aboutsummaryrefslogtreecommitdiffstats
path: root/arch
diff options
context:
space:
mode:
Diffstat (limited to 'arch')
-rw-r--r--arch/x86/kernel/Makefile1
-rw-r--r--arch/x86/kernel/apic_32.c5
-rw-r--r--arch/x86/kernel/apic_64.c26
-rw-r--r--arch/x86/kernel/cpu/Makefile4
-rw-r--r--arch/x86/kernel/cpu/amd.c42
-rw-r--r--arch/x86/kernel/cpu/amd_64.c211
-rw-r--r--arch/x86/kernel/cpu/bugs_64.c (renamed from arch/x86/kernel/bugs_64.c)0
-rw-r--r--arch/x86/kernel/cpu/centaur_64.c43
-rw-r--r--arch/x86/kernel/cpu/cpu.h5
-rw-r--r--arch/x86/kernel/cpu/intel_64.c103
-rw-r--r--arch/x86/kernel/mmconf-fam10h_64.c1
-rw-r--r--arch/x86/kernel/process.c190
-rw-r--r--arch/x86/kernel/process_32.c61
-rw-r--r--arch/x86/kernel/process_64.c35
-rw-r--r--arch/x86/kernel/setup_64.c450
-rw-r--r--arch/x86/kernel/vmlinux_64.lds.S1
-rw-r--r--arch/x86/pci/Makefile_321
-rw-r--r--arch/x86/pci/Makefile_642
-rw-r--r--arch/x86/pci/amd_bus.c (renamed from arch/x86/pci/k8-bus_64.c)32
-rw-r--r--arch/x86/pci/direct.c25
-rw-r--r--arch/x86/pci/pci.h1
21 files changed, 666 insertions, 573 deletions
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 77807d4769c9..3a2a54c950ff 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -25,7 +25,6 @@ obj-$(CONFIG_X86_64) += syscall_64.o vsyscall_64.o setup64.o
25obj-y += bootflag.o e820_$(BITS).o 25obj-y += bootflag.o e820_$(BITS).o
26obj-y += pci-dma.o quirks.o i8237.o topology.o kdebugfs.o 26obj-y += pci-dma.o quirks.o i8237.o topology.o kdebugfs.o
27obj-y += alternative.o i8253.o pci-nommu.o 27obj-y += alternative.o i8253.o pci-nommu.o
28obj-$(CONFIG_X86_64) += bugs_64.o
29obj-y += tsc_$(BITS).o io_delay.o rtc.o 28obj-y += tsc_$(BITS).o io_delay.o rtc.o
30 29
31obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o 30obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o
diff --git a/arch/x86/kernel/apic_32.c b/arch/x86/kernel/apic_32.c
index 4b99b1bdeb6c..c44206e731d4 100644
--- a/arch/x86/kernel/apic_32.c
+++ b/arch/x86/kernel/apic_32.c
@@ -64,9 +64,8 @@ static int enable_local_apic __initdata;
64 64
65/* Local APIC timer verification ok */ 65/* Local APIC timer verification ok */
66static int local_apic_timer_verify_ok; 66static int local_apic_timer_verify_ok;
67/* Disable local APIC timer from the kernel commandline or via dmi quirk 67/* Disable local APIC timer from the kernel commandline or via dmi quirk */
68 or using CPU MSR check */ 68static int local_apic_timer_disabled;
69int local_apic_timer_disabled;
70/* Local APIC timer works in C2 */ 69/* Local APIC timer works in C2 */
71int local_apic_timer_c2_ok; 70int local_apic_timer_c2_ok;
72EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok); 71EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
diff --git a/arch/x86/kernel/apic_64.c b/arch/x86/kernel/apic_64.c
index 0633cfd0dc29..a5cc8447cf4d 100644
--- a/arch/x86/kernel/apic_64.c
+++ b/arch/x86/kernel/apic_64.c
@@ -43,7 +43,7 @@
43#include <mach_ipi.h> 43#include <mach_ipi.h>
44#include <mach_apic.h> 44#include <mach_apic.h>
45 45
46int disable_apic_timer __cpuinitdata; 46static int disable_apic_timer __cpuinitdata;
47static int apic_calibrate_pmtmr __initdata; 47static int apic_calibrate_pmtmr __initdata;
48int disable_apic; 48int disable_apic;
49 49
@@ -422,32 +422,8 @@ void __init setup_boot_APIC_clock(void)
422 setup_APIC_timer(); 422 setup_APIC_timer();
423} 423}
424 424
425/*
426 * AMD C1E enabled CPUs have a real nasty problem: Some BIOSes set the
427 * C1E flag only in the secondary CPU, so when we detect the wreckage
428 * we already have enabled the boot CPU local apic timer. Check, if
429 * disable_apic_timer is set and the DUMMY flag is cleared. If yes,
430 * set the DUMMY flag again and force the broadcast mode in the
431 * clockevents layer.
432 */
433static void __cpuinit check_boot_apic_timer_broadcast(void)
434{
435 if (!disable_apic_timer ||
436 (lapic_clockevent.features & CLOCK_EVT_FEAT_DUMMY))
437 return;
438
439 printk(KERN_INFO "AMD C1E detected late. Force timer broadcast.\n");
440 lapic_clockevent.features |= CLOCK_EVT_FEAT_DUMMY;
441
442 local_irq_enable();
443 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
444 &boot_cpu_physical_apicid);
445 local_irq_disable();
446}
447
448void __cpuinit setup_secondary_APIC_clock(void) 425void __cpuinit setup_secondary_APIC_clock(void)
449{ 426{
450 check_boot_apic_timer_broadcast();
451 setup_APIC_timer(); 427 setup_APIC_timer();
452} 428}
453 429
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index a0c6f8190887..65b1be5fe9ce 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -6,11 +6,15 @@ obj-y := intel_cacheinfo.o addon_cpuid_features.o
6obj-y += proc.o feature_names.o 6obj-y += proc.o feature_names.o
7 7
8obj-$(CONFIG_X86_32) += common.o bugs.o 8obj-$(CONFIG_X86_32) += common.o bugs.o
9obj-$(CONFIG_X86_64) += bugs_64.o
9obj-$(CONFIG_X86_32) += amd.o 10obj-$(CONFIG_X86_32) += amd.o
11obj-$(CONFIG_X86_64) += amd_64.o
10obj-$(CONFIG_X86_32) += cyrix.o 12obj-$(CONFIG_X86_32) += cyrix.o
11obj-$(CONFIG_X86_32) += centaur.o 13obj-$(CONFIG_X86_32) += centaur.o
14obj-$(CONFIG_X86_64) += centaur_64.o
12obj-$(CONFIG_X86_32) += transmeta.o 15obj-$(CONFIG_X86_32) += transmeta.o
13obj-$(CONFIG_X86_32) += intel.o 16obj-$(CONFIG_X86_32) += intel.o
17obj-$(CONFIG_X86_64) += intel_64.o
14obj-$(CONFIG_X86_32) += umc.o 18obj-$(CONFIG_X86_32) += umc.o
15 19
16obj-$(CONFIG_X86_MCE) += mcheck/ 20obj-$(CONFIG_X86_MCE) += mcheck/
diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index 245866828294..81a07ca65d44 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -24,43 +24,6 @@
24extern void vide(void); 24extern void vide(void);
25__asm__(".align 4\nvide: ret"); 25__asm__(".align 4\nvide: ret");
26 26
27#ifdef CONFIG_X86_LOCAL_APIC
28#define ENABLE_C1E_MASK 0x18000000
29#define CPUID_PROCESSOR_SIGNATURE 1
30#define CPUID_XFAM 0x0ff00000
31#define CPUID_XFAM_K8 0x00000000
32#define CPUID_XFAM_10H 0x00100000
33#define CPUID_XFAM_11H 0x00200000
34#define CPUID_XMOD 0x000f0000
35#define CPUID_XMOD_REV_F 0x00040000
36
37/* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
38static __cpuinit int amd_apic_timer_broken(void)
39{
40 u32 lo, hi;
41 u32 eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
42 switch (eax & CPUID_XFAM) {
43 case CPUID_XFAM_K8:
44 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
45 break;
46 case CPUID_XFAM_10H:
47 case CPUID_XFAM_11H:
48 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
49 if (lo & ENABLE_C1E_MASK) {
50 if (smp_processor_id() != boot_cpu_physical_apicid)
51 printk(KERN_INFO "AMD C1E detected late. "
52 " Force timer broadcast.\n");
53 return 1;
54 }
55 break;
56 default:
57 /* err on the side of caution */
58 return 1;
59 }
60 return 0;
61}
62#endif
63
64int force_mwait __cpuinitdata; 27int force_mwait __cpuinitdata;
65 28
66static void __cpuinit early_init_amd(struct cpuinfo_x86 *c) 29static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
@@ -297,11 +260,6 @@ static void __cpuinit init_amd(struct cpuinfo_x86 *c)
297 num_cache_leaves = 3; 260 num_cache_leaves = 3;
298 } 261 }
299 262
300#ifdef CONFIG_X86_LOCAL_APIC
301 if (amd_apic_timer_broken())
302 local_apic_timer_disabled = 1;
303#endif
304
305 /* K6s reports MCEs but don't actually have all the MSRs */ 263 /* K6s reports MCEs but don't actually have all the MSRs */
306 if (c->x86 < 6) 264 if (c->x86 < 6)
307 clear_cpu_cap(c, X86_FEATURE_MCE); 265 clear_cpu_cap(c, X86_FEATURE_MCE);
diff --git a/arch/x86/kernel/cpu/amd_64.c b/arch/x86/kernel/cpu/amd_64.c
new file mode 100644
index 000000000000..30b7557c9641
--- /dev/null
+++ b/arch/x86/kernel/cpu/amd_64.c
@@ -0,0 +1,211 @@
1#include <linux/init.h>
2#include <linux/mm.h>
3
4#include <asm/numa_64.h>
5#include <asm/mmconfig.h>
6#include <asm/cacheflush.h>
7
8#include <mach_apic.h>
9
10#include "cpu.h"
11
12int force_mwait __cpuinitdata;
13
14#ifdef CONFIG_NUMA
15static int __cpuinit nearby_node(int apicid)
16{
17 int i, node;
18
19 for (i = apicid - 1; i >= 0; i--) {
20 node = apicid_to_node[i];
21 if (node != NUMA_NO_NODE && node_online(node))
22 return node;
23 }
24 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
25 node = apicid_to_node[i];
26 if (node != NUMA_NO_NODE && node_online(node))
27 return node;
28 }
29 return first_node(node_online_map); /* Shouldn't happen */
30}
31#endif
32
33/*
34 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
35 * Assumes number of cores is a power of two.
36 */
37static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
38{
39#ifdef CONFIG_SMP
40 unsigned bits;
41#ifdef CONFIG_NUMA
42 int cpu = smp_processor_id();
43 int node = 0;
44 unsigned apicid = hard_smp_processor_id();
45#endif
46 bits = c->x86_coreid_bits;
47
48 /* Low order bits define the core id (index of core in socket) */
49 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
50 /* Convert the initial APIC ID into the socket ID */
51 c->phys_proc_id = c->initial_apicid >> bits;
52
53#ifdef CONFIG_NUMA
54 node = c->phys_proc_id;
55 if (apicid_to_node[apicid] != NUMA_NO_NODE)
56 node = apicid_to_node[apicid];
57 if (!node_online(node)) {
58 /* Two possibilities here:
59 - The CPU is missing memory and no node was created.
60 In that case try picking one from a nearby CPU
61 - The APIC IDs differ from the HyperTransport node IDs
62 which the K8 northbridge parsing fills in.
63 Assume they are all increased by a constant offset,
64 but in the same order as the HT nodeids.
65 If that doesn't result in a usable node fall back to the
66 path for the previous case. */
67
68 int ht_nodeid = c->initial_apicid;
69
70 if (ht_nodeid >= 0 &&
71 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
72 node = apicid_to_node[ht_nodeid];
73 /* Pick a nearby node */
74 if (!node_online(node))
75 node = nearby_node(apicid);
76 }
77 numa_set_node(cpu, node);
78
79 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
80#endif
81#endif
82}
83
84static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
85{
86#ifdef CONFIG_SMP
87 unsigned bits, ecx;
88
89 /* Multi core CPU? */
90 if (c->extended_cpuid_level < 0x80000008)
91 return;
92
93 ecx = cpuid_ecx(0x80000008);
94
95 c->x86_max_cores = (ecx & 0xff) + 1;
96
97 /* CPU telling us the core id bits shift? */
98 bits = (ecx >> 12) & 0xF;
99
100 /* Otherwise recompute */
101 if (bits == 0) {
102 while ((1 << bits) < c->x86_max_cores)
103 bits++;
104 }
105
106 c->x86_coreid_bits = bits;
107
108#endif
109}
110
111static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
112{
113 early_init_amd_mc(c);
114
115 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
116 if (c->x86_power & (1<<8))
117 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
118}
119
120static void __cpuinit init_amd(struct cpuinfo_x86 *c)
121{
122 unsigned level;
123
124#ifdef CONFIG_SMP
125 unsigned long value;
126
127 /*
128 * Disable TLB flush filter by setting HWCR.FFDIS on K8
129 * bit 6 of msr C001_0015
130 *
131 * Errata 63 for SH-B3 steppings
132 * Errata 122 for all steppings (F+ have it disabled by default)
133 */
134 if (c->x86 == 15) {
135 rdmsrl(MSR_K8_HWCR, value);
136 value |= 1 << 6;
137 wrmsrl(MSR_K8_HWCR, value);
138 }
139#endif
140
141 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
142 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
143 clear_cpu_cap(c, 0*32+31);
144
145 /* On C+ stepping K8 rep microcode works well for copy/memset */
146 level = cpuid_eax(1);
147 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
148 level >= 0x0f58))
149 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
150 if (c->x86 == 0x10 || c->x86 == 0x11)
151 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
152
153 /* Enable workaround for FXSAVE leak */
154 if (c->x86 >= 6)
155 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
156
157 level = get_model_name(c);
158 if (!level) {
159 switch (c->x86) {
160 case 15:
161 /* Should distinguish Models here, but this is only
162 a fallback anyways. */
163 strcpy(c->x86_model_id, "Hammer");
164 break;
165 }
166 }
167 display_cacheinfo(c);
168
169 /* Multi core CPU? */
170 if (c->extended_cpuid_level >= 0x80000008)
171 amd_detect_cmp(c);
172
173 if (c->extended_cpuid_level >= 0x80000006 &&
174 (cpuid_edx(0x80000006) & 0xf000))
175 num_cache_leaves = 4;
176 else
177 num_cache_leaves = 3;
178
179 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
180 set_cpu_cap(c, X86_FEATURE_K8);
181
182 /* MFENCE stops RDTSC speculation */
183 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
184
185 if (c->x86 == 0x10)
186 fam10h_check_enable_mmcfg();
187
188 if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
189 unsigned long long tseg;
190
191 /*
192 * Split up direct mapping around the TSEG SMM area.
193 * Don't do it for gbpages because there seems very little
194 * benefit in doing so.
