Linux-2.6.12-rc2
[linux-2.6/next.git] / arch / i386 / kernel / srat.c
blob7b3b27d64409381a9a557a58b6543feade5e6b9f
1 /*
2 * Some of the code in this file has been gleaned from the 64 bit
3 * discontigmem support code base.
5 * Copyright (C) 2002, IBM Corp.
7 * All rights reserved.
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
17 * NON INFRINGEMENT. See the GNU General Public License for more
18 * details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 * Send feedback to Pat Gaughen <gone@us.ibm.com>
26 #include <linux/config.h>
27 #include <linux/mm.h>
28 #include <linux/bootmem.h>
29 #include <linux/mmzone.h>
30 #include <linux/acpi.h>
31 #include <linux/nodemask.h>
32 #include <asm/srat.h>
33 #include <asm/topology.h>
36 * proximity macros and definitions
38 #define NODE_ARRAY_INDEX(x) ((x) / 8) /* 8 bits/char */
39 #define NODE_ARRAY_OFFSET(x) ((x) % 8) /* 8 bits/char */
40 #define BMAP_SET(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit))
41 #define BMAP_TEST(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit)))
42 #define MAX_PXM_DOMAINS 256 /* 1 byte and no promises about values */
43 /* bitmap length; _PXM is at most 255 */
44 #define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8)
45 static u8 pxm_bitmap[PXM_BITMAP_LEN]; /* bitmap of proximity domains */
47 #define MAX_CHUNKS_PER_NODE 4
48 #define MAXCHUNKS (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
49 struct node_memory_chunk_s {
50 unsigned long start_pfn;
51 unsigned long end_pfn;
52 u8 pxm; // proximity domain of node
53 u8 nid; // which cnode contains this chunk?
54 u8 bank; // which mem bank on this node
56 static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS];
58 static int num_memory_chunks; /* total number of memory chunks */
59 static int zholes_size_init;
60 static unsigned long zholes_size[MAX_NUMNODES * MAX_NR_ZONES];
62 extern void * boot_ioremap(unsigned long, unsigned long);
64 /* Identify CPU proximity domains */
65 static void __init parse_cpu_affinity_structure(char *p)
67 struct acpi_table_processor_affinity *cpu_affinity =
68 (struct acpi_table_processor_affinity *) p;
70 if (!cpu_affinity->flags.enabled)
71 return; /* empty entry */
73 /* mark this node as "seen" in node bitmap */
74 BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain);
76 printk("CPU 0x%02X in proximity domain 0x%02X\n",
77 cpu_affinity->apic_id, cpu_affinity->proximity_domain);
81 * Identify memory proximity domains and hot-remove capabilities.
82 * Fill node memory chunk list structure.
84 static void __init parse_memory_affinity_structure (char *sratp)
86 unsigned long long paddr, size;
87 unsigned long start_pfn, end_pfn;
88 u8 pxm;
89 struct node_memory_chunk_s *p, *q, *pend;
90 struct acpi_table_memory_affinity *memory_affinity =
91 (struct acpi_table_memory_affinity *) sratp;
93 if (!memory_affinity->flags.enabled)
94 return; /* empty entry */
96 /* mark this node as "seen" in node bitmap */
97 BMAP_SET(pxm_bitmap, memory_affinity->proximity_domain);
99 /* calculate info for memory chunk structure */
100 paddr = memory_affinity->base_addr_hi;
101 paddr = (paddr << 32) | memory_affinity->base_addr_lo;
102 size = memory_affinity->length_hi;
103 size = (size << 32) | memory_affinity->length_lo;
105 start_pfn = paddr >> PAGE_SHIFT;
106 end_pfn = (paddr + size) >> PAGE_SHIFT;
108 pxm = memory_affinity->proximity_domain;
110 if (num_memory_chunks >= MAXCHUNKS) {
111 printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
112 size/(1024*1024), paddr);
113 return;
116 /* Insertion sort based on base address */
117 pend = &node_memory_chunk[num_memory_chunks];
118 for (p = &node_memory_chunk[0]; p < pend; p++) {
119 if (start_pfn < p->start_pfn)
120 break;
122 if (p < pend) {
123 for (q = pend; q >= p; q--)
124 *(q + 1) = *q;
126 p->start_pfn = start_pfn;
127 p->end_pfn = end_pfn;
128 p->pxm = pxm;
130 num_memory_chunks++;
132 printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
133 start_pfn, end_pfn,
134 memory_affinity->memory_type,
135 memory_affinity->proximity_domain,
136 (memory_affinity->flags.hot_pluggable ?
