aboutsummaryrefslogtreecommitdiffstats
path: root/include/asm-powerpc/pgtable.h
diff options
context:
space:
mode:
Diffstat (limited to 'include/asm-powerpc/pgtable.h')
-rw-r--r--include/asm-powerpc/pgtable.h524
1 files changed, 524 insertions, 0 deletions
diff --git a/include/asm-powerpc/pgtable.h b/include/asm-powerpc/pgtable.h
new file mode 100644
index 000000000000..0303f57366c1
--- /dev/null
+++ b/include/asm-powerpc/pgtable.h
@@ -0,0 +1,524 @@
1#ifndef _ASM_POWERPC_PGTABLE_H
2#define _ASM_POWERPC_PGTABLE_H
3
4#ifndef CONFIG_PPC64
5#include <asm-ppc/pgtable.h>
6#else
7
8/*
9 * This file contains the functions and defines necessary to modify and use
10 * the ppc64 hashed page table.
11 */
12
13#ifndef __ASSEMBLY__
14#include <linux/config.h>
15#include <linux/stddef.h>
16#include <asm/processor.h> /* For TASK_SIZE */
17#include <asm/mmu.h>
18#include <asm/page.h>
19#include <asm/tlbflush.h>
20struct mm_struct;
21#endif /* __ASSEMBLY__ */
22
23#ifdef CONFIG_PPC_64K_PAGES
24#include <asm/pgtable-64k.h>
25#else
26#include <asm/pgtable-4k.h>
27#endif
28
29#define FIRST_USER_ADDRESS 0
30
31/*
32 * Size of EA range mapped by our pagetables.
33 */
34#define PGTABLE_EADDR_SIZE (PTE_INDEX_SIZE + PMD_INDEX_SIZE + \
35 PUD_INDEX_SIZE + PGD_INDEX_SIZE + PAGE_SHIFT)
36#define PGTABLE_RANGE (1UL << PGTABLE_EADDR_SIZE)
37
38#if TASK_SIZE_USER64 > PGTABLE_RANGE
39#error TASK_SIZE_USER64 exceeds pagetable range
40#endif
41
42#if TASK_SIZE_USER64 > (1UL << (USER_ESID_BITS + SID_SHIFT))
43#error TASK_SIZE_USER64 exceeds user VSID range
44#endif
45
46/*
47 * Define the address range of the vmalloc VM area.
48 */
49#define VMALLOC_START (0xD000000000000000ul)
50#define VMALLOC_SIZE (0x80000000000UL)
51#define VMALLOC_END (VMALLOC_START + VMALLOC_SIZE)
52
53/*
54 * Define the address range of the imalloc VM area.
55 */
56#define PHBS_IO_BASE VMALLOC_END
57#define IMALLOC_BASE (PHBS_IO_BASE + 0x80000000ul) /* Reserve 2 gigs for PHBs */
58#define IMALLOC_END (VMALLOC_START + PGTABLE_RANGE)
59
60/*
61 * Common bits in a linux-style PTE. These match the bits in the
62 * (hardware-defined) PowerPC PTE as closely as possible. Additional
63 * bits may be defined in pgtable-*.h
64 */
65#define _PAGE_PRESENT 0x0001 /* software: pte contains a translation */
66#define _PAGE_USER 0x0002 /* matches one of the PP bits */
67#define _PAGE_FILE 0x0002 /* (!present only) software: pte holds file offset */
68#define _PAGE_EXEC 0x0004 /* No execute on POWER4 and newer (we invert) */
69#define _PAGE_GUARDED 0x0008
70#define _PAGE_COHERENT 0x0010 /* M: enforce memory coherence (SMP systems) */
71#define _PAGE_NO_CACHE 0x0020 /* I: cache inhibit */
72#define _PAGE_WRITETHRU 0x0040 /* W: cache write-through */
73#define _PAGE_DIRTY 0x0080 /* C: page changed */
74#define _PAGE_ACCESSED 0x0100 /* R: page referenced */
