xref: /openbmc/linux/arch/riscv/include/asm/pgtable.h (revision 28dce2c4)
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 CONFIG_MMU
15 #define KERNEL_LINK_ADDR	PAGE_OFFSET
16 #else
17 
18 #define ADDRESS_SPACE_END	(UL(-1))
19 
20 #ifdef CONFIG_64BIT
21 /* Leave 2GB for kernel and BPF at the end of the address space */
22 #define KERNEL_LINK_ADDR	(ADDRESS_SPACE_END - SZ_2G + 1)
23 #else
24 #define KERNEL_LINK_ADDR	PAGE_OFFSET
25 #endif
26 
27 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
28 #define VMALLOC_END      (PAGE_OFFSET - 1)
29 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
30 
31 #define BPF_JIT_REGION_SIZE	(SZ_128M)
32 #ifdef CONFIG_64BIT
33 /* KASLR should leave at least 128MB for BPF after the kernel */
34 #define BPF_JIT_REGION_START	PFN_ALIGN((unsigned long)&_end)
35 #define BPF_JIT_REGION_END	(BPF_JIT_REGION_START + BPF_JIT_REGION_SIZE)
36 #else
37 #define BPF_JIT_REGION_START	(PAGE_OFFSET - BPF_JIT_REGION_SIZE)
38 #define BPF_JIT_REGION_END	(VMALLOC_END)
39 #endif
40 
41 /* Modules always live before the kernel */
42 #ifdef CONFIG_64BIT
43 #define MODULES_VADDR	(PFN_ALIGN((unsigned long)&_end) - SZ_2G)
44 #define MODULES_END	(PFN_ALIGN((unsigned long)&_start))
45 #endif
46 
47 /*
48  * Roughly size the vmemmap space to be large enough to fit enough
49  * struct pages to map half the virtual address space. Then
50  * position vmemmap directly below the VMALLOC region.
51  */
52 #define VMEMMAP_SHIFT \
53 	(CONFIG_VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
54 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
55 #define VMEMMAP_END	(VMALLOC_START - 1)
56 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
57 
58 /*
59  * Define vmemmap for pfn_to_page & page_to_pfn calls. Needed if kernel
60  * is configured with CONFIG_SPARSEMEM_VMEMMAP enabled.
61  */
62 #define vmemmap		((struct page *)VMEMMAP_START)
63 
64 #define PCI_IO_SIZE      SZ_16M
65 #define PCI_IO_END       VMEMMAP_START
66 #define PCI_IO_START     (PCI_IO_END - PCI_IO_SIZE)
67 
68 #define FIXADDR_TOP      PCI_IO_START
69 #ifdef CONFIG_64BIT
70 #define FIXADDR_SIZE     PMD_SIZE
71 #else
72 #define FIXADDR_SIZE     PGDIR_SIZE
73 #endif
74 #define FIXADDR_START    (FIXADDR_TOP - FIXADDR_SIZE)
75 
76 #endif
77 
78 #ifdef CONFIG_XIP_KERNEL
79 #define XIP_OFFSET		SZ_8M
80 #endif
81 
82 #ifndef __ASSEMBLY__
83 
84 /* Page Upper Directory not used in RISC-V */
