1 #ifndef _ASM_POWERPC_PGTABLE_PPC64_H_
2 #define _ASM_POWERPC_PGTABLE_PPC64_H_
4 * This file contains the functions and defines necessary to modify and use
5 * the ppc64 hashed page table.
8 #ifdef CONFIG_PPC_64K_PAGES
9 #include <asm/pgtable-ppc64-64k.h>
11 #include <asm/pgtable-ppc64-4k.h>
13 #include <asm/barrier.h>
15 #define FIRST_USER_ADDRESS 0UL
18 * Size of EA range mapped by our pagetables.
20 #define PGTABLE_EADDR_SIZE (PTE_INDEX_SIZE + PMD_INDEX_SIZE + \
21 PUD_INDEX_SIZE + PGD_INDEX_SIZE + PAGE_SHIFT)
22 #define PGTABLE_RANGE (ASM_CONST(1) << PGTABLE_EADDR_SIZE)
24 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
25 #define PMD_CACHE_INDEX (PMD_INDEX_SIZE + 1)
27 #define PMD_CACHE_INDEX PMD_INDEX_SIZE
30 * Define the address range of the kernel non-linear virtual area
33 #ifdef CONFIG_PPC_BOOK3E
34 #define KERN_VIRT_START ASM_CONST(0x8000000000000000)
36 #define KERN_VIRT_START ASM_CONST(0xD000000000000000)
38 #define KERN_VIRT_SIZE ASM_CONST(0x0000100000000000)
41 * The vmalloc space starts at the beginning of that region, and
42 * occupies half of it on hash CPUs and a quarter of it on Book3E
43 * (we keep a quarter for the virtual memmap)
45 #define VMALLOC_START KERN_VIRT_START
46 #ifdef CONFIG_PPC_BOOK3E
47 #define VMALLOC_SIZE (KERN_VIRT_SIZE >> 2)
49 #define VMALLOC_SIZE (KERN_VIRT_SIZE >> 1)
51 #define VMALLOC_END (VMALLOC_START + VMALLOC_SIZE)
54 * The second half of the kernel virtual space is used for IO mappings,
55 * it's itself carved into the PIO region (ISA and PHB IO space) and
58 * ISA_IO_BASE = KERN_IO_START, 64K reserved area
59 * PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
60 * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
62 #define KERN_IO_START (KERN_VIRT_START + (KERN_VIRT_SIZE >> 1))
63 #define FULL_IO_SIZE 0x80000000ul
64 #define ISA_IO_BASE (KERN_IO_START)
65 #define ISA_IO_END (KERN_IO_START + 0x10000ul)
66 #define PHB_IO_BASE (ISA_IO_END)
67 #define PHB_IO_END (KERN_IO_START + FULL_IO_SIZE)
68 #define IOREMAP_BASE (PHB_IO_END)
69 #define IOREMAP_END (KERN_VIRT_START + KERN_VIRT_SIZE)
75 #define REGION_SHIFT 60UL
76 #define REGION_MASK (0xfUL << REGION_SHIFT)
77 #define REGION_ID(ea) (((unsigned long)(ea)) >> REGION_SHIFT)
79 #define VMALLOC_REGION_ID (REGION_ID(VMALLOC_START))
80 #define KERNEL_REGION_ID (REGION_ID(PAGE_OFFSET))
81 #define VMEMMAP_REGION_ID (0xfUL) /* Server only */
82 #define USER_REGION_ID (0UL)
85 * Defines the address of the vmemap area, in its own region on
86 * hash table CPUs and after the vmalloc space on Book3E
88 #ifdef CONFIG_PPC_BOOK3E
89 #define VMEMMAP_BASE VMALLOC_END
90 #define VMEMMAP_END KERN_IO_START
92 #define VMEMMAP_BASE (VMEMMAP_REGION_ID << REGION_SHIFT)
94 #define vmemmap ((struct page *)VMEMMAP_BASE)
98 * Include the PTE bits definitions
100 #ifdef CONFIG_PPC_BOOK3S
101 #include <asm/pte-hash64.h>
103 #include <asm/pte-book3e.h>
105 #include <asm/pte-common.h>
107 #ifdef CONFIG_PPC_MM_SLICES
108 #define HAVE_ARCH_UNMAPPED_AREA
109 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
110 #endif /* CONFIG_PPC_MM_SLICES */
115 * This is the default implementation of various PTE accessors, it's
116 * used in all cases except Book3S with 64K pages where we have a
117 * concept of sub-pages
121 #ifdef CONFIG_STRICT_MM_TYPECHECKS
122 #define __real_pte(e,p) ((real_pte_t){(e)})
123 #define __rpte_to_pte(r) ((r).pte)
125 #define __real_pte(e,p) (e)
126 #define __rpte_to_pte(r) (__pte(r))
128 #define __rpte_to_hidx(r,index) (pte_val(__rpte_to_pte(r)) >> 12)
130 #define pte_iterate_hashed_subpages(rpte, psize, va, index, shift) \
133 shift = mmu_psize_defs[psize].shift; \
135 #define pte_iterate_hashed_end() } while(0)
138 * We expect this to be called only for user addresses or kernel virtual
139 * addresses other than the linear mapping.
