1 #ifndef _PPC64_PGTABLE_H
2 #define _PPC64_PGTABLE_H
5 * This file contains the functions and defines necessary to modify and use
6 * the ppc64 hashed page table.
10 #include <linux/config.h>
11 #include <linux/stddef.h>
12 #include <asm/processor.h> /* For TASK_SIZE */
15 #include <asm/tlbflush.h>
16 #endif /* __ASSEMBLY__ */
18 /* PMD_SHIFT determines what a second-level page table entry can map */
19 #define PMD_SHIFT (PAGE_SHIFT + PAGE_SHIFT - 3)
20 #define PMD_SIZE (1UL << PMD_SHIFT)
21 #define PMD_MASK (~(PMD_SIZE-1))
23 /* PGDIR_SHIFT determines what a third-level page table entry can map */
24 #define PGDIR_SHIFT (PAGE_SHIFT + (PAGE_SHIFT - 3) + (PAGE_SHIFT - 2))
25 #define PGDIR_SIZE (1UL << PGDIR_SHIFT)
26 #define PGDIR_MASK (~(PGDIR_SIZE-1))
29 * Entries per page directory level. The PTE level must use a 64b record
30 * for each page table entry. The PMD and PGD level use a 32b record for
31 * each entry by assuming that each entry is page aligned.
33 #define PTE_INDEX_SIZE 9
34 #define PMD_INDEX_SIZE 10
35 #define PGD_INDEX_SIZE 10
37 #define PTRS_PER_PTE (1 << PTE_INDEX_SIZE)
38 #define PTRS_PER_PMD (1 << PMD_INDEX_SIZE)
39 #define PTRS_PER_PGD (1 << PGD_INDEX_SIZE)
41 #define USER_PTRS_PER_PGD (1024)
42 #define FIRST_USER_PGD_NR 0
44 #define EADDR_SIZE (PTE_INDEX_SIZE + PMD_INDEX_SIZE + \
45 PGD_INDEX_SIZE + PAGE_SHIFT)
48 * Define the address range of the vmalloc VM area.
50 #define VMALLOC_START (0xD000000000000000)
51 #define VMALLOC_END (VMALLOC_START + VALID_EA_BITS)
54 * Define the address range of the imalloc VM area.
57 #define IMALLOC_START (ioremap_bot)
58 #define IMALLOC_VMADDR(x) ((unsigned long)(x))
59 #define PHBS_IO_BASE (0xE000000000000000) /* Reserve 2 gigs for PHBs */
60 #define IMALLOC_BASE (0xE000000080000000)
61 #define IMALLOC_END (IMALLOC_BASE + VALID_EA_BITS)
64 * Define the address range mapped virt <-> physical
66 #define KRANGE_START KERNELBASE
67 #define KRANGE_END (KRANGE_START + VALID_EA_BITS)
70 * Define the user address range
72 #define USER_START (0UL)
73 #define USER_END (USER_START + VALID_EA_BITS)
77 * Bits in a linux-style PTE. These match the bits in the
78 * (hardware-defined) PowerPC PTE as closely as possible.
80 #define _PAGE_PRESENT 0x0001 /* software: pte contains a translation */
81 #define _PAGE_USER 0x0002 /* matches one of the PP bits */
82 #define _PAGE_FILE 0x0002 /* (!present only) software: pte holds file offset */
83 #define _PAGE_RW 0x0004 /* software: user write access allowed */
84 #define _PAGE_GUARDED 0x0008
85 #define _PAGE_COHERENT 0x0010 /* M: enforce memory coherence (SMP systems) */
86 #define _PAGE_NO_CACHE 0x0020 /* I: cache inhibit */
87 #define _PAGE_WRITETHRU 0x0040 /* W: cache write-through */
88 #define _PAGE_DIRTY 0x0080 /* C: page changed */
89 #define _PAGE_ACCESSED 0x0100 /* R: page referenced */
90 #define _PAGE_EXEC 0x0200 /* software: i-cache coherence required */
91 #define _PAGE_HASHPTE 0x0400 /* software: pte has an associated HPTE */
92 #define _PAGE_BUSY 0x0800 /* software: PTE & hash are busy */
93 #define _PAGE_SECONDARY 0x8000 /* software: HPTE is in secondary group */
94 #define _PAGE_GROUP_IX 0x7000 /* software: HPTE index within group */
