From: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
Date: Tue, 8 Apr 2014 15:52:00 +0000 (+0200)
Subject: fio: fix s390 time accounting
X-Git-Tag: fio-2.1.8~7^2~1
X-Git-Url: https://git.kernel.dk/?p=fio.git;a=commitdiff_plain;h=919e789d0999bf13ef1ae670c536d2aef3709894

fio: fix s390 time accounting

The current timer implementation could cause time warps on s390 which ends up
as time bound jobs that would never end, because they always reset themself to
the old time.

When touching this code anyway, we also change it to use the faster stckf
and avoid the calibration as we can control the result to be usecs.
This also eliminates a few calculations cycle->usec in the hot path for the
timer.

In case other architectures have similar improved timers that might not be
usec based, but nsec based or such a thing any architecture can set
ARCH_CPU_CLOCK_CYCLES_PER_USEC to an appropriate per-arch value.
This leaves the infrastructure open for others and the compiler will throw
our division by 1 away anyway.

Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
---

diff --git a/arch/arch-s390.h b/arch/arch-s390.h
index bcd91636..2c205de3 100644
--- a/arch/arch-s390.h
+++ b/arch/arch-s390.h
@@ -22,14 +22,21 @@
 #define read_barrier()	asm volatile("bcr 15,0" : : : "memory")
 #define write_barrier()	asm volatile("bcr 15,0" : : : "memory")
 
+/*
+ * Fio needs monotonic (never lower), but not strict monotonic (never the same)
+ * so store clock fast is enough
+ */
 static inline unsigned long long get_cpu_clock(void)
 {
 	unsigned long long clk;
 
-	__asm__ __volatile__("stck %0" : "=Q" (clk) : : "cc");
-	return clk;
+	__asm__ __volatile__("stckf %0" : "=Q" (clk) : : "cc");
+	return clk>>12;
 }
 
+#define ARCH_CPU_CLOCK_CYCLES_PER_USEC 1
+#define ARCH_HAVE_CPU_CLOCK
+
 #define ARCH_HAVE_INIT
 extern int tsc_reliable;
 static inline int arch_init(char *envp[])
@@ -38,6 +45,4 @@ static inline int arch_init(char *envp[])
 	return 0;
 }
 
-#define ARCH_HAVE_CPU_CLOCK
-
 #endif
diff --git a/gettime.c b/gettime.c
index b89cd46f..c6d45f87 100644
--- a/gettime.c
+++ b/gettime.c
@@ -13,7 +13,7 @@
 #include "hash.h"
 #include "os/os.h"
 
-#ifdef ARCH_HAVE_CPU_CLOCK
+#if defined(ARCH_HAVE_CPU_CLOCK) && !defined(ARCH_CPU_CLOCK_CYCLES_PER_USEC)
 static unsigned long cycles_per_usec;
 static unsigned long inv_cycles_per_usec;
 #endif
@@ -177,7 +177,11 @@ static void *__fio_gettime(struct timeval *tp)
 		} else if (tv)
 			tv->last_cycles = t;
 
+#ifdef ARCH_CPU_CLOCK_CYCLES_PER_USEC
+		usecs = t / ARCH_CPU_CLOCK_CYCLES_PER_USEC;
+#else
 		usecs = (t * inv_cycles_per_usec) / 16777216UL;
+#endif
 		tp->tv_sec = usecs / 1000000;
 		tp->tv_usec = usecs % 1000000;
 		break;
@@ -229,7 +233,7 @@ void fio_gettime(struct timeval *tp, void fio_unused *caller)
 	}
 }
 
-#ifdef ARCH_HAVE_CPU_CLOCK
+#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;
@@ -318,9 +322,13 @@ static int calibrate_cpu_clock(void)
 #else
 static int calibrate_cpu_clock(void)
 {
+#ifdef ARCH_CPU_CLOCK_CYCLES_PER_USEC
+	return 0;
+#else
 	return 1;
-}
 #endif
+}
+#endif // ARCH_HAVE_CPU_CLOCK
 
 #ifndef CONFIG_TLS_THREAD
 void fio_local_clock_init(int is_thread)