Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md
[linux-2.6-block.git] / arch / riscv / include / asm / pgtable.h
CommitLineData
07037db5
PD
1/*
2 * Copyright (C) 2012 Regents of the University of California
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation, version 2.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 */
13
14#ifndef _ASM_RISCV_PGTABLE_H
15#define _ASM_RISCV_PGTABLE_H
16
17#include <linux/mmzone.h>
18
19#include <asm/pgtable-bits.h>
20
21#ifndef __ASSEMBLY__
22
07037db5
PD
23/* Page Upper Directory not used in RISC-V */
24#include <asm-generic/pgtable-nopud.h>
25#include <asm/page.h>
26#include <asm/tlbflush.h>
27#include <linux/mm_types.h>
28
29#ifdef CONFIG_64BIT
30#include <asm/pgtable-64.h>
31#else
32#include <asm/pgtable-32.h>
33#endif /* CONFIG_64BIT */
34
35/* Number of entries in the page global directory */
36#define PTRS_PER_PGD (PAGE_SIZE / sizeof(pgd_t))
37/* Number of entries in the page table */
38#define PTRS_PER_PTE (PAGE_SIZE / sizeof(pte_t))
39
40/* Number of PGD entries that a user-mode program can use */
41#define USER_PTRS_PER_PGD (TASK_SIZE / PGDIR_SIZE)
42#define FIRST_USER_ADDRESS 0
43
44/* Page protection bits */
45#define _PAGE_BASE (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
46
47#define PAGE_NONE __pgprot(0)
48#define PAGE_READ __pgprot(_PAGE_BASE | _PAGE_READ)
49#define PAGE_WRITE __pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_WRITE)
50#define PAGE_EXEC __pgprot(_PAGE_BASE | _PAGE_EXEC)
51#define PAGE_READ_EXEC __pgprot(_PAGE_BASE | _PAGE_READ | _PAGE_EXEC)
52#define PAGE_WRITE_EXEC __pgprot(_PAGE_BASE | _PAGE_READ | \
53 _PAGE_EXEC | _PAGE_WRITE)
54
55#define PAGE_COPY PAGE_READ
56#define PAGE_COPY_EXEC PAGE_EXEC
57#define PAGE_COPY_READ_EXEC PAGE_READ_EXEC
58#define PAGE_SHARED PAGE_WRITE
59#define PAGE_SHARED_EXEC PAGE_WRITE_EXEC
60
61#define _PAGE_KERNEL (_PAGE_READ \
62 | _PAGE_WRITE \
63 | _PAGE_PRESENT \
64 | _PAGE_ACCESSED \
65 | _PAGE_DIRTY)
66
67#define PAGE_KERNEL __pgprot(_PAGE_KERNEL)
68#define PAGE_KERNEL_EXEC __pgprot(_PAGE_KERNEL | _PAGE_EXEC)
69
70extern pgd_t swapper_pg_dir[];
71
72/* MAP_PRIVATE permissions: xwr (copy-on-write) */
73#define __P000 PAGE_NONE
74#define __P001 PAGE_READ
75#define __P010 PAGE_COPY
76#define __P011 PAGE_COPY
77#define __P100 PAGE_EXEC
78#define __P101 PAGE_READ_EXEC
79#define __P110 PAGE_COPY_EXEC
80#define __P111 PAGE_COPY_READ_EXEC
81
82/* MAP_SHARED permissions: xwr */
83#define __S000 PAGE_NONE
84#define __S001 PAGE_READ
85#define __S010 PAGE_SHARED
86#define __S011 PAGE_SHARED
87#define __S100 PAGE_EXEC
88#define __S101 PAGE_READ_EXEC
89#define __S110 PAGE_SHARED_EXEC
90#define __S111 PAGE_SHARED_EXEC
91
92/*
93 * ZERO_PAGE is a global shared page that is always zero,
94 * used for zero-mapped memory areas, etc.