141 #define pte_pagesize_index(mm, addr, pte) MMU_PAGE_4K
143 #endif /* __real_pte */
146 /* pte_clear moved to later in this file */
148 #define PMD_BAD_BITS (PTE_TABLE_SIZE-1)
149 #define PUD_BAD_BITS (PMD_TABLE_SIZE-1)
151 #define pmd_set(pmdp, pmdval) (pmd_val(*(pmdp)) = (pmdval))
152 #define pmd_none(pmd) (!pmd_val(pmd))
153 #define pmd_bad(pmd) (!is_kernel_addr(pmd_val(pmd)) \
154 || (pmd_val(pmd) & PMD_BAD_BITS))
155 #define pmd_present(pmd) (!pmd_none(pmd))
156 #define pmd_clear(pmdp) (pmd_val(*(pmdp)) = 0)
157 #define pmd_page_vaddr(pmd) (pmd_val(pmd) & ~PMD_MASKED_BITS)
158 extern struct page *pmd_page(pmd_t pmd);
160 #define pud_set(pudp, pudval) (pud_val(*(pudp)) = (pudval))
161 #define pud_none(pud) (!pud_val(pud))
162 #define pud_bad(pud) (!is_kernel_addr(pud_val(pud)) \
163 || (pud_val(pud) & PUD_BAD_BITS))
164 #define pud_present(pud) (pud_val(pud) != 0)
165 #define pud_clear(pudp) (pud_val(*(pudp)) = 0)
166 #define pud_page_vaddr(pud) (pud_val(pud) & ~PUD_MASKED_BITS)
168 extern struct page *pud_page(pud_t pud);
170 static inline pte_t pud_pte(pud_t pud)
172 return __pte(pud_val(pud));
175 static inline pud_t pte_pud(pte_t pte)
177 return __pud(pte_val(pte));
179 #define pud_write(pud) pte_write(pud_pte(pud))
180 #define pgd_set(pgdp, pudp) ({pgd_val(*(pgdp)) = (unsigned long)(pudp);})
181 #define pgd_write(pgd) pte_write(pgd_pte(pgd))
184 * Find an entry in a page-table-directory. We combine the address region
185 * (the high order N bits) and the pgd portion of the address.
