2 * ACPI 3.0 based NUMA setup
3 * Copyright 2004 Andi Kleen, SuSE Labs.
5 * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs.
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.
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 <linux/bootmem.h>
19 #include <linux/memblock.h>
21 #include <asm/proto.h>
25 #include <asm/uv/uv.h>
27 int acpi_numa __initdata;
29 static struct acpi_table_slit *acpi_slit;
31 static struct bootnode nodes_add[MAX_NUMNODES];
33 static __init int setup_node(int pxm)
35 return acpi_map_pxm_to_node(pxm);
38 static __init void bad_srat(void)
40 printk(KERN_ERR "SRAT: SRAT not used.\n");
42 memset(nodes_add, 0, sizeof(nodes_add));
45 static __init inline int srat_disabled(void)
50 /* Callback for SLIT parsing */
51 void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
56 length = slit->header.length;
57 phys = memblock_find_in_range(0, max_pfn_mapped<<PAGE_SHIFT, length,
60 if (phys == MEMBLOCK_ERROR)
61 panic(" Can not save slit!\n");
63 acpi_slit = __va(phys);
64 memcpy(acpi_slit, slit, length);
65 memblock_x86_reserve_range(phys, phys + length, "ACPI SLIT");
68 /* Callback for Proximity Domain -> x2APIC mapping */
70 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
77 if (pa->header.length < sizeof(struct acpi_srat_x2apic_cpu_affinity)) {
81 if ((pa->flags & ACPI_SRAT_CPU_ENABLED) == 0)
83 pxm = pa->proximity_domain;
84 node = setup_node(pxm);
86 printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm);
91 apic_id = pa->apic_id;
92 if (apic_id >= MAX_LOCAL_APIC) {
93 printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%04x -> Node %u skipped apicid that is too big\n", pxm, apic_id, node);
96 set_apicid_to_node(apic_id, node);
97 node_set(node, numa_nodes_parsed);
99 printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%04x -> Node %u\n",
103 /* Callback for Proximity Domain -> LAPIC mapping */
105 acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *pa)
112 if (pa->header.length != sizeof(struct acpi_srat_cpu_affinity)) {
116 if ((pa->flags & ACPI_SRAT_CPU_ENABLED) == 0)
118 pxm = pa->proximity_domain_lo;
119 node = setup_node(pxm);
121 printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm);
126 if (get_uv_system_type() >= UV_X2APIC)
127 apic_id = (pa->apic_id << 8) | pa->local_sapic_eid;
129 apic_id = pa->apic_id;
131 if (apic_id >= MAX_LOCAL_APIC) {
132 printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%02x -> Node %u skipped apicid that is too big\n", pxm, apic_id, node);
136 set_apicid_to_node(apic_id, node);
137 node_set(node, numa_nodes_parsed);
139 printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%02x -> Node %u\n",
143 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
144 static inline int save_add_info(void) {return 1;}
146 static inline int save_add_info(void) {return 0;}
150 * This code supports one contiguous hot add area per node
153 update_nodes_add(int node, unsigned long start, unsigned long end)
155 unsigned long s_pfn = start >> PAGE_SHIFT;
156 unsigned long e_pfn = end >> PAGE_SHIFT;
158 struct bootnode *nd = &nodes_add[node];
160 /* I had some trouble with strange memory hotadd regions breaking
161 the boot. Be very strict here and reject anything unexpected.
