[deliverable/linux.git] / arch / x86_64 / mm / srat.c

/*
 * ACPI 3.0 based NUMA setup
 * Copyright 2004 Andi Kleen, SuSE Labs.
 *
 * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs.
 *
 * Called from acpi_numa_init while reading the SRAT and SLIT tables.
 * Assumes all memory regions belonging to a single proximity domain
 * are in one chunk. Holes between them will be included in the node.
 */

#include <linux/kernel.h>
#include <linux/acpi.h>
#include <linux/mmzone.h>
#include <linux/bitmap.h>
#include <linux/module.h>
#include <linux/topology.h>
#include <asm/proto.h>
#include <asm/numa.h>
#include <asm/e820.h>

static struct acpi_table_slit *acpi_slit;

static nodemask_t nodes_parsed __initdata;
static nodemask_t nodes_found __initdata;
static struct node nodes[MAX_NUMNODES] __initdata;
static u8 pxm2node[256] = { [0 ... 255] = 0xff };

static int node_to_pxm(int n);

int pxm_to_node(int pxm)
{
	if ((unsigned)pxm >= 256)
		return -1;
	/* Extend 0xff to (int)-1 */
	return (signed char)pxm2node[pxm];
}

static __init int setup_node(int pxm)
{
	unsigned node = pxm2node[pxm];
	if (node == 0xff) {
		if (nodes_weight(nodes_found) >= MAX_NUMNODES)
			return -1;
		node = first_unset_node(nodes_found); 
		node_set(node, nodes_found);
		pxm2node[pxm] = node;
	}
	return pxm2node[pxm];
}

static __init int conflicting_nodes(unsigned long start, unsigned long end)
{
	int i;
	for_each_node_mask(i, nodes_parsed) {
		struct node *nd = &nodes[i];
		if (nd->start == nd->end)
			continue;
		if (nd->end > start && nd->start < end)
			return i;
		if (nd->end == end && nd->start == start)
			return i;
	}
	return -1;
}

static __init void cutoff_node(int i, unsigned long start, unsigned long end)
{
	struct node *nd = &nodes[i];
	if (nd->start < start) {
		nd->start = start;
		if (nd->end < nd->start)
			nd->start = nd->end;
	}
	if (nd->end > end) {
		nd->end = end;
		if (nd->start > nd->end)
			nd->start = nd->end;
	}
}

static __init void bad_srat(void)
{
	int i;
	printk(KERN_ERR "SRAT: SRAT not used.\n");
	acpi_numa = -1;
	for (i = 0; i < MAX_LOCAL_APIC; i++)
		apicid_to_node[i] = NUMA_NO_NODE;
}

static __init inline int srat_disabled(void)
{
	return numa_off || acpi_numa < 0;
}

/*
 * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
 * up the NUMA heuristics which wants the local node to have a smaller
 * distance than the others.
 * Do some quick checks here and only use the SLIT if it passes.
 */
static __init int slit_valid(struct acpi_table_slit *slit)
{
	int i, j;
	int d = slit->localities;
	for (i = 0; i < d; i++) {
		for (j = 0; j < d; j++)  {
			u8 val = slit->entry[d*i + j];
			if (i == j) {
				if (val != 10)
					return 0;
			} else if (val <= 10)
				return 0;
		}
	}
	return 1;
}

/* Callback for SLIT parsing */
void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
{
	if (!slit_valid(slit)) {
		printk(KERN_INFO "ACPI: SLIT table looks invalid. Not used.\n");
		return;
	}
	acpi_slit = slit;
}

/* Callback for Proximity Domain -> LAPIC mapping */
void __init
acpi_numa_processor_affinity_init(struct acpi_table_processor_affinity *pa)
{
	int pxm, node;
	if (srat_disabled() || pa->flags.enabled == 0)
		return;
	pxm = pa->proximity_domain;
	node = setup_node(pxm);
	if (node < 0) {
		printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm);
		bad_srat();
		return;
	}
	apicid_to_node[pa->apic_id] = node;
	acpi_numa = 1;
	printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n",
	       pxm, pa->apic_id, node);
}

