Recent x86 CPUs come with a constant time stamp counter. If this is

the case we verify if the CPU supports a specific version of the
architectural performance monitoring feature and read out the current
frequency from the fixed-function performance counter of the unhalted
core.

My initial motivation to implement this was the Soekris net6501-70
which comes with an Intel Atom E6xx 1.60GHz CPU.  It has a constant
time stamp counter plus speed step support and boots on the lowest
frequency of 600MHz.  This caused hw.cpuspeed and hw.setperf to
reflect the wrong values.

The diff is a cooperation work with jsg@.  The fixed-function
performance counter read code comes from a former diff of him.

OK jsg@

5 files changed, 130 insertions, 11 deletions

diff --git a/sys/arch/i386/i386/locore.s b/sys/arch/i386/i386/locore.s
index fccbe7e1a75..f2f42b22202 100644
--- a/sys/arch/i386/i386/locore.s
+++ b/sys/arch/i386/i386/locore.s
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.s,v 1.143 2012/10/31 03:30:22 jsg Exp $	*/
+/*	$OpenBSD: locore.s,v 1.144 2012/11/10 09:45:05 mglocker Exp $	*/
 /*	$NetBSD: locore.s,v 1.145 1996/05/03 19:41:19 christos Exp $	*/
 
 /*-
@@ -170,6 +170,10 @@
 	.globl	_C_LABEL(ecpu_feature), _C_LABEL(ecpu_ecxfeature)
 	.globl	_C_LABEL(cpu_cache_eax), _C_LABEL(cpu_cache_ebx)
 	.globl	_C_LABEL(cpu_cache_ecx), _C_LABEL(cpu_cache_edx)
+	.globl	_C_LABEL(cpu_perf_eax)
+	.globl	_C_LABEL(cpu_perf_ebx)
+	.globl	_C_LABEL(cpu_perf_edx)
+	.globl	_C_LABEL(cpu_apmi_edx)
 	.globl	_C_LABEL(cold), _C_LABEL(cnvmem), _C_LABEL(extmem)
 	.globl	_C_LABEL(esym)
 	.globl	_C_LABEL(boothowto), _C_LABEL(bootdev), _C_LABEL(atdevbase)
@@ -213,6 +217,10 @@ _C_LABEL(cpu_cache_eax):.long	0
 _C_LABEL(cpu_cache_ebx):.long	0
 _C_LABEL(cpu_cache_ecx):.long	0
 _C_LABEL(cpu_cache_edx):.long	0
+_C_LABEL(cpu_perf_eax):	.long	0	# arch. perf. mon. flags from 'cpuid'
+_C_LABEL(cpu_perf_ebx):	.long	0	# arch. perf. mon. flags from 'cpuid'
+_C_LABEL(cpu_perf_edx):	.long	0	# arch. perf. mon. flags from 'cpuid'
+_C_LABEL(cpu_apmi_edx):	.long	0	# adv. power management info. 'cpuid'
 _C_LABEL(cpu_vendor): .space 16	# vendor string returned by 'cpuid' instruction
 _C_LABEL(cpu_brandstr):	.space 48 # brand string returned by 'cpuid'
 _C_LABEL(cold):		.long	1	# cold till we are not
@@ -415,6 +423,12 @@ try586:	/* Use the `cpuid' instruction. */
 	movl	%ecx,RELOC(_C_LABEL(cpu_cache_ecx))
 	movl	%edx,RELOC(_C_LABEL(cpu_cache_edx))
 
+	movl	$0x0a,%eax
+	cpuid
+	movl	%eax,RELOC(_C_LABEL(cpu_perf_eax))
+	movl	%ebx,RELOC(_C_LABEL(cpu_perf_ebx))
+	movl	%edx,RELOC(_C_LABEL(cpu_perf_edx))
+
 1:
 	/* Check if brand identification string is supported */
 	movl	$0x80000000,%eax
@@ -445,6 +459,10 @@ try586:	/* Use the `cpuid' instruction. */
 	andl	$0x00ffffff,%edx	/* Shouldn't be necessary */
 	movl	%edx,RELOC(_C_LABEL(cpu_brandstr))+44
 
