sparse-vmemmap.c 4.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159
  1. /*
  2. * Virtual Memory Map support
  3. *
  4. * (C) 2007 sgi. Christoph Lameter.
  5. *
  6. * Virtual memory maps allow VM primitives pfn_to_page, page_to_pfn,
  7. * virt_to_page, page_address() to be implemented as a base offset
  8. * calculation without memory access.
  9. *
  10. * However, virtual mappings need a page table and TLBs. Many Linux
  11. * architectures already map their physical space using 1-1 mappings
  12. * via TLBs. For those arches the virtual memmory map is essentially
  13. * for free if we use the same page size as the 1-1 mappings. In that
  14. * case the overhead consists of a few additional pages that are
  15. * allocated to create a view of memory for vmemmap.
  16. *
  17. * The architecture is expected to provide a vmemmap_populate() function
  18. * to instantiate the mapping.
  19. */
  20. #include <linux/mm.h>
  21. #include <linux/mmzone.h>
  22. #include <linux/bootmem.h>
  23. #include <linux/highmem.h>
  24. #include <linux/module.h>
  25. #include <linux/spinlock.h>
  26. #include <linux/vmalloc.h>
  27. #include <linux/sched.h>
  28. #include <asm/dma.h>
  29. #include <asm/pgalloc.h>
  30. #include <asm/pgtable.h>
  31. /*
  32. * Allocate a block of memory to be used to back the virtual memory map
  33. * or to back the page tables that are used to create the mapping.
  34. * Uses the main allocators if they are available, else bootmem.
  35. */
  36. static void * __init_refok __earlyonly_bootmem_alloc(int node,
  37. unsigned long size,
  38. unsigned long align,
  39. unsigned long goal)
  40. {
  41. return __alloc_bootmem_node(NODE_DATA(node), size, align, goal);
  42. }
  43. void * __meminit vmemmap_alloc_block(unsigned long size, int node)
  44. {
  45. /* If the main allocator is up use that, fallback to bootmem. */
  46. if (slab_is_available()) {
  47. struct page *page = alloc_pages_node(node,
  48. GFP_KERNEL | __GFP_ZERO, get_order(size));
  49. if (page)
  50. return page_address(page);
  51. return NULL;
  52. } else
  53. return __earlyonly_bootmem_alloc(node, size, size,
  54. __pa(MAX_DMA_ADDRESS));
  55. }
  56. void __meminit vmemmap_verify(pte_t *pte, int node,
  57. unsigned long start, unsigned long end)
  58. {
  59. unsigned long pfn = pte_pfn(*pte);
  60. int actual_node = early_pfn_to_nid(pfn);
  61. if (node_distance(actual_node, node) > LOCAL_DISTANCE)
  62. printk(KERN_WARNING "[%lx-%lx] potential offnode "
  63. "page_structs\n", start, end - 1);
  64. }
  65. pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
  66. {
  67. pte_t *pte = pte_offset_kernel(pmd, addr);
  68. if (pte_none(*pte)) {
  69. pte_t entry;
  70. void *p = vmemmap_alloc_block(PAGE_SIZE, node);
  71. if (!p)
  72. return NULL;
  73. entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
  74. set_pte_at(&init_mm, addr, pte, entry);
  75. }
  76. return pte;
  77. }
  78. pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
  79. {
  80. pmd_t *pmd = pmd_offset(pud, addr);
  81. if (pmd_none(*pmd)) {
  82. void *p = vmemmap_alloc_block(PAGE_SIZE, node);
  83. if (!p)
  84. return NULL;
  85. pmd_populate_kernel(&init_mm, pmd, p);
  86. }
  87. return pmd;
  88. }
  89. pud_t * __meminit vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node)
  90. {
  91. pud_t *pud = pud_offset(pgd, addr);
  92. if (pud_none(*pud)) {
  93. void *p = vmemmap_alloc_block(PAGE_SIZE, node);
  94. if (!p)
  95. return NULL;
  96. pud_populate(&init_mm, pud, p);
  97. }
  98. return pud;
  99. }
  100. pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
  101. {
  102. pgd_t *pgd = pgd_offset_k(addr);
  103. if (pgd_none(*pgd)) {
  104. void *p = vmemmap_alloc_block(PAGE_SIZE, node);
  105. if (!p)
  106. return NULL;
  107. pgd_populate(&init_mm, pgd, p);
  108. }
  109. return pgd;
  110. }
  111. int __meminit vmemmap_populate_basepages(struct page *start_page,
  112. unsigned long size, int node)
  113. {
  114. unsigned long addr = (unsigned long)start_page;
  115. unsigned long end = (unsigned long)(start_page + size);
  116. pgd_t *pgd;
  117. pud_t *pud;
  118. pmd_t *pmd;
  119. pte_t *pte;
  120. for (; addr < end; addr += PAGE_SIZE) {
  121. pgd = vmemmap_pgd_populate(addr, node);
  122. if (!pgd)
  123. return -ENOMEM;
  124. pud = vmemmap_pud_populate(pgd, addr, node);
  125. if (!pud)
  126. return -ENOMEM;
  127. pmd = vmemmap_pmd_populate(pud, addr, node);
  128. if (!pmd)
  129. return -ENOMEM;
  130. pte = vmemmap_pte_populate(pmd, addr, node);
  131. if (!pte)
  132. return -ENOMEM;
  133. vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
  134. }
  135. return 0;
  136. }
  137. struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
  138. {
  139. struct page *map = pfn_to_page(pnum * PAGES_PER_SECTION);
  140. int error = vmemmap_populate(map, PAGES_PER_SECTION, nid);
  141. if (error)
  142. return NULL;
  143. return map;
  144. }