[deliverable/linux.git] / arch / x86 / mm / srat_32.c

/*
 * Some of the code in this file has been gleaned from the 64 bit 
 * discontigmem support code base.
 *
 * Copyright (C) 2002, IBM Corp.
 *
 * All rights reserved.          
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 * NON INFRINGEMENT.  See the GNU General Public License for more
 * details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 *
 * Send feedback to Pat Gaughen <gone@us.ibm.com>
 */
#include <linux/mm.h>
#include <linux/bootmem.h>
#include <linux/memblock.h>
#include <linux/mmzone.h>
#include <linux/acpi.h>
#include <linux/nodemask.h>
#include <asm/srat.h>
#include <asm/topology.h>
#include <asm/smp.h>
#include <asm/e820.h>

/*
 * proximity macros and definitions
 */
#define NODE_ARRAY_INDEX(x)	((x) / 8)	/* 8 bits/char */
#define NODE_ARRAY_OFFSET(x)	((x) % 8)	/* 8 bits/char */
#define BMAP_SET(bmap, bit)	((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit))
#define BMAP_TEST(bmap, bit)	((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit)))
/* bitmap length; _PXM is at most 255 */
#define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8) 
static u8 __initdata pxm_bitmap[PXM_BITMAP_LEN];	/* bitmap of proximity domains */

#define MAX_CHUNKS_PER_NODE	3
#define MAXCHUNKS		(MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
struct node_memory_chunk_s {
	unsigned long	start_pfn;
	unsigned long	end_pfn;
	u8	pxm;		// proximity domain of node
	u8	nid;		// which cnode contains this chunk?
	u8	bank;		// which mem bank on this node
};
static struct node_memory_chunk_s __initdata node_memory_chunk[MAXCHUNKS];

static int __initdata num_memory_chunks; /* total number of memory chunks */
static u8 __initdata apicid_to_pxm[MAX_LOCAL_APIC];

int acpi_numa __initdata;

static __init void bad_srat(void)
{
        printk(KERN_ERR "SRAT: SRAT not used.\n");
        acpi_numa = -1;
	num_memory_chunks = 0;
}

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

/* Identify CPU proximity domains */
void __init
acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *cpu_affinity)
{
	if (srat_disabled())
		return;
	if (cpu_affinity->header.length !=
	     sizeof(struct acpi_srat_cpu_affinity)) {
		bad_srat();
		return;
	}

	if ((cpu_affinity->flags & ACPI_SRAT_CPU_ENABLED) == 0)
		return;		/* empty entry */

	/* mark this node as "seen" in node bitmap */
	BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain_lo);

	/* don't need to check apic_id here, because it is always 8 bits */
	apicid_to_pxm[cpu_affinity->apic_id] = cpu_affinity->proximity_domain_lo;

	printk(KERN_DEBUG "CPU %02x in proximity domain %02x\n",
		cpu_affinity->apic_id, cpu_affinity->proximity_domain_lo);
}

/*
 * Identify memory proximity domains and hot-remove capabilities.
 * Fill node memory chunk list structure.
 */
void __init
acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *memory_affinity)
{
	unsigned long long paddr, size;
	unsigned long start_pfn, end_pfn;
	u8 pxm;
	struct node_memory_chunk_s *p, *q, *pend;

	if (srat_disabled())
		return;
	if (memory_affinity->header.length !=
	     sizeof(struct acpi_srat_mem_affinity)) {
		bad_srat();
		return;
	}

	if ((memory_affinity->flags & ACPI_SRAT_MEM_ENABLED) == 0)
		return;		/* empty entry */

	pxm = memory_affinity->proximity_domain & 0xff;

	/* mark this node as "seen" in node bitmap */
	BMAP_SET(pxm_bitmap, pxm);

	/* calculate info for memory chunk structure */
	paddr = memory_affinity->base_address;
	size = memory_affinity->length;

	start_pfn = paddr >> PAGE_SHIFT;
	end_pfn = (paddr + size) >> PAGE_SHIFT;


	if (num_memory_chunks >= MAXCHUNKS) {
		printk(KERN_WARNING "Too many mem chunks in SRAT."
			" Ignoring %lld MBytes at %llx\n",
			size/(1024*1024), paddr);
		return;
	}

