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.
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>
31 #include <linux/module.h>
32 #include <linux/kexec.h>
33 #include <linux/pfn.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 specific 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 highend_pfn, highstart_pfn;
105 extern unsigned long max_low_pfn;
106 extern unsigned long totalram_pages;
107 extern unsigned long totalhigh_pages;
109 #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
111 unsigned long node_remap_start_pfn[MAX_NUMNODES];
112 unsigned long node_remap_size[MAX_NUMNODES];
113 unsigned long node_remap_offset[MAX_NUMNODES];
114 void *node_remap_start_vaddr[MAX_NUMNODES];
115 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
117 void *node_remap_end_vaddr[MAX_NUMNODES];
118 void *node_remap_alloc_vaddr[MAX_NUMNODES];
119 static unsigned long kva_start_pfn;
120 static unsigned long kva_pages;
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 BUG_ON(node_start_pfn[nid] > node_end_pfn[nid]);
159 * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
160 * method. For node zero take this from the bottom of memory, for
161 * subsequent nodes place them at node_remap_start_vaddr which contains
162 * node local data in physically node local memory. See setup_memory()
165 static void __init allocate_pgdat(int nid)
167 if (nid && node_has_online_mem(nid))
168 NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
170 NODE_DATA(nid) = (pg_data_t *)(pfn_to_kaddr(min_low_pfn));
171 min_low_pfn += PFN_UP(sizeof(pg_data_t));
175 void *alloc_remap(int nid, unsigned long size)
177 void *allocation = node_remap_alloc_vaddr[nid];
179 size = ALIGN(size, L1_CACHE_BYTES);
181 if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
184 node_remap_alloc_vaddr[nid] += size;
185 memset(allocation, 0, size);
190 void __init remap_numa_kva(void)
196 for_each_online_node(node) {
197 for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
198 vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
199 set_pmd_pfn((ulong) vaddr,
200 node_remap_start_pfn[node] + pfn,
206 static unsigned long calculate_numa_remap_pages(void)
209 unsigned long size, reserve_pages = 0;
212 for_each_online_node(nid) {
213 unsigned old_end_pfn = node_end_pfn[nid];
216 * The acpi/srat node info can show hot-add memroy zones
217 * where memory could be added but not currently present.
219 if (node_start_pfn[nid] > max_pfn)
221 if (node_end_pfn[nid] > max_pfn)
222 node_end_pfn[nid] = max_pfn;
224 /* ensure the remap includes space for the pgdat. */
225 size = node_remap_size[nid] + sizeof(pg_data_t);
227 /* convert size to large (pmd size) pages, rounding up */
228 size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
229 /* now the roundup is correct, convert to PAGE_SIZE pages */
230 size = size * PTRS_PER_PTE;
233 * Validate the region we are allocating only contains valid
236 for (pfn = node_end_pfn[nid] - size;
237 pfn < node_end_pfn[nid]; pfn++)
238 if (!page_is_ram(pfn))
241 if (pfn != node_end_pfn[nid])
244 printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
246 node_remap_size[nid] = size;
247 node_remap_offset[nid] = reserve_pages;
248 reserve_pages += size;
249 printk("Shrinking node %d from %ld pages to %ld pages\n",
250 nid, node_end_pfn[nid], node_end_pfn[nid] - size);
252 if (node_end_pfn[nid] & (PTRS_PER_PTE-1)) {
254 * Align node_end_pfn[] and node_remap_start_pfn[] to
255 * pmd boundary. remap_numa_kva will barf otherwise.
