2 * linux/mm/memory_hotplug.c
7 #include <linux/config.h>
8 #include <linux/stddef.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 #include <linux/ioport.h>
26 #include <asm/tlbflush.h>
28 extern void zonetable_add(struct zone
*zone
, int nid
, int zid
, unsigned long pfn
,
30 static int __add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
32 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
33 int nr_pages
= PAGES_PER_SECTION
;
34 int nid
= pgdat
->node_id
;
37 zone_type
= zone
- pgdat
->node_zones
;
38 if (!populated_zone(zone
)) {
40 ret
= init_currently_empty_zone(zone
, phys_start_pfn
, nr_pages
);
44 memmap_init_zone(nr_pages
, nid
, zone_type
, phys_start_pfn
);
45 zonetable_add(zone
, nid
, zone_type
, phys_start_pfn
, nr_pages
);
49 extern int sparse_add_one_section(struct zone
*zone
, unsigned long start_pfn
,
51 static int __add_section(struct zone
*zone
, unsigned long phys_start_pfn
)
53 int nr_pages
= PAGES_PER_SECTION
;
56 ret
= sparse_add_one_section(zone
, phys_start_pfn
, nr_pages
);
61 ret
= __add_zone(zone
, phys_start_pfn
);
66 return register_new_memory(__pfn_to_section(phys_start_pfn
));
70 * Reasonably generic function for adding memory. It is
71 * expected that archs that support memory hotplug will
72 * call this function after deciding the zone to which to
75 int __add_pages(struct zone
*zone
, unsigned long phys_start_pfn
,
76 unsigned long nr_pages
)
81 for (i
= 0; i
< nr_pages
; i
+= PAGES_PER_SECTION
) {
82 err
= __add_section(zone
, phys_start_pfn
+ i
);
84 /* We want to keep adding the rest of the
85 * sections if the first ones already exist
87 if (err
&& (err
!= -EEXIST
))
93 EXPORT_SYMBOL_GPL(__add_pages
);
95 static void grow_zone_span(struct zone
*zone
,
96 unsigned long start_pfn
, unsigned long end_pfn
)
98 unsigned long old_zone_end_pfn
;
100 zone_span_writelock(zone
);
102 old_zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
103 if (start_pfn
< zone
->zone_start_pfn
)
104 zone
->zone_start_pfn
= start_pfn
;
106 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
107 zone
->zone_start_pfn
;
109 zone_span_writeunlock(zone
);
112 static void grow_pgdat_span(struct pglist_data
*pgdat
,
113 unsigned long start_pfn
, unsigned long end_pfn
)
115 unsigned long old_pgdat_end_pfn
=
116 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
118 if (start_pfn
< pgdat
->node_start_pfn
)
119 pgdat
->node_start_pfn
= start_pfn
;
121 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
122 pgdat
->node_start_pfn
;
125 int online_pages(unsigned long pfn
, unsigned long nr_pages
)
129 unsigned long onlined_pages
= 0;
131 int need_zonelists_rebuild
= 0;
134 * This doesn't need a lock to do pfn_to_page().
135 * The section can't be removed here because of the
136 * memory_block->state_sem.
138 zone
= page_zone(pfn_to_page(pfn
));
139 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
140 grow_zone_span(zone
, pfn
, pfn
+ nr_pages
);
141 grow_pgdat_span(zone
->zone_pgdat
, pfn
, pfn
+ nr_pages
);
142 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
145 * If this zone is not populated, then it is not in zonelist.
146 * This means the page allocator ignores this zone.
147 * So, zonelist must be updated after online.
149 if (!populated_zone(zone
))
150 need_zonelists_rebuild
= 1;
152 for (i
= 0; i
< nr_pages
; i
++) {
153 struct page
*page
= pfn_to_page(pfn
+ i
);
157 zone
->present_pages
+= onlined_pages
;
158 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
160 setup_per_zone_pages_min();
162 if (need_zonelists_rebuild
)
163 build_all_zonelists();
164 vm_total_pages
= nr_free_pagecache_pages();
168 static pg_data_t
*hotadd_new_pgdat(int nid
, u64 start
)
170 struct pglist_data
*pgdat
;
171 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
172 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
173 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
175 pgdat
= arch_alloc_nodedata(nid
);
179 arch_refresh_nodedata(nid
, pgdat
);
181 /* we can use NODE_DATA(nid) from here */
183 /* init node's zones as empty zones, we don't have any present pages.*/
184 free_area_init_node(nid
, pgdat
, zones_size
, start_pfn
, zholes_size
);
189 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
191 arch_refresh_nodedata(nid
, NULL
);
192 arch_free_nodedata(pgdat
);
196 /* add this memory to iomem resource */
197 static void register_memory_resource(u64 start
, u64 size
)
199 struct resource
*res
;
201 res
= kzalloc(sizeof(struct resource
), GFP_KERNEL
);
204 res
->name
= "System RAM";
206 res
->end
= start
+ size
- 1;
207 res
->flags
= IORESOURCE_MEM
;
208 if (request_resource(&iomem_resource
, res
) < 0) {
209 printk("System RAM resource %llx - %llx cannot be added\n",
210 (unsigned long long)res
->start
, (unsigned long long)res
->end
);
217 int add_memory(int nid
, u64 start
, u64 size
)
219 pg_data_t
*pgdat
= NULL
;
223 if (!node_online(nid
)) {
224 pgdat
= hotadd_new_pgdat(nid
, start
);
228 ret
= kswapd_run(nid
);
233 /* call arch's memory hotadd */
234 ret
= arch_add_memory(nid
, start
, size
);
239 /* we online node here. we have no error path from here. */
240 node_set_online(nid
);
242 /* register this memory as resource */
243 register_memory_resource(start
, size
);
247 /* rollback pgdat allocation and others */
249 rollback_node_hotadd(nid
, pgdat
);
253 EXPORT_SYMBOL_GPL(add_memory
);