Linux 2.6.31.6
[linux/fpc-iii.git] / arch / x86 / mm / srat_64.c
blobdbb5381f7b3b58ec41f026cd65d592bc1c2871ff
1 /*
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/mm.h>
20 #include <asm/proto.h>
21 #include <asm/numa.h>
22 #include <asm/e820.h>
23 #include <asm/apic.h>
24 #include <asm/uv/uv.h>
26 int acpi_numa __initdata;
28 static struct acpi_table_slit *acpi_slit;
30 static nodemask_t nodes_parsed __initdata;
31 static nodemask_t cpu_nodes_parsed __initdata;
32 static struct bootnode nodes[MAX_NUMNODES] __initdata;
33 static struct bootnode nodes_add[MAX_NUMNODES];
35 static int num_node_memblks __initdata;
36 static struct bootnode node_memblk_range[NR_NODE_MEMBLKS] __initdata;
37 static int memblk_nodeid[NR_NODE_MEMBLKS] __initdata;
39 static __init int setup_node(int pxm)
41 return acpi_map_pxm_to_node(pxm);
44 static __init int conflicting_memblks(unsigned long start, unsigned long end)
46 int i;
47 for (i = 0; i < num_node_memblks; i++) {
48 struct bootnode *nd = &node_memblk_range[i];
49 if (nd->start == nd->end)
50 continue;
51 if (nd->end > start && nd->start < end)
52 return memblk_nodeid[i];
53 if (nd->end == end && nd->start == start)
54 return memblk_nodeid[i];
56 return -1;
59 static __init void cutoff_node(int i, unsigned long start, unsigned long end)
61 struct bootnode *nd = &nodes[i];
63 if (nd->start < start) {
64 nd->start = start;
65 if (nd->end < nd->start)
66 nd->start = nd->end;
68 if (nd->end > end) {
69 nd->end = end;
70 if (nd->start > nd->end)
71 nd->start = nd->end;
75 static __init void bad_srat(void)
77 int i;
78 printk(KERN_ERR "SRAT: SRAT not used.\n");
79 acpi_numa = -1;
80 for (i = 0; i < MAX_LOCAL_APIC; i++)
81 apicid_to_node[i] = NUMA_NO_NODE;
82 for (i = 0; i < MAX_NUMNODES; i++) {
83 nodes[i].start = nodes[i].end = 0;
84 nodes_add[i].start = nodes_add[i].end = 0;
86 remove_all_active_ranges();
89 static __init inline int srat_disabled(void)
91 return numa_off || acpi_numa < 0;
94 /* Callback for SLIT parsing */
95 void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
97 unsigned length;
98 unsigned long phys;
100 length = slit->header.length;
101 phys = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, length,
102 PAGE_SIZE);
104 if (phys == -1L)
105 panic(" Can not save slit!\n");
107 acpi_slit = __va(phys);
108 memcpy(acpi_slit, slit, length);
109 reserve_early(phys, phys + length, "ACPI SLIT");
112 /* Callback for Proximity Domain -> x2APIC mapping */
113 void __init
114 acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa)
116 int pxm, node;
117 int apic_id;
119 if (srat_disabled())
120 return;
121 if (pa->header.length < sizeof(struct acpi_srat_x2apic_cpu_affinity)) {
122 bad_srat();
123 return;
125 if ((pa->flags & ACPI_SRAT_CPU_ENABLED) == 0)
126 return;
127 pxm = pa->proximity_domain;
128 node = setup_node(pxm);
129 if (node < 0) {
130 printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm);
131 bad_srat();
132 return;
135 apic_id = pa->apic_id;
136 apicid_to_node[apic_id] = node;
137 node_set(node, cpu_nodes_parsed);
138 acpi_numa = 1;
139 printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n",
140 pxm, apic_id, node);
143 /* Callback for Proximity Domain -> LAPIC mapping */
144 void __init
145 acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *pa)
147 int pxm, node;
148 int apic_id;
150 if (srat_disabled())
151 return;
152 if (pa->header.length != sizeof(struct acpi_srat_cpu_affinity)) {
153 bad_srat();
154 return;
156 if ((pa->flags & ACPI_SRAT_CPU_ENABLED) == 0)
