stat: add a new function to allocate a clat_prio_stat array
[fio.git] / stat.c
diff --git a/stat.c b/stat.c
index eb40bd7f44bb7c4c40c84fc5c70fa9648f996772..c8680a679bf3667bf62c46967c5d44a53a01697d 100644 (file)
--- a/stat.c
+++ b/stat.c
 #include "zbd.h"
 #include "oslib/asprintf.h"
 
+#ifdef WIN32
+#define LOG_MSEC_SLACK 2
+#else
 #define LOG_MSEC_SLACK 1
+#endif
 
 struct fio_sem *stat_sem;
 
@@ -211,7 +215,7 @@ static void show_clat_percentiles(uint64_t *io_u_plat, unsigned long long nr,
 
        len = calc_clat_percentiles(io_u_plat, nr, plist, &ovals, &maxv, &minv);
        if (!len || !ovals)
-               goto out;
+               return;
 
        /*
         * We default to nsecs, but if the value range is such that we
@@ -258,7 +262,6 @@ static void show_clat_percentiles(uint64_t *io_u_plat, unsigned long long nr,
                        log_buf(out, "\n");
        }
 
-out:
        free(ovals);
 }
 
@@ -282,6 +285,47 @@ bool calc_lat(struct io_stat *is, unsigned long long *min,
        return true;
 }
 
+void show_mixed_group_stats(struct group_run_stats *rs, struct buf_output *out) 
+{
+       char *io, *agg, *min, *max;
+       char *ioalt, *aggalt, *minalt, *maxalt;
+       uint64_t io_mix = 0, agg_mix = 0, min_mix = -1, max_mix = 0;
+       uint64_t min_run = -1, max_run = 0;
+       const int i2p = is_power_of_2(rs->kb_base);
+       int i;
+
+       for (i = 0; i < DDIR_RWDIR_CNT; i++) {
+               if (!rs->max_run[i])
+                       continue;
+               io_mix += rs->iobytes[i];
+               agg_mix += rs->agg[i];
+               min_mix = min_mix < rs->min_bw[i] ? min_mix : rs->min_bw[i];
+               max_mix = max_mix > rs->max_bw[i] ? max_mix : rs->max_bw[i];
+               min_run = min_run < rs->min_run[i] ? min_run : rs->min_run[i];
+               max_run = max_run > rs->max_run[i] ? max_run : rs->max_run[i];
+       }
+       io = num2str(io_mix, rs->sig_figs, 1, i2p, N2S_BYTE);
+       ioalt = num2str(io_mix, rs->sig_figs, 1, !i2p, N2S_BYTE);
+       agg = num2str(agg_mix, rs->sig_figs, 1, i2p, rs->unit_base);
+       aggalt = num2str(agg_mix, rs->sig_figs, 1, !i2p, rs->unit_base);
+       min = num2str(min_mix, rs->sig_figs, 1, i2p, rs->unit_base);
+       minalt = num2str(min_mix, rs->sig_figs, 1, !i2p, rs->unit_base);
+       max = num2str(max_mix, rs->sig_figs, 1, i2p, rs->unit_base);
+       maxalt = num2str(max_mix, rs->sig_figs, 1, !i2p, rs->unit_base);
+       log_buf(out, "  MIXED: bw=%s (%s), %s-%s (%s-%s), io=%s (%s), run=%llu-%llumsec\n",
+                       agg, aggalt, min, max, minalt, maxalt, io, ioalt,
+                       (unsigned long long) min_run,
+                       (unsigned long long) max_run);
+       free(io);
+       free(agg);
+       free(min);
+       free(max);
+       free(ioalt);
+       free(aggalt);
+       free(minalt);
+       free(maxalt);
+}
+
 void show_group_stats(struct group_run_stats *rs, struct buf_output *out)
 {
        char *io, *agg, *min, *max;
@@ -306,7 +350,7 @@ void show_group_stats(struct group_run_stats *rs, struct buf_output *out)
                max = num2str(rs->max_bw[i], rs->sig_figs, 1, i2p, rs->unit_base);
                maxalt = num2str(rs->max_bw[i], rs->sig_figs, 1, !i2p, rs->unit_base);
                log_buf(out, "%s: bw=%s (%s), %s-%s (%s-%s), io=%s (%s), run=%llu-%llumsec\n",
-                               rs->unified_rw_rep ? "  MIXED" : str[i],
+                               (rs->unified_rw_rep == UNIFIED_MIXED) ? "  MIXED" : str[i],
                                agg, aggalt, min, max, minalt, maxalt, io, ioalt,
                                (unsigned long long) rs->min_run[i],
                                (unsigned long long) rs->max_run[i]);
@@ -320,6 +364,10 @@ void show_group_stats(struct group_run_stats *rs, struct buf_output *out)
                free(minalt);
                free(maxalt);
        }
+
+       /* Need to aggregate statisitics to show mixed values */
+       if (rs->unified_rw_rep == UNIFIED_BOTH)
+               show_mixed_group_stats(rs, out);
 }
 
 void stat_calc_dist(uint64_t *map, unsigned long total, double *io_u_dist)
@@ -414,11 +462,40 @@ static void display_lat(const char *name, unsigned long long min,
        free(maxp);
 }
 
