blob: fe6eb901326e6610171a927dddd5e01280d7a71b [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
25#include <linux/config.h>
26#include <linux/mm.h>
27#include <linux/bootmem.h>
28#include <linux/mmzone.h>
29#include <linux/highmem.h>
30#include <linux/initrd.h>
31#include <linux/nodemask.h>
Alexey Dobriyan129f6942005-06-23 00:08:33 -070032#include <linux/module.h>
Eric W. Biederman1bc3b912005-06-25 14:58:01 -070033#include <linux/kexec.h>
Dave Hansen22a98352006-03-27 01:16:04 -080034#include <linux/pfn.h>
Eric W. Biederman1bc3b912005-06-25 14:58:01 -070035
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/e820.h>
37#include <asm/setup.h>
38#include <asm/mmzone.h>
39#include <bios_ebda.h>
40
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070041struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
Alexey Dobriyan129f6942005-06-23 00:08:33 -070042EXPORT_SYMBOL(node_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043bootmem_data_t node0_bdata;
44
45/*
46 * numa interface - we expect the numa architecture specfic code to have
47 * populated the following initialisation.
48 *
49 * 1) node_online_map - the map of all nodes configured (online) in the system
Andy Whitcroft05b79bd2005-06-23 00:07:57 -070050 * 2) node_start_pfn - the starting page frame number for a node
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 * 3) node_end_pfn - the ending page fram number for a node
52 */
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070053unsigned long node_start_pfn[MAX_NUMNODES] __read_mostly;
54unsigned long node_end_pfn[MAX_NUMNODES] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Andy Whitcroft05b79bd2005-06-23 00:07:57 -070056
57#ifdef CONFIG_DISCONTIGMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070058/*
Andy Whitcroft05b79bd2005-06-23 00:07:57 -070059 * 4) physnode_map - the mapping between a pfn and owning node
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 * physnode_map keeps track of the physical memory layout of a generic
61 * numa node on a 256Mb break (each element of the array will
62 * represent 256Mb of memory and will be marked by the node id. so,
63 * if the first gig is on node 0, and the second gig is on node 1
64 * physnode_map will contain:
65 *
66 * physnode_map[0-3] = 0;
67 * physnode_map[4-7] = 1;
68 * physnode_map[8- ] = -1;
69 */
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070070s8 physnode_map[MAX_ELEMENTS] __read_mostly = { [0 ... (MAX_ELEMENTS - 1)] = -1};
Alexey Dobriyan129f6942005-06-23 00:08:33 -070071EXPORT_SYMBOL(physnode_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73void memory_present(int nid, unsigned long start, unsigned long end)
74{
75 unsigned long pfn;
76
77 printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
78 nid, start, end);
79 printk(KERN_DEBUG " Setting physnode_map array to node %d for pfns:\n", nid);
80 printk(KERN_DEBUG " ");
81 for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
82 physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
83 printk("%ld ", pfn);
84 }
85 printk("\n");
86}
87
88unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
89 unsigned long end_pfn)
90{
91 unsigned long nr_pages = end_pfn - start_pfn;
92
93 if (!nr_pages)
94 return 0;
95
96 return (nr_pages + 1) * sizeof(struct page);
97}
Andy Whitcroft05b79bd2005-06-23 00:07:57 -070098#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
100extern unsigned long find_max_low_pfn(void);
101extern void find_max_pfn(void);
Dave Hansen05039b92005-10-29 18:16:57 -0700102extern void add_one_highpage_init(struct page *, int, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104extern struct e820map e820;
105extern unsigned long init_pg_tables_end;
106extern unsigned long highend_pfn, highstart_pfn;
107extern unsigned long max_low_pfn;
108extern unsigned long totalram_pages;
109extern unsigned long totalhigh_pages;
110
111#define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
112
113unsigned long node_remap_start_pfn[MAX_NUMNODES];
114unsigned long node_remap_size[MAX_NUMNODES];
115unsigned long node_remap_offset[MAX_NUMNODES];
116void *node_remap_start_vaddr[MAX_NUMNODES];
117void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
118
Dave Hansen6f167ec2005-06-23 00:07:39 -0700119void *node_remap_end_vaddr[MAX_NUMNODES];
120void *node_remap_alloc_vaddr[MAX_NUMNODES];
121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122/*
123 * FLAT - support for basic PC memory model with discontig enabled, essentially
124 * a single node with all available processors in it with a flat
125 * memory map.
