[PATCH] wait_table and zonelist initializing for memory hotadd: update zonelists
[safe/jmp/linux-2.6] / mm / memory_hotplug.c
1 /*
2  *  linux/mm/memory_hotplug.c
3  *
4  *  Copyright (C)
5  */
6
7 #include <linux/config.h>
8 #include <linux/stddef.h>
9 #include <linux/mm.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.h>
13 #include <linux/bootmem.h>
14 #include <linux/compiler.h>
15 #include <linux/module.h>
16 #include <linux/pagevec.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
25 #include <asm/tlbflush.h>
26
27 extern void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
28                           unsigned long size);
29 static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
30 {
31         struct pglist_data *pgdat = zone->zone_pgdat;
32         int nr_pages = PAGES_PER_SECTION;
33         int nid = pgdat->node_id;
34         int zone_type;
35
36         zone_type = zone - pgdat->node_zones;
37         if (!populated_zone(zone)) {
38                 int ret = 0;
39                 ret = init_currently_empty_zone(zone, phys_start_pfn, nr_pages);
40                 if (ret < 0)
41                         return ret;
42         }
43         memmap_init_zone(nr_pages, nid, zone_type, phys_start_pfn);
44         zonetable_add(zone, nid, zone_type, phys_start_pfn, nr_pages);
45         return 0;
46 }
47
48 extern int sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
49                                   int nr_pages);
50 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
51 {
52         int nr_pages = PAGES_PER_SECTION;
53         int ret;
54
55         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
56
57         if (ret < 0)
58                 return ret;
59
60         ret = __add_zone(zone, phys_start_pfn);
61
62         if (ret < 0)
63                 return ret;
64
65         return register_new_memory(__pfn_to_section(phys_start_pfn));
66 }
67
68 /*
69  * Reasonably generic function for adding memory.  It is
70  * expected that archs that support memory hotplug will
71  * call this function after deciding the zone to which to
72  * add the new pages.
73  */
74 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
75                  unsigned long nr_pages)
76 {
77         unsigned long i;
78         int err = 0;
79
80         for (i = 0; i < nr_pages; i += PAGES_PER_SECTION) {
81                 err = __add_section(zone, phys_start_pfn + i);
82
83                 /* We want to keep adding the rest of the
84                  * sections if the first ones already exist
85                  */
86                 if (err && (err != -EEXIST))
87                         break;
88         }
89
90         return err;
91 }
92 EXPORT_SYMBOL_GPL(__add_pages);
93
94 static void grow_zone_span(struct zone *zone,
95                 unsigned long start_pfn, unsigned long end_pfn)
96 {
97         unsigned long old_zone_end_pfn;
98
99         zone_span_writelock(zone);
100
101         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
102         if (start_pfn < zone->zone_start_pfn)
103                 zone->zone_start_pfn = start_pfn;
104
105         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
106                                 zone->zone_start_pfn;
107
108         zone_span_writeunlock(zone);
109 }
110
111 static void grow_pgdat_span(struct pglist_data *pgdat,
112                 unsigned long start_pfn, unsigned long end_pfn)
113 {
114         unsigned long old_pgdat_end_pfn =
115                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
116
117         if (start_pfn < pgdat->node_start_pfn)
118                 pgdat->node_start_pfn = start_pfn;
119
120         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
121                                         pgdat->node_start_pfn;
122 }
123
124 int online_pages(unsigned long pfn, unsigned long nr_pages)
125 {
126         unsigned long i;
127         unsigned long flags;
128         unsigned long onlined_pages = 0;
129         struct zone *zone;
130         int need_zonelists_rebuild = 0;
131
132         /*
133          * This doesn't need a lock to do pfn_to_page().
134          * The section can't be removed here because of the
135          * memory_block->state_sem.
136          */
137         zone = page_zone(pfn_to_page(pfn));
138         pgdat_resize_lock(zone->zone_pgdat, &flags);
139         grow_zone_span(zone, pfn, pfn + nr_pages);
140         grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
141         pgdat_resize_unlock(zone->zone_pgdat, &flags);
142
143         /*
144          * If this zone is not populated, then it is not in zonelist.
145          * This means the page allocator ignores this zone.
146          * So, zonelist must be updated after online.
147          */
148         if (!populated_zone(zone))
149                 need_zonelists_rebuild = 1;
150
151         for (i = 0; i < nr_pages; i++) {
152                 struct page *page = pfn_to_page(pfn + i);
153                 online_page(page);
154                 onlined_pages++;
155         }
156         zone->present_pages += onlined_pages;
157         zone->zone_pgdat->node_present_pages += onlined_pages;
158
159         setup_per_zone_pages_min();
160
161         if (need_zonelists_rebuild)
162                 build_all_zonelists();
163
164         return 0;
165 }