1 /*
2 * Generic VM initialization for x86-64 NUMA setups.
3 * Copyright 2002 Andi Kleen, SuSE Labs.
4 * $Id: numa.c,v 1.6 2003/04/03 12:28:08 ak Exp $
5 */
6 #include <linux/kernel.h>
7 #include <linux/mm.h>
8 #include <linux/string.h>
9 #include <linux/init.h>
10 #include <linux/bootmem.h>
11 #include <linux/mmzone.h>
12 #include <linux/blk.h>
13 #include <asm/e820.h>
14 #include <asm/proto.h>
15 #include <asm/dma.h>
16
17 #undef Dprintk
18 #define Dprintk(...)
19
20 plat_pg_data_t *plat_node_data[MAXNODE];
21 bootmem_data_t plat_node_bdata[MAX_NUMNODES];
22
23 #define ZONE_ALIGN (1UL << (MAX_ORDER+PAGE_SHIFT))
24
25 static int numa_off __initdata;
26
27 unsigned long nodes_present;
28 int maxnode;
29
30 /* Initialize bootmem allocator for a node */
setup_node_bootmem(int nodeid,unsigned long start,unsigned long end)31 void __init setup_node_bootmem(int nodeid, unsigned long start, unsigned long end)
32 {
33 unsigned long start_pfn, end_pfn, bootmap_pages, bootmap_size, bootmap_start;
34 unsigned long nodedata_phys;
35 const int pgdat_size = round_up(sizeof(plat_pg_data_t), PAGE_SIZE);
36
37 start = round_up(start, ZONE_ALIGN);
38
39 printk("Bootmem setup node %d %016lx-%016lx\n", nodeid, start, end);
40
41 start_pfn = start >> PAGE_SHIFT;
42 end_pfn = end >> PAGE_SHIFT;
43
44 nodedata_phys = find_e820_area(start, end, pgdat_size);
45 if (nodedata_phys == -1L)
46 panic("Cannot find memory pgdat in node %d\n", nodeid);
47
48 Dprintk("nodedata_phys %lx\n", nodedata_phys);
49
50 PLAT_NODE_DATA(nodeid) = phys_to_virt(nodedata_phys);
51 memset(PLAT_NODE_DATA(nodeid), 0, sizeof(plat_pg_data_t));
52 NODE_DATA(nodeid)->bdata = &plat_node_bdata[nodeid];
53
54 /* Find a place for the bootmem map */
55 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
56 bootmap_start = round_up(nodedata_phys + pgdat_size, PAGE_SIZE);
57 bootmap_start = find_e820_area(bootmap_start, end, bootmap_pages<<PAGE_SHIFT);
58 if (bootmap_start == -1L)
59 panic("Not enough continuous space for bootmap on node %d", nodeid);
60 Dprintk("bootmap start %lu pages %lu\n", bootmap_start, bootmap_pages);
61
62 bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
63 bootmap_start >> PAGE_SHIFT,
64 start_pfn, end_pfn);
65
66 e820_bootmem_free(NODE_DATA(nodeid), start, end);
67
68 reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys, pgdat_size);
69 reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, bootmap_pages<<PAGE_SHIFT);
70
71 PLAT_NODE_DATA(nodeid)->start_pfn = start_pfn;
72 PLAT_NODE_DATA(nodeid)->end_pfn = end_pfn;
73
74 if (nodeid > maxnode)
75 maxnode = nodeid;
76 nodes_present |= (1UL << nodeid);
77 }
78
79 /* Initialize final allocator for a zone */
setup_node_zones(int nodeid)80 void __init setup_node_zones(int nodeid)
81 {
82 unsigned long start_pfn, end_pfn;
83 unsigned long zones[MAX_NR_ZONES];
84 unsigned long dma_end_pfn;
85 unsigned long lmax_mapnr;
86
87 memset(zones, 0, sizeof(unsigned long) * MAX_NR_ZONES);
88
89 start_pfn = PLAT_NODE_DATA(nodeid)->start_pfn;