187 #define pgd_index(address) (((address) >> (PGDIR_SHIFT)) & (PTRS_PER_PGD - 1))
189 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
191 #define pmd_offset(pudp,addr) \
192 (((pmd_t *) pud_page_vaddr(*(pudp))) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
194 #define pte_offset_kernel(dir,addr) \
195 (((pte_t *) pmd_page_vaddr(*(dir))) + (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)))
197 #define pte_offset_map(dir,addr) pte_offset_kernel((dir), (addr))
198 #define pte_unmap(pte) do { } while(0)
200 /* to find an entry in a kernel page-table-directory */
201 /* This now only contains the vmalloc pages */
202 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
203 extern void hpte_need_flush(struct mm_struct *mm, unsigned long addr,
204 pte_t *ptep, unsigned long pte, int huge);
206 /* Atomic PTE updates */
207 static inline unsigned long pte_update(struct mm_struct *mm,
209 pte_t *ptep, unsigned long clr,
213 #ifdef PTE_ATOMIC_UPDATES
214 unsigned long old, tmp;
216 __asm__ __volatile__(
217 "1: ldarx %0,0,%3 # pte_update\n\
224 : "=&r" (old), "=&r" (tmp), "=m" (*ptep)
225 : "r" (ptep), "r" (clr), "m" (*ptep), "i" (_PAGE_BUSY), "r" (set)
228 unsigned long old = pte_val(*ptep);
229 *ptep = __pte((old & ~clr) | set);
231 /* huge pages use the old page table lock */
233 assert_pte_locked(mm, addr);
235 #ifdef CONFIG_PPC_STD_MMU_64
236 if (old & _PAGE_HASHPTE)
237 hpte_need_flush(mm, addr, ptep, old, huge);
243 static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
244 unsigned long addr, pte_t *ptep)
248 if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
250 old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0, 0);
251 return (old & _PAGE_ACCESSED) != 0;
253 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
254 #define ptep_test_and_clear_young(__vma, __addr, __ptep) \
257 __r = __ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
261 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
262 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
266 if ((pte_val(*ptep) & _PAGE_RW) == 0)
269 pte_update(mm, addr, ptep, _PAGE_RW, 0, 0);
272 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
273 unsigned long addr, pte_t *ptep)
275 if ((pte_val(*ptep) & _PAGE_RW) == 0)
278 pte_update(mm, addr, ptep, _PAGE_RW, 0, 1);
282 * We currently remove entries from the hashtable regardless of whether
283 * the entry was young or dirty. The generic routines only flush if the
284 * entry was young or dirty which is not good enough.
286 * We should be more intelligent about this but for the moment we override
287 * these functions and force a tlb flush unconditionally
289 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
290 #define ptep_clear_flush_young(__vma, __address, __ptep) \
292 int __young = __ptep_test_and_clear_young((__vma)->vm_mm, __address, \
297 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
298 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
299 unsigned long addr, pte_t *ptep)
301 unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0, 0);
305 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
308 pte_update(mm, addr, ptep, ~0UL, 0, 0);
312 /* Set the dirty and/or accessed bits atomically in a linux PTE, this
313 * function doesn't need to flush the hash entry
315 static inline void __ptep_set_access_flags(pte_t *ptep, pte_t entry)
317 unsigned long bits = pte_val(entry) &
318 (_PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_RW | _PAGE_EXEC);
320 #ifdef PTE_ATOMIC_UPDATES
321 unsigned long old, tmp;
323 __asm__ __volatile__(
330 :"=&r" (old), "=&r" (tmp), "=m" (*ptep)
331 :"r" (bits), "r" (ptep), "m" (*ptep), "i" (_PAGE_BUSY)
334 unsigned long old = pte_val(*ptep);
335 *ptep = __pte(old | bits);
339 #define __HAVE_ARCH_PTE_SAME
340 #define pte_same(A,B) (((pte_val(A) ^ pte_val(B)) & ~_PAGE_HPTEFLAGS) == 0)
342 #define pte_ERROR(e) \
343 pr_err("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
344 #define pmd_ERROR(e) \
345 pr_err("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
346 #define pgd_ERROR(e) \
347 pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
349 /* Encode and de-code a swap entry */
350 #define MAX_SWAPFILES_CHECK() do { \
351 BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS); \
353 * Don't have overlapping bits with _PAGE_HPTEFLAGS \
354 * We filter HPTEFLAGS on set_pte. \
356 BUILD_BUG_ON(_PAGE_HPTEFLAGS & (0x1f << _PAGE_BIT_SWAP_TYPE)); \
359 * on pte we don't need handle RADIX_TREE_EXCEPTIONAL_SHIFT;
361 #define SWP_TYPE_BITS 5
362 #define __swp_type(x) (((x).val >> _PAGE_BIT_SWAP_TYPE) \
363 & ((1UL << SWP_TYPE_BITS) - 1))
364 #define __swp_offset(x) ((x).val >> PTE_RPN_SHIFT)
365 #define __swp_entry(type, offset) ((swp_entry_t) { \
366 ((type) << _PAGE_BIT_SWAP_TYPE) \
367 | ((offset) << PTE_RPN_SHIFT) })
369 #define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val((pte)) })
370 #define __swp_entry_to_pte(x) __pte((x).val)
372 void pgtable_cache_add(unsigned shift, void (*ctor)(void *));
373 void pgtable_cache_init(void);
374 #endif /* __ASSEMBLY__ */
377 * THP pages can't be special. So use the _PAGE_SPECIAL
379 #define _PAGE_SPLITTING _PAGE_SPECIAL
382 * We need to differentiate between explicit huge page and THP huge
383 * page, since THP huge page also need to track real subpage details
385 #define _PAGE_THP_HUGE _PAGE_4K_PFN
388 * set of bits not changed in pmd_modify.
