numa.c 7.78 KB
Newer Older
Andi Kleen's avatar
Andi Kleen committed
1 2 3 4 5 6 7 8 9 10 11
/* 
 * Generic VM initialization for x86-64 NUMA setups.
 * Copyright 2002,2003 Andi Kleen, SuSE Labs.
 */ 
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/init.h>
#include <linux/bootmem.h>
#include <linux/mmzone.h>
#include <linux/ctype.h>
Andi Kleen's avatar
Andi Kleen committed
12
#include <linux/module.h>
Matthew Dobson's avatar
Matthew Dobson committed
13 14
#include <linux/nodemask.h>

Andi Kleen's avatar
Andi Kleen committed
15 16 17 18
#include <asm/e820.h>
#include <asm/proto.h>
#include <asm/dma.h>
#include <asm/numa.h>
19
#include <asm/acpi.h>
Andi Kleen's avatar
Andi Kleen committed
20

Andi Kleen's avatar
Andi Kleen committed
21
#ifndef Dprintk
Andi Kleen's avatar
Andi Kleen committed
22
#define Dprintk(x...)
Andi Kleen's avatar
Andi Kleen committed
23
#endif
Andi Kleen's avatar
Andi Kleen committed
24

25
struct pglist_data *node_data[MAX_NUMNODES];
Andi Kleen's avatar
Andi Kleen committed
26 27 28 29 30
bootmem_data_t plat_node_bdata[MAX_NUMNODES];

int memnode_shift;
u8  memnodemap[NODEMAPSIZE];

31 32
#define NUMA_NO_NODE 0xff
unsigned char cpu_to_node[NR_CPUS] = { [0 ... NR_CPUS-1] = NUMA_NO_NODE };
33
cpumask_t     node_to_cpumask[MAX_NUMNODES];
Andi Kleen's avatar
Andi Kleen committed
34

35
int numa_off __initdata;
Andi Kleen's avatar
Andi Kleen committed
36 37 38

unsigned long nodes_present; 

Andi Kleen's avatar
Andi Kleen committed
39
int __init compute_hash_shift(struct node *nodes)
Andi Kleen's avatar
Andi Kleen committed
40 41 42 43 44 45 46 47
{
	int i; 
	int shift = 24;
	u64 addr;
	
	/* When in doubt use brute force. */
	while (shift < 48) { 
		memset(memnodemap,0xff,sizeof(*memnodemap) * NODEMAPSIZE); 
Andi Kleen's avatar
Andi Kleen committed
48 49 50
		for (i = 0; i < numnodes; i++) { 
			if (nodes[i].start == nodes[i].end) 
				continue;
Andi Kleen's avatar
Andi Kleen committed
51 52 53
			for (addr = nodes[i].start; 
			     addr < nodes[i].end; 
			     addr += (1UL << shift)) {
Andi Kleen's avatar
Andi Kleen committed
54 55 56 57
				if (memnodemap[addr >> shift] != 0xff && 
				    memnodemap[addr >> shift] != i) { 
					printk(KERN_INFO 
					    "node %d shift %d addr %Lx conflict %d\n", 
Andi Kleen's avatar
Andi Kleen committed
58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
					       i, shift, addr, memnodemap[addr>>shift]);
					goto next; 
				} 
				memnodemap[addr >> shift] = i; 
			} 
		} 
		return shift; 
	next:
		shift++; 
	} 
	memset(memnodemap,0,sizeof(*memnodemap) * NODEMAPSIZE); 
	return -1; 
}

/* Initialize bootmem allocator for a node */
void __init setup_node_bootmem(int nodeid, unsigned long start, unsigned long end)
{ 
	unsigned long start_pfn, end_pfn, bootmap_pages, bootmap_size, bootmap_start; 
	unsigned long nodedata_phys;
	const int pgdat_size = round_up(sizeof(pg_data_t), PAGE_SIZE);

	start = round_up(start, ZONE_ALIGN); 

	printk("Bootmem setup node %d %016lx-%016lx\n", nodeid, start, end);

	start_pfn = start >> PAGE_SHIFT;
	end_pfn = end >> PAGE_SHIFT;

	nodedata_phys = find_e820_area(start, end, pgdat_size); 
	if (nodedata_phys == -1L) 
		panic("Cannot find memory pgdat in node %d\n", nodeid);

	Dprintk("nodedata_phys %lx\n", nodedata_phys); 

	node_data[nodeid] = phys_to_virt(nodedata_phys);
	memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t));
	NODE_DATA(nodeid)->bdata = &plat_node_bdata[nodeid];
	NODE_DATA(nodeid)->node_start_pfn = start_pfn;
96
	NODE_DATA(nodeid)->node_spanned_pages = end_pfn - start_pfn;
Andi Kleen's avatar
Andi Kleen committed
97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113