195 */
196 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg) &&
197 (tseg >> PMD_SHIFT) <
198 (max_pfn_mapped >> (PMD_SHIFT-PAGE_SHIFT)))
199 set_memory_4k((unsigned long)__va(tseg), 1);
200 }
201}
202
203static struct cpu_dev amd_cpu_dev __cpuinitdata = {
204 .c_vendor = "AMD",
205 .c_ident = { "AuthenticAMD" },
206 .c_early_init = early_init_amd,
207 .c_init = init_amd,
208};
209
210cpu_vendor_dev_register(X86_VENDOR_AMD, &amd_cpu_dev);
211
diff --git a/arch/x86/kernel/bugs_64.c b/arch/x86/kernel/cpu/bugs_64.c
index 9a3ed0649d4e..9a3ed0649d4e 100644
--- a/arch/x86/kernel/bugs_64.c
+++ b/arch/x86/kernel/cpu/bugs_64.c
diff --git a/arch/x86/kernel/cpu/centaur_64.c b/arch/x86/kernel/cpu/centaur_64.c
new file mode 100644
index 000000000000..13526fd5cce1
--- /dev/null
+++ b/arch/x86/kernel/cpu/centaur_64.c
@@ -0,0 +1,43 @@
1#include <linux/init.h>
2#include <linux/smp.h>
3
4#include <asm/cpufeature.h>
5#include <asm/processor.h>
6
7#include "cpu.h"
8
9static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
10{
11 if (c->x86 == 0x6 && c->x86_model >= 0xf)
12 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
13}
14
15static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
16{
17 /* Cache sizes */
18 unsigned n;
19
20 n = c->extended_cpuid_level;
21 if (n >= 0x80000008) {
22 unsigned eax = cpuid_eax(0x80000008);
23 c->x86_virt_bits = (eax >> 8) & 0xff;
24 c->x86_phys_bits = eax & 0xff;
25 }
26
27 if (c->x86 == 0x6 && c->x86_model >= 0xf) {
28 c->x86_cache_alignment = c->x86_clflush_size * 2;
29 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
30 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
31 }
32 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
33}
34
35static struct cpu_dev centaur_cpu_dev __cpuinitdata = {
36 .c_vendor = "Centaur",
37 .c_ident = { "CentaurHauls" },
38 .c_early_init = early_init_centaur,
39 .c_init = init_centaur,
40};
41
42cpu_vendor_dev_register(X86_VENDOR_CENTAUR, &centaur_cpu_dev);
43
diff --git a/arch/x86/kernel/cpu/cpu.h b/arch/x86/kernel/cpu/cpu.h
index 783691b2a738..4d894e8565fe 100644
--- a/arch/x86/kernel/cpu/cpu.h
+++ b/arch/x86/kernel/cpu/cpu.h
@@ -1,3 +1,6 @@
1#ifndef ARCH_X86_CPU_H
2
3#define ARCH_X86_CPU_H
1 4
2struct cpu_model_info { 5struct cpu_model_info {
3 int vendor; 6 int vendor;
@@ -36,3 +39,5 @@ extern struct cpu_vendor_dev __x86cpuvendor_start[], __x86cpuvendor_end[];
36 39
37extern int get_model_name(struct cpuinfo_x86 *c); 40extern int get_model_name(struct cpuinfo_x86 *c);
38extern void display_cacheinfo(struct cpuinfo_x86 *c); 41extern void display_cacheinfo(struct cpuinfo_x86 *c);
42
43#endif
diff --git a/arch/x86/kernel/cpu/intel_64.c b/arch/x86/kernel/cpu/intel_64.c
new file mode 100644
index 000000000000..fcb1cc9d75ca
--- /dev/null
+++ b/arch/x86/kernel/cpu/intel_64.c
@@ -0,0 +1,103 @@
1#include <linux/init.h>
2#include <linux/smp.h>
3#include <asm/processor.h>
4#include <asm/ptrace.h>
5#include <asm/topology.h>
6#include <asm/numa_64.h>
7
8#include "cpu.h"
9
10static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
11{
12 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
13 (c->x86 == 0x6 && c->x86_model >= 0x0e))
14 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
15}
16
17/*
18 * find out the number of processor cores on the die
19 */
20static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
21{
22 unsigned int eax, t;
23
24 if (c->cpuid_level < 4)
25 return 1;
26
27 cpuid_count(4, 0, &eax, &t, &t, &t);
28
29 if (eax & 0x1f)
30 return ((eax >> 26) + 1);
31 else
32 return 1;
33}
34
35static void __cpuinit srat_detect_node(void)
36{
37#ifdef CONFIG_NUMA
38 unsigned node;
39 int cpu = smp_processor_id();
40 int apicid = hard_smp_processor_id();
41
42 /* Don't do the funky fallback heuristics the AMD version employs
43 for now. */
44 node = apicid_to_node[apicid];
45 if (node == NUMA_NO_NODE || !node_online(node))
46 node = first_node(node_online_map);
47 numa_set_node(cpu, node);
48
49 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
50#endif
51}
52
53static void __cpuinit init_intel(struct cpuinfo_x86 *c)
54{
55 /* Cache sizes */
56 unsigned n;
57
58 init_intel_cacheinfo(c);
59 if (c->cpuid_level > 9) {
60 unsigned eax = cpuid_eax(10);
61 /* Check for version and the number of counters */
62 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
63 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
64 }
65
66 if (cpu_has_ds) {
67 unsigned int l1, l2;
68 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
69 if (!(l1 & (1<<11)))
70 set_cpu_cap(c, X86_FEATURE_BTS);
71 if (!(l1 & (1<<12)))
72 set_cpu_cap(c, X86_FEATURE_PEBS);
73 }
74
75
76 if (cpu_has_bts)
77 ds_init_intel(c);
78
79 n = c->extended_cpuid_level;
80 if (n >= 0x80000008) {
81 unsigned eax = cpuid_eax(0x80000008);
82 c->x86_virt_bits = (eax >> 8) & 0xff;
83 c->x86_phys_bits = eax & 0xff;
84 }
85
86 if (c->x86 == 15)
87 c->x86_cache_alignment = c->x86_clflush_size * 2;
88 if (c->x86 == 6)
89 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
90 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
91 c->x86_max_cores = intel_num_cpu_cores(c);
92
93 srat_detect_node();
94}
95
96static struct cpu_dev intel_cpu_dev __cpuinitdata = {
97 .c_vendor = "Intel",
98 .c_ident = { "GenuineIntel" },
99 .c_early_init = early_init_intel,
100 .c_init = init_intel,
101};
102cpu_vendor_dev_register(X86_VENDOR_INTEL, &intel_cpu_dev);
103
diff --git a/arch/x86/kernel/mmconf-fam10h_64.c b/arch/x86/kernel/mmconf-fam10h_64.c
index edc5fbfe85c0..fdfdc550b366 100644
--- a/arch/x86/kernel/mmconf-fam10h_64.c
+++ b/arch/x86/kernel/mmconf-fam10h_64.c
@@ -12,6 +12,7 @@
12#include <asm/io.h> 12#include <asm/io.h>
13#include <asm/msr.h> 13#include <asm/msr.h>
14#include <asm/acpi.h> 14#include <asm/acpi.h>
15#include <asm/mmconfig.h>
15 16
16#include "../pci/pci.h" 17#include "../pci/pci.h"
17 18
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index ba370dc8685b..4061d63aabe7 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -6,6 +6,7 @@
6#include <linux/sched.h> 6#include <linux/sched.h>
7#include <linux/module.h> 7#include <linux/module.h>
8#include <linux/pm.h> 8#include <linux/pm.h>
9#include <linux/clockchips.h>
9 10
10struct kmem_cache *task_xstate_cachep; 11struct kmem_cache *task_xstate_cachep;
11 12
@@ -45,6 +46,76 @@ void arch_task_cache_init(void)
45 SLAB_PANIC, NULL); 46 SLAB_PANIC, NULL);
46} 47}
47 48
49/*
50 * Idle related variables and functions
51 */
52unsigned long boot_option_idle_override = 0;
53EXPORT_SYMBOL(boot_option_idle_override);
54
55/*
56 * Powermanagement idle function, if any..
