X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=gettime.c;h=6ced2f1d7f3fdebf17a834dc9c3cec6c1b2d2e33;hp=8bc84139a92d75abddb4eaa5b1c1c503450382a1;hb=d6bb626ef37d3905221ade2887b422717a07af09;hpb=37e2002175a9585e375ba65a5ffe9a5d49b7bb14

diff --git a/gettime.c b/gettime.c
index 8bc84139..6ced2f1d 100644
--- a/gettime.c
+++ b/gettime.c
@@ -13,12 +13,17 @@
 #include "hash.h"
 #include "os/os.h"
 
-#if defined(ARCH_HAVE_CPU_CLOCK) && !defined(ARCH_CPU_CLOCK_CYCLES_PER_USEC)
+#if defined(ARCH_HAVE_CPU_CLOCK)
+#ifndef ARCH_CPU_CLOCK_CYCLES_PER_USEC
 static unsigned long cycles_per_usec;
-static unsigned long inv_cycles_per_usec;
+static unsigned long inv_cycles_per_nsec;
 static uint64_t max_cycles_for_mult;
+#define NSEC_INV_FACTOR 4096
+#endif
+#ifdef ARCH_CPU_CLOCK_WRAPS
 static unsigned long long cycles_start, cycles_wrap;
 #endif
+#endif
 int tsc_reliable = 0;
 
 struct tv_valid {
@@ -129,7 +134,9 @@ static void fio_init gtod_init(void)
 #ifdef CONFIG_CLOCK_GETTIME
 static int fill_clock_gettime(struct timespec *ts)
 {
-#ifdef CONFIG_CLOCK_MONOTONIC
+#if defined(CONFIG_CLOCK_MONOTONIC_RAW)
+	return clock_gettime(CLOCK_MONOTONIC_RAW, ts);
+#elif defined(CONFIG_CLOCK_MONOTONIC)
 	return clock_gettime(CLOCK_MONOTONIC, ts);
 #else
 	return clock_gettime(CLOCK_REALTIME, ts);
@@ -137,31 +144,31 @@ static int fill_clock_gettime(struct timespec *ts)
 }
 #endif
 
-static void __fio_gettime(struct timeval *tp)
+static void __fio_gettime(struct timespec *tp)
 {
 	switch (fio_clock_source) {
 #ifdef CONFIG_GETTIMEOFDAY
-	case CS_GTOD:
-		gettimeofday(tp, NULL);
+	case CS_GTOD: {
+		struct timeval tv;
+		gettimeofday(&tv, NULL);
+
+		tp->tv_sec = tv.tv_sec;
+		tp->tv_nsec = tv.tv_usec * 1000;
 		break;
+		}
 #endif
 #ifdef CONFIG_CLOCK_GETTIME
 	case CS_CGETTIME: {
-		struct timespec ts;
-
-		if (fill_clock_gettime(&ts) < 0) {
+		if (fill_clock_gettime(tp) < 0) {
 			log_err("fio: clock_gettime fails\n");
 			assert(0);
 		}
-
-		tp->tv_sec = ts.tv_sec;
-		tp->tv_usec = ts.tv_nsec / 1000;
 		break;
 		}
 #endif
 #ifdef ARCH_HAVE_CPU_CLOCK
 	case CS_CPUCLOCK: {
-		uint64_t usecs, t;
+		uint64_t nsecs, t;
 		struct tv_valid *tv;
 
 #ifdef CONFIG_TLS_THREAD
@@ -171,6 +178,7 @@ static void __fio_gettime(struct timeval *tp)
 #endif
 
 		t = get_cpu_clock();
+#ifdef ARCH_CPU_CLOCK_WRAPS
 		if (t < cycles_start && !cycles_wrap)
 			cycles_wrap = 1;
 		else if (cycles_wrap && t >= cycles_start && !tv->warned) {
@@ -179,18 +187,19 @@ static void __fio_gettime(struct timeval *tp)
 		}
 
 		t -= cycles_start;
+#endif
 		tv->last_cycles = t;
 		tv->last_tv_valid = 1;
 #ifdef ARCH_CPU_CLOCK_CYCLES_PER_USEC
-		usecs = t / ARCH_CPU_CLOCK_CYCLES_PER_USEC;
+		nsecs = t * 1000 / ARCH_CPU_CLOCK_CYCLES_PER_USEC;
 #else
 		if (t < max_cycles_for_mult)
-			usecs = (t * inv_cycles_per_usec) / 16777216UL;
+			nsecs = (t * inv_cycles_per_nsec) / NSEC_INV_FACTOR;
 		else
-			usecs = t / cycles_per_usec;
+			nsecs = (t / NSEC_INV_FACTOR) * inv_cycles_per_nsec;
 #endif
-		tp->tv_sec = usecs / 1000000;
-		tp->tv_usec = usecs % 1000000;
+		tp->tv_sec = nsecs / 1000000000ULL;
+		tp->tv_nsec = nsecs % 1000000000ULL;
 		break;
 		}
 #endif
@@ -201,9 +210,9 @@ static void __fio_gettime(struct timeval *tp)
 }
 
