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>
26 #include <linux/config.h>
28 #include <linux/bootmem.h>
29 #include <linux/mmzone.h>
30 #include <linux/acpi.h>
31 #include <linux/nodemask.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 #define MAX_PXM_DOMAINS 256 /* 1 byte and no promises about values */
43 /* bitmap length; _PXM is at most 255 */
44 #define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8)
45 static u8 pxm_bitmap[PXM_BITMAP_LEN]; /* bitmap of proximity domains */
47 #define MAX_CHUNKS_PER_NODE 4
48 #define MAXCHUNKS (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
49 struct node_memory_chunk_s {
50 unsigned long start_pfn;
51 unsigned long end_pfn;
52 u8 pxm; // proximity domain of node
53 u8 nid; // which cnode contains this chunk?
54 u8 bank; // which mem bank on this node
56 static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS];
58 static int num_memory_chunks; /* total number of memory chunks */
59 static int zholes_size_init;
60 static unsigned long zholes_size[MAX_NUMNODES * MAX_NR_ZONES];
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;
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);
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)
123 for (q = pend; q >= p; q--)
126 p->start_pfn = start_pfn;
127 p->end_pfn = end_pfn;
132 printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
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"
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;
155 /* FIXME: MAX_DMA_ADDRESS and max_low_pfn are trying to provide
156 * similarly scoped information and should be handled in a consistant
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) {
169 rend = (cend < max_dma)? cend : max_dma;
171 } else if (cstart < max_low_pfn) {
173 rend = (cend < max_low_pfn)? cend : max_low_pfn;
179 zones[z] += rend - cstart;
185 * The SRAT table always lists ascending addresses, so can always
186 * assume that the first "start" address that you see is the real
187 * start of the node, and that the current "end" address is after
190 static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
193 * Only add present memory as told by the e820.
194 * There is no guarantee from the SRAT that the memory it
195 * enumerates is present at boot time because it represents
196 * *possible* memory hotplug areas the same as normal RAM.
198 if (memory_chunk->start_pfn >= max_pfn) {
199 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
200 memory_chunk->start_pfn, memory_chunk->end_pfn);
203 if (memory_chunk->nid != nid)
206 if (!node_has_online_mem(nid))
207 node_start_pfn[nid] = memory_chunk->start_pfn;
209 if (node_start_pfn[nid] > memory_chunk->start_pfn)
210 node_start_pfn[nid] = memory_chunk->start_pfn;
212 if (node_end_pfn[nid] < memory_chunk->end_pfn)
213 node_end_pfn[nid] = memory_chunk->end_pfn;
216 static u8 pxm_to_nid_map[MAX_PXM_DOMAINS];/* _PXM to logical node ID map */
218 int pxm_to_node(int pxm)
220 return pxm_to_nid_map[pxm];
223 /* Parse the ACPI Static Resource Affinity Table */
224 static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
228 u8 nid_to_pxm_map[MAX_NUMNODES];/* logical node ID to _PXM map */
230 start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
232 end = (u8 *)sratp + sratp->header.length;
234 memset(pxm_bitmap, 0, sizeof(pxm_bitmap)); /* init proximity domain bitmap */
235 memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
236 memset(zholes_size, 0, sizeof(zholes_size));
238 /* -1 in these maps means not available */
239 memset(pxm_to_nid_map, -1, sizeof(pxm_to_nid_map));
240 memset(nid_to_pxm_map, -1, sizeof(nid_to_pxm_map));
242 num_memory_chunks = 0;
245 case ACPI_SRAT_PROCESSOR_AFFINITY:
246 parse_cpu_affinity_structure(p);
248 case ACPI_SRAT_MEMORY_AFFINITY:
249 parse_memory_affinity_structure(p);
252 printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
257 printk("acpi20_parse_srat: Entry length value is zero;"
258 " can't parse any further!\n");
263 if (num_memory_chunks == 0) {
264 printk("could not finy any ACPI SRAT memory areas.\n");
268 /* Calculate total number of nodes in system from PXM bitmap and create
269 * a set of sequential node IDs starting at zero. (ACPI doesn't seem
270 * to specify the range of _PXM values.)
273 * MCD - we no longer HAVE to number nodes sequentially. PXM domain
274 * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
275 * 32, so we will continue numbering them in this manner until MAX_NUMNODES
276 * approaches MAX_PXM_DOMAINS for i386.