126 */
127int __init get_memcfg_numa_flat(void)
128{
129 printk("NUMA - single node, flat memory mode\n");
130
131 /* Run the memory configuration and find the top of memory. */
132 find_max_pfn();
133 node_start_pfn[0] = 0;
134 node_end_pfn[0] = max_pfn;
135 memory_present(0, 0, max_pfn);
136
137 /* Indicate there is one node available. */
138 nodes_clear(node_online_map);
139 node_set_online(0);
140 return 1;
141}
142
143/*
144 * Find the highest page frame number we have available for the node
145 */
146static void __init find_max_pfn_node(int nid)
147{
148 if (node_end_pfn[nid] > max_pfn)
149 node_end_pfn[nid] = max_pfn;
150 /*
151 * if a user has given mem=XXXX, then we need to make sure
152 * that the node _starts_ before that, too, not just ends
153 */
154 if (node_start_pfn[nid] > max_pfn)
155 node_start_pfn[nid] = max_pfn;
156 if (node_start_pfn[nid] > node_end_pfn[nid])
157 BUG();
158}
159
Dave Hansenc2ebaa42005-06-23 00:07:38 -0700160/* Find the owning node for a pfn. */
161int early_pfn_to_nid(unsigned long pfn)
162{
163 int nid;
164
165 for_each_node(nid) {
166 if (node_end_pfn[nid] == 0)
167 break;
168 if (node_start_pfn[nid] <= pfn && node_end_pfn[nid] >= pfn)
169 return nid;
170 }
171
172 return 0;
173}
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175/*
176 * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
177 * method. For node zero take this from the bottom of memory, for
178 * subsequent nodes place them at node_remap_start_vaddr which contains
179 * node local data in physically node local memory. See setup_memory()
180 * for details.
181 */
182static void __init allocate_pgdat(int nid)
183{
184 if (nid && node_has_online_mem(nid))
185 NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
186 else {
187 NODE_DATA(nid) = (pg_data_t *)(__va(min_low_pfn << PAGE_SHIFT));
188 min_low_pfn += PFN_UP(sizeof(pg_data_t));
189 }
190}
191
Dave Hansen6f167ec2005-06-23 00:07:39 -0700192void *alloc_remap(int nid, unsigned long size)
193{
194 void *allocation = node_remap_alloc_vaddr[nid];
195
196 size = ALIGN(size, L1_CACHE_BYTES);
197
198 if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
199 return 0;
200
201 node_remap_alloc_vaddr[nid] += size;
202 memset(allocation, 0, size);
203
204 return allocation;
205}
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207void __init remap_numa_kva(void)
208{
209 void *vaddr;
210 unsigned long pfn;
211 int node;
212
213 for_each_online_node(node) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
215 vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
216 set_pmd_pfn((ulong) vaddr,
217 node_remap_start_pfn[node] + pfn,
218 PAGE_KERNEL_LARGE);
219 }
220 }
221}
222
223static unsigned long calculate_numa_remap_pages(void)
224{
225 int nid;
226 unsigned long size, reserve_pages = 0;
Dave Hansen5b505b92005-06-23 00:07:41 -0700227 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
229 for_each_online_node(nid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 /*
231 * The acpi/srat node info can show hot-add memroy zones
232 * where memory could be added but not currently present.
233 */
234 if (node_start_pfn[nid] > max_pfn)
235 continue;
236 if (node_end_pfn[nid] > max_pfn)
237 node_end_pfn[nid] = max_pfn;
238
239 /* ensure the remap includes space for the pgdat. */
240 size = node_remap_size[nid] + sizeof(pg_data_t);
241
242 /* convert size to large (pmd size) pages, rounding up */
243 size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
244 /* now the roundup is correct, convert to PAGE_SIZE pages */
245 size = size * PTRS_PER_PTE;
Dave Hansen5b505b92005-06-23 00:07:41 -0700246
247 /*
248 * Validate the region we are allocating only contains valid
249 * pages.
250 */
251 for (pfn = node_end_pfn[nid] - size;
252 pfn < node_end_pfn[nid]; pfn++)
253 if (!page_is_ram(pfn))
254 break;
255
256 if (pfn != node_end_pfn[nid])
257 size = 0;
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
260 size, nid);
261 node_remap_size[nid] = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 node_remap_offset[nid] = reserve_pages;
Dave Hansen6f167ec2005-06-23 00:07:39 -0700263 reserve_pages += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 printk("Shrinking node %d from %ld pages to %ld pages\n",
265 nid, node_end_pfn[nid], node_end_pfn[nid] - size);
Ravikiran G Thirumalai4b0271e2005-08-07 09:42:50 -0700266
267 if (node_end_pfn[nid] & (PTRS_PER_PTE-1)) {
268 /*
269 * Align node_end_pfn[] and node_remap_start_pfn[] to
270 * pmd boundary. remap_numa_kva will barf otherwise.
271 */
272 printk("Shrinking node %d further by %ld pages for proper alignment\n",
273 nid, node_end_pfn[nid] & (PTRS_PER_PTE-1));
274 size += node_end_pfn[nid] & (PTRS_PER_PTE-1);
275 }
276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 node_end_pfn[nid] -= size;
278 node_remap_start_pfn[nid] = node_end_pfn[nid];
279 }
280 printk("Reserving total of %ld pages for numa KVA remap\n",
281 reserve_pages);
282 return reserve_pages;
283}
284
285extern void setup_bootmem_allocator(void);
286unsigned long __init setup_memory(void)
287{
288 int nid;
289 unsigned long system_start_pfn, system_max_low_pfn;
290 unsigned long reserve_pages;
291
292 /*
293 * When mapping a NUMA machine we allocate the node_mem_map arrays
294 * from node local memory. They are then mapped directly into KVA
295 * between zone normal and vmalloc space. Calculate the size of
296 * this space and use it to adjust the boundry between ZONE_NORMAL
297 * and ZONE_HIGHMEM.