 #ifdef FIO_DEBUG_TIME
-void fio_gettime(struct timeval *tp, void *caller)
+void fio_gettime(struct timespec *tp, void *caller)
 #else
-void fio_gettime(struct timeval *tp, void fio_unused *caller)
+void fio_gettime(struct timespec *tp, void fio_unused *caller)
 #endif
 {
 #ifdef FIO_DEBUG_TIME
@@ -221,9 +230,10 @@ void fio_gettime(struct timeval *tp, void fio_unused *caller)
 #if defined(ARCH_HAVE_CPU_CLOCK) && !defined(ARCH_CPU_CLOCK_CYCLES_PER_USEC)
 static unsigned long get_cycles_per_usec(void)
 {
-	struct timeval s, e;
+	struct timespec s, e;
 	uint64_t c_s, c_e;
 	enum fio_cs old_cs = fio_clock_source;
+	uint64_t elapsed;
 
 #ifdef CONFIG_CLOCK_GETTIME
 	fio_clock_source = CS_CGETTIME;
@@ -234,8 +244,6 @@ static unsigned long get_cycles_per_usec(void)
 
 	c_s = get_cpu_clock();
 	do {
-		uint64_t elapsed;
-
 		__fio_gettime(&e);
 
 		elapsed = utime_since(&s, &e);
@@ -246,7 +254,7 @@ static unsigned long get_cycles_per_usec(void)
 	} while (1);
 
 	fio_clock_source = old_cs;
-	return (c_e - c_s + 127) >> 7;
+	return (c_e - c_s) / elapsed;
 }
 
 #define NR_TIME_ITERS	50
@@ -292,27 +300,24 @@ static int calibrate_cpu_clock(void)
 	}
 
 	S /= (double) NR_TIME_ITERS;
-	mean /= 10.0;
 
 	for (i = 0; i < NR_TIME_ITERS; i++)
-		dprint(FD_TIME, "cycles[%d]=%llu\n", i,
-					(unsigned long long) cycles[i] / 10);
+		dprint(FD_TIME, "cycles[%d]=%llu\n", i, (unsigned long long) cycles[i]);
 
 	avg /= samples;
-	avg = (avg + 5) / 10;
-	minc /= 10;
-	maxc /= 10;
 	dprint(FD_TIME, "avg: %llu\n", (unsigned long long) avg);
 	dprint(FD_TIME, "min=%llu, max=%llu, mean=%f, S=%f\n",
 			(unsigned long long) minc,
 			(unsigned long long) maxc, mean, S);
 
 	cycles_per_usec = avg;
-	inv_cycles_per_usec = 16777216UL / cycles_per_usec;
-	max_cycles_for_mult = ~0ULL / inv_cycles_per_usec;
-	dprint(FD_TIME, "inv_cycles_per_usec=%lu\n", inv_cycles_per_usec);
+	inv_cycles_per_nsec = NSEC_INV_FACTOR * 1000 / cycles_per_usec;
+	max_cycles_for_mult = ~0ULL / inv_cycles_per_nsec;
+	dprint(FD_TIME, "inv_cycles_per_nsec=%lu\n", inv_cycles_per_nsec);
+#ifdef ARCH_CPU_CLOCK_WRAPS
 	cycles_start = get_cpu_clock();
 	dprint(FD_TIME, "cycles_start=%llu\n", cycles_start);
+#endif
 	return 0;
 }
 #else
@@ -375,13 +380,32 @@ void fio_clock_init(void)
 		log_info("fio: clocksource=cpu may not be reliable\n");
 }
 
-uint64_t utime_since(const struct timeval *s, const struct timeval *e)
+uint64_t ntime_since(const struct timespec *s, const struct timespec *e)
 {
-	long sec, usec;
-	uint64_t ret;
+       int64_t sec, nsec;
+
+       sec = e->tv_sec - s->tv_sec;
+       nsec = e->tv_nsec - s->tv_nsec;
+       if (sec > 0 && nsec < 0) {
+               sec--;
+               nsec += 1000000000LL;
+       }
+
+       /*
+        * time warp bug on some kernels?
+        */
+       if (sec < 0 || (sec == 0 && nsec < 0))
+               return 0;
+
+       return nsec + (sec * 1000000000LL);
+}
+
+uint64_t utime_since(const struct timespec *s, const struct timespec *e)
+{
+	int64_t sec, usec;
 
 	sec = e->tv_sec - s->tv_sec;
-	usec = e->tv_usec - s->tv_usec;
+	usec = (e->tv_nsec - s->tv_nsec) / 1000;
 	if (sec > 0 && usec < 0) {
 		sec--;
 		usec += 1000000;
@@ -393,25 +417,29 @@ uint64_t utime_since(const struct timeval *s, const struct timeval *e)
 	if (sec < 0 || (sec == 0 && usec < 0))
 		return 0;
 
