Christoph Lameter | 8f6aac4 | 2007-10-16 01:24:13 -0700 | [diff] [blame^] | 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 | * Special Kconfig settings: |
| 18 | * |
| 19 | * CONFIG_ARCH_POPULATES_SPARSEMEM_VMEMMAP |
| 20 | * |
| 21 | * The architecture has its own functions to populate the memory |
| 22 | * map and provides a vmemmap_populate function. |
| 23 | * |
| 24 | * CONFIG_ARCH_POPULATES_SPARSEMEM_VMEMMAP_PMD |
| 25 | * |
| 26 | * The architecture provides functions to populate the pmd level |
| 27 | * of the vmemmap mappings. Allowing mappings using large pages |
| 28 | * where available. |
| 29 | * |
| 30 | * If neither are set then PAGE_SIZE mappings are generated which |
| 31 | * require one PTE/TLB per PAGE_SIZE chunk of the virtual memory map. |
| 32 | */ |
| 33 | #include <linux/mm.h> |
| 34 | #include <linux/mmzone.h> |
| 35 | #include <linux/bootmem.h> |
| 36 | #include <linux/highmem.h> |
| 37 | #include <linux/module.h> |
| 38 | #include <linux/spinlock.h> |
| 39 | #include <linux/vmalloc.h> |
| 40 | #include <asm/dma.h> |
| 41 | #include <asm/pgalloc.h> |
| 42 | #include <asm/pgtable.h> |
| 43 | |
| 44 | /* |
| 45 | * Allocate a block of memory to be used to back the virtual memory map |
| 46 | * or to back the page tables that are used to create the mapping. |
| 47 | * Uses the main allocators if they are available, else bootmem. |
| 48 | */ |
| 49 | void * __meminit vmemmap_alloc_block(unsigned long size, int node) |
| 50 | { |
| 51 | /* If the main allocator is up use that, fallback to bootmem. */ |
| 52 | if (slab_is_available()) { |
| 53 | struct page *page = alloc_pages_node(node, |
| 54 | GFP_KERNEL | __GFP_ZERO, get_order(size)); |
| 55 | if (page) |
| 56 | return page_address(page); |
| 57 | return NULL; |
| 58 | } else |
| 59 | return __alloc_bootmem_node(NODE_DATA(node), size, size, |
| 60 | __pa(MAX_DMA_ADDRESS)); |
| 61 | } |
| 62 | |
| 63 | #ifndef CONFIG_ARCH_POPULATES_SPARSEMEM_VMEMMAP |
| 64 | void __meminit vmemmap_verify(pte_t *pte, int node, |
| 65 | unsigned long start, unsigned long end) |
| 66 | { |
| 67 | unsigned long pfn = pte_pfn(*pte); |
| 68 | int actual_node = early_pfn_to_nid(pfn); |
| 69 | |
| 70 | if (actual_node != node) |
| 71 | printk(KERN_WARNING "[%lx-%lx] potential offnode " |
| 72 | "page_structs\n", start, end - 1); |
| 73 | } |
| 74 | |
| 75 | #ifndef CONFIG_ARCH_POPULATES_SPARSEMEM_VMEMMAP_PMD |
| 76 | static int __meminit vmemmap_populate_pte(pmd_t *pmd, unsigned long addr, |
| 77 | unsigned long end, int node) |
| 78 | { |
| 79 | pte_t *pte; |
| 80 | |
| 81 | for (pte = pte_offset_kernel(pmd, addr); addr < end; |
| 82 | pte++, addr += PAGE_SIZE) |
| 83 | if (pte_none(*pte)) { |
| 84 | pte_t entry; |
| 85 | void *p = vmemmap_alloc_block(PAGE_SIZE, node); |
| 86 | if (!p) |
| 87 | return -ENOMEM; |
| 88 | |
| 89 | entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL); |
| 90 | set_pte(pte, entry); |
| 91 | |
