]> Pileus Git - ~andy/linux/blob - mm/memory_hotplug.c
memory unplug: memory hotplug cleanup
[~andy/linux] / mm / memory_hotplug.c
1 /*
2  *  linux/mm/memory_hotplug.c
3  *
4  *  Copyright (C)
5  */
6
7 #include <linux/stddef.h>
8 #include <linux/mm.h>
9 #include <linux/swap.h>
10 #include <linux/interrupt.h>
11 #include <linux/pagemap.h>
12 #include <linux/bootmem.h>
13 #include <linux/compiler.h>
14 #include <linux/module.h>
15 #include <linux/pagevec.h>
16 #include <linux/writeback.h>
17 #include <linux/slab.h>
18 #include <linux/sysctl.h>
19 #include <linux/cpu.h>
20 #include <linux/memory.h>
21 #include <linux/memory_hotplug.h>
22 #include <linux/highmem.h>
23 #include <linux/vmalloc.h>
24 #include <linux/ioport.h>
25 #include <linux/cpuset.h>
26
27 #include <asm/tlbflush.h>
28
29 /* add this memory to iomem resource */
30 static struct resource *register_memory_resource(u64 start, u64 size)
31 {
32         struct resource *res;
33         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
34         BUG_ON(!res);
35
36         res->name = "System RAM";
37         res->start = start;
38         res->end = start + size - 1;
39         res->flags = IORESOURCE_MEM;
40         if (request_resource(&iomem_resource, res) < 0) {
41                 printk("System RAM resource %llx - %llx cannot be added\n",
42                 (unsigned long long)res->start, (unsigned long long)res->end);
43                 kfree(res);
44                 res = NULL;
45         }
46         return res;
47 }
48
49 static void release_memory_resource(struct resource *res)
50 {
51         if (!res)
52                 return;
53         release_resource(res);
54         kfree(res);
55         return;
56 }
57
58
59 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
60 static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
61 {
62         struct pglist_data *pgdat = zone->zone_pgdat;
63         int nr_pages = PAGES_PER_SECTION;
64         int nid = pgdat->node_id;
65         int zone_type;
66
67         zone_type = zone - pgdat->node_zones;
68         if (!zone->wait_table) {
69                 int ret = 0;
70                 ret = init_currently_empty_zone(zone, phys_start_pfn,
71                                                 nr_pages, MEMMAP_HOTPLUG);
72                 if (ret < 0)
73                         return ret;
74         }
75         memmap_init_zone(nr_pages, nid, zone_type,
76                          phys_start_pfn, MEMMAP_HOTPLUG);
77         return 0;
78 }
79
80 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
81 {
82         int nr_pages = PAGES_PER_SECTION;
83         int ret;
84
85         if (pfn_valid(phys_start_pfn))
86                 return -EEXIST;
87
88         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
89
90         if (ret < 0)
91                 return ret;
92
93         ret = __add_zone(zone, phys_start_pfn);
94
95         if (ret < 0)
96                 return ret;
97
98         return register_new_memory(__pfn_to_section(phys_start_pfn));
99 }
100
101 /*
102  * Reasonably generic function for adding memory.  It is
103  * expected that archs that support memory hotplug will
104  * call this function after deciding the zone to which to
105  * add the new pages.
106  */
107 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
108                  unsigned long nr_pages)
109 {
110         unsigned long i;
111         int err = 0;
112         int start_sec, end_sec;
113         /* during initialize mem_map, align hot-added range to section */
114         start_sec = pfn_to_section_nr(phys_start_pfn);
115         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
116
117         for (i = start_sec; i <= end_sec; i++) {
118                 err = __add_section(zone, i << PFN_SECTION_SHIFT);
119
120                 /*
121                  * EEXIST is finally dealed with by ioresource collision
122                  * check. see add_memory() => register_memory_resource()
123                  * Warning will be printed if there is collision.
