On CPUs that have MPERF/APERF support use that information to install a

cpu frequency sensor for each core. This works on many "modern" Intel and
AMD cpus (probably anything that has some kind of turbo mode).
OK kettenis@

4 files changed, 79 insertions, 13 deletions

diff --git a/sys/arch/amd64/amd64/cpu.c b/sys/arch/amd64/amd64/cpu.c
index 30e734e407b..4778347a12e 100644
--- a/sys/arch/amd64/amd64/cpu.c
+++ b/sys/arch/amd64/amd64/cpu.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.c,v 1.155 2022/02/21 11:03:39 mpi Exp $	*/
+/*	$OpenBSD: cpu.c,v 1.156 2022/04/26 08:35:30 claudio Exp $	*/
 /* $NetBSD: cpu.c,v 1.1 2003/04/26 18:39:26 fvdl Exp $ */
 
 /*-
@@ -558,6 +558,11 @@ cpu_attach(struct device *parent, struct device *self, void *aux)
 	ci->ci_func = caa->cpu_func;
 	ci->ci_handled_intr_level = IPL_NONE;
 
+#ifndef SMALL_KERNEL
+	strlcpy(ci->ci_sensordev.xname, ci->ci_dev->dv_xname,
+	    sizeof(ci->ci_sensordev.xname));
+#endif
+
 #if defined(MULTIPROCESSOR)
 	/*
 	 * Allocate UPAGES contiguous pages for the idle PCB and stack.
@@ -663,6 +668,11 @@ cpu_attach(struct device *parent, struct device *self, void *aux)
 #if NVMM > 0
 	cpu_init_vmm(ci);
 #endif /* NVMM > 0 */
+
+#ifndef SMALL_KERNEL
+	if (ci->ci_sensordev.sensors_count > 0)
+		sensordev_install(&ci->ci_sensordev);
+#endif
 }
 
 static void
diff --git a/sys/arch/amd64/amd64/identcpu.c b/sys/arch/amd64/amd64/identcpu.c
index b67e177a686..060e73e7dd5 100644
--- a/sys/arch/amd64/amd64/identcpu.c
+++ b/sys/arch/amd64/amd64/identcpu.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: identcpu.c,v 1.122 2022/01/20 11:06:57 bluhm Exp $	*/
+/*	$OpenBSD: identcpu.c,v 1.123 2022/04/26 08:35:30 claudio Exp $	*/
 /*	$NetBSD: identcpu.c,v 1.1 2003/04/26 18:39:28 fvdl Exp $	*/
 
 /*
@@ -38,6 +38,8 @@
 
 #include <sys/param.h>
 #include <sys/systm.h>
+#include <sys/atomic.h>
+#include <sys/proc.h>
 #include <sys/sysctl.h>
 
 #include "vmm.h"
@@ -246,7 +248,9 @@ cpu_amd64speed(int *freq)
 }
 
 #ifndef SMALL_KERNEL
-void	intelcore_update_sensor(void *args);
+void	intelcore_update_sensor(void *);
+void	cpu_hz_update_sensor(void *);
+
 /*
  * Temperature read on the CPU is relative to the maximum
  * temperature supported by the CPU, Tj(Max).
@@ -299,6 +303,44 @@ intelcore_update_sensor(void *args)
 	}
 }
 
+/*
+ * Effective CPU frequency measurement
+ *
+ * Refer to:
+ *   64-ia-32-architectures-software-developer-vol-3b-part-2-manual.pdf
+ *   Section 14.2 and
+ *   OSRR for AMD Family 17h processors Section 2.1.2
+ * Round to 50Mhz which is the accuracy of this measurement.
+ */
+#define FREQ_50MHZ	(50ULL * 1000000ULL * 1000000ULL)
+void
+cpu_hz_update_sensor(void *args)
+{
+	extern uint64_t	 tsc_frequency;
+	struct cpu_info	*ci = args;
+	uint64_t	 mperf, aperf, mdelta, adelta, val;
+	unsigned long	 s;
+
+	sched_peg_curproc(ci);
+
+	s = intr_disable();
+	mperf = rdmsr(MSR_MPERF);
+	aperf = rdmsr(MSR_APERF);
+	intr_restore(s);
+
+	mdelta = mperf - ci->ci_hz_mperf;
+	adelta = aperf - ci->ci_hz_aperf;
+	ci->ci_hz_mperf = mperf;
+	ci->ci_hz_aperf = aperf;
+
+	if (mdelta > 0) {
+		val = (adelta * 1000000) / mdelta * tsc_frequency;
+		val = ((val + FREQ_50MHZ / 2) / FREQ_50MHZ) * FREQ_50MHZ; 
+		ci->ci_hz_sensor.value = val;
+	}
+
+	atomic_clearbits_int(&curproc->p_flag, P_CPUPEG);
+}
 #endif
 
 void (*setperf_setup)(struct cpu_info *);
@@ -469,7 +511,7 @@ void
 identifycpu(struct cpu_info *ci)
 {
 	uint64_t freq = 0;
-	u_int32_t dummy, val;
+	u_int32_t dummy, val, cpu_tpm_ecxflags = 0;
 	char mycpu_model[48];
 	int i;
 	char *brandstr_from, *brandstr_to;
@@ -619,12 +661,15 @@ identifycpu(struct cpu_info *ci)
 	}
 
 	if (!strcmp(cpu_vendor, "GenuineIntel") && cpuid_level >= 0x06) {
-		CPUID(0x06, ci->ci_feature_tpmflags, dummy, dummy, dummy);
+		CPUID(0x06, ci->ci_feature_tpmflags, dummy, cpu_tpm_ecxflags,
+		    dummy);
 		for (i = 0; i < nitems(cpu_tpm_eaxfeatures); i++)
 			if (ci->ci_feature_tpmflags &
 			    cpu_tpm_eaxfeatures[i].bit)
 				printf(",%s", cpu_tpm_eaxfeatures[i].str);
 	} else if (!strcmp(cpu_vendor, "AuthenticAMD")) {
+		CPUID(0x06, ci->ci_feature_tpmflags, dummy, cpu_tpm_ecxflags,
+		    dummy);
 		if (ci->ci_family >= 0x12)
 			ci->ci_feature_tpmflags |= TPM_ARAT;
 	}
@@ -737,12 +782,9 @@ identifycpu(struct cpu_info *ci)
 
