blob: 1346b7f05397506f0384fb204d9d62f27d838d6e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/init.h>
12#include <linux/fs.h>
13#include <linux/mm.h>
14#include <linux/hugetlb.h>
15#include <linux/pagemap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/slab.h>
17#include <linux/sysctl.h>
18#include <asm/mman.h>
19#include <asm/pgalloc.h>
20#include <asm/tlb.h>
21#include <asm/tlbflush.h>
22
23unsigned int hpage_shift=HPAGE_SHIFT_DEFAULT;
24
David Gibson63551ae2005-06-21 17:14:44 -070025pte_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -070026huge_pte_alloc (struct mm_struct *mm, unsigned long addr)
27{
28 unsigned long taddr = htlbpage_to_page(addr);
29 pgd_t *pgd;
30 pud_t *pud;
31 pmd_t *pmd;
32 pte_t *pte = NULL;
33
34 pgd = pgd_offset(mm, taddr);
35 pud = pud_alloc(mm, pgd, taddr);
36 if (pud) {
37 pmd = pmd_alloc(mm, pud, taddr);
38 if (pmd)
39 pte = pte_alloc_map(mm, pmd, taddr);
40 }
41 return pte;
42}
43
David Gibson63551ae2005-06-21 17:14:44 -070044pte_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -070045huge_pte_offset (struct mm_struct *mm, unsigned long addr)
46{
47 unsigned long taddr = htlbpage_to_page(addr);
48 pgd_t *pgd;
49 pud_t *pud;
50 pmd_t *pmd;
51 pte_t *pte = NULL;
52
53 pgd = pgd_offset(mm, taddr);
54 if (pgd_present(*pgd)) {
55 pud = pud_offset(pgd, taddr);
56 if (pud_present(*pud)) {
57 pmd = pmd_offset(pud, taddr);
58 if (pmd_present(*pmd))
59 pte = pte_offset_map(pmd, taddr);
60 }
61 }
62
63 return pte;
64}
65
Chen, Kenneth W39dde652006-12-06 20:32:03 -080066int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep)
67{
68 return 0;
69}
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#define mk_pte_huge(entry) { pte_val(entry) |= _PAGE_P; }
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073/*
David Gibson42b88be2006-03-22 00:09:01 -080074 * Don't actually need to do any preparation, but need to make sure
75 * the address is in the right region.
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 */
Hugh Dickins68589bc2006-11-14 02:03:32 -080077int prepare_hugepage_range(unsigned long addr, unsigned long len, pgoff_t pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078{
Hugh Dickins68589bc2006-11-14 02:03:32 -080079 if (pgoff & (~HPAGE_MASK >> PAGE_SHIFT))
80 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 if (len & ~HPAGE_MASK)
82 return -EINVAL;
83 if (addr & ~HPAGE_MASK)
84 return -EINVAL;
Peter Chubb0a41e252005-08-16 19:54:00 -070085 if (REGION_NUMBER(addr) != RGN_HPAGE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 return -EINVAL;
87
88 return 0;
89}
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091struct page *follow_huge_addr(struct mm_struct *mm, unsigned long addr, int write)
92{
93 struct page *page;
94 pte_t *ptep;
95
Peter Chubb0a41e252005-08-16 19:54:00 -070096 if (REGION_NUMBER(addr) != RGN_HPAGE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 return ERR_PTR(-EINVAL);
98
99 ptep = huge_pte_offset(mm, addr);
100 if (!ptep || pte_none(*ptep))
101 return NULL;
102 page = pte_page(*ptep);
103 page += ((addr & ~HPAGE_MASK) >> PAGE_SHIFT);
104 return page;
105}
106int pmd_huge(pmd_t pmd)
107{
108 return 0;
109}
110struct page *
111follow_huge_pmd(struct mm_struct *mm, unsigned long address, pmd_t *pmd, int write)
112{
113 return NULL;
114}
115
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700116void hugetlb_free_pgd_range(struct mmu_gather **tlb,
117 unsigned long addr, unsigned long end,
118 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700120 /*
Chen, Kenneth W2332c9a2006-03-22 10:49:00 -0800121 * This is called to free hugetlb page tables.
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700122 *
123 * The offset of these addresses from the base of the hugetlb
124 * region must be scaled down by HPAGE_SIZE/PAGE_SIZE so that
125 * the standard free_pgd_range will free the right page tables.
126 *
127 * If floor and ceiling are also in the hugetlb region, they
128 * must likewise be scaled down; but if outside, left unchanged.
129 */
130
131 addr = htlbpage_to_page(addr);
132 end = htlbpage_to_page(end);
Chen, Kenneth W2332c9a2006-03-22 10:49:00 -0800133 if (REGION_NUMBER(floor) == RGN_HPAGE)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700134 floor = htlbpage_to_page(floor);
Chen, Kenneth W2332c9a2006-03-22 10:49:00 -0800135 if (REGION_NUMBER(ceiling) == RGN_HPAGE)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700136 ceiling = htlbpage_to_page(ceiling);
137
138 free_pgd_range(tlb, addr, end, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139}
140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
142 unsigned long pgoff, unsigned long flags)
143{
144 struct vm_area_struct *vmm;
145
146 if (len > RGN_MAP_LIMIT)
147 return -ENOMEM;
148 if (len & ~HPAGE_MASK)
149 return -EINVAL;
Benjamin Herrenschmidtafa37392007-05-06 14:50:09 -0700150
151 /* Handle MAP_FIXED */
152 if (flags & MAP_FIXED) {
153 if (prepare_hugepage_range(addr, len, pgoff))
154 return -EINVAL;
155 return addr;
156 }
157
Peter Chubb0a41e252005-08-16 19:54:00 -0700158 /* This code assumes that RGN_HPAGE != 0. */
159 if ((REGION_NUMBER(addr) != RGN_HPAGE) || (addr & (HPAGE_SIZE - 1)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 addr = HPAGE_REGION_BASE;
161 else
162 addr = ALIGN(addr, HPAGE_SIZE);
163 for (vmm = find_vma(current->mm, addr); ; vmm = vmm->vm_next) {
164 /* At this point: (!vmm || addr < vmm->vm_end). */
165 if (REGION_OFFSET(addr) + len > RGN_MAP_LIMIT)
166 return -ENOMEM;
167 if (!vmm || (addr + len) <= vmm->vm_start)
168 return addr;
169 addr = ALIGN(vmm->vm_end, HPAGE_SIZE);
170 }
171}
172
173static int __init hugetlb_setup_sz(char *str)
174{
175 u64 tr_pages;
176 unsigned long long size;
177
178 if (ia64_pal_vm_page_size(&tr_pages, NULL) != 0)
179 /*
180 * shouldn't happen, but just in case.
181 */
182 tr_pages = 0x15557000UL;
183
184 size = memparse(str, &str);
185 if (*str || (size & (size-1)) || !(tr_pages & size) ||
186 size <= PAGE_SIZE ||
187 size >= (1UL << PAGE_SHIFT << MAX_ORDER)) {
188 printk(KERN_WARNING "Invalid huge page size specified\n");
189 return 1;
190 }
191
192 hpage_shift = __ffs(size);
193 /*
194 * boot cpu already executed ia64_mmu_init, and has HPAGE_SHIFT_DEFAULT
195 * override here with new page shift.
196 */
197 ia64_set_rr(HPAGE_REGION_BASE, hpage_shift << 2);
198 return 1;
199}
200__setup("hugepagesz=", hugetlb_setup_sz);