LXR linux/mm/sparse-vmemmap.c

   1/*
   2 * Virtual Memory Map support
   3 *
   4 * (C) 2007 sgi. Christoph Lameter <clameter@sgi.com>.
   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 = alloc_pages_node(node,
  52                                GFP_KERNEL | __GFP_ZERO, get_order(size));
  53                if (page)
  54                        return page_address(page);
  55                return NULL;
  56        } else
  57                return __earlyonly_bootmem_alloc(node, size, size,
  58                                __pa(MAX_DMA_ADDRESS));
  59}
  60
  61void __meminit vmemmap_verify(pte_t *pte, int node,
  62                                unsigned long start, unsigned long end)
  63{
  64        unsigned long pfn = pte_pfn(*pte);
  65        int actual_node = early_pfn_to_nid(pfn);
  66
  67        if (actual_node != node)
  68                printk(KERN_WARNING "[%lx-%lx] potential offnode "
  69                        "page_structs\n", start, end - 1);
  70}
  71
  72pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
  73{
  74        pte_t *pte = pte_offset_kernel(pmd, addr);
  75        if (pte_none(*pte)) {
  76                pte_t entry;
  77                void *p = vmemmap_alloc_block(PAGE_SIZE, node);
  78                if (!p)
  79                        return NULL;
  80                entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
  81                set_pte_at(&init_mm, addr, pte, entry);
  82        }
  83        return pte;
  84}
  85
  86pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
  87{
  88        pmd_t *pmd = pmd_offset(pud, addr);
  89        if (pmd_none(*pmd)) {
  90                void *p = vmemmap_alloc_block(PAGE_SIZE, node);
  91                if (!p)
  92                        return NULL;
  93                pmd_populate_kernel(&init_mm, pmd, p);
  94        }
  95        return pmd;
  96}
  97
  98pud_t * __meminit vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node)
  99{
 100        pud_t *pud = pud_offset(pgd, addr);
 101        if (pud_none(*pud)) {
 102                void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 103                if (!p)
 104                        return NULL;
 105                pud_populate(&init_mm, pud, p);
 106        }
 107        return pud;
 108}
 109
 110pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
 111{
 112        pgd_t *pgd = pgd_offset_k(addr);
 113        if (pgd_none(*pgd)) {
 114                void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 115                if (!p)
 116                        return NULL;
 117                pgd_populate(&init_mm, pgd, p);
 118        }
 119        return pgd;
 120}
 121
 122int __meminit vmemmap_populate_basepages(struct page *start_page,
 123                                                unsigned long size, int node)
 124{
 125        unsigned long addr = (unsigned long)start_page;
 126        unsigned long end = (unsigned long)(start_page + size);
 127        pgd_t *pgd;
 128        pud_t *pud;
 129        pmd_t *pmd;
 130        pte_t *pte;
 131
 132        for (; addr < end; addr += PAGE_SIZE) {
 133                pgd = vmemmap_pgd_populate(addr, node);
 134                if (!pgd)
 135                        return -ENOMEM;
 136                pud = vmemmap_pud_populate(pgd, addr, node);
 137                if (!pud)
 138                        return -ENOMEM;
 139                pmd = vmemmap_pmd_populate(pud, addr, node);
 140                if (!pmd)
 141                        return -ENOMEM;
 142                pte = vmemmap_pte_populate(pmd, addr, node);
 143                if (!pte)
 144                        return -ENOMEM;
 145                vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
 146        }
 147
 148        return 0;
 149}
 150
 151struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
 152{
 153        struct page *map = pfn_to_page(pnum * PAGES_PER_SECTION);
 154        int error = vmemmap_populate(map, PAGES_PER_SECTION, nid);
 155        if (error)
 156                return NULL;
 157
 158        return map;
 159}
 160