Merge branch 'timers-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[safe/jmp/linux-2.6] / arch / x86 / kernel / cpu / intel.c
index dc4e081..24ff26a 100644 (file)
@@ -8,8 +8,16 @@
 #include <linux/module.h>
 
 #include <asm/processor.h>
+#include <asm/pgtable.h>
 #include <asm/msr.h>
 #include <asm/uaccess.h>
+#include <asm/ds.h>
+#include <asm/bugs.h>
+
+#ifdef CONFIG_X86_64
+#include <asm/topology.h>
+#include <asm/numa_64.h>
+#endif
 
 #include "cpu.h"
 
 #include <mach_apic.h>
 #endif
 
-extern int trap_init_f00f_bug(void);
+static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
+{
+       /* Unmask CPUID levels if masked: */
+       if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
+               u64 misc_enable;
 
-#ifdef CONFIG_X86_INTEL_USERCOPY
-/*
- * Alignment at which movsl is preferred for bulk memory copies.
- */
-struct movsl_mask movsl_mask __read_mostly;
-#endif
+               rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
 
-void __cpuinit early_intel_workaround(struct cpuinfo_x86 *c)
-{
-       if (c->x86_vendor != X86_VENDOR_INTEL)
-               return;
+               if (misc_enable & MSR_IA32_MISC_ENABLE_LIMIT_CPUID) {
+                       misc_enable &= ~MSR_IA32_MISC_ENABLE_LIMIT_CPUID;
+                       wrmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
+                       c->cpuid_level = cpuid_eax(0);
+               }
+       }
+
+       if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
+               (c->x86 == 0x6 && c->x86_model >= 0x0e))
+               set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
+
+#ifdef CONFIG_X86_64
+       set_cpu_cap(c, X86_FEATURE_SYSENTER32);
+#else
        /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
        if (c->x86 == 15 && c->x86_cache_alignment == 64)
                c->x86_cache_alignment = 128;
+#endif
+
+       /*
+        * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
+        * with P/T states and does not stop in deep C-states
+        */
+       if (c->x86_power & (1 << 8)) {
+               set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
+               set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
+       }
+
 }
 
+#ifdef CONFIG_X86_32
 /*
  *     Early probe support logic for ppro memory erratum #50
  *
  *     This is called before we do cpu ident work
  */
+
 int __cpuinit ppro_with_ram_bug(void)
 {
        /* Uses data from early_cpu_detect now */
@@ -55,18 +84,57 @@ int __cpuinit ppro_with_ram_bug(void)
        }
        return 0;
 }
-       
 
-/*
- * P4 Xeon errata 037 workaround.
- * Hardware prefetcher may cause stale data to be loaded into the cache.
- */
-static void __cpuinit Intel_errata_workarounds(struct cpuinfo_x86 *c)
+#ifdef CONFIG_X86_F00F_BUG
+static void __cpuinit trap_init_f00f_bug(void)
+{
+       __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
+
+       /*
+        * Update the IDT descriptor and reload the IDT so that
+        * it uses the read-only mapped virtual address.
+        */
+       idt_descr.address = fix_to_virt(FIX_F00F_IDT);
+       load_idt(&idt_descr);
+}
+#endif
+
+static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
 {
        unsigned long lo, hi;
 
