2 * Virtual Memory Map support
4 * (C) 2007 sgi. Christoph Lameter.
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.
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.
17 * The architecture is expected to provide a vmemmap_populate() function
18 * to instantiate the mapping.
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>
30 #include <asm/pgalloc.h>
31 #include <asm/pgtable.h>
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.
39 static void * __init_refok __earlyonly_bootmem_alloc(int node,
44 return memblock_virt_alloc_try_nid(size, align, goal,
45 BOOTMEM_ALLOC_ACCESSIBLE, node);
48 static void *vmemmap_buf;
49 static void *vmemmap_buf_end;
51 void * __meminit vmemmap_alloc_block(unsigned long size, int node)
53 /* If the main allocator is up use that, fallback to bootmem. */
54 if (slab_is_available()) {
57 if (node_state(node, N_HIGH_MEMORY))
58 page = alloc_pages_node(
59 node, GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
63 GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
66 return page_address(page);
69 return __earlyonly_bootmem_alloc(node, size, size,
70 __pa(MAX_DMA_ADDRESS));
73 /* need to make sure size is all the same during early stage */
74 static void * __meminit alloc_block_buf(unsigned long size, int node)
79 return vmemmap_alloc_block(size, node);
81 /* take the from buf */
82 ptr = (void *)ALIGN((unsigned long)vmemmap_buf, size);
83 if (ptr + size > vmemmap_buf_end)
84 return vmemmap_alloc_block(size, node);
86 vmemmap_buf = ptr + size;
91 static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
93 return altmap->base_pfn + altmap->reserve + altmap->alloc
97 static unsigned long __meminit vmem_altmap_nr_free(struct vmem_altmap *altmap)
99 unsigned long allocated = altmap->alloc + altmap->align;
101 if (altmap->free > allocated)
102 return altmap->free - allocated;
107 * vmem_altmap_alloc - allocate pages from the vmem_altmap reservation
108 * @altmap - reserved page pool for the allocation
109 * @nr_pfns - size (in pages) of the allocation
111 * Allocations are aligned to the size of the request
113 static unsigned long __meminit vmem_altmap_alloc(struct vmem_altmap *altmap,
114 unsigned long nr_pfns)
116 unsigned long pfn = vmem_altmap_next_pfn(altmap);
117 unsigned long nr_align;
119 nr_align = 1UL << find_first_bit(&nr_pfns, BITS_PER_LONG);
120 nr_align = ALIGN(pfn, nr_align) - pfn;
122 if (nr_pfns + nr_align > vmem_altmap_nr_free(altmap))
124 altmap->alloc += nr_pfns;
125 altmap->align += nr_align;
126 return pfn + nr_align;
129 static void * __meminit altmap_alloc_block_buf(unsigned long size,
130 struct vmem_altmap *altmap)
132 unsigned long pfn, nr_pfns;
135 if (size & ~PAGE_MASK) {
136 pr_warn_once("%s: allocations must be multiple of PAGE_SIZE (%ld)\n",
141 nr_pfns = size >> PAGE_SHIFT;
142 pfn = vmem_altmap_alloc(altmap, nr_pfns);
144 ptr = __va(__pfn_to_phys(pfn));
147 pr_debug("%s: pfn: %#lx alloc: %ld align: %ld nr: %#lx\n",
148 __func__, pfn, altmap->alloc, altmap->align, nr_pfns);
153 /* need to make sure size is all the same during early stage */
154 void * __meminit __vmemmap_alloc_block_buf(unsigned long size, int node,
155 struct vmem_altmap *altmap)
158 return altmap_alloc_block_buf(size, altmap);
159 return alloc_block_buf(size, node);
162 void __meminit vmemmap_verify(pte_t *pte, int node,
163 unsigned long start, unsigned long end)
165 unsigned long pfn = pte_pfn(*pte);
166 int actual_node = early_pfn_to_nid(pfn);
168 if (node_distance(actual_node, node) > LOCAL_DISTANCE)
169 pr_warn("[%lx-%lx] potential offnode page_structs\n",
173 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
175 pte_t *pte = pte_offset_kernel(pmd, addr);
176 if (pte_none(*pte)) {
178 void *p = alloc_block_buf(PAGE_SIZE, node);
181 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
182 set_pte_at(&init_mm, addr, pte, entry);
187 pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
189 pmd_t *pmd = pmd_offset(pud, addr);
190 if (pmd_none(*pmd)) {
191 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
194 pmd_populate_kernel(&init_mm, pmd, p);
199 pud_t * __meminit vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node)
201 pud_t *pud = pud_offset(pgd, addr);
202 if (pud_none(*pud)) {
203 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
206 pud_populate(&init_mm, pud, p);
211 pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
213 pgd_t *pgd = pgd_offset_k(addr);
214 if (pgd_none(*pgd)) {
215 void *p = vmemmap_alloc_block(PAGE_SIZE, node);
218 pgd_populate(&init_mm, pgd, p);
223 int __meminit vmemmap_populate_basepages(unsigned long start,
224 unsigned long end, int node)
226 unsigned long addr = start;
232 for (; addr < end; addr += PAGE_SIZE) {
233 pgd = vmemmap_pgd_populate(addr, node);
236 pud = vmemmap_pud_populate(pgd, addr, node);
239 pmd = vmemmap_pmd_populate(pud, addr, node);
242 pte = vmemmap_pte_populate(pmd, addr, node);
245 vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
251 struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
257 map = pfn_to_page(pnum * PAGES_PER_SECTION);
258 start = (unsigned long)map;
259 end = (unsigned long)(map + PAGES_PER_SECTION);
261 if (vmemmap_populate(start, end, nid))
267 void __init sparse_mem_maps_populate_node(struct page **map_map,
268 unsigned long pnum_begin,
269 unsigned long pnum_end,
270 unsigned long map_count, int nodeid)
273 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
274 void *vmemmap_buf_start;
276 size = ALIGN(size, PMD_SIZE);
277 vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
278 PMD_SIZE, __pa(MAX_DMA_ADDRESS));
280 if (vmemmap_buf_start) {
281 vmemmap_buf = vmemmap_buf_start;
282 vmemmap_buf_end = vmemmap_buf_start + size * map_count;
285 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
286 struct mem_section *ms;
288 if (!present_section_nr(pnum))
291 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
294 ms = __nr_to_section(pnum);
295 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
297 ms->section_mem_map = 0;
300 if (vmemmap_buf_start) {
301 /* need to free left buf */
302 memblock_free_early(__pa(vmemmap_buf),
303 vmemmap_buf_end - vmemmap_buf);
305 vmemmap_buf_end = NULL;