include/asm-generic/tlb.h

   1 /* include/asm-generic/tlb.h
   2  *
   3  *      Generic TLB shootdown code
   4  *
   5  * Copyright 2001 Red Hat, Inc.
   6  * Based on code from mm/memory.c Copyright Linus Torvalds and others.
   7  *
   8  * Copyright 2011 Red Hat, Inc., Peter Zijlstra
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public License
  12  * as published by the Free Software Foundation; either version
  13  * 2 of the License, or (at your option) any later version.
  14  */
  15 #ifndef _ASM_GENERIC__TLB_H
  16 #define _ASM_GENERIC__TLB_H
  17
  18 #include <linux/swap.h>
  19 #include <asm/pgalloc.h>
  20 #include <asm/tlbflush.h>
  21
  22 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
  23 /*
  24  * Semi RCU freeing of the page directories.
  25  *
  26  * This is needed by some architectures to implement software pagetable walkers.
  27  *
  28  * gup_fast() and other software pagetable walkers do a lockless page-table
  29  * walk and therefore needs some synchronization with the freeing of the page
  30  * directories. The chosen means to accomplish that is by disabling IRQs over
  31  * the walk.
  32  *
  33  * Architectures that use IPIs to flush TLBs will then automagically DTRT,
  34  * since we unlink the page, flush TLBs, free the page. Since the disabling of
  35  * IRQs delays the completion of the TLB flush we can never observe an already
  36  * freed page.
  37  *
  38  * Architectures that do not have this (PPC) need to delay the freeing by some
  39  * other means, this is that means.
  40  *
  41  * What we do is batch the freed directory pages (tables) and RCU free them.
  42  * We use the sched RCU variant, as that guarantees that IRQ/preempt disabling
  43  * holds off grace periods.
  44  *
  45  * However, in order to batch these pages we need to allocate storage, this
  46  * allocation is deep inside the MM code and can thus easily fail on memory
  47  * pressure. To guarantee progress we fall back to single table freeing, see
  48  * the implementation of tlb_remove_table_one().
  49  *
  50  */
  51 struct mmu_table_batch {
  52         struct rcu_head         rcu;
  53         unsigned int            nr;
  54         void                    *tables[0];
  55 };
  56
  57 #define MAX_TABLE_BATCH         \
  58         ((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
  59
  60 extern void tlb_table_flush(struct mmu_gather *tlb);
  61 extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
  62
  63 #endif
  64
  65 /*
  66  * If we can't allocate a page to make a big batch of page pointers
  67  * to work on, then just handle a few from the on-stack structure.
  68  */
  69 #define MMU_GATHER_BUNDLE       8
  70
  71 struct mmu_gather_batch {
  72         struct mmu_gather_batch *next;
  73         unsigned int            nr;
  74         unsigned int            max;
  75         struct page             *pages[0];
  76 };
  77
  78 #define MAX_GATHER_BATCH        \
  79         ((PAGE_SIZE - sizeof(struct mmu_gather_batch)) / sizeof(void *))
  80
  81 /*
  82  * Limit the maximum number of mmu_gather batches to reduce a risk of soft
  83  * lockups for non-preemptible kernels on huge machines when a lot of memory
  84  * is zapped during unmapping.
  85  * 10K pages freed at once should be safe even without a preemption point.
  86  */
  87 #define MAX_GATHER_BATCH_COUNT  (10000UL/MAX_GATHER_BATCH)
  88
  89 /* struct mmu_gather is an opaque type used by the mm code for passing around
  90  * any data needed by arch specific code for tlb_remove_page.
  91  */
  92 struct mmu_gather {
  93         struct mm_struct        *mm;
  94 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
  95         struct mmu_table_batch  *batch;
  96 #endif
  97         unsigned long           start;
  98         unsigned long           end;
  99         /* we are in the middle of an operation to clear
 100          * a full mm and can make some optimizations */
 101         unsigned int            fullmm : 1,
 102         /* we have performed an operation which
 103          * requires a complete flush of the tlb */
 104                                 need_flush_all : 1;
 105
 106         struct mmu_gather_batch *active;
 107         struct mmu_gather_batch local;
 108         struct page             *__pages[MMU_GATHER_BUNDLE];
 109         unsigned int            batch_count;
 110         /*
 111          * __tlb_adjust_range  will track the new addr here,
 112          * that that we can adjust the range after the flush
 113          */
 114         unsigned long addr;
 115         int page_size;
 116 };
 117
 118 #define HAVE_GENERIC_MMU_GATHER
 119
 120 void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end);
 121 void tlb_flush_mmu(struct mmu_gather *tlb);
 122 void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start,
 123                                                         unsigned long end);
 124 extern bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page,
 125                                    int page_size);
 126
 127 static inline void __tlb_adjust_range(struct mmu_gather *tlb,
 128                                       unsigned long address)
 129 {
 130         tlb->start = min(tlb->start, address);
 131         tlb->end = max(tlb->end, address + PAGE_SIZE);
 132         /*
 133          * Track the last address with which we adjusted the range. This
 134          * will be used later to adjust again after a mmu_flush due to
 135          * failed __tlb_remove_page
 136          */
 137         tlb->addr = address;
 138 }
 139
 140 static inline void __tlb_reset_range(struct mmu_gather *tlb)
 141 {
 142         if (tlb->fullmm) {
 143                 tlb->start = tlb->end = ~0;
 144         } else {
 145                 tlb->start = TASK_SIZE;
 146                 tlb->end = 0;
 147         }
 148 }
 149
 150 static inline void tlb_remove_page_size(struct mmu_gather *tlb,
 151                                         struct page *page, int page_size)
 152 {
 153         if (__tlb_remove_page_size(tlb, page, page_size)) {
 154                 tlb_flush_mmu(tlb);
 155                 tlb->page_size = page_size;
 156                 __tlb_adjust_range(tlb, tlb->addr);
 157                 __tlb_remove_page_size(tlb, page, page_size);
 158         }
 159 }
 160
 161 static bool __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
 162 {
 163         return __tlb_remove_page_size(tlb, page, PAGE_SIZE);
 164 }
 165
 166 /* tlb_remove_page
 167  *      Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
 168  *      required.
