blob: f6a43c09c322cdb39077392ec1c77e9904c52d82 [file] [log] [blame]
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001/*
2 * sparse memory mappings.
3 */
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004#include <linux/mm.h>
5#include <linux/mmzone.h>
6#include <linux/bootmem.h>
Dave Hansen0b0acbe2005-10-29 18:16:55 -07007#include <linux/highmem.h>
Andy Whitcroftd41dee32005-06-23 00:07:54 -07008#include <linux/module.h>
Dave Hansen28ae55c2005-09-03 15:54:29 -07009#include <linux/spinlock.h>
Dave Hansen0b0acbe2005-10-29 18:16:55 -070010#include <linux/vmalloc.h>
Andy Whitcroftd41dee32005-06-23 00:07:54 -070011#include <asm/dma.h>
Christoph Lameter8f6aac42007-10-16 01:24:13 -070012#include <asm/pgalloc.h>
13#include <asm/pgtable.h>
Andy Whitcroftd41dee32005-06-23 00:07:54 -070014
15/*
16 * Permanent SPARSEMEM data:
17 *
18 * 1) mem_section - memory sections, mem_map's for valid memory
19 */
Bob Picco3e347262005-09-03 15:54:28 -070020#ifdef CONFIG_SPARSEMEM_EXTREME
Bob Picco802f1922005-09-03 15:54:26 -070021struct mem_section *mem_section[NR_SECTION_ROOTS]
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080022 ____cacheline_internodealigned_in_smp;
Bob Picco3e347262005-09-03 15:54:28 -070023#else
24struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080025 ____cacheline_internodealigned_in_smp;
Bob Picco3e347262005-09-03 15:54:28 -070026#endif
27EXPORT_SYMBOL(mem_section);
28
Christoph Lameter89689ae2006-12-06 20:31:45 -080029#ifdef NODE_NOT_IN_PAGE_FLAGS
30/*
31 * If we did not store the node number in the page then we have to
32 * do a lookup in the section_to_node_table in order to find which
33 * node the page belongs to.
34 */
35#if MAX_NUMNODES <= 256
36static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
37#else
38static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
39#endif
40
Andy Whitcroft25ba77c2006-12-06 20:33:03 -080041int page_to_nid(struct page *page)
Christoph Lameter89689ae2006-12-06 20:31:45 -080042{
43 return section_to_node_table[page_to_section(page)];
44}
45EXPORT_SYMBOL(page_to_nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -070046
47static void set_section_nid(unsigned long section_nr, int nid)
48{
49 section_to_node_table[section_nr] = nid;
50}
51#else /* !NODE_NOT_IN_PAGE_FLAGS */
52static inline void set_section_nid(unsigned long section_nr, int nid)
53{
54}
Christoph Lameter89689ae2006-12-06 20:31:45 -080055#endif
56
Bob Picco3e347262005-09-03 15:54:28 -070057#ifdef CONFIG_SPARSEMEM_EXTREME
Sam Ravnborg577a32f2007-05-17 23:29:25 +020058static struct mem_section noinline __init_refok *sparse_index_alloc(int nid)
Bob Picco802f1922005-09-03 15:54:26 -070059{
Dave Hansen28ae55c2005-09-03 15:54:29 -070060 struct mem_section *section = NULL;
61 unsigned long array_size = SECTIONS_PER_ROOT *
62 sizeof(struct mem_section);
Bob Picco802f1922005-09-03 15:54:26 -070063
Mike Kravetz39d24e62006-05-15 09:44:13 -070064 if (slab_is_available())
Mike Kravetz46a66ee2006-05-01 12:16:09 -070065 section = kmalloc_node(array_size, GFP_KERNEL, nid);
66 else
67 section = alloc_bootmem_node(NODE_DATA(nid), array_size);
Bob Picco3e347262005-09-03 15:54:28 -070068
Dave Hansen28ae55c2005-09-03 15:54:29 -070069 if (section)
70 memset(section, 0, array_size);
Bob Picco3e347262005-09-03 15:54:28 -070071
Dave Hansen28ae55c2005-09-03 15:54:29 -070072 return section;
Bob Picco802f1922005-09-03 15:54:26 -070073}
Dave Hansen28ae55c2005-09-03 15:54:29 -070074
