blob: 5247d01329ca96de7e9eacbb3f2511ff8f9c3a59 [file] [log] [blame]
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +01001/*
Hans Rosenfeldeec1d4f2010-10-29 17:14:30 +02002 * AMD NUMA support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 * Discover the memory map and associated nodes.
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +01004 *
Hans Rosenfeldeec1d4f2010-10-29 17:14:30 +02005 * This version reads it directly from the AMD northbridge.
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +01006 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * Copyright 2002,2003 Andi Kleen, SuSE Labs.
8 */
9#include <linux/kernel.h>
10#include <linux/init.h>
11#include <linux/string.h>
12#include <linux/module.h>
13#include <linux/nodemask.h>
Yinghai Lua9ce6bc2010-08-25 13:39:17 -070014#include <linux/memblock.h>
Tejun Heo2706a0b2011-05-02 17:24:48 +020015#include <linux/bootmem.h>
Yinghai Lua9ce6bc2010-08-25 13:39:17 -070016
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <asm/io.h>
18#include <linux/pci_ids.h>
Yinghai Lucbf9bd62008-02-19 03:21:06 -080019#include <linux/acpi.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <asm/types.h>
21#include <asm/mmzone.h>
22#include <asm/proto.h>
23#include <asm/e820.h>
24#include <asm/pci-direct.h>
25#include <asm/numa.h>
Yinghai Lucbf9bd62008-02-19 03:21:06 -080026#include <asm/mpspec.h>
27#include <asm/apic.h>
Andreas Herrmann23ac4ae2010-09-17 18:03:43 +020028#include <asm/amd_nb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
David Rientjesf51bf302010-12-22 17:23:51 -080030static unsigned char __initdata nodeids[8];
David Rientjes8ee2deb2009-09-25 15:20:00 -070031
Linus Torvalds1da177e2005-04-16 15:20:36 -070032static __init int find_northbridge(void)
33{
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +010034 int num;
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +010036 for (num = 0; num < 32; num++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 u32 header;
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +010038
39 header = read_pci_config(0, num, 0, 0x00);
Joachim Deguarabb4a1d62008-01-30 13:34:12 +010040 if (header != (PCI_VENDOR_ID_AMD | (0x1100<<16)) &&
41 header != (PCI_VENDOR_ID_AMD | (0x1200<<16)) &&
42 header != (PCI_VENDOR_ID_AMD | (0x1300<<16)))
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +010043 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +010045 header = read_pci_config(0, num, 1, 0x00);
Joachim Deguarabb4a1d62008-01-30 13:34:12 +010046 if (header != (PCI_VENDOR_ID_AMD | (0x1101<<16)) &&
47 header != (PCI_VENDOR_ID_AMD | (0x1201<<16)) &&
48 header != (PCI_VENDOR_ID_AMD | (0x1301<<16)))
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +010049 continue;
50 return num;
51 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Tejun Heo940fed22011-02-16 12:13:06 +010053 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054}
55
Yinghai Lucbf9bd62008-02-19 03:21:06 -080056static __init void early_get_boot_cpu_id(void)
57{
58 /*
Robert Richterf6e94562010-07-21 19:03:58 +020059 * need to get the APIC ID of the BSP so can use that to
Hans Rosenfeldeec1d4f2010-10-29 17:14:30 +020060 * create apicid_to_node in amd_scan_nodes()
Yinghai Lucbf9bd62008-02-19 03:21:06 -080061 */
Yinghai Lua4caa182008-06-19 12:15:01 -070062#ifdef CONFIG_X86_MPPARSE
Yinghai Lucbf9bd62008-02-19 03:21:06 -080063 /*
64 * get boot-time SMP configuration:
65 */
66 if (smp_found_config)
67 early_get_smp_config();
Yinghai Lua4caa182008-06-19 12:15:01 -070068#endif
David Rientjes8ee2deb2009-09-25 15:20:00 -070069}
70
Tejun Heo940fed22011-02-16 12:13:06 +010071int __init amd_numa_init(void)
David Rientjes8ee2deb2009-09-25 15:20:00 -070072{
Tejun Heo2706a0b2011-05-02 17:24:48 +020073 u64 start = PFN_PHYS(0);
74 u64 end = PFN_PHYS(max_pfn);
David Rientjes8ee2deb2009-09-25 15:20:00 -070075 unsigned numnodes;
Tejun Heo2706a0b2011-05-02 17:24:48 +020076 u64 prevbase;
Tejun Heo45fe6c72011-02-16 12:13:07 +010077 int i, j, nb;
Thomas Gleixnerd34c0892008-05-12 15:43:35 +020078 u32 nodeid, reg;
Tejun Heo45fe6c72011-02-16 12:13:07 +010079 unsigned int bits, cores, apicid_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Andi Kleen0637a702006-09-26 10:52:41 +020081 if (!early_pci_allowed())
Tejun Heo940fed22011-02-16 12:13:06 +010082 return -EINVAL;
Andi Kleen0637a702006-09-26 10:52:41 +020083
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +010084 nb = find_northbridge();
85 if (nb < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 return nb;
87
David Rientjes1af5ba52009-09-25 15:19:47 -070088 pr_info("Scanning NUMA topology in Northbridge %d\n", nb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +010090 reg = read_pci_config(0, nb, 0, 0x60);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 numnodes = ((reg >> 4) & 0xF) + 1;
Andi Kleen3bea9c92007-05-02 19:27:21 +020092 if (numnodes <= 1)
Tejun Heo940fed22011-02-16 12:13:06 +010093 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
David Rientjes8ee2deb2009-09-25 15:20:00 -070095 pr_info("Number of physical nodes %d\n", numnodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 prevbase = 0;
