2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.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>
27 #include <asm/tlbflush.h>
29 static int __add_zone(struct zone
*zone
, unsigned long phys_start_pfn
)
31 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
32 int nr_pages
= PAGES_PER_SECTION
;
33 int nid
= pgdat
->node_id
;
36 zone_type
= zone
- pgdat
->node_zones
;
37 if (!populated_zone(zone
)) {
39 ret
= init_currently_empty_zone(zone
, phys_start_pfn
, nr_pages
);
43 memmap_init_zone(nr_pages
, nid
, zone_type
, phys_start_pfn
);
44 zonetable_add(zone
, nid
, zone_type
, phys_start_pfn
, nr_pages
);
48 static int __add_section(struct zone
*zone
, unsigned long phys_start_pfn
)
50 int nr_pages
= PAGES_PER_SECTION
;
53 if (pfn_valid(phys_start_pfn
))
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
)
80 int start_sec
, end_sec
;
81 /* during initialize mem_map, align hot-added range to section */
82 start_sec
= pfn_to_section_nr(phys_start_pfn
);
83 end_sec
= pfn_to_section_nr(phys_start_pfn
+ nr_pages
- 1);
85 for (i
= start_sec
; i
<= end_sec
; i
++) {
86 err
= __add_section(zone
, i
<< PFN_SECTION_SHIFT
);
89 * EEXIST is finally dealed with by ioresource collision
90 * check. see add_memory() => register_memory_resource()
91 * Warning will be printed if there is collision.
93 if (err
&& (err
!= -EEXIST
))
100 EXPORT_SYMBOL_GPL(__add_pages
);
102 static void grow_zone_span(struct zone
*zone
,
103 unsigned long start_pfn
, unsigned long end_pfn
)
105 unsigned long old_zone_end_pfn
;
107 zone_span_writelock(zone
);
109 old_zone_end_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
110 if (start_pfn
< zone
->zone_start_pfn
)
111 zone
->zone_start_pfn
= start_pfn
;
113 zone
->spanned_pages
= max(old_zone_end_pfn
, end_pfn
) -
114 zone
->zone_start_pfn
;
116 zone_span_writeunlock(zone
);
119 static void grow_pgdat_span(struct pglist_data
*pgdat
,
120 unsigned long start_pfn
, unsigned long end_pfn
)
122 unsigned long old_pgdat_end_pfn
=
123 pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
125 if (start_pfn
< pgdat
->node_start_pfn
)
126 pgdat
->node_start_pfn
= start_pfn
;
128 pgdat
->node_spanned_pages
= max(old_pgdat_end_pfn
, end_pfn
) -
129 pgdat
->node_start_pfn
;
132 int online_pages(unsigned long pfn
, unsigned long nr_pages
)
136 unsigned long onlined_pages
= 0;
139 unsigned long start_pfn
;
141 int need_zonelists_rebuild
= 0;
144 * This doesn't need a lock to do pfn_to_page().
145 * The section can't be removed here because of the
146 * memory_block->state_sem.
148 zone
= page_zone(pfn_to_page(pfn
));
149 pgdat_resize_lock(zone
->zone_pgdat
, &flags
);
150 grow_zone_span(zone
, pfn
, pfn
+ nr_pages
);
151 grow_pgdat_span(zone
->zone_pgdat
, pfn
, pfn
+ nr_pages
);
152 pgdat_resize_unlock(zone
->zone_pgdat
, &flags
);
155 * If this zone is not populated, then it is not in zonelist.
156 * This means the page allocator ignores this zone.
157 * So, zonelist must be updated after online.
159 if (!populated_zone(zone
))
160 need_zonelists_rebuild
= 1;
162 res
.start
= (u64
)pfn
<< PAGE_SHIFT
;
163 res
.end
= res
.start
+ ((u64
)nr_pages
<< PAGE_SHIFT
) - 1;
164 res
.flags
= IORESOURCE_MEM
; /* we just need system ram */
165 section_end
= res
.end
;
167 while ((res
.start
< res
.end
) && (find_next_system_ram(&res
) >= 0)) {
168 start_pfn
= (unsigned long)(res
.start
>> PAGE_SHIFT
);
169 nr_pages
= (unsigned long)
170 ((res
.end
+ 1 - res
.start
) >> PAGE_SHIFT
);
172 if (PageReserved(pfn_to_page(start_pfn
))) {
173 /* this region's page is not onlined now */
174 for (i
= 0; i
< nr_pages
; i
++) {
175 struct page
*page
= pfn_to_page(start_pfn
+ i
);
181 res
.start
= res
.end
+ 1;
182 res
.end
= section_end
;
184 zone
->present_pages
+= onlined_pages
;
185 zone
->zone_pgdat
->node_present_pages
+= onlined_pages
;
187 setup_per_zone_pages_min();
189 if (need_zonelists_rebuild
)
190 build_all_zonelists();
191 vm_total_pages
= nr_free_pagecache_pages();
192 writeback_set_ratelimit();
196 static pg_data_t
*hotadd_new_pgdat(int nid
, u64 start
)
198 struct pglist_data
*pgdat
;
199 unsigned long zones_size
[MAX_NR_ZONES
] = {0};
200 unsigned long zholes_size
[MAX_NR_ZONES
] = {0};
201 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
203 pgdat
= arch_alloc_nodedata(nid
);
207 arch_refresh_nodedata(nid
, pgdat
);
209 /* we can use NODE_DATA(nid) from here */
211 /* init node's zones as empty zones, we don't have any present pages.*/
212 free_area_init_node(nid
, pgdat
, zones_size
, start_pfn
, zholes_size
);
217 static void rollback_node_hotadd(int nid
, pg_data_t
*pgdat
)
219 arch_refresh_nodedata(nid
, NULL
);
220 arch_free_nodedata(pgdat
);
224 /* add this memory to iomem resource */
225 static struct resource
*register_memory_resource(u64 start
, u64 size
)
227 struct resource
*res
;
228 res
= kzalloc(sizeof(struct resource
), GFP_KERNEL
);
231 res
->name
= "System RAM";
233 res
->end
= start
+ size
- 1;
234 res
->flags
= IORESOURCE_MEM
;
235 if (request_resource(&iomem_resource
, res
) < 0) {
236 printk("System RAM resource %llx - %llx cannot be added\n",
237 (unsigned long long)res
->start
, (unsigned long long)res
->end
);
244 static void release_memory_resource(struct resource
*res
)
248 release_resource(res
);
255 int add_memory(int nid
, u64 start
, u64 size
)
257 pg_data_t
*pgdat
= NULL
;
259 struct resource
*res
;
262 res
= register_memory_resource(start
, size
);
266 if (!node_online(nid
)) {
267 pgdat
= hotadd_new_pgdat(nid
, start
);
271 ret
= kswapd_run(nid
);
276 /* call arch's memory hotadd */
277 ret
= arch_add_memory(nid
, start
, size
);
282 /* we online node here. we can't roll back from here. */
283 node_set_online(nid
);
285 cpuset_track_online_nodes();
288 ret
= register_one_node(nid
);
290 * If sysfs file of new node can't create, cpu on the node
291 * can't be hot-added. There is no rollback way now.
292 * So, check by BUG_ON() to catch it reluctantly..
299 /* rollback pgdat allocation and others */
301 rollback_node_hotadd(nid
, pgdat
);
303 release_memory_resource(res
);
307 EXPORT_SYMBOL_GPL(add_memory
);