+	movl	$0x80000007,%eax
+	cpuid
+	movl	%edx,RELOC(_C_LABEL(cpu_apmi_edx))
+
 2:
 	/*
 	 * Finished with old stack; load new %esp now instead of later so we
diff --git a/sys/arch/i386/i386/machdep.c b/sys/arch/i386/i386/machdep.c
index 7795286ea43..00305d9d766 100644
--- a/sys/arch/i386/i386/machdep.c
+++ b/sys/arch/i386/i386/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.516 2012/10/31 03:30:22 jsg Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.517 2012/11/10 09:45:05 mglocker Exp $	*/
 /*	$NetBSD: machdep.c,v 1.214 1996/11/10 03:16:17 thorpej Exp $	*/
 
 /*-
@@ -1069,6 +1069,14 @@ const struct cpu_cpuid_feature cpu_seff0_ebxfeatures[] = {
 	{ SEFF0EBX_SMAP,	"SMAP" },
 };
 
+const struct cpu_cpuid_feature i386_cpuid_eaxperf[] = {
+	{ CPUIDEAX_VERID,	"PERF" },
+};
+
+const struct cpu_cpuid_feature i386_cpuid_edxapmi[] = {
+	{ CPUIDEDX_ITSC,	"ITSC" },
+};
+
 void
 winchip_cpu_setup(struct cpu_info *ci)
 {
@@ -1838,7 +1846,20 @@ identifycpu(struct cpu_info *ci)
 	}
 
 	if (ci->ci_feature_flags && (ci->ci_feature_flags & CPUID_TSC)) {
-		/* Has TSC */
+		/* Has TSC, check if it's constant */
+		switch (vendor) {
+		case CPUVENDOR_INTEL:
+			if ((ci->ci_family == 0x0f && ci->ci_model >= 0x03) ||
+			    (ci->ci_family == 0x06 && ci->ci_model >= 0x0e)) {
+				ci->ci_flags |= CPUF_CONST_TSC;
+			}
+			break;
+		case CPUVENDOR_VIA:
+			if (ci->ci_model >= 0x0f) {
+				ci->ci_flags |= CPUF_CONST_TSC;
+			}
+			break;
+		}
 		calibrate_cyclecounter();
 		if (cpuspeed > 994) {
 			int ghz, fr;
@@ -1900,6 +1921,22 @@ identifycpu(struct cpu_info *ci)
 					numbits++;
 				}
 			}
+			for (i = 0; i < nitems(i386_cpuid_eaxperf); i++) {
+				if (cpu_perf_eax &
+				    i386_cpuid_eaxperf[i].feature_bit) {
+					printf("%s%s", (numbits == 0 ? "" : ","),
+					    i386_cpuid_eaxperf[i].feature_name);
+					numbits++;
+				}
+			}
+			for (i = 0; i < nitems(i386_cpuid_edxapmi); i++) {
+				if (cpu_apmi_edx &
+				    i386_cpuid_edxapmi[i].feature_bit) {
+					printf("%s%s", (numbits == 0 ? "" : ","),
+					    i386_cpuid_edxapmi[i].feature_name);
+					numbits++;
+				}
+			}
 
 			if (cpuid_level >= 0x07) {
 				u_int dummy;
diff --git a/sys/arch/i386/include/cpu.h b/sys/arch/i386/include/cpu.h
index 003ceea564c..b87b89aee46 100644
--- a/sys/arch/i386/include/cpu.h
+++ b/sys/arch/i386/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.123 2012/10/09 04:40:36 jsg Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.124 2012/11/10 09:45:05 mglocker Exp $	*/
 /*	$NetBSD: cpu.h,v 1.35 1996/05/05 19:29:26 christos Exp $	*/
 
