blob: 7c392dc553b89659ce6041f09d5aae62d95ab50a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Written by: Patricia Gaughen <gone@us.ibm.com>, IBM Corporation
3 * August 2002: added remote node KVA remap - Martin J. Bligh
4 *
5 * Copyright (C) 2002, IBM Corp.
6 *
7 * All rights reserved.
8 *
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.
13 *
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.
19 *
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.
23 */
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/mm.h>
26#include <linux/bootmem.h>
27#include <linux/mmzone.h>
28#include <linux/highmem.h>
29#include <linux/initrd.h>
30#include <linux/nodemask.h>
Alexey Dobriyan129f6942005-06-23 00:08:33 -070031#include <linux/module.h>
Eric W. Biederman1bc3b912005-06-25 14:58:01 -070032#include <linux/kexec.h>
Dave Hansen22a98352006-03-27 01:16:04 -080033#include <linux/pfn.h>
Eric W. Biederman1bc3b912005-06-25 14:58:01 -070034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/e820.h>
36#include <asm/setup.h>
37#include <asm/mmzone.h>
38#include <bios_ebda.h>
39
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070040struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
Alexey Dobriyan129f6942005-06-23 00:08:33 -070041EXPORT_SYMBOL(node_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042bootmem_data_t node0_bdata;
43
44/*
Adrian Bunkd254c8f2006-06-30 18:29:51 +020045 * numa interface - we expect the numa architecture specific code to have
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 * populated the following initialisation.
47 *
48 * 1) node_online_map - the map of all nodes configured (online) in the system
Andy Whitcroft05b79bd2005-06-23 00:07:57 -070049 * 2) node_start_pfn - the starting page frame number for a node
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 * 3) node_end_pfn - the ending page fram number for a node
51 */
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070052unsigned long node_start_pfn[MAX_NUMNODES] __read_mostly;
53unsigned long node_end_pfn[MAX_NUMNODES] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Andy Whitcroft05b79bd2005-06-23 00:07:57 -070055
56#ifdef CONFIG_DISCONTIGMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070057/*
Andy Whitcroft05b79bd2005-06-23 00:07:57 -070058 * 4) physnode_map - the mapping between a pfn and owning node
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 * physnode_map keeps track of the physical memory layout of a generic
60 * numa node on a 256Mb break (each element of the array will
61 * represent 256Mb of memory and will be marked by the node id. so,
62 * if the first gig is on node 0, and the second gig is on node 1
63 * physnode_map will contain:
64 *
65 * physnode_map[0-3] = 0;
66 * physnode_map[4-7] = 1;
67 * physnode_map[8- ] = -1;
68 */
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070069s8 physnode_map[MAX_ELEMENTS] __read_mostly = { [0 ... (MAX_ELEMENTS - 1)] = -1};
Alexey Dobriyan129f6942005-06-23 00:08:33 -070070EXPORT_SYMBOL(physnode_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
72void memory_present(int nid, unsigned long start, unsigned long end)
73{
74 unsigned long pfn;
75
76 printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
77 nid, start, end);
78 printk(KERN_DEBUG " Setting physnode_map array to node %d for pfns:\n", nid);
79 printk(KERN_DEBUG " ");
80 for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
81 physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
82 printk("%ld ", pfn);
83 }
84 printk("\n");
85}
86
87unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
88 unsigned long end_pfn)
89{
90 unsigned long nr_pages = end_pfn - start_pfn;
91
92 if (!nr_pages)
93 return 0;
94
95 return (nr_pages + 1) * sizeof(struct page);
96}
Andy Whitcroft05b79bd2005-06-23 00:07:57 -070097#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
99extern unsigned long find_max_low_pfn(void);
100extern void find_max_pfn(void);
Dave Hansen05039b92005-10-29 18:16:57 -0700101extern void add_one_highpage_init(struct page *, int, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
103extern struct e820map e820;
104extern unsigned long init_pg_tables_end;
105extern unsigned long highend_pfn, highstart_pfn;
106extern unsigned long max_low_pfn;
107extern unsigned long totalram_pages;
108extern unsigned long totalhigh_pages;
109
110#define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
111
112unsigned long node_remap_start_pfn[MAX_NUMNODES];
113unsigned long node_remap_size[MAX_NUMNODES];
114unsigned long node_remap_offset[MAX_NUMNODES];
115void *node_remap_start_vaddr[MAX_NUMNODES];
116void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
117
Dave Hansen6f167ec2005-06-23 00:07:39 -0700118void *node_remap_end_vaddr[MAX_NUMNODES];
119void *node_remap_alloc_vaddr[MAX_NUMNODES];
120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121/*
122 * FLAT - support for basic PC memory model with discontig enabled, essentially
123 * a single node with all available processors in it with a flat
124 * memory map.