85 #include <asm-generic/pgtable-nopud.h>
86 #include <asm/page.h>
87 #include <asm/tlbflush.h>
88 #include <linux/mm_types.h>
89 
90 #ifdef CONFIG_64BIT
91 #include <asm/pgtable-64.h>
92 #else
93 #include <asm/pgtable-32.h>
94 #endif /* CONFIG_64BIT */
95 
96 #ifdef CONFIG_XIP_KERNEL
97 #define XIP_FIXUP(addr) ({							\
98 	uintptr_t __a = (uintptr_t)(addr);					\
99 	(__a >= CONFIG_XIP_PHYS_ADDR && __a < CONFIG_XIP_PHYS_ADDR + SZ_16M) ?	\
100 		__a - CONFIG_XIP_PHYS_ADDR + CONFIG_PHYS_RAM_BASE - XIP_OFFSET :\
101 		__a;								\
102 	})
103 #else
104 #define XIP_FIXUP(addr)		(addr)
105 #endif /* CONFIG_XIP_KERNEL */
106 
107 #ifdef CONFIG_MMU
108 /* Number of entries in the page global directory */
109 #define PTRS_PER_PGD    (PAGE_SIZE / sizeof(pgd_t))
110 /* Number of entries in the page table */
111 #define PTRS_PER_PTE    (PAGE_SIZE / sizeof(pte_t))
112 
113 /* Number of PGD entries that a user-mode program can use */
114 #define USER_PTRS_PER_PGD   (TASK_SIZE / PGDIR_SIZE)
115 
116 /* Page protection bits */
117 #define _PAGE_BASE	(_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
118 
119 #define PAGE_NONE		__pgprot(_PAGE_PROT_NONE)
120 #define PAGE_READ		__pgprot(_PAGE_BASE | _PAGE_READ)
121 #define PAGE_WRITE		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE)
122 #define PAGE_EXEC		__pgprot(_PAGE_BASE | _PAGE_EXEC)
123 #define PAGE_READ_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
124 #define PAGE_WRITE_EXEC		__pgprot(_PAGE_BASE | _PAGE_READ |	\
125 					 _PAGE_EXEC | _PAGE_WRITE)
126 
127 #define PAGE_COPY		PAGE_READ
128 #define PAGE_COPY_EXEC		PAGE_EXEC
129 #define PAGE_COPY_READ_EXEC	PAGE_READ_EXEC
130 #define PAGE_SHARED		PAGE_WRITE
131 #define PAGE_SHARED_EXEC	PAGE_WRITE_EXEC
132 
133 #define _PAGE_KERNEL		(_PAGE_READ \
134 				| _PAGE_WRITE \
135 				| _PAGE_PRESENT \
136 				| _PAGE_ACCESSED \
137 				| _PAGE_DIRTY)
138 
139 #define PAGE_KERNEL		__pgprot(_PAGE_KERNEL)
140 #define PAGE_KERNEL_READ	__pgprot(_PAGE_KERNEL & ~_PAGE_WRITE)
141 #define PAGE_KERNEL_EXEC	__pgprot(_PAGE_KERNEL | _PAGE_EXEC)
142 #define PAGE_KERNEL_READ_EXEC	__pgprot((_PAGE_KERNEL & ~_PAGE_WRITE) \
143 					 | _PAGE_EXEC)
144 
145 #define PAGE_TABLE		__pgprot(_PAGE_TABLE)
146 
147 /*
148  * The RISC-V ISA doesn't yet specify how to query or modify PMAs, so we can't
149  * change the properties of memory regions.