137 "enabled and removable" : "enabled" ) );
140 #if MAX_NR_ZONES != 3
141 #error "MAX_NR_ZONES != 3, chunk_to_zone requires review"
142 #endif
143 /* Take a chunk of pages from page frame cstart to cend and count the number
144 * of pages in each zone, returned via zones[].
146 static __init void chunk_to_zones(unsigned long cstart, unsigned long cend,
147 unsigned long *zones)
149 unsigned long max_dma;
150 extern unsigned long max_low_pfn;
152 int z;
153 unsigned long rend;
155 /* FIXME: MAX_DMA_ADDRESS and max_low_pfn are trying to provide
156 * similarly scoped information and should be handled in a consistant
157 * manner.
159 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
161 /* Split the hole into the zones in which it falls. Repeatedly
162 * take the segment in which the remaining hole starts, round it
163 * to the end of that zone.
165 memset(zones, 0, MAX_NR_ZONES * sizeof(long));
166 while (cstart < cend) {
167 if (cstart < max_dma) {
168 z = ZONE_DMA;
169 rend = (cend < max_dma)? cend : max_dma;
171 } else if (cstart < max_low_pfn) {
172 z = ZONE_NORMAL;
173 rend = (cend < max_low_pfn)? cend : max_low_pfn;
175 } else {
176 z = ZONE_HIGHMEM;
177 rend = cend;
179 zones[z] += rend - cstart;
180 cstart = rend;
185 * The SRAT table always lists ascending addresses, so can always
186 * assume that the first "start" address that you see is the real
187 * start of the node, and that the current "end" address is after
188 * the previous one.
190 static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
193 * Only add present memory as told by the e820.
194 * There is no guarantee from the SRAT that the memory it
195 * enumerates is present at boot time because it represents
196 * *possible* memory hotplug areas the same as normal RAM.
198 if (memory_chunk->start_pfn >= max_pfn) {
199 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
200 memory_chunk->start_pfn, memory_chunk->end_pfn);
201 return;
203 if (memory_chunk->nid != nid)
204 return;
206 if (!node_has_online_mem(nid))
207 node_start_pfn[nid] = memory_chunk->start_pfn;
209 if (node_start_pfn[nid] > memory_chunk->start_pfn)
210 node_start_pfn[nid] = memory_chunk->start_pfn;
212 if (node_end_pfn[nid] < memory_chunk->end_pfn)
213 node_end_pfn[nid] = memory_chunk->end_pfn;
216 /* Parse the ACPI Static Resource Affinity Table */
217 static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
219 u8 *start, *end, *p;
220 int i, j, nid;
221 u8 pxm_to_nid_map[MAX_PXM_DOMAINS];/* _PXM to logical node ID map */
222 u8 nid_to_pxm_map[MAX_NUMNODES];/* logical node ID to _PXM map */
224 start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
225 p = start;
226 end = (u8 *)sratp + sratp->header.length;
228 memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */
229 memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
230 memset(zholes_size, 0, sizeof(zholes_size));
232 /* -1 in these maps means not available */
233 memset(pxm_to_nid_map, -1, sizeof(pxm_to_nid_map));
234 memset(nid_to_pxm_map, -1, sizeof(nid_to_pxm_map));
236 num_memory_chunks = 0;
237 while (p < end) {
238 switch (*p) {
239 case ACPI_SRAT_PROCESSOR_AFFINITY:
240 parse_cpu_affinity_structure(p);
241 break;
242 case ACPI_SRAT_MEMORY_AFFINITY:
243 parse_memory_affinity_structure(p);
244 break;
245 default:
246 printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
247 break;
249 p += p[1];
250 if (p[1] == 0) {
251 printk("acpi20_parse_srat: Entry length value is zero;"
252 " can't parse any further!\n");
253 break;
257 if (num_memory_chunks == 0) {
258 printk("could not finy any ACPI SRAT memory areas.\n");
259 goto out_fail;
262 /* Calculate total number of nodes in system from PXM bitmap and create
263 * a set of sequential node IDs starting at zero. (ACPI doesn't seem
264 * to specify the range of _PXM values.)
267 * MCD - we no longer HAVE to number nodes sequentially. PXM domain
268 * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
269 * 32, so we will continue numbering them in this manner until MAX_NUMNODES
270 * approaches MAX_PXM_DOMAINS for i386.