+static struct thread_stat *gen_mixed_ddir_stats_from_ts(struct thread_stat *ts)
+{
+       struct thread_stat *ts_lcl;
+
+       /*
+        * Handle aggregation of Reads (ddir = 0), Writes (ddir = 1), and
+        * Trims (ddir = 2)
+        */
+       ts_lcl = malloc(sizeof(struct thread_stat));
+       if (!ts_lcl) {
+               log_err("fio: failed to allocate local thread stat\n");
+               return NULL;
+       }
+
+       init_thread_stat(ts_lcl);
+
+       /* calculate mixed stats  */
+       ts_lcl->unified_rw_rep = UNIFIED_MIXED;
+       ts_lcl->lat_percentiles = ts->lat_percentiles;
+       ts_lcl->clat_percentiles = ts->clat_percentiles;
+       ts_lcl->slat_percentiles = ts->slat_percentiles;
+       ts_lcl->percentile_precision = ts->percentile_precision;
+       memcpy(ts_lcl->percentile_list, ts->percentile_list, sizeof(ts->percentile_list));
+
+       sum_thread_stats(ts_lcl, ts);
+
+       return ts_lcl;
+}
+
 static double convert_agg_kbytes_percent(struct group_run_stats *rs, int ddir, int mean)
 {
        double p_of_agg = 100.0;
        if (rs && rs->agg[ddir] > 1024) {
-               p_of_agg = mean * 100 / (double) (rs->agg[ddir] / 1024.0);
+               p_of_agg = mean * 100.0 / (double) (rs->agg[ddir] / 1024.0);
 
                if (p_of_agg > 100.0)
                        p_of_agg = 100.0;
@@ -477,7 +554,7 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
        }
 
        log_buf(out, "  %s: IOPS=%s, BW=%s (%s)(%s/%llumsec)%s\n",
-                       rs->unified_rw_rep ? "mixed" : io_ddir_name(ddir),
+                       (ts->unified_rw_rep == UNIFIED_MIXED) ? "mixed" : io_ddir_name(ddir),
                        iops_p, bw_p, bw_p_alt, io_p,
                        (unsigned long long) ts->runtime[ddir],
                        post_st ? : "");
@@ -592,6 +669,18 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
        }
 }
 
+static void show_mixed_ddir_status(struct group_run_stats *rs,
+                                  struct thread_stat *ts,
+                                  struct buf_output *out)
+{
+       struct thread_stat *ts_lcl = gen_mixed_ddir_stats_from_ts(ts);
+
+       if (ts_lcl)
+               show_ddir_status(rs, ts_lcl, DDIR_READ, out);
+
+       free(ts_lcl);
+}
+
 static bool show_lat(double *io_u_lat, int nr, const char **ranges,
                     const char *msg, struct buf_output *out)
 {
@@ -1022,9 +1111,8 @@ void show_disk_util(int terse, struct json_object *parent,
        if (!is_running_backend())
                return;
 
-       if (flist_empty(&disk_list)) {
+       if (flist_empty(&disk_list))
                return;
-       }
 
        if ((output_format & FIO_OUTPUT_JSON) && parent)
                do_json = true;
@@ -1034,9 +1122,9 @@ void show_disk_util(int terse, struct json_object *parent,
        if (!terse && !do_json)
                log_buf(out, "\nDisk stats (read/write):\n");
 
-       if (do_json)
+       if (do_json) {
                json_object_add_disk_utils(parent, &disk_list);
-       else if (output_format & ~(FIO_OUTPUT_JSON | FIO_OUTPUT_JSON_PLUS)) {
+       else if (output_format & ~(FIO_OUTPUT_JSON | FIO_OUTPUT_JSON_PLUS)) {
                flist_for_each(entry, &disk_list) {
                        du = flist_entry(entry, struct disk_util, list);
 
@@ -1083,6 +1171,9 @@ static void show_thread_status_normal(struct thread_stat *ts,
                        show_ddir_status(rs, ts, ddir, out);
        }
 