75#define _PAGE_RW 0x0200 /* software: user write access allowed */
76#define _PAGE_HASHPTE 0x0400 /* software: pte has an associated HPTE */
77#define _PAGE_BUSY 0x0800 /* software: PTE & hash are busy */
78
79#define _PAGE_BASE (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_COHERENT)
80
81#define _PAGE_WRENABLE (_PAGE_RW | _PAGE_DIRTY)
82
83/* __pgprot defined in asm-powerpc/page.h */
84#define PAGE_NONE __pgprot(_PAGE_PRESENT | _PAGE_ACCESSED)
85
86#define PAGE_SHARED __pgprot(_PAGE_BASE | _PAGE_RW | _PAGE_USER)
87#define PAGE_SHARED_X __pgprot(_PAGE_BASE | _PAGE_RW | _PAGE_USER | _PAGE_EXEC)
88#define PAGE_COPY __pgprot(_PAGE_BASE | _PAGE_USER)
89#define PAGE_COPY_X __pgprot(_PAGE_BASE | _PAGE_USER | _PAGE_EXEC)
90#define PAGE_READONLY __pgprot(_PAGE_BASE | _PAGE_USER)
91#define PAGE_READONLY_X __pgprot(_PAGE_BASE | _PAGE_USER | _PAGE_EXEC)
92#define PAGE_KERNEL __pgprot(_PAGE_BASE | _PAGE_WRENABLE)
93#define PAGE_KERNEL_CI __pgprot(_PAGE_PRESENT | _PAGE_ACCESSED | \
94 _PAGE_WRENABLE | _PAGE_NO_CACHE | _PAGE_GUARDED)
95#define PAGE_KERNEL_EXEC __pgprot(_PAGE_BASE | _PAGE_WRENABLE | _PAGE_EXEC)
96
97#define PAGE_AGP __pgprot(_PAGE_BASE | _PAGE_WRENABLE | _PAGE_NO_CACHE)
98#define HAVE_PAGE_AGP
99
100/* PTEIDX nibble */
101#define _PTEIDX_SECONDARY 0x8
102#define _PTEIDX_GROUP_IX 0x7
103
104
105/*
106 * POWER4 and newer have per page execute protection, older chips can only
107 * do this on a segment (256MB) basis.
108 *
109 * Also, write permissions imply read permissions.
110 * This is the closest we can get..
111 *
112 * Note due to the way vm flags are laid out, the bits are XWR
113 */
114#define __P000 PAGE_NONE
115#define __P001 PAGE_READONLY
116#define __P010 PAGE_COPY
117#define __P011 PAGE_COPY
118#define __P100 PAGE_READONLY_X
119#define __P101 PAGE_READONLY_X
120#define __P110 PAGE_COPY_X
121#define __P111 PAGE_COPY_X
122
123#define __S000 PAGE_NONE
124#define __S001 PAGE_READONLY
125#define __S010 PAGE_SHARED
126#define __S011 PAGE_SHARED
127#define __S100 PAGE_READONLY_X
128#define __S101 PAGE_READONLY_X
129#define __S110 PAGE_SHARED_X
130#define __S111 PAGE_SHARED_X
131
132#ifndef __ASSEMBLY__
133
134/*
135 * ZERO_PAGE is a global shared page that is always zero: used
136 * for zero-mapped memory areas etc..
137 */
138extern unsigned long empty_zero_page[PAGE_SIZE/sizeof(unsigned long)];
139#define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
140#endif /* __ASSEMBLY__ */
141
142#ifdef CONFIG_HUGETLB_PAGE
143
144#define HAVE_ARCH_UNMAPPED_AREA
145#define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
146
147#endif
148
149#ifndef __ASSEMBLY__
150
151/*
152 * Conversion functions: convert a page and protection to a page entry,
153 * and a page entry and page directory to the page they refer to.