Yasunori Gotoa3142c82007-05-08 00:23:07 -070075static int __meminit sparse_index_init(unsigned long section_nr, int nid)
Dave Hansen28ae55c2005-09-03 15:54:29 -070076{
Ingo Molnar34af9462006-06-27 02:53:55 -070077 static DEFINE_SPINLOCK(index_init_lock);
Dave Hansen28ae55c2005-09-03 15:54:29 -070078 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
79 struct mem_section *section;
80 int ret = 0;
81
82 if (mem_section[root])
83 return -EEXIST;
84
85 section = sparse_index_alloc(nid);
WANG Congaf0cd5a2007-12-17 16:19:58 -080086 if (!section)
87 return -ENOMEM;
Dave Hansen28ae55c2005-09-03 15:54:29 -070088 /*
89 * This lock keeps two different sections from
90 * reallocating for the same index
91 */
92 spin_lock(&index_init_lock);
93
94 if (mem_section[root]) {
95 ret = -EEXIST;
96 goto out;
97 }
98
99 mem_section[root] = section;
100out:
101 spin_unlock(&index_init_lock);
102 return ret;
103}
104#else /* !SPARSEMEM_EXTREME */
105static inline int sparse_index_init(unsigned long section_nr, int nid)
106{
107 return 0;
108}
109#endif
110
Dave Hansen4ca644d2005-10-29 18:16:51 -0700111/*
112 * Although written for the SPARSEMEM_EXTREME case, this happens
Andy Whitcroftcd881a62007-10-16 01:24:10 -0700113 * to also work for the flat array case because
Dave Hansen4ca644d2005-10-29 18:16:51 -0700114 * NR_SECTION_ROOTS==NR_MEM_SECTIONS.
115 */
116int __section_nr(struct mem_section* ms)
117{
118 unsigned long root_nr;
119 struct mem_section* root;
120
Mike Kravetz12783b02006-05-20 15:00:05 -0700121 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
122 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
Dave Hansen4ca644d2005-10-29 18:16:51 -0700123 if (!root)
124 continue;
125
126 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
127 break;
128 }
129
130 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
131}
132
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700133/*
134 * During early boot, before section_mem_map is used for an actual
135 * mem_map, we use section_mem_map to store the section's NUMA
136 * node. This keeps us from having to use another data structure. The
137 * node information is cleared just before we store the real mem_map.
138 */
139static inline unsigned long sparse_encode_early_nid(int nid)
140{
141 return (nid << SECTION_NID_SHIFT);
142}
143
144static inline int sparse_early_nid(struct mem_section *section)
145{
146 return (section->section_mem_map >> SECTION_NID_SHIFT);
147}
148
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700149/* Record a memory area against a node. */
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700150void __init memory_present(int nid, unsigned long start, unsigned long end)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700151{
152 unsigned long pfn;
153
154 start &= PAGE_SECTION_MASK;
155 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
156 unsigned long section = pfn_to_section_nr(pfn);
Bob Picco802f1922005-09-03 15:54:26 -0700157 struct mem_section *ms;
158
159 sparse_index_init(section, nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -0700160 set_section_nid(section, nid);
Bob Picco802f1922005-09-03 15:54:26 -0700161
162 ms = __nr_to_section(section);
163 if (!ms->section_mem_map)
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700164 ms->section_mem_map = sparse_encode_early_nid(nid) |
165 SECTION_MARKED_PRESENT;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700166 }
167}
168
169/*
170 * Only used by the i386 NUMA architecures, but relatively
171 * generic code.
