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 memory 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/memremap.h>
  24 #include <linux/highmem.h>
  25 #include <linux/slab.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/sched.h>
  29 #include <asm/dma.h>
  30 #include <asm/pgalloc.h>
  31 #include <asm/pgtable.h>
  32
  33 /*
  34  * Allocate a block of memory to be used to back the virtual memory map
  35  * or to back the page tables that are used to create the mapping.
  36  * Uses the main allocators if they are available, else bootmem.
  37  */
  38
  39 static void * __ref __earlyonly_bootmem_alloc(int node,
  40                                 unsigned long size,
  41                                 unsigned long align,
  42                                 unsigned long goal)
  43 {
  44         return memblock_virt_alloc_try_nid(size, align, goal,
  45                                             BOOTMEM_ALLOC_ACCESSIBLE, node);
  46 }
  47
  48 static void *vmemmap_buf;
  49 static void *vmemmap_buf_end;
  50
  51 void * __meminit vmemmap_alloc_block(unsigned long size, int node)
  52 {
  53         /* If the main allocator is up use that, fallback to bootmem. */
  54         if (slab_is_available()) {
  55                 struct page *page;
  56
  57                 page = alloc_pages_node(node,
  58                         GFP_KERNEL | __GFP_ZERO | __GFP_RETRY_MAYFAIL,
  59                         get_order(size));
  60                 if (page)
  61                         return page_address(page);
  62                 return NULL;
  63         } else
  64                 return __earlyonly_bootmem_alloc(node, size, size,
  65                                 __pa(MAX_DMA_ADDRESS));
  66 }
  67
  68 /* need to make sure size is all the same during early stage */
  69 static void * __meminit alloc_block_buf(unsigned long size, int node)
  70 {
  71         void *ptr;
  72
  73         if (!vmemmap_buf)
  74                 return vmemmap_alloc_block(size, node);
  75
  76         /* take the from buf */
  77         ptr = (void *)ALIGN((unsigned long)vmemmap_buf, size);
  78         if (ptr + size > vmemmap_buf_end)
  79                 return vmemmap_alloc_block(size, node);
  80
  81         vmemmap_buf = ptr + size;
  82
  83         return ptr;
  84 }
  85
  86 static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
  87 {
  88         return altmap->base_pfn + altmap->reserve + altmap->alloc
  89                 + altmap->align;
  90 }
  91
  92 static unsigned long __meminit vmem_altmap_nr_free(struct vmem_altmap *altmap)
  93 {
  94         unsigned long allocated = altmap->alloc + altmap->align;
  95
  96         if (altmap->free > allocated)
  97                 return altmap->free - allocated;
  98         return 0;
  99 }
 100
 101 /**
 102  * vmem_altmap_alloc - allocate pages from the vmem_altmap reservation
 103  * @altmap - reserved page pool for the allocation
 104  * @nr_pfns - size (in pages) of the allocation
 105  *
 106  * Allocations are aligned to the size of the request
 107  */
 108 static unsigned long __meminit vmem_altmap_alloc(struct vmem_altmap *altmap,
 109                 unsigned long nr_pfns)
 110 {
 111         unsigned long pfn = vmem_altmap_next_pfn(altmap);
 112         unsigned long nr_align;
 113
 114         nr_align = 1UL << find_first_bit(&nr_pfns, BITS_PER_LONG);
 115         nr_align = ALIGN(pfn, nr_align) - pfn;
 116
 117         if (nr_pfns + nr_align > vmem_altmap_nr_free(altmap))
 118                 return ULONG_MAX;
 119         altmap->alloc += nr_pfns;
 120         altmap->align += nr_align;
 121         return pfn + nr_align;
 122 }
 123
 124 static void * __meminit altmap_alloc_block_buf(unsigned long size,
 125                 struct vmem_altmap *altmap)
 126 {
 127         unsigned long pfn, nr_pfns;
 128         void *ptr;
 129
 130         if (size & ~PAGE_MASK) {
 131                 pr_warn_once("%s: allocations must be multiple of PAGE_SIZE (%ld)\n",
 132                                 __func__, size);
 133                 return NULL;
 134         }
 135
 136         nr_pfns = size >> PAGE_SHIFT;
 137         pfn = vmem_altmap_alloc(altmap, nr_pfns);
 138         if (pfn < ULONG_MAX)
 139                 ptr = __va(__pfn_to_phys(pfn));
 140         else
 141                 ptr = NULL;
 142         pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
 143                         __func__, pfn, altmap->alloc, altmap->align, nr_pfns);
 144
 145         return ptr;
 146 }
 147
 148 /* need to make sure size is all the same during early stage */
 149 void * __meminit __vmemmap_alloc_block_buf(unsigned long size, int node,
 150                 struct vmem_altmap *altmap)
 151 {
 152         if (altmap)
 153                 return altmap_alloc_block_buf(size, altmap);
 154         return alloc_block_buf(size, node);
 155 }
 156
 157 void __meminit vmemmap_verify(pte_t *pte, int node,
 158                                 unsigned long start, unsigned long end)
 159 {
 160         unsigned long pfn = pte_pfn(*pte);
 161         int actual_node = early_pfn_to_nid(pfn);
 162
 163         if (node_distance(actual_node, node) > LOCAL_DISTANCE)
 164                 pr_warn("[%lx-%lx] potential offnode page_structs\n",
 165                         start, end - 1);
 166 }
 167
 168 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
 169 {
 170         pte_t *pte = pte_offset_kernel(pmd, addr);