/* Callback for parsing of the Proximity Domain <-> Memory Area mappings */
void __init
acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma)
{
	struct node *nd;
	unsigned long start, end;
	int node, pxm;
	int i;

	if (srat_disabled() || ma->flags.enabled == 0)
		return;
	pxm = ma->proximity_domain;
	node = setup_node(pxm);
	if (node < 0) {
		printk(KERN_ERR "SRAT: Too many proximity domains.\n");
		bad_srat();
		return;
	}
	start = ma->base_addr_lo | ((u64)ma->base_addr_hi << 32);
	end = start + (ma->length_lo | ((u64)ma->length_hi << 32));
	/* It is fine to add this area to the nodes data it will be used later*/
	if (ma->flags.hot_pluggable == 1)
		printk(KERN_INFO "SRAT: hot plug zone found %lx - %lx \n",
				start, end);
	i = conflicting_nodes(start, end);
	if (i == node) {
		printk(KERN_WARNING
		"SRAT: Warning: PXM %d (%lx-%lx) overlaps with itself (%Lx-%Lx)\n",
			pxm, start, end, nodes[i].start, nodes[i].end);
	} else if (i >= 0) {
		printk(KERN_ERR
		       "SRAT: PXM %d (%lx-%lx) overlaps with PXM %d (%Lx-%Lx)\n",
		       pxm, start, end, node_to_pxm(i),
			nodes[i].start, nodes[i].end);
		bad_srat();
		return;
	}
	nd = &nodes[node];
	if (!node_test_and_set(node, nodes_parsed)) {
		nd->start = start;
		nd->end = end;
	} else {
		if (start < nd->start)
			nd->start = start;
		if (nd->end < end)
			nd->end = end;
	}
	printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm,
	       nd->start, nd->end);
}

/* Sanity check to catch more bad SRATs (they are amazingly common).
   Make sure the PXMs cover all memory. */
static int nodes_cover_memory(void)
{
	int i;
	unsigned long pxmram, e820ram;

	pxmram = 0;
	for_each_node_mask(i, nodes_parsed) {
		unsigned long s = nodes[i].start >> PAGE_SHIFT;
		unsigned long e = nodes[i].end >> PAGE_SHIFT;
		pxmram += e - s;
		pxmram -= e820_hole_size(s, e);
	}

	e820ram = end_pfn - e820_hole_size(0, end_pfn);
	if (pxmram < e820ram) {
		printk(KERN_ERR
	"SRAT: PXMs only cover %luMB of your %luMB e820 RAM. Not used.\n",
			(pxmram << PAGE_SHIFT) >> 20,
			(e820ram << PAGE_SHIFT) >> 20);
		return 0;
	}
	return 1;
}

void __init acpi_numa_arch_fixup(void) {}

/* Use the information discovered above to actually set up the nodes. */
int __init acpi_scan_nodes(unsigned long start, unsigned long end)
{
	int i;

	if (acpi_numa <= 0)
		return -1;

	/* First clean up the node list */
	for_each_node_mask(i, nodes_parsed) {
		cutoff_node(i, start, end);
		if (nodes[i].start == nodes[i].end)
			node_clear(i, nodes_parsed);
	}

	if (!nodes_cover_memory()) {
		bad_srat();
		return -1;
	}

	memnode_shift = compute_hash_shift(nodes, nodes_weight(nodes_parsed));
	if (memnode_shift < 0) {
		printk(KERN_ERR
		     "SRAT: No NUMA node hash function found. Contact maintainer\n");
		bad_srat();
		return -1;
	}