150  */
151 #define _PAGE_IOREMAP _PAGE_KERNEL
152 
153 extern pgd_t swapper_pg_dir[];
154 
155 /* MAP_PRIVATE permissions: xwr (copy-on-write) */
156 #define __P000	PAGE_NONE
157 #define __P001	PAGE_READ
158 #define __P010	PAGE_COPY
159 #define __P011	PAGE_COPY
160 #define __P100	PAGE_EXEC
161 #define __P101	PAGE_READ_EXEC
162 #define __P110	PAGE_COPY_EXEC
163 #define __P111	PAGE_COPY_READ_EXEC
164 
165 /* MAP_SHARED permissions: xwr */
166 #define __S000	PAGE_NONE
167 #define __S001	PAGE_READ
168 #define __S010	PAGE_SHARED
169 #define __S011	PAGE_SHARED
170 #define __S100	PAGE_EXEC
171 #define __S101	PAGE_READ_EXEC
172 #define __S110	PAGE_SHARED_EXEC
173 #define __S111	PAGE_SHARED_EXEC
174 
175 static inline int pmd_present(pmd_t pmd)
176 {
177 	return (pmd_val(pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
178 }
179 
180 static inline int pmd_none(pmd_t pmd)
181 {
182 	return (pmd_val(pmd) == 0);
183 }
184 
185 static inline int pmd_bad(pmd_t pmd)
186 {
187 	return !pmd_present(pmd);
188 }
189 
190 #define pmd_leaf	pmd_leaf
191 static inline int pmd_leaf(pmd_t pmd)
192 {
193 	return pmd_present(pmd) &&
194 	       (pmd_val(pmd) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
195 }
196 
197 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
198 {
199 	*pmdp = pmd;
200 }
201 
202 static inline void pmd_clear(pmd_t *pmdp)
203 {
204 	set_pmd(pmdp, __pmd(0));
205 }
206 
207 static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
208 {
209 	return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
210 }
211 
212 static inline unsigned long _pgd_pfn(pgd_t pgd)
213 {
214 	return pgd_val(pgd) >> _PAGE_PFN_SHIFT;
215 }
216 
217 static inline struct page *pmd_page(pmd_t pmd)
218 {
219 	return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
220 }
221 
222 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
223 {
224 	return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
225 }
226 
227 static inline pte_t pmd_pte(pmd_t pmd)
228 {
229 	return __pte(pmd_val(pmd));
230 }
231 
232 /* Yields the page frame number (PFN) of a page table entry */
233 static inline unsigned long pte_pfn(pte_t pte)
234 {
235 	return (pte_val(pte) >> _PAGE_PFN_SHIFT);
236 }
237 
238 #define pte_page(x)     pfn_to_page(pte_pfn(x))
239 
240 /* Constructs a page table entry */
241 static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
242 {
243 	return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
244 }
245 
246 #define mk_pte(page, prot)       pfn_pte(page_to_pfn(page), prot)
247 
248 static inline int pte_present(pte_t pte)
249 {
250 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
251 }
252 
253 static inline int pte_none(pte_t pte)
254 {
255 	return (pte_val(pte) == 0);
256 }
257 
258 static inline int pte_write(pte_t pte)
259 {
260 	return pte_val(pte) & _PAGE_WRITE;
261 }
262 
263 static inline int pte_exec(pte_t pte)
264 {
265 	return pte_val(pte) & _PAGE_EXEC;
266 }
267 
268 static inline int pte_huge(pte_t pte)
269 {
270 	return pte_present(pte)
271 		&& (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
272 }
273 
274 static inline int pte_dirty(pte_t pte)
275 {
276 	return pte_val(pte) & _PAGE_DIRTY;
277 }
278 
279 static inline int pte_young(pte_t pte)
280 {
281 	return pte_val(pte) & _PAGE_ACCESSED;
282 }
283 
284 static inline int pte_special(pte_t pte)
285 {
286 	return pte_val(pte) & _PAGE_SPECIAL;
287 }
288 
289 /* static inline pte_t pte_rdprotect(pte_t pte) */
290 