272 nodes_clear(node_online_map);
273 for (i = 0; i < MAX_PXM_DOMAINS; i++) {
274 if (BMAP_TEST(pxm_bitmap, i)) {
275 nid = num_online_nodes();
276 pxm_to_nid_map[i] = nid;
277 nid_to_pxm_map[nid] = i;
278 node_set_online(nid);
281 BUG_ON(num_online_nodes() == 0);
283 /* set cnode id in memory chunk structure */
284 for (i = 0; i < num_memory_chunks; i++)
285 node_memory_chunk[i].nid = pxm_to_nid_map[node_memory_chunk[i].pxm];
287 printk("pxm bitmap: ");
288 for (i = 0; i < sizeof(pxm_bitmap); i++) {
289 printk("%02X ", pxm_bitmap[i]);
291 printk("\n");
292 printk("Number of logical nodes in system = %d\n", num_online_nodes());
293 printk("Number of memory chunks in system = %d\n", num_memory_chunks);
295 for (j = 0; j < num_memory_chunks; j++){
296 struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
297 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
298 j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
299 node_read_chunk(chunk->nid, chunk);
302 for_each_online_node(nid) {
303 unsigned long start = node_start_pfn[nid];
304 unsigned long end = node_end_pfn[nid];
306 memory_present(nid, start, end);
307 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
309 return 1;
310 out_fail:
311 return 0;
314 int __init get_memcfg_from_srat(void)
316 struct acpi_table_header *header = NULL;
317 struct acpi_table_rsdp *rsdp = NULL;
318 struct acpi_table_rsdt *rsdt = NULL;
319 struct acpi_pointer *rsdp_address = NULL;
320 struct acpi_table_rsdt saved_rsdt;
321 int tables = 0;
322 int i = 0;
324 acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING, rsdp_address);
326 if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
327 printk("%s: assigning address to rsdp\n", __FUNCTION__);
328 rsdp = (struct acpi_table_rsdp *)
329 (u32)rsdp_address->pointer.physical;
330 } else {
331 printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
332 goto out_err;
334 if (!rsdp) {
335 printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
336 goto out_err;
339 printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
340 rsdp->oem_id);
342 if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
343 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
344 goto out_err;
347 rsdt = (struct acpi_table_rsdt *)
348 boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
350 if (!rsdt) {
351 printk(KERN_WARNING
352 "%s: ACPI: Invalid root system description tables (RSDT)\n",
353 __FUNCTION__);
354 goto out_err;
357 header = & rsdt->header;
359 if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
360 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
361 goto out_err;
365 * The number of tables is computed by taking the
366 * size of all entries (header size minus total
367 * size of RSDT) divided by the size of each entry
368 * (4-byte table pointers).
370 tables = (header->length - sizeof(struct acpi_table_header)) / 4;
372 if (!tables)
373 goto out_err;
375 memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
377 if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
378 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
379 saved_rsdt.header.length);
380 goto out_err;
383 printk("Begin SRAT table scan....\n");
385 for (i = 0; i < tables; i++) {
386 /* Map in header, then map in full table length. */
387 header = (struct acpi_table_header *)
388 boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
389 if (!header)
390 break;
391 header = (struct acpi_table_header *)
392 boot_ioremap(saved_rsdt.entry[i], header->length);
393 if (!header)
394 break;
396 if (strncmp((char *) &header->signature, "SRAT", 4))
397 continue;
399 /* we've found the srat table. don't need to look at any more tables */
400 return acpi20_parse_srat((struct acpi_table_srat *)header);
402 out_err:
403 printk("failed to get NUMA memory information from SRAT table\n");
404 return 0;
407 /* For each node run the memory list to determine whether there are
408 * any memory holes. For each hole determine which ZONE they fall
409 * into.
411 * NOTE#1: this requires knowledge of the zone boundries and so
412 * _cannot_ be performed before those are calculated in setup_memory.
414 * NOTE#2: we rely on the fact that the memory chunks are ordered by
415 * start pfn number during setup.
417 static void __init get_zholes_init(void)
419 int nid;
420 int c;
421 int first;
422 unsigned long end = 0;
424 for_each_online_node(nid) {
425 first = 1;
426 for (c = 0; c < num_memory_chunks; c++){
427 if (node_memory_chunk[c].nid == nid) {
428 if (first) {
429 end = node_memory_chunk[c].end_pfn;
430 first = 0;
432 } else {
433 /* Record any gap between this chunk
434 * and the previous chunk on this node
435 * against the zones it spans.
437 chunk_to_zones(end,
438 node_memory_chunk[c].start_pfn,
439 &zholes_size[nid * MAX_NR_ZONES]);
446 unsigned long * __init get_zholes_size(int nid)
448 if (!zholes_size_init) {
449 zholes_size_init++;
450 get_zholes_init();
452 if (nid >= MAX_NUMNODES || !node_online(nid))
453 printk("%s: nid = %d is invalid/offline. num_online_nodes = %d",
454 __FUNCTION__, nid, num_online_nodes());
455 return &zholes_size[nid * MAX_NR_ZONES];