95 /* Bits 0x7000 identify the index within an HPT Group */
96 #define _PAGE_HPTEFLAGS (_PAGE_BUSY | _PAGE_HASHPTE | _PAGE_SECONDARY | _PAGE_GROUP_IX)
97 /* PAGE_MASK gives the right answer below, but only by accident */
98 /* It should be preserving the high 48 bits and then specifically */
99 /* preserving _PAGE_SECONDARY | _PAGE_GROUP_IX */
100 #define _PAGE_CHG_MASK (PAGE_MASK | _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_HPTEFLAGS)
102 #define _PAGE_BASE (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_COHERENT)
104 #define _PAGE_WRENABLE (_PAGE_RW | _PAGE_DIRTY)
106 /* __pgprot defined in asm-ppc64/page.h */
107 #define PAGE_NONE __pgprot(_PAGE_PRESENT | _PAGE_ACCESSED)
109 #define PAGE_SHARED __pgprot(_PAGE_BASE | _PAGE_RW | _PAGE_USER)
110 #define PAGE_SHARED_X __pgprot(_PAGE_BASE | _PAGE_RW | _PAGE_USER | _PAGE_EXEC)
111 #define PAGE_COPY __pgprot(_PAGE_BASE | _PAGE_USER)
112 #define PAGE_COPY_X __pgprot(_PAGE_BASE | _PAGE_USER | _PAGE_EXEC)
113 #define PAGE_READONLY __pgprot(_PAGE_BASE | _PAGE_USER)
114 #define PAGE_READONLY_X __pgprot(_PAGE_BASE | _PAGE_USER | _PAGE_EXEC)
115 #define PAGE_KERNEL __pgprot(_PAGE_BASE | _PAGE_WRENABLE)
116 #define PAGE_KERNEL_CI __pgprot(_PAGE_PRESENT | _PAGE_ACCESSED | \
117 _PAGE_WRENABLE | _PAGE_NO_CACHE | _PAGE_GUARDED)
120 * The PowerPC can only do execute protection on a segment (256MB) basis,
121 * not on a page basis. So we consider execute permission the same as read.
122 * Also, write permissions imply read permissions.
123 * This is the closest we can get..
125 #define __P000 PAGE_NONE
126 #define __P001 PAGE_READONLY_X
127 #define __P010 PAGE_COPY
128 #define __P011 PAGE_COPY_X
129 #define __P100 PAGE_READONLY
130 #define __P101 PAGE_READONLY_X
131 #define __P110 PAGE_COPY
132 #define __P111 PAGE_COPY_X
134 #define __S000 PAGE_NONE
135 #define __S001 PAGE_READONLY_X
136 #define __S010 PAGE_SHARED
137 #define __S011 PAGE_SHARED_X
138 #define __S100 PAGE_READONLY
139 #define __S101 PAGE_READONLY_X
140 #define __S110 PAGE_SHARED
141 #define __S111 PAGE_SHARED_X
146 * ZERO_PAGE is a global shared page that is always zero: used
147 * for zero-mapped memory areas etc..
149 extern unsigned long empty_zero_page[PAGE_SIZE/sizeof(unsigned long)];
150 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
151 #endif /* __ASSEMBLY__ */
153 /* shift to put page number into pte */
154 #define PTE_SHIFT (16)
156 /* We allow 2^41 bytes of real memory, so we need 29 bits in the PMD
157 * to give the PTE page number. The bottom two bits are for flags. */
158 #define PMD_TO_PTEPAGE_SHIFT (2)
160 #ifdef CONFIG_HUGETLB_PAGE
161 #define _PMD_HUGEPAGE 0x00000001U
162 #define HUGEPTE_BATCH_SIZE (1<<(HPAGE_SHIFT-PMD_SHIFT))
165 int hash_huge_page(struct mm_struct *mm, unsigned long access,
166 unsigned long ea, unsigned long vsid, int local);
167 #endif /* __ASSEMBLY__ */
169 #define HAVE_ARCH_UNMAPPED_AREA
172 #define hash_huge_page(mm,a,ea,vsid,local) -1
173 #define _PMD_HUGEPAGE 0
180 * Conversion functions: convert a page and protection to a page entry,
181 * and a page entry and page directory to the page they refer to.
