2 * Written by: Patricia Gaughen <gone@us.ibm.com>, IBM Corporation
3 * August 2002: added remote node KVA remap - Martin J. Bligh
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.
25 #include <linux/config.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 <linux/module.h>
33 #include <linux/kexec.h>
36 #include <asm/setup.h>
37 #include <asm/mmzone.h>
38 #include <bios_ebda.h>
40 struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
41 EXPORT_SYMBOL(node_data);
42 bootmem_data_t node0_bdata;
45 * numa interface - we expect the numa architecture specfic code to have
46 * populated the following initialisation.
48 * 1) node_online_map - the map of all nodes configured (online) in the system
49 * 2) node_start_pfn - the starting page frame number for a node
50 * 3) node_end_pfn - the ending page fram number for a node
52 unsigned long node_start_pfn[MAX_NUMNODES] __read_mostly;
53 unsigned long node_end_pfn[MAX_NUMNODES] __read_mostly;
56 #ifdef CONFIG_DISCONTIGMEM
58 * 4) physnode_map - the mapping between a pfn and owning node
59 * 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:
65 * physnode_map[0-3] = 0;
66 * physnode_map[4-7] = 1;
67 * physnode_map[8- ] = -1;
69 s8 physnode_map[MAX_ELEMENTS] __read_mostly = { [0 ... (MAX_ELEMENTS - 1)] = -1};
70 EXPORT_SYMBOL(physnode_map);
72 void memory_present(int nid, unsigned long start, unsigned long end)
76 printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
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;
87 unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
88 unsigned long end_pfn)
90 unsigned long nr_pages = end_pfn - start_pfn;
95 return (nr_pages + 1) * sizeof(struct page);
99 extern unsigned long find_max_low_pfn(void);
100 extern void find_max_pfn(void);
101 extern void add_one_highpage_init(struct page *, int, int);
103 extern struct e820map e820;
104 extern unsigned long init_pg_tables_end;
105 extern unsigned long highend_pfn, highstart_pfn;
106 extern unsigned long max_low_pfn;
107 extern unsigned long totalram_pages;
108 extern unsigned long totalhigh_pages;
110 #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
112 unsigned long node_remap_start_pfn[MAX_NUMNODES];
113 unsigned long node_remap_size[MAX_NUMNODES];
114 unsigned long node_remap_offset[MAX_NUMNODES];
115 void *node_remap_start_vaddr[MAX_NUMNODES];
116 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
118 void *node_remap_end_vaddr[MAX_NUMNODES];
119 void *node_remap_alloc_vaddr[MAX_NUMNODES];
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
126 int __init get_memcfg_numa_flat(void)
128 printk("NUMA - single node, flat memory mode\n");
130 /* Run the memory configuration and find the top of memory. */
132 node_start_pfn[0] = 0;
133 node_end_pfn[0] = max_pfn;
134 memory_present(0, 0, max_pfn);
136 /* Indicate there is one node available. */
137 nodes_clear(node_online_map);
143 * Find the highest page frame number we have available for the node
145 static void __init find_max_pfn_node(int nid)
147 if (node_end_pfn[nid] > max_pfn)
148 node_end_pfn[nid] = max_pfn;
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
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])
159 /* Find the owning node for a pfn. */
160 int early_pfn_to_nid(unsigned long pfn)
165 if (node_end_pfn[nid] == 0)
167 if (node_start_pfn[nid] <= pfn && node_end_pfn[nid] >= pfn)
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()
181 static void __init allocate_pgdat(int nid)
183 if (nid && node_has_online_mem(nid))
184 NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
186 NODE_DATA(nid) = (pg_data_t *)(__va(min_low_pfn << PAGE_SHIFT));
187 min_low_pfn += PFN_UP(sizeof(pg_data_t));
191 void *alloc_remap(int nid, unsigned long size)
193 void *allocation = node_remap_alloc_vaddr[nid];
195 size = ALIGN(size, L1_CACHE_BYTES);
197 if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
200 node_remap_alloc_vaddr[nid] += size;
201 memset(allocation, 0, size);
206 void __init remap_numa_kva(void)
212 for_each_online_node(node) {
213 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,
222 static unsigned long calculate_numa_remap_pages(void)
225 unsigned long size, reserve_pages = 0;
228 for_each_online_node(nid) {
230 * The acpi/srat node info can show hot-add memroy zones
231 * where memory could be added but not currently present.
