powerpc/mm: Add option for non-atomic PTE updates to ppc64
[deliverable/linux.git] / arch / powerpc / include / asm / pgtable-ppc64.h
CommitLineData
f88df14b
DG
1#ifndef _ASM_POWERPC_PGTABLE_PPC64_H_
2#define _ASM_POWERPC_PGTABLE_PPC64_H_
3/*
4 * This file contains the functions and defines necessary to modify and use
5 * the ppc64 hashed page table.
6 */
7
8#ifndef __ASSEMBLY__
9#include <linux/stddef.h>
f88df14b 10#include <asm/tlbflush.h>
f88df14b
DG
11#endif /* __ASSEMBLY__ */
12
13#ifdef CONFIG_PPC_64K_PAGES
c605782b 14#include <asm/pgtable-ppc64-64k.h>
f88df14b 15#else
c605782b 16#include <asm/pgtable-ppc64-4k.h>
f88df14b
DG
17#endif
18
19#define FIRST_USER_ADDRESS 0
20
21/*
22 * Size of EA range mapped by our pagetables.
23 */
24#define PGTABLE_EADDR_SIZE (PTE_INDEX_SIZE + PMD_INDEX_SIZE + \
25 PUD_INDEX_SIZE + PGD_INDEX_SIZE + PAGE_SHIFT)
3d5134ee 26#define PGTABLE_RANGE (ASM_CONST(1) << PGTABLE_EADDR_SIZE)
f88df14b 27
c605782b
BH
28
29/* Some sanity checking */
f88df14b
DG
30#if TASK_SIZE_USER64 > PGTABLE_RANGE
31#error TASK_SIZE_USER64 exceeds pagetable range
32#endif
33
34#if TASK_SIZE_USER64 > (1UL << (USER_ESID_BITS + SID_SHIFT))
35#error TASK_SIZE_USER64 exceeds user VSID range
36#endif
37
38/*
39 * Define the address range of the vmalloc VM area.
40 */
41#define VMALLOC_START ASM_CONST(0xD000000000000000)
3d5134ee 42#define VMALLOC_SIZE (PGTABLE_RANGE >> 1)
f88df14b
DG
43#define VMALLOC_END (VMALLOC_START + VMALLOC_SIZE)
44
45/*
3d5134ee
BH
46 * Define the address ranges for MMIO and IO space :
47 *
48 * ISA_IO_BASE = VMALLOC_END, 64K reserved area
49 * PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
50 * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
f88df14b 51 */
3d5134ee
BH
52#define FULL_IO_SIZE 0x80000000ul
53#define ISA_IO_BASE (VMALLOC_END)
54#define ISA_IO_END (VMALLOC_END + 0x10000ul)
55#define PHB_IO_BASE (ISA_IO_END)
56#define PHB_IO_END (VMALLOC_END + FULL_IO_SIZE)
57#define IOREMAP_BASE (PHB_IO_END)
58#define IOREMAP_END (VMALLOC_START + PGTABLE_RANGE)
f88df14b
DG
59
60/*
61 * Region IDs
62 */
63#define REGION_SHIFT 60UL
64#define REGION_MASK (0xfUL << REGION_SHIFT)
65#define REGION_ID(ea) (((unsigned long)(ea)) >> REGION_SHIFT)
66
67#define VMALLOC_REGION_ID (REGION_ID(VMALLOC_START))
68#define KERNEL_REGION_ID (REGION_ID(PAGE_OFFSET))
cec08e7a 69#define VMEMMAP_REGION_ID (0xfUL)
f88df14b
DG
70#define USER_REGION_ID (0UL)
71
d29eff7b 72/*
cec08e7a 73 * Defines the address of the vmemap area, in its own region
d29eff7b 74 */
cec08e7a
BH
75#define VMEMMAP_BASE (VMEMMAP_REGION_ID << REGION_SHIFT)
