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/memblock.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>
34 #include <linux/pfn.h>
35 #include <linux/swap.h>
36 #include <linux/acpi.h>
39 #include <asm/setup.h>
40 #include <asm/mmzone.h>
41 #include <asm/bios_ebda.h>
42 #include <asm/proto.h>
44 struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
45 EXPORT_SYMBOL(node_data);
48 * numa interface - we expect the numa architecture specific code to have
49 * populated the following initialisation.
51 * 1) node_online_map - the map of all nodes configured (online) in the system
52 * 2) node_start_pfn - the starting page frame number for a node
53 * 3) node_end_pfn - the ending page fram number for a node
55 unsigned long node_start_pfn[MAX_NUMNODES] __read_mostly;
56 unsigned long node_end_pfn[MAX_NUMNODES] __read_mostly;
59 #ifdef CONFIG_DISCONTIGMEM
61 * 4) physnode_map - the mapping between a pfn and owning node
62 * physnode_map keeps track of the physical memory layout of a generic
63 * numa node on a 64Mb break (each element of the array will
64 * represent 64Mb of memory and will be marked by the node id. so,
65 * if the first gig is on node 0, and the second gig is on node 1
66 * physnode_map will contain:
68 * physnode_map[0-15] = 0;
69 * physnode_map[16-31] = 1;
70 * physnode_map[32- ] = -1;
72 s8 physnode_map[MAX_ELEMENTS] __read_mostly = { [0 ... (MAX_ELEMENTS - 1)] = -1};
73 EXPORT_SYMBOL(physnode_map);
75 void memory_present(int nid, unsigned long start, unsigned long end)
79 printk(KERN_INFO "Node: %d, start_pfn: %lx, end_pfn: %lx\n",
81 printk(KERN_DEBUG " Setting physnode_map array to node %d for pfns:\n", nid);
82 printk(KERN_DEBUG " ");
83 for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
84 physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
85 printk(KERN_CONT "%lx ", pfn);
87 printk(KERN_CONT "\n");
90 unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
91 unsigned long end_pfn)
93 unsigned long nr_pages = end_pfn - start_pfn;
98 return (nr_pages + 1) * sizeof(struct page);
102 extern unsigned long find_max_low_pfn(void);
103 extern unsigned long highend_pfn, highstart_pfn;
105 #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
107 static void *node_remap_start_vaddr[MAX_NUMNODES];
108 void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
111 * FLAT - support for basic PC memory model with discontig enabled, essentially
112 * a single node with all available processors in it with a flat
115 static int __init get_memcfg_numa_flat(void)
117 printk(KERN_DEBUG "NUMA - single node, flat memory mode\n");
119 node_start_pfn[0] = 0;
120 node_end_pfn[0] = max_pfn;
121 memblock_x86_register_active_regions(0, 0, max_pfn);
123 /* Indicate there is one node available. */
124 nodes_clear(node_online_map);
130 * Find the highest page frame number we have available for the node
132 static void __init propagate_e820_map_node(int nid)
134 if (node_end_pfn[nid] > max_pfn)
135 node_end_pfn[nid] = max_pfn;
137 * if a user has given mem=XXXX, then we need to make sure
138 * that the node _starts_ before that, too, not just ends
140 if (node_start_pfn[nid] > max_pfn)
141 node_start_pfn[nid] = max_pfn;
142 BUG_ON(node_start_pfn[nid] > node_end_pfn[nid]);
146 * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
147 * method. For node zero take this from the bottom of memory, for
148 * subsequent nodes place them at node_remap_start_vaddr which contains
149 * node local data in physically node local memory. See setup_memory()
152 static void __init allocate_pgdat(int nid)
156 NODE_DATA(nid) = alloc_remap(nid, ALIGN(sizeof(pg_data_t), PAGE_SIZE));
157 if (!NODE_DATA(nid)) {
158 unsigned long pgdat_phys;
159 pgdat_phys = memblock_find_in_range(min_low_pfn<<PAGE_SHIFT,
160 max_pfn_mapped<<PAGE_SHIFT,
163 NODE_DATA(nid) = (pg_data_t *)(pfn_to_kaddr(pgdat_phys>>PAGE_SHIFT));
164 memset(buf, 0, sizeof(buf));
165 sprintf(buf, "NODE_DATA %d", nid);
166 memblock_x86_reserve_range(pgdat_phys, pgdat_phys + sizeof(pg_data_t), buf);
168 printk(KERN_DEBUG "allocate_pgdat: node %d NODE_DATA %08lx\n",
169 nid, (unsigned long)NODE_DATA(nid));
173 * Remap memory allocator
175 static unsigned long node_remap_start_pfn[MAX_NUMNODES];
176 static void *node_remap_end_vaddr[MAX_NUMNODES];
177 static void *node_remap_alloc_vaddr[MAX_NUMNODES];
180 * alloc_remap - Allocate remapped memory
181 * @nid: NUMA node to allocate memory from
182 * @size: The size of allocation
184 * Allocate @size bytes from the remap area of NUMA node @nid. The
185 * size of the remap area is predetermined by init_alloc_remap() and
186 * only the callers considered there should call this function. For
187 * more info, please read the comment on top of init_alloc_remap().