291 static inline pte_t pte_wrprotect(pte_t pte)
292 {
293 	return __pte(pte_val(pte) & ~(_PAGE_WRITE));
294 }
295 
296 /* static inline pte_t pte_mkread(pte_t pte) */
297 
298 static inline pte_t pte_mkwrite(pte_t pte)
299 {
300 	return __pte(pte_val(pte) | _PAGE_WRITE);
301 }
302 
303 /* static inline pte_t pte_mkexec(pte_t pte) */
304 
305 static inline pte_t pte_mkdirty(pte_t pte)
306 {
307 	return __pte(pte_val(pte) | _PAGE_DIRTY);
308 }
309 
310 static inline pte_t pte_mkclean(pte_t pte)
311 {
312 	return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
313 }
314 
315 static inline pte_t pte_mkyoung(pte_t pte)
316 {
317 	return __pte(pte_val(pte) | _PAGE_ACCESSED);
318 }
319 
320 static inline pte_t pte_mkold(pte_t pte)
321 {
322 	return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
323 }
324 
325 static inline pte_t pte_mkspecial(pte_t pte)
326 {
327 	return __pte(pte_val(pte) | _PAGE_SPECIAL);
328 }
329 
330 static inline pte_t pte_mkhuge(pte_t pte)
331 {
332 	return pte;
333 }
334 
335 #ifdef CONFIG_NUMA_BALANCING
336 /*
337  * See the comment in include/asm-generic/pgtable.h
338  */
339 static inline int pte_protnone(pte_t pte)
340 {
341 	return (pte_val(pte) & (_PAGE_PRESENT | _PAGE_PROT_NONE)) == _PAGE_PROT_NONE;
342 }
343 
344 static inline int pmd_protnone(pmd_t pmd)
345 {
346 	return pte_protnone(pmd_pte(pmd));
347 }
348 #endif
349 
350 /* Modify page protection bits */
351 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
352 {
353 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
354 }
355 
356 #define pgd_ERROR(e) \
357 	pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
358 
359 
360 /* Commit new configuration to MMU hardware */
361 static inline void update_mmu_cache(struct vm_area_struct *vma,
362 	unsigned long address, pte_t *ptep)
363 {
364 	/*
365 	 * The kernel assumes that TLBs don't cache invalid entries, but
366 	 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
367 	 * cache flush; it is necessary even after writing invalid entries.
368 	 * Relying on flush_tlb_fix_spurious_fault would suffice, but
369 	 * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
370 	 */
371 	local_flush_tlb_page(address);
372 }
373 
374 #define __HAVE_ARCH_PTE_SAME
375 static inline int pte_same(pte_t pte_a, pte_t pte_b)
376 {
377 	return pte_val(pte_a) == pte_val(pte_b);
378 }
379 
380 /*
381  * Certain architectures need to do special things when PTEs within
382  * a page table are directly modified.  Thus, the following hook is
383  * made available.
384  */
385 static inline void set_pte(pte_t *ptep, pte_t pteval)
386 {
387 	*ptep = pteval;
388 }
389 
390 void flush_icache_pte(pte_t pte);
391 
392 static inline void set_pte_at(struct mm_struct *mm,
393 	unsigned long addr, pte_t *ptep, pte_t pteval)
394 {
395 	if (pte_present(pteval) && pte_exec(pteval))
396 		flush_icache_pte(pteval);
397 
398 	set_pte(ptep, pteval);
399 }
400 
401 static inline void pte_clear(struct mm_struct *mm,
402 	unsigned long addr, pte_t *ptep)
403 {
404 	set_pte_at(mm, addr, ptep, __pte(0));
405 }
406 
407 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
408 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
409 					unsigned long address, pte_t *ptep,
410 					pte_t entry, int dirty)
411 {
412 	if (!pte_same(*ptep, entry))
413 		set_pte_at(vma->vm_mm, address, ptep, entry);
414 	/*
415 	 * update_mmu_cache will unconditionally execute, handling both
416 	 * the case that the PTE changed and the spurious fault case.
