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 <pzijlstr@redhat.com>
   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         unsigned int            need_flush : 1, /* Did free PTEs */
 100                                 fast_mode  : 1; /* No batching   */
 101
 102         /* we are in the middle of an operation to clear
 103          * a full mm and can make some optimizations */
 104         unsigned int            fullmm : 1,
 105         /* we have performed an operation which
 106          * requires a complete flush of the tlb */
 107                                 need_flush_all : 1;
 108
 109         struct mmu_gather_batch *active;
 110         struct mmu_gather_batch local;
 111         struct page             *__pages[MMU_GATHER_BUNDLE];
 112         unsigned int            batch_count;
 113 };
 114
 115 #define HAVE_GENERIC_MMU_GATHER
 116
 117 static inline int tlb_fast_mode(struct mmu_gather *tlb)
 118 {
 119 #ifdef CONFIG_SMP
 120         return tlb->fast_mode;
 121 #else
 122         /*
 123          * For UP we don't need to worry about TLB flush
 124          * and page free order so much..
 125          */
 126         return 1;
 127 #endif
 128 }
 129
 130 void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, bool fullmm);
 131 void tlb_flush_mmu(struct mmu_gather *tlb);
 132 void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start,
 133                                                         unsigned long end);
 134 int __tlb_remove_page(struct mmu_gather *tlb, struct page *page);
 135
 136 /* tlb_remove_page
 137  *      Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
 138  *      required.
 139  */
 140 static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
 141 {
 142         if (!__tlb_remove_page(tlb, page))
 143                 tlb_flush_mmu(tlb);
 144 }
 145
 146 /**
 147  * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
 148  *
 149  * Record the fact that pte's were really umapped in ->need_flush, so we can
 150  * later optimise away the tlb invalidate.   This helps when userspace is
 151  * unmapping already-unmapped pages, which happens quite a lot.
 152  */
 153 #define tlb_remove_tlb_entry(tlb, ptep, address)                \
 154         do {                                                    \
 155                 tlb->need_flush = 1;                            \
 156                 __tlb_remove_tlb_entry(tlb, ptep, address);     \
 157         } while (0)
 158
 159 /**
 160  * tlb_remove_pmd_tlb_entry - remember a pmd mapping for later tlb invalidation
 161  * This is a nop so far, because only x86 needs it.
 162  */
 163 #ifndef __tlb_remove_pmd_tlb_entry
 164 #define __tlb_remove_pmd_tlb_entry(tlb, pmdp, address) do {} while (0)
 165 #endif
 166
 167 #define tlb_remove_pmd_tlb_entry(tlb, pmdp, address)            \
 168         do {                                                    \
 169                 tlb->need_flush = 1;                            \
 170                 __tlb_remove_pmd_tlb_entry(tlb, pmdp, address); \
 171         } while (0)
 172
 173 #define pte_free_tlb(tlb, ptep, address)                        \
 174         do {                                                    \
 175                 tlb->need_flush = 1;                            \
 176                 __pte_free_tlb(tlb, ptep, address);             \
 177         } while (0)
 178
 179 #ifndef __ARCH_HAS_4LEVEL_HACK
 180 #define pud_free_tlb(tlb, pudp, address)                        \
 181         do {                                                    \
 182                 tlb->need_flush = 1;                            \
 183                 __pud_free_tlb(tlb, pudp, address);             \
 184         } while (0)
 185 #endif
 186
 187 #define pmd_free_tlb(tlb, pmdp, address)                        \
 188         do {                                                    \
 189                 tlb->need_flush = 1;                            \
 190                 __pmd_free_tlb(tlb, pmdp, address);             \
 191         } while (0)
 192
 193 #define tlb_migrate_finish(mm) do {} while (0)
 194
 195 #endif /* _ASM_GENERIC__TLB_H */