95 */
96extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
97#define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
98
99static inline int pmd_present(pmd_t pmd)
100{
101 return (pmd_val(pmd) & _PAGE_PRESENT);
102}
103
104static inline int pmd_none(pmd_t pmd)
105{
106 return (pmd_val(pmd) == 0);
107}
108
109static inline int pmd_bad(pmd_t pmd)
110{
111 return !pmd_present(pmd);
112}
113
114static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
115{
116 *pmdp = pmd;
117}
118
119static inline void pmd_clear(pmd_t *pmdp)
120{
121 set_pmd(pmdp, __pmd(0));
122}
123
124
125static inline pgd_t pfn_pgd(unsigned long pfn, pgprot_t prot)
126{
127 return __pgd((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
128}
129
130#define pgd_index(addr) (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
131
132/* Locate an entry in the page global directory */
133static inline pgd_t *pgd_offset(const struct mm_struct *mm, unsigned long addr)
134{
135 return mm->pgd + pgd_index(addr);
136}
137/* Locate an entry in the kernel page global directory */
138#define pgd_offset_k(addr) pgd_offset(&init_mm, (addr))
139
140static inline struct page *pmd_page(pmd_t pmd)
141{
142 return pfn_to_page(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
143}
144
145static inline unsigned long pmd_page_vaddr(pmd_t pmd)
146{
147 return (unsigned long)pfn_to_virt(pmd_val(pmd) >> _PAGE_PFN_SHIFT);
148}
149
150/* Yields the page frame number (PFN) of a page table entry */
151static inline unsigned long pte_pfn(pte_t pte)
152{
153 return (pte_val(pte) >> _PAGE_PFN_SHIFT);
154}
155
156#define pte_page(x) pfn_to_page(pte_pfn(x))
157
158/* Constructs a page table entry */
159static inline pte_t pfn_pte(unsigned long pfn, pgprot_t prot)
160{
161 return __pte((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
162}
163
164static inline pte_t mk_pte(struct page *page, pgprot_t prot)
165{
166 return pfn_pte(page_to_pfn(page), prot);
167}
168
169#define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
170
171static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long addr)
172{
173 return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(addr);
174}
175
176#define pte_offset_map(dir, addr) pte_offset_kernel((dir), (addr))
177#define pte_unmap(pte) ((void)(pte))
178
07037db5
PD
179static inline int pte_present(pte_t pte)
180{
181 return (pte_val(pte) & _PAGE_PRESENT);
182}
183
184static inline int pte_none(pte_t pte)
185{
186 return (pte_val(pte) == 0);
187}
188
07037db5
PD
189static inline int pte_write(pte_t pte)
190{
191 return pte_val(pte) & _PAGE_WRITE;
192}
193
08f051ed
AW
194static inline int pte_exec(pte_t pte)
195{
196 return pte_val(pte) & _PAGE_EXEC;
197}
198
07037db5
PD
199static inline int pte_huge(pte_t pte)
200{
201 return pte_present(pte)
202 && (pte_val(pte) & (_PAGE_READ | _PAGE_WRITE | _PAGE_EXEC));
203}
204
07037db5
PD
205static inline int pte_dirty(pte_t pte)
206{
207 return pte_val(pte) & _PAGE_DIRTY;
208}
209
210static inline int pte_young(pte_t pte)
211{
212 return pte_val(pte) & _PAGE_ACCESSED;
213}
214
215static inline int pte_special(pte_t pte)
216{
217 return pte_val(pte) & _PAGE_SPECIAL;
218}
219
220/* static inline pte_t pte_rdprotect(pte_t pte) */
221
222static inline pte_t pte_wrprotect(pte_t pte)
223{
224 return __pte(pte_val(pte) & ~(_PAGE_WRITE));
225}
226
227/* static inline pte_t pte_mkread(pte_t pte) */
228
229static inline pte_t pte_mkwrite(pte_t pte)
230{
231 return __pte(pte_val(pte) | _PAGE_WRITE);
232}
233
234/* static inline pte_t pte_mkexec(pte_t pte) */
235
236static inline pte_t pte_mkdirty(pte_t pte)
237{
238 return __pte(pte_val(pte) | _PAGE_DIRTY);
239}
240
241static inline pte_t pte_mkclean(pte_t pte)
242{
243 return __pte(pte_val(pte) & ~(_PAGE_DIRTY));
244}
245
246static inline pte_t pte_mkyoung(pte_t pte)
247{
248 return __pte(pte_val(pte) | _PAGE_ACCESSED);
249}
250
251static inline pte_t pte_mkold(pte_t pte)
252{
253 return __pte(pte_val(pte) & ~(_PAGE_ACCESSED));
254}
255
256static inline pte_t pte_mkspecial(pte_t pte)
257{
258 return __pte(pte_val(pte) | _PAGE_SPECIAL);
259}
260
261/* Modify page protection bits */
262static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
263{
264 return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
265}
266
267#define pgd_ERROR(e) \
268 pr_err("%s:%d: bad pgd " PTE_FMT ".\n", __FILE__, __LINE__, pgd_val(e))
269
270
271/* Commit new configuration to MMU hardware */
272static inline void update_mmu_cache(struct vm_area_struct *vma,
273 unsigned long address, pte_t *ptep)
274{
275 /*
276 * The kernel assumes that TLBs don't cache invalid entries, but
277 * in RISC-V, SFENCE.VMA specifies an ordering constraint, not a
278 * cache flush; it is necessary even after writing invalid entries.
279 * Relying on flush_tlb_fix_spurious_fault would suffice, but
280 * the extra traps reduce performance. So, eagerly SFENCE.VMA.