154 *
155 * mk_pte takes a (struct page *) as input
156 */
157#define mk_pte(page, pgprot) pfn_pte(page_to_pfn(page), (pgprot))
158
159static inline pte_t pfn_pte(unsigned long pfn, pgprot_t pgprot)
160{
161 pte_t pte;
162
163
164 pte_val(pte) = (pfn << PTE_RPN_SHIFT) | pgprot_val(pgprot);
165 return pte;
166}
167
168#define pte_modify(_pte, newprot) \
169 (__pte((pte_val(_pte) & _PAGE_CHG_MASK) | pgprot_val(newprot)))
170
171#define pte_none(pte) ((pte_val(pte) & ~_PAGE_HPTEFLAGS) == 0)
172#define pte_present(pte) (pte_val(pte) & _PAGE_PRESENT)
173
174/* pte_clear moved to later in this file */
175
176#define pte_pfn(x) ((unsigned long)((pte_val(x)>>PTE_RPN_SHIFT)))
177#define pte_page(x) pfn_to_page(pte_pfn(x))
178
179#define pmd_set(pmdp, pmdval) (pmd_val(*(pmdp)) = (pmdval))
180#define pmd_none(pmd) (!pmd_val(pmd))
181#define pmd_bad(pmd) (pmd_val(pmd) == 0)
182#define pmd_present(pmd) (pmd_val(pmd) != 0)
183#define pmd_clear(pmdp) (pmd_val(*(pmdp)) = 0)
184#define pmd_page_kernel(pmd) (pmd_val(pmd) & ~PMD_MASKED_BITS)
185#define pmd_page(pmd) virt_to_page(pmd_page_kernel(pmd))
186
187#define pud_set(pudp, pudval) (pud_val(*(pudp)) = (pudval))
188#define pud_none(pud) (!pud_val(pud))
189#define pud_bad(pud) ((pud_val(pud)) == 0)
190#define pud_present(pud) (pud_val(pud) != 0)
191#define pud_clear(pudp) (pud_val(*(pudp)) = 0)
192#define pud_page(pud) (pud_val(pud) & ~PUD_MASKED_BITS)
193
194#define pgd_set(pgdp, pudp) ({pgd_val(*(pgdp)) = (unsigned long)(pudp);})
195
196/*
197 * Find an entry in a page-table-directory. We combine the address region
198 * (the high order N bits) and the pgd portion of the address.
199 */
200/* to avoid overflow in free_pgtables we don't use PTRS_PER_PGD here */
201#define pgd_index(address) (((address) >> (PGDIR_SHIFT)) & 0x1ff)
202
203#define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
204
205#define pmd_offset(pudp,addr) \
206 (((pmd_t *) pud_page(*(pudp))) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
207
208#define pte_offset_kernel(dir,addr) \
209 (((pte_t *) pmd_page_kernel(*(dir))) + (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)))
210
211#define pte_offset_map(dir,addr) pte_offset_kernel((dir), (addr))
212#define pte_offset_map_nested(dir,addr) pte_offset_kernel((dir), (addr))
213#define pte_unmap(pte) do { } while(0)
214#define pte_unmap_nested(pte) do { } while(0)
215
216/* to find an entry in a kernel page-table-directory */
217/* This now only contains the vmalloc pages */
218#define pgd_offset_k(address) pgd_offset(&init_mm, address)
219
220/*
221 * The following only work if pte_present() is true.
222 * Undefined behaviour if not..
