Merge branch 'generic-dispatch' of git://brick.kernel.dk/data/git/linux-2.6-block
[linux-2.6] / arch / ia64 / mm / hugetlbpage.c
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
27 pte_t *
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
46 pte_t *
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
70 /*
71  * This function checks for proper alignment of input addr and len parameters.
72  */
73 int is_aligned_hugepage_range(unsigned long addr, unsigned long len)
74 {
75         if (len & ~HPAGE_MASK)
76                 return -EINVAL;
77         if (addr & ~HPAGE_MASK)
78                 return -EINVAL;
79         if (REGION_NUMBER(addr) != RGN_HPAGE)
80                 return -EINVAL;
81
82         return 0;
83 }
84
85 struct page *follow_huge_addr(struct mm_struct *mm, unsigned long addr, int write)
86 {
87         struct page *page;
88         pte_t *ptep;
89
90         if (REGION_NUMBER(addr) != RGN_HPAGE)
91                 return ERR_PTR(-EINVAL);
92
93         ptep = huge_pte_offset(mm, addr);
94         if (!ptep || pte_none(*ptep))
95                 return NULL;
96         page = pte_page(*ptep);
97         page += ((addr & ~HPAGE_MASK) >> PAGE_SHIFT);
98         return page;
99 }
100 int pmd_huge(pmd_t pmd)
101 {
102         return 0;
103 }
104 struct page *
105 follow_huge_pmd(struct mm_struct *mm, unsigned long address, pmd_t *pmd, int write)
106 {
107         return NULL;
108 }
109
110 void hugetlb_free_pgd_range(struct mmu_gather **tlb,
111                         unsigned long addr, unsigned long end,
112                         unsigned long floor, unsigned long ceiling)
113 {
114         /*
115          * This is called only when is_hugepage_only_range(addr,),
116          * and it follows that is_hugepage_only_range(end,) also.
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);
128         if (is_hugepage_only_range(tlb->mm, floor, HPAGE_SIZE))
129                 floor = htlbpage_to_page(floor);
130         if (is_hugepage_only_range(tlb->mm, ceiling, HPAGE_SIZE))
131                 ceiling = htlbpage_to_page(ceiling);
132
133         free_pgd_range(tlb, addr, end, floor, ceiling);
134 }
135
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;
145         /* This code assumes that RGN_HPAGE != 0. */
146         if ((REGION_NUMBER(addr) != RGN_HPAGE) || (addr & (HPAGE_SIZE - 1)))
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);