blob: b0f615759e97461052eb154c431487c90d4cbad3 [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>
Jack Steiner34d8a382008-07-29 22:33:54 -070016#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/slab.h>
18#include <linux/sysctl.h>
vignesh babu9be26f42007-06-07 15:27:46 +053019#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <asm/mman.h>
21#include <asm/pgalloc.h>
22#include <asm/tlb.h>
23#include <asm/tlbflush.h>
24
Jack Steiner34d8a382008-07-29 22:33:54 -070025unsigned int hpage_shift = HPAGE_SHIFT_DEFAULT;
26EXPORT_SYMBOL(hpage_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
David Gibson63551ae2005-06-21 17:14:44 -070028pte_t *
Andi Kleena5516432008-07-23 21:27:41 -070029huge_pte_alloc(struct mm_struct *mm, unsigned long addr, unsigned long sz)
Linus Torvalds1da177e2005-04-16 15:20:36 -070030{
31 unsigned long taddr = htlbpage_to_page(addr);
32 pgd_t *pgd;
33 pud_t *pud;
34 pmd_t *pmd;
35 pte_t *pte = NULL;
36
37 pgd = pgd_offset(mm, taddr);
38 pud = pud_alloc(mm, pgd, taddr);
39 if (pud) {
40 pmd = pmd_alloc(mm, pud, taddr);
41 if (pmd)
42 pte = pte_alloc_map(mm, pmd, taddr);
43 }
44 return pte;
45}
46
David Gibson63551ae2005-06-21 17:14:44 -070047pte_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -070048huge_pte_offset (struct mm_struct *mm, unsigned long addr)
49{
50 unsigned long taddr = htlbpage_to_page(addr);
51 pgd_t *pgd;
52 pud_t *pud;
53 pmd_t *pmd;
54 pte_t *pte = NULL;
55
56 pgd = pgd_offset(mm, taddr);
57 if (pgd_present(*pgd)) {
58 pud = pud_offset(pgd, taddr);
59 if (pud_present(*pud)) {
60 pmd = pmd_offset(pud, taddr);
61 if (pmd_present(*pmd))
62 pte = pte_offset_map(pmd, taddr);
63 }
64 }
65
66 return pte;
67}
68
Chen, Kenneth W39dde652006-12-06 20:32:03 -080069int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep)
70{
71 return 0;
72}
73
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#define mk_pte_huge(entry) { pte_val(entry) |= _PAGE_P; }
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/*
David Gibson42b88be2006-03-22 00:09:01 -080077 * Don't actually need to do any preparation, but need to make sure
78 * the address is in the right region.
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 */
Andi Kleena5516432008-07-23 21:27:41 -070080int prepare_hugepage_range(struct file *file,
81 unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
83 if (len & ~HPAGE_MASK)
84 return -EINVAL;
85 if (addr & ~HPAGE_MASK)
86 return -EINVAL;
Peter Chubb0a41e252005-08-16 19:54:00 -070087 if (REGION_NUMBER(addr) != RGN_HPAGE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 return -EINVAL;
89
90 return 0;
91}
92
Linus Torvalds1da177e2005-04-16 15:20:36 -070093struct page *follow_huge_addr(struct mm_struct *mm, unsigned long addr, int write)
94{
95 struct page *page;
96 pte_t *ptep;
97
Peter Chubb0a41e252005-08-16 19:54:00 -070098 if (REGION_NUMBER(addr) != RGN_HPAGE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 return ERR_PTR(-EINVAL);
100
101 ptep = huge_pte_offset(mm, addr);
102 if (!ptep || pte_none(*ptep))
103 return NULL;
104 page = pte_page(*ptep);
105 page += ((addr & ~HPAGE_MASK) >> PAGE_SHIFT);
106 return page;
107}
108int pmd_huge(pmd_t pmd)
109{
110 return 0;
111}
Andi Kleenceb86872008-07-23 21:27:50 -0700112
113int pud_huge(pud_t pud)
114{
115 return 0;
116}
117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118struct page *
119follow_huge_pmd(struct mm_struct *mm, unsigned long address, pmd_t *pmd, int write)
120{
121 return NULL;
122}
123
Jan Beulich42b77722008-07-23 21:27:10 -0700124void hugetlb_free_pgd_range(struct mmu_gather *tlb,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700125 unsigned long addr, unsigned long end,
126 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127{
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700128 /*
Chen, Kenneth W2332c9a2006-03-22 10:49:00 -0800129 * This is called to free hugetlb page tables.
