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; | |
e4e94072 | 26 | static u8 pxm2node[256] = { [0 ... 255] = 0xff }; |
1da177e4 | 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) | |
e4e94072 AK |
33 | return -1; |
34 | /* Extend 0xff to (int)-1 */ | |
35 | return (signed char)pxm2node[pxm]; | |
69e1a33f AK |
36 | } |
37 | ||
1da177e4 LT |
38 | static __init int setup_node(int pxm) |
39 | { | |
40 | unsigned node = pxm2node[pxm]; | |
41 | if (node == 0xff) { | |
42 | if (nodes_weight(nodes_found) >= MAX_NUMNODES) | |
43 | return -1; | |
44 | node = first_unset_node(nodes_found); | |
45 | node_set(node, nodes_found); | |
46 | pxm2node[pxm] = node; | |
47 | } | |
48 | return pxm2node[pxm]; | |
49 | } | |
50 | ||
51 | static __init int conflicting_nodes(unsigned long start, unsigned long end) | |
52 | { | |
53 | int i; | |
4b6a455c | 54 | for_each_node_mask(i, nodes_parsed) { |
1da177e4 LT |
55 | struct node *nd = &nodes[i]; |
56 | if (nd->start == nd->end) | |
57 | continue; | |
58 | if (nd->end > start && nd->start < end) | |
05d1fa4b | 59 | return i; |
1da177e4 | 60 | if (nd->end == end && nd->start == start) |
05d1fa4b | 61 | return i; |
1da177e4 LT |
62 | } |
63 | return -1; | |
64 | } | |
65 | ||
66 | static __init void cutoff_node(int i, unsigned long start, unsigned long end) | |
67 | { | |
68 | struct node *nd = &nodes[i]; | |
69 | if (nd->start < start) { | |
70 | nd->start = start; | |
71 | if (nd->end < nd->start) | |
72 | nd->start = nd->end; | |
73 | } | |
74 | if (nd->end > end) { | |
1da177e4 LT |
75 | nd->end = end; |
76 | if (nd->start > nd->end) | |
77 | nd->start = nd->end; | |
78 | } | |
79 | } | |
80 | ||
81 | static __init void bad_srat(void) | |
82 | { | |
2bce2b54 | 83 | int i; |
1da177e4 LT |
84 | printk(KERN_ERR "SRAT: SRAT not used.\n"); |
85 | acpi_numa = -1; | |
2bce2b54 AK |
86 | for (i = 0; i < MAX_LOCAL_APIC; i++) |
87 | apicid_to_node[i] = NUMA_NO_NODE; | |
1da177e4 LT |
88 | } |
89 | ||
90 | static __init inline int srat_disabled(void) | |
91 | { | |
92 | return numa_off || acpi_numa < 0; | |
93 | } | |
94 | ||
1584b89c AK |
95 | /* |
96 | * A lot of BIOS fill in 10 (= no distance) everywhere. This messes | |
97 | * up the NUMA heuristics which wants the local node to have a smaller | |
98 | * distance than the others. | |
99 | * Do some quick checks here and only use the SLIT if it passes. | |
100 | */ | |
101 | static __init int slit_valid(struct acpi_table_slit *slit) | |
102 | { | |
103 | int i, j; | |
104 | int d = slit->localities; | |
105 | for (i = 0; i < d; i++) { | |
106 | for (j = 0; j < d; j++) { | |
107 | u8 val = slit->entry[d*i + j]; | |
108 | if (i == j) { | |
109 | if (val != 10) | |
110 | return 0; | |
111 | } else if (val <= 10) | |
112 | return 0; | |
113 | } | |
114 | } | |
115 | return 1; | |
116 | } | |
117 | ||
1da177e4 LT |
118 | /* Callback for SLIT parsing */ |
119 | void __init acpi_numa_slit_init(struct acpi_table_slit *slit) | |
120 | { | |
1584b89c AK |
121 | if (!slit_valid(slit)) { |
122 | printk(KERN_INFO "ACPI: SLIT table looks invalid. Not used.\n"); | |
123 | return; | |
124 | } | |
1da177e4 LT |
125 | acpi_slit = slit; |
126 | } | |
127 | ||
128 | /* Callback for Proximity Domain -> LAPIC mapping */ | |
129 | void __init | |
130 | acpi_numa_processor_affinity_init(struct acpi_table_processor_affinity *pa) | |
131 | { | |
132 | int pxm, node; | |
133 | if (srat_disabled() || pa->flags.enabled == 0) | |
134 | return; | |
135 | pxm = pa->proximity_domain; | |
136 | node = setup_node(pxm); | |
137 | if (node < 0) { | |
138 | printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm); | |
139 | bad_srat(); | |
140 | return; | |
141 | } | |
0b07e984 | 142 | apicid_to_node[pa->apic_id] = node; |
1da177e4 | 143 | acpi_numa = 1; |
0b07e984 AK |
144 | printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n", |
145 | pxm, pa->apic_id, node); | |
1da177e4 LT |
