initial commit with v2.6.9
[linux-2.6.9-moxart.git] / arch / i386 / kernel / srat.c
blobe80cd8e417e085856d58f810e68b97e38409ef33
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 <asm/srat.h>
34 * proximity macros and definitions
36 #define NODE_ARRAY_INDEX(x) ((x) / 8) /* 8 bits/char */
37 #define NODE_ARRAY_OFFSET(x) ((x) % 8) /* 8 bits/char */
38 #define BMAP_SET(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit))
39 #define BMAP_TEST(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit)))
40 #define MAX_PXM_DOMAINS 256 /* 1 byte and no promises about values */
41 /* bitmap length; _PXM is at most 255 */
42 #define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8)
43 static u8 pxm_bitmap[PXM_BITMAP_LEN]; /* bitmap of proximity domains */
45 #define MAX_CHUNKS_PER_NODE 4
46 #define MAXCHUNKS (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
47 struct node_memory_chunk_s {
48 unsigned long start_pfn;
49 unsigned long end_pfn;
50 u8 pxm; // proximity domain of node
51 u8 nid; // which cnode contains this chunk?
52 u8 bank; // which mem bank on this node
54 static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS];
56 static int num_memory_chunks; /* total number of memory chunks */
57 static int zholes_size_init;
58 static unsigned long zholes_size[MAX_NUMNODES * MAX_NR_ZONES];
60 extern unsigned long node_start_pfn[], node_end_pfn[];
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;
184 /* Parse the ACPI Static Resource Affinity Table */
185 static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
187 u8 *start, *end, *p;
188 int i, j, nid;
189 u8 pxm_to_nid_map[MAX_PXM_DOMAINS];/* _PXM to logical node ID map */
190 u8 nid_to_pxm_map[MAX_NUMNODES];/* logical node ID to _PXM map */
192 start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
193 p = start;
194 end = (u8 *)sratp + sratp->header.length;
196 memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */
197 memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
198 memset(zholes_size, 0, sizeof(zholes_size));
200 /* -1 in these maps means not available */
201 memset(pxm_to_nid_map, -1, sizeof(pxm_to_nid_map));
202 memset(nid_to_pxm_map, -1, sizeof(nid_to_pxm_map));
204 num_memory_chunks = 0;
205 while (p < end) {
206 switch (*p) {
207 case ACPI_SRAT_PROCESSOR_AFFINITY:
208 parse_cpu_affinity_structure(p);
209 break;
210 case ACPI_SRAT_MEMORY_AFFINITY:
211 parse_memory_affinity_structure(p);
212 break;
213 default:
214 printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
215 break;
217 p += p[1];
218 if (p[1] == 0) {
219 printk("acpi20_parse_srat: Entry length value is zero;"
220 " can't parse any further!\n");
221 break;
225 if (num_memory_chunks == 0) {
226 printk("could not finy any ACPI SRAT memory areas.\n");
227 goto out_fail;
230 /* Calculate total number of nodes in system from PXM bitmap and create
231 * a set of sequential node IDs starting at zero. (ACPI doesn't seem
232 * to specify the range of _PXM values.)
