Merge branch 'linus' into x86/xen
[cascardo/linux.git] / arch / x86 / kernel / cpu / common.c
index 9983bc3..fb789dd 100644 (file)
@@ -1,28 +1,62 @@
 #include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
 #include <linux/string.h>
+#include <linux/bootmem.h>
+#include <linux/bitops.h>
+#include <linux/module.h>
+#include <linux/kgdb.h>
+#include <linux/topology.h>
 #include <linux/delay.h>
 #include <linux/smp.h>
-#include <linux/module.h>
 #include <linux/percpu.h>
-#include <linux/bootmem.h>
-#include <asm/processor.h>
 #include <asm/i387.h>
 #include <asm/msr.h>
 #include <asm/io.h>
+#include <asm/linkage.h>
 #include <asm/mmu_context.h>
 #include <asm/mtrr.h>
 #include <asm/mce.h>
 #include <asm/pat.h>
 #include <asm/asm.h>
+#include <asm/numa.h>
 #ifdef CONFIG_X86_LOCAL_APIC
 #include <asm/mpspec.h>
 #include <asm/apic.h>
 #include <mach_apic.h>
+#include <asm/genapic.h>
 #endif
 
+#include <asm/pda.h>
+#include <asm/pgtable.h>
+#include <asm/processor.h>
+#include <asm/desc.h>
+#include <asm/atomic.h>
+#include <asm/proto.h>
+#include <asm/sections.h>
+#include <asm/setup.h>
+
 #include "cpu.h"
 
+static struct cpu_dev *this_cpu __cpuinitdata;
+
+#ifdef CONFIG_X86_64
+/* We need valid kernel segments for data and code in long mode too
+ * IRET will check the segment types  kkeil 2000/10/28
+ * Also sysret mandates a special GDT layout
+ */
+/* The TLS descriptors are currently at a different place compared to i386.
+   Hopefully nobody expects them at a fixed place (Wine?) */
 DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
+       [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
+       [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
+       [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
+       [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
+       [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
+       [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
+} };
+#else
+DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
        [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
        [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
        [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
@@ -56,17 +90,150 @@ DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
        [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
        [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } },
 } };
+#endif
 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
 
-__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
-
+#ifdef CONFIG_X86_32
 static int cachesize_override __cpuinitdata = -1;
 static int disable_x86_serial_nr __cpuinitdata = 1;
 
-struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
+static int __init cachesize_setup(char *str)
+{
+       get_option(&str, &cachesize_override);
+       return 1;
+}
+__setup("cachesize=", cachesize_setup);
+
+static int __init x86_fxsr_setup(char *s)
+{
+       setup_clear_cpu_cap(X86_FEATURE_FXSR);
+       setup_clear_cpu_cap(X86_FEATURE_XMM);
+       return 1;
+}
+__setup("nofxsr", x86_fxsr_setup);
+
+static int __init x86_sep_setup(char *s)
+{
+       setup_clear_cpu_cap(X86_FEATURE_SEP);
+       return 1;
+}
+__setup("nosep", x86_sep_setup);
+
+/* Standard macro to see if a specific flag is changeable */
+static inline int flag_is_changeable_p(u32 flag)
+{
+       u32 f1, f2;
+
+       asm("pushfl\n\t"
+           "pushfl\n\t"
+           "popl %0\n\t"
+           "movl %0,%1\n\t"
+           "xorl %2,%0\n\t"
+           "pushl %0\n\t"
+           "popfl\n\t"
+           "pushfl\n\t"
+           "popl %0\n\t"
+           "popfl\n\t"
+           : "=&r" (f1), "=&r" (f2)
+           : "ir" (flag));
+
+       return ((f1^f2) & flag) != 0;
+}
+
+/* Probe for the CPUID instruction */
+static int __cpuinit have_cpuid_p(void)
+{
+       return flag_is_changeable_p(X86_EFLAGS_ID);
+}
+
+static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
+{
+       if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
+               /* Disable processor serial number */
+               unsigned long lo, hi;
+               rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
+               lo |= 0x200000;
+               wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
+               printk(KERN_NOTICE "CPU serial number disabled.\n");
+               clear_cpu_cap(c, X86_FEATURE_PN);
+
+               /* Disabling the serial number may affect the cpuid level */
+               c->cpuid_level = cpuid_eax(0);
+       }
+}
+
+static int __init x86_serial_nr_setup(char *s)
+{
+       disable_x86_serial_nr = 0;
+       return 1;
+}
+__setup("serialnumber", x86_serial_nr_setup);
+#else
+static inline int flag_is_changeable_p(u32 flag)
+{
+       return 1;
+}
+/* Probe for the CPUID instruction */
+static inline int have_cpuid_p(void)
+{
+       return 1;
+}
+static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
+{
+}
+#endif
+
+/*
+ * Naming convention should be: <Name> [(<Codename>)]
+ * This table only is used unless init_<vendor>() below doesn't set it;
+ * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
+ *
+ */
+
+/* Look up CPU names by table lookup. */
+static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
+{
+       struct cpu_model_info *info;
+
+       if (c->x86_model >= 16)
+               return NULL;    /* Range check */
+
+       if (!this_cpu)
+               return NULL;
+
+       info = this_cpu->c_models;
+
+       while (info && info->family) {
+               if (info->family == c->x86)
+                       return info->model_names[c->x86_model];
+               info++;
+       }
+       return NULL;            /* Not found */
+}
+
+__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
+
+/* Current gdt points %fs at the "master" per-cpu area: after this,
+ * it's on the real one. */
+void switch_to_new_gdt(void)
+{
+       struct desc_ptr gdt_descr;
+
+       gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
+       gdt_descr.size = GDT_SIZE - 1;
+       load_gdt(&gdt_descr);
+#ifdef CONFIG_X86_32
+       asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
+#endif
+}
+
+static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
 
