mm/sparse-vmemmap.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Virtual Memory Map support
   4  *
   5  * (C) 2007 sgi. Christoph Lameter.
   6  *
   7  * Virtual memory maps allow VM primitives pfn_to_page, page_to_pfn,
   8  * virt_to_page, page_address() to be implemented as a base offset
   9  * calculation without memory access.
  10  *
  11  * However, virtual mappings need a page table and TLBs. Many Linux
  12  * architectures already map their physical space using 1-1 mappings
  13  * via TLBs. For those arches the virtual memory map is essentially
  14  * for free if we use the same page size as the 1-1 mappings. In that
  15  * case the overhead consists of a few additional pages that are
  16  * allocated to create a view of memory for vmemmap.
  17  *
  18  * The architecture is expected to provide a vmemmap_populate() function
  19  * to instantiate the mapping.
  20  */
  21 #include <linux/mm.h>
  22 #include <linux/mmzone.h>
  23 #include <linux/bootmem.h>
  24 #include <linux/memremap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/slab.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/sched.h>
  30 #include <asm/dma.h>
  31 #include <asm/pgalloc.h>
  32 #include <asm/pgtable.h>
  33
  34 /*
  35  * Allocate a block of memory to be used to back the virtual memory map
  36  * or to back the page tables that are used to create the mapping.
  37  * Uses the main allocators if they are available, else bootmem.
  38  */
  39
  40 static void * __ref __earlyonly_bootmem_alloc(int node,
  41                                 unsigned long size,
  42                                 unsigned long align,
  43                                 unsigned long goal)
  44 {
  45         return memblock_virt_alloc_try_nid(size, align, goal,
  46                                             BOOTMEM_ALLOC_ACCESSIBLE, node);
  47 }
  48
  49 static void *vmemmap_buf;
  50 static void *vmemmap_buf_end;
  51
  52 void * __meminit vmemmap_alloc_block(unsigned long size, int node)
  53 {
  54         /* If the main allocator is up use that, fallback to bootmem. */
  55         if (slab_is_available()) {
  56                 struct page *page;
  57
  58                 page = alloc_pages_node(node,
  59                         GFP_KERNEL | __GFP_ZERO | __GFP_RETRY_MAYFAIL,
  60                         get_order(size));
  61                 if (page)
  62                         return page_address(page);
  63                 return NULL;
  64         } else
  65                 return __earlyonly_bootmem_alloc(node, size, size,
  66                                 __pa(MAX_DMA_ADDRESS));
  67 }
  68
  69 /* need to make sure size is all the same during early stage */
  70 static void * __meminit alloc_block_buf(unsigned long size, int node)
  71 {
  72         void *ptr;
  73
  74         if (!vmemmap_buf)
  75                 return vmemmap_alloc_block(size, node);
  76
  77         /* take the from buf */
  78         ptr = (void *)ALIGN((unsigned long)vmemmap_buf, size);
  79         if (ptr + size > vmemmap_buf_end)
  80                 return vmemmap_alloc_block(size, node);
  81
  82         vmemmap_buf = ptr + size;
  83
  84         return ptr;
  85 }
  86
  87 static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
  88 {
  89         return altmap->base_pfn + altmap->reserve + altmap->alloc
  90                 + altmap->align;
  91 }
  92
  93 static unsigned long __meminit vmem_altmap_nr_free(struct vmem_altmap *altmap)
  94 {
  95         unsigned long allocated = altmap->alloc + altmap->align;
  96
  97         if (altmap->free > allocated)
  98                 return altmap->free - allocated;
  99         return 0;
 100 }
 101
 102 /**
 103  * vmem_altmap_alloc - allocate pages from the vmem_altmap reservation
 104  * @altmap - reserved page pool for the allocation
 105  * @nr_pfns - size (in pages) of the allocation
 106  *
 107  * Allocations are aligned to the size of the request
 108  */
 109 static unsigned long __meminit vmem_altmap_alloc(struct vmem_altmap *altmap,
 110                 unsigned long nr_pfns)
 111 {
 112         unsigned long pfn = vmem_altmap_next_pfn(altmap);
 113         unsigned long nr_align;
 114
 115         nr_align = 1UL << find_first_bit(&nr_pfns, BITS_PER_LONG);
 116         nr_align = ALIGN(pfn, nr_align) - pfn;
 117
 118         if (nr_pfns + nr_align > vmem_altmap_nr_free(altmap))
 119                 return ULONG_MAX;
 120         altmap->alloc += nr_pfns;
 121         altmap->align += nr_align;
 122         return pfn + nr_align;
 123 }
 124
 125 static void * __meminit altmap_alloc_block_buf(unsigned long size,
 126                 struct vmem_altmap *altmap)
 127 {
 128         unsigned long pfn, nr_pfns;
 129         void *ptr;
 130
 131         if (size & ~PAGE_MASK) {
 132                 pr_warn_once("%s: allocations must be multiple of PAGE_SIZE (%ld)\n",
 133                                 __func__, size);
 134                 return NULL;
 135         }
 136
 137         nr_pfns = size >> PAGE_SHIFT;
 138         pfn = vmem_altmap_alloc(altmap, nr_pfns);
 139         if (pfn < ULONG_MAX)
 140                 ptr = __va(__pfn_to_phys(pfn));
 141         else
 142                 ptr = NULL;
 143         pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
 144                         __func__, pfn, altmap->alloc, altmap->align, nr_pfns);
 145
 146         return ptr;
 147 }
 148
 149 /* need to make sure size is all the same during early stage */
 150 void * __meminit __vmemmap_alloc_block_buf(unsigned long size, int node,
 151                 struct vmem_altmap *altmap)
 152 {
 153         if (altmap)
 154                 return altmap_alloc_block_buf(size, altmap);
 155         return alloc_block_buf(size, node);
 156 }
 157
 158 void __meminit vmemmap_verify(pte_t *pte, int node,
 159                                 unsigned long start, unsigned long end)
