blob: 14a95054d4e058a85f6b8d80c162aa2617e7f848 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * IA-32 Huge TLB Page Support for Kernel.
3 *
4 * Copyright (C) 2002, Rohit Seth <rohit.seth@intel.com>
5 */
6
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#include <linux/init.h>
8#include <linux/fs.h>
9#include <linux/mm.h>
10#include <linux/hugetlb.h>
11#include <linux/pagemap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/err.h>
13#include <linux/sysctl.h>
14#include <asm/mman.h>
15#include <asm/tlb.h>
16#include <asm/tlbflush.h>
Jeremy Fitzhardingea5a19c62008-01-30 13:33:39 +010017#include <asm/pgalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#if 0 /* This is just for testing */
20struct page *
21follow_huge_addr(struct mm_struct *mm, unsigned long address, int write)
22{
23 unsigned long start = address;
24 int length = 1;
25 int nr;
26 struct page *page;
27 struct vm_area_struct *vma;
28
29 vma = find_vma(mm, addr);
30 if (!vma || !is_vm_hugetlb_page(vma))
31 return ERR_PTR(-EINVAL);
32
33 pte = huge_pte_offset(mm, address);
34
35 /* hugetlb should be locked, and hence, prefaulted */
36 WARN_ON(!pte || pte_none(*pte));
37
38 page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)];
39
Christoph Lameter25e59882008-03-26 21:03:04 -070040 WARN_ON(!PageHead(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42 return page;
43}
44
45int pmd_huge(pmd_t pmd)
46{
47 return 0;
48}
49
Andi Kleenceb86872008-07-23 21:27:50 -070050int pud_huge(pud_t pud)
51{
52 return 0;
53}
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#else
56
Naoya Horiguchicbef8472015-02-11 15:25:19 -080057/*
58 * pmd_huge() returns 1 if @pmd is hugetlb related entry, that is normal
59 * hugetlb entry or non-present (migration or hwpoisoned) hugetlb entry.
60 * Otherwise, returns 0.
61 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070062int pmd_huge(pmd_t pmd)
63{
Naoya Horiguchicbef8472015-02-11 15:25:19 -080064 return !pmd_none(pmd) &&
65 (pmd_val(pmd) & (_PAGE_PRESENT|_PAGE_PSE)) != _PAGE_PRESENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066}
67
Andi Kleenceb86872008-07-23 21:27:50 -070068int pud_huge(pud_t pud)
69{
Andi Kleen39c11e62008-07-23 21:27:50 -070070 return !!(pud_val(pud) & _PAGE_PSE);
Andi Kleenceb86872008-07-23 21:27:50 -070071}
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#endif
73
Kirill A. Shutemovfd8526a2013-11-19 15:17:50 +020074#ifdef CONFIG_HUGETLB_PAGE
Linus Torvalds1da177e2005-04-16 15:20:36 -070075static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *file,
76 unsigned long addr, unsigned long len,
77 unsigned long pgoff, unsigned long flags)
78{
Andi Kleen39c11e62008-07-23 21:27:50 -070079 struct hstate *h = hstate_file(file);
Michel Lespinassecdc17342012-12-11 16:02:02 -080080 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Michel Lespinassecdc17342012-12-11 16:02:02 -080082 info.flags = 0;
83 info.length = len;
Kirill A. Shutemovfd8526a2013-11-19 15:17:50 +020084 info.low_limit = current->mm->mmap_legacy_base;
Michel Lespinassecdc17342012-12-11 16:02:02 -080085 info.high_limit = TASK_SIZE;
86 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
87 info.align_offset = 0;
88 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089}
90
91static unsigned long hugetlb_get_unmapped_area_topdown(struct file *file,
92 unsigned long addr0, unsigned long len,
93 unsigned long pgoff, unsigned long flags)
94{
Andi Kleen39c11e62008-07-23 21:27:50 -070095 struct hstate *h = hstate_file(file);
Michel Lespinassecdc17342012-12-11 16:02:02 -080096 struct vm_unmapped_area_info info;
97 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
Michel Lespinassecdc17342012-12-11 16:02:02 -080099 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
100 info.length = len;
101 info.low_limit = PAGE_SIZE;
102 info.high_limit = current->mm->mmap_base;
103 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
104 info.align_offset = 0;
105 addr = vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 /*
108 * A failed mmap() very likely causes application failure,
109 * so fall back to the bottom-up function here. This scenario
110 * can happen with large stack limits and large mmap()
111 * allocations.