	/* Insertion sort based on base address */
	pend = &node_memory_chunk[num_memory_chunks];
	for (p = &node_memory_chunk[0]; p < pend; p++) {
		if (start_pfn < p->start_pfn)
			break;
	}
	if (p < pend) {
		for (q = pend; q >= p; q--)
			*(q + 1) = *q;
	}
	p->start_pfn = start_pfn;
	p->end_pfn = end_pfn;
	p->pxm = pxm;

	num_memory_chunks++;

	printk(KERN_DEBUG "Memory range %08lx to %08lx"
			  " in proximity domain %02x %s\n",
		start_pfn, end_pfn,
		pxm,
		((memory_affinity->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ?
		 "enabled and removable" : "enabled" ) );
}

/* Callback for SLIT parsing */
void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
{
}

void acpi_numa_arch_fixup(void)
{
}
/*
 * The SRAT table always lists ascending addresses, so can always
 * assume that the first "start" address that you see is the real
 * start of the node, and that the current "end" address is after
 * the previous one.
 */
static __init int node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
{
	/*
	 * Only add present memory as told by the e820.
	 * There is no guarantee from the SRAT that the memory it
	 * enumerates is present at boot time because it represents
	 * *possible* memory hotplug areas the same as normal RAM.
	 */
	if (memory_chunk->start_pfn >= max_pfn) {
		printk(KERN_INFO "Ignoring SRAT pfns: %08lx - %08lx\n",
			memory_chunk->start_pfn, memory_chunk->end_pfn);
		return -1;
	}
	if (memory_chunk->nid != nid)
		return -1;

	if (!node_has_online_mem(nid))
		node_start_pfn[nid] = memory_chunk->start_pfn;

	if (node_start_pfn[nid] > memory_chunk->start_pfn)
		node_start_pfn[nid] = memory_chunk->start_pfn;

	if (node_end_pfn[nid] < memory_chunk->end_pfn)
		node_end_pfn[nid] = memory_chunk->end_pfn;

	return 0;
}

int __init get_memcfg_from_srat(void)
{
	int i, j;

	if (srat_disabled())
		goto out_fail;

	if (acpi_numa_init() < 0)
		goto out_fail;

	if (num_memory_chunks == 0) {
		printk(KERN_DEBUG
			 "could not find any ACPI SRAT memory areas.\n");
		goto out_fail;
	}

	/* Calculate total number of nodes in system from PXM bitmap and create
	 * a set of sequential node IDs starting at zero.  (ACPI doesn't seem
	 * to specify the range of _PXM values.)
	 */
	/*
	 * MCD - we no longer HAVE to number nodes sequentially.  PXM domain
	 * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
	 * 32, so we will continue numbering them in this manner until MAX_NUMNODES
	 * approaches MAX_PXM_DOMAINS for i386.
	 */
	nodes_clear(node_online_map);
	for (i = 0; i < MAX_PXM_DOMAINS; i++) {
		if (BMAP_TEST(pxm_bitmap, i)) {
			int nid = acpi_map_pxm_to_node(i);
			node_set_online(nid);
		}
	}
	BUG_ON(num_online_nodes() == 0);

	/* set cnode id in memory chunk structure */
	for (i = 0; i < num_memory_chunks; i++)
		node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);

	printk(KERN_DEBUG "pxm bitmap: ");
	for (i = 0; i < sizeof(pxm_bitmap); i++) {
		printk(KERN_CONT "%02x ", pxm_bitmap[i]);
	}
	printk(KERN_CONT "\n");
	printk(KERN_DEBUG "Number of logical nodes in system = %d\n",
			 num_online_nodes());
	printk(KERN_DEBUG "Number of memory chunks in system = %d\n",
			 num_memory_chunks);

	for (i = 0; i < MAX_LOCAL_APIC; i++)
		set_apicid_to_node(i, pxm_to_node(apicid_to_pxm[i]));

	for (j = 0; j < num_memory_chunks; j++){
		struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
		printk(KERN_DEBUG
			"chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
		       j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
		if (node_read_chunk(chunk->nid, chunk))
			continue;