172 */
173unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
174 unsigned long end_pfn)
175{
176 unsigned long pfn;
177 unsigned long nr_pages = 0;
178
179 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
180 if (nid != early_pfn_to_nid(pfn))
181 continue;
182
Andy Whitcroft540557b2007-10-16 01:24:11 -0700183 if (pfn_present(pfn))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700184 nr_pages += PAGES_PER_SECTION;
185 }
186
187 return nr_pages * sizeof(struct page);
188}
189
190/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700191 * Subtle, we encode the real pfn into the mem_map such that
192 * the identity pfn - section_mem_map will return the actual
193 * physical page frame number.
194 */
195static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
196{
197 return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
198}
199
200/*
201 * We need this if we ever free the mem_maps. While not implemented yet,
202 * this function is included for parity with its sibling.
203 */
204static __attribute((unused))
205struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
206{
207 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
208}
209
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700210static int __meminit sparse_init_one_section(struct mem_section *ms,
Mel Gorman5c0e3062007-10-16 01:25:56 -0700211 unsigned long pnum, struct page *mem_map,
212 unsigned long *pageblock_bitmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700213{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700214 if (!present_section(ms))
Andy Whitcroft29751f62005-06-23 00:08:00 -0700215 return -EINVAL;
216
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700217 ms->section_mem_map &= ~SECTION_MAP_MASK;
Andy Whitcroft540557b2007-10-16 01:24:11 -0700218 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
219 SECTION_HAS_MEM_MAP;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700220 ms->pageblock_flags = pageblock_bitmap;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700221
222 return 1;
223}
224
Mel Gorman5c0e3062007-10-16 01:25:56 -0700225static unsigned long usemap_size(void)
226{
227 unsigned long size_bytes;
228 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
229 size_bytes = roundup(size_bytes, sizeof(unsigned long));
230 return size_bytes;
231}
232
233#ifdef CONFIG_MEMORY_HOTPLUG
234static unsigned long *__kmalloc_section_usemap(void)
235{
236 return kmalloc(usemap_size(), GFP_KERNEL);
237}
238#endif /* CONFIG_MEMORY_HOTPLUG */
239
Sam Ravnborga322f8a2008-02-04 22:29:35 -0800240static unsigned long *__init sparse_early_usemap_alloc(unsigned long pnum)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700241{
242 unsigned long *usemap;
243 struct mem_section *ms = __nr_to_section(pnum);
244 int nid = sparse_early_nid(ms);
245
246 usemap = alloc_bootmem_node(NODE_DATA(nid), usemap_size());
247 if (usemap)
248 return usemap;
249
250 /* Stupid: suppress gcc warning for SPARSEMEM && !NUMA */
251 nid = 0;
252
253 printk(KERN_WARNING "%s: allocation failed\n", __FUNCTION__);
254 return NULL;
255}
256
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700257#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700258struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700259{
260 struct page *map;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700261
262 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
263 if (map)
264 return map;
265
266 map = alloc_bootmem_node(NODE_DATA(nid),
267 sizeof(struct page) * PAGES_PER_SECTION);
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700268 return map;
269}
270#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
271
272struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
273{
274 struct page *map;
275 struct mem_section *ms = __nr_to_section(pnum);
276 int nid = sparse_early_nid(ms);
277
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700278 map = sparse_mem_map_populate(pnum, nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700279 if (map)
280 return map;
281
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700282 printk(KERN_ERR "%s: sparsemem memory map backing failed "
283 "some memory will not be available.\n", __FUNCTION__);
Bob Picco802f1922005-09-03 15:54:26 -0700284 ms->section_mem_map = 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700285 return NULL;
286}
287
Stephen Rothwell193faea2007-06-08 13:46:51 -0700288/*
289 * Allocate the accumulated non-linear sections, allocate a mem_map
290 * for each and record the physical to section mapping.
