fio: fix s390 time accounting

[fio.git] / gettime.c

diff --git a/gettime.c b/gettime.c
index 352c1d3d7611be0277d3e038d737c1465f9ec820..c6d45f8761f799e9e324b2f95f9b4bff40a56158 100644 (file)
--- 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
@@ -25,14 +25,14 @@ struct tv_valid {
        int last_tv_valid;
 };
 #ifdef CONFIG_TLS_THREAD
-static struct tv_valid __thread static_tv_valid;
+static __thread struct tv_valid static_tv_valid;
 #else
 static pthread_key_t tv_tls_key;
 #endif
 
 enum fio_cs fio_clock_source = FIO_PREFERRED_CLOCK_SOURCE;
 int fio_clock_source_set = 0;
-enum fio_cs fio_clock_source_inited = CS_INVAL;
+static enum fio_cs fio_clock_source_inited = CS_INVAL;
 
 #ifdef FIO_DEBUG_TIME
 
@@ -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;
@@ -205,7 +209,7 @@ void fio_gettime(struct timeval *tp, void fio_unused *caller)
 
        gtod_log_caller(caller);
 #endif
-       if (fio_tv) {
+       if (fio_unlikely(fio_tv)) {
                memcpy(tp, fio_tv, sizeof(*tp));
                return;
        }
@@ -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;
@@ -318,16 +322,20 @@ 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)
 {
        struct tv_valid *t;
 
-       t = calloc(sizeof(*t), 1);
+       t = calloc(1, sizeof(*t));
        if (pthread_setspecific(tv_tls_key, t))
                log_err("fio: can't set TLS key\n");
 }
@@ -441,9 +449,9 @@ uint64_t time_since_now(struct timeval *s)
 #define CLOCK_ENTRIES  100000
 
 struct clock_entry {
-       uint64_t seq;
+       uint32_t seq;
+       uint32_t cpu;
        uint64_t tsc;
-       uint64_t cpu;
 };
 
 struct clock_thread {
@@ -451,11 +459,11 @@ struct clock_thread {
        int cpu;
        pthread_mutex_t lock;
        pthread_mutex_t started;
-       uint64_t *seq;
+       uint32_t *seq;
        struct clock_entry *entries;
 };
 
-static inline uint64_t atomic64_inc_return(uint64_t *seq)
+static inline uint32_t atomic32_inc_return(uint32_t *seq)
 {
        return 1 + __sync_fetch_and_add(seq, 1);
 }
@@ -465,6 +473,7 @@ static void *clock_thread_fn(void *data)
        struct clock_thread *t = data;
        struct clock_entry *c;
        os_cpu_mask_t cpu_mask;
+       uint32_t last_seq;
        int i;
 
        memset(&cpu_mask, 0, sizeof(cpu_mask));
@@ -478,13 +487,17 @@ static void *clock_thread_fn(void *data)
        pthread_mutex_lock(&t->lock);
        pthread_mutex_unlock(&t->started);
 
+       last_seq = 0;
        c = &t->entries[0];
        for (i = 0; i < CLOCK_ENTRIES; i++, c++) {
-               uint64_t seq, tsc;
+               uint32_t seq;
+               uint64_t tsc;
 
                c->cpu = t->cpu;
                do {
-                       seq = atomic64_inc_return(t->seq);
+                       seq = atomic32_inc_return(t->seq);
+                       if (seq < last_seq)
+                               break;
                        tsc = get_cpu_clock();
                } while (seq != *t->seq);
 
@@ -492,12 +505,14 @@ static void *clock_thread_fn(void *data)
                c->tsc = tsc;
        }
 
-       log_info("cs: cpu%3d: %lu clocks seen\n", t->cpu, t->entries[CLOCK_ENTRIES - 1].tsc - t->entries[0].tsc);
+       log_info("cs: cpu%3d: %llu clocks seen\n", t->cpu,
+               (unsigned long long) t->entries[i - 1].tsc - t->entries[0].tsc);
+
        /*
         * The most common platform clock breakage is returning zero
         * indefinitely. Check for that and return failure.
         */
-       if (!t->entries[CLOCK_ENTRIES - 1].tsc && !t->entries[0].tsc)
+       if (!t->entries[i - 1].tsc && !t->entries[0].tsc)
                return (void *) 1;
 
        return NULL;
@@ -521,7 +536,7 @@ int fio_monotonic_clocktest(void)
        struct clock_entry *entries;
        unsigned long tentries, failed;
        struct clock_entry *prev, *this;
-       uint64_t seq = 0;
+       uint32_t seq = 0;
        int i;
 
        log_info("cs: reliable_tsc: %s\n", tsc_reliable ? "yes" : "no");
@@ -571,7 +586,7 @@ int fio_monotonic_clocktest(void)
        free(threads);
 
        if (failed) {
-               log_err("Clocksource test: %u threads failed\n", failed);
+               log_err("Clocksource test: %lu threads failed\n", failed);
                goto err;
        }
 
@@ -588,9 +603,10 @@ int fio_monotonic_clocktest(void)
                if (prev->tsc > this->tsc) {
                        uint64_t diff = prev->tsc - this->tsc;
 
-                       log_info("cs: CPU clock mismatch (diff=%lu):\n", diff);
-                       log_info("\t CPU%3lu: TSC=%lu, SEQ=%lu\n", prev->cpu, prev->tsc, prev->seq);
-                       log_info("\t CPU%3lu: TSC=%lu, SEQ=%lu\n", this->cpu, this->tsc, this->seq);
+                       log_info("cs: CPU clock mismatch (diff=%llu):\n",
+                                               (unsigned long long) diff);
+                       log_info("\t CPU%3u: TSC=%llu, SEQ=%u\n", prev->cpu, (unsigned long long) prev->tsc, prev->seq);
+                       log_info("\t CPU%3u: TSC=%llu, SEQ=%u\n", this->cpu, (unsigned long long) this->tsc, this->seq);
                        failed++;
                }