90 end_pfn = PLAT_NODE_DATA(nodeid)->end_pfn;
91
92 printk("setting up node %d %lx-%lx\n", nodeid, start_pfn, end_pfn);
93
94 /* All nodes > 0 have a zero length zone DMA */
95 dma_end_pfn = __pa(MAX_DMA_ADDRESS) >> PAGE_SHIFT;
96 if (start_pfn < dma_end_pfn) {
97 zones[ZONE_DMA] = dma_end_pfn - start_pfn;
98 zones[ZONE_NORMAL] = end_pfn - dma_end_pfn;
99 } else {
100 zones[ZONE_NORMAL] = end_pfn - start_pfn;
101 }
102
103 free_area_init_node(nodeid, NODE_DATA(nodeid), NULL, zones,
104 start_pfn<<PAGE_SHIFT, NULL);
105 lmax_mapnr = PLAT_NODE_DATA_STARTNR(nodeid) + PLAT_NODE_DATA_SIZE(nodeid);
106 if (lmax_mapnr > max_mapnr)
107 max_mapnr = lmax_mapnr;
108 }
109
110 int fake_node;
111
numa_initmem_init(unsigned long start_pfn,unsigned long end_pfn)112 int __init numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
113 {
114 #ifdef CONFIG_K8_NUMA
115 if (!numa_off && !k8_scan_nodes(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT))
116 return 0;
117 #endif
118 printk(KERN_INFO "%s\n",
119 numa_off ? "NUMA turned off" : "No NUMA configuration found");
120
121 printk(KERN_INFO "Faking a node at %016lx-%016lx\n",
122 start_pfn << PAGE_SHIFT,
123 end_pfn << PAGE_SHIFT);
124 /* setup dummy node covering all memory */
125 fake_node = 1;
126 memnode_shift = 63;
127 memnodemap[0] = 0;
128 setup_node_bootmem(0, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
129 return -1;
130 }
131
132 #define for_all_nodes(x) for ((x) = 0; (x) <= maxnode; (x)++) \
133 if ((1UL << (x)) & nodes_present)
134
numa_free_all_bootmem(void)135 unsigned long __init numa_free_all_bootmem(void)
136 {
137 int i;
138 unsigned long pages = 0;
139 for_all_nodes(i) {
140 pages += free_all_bootmem_node(NODE_DATA(i));
141 }
142 return pages;
143 }
144
paging_init(void)145 void __init paging_init(void)
146 {
147 int i;
148 for_all_nodes(i) {
149 setup_node_zones(i);
150 }
151 }
152
show_mem(void)153 void show_mem(void)
154 {
155 long i,free = 0,total = 0,reserved = 0;
156 long shared = 0, cached = 0;
157 int nid;
158
159 printk("\nMem-info:\n");
160 show_free_areas();
161 printk("Free swap: %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10));
162 for_all_nodes (nid) {
163 mem_map_t * lmem_map = NODE_MEM_MAP(nid);
164 i = PLAT_NODE_DATA_SIZE(nid);
165 while (i-- > 0) {
166 total++;
167 if (PageReserved(lmem_map+i))
168 reserved++;
169 else if (PageSwapCache(lmem_map+i))
170 cached++;
171 else if (!page_count(lmem_map+i))
172 free++;
173 else
174 shared += atomic_read(&lmem_map[i].count) - 1;
175 }
176 }
177 printk("%ld pages of RAM\n",total);
178 printk("%ld free pages\n",free);
179 printk("%ld reserved pages\n",reserved);
180 printk("%ld pages shared\n",shared);
181 printk("%ld pages swap cached\n",cached);
182 show_buffers();
183 }
184
185 /* [numa=off] */
numa_setup(char * opt)186 __init int numa_setup(char *opt)
187 {
188 if (!strncmp(opt,"off",3))
189 numa_off = 1;
190 return 1;
191 }
192
193 __setup("numa=", numa_setup);
194
195