Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * ACPI 3.0 based NUMA setup | |
3 | * Copyright 2004 Andi Kleen, SuSE Labs. | |
4 | * | |
5 | * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs. | |
6 | * | |
7 | * Called from acpi_numa_init while reading the SRAT and SLIT tables. | |
8 | * Assumes all memory regions belonging to a single proximity domain | |
9 | * are in one chunk. Holes between them will be included in the node. | |
10 | */ | |
11 | ||
12 | #include <linux/kernel.h> | |
13 | #include <linux/acpi.h> | |
14 | #include <linux/mmzone.h> | |
15 | #include <linux/bitmap.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/topology.h> | |
18 | #include <asm/proto.h> | |
19 | #include <asm/numa.h> | |
20 | ||
21 | static struct acpi_table_slit *acpi_slit; | |
22 | ||
23 | static nodemask_t nodes_parsed __initdata; | |
24 | static nodemask_t nodes_found __initdata; | |
25 | static struct node nodes[MAX_NUMNODES] __initdata; | |
26 | static __u8 pxm2node[256] = { [0 ... 255] = 0xff }; | |
27 | ||
05d1fa4b AK |
28 | static int node_to_pxm(int n); |
29 | ||
69e1a33f AK |
30 | int pxm_to_node(int pxm) |
31 | { | |
32 | if ((unsigned)pxm >= 256) | |
33 | return 0; | |
34 | return pxm2node[pxm]; | |
35 | } | |
36 | ||
1da177e4 LT |
37 | static __init int setup_node(int pxm) |
38 | { | |
39 | unsigned node = pxm2node[pxm]; | |
40 | if (node == 0xff) { | |
41 | if (nodes_weight(nodes_found) >= MAX_NUMNODES) | |
42 | return -1; | |
43 | node = first_unset_node(nodes_found); | |
44 | node_set(node, nodes_found); | |
45 | pxm2node[pxm] = node; | |
46 | } | |
47 | return pxm2node[pxm]; | |
48 | } | |
49 | ||
50 | static __init int conflicting_nodes(unsigned long start, unsigned long end) | |
51 | { | |
52 | int i; | |
53 | for_each_online_node(i) { | |
54 | struct node *nd = &nodes[i]; | |
55 | if (nd->start == nd->end) | |
56 | continue; | |
57 | if (nd->end > start && nd->start < end) | |
05d1fa4b | 58 | return i; |
1da177e4 | 59 | if (nd->end == end && nd->start == start) |
05d1fa4b | 60 | return i; |
1da177e4 LT |
61 | } |
62 | return -1; | |
63 | } | |
64 | ||
65 | static __init void cutoff_node(int i, unsigned long start, unsigned long end) | |
66 | { | |
67 | struct node *nd = &nodes[i]; | |
68 | if (nd->start < start) { | |
69 | nd->start = start; | |
70 | if (nd->end < nd->start) | |
71 | nd->start = nd->end; | |
72 | } | |
73 | if (nd->end > end) { | |
74 | if (!(end & 0xfff)) | |
75 | end--; | |
76 | nd->end = end; | |
77 | if (nd->start > nd->end) | |
78 | nd->start = nd->end; | |
79 | } | |
80 | } | |
81 | ||
82 | static __init void bad_srat(void) | |
83 | { | |
2bce2b54 | 84 | int i; |
1da177e4 LT |
85 | printk(KERN_ERR "SRAT: SRAT not used.\n"); |
86 | acpi_numa = -1; | |
2bce2b54 AK |
87 | for (i = 0; i < MAX_LOCAL_APIC; i++) |
88 | apicid_to_node[i] = NUMA_NO_NODE; | |
1da177e4 LT |
89 | } |
90 | ||
91 | static __init inline int srat_disabled(void) | |
92 | { | |
93 | return numa_off || acpi_numa < 0; | |
94 | } | |
95 | ||
96 | /* Callback for SLIT parsing */ | |
97 | void __init acpi_numa_slit_init(struct acpi_table_slit *slit) | |
98 | { | |
99 | acpi_slit = slit; | |
100 | } | |
101 | ||
102 | /* Callback for Proximity Domain -> LAPIC mapping */ | |
103 | void __init | |
104 | acpi_numa_processor_affinity_init(struct acpi_table_processor_affinity *pa) | |
105 | { | |
106 | int pxm, node; | |
107 | if (srat_disabled() || pa->flags.enabled == 0) | |
108 | return; | |
109 | pxm = pa->proximity_domain; | |
110 | node = setup_node(pxm); | |
111 | if (node < 0) { | |
112 | printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm); | |
113 | bad_srat(); | |
114 | return; | |
115 | } | |
0b07e984 | 116 | apicid_to_node[pa->apic_id] = node; |
1da177e4 | 117 | acpi_numa = 1; |
0b07e984 AK |
118 | printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n", |
119 | pxm, pa->apic_id, node); | |
1da177e4 LT |
120 | } |