76#define vmemmap ((struct page *)VMEMMAP_BASE)
77
d29eff7b 78
f88df14b 79/*
c605782b 80 * Include the PTE bits definitions
f88df14b 81 */
c605782b 82#include <asm/pte-hash64.h>
71087002 83#include <asm/pte-common.h>
c605782b 84
f88df14b 85
94ee815c 86#ifdef CONFIG_PPC_MM_SLICES
f88df14b
DG
87#define HAVE_ARCH_UNMAPPED_AREA
88#define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
94ee815c 89#endif /* CONFIG_PPC_MM_SLICES */
f88df14b
DG
90
91#ifndef __ASSEMBLY__
92
c605782b
BH
93/*
94 * This is the default implementation of various PTE accessors, it's
95 * used in all cases except Book3S with 64K pages where we have a
96 * concept of sub-pages
97 */
98#ifndef __real_pte
99
100#ifdef STRICT_MM_TYPECHECKS
101#define __real_pte(e,p) ((real_pte_t){(e)})
102#define __rpte_to_pte(r) ((r).pte)
103#else
104#define __real_pte(e,p) (e)
105#define __rpte_to_pte(r) (__pte(r))
106#endif
107#define __rpte_to_hidx(r,index) (pte_val(__rpte_to_pte(r)) >> 12)
108
109#define pte_iterate_hashed_subpages(rpte, psize, va, index, shift) \
110 do { \
111 index = 0; \
112 shift = mmu_psize_defs[psize].shift; \
113
114#define pte_iterate_hashed_end() } while(0)
115
116#ifdef CONFIG_PPC_HAS_HASH_64K
117#define pte_pagesize_index(mm, addr, pte) get_slice_psize(mm, addr)
118#else
119#define pte_pagesize_index(mm, addr, pte) MMU_PAGE_4K
120#endif
121
122#endif /* __real_pte */
123
124
f88df14b
DG
125/* pte_clear moved to later in this file */
126
f88df14b
DG
127#define PMD_BAD_BITS (PTE_TABLE_SIZE-1)
128#define PUD_BAD_BITS (PMD_TABLE_SIZE-1)
129
130#define pmd_set(pmdp, pmdval) (pmd_val(*(pmdp)) = (pmdval))
131#define pmd_none(pmd) (!pmd_val(pmd))
132#define pmd_bad(pmd) (!is_kernel_addr(pmd_val(pmd)) \
133 || (pmd_val(pmd) & PMD_BAD_BITS))
134#define pmd_present(pmd) (pmd_val(pmd) != 0)
135#define pmd_clear(pmdp) (pmd_val(*(pmdp)) = 0)
136#define pmd_page_vaddr(pmd) (pmd_val(pmd) & ~PMD_MASKED_BITS)
137#define pmd_page(pmd) virt_to_page(pmd_page_vaddr(pmd))
138
139#define pud_set(pudp, pudval) (pud_val(*(pudp)) = (pudval))
140#define pud_none(pud) (!pud_val(pud))
141#define pud_bad(pud) (!is_kernel_addr(pud_val(pud)) \
142 || (pud_val(pud) & PUD_BAD_BITS))
143#define pud_present(pud) (pud_val(pud) != 0)
144#define pud_clear(pudp) (pud_val(*(pudp)) = 0)
145#define pud_page_vaddr(pud) (pud_val(pud) & ~PUD_MASKED_BITS)
146#define pud_page(pud) virt_to_page(pud_page_vaddr(pud))
147
148#define pgd_set(pgdp, pudp) ({pgd_val(*(pgdp)) = (unsigned long)(pudp);})
149
150/*
151 * Find an entry in a page-table-directory. We combine the address region
152 * (the high order N bits) and the pgd portion of the address.