	/* Find a place for the bootmem map */
	bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); 
	bootmap_start = round_up(nodedata_phys + pgdat_size, PAGE_SIZE);
	bootmap_start = find_e820_area(bootmap_start, end, bootmap_pages<<PAGE_SHIFT);
	if (bootmap_start == -1L) 
		panic("Not enough continuous space for bootmap on node %d", nodeid); 
	Dprintk("bootmap start %lu pages %lu\n", bootmap_start, bootmap_pages); 
	
	bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
					 bootmap_start >> PAGE_SHIFT, 
					 start_pfn, end_pfn); 

	e820_bootmem_free(NODE_DATA(nodeid), start, end);

	reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys, pgdat_size); 
	reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, bootmap_pages<<PAGE_SHIFT);
114
	if (nodeid + 1 > numnodes)
Andi Kleen's avatar
Andi Kleen committed
115
		numnodes = nodeid + 1;
116
	node_set_online(nodeid);
Andi Kleen's avatar
Andi Kleen committed
117 118 119 120 121 122 123 124 125 126 127 128 129 130
} 

/* Initialize final allocator for a zone */
void __init setup_node_zones(int nodeid)
{ 
	unsigned long start_pfn, end_pfn; 
	unsigned long zones[MAX_NR_ZONES];
	unsigned long dma_end_pfn;

	memset(zones, 0, sizeof(unsigned long) * MAX_NR_ZONES); 

	start_pfn = node_start_pfn(nodeid);
	end_pfn = node_end_pfn(nodeid);

Andi Kleen's avatar
Andi Kleen committed
131
	Dprintk(KERN_INFO "setting up node %d %lx-%lx\n", nodeid, start_pfn, end_pfn);
Andi Kleen's avatar
Andi Kleen committed
132 133 134 135 136 137 138 139 140 141
	
	/* All nodes > 0 have a zero length zone DMA */ 
	dma_end_pfn = __pa(MAX_DMA_ADDRESS) >> PAGE_SHIFT; 
	if (start_pfn < dma_end_pfn) { 
		zones[ZONE_DMA] = dma_end_pfn - start_pfn;
		zones[ZONE_NORMAL] = end_pfn - dma_end_pfn; 
	} else { 
		zones[ZONE_NORMAL] = end_pfn - start_pfn; 
	} 
    
142
	free_area_init_node(nodeid, NODE_DATA(nodeid), zones,
Andi Kleen's avatar
Andi Kleen committed
143 144 145
			    start_pfn, NULL); 
} 

Andi Kleen's avatar
Andi Kleen committed
146 147 148 149 150 151 152 153 154
void __init numa_init_array(void)
{
	int rr, i;
	/* There are unfortunately some poorly designed mainboards around
	   that only connect memory to a single CPU. This breaks the 1:1 cpu->node
	   mapping. To avoid this fill in the mapping for all possible
	   CPUs, as the number of CPUs is not known yet. 
	   We round robin the existing nodes. */
	rr = 0;
155
	for (i = 0; i < NR_CPUS; i++) {
156 157
		if (cpu_to_node[i] != NUMA_NO_NODE)
			continue;
Matthew Dobson's avatar
Matthew Dobson committed
158
		rr = next_node(rr, node_online_map);
Andi Kleen's avatar
Andi Kleen committed
159
		if (rr == MAX_NUMNODES)
Matthew Dobson's avatar
Matthew Dobson committed
160
			rr = first_node(node_online_map);
Andi Kleen's avatar
Andi Kleen committed
161 162 163 164 165 166
		cpu_to_node[i] = rr;
		rr++; 
	}

	set_bit(0, &node_to_cpumask[cpu_to_node(0)]);
}
Andi Kleen's avatar
Andi Kleen committed
167

168
#ifdef CONFIG_NUMA_EMU
169 170 171 172 173 174
int numa_fake __initdata = 0;

/* Numa emulation */
static int numa_emulation(unsigned long start_pfn, unsigned long end_pfn)
{
 	int i;
175
 	struct node nodes[MAX_NUMNODES];
176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212
 	unsigned long sz = ((end_pfn - start_pfn)<<PAGE_SHIFT) / numa_fake;