+#ifdef CONFIG_X86_F00F_BUG
+       /*
+        * All current models of Pentium and Pentium with MMX technology CPUs
+        * have the F0 0F bug, which lets nonprivileged users lock up the system.
+        * Note that the workaround only should be initialized once...
+        */
+       c->f00f_bug = 0;
+       if (!paravirt_enabled() && c->x86 == 5) {
+               static int f00f_workaround_enabled;
+
+               c->f00f_bug = 1;
+               if (!f00f_workaround_enabled) {
+                       trap_init_f00f_bug();
+                       printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
+                       f00f_workaround_enabled = 1;
+               }
+       }
+#endif
+
+       /*
+        * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
+        * model 3 mask 3
+        */
+       if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
+               clear_cpu_cap(c, X86_FEATURE_SEP);
+
+       /*
+        * P4 Xeon errata 037 workaround.
+        * Hardware prefetcher may cause stale data to be loaded into the cache.
+        */
        if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
-               rdmsr (MSR_IA32_MISC_ENABLE, lo, hi);
+               rdmsr(MSR_IA32_MISC_ENABLE, lo, hi);
                if ((lo & (1<<9)) == 0) {
                        printk (KERN_INFO "CPU: C0 stepping P4 Xeon detected.\n");
                        printk (KERN_INFO "CPU: Disabling hardware prefetching (Errata 037)\n");
@@ -74,13 +142,68 @@ static void __cpuinit Intel_errata_workarounds(struct cpuinfo_x86 *c)
                        wrmsr (MSR_IA32_MISC_ENABLE, lo, hi);
                }
        }
+
+       /*
+        * See if we have a good local APIC by checking for buggy Pentia,
+        * i.e. all B steppings and the C2 stepping of P54C when using their
+        * integrated APIC (see 11AP erratum in "Pentium Processor
+        * Specification Update").
+        */
+       if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
+           (c->x86_mask < 0x6 || c->x86_mask == 0xb))
+               set_cpu_cap(c, X86_FEATURE_11AP);
+
+
+#ifdef CONFIG_X86_INTEL_USERCOPY
+       /*
+        * Set up the preferred alignment for movsl bulk memory moves
+        */
+       switch (c->x86) {
+       case 4:         /* 486: untested */
+               break;
+       case 5:         /* Old Pentia: untested */
+               break;
+       case 6:         /* PII/PIII only like movsl with 8-byte alignment */
+               movsl_mask.mask = 7;
+               break;
+       case 15:        /* P4 is OK down to 8-byte alignment */
+               movsl_mask.mask = 7;
+               break;
+       }
+#endif
+
+#ifdef CONFIG_X86_NUMAQ
+       numaq_tsc_disable();
+#endif
 }
+#else
+static void __cpuinit intel_workarounds(struct cpuinfo_x86 *c)
+{
+}
+#endif
 
+static void __cpuinit srat_detect_node(void)
+{
+#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
+       unsigned node;
+       int cpu = smp_processor_id();
+       int apicid = hard_smp_processor_id();
+
+       /* Don't do the funky fallback heuristics the AMD version employs
+          for now. */
+       node = apicid_to_node[apicid];
+       if (node == NUMA_NO_NODE || !node_online(node))
+               node = first_node(node_online_map);
+       numa_set_node(cpu, node);
+
+       printk(KERN_INFO "CPU %d/0x%x -> Node %d\n", cpu, apicid, node);
+#endif
+}
 
 /*
  * find out the number of processor cores on the die
  */
-static int __cpuinit num_cpu_cores(struct cpuinfo_x86 *c)
+static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
 {
        unsigned int eax, ebx, ecx, edx;
 
@@ -95,47 +218,96 @@ static int __cpuinit num_cpu_cores(struct cpuinfo_x86 *c)
                return 1;
 }
 
+static void __cpuinit detect_vmx_virtcap(struct cpuinfo_x86 *c)
+{
+       /* Intel VMX MSR indicated features */
+#define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW   0x00200000
+#define X86_VMX_FEATURE_PROC_CTLS_VNMI         0x00400000
+#define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS     0x80000000
+#define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC   0x00000001
+#define X86_VMX_FEATURE_PROC_CTLS2_EPT         0x00000002
+#define X86_VMX_FEATURE_PROC_CTLS2_VPID                0x00000020
+
+       u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
+
+       clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
+       clear_cpu_cap(c, X86_FEATURE_VNMI);
+       clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
+       clear_cpu_cap(c, X86_FEATURE_EPT);
+       clear_cpu_cap(c, X86_FEATURE_VPID);
+
+       rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
+       msr_ctl = vmx_msr_high | vmx_msr_low;
+       if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
+               set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
+       if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
+               set_cpu_cap(c, X86_FEATURE_VNMI);
+       if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
+               rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
+                     vmx_msr_low, vmx_msr_high);
+               msr_ctl2 = vmx_msr_high | vmx_msr_low;
+               if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
+                   (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
+                       set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
+               if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
+                       set_cpu_cap(c, X86_FEATURE_EPT);
+               if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
+                       set_cpu_cap(c, X86_FEATURE_VPID);
+       }
+}
+
 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
 {
        unsigned int l2 = 0;
-       char *p = NULL;
 