112 */
Michel Lespinassecdc17342012-12-11 16:02:02 -0800113 if (addr & ~PAGE_MASK) {
114 VM_BUG_ON(addr != -ENOMEM);
115 info.flags = 0;
116 info.low_limit = TASK_UNMAPPED_BASE;
117 info.high_limit = TASK_SIZE;
118 addr = vm_unmapped_area(&info);
119 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
121 return addr;
122}
123
124unsigned long
125hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
126 unsigned long len, unsigned long pgoff, unsigned long flags)
127{
Andi Kleen39c11e62008-07-23 21:27:50 -0700128 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 struct mm_struct *mm = current->mm;
130 struct vm_area_struct *vma;
131
Andi Kleen39c11e62008-07-23 21:27:50 -0700132 if (len & ~huge_page_mask(h))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 return -EINVAL;
134 if (len > TASK_SIZE)
135 return -ENOMEM;
136
Benjamin Herrenschmidt5a8130f2007-05-06 14:50:08 -0700137 if (flags & MAP_FIXED) {
Andi Kleena5516432008-07-23 21:27:41 -0700138 if (prepare_hugepage_range(file, addr, len))
Benjamin Herrenschmidt5a8130f2007-05-06 14:50:08 -0700139 return -EINVAL;
140 return addr;
141 }
142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 if (addr) {
Andi Kleen39c11e62008-07-23 21:27:50 -0700144 addr = ALIGN(addr, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 vma = find_vma(mm, addr);
146 if (TASK_SIZE - len >= addr &&
147 (!vma || addr + len <= vma->vm_start))
148 return addr;
149 }
150 if (mm->get_unmapped_area == arch_get_unmapped_area)
151 return hugetlb_get_unmapped_area_bottomup(file, addr, len,
152 pgoff, flags);
153 else
154 return hugetlb_get_unmapped_area_topdown(file, addr, len,
155 pgoff, flags);
156}
Kirill A. Shutemovfd8526a2013-11-19 15:17:50 +0200157#endif /* CONFIG_HUGETLB_PAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
Andi Kleenb4718e62008-07-23 21:27:51 -0700159#ifdef CONFIG_X86_64
160static __init int setup_hugepagesz(char *opt)
161{
162 unsigned long ps = memparse(opt, &opt);
163 if (ps == PMD_SIZE) {
164 hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT);
Borislav Petkovb8291adc2016-03-29 17:41:58 +0200165 } else if (ps == PUD_SIZE && boot_cpu_has(X86_FEATURE_GBPAGES)) {
Andi Kleenb4718e62008-07-23 21:27:51 -0700166 hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
167 } else {
168 printk(KERN_ERR "hugepagesz: Unsupported page size %lu M\n",
169 ps >> 20);
170 return 0;
171 }
172 return 1;
173}
174__setup("hugepagesz=", setup_hugepagesz);
Kirill A. Shutemovece84b32015-02-10 14:08:19 -0800175
Vlastimil Babka080fe202016-02-05 15:36:41 -0800176#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Kirill A. Shutemovece84b32015-02-10 14:08:19 -0800177static __init int gigantic_pages_init(void)
178{
Vlastimil Babka080fe202016-02-05 15:36:41 -0800179 /* With compaction or CMA we can allocate gigantic pages at runtime */
Borislav Petkovb8291adc2016-03-29 17:41:58 +0200180 if (boot_cpu_has(X86_FEATURE_GBPAGES) && !size_to_hstate(1UL << PUD_SHIFT))
Kirill A. Shutemovece84b32015-02-10 14:08:19 -0800181 hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
182 return 0;
183}
184arch_initcall(gigantic_pages_init);
185#endif
Andi Kleenb4718e62008-07-23 21:27:51 -0700186#endif