125 */
126int __init get_memcfg_numa_flat(void)
127{
128 printk("NUMA - single node, flat memory mode\n");
129
130 /* Run the memory configuration and find the top of memory. */
131 find_max_pfn();
132 node_start_pfn[0] = 0;
133 node_end_pfn[0] = max_pfn;
134 memory_present(0, 0, max_pfn);
135
136 /* Indicate there is one node available. */
137 nodes_clear(node_online_map);
138 node_set_online(0);
139 return 1;
140}
141
142/*
143 * Find the highest page frame number we have available for the node
144 */
145static void __init find_max_pfn_node(int nid)
146{
147 if (node_end_pfn[nid] > max_pfn)
148 node_end_pfn[nid] = max_pfn;
149 /*
150 * if a user has given mem=XXXX, then we need to make sure
151 * that the node _starts_ before that, too, not just ends
152 */
153 if (node_start_pfn[nid] > max_pfn)
154 node_start_pfn[nid] = max_pfn;
155 if (node_start_pfn[nid] > node_end_pfn[nid])
156 BUG();
157}
158
Dave Hansenc2ebaa42005-06-23 00:07:38 -0700159/* Find the owning node for a pfn. */
160int early_pfn_to_nid(unsigned long pfn)
161{
162 int nid;
163
164 for_each_node(nid) {
165 if (node_end_pfn[nid] == 0)
166 break;
167 if (node_start_pfn[nid] <= pfn && node_end_pfn[nid] >= pfn)
168 return nid;
169 }
170
171 return 0;
172}
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174/*
175 * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
176 * method. For node zero take this from the bottom of memory, for
177 * subsequent nodes place them at node_remap_start_vaddr which contains
178 * node local data in physically node local memory. See setup_memory()
179 * for details.
180 */
181static void __init allocate_pgdat(int nid)
182{
183 if (nid && node_has_online_mem(nid))
184 NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
185 else {
186 NODE_DATA(nid) = (pg_data_t *)(__va(min_low_pfn << PAGE_SHIFT));
187 min_low_pfn += PFN_UP(sizeof(pg_data_t));
188 }
189}
190
Dave Hansen6f167ec2005-06-23 00:07:39 -0700191void *alloc_remap(int nid, unsigned long size)
192{
193 void *allocation = node_remap_alloc_vaddr[nid];
194
195 size = ALIGN(size, L1_CACHE_BYTES);
196
197 if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
198 return 0;
199
200 node_remap_alloc_vaddr[nid] += size;
201 memset(allocation, 0, size);
202
203 return allocation;
204}
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206void __init remap_numa_kva(void)
207{
208 void *vaddr;
209 unsigned long pfn;
210 int node;
211
212 for_each_online_node(node) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
214 vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
215 set_pmd_pfn((ulong) vaddr,
216 node_remap_start_pfn[node] + pfn,
217 PAGE_KERNEL_LARGE);
218 }
219 }
220}
221
222static unsigned long calculate_numa_remap_pages(void)
223{
224 int nid;
225 unsigned long size, reserve_pages = 0;
Dave Hansen5b505b92005-06-23 00:07:41 -0700226 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
228 for_each_online_node(nid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 /*
230 * The acpi/srat node info can show hot-add memroy zones
231 * where memory could be added but not currently present.
