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>
35 * proximity macros and definitions
37 #define NODE_ARRAY_INDEX(x) ((x) / 8) /* 8 bits/char */
38 #define NODE_ARRAY_OFFSET(x) ((x) % 8) /* 8 bits/char */
39 #define BMAP_SET(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit))
40 #define BMAP_TEST(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit)))
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 3
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 */
58 extern void * boot_ioremap(unsigned long, unsigned long);
60 /* Identify CPU proximity domains */
61 static void __init parse_cpu_affinity_structure(char *p)
63 struct acpi_table_processor_affinity *cpu_affinity =
64 (struct acpi_table_processor_affinity *) p;
66 if (!cpu_affinity->flags.enabled)
67 return; /* empty entry */
69 /* mark this node as "seen" in node bitmap */
70 BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain);
72 printk("CPU 0x%02X in proximity domain 0x%02X\n",
73 cpu_affinity->apic_id, cpu_affinity->proximity_domain);
77 * Identify memory proximity domains and hot-remove capabilities.
78 * Fill node memory chunk list structure.
80 static void __init parse_memory_affinity_structure (char *sratp)
82 unsigned long long paddr, size;
83 unsigned long start_pfn, end_pfn;
85 struct node_memory_chunk_s *p, *q, *pend;
86 struct acpi_table_memory_affinity *memory_affinity =
87 (struct acpi_table_memory_affinity *) sratp;
89 if (!memory_affinity->flags.enabled)
90 return; /* empty entry */
92 /* mark this node as "seen" in node bitmap */
93 BMAP_SET(pxm_bitmap, memory_affinity->proximity_domain);
95 /* calculate info for memory chunk structure */
96 paddr = memory_affinity->base_addr_hi;
97 paddr = (paddr << 32) | memory_affinity->base_addr_lo;
98 size = memory_affinity->length_hi;
99 size = (size << 32) | memory_affinity->length_lo;
101 start_pfn = paddr >> PAGE_SHIFT;
102 end_pfn = (paddr + size) >> PAGE_SHIFT;
104 pxm = memory_affinity->proximity_domain;
106 if (num_memory_chunks >= MAXCHUNKS) {
107 printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
108 size/(1024*1024), paddr);
112 /* Insertion sort based on base address */
113 pend = &node_memory_chunk[num_memory_chunks];
114 for (p = &node_memory_chunk[0]; p < pend; p++) {
115 if (start_pfn < p->start_pfn)
119 for (q = pend; q >= p; q--)
122 p->start_pfn = start_pfn;
123 p->end_pfn = end_pfn;
128 printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
130 memory_affinity->memory_type,
131 memory_affinity->proximity_domain,
132 (memory_affinity->flags.hot_pluggable ?
133 "enabled and removable" : "enabled" ) );
137 * The SRAT table always lists ascending addresses, so can always
138 * assume that the first "start" address that you see is the real
139 * start of the node, and that the current "end" address is after
142 static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
145 * Only add present memory as told by the e820.
146 * There is no guarantee from the SRAT that the memory it
147 * enumerates is present at boot time because it represents
148 * *possible* memory hotplug areas the same as normal RAM.
150 if (memory_chunk->start_pfn >= max_pfn) {
151 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
152 memory_chunk->start_pfn, memory_chunk->end_pfn);
155 if (memory_chunk->nid != nid)
158 if (!node_has_online_mem(nid))
159 node_start_pfn[nid] = memory_chunk->start_pfn;
161 if (node_start_pfn[nid] > memory_chunk->start_pfn)
162 node_start_pfn[nid] = memory_chunk->start_pfn;
164 if (node_end_pfn[nid] < memory_chunk->end_pfn)
165 node_end_pfn[nid] = memory_chunk->end_pfn;
168 /* Parse the ACPI Static Resource Affinity Table */
169 static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
174 start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
176 end = (u8 *)sratp + sratp->header.length;
178 memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */
179 memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
181 num_memory_chunks = 0;
184 case ACPI_SRAT_PROCESSOR_AFFINITY:
185 parse_cpu_affinity_structure(p);
187 case ACPI_SRAT_MEMORY_AFFINITY:
188 parse_memory_affinity_structure(p);
191 printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
196 printk("acpi20_parse_srat: Entry length value is zero;"
197 " can't parse any further!\n");
202 if (num_memory_chunks == 0) {
203 printk("could not finy any ACPI SRAT memory areas.\n");
207 /* Calculate total number of nodes in system from PXM bitmap and create
208 * a set of sequential node IDs starting at zero. (ACPI doesn't seem
209 * to specify the range of _PXM values.)
