Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * IA-64 Huge TLB Page Support for Kernel. |
| 3 | * |
| 4 | * Copyright (C) 2002-2004 Rohit Seth <rohit.seth@intel.com> |
| 5 | * Copyright (C) 2003-2004 Ken Chen <kenneth.w.chen@intel.com> |
| 6 | * |
| 7 | * Sep, 2003: add numa support |
| 8 | * Feb, 2004: dynamic hugetlb page size via boot parameter |
| 9 | */ |
| 10 | |
| 11 | #include <linux/config.h> |
| 12 | #include <linux/init.h> |
| 13 | #include <linux/fs.h> |
| 14 | #include <linux/mm.h> |
| 15 | #include <linux/hugetlb.h> |
| 16 | #include <linux/pagemap.h> |
| 17 | #include <linux/smp_lock.h> |
| 18 | #include <linux/slab.h> |
| 19 | #include <linux/sysctl.h> |
| 20 | #include <asm/mman.h> |
| 21 | #include <asm/pgalloc.h> |
| 22 | #include <asm/tlb.h> |
| 23 | #include <asm/tlbflush.h> |
| 24 | |
| 25 | unsigned int hpage_shift=HPAGE_SHIFT_DEFAULT; |
| 26 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 27 | pte_t * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | huge_pte_alloc (struct mm_struct *mm, unsigned long addr) |
| 29 | { |
| 30 | unsigned long taddr = htlbpage_to_page(addr); |
| 31 | pgd_t *pgd; |
| 32 | pud_t *pud; |
| 33 | pmd_t *pmd; |
| 34 | pte_t *pte = NULL; |
| 35 | |
| 36 | pgd = pgd_offset(mm, taddr); |
| 37 | pud = pud_alloc(mm, pgd, taddr); |
| 38 | if (pud) { |
| 39 | pmd = pmd_alloc(mm, pud, taddr); |
| 40 | if (pmd) |
| 41 | pte = pte_alloc_map(mm, pmd, taddr); |
| 42 | } |
| 43 | return pte; |
| 44 | } |
| 45 | |
David Gibson | 63551ae | 2005-06-21 17:14:44 -0700 | [diff] [blame] | 46 | pte_t * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | huge_pte_offset (struct mm_struct *mm, unsigned long addr) |
| 48 | { |
| 49 | unsigned long taddr = htlbpage_to_page(addr); |
| 50 | pgd_t *pgd; |
| 51 | pud_t *pud; |
| 52 | pmd_t *pmd; |
| 53 | pte_t *pte = NULL; |
| 54 | |
| 55 | pgd = pgd_offset(mm, taddr); |
| 56 | if (pgd_present(*pgd)) { |
| 57 | pud = pud_offset(pgd, taddr); |
| 58 | if (pud_present(*pud)) { |
| 59 | pmd = pmd_offset(pud, taddr); |
| 60 | if (pmd_present(*pmd)) |
| 61 | pte = pte_offset_map(pmd, taddr); |
| 62 | } |
| 63 | } |
| 64 | |
| 65 | return pte; |
| 66 | } |
| 67 | |
| 68 | #define mk_pte_huge(entry) { pte_val(entry) |= _PAGE_P; } |
| 69 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | /* |
David Gibson | 42b88be | 2006-03-22 00:09:01 -0800 | [diff] [blame] | 71 | * Don't actually need to do any preparation, but need to make sure |
| 72 | * the address is in the right region. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | */ |
David Gibson | 42b88be | 2006-03-22 00:09:01 -0800 | [diff] [blame] | 74 | int prepare_hugepage_range(unsigned long addr, unsigned long len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | { |
| 76 | if (len & ~HPAGE_MASK) |
| 77 | return -EINVAL; |
| 78 | if (addr & ~HPAGE_MASK) |
| 79 | return -EINVAL; |
Peter Chubb | 0a41e25 | 2005-08-16 19:54:00 -0700 | [diff] [blame] | 80 | if (REGION_NUMBER(addr) != RGN_HPAGE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | return -EINVAL; |
| 82 | |
| 83 | return 0; |
| 84 | } |
| 85 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | struct page *follow_huge_addr(struct mm_struct *mm, unsigned long addr, int write) |
| 87 | { |
| 88 | struct page *page; |
| 89 | pte_t *ptep; |
| 90 | |
Peter Chubb | 0a41e25 | 2005-08-16 19:54:00 -0700 | [diff] [blame] | 91 | if (REGION_NUMBER(addr) != RGN_HPAGE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | return ERR_PTR(-EINVAL); |
| 93 | |
| 94 | ptep = huge_pte_offset(mm, addr); |
| 95 | if (!ptep || pte_none(*ptep)) |
| 96 | return NULL; |
| 97 | page = pte_page(*ptep); |
| 98 | page += ((addr & ~HPAGE_MASK) >> PAGE_SHIFT); |
| 99 | return page; |
| 100 | } |
| 101 | int pmd_huge(pmd_t pmd) |