121 | ||
122 | /* Callback for parsing of the Proximity Domain <-> Memory Area mappings */ | |
123 | void __init | |
124 | acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma) | |
125 | { | |
126 | struct node *nd; | |
127 | unsigned long start, end; | |
128 | int node, pxm; | |
129 | int i; | |
130 | ||
131 | if (srat_disabled() || ma->flags.enabled == 0) | |
132 | return; | |
1da177e4 LT |
133 | pxm = ma->proximity_domain; |
134 | node = setup_node(pxm); | |
135 | if (node < 0) { | |
136 | printk(KERN_ERR "SRAT: Too many proximity domains.\n"); | |
137 | bad_srat(); | |
138 | return; | |
139 | } | |
140 | start = ma->base_addr_lo | ((u64)ma->base_addr_hi << 32); | |
141 | end = start + (ma->length_lo | ((u64)ma->length_hi << 32)); | |
69cb62eb AK |
142 | /* It is fine to add this area to the nodes data it will be used later*/ |
143 | if (ma->flags.hot_pluggable == 1) | |
144 | printk(KERN_INFO "SRAT: hot plug zone found %lx - %lx \n", | |
145 | start, end); | |
1da177e4 | 146 | i = conflicting_nodes(start, end); |
05d1fa4b AK |
147 | if (i == node) { |
148 | printk(KERN_WARNING | |
149 | "SRAT: Warning: PXM %d (%lx-%lx) overlaps with itself (%Lx-%Lx)\n", | |
150 | pxm, start, end, nodes[i].start, nodes[i].end); | |
151 | } else if (i >= 0) { | |
1da177e4 | 152 | printk(KERN_ERR |
05d1fa4b AK |
153 | "SRAT: PXM %d (%lx-%lx) overlaps with PXM %d (%Lx-%Lx)\n", |
154 | pxm, start, end, node_to_pxm(i), | |
155 | nodes[i].start, nodes[i].end); | |
1da177e4 LT |
156 | bad_srat(); |
157 | return; | |
158 | } | |
159 | nd = &nodes[node]; | |
160 | if (!node_test_and_set(node, nodes_parsed)) { | |
161 | nd->start = start; | |
162 | nd->end = end; | |
163 | } else { | |
164 | if (start < nd->start) | |
165 | nd->start = start; | |
166 | if (nd->end < end) | |
167 | nd->end = end; | |
168 | } | |
169 | if (!(nd->end & 0xfff)) | |
170 | nd->end--; | |
171 | printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm, | |
172 | nd->start, nd->end); | |
173 | } | |
174 | ||
175 | void __init acpi_numa_arch_fixup(void) {} | |
176 | ||
177 | /* Use the information discovered above to actually set up the nodes. */ | |
178 | int __init acpi_scan_nodes(unsigned long start, unsigned long end) | |
179 | { | |
180 | int i; | |
181 | if (acpi_numa <= 0) | |
182 | return -1; | |
e58e0d03 AK |
183 | |
184 | /* First clean up the node list */ | |
185 | for_each_node_mask(i, nodes_parsed) { | |
186 | cutoff_node(i, start, end); | |
187 | if (nodes[i].start == nodes[i].end) | |
188 | node_clear(i, nodes_parsed); | |
189 | } | |
190 | ||
1da177e4 LT |
191 | memnode_shift = compute_hash_shift(nodes, nodes_weight(nodes_parsed)); |
192 | if (memnode_shift < 0) { | |
193 | printk(KERN_ERR | |
194 | "SRAT: No NUMA node hash function found. Contact maintainer\n"); | |
195 | bad_srat(); | |
196 | return -1; | |
197 | } | |
e58e0d03 AK |
198 | |
199 | /* Finally register nodes */ | |
200 | for_each_node_mask(i, nodes_parsed) | |
1da177e4 | 201 | setup_node_bootmem(i, nodes[i].start, nodes[i].end); |
1da177e4 LT |
202 | for (i = 0; i < NR_CPUS; i++) { |
203 | if (cpu_to_node[i] == NUMA_NO_NODE) | |
204 | continue; | |
205 | if (!node_isset(cpu_to_node[i], nodes_parsed)) | |
206 | cpu_to_node[i] = NUMA_NO_NODE; | |
207 | } | |
208 | numa_init_array(); | |
209 | return 0; | |
210 | } | |
211 | ||
05d1fa4b | 212 | static int node_to_pxm(int n) |
1da177e4 LT |
213 | { |
214 | int i; | |
215 | if (pxm2node[n] == n) | |
216 | return n; | |
217 | for (i = 0; i < 256; i++) | |
218 | if (pxm2node[i] == n) | |
219 | return i; | |
220 | return 0; | |
221 | } | |
222 | ||
223 | int __node_distance(int a, int b) | |
224 | { | |
225 | int index; | |
226 | ||
227 | if (!acpi_slit) | |
228 | return a == b ? 10 : 20; | |
229 | index = acpi_slit->localities * node_to_pxm(a); | |
230 | return acpi_slit->entry[index + node_to_pxm(b)]; | |
231 | } | |
232 | ||
233 | EXPORT_SYMBOL(__node_distance); |