390 #define _HPAGE_CHG_MASK (PTE_RPN_MASK | _PAGE_HPTEFLAGS | \
391 _PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_SPLITTING | \
396 * The linux hugepage PMD now include the pmd entries followed by the address
397 * to the stashed pgtable_t. The stashed pgtable_t contains the hpte bits.
398 * [ 1 bit secondary | 3 bit hidx | 1 bit valid | 000]. We use one byte per
399 * each HPTE entry. With 16MB hugepage and 64K HPTE we need 256 entries and
400 * with 4K HPTE we need 4096 entries. Both will fit in a 4K pgtable_t.
402 * The last three bits are intentionally left to zero. This memory location
403 * are also used as normal page PTE pointers. So if we have any pointers
404 * left around while we collapse a hugepage, we need to make sure
405 * _PAGE_PRESENT bit of that is zero when we look at them
407 static inline unsigned int hpte_valid(unsigned char *hpte_slot_array, int index)
409 return (hpte_slot_array[index] >> 3) & 0x1;
412 static inline unsigned int hpte_hash_index(unsigned char *hpte_slot_array,
415 return hpte_slot_array[index] >> 4;
418 static inline void mark_hpte_slot_valid(unsigned char *hpte_slot_array,
419 unsigned int index, unsigned int hidx)
421 hpte_slot_array[index] = hidx << 4 | 0x1 << 3;
424 struct page *realmode_pfn_to_page(unsigned long pfn);
426 static inline char *get_hpte_slot_array(pmd_t *pmdp)
429 * The hpte hindex is stored in the pgtable whose address is in the
430 * second half of the PMD
432 * Order this load with the test for pmd_trans_huge in the caller
435 return *(char **)(pmdp + PTRS_PER_PMD);
440 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
441 extern void hpte_do_hugepage_flush(struct mm_struct *mm, unsigned long addr,
442 pmd_t *pmdp, unsigned long old_pmd);
443 extern pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot);
444 extern pmd_t mk_pmd(struct page *page, pgprot_t pgprot);
445 extern pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot);
446 extern void set_pmd_at(struct mm_struct *mm, unsigned long addr,
447 pmd_t *pmdp, pmd_t pmd);
448 extern void update_mmu_cache_pmd(struct vm_area_struct *vma, unsigned long addr,
452 * For core kernel code by design pmd_trans_huge is never run on any hugetlbfs
453 * page. The hugetlbfs page table walking and mangling paths are totally
454 * separated form the core VM paths and they're differentiated by
455 * VM_HUGETLB being set on vm_flags well before any pmd_trans_huge could run.
457 * pmd_trans_huge() is defined as false at build time if
458 * CONFIG_TRANSPARENT_HUGEPAGE=n to optimize away code blocks at build
461 * For ppc64 we need to differntiate from explicit hugepages from THP, because
462 * for THP we also track the subpage details at the pmd level. We don't do
463 * that for explicit huge pages.