257 printk("Shrinking node %d further by %ld pages for proper alignment\n",
258 nid, node_end_pfn[nid] & (PTRS_PER_PTE-1));
259 size += node_end_pfn[nid] & (PTRS_PER_PTE-1);
262 node_end_pfn[nid] -= size;
263 node_remap_start_pfn[nid] = node_end_pfn[nid];
264 shrink_active_range(nid, old_end_pfn, node_end_pfn[nid]);
266 printk("Reserving total of %ld pages for numa KVA remap\n",
268 return reserve_pages;
271 extern void setup_bootmem_allocator(void);
272 unsigned long __init setup_memory(void)
275 unsigned long system_start_pfn, system_max_low_pfn;
278 * When mapping a NUMA machine we allocate the node_mem_map arrays
279 * from node local memory. They are then mapped directly into KVA
280 * between zone normal and vmalloc space. Calculate the size of
281 * this space and use it to adjust the boundry between ZONE_NORMAL
287 kva_pages = calculate_numa_remap_pages();
289 /* partially used pages are not usable - thus round upwards */
290 system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
292 kva_start_pfn = find_max_low_pfn() - kva_pages;
294 #ifdef CONFIG_BLK_DEV_INITRD
295 /* Numa kva area is below the initrd */
296 if (LOADER_TYPE && INITRD_START)
297 kva_start_pfn = PFN_DOWN(INITRD_START) - kva_pages;
299 kva_start_pfn -= kva_start_pfn & (PTRS_PER_PTE-1);
301 system_max_low_pfn = max_low_pfn = find_max_low_pfn();
302 printk("kva_start_pfn ~ %ld find_max_low_pfn() ~ %ld\n",
303 kva_start_pfn, max_low_pfn);
304 printk("max_pfn = %ld\n", max_pfn);
305 #ifdef CONFIG_HIGHMEM
306 highstart_pfn = highend_pfn = max_pfn;
307 if (max_pfn > system_max_low_pfn)
308 highstart_pfn = system_max_low_pfn;
309 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
310 pages_to_mb(highend_pfn - highstart_pfn));
311 num_physpages = highend_pfn;
312 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
314 num_physpages = system_max_low_pfn;
315 high_memory = (void *) __va(system_max_low_pfn * PAGE_SIZE - 1) + 1;
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 kva_start_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 for_each_online_node(nid)
342 find_max_pfn_node(nid);
344 memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
345 NODE_DATA(0)->bdata = &node0_bdata;
346 setup_bootmem_allocator();
350 void __init numa_kva_reserve(void)
352 reserve_bootmem(PFN_PHYS(kva_start_pfn),PFN_PHYS(kva_pages));
355 void __init zone_sizes_init(void)
358 unsigned long max_zone_pfns[MAX_NR_ZONES];
359 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
360 max_zone_pfns[ZONE_DMA] =
361 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
362 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
363 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
365 /* If SRAT has not registered memory, register it now */
366 if (find_max_pfn_with_active_regions() == 0) {
367 for_each_online_node(nid) {
368 if (node_has_online_mem(nid))
369 add_active_range(nid, node_start_pfn[nid],
374 free_area_init_nodes(max_zone_pfns);
378 void __init set_highmem_pages_init(int bad_ppro)
380 #ifdef CONFIG_HIGHMEM
384 for_each_zone(zone) {
385 unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
387 if (!is_highmem(zone))
390 zone_start_pfn = zone->zone_start_pfn;
391 zone_end_pfn = zone_start_pfn + zone->spanned_pages;
393 printk("Initializing %s for node %d (%08lx:%08lx)\n",
394 zone->name, zone_to_nid(zone),
395 zone_start_pfn, zone_end_pfn);
397 for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
398 if (!pfn_valid(node_pfn))
400 page = pfn_to_page(node_pfn);
401 add_one_highpage_init(page, node_pfn, bad_ppro);
404 totalram_pages += totalhigh_pages;
408 #ifdef CONFIG_MEMORY_HOTPLUG
409 int paddr_to_nid(u64 addr)
412 unsigned long pfn = PFN_DOWN(addr);
415 if (node_start_pfn[nid] <= pfn &&
416 pfn < node_end_pfn[nid])
423 * This function is used to ask node id BEFORE memmap and mem_section's
424 * initialization (pfn_to_nid() can't be used yet).
425 * If _PXM is not defined on ACPI's DSDT, node id must be found by this.
427 int memory_add_physaddr_to_nid(u64 addr)
429 int nid = paddr_to_nid(addr);
430 return (nid >= 0) ? nid : 0;
433 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);