-#ifdef CONFIG_X86_F00F_BUG
+       early_init_intel(c);
+
+       intel_workarounds(c);
+
        /*
-        * All current models of Pentium and Pentium with MMX technology CPUs
-        * have the F0 0F bug, which lets nonprivileged users lock up the system.
-        * Note that the workaround only should be initialized once...
+        * Detect the extended topology information if available. This
+        * will reinitialise the initial_apicid which will be used
+        * in init_intel_cacheinfo()
         */
-       c->f00f_bug = 0;
-       if (!paravirt_enabled() && c->x86 == 5) {
-               static int f00f_workaround_enabled = 0;
-
-               c->f00f_bug = 1;
-               if ( !f00f_workaround_enabled ) {
-                       trap_init_f00f_bug();
-                       printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n");
-                       f00f_workaround_enabled = 1;
-               }
-       }
-#endif
+       detect_extended_topology(c);
 
-       select_idle_routine(c);
        l2 = init_intel_cacheinfo(c);
-       if (c->cpuid_level > 9 ) {
+       if (c->cpuid_level > 9) {
                unsigned eax = cpuid_eax(10);
                /* Check for version and the number of counters */
                if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
-                       set_bit(X86_FEATURE_ARCH_PERFMON, c->x86_capability);
+                       set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
        }
 
-       /* SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until model 3 mask 3 */
-       if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
-               clear_bit(X86_FEATURE_SEP, c->x86_capability);
+       if (cpu_has_xmm2)
+               set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
+       if (cpu_has_ds) {
+               unsigned int l1;
+               rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
+               if (!(l1 & (1<<11)))
+                       set_cpu_cap(c, X86_FEATURE_BTS);
+               if (!(l1 & (1<<12)))
+                       set_cpu_cap(c, X86_FEATURE_PEBS);
+               ds_init_intel(c);
+       }
 
-       /* Names for the Pentium II/Celeron processors 
-          detectable only by also checking the cache size.
-          Dixon is NOT a Celeron. */
+       if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush)
+               set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR);
+
+#ifdef CONFIG_X86_64
+       if (c->x86 == 15)
+               c->x86_cache_alignment = c->x86_clflush_size * 2;
+       if (c->x86 == 6)
+               set_cpu_cap(c, X86_FEATURE_REP_GOOD);
+#else
+       /*
+        * Names for the Pentium II/Celeron processors
+        * detectable only by also checking the cache size.
+        * Dixon is NOT a Celeron.
+        */
        if (c->x86 == 6) {
+               char *p = NULL;
+
                switch (c->x86_model) {
                case 5:
                        if (c->x86_mask == 0) {
@@ -145,72 +317,53 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
                                        p = "Mobile Pentium II (Dixon)";
                        }
                        break;
-                       
+
                case 6:
                        if (l2 == 128)
                                p = "Celeron (Mendocino)";
                        else if (c->x86_mask == 0 || c->x86_mask == 5)
                                p = "Celeron-A";
                        break;
-                       
+
                case 8:
                        if (l2 == 128)
                                p = "Celeron (Coppermine)";
                        break;
                }
-       }
-
-       if ( p )
-               strcpy(c->x86_model_id, p);
-       
-       c->x86_max_cores = num_cpu_cores(c);
 