466 static inline int pmd_trans_huge(pmd_t pmd)
469 * leaf pte for huge page, bottom two bits != 00
471 return (pmd_val(pmd) & 0x3) && (pmd_val(pmd) & _PAGE_THP_HUGE);
474 static inline int pmd_trans_splitting(pmd_t pmd)
476 if (pmd_trans_huge(pmd))
477 return pmd_val(pmd) & _PAGE_SPLITTING;
481 extern int has_transparent_hugepage(void);
483 static inline void hpte_do_hugepage_flush(struct mm_struct *mm,
484 unsigned long addr, pmd_t *pmdp,
485 unsigned long old_pmd)
488 WARN(1, "%s called with THP disabled\n", __func__);
490 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
492 static inline int pmd_large(pmd_t pmd)
495 * leaf pte for huge page, bottom two bits != 00
497 return ((pmd_val(pmd) & 0x3) != 0x0);
500 static inline pte_t pmd_pte(pmd_t pmd)
502 return __pte(pmd_val(pmd));
505 static inline pmd_t pte_pmd(pte_t pte)
507 return __pmd(pte_val(pte));
510 static inline pte_t *pmdp_ptep(pmd_t *pmd)
515 #define pmd_pfn(pmd) pte_pfn(pmd_pte(pmd))
516 #define pmd_dirty(pmd) pte_dirty(pmd_pte(pmd))
517 #define pmd_young(pmd) pte_young(pmd_pte(pmd))
518 #define pmd_mkold(pmd) pte_pmd(pte_mkold(pmd_pte(pmd)))
519 #define pmd_wrprotect(pmd) pte_pmd(pte_wrprotect(pmd_pte(pmd)))
520 #define pmd_mkdirty(pmd) pte_pmd(pte_mkdirty(pmd_pte(pmd)))
521 #define pmd_mkyoung(pmd) pte_pmd(pte_mkyoung(pmd_pte(pmd)))
522 #define pmd_mkwrite(pmd) pte_pmd(pte_mkwrite(pmd_pte(pmd)))
524 #define __HAVE_ARCH_PMD_WRITE
525 #define pmd_write(pmd) pte_write(pmd_pte(pmd))
527 static inline pmd_t pmd_mkhuge(pmd_t pmd)
529 /* Do nothing, mk_pmd() does this part. */
533 static inline pmd_t pmd_mknotpresent(pmd_t pmd)
535 pmd_val(pmd) &= ~_PAGE_PRESENT;
539 static inline pmd_t pmd_mksplitting(pmd_t pmd)
541 pmd_val(pmd) |= _PAGE_SPLITTING;
545 #define __HAVE_ARCH_PMD_SAME
546 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
548 return (((pmd_val(pmd_a) ^ pmd_val(pmd_b)) & ~_PAGE_HPTEFLAGS) == 0);
551 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
552 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
553 unsigned long address, pmd_t *pmdp,
554 pmd_t entry, int dirty);
556 extern unsigned long pmd_hugepage_update(struct mm_struct *mm,
562 static inline int __pmdp_test_and_clear_young(struct mm_struct *mm,
563 unsigned long addr, pmd_t *pmdp)
567 if ((pmd_val(*pmdp) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
569 old = pmd_hugepage_update(mm, addr, pmdp, _PAGE_ACCESSED, 0);
570 return ((old & _PAGE_ACCESSED) != 0);
573 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
574 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
575 unsigned long address, pmd_t *pmdp);
576 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
577 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
578 unsigned long address, pmd_t *pmdp);
580 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
581 extern pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
582 unsigned long addr, pmd_t *pmdp);
584 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
585 static inline void pmdp_set_wrprotect(struct mm_struct *mm, unsigned long addr,
589 if ((pmd_val(*pmdp) & _PAGE_RW) == 0)
592 pmd_hugepage_update(mm, addr, pmdp, _PAGE_RW, 0);
595 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
596 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
597 unsigned long address, pmd_t *pmdp);
599 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
600 unsigned long address, pmd_t *pmdp);
601 #define pmdp_collapse_flush pmdp_collapse_flush
603 #define __HAVE_ARCH_PGTABLE_DEPOSIT
604 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
606 #define __HAVE_ARCH_PGTABLE_WITHDRAW
607 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
609 #define __HAVE_ARCH_PMDP_INVALIDATE
610 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
613 #define pmd_move_must_withdraw pmd_move_must_withdraw
615 static inline int pmd_move_must_withdraw(struct spinlock *new_pmd_ptl,
616 struct spinlock *old_pmd_ptl)
619 * Archs like ppc64 use pgtable to store per pmd
620 * specific information. So when we switch the pmd,
621 * we should also withdraw and deposit the pgtable
625 #endif /* __ASSEMBLY__ */
626 #endif /* _ASM_POWERPC_PGTABLE_PPC64_H_ */