278 nodes_clear(node_online_map);
279 for (i = 0; i < MAX_PXM_DOMAINS; i++) {
280 if (BMAP_TEST(pxm_bitmap, i)) {
281 nid = num_online_nodes();
282 pxm_to_nid_map[i] = nid;
283 nid_to_pxm_map[nid] = i;
284 node_set_online(nid);
287 BUG_ON(num_online_nodes() == 0);
289 /* set cnode id in memory chunk structure */
290 for (i = 0; i < num_memory_chunks; i++)
291 node_memory_chunk[i].nid = pxm_to_nid_map[node_memory_chunk[i].pxm];
293 printk("pxm bitmap: ");
294 for (i = 0; i < sizeof(pxm_bitmap); i++) {
295 printk("%02X ", pxm_bitmap[i]);
298 printk("Number of logical nodes in system = %d\n", num_online_nodes());
299 printk("Number of memory chunks in system = %d\n", num_memory_chunks);
301 for (j = 0; j < num_memory_chunks; j++){
302 struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
303 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
304 j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
305 node_read_chunk(chunk->nid, chunk);
308 for_each_online_node(nid) {
309 unsigned long start = node_start_pfn[nid];
310 unsigned long end = node_end_pfn[nid];
312 memory_present(nid, start, end);
313 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
320 int __init get_memcfg_from_srat(void)
322 struct acpi_table_header *header = NULL;
323 struct acpi_table_rsdp *rsdp = NULL;
324 struct acpi_table_rsdt *rsdt = NULL;
325 struct acpi_pointer *rsdp_address = NULL;
326 struct acpi_table_rsdt saved_rsdt;
330 acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING, rsdp_address);
332 if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
333 printk("%s: assigning address to rsdp\n", __FUNCTION__);
334 rsdp = (struct acpi_table_rsdp *)
335 (u32)rsdp_address->pointer.physical;
337 printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
341 printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
345 printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
348 if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
349 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
353 rsdt = (struct acpi_table_rsdt *)
354 boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
358 "%s: ACPI: Invalid root system description tables (RSDT)\n",
363 header = & rsdt->header;
365 if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
366 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
371 * The number of tables is computed by taking the
372 * size of all entries (header size minus total
373 * size of RSDT) divided by the size of each entry
374 * (4-byte table pointers).
376 tables = (header->length - sizeof(struct acpi_table_header)) / 4;
381 memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
383 if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
384 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
385 saved_rsdt.header.length);
389 printk("Begin SRAT table scan....\n");
391 for (i = 0; i < tables; i++) {
392 /* Map in header, then map in full table length. */
393 header = (struct acpi_table_header *)
394 boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
397 header = (struct acpi_table_header *)
398 boot_ioremap(saved_rsdt.entry[i], header->length);
402 if (strncmp((char *) &header->signature, "SRAT", 4))
405 /* we've found the srat table. don't need to look at any more tables */
406 return acpi20_parse_srat((struct acpi_table_srat *)header);
409 printk("failed to get NUMA memory information from SRAT table\n");
413 /* For each node run the memory list to determine whether there are
414 * any memory holes. For each hole determine which ZONE they fall
417 * NOTE#1: this requires knowledge of the zone boundries and so
418 * _cannot_ be performed before those are calculated in setup_memory.
420 * NOTE#2: we rely on the fact that the memory chunks are ordered by
421 * start pfn number during setup.
423 static void __init get_zholes_init(void)
428 unsigned long end = 0;
430 for_each_online_node(nid) {
432 for (c = 0; c < num_memory_chunks; c++){
433 if (node_memory_chunk[c].nid == nid) {
435 end = node_memory_chunk[c].end_pfn;
439 /* Record any gap between this chunk
440 * and the previous chunk on this node
441 * against the zones it spans.
444 node_memory_chunk[c].start_pfn,
445 &zholes_size[nid * MAX_NR_ZONES]);
452 unsigned long * __init get_zholes_size(int nid)
454 if (!zholes_size_init) {
458 if (nid >= MAX_NUMNODES || !node_online(nid))
459 printk("%s: nid = %d is invalid/offline. num_online_nodes = %d",
460 __FUNCTION__, nid, num_online_nodes());
461 return &zholes_size[nid * MAX_NR_ZONES];