223 */
224static inline int pte_read(pte_t pte) { return pte_val(pte) & _PAGE_USER;}
225static inline int pte_write(pte_t pte) { return pte_val(pte) & _PAGE_RW;}
226static inline int pte_exec(pte_t pte) { return pte_val(pte) & _PAGE_EXEC;}
227static inline int pte_dirty(pte_t pte) { return pte_val(pte) & _PAGE_DIRTY;}
228static inline int pte_young(pte_t pte) { return pte_val(pte) & _PAGE_ACCESSED;}
229static inline int pte_file(pte_t pte) { return pte_val(pte) & _PAGE_FILE;}
230
231static inline void pte_uncache(pte_t pte) { pte_val(pte) |= _PAGE_NO_CACHE; }
232static inline void pte_cache(pte_t pte) { pte_val(pte) &= ~_PAGE_NO_CACHE; }
233
234static inline pte_t pte_rdprotect(pte_t pte) {
235 pte_val(pte) &= ~_PAGE_USER; return pte; }
236static inline pte_t pte_exprotect(pte_t pte) {
237 pte_val(pte) &= ~_PAGE_EXEC; return pte; }
238static inline pte_t pte_wrprotect(pte_t pte) {
239 pte_val(pte) &= ~(_PAGE_RW); return pte; }
240static inline pte_t pte_mkclean(pte_t pte) {
241 pte_val(pte) &= ~(_PAGE_DIRTY); return pte; }
242static inline pte_t pte_mkold(pte_t pte) {
243 pte_val(pte) &= ~_PAGE_ACCESSED; return pte; }
244static inline pte_t pte_mkread(pte_t pte) {
245 pte_val(pte) |= _PAGE_USER; return pte; }
246static inline pte_t pte_mkexec(pte_t pte) {
247 pte_val(pte) |= _PAGE_USER | _PAGE_EXEC; return pte; }
248static inline pte_t pte_mkwrite(pte_t pte) {
249 pte_val(pte) |= _PAGE_RW; return pte; }
250static inline pte_t pte_mkdirty(pte_t pte) {
251 pte_val(pte) |= _PAGE_DIRTY; return pte; }
252static inline pte_t pte_mkyoung(pte_t pte) {
253 pte_val(pte) |= _PAGE_ACCESSED; return pte; }
254static inline pte_t pte_mkhuge(pte_t pte) {
255 return pte; }
256
257/* Atomic PTE updates */
258static inline unsigned long pte_update(pte_t *p, unsigned long clr)
259{
260 unsigned long old, tmp;
261
262 __asm__ __volatile__(
263 "1: ldarx %0,0,%3 # pte_update\n\
264 andi. %1,%0,%6\n\
265 bne- 1b \n\
266 andc %1,%0,%4 \n\
267 stdcx. %1,0,%3 \n\
268 bne- 1b"
269 : "=&r" (old), "=&r" (tmp), "=m" (*p)
270 : "r" (p), "r" (clr), "m" (*p), "i" (_PAGE_BUSY)
271 : "cc" );
272 return old;
273}
274
275/* PTE updating functions, this function puts the PTE in the
276 * batch, doesn't actually triggers the hash flush immediately,
277 * you need to call flush_tlb_pending() to do that.
278 * Pass -1 for "normal" size (4K or 64K)
279 */
280extern void hpte_update(struct mm_struct *mm, unsigned long addr,
281 pte_t *ptep, unsigned long pte, int huge);
282
283static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
284 unsigned long addr, pte_t *ptep)
285{
286 unsigned long old;
287
288 if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
289 return 0;
290 old = pte_update(ptep, _PAGE_ACCESSED);
291 if (old & _PAGE_HASHPTE) {
292 hpte_update(mm, addr, ptep, old, 0);
293 flush_tlb_pending();
294 }
295 return (old & _PAGE_ACCESSED) != 0;
296}
297#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
298#define ptep_test_and_clear_young(__vma, __addr, __ptep) \
299({ \
300 int __r; \
301 __r = __ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
302 __r; \
303})
304
305/*
306 * On RW/DIRTY bit transitions we can avoid flushing the hpte. For the
307 * moment we always flush but we need to fix hpte_update and test if the
308 * optimisation is worth it.