212 * MCD - we no longer HAVE to number nodes sequentially. PXM domain
213 * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
214 * 32, so we will continue numbering them in this manner until MAX_NUMNODES
215 * approaches MAX_PXM_DOMAINS for i386.
217 nodes_clear(node_online_map);
218 for (i = 0; i < MAX_PXM_DOMAINS; i++) {
219 if (BMAP_TEST(pxm_bitmap, i)) {
220 int nid = acpi_map_pxm_to_node(i);
221 node_set_online(nid);
224 BUG_ON(num_online_nodes() == 0);
226 /* set cnode id in memory chunk structure */
227 for (i = 0; i < num_memory_chunks; i++)
228 node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);
230 printk("pxm bitmap: ");
231 for (i = 0; i < sizeof(pxm_bitmap); i++) {
232 printk("%02X ", pxm_bitmap[i]);
235 printk("Number of logical nodes in system = %d\n", num_online_nodes());
236 printk("Number of memory chunks in system = %d\n", num_memory_chunks);
238 for (j = 0; j < num_memory_chunks; j++){
239 struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
240 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
241 j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
242 node_read_chunk(chunk->nid, chunk);
243 add_active_range(chunk->nid, chunk->start_pfn, chunk->end_pfn);
246 for_each_online_node(nid) {
247 unsigned long start = node_start_pfn[nid];
248 unsigned long end = node_end_pfn[nid];
250 memory_present(nid, start, end);
251 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
258 int __init get_memcfg_from_srat(void)
260 struct acpi_table_header *header = NULL;
261 struct acpi_table_rsdp *rsdp = NULL;
262 struct acpi_table_rsdt *rsdt = NULL;
263 struct acpi_pointer *rsdp_address = NULL;
264 struct acpi_table_rsdt saved_rsdt;
268 if (ACPI_FAILURE(acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING,
270 printk("%s: System description tables not found\n",
275 if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
276 printk("%s: assigning address to rsdp\n", __FUNCTION__);
277 rsdp = (struct acpi_table_rsdp *)
278 (u32)rsdp_address->pointer.physical;
280 printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
284 printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
288 printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
291 if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
292 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
296 rsdt = (struct acpi_table_rsdt *)
297 boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
301 "%s: ACPI: Invalid root system description tables (RSDT)\n",
306 header = & rsdt->header;
308 if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
309 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
314 * The number of tables is computed by taking the
315 * size of all entries (header size minus total
316 * size of RSDT) divided by the size of each entry
317 * (4-byte table pointers).
319 tables = (header->length - sizeof(struct acpi_table_header)) / 4;
324 memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
326 if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
327 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
328 saved_rsdt.header.length);
332 printk("Begin SRAT table scan....\n");
334 for (i = 0; i < tables; i++) {
335 /* Map in header, then map in full table length. */
336 header = (struct acpi_table_header *)
337 boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
340 header = (struct acpi_table_header *)
341 boot_ioremap(saved_rsdt.entry[i], header->length);
345 if (strncmp((char *) &header->signature, "SRAT", 4))
348 /* we've found the srat table. don't need to look at any more tables */
349 return acpi20_parse_srat((struct acpi_table_srat *)header);
352 remove_all_active_ranges();
353 printk("failed to get NUMA memory information from SRAT table\n");