 171         if (pte_none(*pte)) {
 172                 pte_t entry;
 173                 void *p = alloc_block_buf(PAGE_SIZE, node);
 174                 if (!p)
 175                         return NULL;
 176                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
 177                 set_pte_at(&init_mm, addr, pte, entry);
 178         }
 179         return pte;
 180 }
 181
 182 pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
 183 {
 184         pmd_t *pmd = pmd_offset(pud, addr);
 185         if (pmd_none(*pmd)) {
 186                 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 187                 if (!p)
 188                         return NULL;
 189                 pmd_populate_kernel(&init_mm, pmd, p);
 190         }
 191         return pmd;
 192 }
 193
 194 pud_t * __meminit vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node)
 195 {
 196         pud_t *pud = pud_offset(p4d, addr);
 197         if (pud_none(*pud)) {
 198                 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 199                 if (!p)
 200                         return NULL;
 201                 pud_populate(&init_mm, pud, p);
 202         }
 203         return pud;
 204 }
 205
 206 p4d_t * __meminit vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node)
 207 {
 208         p4d_t *p4d = p4d_offset(pgd, addr);
 209         if (p4d_none(*p4d)) {
 210                 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 211                 if (!p)
 212                         return NULL;
 213                 p4d_populate(&init_mm, p4d, p);
 214         }
 215         return p4d;
 216 }
 217
 218 pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
 219 {
 220         pgd_t *pgd = pgd_offset_k(addr);
 221         if (pgd_none(*pgd)) {
 222                 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 223                 if (!p)
 224                         return NULL;
 225                 pgd_populate(&init_mm, pgd, p);
 226         }
 227         return pgd;
 228 }
 229
 230 int __meminit vmemmap_populate_basepages(unsigned long start,
 231                                          unsigned long end, int node)
 232 {
 233         unsigned long addr = start;
 234         pgd_t *pgd;
 235         p4d_t *p4d;
 236         pud_t *pud;
 237         pmd_t *pmd;
 238         pte_t *pte;
 239
 240         for (; addr < end; addr += PAGE_SIZE) {
 241                 pgd = vmemmap_pgd_populate(addr, node);
 242                 if (!pgd)
 243                         return -ENOMEM;
 244                 p4d = vmemmap_p4d_populate(pgd, addr, node);
 245                 if (!p4d)
 246                         return -ENOMEM;
 247                 pud = vmemmap_pud_populate(p4d, addr, node);
 248                 if (!pud)
 249                         return -ENOMEM;
 250                 pmd = vmemmap_pmd_populate(pud, addr, node);
 251                 if (!pmd)
 252                         return -ENOMEM;
 253                 pte = vmemmap_pte_populate(pmd, addr, node);
 254                 if (!pte)
 255                         return -ENOMEM;
 256                 vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
 257         }
 258
 259         return 0;
 260 }
 261
 262 struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
 263 {
 264         unsigned long start;
 265         unsigned long end;
 266         struct page *map;
 267
 268         map = pfn_to_page(pnum * PAGES_PER_SECTION);
 269         start = (unsigned long)map;
 270         end = (unsigned long)(map + PAGES_PER_SECTION);
 271
 272         if (vmemmap_populate(start, end, nid))
 273                 return NULL;
 274
 275         return map;
 276 }
 277
 278 void __init sparse_mem_maps_populate_node(struct page **map_map,
 279                                           unsigned long pnum_begin,
 280                                           unsigned long pnum_end,
 281                                           unsigned long map_count, int nodeid)
 282 {
 283         unsigned long pnum;
 284         unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
 285         void *vmemmap_buf_start;
 286
 287         size = ALIGN(size, PMD_SIZE);
 288         vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
 289                          PMD_SIZE, __pa(MAX_DMA_ADDRESS));
 290
 291         if (vmemmap_buf_start) {
 292                 vmemmap_buf = vmemmap_buf_start;
 293                 vmemmap_buf_end = vmemmap_buf_start + size * map_count;
 294         }
 295
 296         for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
 297                 struct mem_section *ms;
 298
 299                 if (!present_section_nr(pnum))
 300                         continue;
 301
 302                 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
 303                 if (map_map[pnum])
 304                         continue;
 305                 ms = __nr_to_section(pnum);
 306                 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
 307                        __func__);
 308                 ms->section_mem_map = 0;
 309         }
 310
 311         if (vmemmap_buf_start) {
 312                 /* need to free left buf */
 313                 memblock_free_early(__pa(vmemmap_buf),
 314                                     vmemmap_buf_end - vmemmap_buf);
 315                 vmemmap_buf = NULL;
 316                 vmemmap_buf_end = NULL;
 317         }
 318 }