162 If you want working memory hotadd write correct SRATs.
164 The node size check is a basic sanity check to guard against
166 if ((signed long)(end - start) < NODE_MIN_SIZE) {
167 printk(KERN_ERR "SRAT: Hotplug area too small\n");
171 /* This check might be a bit too strict, but I'm keeping it for now. */
172 if (absent_pages_in_range(s_pfn, e_pfn) != e_pfn - s_pfn) {
174 "SRAT: Hotplug area %lu -> %lu has existing memory\n",
181 if (nd->start == nd->end) {
186 if (nd->start == end) {
190 if (nd->end == start) {
195 printk(KERN_ERR "SRAT: Hotplug zone not continuous. Partly ignored\n");
199 node_set(node, numa_nodes_parsed);
200 printk(KERN_INFO "SRAT: hot plug zone found %Lx - %Lx\n",
205 /* Callback for parsing of the Proximity Domain <-> Memory Area mappings */
207 acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
209 unsigned long start, end;
214 if (ma->header.length != sizeof(struct acpi_srat_mem_affinity)) {
218 if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0)
221 if ((ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) && !save_add_info())
223 start = ma->base_address;
224 end = start + ma->length;
225 pxm = ma->proximity_domain;
226 node = setup_node(pxm);
228 printk(KERN_ERR "SRAT: Too many proximity domains.\n");
233 if (numa_add_memblk(node, start, end) < 0) {
238 printk(KERN_INFO "SRAT: Node %u PXM %u %lx-%lx\n", node, pxm,
241 if (ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE)
242 update_nodes_add(node, start, end);
245 void __init acpi_numa_arch_fixup(void) {}
247 int __init x86_acpi_numa_init(void)
251 ret = acpi_numa_init();
254 return srat_disabled() ? -EINVAL : 0;
257 #ifdef CONFIG_NUMA_EMU
258 static int fake_node_to_pxm_map[MAX_NUMNODES] __initdata = {
259 [0 ... MAX_NUMNODES-1] = PXM_INVAL
261 static s16 fake_apicid_to_node[MAX_LOCAL_APIC] __initdata = {
262 [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE
266 * In NUMA emulation, we need to setup proximity domain (_PXM) to node ID
267 * mappings that respect the real ACPI topology but reflect our emulated
268 * environment. For each emulated node, we find which real node it appears on
269 * and create PXM to NID mappings for those fake nodes which mirror that
270 * locality. SLIT will now represent the correct distances between emulated
271 * nodes as a result of the real topology.
273 void __init acpi_fake_nodes(const struct bootnode *fake_nodes, int num_nodes)
277 for (i = 0; i < num_nodes; i++) {
280 nid = find_node_by_addr(fake_nodes[i].start);
281 if (nid == NUMA_NO_NODE)
283 pxm = node_to_pxm(nid);
284 if (pxm == PXM_INVAL)
286 fake_node_to_pxm_map[i] = pxm;
288 * For each apicid_to_node mapping that exists for this real
289 * node, it must now point to the fake node ID.
291 for (j = 0; j < MAX_LOCAL_APIC; j++)
292 if (__apicid_to_node[j] == nid &&
293 fake_apicid_to_node[j] == NUMA_NO_NODE)
294 fake_apicid_to_node[j] = i;
298 * If there are apicid-to-node mappings for physical nodes that do not
299 * have a corresponding emulated node, it should default to a guaranteed
302 for (i = 0; i < MAX_LOCAL_APIC; i++)
303 if (__apicid_to_node[i] != NUMA_NO_NODE &&
304 fake_apicid_to_node[i] == NUMA_NO_NODE)
305 fake_apicid_to_node[i] = 0;
307 for (i = 0; i < num_nodes; i++)
308 __acpi_map_pxm_to_node(fake_node_to_pxm_map[i], i);
309 memcpy(__apicid_to_node, fake_apicid_to_node, sizeof(__apicid_to_node));
311 for (i = 0; i < num_nodes; i++)
312 if (fake_nodes[i].start != fake_nodes[i].end)
313 node_set(i, numa_nodes_parsed);
316 static int null_slit_node_compare(int a, int b)
318 return node_to_pxm(a) == node_to_pxm(b);
321 static int null_slit_node_compare(int a, int b)
325 #endif /* CONFIG_NUMA_EMU */
327 int __node_distance(int a, int b)
332 return null_slit_node_compare(a, b) ? LOCAL_DISTANCE :
334 index = acpi_slit->locality_count * node_to_pxm(a);
335 return acpi_slit->entry[index + node_to_pxm(b)];
338 EXPORT_SYMBOL(__node_distance);
340 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || defined(CONFIG_ACPI_HOTPLUG_MEMORY)
341 int memory_add_physaddr_to_nid(u64 start)
346 if (nodes_add[i].start <= start && nodes_add[i].end > start)
351 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);