 static void __cpuinit default_init(struct cpuinfo_x86 *c)
 {
+#ifdef CONFIG_X86_64
+       display_cacheinfo(c);
+#else
        /* Not much we can do here... */
        /* Check if at least it has cpuid */
        if (c->cpuid_level == -1) {
@@ -76,28 +243,22 @@ static void __cpuinit default_init(struct cpuinfo_x86 *c)
                else if (c->x86 == 3)
                        strcpy(c->x86_model_id, "386");
        }
+#endif
 }
 
 static struct cpu_dev __cpuinitdata default_cpu = {
        .c_init = default_init,
        .c_vendor = "Unknown",
+       .c_x86_vendor = X86_VENDOR_UNKNOWN,
 };
-static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
 
-static int __init cachesize_setup(char *str)
-{
-       get_option(&str, &cachesize_override);
-       return 1;
-}
-__setup("cachesize=", cachesize_setup);
-
-int __cpuinit get_model_name(struct cpuinfo_x86 *c)
+static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
 {
        unsigned int *v;
        char *p, *q;
 
-       if (cpuid_eax(0x80000000) < 0x80000004)
-               return 0;
+       if (c->extended_cpuid_level < 0x80000004)
+               return;
 
        v = (unsigned int *) c->x86_model_id;
        cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
@@ -116,30 +277,34 @@ int __cpuinit get_model_name(struct cpuinfo_x86 *c)
             while (q <= &c->x86_model_id[48])
                  *q++ = '\0';  /* Zero-pad the rest */
        }
-
-       return 1;
 }
 
-
 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
 {
-       unsigned int n, dummy, ecx, edx, l2size;
+       unsigned int n, dummy, ebx, ecx, edx, l2size;
 
-       n = cpuid_eax(0x80000000);
+       n = c->extended_cpuid_level;
 
        if (n >= 0x80000005) {
-               cpuid(0x80000005, &dummy, &dummy, &ecx, &edx);
+               cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
                printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
-                       edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
-               c->x86_cache_size = (ecx>>24)+(edx>>24);
+                               edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
+               c->x86_cache_size = (ecx>>24) + (edx>>24);
+#ifdef CONFIG_X86_64
+               /* On K8 L1 TLB is inclusive, so don't count it */
+               c->x86_tlbsize = 0;
+#endif
        }
 
        if (n < 0x80000006)     /* Some chips just has a large L1. */
                return;
 
-       ecx = cpuid_ecx(0x80000006);
+       cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
        l2size = ecx >> 16;
 
+#ifdef CONFIG_X86_64
+       c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
+#else
        /* do processor-specific cache resizing */
        if (this_cpu->c_size_cache)
                l2size = this_cpu->c_size_cache(c, l2size);
@@ -150,116 +315,106 @@ void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
 
        if (l2size == 0)
                return;         /* Again, no L2 cache is possible */
+#endif
 
        c->x86_cache_size = l2size;
 
        printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
-              l2size, ecx & 0xFF);
+                       l2size, ecx & 0xFF);
 }
 
-/*
- * Naming convention should be: <Name> [(<Codename>)]
- * This table only is used unless init_<vendor>() below doesn't set it;
- * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
- *
- */
-
-/* Look up CPU names by table lookup. */
-static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
+void __cpuinit detect_ht(struct cpuinfo_x86 *c)
 {
-       struct cpu_model_info *info;
+#ifdef CONFIG_X86_HT
+       u32 eax, ebx, ecx, edx;
+       int index_msb, core_bits;
 
-       if (c->x86_model >= 16)
-               return NULL;    /* Range check */
+       if (!cpu_has(c, X86_FEATURE_HT))
+               return;
 
-       if (!this_cpu)
-               return NULL;
+       if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
+               goto out;
 
-       info = this_cpu->c_models;
+       if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
+               return;
 
