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 range_start; 47 unsigned long range_end; 48 unsigned int nr; 49 unsigned int max; 50 struct page **pages; 51 struct page *local[MMU_GATHER_BUNDLE]; 52 }; 53 54 DECLARE_PER_CPU(struct mmu_gather, mmu_gathers); 55 56 /* 57 * This is unnecessarily complex. There's three ways the TLB shootdown 58 * code is used: 59 * 1. Unmapping a range of vmas. See zap_page_range(), unmap_region(). 60 * tlb->fullmm = 0, and tlb_start_vma/tlb_end_vma will be called. 61 * tlb->vma will be non-NULL. 62 * 2. Unmapping all vmas. See exit_mmap(). 63 * tlb->fullmm = 1, and tlb_start_vma/tlb_end_vma will be called. 64 * tlb->vma will be non-NULL. Additionally, page tables will be freed. 65 * 3. Unmapping argument pages. See shift_arg_pages(). 66 * tlb->fullmm = 0, but tlb_start_vma/tlb_end_vma will not be called. 67 * tlb->vma will be NULL. 68 */ 69 static inline void tlb_flush(struct mmu_gather *tlb) 70 { 71 if (tlb->fullmm || !tlb->vma) 72 flush_tlb_mm(tlb->mm); 73 else if (tlb->range_end > 0) { 74 flush_tlb_range(tlb->vma, tlb->range_start, tlb->range_end); 75 tlb->range_start = TASK_SIZE; 76 tlb->range_end = 0; 77 } 78 } 79 80 static inline void tlb_add_flush(struct mmu_gather *tlb, unsigned long addr) 81 { 82 if (!tlb->fullmm) { 83 if (addr < tlb->range_start) 84 tlb->range_start = addr; 85 if (addr + PAGE_SIZE > tlb->range_end) 86 tlb->range_end = addr + PAGE_SIZE; 87 } 88 } 89 90 static inline void __tlb_alloc_page(struct mmu_gather *tlb) 91 { 92 unsigned long addr = __get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); 93 94 if (addr) { 95 tlb->pages = (void *)addr; 96 tlb->max = PAGE_SIZE / sizeof(struct page *); 97 } 98 } 99 100 static inline void tlb_flush_mmu(struct mmu_gather *tlb) 101 { 102 tlb_flush(tlb); 103 free_pages_and_swap_cache(tlb->pages, tlb->nr); 104 tlb->nr = 0; 105 if (tlb->pages == tlb->local) 106 __tlb_alloc_page(tlb); 107 } 108 109 static inline void 110 tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned int fullmm) 111 { 112 tlb->mm = mm; 113 tlb->fullmm = fullmm; 114 tlb->vma = NULL; 115 tlb->max = ARRAY_SIZE(tlb->local); 116 tlb->pages = tlb->local; 117 tlb->nr = 0; 118 __tlb_alloc_page(tlb); 119 } 120 121 static inline void 122 tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) 123 { 124 tlb_flush_mmu(tlb); 125 126 /* keep the page table cache within bounds */ 127 check_pgt_cache(); 128 129 if (tlb->pages != tlb->local) 130 free_pages((unsigned long)tlb->pages, 0); 131 } 132 133 /* 134 * Memorize the range for the TLB flush. 135 */ 136 static inline void 137 tlb_remove_tlb_entry(struct mmu_gather *tlb, pte_t *ptep, unsigned long addr) 138 { 139 tlb_add_flush(tlb, addr); 140 } 141 142 /* 143 * In the case of tlb vma handling, we can optimise these away in the 144 * case where we're doing a full MM flush. When we're doing a munmap, 145 * the vmas are adjusted to only cover the region to be torn down. 146 */ 147 static inline void 148 tlb_start_vma(struct mmu_gather *tlb, struct vm_area_struct *vma) 149 { 150 if (!tlb->fullmm) { 151 flush_cache_range(vma, vma->vm_start, vma->vm_end); 152 tlb->vma = vma; 153 tlb->range_start = TASK_SIZE; 154 tlb->range_end = 0; 155 } 156 } 157 158 static inline void 159 tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma) 160 { 161 if (!tlb->fullmm) 162 tlb_flush(tlb); 163 } 164 165 static inline int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) 166 { 167 tlb->pages[tlb->nr++] = page; 168 VM_BUG_ON(tlb->nr > tlb->max); 169 return tlb->max - tlb->nr; 170 } 171 172 static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page) 173 { 174 if (!__tlb_remove_page(tlb, page)) 175 tlb_flush_mmu(tlb); 176 } 177 178 static inline void __pte_free_tlb(struct mmu_gather *tlb, pgtable_t pte, 179 unsigned long addr) 180 { 181 pgtable_page_dtor(pte); 182 183 #ifdef CONFIG_ARM_LPAE 184 tlb_add_flush(tlb, addr); 185 #else 186 /* 187 * With the classic ARM MMU, a pte page has two corresponding pmd 188 * entries, each covering 1MB. 189 */ 190 addr &= PMD_MASK; 191 tlb_add_flush(tlb, addr + SZ_1M - PAGE_SIZE); 192 tlb_add_flush(tlb, addr + SZ_1M); 193 #endif 194 195 tlb_remove_page(tlb, pte); 196 } 197 198 static inline void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmdp, 199 unsigned long addr) 200 { 201 #ifdef CONFIG_ARM_LPAE 202 tlb_add_flush(tlb, addr); 203 tlb_remove_page(tlb, virt_to_page(pmdp)); 204 #endif 205 } 206 207 #define pte_free_tlb(tlb, ptep, addr) __pte_free_tlb(tlb, ptep, addr) 208 #define pmd_free_tlb(tlb, pmdp, addr) __pmd_free_tlb(tlb, pmdp, addr) 209 #define pud_free_tlb(tlb, pudp, addr) pud_free((tlb)->mm, pudp) 210 211 #define tlb_migrate_finish(mm) do { } while (0) 212 213 #endif /* CONFIG_MMU */ 214 #endif 215