]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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> | |
32 | #include <asm/e820.h> | |
33 | #include <asm/setup.h> | |
34 | #include <asm/mmzone.h> | |
35 | #include <bios_ebda.h> | |
36 | ||
37 | struct pglist_data *node_data[MAX_NUMNODES]; | |
38 | bootmem_data_t node0_bdata; | |
39 | ||
40 | /* | |
41 | * numa interface - we expect the numa architecture specfic code to have | |
42 | * populated the following initialisation. | |
43 | * | |
44 | * 1) node_online_map - the map of all nodes configured (online) in the system | |
05b79bdc | 45 | * 2) node_start_pfn - the starting page frame number for a node |
1da177e4 LT |
46 | * 3) node_end_pfn - the ending page fram number for a node |
47 | */ | |
05b79bdc AW |
48 | unsigned long node_start_pfn[MAX_NUMNODES]; |
49 | unsigned long node_end_pfn[MAX_NUMNODES]; | |
50 | ||
1da177e4 | 51 | |
05b79bdc | 52 | #ifdef CONFIG_DISCONTIGMEM |
1da177e4 | 53 | /* |
05b79bdc | 54 | * 4) physnode_map - the mapping between a pfn and owning node |
1da177e4 LT |
55 | * physnode_map keeps track of the physical memory layout of a generic |
56 | * numa node on a 256Mb break (each element of the array will | |
57 | * represent 256Mb of memory and will be marked by the node id. so, | |
58 | * if the first gig is on node 0, and the second gig is on node 1 | |
59 | * physnode_map will contain: | |
60 | * | |
61 | * physnode_map[0-3] = 0; | |
62 | * physnode_map[4-7] = 1; | |
63 | * physnode_map[8- ] = -1; | |
64 | */ | |
65 | s8 physnode_map[MAX_ELEMENTS] = { [0 ... (MAX_ELEMENTS - 1)] = -1}; | |
66 | ||
67 | void memory_present(int nid, unsigned long start, unsigned long end) | |
68 | { | |
69 | unsigned long pfn; | |
70 | ||
71 | printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n", | |
72 | nid, start, end); | |
73 | printk(KERN_DEBUG " Setting physnode_map array to node %d for pfns:\n", nid); | |
74 | printk(KERN_DEBUG " "); | |
75 | for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) { | |
76 | physnode_map[pfn / PAGES_PER_ELEMENT] = nid; | |
77 | printk("%ld ", pfn); | |
78 | } | |
79 | printk("\n"); | |
80 | } | |
81 | ||
82 | unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn, | |
83 | unsigned long end_pfn) | |
84 | { | |
85 | unsigned long nr_pages = end_pfn - start_pfn; | |
86 | ||
87 | if (!nr_pages) | |
88 | return 0; | |
89 | ||
90 | return (nr_pages + 1) * sizeof(struct page); | |
91 | } | |
05b79bdc | 92 | #endif |
1da177e4 LT |
93 | |
94 | extern unsigned long find_max_low_pfn(void); | |
95 | extern void find_max_pfn(void); | |
96 | extern void one_highpage_init(struct page *, int, int); | |
97 | ||
98 | extern struct e820map e820; | |
99 | extern unsigned long init_pg_tables_end; | |
100 | extern unsigned long highend_pfn, highstart_pfn; | |
101 | extern unsigned long max_low_pfn; | |
102 | extern unsigned long totalram_pages; | |
103 | extern unsigned long totalhigh_pages; | |
104 | ||
105 | #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE) | |
106 | ||
107 | unsigned long node_remap_start_pfn[MAX_NUMNODES]; | |
108 | unsigned long node_remap_size[MAX_NUMNODES]; | |
109 | unsigned long node_remap_offset[MAX_NUMNODES]; | |
110 | void *node_remap_start_vaddr[MAX_NUMNODES]; | |
111 | void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags); | |
112 | ||
6f167ec7 DH |
113 | void *node_remap_end_vaddr[MAX_NUMNODES]; |
114 | void *node_remap_alloc_vaddr[MAX_NUMNODES]; | |
115 | ||
1da177e4 LT |
116 | /* |
117 | * FLAT - support for basic PC memory model with discontig enabled, essentially | |
118 | * a single node with all available processors in it with a flat | |
