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 */
57 static int zholes_size_init;
58 static unsigned long zholes_size[MAX_NUMNODES * MAX_NR_ZONES];
60 extern void * boot_ioremap(unsigned long, unsigned long);
62 /* Identify CPU proximity domains */
63 static void __init parse_cpu_affinity_structure(char *p)
65 struct acpi_table_processor_affinity *cpu_affinity =
66 (struct acpi_table_processor_affinity *) p;
68 if (!cpu_affinity->flags.enabled)
69 return; /* empty entry */
71 /* mark this node as "seen" in node bitmap */
72 BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain);
74 printk("CPU 0x%02X in proximity domain 0x%02X\n",
75 cpu_affinity->apic_id, cpu_affinity->proximity_domain);
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_table_memory_affinity *memory_affinity =
89 (struct acpi_table_memory_affinity *) sratp;
91 if (!memory_affinity->flags.enabled)
92 return; /* empty entry */
94 /* mark this node as "seen" in node bitmap */
95 BMAP_SET(pxm_bitmap, memory_affinity->proximity_domain);
97 /* calculate info for memory chunk structure */
98 paddr = memory_affinity->base_addr_hi;
99 paddr = (paddr << 32) | memory_affinity->base_addr_lo;
100 size = memory_affinity->length_hi;
101 size = (size << 32) | memory_affinity->length_lo;
103 start_pfn = paddr >> PAGE_SHIFT;
104 end_pfn = (paddr + size) >> PAGE_SHIFT;
106 pxm = memory_affinity->proximity_domain;
108 if (num_memory_chunks >= MAXCHUNKS) {
109 printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
110 size/(1024*1024), paddr);
114 /* Insertion sort based on base address */
115 pend = &node_memory_chunk[num_memory_chunks];
116 for (p = &node_memory_chunk[0]; p < pend; p++) {
117 if (start_pfn < p->start_pfn)
121 for (q = pend; q >= p; q--)
124 p->start_pfn = start_pfn;
125 p->end_pfn = end_pfn;
130 printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
132 memory_affinity->memory_type,
133 memory_affinity->proximity_domain,
134 (memory_affinity->flags.hot_pluggable ?
135 "enabled and removable" : "enabled" ) );
138 /* Take a chunk of pages from page frame cstart to cend and count the number
139 * of pages in each zone, returned via zones[].
141 static __init void chunk_to_zones(unsigned long cstart, unsigned long cend,
142 unsigned long *zones)
144 unsigned long max_dma;
145 extern unsigned long max_low_pfn;
150 /* FIXME: MAX_DMA_ADDRESS and max_low_pfn are trying to provide
151 * similarly scoped information and should be handled in a consistant
154 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
156 /* Split the hole into the zones in which it falls. Repeatedly
157 * take the segment in which the remaining hole starts, round it
158 * to the end of that zone.
160 memset(zones, 0, MAX_NR_ZONES * sizeof(long));
161 while (cstart < cend) {
162 if (cstart < max_dma) {
164 rend = (cend < max_dma)? cend : max_dma;
166 } else if (cstart < max_low_pfn) {
168 rend = (cend < max_low_pfn)? cend : max_low_pfn;
174 zones[z] += rend - cstart;
180 * The SRAT table always lists ascending addresses, so can always
181 * assume that the first "start" address that you see is the real
182 * start of the node, and that the current "end" address is after
185 static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
188 * Only add present memory as told by the e820.
189 * There is no guarantee from the SRAT that the memory it
190 * enumerates is present at boot time because it represents
191 * *possible* memory hotplug areas the same as normal RAM.
193 if (memory_chunk->start_pfn >= max_pfn) {
194 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
195 memory_chunk->start_pfn, memory_chunk->end_pfn);
198 if (memory_chunk->nid != nid)
201 if (!node_has_online_mem(nid))
202 node_start_pfn[nid] = memory_chunk->start_pfn;
204 if (node_start_pfn[nid] > memory_chunk->start_pfn)
205 node_start_pfn[nid] = memory_chunk->start_pfn;
207 if (node_end_pfn[nid] < memory_chunk->end_pfn)
208 node_end_pfn[nid] = memory_chunk->end_pfn;
211 /* Parse the ACPI Static Resource Affinity Table */
212 static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
217 start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
219 end = (u8 *)sratp + sratp->header.length;
221 memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */
222 memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
223 memset(zholes_size, 0, sizeof(zholes_size));
225 num_memory_chunks = 0;
228 case ACPI_SRAT_PROCESSOR_AFFINITY:
229 parse_cpu_affinity_structure(p);
231 case ACPI_SRAT_MEMORY_AFFINITY:
232 parse_memory_affinity_structure(p);
235 printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
240 printk("acpi20_parse_srat: Entry length value is zero;"
241 " can't parse any further!\n");
246 if (num_memory_chunks == 0) {
247 printk("could not finy any ACPI SRAT memory areas.\n");
251 /* Calculate total number of nodes in system from PXM bitmap and create
252 * a set of sequential node IDs starting at zero. (ACPI doesn't seem
253 * to specify the range of _PXM values.)