-	ret = sec * 1000000ULL + usec;
-
-	return ret;
+	return usec + (sec * 1000000);
 }
 
-uint64_t utime_since_now(const struct timeval *s)
+uint64_t utime_since_now(const struct timespec *s)
 {
-	struct timeval t;
+	struct timespec t;
+#ifdef FIO_DEBUG_TIME
+	void *p = __builtin_return_address(0);
 
+	fio_gettime(&t, p);
+#else
 	fio_gettime(&t, NULL);
+#endif
+
 	return utime_since(s, &t);
 }
 
-uint64_t mtime_since(const struct timeval *s, const struct timeval *e)
+uint64_t mtime_since_tv(const struct timeval *s, const struct timeval *e)
 {
-	long sec, usec, ret;
+	int64_t sec, usec;
 
 	sec = e->tv_sec - s->tv_sec;
-	usec = e->tv_usec - s->tv_usec;
+	usec = (e->tv_usec - s->tv_usec);
 	if (sec > 0 && usec < 0) {
 		sec--;
 		usec += 1000000;
@@ -420,23 +448,45 @@ uint64_t mtime_since(const struct timeval *s, const struct timeval *e)
 	if (sec < 0 || (sec == 0 && usec < 0))
 		return 0;
 
-	sec *= 1000UL;
-	usec /= 1000UL;
-	ret = sec + usec;
-
-	return ret;
+	sec *= 1000;
+	usec /= 1000;
+	return sec + usec;
 }
 
-uint64_t mtime_since_now(const struct timeval *s)
+uint64_t mtime_since_now(const struct timespec *s)
 {
-	struct timeval t;
+	struct timespec t;
+#ifdef FIO_DEBUG_TIME
 	void *p = __builtin_return_address(0);
 
 	fio_gettime(&t, p);
+#else
+	fio_gettime(&t, NULL);
+#endif
+
 	return mtime_since(s, &t);
 }
 
-uint64_t time_since_now(const struct timeval *s)
+uint64_t mtime_since(const struct timespec *s, const struct timespec *e)
+{
+	int64_t sec, usec;
+
+	sec = e->tv_sec - s->tv_sec;
+	usec = (e->tv_nsec - s->tv_nsec) / 1000;
+	if (sec > 0 && usec < 0) {
+		sec--;
+		usec += 1000000;
+	}
+
+	if (sec < 0 || (sec == 0 && usec < 0))
+		return 0;
+
+	sec *= 1000;
+	usec /= 1000;
+	return sec + usec;
+}
+
+uint64_t time_since_now(const struct timespec *s)
 {
 	return mtime_since_now(s) / 1000;
 }
@@ -475,6 +525,7 @@ static void *clock_thread_fn(void *data)
 	struct clock_entry *c;
 	os_cpu_mask_t cpu_mask;
 	uint32_t last_seq;
+	unsigned long long first;
 	int i;
 
 	if (fio_cpuset_init(&cpu_mask)) {
@@ -496,6 +547,7 @@ static void *clock_thread_fn(void *data)
 	pthread_mutex_lock(&t->lock);
 	pthread_mutex_unlock(&t->started);
 
+	first = get_cpu_clock();
 	last_seq = 0;
 	c = &t->entries[0];
 	for (i = 0; i < t->nr_entries; i++, c++) {
@@ -518,7 +570,8 @@ static void *clock_thread_fn(void *data)
 		unsigned long long clocks;
 
 		clocks = t->entries[i - 1].tsc - t->entries[0].tsc;
-		log_info("cs: cpu%3d: %llu clocks seen\n", t->cpu, clocks);
+		log_info("cs: cpu%3d: %llu clocks seen, first %llu\n", t->cpu,
+							clocks, first);
 	}
 
 	/*
@@ -528,6 +581,7 @@ static void *clock_thread_fn(void *data)
 	if (!t->entries[i - 1].tsc && !t->entries[0].tsc)
 		goto err;
 
+	fio_cpuset_exit(&cpu_mask);
 	return NULL;
 err:
 	fio_cpuset_exit(&cpu_mask);
@@ -629,6 +683,8 @@ int fio_monotonic_clocktest(int debug)
 
 	qsort(entries, tentries, sizeof(struct clock_entry), clock_cmp);
 
+	/* silence silly gcc */
+	prev = NULL;
 	for (failed = i = 0; i < tentries; i++) {
 		this = &entries[i];