146 | } |
147 | ||
148 | /* Callback for parsing of the Proximity Domain <-> Memory Area mappings */ | |
149 | void __init | |
150 | acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma) | |
151 | { | |
152 | struct node *nd; | |
153 | unsigned long start, end; | |
154 | int node, pxm; | |
155 | int i; | |
156 | ||
157 | if (srat_disabled() || ma->flags.enabled == 0) | |
158 | return; | |
1da177e4 LT |
159 | pxm = ma->proximity_domain; |
160 | node = setup_node(pxm); | |
161 | if (node < 0) { | |
162 | printk(KERN_ERR "SRAT: Too many proximity domains.\n"); | |
163 | bad_srat(); | |
164 | return; | |
165 | } | |
166 | start = ma->base_addr_lo | ((u64)ma->base_addr_hi << 32); | |
167 | end = start + (ma->length_lo | ((u64)ma->length_hi << 32)); | |
69cb62eb AK |
168 | /* It is fine to add this area to the nodes data it will be used later*/ |
169 | if (ma->flags.hot_pluggable == 1) | |
170 | printk(KERN_INFO "SRAT: hot plug zone found %lx - %lx \n", | |
171 | start, end); | |
1da177e4 | 172 | i = conflicting_nodes(start, end); |
05d1fa4b AK |
173 | if (i == node) { |
174 | printk(KERN_WARNING | |
175 | "SRAT: Warning: PXM %d (%lx-%lx) overlaps with itself (%Lx-%Lx)\n", | |
176 | pxm, start, end, nodes[i].start, nodes[i].end); | |
177 | } else if (i >= 0) { | |
1da177e4 | 178 | printk(KERN_ERR |
05d1fa4b AK |
179 | "SRAT: PXM %d (%lx-%lx) overlaps with PXM %d (%Lx-%Lx)\n", |
180 | pxm, start, end, node_to_pxm(i), | |
181 | nodes[i].start, nodes[i].end); | |
1da177e4 LT |
182 | bad_srat(); |
183 | return; | |
184 | } | |
185 | nd = &nodes[node]; | |
186 | if (!node_test_and_set(node, nodes_parsed)) { | |
187 | nd->start = start; | |
188 | nd->end = end; | |
189 | } else { | |
190 | if (start < nd->start) | |
191 | nd->start = start; | |
192 | if (nd->end < end) | |
193 | nd->end = end; | |
194 | } | |
1da177e4 LT |
195 | printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm, |
196 | nd->start, nd->end); | |
197 | } | |
198 | ||
199 | void __init acpi_numa_arch_fixup(void) {} | |
200 | ||
201 | /* Use the information discovered above to actually set up the nodes. */ | |
202 | int __init acpi_scan_nodes(unsigned long start, unsigned long end) | |
203 | { | |
204 | int i; | |
205 | if (acpi_numa <= 0) | |
206 | return -1; | |
e58e0d03 AK |
207 | |
208 | /* First clean up the node list */ | |
209 | for_each_node_mask(i, nodes_parsed) { | |
210 | cutoff_node(i, start, end); | |
211 | if (nodes[i].start == nodes[i].end) | |
212 | node_clear(i, nodes_parsed); | |
213 | } | |
214 | ||
1da177e4 LT |
215 | memnode_shift = compute_hash_shift(nodes, nodes_weight(nodes_parsed)); |
216 | if (memnode_shift < 0) { | |
217 | printk(KERN_ERR | |
218 | "SRAT: No NUMA node hash function found. Contact maintainer\n"); | |
219 | bad_srat(); | |
220 | return -1; | |
221 | } | |
e58e0d03 AK |
222 | |
223 | /* Finally register nodes */ | |
224 | for_each_node_mask(i, nodes_parsed) | |
1da177e4 | 225 | setup_node_bootmem(i, nodes[i].start, nodes[i].end); |
1da177e4 LT |
226 | for (i = 0; i < NR_CPUS; i++) { |
227 | if (cpu_to_node[i] == NUMA_NO_NODE) | |
228 | continue; | |
229 | if (!node_isset(cpu_to_node[i], nodes_parsed)) | |
69d81fcd | 230 | numa_set_node(i, NUMA_NO_NODE); |
1da177e4 LT |
231 | } |
232 | numa_init_array(); | |
233 | return 0; | |
234 | } | |
235 | ||
05d1fa4b | 236 | static int node_to_pxm(int n) |
1da177e4 LT |
237 | { |
238 | int i; | |
239 | if (pxm2node[n] == n) | |
240 | return n; | |
241 | for (i = 0; i < 256; i++) | |
242 | if (pxm2node[i] == n) | |
243 | return i; | |
244 | return 0; | |
245 | } | |
246 | ||
247 | int __node_distance(int a, int b) | |
248 | { | |
249 | int index; | |
250 | ||
251 | if (!acpi_slit) | |
252 | return a == b ? 10 : 20; | |
253 | index = acpi_slit->localities * node_to_pxm(a); | |
254 | return acpi_slit->entry[index + node_to_pxm(b)]; | |
255 | } | |
256 | ||
257 | EXPORT_SYMBOL(__node_distance); |