| 102 | { |
| 103 | return 0; |
| 104 | } |
| 105 | struct page * |
| 106 | follow_huge_pmd(struct mm_struct *mm, unsigned long address, pmd_t *pmd, int write) |
| 107 | { |
| 108 | return NULL; |
| 109 | } |
| 110 | |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 111 | void hugetlb_free_pgd_range(struct mmu_gather **tlb, |
| 112 | unsigned long addr, unsigned long end, |
| 113 | unsigned long floor, unsigned long ceiling) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | { |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 115 | /* |
Chen, Kenneth W | 2332c9a | 2006-03-22 10:49:00 -0800 | [diff] [blame] | 116 | * This is called to free hugetlb page tables. |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 117 | * |
| 118 | * The offset of these addresses from the base of the hugetlb |
| 119 | * region must be scaled down by HPAGE_SIZE/PAGE_SIZE so that |
| 120 | * the standard free_pgd_range will free the right page tables. |
| 121 | * |
| 122 | * If floor and ceiling are also in the hugetlb region, they |
| 123 | * must likewise be scaled down; but if outside, left unchanged. |
| 124 | */ |
| 125 | |
| 126 | addr = htlbpage_to_page(addr); |
| 127 | end = htlbpage_to_page(end); |
Chen, Kenneth W | 2332c9a | 2006-03-22 10:49:00 -0800 | [diff] [blame] | 128 | if (REGION_NUMBER(floor) == RGN_HPAGE) |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 129 | floor = htlbpage_to_page(floor); |
Chen, Kenneth W | 2332c9a | 2006-03-22 10:49:00 -0800 | [diff] [blame] | 130 | if (REGION_NUMBER(ceiling) == RGN_HPAGE) |
Hugh Dickins | 3bf5ee9 | 2005-04-19 13:29:16 -0700 | [diff] [blame] | 131 | ceiling = htlbpage_to_page(ceiling); |
| 132 | |
| 133 | free_pgd_range(tlb, addr, end, floor, ceiling); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | } |
| 135 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, |
| 137 | unsigned long pgoff, unsigned long flags) |
| 138 | { |
| 139 | struct vm_area_struct *vmm; |
| 140 | |
| 141 | if (len > RGN_MAP_LIMIT) |
| 142 | return -ENOMEM; |
| 143 | if (len & ~HPAGE_MASK) |
| 144 | return -EINVAL; |
Peter Chubb | 0a41e25 | 2005-08-16 19:54:00 -0700 | [diff] [blame] | 145 | /* This code assumes that RGN_HPAGE != 0. */ |
| 146 | if ((REGION_NUMBER(addr) != RGN_HPAGE) || (addr & (HPAGE_SIZE - 1))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | addr = HPAGE_REGION_BASE; |
| 148 | else |
| 149 | addr = ALIGN(addr, HPAGE_SIZE); |
| 150 | for (vmm = find_vma(current->mm, addr); ; vmm = vmm->vm_next) { |
| 151 | /* At this point: (!vmm || addr < vmm->vm_end). */ |
| 152 | if (REGION_OFFSET(addr) + len > RGN_MAP_LIMIT) |
| 153 | return -ENOMEM; |
| 154 | if (!vmm || (addr + len) <= vmm->vm_start) |
| 155 | return addr; |
| 156 | addr = ALIGN(vmm->vm_end, HPAGE_SIZE); |
| 157 | } |
| 158 | } |
| 159 | |
| 160 | static int __init hugetlb_setup_sz(char *str) |
| 161 | { |
| 162 | u64 tr_pages; |
| 163 | unsigned long long size; |
| 164 | |
| 165 | if (ia64_pal_vm_page_size(&tr_pages, NULL) != 0) |
| 166 | /* |
| 167 | * shouldn't happen, but just in case. |
| 168 | */ |
| 169 | tr_pages = 0x15557000UL; |
| 170 | |
| 171 | size = memparse(str, &str); |
| 172 | if (*str || (size & (size-1)) || !(tr_pages & size) || |
| 173 | size <= PAGE_SIZE || |
| 174 | size >= (1UL << PAGE_SHIFT << MAX_ORDER)) { |
| 175 | printk(KERN_WARNING "Invalid huge page size specified\n"); |
| 176 | return 1; |
| 177 | } |
| 178 | |
| 179 | hpage_shift = __ffs(size); |
| 180 | /* |
| 181 | * boot cpu already executed ia64_mmu_init, and has HPAGE_SHIFT_DEFAULT |
| 182 | * override here with new page shift. |
| 183 | */ |
| 184 | ia64_set_rr(HPAGE_REGION_BASE, hpage_shift << 2); |
| 185 | return 1; |
| 186 | } |
| 187 | __setup("hugepagesz=", hugetlb_setup_sz); |