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/mm.h>
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>
34#include <linux/swap.h>
35
36#include <asm/e820.h>
37#include <asm/setup.h>
38#include <asm/mmzone.h>
39#include <bios_ebda.h>
40
41struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
42EXPORT_SYMBOL(node_data);
43bootmem_data_t node0_bdata;
44
45/*
46 * numa interface - we expect the numa architecture specific code to have
47 *                  populated the following initialisation.
48 *
49 * 1) node_online_map  - the map of all nodes configured (online) in the system
50 * 2) node_start_pfn   - the starting page frame number for a node
51 * 3) node_end_pfn     - the ending page fram number for a node
52 */
53unsigned long node_start_pfn[MAX_NUMNODES] __read_mostly;
54unsigned long node_end_pfn[MAX_NUMNODES] __read_mostly;
55
56
57#ifdef CONFIG_DISCONTIGMEM
58/*
59 * 4) physnode_map     - the mapping between a pfn and owning node
60 * 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 */
70s8 physnode_map[MAX_ELEMENTS] __read_mostly = { [0 ... (MAX_ELEMENTS - 1)] = -1};
71EXPORT_SYMBOL(physnode_map);
72
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}
98#endif
99
100extern unsigned long find_max_low_pfn(void);
101extern void add_one_highpage_init(struct page *, int, int);
102extern unsigned long highend_pfn, highstart_pfn;
103
104#define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
105
106unsigned long node_remap_start_pfn[MAX_NUMNODES];
107unsigned long node_remap_size[MAX_NUMNODES];
108unsigned long node_remap_offset[MAX_NUMNODES];
109void *node_remap_start_vaddr[MAX_NUMNODES];
110void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
111
112void *node_remap_end_vaddr[MAX_NUMNODES];
113void *node_remap_alloc_vaddr[MAX_NUMNODES];
114static unsigned long kva_start_pfn;
115static unsigned long kva_pages;
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 */
121int __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 */
140static 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	if (node_start_pfn[nid] > max_pfn)
145		node_start_pfn[nid] = max_pfn;
146	BUG_ON(node_start_pfn[nid] > node_end_pfn[nid]);
147}
148
149/*
150 * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
151 * method.  For node zero take this from the bottom of memory, for
152 * subsequent nodes place them at node_remap_start_vaddr which contains
153 * node local data in physically node local memory.  See setup_memory()
154 * for details.
155 */
156static void __init allocate_pgdat(int nid)
157{
158	if (nid && node_has_online_mem(nid))
159		NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
160	else {
161		NODE_DATA(nid) = (pg_data_t *)(pfn_to_kaddr(min_low_pfn));
162		min_low_pfn += PFN_UP(sizeof(pg_data_t));
163	}
164}
165
166void *alloc_remap(int nid, unsigned long size)
167{
168	void *allocation = node_remap_alloc_vaddr[nid];
169
170	size = ALIGN(size, L1_CACHE_BYTES);
171
172	if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
173		return 0;
174
175	node_remap_alloc_vaddr[nid] += size;
176	memset(allocation, 0, size);
177
178	return allocation;
179}
180
181void __init remap_numa_kva(void)
182{
183	void *vaddr;
184	unsigned long pfn;
185	int node;
186
187	for_each_online_node(node) {
188		for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
189			vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
190			set_pmd_pfn((ulong) vaddr,
191				node_remap_start_pfn[node] + pfn,
192				PAGE_KERNEL_LARGE);
193		}
194	}
195}
196
197static unsigned long calculate_numa_remap_pages(void)
198{
199	int nid;
200	unsigned long size, reserve_pages = 0;
201	unsigned long pfn;
202
203	for_each_online_node(nid) {
204		unsigned old_end_pfn = node_end_pfn[nid];
205
206		/*
207		 * The acpi/srat node info can show hot-add memroy zones
208		 * where memory could be added but not currently present.