-       while (info && info->family) {
-               if (info->family == c->x86)
-                       return info->model_names[c->x86_model];
-               info++;
+       cpuid(1, &eax, &ebx, &ecx, &edx);
+
+       smp_num_siblings = (ebx & 0xff0000) >> 16;
+
+       if (smp_num_siblings == 1) {
+               printk(KERN_INFO  "CPU: Hyper-Threading is disabled\n");
+       } else if (smp_num_siblings > 1) {
+
+               if (smp_num_siblings > NR_CPUS) {
+                       printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
+                                       smp_num_siblings);
+                       smp_num_siblings = 1;
+                       return;
+               }
+
+               index_msb = get_count_order(smp_num_siblings);
+#ifdef CONFIG_X86_64
+               c->phys_proc_id = phys_pkg_id(index_msb);
+#else
+               c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb);
+#endif
+
+               smp_num_siblings = smp_num_siblings / c->x86_max_cores;
+
+               index_msb = get_count_order(smp_num_siblings);
+
+               core_bits = get_count_order(c->x86_max_cores);
+
+#ifdef CONFIG_X86_64
+               c->cpu_core_id = phys_pkg_id(index_msb) &
+                                              ((1 << core_bits) - 1);
+#else
+               c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) &
+                                              ((1 << core_bits) - 1);
+#endif
        }
-       return NULL;            /* Not found */
-}
 
+out:
+       if ((c->x86_max_cores * smp_num_siblings) > 1) {
+               printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
+                      c->phys_proc_id);
+               printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
+                      c->cpu_core_id);
+       }
+#endif
+}
 
-static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c, int early)
+static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
 {
        char *v = c->x86_vendor_id;
        int i;
        static int printed;
 
        for (i = 0; i < X86_VENDOR_NUM; i++) {
-               if (cpu_devs[i]) {
-                       if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
-                           (cpu_devs[i]->c_ident[1] &&
-                            !strcmp(v, cpu_devs[i]->c_ident[1]))) {
-                               c->x86_vendor = i;
-                               if (!early)
-                                       this_cpu = cpu_devs[i];
-                               return;
-                       }
+               if (!cpu_devs[i])
+                       break;
+
+               if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
+                   (cpu_devs[i]->c_ident[1] &&
+                    !strcmp(v, cpu_devs[i]->c_ident[1]))) {
+                       this_cpu = cpu_devs[i];
+                       c->x86_vendor = this_cpu->c_x86_vendor;
+                       return;
                }
        }
+
        if (!printed) {
                printed++;
-               printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
+               printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v);
                printk(KERN_ERR "CPU: Your system may be unstable.\n");
        }
+
        c->x86_vendor = X86_VENDOR_UNKNOWN;
        this_cpu = &default_cpu;
 }
 
-
-static int __init x86_fxsr_setup(char *s)
-{
-       setup_clear_cpu_cap(X86_FEATURE_FXSR);
-       setup_clear_cpu_cap(X86_FEATURE_XMM);
-       return 1;
-}
-__setup("nofxsr", x86_fxsr_setup);
-
-
-static int __init x86_sep_setup(char *s)
-{
-       setup_clear_cpu_cap(X86_FEATURE_SEP);
-       return 1;
-}
-__setup("nosep", x86_sep_setup);
-
-
-/* Standard macro to see if a specific flag is changeable */
-static inline int flag_is_changeable_p(u32 flag)
-{
-       u32 f1, f2;
-
-       asm("pushfl\n\t"
-           "pushfl\n\t"
-           "popl %0\n\t"
-           "movl %0,%1\n\t"
-           "xorl %2,%0\n\t"
-           "pushl %0\n\t"
-           "popfl\n\t"
-           "pushfl\n\t"
-           "popl %0\n\t"
-           "popfl\n\t"
-           : "=&r" (f1), "=&r" (f2)
-           : "ir" (flag));
-
-       return ((f1^f2) & flag) != 0;
-}
-
-
-/* Probe for the CPUID instruction */
-static int __cpuinit have_cpuid_p(void)
-{
-       return flag_is_changeable_p(X86_EFLAGS_ID);
-}
-
-void __init cpu_detect(struct cpuinfo_x86 *c)
+void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
 {
        /* Get vendor name */
        cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
@@ -268,50 +423,87 @@ void __init cpu_detect(struct cpuinfo_x86 *c)
              (unsigned int *)&c->x86_vendor_id[4]);
 
        c->x86 = 4;
+       /* Intel-defined flags: level 0x00000001 */
        if (c->cpuid_level >= 0x00000001) {
                u32 junk, tfms, cap0, misc;
                cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
-               c->x86 = (tfms >> 8) & 15;
-               c->x86_model = (tfms >> 4) & 15;
+               c->x86 = (tfms >> 8) & 0xf;
+               c->x86_model = (tfms >> 4) & 0xf;
+               c->x86_mask = tfms & 0xf;
                if (c->x86 == 0xf)
                        c->x86 += (tfms >> 20) & 0xff;
                if (c->x86 >= 0x6)
-                       c->x86_model += ((tfms >> 16) & 0xF) << 4;
-               c->x86_mask = tfms & 15;
+                       c->x86_model += ((tfms >> 16) & 0xf) << 4;
                if (cap0 & (1<<19)) {
-                       c->x86_cache_alignment = ((misc >> 8) & 0xff) * 8;
                        c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
+                       c->x86_cache_alignment = c->x86_clflush_size;
                }
        }
 }
-static void __cpuinit early_get_cap(struct cpuinfo_x86 *c)
+
+static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
 {
        u32 tfms, xlvl;
-       unsigned int ebx;
+       u32 ebx;
 