57 */
58void (*pm_idle)(void);
59EXPORT_SYMBOL(pm_idle);
60
61#ifdef CONFIG_X86_32
62/*
63 * This halt magic was a workaround for ancient floppy DMA
64 * wreckage. It should be safe to remove.
65 */
66static int hlt_counter;
67void disable_hlt(void)
68{
69 hlt_counter++;
70}
71EXPORT_SYMBOL(disable_hlt);
72
73void enable_hlt(void)
74{
75 hlt_counter--;
76}
77EXPORT_SYMBOL(enable_hlt);
78
79static inline int hlt_use_halt(void)
80{
81 return (!hlt_counter && boot_cpu_data.hlt_works_ok);
82}
83#else
84static inline int hlt_use_halt(void)
85{
86 return 1;
87}
88#endif
89
90/*
91 * We use this if we don't have any better
92 * idle routine..
93 */
94void default_idle(void)
95{
96 if (hlt_use_halt()) {
97 current_thread_info()->status &= ~TS_POLLING;
98 /*
99 * TS_POLLING-cleared state must be visible before we
100 * test NEED_RESCHED:
101 */
102 smp_mb();
103
104 if (!need_resched())
105 safe_halt(); /* enables interrupts racelessly */
106 else
107 local_irq_enable();
108 current_thread_info()->status |= TS_POLLING;
109 } else {
110 local_irq_enable();
111 /* loop is done by the caller */
112 cpu_relax();
113 }
114}
115#ifdef CONFIG_APM_MODULE
116EXPORT_SYMBOL(default_idle);
117#endif
118
48static void do_nothing(void *unused) 119static void do_nothing(void *unused)
49{ 120{
50} 121}
@@ -122,44 +193,129 @@ static void poll_idle(void)
122 * 193 *
123 * idle=mwait overrides this decision and forces the usage of mwait. 194 * idle=mwait overrides this decision and forces the usage of mwait.
124 */ 195 */
196
197#define MWAIT_INFO 0x05
198#define MWAIT_ECX_EXTENDED_INFO 0x01
199#define MWAIT_EDX_C1 0xf0
200
125static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c) 201static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
126{ 202{
203 u32 eax, ebx, ecx, edx;
204
127 if (force_mwait) 205 if (force_mwait)
128 return 1; 206 return 1;
129 207
130 if (c->x86_vendor == X86_VENDOR_AMD) { 208 if (c->cpuid_level < MWAIT_INFO)
131 switch(c->x86) { 209 return 0;
132 case 0x10: 210
133 case 0x11: 211 cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
134 return 0; 212 /* Check, whether EDX has extended info about MWAIT */
135 } 213 if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
136 } 214 return 1;
215
216 /*
217 * edx enumeratios MONITOR/MWAIT extensions. Check, whether
218 * C1 supports MWAIT
219 */
220 return (edx & MWAIT_EDX_C1);
221}
222
223/*
224 * Check for AMD CPUs, which have potentially C1E support
225 */
226static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
227{
228 if (c->x86_vendor != X86_VENDOR_AMD)
229 return 0;
230
231 if (c->x86 < 0x0F)
232 return 0;
233
234 /* Family 0x0f models < rev F do not have C1E */
235 if (c->x86 == 0x0f && c->x86_model < 0x40)
236 return 0;
237
137 return 1; 238 return 1;
138} 239}
139 240
140void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c) 241/*
242 * C1E aware idle routine. We check for C1E active in the interrupt
243 * pending message MSR. If we detect C1E, then we handle it the same
244 * way as C3 power states (local apic timer and TSC stop)
245 */
246static void c1e_idle(void)
141{ 247{
142 static int selected; 248 static cpumask_t c1e_mask = CPU_MASK_NONE;
249 static int c1e_detected;
143 250
144 if (selected) 251 if (need_resched())
145 return; 252 return;
253
254 if (!c1e_detected) {
255 u32 lo, hi;
256
257 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
258 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
259 c1e_detected = 1;
260 mark_tsc_unstable("TSC halt in C1E");
261 printk(KERN_INFO "System has C1E enabled\n");
262 }
263 }
264
265 if (c1e_detected) {
266 int cpu = smp_processor_id();
267
268 if (!cpu_isset(cpu, c1e_mask)) {
269 cpu_set(cpu, c1e_mask);
270 /*
271 * Force broadcast so ACPI can not interfere. Needs
272 * to run with interrupts enabled as it uses
273 * smp_function_call.
274 */
275 local_irq_enable();
276 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
277 &cpu);
278 printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
279 cpu);
280 local_irq_disable();
281 }
282 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
283
284 default_idle();
285
286 /*
287 * The switch back from broadcast mode needs to be
288 * called with interrupts disabled.
289 */
290 local_irq_disable();
291 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
292 local_irq_enable();
293 } else
294 default_idle();
295}
296
297void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
298{
146#ifdef CONFIG_X86_SMP 299#ifdef CONFIG_X86_SMP
147 if (pm_idle == poll_idle && smp_num_siblings > 1) { 300 if (pm_idle == poll_idle && smp_num_siblings > 1) {
148 printk(KERN_WARNING "WARNING: polling idle and HT enabled," 301 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
149 " performance may degrade.\n"); 302 " performance may degrade.\n");
150 } 303 }
151#endif 304#endif
305 if (pm_idle)
306 return;
307
152 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) { 308 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
153 /* 309 /*
154 * Skip, if setup has overridden idle.
155 * One CPU supports mwait => All CPUs supports mwait 310 * One CPU supports mwait => All CPUs supports mwait
156 */ 311 */
157 if (!pm_idle) { 312 printk(KERN_INFO "using mwait in idle threads.\n");
158 printk(KERN_INFO "using mwait in idle threads.\n"); 313 pm_idle = mwait_idle;
159 pm_idle = mwait_idle; 314 } else if (check_c1e_idle(c)) {
160 } 315 printk(KERN_INFO "using C1E aware idle routine\n");
161 } 316 pm_idle = c1e_idle;
162 selected = 1; 317 } else
318 pm_idle = default_idle;
163} 319}
164 320
165static int __init idle_setup(char *str) 321static int __init idle_setup(char *str)
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index e2db9ac5c61c..c2a11d77b1b5 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -58,11 +58,6 @@
58 58
59asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); 59asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
60 60
61static int hlt_counter;
62
63unsigned long boot_option_idle_override = 0;
64EXPORT_SYMBOL(boot_option_idle_override);
65
66DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task; 61DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
67EXPORT_PER_CPU_SYMBOL(current_task); 62EXPORT_PER_CPU_SYMBOL(current_task);
68 63
@@ -77,55 +72,6 @@ unsigned long thread_saved_pc(struct task_struct *tsk)
77 return ((unsigned long *)tsk->thread.sp)[3]; 72 return ((unsigned long *)tsk->thread.sp)[3];
78} 73}
79 74
80/*
81 * Powermanagement idle function, if any..
82 */
83void (*pm_idle)(void);
84EXPORT_SYMBOL(pm_idle);
85
86void disable_hlt(void)
87{
88 hlt_counter++;
89}
90
91EXPORT_SYMBOL(disable_hlt);
92
93void enable_hlt(void)
94{
95 hlt_counter--;
96}
97
98EXPORT_SYMBOL(enable_hlt);
99
100/*
101 * We use this if we don't have any better
102 * idle routine..