	/* Finally register nodes */
	for_each_node_mask(i, nodes_parsed)
		setup_node_bootmem(i, nodes[i].start, nodes[i].end);
	for (i = 0; i < NR_CPUS; i++) { 
		if (cpu_to_node[i] == NUMA_NO_NODE)
			continue;
		if (!node_isset(cpu_to_node[i], nodes_parsed))
			numa_set_node(i, NUMA_NO_NODE);
	}
	numa_init_array();
	return 0;
}

static int node_to_pxm(int n)
{
       int i;
       if (pxm2node[n] == n)
               return n;
       for (i = 0; i < 256; i++)
               if (pxm2node[i] == n)
                       return i;
       return 0;
}

int __node_distance(int a, int b)
{
	int index;

	if (!acpi_slit)
		return a == b ? 10 : 20;
	index = acpi_slit->localities * node_to_pxm(a);
	return acpi_slit->entry[index + node_to_pxm(b)];
}

EXPORT_SYMBOL(__node_distance);
Commit	Line	Data
1da177e4 LT	1	/*
	2	* ACPI 3.0 based NUMA setup
	3	* Copyright 2004 Andi Kleen, SuSE Labs.
	4	*
	5	* Reads the ACPI SRAT table to figure out what memory belongs to which CPUs.
	6	*
	7	* Called from acpi_numa_init while reading the SRAT and SLIT tables.
	8	* Assumes all memory regions belonging to a single proximity domain
	9	* are in one chunk. Holes between them will be included in the node.
	10	*/
	11
	12	#include <linux/kernel.h>
	13	#include <linux/acpi.h>
	14	#include <linux/mmzone.h>
	15	#include <linux/bitmap.h>
	16	#include <linux/module.h>
	17	#include <linux/topology.h>
	18	#include <asm/proto.h>
	19	#include <asm/numa.h>
8a6fdd3e	20	#include <asm/e820.h>
1da177e4 LT	21
	22	static struct acpi_table_slit *acpi_slit;
	23
	24	static nodemask_t nodes_parsed __initdata;
	25	static nodemask_t nodes_found __initdata;
	26	static struct node nodes[MAX_NUMNODES] __initdata;
e4e94072	27	static u8 pxm2node[256] = { [0 ... 255] = 0xff };
1da177e4	28
05d1fa4b AK	29	static int node_to_pxm(int n);
05d1fa4b AK	30
69e1a33f AK	31	int pxm_to_node(int pxm)
	32	{
	33	if ((unsigned)pxm >= 256)
e4e94072 AK	34	return -1;
	35	/* Extend 0xff to (int)-1 */
	36	return (signed char)pxm2node[pxm];
69e1a33f AK	37	}
69e1a33f AK	38
1da177e4 LT	39	static __init int setup_node(int pxm)
	40	{
	41	unsigned node = pxm2node[pxm];
	42	if (node == 0xff) {
	43	if (nodes_weight(nodes_found) >= MAX_NUMNODES)
	44	return -1;
	45	node = first_unset_node(nodes_found);
	46	node_set(node, nodes_found);
	47	pxm2node[pxm] = node;
	48	}
	49	return pxm2node[pxm];
	50	}
	51
	52	static __init int conflicting_nodes(unsigned long start, unsigned long end)
	53	{
	54	int i;
4b6a455c	55	for_each_node_mask(i, nodes_parsed) {
1da177e4 LT	56	struct node *nd = &nodes[i];
	57	if (nd->start == nd->end)
	58	continue;
	59	if (nd->end > start && nd->start < end)
05d1fa4b	60	return i;
1da177e4	61	if (nd->end == end && nd->start == start)
05d1fa4b	62	return i;
1da177e4 LT	63	}
	64	return -1;
	65	}
	66
	67	static __init void cutoff_node(int i, unsigned long start, unsigned long end)
	68	{
	69	struct node *nd = &nodes[i];
	70	if (nd->start < start) {
	71	nd->start = start;
	72	if (nd->end < nd->start)
	73	nd->start = nd->end;
	74	}
	75	if (nd->end > end) {
1da177e4 LT	76	nd->end = end;
	77	if (nd->start > nd->end)
	78	nd->start = nd->end;
