Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright 2002 Andi Kleen, SuSE Labs. | |
3 | * Thanks to Ben LaHaise for precious feedback. | |
4 | */ | |
5 | ||
6 | #include <linux/config.h> | |
7 | #include <linux/mm.h> | |
8 | #include <linux/sched.h> | |
9 | #include <linux/highmem.h> | |
10 | #include <linux/module.h> | |
11 | #include <linux/slab.h> | |
12 | #include <asm/uaccess.h> | |
13 | #include <asm/processor.h> | |
14 | #include <asm/tlbflush.h> | |
15 | #include <asm/io.h> | |
16 | ||
17 | static inline pte_t *lookup_address(unsigned long address) | |
18 | { | |
19 | pgd_t *pgd = pgd_offset_k(address); | |
20 | pud_t *pud; | |
21 | pmd_t *pmd; | |
22 | pte_t *pte; | |
23 | if (pgd_none(*pgd)) | |
24 | return NULL; | |
25 | pud = pud_offset(pgd, address); | |
26 | if (!pud_present(*pud)) | |
27 | return NULL; | |
28 | pmd = pmd_offset(pud, address); | |
29 | if (!pmd_present(*pmd)) | |
30 | return NULL; | |
31 | if (pmd_large(*pmd)) | |
32 | return (pte_t *)pmd; | |
33 | pte = pte_offset_kernel(pmd, address); | |
34 | if (pte && !pte_present(*pte)) | |
35 | pte = NULL; | |
36 | return pte; | |
37 | } | |
38 | ||
39 | static struct page *split_large_page(unsigned long address, pgprot_t prot, | |
40 | pgprot_t ref_prot) | |
41 | { | |
42 | int i; | |
43 | unsigned long addr; | |
44 | struct page *base = alloc_pages(GFP_KERNEL, 0); | |
45 | pte_t *pbase; | |
46 | if (!base) | |
47 | return NULL; | |
48 | address = __pa(address); | |
49 | addr = address & LARGE_PAGE_MASK; | |
50 | pbase = (pte_t *)page_address(base); | |
51 | for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE) { | |
52 | pbase[i] = pfn_pte(addr >> PAGE_SHIFT, | |
53 | addr == address ? prot : ref_prot); | |
54 | } | |
55 | return base; | |
56 | } | |
57 | ||
58 | ||
59 | static void flush_kernel_map(void *address) | |
60 | { | |
61 | if (0 && address && cpu_has_clflush) { | |
62 | /* is this worth it? */ | |
63 | int i; | |
64 | for (i = 0; i < PAGE_SIZE; i += boot_cpu_data.x86_clflush_size) | |
65 | asm volatile("clflush (%0)" :: "r" (address + i)); | |
66 | } else | |
67 | asm volatile("wbinvd":::"memory"); | |
68 | if (address) | |
69 | __flush_tlb_one(address); | |
70 | else | |
71 | __flush_tlb_all(); | |
72 | } | |
73 | ||
74 | ||
75 | static inline void flush_map(unsigned long address) | |
76 | { | |
77 | on_each_cpu(flush_kernel_map, (void *)address, 1, 1); | |
78 | } | |
79 | ||
20aaffd6 NP |
80 | static struct page *deferred_pages; /* protected by init_mm.mmap_sem */ |
81 | ||
82 | static inline void save_page(struct page *fpage) | |
1da177e4 | 83 | { |
20aaffd6 NP |
84 | fpage->lru.next = (struct list_head *)deferred_pages; |
85 | deferred_pages = fpage; | |
1da177e4 LT |
86 | } |
87 | ||
88 | /* | |
89 | * No more special protections in this 2/4MB area - revert to a | |
90 | * large page again. | |
91 | */ | |
92 | static void revert_page(unsigned long address, pgprot_t ref_prot) | |
93 | { | |
94 | pgd_t *pgd; | |
95 | pud_t *pud; | |
96 | pmd_t *pmd; | |
97 | pte_t large_pte; | |
98 | ||
99 | pgd = pgd_offset_k(address); | |
100 | BUG_ON(pgd_none(*pgd)); | |
101 | pud = pud_offset(pgd,address); | |
102 | BUG_ON(pud_none(*pud)); | |
103 | pmd = pmd_offset(pud, address); | |
104 | BUG_ON(pmd_val(*pmd) & _PAGE_PSE); | |
105 | pgprot_val(ref_prot) |= _PAGE_PSE; | |
106 | large_pte = mk_pte_phys(__pa(address) & LARGE_PAGE_MASK, ref_prot); | |
107 | set_pte((pte_t *)pmd, large_pte); | |
108 | } | |
109 | ||
110 | static int | |
111 | __change_page_attr(unsigned long address, unsigned long pfn, pgprot_t prot, | |
112 | pgprot_t ref_prot) | |
113 | { | |
114 | pte_t *kpte; | |
115 | struct page *kpte_page; | |
116 | unsigned kpte_flags; | |
c728252c | 117 | pgprot_t ref_prot2; |
1da177e4 LT |
118 | kpte = lookup_address(address); |
119 | if (!kpte) return 0; | |
120 | kpte_page = virt_to_page(((unsigned long)kpte) & PAGE_MASK); | |