103 */
104void default_idle(void)
105{
106 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
107 current_thread_info()->status &= ~TS_POLLING;
108 /*
109 * TS_POLLING-cleared state must be visible before we
110 * test NEED_RESCHED:
111 */
112 smp_mb();
113
114 if (!need_resched())
115 safe_halt(); /* enables interrupts racelessly */
116 else
117 local_irq_enable();
118 current_thread_info()->status |= TS_POLLING;
119 } else {
120 local_irq_enable();
121 /* loop is done by the caller */
122 cpu_relax();
123 }
124}
125#ifdef CONFIG_APM_MODULE
126EXPORT_SYMBOL(default_idle);
127#endif
128
129#ifdef CONFIG_HOTPLUG_CPU 75#ifdef CONFIG_HOTPLUG_CPU
130#include <asm/nmi.h> 76#include <asm/nmi.h>
131/* We don't actually take CPU down, just spin without interrupts. */ 77/* We don't actually take CPU down, just spin without interrupts. */
@@ -168,24 +114,19 @@ void cpu_idle(void)
168 while (1) { 114 while (1) {
169 tick_nohz_stop_sched_tick(); 115 tick_nohz_stop_sched_tick();
170 while (!need_resched()) { 116 while (!need_resched()) {
171 void (*idle)(void);
172 117
173 check_pgt_cache(); 118 check_pgt_cache();
174 rmb(); 119 rmb();
175 idle = pm_idle;
176 120
177 if (rcu_pending(cpu)) 121 if (rcu_pending(cpu))
178 rcu_check_callbacks(cpu, 0); 122 rcu_check_callbacks(cpu, 0);
179 123
180 if (!idle)
181 idle = default_idle;
182
183 if (cpu_is_offline(cpu)) 124 if (cpu_is_offline(cpu))
184 play_dead(); 125 play_dead();
185 126
186 local_irq_disable(); 127 local_irq_disable();
187 __get_cpu_var(irq_stat).idle_timestamp = jiffies; 128 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
188 idle(); 129 pm_idle();
189 } 130 }
190 tick_nohz_restart_sched_tick(); 131 tick_nohz_restart_sched_tick();
191 preempt_enable_no_resched(); 132 preempt_enable_no_resched();
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index c6eb5c91e5f6..290183e9731a 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -56,15 +56,6 @@ asmlinkage extern void ret_from_fork(void);
56 56
57unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED; 57unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
58 58
59unsigned long boot_option_idle_override = 0;
60EXPORT_SYMBOL(boot_option_idle_override);
61
62/*
63 * Powermanagement idle function, if any..
64 */
65void (*pm_idle)(void);
66EXPORT_SYMBOL(pm_idle);
67
68static ATOMIC_NOTIFIER_HEAD(idle_notifier); 59static ATOMIC_NOTIFIER_HEAD(idle_notifier);
69 60
70void idle_notifier_register(struct notifier_block *n) 61void idle_notifier_register(struct notifier_block *n)
@@ -94,25 +85,6 @@ void exit_idle(void)
94 __exit_idle(); 85 __exit_idle();
95} 86}
96 87
97/*
98 * We use this if we don't have any better
99 * idle routine..
100 */
101void default_idle(void)
102{
103 current_thread_info()->status &= ~TS_POLLING;
104 /*
105 * TS_POLLING-cleared state must be visible before we
106 * test NEED_RESCHED:
107 */
108 smp_mb();
109 if (!need_resched())
110 safe_halt(); /* enables interrupts racelessly */
111 else
112 local_irq_enable();
113 current_thread_info()->status |= TS_POLLING;
114}
115
116#ifdef CONFIG_HOTPLUG_CPU 88#ifdef CONFIG_HOTPLUG_CPU
117DECLARE_PER_CPU(int, cpu_state); 89DECLARE_PER_CPU(int, cpu_state);
118 90
@@ -150,12 +122,9 @@ void cpu_idle(void)
150 while (1) { 122 while (1) {
151 tick_nohz_stop_sched_tick(); 123 tick_nohz_stop_sched_tick();
152 while (!need_resched()) { 124 while (!need_resched()) {
153 void (*idle)(void);
154 125
155 rmb(); 126 rmb();
156 idle = pm_idle; 127
157 if (!idle)
158 idle = default_idle;
159 if (cpu_is_offline(smp_processor_id())) 128 if (cpu_is_offline(smp_processor_id()))
160 play_dead(); 129 play_dead();
161 /* 130 /*
@@ -165,7 +134,7 @@ void cpu_idle(void)
165 */ 134 */
166 local_irq_disable(); 135 local_irq_disable();
167 enter_idle(); 136 enter_idle();
168 idle(); 137 pm_idle();
169 /* In many cases the interrupt that ended idle 138 /* In many cases the interrupt that ended idle
170 has already called exit_idle. But some idle 139 has already called exit_idle. But some idle
171 loops can be woken up without interrupt. */ 140 loops can be woken up without interrupt. */
diff --git a/arch/x86/kernel/setup_64.c b/arch/x86/kernel/setup_64.c
index 6dff1286ad8a..b789ec599923 100644
--- a/arch/x86/kernel/setup_64.c
+++ b/arch/x86/kernel/setup_64.c
@@ -71,6 +71,7 @@
71#include <asm/topology.h> 71#include <asm/topology.h>
72#include <asm/trampoline.h> 72#include <asm/trampoline.h>
73#include <asm/pat.h> 73#include <asm/pat.h>
74#include <asm/mmconfig.h>
74 75
75#include <mach_apic.h> 76#include <mach_apic.h>
76#ifdef CONFIG_PARAVIRT 77#ifdef CONFIG_PARAVIRT
@@ -79,6 +80,8 @@
79#define ARCH_SETUP 80#define ARCH_SETUP
80#endif 81#endif
81 82
83#include "cpu/cpu.h"
84
82/* 85/*
83 * Machine setup.. 86 * Machine setup..
84 */ 87 */
@@ -95,8 +98,6 @@ int bootloader_type;
95 98
96unsigned long saved_video_mode; 99unsigned long saved_video_mode;
97 100
98int force_mwait __cpuinitdata;
99
100/* 101/*
101 * Early DMI memory 102 * Early DMI memory
102 */ 103 */
@@ -164,6 +165,7 @@ static struct resource bss_resource = {
164 .flags = IORESOURCE_RAM, 165 .flags = IORESOURCE_RAM,
165}; 166};
166 167
168static void __init early_cpu_init(void);
167static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c); 169static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
168 170
169#ifdef CONFIG_PROC_VMCORE 171#ifdef CONFIG_PROC_VMCORE
@@ -293,18 +295,6 @@ static void __init parse_setup_data(void)
293 } 295 }
294} 296}
295 297
296#ifdef CONFIG_PCI_MMCONFIG
297extern void __cpuinit fam10h_check_enable_mmcfg(void);
298extern void __init check_enable_amd_mmconf_dmi(void);
299#else
300void __cpuinit fam10h_check_enable_mmcfg(void)
301{
302}
303void __init check_enable_amd_mmconf_dmi(void)
304{
305}
306#endif
307
308/* 298/*
309 * setup_arch - architecture-specific boot-time initializations 299 * setup_arch - architecture-specific boot-time initializations
310 * 300 *
@@ -352,6 +342,7 @@ void __init setup_arch(char **cmdline_p)
352 bss_resource.start = virt_to_phys(&__bss_start); 342 bss_resource.start = virt_to_phys(&__bss_start);
353 bss_resource.end = virt_to_phys(&__bss_stop)-1; 343 bss_resource.end = virt_to_phys(&__bss_stop)-1;
354 344
345 early_cpu_init();
355 early_identify_cpu(&boot_cpu_data); 346 early_identify_cpu(&boot_cpu_data);
356 347
357 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 348 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
@@ -537,7 +528,20 @@ void __init setup_arch(char **cmdline_p)
537 check_enable_amd_mmconf_dmi(); 528 check_enable_amd_mmconf_dmi();
538} 529}
539 530
540static int __cpuinit get_model_name(struct cpuinfo_x86 *c) 531struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
532
533static void __cpuinit default_init(struct cpuinfo_x86 *c)
534{
535 display_cacheinfo(c);
536}
537
538static struct cpu_dev __cpuinitdata default_cpu = {
539 .c_init = default_init,
540 .c_vendor = "Unknown",
541};
542static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
543
544int __cpuinit get_model_name(struct cpuinfo_x86 *c)
541{ 545{
542 unsigned int *v; 546 unsigned int *v;
543 547
@@ -553,7 +557,7 @@ static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
553} 557}
554 558
555 559
556static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c) 560void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
557{ 561{
558 unsigned int n, dummy, eax, ebx, ecx, edx; 562 unsigned int n, dummy, eax, ebx, ecx, edx;
559 563
@@ -585,228 +589,6 @@ static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
585 } 589 }
586} 590}
587 591
588#ifdef CONFIG_NUMA
589static int __cpuinit nearby_node(int apicid)
590{
591 int i, node;
592
593 for (i = apicid - 1; i >= 0; i--) {
594 node = apicid_to_node[i];
595 if (node != NUMA_NO_NODE && node_online(node))
596 return node;
597 }
598 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
599 node = apicid_to_node[i];
600 if (node != NUMA_NO_NODE && node_online(node))
601 return node;
602 }
603 return first_node(node_online_map); /* Shouldn't happen */
604}
605#endif
606
607/*
608 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
609 * Assumes number of cores is a power of two.