189 * The caller must be ready to handle allocation failure from this
190 * function and fall back to regular memory allocator in such cases.
193 * Single CPU early boot context.
196 * Pointer to the allocated memory on success, %NULL on failure.
198 void *alloc_remap(int nid, unsigned long size)
200 void *allocation = node_remap_alloc_vaddr[nid];
202 size = ALIGN(size, L1_CACHE_BYTES);
204 if (!allocation || (allocation + size) > node_remap_end_vaddr[nid])
207 node_remap_alloc_vaddr[nid] += size;
208 memset(allocation, 0, size);
213 #ifdef CONFIG_HIBERNATION
215 * resume_map_numa_kva - add KVA mapping to the temporary page tables created
216 * during resume from hibernation
217 * @pgd_base - temporary resume page directory
219 void resume_map_numa_kva(pgd_t *pgd_base)
223 for_each_online_node(node) {
224 unsigned long start_va, start_pfn, nr_pages, pfn;
226 start_va = (unsigned long)node_remap_start_vaddr[node];
227 start_pfn = node_remap_start_pfn[node];
228 nr_pages = (node_remap_end_vaddr[node] -
229 node_remap_start_vaddr[node]) >> PAGE_SHIFT;
231 printk(KERN_DEBUG "%s: node %d\n", __func__, node);
233 for (pfn = 0; pfn < nr_pages; pfn += PTRS_PER_PTE) {
234 unsigned long vaddr = start_va + (pfn << PAGE_SHIFT);
235 pgd_t *pgd = pgd_base + pgd_index(vaddr);
236 pud_t *pud = pud_offset(pgd, vaddr);
237 pmd_t *pmd = pmd_offset(pud, vaddr);
239 set_pmd(pmd, pfn_pmd(start_pfn + pfn,
240 PAGE_KERNEL_LARGE_EXEC));
242 printk(KERN_DEBUG "%s: %08lx -> pfn %08lx\n",
243 __func__, vaddr, start_pfn + pfn);
250 * init_alloc_remap - Initialize remap allocator for a NUMA node
251 * @nid: NUMA node to initizlie remap allocator for
253 * NUMA nodes may end up without any lowmem. As allocating pgdat and
254 * memmap on a different node with lowmem is inefficient, a special
255 * remap allocator is implemented which can be used by alloc_remap().
257 * For each node, the amount of memory which will be necessary for
258 * pgdat and memmap is calculated and two memory areas of the size are
259 * allocated - one in the node and the other in lowmem; then, the area
260 * in the node is remapped to the lowmem area.
262 * As pgdat and memmap must be allocated in lowmem anyway, this
263 * doesn't waste lowmem address space; however, the actual lowmem
264 * which gets remapped over is wasted. The amount shouldn't be
265 * problematic on machines this feature will be used.
267 * Initialization failure isn't fatal. alloc_remap() is used
268 * opportunistically and the callers will fall back to other memory
269 * allocation mechanisms on failure.
271 static __init void init_alloc_remap(int nid)
273 unsigned long size, pfn;
274 u64 node_pa, remap_pa;
278 * The acpi/srat node info can show hot-add memroy zones where
279 * memory could be added but not currently present.
281 printk(KERN_DEBUG "node %d pfn: [%lx - %lx]\n",
282 nid, node_start_pfn[nid], node_end_pfn[nid]);
283 if (node_start_pfn[nid] > max_pfn)
285 if (!node_end_pfn[nid])
287 if (node_end_pfn[nid] > max_pfn)
288 node_end_pfn[nid] = max_pfn;
290 /* calculate the necessary space aligned to large page size */
291 size = node_memmap_size_bytes(nid, node_start_pfn[nid],
292 min(node_end_pfn[nid], max_pfn));
293 size += ALIGN(sizeof(pg_data_t), PAGE_SIZE);
294 size = ALIGN(size, LARGE_PAGE_BYTES);
296 /* allocate node memory and the lowmem remap area */
297 node_pa = memblock_find_in_range(node_start_pfn[nid] << PAGE_SHIFT,
298 (u64)node_end_pfn[nid] << PAGE_SHIFT,
299 size, LARGE_PAGE_BYTES);
300 if (node_pa == MEMBLOCK_ERROR) {
301 pr_warning("remap_alloc: failed to allocate %lu bytes for node %d\n",
305 memblock_x86_reserve_range(node_pa, node_pa + size, "KVA RAM");
307 remap_pa = memblock_find_in_range(min_low_pfn << PAGE_SHIFT,
308 max_low_pfn << PAGE_SHIFT,
309 size, LARGE_PAGE_BYTES);
310 if (remap_pa == MEMBLOCK_ERROR) {
311 pr_warning("remap_alloc: failed to allocate %lu bytes remap area for node %d\n",