153 */
154/* to avoid overflow in free_pgtables we don't use PTRS_PER_PGD here */
155#define pgd_index(address) (((address) >> (PGDIR_SHIFT)) & 0x1ff)
156
157#define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
158
159#define pmd_offset(pudp,addr) \
160 (((pmd_t *) pud_page_vaddr(*(pudp))) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
161
162#define pte_offset_kernel(dir,addr) \
163 (((pte_t *) pmd_page_vaddr(*(dir))) + (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)))
164
165#define pte_offset_map(dir,addr) pte_offset_kernel((dir), (addr))
166#define pte_offset_map_nested(dir,addr) pte_offset_kernel((dir), (addr))
167#define pte_unmap(pte) do { } while(0)
168#define pte_unmap_nested(pte) do { } while(0)
169
170/* to find an entry in a kernel page-table-directory */
171/* This now only contains the vmalloc pages */
172#define pgd_offset_k(address) pgd_offset(&init_mm, address)
173
f88df14b
DG
174
175/* Atomic PTE updates */
176static inline unsigned long pte_update(struct mm_struct *mm,
177 unsigned long addr,
178 pte_t *ptep, unsigned long clr,
179 int huge)
180{
a033a487 181#ifdef PTE_ATOMIC_UPDATES
f88df14b
DG
182 unsigned long old, tmp;
183
184 __asm__ __volatile__(
185 "1: ldarx %0,0,%3 # pte_update\n\
186 andi. %1,%0,%6\n\
187 bne- 1b \n\
188 andc %1,%0,%4 \n\
189 stdcx. %1,0,%3 \n\
190 bne- 1b"
191 : "=&r" (old), "=&r" (tmp), "=m" (*ptep)
192 : "r" (ptep), "r" (clr), "m" (*ptep), "i" (_PAGE_BUSY)
193 : "cc" );
a033a487
BH
194#else
195 unsigned long old = pte_val(*ptep);
196 *ptep = __pte(old & ~clr);
197#endif
8d30c14c
BH
198 /* huge pages use the old page table lock */
199 if (!huge)
200 assert_pte_locked(mm, addr);
201
f88df14b
DG
202 if (old & _PAGE_HASHPTE)
203 hpte_need_flush(mm, addr, ptep, old, huge);
204 return old;
205}
206
207static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
208 unsigned long addr, pte_t *ptep)
209{
210 unsigned long old;
211
212 if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
213 return 0;
214 old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0);
215 return (old & _PAGE_ACCESSED) != 0;
216}
217#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
218#define ptep_test_and_clear_young(__vma, __addr, __ptep) \
219({ \
220 int __r; \
221 __r = __ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
222 __r; \
223})
224
f88df14b
DG
225#define __HAVE_ARCH_PTEP_SET_WRPROTECT
226static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
227 pte_t *ptep)
228{
229 unsigned long old;
230
231 if ((pte_val(*ptep) & _PAGE_RW) == 0)
232 return;
233 old = pte_update(mm, addr, ptep, _PAGE_RW, 0);
234}
235
016b33c4
AW
236static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
237 unsigned long addr, pte_t *ptep)
238{
239 unsigned long old;
240
86df8642
DG
241 if ((pte_val(*ptep) & _PAGE_RW) == 0)
242 return;
016b33c4
AW
243 old = pte_update(mm, addr, ptep, _PAGE_RW, 1);
244}
f88df14b
DG
245
246/*
247 * We currently remove entries from the hashtable regardless of whether
248 * the entry was young or dirty. The generic routines only flush if the
249 * entry was young or dirty which is not good enough.