| 92 | } else |
| 93 | vmemmap_verify(pte, node, addr + PAGE_SIZE, end); |
| 94 | |
| 95 | return 0; |
| 96 | } |
| 97 | |
| 98 | int __meminit vmemmap_populate_pmd(pud_t *pud, unsigned long addr, |
| 99 | unsigned long end, int node) |
| 100 | { |
| 101 | pmd_t *pmd; |
| 102 | int error = 0; |
| 103 | unsigned long next; |
| 104 | |
| 105 | for (pmd = pmd_offset(pud, addr); addr < end && !error; |
| 106 | pmd++, addr = next) { |
| 107 | if (pmd_none(*pmd)) { |
| 108 | void *p = vmemmap_alloc_block(PAGE_SIZE, node); |
| 109 | if (!p) |
| 110 | return -ENOMEM; |
| 111 | |
| 112 | pmd_populate_kernel(&init_mm, pmd, p); |
| 113 | } else |
| 114 | vmemmap_verify((pte_t *)pmd, node, |
| 115 | pmd_addr_end(addr, end), end); |
| 116 | next = pmd_addr_end(addr, end); |
| 117 | error = vmemmap_populate_pte(pmd, addr, next, node); |
| 118 | } |
| 119 | return error; |
| 120 | } |
| 121 | #endif /* CONFIG_ARCH_POPULATES_SPARSEMEM_VMEMMAP_PMD */ |
| 122 | |
| 123 | static int __meminit vmemmap_populate_pud(pgd_t *pgd, unsigned long addr, |
| 124 | unsigned long end, int node) |
| 125 | { |
| 126 | pud_t *pud; |
| 127 | int error = 0; |
| 128 | unsigned long next; |
| 129 | |
| 130 | for (pud = pud_offset(pgd, addr); addr < end && !error; |
| 131 | pud++, addr = next) { |
| 132 | if (pud_none(*pud)) { |
| 133 | void *p = vmemmap_alloc_block(PAGE_SIZE, node); |
| 134 | if (!p) |
| 135 | return -ENOMEM; |
| 136 | |
| 137 | pud_populate(&init_mm, pud, p); |
| 138 | } |
| 139 | next = pud_addr_end(addr, end); |
| 140 | error = vmemmap_populate_pmd(pud, addr, next, node); |
| 141 | } |
| 142 | return error; |
| 143 | } |
| 144 | |
| 145 | int __meminit vmemmap_populate(struct page *start_page, |
| 146 | unsigned long nr, int node) |
| 147 | { |
| 148 | pgd_t *pgd; |
| 149 | unsigned long addr = (unsigned long)start_page; |
| 150 | unsigned long end = (unsigned long)(start_page + nr); |
| 151 | unsigned long next; |
| 152 | int error = 0; |
| 153 | |
| 154 | printk(KERN_DEBUG "[%lx-%lx] Virtual memory section" |
| 155 | " (%ld pages) node %d\n", addr, end - 1, nr, node); |
| 156 | |
| 157 | for (pgd = pgd_offset_k(addr); addr < end && !error; |
| 158 | pgd++, addr = next) { |
| 159 | if (pgd_none(*pgd)) { |
| 160 | void *p = vmemmap_alloc_block(PAGE_SIZE, node); |
| 161 | if (!p) |
| 162 | return -ENOMEM; |
| 163 | |
| 164 | pgd_populate(&init_mm, pgd, p); |
| 165 | } |
| 166 | next = pgd_addr_end(addr,end); |
| 167 | error = vmemmap_populate_pud(pgd, addr, next, node); |
| 168 | } |
| 169 | return error; |
| 170 | } |
| 171 | #endif /* !CONFIG_ARCH_POPULATES_SPARSEMEM_VMEMMAP */ |
| 172 | |
| 173 | struct page __init *sparse_early_mem_map_populate(unsigned long pnum, int nid) |
| 174 | { |
| 175 | struct page *map = pfn_to_page(pnum * PAGES_PER_SECTION); |
| 176 | int error = vmemmap_populate(map, PAGES_PER_SECTION, nid); |
| 177 | if (error) |
| 178 | return NULL; |
| 179 | |
| 180 | return map; |
| 181 | } |