157 return;
158 pxm = pa->proximity_domain_lo;
159 node = setup_node(pxm);
160 if (node < 0) {
161 printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm);
162 bad_srat();
163 return;
166 if (get_uv_system_type() >= UV_X2APIC)
167 apic_id = (pa->apic_id << 8) | pa->local_sapic_eid;
168 else
169 apic_id = pa->apic_id;
170 apicid_to_node[apic_id] = node;
171 node_set(node, cpu_nodes_parsed);
172 acpi_numa = 1;
173 printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n",
174 pxm, apic_id, node);
177 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
178 static inline int save_add_info(void) {return 1;}
179 #else
180 static inline int save_add_info(void) {return 0;}
181 #endif
183 * Update nodes_add[]
184 * This code supports one contiguous hot add area per node
186 static void __init
187 update_nodes_add(int node, unsigned long start, unsigned long end)
189 unsigned long s_pfn = start >> PAGE_SHIFT;
190 unsigned long e_pfn = end >> PAGE_SHIFT;
191 int changed = 0;
192 struct bootnode *nd = &nodes_add[node];
194 /* I had some trouble with strange memory hotadd regions breaking
195 the boot. Be very strict here and reject anything unexpected.
196 If you want working memory hotadd write correct SRATs.
198 The node size check is a basic sanity check to guard against
199 mistakes */
200 if ((signed long)(end - start) < NODE_MIN_SIZE) {
201 printk(KERN_ERR "SRAT: Hotplug area too small\n");
202 return;
205 /* This check might be a bit too strict, but I'm keeping it for now. */
206 if (absent_pages_in_range(s_pfn, e_pfn) != e_pfn - s_pfn) {
207 printk(KERN_ERR
208 "SRAT: Hotplug area %lu -> %lu has existing memory\n",
209 s_pfn, e_pfn);
210 return;
213 /* Looks good */
215 if (nd->start == nd->end) {
216 nd->start = start;
217 nd->end = end;
218 changed = 1;
219 } else {
220 if (nd->start == end) {
221 nd->start = start;
222 changed = 1;
224 if (nd->end == start) {
225 nd->end = end;
226 changed = 1;
228 if (!changed)
229 printk(KERN_ERR "SRAT: Hotplug zone not continuous. Partly ignored\n");
232 if (changed)
233 printk(KERN_INFO "SRAT: hot plug zone found %Lx - %Lx\n",
234 nd->start, nd->end);
237 /* Callback for parsing of the Proximity Domain <-> Memory Area mappings */
238 void __init
239 acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
241 struct bootnode *nd, oldnode;
242 unsigned long start, end;
243 int node, pxm;
244 int i;
246 if (srat_disabled())
247 return;
248 if (ma->header.length != sizeof(struct acpi_srat_mem_affinity)) {
249 bad_srat();
250 return;
252 if ((ma->flags & ACPI_SRAT_MEM_ENABLED) == 0)
253 return;
255 if ((ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) && !save_add_info())
256 return;
257 start = ma->base_address;
258 end = start + ma->length;
259 pxm = ma->proximity_domain;
260 node = setup_node(pxm);
261 if (node < 0) {
262 printk(KERN_ERR "SRAT: Too many proximity domains.\n");
263 bad_srat();
264 return;
266 i = conflicting_memblks(start, end);
267 if (i == node) {
268 printk(KERN_WARNING
269 "SRAT: Warning: PXM %d (%lx-%lx) overlaps with itself (%Lx-%Lx)\n",
270 pxm, start, end, nodes[i].start, nodes[i].end);
271 } else if (i >= 0) {
272 printk(KERN_ERR
273 "SRAT: PXM %d (%lx-%lx) overlaps with PXM %d (%Lx-%Lx)\n",
274 pxm, start, end, node_to_pxm(i),
275 nodes[i].start, nodes[i].end);
276 bad_srat();
277 return;
279 nd = &nodes[node];
280 oldnode = *nd;
281 if (!node_test_and_set(node, nodes_parsed)) {
282 nd->start = start;
283 nd->end = end;
284 } else {