309 */
310static inline int __ptep_test_and_clear_dirty(struct mm_struct *mm,
311 unsigned long addr, pte_t *ptep)
312{
313 unsigned long old;
314
315 if ((pte_val(*ptep) & _PAGE_DIRTY) == 0)
316 return 0;
317 old = pte_update(ptep, _PAGE_DIRTY);
318 if (old & _PAGE_HASHPTE)
319 hpte_update(mm, addr, ptep, old, 0);
320 return (old & _PAGE_DIRTY) != 0;
321}
322#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
323#define ptep_test_and_clear_dirty(__vma, __addr, __ptep) \
324({ \
325 int __r; \
326 __r = __ptep_test_and_clear_dirty((__vma)->vm_mm, __addr, __ptep); \
327 __r; \
328})
329
330#define __HAVE_ARCH_PTEP_SET_WRPROTECT
331static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
332 pte_t *ptep)
333{
334 unsigned long old;
335
336 if ((pte_val(*ptep) & _PAGE_RW) == 0)
337 return;
338 old = pte_update(ptep, _PAGE_RW);
339 if (old & _PAGE_HASHPTE)
340 hpte_update(mm, addr, ptep, old, 0);
341}
342
343/*
344 * We currently remove entries from the hashtable regardless of whether
345 * the entry was young or dirty. The generic routines only flush if the
346 * entry was young or dirty which is not good enough.
347 *
348 * We should be more intelligent about this but for the moment we override
349 * these functions and force a tlb flush unconditionally
350 */
351#define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
352#define ptep_clear_flush_young(__vma, __address, __ptep) \
353({ \
354 int __young = __ptep_test_and_clear_young((__vma)->vm_mm, __address, \
355 __ptep); \
356 __young; \
357})
358
359#define __HAVE_ARCH_PTEP_CLEAR_DIRTY_FLUSH
360#define ptep_clear_flush_dirty(__vma, __address, __ptep) \
361({ \
362 int __dirty = __ptep_test_and_clear_dirty((__vma)->vm_mm, __address, \
363 __ptep); \
364 flush_tlb_page(__vma, __address); \
365 __dirty; \
366})
367
368#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
369static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
370 unsigned long addr, pte_t *ptep)
371{
372 unsigned long old = pte_update(ptep, ~0UL);
373
374 if (old & _PAGE_HASHPTE)
375 hpte_update(mm, addr, ptep, old, 0);
376 return __pte(old);
377}
378
379static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
380 pte_t * ptep)
381{
382 unsigned long old = pte_update(ptep, ~0UL);
383
384 if (old & _PAGE_HASHPTE)
385 hpte_update(mm, addr, ptep, old, 0);
386}
387
388/*
389 * set_pte stores a linux PTE into the linux page table.
390 */
391static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
392 pte_t *ptep, pte_t pte)
393{
394 if (pte_present(*ptep)) {
395 pte_clear(mm, addr, ptep);
396 flush_tlb_pending();
397 }
398 pte = __pte(pte_val(pte) & ~_PAGE_HPTEFLAGS);
399
400#ifdef CONFIG_PPC_64K_PAGES
401 if (mmu_virtual_psize != MMU_PAGE_64K)
402 pte = __pte(pte_val(pte) | _PAGE_COMBO);
403#endif /* CONFIG_PPC_64K_PAGES */
404
405 *ptep = pte;
406}
407
408/* Set the dirty and/or accessed bits atomically in a linux PTE, this
409 * function doesn't need to flush the hash entry
410 */
411#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
412static inline void __ptep_set_access_flags(pte_t *ptep, pte_t entry, int dirty)
413{
414 unsigned long bits = pte_val(entry) &
415 (_PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_RW | _PAGE_EXEC);
416 unsigned long old, tmp;
417
418 __asm__ __volatile__(
419 "1: ldarx %0,0,%4\n\
420 andi. %1,%0,%6\n\
421 bne- 1b \n\
422 or %0,%3,%0\n\
423 stdcx. %0,0,%4\n\
424 bne- 1b"
425 :"=&r" (old), "=&r" (tmp), "=m" (*ptep)
426 :"r" (bits), "r" (ptep), "m" (*ptep), "i" (_PAGE_BUSY)
427 :"cc");
428}
429#define ptep_set_access_flags(__vma, __address, __ptep, __entry, __dirty) \
430 do { \
431 __ptep_set_access_flags(__ptep, __entry, __dirty); \
432 flush_tlb_page_nohash(__vma, __address); \
433 } while(0)
434
435/*
436 * Macro to mark a page protection value as "uncacheable".
