE_KERNEL
114 
115 extern pgd_t swapper_pg_dir[];
116 
117 /* MAP_PRIVATE permissions: xwr (copy-on-write) */
118 #define __P000	PAGE_NONE
119 #define __P001	PAGE_READ
120 #define __P010	PAGE_COPY
121 #define __P011	PAGE_COPY
122 #define __P100	PAGE_EXEC
123 #define __P101	PAGE_READ_EXEC
124 #define __P110	PAGE_COPY_EXEC
125 #define __P111	PAGE_COPY_READ_EXEC
126 
127 /* MAP_SHARED permissions: xwr */
128 #define __S000	PAGE_NONE
129 #define __S001	PAGE_READ
130 #define __S010	PAGE_SHARED
131 #define __S011	PAGE_SHARED
132 #define __S100	PAGE_EXEC
133 #define __S101	PAGE_READ_EXEC
134 #define __S110	PAGE_SHARED_EXEC
135 #define __S111	PAGE_SHARED_EXEC
136 
137 static inline int pmd_present(pmd_t pmd)
138 {
139 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
140 }
141 
142 static inline int pmd_none(pmd_t pmd)
143 {
144 	return (pmd_val(pmd) == 0);
145 }
146 
147 static inline int pmd_bad(pmd_t pmd)
148 {
149 	return !pmd_present(pmd);
150 }
151 
152 #define pmd_leaf	pmd_leaf
153 static inline int pmd_leaf(pmd_t pmd)
154 {
155 	return pmd_present(pmd) &&
156 	       (pmd_val(pmd) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
157 }
158 
159 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
160 {
161 	*pmdp = pmd;
162 }
163 
164 static inline void pmd_clear(pmd_t *pmdp)
165 {
166 	set_pmd(pmdp, __pmd(0));
167 }
168 
169 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
170 {
171 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
172 }
173 
174 static inline unsigned long _pgd_pfn(pgd_t pgd)
175 {
176 	return pgd_val(pgd) >> _PAGE_PFN_SHIFT;
177 }
178 
179 static inline struct page *pmd_page(pmd_t pmd)
180 {
181 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
182 }
183 
184 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
185 {
186 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
187 }
188 
189 static inline pte_t pmd_pte(pmd_t pmd)
190 {
191 	return __pte(pmd_val(pmd));
192 }
193 
194 /* Yields the page frame number (PFN) of a page table entry */
195 static inline unsigned long pte_pfn(pte_t pte)
196 {
197 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
198 }
199 
200 #define pte_page(x)     pfn_to_page(pte_pfn(x))
201 
202 /* Constructs a page table entry */
203 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
204 {
205 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
206 }
207 
208 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
209 
210 static inline int pte_present(pte_t pte)
211 {
212 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
213 }
214 
215 static inline int pte_none(pte_t pte)
216 {
217 	return (pte_val(pte) == 0);
218 }
219 
220 static inline int pte_write(pte_t pte)
221 {
222 	return pte_val(pte) & _PAGE_WRITE;
223 }
224 
225 static inline int pte_exec(pte_t pte)
226 {
227 	return pte_val(pte) & _PAGE_EXEC;
228 }
229 
230 static inline int pte_huge(pte_t pte)
231 {
232 	return pte_present(pte)
233 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
234 }
235 
236 static inline int pte_dirty(pte_t pte)
237 {
238 	return pte_val(pte) & _PAGE_DIRTY;
239 }
240 
241 static inline int pte_young(pte_t pte)
242 {
243 	return pte_val(pte) & _PAGE_ACCESSED;
244 }
245 
246 static inline int pte_special(pte_t pte)
247 {
248 	return pte_val(pte) & _PAGE_SPECIAL;
249 }
250 
251 /* static inline pte_t pte_rdprotect(pte_t pte) */
252 
253 static inline pte_t pte_wrprotect(pte_t pte)
254 {
255 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
256 }
257 
258 /* static inline pte_t pte_mkread(pte_t pte) */
259 
260 static inline pte_t pte_mkwrite(pte_t pte)
261 {
262 	return __pte(pte_val(pte) | _PAGE_WRITE);
263 }
264 
265 /* static inline pte_t pte_mkexec(pte_t pte) */
266 
267 static inline pte_t pte_mkdirty(pte_t pte)
268 {
269 	return __pte(pte_val(pte) | _PAGE_DIRTY);
270 }
271 
272 static inline pte_t pte_mkclean(pte_t pte)
273 {
274 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
275 }
276 
277 static inline pte_t pte_mkyoung(pte_t pte)
278 {
279 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
280 }
281 
282 static inline pte_t pte_mkold(pte_t pte)
283 {
284 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
285 }
286 
287 static inline pte_t pte_mkspecial(pte_t pte)
288 {
289 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
290 }
291 
292 static inline pte_t pte_mkhuge(pte_t pte)
293 {
294 	return pte;
295 }
296 
297 #ifdef CONFIG_NUMA_BALANCING
298 /*
299  * See the comment in include/asm-generic/pgtable.h
300  */
301 static inline int pte_protnone(pte_t pte)
302 {
303 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)) == _PAGE_PROT_NONE;
304 }
305 
306 static inline int pmd_protnone(pmd_t pmd)
307 {
308 	return pte_protnone(pmd_pte(pmd));
309 }
310 #endif
311 
312 /* Modify page protection bits */
313 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
314 {
315 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
316 }
317 
318 #define pgd_ERROR(e) \
319 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
320 
321 
322 /* Commit new configuration to MMU hardware */
323 static inline void update_mmu_cache(struct vm_area_struct *vma,
324 	unsigned long address, pte_t *ptep)
325 {
326 	/*
327 	 * The kernel assumes that TLBs don't cache invalid entries, but
328 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
329 	 * cache flush; it is necessary even after writing invalid entries.
