[PATCH] x86_64: i386/x86-64 Add nmi watchdog support for new Intel CPUs

Intel now has support for Architectural Performance Monitoring Counters
( Refer to IA-32 Intel Architecture Software Developer's Manual
http://www.intel.com/design/pentium4/manuals/253669.htm ). This
feature is present starting from Intel Core Duo and Intel Core Solo processors.

What this means is, the performance monitoring counters and some performance
monitoring events are now defined in an architectural way (using cpuid).
And there will be no need to check for family/model etc for these architectural
events.

Below is the patch to use this performance counters in nmi watchdog driver.
Patch handles both i386 and x86-64 kernels.

Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

2 files changed, 70 insertions, 1 deletions

diff --git a/arch/i386/kernel/cpu/intel.c b/arch/i386/kernel/cpu/intel.c
index 5386b29bb5a5..10afc645c540 100644
--- a/arch/i386/kernel/cpu/intel.c
+++ b/arch/i386/kernel/cpu/intel.c
@@ -122,6 +122,12 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
 
         select_idle_routine(c);
         l2 = init_intel_cacheinfo(c);
+        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);
+        }
 
         /* 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)
diff --git a/arch/i386/kernel/nmi.c b/arch/i386/kernel/nmi.c
index bd3875419630..a76e93146585 100644
--- a/arch/i386/kernel/nmi.c
+++ b/arch/i386/kernel/nmi.c
@@ -24,6 +24,7 @@
 
 #include <asm/smp.h>
 #include <asm/nmi.h>
+#include <asm/intel_arch_perfmon.h>
 
 #include "mach_traps.h"
 
@@ -95,6 +96,9 @@ int nmi_active;
         (P4_CCCR_OVF_PMI0|P4_CCCR_THRESHOLD(15)|P4_CCCR_COMPLEMENT|     \
          P4_CCCR_COMPARE|P4_CCCR_REQUIRED|P4_CCCR_ESCR_SELECT(4)|P4_CCCR_ENABLE)
 
+#define ARCH_PERFMON_NMI_EVENT_SEL      ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL
+#define ARCH_PERFMON_NMI_EVENT_UMASK    ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK
+
 #ifdef CONFIG_SMP
 /* The performance counters used by NMI_LOCAL_APIC don't trigger when
  * the CPU is idle. To make sure the NMI watchdog really ticks on all
@@ -207,6 +211,8 @@ static int __init setup_nmi_watchdog(char *str)
 
 __setup("nmi_watchdog=", setup_nmi_watchdog);
 
+static void disable_intel_arch_watchdog(void);
+
 static void disable_lapic_nmi_watchdog(void)
 {
         if (nmi_active <= 0)
@@ -216,6 +222,10 @@ static void disable_lapic_nmi_watchdog(void)
                 wrmsr(MSR_K7_EVNTSEL0, 0, 0);
                 break;
         case X86_VENDOR_INTEL:
+                if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) {
+                        disable_intel_arch_watchdog();
+                        break;
+                }
                 switch (boot_cpu_data.x86) {
                 case 6:
                         if (boot_cpu_data.x86_model > 0xd)
@@ -444,6 +454,53 @@ static int setup_p4_watchdog(void)
         return 1;
 }
 
+static void disable_intel_arch_watchdog(void)
+{
+        unsigned ebx;
+
+        /*
+         * Check whether the Architectural PerfMon supports
+         * Unhalted Core Cycles Event or not.
+         * NOTE: Corresponding bit = 0 in ebp indicates event present.
+         */
+        ebx = cpuid_ebx(10);
+        if (!(ebx & ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT))
+                wrmsr(MSR_ARCH_PERFMON_EVENTSEL0, 0, 0);
+}
+
+static int setup_intel_arch_watchdog(void)
+{
+        unsigned int evntsel;
+        unsigned ebx;
+
+        /*
+         * Check whether the Architectural PerfMon supports
+         * Unhalted Core Cycles Event or not.
+         * NOTE: Corresponding bit = 0 in ebp indicates event present.
+         */
+        ebx = cpuid_ebx(10);
+        if ((ebx & ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT))
+                return 0;
+
+        nmi_perfctr_msr = MSR_ARCH_PERFMON_PERFCTR0;
+
+        clear_msr_range(MSR_ARCH_PERFMON_EVENTSEL0, 2);
+        clear_msr_range(MSR_ARCH_PERFMON_PERFCTR0, 2);
+
+        evntsel = ARCH_PERFMON_EVENTSEL_INT
+                | ARCH_PERFMON_EVENTSEL_OS
+                | ARCH_PERFMON_EVENTSEL_USR
+                | ARCH_PERFMON_NMI_EVENT_SEL
+                | ARCH_PERFMON_NMI_EVENT_UMASK;
+
+        wrmsr(MSR_ARCH_PERFMON_EVENTSEL0, evntsel, 0);
+        write_watchdog_counter("INTEL_ARCH_PERFCTR0");
+        apic_write(APIC_LVTPC, APIC_DM_NMI);
+        evntsel |= ARCH_PERFMON_EVENTSEL0_ENABLE;
+        wrmsr(MSR_ARCH_PERFMON_EVENTSEL0, evntsel, 0);
+        return 1;
+}
+
 void setup_apic_nmi_watchdog (void)
 {
         switch (boot_cpu_data.x86_vendor) {
@@ -453,6 +510,11 @@ void setup_apic_nmi_watchdog (void)
                 setup_k7_watchdog();
                 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)
@@ -556,7 +618,8 @@ void nmi_watchdog_tick (struct pt_regs * regs)
                         wrmsr(MSR_P4_IQ_CCCR0, nmi_p4_cccr_val, 0);
                         apic_write(APIC_LVTPC, APIC_DM_NMI);
                 }
-                else if (nmi_perfctr_msr == MSR_P6_PERFCTR0) {
+                else if (nmi_perfctr_msr == MSR_P6_PERFCTR0 ||
+                         nmi_perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0) {
                         /* Only P6 based Pentium M need to re-unmask
                          * the apic vector but it doesn't hurt
                          * other P6 variant */