124                  */
125                 if (err && (err != -EEXIST))
126                         break;
127                 err = 0;
128         }
129
130         return err;
131 }
132 EXPORT_SYMBOL_GPL(__add_pages);
133
134 static void grow_zone_span(struct zone *zone,
135                 unsigned long start_pfn, unsigned long end_pfn)
136 {
137         unsigned long old_zone_end_pfn;
138
139         zone_span_writelock(zone);
140
141         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
142         if (start_pfn < zone->zone_start_pfn)
143                 zone->zone_start_pfn = start_pfn;
144
145         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
146                                 zone->zone_start_pfn;
147
148         zone_span_writeunlock(zone);
149 }
150
151 static void grow_pgdat_span(struct pglist_data *pgdat,
152                 unsigned long start_pfn, unsigned long end_pfn)
153 {
154         unsigned long old_pgdat_end_pfn =
155                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
156
157         if (start_pfn < pgdat->node_start_pfn)
158                 pgdat->node_start_pfn = start_pfn;
159
160         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
161                                         pgdat->node_start_pfn;
162 }
163
164 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
165                         void *arg)
166 {
167         unsigned long i;
168         unsigned long onlined_pages = *(unsigned long *)arg;
169         struct page *page;
170         if (PageReserved(pfn_to_page(start_pfn)))
171                 for (i = 0; i < nr_pages; i++) {
172                         page = pfn_to_page(start_pfn + i);
173                         online_page(page);
174                         onlined_pages++;
175                 }
176         *(unsigned long *)arg = onlined_pages;
177         return 0;
178 }
179
180
181 int online_pages(unsigned long pfn, unsigned long nr_pages)
182 {
183         unsigned long flags;
184         unsigned long onlined_pages = 0;
185         struct zone *zone;
186         int need_zonelists_rebuild = 0;
187
188         /*
189          * This doesn't need a lock to do pfn_to_page().
190          * The section can't be removed here because of the
191          * memory_block->state_sem.
192          */
193         zone = page_zone(pfn_to_page(pfn));
194         pgdat_resize_lock(zone->zone_pgdat, &flags);
195         grow_zone_span(zone, pfn, pfn + nr_pages);
196         grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
197         pgdat_resize_unlock(zone->zone_pgdat, &flags);
198
199         /*
200          * If this zone is not populated, then it is not in zonelist.
201          * This means the page allocator ignores this zone.
202          * So, zonelist must be updated after online.
203          */
204         if (!populated_zone(zone))
205                 need_zonelists_rebuild = 1;
206
207         walk_memory_resource(pfn, nr_pages, &onlined_pages,
208                 online_pages_range);
209         zone->present_pages += onlined_pages;
210         zone->zone_pgdat->node_present_pages += onlined_pages;
211
212         setup_per_zone_pages_min();
213         if (onlined_pages) {
214                 kswapd_run(zone_to_nid(zone));
215                 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
216         }
217
218         if (need_zonelists_rebuild)
219                 build_all_zonelists();
220         vm_total_pages = nr_free_pagecache_pages();
221         writeback_set_ratelimit();
222         return 0;
223 }
224 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
225
226 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
227 {
228         struct pglist_data *pgdat;
229         unsigned long zones_size[MAX_NR_ZONES] = {0};
230         unsigned long zholes_size[MAX_NR_ZONES] = {0};
231         unsigned long start_pfn = start >> PAGE_SHIFT;
232
233         pgdat = arch_alloc_nodedata(nid);
234         if (!pgdat)
235                 return NULL;
236
237         arch_refresh_nodedata(nid, pgdat);
238
239         /* we can use NODE_DATA(nid) from here */
240
241         /* init node's zones as empty zones, we don't have any present pages.*/
242         free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
243
244         return pgdat;
245 }
246
247 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
248 {
249         arch_refresh_nodedata(nid, NULL);
250         arch_free_nodedata(pgdat);
251         return;
252 }
253
254
255 int add_memory(int nid, u64 start, u64 size)
256 {
257         pg_data_t *pgdat = NULL;
258         int new_pgdat = 0;
259         struct resource *res;
260         int ret;
261
262         res = register_memory_resource(start, size);
263         if (!res)
264                 return -EEXIST;
265
266         if (!node_online(nid)) {
267                 pgdat = hotadd_new_pgdat(nid, start);
268                 if (!pgdat)
269                         return -ENOMEM;
270                 new_pgdat = 1;
271         }
272
273         /* call arch's memory hotadd */
274         ret = arch_add_memory(nid, start, size);
275
276         if (ret < 0)
277                 goto error;
278
279         /* we online node here. we can't roll back from here. */
280         node_set_online(nid);
281
282         cpuset_track_online_nodes();
283
284         if (new_pgdat) {
285                 ret = register_one_node(nid);
286                 /*
287                  * If sysfs file of new node can't create, cpu on the node
288                  * can't be hot-added. There is no rollback way now.
289                  * So, check by BUG_ON() to catch it reluctantly..
290                  */
291                 BUG_ON(ret);
292         }
293
294         return ret;
295 error:
296         /* rollback pgdat allocation and others */
297         if (new_pgdat)
298                 rollback_node_hotadd(nid, pgdat);
299         if (res)
300                 release_memory_resource(res);
301
302         return ret;
303 }
304 EXPORT_SYMBOL_GPL(add_memory);