-       memset(&c->x86_capability, 0, sizeof c->x86_capability);
-       if (have_cpuid_p()) {
-               /* Intel-defined flags: level 0x00000001 */
-               if (c->cpuid_level >= 0x00000001) {
-                       u32 capability, excap;
-                       cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
-                       c->x86_capability[0] = capability;
-                       c->x86_capability[4] = excap;
-               }
+       /* Intel-defined flags: level 0x00000001 */
+       if (c->cpuid_level >= 0x00000001) {
+               u32 capability, excap;
+               cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
+               c->x86_capability[0] = capability;
+               c->x86_capability[4] = excap;
+       }
 
-               /* AMD-defined flags: level 0x80000001 */
-               xlvl = cpuid_eax(0x80000000);
-               if ((xlvl & 0xffff0000) == 0x80000000) {
-                       if (xlvl >= 0x80000001) {
-                               c->x86_capability[1] = cpuid_edx(0x80000001);
-                               c->x86_capability[6] = cpuid_ecx(0x80000001);
-                       }
+       /* AMD-defined flags: level 0x80000001 */
+       xlvl = cpuid_eax(0x80000000);
+       c->extended_cpuid_level = xlvl;
+       if ((xlvl & 0xffff0000) == 0x80000000) {
+               if (xlvl >= 0x80000001) {
+                       c->x86_capability[1] = cpuid_edx(0x80000001);
+                       c->x86_capability[6] = cpuid_ecx(0x80000001);
                }
+       }
 
+#ifdef CONFIG_X86_64
+       if (c->extended_cpuid_level >= 0x80000008) {
+               u32 eax = cpuid_eax(0x80000008);
+
+               c->x86_virt_bits = (eax >> 8) & 0xff;
+               c->x86_phys_bits = eax & 0xff;
        }
+#endif
+
+       if (c->extended_cpuid_level >= 0x80000007)
+               c->x86_power = cpuid_edx(0x80000007);
 
 }
 
+static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
+{
+#ifdef CONFIG_X86_32
+       int i;
+
+       /*
+        * First of all, decide if this is a 486 or higher
+        * It's a 486 if we can modify the AC flag
+        */
+       if (flag_is_changeable_p(X86_EFLAGS_AC))
+               c->x86 = 4;
+       else
+               c->x86 = 3;
+
+       for (i = 0; i < X86_VENDOR_NUM; i++)
+               if (cpu_devs[i] && cpu_devs[i]->c_identify) {
+                       c->x86_vendor_id[0] = 0;
+                       cpu_devs[i]->c_identify(c);
+                       if (c->x86_vendor_id[0]) {
+                               get_cpu_vendor(c);
+                               break;
+                       }
+               }
+#endif
+}
+
 /*
  * Do minimum CPU detection early.
  * Fields really needed: vendor, cpuid_level, family, model, mask,
@@ -321,25 +513,61 @@ static void __cpuinit early_get_cap(struct cpuinfo_x86 *c)
  * WARNING: this function is only called on the BP.  Don't add code here
  * that is supposed to run on all CPUs.
  */
-static void __init early_cpu_detect(void)
+static void __init early_identify_cpu(struct cpuinfo_x86 *c)
 {
-       struct cpuinfo_x86 *c = &boot_cpu_data;
-
-       c->x86_cache_alignment = 32;
+#ifdef CONFIG_X86_64
+       c->x86_clflush_size = 64;
+#else
        c->x86_clflush_size = 32;
+#endif
+       c->x86_cache_alignment = c->x86_clflush_size;
+
+       memset(&c->x86_capability, 0, sizeof c->x86_capability);
+       c->extended_cpuid_level = 0;
 
+       if (!have_cpuid_p())
+               identify_cpu_without_cpuid(c);
+
+       /* cyrix could have cpuid enabled via c_identify()*/
        if (!have_cpuid_p())
                return;
 
        cpu_detect(c);
 
-       get_cpu_vendor(c, 1);
+       get_cpu_vendor(c);
 
-       early_get_cap(c);
+       get_cpu_cap(c);
 
