tlb.h 6.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221
  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. #include <linux/swap.h>
  18. #include <asm/pgalloc.h>
  19. #include <asm/tlbflush.h>
  20. #ifdef CONFIG_HAVE_RCU_TABLE_FREE
  21. /*
  22. * Semi RCU freeing of the page directories.
  23. *
  24. * This is needed by some architectures to implement software pagetable walkers.
  25. *
  26. * gup_fast() and other software pagetable walkers do a lockless page-table
  27. * walk and therefore needs some synchronization with the freeing of the page
  28. * directories. The chosen means to accomplish that is by disabling IRQs over
  29. * the walk.
  30. *
  31. * Architectures that use IPIs to flush TLBs will then automagically DTRT,
  32. * since we unlink the page, flush TLBs, free the page. Since the disabling of
  33. * IRQs delays the completion of the TLB flush we can never observe an already
  34. * freed page.
  35. *
  36. * Architectures that do not have this (PPC) need to delay the freeing by some
  37. * other means, this is that means.
  38. *
  39. * What we do is batch the freed directory pages (tables) and RCU free them.
  40. * We use the sched RCU variant, as that guarantees that IRQ/preempt disabling
  41. * holds off grace periods.
  42. *
  43. * However, in order to batch these pages we need to allocate storage, this
  44. * allocation is deep inside the MM code and can thus easily fail on memory
  45. * pressure. To guarantee progress we fall back to single table freeing, see
  46. * the implementation of tlb_remove_table_one().
  47. *
  48. */
  49. struct mmu_table_batch {
  50. struct rcu_head rcu;
  51. unsigned int nr;
  52. void *tables[0];
  53. };
  54. #define MAX_TABLE_BATCH \
  55. ((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
  56. extern void tlb_table_flush(struct mmu_gather *tlb);
  57. extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
  58. #endif
  59. /*
  60. * If we can't allocate a page to make a big batch of page pointers
  61. * to work on, then just handle a few from the on-stack structure.
  62. */
  63. #define MMU_GATHER_BUNDLE 8
  64. struct mmu_gather_batch {
  65. struct mmu_gather_batch *next;
  66. unsigned int nr;
  67. unsigned int max;
  68. struct page *pages[0];
  69. };
  70. #define MAX_GATHER_BATCH \
  71. ((PAGE_SIZE - sizeof(struct mmu_gather_batch)) / sizeof(void *))
  72. /*
  73. * Limit the maximum number of mmu_gather batches to reduce a risk of soft
  74. * lockups for non-preemptible kernels on huge machines when a lot of memory
  75. * is zapped during unmapping.
  76. * 10K pages freed at once should be safe even without a preemption point.
  77. */
  78. #define MAX_GATHER_BATCH_COUNT (10000UL/MAX_GATHER_BATCH)
  79. /* struct mmu_gather is an opaque type used by the mm code for passing around
  80. * any data needed by arch specific code for tlb_remove_page.
  81. */
  82. struct mmu_gather {
  83. struct mm_struct *mm;
  84. #ifdef CONFIG_HAVE_RCU_TABLE_FREE
  85. struct mmu_table_batch *batch;
  86. #endif
  87. unsigned long start;
  88. unsigned long end;
  89. /* we are in the middle of an operation to clear
  90. * a full mm and can make some optimizations */
  91. unsigned int fullmm : 1,
  92. /* we have performed an operation which
  93. * requires a complete flush of the tlb */
  94. need_flush_all : 1;
  95. struct mmu_gather_batch *active;
  96. struct mmu_gather_batch local;
  97. struct page *__pages[MMU_GATHER_BUNDLE];
  98. unsigned int batch_count;
  99. };
  100. #define HAVE_GENERIC_MMU_GATHER
  101. void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end);
  102. void tlb_flush_mmu(struct mmu_gather *tlb);
  103. void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start,
  104. unsigned long end);
  105. int __tlb_remove_page(struct mmu_gather *tlb, struct page *page);
  106. /* tlb_remove_page
  107. * Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
  108. * required.
  109. */
  110. static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
  111. {
  112. if (!__tlb_remove_page(tlb, page))
  113. tlb_flush_mmu(tlb);
  114. }
  115. static inline void __tlb_adjust_range(struct mmu_gather *tlb,
  116. unsigned long address)
  117. {
  118. tlb->start = min(tlb->start, address);
  119. tlb->end = max(tlb->end, address + PAGE_SIZE);
  120. }
  121. static inline void __tlb_reset_range(struct mmu_gather *tlb)
  122. {
  123. if (tlb->fullmm) {
  124. tlb->start = tlb->end = ~0;
  125. } else {
  126. tlb->start = TASK_SIZE;
  127. tlb->end = 0;
  128. }
  129. }
  130. /*
  131. * In the case of tlb vma handling, we can optimise these away in the
  132. * case where we're doing a full MM flush. When we're doing a munmap,
  133. * the vmas are adjusted to only cover the region to be torn down.
  134. */
  135. #ifndef tlb_start_vma
  136. #define tlb_start_vma(tlb, vma) do { } while (0)
  137. #endif
  138. #define __tlb_end_vma(tlb, vma) \
  139. do { \
  140. if (!tlb->fullmm && tlb->end) { \
  141. tlb_flush(tlb); \
  142. __tlb_reset_range(tlb); \
  143. } \
  144. } while (0)
  145. #ifndef tlb_end_vma
  146. #define tlb_end_vma __tlb_end_vma
  147. #endif
  148. #ifndef __tlb_remove_tlb_entry
  149. #define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
  150. #endif
  151. /**
  152. * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
  153. *
  154. * Record the fact that pte's were really unmapped by updating the range,
  155. * so we can later optimise away the tlb invalidate. This helps when
  156. * userspace is unmapping already-unmapped pages, which happens quite a lot.
  157. */
  158. #define tlb_remove_tlb_entry(tlb, ptep, address) \
  159. do { \
  160. __tlb_adjust_range(tlb, address); \
  161. __tlb_remove_tlb_entry(tlb, ptep, address); \
  162. } while (0)
  163. /**
  164. * tlb_remove_pmd_tlb_entry - remember a pmd mapping for later tlb invalidation
  165. * This is a nop so far, because only x86 needs it.
  166. */
  167. #ifndef __tlb_remove_pmd_tlb_entry
  168. #define __tlb_remove_pmd_tlb_entry(tlb, pmdp, address) do {} while (0)
  169. #endif
  170. #define tlb_remove_pmd_tlb_entry(tlb, pmdp, address) \
  171. do { \
  172. __tlb_adjust_range(tlb, address); \
  173. __tlb_remove_pmd_tlb_entry(tlb, pmdp, address); \
  174. } while (0)
  175. #define pte_free_tlb(tlb, ptep, address) \
  176. do { \
  177. __tlb_adjust_range(tlb, address); \
  178. __pte_free_tlb(tlb, ptep, address); \
  179. } while (0)
  180. #ifndef __ARCH_HAS_4LEVEL_HACK
  181. #define pud_free_tlb(tlb, pudp, address) \
  182. do { \
  183. __tlb_adjust_range(tlb, address); \
  184. __pud_free_tlb(tlb, pudp, address); \
  185. } while (0)
  186. #endif
  187. #define pmd_free_tlb(tlb, pmdp, address) \
  188. do { \
  189. __tlb_adjust_range(tlb, address); \
  190. __pmd_free_tlb(tlb, pmdp, address); \
  191. } while (0)
  192. #define tlb_migrate_finish(mm) do {} while (0)
  193. #endif /* _ASM_GENERIC__TLB_H */