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.
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
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>
27 #include <linux/bootmem.h>
28 #include <linux/mmzone.h>
29 #include <linux/acpi.h>
30 #include <linux/nodemask.h>
32 #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 3
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 u8 __initdata apicid_to_pxm[MAX_APICID];
60 /* Identify CPU proximity domains */
61 static void __init parse_cpu_affinity_structure(char *p)
63 struct acpi_srat_cpu_affinity *cpu_affinity =
64 (struct acpi_srat_cpu_affinity *) p;
66 if ((cpu_affinity->flags & ACPI_SRAT_CPU_ENABLED) == 0)
67 return; /* empty entry */
69 /* mark this node as "seen" in node bitmap */
70 BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain_lo);
72 apicid_to_pxm[cpu_affinity->apic_id] = cpu_affinity->proximity_domain_lo;
74 printk("CPU 0x%02X in proximity domain 0x%02X\n",
75 cpu_affinity->apic_id, cpu_affinity->proximity_domain_lo);
79 * Identify memory proximity domains and hot-remove capabilities.
80 * Fill node memory chunk list structure.
82 static void __init parse_memory_affinity_structure (char *sratp)
84 unsigned long long paddr, size;
85 unsigned long start_pfn, end_pfn;
87 struct node_memory_chunk_s *p, *q, *pend;
88 struct acpi_srat_mem_affinity *memory_affinity =
89 (struct acpi_srat_mem_affinity *) sratp;
91 if ((memory_affinity->flags & ACPI_SRAT_MEM_ENABLED) == 0)
92 return; /* empty entry */
94 pxm = memory_affinity->proximity_domain & 0xff;
96 /* mark this node as "seen" in node bitmap */
97 BMAP_SET(pxm_bitmap, pxm);
99 /* calculate info for memory chunk structure */
100 paddr = memory_affinity->base_address;
101 size = memory_affinity->length;
103 start_pfn = paddr >> PAGE_SHIFT;
104 end_pfn = (paddr + size) >> PAGE_SHIFT;
107 if (num_memory_chunks >= MAXCHUNKS) {
108 printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
109 size/(1024*1024), paddr);
113 /* Insertion sort based on base address */
114 pend = &node_memory_chunk[num_memory_chunks];
115 for (p = &node_memory_chunk[0]; p < pend; p++) {
116 if (start_pfn < p->start_pfn)
120 for (q = pend; q >= p; q--)
123 p->start_pfn = start_pfn;
124 p->end_pfn = end_pfn;
129 printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
131 memory_affinity->memory_type,
133 ((memory_affinity->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ?
134 "enabled and removable" : "enabled" ) );
138 * The SRAT table always lists ascending addresses, so can always
139 * assume that the first "start" address that you see is the real
140 * start of the node, and that the current "end" address is after
143 static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
146 * Only add present memory as told by the e820.
147 * There is no guarantee from the SRAT that the memory it
148 * enumerates is present at boot time because it represents
149 * *possible* memory hotplug areas the same as normal RAM.
151 if (memory_chunk->start_pfn >= max_pfn) {
152 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
153 memory_chunk->start_pfn, memory_chunk->end_pfn);
156 if (memory_chunk->nid != nid)
159 if (!node_has_online_mem(nid))
160 node_start_pfn[nid] = memory_chunk->start_pfn;
162 if (node_start_pfn[nid] > memory_chunk->start_pfn)
163 node_start_pfn[nid] = memory_chunk->start_pfn;
165 if (node_end_pfn[nid] < memory_chunk->end_pfn)
166 node_end_pfn[nid] = memory_chunk->end_pfn;
169 /* Parse the ACPI Static Resource Affinity Table */
170 static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
175 start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
177 end = (u8 *)sratp + sratp->header.length;
179 memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */
180 memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
182 num_memory_chunks = 0;
185 case ACPI_SRAT_TYPE_CPU_AFFINITY:
186 parse_cpu_affinity_structure(p);
188 case ACPI_SRAT_TYPE_MEMORY_AFFINITY:
189 parse_memory_affinity_structure(p);
192 printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
197 printk("acpi20_parse_srat: Entry length value is zero;"
198 " can't parse any further!\n");
203 if (num_memory_chunks == 0) {
204 printk("could not finy any ACPI SRAT memory areas.\n");
208 /* Calculate total number of nodes in system from PXM bitmap and create
209 * a set of sequential node IDs starting at zero. (ACPI doesn't seem
210 * to specify the range of _PXM values.)