+       if (ts->unified_rw_rep == UNIFIED_BOTH)
+               show_mixed_ddir_status(rs, ts, out);
+
        show_latencies(ts, out);
 
        if (ts->sync_stat.samples)
@@ -1193,18 +1284,19 @@ static void show_ddir_status_terse(struct thread_stat *ts,
        else
                log_buf(out, ";%llu;%llu;%f;%f", 0ULL, 0ULL, 0.0, 0.0);
 
-       if (ts->lat_percentiles)
+       if (ts->lat_percentiles) {
                len = calc_clat_percentiles(ts->io_u_plat[FIO_LAT][ddir],
                                        ts->lat_stat[ddir].samples,
                                        ts->percentile_list, &ovals, &maxv,
                                        &minv);
-       else if (ts->clat_percentiles)
+       } else if (ts->clat_percentiles) {
                len = calc_clat_percentiles(ts->io_u_plat[FIO_CLAT][ddir],
                                        ts->clat_stat[ddir].samples,
                                        ts->percentile_list, &ovals, &maxv,
                                        &minv);
-       else
+       } else {
                len = 0;
+       }
 
        for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
                if (i >= len) {
@@ -1232,8 +1324,9 @@ static void show_ddir_status_terse(struct thread_stat *ts,
                }
 
                log_buf(out, ";%llu;%llu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
-       } else
+       } else {
                log_buf(out, ";%llu;%llu;%f%%;%f;%f", 0ULL, 0ULL, 0.0, 0.0, 0.0);
+       }
 
        if (ver == 5) {
                if (bw_stat)
@@ -1249,8 +1342,22 @@ static void show_ddir_status_terse(struct thread_stat *ts,
        }
 }
 
-static struct json_object *add_ddir_lat_json(struct thread_stat *ts, uint32_t percentiles,
-               struct io_stat *lat_stat, uint64_t *io_u_plat)
+static void show_mixed_ddir_status_terse(struct thread_stat *ts,
+                                  struct group_run_stats *rs,
+                                  int ver, struct buf_output *out)
+{
+       struct thread_stat *ts_lcl = gen_mixed_ddir_stats_from_ts(ts);
+
+       if (ts_lcl)
+               show_ddir_status_terse(ts_lcl, rs, DDIR_READ, ver, out);
+
+       free(ts_lcl);
+}
+
+static struct json_object *add_ddir_lat_json(struct thread_stat *ts,
+                                            uint32_t percentiles,
+                                            struct io_stat *lat_stat,
+                                            uint64_t *io_u_plat)
 {
        char buf[120];
        double mean, dev;
@@ -1310,12 +1417,12 @@ static void add_ddir_status_json(struct thread_stat *ts,
 
        assert(ddir_rw(ddir) || ddir_sync(ddir));
 
-       if (ts->unified_rw_rep && ddir != DDIR_READ)
+       if ((ts->unified_rw_rep == UNIFIED_MIXED) && ddir != DDIR_READ)
                return;
 
        dir_object = json_create_object();
        json_object_add_value_object(parent,
-               ts->unified_rw_rep ? "mixed" : io_ddir_name(ddir), dir_object);
+               (ts->unified_rw_rep == UNIFIED_MIXED) ? "mixed" : io_ddir_name(ddir), dir_object);
 
        if (ddir_rw(ddir)) {
                bw_bytes = 0;
@@ -1418,6 +1525,18 @@ static void add_ddir_status_json(struct thread_stat *ts,
        }
 }
 
+static void add_mixed_ddir_status_json(struct thread_stat *ts,
+               struct group_run_stats *rs, struct json_object *parent)
+{
+       struct thread_stat *ts_lcl = gen_mixed_ddir_stats_from_ts(ts);
+
+       /* add the aggregated stats to json parent */
+       if (ts_lcl)
+               add_ddir_status_json(ts_lcl, rs, DDIR_READ, parent);
+
+       free(ts_lcl);
+}
+
 static void show_thread_status_terse_all(struct thread_stat *ts,
                                         struct group_run_stats *rs, int ver,
                                         struct buf_output *out)
@@ -1435,14 +1554,17 @@ static void show_thread_status_terse_all(struct thread_stat *ts,
                log_buf(out, "%d;%s;%s;%d;%d", ver, fio_version_string,
                        ts->name, ts->groupid, ts->error);
 