256 * MCD - we no longer HAVE to number nodes sequentially. PXM domain
257 * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
258 * 32, so we will continue numbering them in this manner until MAX_NUMNODES
259 * approaches MAX_PXM_DOMAINS for i386.
261 nodes_clear(node_online_map);
262 for (i = 0; i < MAX_PXM_DOMAINS; i++) {
263 if (BMAP_TEST(pxm_bitmap, i)) {
264 int nid = acpi_map_pxm_to_node(i);
265 node_set_online(nid);
268 BUG_ON(num_online_nodes() == 0);
270 /* set cnode id in memory chunk structure */
271 for (i = 0; i < num_memory_chunks; i++)
272 node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);
274 printk("pxm bitmap: ");
275 for (i = 0; i < sizeof(pxm_bitmap); i++) {
276 printk("%02X ", pxm_bitmap[i]);
279 printk("Number of logical nodes in system = %d\n", num_online_nodes());
280 printk("Number of memory chunks in system = %d\n", num_memory_chunks);
282 for (j = 0; j < num_memory_chunks; j++){
283 struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
284 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
285 j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
286 node_read_chunk(chunk->nid, chunk);
289 for_each_online_node(nid) {
290 unsigned long start = node_start_pfn[nid];
291 unsigned long end = node_end_pfn[nid];
293 memory_present(nid, start, end);
294 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
301 int __init get_memcfg_from_srat(void)
303 struct acpi_table_header *header = NULL;
304 struct acpi_table_rsdp *rsdp = NULL;
305 struct acpi_table_rsdt *rsdt = NULL;
306 struct acpi_pointer *rsdp_address = NULL;
307 struct acpi_table_rsdt saved_rsdt;
311 if (ACPI_FAILURE(acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING,
313 printk("%s: System description tables not found\n",
318 if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
319 printk("%s: assigning address to rsdp\n", __FUNCTION__);
320 rsdp = (struct acpi_table_rsdp *)
321 (u32)rsdp_address->pointer.physical;
323 printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
327 printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
331 printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
334 if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
335 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
339 rsdt = (struct acpi_table_rsdt *)
340 boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
344 "%s: ACPI: Invalid root system description tables (RSDT)\n",
349 header = & rsdt->header;
351 if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
352 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
357 * The number of tables is computed by taking the
358 * size of all entries (header size minus total
359 * size of RSDT) divided by the size of each entry
360 * (4-byte table pointers).
362 tables = (header->length - sizeof(struct acpi_table_header)) / 4;
367 memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
369 if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
370 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
371 saved_rsdt.header.length);
375 printk("Begin SRAT table scan....\n");
377 for (i = 0; i < tables; i++) {
378 /* Map in header, then map in full table length. */
379 header = (struct acpi_table_header *)
380 boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
383 header = (struct acpi_table_header *)
384 boot_ioremap(saved_rsdt.entry[i], header->length);
388 if (strncmp((char *) &header->signature, "SRAT", 4))
391 /* we've found the srat table. don't need to look at any more tables */
392 return acpi20_parse_srat((struct acpi_table_srat *)header);
395 printk("failed to get NUMA memory information from SRAT table\n");
399 /* For each node run the memory list to determine whether there are
400 * any memory holes. For each hole determine which ZONE they fall
403 * NOTE#1: this requires knowledge of the zone boundries and so
404 * _cannot_ be performed before those are calculated in setup_memory.
406 * NOTE#2: we rely on the fact that the memory chunks are ordered by
407 * start pfn number during setup.
409 static void __init get_zholes_init(void)
414 unsigned long end = 0;
416 for_each_online_node(nid) {
418 for (c = 0; c < num_memory_chunks; c++){
419 if (node_memory_chunk[c].nid == nid) {
421 end = node_memory_chunk[c].end_pfn;
425 /* Record any gap between this chunk
426 * and the previous chunk on this node
427 * against the zones it spans.
430 node_memory_chunk[c].start_pfn,
431 &zholes_size[nid * MAX_NR_ZONES]);
438 unsigned long * __init get_zholes_size(int nid)
440 if (!zholes_size_init) {
444 if (nid >= MAX_NUMNODES || !node_online(nid))
445 printk("%s: nid = %d is invalid/offline. num_online_nodes = %d",
446 __FUNCTION__, nid, num_online_nodes());
447 return &zholes_size[nid * MAX_NR_ZONES];