Merge branch 'master' into gfs2
[pandora-kernel.git] / arch / i386 / kernel / nmi.c
index 6e5085d..dbda706 100644 (file)
 #include <linux/sysctl.h>
 #include <linux/percpu.h>
 #include <linux/dmi.h>
+#include <linux/kprobes.h>
 
 #include <asm/smp.h>
 #include <asm/nmi.h>
 #include <asm/kdebug.h>
+#include <asm/intel_arch_perfmon.h>
 
 #include "mach_traps.h"
 
@@ -77,6 +79,9 @@ static inline unsigned int nmi_perfctr_msr_to_bit(unsigned int msr)
        case X86_VENDOR_AMD:
                return (msr - MSR_K7_PERFCTR0);
        case X86_VENDOR_INTEL:
+               if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
+                       return (msr - MSR_ARCH_PERFMON_PERFCTR0);
+
                switch (boot_cpu_data.x86) {
                case 6:
                        return (msr - MSR_P6_PERFCTR0);
@@ -95,6 +100,9 @@ static inline unsigned int nmi_evntsel_msr_to_bit(unsigned int msr)
        case X86_VENDOR_AMD:
                return (msr - MSR_K7_EVNTSEL0);
        case X86_VENDOR_INTEL:
+               if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
+                       return (msr - MSR_ARCH_PERFMON_EVENTSEL0);
+
                switch (boot_cpu_data.x86) {
                case 6:
                        return (msr - MSR_P6_EVNTSEL0);
@@ -174,7 +182,10 @@ static __cpuinit inline int nmi_known_cpu(void)
        case X86_VENDOR_AMD:
                return ((boot_cpu_data.x86 == 15) || (boot_cpu_data.x86 == 6));
        case X86_VENDOR_INTEL:
-               return ((boot_cpu_data.x86 == 15) || (boot_cpu_data.x86 == 6));
+               if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
+                       return 1;
+               else
+                       return ((boot_cpu_data.x86 == 15) || (boot_cpu_data.x86 == 6));
        }
        return 0;
 }
@@ -261,8 +272,24 @@ static int __init check_nmi_watchdog(void)
 
        /* now that we know it works we can reduce NMI frequency to
           something more reasonable; makes a difference in some configs */
-       if (nmi_watchdog == NMI_LOCAL_APIC)
+       if (nmi_watchdog == NMI_LOCAL_APIC) {
+               struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
+
                nmi_hz = 1;
+               /*
+                * On Intel CPUs with ARCH_PERFMON only 32 bits in the counter
+                * are writable, with higher bits sign extending from bit 31.
+                * So, we can only program the counter with 31 bit values and
+                * 32nd bit should be 1, for 33.. to be 1.
+                * Find the appropriate nmi_hz
+                */
+               if (wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0 &&
+                       ((u64)cpu_khz * 1000) > 0x7fffffffULL) {
+                       u64 count = (u64)cpu_khz * 1000;
+                       do_div(count, 0x7fffffffUL);
+                       nmi_hz = count + 1;
+               }
+       }
 
        kfree(prev_nmi_count);
        return 0;
@@ -637,6 +664,85 @@ static void stop_p4_watchdog(void)
        release_perfctr_nmi(wd->perfctr_msr);
 }
 