-       /* Log Read Status */
+       /* Log Read Status, or mixed if unified_rw_rep = 1 */
        show_ddir_status_terse(ts, rs, DDIR_READ, ver, out);
-       /* Log Write Status */
-       show_ddir_status_terse(ts, rs, DDIR_WRITE, ver, out);
-       /* Log Trim Status */
-       if (ver == 2 || ver == 4 || ver == 5)
-               show_ddir_status_terse(ts, rs, DDIR_TRIM, ver, out);
-
+       if (ts->unified_rw_rep != UNIFIED_MIXED) {
+               /* Log Write Status */
+               show_ddir_status_terse(ts, rs, DDIR_WRITE, ver, out);
+               /* Log Trim Status */
+               if (ver == 2 || ver == 4 || ver == 5)
+                       show_ddir_status_terse(ts, rs, DDIR_TRIM, ver, out);
+       }
+       if (ts->unified_rw_rep == UNIFIED_BOTH)
+               show_mixed_ddir_status_terse(ts, rs, ver, out);
        /* CPU Usage */
        if (ts->total_run_time) {
                double runt = (double) ts->total_run_time;
@@ -1547,6 +1669,9 @@ static struct json_object *show_thread_status_json(struct thread_stat *ts,
        add_ddir_status_json(ts, rs, DDIR_TRIM, root);
        add_ddir_status_json(ts, rs, DDIR_SYNC, root);
 
+       if (ts->unified_rw_rep == UNIFIED_BOTH)
+               add_mixed_ddir_status_json(ts, rs, root);
+
        /* CPU Usage */
        if (ts->total_run_time) {
                double runt = (double) ts->total_run_time;
@@ -1817,9 +1942,10 @@ static void __sum_stat(struct io_stat *dst, struct io_stat *src, bool first)
  * numbers. For group_reporting, we should just add those up, not make
  * them the mean of everything.
  */
-static void sum_stat(struct io_stat *dst, struct io_stat *src, bool first,
-                    bool pure_sum)
+static void sum_stat(struct io_stat *dst, struct io_stat *src, bool pure_sum)
 {
+       bool first = dst->samples == 0;
+
        if (src->samples == 0)
                return;
 
@@ -1869,48 +1995,81 @@ void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src)
                dst->sig_figs = src->sig_figs;
 }
 
-void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src,
-                     bool first)
+/*
+ * Free the clat_prio_stat arrays allocated by alloc_clat_prio_stat_ddir().
+ */
+void free_clat_prio_stats(struct thread_stat *ts)
+{
+       enum fio_ddir ddir;
+
+       for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
+               sfree(ts->clat_prio[ddir]);
+               ts->clat_prio[ddir] = NULL;
+               ts->nr_clat_prio[ddir] = 0;
+       }
+}
+
+/*
+ * Allocate a clat_prio_stat array. The array has to be allocated/freed using
+ * smalloc/sfree, so that it is accessible by the process/thread summing the
+ * thread_stats.
+ */
+int alloc_clat_prio_stat_ddir(struct thread_stat *ts, enum fio_ddir ddir,
+                             int nr_prios)
+{
+       struct clat_prio_stat *clat_prio;
+       int i;
+
+       clat_prio = scalloc(nr_prios, sizeof(*ts->clat_prio[ddir]));
+       if (!clat_prio) {
+               log_err("fio: failed to allocate ts clat data\n");
+               return 1;
+       }
+
+       for (i = 0; i < nr_prios; i++)
+               clat_prio[i].clat_stat.min_val = ULONG_MAX;
+
+       ts->clat_prio[ddir] = clat_prio;
+       ts->nr_clat_prio[ddir] = nr_prios;
+
+       return 0;
+}
+
+void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src)
 {
        int k, l, m;
 
        for (l = 0; l < DDIR_RWDIR_CNT; l++) {
-               if (!dst->unified_rw_rep) {
-                       sum_stat(&dst->clat_stat[l], &src->clat_stat[l], first, false);
-                       sum_stat(&dst->clat_high_prio_stat[l], &src->clat_high_prio_stat[l], first, false);
-                       sum_stat(&dst->clat_low_prio_stat[l], &src->clat_low_prio_stat[l], first, false);
-                       sum_stat(&dst->slat_stat[l], &src->slat_stat[l], first, false);
-                       sum_stat(&dst->lat_stat[l], &src->lat_stat[l], first, false);
-                       sum_stat(&dst->bw_stat[l], &src->bw_stat[l], first, true);
-                       sum_stat(&dst->iops_stat[l], &src->iops_stat[l], first, true);
+               if (dst->unified_rw_rep != UNIFIED_MIXED) {
+                       sum_stat(&dst->clat_stat[l], &src->clat_stat[l], false);
+                       sum_stat(&dst->clat_high_prio_stat[l], &src->clat_high_prio_stat[l], false);
+                       sum_stat(&dst->clat_low_prio_stat[l], &src->clat_low_prio_stat[l], false);
+                       sum_stat(&dst->slat_stat[l], &src->slat_stat[l], false);
+                       sum_stat(&dst->lat_stat[l], &src->lat_stat[l], false);
+                       sum_stat(&dst->bw_stat[l], &src->bw_stat[l], true);
+                       sum_stat(&dst->iops_stat[l], &src->iops_stat[l], true);
 