 /*-
@@ -163,6 +163,7 @@ struct cpu_info {
 #define	CPUF_SP		0x0004		/* CPU is only processor */
 #define	CPUF_PRIMARY	0x0008		/* CPU is active primary processor */
 #define	CPUF_APIC_CD	0x0010		/* CPU has apic configured */
+#define	CPUF_CONST_TSC	0x0020		/* CPU has constant TSC */
 
 #define	CPUF_PRESENT	0x1000		/* CPU is present */
 #define	CPUF_RUNNING	0x2000		/* CPU is running */
@@ -319,6 +320,10 @@ extern int cpu_cache_eax;
 extern int cpu_cache_ebx;
 extern int cpu_cache_ecx;
 extern int cpu_cache_edx;
+extern int cpu_perf_eax;
+extern int cpu_perf_ebx;
+extern int cpu_perf_edx;
+extern int cpu_apmi_edx;
 
 /* machdep.c */
 extern int cpu_apmhalt;
diff --git a/sys/arch/i386/include/specialreg.h b/sys/arch/i386/include/specialreg.h
index d693449f881..5f5ed76b9b1 100644
--- a/sys/arch/i386/include/specialreg.h
+++ b/sys/arch/i386/include/specialreg.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: specialreg.h,v 1.43 2012/10/09 09:16:09 jsg Exp $	*/
+/*	$OpenBSD: specialreg.h,v 1.44 2012/11/10 09:45:05 mglocker Exp $	*/
 /*	$NetBSD: specialreg.h,v 1.7 1994/10/27 04:16:26 cgd Exp $	*/
 
 /*-
@@ -177,6 +177,27 @@
 #define	SEFF0EBX_SMAP		0x00100000 /* Supervisor mode access prevent */
 
 /*
+ * "Architectural Performance Monitoring" bits (CPUID function 0x0a):
+ * EAX bits
+ */
+
+#define CPUIDEAX_VERID			0x000000ff
+#define CPUIDEAX_NUM_GC(cpuid)		(((cpuid) >>  8) & 0x000000ff)
+#define CPUIDEAX_BIT_GC(cpuid)		(((cpuid) >> 16) & 0x000000ff)
+#define CPUIDEAX_LEN_EBX(cpuid)		(((cpuid) >> 24) & 0x000000ff)
+
+#define CPUIDEBX_EVT_CORE		(1 << 0) /* Core cycle */
+#define CPUIDEBX_EVT_INST		(1 << 1) /* Instruction retired */
+#define CPUIDEBX_EVT_REFR		(1 << 2) /* Reference cycles */
+#define CPUIDEBX_EVT_CACHE_REF		(1 << 3) /* Last-level cache ref. */
+#define CPUIDEBX_EVT_CACHE_MIS		(1 << 4) /* Last-level cache miss. */
+#define CPUIDEBX_EVT_BRANCH_INST	(1 << 5) /* Branch instruction ret. */
+#define CPUIDEBX_EVT_BRANCH_MISP	(1 << 6) /* Branch mispredict ret. */
+
+#define CPUIDEDX_NUM_FC(cpuid)		(((cpuid) >> 0) & 0x0000001f)
+#define CPUIDEDX_BIT_FC(cpuid)		(((cpuid) >> 5) & 0x000000ff)
+
+/*
  * CPUID "extended features" bits (CPUID function 0x80000001):
  * EDX bits, then ECX bits
  */
@@ -213,6 +234,13 @@
 #define	CPUIDECX_TBM		0x00200000 /* Trailing bit manipulation instruction */
 #define	CPUIDECX_TOPEXT		0x00400000 /* Topology extensions support */
 