232 */
233 if (node_start_pfn[nid] > max_pfn)
234 continue;
235 if (node_end_pfn[nid] > max_pfn)
236 node_end_pfn[nid] = max_pfn;
237
238 /* ensure the remap includes space for the pgdat. */
239 size = node_remap_size[nid] + sizeof(pg_data_t);
240
241 /* convert size to large (pmd size) pages, rounding up */
242 size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
243 /* now the roundup is correct, convert to PAGE_SIZE pages */
244 size = size * PTRS_PER_PTE;
Dave Hansen5b505b92005-06-23 00:07:41 -0700245
246 /*
247 * Validate the region we are allocating only contains valid
248 * pages.
249 */
250 for (pfn = node_end_pfn[nid] - size;
251 pfn < node_end_pfn[nid]; pfn++)
252 if (!page_is_ram(pfn))
253 break;
254
255 if (pfn != node_end_pfn[nid])
256 size = 0;
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
259 size, nid);
260 node_remap_size[nid] = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 node_remap_offset[nid] = reserve_pages;
Dave Hansen6f167ec2005-06-23 00:07:39 -0700262 reserve_pages += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 printk("Shrinking node %d from %ld pages to %ld pages\n",
264 nid, node_end_pfn[nid], node_end_pfn[nid] - size);
Ravikiran G Thirumalai4b0271e2005-08-07 09:42:50 -0700265
266 if (node_end_pfn[nid] & (PTRS_PER_PTE-1)) {
267 /*
268 * Align node_end_pfn[] and node_remap_start_pfn[] to
269 * pmd boundary. remap_numa_kva will barf otherwise.
270 */
271 printk("Shrinking node %d further by %ld pages for proper alignment\n",
272 nid, node_end_pfn[nid] & (PTRS_PER_PTE-1));
273 size += node_end_pfn[nid] & (PTRS_PER_PTE-1);
274 }
275
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 node_end_pfn[nid] -= size;
277 node_remap_start_pfn[nid] = node_end_pfn[nid];
278 }
279 printk("Reserving total of %ld pages for numa KVA remap\n",
280 reserve_pages);
281 return reserve_pages;
282}
283
284extern void setup_bootmem_allocator(void);
285unsigned long __init setup_memory(void)
286{
287 int nid;
288 unsigned long system_start_pfn, system_max_low_pfn;
289 unsigned long reserve_pages;
290
291 /*
292 * When mapping a NUMA machine we allocate the node_mem_map arrays
293 * from node local memory. They are then mapped directly into KVA
294 * between zone normal and vmalloc space. Calculate the size of
295 * this space and use it to adjust the boundry between ZONE_NORMAL
296 * and ZONE_HIGHMEM.
