From fd11d7afaecf030c6c9a4a9c6953037a2ff18353 Mon Sep 17 00:00:00 2001
From: Jens Axboe <axboe@suse.de>
Date: Tue, 8 Nov 2005 12:57:07 +0100
Subject: [PATCH] [PATCH] fio: improve timing on low number of usecs

---
 arch.h | 63 +++++++++++++++++++++++++++++++++++++++++++++++++++
 fio.c  | 71 +++++++++++++++++++++-------------------------------------
 2 files changed, 89 insertions(+), 45 deletions(-)
 create mode 100644 arch.h

diff --git a/arch.h b/arch.h
new file mode 100644
index 0000000..08f8940
--- /dev/null
+++ b/arch.h
@@ -0,0 +1,63 @@
+#ifndef ARCH_H
+#define ARCH_H
+
+/*
+ * assume we don't have _get either, if _set isn't defined
+ */
+#ifndef __NR_ioprio_set
+#if defined(__i386__)
+#define __NR_ioprio_set		289
+#define __NR_ioprio_get		290
+#elif defined(__powerpc__) || defined(__powerpc64__)
+#define __NR_ioprio_set		273
+#define __NR_ioprio_get		274
+#elif defined(__x86_64__)
+#define __NR_ioprio_set		251
+#define __NR_ioprio_get		252
+#elif defined(__ia64__)
+#define __NR_ioprio_set		1274
+#define __NR_ioprio_get		1275
+#elif defined(__alpha__)
+#define __NR_ioprio_set		442
+#define __NR_ioprio_get		443
+#elif defined(__s390x__) || defined(__s390__)
+#define __NR_ioprio_set		282
+#define __NR_ioprio_get		283
+#else
+#error "Unsupported arch"
+#endif
+#endif
+
+#ifndef __NR_fadvise64
+#if defined(__i386__)
+#define __NR_fadvise64		250
+#elif defined(__powerpc__) || defined(__powerpc64__)
+#define __NR_fadvise64		233
+#elif defined(__x86_64__)
+#define __NR_fadvise64		221
+#elif defined(__ia64__)
+#define __NR_fadvise64		1234
+#elif defined(__alpha__)
+#define __NR_fadvise64		413
+#elif defined(__s390x__) || defined(__s390__)
+#define __NR_fadvise64		253
+#else
+#error "Unsupported arch"
+#endif
+#endif
+
+#if defined(__i386__) || defined(__x86_64)
+#define nop	__asm__ __volatile__("rep;nop": : :"memory")
+#elif defined defined(__powerpc__) || defined(__powerpc64__)
+#define nop	do { } while (0)
+#elif defined(__ia64__)
+#define nop	asm volatile ("hint @pause" ::: "memory");
+#elif defined(__alpha__)
+#define nop	do { } while (0)
+#elif defined(__s390x__) || defined(__s390__)
+#define nop	asm volatile ("diag 0,0,68" : : : "memory")
+#else
+#error "Unsupported arch"
+#endif
+
+#endif
diff --git a/fio.c b/fio.c
index 0e4f0f0..ef2459a 100644
--- a/fio.c
+++ b/fio.c
@@ -44,6 +44,7 @@
 #include <asm/types.h>
 #include <asm/bitops.h>
 
+#include "arch.h"
 #include "list.h"
 #include "md5.h"
 
@@ -53,51 +54,6 @@
 
 #define MAX_JOBS	(1024)
 
-/*
- * assume we don't have _get either, if _set isn't defined
- */
-#ifndef __NR_ioprio_set
-#if defined(__i386__)
-#define __NR_ioprio_set		289
-#define __NR_ioprio_get		290
-#elif defined(__powerpc__) || defined(__powerpc64__)
-#define __NR_ioprio_set		273
-#define __NR_ioprio_get		274
-#elif defined(__x86_64__)
-#define __NR_ioprio_set		251
-#define __NR_ioprio_get		252
-#elif defined(__ia64__)
-#define __NR_ioprio_set		1274
-#define __NR_ioprio_get		1275
-#elif defined(__alpha__)
-#define __NR_ioprio_set		442
-#define __NR_ioprio_get		443
-#elif defined(__s390x__) || defined(__s390__)
-#define __NR_ioprio_set		282
-#define __NR_ioprio_get		283
-#else
-#error "Unsupported arch"
-#endif
-#endif
-
-#ifndef __NR_fadvise64
-#if defined(__i386__)
-#define __NR_fadvise64		250
-#elif defined(__powerpc__) || defined(__powerpc64__)
-#define __NR_fadvise64		233
-#elif defined(__x86_64__)
-#define __NR_fadvise64		221
-#elif defined(__ia64__)
-#define __NR_fadvise64		1234
-#elif defined(__alpha__)
-#define __NR_fadvise64		413
-#elif defined(__s390x__) || defined(__s390__)
-#define __NR_fadvise64		253
-#else
-#error "Unsupported arch"
-#endif
-#endif
-
 static int ioprio_set(int which, int who, int ioprio)
 {
 	return syscall(__NR_ioprio_set, which, who, ioprio);
@@ -402,6 +358,14 @@ static unsigned long utime_since(struct timeval *s, struct timeval *e)
 	return sec + usec;
 }
 
+static unsigned long utime_since_now(struct timeval *s)
+{
+	struct timeval t;
+
+	gettimeofday(&t, NULL);
+	return utime_since(s, &t);
+}
+
 static unsigned long mtime_since(struct timeval *s, struct timeval *e)
 {
 	double sec, usec;
@@ -588,18 +552,35 @@ static void add_bw_sample(struct thread_data *td)
 	td->stat_io_bytes = td->this_io_bytes;
 }
 
+/*
+ * busy looping version for the last few usec
+ */
+static void __usec_sleep(int usec)
+{
+	struct timeval start;
+
+	gettimeofday(&start, NULL);
+	while (utime_since_now(&start) < usec)
+		 __asm__ __volatile__("rep;nop": : :"memory");
+}
+
 static void usec_sleep(int usec)
 {
 	struct timespec req = { .tv_sec = 0, .tv_nsec = usec * 1000 };
 	struct timespec rem;
 
 	do {
+		if (usec < 5000) {
+			__usec_sleep(usec);
+			break;
+		}
 		rem.tv_sec = rem.tv_nsec = 0;
 		nanosleep(&req, &rem);
 		if (!rem.tv_nsec)
 			break;
 
 		req.tv_nsec = rem.tv_nsec;
+		usec = rem.tv_nsec * 1000;
 	} while (1);
 }
 
-- 
2.25.1