                        dst->io_bytes[l] += src->io_bytes[l];
 
                        if (dst->runtime[l] < src->runtime[l])
                                dst->runtime[l] = src->runtime[l];
                } else {
-                       sum_stat(&dst->clat_stat[0], &src->clat_stat[l], first, false);
-                       sum_stat(&dst->clat_high_prio_stat[0], &src->clat_high_prio_stat[l], first, false);
-                       sum_stat(&dst->clat_low_prio_stat[0], &src->clat_low_prio_stat[l], first, false);
-                       sum_stat(&dst->slat_stat[0], &src->slat_stat[l], first, false);
-                       sum_stat(&dst->lat_stat[0], &src->lat_stat[l], first, false);
-                       sum_stat(&dst->bw_stat[0], &src->bw_stat[l], first, true);
-                       sum_stat(&dst->iops_stat[0], &src->iops_stat[l], first, true);
+                       sum_stat(&dst->clat_stat[0], &src->clat_stat[l], false);
+                       sum_stat(&dst->clat_high_prio_stat[0], &src->clat_high_prio_stat[l], false);
+                       sum_stat(&dst->clat_low_prio_stat[0], &src->clat_low_prio_stat[l], false);
+                       sum_stat(&dst->slat_stat[0], &src->slat_stat[l], false);
+                       sum_stat(&dst->lat_stat[0], &src->lat_stat[l], false);
+                       sum_stat(&dst->bw_stat[0], &src->bw_stat[l], true);
+                       sum_stat(&dst->iops_stat[0], &src->iops_stat[l], true);
 
                        dst->io_bytes[0] += src->io_bytes[l];
 
                        if (dst->runtime[0] < src->runtime[l])
                                dst->runtime[0] = src->runtime[l];
-
-                       /*
-                        * We're summing to the same destination, so override
-                        * 'first' after the first iteration of the loop
-                        */
-                       first = false;
                }
        }
 
-       sum_stat(&dst->sync_stat, &src->sync_stat, first, false);
+       sum_stat(&dst->sync_stat, &src->sync_stat, false);
        dst->usr_time += src->usr_time;
        dst->sys_time += src->sys_time;
        dst->ctx += src->ctx;
@@ -1931,7 +2090,7 @@ void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src,
                dst->io_u_lat_m[k] += src->io_u_lat_m[k];
 
        for (k = 0; k < DDIR_RWDIR_CNT; k++) {
-               if (!dst->unified_rw_rep) {
+               if (dst->unified_rw_rep != UNIFIED_MIXED) {
                        dst->total_io_u[k] += src->total_io_u[k];
                        dst->short_io_u[k] += src->short_io_u[k];
                        dst->drop_io_u[k] += src->drop_io_u[k];
@@ -1947,7 +2106,7 @@ void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src,
        for (k = 0; k < FIO_LAT_CNT; k++)
                for (l = 0; l < DDIR_RWDIR_CNT; l++)
                        for (m = 0; m < FIO_IO_U_PLAT_NR; m++)
-                               if (!dst->unified_rw_rep)
+                               if (dst->unified_rw_rep != UNIFIED_MIXED)
                                        dst->io_u_plat[k][l][m] += src->io_u_plat[k][l][m];
                                else
                                        dst->io_u_plat[k][0][m] += src->io_u_plat[k][l][m];
@@ -1957,7 +2116,7 @@ void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src,
 
        for (k = 0; k < DDIR_RWDIR_CNT; k++) {
                for (m = 0; m < FIO_IO_U_PLAT_NR; m++) {
-                       if (!dst->unified_rw_rep) {
+                       if (dst->unified_rw_rep != UNIFIED_MIXED) {
                                dst->io_u_plat_high_prio[k][m] += src->io_u_plat_high_prio[k][m];
                                dst->io_u_plat_low_prio[k][m] += src->io_u_plat_low_prio[k][m];
                        } else {
@@ -1985,22 +2144,27 @@ void init_group_run_stat(struct group_run_stats *gs)
                gs->min_bw[i] = gs->min_run[i] = ~0UL;
 }
 
-void init_thread_stat(struct thread_stat *ts)
+void init_thread_stat_min_vals(struct thread_stat *ts)
 {
-       int j;
+       int i;
 