119 | * memory map. | |
120 | */ | |
121 | int __init get_memcfg_numa_flat(void) | |
122 | { | |
123 | printk("NUMA - single node, flat memory mode\n"); | |
124 | ||
125 | /* Run the memory configuration and find the top of memory. */ | |
126 | find_max_pfn(); | |
127 | node_start_pfn[0] = 0; | |
128 | node_end_pfn[0] = max_pfn; | |
129 | memory_present(0, 0, max_pfn); | |
130 | ||
131 | /* Indicate there is one node available. */ | |
132 | nodes_clear(node_online_map); | |
133 | node_set_online(0); | |
134 | return 1; | |
135 | } | |
136 | ||
137 | /* | |
138 | * Find the highest page frame number we have available for the node | |
139 | */ | |
140 | static void __init find_max_pfn_node(int nid) | |
141 | { | |
142 | if (node_end_pfn[nid] > max_pfn) | |
143 | node_end_pfn[nid] = max_pfn; | |
144 | /* | |
145 | * if a user has given mem=XXXX, then we need to make sure | |
146 | * that the node _starts_ before that, too, not just ends | |
147 | */ | |
148 | if (node_start_pfn[nid] > max_pfn) | |
149 | node_start_pfn[nid] = max_pfn; | |
150 | if (node_start_pfn[nid] > node_end_pfn[nid]) | |
151 | BUG(); | |
152 | } | |
153 | ||
c2ebaa42 DH |
154 | /* Find the owning node for a pfn. */ |
155 | int early_pfn_to_nid(unsigned long pfn) | |
156 | { | |
157 | int nid; | |
158 | ||
159 | for_each_node(nid) { | |
160 | if (node_end_pfn[nid] == 0) | |
161 | break; | |
162 | if (node_start_pfn[nid] <= pfn && node_end_pfn[nid] >= pfn) | |
163 | return nid; | |
164 | } | |
165 | ||
166 | return 0; | |
167 | } | |
168 | ||
1da177e4 LT |
169 | /* |
170 | * Allocate memory for the pg_data_t for this node via a crude pre-bootmem | |
171 | * method. For node zero take this from the bottom of memory, for | |
172 | * subsequent nodes place them at node_remap_start_vaddr which contains | |
173 | * node local data in physically node local memory. See setup_memory() | |
174 | * for details. | |
175 | */ | |
176 | static void __init allocate_pgdat(int nid) | |
177 | { | |
178 | if (nid && node_has_online_mem(nid)) | |
179 | NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid]; | |
180 | else { | |
181 | NODE_DATA(nid) = (pg_data_t *)(__va(min_low_pfn << PAGE_SHIFT)); | |
182 | min_low_pfn += PFN_UP(sizeof(pg_data_t)); | |
183 | } | |
184 | } | |
185 | ||
6f167ec7 DH |
186 | void *alloc_remap(int nid, unsigned long size) |
187 | { | |
188 | void *allocation = node_remap_alloc_vaddr[nid]; | |
189 | ||
190 | size = ALIGN(size, L1_CACHE_BYTES); | |
191 | ||
192 | if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid]) | |
193 | return 0; | |
194 | ||
195 | node_remap_alloc_vaddr[nid] += size; | |
196 | memset(allocation, 0, size); | |
197 | ||
198 | return allocation; | |
199 | } | |
200 | ||
1da177e4 LT |
201 | void __init remap_numa_kva(void) |
202 | { | |
203 | void *vaddr; | |
204 | unsigned long pfn; | |
205 | int node; | |
206 | ||
207 | for_each_online_node(node) { | |
1da177e4 LT |
208 | for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) { |
209 | vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT); | |
210 | set_pmd_pfn((ulong) vaddr, | |
211 | node_remap_start_pfn[node] + pfn, | |
212 | PAGE_KERNEL_LARGE); | |
213 | } | |
214 | } | |
215 | } | |
216 | ||
217 | static unsigned long calculate_numa_remap_pages(void) | |
218 | { | |
219 | int nid; | |
220 | unsigned long size, reserve_pages = 0; | |
5b505b90 | 221 | unsigned long pfn; |
1da177e4 LT |
222 | |
223 | for_each_online_node(nid) { | |
1da177e4 LT |
224 | /* |
225 | * The acpi/srat node info can show hot-add memroy zones | |
226 | * where memory could be added but not currently present. | |
227 | */ | |