 #ifndef SMALL_KERNEL
 	if (CPU_IS_PRIMARY(ci) && (ci->ci_feature_tpmflags & TPM_SENSOR)) {
-		strlcpy(ci->ci_sensordev.xname, ci->ci_dev->dv_xname,
-		    sizeof(ci->ci_sensordev.xname));
 		ci->ci_sensor.type = SENSOR_TEMP;
 		sensor_task_register(ci, intelcore_update_sensor, 5);
 		sensor_attach(&ci->ci_sensordev, &ci->ci_sensor);
-		sensordev_install(&ci->ci_sensordev);
 	}
 #endif
 
@@ -762,12 +804,9 @@ identifycpu(struct cpu_info *ci)
 	if (CPU_IS_PRIMARY(ci) && !strcmp(cpu_vendor, "CentaurHauls")) {
 		ci->cpu_setup = via_nano_setup;
 #ifndef SMALL_KERNEL
-		strlcpy(ci->ci_sensordev.xname, ci->ci_dev->dv_xname,
-		    sizeof(ci->ci_sensordev.xname));
 		ci->ci_sensor.type = SENSOR_TEMP;
 		sensor_task_register(ci, via_update_sensor, 5);
 		sensor_attach(&ci->ci_sensordev, &ci->ci_sensor);
-		sensordev_install(&ci->ci_sensordev);
 #endif
 	}
 
@@ -777,6 +816,16 @@ identifycpu(struct cpu_info *ci)
 #if NVMM > 0
 	cpu_check_vmm_cap(ci);
 #endif /* NVMM > 0 */
+
+	/* Check for effective frequency via MPERF, APERF */
+	if ((cpu_tpm_ecxflags & TPM_EFFFREQ) &&
+	    ci->ci_smt_id == 0) {
+#ifndef SMALL_KERNEL
+		ci->ci_hz_sensor.type = SENSOR_FREQ;
+		sensor_task_register(ci, cpu_hz_update_sensor, 1);
+		sensor_attach(&ci->ci_sensordev, &ci->ci_hz_sensor);
+#endif
+	}
 }
 
 #ifndef SMALL_KERNEL
diff --git a/sys/arch/amd64/include/cpu.h b/sys/arch/amd64/include/cpu.h
index 4bc8627b554..0bf6cc41900 100644
--- a/sys/arch/amd64/include/cpu.h
+++ b/sys/arch/amd64/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.141 2021/08/31 17:40:59 dv Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.142 2022/04/26 08:35:30 claudio Exp $	*/
 /*	$NetBSD: cpu.h,v 1.1 2003/04/26 18:39:39 fvdl Exp $	*/
 
 /*-
@@ -196,6 +196,9 @@ struct cpu_info {
 
 	struct ksensordev	ci_sensordev;
 	struct ksensor		ci_sensor;
+	struct ksensor		ci_hz_sensor;
+	u_int64_t		ci_hz_mperf;
+	u_int64_t		ci_hz_aperf;
 #if defined(GPROF) || defined(DDBPROF)
 	struct gmonparam	*ci_gmon;
 #endif
diff --git a/sys/arch/amd64/include/specialreg.h b/sys/arch/amd64/include/specialreg.h
index 6b32f07e2f4..34701ffb864 100644
--- a/sys/arch/amd64/include/specialreg.h
+++ b/sys/arch/amd64/include/specialreg.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: specialreg.h,v 1.91 2021/11/19 04:00:53 guenther Exp $	*/
+/*	$OpenBSD: specialreg.h,v 1.92 2022/04/26 08:35:30 claudio Exp $	*/
 /*	$NetBSD: specialreg.h,v 1.1 2003/04/26 18:39:48 fvdl Exp $	*/
 /*	$NetBSD: x86/specialreg.h,v 1.2 2003/04/25 21:54:30 fvdl Exp $	*/
 
@@ -235,6 +235,8 @@
  */
 #define	TPM_SENSOR	0x00000001	 /* Digital temp sensor */
 #define	TPM_ARAT	0x00000004	 /* APIC Timer Always Running */
+/* Thermal and Power Management (CPUID function 0x6) ECX bits */
+#define	TPM_EFFFREQ	0x00000001	 /* APERF & MPERF MSR present */
 
  /*
   * "Architectural Performance Monitoring" bits (CPUID function 0x0a):
@@ -369,6 +371,8 @@
 #define MSR_PERFCTR0		0x0c1
 #define MSR_PERFCTR1		0x0c2
 #define MSR_FSB_FREQ		0x0cd	/* Core Duo/Solo only */
+#define MSR_MPERF		0x0e7
+#define MSR_APERF		0x0e8
 #define MSR_MTRRcap		0x0fe
 #define MTRRcap_FIXED		0x100	/* bit 8 - fixed MTRRs supported */
 #define MTRRcap_WC		0x400	/* bit 10 - WC type supported */