+       for (i = 0; i < DDIR_RWDIR_CNT; i++) {
+               ts->clat_stat[i].min_val = ULONG_MAX;
+               ts->slat_stat[i].min_val = ULONG_MAX;
+               ts->lat_stat[i].min_val = ULONG_MAX;
+               ts->bw_stat[i].min_val = ULONG_MAX;
+               ts->iops_stat[i].min_val = ULONG_MAX;
+               ts->clat_high_prio_stat[i].min_val = ULONG_MAX;
+               ts->clat_low_prio_stat[i].min_val = ULONG_MAX;
+       }
+       ts->sync_stat.min_val = ULONG_MAX;
+}
+
+void init_thread_stat(struct thread_stat *ts)
+{
        memset(ts, 0, sizeof(*ts));
 
-       for (j = 0; j < DDIR_RWDIR_CNT; j++) {
-               ts->lat_stat[j].min_val = -1UL;
-               ts->clat_stat[j].min_val = -1UL;
-               ts->slat_stat[j].min_val = -1UL;
-               ts->bw_stat[j].min_val = -1UL;
-               ts->iops_stat[j].min_val = -1UL;
-               ts->clat_high_prio_stat[j].min_val = -1UL;
-               ts->clat_low_prio_stat[j].min_val = -1UL;
-       }
-       ts->sync_stat.min_val = -1UL;
+       init_thread_stat_min_vals(ts);
        ts->groupid = -1;
 }
 
@@ -2139,7 +2303,7 @@ void __show_run_stats(void)
                for (k = 0; k < ts->nr_block_infos; k++)
                        ts->block_infos[k] = td->ts.block_infos[k];
 
-               sum_thread_stats(ts, &td->ts, idx == 1);
+               sum_thread_stats(ts, &td->ts);
 
                if (td->o.ss_dur) {
                        ts->ss_state = td->ss.state;
@@ -2166,7 +2330,7 @@ void __show_run_stats(void)
                rs->kb_base = ts->kb_base;
                rs->unit_base = ts->unit_base;
                rs->sig_figs = ts->sig_figs;
-               rs->unified_rw_rep += ts->unified_rw_rep;
+               rs->unified_rw_rep |= ts->unified_rw_rep;
 
                for (j = 0; j < DDIR_RWDIR_CNT; j++) {
                        if (!ts->runtime[j])
@@ -2427,27 +2591,25 @@ static inline void add_stat_sample(struct io_stat *is, unsigned long long data)
  */
 static struct io_logs *get_new_log(struct io_log *iolog)
 {
-       size_t new_size, new_samples;
+       size_t new_samples;
        struct io_logs *cur_log;
 
        /*
         * Cap the size at MAX_LOG_ENTRIES, so we don't keep doubling
         * forever
         */
-       if (!iolog->cur_log_max)
-               new_samples = DEF_LOG_ENTRIES;
-       else {
+       if (!iolog->cur_log_max) {
+               new_samples = iolog->td->o.log_entries;
+       else {
                new_samples = iolog->cur_log_max * 2;
                if (new_samples > MAX_LOG_ENTRIES)
                        new_samples = MAX_LOG_ENTRIES;
        }
 
-       new_size = new_samples * log_entry_sz(iolog);
-
        cur_log = smalloc(sizeof(*cur_log));
        if (cur_log) {
                INIT_FLIST_HEAD(&cur_log->list);
-               cur_log->log = malloc(new_size);
+               cur_log->log = calloc(new_samples, log_entry_sz(iolog));
                if (cur_log->log) {
                        cur_log->nr_samples = 0;
                        cur_log->max_samples = new_samples;
@@ -2536,6 +2698,14 @@ void regrow_logs(struct thread_data *td)
        td->flags &= ~TD_F_REGROW_LOGS;
 }
 
+void regrow_agg_logs(void)
+{
+       enum fio_ddir ddir;
+
+       for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
+               regrow_log(agg_io_log[ddir]);
+}
+
 static struct io_logs *get_cur_log(struct io_log *iolog)
 {
        struct io_logs *cur_log;
@@ -2570,7 +2740,8 @@ static struct io_logs *get_cur_log(struct io_log *iolog)
 
 static void __add_log_sample(struct io_log *iolog, union io_sample_data data,
                             enum fio_ddir ddir, unsigned long long bs,
-                            unsigned long t, uint64_t offset, uint8_t priority_bit)
+                            unsigned long t, uint64_t offset,
+                            unsigned int priority)
 {
        struct io_logs *cur_log;
 