228 | if (node_start_pfn[nid] > max_pfn) | |
229 | continue; | |
230 | if (node_end_pfn[nid] > max_pfn) | |
231 | node_end_pfn[nid] = max_pfn; | |
232 | ||
233 | /* ensure the remap includes space for the pgdat. */ | |
234 | size = node_remap_size[nid] + sizeof(pg_data_t); | |
235 | ||
236 | /* convert size to large (pmd size) pages, rounding up */ | |
237 | size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES; | |
238 | /* now the roundup is correct, convert to PAGE_SIZE pages */ | |
239 | size = size * PTRS_PER_PTE; | |
5b505b90 DH |
240 | |
241 | /* | |
242 | * Validate the region we are allocating only contains valid | |
243 | * pages. | |
244 | */ | |
245 | for (pfn = node_end_pfn[nid] - size; | |
246 | pfn < node_end_pfn[nid]; pfn++) | |
247 | if (!page_is_ram(pfn)) | |
248 | break; | |
249 | ||
250 | if (pfn != node_end_pfn[nid]) | |
251 | size = 0; | |
252 | ||
1da177e4 LT |
253 | printk("Reserving %ld pages of KVA for lmem_map of node %d\n", |
254 | size, nid); | |
255 | node_remap_size[nid] = size; | |
1da177e4 | 256 | node_remap_offset[nid] = reserve_pages; |
6f167ec7 | 257 | reserve_pages += size; |
1da177e4 LT |
258 | printk("Shrinking node %d from %ld pages to %ld pages\n", |
259 | nid, node_end_pfn[nid], node_end_pfn[nid] - size); | |
260 | node_end_pfn[nid] -= size; | |
261 | node_remap_start_pfn[nid] = node_end_pfn[nid]; | |
262 | } | |
263 | printk("Reserving total of %ld pages for numa KVA remap\n", | |
264 | reserve_pages); | |
265 | return reserve_pages; | |
266 | } | |
267 | ||
268 | extern void setup_bootmem_allocator(void); | |
269 | unsigned long __init setup_memory(void) | |
270 | { | |
271 | int nid; | |
272 | unsigned long system_start_pfn, system_max_low_pfn; | |
273 | unsigned long reserve_pages; | |
274 | ||
275 | /* | |
276 | * When mapping a NUMA machine we allocate the node_mem_map arrays | |
277 | * from node local memory. They are then mapped directly into KVA | |
278 | * between zone normal and vmalloc space. Calculate the size of | |
279 | * this space and use it to adjust the boundry between ZONE_NORMAL | |
280 | * and ZONE_HIGHMEM. | |
281 | */ | |
282 | find_max_pfn(); | |
283 | get_memcfg_numa(); | |
284 | ||
285 | reserve_pages = calculate_numa_remap_pages(); | |
286 | ||
287 | /* partially used pages are not usable - thus round upwards */ | |
288 | system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end); | |
289 | ||
290 | system_max_low_pfn = max_low_pfn = find_max_low_pfn() - reserve_pages; | |
291 | printk("reserve_pages = %ld find_max_low_pfn() ~ %ld\n", | |
292 | reserve_pages, max_low_pfn + reserve_pages); | |
293 | printk("max_pfn = %ld\n", max_pfn); | |
294 | #ifdef CONFIG_HIGHMEM | |
295 | highstart_pfn = highend_pfn = max_pfn; | |
296 | if (max_pfn > system_max_low_pfn) | |
297 | highstart_pfn = system_max_low_pfn; | |
298 | printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", | |
299 | pages_to_mb(highend_pfn - highstart_pfn)); | |
300 | #endif | |
301 | printk(KERN_NOTICE "%ldMB LOWMEM available.\n", | |
302 | pages_to_mb(system_max_low_pfn)); | |
303 | printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n", | |
304 | min_low_pfn, max_low_pfn, highstart_pfn); | |
305 | ||
306 | printk("Low memory ends at vaddr %08lx\n", | |
307 | (ulong) pfn_to_kaddr(max_low_pfn)); | |
308 | for_each_online_node(nid) { | |
309 | node_remap_start_vaddr[nid] = pfn_to_kaddr( | |
6f167ec7 DH |
310 | highstart_pfn + node_remap_offset[nid]); |
311 | /* Init the node remap allocator */ | |
312 | node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] + | |
313 | (node_remap_size[nid] * PAGE_SIZE); | |
