blob: f50d4b3f501ad6a5cd0fcf10dfd66b82500b0e54 [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/sysctl.h>
vignesh babu9be26f42007-06-07 15:27:46 +053018#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <asm/mman.h>
20#include <asm/pgalloc.h>
21#include <asm/tlb.h>
22#include <asm/tlbflush.h>
23
Jack Steiner34d8a382008-07-29 22:33:54 -070024unsigned int hpage_shift = HPAGE_SHIFT_DEFAULT;
25EXPORT_SYMBOL(hpage_shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
David Gibson63551ae2005-06-21 17:14:44 -070027pte_t *
Andi Kleena5516432008-07-23 21:27:41 -070028huge_pte_alloc(struct mm_struct *mm, unsigned long addr, unsigned long sz)
Linus Torvalds1da177e2005-04-16 15:20:36 -070029{
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)
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080041 pte = pte_alloc_map(mm, NULL, pmd, taddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 }
43 return pte;
44}
45
David Gibson63551ae2005-06-21 17:14:44 -070046pte_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -070047huge_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 Torvalds1da177e2005-04-16 15:20:36 -070070/*
David Gibson42b88be2006-03-22 00:09:01 -080071 * Don't actually need to do any preparation, but need to make sure
72 * the address is in the right region.
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 */
Andi Kleena5516432008-07-23 21:27:41 -070074int prepare_hugepage_range(struct file *file,
75 unsigned long addr, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
77 if (len & ~HPAGE_MASK)
78 return -EINVAL;
79 if (addr & ~HPAGE_MASK)
80 return -EINVAL;
Peter Chubb0a41e252005-08-16 19:54:00 -070081 if (REGION_NUMBER(addr) != RGN_HPAGE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 return -EINVAL;
83
84 return 0;
85}
86
Linus Torvalds1da177e2005-04-16 15:20:36 -070087struct page *follow_huge_addr(struct mm_struct *mm, unsigned long addr, int write)
88{
89 struct page *page;
90 pte_t *ptep;
91
Peter Chubb0a41e252005-08-16 19:54:00 -070092 if (REGION_NUMBER(addr) != RGN_HPAGE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 return ERR_PTR(-EINVAL);
94
95 ptep = huge_pte_offset(mm, addr);
96 if (!ptep || pte_none(*ptep))
97 return NULL;
98 page = pte_page(*ptep);
99 page += ((addr & ~HPAGE_MASK) >> PAGE_SHIFT);
100 return page;
101}
102int pmd_huge(pmd_t pmd)
103{
104 return 0;
105}
Andi Kleenceb86872008-07-23 21:27:50 -0700106
107int pud_huge(pud_t pud)
108{
109 return 0;
110}
111
Jan Beulich42b77722008-07-23 21:27:10 -0700112void hugetlb_free_pgd_range(struct mmu_gather *tlb,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700113 unsigned long addr, unsigned long end,
114 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115{
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700116 /*
Chen, Kenneth W2332c9a2006-03-22 10:49:00 -0800117 * This is called to free hugetlb page tables.
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700118 *
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);
Chen, Kenneth W2332c9a2006-03-22 10:49:00 -0800129 if (REGION_NUMBER(floor) == RGN_HPAGE)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700130 floor = htlbpage_to_page(floor);
Chen, Kenneth W2332c9a2006-03-22 10:49:00 -0800131 if (REGION_NUMBER(ceiling) == RGN_HPAGE)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700132 ceiling = htlbpage_to_page(ceiling);
133
134 free_pgd_range(tlb, addr, end, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135}
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
138 unsigned long pgoff, unsigned long flags)
139{
Michel Lespinassee31048a2013-02-21 15:10:29 -0800140 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
142 if (len > RGN_MAP_LIMIT)
143 return -ENOMEM;
144 if (len & ~HPAGE_MASK)
145 return -EINVAL;
Benjamin Herrenschmidtafa37392007-05-06 14:50:09 -0700146
147 /* Handle MAP_FIXED */
148 if (flags & MAP_FIXED) {
Andi Kleena5516432008-07-23 21:27:41 -0700149 if (prepare_hugepage_range(file, addr, len))
Benjamin Herrenschmidtafa37392007-05-06 14:50:09 -0700150 return -EINVAL;
151 return addr;
152 }
153
Peter Chubb0a41e252005-08-16 19:54:00 -0700154 /* This code assumes that RGN_HPAGE != 0. */
155 if ((REGION_NUMBER(addr) != RGN_HPAGE) || (addr & (HPAGE_SIZE - 1)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 addr = HPAGE_REGION_BASE;
Michel Lespinassee31048a2013-02-21 15:10:29 -0800157
158 info.flags = 0;
159 info.length = len;
160 info.low_limit = addr;
161 info.high_limit = HPAGE_REGION_BASE + RGN_MAP_LIMIT;
162 info.align_mask = PAGE_MASK & (HPAGE_SIZE - 1);
163 info.align_offset = 0;
164 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165}
166
167static int __init hugetlb_setup_sz(char *str)
168{
169 u64 tr_pages;
170 unsigned long long size;
171
172 if (ia64_pal_vm_page_size(&tr_pages, NULL) != 0)
173 /*
174 * shouldn't happen, but just in case.
175 */
176 tr_pages = 0x15557000UL;
177
178 size = memparse(str, &str);
vignesh babu9be26f42007-06-07 15:27:46 +0530179 if (*str || !is_power_of_2(size) || !(tr_pages & size) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 size <= PAGE_SIZE ||
181 size >= (1UL << PAGE_SHIFT << MAX_ORDER)) {
182 printk(KERN_WARNING "Invalid huge page size specified\n");
183 return 1;
184 }
185
186 hpage_shift = __ffs(size);
187 /*
188 * boot cpu already executed ia64_mmu_init, and has HPAGE_SHIFT_DEFAULT
189 * override here with new page shift.
190 */
191 ia64_set_rr(HPAGE_REGION_BASE, hpage_shift << 2);
Mel Gormand9c23402007-10-16 01:26:01 -0700192 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193}
Mel Gormand9c23402007-10-16 01:26:01 -0700194early_param("hugepagesz", hugetlb_setup_sz);