 	/* Kludge needed for the hash function */
 	if (hweight64(sz) > 1) {
 		unsigned long x = 1;
 		while ((x << 1) < sz)
 			x <<= 1;
 		if (x < sz/2)
 			printk("Numa emulation unbalanced. Complain to maintainer\n");
 		sz = x;
 	}

 	memset(&nodes,0,sizeof(nodes));
 	for (i = 0; i < numa_fake; i++) {
 		nodes[i].start = (start_pfn<<PAGE_SHIFT) + i*sz;
 		if (i == numa_fake-1)
 			sz = (end_pfn<<PAGE_SHIFT) - nodes[i].start;
 		nodes[i].end = nodes[i].start + sz;
 		if (i != numa_fake-1)
 			nodes[i].end--;
 		printk(KERN_INFO "Faking node %d at %016Lx-%016Lx (%LuMB)\n",
 		       i,
 		       nodes[i].start, nodes[i].end,
 		       (nodes[i].end - nodes[i].start) >> 20);
 	}
 	numnodes = numa_fake;
 	memnode_shift = compute_hash_shift(nodes);
 	if (memnode_shift < 0) {
 		memnode_shift = 0;
 		printk(KERN_ERR "No NUMA hash function found. Emulation disabled.\n");
 		return -1;
 	}
 	for (i = 0; i < numa_fake; i++)
 		setup_node_bootmem(i, nodes[i].start, nodes[i].end);
 	numa_init_array();
 	return 0;
}
213
#endif
214

Andi Kleen's avatar
Andi Kleen committed
215
void __init numa_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
Andi Kleen's avatar
Andi Kleen committed
216
{ 
Andi Kleen's avatar
Andi Kleen committed
217 218
	int i;

219
#ifdef CONFIG_NUMA_EMU
220 221
	if (numa_fake && !numa_emulation(start_pfn, end_pfn))
 		return;
222
#endif
223

224 225 226 227 228 229
#ifdef CONFIG_ACPI_NUMA
	if (!numa_off && !acpi_scan_nodes(start_pfn << PAGE_SHIFT,
					  end_pfn << PAGE_SHIFT))
 		return;
#endif

Andi Kleen's avatar
Andi Kleen committed
230 231
#ifdef CONFIG_K8_NUMA
	if (!numa_off && !k8_scan_nodes(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT))
Andi Kleen's avatar
Andi Kleen committed
232
		return;
Andi Kleen's avatar
Andi Kleen committed
233 234 235 236 237 238 239 240 241 242
#endif
	printk(KERN_INFO "%s\n",
	       numa_off ? "NUMA turned off" : "No NUMA configuration found");

	printk(KERN_INFO "Faking a node at %016lx-%016lx\n", 
	       start_pfn << PAGE_SHIFT,
	       end_pfn << PAGE_SHIFT); 
		/* setup dummy node covering all memory */ 
	memnode_shift = 63; 
	memnodemap[0] = 0;
243
	numnodes = 1;
Andi Kleen's avatar
Andi Kleen committed
244 245
	for (i = 0; i < NR_CPUS; i++)
		cpu_to_node[i] = 0;
246
	node_to_cpumask[0] = cpumask_of_cpu(0);
247
	setup_node_bootmem(0, start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT);
Andi Kleen's avatar
Andi Kleen committed
248 249 250 251 252 253 254
}

__init void numa_add_cpu(int cpu)
{
	/* BP is initialized elsewhere */
	if (cpu) 
		set_bit(cpu, &node_to_cpumask[cpu_to_node(cpu)]);
Andi Kleen's avatar
Andi Kleen committed
255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277
} 

unsigned long __init numa_free_all_bootmem(void) 
{ 
	int i;
	unsigned long pages = 0;
	for_all_nodes(i) {
		pages += free_all_bootmem_node(NODE_DATA(i));
	}
	return pages;
} 

void __init paging_init(void)
{ 
	int i;
	for_all_nodes(i) { 
		setup_node_zones(i); 
	}
} 

/* [numa=off] */
__init int numa_setup(char *opt) 
{ 
278
	if (!strcmp(opt,"off"))
Andi Kleen's avatar
Andi Kleen committed
279
		numa_off = 1;
280
#ifdef CONFIG_NUMA_EMU
281 282 283 284 285
	if(!strncmp(opt, "fake=", 5)) {
		numa_fake = simple_strtoul(opt+5,NULL,0); ;
		if (numa_fake >= MAX_NUMNODES)
			numa_fake = MAX_NUMNODES;
	}
286 287 288 289
#endif
#ifdef CONFIG_ACPI_NUMA
 	if (!strncmp(opt,"noacpi",6))
 		acpi_numa = -1;
290
#endif
Andi Kleen's avatar
Andi Kleen committed
291 292 293
	return 1;
} 

Andi Kleen's avatar
Andi Kleen committed
294 295 296 297 298
EXPORT_SYMBOL(cpu_to_node);
EXPORT_SYMBOL(node_to_cpumask);
EXPORT_SYMBOL(memnode_shift);
EXPORT_SYMBOL(memnodemap);
EXPORT_SYMBOL(node_data);