blob: 302f43fd9c28c6d14652981d672569d57e3dea35 [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>
Ingo Molnar01042602017-02-08 18:51:31 +010010#include <linux/sched/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/hugetlb.h>
12#include <linux/pagemap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/err.h>
14#include <linux/sysctl.h>
Dmitry Safonove13b73d2017-03-14 14:41:26 +030015#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <asm/mman.h>
17#include <asm/tlb.h>
18#include <asm/tlbflush.h>
Jeremy Fitzhardingea5a19c62008-01-30 13:33:39 +010019#include <asm/pgalloc.h>
Dmitry Safonove13b73d2017-03-14 14:41:26 +030020#include <asm/elf.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#if 0 /* This is just for testing */
23struct page *
24follow_huge_addr(struct mm_struct *mm, unsigned long address, int write)
25{
26 unsigned long start = address;
27 int length = 1;
28 int nr;
29 struct page *page;
30 struct vm_area_struct *vma;
31
32 vma = find_vma(mm, addr);
33 if (!vma || !is_vm_hugetlb_page(vma))
34 return ERR_PTR(-EINVAL);
35
36 pte = huge_pte_offset(mm, address);
37
38 /* hugetlb should be locked, and hence, prefaulted */
39 WARN_ON(!pte || pte_none(*pte));
40
41 page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)];
42
Christoph Lameter25e59882008-03-26 21:03:04 -070043 WARN_ON(!PageHead(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45 return page;
46}
47
48int pmd_huge(pmd_t pmd)
49{
50 return 0;
51}
52
Andi Kleenceb86872008-07-23 21:27:50 -070053int pud_huge(pud_t pud)
54{
55 return 0;
56}
57
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#else
59
Naoya Horiguchicbef8472015-02-11 15:25:19 -080060/*
61 * pmd_huge() returns 1 if @pmd is hugetlb related entry, that is normal
62 * hugetlb entry or non-present (migration or hwpoisoned) hugetlb entry.
63 * Otherwise, returns 0.
64 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070065int pmd_huge(pmd_t pmd)
66{
Naoya Horiguchicbef8472015-02-11 15:25:19 -080067 return !pmd_none(pmd) &&
68 (pmd_val(pmd) & (_PAGE_PRESENT|_PAGE_PSE)) != _PAGE_PRESENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069}
70
Andi Kleenceb86872008-07-23 21:27:50 -070071int pud_huge(pud_t pud)
72{
Andi Kleen39c11e62008-07-23 21:27:50 -070073 return !!(pud_val(pud) & _PAGE_PSE);
Andi Kleenceb86872008-07-23 21:27:50 -070074}
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#endif
76
Kirill A. Shutemovfd8526a2013-11-19 15:17:50 +020077#ifdef CONFIG_HUGETLB_PAGE
Linus Torvalds1da177e2005-04-16 15:20:36 -070078static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *file,
79 unsigned long addr, unsigned long len,
80 unsigned long pgoff, unsigned long flags)
81{
Andi Kleen39c11e62008-07-23 21:27:50 -070082 struct hstate *h = hstate_file(file);
Michel Lespinassecdc17342012-12-11 16:02:02 -080083 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Michel Lespinassecdc17342012-12-11 16:02:02 -080085 info.flags = 0;
86 info.length = len;
Dmitry Safonove13b73d2017-03-14 14:41:26 +030087 info.low_limit = get_mmap_base(1);
88 info.high_limit = in_compat_syscall() ?
89 tasksize_32bit() : tasksize_64bit();
Michel Lespinassecdc17342012-12-11 16:02:02 -080090 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
91 info.align_offset = 0;
92 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093}
94
95static unsigned long hugetlb_get_unmapped_area_topdown(struct file *file,
96 unsigned long addr0, unsigned long len,
97 unsigned long pgoff, unsigned long flags)
98{
Andi Kleen39c11e62008-07-23 21:27:50 -070099 struct hstate *h = hstate_file(file);
Michel Lespinassecdc17342012-12-11 16:02:02 -0800100 struct vm_unmapped_area_info info;
101 unsigned long addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
Michel Lespinassecdc17342012-12-11 16:02:02 -0800103 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
104 info.length = len;
105 info.low_limit = PAGE_SIZE;
Dmitry Safonove13b73d2017-03-14 14:41:26 +0300106 info.high_limit = get_mmap_base(0);
Michel Lespinassecdc17342012-12-11 16:02:02 -0800107 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
108 info.align_offset = 0;
109 addr = vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 /*
112 * A failed mmap() very likely causes application failure,
113 * so fall back to the bottom-up function here. This scenario
114 * can happen with large stack limits and large mmap()
115 * allocations.