291 */
292void __init sparse_init(void)
293{
294 unsigned long pnum;
295 struct page *map;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700296 unsigned long *usemap;
Stephen Rothwell193faea2007-06-08 13:46:51 -0700297
298 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
Andy Whitcroft540557b2007-10-16 01:24:11 -0700299 if (!present_section_nr(pnum))
Stephen Rothwell193faea2007-06-08 13:46:51 -0700300 continue;
301
302 map = sparse_early_mem_map_alloc(pnum);
303 if (!map)
304 continue;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700305
306 usemap = sparse_early_usemap_alloc(pnum);
307 if (!usemap)
308 continue;
309
310 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
311 usemap);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700312 }
313}
314
315#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700316#ifdef CONFIG_SPARSEMEM_VMEMMAP
317static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
318 unsigned long nr_pages)
319{
320 /* This will make the necessary allocations eventually. */
321 return sparse_mem_map_populate(pnum, nid);
322}
323static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
324{
325 return; /* XXX: Not implemented yet */
326}
327#else
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700328static struct page *__kmalloc_section_memmap(unsigned long nr_pages)
329{
330 struct page *page, *ret;
331 unsigned long memmap_size = sizeof(struct page) * nr_pages;
332
Yasunori Gotof2d0aa52006-10-28 10:38:32 -0700333 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700334 if (page)
335 goto got_map_page;
336
337 ret = vmalloc(memmap_size);
338 if (ret)
339 goto got_map_ptr;
340
341 return NULL;
342got_map_page:
343 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
344got_map_ptr:
345 memset(ret, 0, memmap_size);
346
347 return ret;
348}
349
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700350static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
351 unsigned long nr_pages)
352{
353 return __kmalloc_section_memmap(nr_pages);
354}
355
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700356static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
357{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800358 if (is_vmalloc_addr(memmap))
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700359 vfree(memmap);
360 else
361 free_pages((unsigned long)memmap,
362 get_order(sizeof(struct page) * nr_pages));
363}
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700364#endif /* CONFIG_SPARSEMEM_VMEMMAP */
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700365
Andy Whitcroft29751f62005-06-23 00:08:00 -0700366/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700367 * returns the number of sections whose mem_maps were properly
368 * set. If this is <=0, then that means that the passed-in
369 * map was not consumed and must be freed.
370 */
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700371int sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
372 int nr_pages)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700373{
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700374 unsigned long section_nr = pfn_to_section_nr(start_pfn);
375 struct pglist_data *pgdat = zone->zone_pgdat;
376 struct mem_section *ms;
377 struct page *memmap;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700378 unsigned long *usemap;
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700379 unsigned long flags;
380 int ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700381
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700382 /*
383 * no locking for this, because it does its own
384 * plus, it does a kmalloc
385 */
WANG Congbbd06822007-12-17 16:19:59 -0800386 ret = sparse_index_init(section_nr, pgdat->node_id);
387 if (ret < 0 && ret != -EEXIST)
388 return ret;
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700389 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800390 if (!memmap)
391 return -ENOMEM;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700392 usemap = __kmalloc_section_usemap();
WANG Congbbd06822007-12-17 16:19:59 -0800393 if (!usemap) {
394 __kfree_section_memmap(memmap, nr_pages);
395 return -ENOMEM;
396 }
Andy Whitcroft29751f62005-06-23 00:08:00 -0700397
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700398 pgdat_resize_lock(pgdat, &flags);
399
400 ms = __pfn_to_section(start_pfn);
401 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
402 ret = -EEXIST;
403 goto out;
404 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700405
Andy Whitcroft29751f62005-06-23 00:08:00 -0700406 ms->section_mem_map |= SECTION_MARKED_PRESENT;
407
Mel Gorman5c0e3062007-10-16 01:25:56 -0700408 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700409
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700410out:
411 pgdat_resize_unlock(pgdat, &flags);
WANG Congbbd06822007-12-17 16:19:59 -0800412 if (ret <= 0) {
413 kfree(usemap);
Mike Kravetz46a66ee2006-05-01 12:16:09 -0700414 __kfree_section_memmap(memmap, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800415 }
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700416 return ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700417}
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700418#endif