417 	 */
418 	return true;
419 }
420 
421 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
422 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
423 				       unsigned long address, pte_t *ptep)
424 {
425 	return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
426 }
427 
428 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
429 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
430 					    unsigned long address,
431 					    pte_t *ptep)
432 {
433 	if (!pte_young(*ptep))
434 		return 0;
435 	return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
436 }
437 
438 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
439 static inline void ptep_set_wrprotect(struct mm_struct *mm,
440 				      unsigned long address, pte_t *ptep)
441 {
442 	atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
443 }
444 
445 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
446 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
447 					 unsigned long address, pte_t *ptep)
448 {
449 	/*
450 	 * This comment is borrowed from x86, but applies equally to RISC-V:
451 	 *
452 	 * Clearing the accessed bit without a TLB flush
453 	 * doesn't cause data corruption. [ It could cause incorrect
454 	 * page aging and the (mistaken) reclaim of hot pages, but the
455 	 * chance of that should be relatively low. ]
456 	 *
457 	 * So as a performance optimization don't flush the TLB when
458 	 * clearing the accessed bit, it will eventually be flushed by
459 	 * a context switch or a VM operation anyway. [ In the rare
460 	 * event of it not getting flushed for a long time the delay
461 	 * shouldn't really matter because there's no real memory
462 	 * pressure for swapout to react to. ]
463 	 */
464 	return ptep_test_and_clear_young(vma, address, ptep);
465 }
466 
467 /*
468  * Encode and decode a swap entry
469  *
470  * Format of swap PTE:
471  *	bit            0:	_PAGE_PRESENT (zero)
472  *	bit            1:	_PAGE_PROT_NONE (zero)
473  *	bits      2 to 6:	swap type
474  *	bits 7 to XLEN-1:	swap offset
475  */
476 #define __SWP_TYPE_SHIFT	2
477 #define __SWP_TYPE_BITS		5
478 #define __SWP_TYPE_MASK		((1UL << __SWP_TYPE_BITS) - 1)
479 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
480 
481 #define MAX_SWAPFILES_CHECK()	\
482 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
483 
484 #define __swp_type(x)	(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
485 #define __swp_offset(x)	((x).val >> __SWP_OFFSET_SHIFT)
486 #define __swp_entry(type, offset) ((swp_entry_t) \
487 	{ ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
488 
489 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
490 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
491 
492 /*
493  * In the RV64 Linux scheme, we give the user half of the virtual-address space
494  * and give the kernel the other (upper) half.
495  */
496 #ifdef CONFIG_64BIT
497 #define KERN_VIRT_START	(-(BIT(CONFIG_VA_BITS)) + TASK_SIZE)
498 #else
499 #define KERN_VIRT_START	FIXADDR_START
500 #endif
501 
502 /*
503  * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
504  * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
505  */
506 #ifdef CONFIG_64BIT
507 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
508 #else
509 #define TASK_SIZE FIXADDR_START
510 #endif
511 
512 #else /* CONFIG_MMU */
513 
514 #define PAGE_SHARED		__pgprot(0)
515 #define PAGE_KERNEL		__pgprot(0)
516 #define swapper_pg_dir		NULL
517 #define TASK_SIZE		0xffffffffUL
518 #define VMALLOC_START		0
519 #define VMALLOC_END		TASK_SIZE
520 
521 #endif /* !CONFIG_MMU */
522 
523 #define kern_addr_valid(addr)   (1) /* FIXME */
524 
525 extern char _start[];
526 extern void *_dtb_early_va;
527 extern uintptr_t _dtb_early_pa;
528 #if defined(CONFIG_XIP_KERNEL) && defined(CONFIG_MMU)
529 #define dtb_early_va	(*(void **)XIP_FIXUP(&_dtb_early_va))
530 #define dtb_early_pa	(*(uintptr_t *)XIP_FIXUP(&_dtb_early_pa))
531 #else
532 #define dtb_early_va	_dtb_early_va
533 #define dtb_early_pa	_dtb_early_pa
534 #endif /* CONFIG_XIP_KERNEL */
535 
536 void setup_bootmem(void);
537 void paging_init(void);
538 void misc_mem_init(void);
539 
540 #define FIRST_USER_ADDRESS  0
541 
542 /*
543  * ZERO_PAGE is a global shared page that is always zero,
544  * used for zero-mapped memory areas, etc.
545  */
546 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
547 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
548 
549 #endif /* !__ASSEMBLY__ */
550 
551 #endif /* _ASM_RISCV_PGTABLE_H */
552