1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * AMD NUMA support.
4 * Discover the memory map and associated nodes.
5 *
6 * This version reads it directly from the AMD northbridge.
7 *
8 * Copyright 2002,2003 Andi Kleen, SuSE Labs.
9 */
10 #include <linux/kernel.h>
11 #include <linux/init.h>
12 #include <linux/string.h>
13 #include <linux/nodemask.h>
14 #include <linux/memblock.h>
15 #include <linux/numa_memblks.h>
16
17 #include <asm/io.h>
18 #include <linux/pci_ids.h>
19 #include <linux/acpi.h>
20 #include <asm/types.h>
21 #include <asm/mmzone.h>
22 #include <asm/proto.h>
23 #include <asm/e820/api.h>
24 #include <asm/pci-direct.h>
25 #include <asm/numa.h>
26 #include <asm/mpspec.h>
27 #include <asm/apic.h>
28 #include <asm/amd_nb.h>
29
30 static unsigned char __initdata nodeids[8];
31
find_northbridge(void)32 static __init int find_northbridge(void)
33 {
34 int num;
35
36 for (num = 0; num < 32; num++) {
37 u32 header;
38
39 header = read_pci_config(0, num, 0, 0x00);
40 if (header != (PCI_VENDOR_ID_AMD | (0x1100<<16)) &&
41 header != (PCI_VENDOR_ID_AMD | (0x1200<<16)) &&
42 header != (PCI_VENDOR_ID_AMD | (0x1300<<16)))
43 continue;
44
45 header = read_pci_config(0, num, 1, 0x00);
46 if (header != (PCI_VENDOR_ID_AMD | (0x1101<<16)) &&
47 header != (PCI_VENDOR_ID_AMD | (0x1201<<16)) &&
48 header != (PCI_VENDOR_ID_AMD | (0x1301<<16)))
49 continue;
50 return num;
51 }
52
53 return -ENOENT;
54 }
55
amd_numa_init(void)56 int __init amd_numa_init(void)
57 {
58 unsigned int numnodes, cores, apicid;
59 u64 prevbase, start = PFN_PHYS(0);
60 u64 end = PFN_PHYS(max_pfn);
61 u32 nodeid, reg;
62 int i, j, nb;
63
64 if (!early_pci_allowed())
65 return -EINVAL;
66
67 nb = find_northbridge();
68 if (nb < 0)
69 return nb;
70
71 pr_info("Scanning NUMA topology in Northbridge %d\n", nb);
72
73 reg = read_pci_config(0, nb, 0, 0x60);
74 numnodes = ((reg >> 4) & 0xF) + 1;
75 if (numnodes <= 1)
76 return -ENOENT;
77
78 pr_info("Number of physical nodes %d\n", numnodes);
79
80 prevbase = 0;
81 for (i = 0; i < 8; i++) {
82 u64 base, limit;
83
84 base = read_pci_config(0, nb, 1, 0x40 + i*8);
85 limit = read_pci_config(0, nb, 1, 0x44 + i*8);
86
87 nodeids[i] = nodeid = limit & 7;
88 if ((base & 3) == 0) {
89 if (i < numnodes)
90 pr_info("Skipping disabled node %d\n", i);
91 continue;
92 }
93 if (nodeid >= numnodes) {
94 pr_info("Ignoring excess node %d (%Lx:%Lx)\n", nodeid,
95 base, limit);
96 continue;
97 }
98
99 if (!limit) {
100 pr_info("Skipping node entry %d (base %Lx)\n",
101 i, base);
102 continue;
103 }
104 if ((base >> 8) & 3 || (limit >> 8) & 3) {
105 pr_err("Node %d using interleaving mode %Lx/%Lx\n",
106 nodeid, (base >> 8) & 3, (limit >> 8) & 3);
107 return -EINVAL;
108 }
109 if (node_isset(nodeid, numa_nodes_parsed)) {
110 pr_info("Node %d already present, skipping\n",
111 nodeid);
112 continue;
113 }
114
115 limit >>= 16;
116 limit++;
117 limit <<= 24;
118
119 if (limit > end)
120 limit = end;
121 if (limit <= base)
122 continue;
123
124 base >>= 16;
125 base <<= 24;
126
127 if (base < start)
128 base = start;
129 if (limit > end)
130 limit = end;
131 if (limit == base) {
132 pr_err("Empty node %d\n", nodeid);
133 continue;
134 }
135 if (limit < base) {
136 pr_err("Node %d bogus settings %Lx-%Lx.\n",
137 nodeid, base, limit);
138 continue;
139 }
140
141 /* Could sort here, but pun for now. Should not happen anyroads. */
142 if (prevbase > base) {
143 pr_err("Node map not sorted %Lx,%Lx\n",
144 prevbase, base);
145 return -EINVAL;
146 }
147
148 pr_info("Node %d MemBase %016Lx Limit %016Lx\n",
149 nodeid, base, limit);
150
151 prevbase = base;
152 numa_add_memblk(nodeid, base, limit);
153 node_set(nodeid, numa_nodes_parsed);
154 }
155
156 if (nodes_empty(numa_nodes_parsed))
157 return -ENOENT;
158
159 /*
160 * We seem to have valid NUMA configuration. Map apicids to nodes
161 * using the size of the core domain in the APIC space.
162 */
163 cores = topology_get_domain_size(TOPO_CORE_DOMAIN);
164
165 apicid = boot_cpu_physical_apicid;
166 if (apicid > 0)
167 pr_info("BSP APIC ID: %02x\n", apicid);
168
169 for_each_node_mask(i, numa_nodes_parsed) {
170 for (j = 0; j < cores; j++, apicid++)
171 set_apicid_to_node(apicid, i);
172 }
173 return 0;
174 }
175