330 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
331 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
332 	 */
333 	local_flush_tlb_page(address);
334 }
335 
336 #define __HAVE_ARCH_PTE_SAME
337 static inline int pte_same(pte_t pte_a, pte_t pte_b)
338 {
339 	return pte_val(pte_a) == pte_val(pte_b);
340 }
341 
342 /*
343  * Certain architectures need to do special things when PTEs within
344  * a page table are directly modified.  Thus, the following hook is
345  * made available.
346  */
347 static inline void set_pte(pte_t *ptep, pte_t pteval)
348 {
349 	*ptep = pteval;
350 }
351 
352 void flush_icache_pte(pte_t pte);
353 
354 static inline void set_pte_at(struct mm_struct *mm,
355 	unsigned long addr, pte_t *ptep, pte_t pteval)
356 {
357 	if (pte_present(pteval) && pte_exec(pteval))
358 		flush_icache_pte(pteval);
359 
360 	set_pte(ptep, pteval);
361 }
362 
363 static inline void pte_clear(struct mm_struct *mm,
364 	unsigned long addr, pte_t *ptep)
365 {
366 	set_pte_at(mm, addr, ptep, __pte(0));
367 }
368 
369 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
370 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
371 					unsigned long address, pte_t *ptep,
372 					pte_t entry, int dirty)
373 {
374 	if (!pte_same(*ptep, entry))
375 		set_pte_at(vma->vm_mm, address, ptep, entry);
376 	/*
377 	 * update_mmu_cache will unconditionally execute, handling both
378 	 * the case that the PTE changed and the spurious fault case.
379 	 */
380 	return true;
381 }
382 
383 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
384 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
385 				       unsigned long address, pte_t *ptep)
386 {
387 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
388 }
389 
390 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
391 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
392 					    unsigned long address,
393 					    pte_t *ptep)
394 {
395 	if (!pte_young(*ptep))
396 		return 0;
397 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
398 }
399 
400 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
401 static inline void ptep_set_wrprotect(struct mm_struct *mm,
402 				      unsigned long address, pte_t *ptep)
403 {
404 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
405 }
406 
407 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
408 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
409 					 unsigned long address, pte_t *ptep)
410 {
411 	/*
412 	 * This comment is borrowed from x86, but applies equally to RISC-V:
413 	 *
414 	 * Clearing the accessed bit without a TLB flush
415 	 * doesn't cause data corruption. [ It could cause incorrect
416 	 * page aging and the (mistaken) reclaim of hot pages, but the
417 	 * chance of that should be relatively low. ]
418 	 *
419 	 * So as a performance optimization don't flush the TLB when
420 	 * clearing the accessed bit, it will eventually be flushed by
421 	 * a context switch or a VM operation anyway. [ In the rare
422 	 * event of it not getting flushed for a long time the delay
423 	 * shouldn't really matter because there's no real memory
424 	 * pressure for swapout to react to. ]
425 	 */
426 	return ptep_test_and_clear_young(vma, address, ptep);
427 }
428 
429 /*
430  * Encode and decode a swap entry
431  *
432  * Format of swap PTE:
433  *	bit            0:	_PAGE_PRESENT (zero)
434  *	bit            1:	_PAGE_PROT_NONE (zero)
435  *	bits      2 to 6:	swap type
436  *	bits 7 to XLEN-1:	swap offset
437  */
438 #define __SWP_TYPE_SHIFT	2
439 #define __SWP_TYPE_BITS		5
440 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
441 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
442 
443 #define MAX_SWAPFILES_CHECK()	\
444 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
445 
446 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
447 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
448 #define __swp_entry(type, offset) ((swp_entry_t) \
449 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
450 
451 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
452 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
453 
454 /*
455  * In the RV64 Linux scheme, we give the user half of the virtual-address space
456  * and give the kernel the other (upper) half.
457  */
458 #ifdef CONFIG_64BIT
459 #define KERN_VIRT_START	(-(BIT(CONFIG_VA_BITS)) + TASK_SIZE)
460 #else
461 #define KERN_VIRT_START	FIXADDR_START
462 #endif
463 
464 /*
465  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
466  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
467  */
468 #ifdef CONFIG_64BIT
469 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
470 #else
471 #define TASK_SIZE FIXADDR_START
472 #endif
473 
474 #else /* CONFIG_MMU */
475 
476 #define PAGE_SHARED		__pgprot(0)
477 #define PAGE_KERNEL		__pgprot(0)
478 #define swapper_pg_dir		NULL
479 #define TASK_SIZE		0xffffffffUL
480 #define VMALLOC_START		0
481 #define VMALLOC_END		TASK_SIZE
482 
483 #endif /* !CONFIG_MMU */
484 
485 #define kern_addr_valid(addr)   (1) /* FIXME */
486 
487 extern void *dtb_early_va;
488 extern uintptr_t dtb_early_pa;
489 void setup_bootmem(void);
490 void paging_init(void);
491 void misc_mem_init(void);
492 
493 #define FIRST_USER_ADDRESS  0
494 
495 /*
496  * ZERO_PAGE is a global shared page that is always zero,
497  * used for zero-mapped memory areas, etc.
498  */
499 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
500 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
501 
502 #endif /* !__ASSEMBLY__ */
503 
504 #endif /* _ASM_RISCV_PGTABLE_H */
505