183 * mk_pte takes a (struct page *) as input
185 #define mk_pte(page, pgprot) pfn_pte(page_to_pfn(page), (pgprot))
187 #define pfn_pte(pfn,pgprot) \
190 pte_val(pte) = ((unsigned long)(pfn) << PTE_SHIFT) | \
191 pgprot_val(pgprot); \
195 #define pte_modify(_pte, newprot) \
196 (__pte((pte_val(_pte) & _PAGE_CHG_MASK) | pgprot_val(newprot)))
198 #define pte_none(pte) ((pte_val(pte) & ~_PAGE_HPTEFLAGS) == 0)
199 #define pte_present(pte) (pte_val(pte) & _PAGE_PRESENT)
201 /* pte_clear moved to later in this file */
203 #define pte_pfn(x) ((unsigned long)((pte_val(x) >> PTE_SHIFT)))
204 #define pte_page(x) pfn_to_page(pte_pfn(x))
206 #define pmd_set(pmdp, ptep) \
207 (pmd_val(*(pmdp)) = (__ba_to_bpn(ptep) << PMD_TO_PTEPAGE_SHIFT))
208 #define pmd_none(pmd) (!pmd_val(pmd))
209 #define pmd_hugepage(pmd) (!!(pmd_val(pmd) & _PMD_HUGEPAGE))
210 #define pmd_bad(pmd) (((pmd_val(pmd)) == 0) || pmd_hugepage(pmd))
211 #define pmd_present(pmd) ((!pmd_hugepage(pmd)) \
212 && (pmd_val(pmd) & ~_PMD_HUGEPAGE) != 0)
213 #define pmd_clear(pmdp) (pmd_val(*(pmdp)) = 0)
214 #define pmd_page_kernel(pmd) \
215 (__bpn_to_ba(pmd_val(pmd) >> PMD_TO_PTEPAGE_SHIFT))
216 #define pmd_page(pmd) virt_to_page(pmd_page_kernel(pmd))
217 #define pgd_set(pgdp, pmdp) (pgd_val(*(pgdp)) = (__ba_to_bpn(pmdp)))
218 #define pgd_none(pgd) (!pgd_val(pgd))
219 #define pgd_bad(pgd) ((pgd_val(pgd)) == 0)
220 #define pgd_present(pgd) (pgd_val(pgd) != 0UL)
221 #define pgd_clear(pgdp) (pgd_val(*(pgdp)) = 0UL)
222 #define pgd_page(pgd) (__bpn_to_ba(pgd_val(pgd)))
225 * Find an entry in a page-table-directory. We combine the address region
226 * (the high order N bits) and the pgd portion of the address.
228 /* to avoid overflow in free_pgtables we don't use PTRS_PER_PGD here */
229 #define pgd_index(address) (((address) >> (PGDIR_SHIFT)) & 0x7ff)
231 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
233 /* Find an entry in the second-level page table.. */
234 #define pmd_offset(dir,addr) \
235 ((pmd_t *) pgd_page(*(dir)) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
237 /* Find an entry in the third-level page table.. */
238 #define pte_offset_kernel(dir,addr) \
239 ((pte_t *) pmd_page_kernel(*(dir)) + (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)))
241 #define pte_offset_map(dir,addr) pte_offset_kernel((dir), (addr))
242 #define pte_offset_map_nested(dir,addr) pte_offset_kernel((dir), (addr))
243 #define pte_unmap(pte) do { } while(0)
244 #define pte_unmap_nested(pte) do { } while(0)
246 /* to find an entry in a kernel page-table-directory */
247 /* This now only contains the vmalloc pages */
248 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
250 /* to find an entry in the ioremap page-table-directory */
251 #define pgd_offset_i(address) (ioremap_pgd + pgd_index(address))
253 #define pages_to_mb(x) ((x) >> (20-PAGE_SHIFT))
256 * The following only work if pte_present() is true.
257 * Undefined behaviour if not..