298 */
299 find_max_pfn();
300 get_memcfg_numa();
301
302 reserve_pages = calculate_numa_remap_pages();
303
304 /* partially used pages are not usable - thus round upwards */
305 system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
306
307 system_max_low_pfn = max_low_pfn = find_max_low_pfn() - reserve_pages;
308 printk("reserve_pages = %ld find_max_low_pfn() ~ %ld\n",
309 reserve_pages, max_low_pfn + reserve_pages);
310 printk("max_pfn = %ld\n", max_pfn);
311#ifdef CONFIG_HIGHMEM
312 highstart_pfn = highend_pfn = max_pfn;
313 if (max_pfn > system_max_low_pfn)
314 highstart_pfn = system_max_low_pfn;
315 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
316 pages_to_mb(highend_pfn - highstart_pfn));
317#endif
318 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
319 pages_to_mb(system_max_low_pfn));
320 printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
321 min_low_pfn, max_low_pfn, highstart_pfn);
322
323 printk("Low memory ends at vaddr %08lx\n",
324 (ulong) pfn_to_kaddr(max_low_pfn));
325 for_each_online_node(nid) {
326 node_remap_start_vaddr[nid] = pfn_to_kaddr(
Dave Hansen6f167ec2005-06-23 00:07:39 -0700327 highstart_pfn + node_remap_offset[nid]);
328 /* Init the node remap allocator */
329 node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
330 (node_remap_size[nid] * PAGE_SIZE);
331 node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
332 ALIGN(sizeof(pg_data_t), PAGE_SIZE);
333
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 allocate_pgdat(nid);
335 printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
336 (ulong) node_remap_start_vaddr[nid],
Dave Hansen6f167ec2005-06-23 00:07:39 -0700337 (ulong) pfn_to_kaddr(highstart_pfn
338 + node_remap_offset[nid] + node_remap_size[nid]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 }
340 printk("High memory starts at vaddr %08lx\n",
341 (ulong) pfn_to_kaddr(highstart_pfn));
342 vmalloc_earlyreserve = reserve_pages * PAGE_SIZE;
343 for_each_online_node(nid)
344 find_max_pfn_node(nid);
345
346 memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
347 NODE_DATA(0)->bdata = &node0_bdata;
348 setup_bootmem_allocator();
349 return max_low_pfn;
350}
351
352void __init zone_sizes_init(void)
353{
354 int nid;
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
357 for_each_online_node(nid) {
358 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
359 unsigned long *zholes_size;
360 unsigned int max_dma;
361
362 unsigned long low = max_low_pfn;
363 unsigned long start = node_start_pfn[nid];
364 unsigned long high = node_end_pfn[nid];
365
366 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
367
368 if (node_has_online_mem(nid)){
369 if (start > low) {
370#ifdef CONFIG_HIGHMEM
371 BUG_ON(start > high);
372 zones_size[ZONE_HIGHMEM] = high - start;
373#endif
374 } else {
375 if (low < max_dma)
376 zones_size[ZONE_DMA] = low;
377 else {
378 BUG_ON(max_dma > low);
379 BUG_ON(low > high);
380 zones_size[ZONE_DMA] = max_dma;
381 zones_size[ZONE_NORMAL] = low - max_dma;
382#ifdef CONFIG_HIGHMEM
383 zones_size[ZONE_HIGHMEM] = high - low;
384#endif
385 }
386 }
387 }
388
389 zholes_size = get_zholes_size(nid);
Dave Hansen6f167ec2005-06-23 00:07:39 -0700390
391 free_area_init_node(nid, NODE_DATA(nid), zones_size, start,
392 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 }
394 return;
395}
396
397void __init set_highmem_pages_init(int bad_ppro)
398{
399#ifdef CONFIG_HIGHMEM
400 struct zone *zone;
Andy Whitcroft05b79bd2005-06-23 00:07:57 -0700401 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
403 for_each_zone(zone) {
Andy Whitcroft05b79bd2005-06-23 00:07:57 -0700404 unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
405
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 if (!is_highmem(zone))
407 continue;
408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 zone_start_pfn = zone->zone_start_pfn;
Andy Whitcroft05b79bd2005-06-23 00:07:57 -0700410 zone_end_pfn = zone_start_pfn + zone->spanned_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Andy Whitcroft05b79bd2005-06-23 00:07:57 -0700412 printk("Initializing %s for node %d (%08lx:%08lx)\n",
413 zone->name, zone->zone_pgdat->node_id,
414 zone_start_pfn, zone_end_pfn);
415
416 for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
417 if (!pfn_valid(node_pfn))
418 continue;
419 page = pfn_to_page(node_pfn);
Dave Hansen05039b92005-10-29 18:16:57 -0700420 add_one_highpage_init(page, node_pfn, bad_ppro);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
422 }
423 totalram_pages += totalhigh_pages;
424#endif
425}