209		 */
210		if (node_start_pfn[nid] > max_pfn)
211			continue;
212		if (node_end_pfn[nid] > max_pfn)
213			node_end_pfn[nid] = max_pfn;
214
215		/* ensure the remap includes space for the pgdat. */
216		size = node_remap_size[nid] + sizeof(pg_data_t);
217
218		/* convert size to large (pmd size) pages, rounding up */
219		size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
220		/* now the roundup is correct, convert to PAGE_SIZE pages */
221		size = size * PTRS_PER_PTE;
222
223		/*
224		 * Validate the region we are allocating only contains valid
225		 * pages.
226		 */
227		for (pfn = node_end_pfn[nid] - size;
228		     pfn < node_end_pfn[nid]; pfn++)
229			if (!page_is_ram(pfn))
230				break;
231
232		if (pfn != node_end_pfn[nid])
233			size = 0;
234
235		printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
236				size, nid);
237		node_remap_size[nid] = size;
238		node_remap_offset[nid] = reserve_pages;
239		reserve_pages += size;
240		printk("Shrinking node %d from %ld pages to %ld pages\n",
241			nid, node_end_pfn[nid], node_end_pfn[nid] - size);
242
243		if (node_end_pfn[nid] & (PTRS_PER_PTE-1)) {
244			/*
245			 * Align node_end_pfn[] and node_remap_start_pfn[] to
246			 * pmd boundary. remap_numa_kva will barf otherwise.
247			 */
248			printk("Shrinking node %d further by %ld pages for proper alignment\n",
249				nid, node_end_pfn[nid] & (PTRS_PER_PTE-1));
250			size +=  node_end_pfn[nid] & (PTRS_PER_PTE-1);
251		}
252
253		node_end_pfn[nid] -= size;
254		node_remap_start_pfn[nid] = node_end_pfn[nid];
255		shrink_active_range(nid, old_end_pfn, node_end_pfn[nid]);
256	}
257	printk("Reserving total of %ld pages for numa KVA remap\n",
258			reserve_pages);
259	return reserve_pages;
260}
261
262extern void setup_bootmem_allocator(void);
263unsigned long __init setup_memory(void)
264{
265	int nid;
266	unsigned long system_start_pfn, system_max_low_pfn;
267
268	/*
269	 * When mapping a NUMA machine we allocate the node_mem_map arrays
270	 * from node local memory.  They are then mapped directly into KVA
271	 * between zone normal and vmalloc space.  Calculate the size of
272	 * this space and use it to adjust the boundry between ZONE_NORMAL
273	 * and ZONE_HIGHMEM.
274	 */
275	find_max_pfn();
276	get_memcfg_numa();
277
278	kva_pages = calculate_numa_remap_pages();
279
280	/* partially used pages are not usable - thus round upwards */
281	system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
282
283	kva_start_pfn = find_max_low_pfn() - kva_pages;
284
285#ifdef CONFIG_BLK_DEV_INITRD
286	/* Numa kva area is below the initrd */
287	if (LOADER_TYPE && INITRD_START)
288		kva_start_pfn = PFN_DOWN(INITRD_START)  - kva_pages;
289#endif
290	kva_start_pfn -= kva_start_pfn & (PTRS_PER_PTE-1);
291
292	system_max_low_pfn = max_low_pfn = find_max_low_pfn();
293	printk("kva_start_pfn ~ %ld find_max_low_pfn() ~ %ld\n",
294		kva_start_pfn, max_low_pfn);
295	printk("max_pfn = %ld\n", max_pfn);
296#ifdef CONFIG_HIGHMEM
297	highstart_pfn = highend_pfn = max_pfn;
298	if (max_pfn > system_max_low_pfn)
299		highstart_pfn = system_max_low_pfn;
300	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
301	       pages_to_mb(highend_pfn - highstart_pfn));
302	num_physpages = highend_pfn;
303	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
304#else
305	num_physpages = system_max_low_pfn;
306	high_memory = (void *) __va(system_max_low_pfn * PAGE_SIZE - 1) + 1;
307#endif
308	printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
