LXR linux/mm/sparse-vmemmap.c

   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/*
  33 * Allocate a block of memory to be used to back the virtual memory map
  34 * or to back the page tables that are used to create the mapping.
  35 * Uses the main allocators if they are available, else bootmem.
  36 */
  37
  38static void * __init_refok __earlyonly_bootmem_alloc(int node,
  39                                unsigned long size,
  40                                unsigned long align,
  41                                unsigned long goal)
  42{
  43        return __alloc_bootmem_node(NODE_DATA(node), size, align, goal);
  44}
  45
  46
  47void * __meminit vmemmap_alloc_block(unsigned long size, int node)
  48{
  49        /* If the main allocator is up use that, fallback to bootmem. */
  50        if (slab_is_available()) {
  51                struct page *page;
  52
  53                if (node_state(node, N_HIGH_MEMORY))
  54                        page = alloc_pages_node(node,
  55                                GFP_KERNEL | __GFP_ZERO, get_order(size));
  56                else
  57                        page = alloc_pages(GFP_KERNEL | __GFP_ZERO,
  58                                get_order(size));
  59                if (page)
  60                        return page_address(page);
  61                return NULL;
  62        } else
  63                return __earlyonly_bootmem_alloc(node, size, size,
  64                                __pa(MAX_DMA_ADDRESS));
  65}
  66
  67void __meminit vmemmap_verify(pte_t *pte, int node,
  68                                unsigned long start, unsigned long end)
  69{
  70        unsigned long pfn = pte_pfn(*pte);
  71        int actual_node = early_pfn_to_nid(pfn);
  72
  73        if (node_distance(actual_node, node) > LOCAL_DISTANCE)
  74                printk(KERN_WARNING "[%lx-%lx] potential offnode "
  75                        "page_structs\n", start, end - 1);
  76}
  77
  78pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
  79{
  80        pte_t *pte = pte_offset_kernel(pmd, addr);
  81        if (pte_none(*pte)) {
  82                pte_t entry;
  83                void *p = vmemmap_alloc_block(PAGE_SIZE, node);
  84                if (!p)
  85                        return NULL;
  86                entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
  87                set_pte_at(&init_mm, addr, pte, entry);
  88        }
  89        return pte;
  90}
  91
  92pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
  93{
  94        pmd_t *pmd = pmd_offset(pud, addr);
  95        if (pmd_none(*pmd)) {
  96                void *p = vmemmap_alloc_block(PAGE_SIZE, node);
  97                if (!p)
  98                        return NULL;
  99                pmd_populate_kernel(&init_mm, pmd, p);
 100        }
 101        return pmd;
 102}
 103
 104pud_t * __meminit vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node)
 105{
 106        pud_t *pud = pud_offset(pgd, addr);
 107        if (pud_none(*pud)) {
 108                void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 109                if (!p)
 110                        return NULL;
 111                pud_populate(&init_mm, pud, p);
 112        }
 113        return pud;
 114}
 115
 116pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
 117{
 118        pgd_t *pgd = pgd_offset_k(addr);
 119        if (pgd_none(*pgd)) {
 120                void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 121                if (!p)
 122                        return NULL;
 123                pgd_populate(&init_mm, pgd, p);
 124        }
 125        return pgd;
 126}
 127
 128int __meminit vmemmap_populate_basepages(struct page *start_page,
 129                                                unsigned long size, int node)
 130{
 131        unsigned long addr = (unsigned long)start_page;
 132        unsigned long end = (unsigned long)(start_page + size);
 133        pgd_t *pgd;
 134        pud_t *pud;
 135        pmd_t *pmd;
 136        pte_t *pte;
 137
 138        for (; addr < end; addr += PAGE_SIZE) {
 139                pgd = vmemmap_pgd_populate(addr, node);
 140                if (!pgd)
 141                        return -ENOMEM;
 142                pud = vmemmap_pud_populate(pgd, addr, node);
 143                if (!pud)
 144                        return -ENOMEM;
 145                pmd = vmemmap_pmd_populate(pud, addr, node);
 146                if (!pmd)
 147                        return -ENOMEM;
 148                pte = vmemmap_pte_populate(pmd, addr, node);
 149                if (!pte)
 150                        return -ENOMEM;
 151                vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
 152        }
 153
 154        return 0;
 155}
 156
 157struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
 158{
 159        struct page *map = pfn_to_page(pnum * PAGES_PER_SECTION);
 160        int error = vmemmap_populate(map, PAGES_PER_SECTION, nid);
 161        if (error)
 162                return NULL;
 163
 164        return map;
 165}
 166