xref: /openbmc/linux/mm/sparse-vmemmap.c (revision 063f4661)
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 __alloc_bootmem_node_high(NODE_DATA(node), size, align, goal);
44 }
45 
46 static void *vmemmap_buf;
47 static void *vmemmap_buf_end;
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;
54 
55 		if (node_state(node, N_HIGH_MEMORY))
56 			page = alloc_pages_node(
57 				node, GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
58 				get_order(size));
59 		else
60 			page = alloc_pages(
61 				GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
62 				get_order(size));
63 		if (page)
64 			return page_address(page);
65 		return NULL;
66 	} else
67 		return __earlyonly_bootmem_alloc(node, size, size,
68 				__pa(MAX_DMA_ADDRESS));
69 }
70 
71 /* need to make sure size is all the same during early stage */
72 void * __meminit vmemmap_alloc_block_buf(unsigned long size, int node)
73 {
74 	void *ptr;
75 
76 	if (!vmemmap_buf)
77 		return vmemmap_alloc_block(size, node);
78 
79 	/* take the from buf */
80 	ptr = (void *)ALIGN((unsigned long)vmemmap_buf, size);
81 	if (ptr + size > vmemmap_buf_end)
82 		return vmemmap_alloc_block(size, node);
83 
84 	vmemmap_buf = ptr + size;
85 
86 	return ptr;
87 }
88 
89 void __meminit vmemmap_verify(pte_t *pte, int node,
90 				unsigned long start, unsigned long end)
91 {
92 	unsigned long pfn = pte_pfn(*pte);
93 	int actual_node = early_pfn_to_nid(pfn);
94 
95 	if (node_distance(actual_node, node) > LOCAL_DISTANCE)
96 		printk(KERN_WARNING "[%lx-%lx] potential offnode "
97 			"page_structs\n", start, end - 1);
98 }
99 
100 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
101 {
102 	pte_t *pte = pte_offset_kernel(pmd, addr);
103 	if (pte_none(*pte)) {
104 		pte_t entry;
105 		void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
106 		if (!p)
107 			return NULL;
108 		entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
109 		set_pte_at(&init_mm, addr, pte, entry);
110 	}
111 	return pte;
112 }
113 
114 pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
115 {
116 	pmd_t *pmd = pmd_offset(pud, addr);
117 	if (pmd_none(*pmd)) {
118 		void *p = vmemmap_alloc_block(PAGE_SIZE, node);
119 		if (!p)
120 			return NULL;
121 		pmd_populate_kernel(&init_mm, pmd, p);
122 	}
123 	return pmd;
124 }
125 
126 pud_t * __meminit vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node)
127 {
128 	pud_t *pud = pud_offset(pgd, addr);
129 	if (pud_none(*pud)) {
130 		void *p = vmemmap_alloc_block(PAGE_SIZE, node);
131 		if (!p)
132 			return NULL;
133 		pud_populate(&init_mm, pud, p);
134 	}
135 	return pud;
136 }
137 
138 pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
139 {
140 	pgd_t *pgd = pgd_offset_k(addr);
141 	if (pgd_none(*pgd)) {
142 		void *p = vmemmap_alloc_block(PAGE_SIZE, node);
143 		if (!p)
144 			return NULL;
145 		pgd_populate(&init_mm, pgd, p);
146 	}
147 	return pgd;
148 }
149 
150 int __meminit vmemmap_populate_basepages(unsigned long start,
151 					 unsigned long end, int node)
152 {
153 	unsigned long addr = start;
154 	pgd_t *pgd;
155 	pud_t *pud;
156 	pmd_t *pmd;
157 	pte_t *pte;
158 
159 	for (; addr < end; addr += PAGE_SIZE) {
160 		pgd = vmemmap_pgd_populate(addr, node);
161 		if (!pgd)
162 			return -ENOMEM;
163 		pud = vmemmap_pud_populate(pgd, addr, node);
164 		if (!pud)
165 			return -ENOMEM;
166 		pmd = vmemmap_pmd_populate(pud, addr, node);
167 		if (!pmd)
168 			return -ENOMEM;
169 		pte = vmemmap_pte_populate(pmd, addr, node);
170 		if (!pte)
171 			return -ENOMEM;
172 		vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
173 	}
174 
175 	return 0;
176 }
177 
178 struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
179 {
180 	unsigned long start;
181 	unsigned long end;
182 	struct page *map;
183 
184 	map = pfn_to_page(pnum * PAGES_PER_SECTION);
185 	start = (unsigned long)map;
186 	end = (unsigned long)(map + PAGES_PER_SECTION);
187 
188 	if (vmemmap_populate(start, end, nid))
189 		return NULL;
190 
191 	return map;
192 }
193 
194 void __init sparse_mem_maps_populate_node(struct page **map_map,
195 					  unsigned long pnum_begin,
196 					  unsigned long pnum_end,
197 					  unsigned long map_count, int nodeid)
198 {
199 	unsigned long pnum;
200 	unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
201 	void *vmemmap_buf_start;
202 
203 	size = ALIGN(size, PMD_SIZE);
204 	vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
205 			 PMD_SIZE, __pa(MAX_DMA_ADDRESS));
206 
207 	if (vmemmap_buf_start) {
208 		vmemmap_buf = vmemmap_buf_start;
209 		vmemmap_buf_end = vmemmap_buf_start + size * map_count;
210 	}
211 
212 	for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
213 		struct mem_section *ms;
214 
215 		if (!present_section_nr(pnum))
216 			continue;
217 
218 		map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
219 		if (map_map[pnum])
220 			continue;
221 		ms = __nr_to_section(pnum);
222 		printk(KERN_ERR "%s: sparsemem memory map backing failed "
223 			"some memory will not be available.\n", __func__);
224 		ms->section_mem_map = 0;
225 	}
226 
227 	if (vmemmap_buf_start) {
228 		/* need to free left buf */
229 		free_bootmem(__pa(vmemmap_buf), vmemmap_buf_end - vmemmap_buf);
230 		vmemmap_buf = NULL;
231 		vmemmap_buf_end = NULL;
232 	}
233 }
234