610 */
611static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
612{
613#ifdef CONFIG_SMP
614 unsigned bits;
615#ifdef CONFIG_NUMA
616 int cpu = smp_processor_id();
617 int node = 0;
618 unsigned apicid = hard_smp_processor_id();
619#endif
620 bits = c->x86_coreid_bits;
621
622 /* Low order bits define the core id (index of core in socket) */
623 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
624 /* Convert the initial APIC ID into the socket ID */
625 c->phys_proc_id = c->initial_apicid >> bits;
626
627#ifdef CONFIG_NUMA
628 node = c->phys_proc_id;
629 if (apicid_to_node[apicid] != NUMA_NO_NODE)
630 node = apicid_to_node[apicid];
631 if (!node_online(node)) {
632 /* Two possibilities here:
633 - The CPU is missing memory and no node was created.
634 In that case try picking one from a nearby CPU
635 - The APIC IDs differ from the HyperTransport node IDs
636 which the K8 northbridge parsing fills in.
637 Assume they are all increased by a constant offset,
638 but in the same order as the HT nodeids.
639 If that doesn't result in a usable node fall back to the
640 path for the previous case. */
641
642 int ht_nodeid = c->initial_apicid;
643
644 if (ht_nodeid >= 0 &&
645 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
646 node = apicid_to_node[ht_nodeid];
647 /* Pick a nearby node */
648 if (!node_online(node))
649 node = nearby_node(apicid);
650 }
651 numa_set_node(cpu, node);
652
653 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
654#endif
655#endif
656}
657
658static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
659{
660#ifdef CONFIG_SMP
661 unsigned bits, ecx;
662
663 /* Multi core CPU? */
664 if (c->extended_cpuid_level < 0x80000008)
665 return;
666
667 ecx = cpuid_ecx(0x80000008);
668
669 c->x86_max_cores = (ecx & 0xff) + 1;
670
671 /* CPU telling us the core id bits shift? */
672 bits = (ecx >> 12) & 0xF;
673
674 /* Otherwise recompute */
675 if (bits == 0) {
676 while ((1 << bits) < c->x86_max_cores)
677 bits++;
678 }
679
680 c->x86_coreid_bits = bits;
681
682#endif
683}
684
685#define ENABLE_C1E_MASK 0x18000000
686#define CPUID_PROCESSOR_SIGNATURE 1
687#define CPUID_XFAM 0x0ff00000
688#define CPUID_XFAM_K8 0x00000000
689#define CPUID_XFAM_10H 0x00100000
690#define CPUID_XFAM_11H 0x00200000
691#define CPUID_XMOD 0x000f0000
692#define CPUID_XMOD_REV_F 0x00040000
693
694/* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
695static __cpuinit int amd_apic_timer_broken(void)
696{
697 u32 lo, hi, eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
698
699 switch (eax & CPUID_XFAM) {
700 case CPUID_XFAM_K8:
701 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
702 break;
703 case CPUID_XFAM_10H:
704 case CPUID_XFAM_11H:
705 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
706 if (lo & ENABLE_C1E_MASK)
707 return 1;
708 break;
709 default:
710 /* err on the side of caution */
711 return 1;
712 }
713 return 0;
714}
715
716static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
717{
718 early_init_amd_mc(c);
719
720 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
721 if (c->x86_power & (1<<8))
722 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
723}
724
725static void __cpuinit init_amd(struct cpuinfo_x86 *c)
726{
727 unsigned level;
728
729#ifdef CONFIG_SMP
730 unsigned long value;
731
732 /*
733 * Disable TLB flush filter by setting HWCR.FFDIS on K8
734 * bit 6 of msr C001_0015
735 *
736 * Errata 63 for SH-B3 steppings
737 * Errata 122 for all steppings (F+ have it disabled by default)
738 */
739 if (c->x86 == 15) {
740 rdmsrl(MSR_K8_HWCR, value);
741 value |= 1 << 6;
742 wrmsrl(MSR_K8_HWCR, value);
743 }
744#endif
745
746 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
747 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
748 clear_cpu_cap(c, 0*32+31);
749
750 /* On C+ stepping K8 rep microcode works well for copy/memset */
751 level = cpuid_eax(1);
752 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
753 level >= 0x0f58))
754 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
755 if (c->x86 == 0x10 || c->x86 == 0x11)
756 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
757
758 /* Enable workaround for FXSAVE leak */
759 if (c->x86 >= 6)
760 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
761
762 level = get_model_name(c);
763 if (!level) {
764 switch (c->x86) {
765 case 15:
766 /* Should distinguish Models here, but this is only
767 a fallback anyways. */
768 strcpy(c->x86_model_id, "Hammer");
769 break;
770 }
771 }
772 display_cacheinfo(c);
773
774 /* Multi core CPU? */
775 if (c->extended_cpuid_level >= 0x80000008)
776 amd_detect_cmp(c);
777
778 if (c->extended_cpuid_level >= 0x80000006 &&
779 (cpuid_edx(0x80000006) & 0xf000))
780 num_cache_leaves = 4;
781 else
782 num_cache_leaves = 3;
783
784 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
785 set_cpu_cap(c, X86_FEATURE_K8);
786
787 /* MFENCE stops RDTSC speculation */
788 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
789
790 if (c->x86 == 0x10)
791 fam10h_check_enable_mmcfg();
792
793 if (amd_apic_timer_broken())
794 disable_apic_timer = 1;
795
796 if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
797 unsigned long long tseg;
798
799 /*
800 * Split up direct mapping around the TSEG SMM area.
801 * Don't do it for gbpages because there seems very little
802 * benefit in doing so.
