Merge branch 'x86/paravirt' into x86/apic
[linux-2.6] / arch / x86 / mm / k8topology_64.c
1 /*
2  * AMD K8 NUMA support.
3  * Discover the memory map and associated nodes.
4  *
5  * This version reads it directly from the K8 northbridge.
6  *
7  * 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>
14 #include <asm/io.h>
15 #include <linux/pci_ids.h>
16 #include <linux/acpi.h>
17 #include <asm/types.h>
18 #include <asm/mmzone.h>
19 #include <asm/proto.h>
20 #include <asm/e820.h>
21 #include <asm/pci-direct.h>
22 #include <asm/numa.h>
23 #include <asm/mpspec.h>
24 #include <asm/apic.h>
25 #include <asm/k8.h>
26
27 static __init int find_northbridge(void)
28 {
29         int num;
30
31         for (num = 0; num < 32; num++) {
32                 u32 header;
33
34                 header = read_pci_config(0, num, 0, 0x00);
35                 if (header != (PCI_VENDOR_ID_AMD | (0x1100<<16)) &&
36                         header != (PCI_VENDOR_ID_AMD | (0x1200<<16)) &&
37                         header != (PCI_VENDOR_ID_AMD | (0x1300<<16)))
38                         continue;
39
40                 header = read_pci_config(0, num, 1, 0x00);
41                 if (header != (PCI_VENDOR_ID_AMD | (0x1101<<16)) &&
42                         header != (PCI_VENDOR_ID_AMD | (0x1201<<16)) &&
43                         header != (PCI_VENDOR_ID_AMD | (0x1301<<16)))
44                         continue;
45                 return num;
46         }
47
48         return -1;
49 }
50
51 static __init void early_get_boot_cpu_id(void)
52 {
53         /*
54          * need to get boot_cpu_id so can use that to create apicid_to_node
55          * in k8_scan_nodes()
56          */
57         /*
58          * Find possible boot-time SMP configuration:
59          */
60 #ifdef CONFIG_X86_MPPARSE
61         early_find_smp_config();
62 #endif
63 #ifdef CONFIG_ACPI
64         /*
65          * Read APIC information from ACPI tables.
66          */
67         early_acpi_boot_init();
68 #endif
69 #ifdef CONFIG_X86_MPPARSE
70         /*
71          * get boot-time SMP configuration:
72          */
73         if (smp_found_config)
74                 early_get_smp_config();
75 #endif
76         early_init_lapic_mapping();
77 }
78
79 int __init k8_scan_nodes(unsigned long start, unsigned long end)
80 {
81         unsigned numnodes, cores, bits, apicid_base;
82         unsigned long prevbase;
83         struct bootnode nodes[8];
84         int i, j, nb, found = 0;
85         u32 nodeid, reg;
86
87         if (!early_pci_allowed())
88                 return -1;
89
90         nb = find_northbridge();
91         if (nb < 0)
92                 return nb;
93
94         printk(KERN_INFO "Scanning NUMA topology in Northbridge %d\n", nb);
95
96         reg = read_pci_config(0, nb, 0, 0x60);
97         numnodes = ((reg >> 4) & 0xF) + 1;
98         if (numnodes <= 1)
99                 return -1;
100
101         printk(KERN_INFO "Number of nodes %d\n", numnodes);
102
103         memset(&nodes, 0, sizeof(nodes));
104         prevbase = 0;
105         for (i = 0; i < 8; i++) {
106                 unsigned long base, limit;
107
108                 base = read_pci_config(0, nb, 1, 0x40 + i*8);
109                 limit = read_pci_config(0, nb, 1, 0x44 + i*8);
110
111                 nodeid = limit & 7;
112                 if ((base & 3) == 0) {
113                         if (i < numnodes)
114                                 printk("Skipping disabled node %d\n", i);
115                         continue;
116                 }
117                 if (nodeid >= numnodes) {
118                         printk("Ignoring excess node %d (%lx:%lx)\n", nodeid,
119                                base, limit);
120                         continue;
121                 }
122
123                 if (!limit) {
124                         printk(KERN_INFO "Skipping node entry %d (base %lx)\n",
125                                i, base);
126                         continue;
127                 }
128                 if ((base >> 8) & 3 || (limit >> 8) & 3) {
129                         printk(KERN_ERR "Node %d using interleaving mode %lx/%lx\n",
130                                nodeid, (base>>8)&3, (limit>>8) & 3);
131                         return -1;
132                 }
133                 if (node_isset(nodeid, node_possible_map)) {
134                         printk(KERN_INFO "Node %d already present. Skipping\n",
135                                nodeid);
136                         continue;
137                 }
138
139                 limit >>= 16;
140                 limit <<= 24;
141                 limit |= (1<<24)-1;
142                 limit++;
143
144                 if (limit > max_pfn << PAGE_SHIFT)
145                         limit = max_pfn << PAGE_SHIFT;
146                 if (limit <= base)
147                         continue;
148
149                 base >>= 16;
150                 base <<= 24;
151
152                 if (base < start)
153                         base = start;
154                 if (limit > end)
155                         limit = end;
156                 if (limit == base) {
157                         printk(KERN_ERR "Empty node %d\n", nodeid);
158                         continue;
159                 }
160                 if (limit < base) {
161                         printk(KERN_ERR "Node %d bogus settings %lx-%lx.\n",
162                                nodeid, base, limit);
163                         continue;
164                 }
165
166                 /* Could sort here, but pun for now. Should not happen anyroads. */
167                 if (prevbase > base) {
168                         printk(KERN_ERR "Node map not sorted %lx,%lx\n",
169                                prevbase, base);
170                         return -1;
171                 }
172
173                 printk(KERN_INFO "Node %d MemBase %016lx Limit %016lx\n",
174                        nodeid, base, limit);
175
176                 found++;
177
178                 nodes[nodeid].start = base;
179                 nodes[nodeid].end = limit;
180
181                 prevbase = base;
182
183                 node_set(nodeid, node_possible_map);
184         }
185
186         if (!found)
187                 return -1;
188
189         memnode_shift = compute_hash_shift(nodes, 8, NULL);
190         if (memnode_shift < 0) {
191                 printk(KERN_ERR "No NUMA node hash function found. Contact maintainer\n");
192                 return -1;
193         }
194         printk(KERN_INFO "Using node hash shift of %d\n", memnode_shift);
195
196         /* use the coreid bits from early_identify_cpu */
197         bits = boot_cpu_data.x86_coreid_bits;
198         cores = (1<<bits);
199         apicid_base = 0;
200         /* need to get boot_cpu_id early for system with apicid lifting */
201         early_get_boot_cpu_id();
202         if (boot_cpu_physical_apicid > 0) {
203                 printk(KERN_INFO "BSP APIC ID: %02x\n",
204                                  boot_cpu_physical_apicid);
205                 apicid_base = boot_cpu_physical_apicid;
206         }
207
208         for (i = 0; i < 8; i++) {
209                 if (nodes[i].start == nodes[i].end)
210                         continue;
211
212                 e820_register_active_regions(i,
213                                 nodes[i].start >> PAGE_SHIFT,
214                                 nodes[i].end >> PAGE_SHIFT);
215                 for (j = apicid_base; j < cores + apicid_base; j++)
216                         apicid_to_node[(i << bits) + j] = i;
217                 setup_node_bootmem(i, nodes[i].start, nodes[i].end);
218         }
219
220         numa_init_array();
221         return 0;
222 }