309			pages_to_mb(system_max_low_pfn));
310	printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
311			min_low_pfn, max_low_pfn, highstart_pfn);
312
313	printk("Low memory ends at vaddr %08lx\n",
314			(ulong) pfn_to_kaddr(max_low_pfn));
315	for_each_online_node(nid) {
316		node_remap_start_vaddr[nid] = pfn_to_kaddr(
317				kva_start_pfn + node_remap_offset[nid]);
318		/* Init the node remap allocator */
319		node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
320			(node_remap_size[nid] * PAGE_SIZE);
321		node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
322			ALIGN(sizeof(pg_data_t), PAGE_SIZE);
323
324		allocate_pgdat(nid);
325		printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
326			(ulong) node_remap_start_vaddr[nid],
327			(ulong) pfn_to_kaddr(highstart_pfn
328			   + node_remap_offset[nid] + node_remap_size[nid]));
329	}
330	printk("High memory starts at vaddr %08lx\n",
331			(ulong) pfn_to_kaddr(highstart_pfn));
332	for_each_online_node(nid)
333		find_max_pfn_node(nid);
334
335	memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
336	NODE_DATA(0)->bdata = &node0_bdata;
337	setup_bootmem_allocator();
338	return max_low_pfn;
339}
340
341void __init numa_kva_reserve(void)
342{
343	reserve_bootmem(PFN_PHYS(kva_start_pfn),PFN_PHYS(kva_pages));
344}
345
346void __init zone_sizes_init(void)
347{
348	int nid;
349	unsigned long max_zone_pfns[MAX_NR_ZONES];
350	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
351	max_zone_pfns[ZONE_DMA] =
352		virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
353	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
354#ifdef CONFIG_HIGHMEM
355	max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
356#endif
357
358	/* If SRAT has not registered memory, register it now */
359	if (find_max_pfn_with_active_regions() == 0) {
360		for_each_online_node(nid) {
361			if (node_has_online_mem(nid))
362				add_active_range(nid, node_start_pfn[nid],
363							node_end_pfn[nid]);
364		}
365	}
366
367	free_area_init_nodes(max_zone_pfns);
368	return;
369}
370
371void __init set_highmem_pages_init(int bad_ppro)
372{
373#ifdef CONFIG_HIGHMEM
374	struct zone *zone;
375	struct page *page;
376
377	for_each_zone(zone) {
378		unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
379
380		if (!is_highmem(zone))
381			continue;
382
383		zone_start_pfn = zone->zone_start_pfn;
384		zone_end_pfn = zone_start_pfn + zone->spanned_pages;
385
386		printk("Initializing %s for node %d (%08lx:%08lx)\n",
387				zone->name, zone_to_nid(zone),
388				zone_start_pfn, zone_end_pfn);
389
390		for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
391			if (!pfn_valid(node_pfn))
392				continue;
393			page = pfn_to_page(node_pfn);
394			add_one_highpage_init(page, node_pfn, bad_ppro);
395		}
396	}
397	totalram_pages += totalhigh_pages;
398#endif
399}
400
401#ifdef CONFIG_MEMORY_HOTPLUG
402int paddr_to_nid(u64 addr)
403{
404	int nid;
405	unsigned long pfn = PFN_DOWN(addr);
406
407	for_each_node(nid)
408		if (node_start_pfn[nid] <= pfn &&
409		    pfn < node_end_pfn[nid])
410			return nid;
411
412	return -1;
413}
414
415/*
416 * This function is used to ask node id BEFORE memmap and mem_section's
417 * initialization (pfn_to_nid() can't be used yet).
418 * If _PXM is not defined on ACPI's DSDT, node id must be found by this.
419 */
420int memory_add_physaddr_to_nid(u64 addr)
421{
422	int nid = paddr_to_nid(addr);
423	return (nid >= 0) ? nid : 0;
424}
425
426EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
427#endif
428