250 *
251 * We should be more intelligent about this but for the moment we override
252 * these functions and force a tlb flush unconditionally
253 */
254#define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
255#define ptep_clear_flush_young(__vma, __address, __ptep) \
256({ \
257 int __young = __ptep_test_and_clear_young((__vma)->vm_mm, __address, \
258 __ptep); \
259 __young; \
260})
261
f88df14b
DG
262#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
263static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
264 unsigned long addr, pte_t *ptep)
265{
266 unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0);
267 return __pte(old);
268}
269
270static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
271 pte_t * ptep)
272{
273 pte_update(mm, addr, ptep, ~0UL, 0);
274}
275
f88df14b
DG
276
277/* Set the dirty and/or accessed bits atomically in a linux PTE, this
278 * function doesn't need to flush the hash entry
279 */
8d30c14c 280static inline void __ptep_set_access_flags(pte_t *ptep, pte_t entry)
f88df14b
DG
281{
282 unsigned long bits = pte_val(entry) &
8d1cf34e
BH
283 (_PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_RW |
284 _PAGE_EXEC | _PAGE_HWEXEC);
a033a487
BH
285
286#ifdef PTE_ATOMIC_UPDATES
f88df14b
DG
287 unsigned long old, tmp;
288
289 __asm__ __volatile__(
290 "1: ldarx %0,0,%4\n\
291 andi. %1,%0,%6\n\
292 bne- 1b \n\
293 or %0,%3,%0\n\
294 stdcx. %0,0,%4\n\
295 bne- 1b"
296 :"=&r" (old), "=&r" (tmp), "=m" (*ptep)
297 :"r" (bits), "r" (ptep), "m" (*ptep), "i" (_PAGE_BUSY)
298 :"cc");
a033a487
BH
299#else
300 unsigned long old = pte_val(*ptep);
301 *ptep = __pte(old | bits);
302#endif
f88df14b 303}
f88df14b 304
f88df14b
DG
305#define __HAVE_ARCH_PTE_SAME
306#define pte_same(A,B) (((pte_val(A) ^ pte_val(B)) & ~_PAGE_HPTEFLAGS) == 0)
307
308#define pte_ERROR(e) \
309 printk("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
310#define pmd_ERROR(e) \
311 printk("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
312#define pgd_ERROR(e) \
313 printk("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
314
f88df14b
DG
315/* Encode and de-code a swap entry */
316#define __swp_type(entry) (((entry).val >> 1) & 0x3f)
317#define __swp_offset(entry) ((entry).val >> 8)
318#define __swp_entry(type, offset) ((swp_entry_t){((type)<< 1)|((offset)<<8)})
319#define __pte_to_swp_entry(pte) ((swp_entry_t){pte_val(pte) >> PTE_RPN_SHIFT})
320#define __swp_entry_to_pte(x) ((pte_t) { (x).val << PTE_RPN_SHIFT })
321#define pte_to_pgoff(pte) (pte_val(pte) >> PTE_RPN_SHIFT)
322#define pgoff_to_pte(off) ((pte_t) {((off) << PTE_RPN_SHIFT)|_PAGE_FILE})
323#define PTE_FILE_MAX_BITS (BITS_PER_LONG - PTE_RPN_SHIFT)
324
f88df14b
DG
325void pgtable_cache_init(void);
326
327/*
328 * find_linux_pte returns the address of a linux pte for a given
329 * effective address and directory. If not found, it returns zero.
330 */static inline pte_t *find_linux_pte(pgd_t *pgdir, unsigned long ea)
331{
332 pgd_t *pg;
333 pud_t *pu;
334 pmd_t *pm;
335 pte_t *pt = NULL;
336
337 pg = pgdir + pgd_index(ea);
338 if (!pgd_none(*pg)) {
339 pu = pud_offset(pg, ea);
340 if (!pud_none(*pu)) {
341 pm = pmd_offset(pu, ea);
342 if (pmd_present(*pm))
343 pt = pte_offset_kernel(pm, ea);
344 }
345 }
346 return pt;
347}
348
ce0ad7f0
NP
349pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long address);
350
f88df14b
DG
351#endif /* __ASSEMBLY__ */
352
353#endif /* _ASM_POWERPC_PGTABLE_PPC64_H_ */
This page took 0.229635 seconds and 5 git commands to generate.