lightnvm: missing free on init error
[linux-2.6-block.git] / mm / sparse-vmemmap.c
CommitLineData
8f6aac41
CL
1/*
2 * Virtual Memory Map support
3 *
cde53535 4 * (C) 2007 sgi. Christoph Lameter.
8f6aac41
CL
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
b595076a 12 * via TLBs. For those arches the virtual memory map is essentially
8f6aac41
CL
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 *
29c71111
AW
17 * The architecture is expected to provide a vmemmap_populate() function
18 * to instantiate the mapping.
8f6aac41
CL
19 */
20#include <linux/mm.h>
21#include <linux/mmzone.h>
22#include <linux/bootmem.h>
23#include <linux/highmem.h>
5a0e3ad6 24#include <linux/slab.h>
8f6aac41
CL
25#include <linux/spinlock.h>
26#include <linux/vmalloc.h>
8bca44bb 27#include <linux/sched.h>
8f6aac41
CL
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 */
e0dc3a53
KH
37
38static void * __init_refok __earlyonly_bootmem_alloc(int node,
39 unsigned long size,
40 unsigned long align,
41 unsigned long goal)
42{
bb016b84
SS
43 return memblock_virt_alloc_try_nid(size, align, goal,
44 BOOTMEM_ALLOC_ACCESSIBLE, node);
e0dc3a53
KH
45}
46
9bdac914
YL
47static void *vmemmap_buf;
48static void *vmemmap_buf_end;
e0dc3a53 49
8f6aac41
CL
50void * __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()) {
f52407ce
SL
54 struct page *page;
55
56 if (node_state(node, N_HIGH_MEMORY))
055e4fd9
BH
57 page = alloc_pages_node(
58 node, GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
59 get_order(size));
f52407ce 60 else
055e4fd9
BH
61 page = alloc_pages(
62 GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT,
f52407ce 63 get_order(size));
8f6aac41
CL
64 if (page)
65 return page_address(page);
66 return NULL;
67 } else
e0dc3a53 68 return __earlyonly_bootmem_alloc(node, size, size,
8f6aac41
CL
69 __pa(MAX_DMA_ADDRESS));
70}
71
9bdac914
YL
72/* need to make sure size is all the same during early stage */
73void * __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
8f6aac41
CL
90void __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
b41ad14c 96 if (node_distance(actual_node, node) > LOCAL_DISTANCE)
8f6aac41
CL
97 printk(KERN_WARNING "[%lx-%lx] potential offnode "
98 "page_structs\n", start, end - 1);
99}
100
29c71111 101pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
8f6aac41 102{
29c71111
AW
103 pte_t *pte = pte_offset_kernel(pmd, addr);
104 if (pte_none(*pte)) {
105 pte_t entry;
9bdac914 106 void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
29c71111 107 if (!p)
9dce07f1 108 return NULL;
29c71111
AW
109 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
110 set_pte_at(&init_mm, addr, pte, entry);
111 }
112 return pte;
8f6aac41
CL
113}
114
29c71111 115pmd_t * __meminit vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node)
8f6aac41 116{
29c71111
AW
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)
9dce07f1 121 return NULL;
29c71111 122 pmd_populate_kernel(&init_mm, pmd, p);
8f6aac41 123 }
29c71111 124 return pmd;
8f6aac41 125}
8f6aac41 126
29c71111 127pud_t * __meminit vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node)
8f6aac41 128{
29c71111
AW
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)
9dce07f1 133 return NULL;
29c71111
AW
134 pud_populate(&init_mm, pud, p);
135 }
136 return pud;
137}
8f6aac41 138
29c71111
AW
139pgd_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)
9dce07f1 145 return NULL;
29c71111 146 pgd_populate(&init_mm, pgd, p);
8f6aac41 147 }
29c71111 148 return pgd;
8f6aac41
CL
149}
150
0aad818b
JW
151int __meminit vmemmap_populate_basepages(unsigned long start,
152 unsigned long end, int node)
8f6aac41 153{
0aad818b 154 unsigned long addr = start;
29c71111
AW
155 pgd_t *pgd;
156 pud_t *pud;
157 pmd_t *pmd;
158 pte_t *pte;
8f6aac41 159
29c71111
AW
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);
8f6aac41 174 }
29c71111
AW
175
176 return 0;
8f6aac41 177}
8f6aac41 178
98f3cfc1 179struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
8f6aac41 180{
0aad818b
JW
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))
8f6aac41
CL
190 return NULL;
191
192 return map;
193}
9bdac914
YL
194
195void __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 */
bb016b84
SS
230 memblock_free_early(__pa(vmemmap_buf),
231 vmemmap_buf_end - vmemmap_buf);
9bdac914
YL
232 vmemmap_buf = NULL;
233 vmemmap_buf_end = NULL;
234 }
235}