281 */
282 local_flush_tlb_page(address);
283}
284
285#define __HAVE_ARCH_PTE_SAME
286static inline int pte_same(pte_t pte_a, pte_t pte_b)
287{
288 return pte_val(pte_a) == pte_val(pte_b);
289}
290
08f051ed
AW
291/*
292 * Certain architectures need to do special things when PTEs within
293 * a page table are directly modified. Thus, the following hook is
294 * made available.
295 */
296static inline void set_pte(pte_t *ptep, pte_t pteval)
297{
298 *ptep = pteval;
299}
300
301void flush_icache_pte(pte_t pte);
302
303static inline void set_pte_at(struct mm_struct *mm,
304 unsigned long addr, pte_t *ptep, pte_t pteval)
305{
306 if (pte_present(pteval) && pte_exec(pteval))
307 flush_icache_pte(pteval);
308
309 set_pte(ptep, pteval);
310}
311
312static inline void pte_clear(struct mm_struct *mm,
313 unsigned long addr, pte_t *ptep)
314{
315 set_pte_at(mm, addr, ptep, __pte(0));
316}
317
07037db5
PD
318#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
319static inline int ptep_set_access_flags(struct vm_area_struct *vma,
320 unsigned long address, pte_t *ptep,
321 pte_t entry, int dirty)
322{
323 if (!pte_same(*ptep, entry))
324 set_pte_at(vma->vm_mm, address, ptep, entry);
325 /*
326 * update_mmu_cache will unconditionally execute, handling both
327 * the case that the PTE changed and the spurious fault case.
328 */
329 return true;
330}
331
332#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
333static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
334 unsigned long address, pte_t *ptep)
335{
336 return __pte(atomic_long_xchg((atomic_long_t *)ptep, 0));
337}
338
339#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
340static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
341 unsigned long address,
342 pte_t *ptep)
343{
344 if (!pte_young(*ptep))
345 return 0;
346 return test_and_clear_bit(_PAGE_ACCESSED_OFFSET, &pte_val(*ptep));
347}
348
349#define __HAVE_ARCH_PTEP_SET_WRPROTECT
350static inline void ptep_set_wrprotect(struct mm_struct *mm,
351 unsigned long address, pte_t *ptep)
352{
353 atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
354}
355
356#define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
357static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
358 unsigned long address, pte_t *ptep)
359{
360 /*
361 * This comment is borrowed from x86, but applies equally to RISC-V:
362 *
363 * Clearing the accessed bit without a TLB flush
364 * doesn't cause data corruption. [ It could cause incorrect
365 * page aging and the (mistaken) reclaim of hot pages, but the
366 * chance of that should be relatively low. ]
367 *
368 * So as a performance optimization don't flush the TLB when
369 * clearing the accessed bit, it will eventually be flushed by
370 * a context switch or a VM operation anyway. [ In the rare
371 * event of it not getting flushed for a long time the delay
372 * shouldn't really matter because there's no real memory
373 * pressure for swapout to react to. ]
374 */
375 return ptep_test_and_clear_young(vma, address, ptep);
376}
377
378/*
379 * Encode and decode a swap entry
380 *
381 * Format of swap PTE:
382 * bit 0: _PAGE_PRESENT (zero)
383 * bit 1: reserved for future use (zero)
384 * bits 2 to 6: swap type
385 * bits 7 to XLEN-1: swap offset
386 */
387#define __SWP_TYPE_SHIFT 2
388#define __SWP_TYPE_BITS 5
389#define __SWP_TYPE_MASK ((1UL << __SWP_TYPE_BITS) - 1)
390#define __SWP_OFFSET_SHIFT (__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
391
392#define MAX_SWAPFILES_CHECK() \
393 BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
394
395#define __swp_type(x) (((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
396#define __swp_offset(x) ((x).val >> __SWP_OFFSET_SHIFT)
397#define __swp_entry(type, offset) ((swp_entry_t) \
398 { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
399
400#define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) })
401#define __swp_entry_to_pte(x) ((pte_t) { (x).val })
402
403#ifdef CONFIG_FLATMEM
404#define kern_addr_valid(addr) (1) /* FIXME */
405#endif
406
407extern void paging_init(void);
408
409static inline void pgtable_cache_init(void)
410{
411 /* No page table caches to initialize */
412}
413
07037db5
PD
414#define VMALLOC_SIZE (KERN_VIRT_SIZE >> 1)
415#define VMALLOC_END (PAGE_OFFSET - 1)
416#define VMALLOC_START (PAGE_OFFSET - VMALLOC_SIZE)
417
418/*
419 * Task size is 0x40000000000 for RV64 or 0xb800000 for RV32.
420 * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
421 */
422#ifdef CONFIG_64BIT
423#define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
424#else
425#define TASK_SIZE VMALLOC_START
426#endif
427
428#include <asm-generic/pgtable.h>
429
430#endif /* !__ASSEMBLY__ */
431
432#endif /* _ASM_RISCV_PGTABLE_H */