		memblock_x86_register_active_regions(chunk->nid, chunk->start_pfn,
					     min(chunk->end_pfn, max_pfn));
	}
	/* for out of order entries in SRAT */
	sort_node_map();

	return 1;
out_fail:
	printk(KERN_DEBUG "failed to get NUMA memory information from SRAT"
			" table\n");
	return 0;
}
Commit	Line	Data
1da177e4 LT	1	/*
	2	* Some of the code in this file has been gleaned from the 64 bit
	3	* discontigmem support code base.
	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	* Send feedback to Pat Gaughen <gone@us.ibm.com>
	25	*/
1da177e4 LT	26	#include <linux/mm.h>
1da177e4 LT	27	#include <linux/bootmem.h>
a9ce6bc1	28	#include <linux/memblock.h>
1da177e4 LT	29	#include <linux/mmzone.h>
	30	#include <linux/acpi.h>
	31	#include <linux/nodemask.h>
	32	#include <asm/srat.h>
	33	#include <asm/topology.h>
3b08606d	34	#include <asm/smp.h>
7b2a0a6c	35	#include <asm/e820.h>
1da177e4 LT	36
	37	/*
	38	* proximity macros and definitions
	39	*/
	40	#define NODE_ARRAY_INDEX(x) ((x) / 8) /* 8 bits/char */
	41	#define NODE_ARRAY_OFFSET(x) ((x) % 8) /* 8 bits/char */
	42	#define BMAP_SET(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] \|= 1 << NODE_ARRAY_OFFSET(bit))
	43	#define BMAP_TEST(bmap, bit) ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit)))
1da177e4 LT	44	/* bitmap length; _PXM is at most 255 */
1da177e4 LT	45	#define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8)
1c6e5503	46	static u8 __initdata pxm_bitmap[PXM_BITMAP_LEN]; /* bitmap of proximity domains */
1da177e4	47
b9b15780	48	#define MAX_CHUNKS_PER_NODE 3
1da177e4 LT	49	#define MAXCHUNKS (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
	50	struct node_memory_chunk_s {
	51	unsigned long start_pfn;
	52	unsigned long end_pfn;
	53	u8 pxm; // proximity domain of node
	54	u8 nid; // which cnode contains this chunk?
	55	u8 bank; // which mem bank on this node
	56	};
1c6e5503	57	static struct node_memory_chunk_s __initdata node_memory_chunk[MAXCHUNKS];
1da177e4	58
1c6e5503	59	static int __initdata num_memory_chunks; /* total number of memory chunks */
b78aa66b	60	static u8 __initdata apicid_to_pxm[MAX_LOCAL_APIC];
1da177e4	61
1c6e5503 YL	62	int acpi_numa __initdata;
	63
	64	static __init void bad_srat(void)
	65	{
	66	printk(KERN_ERR "SRAT: SRAT not used.\n");
	67	acpi_numa = -1;
	68	num_memory_chunks = 0;
	69	}
	70
	71	static __init inline int srat_disabled(void)
	72	{
	73	return numa_off \|\| acpi_numa < 0;
	74	}
	75
1da177e4	76	/* Identify CPU proximity domains */
1c6e5503 YL	77	void __init
1c6e5503 YL	78	acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *cpu_affinity)
1da177e4	79	{
1c6e5503 YL	80	if (srat_disabled())
	81	return;
	82	if (cpu_affinity->header.length !=
	83	sizeof(struct acpi_srat_cpu_affinity)) {
	84	bad_srat();
	85	return;
	86	}
1da177e4	87
0e568335	88	if ((cpu_affinity->flags & ACPI_SRAT_CPU_ENABLED) == 0)
1da177e4 LT	89	return; /* empty entry */
	90
	91	/* mark this node as "seen" in node bitmap */
0e568335	92	BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain_lo);
1da177e4	93
d3bd0588	94	/* don't need to check apic_id here, because it is always 8 bits */
0e568335	95	apicid_to_pxm[cpu_affinity->apic_id] = cpu_affinity->proximity_domain_lo;
3b08606d	96
c0943457	97	printk(KERN_DEBUG "CPU %02x in proximity domain %02x\n",