+/*
+ * "Advanced Power Management Information" bits (CPUID function 0x80000007):
+ * EDX bits.
+ */
+
+#define CPUIDEDX_ITSC		(1 << 8)	/* Invariant TSC */
+
 #define	CPUID2FAMILY(cpuid)	(((cpuid) >> 8) & 15)
 #define	CPUID2MODEL(cpuid)	(((cpuid) >> 4) & 15)
 #define	CPUID2STEPPING(cpuid)	((cpuid) & 15)
@@ -309,6 +337,14 @@
 #define	MSR_MTRRfix4K_F8000	0x26f
 #define MSR_CR_PAT		0x277
 #define MSR_MTRRdefType		0x2ff
+#define MSR_PERF_FIXED_CTR1	0x30a	/* CPU_CLK_Unhalted.Core */
+#define MSR_PERF_FIXED_CTR2	0x30b	/* CPU_CLK.Unhalted.Ref */
+#define MSR_PERF_FIXED_CTR_CTRL	0x38d
+#define MSR_PERF_FIXED_CTR1_EN	(1 << 4)
+#define MSR_PERF_FIXED_CTR2_EN	(1 << 8)
+#define MSR_PERF_GLOBAL_CTRL	0x38f
+#define MSR_PERF_GLOBAL_CTR1_EN	(1ULL << 33)
+#define MSR_PERF_GLOBAL_CTR2_EN	(1ULL << 34)
 #define MSR_MC0_CTL		0x400
 #define MSR_MC0_STATUS		0x401
 #define MSR_MC0_ADDR		0x402
diff --git a/sys/arch/i386/isa/clock.c b/sys/arch/i386/isa/clock.c
index c9ac92eb481..8e674950b6d 100644
--- a/sys/arch/i386/isa/clock.c
+++ b/sys/arch/i386/isa/clock.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: clock.c,v 1.46 2011/07/05 17:11:07 oga Exp $	*/
+/*	$OpenBSD: clock.c,v 1.47 2012/11/10 09:45:05 mglocker Exp $	*/
 /*	$NetBSD: clock.c,v 1.39 1996/05/12 23:11:54 mycroft Exp $	*/
 
 /*-
@@ -376,11 +376,34 @@ i8254_delay(int n)
 void
 calibrate_cyclecounter(void)
 {
-	unsigned long long count, last_count;
-
-	__asm __volatile("rdtsc" : "=A" (last_count));
-	delay(1000000);
-	__asm __volatile("rdtsc" : "=A" (count));
+	struct cpu_info *ci;
+	unsigned long long count, last_count, msr;
+
+	ci = curcpu();
+
+	if ((ci->ci_flags & CPUF_CONST_TSC) &&
+	    (cpu_perf_eax & CPUIDEAX_VERID) > 1 &&
+	    CPUIDEDX_NUM_FC(cpu_perf_edx) > 1) {
+		msr = rdmsr(MSR_PERF_FIXED_CTR_CTRL) | MSR_PERF_FIXED_CTR1_EN;
+		wrmsr(MSR_PERF_FIXED_CTR_CTRL, msr);
+		msr = rdmsr(MSR_PERF_GLOBAL_CTRL) | MSR_PERF_GLOBAL_CTR1_EN;
+		wrmsr(MSR_PERF_GLOBAL_CTRL, msr);
+
+		last_count = rdmsr(MSR_PERF_FIXED_CTR1);
+		delay(1000000);
+		count = rdmsr(MSR_PERF_FIXED_CTR1);
+
+		msr = rdmsr(MSR_PERF_FIXED_CTR_CTRL);
+		msr &= ~MSR_PERF_FIXED_CTR1_EN;
+		wrmsr(MSR_PERF_FIXED_CTR_CTRL, msr);
+		msr = rdmsr(MSR_PERF_GLOBAL_CTRL);
+		msr &= ~MSR_PERF_GLOBAL_CTR1_EN;
+		wrmsr(MSR_PERF_GLOBAL_CTRL, msr);
+	} else {
+		__asm __volatile("rdtsc" : "=A" (last_count));
+		delay(1000000);
+		__asm __volatile("rdtsc" : "=A" (count));
+	}
 	cpuspeed = ((count - last_count) + 999999) / 1000000;
 }