-       detect_ht(c);
-
-       /* Work around errata */
-       Intel_errata_workarounds(c);
-
-#ifdef CONFIG_X86_INTEL_USERCOPY
-       /*
-        * Set up the preferred alignment for movsl bulk memory moves
-        */
-       switch (c->x86) {
-       case 4:         /* 486: untested */
-               break;
-       case 5:         /* Old Pentia: untested */
-               break;
-       case 6:         /* PII/PIII only like movsl with 8-byte alignment */
-               movsl_mask.mask = 7;
-               break;
-       case 15:        /* P4 is OK down to 8-byte alignment */
-               movsl_mask.mask = 7;
-               break;
+               if (p)
+                       strcpy(c->x86_model_id, p);
        }
+
+       if (c->x86 == 15)
+               set_cpu_cap(c, X86_FEATURE_P4);
+       if (c->x86 == 6)
+               set_cpu_cap(c, X86_FEATURE_P3);
 #endif
 
-       if (c->x86 == 15) {
-               set_bit(X86_FEATURE_P4, c->x86_capability);
-               set_bit(X86_FEATURE_SYNC_RDTSC, c->x86_capability);
+       if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
+               /*
+                * let's use the legacy cpuid vector 0x1 and 0x4 for topology
+                * detection.
+                */
+               c->x86_max_cores = intel_num_cpu_cores(c);
+#ifdef CONFIG_X86_32
+               detect_ht(c);
+#endif
        }
-       if (c->x86 == 6) 
-               set_bit(X86_FEATURE_P3, c->x86_capability);
-       if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
-               (c->x86 == 0x6 && c->x86_model >= 0x0e))
-               set_bit(X86_FEATURE_CONSTANT_TSC, c->x86_capability);
 
-       if (cpu_has_ds) {
-               unsigned int l1;
-               rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
-               if (!(l1 & (1<<11)))
-                       set_bit(X86_FEATURE_BTS, c->x86_capability);
-               if (!(l1 & (1<<12)))
-                       set_bit(X86_FEATURE_PEBS, c->x86_capability);
-       }
+       /* Work around errata */
+       srat_detect_node();
+
+       if (cpu_has(c, X86_FEATURE_VMX))
+               detect_vmx_virtcap(c);
 }
 
-static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 * c, unsigned int size)
+#ifdef CONFIG_X86_32
+static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
 {
-       /* Intel PIII Tualatin. This comes in two flavours.
+       /*
+        * Intel PIII Tualatin. This comes in two flavours.
         * One has 256kb of cache, the other 512. We have no way
         * to determine which, so we use a boottime override
         * for the 512kb model, and assume 256 otherwise.
@@ -219,45 +372,47 @@ static unsigned int __cpuinit intel_size_cache(struct cpuinfo_x86 * c, unsigned
                size = 256;
        return size;
 }
+#endif
 