213 * MCD - we no longer HAVE to number nodes sequentially. PXM domain
214 * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
215 * 32, so we will continue numbering them in this manner until MAX_NUMNODES
216 * approaches MAX_PXM_DOMAINS for i386.
218 nodes_clear(node_online_map);
219 for (i = 0; i < MAX_PXM_DOMAINS; i++) {
220 if (BMAP_TEST(pxm_bitmap, i)) {
221 int nid = acpi_map_pxm_to_node(i);
222 node_set_online(nid);
225 BUG_ON(num_online_nodes() == 0);
227 /* set cnode id in memory chunk structure */
228 for (i = 0; i < num_memory_chunks; i++)
229 node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);
231 printk("pxm bitmap: ");
232 for (i = 0; i < sizeof(pxm_bitmap); i++) {
233 printk("%02X ", pxm_bitmap[i]);
236 printk("Number of logical nodes in system = %d\n", num_online_nodes());
237 printk("Number of memory chunks in system = %d\n", num_memory_chunks);
239 for (i = 0; i < MAX_APICID; i++)
240 apicid_2_node[i] = pxm_to_node(apicid_to_pxm[i]);
242 for (j = 0; j < num_memory_chunks; j++){
243 struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
244 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
245 j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
246 node_read_chunk(chunk->nid, chunk);
247 add_active_range(chunk->nid, chunk->start_pfn, chunk->end_pfn);
250 for_each_online_node(nid) {
251 unsigned long start = node_start_pfn[nid];
252 unsigned long end = node_end_pfn[nid];
254 memory_present(nid, start, end);
255 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
262 struct acpi_static_rsdt {
263 struct acpi_table_rsdt table;
264 u32 padding[7]; /* Allow for 7 more table entries */
267 int __init get_memcfg_from_srat(void)
269 struct acpi_table_header *header = NULL;
270 struct acpi_table_rsdp *rsdp = NULL;
271 struct acpi_table_rsdt *rsdt = NULL;
272 acpi_native_uint rsdp_address = 0;
273 struct acpi_static_rsdt saved_rsdt;
277 rsdp_address = acpi_find_rsdp();
279 printk("%s: System description tables not found\n",
284 printk("%s: assigning address to rsdp\n", __FUNCTION__);
285 rsdp = (struct acpi_table_rsdp *)(u32)rsdp_address;
287 printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
291 printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
294 if (strncmp(rsdp->signature, ACPI_SIG_RSDP,strlen(ACPI_SIG_RSDP))) {
295 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
299 rsdt = (struct acpi_table_rsdt *)
300 early_ioremap(rsdp->rsdt_physical_address, sizeof(struct acpi_table_rsdt));
304 "%s: ACPI: Invalid root system description tables (RSDT)\n",
309 header = &rsdt->header;
311 if (strncmp(header->signature, ACPI_SIG_RSDT, strlen(ACPI_SIG_RSDT))) {
312 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
317 * The number of tables is computed by taking the
318 * size of all entries (header size minus total
319 * size of RSDT) divided by the size of each entry
320 * (4-byte table pointers).
322 tables = (header->length - sizeof(struct acpi_table_header)) / 4;
327 memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
329 if (saved_rsdt.table.header.length > sizeof(saved_rsdt)) {
330 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
331 saved_rsdt.table.header.length);
335 printk("Begin SRAT table scan....\n");
337 for (i = 0; i < tables; i++) {
338 /* Map in header, then map in full table length. */
339 header = (struct acpi_table_header *)
340 early_ioremap(saved_rsdt.table.table_offset_entry[i], sizeof(struct acpi_table_header));
343 header = (struct acpi_table_header *)
344 early_ioremap(saved_rsdt.table.table_offset_entry[i], header->length);
348 if (strncmp((char *) &header->signature, ACPI_SIG_SRAT, 4))
351 /* we've found the srat table. don't need to look at any more tables */
352 return acpi20_parse_srat((struct acpi_table_srat *)header);
355 remove_all_active_ranges();
356 printk("failed to get NUMA memory information from SRAT table\n");