116 */
Michel Lespinassecdc17342012-12-11 16:02:02 -0800117 if (addr & ~PAGE_MASK) {
118 VM_BUG_ON(addr != -ENOMEM);
119 info.flags = 0;
120 info.low_limit = TASK_UNMAPPED_BASE;
121 info.high_limit = TASK_SIZE;
122 addr = vm_unmapped_area(&info);
123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
125 return addr;
126}
127
128unsigned long
129hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
130 unsigned long len, unsigned long pgoff, unsigned long flags)
131{
Andi Kleen39c11e62008-07-23 21:27:50 -0700132 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 struct mm_struct *mm = current->mm;
134 struct vm_area_struct *vma;
135
Andi Kleen39c11e62008-07-23 21:27:50 -0700136 if (len & ~huge_page_mask(h))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 return -EINVAL;
138 if (len > TASK_SIZE)
139 return -ENOMEM;
140
Benjamin Herrenschmidt5a8130f2007-05-06 14:50:08 -0700141 if (flags & MAP_FIXED) {
Andi Kleena5516432008-07-23 21:27:41 -0700142 if (prepare_hugepage_range(file, addr, len))
Benjamin Herrenschmidt5a8130f2007-05-06 14:50:08 -0700143 return -EINVAL;
144 return addr;
145 }
146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 if (addr) {
Andi Kleen39c11e62008-07-23 21:27:50 -0700148 addr = ALIGN(addr, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 vma = find_vma(mm, addr);
150 if (TASK_SIZE - len >= addr &&
151 (!vma || addr + len <= vma->vm_start))
152 return addr;
153 }
154 if (mm->get_unmapped_area == arch_get_unmapped_area)
155 return hugetlb_get_unmapped_area_bottomup(file, addr, len,
156 pgoff, flags);
157 else
158 return hugetlb_get_unmapped_area_topdown(file, addr, len,
159 pgoff, flags);
160}
Kirill A. Shutemovfd8526a2013-11-19 15:17:50 +0200161#endif /* CONFIG_HUGETLB_PAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Andi Kleenb4718e62008-07-23 21:27:51 -0700163#ifdef CONFIG_X86_64
164static __init int setup_hugepagesz(char *opt)
165{
166 unsigned long ps = memparse(opt, &opt);
167 if (ps == PMD_SIZE) {
168 hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT);
Borislav Petkovb8291adc2016-03-29 17:41:58 +0200169 } else if (ps == PUD_SIZE && boot_cpu_has(X86_FEATURE_GBPAGES)) {
Andi Kleenb4718e62008-07-23 21:27:51 -0700170 hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
171 } else {
Vaishali Thakkar2b18e532016-05-19 17:11:20 -0700172 hugetlb_bad_size();
Andi Kleenb4718e62008-07-23 21:27:51 -0700173 printk(KERN_ERR "hugepagesz: Unsupported page size %lu M\n",
174 ps >> 20);
175 return 0;
176 }
177 return 1;
178}
179__setup("hugepagesz=", setup_hugepagesz);
Kirill A. Shutemovece84b32015-02-10 14:08:19 -0800180
Vlastimil Babka080fe202016-02-05 15:36:41 -0800181#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Kirill A. Shutemovece84b32015-02-10 14:08:19 -0800182static __init int gigantic_pages_init(void)
183{
Vlastimil Babka080fe202016-02-05 15:36:41 -0800184 /* With compaction or CMA we can allocate gigantic pages at runtime */
Borislav Petkovb8291adc2016-03-29 17:41:58 +0200185 if (boot_cpu_has(X86_FEATURE_GBPAGES) && !size_to_hstate(1UL << PUD_SHIFT))
Kirill A. Shutemovece84b32015-02-10 14:08:19 -0800186 hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
187 return 0;
188}
189arch_initcall(gigantic_pages_init);
190#endif
Andi Kleenb4718e62008-07-23 21:27:51 -0700191#endif