xref: /linux/mm/sparse-vmemmap.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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/highmem.h>
24 #include <linux/slab.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 
38 static void * __init_refok __earlyonly_bootmem_alloc(int node,
39 				unsigned long size,
40 				unsigned long align,
41 				unsigned long goal)
42 {
43 	return memblock_virt_alloc_try_nid(size, align, goal,
44 					    BOOTMEM_ALLOC_ACCESSIBLE, node);
45 }
46 
47 static void *vmemmap_buf;
48 static void *vmemmap_buf_end;
49 
50 void * __meminit vmemmap_alloc_block(unsigned long size, int node)
51 {
52 	/* If the main allocator is up use that, fallback to bootmem. */
53 	if (slab_is_available()) {
54 		struct page *page;
55 
56 		if (node_state(node, N_HIGH_MEMORY))
57 			page = alloc_pages_node(
58 				node, GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
59 				get_order(size));
60 		else
61 			page = alloc_pages(
62 				GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
63 				get_order(size));
64 		if (page)
65 			return page_address(page);
66 		return NULL;
67 	} else
68 		return __earlyonly_bootmem_alloc(node, size, size,
69 				__pa(MAX_DMA_ADDRESS));
70 }
71 
72 /* need to make sure size is all the same during early stage */
73 void * __meminit vmemmap_alloc_block_buf(unsigned long size, int node)
74 {
75 	void *ptr;
76 
77 	if (!vmemmap_buf)
78 		return vmemmap_alloc_block(size, node);
79 
80 	/* take the from buf */
81 	ptr = (void *)ALIGN((unsigned long)vmemmap_buf, size);
82 	if (ptr + size > vmemmap_buf_end)
83 		return vmemmap_alloc_block(size, node);
84 
85 	vmemmap_buf = ptr + size;
86 
87 	return ptr;
88 }
89 
90 void __meminit vmemmap_verify(pte_t *pte, int node,
91 				unsigned long start, unsigned long end)
92 {
93 	unsigned long pfn = pte_pfn(*pte);
94 	int actual_node = early_pfn_to_nid(pfn);
95 
96 	if (node_distance(actual_node, node) > LOCAL_DISTANCE)
97 		printk(KERN_WARNING "[%lx-%lx] potential offnode "
98 			"page_structs\n", start, end - 1);
99 }
100 
101 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
102 {
103 	pte_t *pte = pte_offset_kernel(pmd, addr);
104 	if (pte_none(*pte)) {
105 		pte_t entry;
106 		void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
107 		if (!p)
108 			return NULL;
109 		entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
110 		set_pte_at(&init_mm, addr, pte, entry);
111 	}
112 	return pte;
113 }
114 
115 pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
116 {
117 	pmd_t *pmd = pmd_offset(pud, addr);
118 	if (pmd_none(*pmd)) {
119 		void *p = vmemmap_alloc_block(PAGE_SIZE, node);
120 		if (!p)
121 			return NULL;
122 		pmd_populate_kernel(&init_mm, pmd, p);
123 	}
124 	return pmd;
125 }
126 
127 pud_t * __meminit vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node)
128 {
129 	pud_t *pud = pud_offset(pgd, addr);
130 	if (pud_none(*pud)) {
131 		void *p = vmemmap_alloc_block(PAGE_SIZE, node);
132 		if (!p)
133 			return NULL;
134 		pud_populate(&init_mm, pud, p);
135 	}
136 	return pud;
137 }
138 
139 pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
140 {
141 	pgd_t *pgd = pgd_offset_k(addr);
142 	if (pgd_none(*pgd)) {
143 		void *p = vmemmap_alloc_block(PAGE_SIZE, node);
144 		if (!p)
145 			return NULL;
146 		pgd_populate(&init_mm, pgd, p);
147 	}
148 	return pgd;
149 }
150 
151 int __meminit vmemmap_populate_basepages(unsigned long start,
152 					 unsigned long end, int node)
153 {
154 	unsigned long addr = start;
155 	pgd_t *pgd;
156 	pud_t *pud;
157 	pmd_t *pmd;
158 	pte_t *pte;
159 
160 	for (; addr < end; addr += PAGE_SIZE) {
161 		pgd = vmemmap_pgd_populate(addr, node);
162 		if (!pgd)
163 			return -ENOMEM;
164 		pud = vmemmap_pud_populate(pgd, addr, node);
165 		if (!pud)
166 			return -ENOMEM;
167 		pmd = vmemmap_pmd_populate(pud, addr, node);
168 		if (!pmd)
169 			return -ENOMEM;
170 		pte = vmemmap_pte_populate(pmd, addr, node);
171 		if (!pte)
172 			return -ENOMEM;
173 		vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
174 	}
175 
176 	return 0;
177 }
178 
179 struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
180 {
181 	unsigned long start;
182 	unsigned long end;
183 	struct page *map;
184 
185 	map = pfn_to_page(pnum * PAGES_PER_SECTION);
186 	start = (unsigned long)map;
187 	end = (unsigned long)(map + PAGES_PER_SECTION);
188 
189 	if (vmemmap_populate(start, end, nid))
190 		return NULL;
191 
192 	return map;
193 }
194 
195 void __init sparse_mem_maps_populate_node(struct page **map_map,
196 					  unsigned long pnum_begin,
197 					  unsigned long pnum_end,
198 					  unsigned long map_count, int nodeid)
199 {
200 	unsigned long pnum;
201 	unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
202 	void *vmemmap_buf_start;
203 
204 	size = ALIGN(size, PMD_SIZE);
205 	vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
206 			 PMD_SIZE, __pa(MAX_DMA_ADDRESS));
207 
208 	if (vmemmap_buf_start) {
209 		vmemmap_buf = vmemmap_buf_start;
210 		vmemmap_buf_end = vmemmap_buf_start + size * map_count;
211 	}
212 
213 	for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
214 		struct mem_section *ms;
215 
216 		if (!present_section_nr(pnum))
217 			continue;
218 
219 		map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
220 		if (map_map[pnum])
221 			continue;
222 		ms = __nr_to_section(pnum);
223 		printk(KERN_ERR "%s: sparsemem memory map backing failed "
224 			"some memory will not be available.\n", __func__);
225 		ms->section_mem_map = 0;
226 	}
227 
228 	if (vmemmap_buf_start) {
229 		/* need to free left buf */
230 		memblock_free_early(__pa(vmemmap_buf),
231 				    vmemmap_buf_end - vmemmap_buf);
232 		vmemmap_buf = NULL;
233 		vmemmap_buf_end = NULL;
234 	}
235 }
236