0e568335	98	cpu_affinity->apic_id, cpu_affinity->proximity_domain_lo);
1da177e4 LT	99	}
	100
	101	/*
	102	* Identify memory proximity domains and hot-remove capabilities.
	103	* Fill node memory chunk list structure.
	104	*/
1c6e5503 YL	105	void __init
1c6e5503 YL	106	acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *memory_affinity)
1da177e4 LT	107	{
1da177e4 LT	108	unsigned long long paddr, size;
0e568335	109	unsigned long start_pfn, end_pfn;
1da177e4 LT	110	u8 pxm;
1da177e4 LT	111	struct node_memory_chunk_s p, q, *pend;
1c6e5503 YL	112
	113	if (srat_disabled())
	114	return;
	115	if (memory_affinity->header.length !=
	116	sizeof(struct acpi_srat_mem_affinity)) {
	117	bad_srat();
	118	return;
	119	}
1da177e4	120
0e568335	121	if ((memory_affinity->flags & ACPI_SRAT_MEM_ENABLED) == 0)
1da177e4 LT	122	return; /* empty entry */
1da177e4 LT	123
0e568335 AS	124	pxm = memory_affinity->proximity_domain & 0xff;
0e568335 AS	125
1da177e4	126	/* mark this node as "seen" in node bitmap */
0e568335	127	BMAP_SET(pxm_bitmap, pxm);
1da177e4 LT	128
1da177e4 LT	129	/* calculate info for memory chunk structure */
0e568335 AS	130	paddr = memory_affinity->base_address;
	131	size = memory_affinity->length;
	132
1da177e4 LT	133	start_pfn = paddr >> PAGE_SHIFT;
1da177e4 LT	134	end_pfn = (paddr + size) >> PAGE_SHIFT;
0e568335	135
1da177e4 LT	136
1da177e4 LT	137	if (num_memory_chunks >= MAXCHUNKS) {
c0943457 YL	138	printk(KERN_WARNING "Too many mem chunks in SRAT."
c0943457 YL	139	" Ignoring %lld MBytes at %llx\n",
1da177e4 LT	140	size/(1024*1024), paddr);
	141	return;
	142	}
	143
	144	/* Insertion sort based on base address */
	145	pend = &node_memory_chunk[num_memory_chunks];
	146	for (p = &node_memory_chunk[0]; p < pend; p++) {
	147	if (start_pfn < p->start_pfn)
	148	break;
	149	}
	150	if (p < pend) {
	151	for (q = pend; q >= p; q--)
	152	(q + 1) = q;
	153	}
	154	p->start_pfn = start_pfn;
	155	p->end_pfn = end_pfn;
	156	p->pxm = pxm;
	157
	158	num_memory_chunks++;
	159
19d0cfe9	160	printk(KERN_DEBUG "Memory range %08lx to %08lx"
c0943457	161	" in proximity domain %02x %s\n",
1da177e4	162	start_pfn, end_pfn,
0e568335 AS	163	pxm,
0e568335 AS	164	((memory_affinity->flags & ACPI_SRAT_MEM_HOT_PLUGGABLE) ?
1da177e4 LT	165	"enabled and removable" : "enabled" ) );
	166	}
	167
1c6e5503 YL	168	/* Callback for SLIT parsing */
	169	void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
	170	{
	171	}
	172
	173	void acpi_numa_arch_fixup(void)
	174	{
	175	}
1da177e4 LT	176	/*
	177	* The SRAT table always lists ascending addresses, so can always
	178	* assume that the first "start" address that you see is the real
	179	* start of the node, and that the current "end" address is after
	180	* the previous one.
	181	*/
858f7747	182	static __init int node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
1da177e4 LT	183	{
	184	/*
	185	* Only add present memory as told by the e820.
	186	* There is no guarantee from the SRAT that the memory it
	187	* enumerates is present at boot time because it represents
	188	* possible memory hotplug areas the same as normal RAM.
	189	*/
	190	if (memory_chunk->start_pfn >= max_pfn) {
c0943457	191	printk(KERN_INFO "Ignoring SRAT pfns: %08lx - %08lx\n",
