Fix TLB gather virtual address range invalidation corner cases
[deliverable/linux.git] / arch / arm / include / asm / tlb.h
1 /*
2 * arch/arm/include/asm/tlb.h
3 *
4 * Copyright (C) 2002 Russell King
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * Experimentation shows that on a StrongARM, it appears to be faster
11 * to use the "invalidate whole tlb" rather than "invalidate single
12 * tlb" for this.
13 *
14 * This appears true for both the process fork+exit case, as well as
15 * the munmap-large-area case.
16 */
17 #ifndef __ASMARM_TLB_H
18 #define __ASMARM_TLB_H
19
20 #include <asm/cacheflush.h>
21
22 #ifndef CONFIG_MMU
23
24 #include <linux/pagemap.h>
25
26 #define tlb_flush(tlb) ((void) tlb)
27
28 #include <asm-generic/tlb.h>
29
30 #else /* !CONFIG_MMU */
31
32 #include <linux/swap.h>
33 #include <asm/pgalloc.h>
34 #include <asm/tlbflush.h>
35
36 #define MMU_GATHER_BUNDLE 8
37
38 /*
39 * TLB handling. This allows us to remove pages from the page
40 * tables, and efficiently handle the TLB issues.
41 */
42 struct mmu_gather {
43 struct mm_struct *mm;
44 unsigned int fullmm;
45 struct vm_area_struct *vma;
46 unsigned long start, end;
47 unsigned long range_start;
48 unsigned long range_end;
49 unsigned int nr;
50 unsigned int max;
51 struct page **pages;
52 struct page *local[MMU_GATHER_BUNDLE];
53 };
54
55 DECLARE_PER_CPU(struct mmu_gather, mmu_gathers);
56
57 /*
58 * This is unnecessarily complex. There's three ways the TLB shootdown
59 * code is used:
60 * 1. Unmapping a range of vmas. See zap_page_range(), unmap_region().
61 * tlb->fullmm = 0, and tlb_start_vma/tlb_end_vma will be called.
62 * tlb->vma will be non-NULL.
63 * 2. Unmapping all vmas. See exit_mmap().
64 * tlb->fullmm = 1, and tlb_start_vma/tlb_end_vma will be called.
65 * tlb->vma will be non-NULL. Additionally, page tables will be freed.
66 * 3. Unmapping argument pages. See shift_arg_pages().
67 * tlb->fullmm = 0, but tlb_start_vma/tlb_end_vma will not be called.
68 * tlb->vma will be NULL.
69 */
70 static inline void tlb_flush(struct mmu_gather *tlb)
71 {
72 if (tlb->fullmm || !tlb->vma)
73 flush_tlb_mm(tlb->mm);
74 else if (tlb->range_end > 0) {
75 flush_tlb_range(tlb->vma, tlb->range_start, tlb->range_end);
76 tlb->range_start = TASK_SIZE;
77 tlb->range_end = 0;
78 }
79 }
80
81 static inline void tlb_add_flush(struct mmu_gather *tlb, unsigned long addr)
82 {
83 if (!tlb->fullmm) {
84 if (addr < tlb->range_start)
85 tlb->range_start = addr;
86 if (addr + PAGE_SIZE > tlb->range_end)
87 tlb->range_end = addr + PAGE_SIZE;
88 }
89 }
90
91 static inline void __tlb_alloc_page(struct mmu_gather *tlb)
92 {
93 unsigned long addr = __get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
94
95 if (addr) {
96 tlb->pages = (void *)addr;
97 tlb->max = PAGE_SIZE / sizeof(struct page *);
98 }
99 }
100
101 static inline void tlb_flush_mmu(struct mmu_gather *tlb)
102 {
103 tlb_flush(tlb);
104 free_pages_and_swap_cache(tlb->pages, tlb->nr);
105 tlb->nr = 0;
106 if (tlb->pages == tlb->local)
107 __tlb_alloc_page(tlb);
108 }
109
110 static inline void
111 tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end)
112 {
113 tlb->mm = mm;
114 tlb->fullmm = !(start | (end+1));
115 tlb->start = start;
116 tlb->end = end;
117 tlb->vma = NULL;
118 tlb->max = ARRAY_SIZE(tlb->local);
119 tlb->pages = tlb->local;
120 tlb->nr = 0;
121 __tlb_alloc_page(tlb);
122 }
123
124 static inline void
125 tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
126 {
127 tlb_flush_mmu(tlb);
128
129 /* keep the page table cache within bounds */
130 check_pgt_cache();
131
132 if (tlb->pages != tlb->local)
133 free_pages((unsigned long)tlb->pages, 0);
134 }
135
136 /*
137 * Memorize the range for the TLB flush.
138 */
139 static inline void
140 tlb_remove_tlb_entry(struct mmu_gather *tlb, pte_t *ptep, unsigned long addr)
141 {
142 tlb_add_flush(tlb, addr);
143 }
144
145 /*
146 * In the case of tlb vma handling, we can optimise these away in the
147 * case where we're doing a full MM flush. When we're doing a munmap,
148 * the vmas are adjusted to only cover the region to be torn down.
149 */
150 static inline void
151 tlb_start_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
152 {
153 if (!tlb->fullmm) {
154 flush_cache_range(vma, vma->vm_start, vma->vm_end);
155 tlb->vma = vma;
156 tlb->range_start = TASK_SIZE;
157 tlb->range_end = 0;
158 }
159 }
160
161 static inline void
162 tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
163 {
164 if (!tlb->fullmm)
165 tlb_flush(tlb);
166 }
167
168 static inline int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
169 {
170 tlb->pages[tlb->nr++] = page;
171 VM_BUG_ON(tlb->nr > tlb->max);
172 return tlb->max - tlb->nr;
173 }
174
175 static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
176 {
177 if (!__tlb_remove_page(tlb, page))
178 tlb_flush_mmu(tlb);
179 }
180
181 static inline void __pte_free_tlb(struct mmu_gather *tlb, pgtable_t pte,
182 unsigned long addr)
183 {
184 pgtable_page_dtor(pte);
185
186 #ifdef CONFIG_ARM_LPAE
187 tlb_add_flush(tlb, addr);
188 #else
189 /*
190 * With the classic ARM MMU, a pte page has two corresponding pmd
191 * entries, each covering 1MB.
192 */
193 addr &= PMD_MASK;
194 tlb_add_flush(tlb, addr + SZ_1M - PAGE_SIZE);
195 tlb_add_flush(tlb, addr + SZ_1M);
196 #endif
197
198 tlb_remove_page(tlb, pte);
199 }
200
201 static inline void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmdp,
202 unsigned long addr)
203 {
204 #ifdef CONFIG_ARM_LPAE
205 tlb_add_flush(tlb, addr);
206 tlb_remove_page(tlb, virt_to_page(pmdp));
207 #endif
208 }
209
210 static inline void
211 tlb_remove_pmd_tlb_entry(struct mmu_gather *tlb, pmd_t *pmdp, unsigned long addr)
212 {
213 tlb_add_flush(tlb, addr);
214 }
215
216 #define pte_free_tlb(tlb, ptep, addr) __pte_free_tlb(tlb, ptep, addr)
217 #define pmd_free_tlb(tlb, pmdp, addr) __pmd_free_tlb(tlb, pmdp, addr)
218 #define pud_free_tlb(tlb, pudp, addr) pud_free((tlb)->mm, pudp)
219
220 #define tlb_migrate_finish(mm) do { } while (0)
221
222 #endif /* CONFIG_MMU */
223 #endif
This page took 0.035367 seconds and 5 git commands to generate.