-       if (c->x86_vendor != X86_VENDOR_UNKNOWN &&
-           cpu_devs[c->x86_vendor]->c_early_init)
-               cpu_devs[c->x86_vendor]->c_early_init(c);
+       if (this_cpu->c_early_init)
+               this_cpu->c_early_init(c);
+
+       validate_pat_support(c);
+}
+
+void __init early_cpu_init(void)
+{
+       struct cpu_dev **cdev;
+       int count = 0;
+
+       printk("KERNEL supported cpus:\n");
+       for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
+               struct cpu_dev *cpudev = *cdev;
+               unsigned int j;
+
+               if (count >= X86_VENDOR_NUM)
+                       break;
+               cpu_devs[count] = cpudev;
+               count++;
+
+               for (j = 0; j < 2; j++) {
+                       if (!cpudev->c_ident[j])
+                               continue;
+                       printk("  %s %s\n", cpudev->c_vendor,
+                               cpudev->c_ident[j]);
+               }
+       }
+
+       early_identify_cpu(&boot_cpu_data);
 }
 
 /*
@@ -357,86 +585,41 @@ static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
 
 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
 {
-       u32 tfms, xlvl;
-       unsigned int ebx;
-
-       if (have_cpuid_p()) {
-               /* Get vendor name */
-               cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
-                     (unsigned int *)&c->x86_vendor_id[0],
-                     (unsigned int *)&c->x86_vendor_id[8],
-                     (unsigned int *)&c->x86_vendor_id[4]);
-
-               get_cpu_vendor(c, 0);
-               /* Initialize the standard set of capabilities */
-               /* Note that the vendor-specific code below might override */
-               /* Intel-defined flags: level 0x00000001 */
-               if (c->cpuid_level >= 0x00000001) {
-                       u32 capability, excap;
-                       cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
-                       c->x86_capability[0] = capability;
-                       c->x86_capability[4] = excap;
-                       c->x86 = (tfms >> 8) & 15;
-                       c->x86_model = (tfms >> 4) & 15;
-                       if (c->x86 == 0xf)
-                               c->x86 += (tfms >> 20) & 0xff;
-                       if (c->x86 >= 0x6)
-                               c->x86_model += ((tfms >> 16) & 0xF) << 4;
-                       c->x86_mask = tfms & 15;
-                       c->initial_apicid = (ebx >> 24) & 0xFF;
-#ifdef CONFIG_X86_HT
-                       c->apicid = phys_pkg_id(c->initial_apicid, 0);
-                       c->phys_proc_id = c->initial_apicid;
-#else
-                       c->apicid = c->initial_apicid;
-#endif
-                       if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
-                               c->x86_clflush_size = ((ebx >> 8) & 0xff) * 8;
-               } else {
-                       /* Have CPUID level 0 only - unheard of */
-                       c->x86 = 4;
-               }
+       c->extended_cpuid_level = 0;
 
-               /* AMD-defined flags: level 0x80000001 */
-               xlvl = cpuid_eax(0x80000000);
-               if ((xlvl & 0xffff0000) == 0x80000000) {
-                       if (xlvl >= 0x80000001) {
-                               c->x86_capability[1] = cpuid_edx(0x80000001);
-                               c->x86_capability[6] = cpuid_ecx(0x80000001);
-                       }
-                       if (xlvl >= 0x80000004)
-                               get_model_name(c); /* Default name */
-               }
+       if (!have_cpuid_p())
+               identify_cpu_without_cpuid(c);
 
-               init_scattered_cpuid_features(c);
-               detect_nopl(c);
-       }
-}
+       /* cyrix could have cpuid enabled via c_identify()*/
+       if (!have_cpuid_p())
+               return;
 
-static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
-{
-       if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
-               /* Disable processor serial number */
-               unsigned long lo, hi;
-               rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
-               lo |= 0x200000;
-               wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
-               printk(KERN_NOTICE "CPU serial number disabled.\n");
-               clear_cpu_cap(c, X86_FEATURE_PN);
+       cpu_detect(c);
 
-               /* Disabling the serial number may affect the cpuid level */
-               c->cpuid_level = cpuid_eax(0);
-       }
-}
+       get_cpu_vendor(c);
 
-static int __init x86_serial_nr_setup(char *s)
-{
-       disable_x86_serial_nr = 0;
-       return 1;
-}
-__setup("serialnumber", x86_serial_nr_setup);
+       get_cpu_cap(c);
 
+       if (c->cpuid_level >= 0x00000001) {
+               c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
+#ifdef CONFIG_X86_32
+# ifdef CONFIG_X86_HT
+               c->apicid = phys_pkg_id(c->initial_apicid, 0);
+# else
+               c->apicid = c->initial_apicid;
+# endif
+#endif
 
+#ifdef CONFIG_X86_HT
+               c->phys_proc_id = c->initial_apicid;
+#endif
+       }
+
+       get_model_name(c); /* Default name */
+
+       init_scattered_cpuid_features(c);
+       detect_nopl(c);
+}
 