1da177e4	192	memory_chunk->start_pfn, memory_chunk->end_pfn);
858f7747	193	return -1;
1da177e4 LT	194	}
1da177e4 LT	195	if (memory_chunk->nid != nid)
858f7747	196	return -1;
1da177e4 LT	197
	198	if (!node_has_online_mem(nid))
	199	node_start_pfn[nid] = memory_chunk->start_pfn;
	200
	201	if (node_start_pfn[nid] > memory_chunk->start_pfn)
	202	node_start_pfn[nid] = memory_chunk->start_pfn;
	203
	204	if (node_end_pfn[nid] < memory_chunk->end_pfn)
	205	node_end_pfn[nid] = memory_chunk->end_pfn;
858f7747 YL	206
858f7747 YL	207	return 0;
1da177e4 LT	208	}
1da177e4 LT	209
1c6e5503	210	int __init get_memcfg_from_srat(void)
1da177e4	211	{
797390d8	212	int i, j;
1da177e4	213
1c6e5503 YL	214	if (srat_disabled())
1c6e5503 YL	215	goto out_fail;
1da177e4	216
765af22d TH	217	if (acpi_numa_init() < 0)
	218	goto out_fail;
	219
1da177e4	220	if (num_memory_chunks == 0) {
23b6c52c	221	printk(KERN_DEBUG
e0e5ea32	222	"could not find any ACPI SRAT memory areas.\n");
1da177e4 LT	223	goto out_fail;
	224	}
	225
	226	/* Calculate total number of nodes in system from PXM bitmap and create
	227	* a set of sequential node IDs starting at zero. (ACPI doesn't seem
	228	* to specify the range of _PXM values.)
	229	*/
	230	/*
	231	* MCD - we no longer HAVE to number nodes sequentially. PXM domain
	232	* numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
	233	* 32, so we will continue numbering them in this manner until MAX_NUMNODES
	234	* approaches MAX_PXM_DOMAINS for i386.
	235	*/
	236	nodes_clear(node_online_map);
	237	for (i = 0; i < MAX_PXM_DOMAINS; i++) {
	238	if (BMAP_TEST(pxm_bitmap, i)) {
762834e8	239	int nid = acpi_map_pxm_to_node(i);
1da177e4 LT	240	node_set_online(nid);
	241	}
	242	}
	243	BUG_ON(num_online_nodes() == 0);
	244
	245	/* set cnode id in memory chunk structure */
	246	for (i = 0; i < num_memory_chunks; i++)
762834e8	247	node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);
1da177e4	248
c0943457	249	printk(KERN_DEBUG "pxm bitmap: ");
1da177e4	250	for (i = 0; i < sizeof(pxm_bitmap); i++) {
c0943457	251	printk(KERN_CONT "%02x ", pxm_bitmap[i]);
1da177e4	252	}
c0943457 YL	253	printk(KERN_CONT "\n");
	254	printk(KERN_DEBUG "Number of logical nodes in system = %d\n",
	255	num_online_nodes());
	256	printk(KERN_DEBUG "Number of memory chunks in system = %d\n",
	257	num_memory_chunks);
1da177e4	258
b78aa66b	259	for (i = 0; i < MAX_LOCAL_APIC; i++)
bbc9e2f4	260	set_apicid_to_node(i, pxm_to_node(apicid_to_pxm[i]));
3b08606d	261
1da177e4 LT	262	for (j = 0; j < num_memory_chunks; j++){
1da177e4 LT	263	struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
c0943457 YL	264	printk(KERN_DEBUG
c0943457 YL	265	"chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
1da177e4	266	j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
858f7747 YL	267	if (node_read_chunk(chunk->nid, chunk))
	268	continue;
	269
a9ce6bc1	270	memblock_x86_register_active_regions(chunk->nid, chunk->start_pfn,
7b2a0a6c	271	min(chunk->end_pfn, max_pfn));
23774942	272	}
32996250 YL	273	/* for out of order entries in SRAT */
32996250 YL	274	sort_node_map();
1c6e5503	275
1da177e4 LT	276	return 1;
1da177e4 LT	277	out_fail:
23b6c52c	278	printk(KERN_DEBUG "failed to get NUMA memory information from SRAT"
c0943457	279	" table\n");
1da177e4 LT	280	return 0;
1da177e4 LT	281	}