	79	}
	80	}
	81
	82	static __init void bad_srat(void)
	83	{
2bce2b54	84	int i;
1da177e4 LT	85	printk(KERN_ERR "SRAT: SRAT not used.\n");
1da177e4 LT	86	acpi_numa = -1;
2bce2b54 AK	87	for (i = 0; i < MAX_LOCAL_APIC; i++)
2bce2b54 AK	88	apicid_to_node[i] = NUMA_NO_NODE;
1da177e4 LT	89	}
	90
	91	static __init inline int srat_disabled(void)
	92	{
	93	return numa_off \|\| acpi_numa < 0;
	94	}
	95
1584b89c AK	96	/*
	97	* A lot of BIOS fill in 10 (= no distance) everywhere. This messes
	98	* up the NUMA heuristics which wants the local node to have a smaller
	99	* distance than the others.
	100	* Do some quick checks here and only use the SLIT if it passes.
	101	*/
	102	static __init int slit_valid(struct acpi_table_slit *slit)
	103	{
	104	int i, j;
	105	int d = slit->localities;
	106	for (i = 0; i < d; i++) {
	107	for (j = 0; j < d; j++) {
	108	u8 val = slit->entry[d*i + j];
	109	if (i == j) {
	110	if (val != 10)
	111	return 0;
	112	} else if (val <= 10)
	113	return 0;
	114	}
	115	}
	116	return 1;
	117	}
	118
1da177e4 LT	119	/* Callback for SLIT parsing */
	120	void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
	121	{
1584b89c AK	122	if (!slit_valid(slit)) {
	123	printk(KERN_INFO "ACPI: SLIT table looks invalid. Not used.\n");
	124	return;
	125	}
1da177e4 LT	126	acpi_slit = slit;
	127	}
	128
	129	/* Callback for Proximity Domain -> LAPIC mapping */
	130	void __init
	131	acpi_numa_processor_affinity_init(struct acpi_table_processor_affinity *pa)
	132	{
	133	int pxm, node;
	134	if (srat_disabled() \|\| pa->flags.enabled == 0)
	135	return;
	136	pxm = pa->proximity_domain;
	137	node = setup_node(pxm);
	138	if (node < 0) {
	139	printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm);
	140	bad_srat();
	141	return;
	142	}
0b07e984	143	apicid_to_node[pa->apic_id] = node;
1da177e4	144	acpi_numa = 1;
0b07e984 AK	145	printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n",
0b07e984 AK	146	pxm, pa->apic_id, node);
1da177e4 LT	147	}
	148
	149	/* Callback for parsing of the Proximity Domain <-> Memory Area mappings */
	150	void __init
	151	acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma)
	152	{
	153	struct node *nd;
	154	unsigned long start, end;
	155	int node, pxm;
	156	int i;
	157
	158	if (srat_disabled() \|\| ma->flags.enabled == 0)
	159	return;
1da177e4 LT	160	pxm = ma->proximity_domain;
	161	node = setup_node(pxm);
	162	if (node < 0) {
	163	printk(KERN_ERR "SRAT: Too many proximity domains.\n");
	164	bad_srat();
	165	return;
	166	}
	167	start = ma->base_addr_lo \| ((u64)ma->base_addr_hi << 32);
	168	end = start + (ma->length_lo \| ((u64)ma->length_hi << 32));
69cb62eb AK	169	/* It is fine to add this area to the nodes data it will be used later*/
	170	if (ma->flags.hot_pluggable == 1)
	171	printk(KERN_INFO "SRAT: hot plug zone found %lx - %lx \n",
	172	start, end);
1da177e4	173	i = conflicting_nodes(start, end);
05d1fa4b AK	174	if (i == node) {
	175	printk(KERN_WARNING
	176	"SRAT: Warning: PXM %d (%lx-%lx) overlaps with itself (%Lx-%Lx)\n",
	177	pxm, start, end, nodes[i].start, nodes[i].end);
	178	} else if (i >= 0) {