 static struct cpu_dev intel_cpu_dev __cpuinitdata = {
        .c_vendor       = "Intel",
-       .c_ident        = { "GenuineIntel" },
+       .c_ident        = { "GenuineIntel" },
+#ifdef CONFIG_X86_32
        .c_models = {
-               { .vendor = X86_VENDOR_INTEL, .family = 4, .model_names = 
-                 { 
-                         [0] = "486 DX-25/33", 
-                         [1] = "486 DX-50", 
-                         [2] = "486 SX", 
-                         [3] = "486 DX/2", 
-                         [4] = "486 SL", 
-                         [5] = "486 SX/2", 
-                         [7] = "486 DX/2-WB", 
-                         [8] = "486 DX/4", 
+               { .vendor = X86_VENDOR_INTEL, .family = 4, .model_names =
+                 {
+                         [0] = "486 DX-25/33",
+                         [1] = "486 DX-50",
+                         [2] = "486 SX",
+                         [3] = "486 DX/2",
+                         [4] = "486 SL",
+                         [5] = "486 SX/2",
+                         [7] = "486 DX/2-WB",
+                         [8] = "486 DX/4",
                          [9] = "486 DX/4-WB"
                  }
                },
                { .vendor = X86_VENDOR_INTEL, .family = 5, .model_names =
-                 { 
-                         [0] = "Pentium 60/66 A-step", 
-                         [1] = "Pentium 60/66", 
+                 {
+                         [0] = "Pentium 60/66 A-step",
+                         [1] = "Pentium 60/66",
                          [2] = "Pentium 75 - 200",
-                         [3] = "OverDrive PODP5V83", 
+                         [3] = "OverDrive PODP5V83",
                          [4] = "Pentium MMX",
-                         [7] = "Mobile Pentium 75 - 200", 
+                         [7] = "Mobile Pentium 75 - 200",
                          [8] = "Mobile Pentium MMX"
                  }
                },
                { .vendor = X86_VENDOR_INTEL, .family = 6, .model_names =
-                 { 
+                 {
                          [0] = "Pentium Pro A-step",
-                         [1] = "Pentium Pro", 
-                         [3] = "Pentium II (Klamath)", 
-                         [4] = "Pentium II (Deschutes)", 
-                         [5] = "Pentium II (Deschutes)", 
+                         [1] = "Pentium Pro",
+                         [3] = "Pentium II (Klamath)",
+                         [4] = "Pentium II (Deschutes)",
+                         [5] = "Pentium II (Deschutes)",
                          [6] = "Mobile Pentium II",
-                         [7] = "Pentium III (Katmai)", 
-                         [8] = "Pentium III (Coppermine)", 
+                         [7] = "Pentium III (Katmai)",
+                         [8] = "Pentium III (Coppermine)",
                          [10] = "Pentium III (Cascades)",
                          [11] = "Pentium III (Tualatin)",
                  }
@@ -272,62 +427,12 @@ static struct cpu_dev intel_cpu_dev __cpuinitdata = {
                  }
                },
        },
-       .c_init         = init_intel,
        .c_size_cache   = intel_size_cache,
-};
-
-__init int intel_cpu_init(void)
-{
-       cpu_devs[X86_VENDOR_INTEL] = &intel_cpu_dev;
-       return 0;
-}
-
-#ifndef CONFIG_X86_CMPXCHG
-unsigned long cmpxchg_386_u8(volatile void *ptr, u8 old, u8 new)
-{
-       u8 prev;
-       unsigned long flags;
-
-       /* Poor man's cmpxchg for 386. Unsuitable for SMP */
-       local_irq_save(flags);
-       prev = *(u8 *)ptr;
-       if (prev == old)
-               *(u8 *)ptr = new;
-       local_irq_restore(flags);
-       return prev;
-}
-EXPORT_SYMBOL(cmpxchg_386_u8);
-
-unsigned long cmpxchg_386_u16(volatile void *ptr, u16 old, u16 new)
-{
-       u16 prev;
-       unsigned long flags;
-
-       /* Poor man's cmpxchg for 386. Unsuitable for SMP */
-       local_irq_save(flags);
-       prev = *(u16 *)ptr;
-       if (prev == old)
-               *(u16 *)ptr = new;
-       local_irq_restore(flags);
-       return prev;
-}
-EXPORT_SYMBOL(cmpxchg_386_u16);
-
-unsigned long cmpxchg_386_u32(volatile void *ptr, u32 old, u32 new)
-{
-       u32 prev;
-       unsigned long flags;
-
-       /* Poor man's cmpxchg for 386. Unsuitable for SMP */
-       local_irq_save(flags);
-       prev = *(u32 *)ptr;
-       if (prev == old)
-               *(u32 *)ptr = new;
-       local_irq_restore(flags);
-       return prev;
-}
-EXPORT_SYMBOL(cmpxchg_386_u32);
 #endif
+       .c_early_init   = early_init_intel,
+       .c_init         = init_intel,
+       .c_x86_vendor   = X86_VENDOR_INTEL,
+};
 
-// arch_initcall(intel_cpu_init);
+cpu_dev_register(intel_cpu_dev);