121 | kpte_flags = pte_val(*kpte); | |
122 | if (pgprot_val(prot) != pgprot_val(ref_prot)) { | |
123 | if ((kpte_flags & _PAGE_PSE) == 0) { | |
124 | set_pte(kpte, pfn_pte(pfn, prot)); | |
125 | } else { | |
126 | /* | |
127 | * split_large_page will take the reference for this change_page_attr | |
128 | * on the split page. | |
129 | */ | |
c728252c AV |
130 | |
131 | struct page *split; | |
132 | ref_prot2 = __pgprot(pgprot_val(pte_pgprot(*lookup_address(address))) & ~(1<<_PAGE_BIT_PSE)); | |
133 | ||
134 | split = split_large_page(address, prot, ref_prot2); | |
1da177e4 LT |
135 | if (!split) |
136 | return -ENOMEM; | |
c728252c | 137 | set_pte(kpte,mk_pte(split, ref_prot2)); |
1da177e4 LT |
138 | kpte_page = split; |
139 | } | |
140 | get_page(kpte_page); | |
141 | } else if ((kpte_flags & _PAGE_PSE) == 0) { | |
142 | set_pte(kpte, pfn_pte(pfn, ref_prot)); | |
143 | __put_page(kpte_page); | |
144 | } else | |
145 | BUG(); | |
146 | ||
147 | /* on x86-64 the direct mapping set at boot is not using 4k pages */ | |
148 | BUG_ON(PageReserved(kpte_page)); | |
149 | ||
150 | switch (page_count(kpte_page)) { | |
151 | case 1: | |
20aaffd6 | 152 | save_page(kpte_page); |
1da177e4 LT |
153 | revert_page(address, ref_prot); |
154 | break; | |
155 | case 0: | |
156 | BUG(); /* memleak and failed 2M page regeneration */ | |
157 | } | |
158 | return 0; | |
159 | } | |
160 | ||
161 | /* | |
162 | * Change the page attributes of an page in the linear mapping. | |
163 | * | |
164 | * This should be used when a page is mapped with a different caching policy | |
165 | * than write-back somewhere - some CPUs do not like it when mappings with | |
166 | * different caching policies exist. This changes the page attributes of the | |
167 | * in kernel linear mapping too. | |
168 | * | |
169 | * The caller needs to ensure that there are no conflicting mappings elsewhere. | |
170 | * This function only deals with the kernel linear map. | |
171 | * | |
172 | * Caller must call global_flush_tlb() after this. | |
173 | */ | |
174 | int change_page_attr_addr(unsigned long address, int numpages, pgprot_t prot) | |
175 | { | |
176 | int err = 0; | |
177 | int i; | |
178 | ||
179 | down_write(&init_mm.mmap_sem); | |
180 | for (i = 0; i < numpages; i++, address += PAGE_SIZE) { | |
181 | unsigned long pfn = __pa(address) >> PAGE_SHIFT; | |
182 | ||
183 | err = __change_page_attr(address, pfn, prot, PAGE_KERNEL); | |
184 | if (err) | |
185 | break; | |
186 | /* Handle kernel mapping too which aliases part of the | |
187 | * lowmem */ | |
188 | if (__pa(address) < KERNEL_TEXT_SIZE) { | |
189 | unsigned long addr2; | |
190 | pgprot_t prot2 = prot; | |
191 | addr2 = __START_KERNEL_map + __pa(address); | |
192 | pgprot_val(prot2) &= ~_PAGE_NX; | |
193 | err = __change_page_attr(addr2, pfn, prot2, PAGE_KERNEL_EXEC); | |
194 | } | |
195 | } | |
196 | up_write(&init_mm.mmap_sem); | |
197 | return err; | |
198 | } | |
199 | ||
200 | /* Don't call this for MMIO areas that may not have a mem_map entry */ | |
201 | int change_page_attr(struct page *page, int numpages, pgprot_t prot) | |
202 | { | |
203 | unsigned long addr = (unsigned long)page_address(page); | |
204 | return change_page_attr_addr(addr, numpages, prot); | |
205 | } | |
206 | ||
207 | void global_flush_tlb(void) | |
208 | { | |
20aaffd6 | 209 | struct page *dpage; |
1da177e4 LT |
210 | |
211 | down_read(&init_mm.mmap_sem); | |
20aaffd6 | 212 | dpage = xchg(&deferred_pages, NULL); |
1da177e4 | 213 | up_read(&init_mm.mmap_sem); |
20aaffd6 NP |
214 | |
215 | flush_map((dpage && !dpage->lru.next) ? (unsigned long)page_address(dpage) : 0); | |
216 | while (dpage) { | |
217 | struct page *tmp = dpage; | |
218 | dpage = (struct page *)dpage->lru.next; | |
219 | __free_page(tmp); | |
1da177e4 LT |
220 | } |
221 | } | |
222 | ||
223 | EXPORT_SYMBOL(change_page_attr); | |
224 | EXPORT_SYMBOL(global_flush_tlb); |