@@ -2589,7 +2760,7 @@ static void __add_log_sample(struct io_log *iolog, union io_sample_data data,
                s->time = t + (iolog->td ? iolog->td->unix_epoch : 0);
                io_sample_set_ddir(iolog, s, ddir);
                s->bs = bs;
-               s->priority_bit = priority_bit;
+               s->priority = priority;
 
                if (iolog->log_offset) {
                        struct io_sample_offset *so = (void *) s;
@@ -2666,7 +2837,7 @@ void reset_io_stats(struct thread_data *td)
 }
 
 static void __add_stat_to_log(struct io_log *iolog, enum fio_ddir ddir,
-                             unsigned long elapsed, bool log_max, uint8_t priority_bit)
+                             unsigned long elapsed, bool log_max)
 {
        /*
         * Note an entry in the log. Use the mean from the logged samples,
@@ -2681,26 +2852,26 @@ static void __add_stat_to_log(struct io_log *iolog, enum fio_ddir ddir,
                else
                        data.val = iolog->avg_window[ddir].mean.u.f + 0.50;
 
-               __add_log_sample(iolog, data, ddir, 0, elapsed, 0, priority_bit);
+               __add_log_sample(iolog, data, ddir, 0, elapsed, 0, 0);
        }
 
        reset_io_stat(&iolog->avg_window[ddir]);
 }
 
 static void _add_stat_to_log(struct io_log *iolog, unsigned long elapsed,
-                            bool log_max, uint8_t priority_bit)
+                            bool log_max)
 {
        int ddir;
 
        for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
-               __add_stat_to_log(iolog, ddir, elapsed, log_max, priority_bit);
+               __add_stat_to_log(iolog, ddir, elapsed, log_max);
 }
 
 static unsigned long add_log_sample(struct thread_data *td,
                                    struct io_log *iolog,
                                    union io_sample_data data,
                                    enum fio_ddir ddir, unsigned long long bs,
-                                   uint64_t offset, uint8_t priority_bit)
+                                   uint64_t offset, unsigned int ioprio)
 {
        unsigned long elapsed, this_window;
 
@@ -2713,7 +2884,8 @@ static unsigned long add_log_sample(struct thread_data *td,
         * If no time averaging, just add the log sample.
         */
        if (!iolog->avg_msec) {
-               __add_log_sample(iolog, data, ddir, bs, elapsed, offset, priority_bit);
+               __add_log_sample(iolog, data, ddir, bs, elapsed, offset,
+                                ioprio);
                return 0;
        }
 
@@ -2737,9 +2909,10 @@ static unsigned long add_log_sample(struct thread_data *td,
                        return diff;
        }
 
-       __add_stat_to_log(iolog, ddir, elapsed, td->o.log_max != 0, priority_bit);
+       __add_stat_to_log(iolog, ddir, elapsed, td->o.log_max != 0);
+
+       iolog->avg_last[ddir] = elapsed - (elapsed % iolog->avg_msec);
 
-       iolog->avg_last[ddir] = elapsed - (this_window - iolog->avg_msec);
        return iolog->avg_msec;
 }
 
@@ -2750,19 +2923,19 @@ void finalize_logs(struct thread_data *td, bool unit_logs)
        elapsed = mtime_since_now(&td->epoch);
 
        if (td->clat_log && unit_logs)
-               _add_stat_to_log(td->clat_log, elapsed, td->o.log_max != 0, 0);
+               _add_stat_to_log(td->clat_log, elapsed, td->o.log_max != 0);
        if (td->slat_log && unit_logs)
-               _add_stat_to_log(td->slat_log, elapsed, td->o.log_max != 0, 0);
+               _add_stat_to_log(td->slat_log, elapsed, td->o.log_max != 0);
        if (td->lat_log && unit_logs)
-               _add_stat_to_log(td->lat_log, elapsed, td->o.log_max != 0, 0);
+               _add_stat_to_log(td->lat_log, elapsed, td->o.log_max != 0);
        if (td->bw_log && (unit_logs == per_unit_log(td->bw_log)))
-               _add_stat_to_log(td->bw_log, elapsed, td->o.log_max != 0, 0);
+               _add_stat_to_log(td->bw_log, elapsed, td->o.log_max != 0);
        if (td->iops_log && (unit_logs == per_unit_log(td->iops_log)))
-               _add_stat_to_log(td->iops_log, elapsed, td->o.log_max != 0, 0);
+               _add_stat_to_log(td->iops_log, elapsed, td->o.log_max != 0);
 }
 