314 | node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] + | |
315 | ALIGN(sizeof(pg_data_t), PAGE_SIZE); | |
316 | ||
1da177e4 LT |
317 | allocate_pgdat(nid); |
318 | printk ("node %d will remap to vaddr %08lx - %08lx\n", nid, | |
319 | (ulong) node_remap_start_vaddr[nid], | |
6f167ec7 DH |
320 | (ulong) pfn_to_kaddr(highstart_pfn |
321 | + node_remap_offset[nid] + node_remap_size[nid])); | |
1da177e4 LT |
322 | } |
323 | printk("High memory starts at vaddr %08lx\n", | |
324 | (ulong) pfn_to_kaddr(highstart_pfn)); | |
325 | vmalloc_earlyreserve = reserve_pages * PAGE_SIZE; | |
326 | for_each_online_node(nid) | |
327 | find_max_pfn_node(nid); | |
328 | ||
329 | memset(NODE_DATA(0), 0, sizeof(struct pglist_data)); | |
330 | NODE_DATA(0)->bdata = &node0_bdata; | |
331 | setup_bootmem_allocator(); | |
332 | return max_low_pfn; | |
333 | } | |
334 | ||
335 | void __init zone_sizes_init(void) | |
336 | { | |
337 | int nid; | |
338 | ||
339 | /* | |
340 | * Insert nodes into pgdat_list backward so they appear in order. | |
341 | * Clobber node 0's links and NULL out pgdat_list before starting. | |
342 | */ | |
343 | pgdat_list = NULL; | |
344 | for (nid = MAX_NUMNODES - 1; nid >= 0; nid--) { | |
345 | if (!node_online(nid)) | |
346 | continue; | |
347 | NODE_DATA(nid)->pgdat_next = pgdat_list; | |
348 | pgdat_list = NODE_DATA(nid); | |
349 | } | |
350 | ||
351 | for_each_online_node(nid) { | |
352 | unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0}; | |
353 | unsigned long *zholes_size; | |
354 | unsigned int max_dma; | |
355 | ||
356 | unsigned long low = max_low_pfn; | |
357 | unsigned long start = node_start_pfn[nid]; | |
358 | unsigned long high = node_end_pfn[nid]; | |
359 | ||
360 | max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT; | |
361 | ||
362 | if (node_has_online_mem(nid)){ | |
363 | if (start > low) { | |
364 | #ifdef CONFIG_HIGHMEM | |
365 | BUG_ON(start > high); | |
366 | zones_size[ZONE_HIGHMEM] = high - start; | |
367 | #endif | |
368 | } else { | |
369 | if (low < max_dma) | |
370 | zones_size[ZONE_DMA] = low; | |
371 | else { | |
372 | BUG_ON(max_dma > low); | |
373 | BUG_ON(low > high); | |
374 | zones_size[ZONE_DMA] = max_dma; | |
375 | zones_size[ZONE_NORMAL] = low - max_dma; | |
376 | #ifdef CONFIG_HIGHMEM | |
377 | zones_size[ZONE_HIGHMEM] = high - low; | |
378 | #endif | |
379 | } | |
380 | } | |
381 | } | |
382 | ||
383 | zholes_size = get_zholes_size(nid); | |
6f167ec7 DH |
384 | |
385 | free_area_init_node(nid, NODE_DATA(nid), zones_size, start, | |
386 | zholes_size); | |
1da177e4 LT |
387 | } |
388 | return; | |
389 | } | |
390 | ||
391 | void __init set_highmem_pages_init(int bad_ppro) | |
392 | { | |
393 | #ifdef CONFIG_HIGHMEM | |
394 | struct zone *zone; | |
05b79bdc | 395 | struct page *page; |
1da177e4 LT |
396 | |
397 | for_each_zone(zone) { | |
05b79bdc AW |
398 | unsigned long node_pfn, zone_start_pfn, zone_end_pfn; |
399 | ||
1da177e4 LT |
400 | if (!is_highmem(zone)) |
401 | continue; | |
402 | ||
1da177e4 | 403 | zone_start_pfn = zone->zone_start_pfn; |
05b79bdc AW |
404 | zone_end_pfn = zone_start_pfn + zone->spanned_pages; |
405 | ||
406 | printk("Initializing %s for node %d (%08lx:%08lx)\n", | |
407 | zone->name, zone->zone_pgdat->node_id, | |
408 | zone_start_pfn, zone_end_pfn); | |
1da177e4 | 409 | |
05b79bdc AW |
410 | for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) { |
411 | if (!pfn_valid(node_pfn)) | |
412 | continue; | |
413 | page = pfn_to_page(node_pfn); | |
414 | one_highpage_init(page, node_pfn, bad_ppro); | |
1da177e4 LT |
415 | } |
416 | } | |
417 | totalram_pages += totalhigh_pages; | |
418 | #endif | |
419 | } |