234 numnodes = 0; /* init total nodes in system */
235 for (i = 0; i < MAX_PXM_DOMAINS; i++) {
236 if (BMAP_TEST(pxm_bitmap, i)) {
237 pxm_to_nid_map[i] = numnodes;
238 nid_to_pxm_map[numnodes] = i;
239 node_set_online(numnodes);
240 ++numnodes;
244 if (numnodes == 0)
245 BUG();
247 /* set cnode id in memory chunk structure */
248 for (i = 0; i < num_memory_chunks; i++)
249 node_memory_chunk[i].nid = pxm_to_nid_map[node_memory_chunk[i].pxm];
251 printk("pxm bitmap: ");
252 for (i = 0; i < sizeof(pxm_bitmap); i++) {
253 printk("%02X ", pxm_bitmap[i]);
255 printk("\n");
256 printk("Number of logical nodes in system = %d\n", numnodes);
257 printk("Number of memory chunks in system = %d\n", num_memory_chunks);
259 for (j = 0; j < num_memory_chunks; j++){
260 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
261 j, node_memory_chunk[j].nid,
262 node_memory_chunk[j].start_pfn,
263 node_memory_chunk[j].end_pfn);
266 /*calculate node_start_pfn/node_end_pfn arrays*/
267 for (nid = 0; nid < numnodes; nid++) {
268 int been_here_before = 0;
270 for (j = 0; j < num_memory_chunks; j++){
271 if (node_memory_chunk[j].nid == nid) {
272 if (been_here_before == 0) {
273 node_start_pfn[nid] = node_memory_chunk[j].start_pfn;
274 node_end_pfn[nid] = node_memory_chunk[j].end_pfn;
275 been_here_before = 1;
276 } else { /* We've found another chunk of memory for the node */
277 if (node_start_pfn[nid] < node_memory_chunk[j].start_pfn) {
278 node_end_pfn[nid] = node_memory_chunk[j].end_pfn;
284 return 1;
285 out_fail:
286 return 0;
289 int __init get_memcfg_from_srat(void)
291 struct acpi_table_header *header = NULL;
292 struct acpi_table_rsdp *rsdp = NULL;
293 struct acpi_table_rsdt *rsdt = NULL;
294 struct acpi_pointer *rsdp_address = NULL;
295 struct acpi_table_rsdt saved_rsdt;
296 int tables = 0;
297 int i = 0;
299 acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING, rsdp_address);
301 if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
302 printk("%s: assigning address to rsdp\n", __FUNCTION__);
303 rsdp = (struct acpi_table_rsdp *)
304 (u32)rsdp_address->pointer.physical;
305 } else {
306 printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
307 goto out_err;
309 if (!rsdp) {
310 printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
311 goto out_err;
314 printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
315 rsdp->oem_id);
317 if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
318 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
319 goto out_err;
322 rsdt = (struct acpi_table_rsdt *)
323 boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
325 if (!rsdt) {
326 printk(KERN_WARNING
327 "%s: ACPI: Invalid root system description tables (RSDT)\n",
328 __FUNCTION__);
329 goto out_err;
332 header = & rsdt->header;
334 if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
335 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
336 goto out_err;
340 * The number of tables is computed by taking the
341 * size of all entries (header size minus total
342 * size of RSDT) divided by the size of each entry
343 * (4-byte table pointers).
345 tables = (header->length - sizeof(struct acpi_table_header)) / 4;
347 if (!tables)
348 goto out_err;
350 memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
352 if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
353 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
354 saved_rsdt.header.length);
355 goto out_err;
358 printk("Begin SRAT table scan....\n");
360 for (i = 0; i < tables; i++) {
361 /* Map in header, then map in full table length. */
362 header = (struct acpi_table_header *)
363 boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
364 if (!header)
365 break;
366 header = (struct acpi_table_header *)
367 boot_ioremap(saved_rsdt.entry[i], header->length);
368 if (!header)
369 break;
371 if (strncmp((char *) &header->signature, "SRAT", 4))
372 continue;
374 /* we've found the srat table. don't need to look at any more tables */
375 return acpi20_parse_srat((struct acpi_table_srat *)header);
377 out_err:
378 printk("failed to get NUMA memory information from SRAT table\n");
379 return 0;
382 /* For each node run the memory list to determine whether there are
383 * any memory holes. For each hole determine which ZONE they fall
384 * into.
386 * NOTE#1: this requires knowledge of the zone boundries and so
387 * _cannot_ be performed before those are calculated in setup_memory.
389 * NOTE#2: we rely on the fact that the memory chunks are ordered by
390 * start pfn number during setup.
392 static void __init get_zholes_init(void)
394 int nid;
395 int c;
396 int first;
397 unsigned long end = 0;
399 for (nid = 0; nid < numnodes; nid++) {
400 first = 1;
401 for (c = 0; c < num_memory_chunks; c++){
402 if (node_memory_chunk[c].nid == nid) {
403 if (first) {
404 end = node_memory_chunk[c].end_pfn;
405 first = 0;
407 } else {
408 /* Record any gap between this chunk
409 * and the previous chunk on this node
410 * against the zones it spans.
412 chunk_to_zones(end,
413 node_memory_chunk[c].start_pfn,
414 &zholes_size[nid * MAX_NR_ZONES]);
421 unsigned long * __init get_zholes_size(int nid)
423 if (!zholes_size_init) {
424 zholes_size_init++;
425 get_zholes_init();
427 if((nid >= numnodes) | (nid >= MAX_NUMNODES))
428 printk("%s: nid = %d is invalid. numnodes = %d",
429 __FUNCTION__, nid, numnodes);
430 return &zholes_size[nid * MAX_NR_ZONES];