233 if (node_start_pfn[nid] > max_pfn)
235 if (node_end_pfn[nid] > max_pfn)
236 node_end_pfn[nid] = max_pfn;
238 /* ensure the remap includes space for the pgdat. */
239 size = node_remap_size[nid] + sizeof(pg_data_t);
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;
247 * Validate the region we are allocating only contains valid
250 for (pfn = node_end_pfn[nid] - size;
251 pfn < node_end_pfn[nid]; pfn++)
252 if (!page_is_ram(pfn))
255 if (pfn != node_end_pfn[nid])
258 printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
260 node_remap_size[nid] = size;
261 node_remap_offset[nid] = reserve_pages;
262 reserve_pages += size;
263 printk("Shrinking node %d from %ld pages to %ld pages\n",
264 nid, node_end_pfn[nid], node_end_pfn[nid] - size);
266 if (node_end_pfn[nid] & (PTRS_PER_PTE-1)) {
268 * Align node_end_pfn[] and node_remap_start_pfn[] to
269 * pmd boundary. remap_numa_kva will barf otherwise.
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);
276 node_end_pfn[nid] -= size;
277 node_remap_start_pfn[nid] = node_end_pfn[nid];
279 printk("Reserving total of %ld pages for numa KVA remap\n",
281 return reserve_pages;
284 extern void setup_bootmem_allocator(void);
285 unsigned long __init setup_memory(void)
288 unsigned long system_start_pfn, system_max_low_pfn;
289 unsigned long reserve_pages;
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
301 reserve_pages = calculate_numa_remap_pages();
303 /* partially used pages are not usable - thus round upwards */
304 system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
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));
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);
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(
326 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);
334 printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
335 (ulong) node_remap_start_vaddr[nid],
336 (ulong) pfn_to_kaddr(highstart_pfn
337 + node_remap_offset[nid] + node_remap_size[nid]));
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);
345 memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
346 NODE_DATA(0)->bdata = &node0_bdata;
347 setup_bootmem_allocator();
351 void __init zone_sizes_init(void)
356 * Insert nodes into pgdat_list backward so they appear in order.
357 * Clobber node 0's links and NULL out pgdat_list before starting.
360 for (nid = MAX_NUMNODES - 1; nid >= 0; nid--) {
361 if (!node_online(nid))
363 NODE_DATA(nid)->pgdat_next = pgdat_list;
364 pgdat_list = NODE_DATA(nid);
367 for_each_online_node(nid) {
368 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
369 unsigned long *zholes_size;
370 unsigned int max_dma;
372 unsigned long low = max_low_pfn;
373 unsigned long start = node_start_pfn[nid];
374 unsigned long high = node_end_pfn[nid];
376 max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
378 if (node_has_online_mem(nid)){
380 #ifdef CONFIG_HIGHMEM
381 BUG_ON(start > high);
382 zones_size[ZONE_HIGHMEM] = high - start;
386 zones_size[ZONE_DMA] = low;
388 BUG_ON(max_dma > low);
390 zones_size[ZONE_DMA] = max_dma;
391 zones_size[ZONE_NORMAL] = low - max_dma;
392 #ifdef CONFIG_HIGHMEM
393 zones_size[ZONE_HIGHMEM] = high - low;
399 zholes_size = get_zholes_size(nid);
401 free_area_init_node(nid, NODE_DATA(nid), zones_size, start,
407 void __init set_highmem_pages_init(int bad_ppro)
409 #ifdef CONFIG_HIGHMEM
413 for_each_zone(zone) {
414 unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
416 if (!is_highmem(zone))
419 zone_start_pfn = zone->zone_start_pfn;
420 zone_end_pfn = zone_start_pfn + zone->spanned_pages;
422 printk("Initializing %s for node %d (%08lx:%08lx)\n",
423 zone->name, zone->zone_pgdat->node_id,
424 zone_start_pfn, zone_end_pfn);
426 for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
427 if (!pfn_valid(node_pfn))
429 page = pfn_to_page(node_pfn);
430 add_one_highpage_init(page, node_pfn, bad_ppro);
433 totalram_pages += totalhigh_pages;