803 */
804 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg) &&
805 (tseg >> PMD_SHIFT) < (max_pfn_mapped >> (PMD_SHIFT-PAGE_SHIFT)))
806 set_memory_4k((unsigned long)__va(tseg), 1);
807 }
808}
809
810void __cpuinit detect_ht(struct cpuinfo_x86 *c) 592void __cpuinit detect_ht(struct cpuinfo_x86 *c)
811{ 593{
812#ifdef CONFIG_SMP 594#ifdef CONFIG_SMP
@@ -857,135 +639,59 @@ out:
857#endif 639#endif
858} 640}
859 641
860/* 642static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
861 * find out the number of processor cores on the die
862 */
863static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
864{
865 unsigned int eax, t;
866
867 if (c->cpuid_level < 4)
868 return 1;
869
870 cpuid_count(4, 0, &eax, &t, &t, &t);
871
872 if (eax & 0x1f)
873 return ((eax >> 26) + 1);
874 else
875 return 1;
876}
877
878static void __cpuinit srat_detect_node(void)
879{
880#ifdef CONFIG_NUMA
881 unsigned node;
882 int cpu = smp_processor_id();
883 int apicid = hard_smp_processor_id();
884
885 /* Don't do the funky fallback heuristics the AMD version employs
886 for now. */
887 node = apicid_to_node[apicid];
888 if (node == NUMA_NO_NODE || !node_online(node))
889 node = first_node(node_online_map);
890 numa_set_node(cpu, node);
891
892 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
893#endif
894}
895
896static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
897{
898 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
899 (c->x86 == 0x6 && c->x86_model >= 0x0e))
900 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
901}
902
903static void __cpuinit init_intel(struct cpuinfo_x86 *c)
904{ 643{
905 /* Cache sizes */ 644 char *v = c->x86_vendor_id;
906 unsigned n; 645 int i;
907 646 static int printed;
908 init_intel_cacheinfo(c); 647
909 if (c->cpuid_level > 9) { 648 for (i = 0; i < X86_VENDOR_NUM; i++) {
910 unsigned eax = cpuid_eax(10); 649 if (cpu_devs[i]) {
911 /* Check for version and the number of counters */ 650 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
912 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1)) 651 (cpu_devs[i]->c_ident[1] &&
913 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON); 652 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
653 c->x86_vendor = i;
654 this_cpu = cpu_devs[i];
655 return;
656 }
657 }
914 } 658 }
915 659 if (!printed) {
916 if (cpu_has_ds) { 660 printed++;
917 unsigned int l1, l2; 661 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
918 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2); 662 printk(KERN_ERR "CPU: Your system may be unstable.\n");
919 if (!(l1 & (1<<11)))
920 set_cpu_cap(c, X86_FEATURE_BTS);
921 if (!(l1 & (1<<12)))
922 set_cpu_cap(c, X86_FEATURE_PEBS);
923 } 663 }
924 664 c->x86_vendor = X86_VENDOR_UNKNOWN;
925
926 if (cpu_has_bts)
927 ds_init_intel(c);
928
929 n = c->extended_cpuid_level;
930 if (n >= 0x80000008) {
931 unsigned eax = cpuid_eax(0x80000008);
932 c->x86_virt_bits = (eax >> 8) & 0xff;
933 c->x86_phys_bits = eax & 0xff;
934 /* CPUID workaround for Intel 0F34 CPU */
935 if (c->x86_vendor == X86_VENDOR_INTEL &&
936 c->x86 == 0xF && c->x86_model == 0x3 &&
937 c->x86_mask == 0x4)
938 c->x86_phys_bits = 36;
939 }
940
941 if (c->x86 == 15)
942 c->x86_cache_alignment = c->x86_clflush_size * 2;
943 if (c->x86 == 6)
944 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
945 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
946 c->x86_max_cores = intel_num_cpu_cores(c);
947
948 srat_detect_node();
949}
950
951static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
952{
953 if (c->x86 == 0x6 && c->x86_model >= 0xf)
954 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
955} 665}
956 666
957static void __cpuinit init_centaur(struct cpuinfo_x86 *c) 667static void __init early_cpu_support_print(void)
958{ 668{
959 /* Cache sizes */ 669 int i,j;
960 unsigned n; 670 struct cpu_dev *cpu_devx;
961 671
962 n = c->extended_cpuid_level; 672 printk("KERNEL supported cpus:\n");
963 if (n >= 0x80000008) { 673 for (i = 0; i < X86_VENDOR_NUM; i++) {
964 unsigned eax = cpuid_eax(0x80000008); 674 cpu_devx = cpu_devs[i];
965 c->x86_virt_bits = (eax >> 8) & 0xff; 675 if (!cpu_devx)
966 c->x86_phys_bits = eax & 0xff; 676 continue;
967 } 677 for (j = 0; j < 2; j++) {
968 678 if (!cpu_devx->c_ident[j])
969 if (c->x86 == 0x6 && c->x86_model >= 0xf) { 679 continue;
970 c->x86_cache_alignment = c->x86_clflush_size * 2; 680 printk(" %s %s\n", cpu_devx->c_vendor,
971 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 681 cpu_devx->c_ident[j]);
972 set_cpu_cap(c, X86_FEATURE_REP_GOOD); 682 }
973 } 683 }
974 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
975} 684}
976 685
977static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c) 686static void __init early_cpu_init(void)
978{ 687{
979 char *v = c->x86_vendor_id; 688 struct cpu_vendor_dev *cvdev;
980 689
981 if (!strcmp(v, "AuthenticAMD")) 690 for (cvdev = __x86cpuvendor_start ;
982 c->x86_vendor = X86_VENDOR_AMD; 691 cvdev < __x86cpuvendor_end ;
983 else if (!strcmp(v, "GenuineIntel")) 692 cvdev++)
984 c->x86_vendor = X86_VENDOR_INTEL; 693 cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
985 else if (!strcmp(v, "CentaurHauls")) 694 early_cpu_support_print();
986 c->x86_vendor = X86_VENDOR_CENTAUR;
987 else
988 c->x86_vendor = X86_VENDOR_UNKNOWN;
989} 695}
990 696
991/* Do some early cpuid on the boot CPU to get some parameter that are 697/* Do some early cpuid on the boot CPU to get some parameter that are
@@ -1066,17 +772,9 @@ static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1066 if (c->extended_cpuid_level >= 0x80000007) 772 if (c->extended_cpuid_level >= 0x80000007)
1067 c->x86_power = cpuid_edx(0x80000007); 773 c->x86_power = cpuid_edx(0x80000007);
1068 774
1069 switch (c->x86_vendor) { 775 if (c->x86_vendor != X86_VENDOR_UNKNOWN &&
1070 case X86_VENDOR_AMD: 776 cpu_devs[c->x86_vendor]->c_early_init)
1071 early_init_amd(c); 777 cpu_devs[c->x86_vendor]->c_early_init(c);
1072 break;
1073 case X86_VENDOR_INTEL:
1074 early_init_intel(c);
1075 break;
1076 case X86_VENDOR_CENTAUR:
1077 early_init_centaur(c);
1078 break;
1079 }
1080 778
1081 validate_pat_support(c); 779 validate_pat_support(c);
1082} 780}
@@ -1104,24 +802,8 @@ void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1104 * At the end of this section, c->x86_capability better 802 * At the end of this section, c->x86_capability better
1105 * indicate the features this CPU genuinely supports! 803 * indicate the features this CPU genuinely supports!
