Merge master.kernel.org:/home/rmk/linux-2.6-serial
[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  * Don't actually need to do any preparation, but need to make sure
72  * the address is in the right region.
73  */
74 int prepare_hugepage_range(unsigned long addr, unsigned long len)
75 {
76         if (len & ~HPAGE_MASK)
77                 return -EINVAL;
78         if (addr & ~HPAGE_MASK)
79                 return -EINVAL;
80         if (REGION_NUMBER(addr) != RGN_HPAGE)
81                 return -EINVAL;
82
83         return 0;
84 }
85
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
91         if (REGION_NUMBER(addr) != RGN_HPAGE)
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
111 void hugetlb_free_pgd_range(struct mmu_gather **tlb,
112                         unsigned long addr, unsigned long end,
113                         unsigned long floor, unsigned long ceiling)
114 {
115         /*
116          * This is called only when is_hugepage_only_range(addr,),
117          * and it follows that is_hugepage_only_range(end,) also.
118          *
119          * The offset of these addresses from the base of the hugetlb
120          * region must be scaled down by HPAGE_SIZE/PAGE_SIZE so that
121          * the standard free_pgd_range will free the right page tables.
122          *
123          * If floor and ceiling are also in the hugetlb region, they
124          * must likewise be scaled down; but if outside, left unchanged.
125          */
126
127         addr = htlbpage_to_page(addr);
128         end  = htlbpage_to_page(end);
129         if (is_hugepage_only_range(tlb->mm, floor, HPAGE_SIZE))
130                 floor = htlbpage_to_page(floor);
131         if (is_hugepage_only_range(tlb->mm, ceiling, HPAGE_SIZE))
132                 ceiling = htlbpage_to_page(ceiling);
133
134         free_pgd_range(tlb, addr, end, floor, ceiling);
135 }
136
137 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
138                 unsigned long pgoff, unsigned long flags)
139 {
140         struct vm_area_struct *vmm;
141
142         if (len > RGN_MAP_LIMIT)
143                 return -ENOMEM;
144         if (len & ~HPAGE_MASK)
145                 return -EINVAL;
146         /* This code assumes that RGN_HPAGE != 0. */
147         if ((REGION_NUMBER(addr) != RGN_HPAGE) || (addr & (HPAGE_SIZE - 1)))
148                 addr = HPAGE_REGION_BASE;
149         else
150                 addr = ALIGN(addr, HPAGE_SIZE);
151         for (vmm = find_vma(current->mm, addr); ; vmm = vmm->vm_next) {
152                 /* At this point:  (!vmm || addr < vmm->vm_end). */
153                 if (REGION_OFFSET(addr) + len > RGN_MAP_LIMIT)
154                         return -ENOMEM;
155                 if (!vmm || (addr + len) <= vmm->vm_start)
156                         return addr;
157                 addr = ALIGN(vmm->vm_end, HPAGE_SIZE);
158         }
159 }
160
161 static int __init hugetlb_setup_sz(char *str)
162 {
163         u64 tr_pages;
164         unsigned long long size;
165
166         if (ia64_pal_vm_page_size(&tr_pages, NULL) != 0)
167                 /*
168                  * shouldn't happen, but just in case.
169                  */
170                 tr_pages = 0x15557000UL;
171
172         size = memparse(str, &str);
173         if (*str || (size & (size-1)) || !(tr_pages & size) ||
174                 size <= PAGE_SIZE ||
175                 size >= (1UL << PAGE_SHIFT << MAX_ORDER)) {
176                 printk(KERN_WARNING "Invalid huge page size specified\n");
177                 return 1;
178         }
179
180         hpage_shift = __ffs(size);
181         /*
182          * boot cpu already executed ia64_mmu_init, and has HPAGE_SHIFT_DEFAULT
183          * override here with new page shift.
184          */
185         ia64_set_rr(HPAGE_REGION_BASE, hpage_shift << 2);
186         return 1;
187 }
188 __setup("hugepagesz=", hugetlb_setup_sz);