1da177e4	179	printk(KERN_ERR
05d1fa4b AK	180	"SRAT: PXM %d (%lx-%lx) overlaps with PXM %d (%Lx-%Lx)\n",
	181	pxm, start, end, node_to_pxm(i),
	182	nodes[i].start, nodes[i].end);
1da177e4 LT	183	bad_srat();
	184	return;
	185	}
	186	nd = &nodes[node];
	187	if (!node_test_and_set(node, nodes_parsed)) {
	188	nd->start = start;
	189	nd->end = end;
	190	} else {
	191	if (start < nd->start)
	192	nd->start = start;
	193	if (nd->end < end)
	194	nd->end = end;
	195	}
1da177e4 LT	196	printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm,
	197	nd->start, nd->end);
	198	}
	199
8a6fdd3e AK	200	/* Sanity check to catch more bad SRATs (they are amazingly common).
	201	Make sure the PXMs cover all memory. */
	202	static int nodes_cover_memory(void)
	203	{
	204	int i;
	205	unsigned long pxmram, e820ram;
	206
	207	pxmram = 0;
	208	for_each_node_mask(i, nodes_parsed) {
	209	unsigned long s = nodes[i].start >> PAGE_SHIFT;
	210	unsigned long e = nodes[i].end >> PAGE_SHIFT;
	211	pxmram += e - s;
	212	pxmram -= e820_hole_size(s, e);
	213	}
	214
	215	e820ram = end_pfn - e820_hole_size(0, end_pfn);
	216	if (pxmram < e820ram) {
	217	printk(KERN_ERR
	218	"SRAT: PXMs only cover %luMB of your %luMB e820 RAM. Not used.\n",
	219	(pxmram << PAGE_SHIFT) >> 20,
	220	(e820ram << PAGE_SHIFT) >> 20);
	221	return 0;
	222	}
	223	return 1;
	224	}
	225
1da177e4 LT	226	void __init acpi_numa_arch_fixup(void) {}
	227
	228	/* Use the information discovered above to actually set up the nodes. */
	229	int __init acpi_scan_nodes(unsigned long start, unsigned long end)
	230	{
	231	int i;
8a6fdd3e	232
1da177e4 LT	233	if (acpi_numa <= 0)
1da177e4 LT	234	return -1;
e58e0d03 AK	235
	236	/* First clean up the node list */
	237	for_each_node_mask(i, nodes_parsed) {
	238	cutoff_node(i, start, end);
	239	if (nodes[i].start == nodes[i].end)
	240	node_clear(i, nodes_parsed);
	241	}
	242
8a6fdd3e AK	243	if (!nodes_cover_memory()) {
	244	bad_srat();
	245	return -1;
	246	}
	247
1da177e4 LT	248	memnode_shift = compute_hash_shift(nodes, nodes_weight(nodes_parsed));
	249	if (memnode_shift < 0) {
	250	printk(KERN_ERR
	251	"SRAT: No NUMA node hash function found. Contact maintainer\n");
	252	bad_srat();
	253	return -1;
	254	}
e58e0d03 AK	255
	256	/* Finally register nodes */
	257	for_each_node_mask(i, nodes_parsed)
1da177e4	258	setup_node_bootmem(i, nodes[i].start, nodes[i].end);
1da177e4 LT	259	for (i = 0; i < NR_CPUS; i++) {
	260	if (cpu_to_node[i] == NUMA_NO_NODE)
	261	continue;
	262	if (!node_isset(cpu_to_node[i], nodes_parsed))
69d81fcd	263	numa_set_node(i, NUMA_NO_NODE);
1da177e4 LT	264	}
	265	numa_init_array();
	266	return 0;
	267	}
	268
05d1fa4b	269	static int node_to_pxm(int n)
1da177e4 LT	270	{
	271	int i;
	272	if (pxm2node[n] == n)
	273	return n;
	274	for (i = 0; i < 256; i++)
	275	if (pxm2node[i] == n)
	276	return i;
	277	return 0;
	278	}
	279
	280	int __node_distance(int a, int b)
	281	{
	282	int index;
	283
	284	if (!acpi_slit)
	285	return a == b ? 10 : 20;
	286	index = acpi_slit->localities * node_to_pxm(a);
	287	return acpi_slit->entry[index + node_to_pxm(b)];
	288	}
	289
	290	EXPORT_SYMBOL(__node_distance);