+#define ARCH_PERFMON_NMI_EVENT_SEL     ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL
+#define ARCH_PERFMON_NMI_EVENT_UMASK   ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK
+
+static int setup_intel_arch_watchdog(void)
+{
+       unsigned int ebx;
+       union cpuid10_eax eax;
+       unsigned int unused;
+       unsigned int perfctr_msr, evntsel_msr;
+       unsigned int evntsel;
+       struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
+
+       /*
+        * Check whether the Architectural PerfMon supports
+        * Unhalted Core Cycles Event or not.
+        * NOTE: Corresponding bit = 0 in ebx indicates event present.
+        */
+       cpuid(10, &(eax.full), &ebx, &unused, &unused);
+       if ((eax.split.mask_length < (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX+1)) ||
+           (ebx & ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT))
+               goto fail;
+
+       perfctr_msr = MSR_ARCH_PERFMON_PERFCTR0;
+       evntsel_msr = MSR_ARCH_PERFMON_EVENTSEL0;
+
+       if (!reserve_perfctr_nmi(perfctr_msr))
+               goto fail;
+
+       if (!reserve_evntsel_nmi(evntsel_msr))
+               goto fail1;
+
+       wrmsrl(perfctr_msr, 0UL);
+
+       evntsel = ARCH_PERFMON_EVENTSEL_INT
+               | ARCH_PERFMON_EVENTSEL_OS
+               | ARCH_PERFMON_EVENTSEL_USR
+               | ARCH_PERFMON_NMI_EVENT_SEL
+               | ARCH_PERFMON_NMI_EVENT_UMASK;
+
+       /* setup the timer */
+       wrmsr(evntsel_msr, evntsel, 0);
+       write_watchdog_counter(perfctr_msr, "INTEL_ARCH_PERFCTR0");
+       apic_write(APIC_LVTPC, APIC_DM_NMI);
+       evntsel |= ARCH_PERFMON_EVENTSEL0_ENABLE;
+       wrmsr(evntsel_msr, evntsel, 0);
+
+       wd->perfctr_msr = perfctr_msr;
+       wd->evntsel_msr = evntsel_msr;
+       wd->cccr_msr = 0;  //unused
+       wd->check_bit = 1ULL << (eax.split.bit_width - 1);
+       return 1;
+fail1:
+       release_perfctr_nmi(perfctr_msr);
+fail:
+       return 0;
+}
+
+static void stop_intel_arch_watchdog(void)
+{
+       unsigned int ebx;
+       union cpuid10_eax eax;
+       unsigned int unused;
+       struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
+
+       /*
+        * Check whether the Architectural PerfMon supports
+        * Unhalted Core Cycles Event or not.
+        * NOTE: Corresponding bit = 0 in ebx indicates event present.
+        */
+       cpuid(10, &(eax.full), &ebx, &unused, &unused);
+       if ((eax.split.mask_length < (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX+1)) ||
+           (ebx & ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT))
+               return;
+
+       wrmsr(wd->evntsel_msr, 0, 0);
+       release_evntsel_nmi(wd->evntsel_msr);
+       release_perfctr_nmi(wd->perfctr_msr);
+}
+
 void setup_apic_nmi_watchdog (void *unused)
 {
        struct nmi_watchdog_ctlblk *wd = &__get_cpu_var(nmi_watchdog_ctlblk);
@@ -663,6 +769,11 @@ void setup_apic_nmi_watchdog (void *unused)
                                return;
                        break;
                case X86_VENDOR_INTEL:
+                       if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) {
+                               if (!setup_intel_arch_watchdog())
+                                       return;
+                               break;
+                       }
                        switch (boot_cpu_data.x86) {
                        case 6:
                                if (boot_cpu_data.x86_model > 0xd)
@@ -708,6 +819,10 @@ void stop_apic_nmi_watchdog(void *unused)
                        stop_k7_watchdog();
                        break;
                case X86_VENDOR_INTEL:
+                       if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) {
+                               stop_intel_arch_watchdog();
+                               break;
+                       }
                        switch (boot_cpu_data.x86) {
                        case 6:
                                if (boot_cpu_data.x86_model > 0xd)
@@ -768,7 +883,7 @@ EXPORT_SYMBOL(touch_nmi_watchdog);
 
 extern void die_nmi(struct pt_regs *, const char *msg);
 
-int nmi_watchdog_tick (struct pt_regs * regs, unsigned reason)
+__kprobes int nmi_watchdog_tick(struct pt_regs * regs, unsigned reason)
 {
 
        /*
@@ -831,10 +946,12 @@ int nmi_watchdog_tick (struct pt_regs * regs, unsigned reason)
                                wrmsrl(wd->cccr_msr, dummy);
                                apic_write(APIC_LVTPC, APIC_DM_NMI);
                        }
-                       else if (wd->perfctr_msr == MSR_P6_PERFCTR0) {
-                               /* Only P6 based Pentium M need to re-unmask
+                       else if (wd->perfctr_msr == MSR_P6_PERFCTR0 ||
+                                wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0) {
+                               /* P6 based Pentium M need to re-unmask
                                 * the apic vector but it doesn't hurt
-                                * other P6 variant */
+                                * other P6 variant.
+                                * ArchPerfom/Core Duo also needs this */
                                apic_write(APIC_LVTPC, APIC_DM_NMI);
                        }
                        /* start the cycle over again */
@@ -846,8 +963,7 @@ int nmi_watchdog_tick (struct pt_regs * regs, unsigned reason)
                         * This matches the old behaviour.
                         */
                        rc = 1;
-               } else
-                       printk(KERN_WARNING "Unknown enabled NMI hardware?!\n");
+               }
        }
 done:
        return rc;