285 if (start < nd->start)
286 nd->start = start;
287 if (nd->end < end)
288 nd->end = end;
291 printk(KERN_INFO "SRAT: Node %u PXM %u %lx-%lx\n", node, pxm,
292 start, end);
293 e820_register_active_regions(node, start >> PAGE_SHIFT,
294 end >> PAGE_SHIFT);
296 if (ma->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) {
297 update_nodes_add(node, start, end);
298 /* restore nodes[node] */
299 *nd = oldnode;
300 if ((nd->start | nd->end) == 0)
301 node_clear(node, nodes_parsed);
304 node_memblk_range[num_node_memblks].start = start;
305 node_memblk_range[num_node_memblks].end = end;
306 memblk_nodeid[num_node_memblks] = node;
307 num_node_memblks++;
310 /* Sanity check to catch more bad SRATs (they are amazingly common).
311 Make sure the PXMs cover all memory. */
312 static int __init nodes_cover_memory(const struct bootnode *nodes)
314 int i;
315 unsigned long pxmram, e820ram;
317 pxmram = 0;
318 for_each_node_mask(i, nodes_parsed) {
319 unsigned long s = nodes[i].start >> PAGE_SHIFT;
320 unsigned long e = nodes[i].end >> PAGE_SHIFT;
321 pxmram += e - s;
322 pxmram -= absent_pages_in_range(s, e);
323 if ((long)pxmram < 0)
324 pxmram = 0;
327 e820ram = max_pfn - (e820_hole_size(0, max_pfn<<PAGE_SHIFT)>>PAGE_SHIFT);
328 /* We seem to lose 3 pages somewhere. Allow 1M of slack. */
329 if ((long)(e820ram - pxmram) >= (1<<(20 - PAGE_SHIFT))) {
330 printk(KERN_ERR
331 "SRAT: PXMs only cover %luMB of your %luMB e820 RAM. Not used.\n",
332 (pxmram << PAGE_SHIFT) >> 20,
333 (e820ram << PAGE_SHIFT) >> 20);
334 return 0;
336 return 1;
339 void __init acpi_numa_arch_fixup(void) {}
341 /* Use the information discovered above to actually set up the nodes. */
342 int __init acpi_scan_nodes(unsigned long start, unsigned long end)
344 int i;
346 if (acpi_numa <= 0)
347 return -1;
349 /* First clean up the node list */
350 for (i = 0; i < MAX_NUMNODES; i++)
351 cutoff_node(i, start, end);
353 if (!nodes_cover_memory(nodes)) {
354 bad_srat();
355 return -1;
358 memnode_shift = compute_hash_shift(node_memblk_range, num_node_memblks,
359 memblk_nodeid);
360 if (memnode_shift < 0) {
361 printk(KERN_ERR
362 "SRAT: No NUMA node hash function found. Contact maintainer\n");
363 bad_srat();
364 return -1;
367 /* Account for nodes with cpus and no memory */
368 nodes_or(node_possible_map, nodes_parsed, cpu_nodes_parsed);
370 /* Finally register nodes */
371 for_each_node_mask(i, node_possible_map)
372 setup_node_bootmem(i, nodes[i].start, nodes[i].end);
373 /* Try again in case setup_node_bootmem missed one due
374 to missing bootmem */
375 for_each_node_mask(i, node_possible_map)
376 if (!node_online(i))
377 setup_node_bootmem(i, nodes[i].start, nodes[i].end);
379 for (i = 0; i < nr_cpu_ids; i++) {
380 int node = early_cpu_to_node(i);
382 if (node == NUMA_NO_NODE)
383 continue;
384 if (!node_online(node))
385 numa_clear_node(i);
387 numa_init_array();
388 return 0;
391 #ifdef CONFIG_NUMA_EMU
392 static int fake_node_to_pxm_map[MAX_NUMNODES] __initdata = {
393 [0 ... MAX_NUMNODES-1] = PXM_INVAL
395 static s16 fake_apicid_to_node[MAX_LOCAL_APIC] __initdata = {
396 [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE
398 static int __init find_node_by_addr(unsigned long addr)
400 int ret = NUMA_NO_NODE;
401 int i;
403 for_each_node_mask(i, nodes_parsed) {
405 * Find the real node that this emulated node appears on. For
406 * the sake of simplicity, we only use a real node's starting
407 * address to determine which emulated node it appears on.
