[PATCH] x86: cyrix code CONFIG_PCI fix / add __initdata
[linux-2.6/linux-2.6-openrd.git] / arch / i386 / kernel / srat.c
blob989c85255dbe53a5ca33fd41cc6aae8423b7a8a2
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 /* bitmap length; _PXM is at most 255 */
43 #define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8)
44 static u8 pxm_bitmap[PXM_BITMAP_LEN]; /* bitmap of proximity domains */
46 #define MAX_CHUNKS_PER_NODE 4
47 #define MAXCHUNKS (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
48 struct node_memory_chunk_s {
49 unsigned long start_pfn;
50 unsigned long end_pfn;
51 u8 pxm; // proximity domain of node
52 u8 nid; // which cnode contains this chunk?
53 u8 bank; // which mem bank on this node
55 static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS];
57 static int num_memory_chunks; /* total number of memory chunks */
58 static int zholes_size_init;
59 static unsigned long zholes_size[MAX_NUMNODES * MAX_NR_ZONES];
61 extern void * boot_ioremap(unsigned long, unsigned long);
63 /* Identify CPU proximity domains */
64 static void __init parse_cpu_affinity_structure(char *p)
66 struct acpi_table_processor_affinity *cpu_affinity =
67 (struct acpi_table_processor_affinity *) p;
69 if (!cpu_affinity->flags.enabled)
70 return; /* empty entry */
72 /* mark this node as "seen" in node bitmap */
73 BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain);
75 printk("CPU 0x%02X in proximity domain 0x%02X\n",
76 cpu_affinity->apic_id, cpu_affinity->proximity_domain);
80 * Identify memory proximity domains and hot-remove capabilities.
81 * Fill node memory chunk list structure.
83 static void __init parse_memory_affinity_structure (char *sratp)
85 unsigned long long paddr, size;
86 unsigned long start_pfn, end_pfn;
87 u8 pxm;
88 struct node_memory_chunk_s *p, *q, *pend;
89 struct acpi_table_memory_affinity *memory_affinity =
90 (struct acpi_table_memory_affinity *) sratp;
92 if (!memory_affinity->flags.enabled)
93 return; /* empty entry */
95 /* mark this node as "seen" in node bitmap */
96 BMAP_SET(pxm_bitmap, memory_affinity->proximity_domain);
98 /* calculate info for memory chunk structure */
99 paddr = memory_affinity->base_addr_hi;
100 paddr = (paddr << 32) | memory_affinity->base_addr_lo;
101 size = memory_affinity->length_hi;
102 size = (size << 32) | memory_affinity->length_lo;
104 start_pfn = paddr >> PAGE_SHIFT;
105 end_pfn = (paddr + size) >> PAGE_SHIFT;
107 pxm = memory_affinity->proximity_domain;
109 if (num_memory_chunks >= MAXCHUNKS) {
110 printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
111 size/(1024*1024), paddr);
112 return;
115 /* Insertion sort based on base address */
116 pend = &node_memory_chunk[num_memory_chunks];
117 for (p = &node_memory_chunk[0]; p < pend; p++) {
118 if (start_pfn < p->start_pfn)
119 break;
121 if (p < pend) {
122 for (q = pend; q >= p; q--)
123 *(q + 1) = *q;
125 p->start_pfn = start_pfn;
126 p->end_pfn = end_pfn;
127 p->pxm = pxm;
129 num_memory_chunks++;
131 printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
132 start_pfn, end_pfn,
133 memory_affinity->memory_type,
134 memory_affinity->proximity_domain,
135 (memory_affinity->flags.hot_pluggable ?
136 "enabled and removable" : "enabled" ) );
139 #if MAX_NR_ZONES != 4
140 #error "MAX_NR_ZONES != 4, chunk_to_zone requires review"
141 #endif
142 /* Take a chunk of pages from page frame cstart to cend and count the number
143 * of pages in each zone, returned via zones[].
145 static __init void chunk_to_zones(unsigned long cstart, unsigned long cend,
146 unsigned long *zones)
148 unsigned long max_dma;
149 extern unsigned long max_low_pfn;
151 int z;
152 unsigned long rend;
154 /* FIXME: MAX_DMA_ADDRESS and max_low_pfn are trying to provide
155 * similarly scoped information and should be handled in a consistant
156 * manner.