 160 {
 161         unsigned long pfn = pte_pfn(*pte);
 162         int actual_node = early_pfn_to_nid(pfn);
 163
 164         if (node_distance(actual_node, node) > LOCAL_DISTANCE)
 165                 pr_warn("[%lx-%lx] potential offnode page_structs\n",
 166                         start, end - 1);
 167 }
 168
 169 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
 170 {
 171         pte_t *pte = pte_offset_kernel(pmd, addr);
 172         if (pte_none(*pte)) {
 173                 pte_t entry;
 174                 void *p = alloc_block_buf(PAGE_SIZE, node);
 175                 if (!p)
 176                         return NULL;
 177                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
 178                 set_pte_at(&init_mm, addr, pte, entry);
 179         }
 180         return pte;
 181 }
 182
 183 pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
 184 {
 185         pmd_t *pmd = pmd_offset(pud, addr);
 186         if (pmd_none(*pmd)) {
 187                 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 188                 if (!p)
 189                         return NULL;
 190                 pmd_populate_kernel(&init_mm, pmd, p);
 191         }
 192         return pmd;
 193 }
 194
 195 pud_t * __meminit vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node)
 196 {
 197         pud_t *pud = pud_offset(p4d, addr);
 198         if (pud_none(*pud)) {
 199                 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 200                 if (!p)
 201                         return NULL;
 202                 pud_populate(&init_mm, pud, p);
 203         }
 204         return pud;
 205 }
 206
 207 p4d_t * __meminit vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node)
 208 {
 209         p4d_t *p4d = p4d_offset(pgd, addr);
 210         if (p4d_none(*p4d)) {
 211                 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 212                 if (!p)
 213                         return NULL;
 214                 p4d_populate(&init_mm, p4d, p);
 215         }
 216         return p4d;
 217 }
 218
 219 pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
 220 {
 221         pgd_t *pgd = pgd_offset_k(addr);
 222         if (pgd_none(*pgd)) {
 223                 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
 224                 if (!p)
 225                         return NULL;
 226                 pgd_populate(&init_mm, pgd, p);
 227         }
 228         return pgd;
 229 }
 230
 231 int __meminit vmemmap_populate_basepages(unsigned long start,
 232                                          unsigned long end, int node)
 233 {
 234         unsigned long addr = start;
 235         pgd_t *pgd;
 236         p4d_t *p4d;
 237         pud_t *pud;
 238         pmd_t *pmd;
 239         pte_t *pte;
 240
 241         for (; addr < end; addr += PAGE_SIZE) {
 242                 pgd = vmemmap_pgd_populate(addr, node);
 243                 if (!pgd)
 244                         return -ENOMEM;
 245                 p4d = vmemmap_p4d_populate(pgd, addr, node);
 246                 if (!p4d)
 247                         return -ENOMEM;
 248                 pud = vmemmap_pud_populate(p4d, addr, node);
 249                 if (!pud)
 250                         return -ENOMEM;
 251                 pmd = vmemmap_pmd_populate(pud, addr, node);
 252                 if (!pmd)
 253                         return -ENOMEM;
 254                 pte = vmemmap_pte_populate(pmd, addr, node);
 255                 if (!pte)
 256                         return -ENOMEM;
 257                 vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
 258         }
 259
 260         return 0;
 261 }
 262
 263 struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
 264 {
 265         unsigned long start;
 266         unsigned long end;
 267         struct page *map;
 268
 269         map = pfn_to_page(pnum * PAGES_PER_SECTION);
 270         start = (unsigned long)map;
 271         end = (unsigned long)(map + PAGES_PER_SECTION);
 272
 273         if (vmemmap_populate(start, end, nid))
 274                 return NULL;
 275
 276         return map;
 277 }
 278
 279 void __init sparse_mem_maps_populate_node(struct page **map_map,
 280                                           unsigned long pnum_begin,
 281                                           unsigned long pnum_end,
 282                                           unsigned long map_count, int nodeid)
 283 {
 284         unsigned long pnum;
 285         unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
 286         void *vmemmap_buf_start;
 287
 288         size = ALIGN(size, PMD_SIZE);
 289         vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
 290                          PMD_SIZE, __pa(MAX_DMA_ADDRESS));
 291
 292         if (vmemmap_buf_start) {
 293                 vmemmap_buf = vmemmap_buf_start;
 294                 vmemmap_buf_end = vmemmap_buf_start + size * map_count;
 295         }
 296
 297         for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
 298                 struct mem_section *ms;
 299
 300                 if (!present_section_nr(pnum))
 301                         continue;
 302
 303                 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
 304                 if (map_map[pnum])
 305                         continue;
 306                 ms = __nr_to_section(pnum);
 307                 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
 308                        __func__);
 309                 ms->section_mem_map = 0;
 310         }
 311
 312         if (vmemmap_buf_start) {
 313                 /* need to free left buf */
 314                 memblock_free_early(__pa(vmemmap_buf),
 315                                     vmemmap_buf_end - vmemmap_buf);
 316                 vmemmap_buf = NULL;
 317                 vmemmap_buf_end = NULL;
 318         }
 319 }