297 */
298 find_max_pfn();
299 get_memcfg_numa();
300
301 reserve_pages = calculate_numa_remap_pages();
302
303 /* partially used pages are not usable - thus round upwards */
304 system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
305
306 system_max_low_pfn = max_low_pfn = find_max_low_pfn() - reserve_pages;
307 printk("reserve_pages = %ld find_max_low_pfn() ~ %ld\n",
308 reserve_pages, max_low_pfn + reserve_pages);
309 printk("max_pfn = %ld\n", max_pfn);
310#ifdef CONFIG_HIGHMEM
311 highstart_pfn = highend_pfn = max_pfn;
312 if (max_pfn > system_max_low_pfn)
313 highstart_pfn = system_max_low_pfn;
314 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
315 pages_to_mb(highend_pfn - highstart_pfn));
316#endif
317 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
318 pages_to_mb(system_max_low_pfn));
319 printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
320 min_low_pfn, max_low_pfn, highstart_pfn);
321
322 printk("Low memory ends at vaddr %08lx\n",
323 (ulong) pfn_to_kaddr(max_low_pfn));
324 for_each_online_node(nid) {
325 node_remap_start_vaddr[nid] = pfn_to_kaddr(
Dave Hansen6f167ec2005-06-23 00:07:39 -0700326 highstart_pfn + node_remap_offset[nid]);
327 /* Init the node remap allocator */
328 node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
329 (node_remap_size[nid] * PAGE_SIZE);
330 node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
331 ALIGN(sizeof(pg_data_t), PAGE_SIZE);
332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 allocate_pgdat(nid);
334 printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
335 (ulong) node_remap_start_vaddr[nid],
Dave Hansen6f167ec2005-06-23 00:07:39 -0700336 (ulong) pfn_to_kaddr(highstart_pfn
337 + node_remap_offset[nid] + node_remap_size[nid]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 }
339 printk("High memory starts at vaddr %08lx\n",
340 (ulong) pfn_to_kaddr(highstart_pfn));
341 vmalloc_earlyreserve = reserve_pages * PAGE_SIZE;
342 for_each_online_node(nid)
343 find_max_pfn_node(nid);
344
345 memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
346 NODE_DATA(0)->bdata = &node0_bdata;
347 setup_bootmem_allocator();
348 return max_low_pfn;
349}
350
351void __init zone_sizes_init(void)
352{
353 int nid;
354
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
356 for_each_online_node(nid) {
357 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
358 unsigned long *zholes_size;
359 unsigned int max_dma;
360
361 unsigned long low = max_low_pfn;
362 unsigned long start = node_start_pfn[nid];
363 unsigned long high = node_end_pfn[nid];
364
365 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
366
367 if (node_has_online_mem(nid)){
368 if (start > low) {
369#ifdef CONFIG_HIGHMEM
370 BUG_ON(start > high);
371 zones_size[ZONE_HIGHMEM] = high - start;
372#endif
373 } else {
374 if (low < max_dma)
375 zones_size[ZONE_DMA] = low;
376 else {
377 BUG_ON(max_dma > low);
378 BUG_ON(low > high);
379 zones_size[ZONE_DMA] = max_dma;
380 zones_size[ZONE_NORMAL] = low - max_dma;
381#ifdef CONFIG_HIGHMEM
382 zones_size[ZONE_HIGHMEM] = high - low;
383#endif
384 }
385 }
386 }
387
388 zholes_size = get_zholes_size(nid);
Dave Hansen6f167ec2005-06-23 00:07:39 -0700389
390 free_area_init_node(nid, NODE_DATA(nid), zones_size, start,
391 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 }
393 return;
394}
395
396void __init set_highmem_pages_init(int bad_ppro)
397{
398#ifdef CONFIG_HIGHMEM
399 struct zone *zone;
Andy Whitcroft05b79bd2005-06-23 00:07:57 -0700400 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
402 for_each_zone(zone) {
Andy Whitcroft05b79bd2005-06-23 00:07:57 -0700403 unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
404
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 if (!is_highmem(zone))
406 continue;
407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 zone_start_pfn = zone->zone_start_pfn;
Andy Whitcroft05b79bd2005-06-23 00:07:57 -0700409 zone_end_pfn = zone_start_pfn + zone->spanned_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Andy Whitcroft05b79bd2005-06-23 00:07:57 -0700411 printk("Initializing %s for node %d (%08lx:%08lx)\n",
412 zone->name, zone->zone_pgdat->node_id,
413 zone_start_pfn, zone_end_pfn);
414
415 for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
416 if (!pfn_valid(node_pfn))
417 continue;
418 page = pfn_to_page(node_pfn);
Dave Hansen05039b92005-10-29 18:16:57 -0700419 add_one_highpage_init(page, node_pfn, bad_ppro);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 }
421 }
422 totalram_pages += totalhigh_pages;
423#endif
424}