 169  */
 170 static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
 171 {
 172         return tlb_remove_page_size(tlb, page, PAGE_SIZE);
 173 }
 174
 175 static inline bool __tlb_remove_pte_page(struct mmu_gather *tlb, struct page *page)
 176 {
 177         /* active->nr should be zero when we call this */
 178         VM_BUG_ON_PAGE(tlb->active->nr, page);
 179         tlb->page_size = PAGE_SIZE;
 180         __tlb_adjust_range(tlb, tlb->addr);
 181         return __tlb_remove_page(tlb, page);
 182 }
 183
 184 /*
 185  * In the case of tlb vma handling, we can optimise these away in the
 186  * case where we're doing a full MM flush.  When we're doing a munmap,
 187  * the vmas are adjusted to only cover the region to be torn down.
 188  */
 189 #ifndef tlb_start_vma
 190 #define tlb_start_vma(tlb, vma) do { } while (0)
 191 #endif
 192
 193 #define __tlb_end_vma(tlb, vma)                                 \
 194         do {                                                    \
 195                 if (!tlb->fullmm && tlb->end) {                 \
 196                         tlb_flush(tlb);                         \
 197                         __tlb_reset_range(tlb);                 \
 198                 }                                               \
 199         } while (0)
 200
 201 #ifndef tlb_end_vma
 202 #define tlb_end_vma     __tlb_end_vma
 203 #endif
 204
 205 #ifndef __tlb_remove_tlb_entry
 206 #define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
 207 #endif
 208
 209 /**
 210  * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
 211  *
 212  * Record the fact that pte's were really unmapped by updating the range,
 213  * so we can later optimise away the tlb invalidate.   This helps when
 214  * userspace is unmapping already-unmapped pages, which happens quite a lot.
 215  */
 216 #define tlb_remove_tlb_entry(tlb, ptep, address)                \
 217         do {                                                    \
 218                 __tlb_adjust_range(tlb, address);               \
 219                 __tlb_remove_tlb_entry(tlb, ptep, address);     \
 220         } while (0)
 221
 222 /**
 223  * tlb_remove_pmd_tlb_entry - remember a pmd mapping for later tlb invalidation
 224  * This is a nop so far, because only x86 needs it.
 225  */
 226 #ifndef __tlb_remove_pmd_tlb_entry
 227 #define __tlb_remove_pmd_tlb_entry(tlb, pmdp, address) do {} while (0)
 228 #endif
 229
 230 #define tlb_remove_pmd_tlb_entry(tlb, pmdp, address)            \
 231         do {                                                    \
 232                 __tlb_adjust_range(tlb, address);               \
 233                 __tlb_remove_pmd_tlb_entry(tlb, pmdp, address); \
 234         } while (0)
 235
 236 #define pte_free_tlb(tlb, ptep, address)                        \
 237         do {                                                    \
 238                 __tlb_adjust_range(tlb, address);               \
 239                 __pte_free_tlb(tlb, ptep, address);             \
 240         } while (0)
 241
 242 #ifndef __ARCH_HAS_4LEVEL_HACK
 243 #define pud_free_tlb(tlb, pudp, address)                        \
 244         do {                                                    \
 245                 __tlb_adjust_range(tlb, address);               \
 246                 __pud_free_tlb(tlb, pudp, address);             \
 247         } while (0)
 248 #endif
 249
 250 #define pmd_free_tlb(tlb, pmdp, address)                        \
 251         do {                                                    \
 252                 __tlb_adjust_range(tlb, address);               \
 253                 __pmd_free_tlb(tlb, pmdp, address);             \
 254         } while (0)
 255
 256 #define tlb_migrate_finish(mm) do {} while (0)
 257
 258 #endif /* _ASM_GENERIC__TLB_H */