158 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
160 /* Split the hole into the zones in which it falls. Repeatedly
161 * take the segment in which the remaining hole starts, round it
162 * to the end of that zone.
164 memset(zones, 0, MAX_NR_ZONES * sizeof(long));
165 while (cstart < cend) {
166 if (cstart < max_dma) {
167 z = ZONE_DMA;
168 rend = (cend < max_dma)? cend : max_dma;
170 } else if (cstart < max_low_pfn) {
171 z = ZONE_NORMAL;
172 rend = (cend < max_low_pfn)? cend : max_low_pfn;
174 } else {
175 z = ZONE_HIGHMEM;
176 rend = cend;
178 zones[z] += rend - cstart;
179 cstart = rend;
184 * The SRAT table always lists ascending addresses, so can always
185 * assume that the first "start" address that you see is the real
186 * start of the node, and that the current "end" address is after
187 * the previous one.
189 static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
192 * Only add present memory as told by the e820.
193 * There is no guarantee from the SRAT that the memory it
194 * enumerates is present at boot time because it represents
195 * *possible* memory hotplug areas the same as normal RAM.
197 if (memory_chunk->start_pfn >= max_pfn) {
198 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
199 memory_chunk->start_pfn, memory_chunk->end_pfn);
200 return;
202 if (memory_chunk->nid != nid)
203 return;
205 if (!node_has_online_mem(nid))
206 node_start_pfn[nid] = memory_chunk->start_pfn;
208 if (node_start_pfn[nid] > memory_chunk->start_pfn)
209 node_start_pfn[nid] = memory_chunk->start_pfn;
211 if (node_end_pfn[nid] < memory_chunk->end_pfn)
212 node_end_pfn[nid] = memory_chunk->end_pfn;
215 /* Parse the ACPI Static Resource Affinity Table */
216 static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
218 u8 *start, *end, *p;
219 int i, j, nid;
221 start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
222 p = start;
223 end = (u8 *)sratp + sratp->header.length;
225 memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */
226 memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
227 memset(zholes_size, 0, sizeof(zholes_size));
229 num_memory_chunks = 0;
230 while (p < end) {
231 switch (*p) {
232 case ACPI_SRAT_PROCESSOR_AFFINITY:
233 parse_cpu_affinity_structure(p);
234 break;
235 case ACPI_SRAT_MEMORY_AFFINITY:
236 parse_memory_affinity_structure(p);
237 break;
238 default:
239 printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
240 break;
242 p += p[1];
243 if (p[1] == 0) {
244 printk("acpi20_parse_srat: Entry length value is zero;"
245 " can't parse any further!\n");
246 break;
250 if (num_memory_chunks == 0) {
251 printk("could not finy any ACPI SRAT memory areas.\n");
252 goto out_fail;
255 /* Calculate total number of nodes in system from PXM bitmap and create
256 * a set of sequential node IDs starting at zero. (ACPI doesn't seem
257 * to specify the range of _PXM values.)
260 * MCD - we no longer HAVE to number nodes sequentially. PXM domain
261 * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
262 * 32, so we will continue numbering them in this manner until MAX_NUMNODES
263 * approaches MAX_PXM_DOMAINS for i386.