259 static inline int pte_read(pte_t pte) { return pte_val(pte) & _PAGE_USER;}
260 static inline int pte_write(pte_t pte) { return pte_val(pte) & _PAGE_RW;}
261 static inline int pte_exec(pte_t pte) { return pte_val(pte) & _PAGE_EXEC;}
262 static inline int pte_dirty(pte_t pte) { return pte_val(pte) & _PAGE_DIRTY;}
263 static inline int pte_young(pte_t pte) { return pte_val(pte) & _PAGE_ACCESSED;}
264 static inline int pte_file(pte_t pte) { return pte_val(pte) & _PAGE_FILE;}
266 static inline void pte_uncache(pte_t pte) { pte_val(pte) |= _PAGE_NO_CACHE; }
267 static inline void pte_cache(pte_t pte) { pte_val(pte) &= ~_PAGE_NO_CACHE; }
269 static inline pte_t pte_rdprotect(pte_t pte) {
270 pte_val(pte) &= ~_PAGE_USER; return pte; }
271 static inline pte_t pte_exprotect(pte_t pte) {
272 pte_val(pte) &= ~_PAGE_EXEC; return pte; }
273 static inline pte_t pte_wrprotect(pte_t pte) {
274 pte_val(pte) &= ~(_PAGE_RW); return pte; }
275 static inline pte_t pte_mkclean(pte_t pte) {
276 pte_val(pte) &= ~(_PAGE_DIRTY); return pte; }
277 static inline pte_t pte_mkold(pte_t pte) {
278 pte_val(pte) &= ~_PAGE_ACCESSED; return pte; }
280 static inline pte_t pte_mkread(pte_t pte) {
281 pte_val(pte) |= _PAGE_USER; return pte; }
282 static inline pte_t pte_mkexec(pte_t pte) {
283 pte_val(pte) |= _PAGE_USER | _PAGE_EXEC; return pte; }
284 static inline pte_t pte_mkwrite(pte_t pte) {
285 pte_val(pte) |= _PAGE_RW; return pte; }
286 static inline pte_t pte_mkdirty(pte_t pte) {
287 pte_val(pte) |= _PAGE_DIRTY; return pte; }
288 static inline pte_t pte_mkyoung(pte_t pte) {
289 pte_val(pte) |= _PAGE_ACCESSED; return pte; }
291 /* Atomic PTE updates */
292 static inline unsigned long pte_update(pte_t *p, unsigned long clr)
294 unsigned long old, tmp;
296 __asm__ __volatile__(
297 "1: ldarx %0,0,%3 # pte_update\n\
303 : "=&r" (old), "=&r" (tmp), "=m" (*p)
304 : "r" (p), "r" (clr), "m" (*p), "i" (_PAGE_BUSY)
309 /* PTE updating functions */
310 extern void hpte_update(pte_t *ptep, unsigned long pte, int wrprot);
312 static inline int ptep_test_and_clear_young(pte_t *ptep)
316 if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
318 old = pte_update(ptep, _PAGE_ACCESSED);
319 if (old & _PAGE_HASHPTE) {
320 hpte_update(ptep, old, 0);
321 flush_tlb_pending(); /* XXX generic code doesn't flush */
323 return (old & _PAGE_ACCESSED) != 0;
327 * On RW/DIRTY bit transitions we can avoid flushing the hpte. For the
328 * moment we always flush but we need to fix hpte_update and test if the
329 * optimisation is worth it.
331 static inline int ptep_test_and_clear_dirty(pte_t *ptep)
335 if ((pte_val(*ptep) & _PAGE_DIRTY) == 0)
337 old = pte_update(ptep, _PAGE_DIRTY);
338 if (old & _PAGE_HASHPTE)
339 hpte_update(ptep, old, 0);
340 return (old & _PAGE_DIRTY) != 0;
343 static inline void ptep_set_wrprotect(pte_t *ptep)
347 if ((pte_val(*ptep) & _PAGE_RW) == 0)
349 old = pte_update(ptep, _PAGE_RW);
350 if (old & _PAGE_HASHPTE)
351 hpte_update(ptep, old, 0);
355 * We currently remove entries from the hashtable regardless of whether
356 * the entry was young or dirty. The generic routines only flush if the
357 * entry was young or dirty which is not good enough.
359 * We should be more intelligent about this but for the moment we override
360 * these functions and force a tlb flush unconditionally
362 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
363 #define ptep_clear_flush_young(__vma, __address, __ptep) \
365 int __young = ptep_test_and_clear_young(__ptep); \
369 #define __HAVE_ARCH_PTEP_CLEAR_DIRTY_FLUSH
370 #define ptep_clear_flush_dirty(__vma, __address, __ptep) \
372 int __dirty = ptep_test_and_clear_dirty(__ptep); \
373 flush_tlb_page(__vma, __address); \
377 static inline pte_t ptep_get_and_clear(pte_t *ptep)
379 unsigned long old = pte_update(ptep, ~0UL);
381 if (old & _PAGE_HASHPTE)
382 hpte_update(ptep, old, 0);
386 static inline void pte_clear(pte_t * ptep)
388 unsigned long old = pte_update(ptep, ~0UL);
390 if (old & _PAGE_HASHPTE)
391 hpte_update(ptep, old, 0);
395 * set_pte stores a linux PTE into the linux page table.
397 static inline void set_pte(pte_t *ptep, pte_t pte)
399 if (pte_present(*ptep))
401 *ptep = __pte(pte_val(pte)) & ~_PAGE_HPTEFLAGS;
405 * Macro to mark a page protection value as "uncacheable".
