[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>

4 files changed, 153 insertions, 6 deletions

diff --git a/arch/i386/kernel/cpu/intel.c b/arch/i386/kernel/cpu/intel.c
index 5386b29bb5a..10afc645c54 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 bd387541963..a76e9314658 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 */
diff --git a/arch/x86_64/kernel/nmi.c b/arch/x86_64/kernel/nmi.c
index ab421e22fa6..399489c9313 100644
--- a/arch/x86_64/kernel/nmi.c
+++ b/arch/x86_64/kernel/nmi.c
@@ -27,6 +27,7 @@
 #include <asm/proto.h>
 #include <asm/kdebug.h>
 #include <asm/mce.h>
+#include <asm/intel_arch_perfmon.h>
 
 /*
  * lapic_nmi_owner tracks the ownership of the lapic NMI hardware:
@@ -66,6 +67,9 @@ static unsigned int nmi_p4_cccr_val;
 #define K7_EVENT_CYCLES_PROCESSOR_IS_RUNNING	0x76
 #define K7_NMI_EVENT		K7_EVENT_CYCLES_PROCESSOR_IS_RUNNING
 
+#define ARCH_PERFMON_NMI_EVENT_SEL	ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL
+#define ARCH_PERFMON_NMI_EVENT_UMASK	ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK
+
 #define MSR_P4_MISC_ENABLE	0x1A0
 #define MSR_P4_MISC_ENABLE_PERF_AVAIL	(1<<7)
 #define MSR_P4_MISC_ENABLE_PEBS_UNAVAIL	(1<<12)
@@ -97,7 +101,10 @@ static __cpuinit inline int nmi_known_cpu(void)
 	case X86_VENDOR_AMD:
 		return boot_cpu_data.x86 == 15;
 	case X86_VENDOR_INTEL:
-		return boot_cpu_data.x86 == 15;
+		if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
+			return 1;
+		else
+			return (boot_cpu_data.x86 == 15);
 	}
 	return 0;
 }
@@ -203,6 +210,8 @@ 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)
@@ -215,6 +224,8 @@ static void disable_lapic_nmi_watchdog(void)
 		if (boot_cpu_data.x86 == 15) {
 			wrmsr(MSR_P4_IQ_CCCR0, 0, 0);
 			wrmsr(MSR_P4_CRU_ESCR0, 0, 0);
+		} else if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) {
+			disable_intel_arch_watchdog();
 		}
 		break;
 	}
@@ -367,6 +378,53 @@ static void setup_k7_watchdog(void)
 	wrmsr(MSR_K7_EVNTSEL0, evntsel, 0);
 }
 
+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);
+	wrmsrl(MSR_ARCH_PERFMON_PERFCTR0, -((u64)cpu_khz * 1000 / nmi_hz));
+	apic_write(APIC_LVTPC, APIC_DM_NMI);
+	evntsel |= ARCH_PERFMON_EVENTSEL0_ENABLE;
+	wrmsr(MSR_ARCH_PERFMON_EVENTSEL0, evntsel, 0);
+	return 1;
+}
+
 
 static int setup_p4_watchdog(void)
 {
@@ -420,10 +478,16 @@ void setup_apic_nmi_watchdog(void)
 		setup_k7_watchdog();
 		break;
 	case X86_VENDOR_INTEL:
-		if (boot_cpu_data.x86 != 15)
-			return;
-		if (!setup_p4_watchdog())
+		if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) {
+			if (!setup_intel_arch_watchdog())
+				return;
+		} else if (boot_cpu_data.x86 == 15) {
+			if (!setup_p4_watchdog())
+				return;
+		} else {
 			return;
+		}
+
 		break;
 
 	default:
@@ -508,7 +572,14 @@ void __kprobes nmi_watchdog_tick(struct pt_regs * regs, unsigned reason)
  			 */
  			wrmsr(MSR_P4_IQ_CCCR0, nmi_p4_cccr_val, 0);
  			apic_write(APIC_LVTPC, APIC_DM_NMI);
- 		}
+ 		} else if (nmi_perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0) {
+			/*
+			 * For Intel based architectural perfmon
+			 * - LVTPC is masked on interrupt and must be
+			 *   unmasked by the LVTPC handler.
+			 */
+			apic_write(APIC_LVTPC, APIC_DM_NMI);
+		}
 		wrmsrl(nmi_perfctr_msr, -((u64)cpu_khz * 1000 / nmi_hz));
 	}
 }
diff --git a/arch/x86_64/kernel/setup.c b/arch/x86_64/kernel/setup.c
index 2a5fce0fd1c..594642a6c03 100644
--- a/arch/x86_64/kernel/setup.c
+++ b/arch/x86_64/kernel/setup.c
@@ -988,6 +988,13 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
 	unsigned n;
 
 	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);
+	}
+
 	n = c->extended_cpuid_level;
 	if (n >= 0x80000008) {
 		unsigned eax = cpuid_eax(0x80000008);