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700130 *
131 * The offset of these addresses from the base of the hugetlb
132 * region must be scaled down by HPAGE_SIZE/PAGE_SIZE so that
133 * the standard free_pgd_range will free the right page tables.
134 *
135 * If floor and ceiling are also in the hugetlb region, they
136 * must likewise be scaled down; but if outside, left unchanged.
137 */
138
139 addr = htlbpage_to_page(addr);
140 end = htlbpage_to_page(end);
Chen, Kenneth W2332c9a2006-03-22 10:49:00 -0800141 if (REGION_NUMBER(floor) == RGN_HPAGE)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700142 floor = htlbpage_to_page(floor);
Chen, Kenneth W2332c9a2006-03-22 10:49:00 -0800143 if (REGION_NUMBER(ceiling) == RGN_HPAGE)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700144 ceiling = htlbpage_to_page(ceiling);
145
146 free_pgd_range(tlb, addr, end, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
150 unsigned long pgoff, unsigned long flags)
151{
152 struct vm_area_struct *vmm;
153
154 if (len > RGN_MAP_LIMIT)
155 return -ENOMEM;
156 if (len & ~HPAGE_MASK)
157 return -EINVAL;
Benjamin Herrenschmidtafa37392007-05-06 14:50:09 -0700158
159 /* Handle MAP_FIXED */
160 if (flags & MAP_FIXED) {
Andi Kleena5516432008-07-23 21:27:41 -0700161 if (prepare_hugepage_range(file, addr, len))
Benjamin Herrenschmidtafa37392007-05-06 14:50:09 -0700162 return -EINVAL;
163 return addr;
164 }
165
Peter Chubb0a41e252005-08-16 19:54:00 -0700166 /* This code assumes that RGN_HPAGE != 0. */
167 if ((REGION_NUMBER(addr) != RGN_HPAGE) || (addr & (HPAGE_SIZE - 1)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 addr = HPAGE_REGION_BASE;
169 else
170 addr = ALIGN(addr, HPAGE_SIZE);
171 for (vmm = find_vma(current->mm, addr); ; vmm = vmm->vm_next) {
172 /* At this point: (!vmm || addr < vmm->vm_end). */
173 if (REGION_OFFSET(addr) + len > RGN_MAP_LIMIT)
174 return -ENOMEM;
175 if (!vmm || (addr + len) <= vmm->vm_start)
176 return addr;
177 addr = ALIGN(vmm->vm_end, HPAGE_SIZE);
178 }
179}
180
181static int __init hugetlb_setup_sz(char *str)
182{
183 u64 tr_pages;
184 unsigned long long size;
185
186 if (ia64_pal_vm_page_size(&tr_pages, NULL) != 0)
187 /*
188 * shouldn't happen, but just in case.
189 */
190 tr_pages = 0x15557000UL;
191
192 size = memparse(str, &str);
vignesh babu9be26f42007-06-07 15:27:46 +0530193 if (*str || !is_power_of_2(size) || !(tr_pages & size) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 size <= PAGE_SIZE ||
195 size >= (1UL << PAGE_SHIFT << MAX_ORDER)) {
196 printk(KERN_WARNING "Invalid huge page size specified\n");
197 return 1;
198 }
199
200 hpage_shift = __ffs(size);
201 /*
202 * boot cpu already executed ia64_mmu_init, and has HPAGE_SHIFT_DEFAULT
203 * override here with new page shift.
204 */
205 ia64_set_rr(HPAGE_REGION_BASE, hpage_shift << 2);
Mel Gormand9c23402007-10-16 01:26:01 -0700206 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
Mel Gormand9c23402007-10-16 01:26:01 -0700208early_param("hugepagesz", hugetlb_setup_sz);