 /*
  * This does the hard work of actually picking apart the CPU stuff...
@@ -448,30 +631,29 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
        c->loops_per_jiffy = loops_per_jiffy;
        c->x86_cache_size = -1;
        c->x86_vendor = X86_VENDOR_UNKNOWN;
-       c->cpuid_level = -1;    /* CPUID not detected */
        c->x86_model = c->x86_mask = 0; /* So far unknown... */
        c->x86_vendor_id[0] = '\0'; /* Unset */
        c->x86_model_id[0] = '\0';  /* Unset */
        c->x86_max_cores = 1;
+       c->x86_coreid_bits = 0;
+#ifdef CONFIG_X86_64
+       c->x86_clflush_size = 64;
+#else
+       c->cpuid_level = -1;    /* CPUID not detected */
        c->x86_clflush_size = 32;
+#endif
+       c->x86_cache_alignment = c->x86_clflush_size;
        memset(&c->x86_capability, 0, sizeof c->x86_capability);
 
-       if (!have_cpuid_p()) {
-               /*
-                * First of all, decide if this is a 486 or higher
-                * It's a 486 if we can modify the AC flag
-                */
-               if (flag_is_changeable_p(X86_EFLAGS_AC))
-                       c->x86 = 4;
-               else
-                       c->x86 = 3;
-       }
-
        generic_identify(c);
 
        if (this_cpu->c_identify)
                this_cpu->c_identify(c);
 
+#ifdef CONFIG_X86_64
+       c->apicid = phys_pkg_id(0);
+#endif
+
        /*
         * Vendor-specific initialization.  In this section we
         * canonicalize the feature flags, meaning if there are
@@ -505,6 +687,10 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
                                c->x86, c->x86_model);
        }
 
+#ifdef CONFIG_X86_64
+       detect_ht(c);
+#endif
+
        /*
         * On SMP, boot_cpu_data holds the common feature set between
         * all CPUs; so make sure that we indicate which features are
@@ -513,7 +699,7 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
         */
        if (c != &boot_cpu_data) {
                /* AND the already accumulated flags with these */
-               for (i = 0 ; i < NCAPINTS ; i++)
+               for (i = 0; i < NCAPINTS; i++)
                        boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
        }
 
@@ -521,72 +707,79 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
        for (i = 0; i < NCAPINTS; i++)
                c->x86_capability[i] &= ~cleared_cpu_caps[i];
 
+#ifdef CONFIG_X86_MCE
        /* Init Machine Check Exception if available. */
        mcheck_init(c);
+#endif
 
        select_idle_routine(c);
+
+#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
+       numa_add_cpu(smp_processor_id());
+#endif
 }
 
 void __init identify_boot_cpu(void)
 {
        identify_cpu(&boot_cpu_data);
+#ifdef CONFIG_X86_32
        sysenter_setup();
        enable_sep_cpu();
+#endif
 }
 
 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
 {
        BUG_ON(c == &boot_cpu_data);
        identify_cpu(c);
+#ifdef CONFIG_X86_32
        enable_sep_cpu();
+#endif
        mtrr_ap_init();
 }
 
-#ifdef CONFIG_X86_HT
-void __cpuinit detect_ht(struct cpuinfo_x86 *c)
-{
-       u32     eax, ebx, ecx, edx;
-       int     index_msb, core_bits;
-
-       cpuid(1, &eax, &ebx, &ecx, &edx);
-
-       if (!cpu_has(c, X86_FEATURE_HT) || cpu_has(c, X86_FEATURE_CMP_LEGACY))
-               return;
-
-       smp_num_siblings = (ebx & 0xff0000) >> 16;
+struct msr_range {
+       unsigned min;
+       unsigned max;
+};
 
-       if (smp_num_siblings == 1) {
-               printk(KERN_INFO  "CPU: Hyper-Threading is disabled\n");
-       } else if (smp_num_siblings > 1) {
+static struct msr_range msr_range_array[] __cpuinitdata = {
+       { 0x00000000, 0x00000418},
+       { 0xc0000000, 0xc000040b},
+       { 0xc0010000, 0xc0010142},
+       { 0xc0011000, 0xc001103b},
+};
 
-               if (smp_num_siblings > NR_CPUS) {
-                       printk(KERN_WARNING "CPU: Unsupported number of the "
-                                       "siblings %d", smp_num_siblings);
-                       smp_num_siblings = 1;
-                       return;
+static void __cpuinit print_cpu_msr(void)
+{
+       unsigned index;
+       u64 val;
+       int i;
+       unsigned index_min, index_max;
+
+       for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
+               index_min = msr_range_array[i].min;
+               index_max = msr_range_array[i].max;
+               for (index = index_min; index < index_max; index++) {
+                       if (rdmsrl_amd_safe(index, &val))
+                               continue;
+                       printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
                }
+       }
+}
 
-               index_msb = get_count_order(smp_num_siblings);
-               c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb);
-
-               printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
-                      c->phys_proc_id);
-
-               smp_num_siblings = smp_num_siblings / c->x86_max_cores;
-
-               index_msb = get_count_order(smp_num_siblings) ;
+static int show_msr __cpuinitdata;
+static __init int setup_show_msr(char *arg)
+{
+       int num;
 