313 memblock_x86_free_range(node_pa, node_pa + size);
316 memblock_x86_reserve_range(remap_pa, remap_pa + size, "KVA PG");
317 remap_va = phys_to_virt(remap_pa);
319 /* perform actual remap */
320 for (pfn = 0; pfn < size >> PAGE_SHIFT; pfn += PTRS_PER_PTE)
321 set_pmd_pfn((unsigned long)remap_va + (pfn << PAGE_SHIFT),
322 (node_pa >> PAGE_SHIFT) + pfn,
325 /* initialize remap allocator parameters */
326 node_remap_start_pfn[nid] = node_pa >> PAGE_SHIFT;
327 node_remap_start_vaddr[nid] = remap_va;
328 node_remap_end_vaddr[nid] = remap_va + size;
329 node_remap_alloc_vaddr[nid] = remap_va;
331 printk(KERN_DEBUG "remap_alloc: node %d [%08llx-%08llx) -> [%p-%p)\n",
332 nid, node_pa, node_pa + size, remap_va, remap_va + size);
335 static int get_memcfg_numaq(void)
337 #ifdef CONFIG_X86_NUMAQ
343 if (numaq_numa_init() < 0) {
344 nodes_clear(numa_nodes_parsed);
345 remove_all_active_ranges();
349 for_each_node_mask(nid, numa_nodes_parsed)
350 node_set_online(nid);
358 static int get_memcfg_from_srat(void)
360 #ifdef CONFIG_ACPI_NUMA
366 if (x86_acpi_numa_init() < 0) {
367 nodes_clear(numa_nodes_parsed);
368 remove_all_active_ranges();
372 for_each_node_mask(nid, numa_nodes_parsed)
373 node_set_online(nid);
381 static void get_memcfg_numa(void)
383 if (get_memcfg_numaq())
385 if (get_memcfg_from_srat())
387 get_memcfg_numa_flat();
390 void __init initmem_init(void)
397 for_each_online_node(nid)
398 init_alloc_remap(nid);
400 #ifdef CONFIG_HIGHMEM
401 highstart_pfn = highend_pfn = max_pfn;
402 if (max_pfn > max_low_pfn)
403 highstart_pfn = max_low_pfn;
404 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
405 pages_to_mb(highend_pfn - highstart_pfn));
406 num_physpages = highend_pfn;
407 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
409 num_physpages = max_low_pfn;
410 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
412 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
413 pages_to_mb(max_low_pfn));
414 printk(KERN_DEBUG "max_low_pfn = %lx, highstart_pfn = %lx\n",
415 max_low_pfn, highstart_pfn);
417 printk(KERN_DEBUG "Low memory ends at vaddr %08lx\n",
418 (ulong) pfn_to_kaddr(max_low_pfn));
419 for_each_online_node(nid)
422 printk(KERN_DEBUG "High memory starts at vaddr %08lx\n",
423 (ulong) pfn_to_kaddr(highstart_pfn));
424 for_each_online_node(nid)
425 propagate_e820_map_node(nid);
427 for_each_online_node(nid) {
428 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
429 NODE_DATA(nid)->node_id = nid;
432 setup_bootmem_allocator();
435 #ifdef CONFIG_MEMORY_HOTPLUG
436 static int paddr_to_nid(u64 addr)
439 unsigned long pfn = PFN_DOWN(addr);
442 if (node_start_pfn[nid] <= pfn &&
443 pfn < node_end_pfn[nid])
450 * This function is used to ask node id BEFORE memmap and mem_section's
451 * initialization (pfn_to_nid() can't be used yet).
452 * If _PXM is not defined on ACPI's DSDT, node id must be found by this.
454 int memory_add_physaddr_to_nid(u64 addr)
456 int nid = paddr_to_nid(addr);
457 return (nid >= 0) ? nid : 0;
460 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
463 /* temporary shim, will go away soon */
464 int __init numa_add_memblk(int nid, u64 start, u64 end)
466 unsigned long start_pfn = start >> PAGE_SHIFT;
467 unsigned long end_pfn = end >> PAGE_SHIFT;
469 printk(KERN_DEBUG "nid %d start_pfn %08lx end_pfn %08lx\n",
470 nid, start_pfn, end_pfn);
472 if (start >= (u64)max_pfn << PAGE_SHIFT) {
473 printk(KERN_INFO "Ignoring SRAT pfns: %08lx - %08lx\n",
478 node_set_online(nid);
479 memblock_x86_register_active_regions(nid, start_pfn,
480 min(end_pfn, max_pfn));
482 if (!node_has_online_mem(nid)) {
483 node_start_pfn[nid] = start_pfn;
484 node_end_pfn[nid] = end_pfn;
486 node_start_pfn[nid] = min(node_start_pfn[nid], start_pfn);
487 node_end_pfn[nid] = max(node_end_pfn[nid], end_pfn);
492 /* temporary shim, will go away soon */
493 void __init numa_set_distance(int from, int to, int distance)