265 nodes_clear(node_online_map);
266 for (i = 0; i < MAX_PXM_DOMAINS; i++) {
267 if (BMAP_TEST(pxm_bitmap, i)) {
268 int nid = acpi_map_pxm_to_node(i);
269 node_set_online(nid);
272 BUG_ON(num_online_nodes() == 0);
274 /* set cnode id in memory chunk structure */
275 for (i = 0; i < num_memory_chunks; i++)
276 node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);
278 printk("pxm bitmap: ");
279 for (i = 0; i < sizeof(pxm_bitmap); i++) {
280 printk("%02X ", pxm_bitmap[i]);
282 printk("\n");
283 printk("Number of logical nodes in system = %d\n", num_online_nodes());
284 printk("Number of memory chunks in system = %d\n", num_memory_chunks);
286 for (j = 0; j < num_memory_chunks; j++){
287 struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
288 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
289 j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
290 node_read_chunk(chunk->nid, chunk);
293 for_each_online_node(nid) {
294 unsigned long start = node_start_pfn[nid];
295 unsigned long end = node_end_pfn[nid];
297 memory_present(nid, start, end);
298 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
300 return 1;
301 out_fail:
302 return 0;
305 int __init get_memcfg_from_srat(void)
307 struct acpi_table_header *header = NULL;
308 struct acpi_table_rsdp *rsdp = NULL;
309 struct acpi_table_rsdt *rsdt = NULL;
310 struct acpi_pointer *rsdp_address = NULL;
311 struct acpi_table_rsdt saved_rsdt;
312 int tables = 0;
313 int i = 0;
315 if (ACPI_FAILURE(acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING,
316 rsdp_address))) {
317 printk("%s: System description tables not found\n",
318 __FUNCTION__);
319 goto out_err;
322 if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
323 printk("%s: assigning address to rsdp\n", __FUNCTION__);
324 rsdp = (struct acpi_table_rsdp *)
325 (u32)rsdp_address->pointer.physical;
326 } else {
327 printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
328 goto out_err;
330 if (!rsdp) {
331 printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
332 goto out_err;
335 printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
336 rsdp->oem_id);
338 if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
339 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
340 goto out_err;
343 rsdt = (struct acpi_table_rsdt *)
344 boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
346 if (!rsdt) {
347 printk(KERN_WARNING
348 "%s: ACPI: Invalid root system description tables (RSDT)\n",
349 __FUNCTION__);
350 goto out_err;
353 header = & rsdt->header;
355 if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
356 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
357 goto out_err;
361 * The number of tables is computed by taking the
362 * size of all entries (header size minus total
363 * size of RSDT) divided by the size of each entry
364 * (4-byte table pointers).
366 tables = (header->length - sizeof(struct acpi_table_header)) / 4;
368 if (!tables)
369 goto out_err;
371 memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
373 if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
374 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
375 saved_rsdt.header.length);
376 goto out_err;
379 printk("Begin SRAT table scan....\n");
381 for (i = 0; i < tables; i++) {
382 /* Map in header, then map in full table length. */
383 header = (struct acpi_table_header *)
384 boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
385 if (!header)
386 break;
387 header = (struct acpi_table_header *)
388 boot_ioremap(saved_rsdt.entry[i], header->length);
389 if (!header)
390 break;
392 if (strncmp((char *) &header->signature, "SRAT", 4))
393 continue;
395 /* we've found the srat table. don't need to look at any more tables */
396 return acpi20_parse_srat((struct acpi_table_srat *)header);
398 out_err:
399 printk("failed to get NUMA memory information from SRAT table\n");
400 return 0;
403 /* For each node run the memory list to determine whether there are
404 * any memory holes. For each hole determine which ZONE they fall
405 * into.
407 * NOTE#1: this requires knowledge of the zone boundries and so
408 * _cannot_ be performed before those are calculated in setup_memory.
410 * NOTE#2: we rely on the fact that the memory chunks are ordered by
411 * start pfn number during setup.
413 static void __init get_zholes_init(void)
415 int nid;
416 int c;
417 int first;
418 unsigned long end = 0;
420 for_each_online_node(nid) {
421 first = 1;
422 for (c = 0; c < num_memory_chunks; c++){
423 if (node_memory_chunk[c].nid == nid) {
424 if (first) {
425 end = node_memory_chunk[c].end_pfn;
426 first = 0;
428 } else {
429 /* Record any gap between this chunk
430 * and the previous chunk on this node
431 * against the zones it spans.
433 chunk_to_zones(end,
434 node_memory_chunk[c].start_pfn,
435 &zholes_size[nid * MAX_NR_ZONES]);
442 unsigned long * __init get_zholes_size(int nid)
444 if (!zholes_size_init) {
445 zholes_size_init++;
446 get_zholes_init();
448 if (nid >= MAX_NUMNODES || !node_online(nid))
449 printk("%s: nid = %d is invalid/offline. num_online_nodes = %d",
450 __FUNCTION__, nid, num_online_nodes());
451 return &zholes_size[nid * MAX_NR_ZONES];