409 if (addr >= nodes[i].start && addr < nodes[i].end) {
410 ret = i;
411 break;
414 return ret;
418 * In NUMA emulation, we need to setup proximity domain (_PXM) to node ID
419 * mappings that respect the real ACPI topology but reflect our emulated
420 * environment. For each emulated node, we find which real node it appears on
421 * and create PXM to NID mappings for those fake nodes which mirror that
422 * locality. SLIT will now represent the correct distances between emulated
423 * nodes as a result of the real topology.
425 void __init acpi_fake_nodes(const struct bootnode *fake_nodes, int num_nodes)
427 int i, j;
429 printk(KERN_INFO "Faking PXM affinity for fake nodes on real "
430 "topology.\n");
431 for (i = 0; i < num_nodes; i++) {
432 int nid, pxm;
434 nid = find_node_by_addr(fake_nodes[i].start);
435 if (nid == NUMA_NO_NODE)
436 continue;
437 pxm = node_to_pxm(nid);
438 if (pxm == PXM_INVAL)
439 continue;
440 fake_node_to_pxm_map[i] = pxm;
442 * For each apicid_to_node mapping that exists for this real
443 * node, it must now point to the fake node ID.
445 for (j = 0; j < MAX_LOCAL_APIC; j++)
446 if (apicid_to_node[j] == nid)
447 fake_apicid_to_node[j] = i;
449 for (i = 0; i < num_nodes; i++)
450 __acpi_map_pxm_to_node(fake_node_to_pxm_map[i], i);
451 memcpy(apicid_to_node, fake_apicid_to_node, sizeof(apicid_to_node));
453 nodes_clear(nodes_parsed);
454 for (i = 0; i < num_nodes; i++)
455 if (fake_nodes[i].start != fake_nodes[i].end)
456 node_set(i, nodes_parsed);
457 WARN_ON(!nodes_cover_memory(fake_nodes));
460 static int null_slit_node_compare(int a, int b)
462 return node_to_pxm(a) == node_to_pxm(b);
464 #else
465 static int null_slit_node_compare(int a, int b)
467 return a == b;
469 #endif /* CONFIG_NUMA_EMU */
471 int __node_distance(int a, int b)
473 int index;
475 if (!acpi_slit)
476 return null_slit_node_compare(a, b) ? LOCAL_DISTANCE :
477 REMOTE_DISTANCE;
478 index = acpi_slit->locality_count * node_to_pxm(a);
479 return acpi_slit->entry[index + node_to_pxm(b)];
482 EXPORT_SYMBOL(__node_distance);
484 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) || defined(CONFIG_ACPI_HOTPLUG_MEMORY)
485 int memory_add_physaddr_to_nid(u64 start)
487 int i, ret = 0;
489 for_each_node(i)
490 if (nodes_add[i].start <= start && nodes_add[i].end > start)
491 ret = i;
493 return ret;
495 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
496 #endif