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