437 */
438#define pgprot_noncached(prot) (__pgprot(pgprot_val(prot) | _PAGE_NO_CACHE | _PAGE_GUARDED))
439
440struct file;
441extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
442 unsigned long size, pgprot_t vma_prot);
443#define __HAVE_PHYS_MEM_ACCESS_PROT
444
445#define __HAVE_ARCH_PTE_SAME
446#define pte_same(A,B) (((pte_val(A) ^ pte_val(B)) & ~_PAGE_HPTEFLAGS) == 0)
447
448#define pte_ERROR(e) \
449 printk("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
450#define pmd_ERROR(e) \
451 printk("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
452#define pgd_ERROR(e) \
453 printk("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
454
455extern pgd_t swapper_pg_dir[];
456
457extern void paging_init(void);
458
459#ifdef CONFIG_HUGETLB_PAGE
460#define hugetlb_free_pgd_range(tlb, addr, end, floor, ceiling) \
461 free_pgd_range(tlb, addr, end, floor, ceiling)
462#endif
463
464/*
465 * This gets called at the end of handling a page fault, when
466 * the kernel has put a new PTE into the page table for the process.
467 * We use it to put a corresponding HPTE into the hash table
468 * ahead of time, instead of waiting for the inevitable extra
469 * hash-table miss exception.
470 */
471struct vm_area_struct;
472extern void update_mmu_cache(struct vm_area_struct *, unsigned long, pte_t);
473
474/* Encode and de-code a swap entry */
475#define __swp_type(entry) (((entry).val >> 1) & 0x3f)
476#define __swp_offset(entry) ((entry).val >> 8)
477#define __swp_entry(type, offset) ((swp_entry_t){((type)<< 1)|((offset)<<8)})
478#define __pte_to_swp_entry(pte) ((swp_entry_t){pte_val(pte) >> PTE_RPN_SHIFT})
479#define __swp_entry_to_pte(x) ((pte_t) { (x).val << PTE_RPN_SHIFT })
480#define pte_to_pgoff(pte) (pte_val(pte) >> PTE_RPN_SHIFT)
481#define pgoff_to_pte(off) ((pte_t) {((off) << PTE_RPN_SHIFT)|_PAGE_FILE})
482#define PTE_FILE_MAX_BITS (BITS_PER_LONG - PTE_RPN_SHIFT)
483
484/*
485 * kern_addr_valid is intended to indicate whether an address is a valid
486 * kernel address. Most 32-bit archs define it as always true (like this)
487 * but most 64-bit archs actually perform a test. What should we do here?
488 * The only use is in fs/ncpfs/dir.c
489 */
490#define kern_addr_valid(addr) (1)
491
492#define io_remap_pfn_range(vma, vaddr, pfn, size, prot) \
493 remap_pfn_range(vma, vaddr, pfn, size, prot)
494
495void pgtable_cache_init(void);
496
497/*
498 * find_linux_pte returns the address of a linux pte for a given
499 * effective address and directory. If not found, it returns zero.
500 */static inline pte_t *find_linux_pte(pgd_t *pgdir, unsigned long ea)
501{
502 pgd_t *pg;
503 pud_t *pu;
504 pmd_t *pm;
505 pte_t *pt = NULL;
506
507 pg = pgdir + pgd_index(ea);
508 if (!pgd_none(*pg)) {
509 pu = pud_offset(pg, ea);
510 if (!pud_none(*pu)) {
511 pm = pmd_offset(pu, ea);
512 if (pmd_present(*pm))
513 pt = pte_offset_kernel(pm, ea);
514 }
515 }
516 return pt;
517}
518
519#include <asm-generic/pgtable.h>
520
521#endif /* __ASSEMBLY__ */
522
523#endif /* CONFIG_PPC64 */
524#endif /* _ASM_POWERPC_PGTABLE_H */