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +010098 for (i = 0; i < 8; i++) {
Tejun Heo2706a0b2011-05-02 17:24:48 +020099 u64 base, limit;
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 base = read_pci_config(0, nb, 1, 0x40 + i*8);
102 limit = read_pci_config(0, nb, 1, 0x44 + i*8);
103
David Rientjesf51bf302010-12-22 17:23:51 -0800104 nodeids[i] = nodeid = limit & 7;
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100105 if ((base & 3) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 if (i < numnodes)
David Rientjes1af5ba52009-09-25 15:19:47 -0700107 pr_info("Skipping disabled node %d\n", i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 continue;
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 if (nodeid >= numnodes) {
Tejun Heo2706a0b2011-05-02 17:24:48 +0200111 pr_info("Ignoring excess node %d (%Lx:%Lx)\n", nodeid,
David Rientjes1af5ba52009-09-25 15:19:47 -0700112 base, limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 continue;
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100116 if (!limit) {
Tejun Heo2706a0b2011-05-02 17:24:48 +0200117 pr_info("Skipping node entry %d (base %Lx)\n",
David Rientjes1af5ba52009-09-25 15:19:47 -0700118 i, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 continue;
120 }
121 if ((base >> 8) & 3 || (limit >> 8) & 3) {
Tejun Heo2706a0b2011-05-02 17:24:48 +0200122 pr_err("Node %d using interleaving mode %Lx/%Lx\n",
David Rientjes1af5ba52009-09-25 15:19:47 -0700123 nodeid, (base >> 8) & 3, (limit >> 8) & 3);
Tejun Heo940fed22011-02-16 12:13:06 +0100124 return -EINVAL;
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100125 }
Tejun Heo4697bdc2011-02-16 17:11:09 +0100126 if (node_isset(nodeid, numa_nodes_parsed)) {
David Rientjes1af5ba52009-09-25 15:19:47 -0700127 pr_info("Node %d already present, skipping\n",
128 nodeid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 continue;
130 }
131
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100132 limit >>= 16;
133 limit <<= 24;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 limit |= (1<<24)-1;
Magnus Dammffd10a22005-11-05 17:25:54 +0100135 limit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
David Rientjes8ee2deb2009-09-25 15:20:00 -0700137 if (limit > end)
138 limit = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 if (limit <= base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 continue;
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100141
142 base >>= 16;
143 base <<= 24;
144
145 if (base < start)
146 base = start;
147 if (limit > end)
148 limit = end;
149 if (limit == base) {
David Rientjes1af5ba52009-09-25 15:19:47 -0700150 pr_err("Empty node %d\n", nodeid);
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100151 continue;
152 }
153 if (limit < base) {
Tejun Heo2706a0b2011-05-02 17:24:48 +0200154 pr_err("Node %d bogus settings %Lx-%Lx.\n",
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100155 nodeid, base, limit);
156 continue;
157 }
158
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 /* Could sort here, but pun for now. Should not happen anyroads. */
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100160 if (prevbase > base) {
Tejun Heo2706a0b2011-05-02 17:24:48 +0200161 pr_err("Node map not sorted %Lx,%Lx\n",
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100162 prevbase, base);
Tejun Heo940fed22011-02-16 12:13:06 +0100163 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 }
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100165
Tejun Heo2706a0b2011-05-02 17:24:48 +0200166 pr_info("Node %d MemBase %016Lx Limit %016Lx\n",
David Rientjes1af5ba52009-09-25 15:19:47 -0700167 nodeid, base, limit);
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100168
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 prevbase = base;
Tejun Heo91556232011-02-16 17:11:09 +0100170 numa_add_memblk(nodeid, base, limit);
Tejun Heo92d4a432011-02-16 17:11:09 +0100171 node_set(nodeid, numa_nodes_parsed);
Carlos R. Mafra3aa88cd2008-01-30 13:32:36 +0100172 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Tejun Heo4697bdc2011-02-16 17:11:09 +0100174 if (!nodes_weight(numa_nodes_parsed))
Tejun Heo940fed22011-02-16 12:13:06 +0100175 return -ENOENT;
Tejun Heo45fe6c72011-02-16 12:13:07 +0100176
177 /*
178 * We seem to have valid NUMA configuration. Map apicids to nodes
179 * using the coreid bits from early_identify_cpu.
180 */
181 bits = boot_cpu_data.x86_coreid_bits;
182 cores = 1 << bits;
183 apicid_base = 0;
184
185 /* get the APIC ID of the BSP early for systems with apicid lifting */
186 early_get_boot_cpu_id();
187 if (boot_cpu_physical_apicid > 0) {
188 pr_info("BSP APIC ID: %02x\n", boot_cpu_physical_apicid);
189 apicid_base = boot_cpu_physical_apicid;
190 }
191
Tejun Heo92d4a432011-02-16 17:11:09 +0100192 for_each_node_mask(i, numa_nodes_parsed)
Tejun Heo45fe6c72011-02-16 12:13:07 +0100193 for (j = apicid_base; j < cores + apicid_base; j++)
194 set_apicid_to_node((i << bits) + j, i);
195
David Rientjes8ee2deb2009-09-25 15:20:00 -0700196 return 0;
197}