407 #define pgprot_noncached(prot) (__pgprot(pgprot_val(prot) | _PAGE_NO_CACHE | _PAGE_GUARDED))
409 #define pte_same(A,B) (((pte_val(A) ^ pte_val(B)) & ~_PAGE_HPTEFLAGS) == 0)
411 extern unsigned long ioremap_bot, ioremap_base;
413 #define USER_PGD_PTRS (PAGE_OFFSET >> PGDIR_SHIFT)
414 #define KERNEL_PGD_PTRS (PTRS_PER_PGD-USER_PGD_PTRS)
416 #define pte_ERROR(e) \
417 printk("%s:%d: bad pte %016lx.\n", __FILE__, __LINE__, pte_val(e))
418 #define pmd_ERROR(e) \
419 printk("%s:%d: bad pmd %08x.\n", __FILE__, __LINE__, pmd_val(e))
420 #define pgd_ERROR(e) \
421 printk("%s:%d: bad pgd %08x.\n", __FILE__, __LINE__, pgd_val(e))
423 extern pgd_t swapper_pg_dir[1024];
424 extern pgd_t ioremap_dir[1024];
426 extern void paging_init(void);
429 * This gets called at the end of handling a page fault, when
430 * the kernel has put a new PTE into the page table for the process.
431 * We use it to put a corresponding HPTE into the hash table
432 * ahead of time, instead of waiting for the inevitable extra
433 * hash-table miss exception.
435 struct vm_area_struct;
436 extern void update_mmu_cache(struct vm_area_struct *, unsigned long, pte_t);
438 /* Encode and de-code a swap entry */
439 #define __swp_type(entry) (((entry).val >> 1) & 0x3f)
440 #define __swp_offset(entry) ((entry).val >> 8)
441 #define __swp_entry(type, offset) ((swp_entry_t) { ((type) << 1) | ((offset) << 8) })
442 #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) >> PTE_SHIFT })
443 #define __swp_entry_to_pte(x) ((pte_t) { (x).val << PTE_SHIFT })
444 #define pte_to_pgoff(pte) (pte_val(pte) >> PTE_SHIFT)
445 #define pgoff_to_pte(off) ((pte_t) {((off) << PTE_SHIFT)|_PAGE_FILE})
446 #define PTE_FILE_MAX_BITS (BITS_PER_LONG - PTE_SHIFT)
449 * kern_addr_valid is intended to indicate whether an address is a valid
450 * kernel address. Most 32-bit archs define it as always true (like this)
451 * but most 64-bit archs actually perform a test. What should we do here?
452 * The only use is in fs/ncpfs/dir.c
454 #define kern_addr_valid(addr) (1)
456 #define io_remap_page_range remap_page_range
458 void pgtable_cache_init(void);
460 extern void hpte_init_pSeries(void);
461 extern void hpte_init_iSeries(void);
463 /* imalloc region types */
464 #define IM_REGION_UNUSED 0x1
465 #define IM_REGION_SUBSET 0x2
466 #define IM_REGION_EXISTS 0x4
467 #define IM_REGION_OVERLAP 0x8
469 extern struct vm_struct * im_get_free_area(unsigned long size);
470 extern struct vm_struct * im_get_area(unsigned long v_addr, unsigned long size,
472 unsigned long im_free(void *addr);
474 typedef pte_t *pte_addr_t;
476 long pSeries_lpar_hpte_insert(unsigned long hpte_group,
477 unsigned long va, unsigned long prpn,
478 int secondary, unsigned long hpteflags,
479 int bolted, int large);
481 long pSeries_hpte_insert(unsigned long hpte_group, unsigned long va,
482 unsigned long prpn, int secondary,
483 unsigned long hpteflags, int bolted, int large);
486 * find_linux_pte returns the address of a linux pte for a given
487 * effective address and directory. If not found, it returns zero.
489 static inline pte_t *find_linux_pte(pgd_t *pgdir, unsigned long ea)
496 pg = pgdir + pgd_index(ea);
497 if (!pgd_none(*pg)) {
499 pm = pmd_offset(pg, ea);
500 if (pmd_present(*pm)) {
501 pt = pte_offset_kernel(pm, ea);
503 if (!pte_present(pte))
511 #endif /* __ASSEMBLY__ */
513 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
514 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
515 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
516 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
517 #define __HAVE_ARCH_PTEP_MKDIRTY
518 #define __HAVE_ARCH_PTE_SAME
519 #include <asm-generic/pgtable.h>
521 #endif /* _PPC64_PGTABLE_H */