-               core_bits = get_count_order(c->x86_max_cores);
+       get_option(&arg, &num);
 
-               c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) &
-                                              ((1 << core_bits) - 1);
-
-               if (c->x86_max_cores > 1)
-                       printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
-                              c->cpu_core_id);
-       }
+       if (num > 0)
+               show_msr = num;
+       return 1;
 }
-#endif
+__setup("show_msr=", setup_show_msr);
 
 static __init int setup_noclflush(char *arg)
 {
@@ -605,17 +798,25 @@ void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
                vendor = c->x86_vendor_id;
 
        if (vendor && strncmp(c->x86_model_id, vendor, strlen(vendor)))
-               printk("%s ", vendor);
+               printk(KERN_CONT "%s ", vendor);
 
-       if (!c->x86_model_id[0])
-               printk("%d86", c->x86);
+       if (c->x86_model_id[0])
+               printk(KERN_CONT "%s", c->x86_model_id);
        else
-               printk("%s", c->x86_model_id);
+               printk(KERN_CONT "%d86", c->x86);
 
        if (c->x86_mask || c->cpuid_level >= 0)
-               printk(" stepping %02x\n", c->x86_mask);
+               printk(KERN_CONT " stepping %02x\n", c->x86_mask);
        else
-               printk("\n");
+               printk(KERN_CONT "\n");
+
+#ifdef CONFIG_SMP
+       if (c->cpu_index < show_msr)
+               print_cpu_msr();
+#else
+       if (show_msr)
+               print_cpu_msr();
+#endif
 }
 
 static __init int setup_disablecpuid(char *arg)
@@ -631,19 +832,89 @@ __setup("clearcpuid=", setup_disablecpuid);
 
 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
 
-void __init early_cpu_init(void)
+#ifdef CONFIG_X86_64
+struct x8664_pda **_cpu_pda __read_mostly;
+EXPORT_SYMBOL(_cpu_pda);
+
+struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
+
+char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
+
+void __cpuinit pda_init(int cpu)
+{
+       struct x8664_pda *pda = cpu_pda(cpu);
+
+       /* Setup up data that may be needed in __get_free_pages early */
+       loadsegment(fs, 0);
+       loadsegment(gs, 0);
+       /* Memory clobbers used to order PDA accessed */
+       mb();
+       wrmsrl(MSR_GS_BASE, pda);
+       mb();
+
+       pda->cpunumber = cpu;
+       pda->irqcount = -1;
+       pda->kernelstack = (unsigned long)stack_thread_info() -
+                                PDA_STACKOFFSET + THREAD_SIZE;
+       pda->active_mm = &init_mm;
+       pda->mmu_state = 0;
+
+       if (cpu == 0) {
+               /* others are initialized in smpboot.c */
+               pda->pcurrent = &init_task;
+               pda->irqstackptr = boot_cpu_stack;
+               pda->irqstackptr += IRQSTACKSIZE - 64;
+       } else {
+               if (!pda->irqstackptr) {
+                       pda->irqstackptr = (char *)
+                               __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
+                       if (!pda->irqstackptr)
+                               panic("cannot allocate irqstack for cpu %d",
+                                     cpu);
+                       pda->irqstackptr += IRQSTACKSIZE - 64;
+               }
+
+               if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
+                       pda->nodenumber = cpu_to_node(cpu);
+       }
+}
+
+char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
+                          DEBUG_STKSZ] __page_aligned_bss;
+
+extern asmlinkage void ignore_sysret(void);
+
+/* May not be marked __init: used by software suspend */
+void syscall_init(void)
 {
-       struct cpu_vendor_dev *cvdev;
+       /*
+        * LSTAR and STAR live in a bit strange symbiosis.
+        * They both write to the same internal register. STAR allows to
+        * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
+        */
+       wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
+       wrmsrl(MSR_LSTAR, system_call);
+       wrmsrl(MSR_CSTAR, ignore_sysret);
 
-       for (cvdev = __x86cpuvendor_start ;
-            cvdev < __x86cpuvendor_end   ;
-            cvdev++)
-               cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
+#ifdef CONFIG_IA32_EMULATION
+       syscall32_cpu_init();
+#endif
 
-       early_cpu_detect();
-       validate_pat_support(&boot_cpu_data);
+       /* Flags to clear on syscall */
+       wrmsrl(MSR_SYSCALL_MASK,
+              X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
 }
 
+unsigned long kernel_eflags;
+
+/*
+ * Copies of the original ist values from the tss are only accessed during
+ * debugging, no special alignment required.
+ */
+DEFINE_PER_CPU(struct orig_ist, orig_ist);
+
+#else
+
 /* Make sure %fs is initialized properly in idle threads */
 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
 {
@@ -651,25 +922,136 @@ struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
        regs->fs = __KERNEL_PERCPU;
        return regs;
 }
-
-/* Current gdt points %fs at the "master" per-cpu area: after this,
- * it's on the real one. */
-void switch_to_new_gdt(void)
-{
-       struct desc_ptr gdt_descr;
-
-       gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
-       gdt_descr.size = GDT_SIZE - 1;
-       load_gdt(&gdt_descr);
-       asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
-}
+#endif
 