1106 */ 804 */
1107 switch (c->x86_vendor) { 805 if (this_cpu->c_init)
1108 case X86_VENDOR_AMD: 806 this_cpu->c_init(c);
1109 init_amd(c);
1110 break;
1111
1112 case X86_VENDOR_INTEL:
1113 init_intel(c);
1114 break;
1115
1116 case X86_VENDOR_CENTAUR:
1117 init_centaur(c);
1118 break;
1119
1120 case X86_VENDOR_UNKNOWN:
1121 default:
1122 display_cacheinfo(c);
1123 break;
1124 }
1125 807
1126 detect_ht(c); 808 detect_ht(c);
1127 809
diff --git a/arch/x86/kernel/vmlinux_64.lds.S b/arch/x86/kernel/vmlinux_64.lds.S
index fad3674b06a5..b29f63bdff5e 100644
--- a/arch/x86/kernel/vmlinux_64.lds.S
+++ b/arch/x86/kernel/vmlinux_64.lds.S
@@ -177,6 +177,7 @@ SECTIONS
177 *(.con_initcall.init) 177 *(.con_initcall.init)
178 } 178 }
179 __con_initcall_end = .; 179 __con_initcall_end = .;
180 . = ALIGN(16);
180 __x86cpuvendor_start = .; 181 __x86cpuvendor_start = .;
181 .x86cpuvendor.init : AT(ADDR(.x86cpuvendor.init) - LOAD_OFFSET) { 182 .x86cpuvendor.init : AT(ADDR(.x86cpuvendor.init) - LOAD_OFFSET) {
182 *(.x86cpuvendor.init) 183 *(.x86cpuvendor.init)
diff --git a/arch/x86/pci/Makefile_32 b/arch/x86/pci/Makefile_32
index 89ec35d00efd..f647e7e56da4 100644
--- a/arch/x86/pci/Makefile_32
+++ b/arch/x86/pci/Makefile_32
@@ -22,3 +22,4 @@ pci-$(CONFIG_X86_NUMAQ) := numa.o irq.o
22pci-$(CONFIG_NUMA) += mp_bus_to_node.o 22pci-$(CONFIG_NUMA) += mp_bus_to_node.o
23 23
24obj-y += $(pci-y) common.o early.o 24obj-y += $(pci-y) common.o early.o
25obj-y += amd_bus.o
diff --git a/arch/x86/pci/Makefile_64 b/arch/x86/pci/Makefile_64
index 8fbd19832cf6..fd47068c95de 100644
--- a/arch/x86/pci/Makefile_64
+++ b/arch/x86/pci/Makefile_64
@@ -13,5 +13,5 @@ obj-y += legacy.o irq.o common.o early.o
13# mmconfig has a 64bit special 13# mmconfig has a 64bit special
14obj-$(CONFIG_PCI_MMCONFIG) += mmconfig_64.o direct.o mmconfig-shared.o 14obj-$(CONFIG_PCI_MMCONFIG) += mmconfig_64.o direct.o mmconfig-shared.o
15 15
16obj-y += k8-bus_64.o 16obj-y += amd_bus.o
17 17
diff --git a/arch/x86/pci/k8-bus_64.c b/arch/x86/pci/amd_bus.c
index 5c2799c20e47..15f505d3a78e 100644
--- a/arch/x86/pci/k8-bus_64.c
+++ b/arch/x86/pci/amd_bus.c
@@ -1,5 +1,9 @@
1#include <linux/init.h> 1#include <linux/init.h>
2#include <linux/pci.h> 2#include <linux/pci.h>
3#include "pci.h"
4
5#ifdef CONFIG_X86_64
6
3#include <asm/pci-direct.h> 7#include <asm/pci-direct.h>
4#include <asm/mpspec.h> 8#include <asm/mpspec.h>
5#include <linux/cpumask.h> 9#include <linux/cpumask.h>
@@ -526,3 +530,31 @@ static int __init early_fill_mp_bus_info(void)
526} 530}
527 531
528postcore_initcall(early_fill_mp_bus_info); 532postcore_initcall(early_fill_mp_bus_info);
533
534#endif
535
536/* common 32/64 bit code */
537
538#define ENABLE_CF8_EXT_CFG (1ULL << 46)
539
540static void enable_pci_io_ecs_per_cpu(void *unused)
541{
542 u64 reg;
543 rdmsrl(MSR_AMD64_NB_CFG, reg);
544 if (!(reg & ENABLE_CF8_EXT_CFG)) {
545 reg |= ENABLE_CF8_EXT_CFG;
546 wrmsrl(MSR_AMD64_NB_CFG, reg);
547 }
548}
549
550static int __init enable_pci_io_ecs(void)
551{
552 /* assume all cpus from fam10h have IO ECS */
553 if (boot_cpu_data.x86 < 0x10)
554 return 0;
555 on_each_cpu(enable_pci_io_ecs_per_cpu, NULL, 1, 1);
556 pci_probe |= PCI_HAS_IO_ECS;
557 return 0;
558}
559
560postcore_initcall(enable_pci_io_ecs);
diff --git a/arch/x86/pci/direct.c b/arch/x86/pci/direct.c
index 21d1e0e0d535..9915293500fb 100644
--- a/arch/x86/pci/direct.c
+++ b/arch/x86/pci/direct.c
@@ -8,18 +8,21 @@
8#include "pci.h" 8#include "pci.h"
9 9
10/* 10/*
11 * Functions for accessing PCI configuration space with type 1 accesses 11 * Functions for accessing PCI base (first 256 bytes) and extended
12 * (4096 bytes per PCI function) configuration space with type 1
13 * accesses.
12 */ 14 */
13 15
14#define PCI_CONF1_ADDRESS(bus, devfn, reg) \ 16#define PCI_CONF1_ADDRESS(bus, devfn, reg) \
15 (0x80000000 | (bus << 16) | (devfn << 8) | (reg & ~3)) 17 (0x80000000 | ((reg & 0xF00) << 16) | (bus << 16) \
18 | (devfn << 8) | (reg & 0xFC))
16 19
17static int pci_conf1_read(unsigned int seg, unsigned int bus, 20static int pci_conf1_read(unsigned int seg, unsigned int bus,
18 unsigned int devfn, int reg, int len, u32 *value) 21 unsigned int devfn, int reg, int len, u32 *value)
19{ 22{
20 unsigned long flags; 23 unsigned long flags;
21 24
22 if ((bus > 255) || (devfn > 255) || (reg > 255)) { 25 if ((bus > 255) || (devfn > 255) || (reg > 4095)) {
23 *value = -1; 26 *value = -1;
24 return -EINVAL; 27 return -EINVAL;
25 } 28 }
@@ -50,7 +53,7 @@ static int pci_conf1_write(unsigned int seg, unsigned int bus,
50{ 53{
51 unsigned long flags; 54 unsigned long flags;
52 55
53 if ((bus > 255) || (devfn > 255) || (reg > 255)) 56 if ((bus > 255) || (devfn > 255) || (reg > 4095))
54 return -EINVAL; 57 return -EINVAL;
55 58
56 spin_lock_irqsave(&pci_config_lock, flags); 59 spin_lock_irqsave(&pci_config_lock, flags);
@@ -260,10 +263,18 @@ void __init pci_direct_init(int type)
260 return; 263 return;
261 printk(KERN_INFO "PCI: Using configuration type %d for base access\n", 264 printk(KERN_INFO "PCI: Using configuration type %d for base access\n",
262 type); 265 type);
263 if (type == 1) 266 if (type == 1) {
264 raw_pci_ops = &pci_direct_conf1; 267 raw_pci_ops = &pci_direct_conf1;
265 else 268 if (raw_pci_ext_ops)
266 raw_pci_ops = &pci_direct_conf2; 269 return;
270 if (!(pci_probe & PCI_HAS_IO_ECS))
271 return;
272 printk(KERN_INFO "PCI: Using configuration type 1 "
273 "for extended access\n");
274 raw_pci_ext_ops = &pci_direct_conf1;
275 return;
276 }
277 raw_pci_ops = &pci_direct_conf2;
267} 278}
268 279
269int __init pci_direct_probe(void) 280int __init pci_direct_probe(void)
diff --git a/arch/x86/pci/pci.h b/arch/x86/pci/pci.h
index 720c4c554534..ba263e626a68 100644
--- a/arch/x86/pci/pci.h
+++ b/arch/x86/pci/pci.h
@@ -27,6 +27,7 @@
27#define PCI_CAN_SKIP_ISA_ALIGN 0x8000 27#define PCI_CAN_SKIP_ISA_ALIGN 0x8000
28#define PCI_USE__CRS 0x10000 28#define PCI_USE__CRS 0x10000
29#define PCI_CHECK_ENABLE_AMD_MMCONF 0x20000 29#define PCI_CHECK_ENABLE_AMD_MMCONF 0x20000
30#define PCI_HAS_IO_ECS 0x40000
30 31
31extern unsigned int pci_probe; 32extern unsigned int pci_probe;
32extern unsigned long pirq_table_addr; 33extern unsigned long pirq_table_addr;