Import a hackish tool I use to examine the skew of the TSC across

CPUs in a system.  The tool queries the kernel for its set of CPUs
and compares TSC values on each of the additional CPUs to the first
CPU in turn.  It then outputs a table of simple statistics.

tools/tools/tscdrift/Makefile

@@ -0,0 +1,10 @@
+# $FreeBSD$
+
+PROG=	tscdrift
+MAN=
+WARNS?=	6
+
+LDADD=	-lpthread -lm
+DPADD=	${LIBPTHREAD} ${LIBM}
+
+.include <bsd.prog.mk>

tools/tools/tscdrift/tscdrift.c

@@ -0,0 +1,193 @@
+/*-
+ * Copyright (c) 2014 Advanced Computing Technologies LLC
+ * Written by: John H. Baldwin <jhb@FreeBSD.org>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+#include <sys/param.h>
+#include <sys/cpuset.h>
+#include <machine/atomic.h>
+#include <machine/cpu.h>
+#include <machine/cpufunc.h>
+#include <assert.h>
+#include <err.h>
+#include <errno.h>
+#include <math.h>
+#include <pthread.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <stdio.h>
+
+#define	barrier()	__asm __volatile("" ::: "memory")
+
+#define	TESTS		1024
+
+static volatile int gate;
+static volatile uint64_t thread_tsc;
+
+/* Bind the current thread to the specified CPU. */
+static void
+bind_cpu(int cpu)
+{
+	cpuset_t set;
+
+	CPU_ZERO(&set);
+	CPU_SET(cpu, &set);
+	if (cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_TID, -1, sizeof(set),
+	    &set) < 0)
+		err(1, "cpuset_setaffinity(%d)", cpu);
+}
+
+static void *
+thread_main(void *arg)
+{
+	int cpu, i;
+
+	cpu = (intptr_t)arg;
+	bind_cpu(cpu);
+	for (i = 0; i < TESTS; i++) {
+		gate = 1;
+		while (gate == 1)
+			cpu_spinwait();
+		barrier();
+
+		__asm __volatile("lfence");
+		thread_tsc = rdtsc();
+
+		barrier();
+		gate = 3;
+		while (gate == 3)
+			cpu_spinwait();
+	}
+	return (NULL);
+}
+
+int
+main(int ac __unused, char **av __unused)
+{
+	cpuset_t all_cpus;
+	int64_t **skew, *aveskew, *minskew, *maxskew;
+	float *stddev;
+	double sumsq;
+	pthread_t child;
+	uint64_t tsc;
+	int *cpus;
+	int error, i, j, ncpu;
+
+	/*
+	 * Find all the CPUs this program is eligible to run on and use
+	 * this as our global set.  This means you can use cpuset to
+	 * restrict this program to only run on a subset of CPUs.
+	 */
+	if (cpuset_getaffinity(CPU_LEVEL_WHICH, CPU_WHICH_PID, -1,
+	    sizeof(all_cpus), &all_cpus) < 0)
+		err(1, "cpuset_getaffinity");
+	for (ncpu = 0, i = 0; i < CPU_SETSIZE; i++) {
+		if (CPU_ISSET(i, &all_cpus))
+			ncpu++;
+	}
+	if (ncpu < 2)
+		errx(1, "Only one available CPU");
+	cpus = calloc(ncpu, sizeof(*cpus));
+	skew = calloc(ncpu, sizeof(*skew));
+	for (i = 0; i < ncpu; i++)
+		skew[i] = calloc(TESTS, sizeof(*skew[i]));
+	for (i = 0, j = 0; i < CPU_SETSIZE; i++)
+		if (CPU_ISSET(i, &all_cpus)) {
+			assert(j < ncpu);
+			cpus[j] = i;
+			j++;
+		}
+
+	/*
+	 * We bind this thread to the first CPU and then bind all the
+	 * other threads to other CPUs in turn saving TESTS counts of
+	 * skew calculations.
+	 */
+	bind_cpu(cpus[0]);
+	for (i = 1; i < ncpu; i++) {
+		error = pthread_create(&child, NULL, thread_main,
+		    (void *)(intptr_t)cpus[i]);
+		if (error)
+			errc(1, error, "pthread_create");
+
+		for (j = 0; j < TESTS; j++) {
+			while (gate != 1)
+				cpu_spinwait();
+			gate = 2;
+			barrier();
+
+			tsc = rdtsc();
+
+			barrier();
+			while (gate != 3)
+				cpu_spinwait();
+			gate = 4;
+
+			skew[i][j] = thread_tsc - tsc;
+		}
+
+		error = pthread_join(child, NULL);
+		if (error)
+			errc(1, error, "pthread_join");
+	}
+
+	/*
+	 * Compute average skew for each CPU and output a summary of
+	 * the results.
+	 */
+	aveskew = calloc(ncpu, sizeof(*aveskew));
+	minskew = calloc(ncpu, sizeof(*minskew));
+	maxskew = calloc(ncpu, sizeof(*maxskew));
+	stddev = calloc(ncpu, sizeof(*stddev));
+	stddev[0] = 0.0;
+	for (i = 1; i < ncpu; i++) {
+		sumsq = 0;
+		minskew[i] = maxskew[i] = skew[i][0];
+		for (j = 0; j < TESTS; j++) {
+			aveskew[i] += skew[i][j];
+			if (skew[i][j] < minskew[i])
+				minskew[i] = skew[i][j];
+			if (skew[i][j] > maxskew[i])
+				maxskew[i] = skew[i][j];
+			sumsq += (skew[i][j] * skew[i][j]);
+		}
+		aveskew[i] /= TESTS;
+		sumsq /= TESTS;
+		sumsq -= aveskew[i] * aveskew[i];
+		stddev[i] = sqrt(sumsq);
+	}
+
+	printf("CPU | TSC skew (min/avg/max/stddev)\n");
+	printf("----+------------------------------\n");
+	for (i = 0; i < ncpu; i++)
+		printf("%3d | %5jd %5jd %5jd   %6.3f\n", cpus[i],
+		    (intmax_t)minskew[i], (intmax_t)aveskew[i],
+		    (intmax_t)maxskew[i], stddev[i]);
+	return (0);
+}