 /*
  * cpu_init() initializes state that is per-CPU. Some data is already
  * initialized (naturally) in the bootstrap process, such as the GDT
  * and IDT. We reload them nevertheless, this function acts as a
  * 'CPU state barrier', nothing should get across.
+ * A lot of state is already set up in PDA init for 64 bit
  */
+#ifdef CONFIG_X86_64
+void __cpuinit cpu_init(void)
+{
+       int cpu = stack_smp_processor_id();
+       struct tss_struct *t = &per_cpu(init_tss, cpu);
+       struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
+       unsigned long v;
+       char *estacks = NULL;
+       struct task_struct *me;
+       int i;
+
+       /* CPU 0 is initialised in head64.c */
+       if (cpu != 0)
+               pda_init(cpu);
+       else
+               estacks = boot_exception_stacks;
+
+       me = current;
+
+       if (cpu_test_and_set(cpu, cpu_initialized))
+               panic("CPU#%d already initialized!\n", cpu);
+
+       printk(KERN_INFO "Initializing CPU#%d\n", cpu);
+
+       clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
+
+       /*
+        * Initialize the per-CPU GDT with the boot GDT,
+        * and set up the GDT descriptor:
+        */
+
+       switch_to_new_gdt();
+       load_idt((const struct desc_ptr *)&idt_descr);
+
+       memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
+       syscall_init();
+
+       wrmsrl(MSR_FS_BASE, 0);
+       wrmsrl(MSR_KERNEL_GS_BASE, 0);
+       barrier();
+
+       check_efer();
+       if (cpu != 0 && x2apic)
+               enable_x2apic();
+
+       /*
+        * set up and load the per-CPU TSS
+        */
+       if (!orig_ist->ist[0]) {
+               static const unsigned int order[N_EXCEPTION_STACKS] = {
+                 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
+                 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
+               };
+               for (v = 0; v < N_EXCEPTION_STACKS; v++) {
+                       if (cpu) {
+                               estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
+                               if (!estacks)
+                                       panic("Cannot allocate exception "
+                                             "stack %ld %d\n", v, cpu);
+                       }
+                       estacks += PAGE_SIZE << order[v];
+                       orig_ist->ist[v] = t->x86_tss.ist[v] =
+                                       (unsigned long)estacks;
+               }
+       }
+
+       t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
+       /*
+        * <= is required because the CPU will access up to
+        * 8 bits beyond the end of the IO permission bitmap.
+        */
+       for (i = 0; i <= IO_BITMAP_LONGS; i++)
+               t->io_bitmap[i] = ~0UL;
+
+       atomic_inc(&init_mm.mm_count);
+       me->active_mm = &init_mm;
+       if (me->mm)
+               BUG();
+       enter_lazy_tlb(&init_mm, me);
+
+       load_sp0(t, &current->thread);
+       set_tss_desc(cpu, t);
+       load_TR_desc();
+       load_LDT(&init_mm.context);
+
+#ifdef CONFIG_KGDB
+       /*
+        * If the kgdb is connected no debug regs should be altered.  This
+        * is only applicable when KGDB and a KGDB I/O module are built
+        * into the kernel and you are using early debugging with
+        * kgdbwait. KGDB will control the kernel HW breakpoint registers.
+        */
+       if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
+               arch_kgdb_ops.correct_hw_break();
+       else {
+#endif
+       /*
+        * Clear all 6 debug registers:
+        */
+
+       set_debugreg(0UL, 0);
+       set_debugreg(0UL, 1);
+       set_debugreg(0UL, 2);
+       set_debugreg(0UL, 3);
+       set_debugreg(0UL, 6);
+       set_debugreg(0UL, 7);
+#ifdef CONFIG_KGDB
+       /* If the kgdb is connected no debug regs should be altered. */
+       }
+#endif
+
+       fpu_init();
+
+       raw_local_save_flags(kernel_eflags);
+
+       if (is_uv_system())
+               uv_cpu_init();
+}
+
+#else
+
 void __cpuinit cpu_init(void)
 {
        int cpu = smp_processor_id();
@@ -723,8 +1105,21 @@ void __cpuinit cpu_init(void)
        /*
         * Force FPU initialization:
         */
-       current_thread_info()->status = 0;
+       if (cpu_has_xsave)
+               current_thread_info()->status = TS_XSAVE;
+       else
+               current_thread_info()->status = 0;
        clear_used_math();
        mxcsr_feature_mask_init();
+
+       /*
+        * Boot processor to setup the FP and extended state context info.
+        */
+       if (!smp_processor_id())
+               init_thread_xstate();
+
+       xsave_init();
 }
 
+
+#endif