-void add_agg_sample(union io_sample_data data, enum fio_ddir ddir, unsigned long long bs,
-                                       uint8_t priority_bit)
+void add_agg_sample(union io_sample_data data, enum fio_ddir ddir,
+                   unsigned long long bs)
 {
        struct io_log *iolog;
 
@@ -2770,7 +2943,7 @@ void add_agg_sample(union io_sample_data data, enum fio_ddir ddir, unsigned long
                return;
 
        iolog = agg_io_log[ddir];
-       __add_log_sample(iolog, data, ddir, bs, mtime_since_genesis(), 0, priority_bit);
+       __add_log_sample(iolog, data, ddir, bs, mtime_since_genesis(), 0, 0);
 }
 
 void add_sync_clat_sample(struct thread_stat *ts, unsigned long long nsec)
@@ -2782,8 +2955,10 @@ void add_sync_clat_sample(struct thread_stat *ts, unsigned long long nsec)
        add_stat_sample(&ts->sync_stat, nsec);
 }
 
-static void add_lat_percentile_sample_noprio(struct thread_stat *ts,
-                               unsigned long long nsec, enum fio_ddir ddir, enum fio_lat lat)
+static inline void add_lat_percentile_sample(struct thread_stat *ts,
+                                            unsigned long long nsec,
+                                            enum fio_ddir ddir,
+                                            enum fio_lat lat)
 {
        unsigned int idx = plat_val_to_idx(nsec);
        assert(idx < FIO_IO_U_PLAT_NR);
@@ -2791,15 +2966,14 @@ static void add_lat_percentile_sample_noprio(struct thread_stat *ts,
        ts->io_u_plat[lat][ddir][idx]++;
 }
 
-static void add_lat_percentile_sample(struct thread_stat *ts,
-                               unsigned long long nsec, enum fio_ddir ddir, uint8_t priority_bit,
-                               enum fio_lat lat)
+static inline void add_lat_percentile_prio_sample(struct thread_stat *ts,
+                                                 unsigned long long nsec,
+                                                 enum fio_ddir ddir,
+                                                 bool high_prio)
 {
        unsigned int idx = plat_val_to_idx(nsec);
 
-       add_lat_percentile_sample_noprio(ts, nsec, ddir, lat);
-
-       if (!priority_bit)
+       if (!high_prio)
                ts->io_u_plat_low_prio[ddir][idx]++;
        else
                ts->io_u_plat_high_prio[ddir][idx]++;
@@ -2807,7 +2981,7 @@ static void add_lat_percentile_sample(struct thread_stat *ts,
 
 void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
                     unsigned long long nsec, unsigned long long bs,
-                    uint64_t offset, uint8_t priority_bit)
+                    uint64_t offset, unsigned int ioprio, bool high_prio)
 {
        const bool needs_lock = td_async_processing(td);
        unsigned long elapsed, this_window;
@@ -2819,8 +2993,17 @@ void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
 
        add_stat_sample(&ts->clat_stat[ddir], nsec);
 
+       /*
+        * When lat_percentiles=1 (default 0), the reported high/low priority
+        * percentiles and stats are used for describing total latency values,
+        * even though the variable names themselves start with clat_.
+        *
+        * Because of the above definition, add a prio stat sample only when
+        * lat_percentiles=0. add_lat_sample() will add the prio stat sample
+        * when lat_percentiles=1.
+        */
        if (!ts->lat_percentiles) {
-               if (priority_bit)
+               if (high_prio)
                        add_stat_sample(&ts->clat_high_prio_stat[ddir], nsec);
                else
                        add_stat_sample(&ts->clat_low_prio_stat[ddir], nsec);
@@ -2828,13 +3011,18 @@ void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
 
        if (td->clat_log)
                add_log_sample(td, td->clat_log, sample_val(nsec), ddir, bs,
-                              offset, priority_bit);
+                              offset, ioprio);
 
        if (ts->clat_percentiles) {
-               if (ts->lat_percentiles)
-                       add_lat_percentile_sample_noprio(ts, nsec, ddir, FIO_CLAT);
-               else
-                       add_lat_percentile_sample(ts, nsec, ddir, priority_bit, FIO_CLAT);
+               /*
+                * Because of the above definition, add a prio lat percentile
+                * sample only when lat_percentiles=0. add_lat_sample() will add
+                * the prio lat percentile sample when lat_percentiles=1.
+                */
+               add_lat_percentile_sample(ts, nsec, ddir, FIO_CLAT);
+               if (!ts->lat_percentiles)
+                       add_lat_percentile_prio_sample(ts, nsec, ddir,
+                                                      high_prio);
        }
 
        if (iolog && iolog->hist_msec) {
@@ -2863,7 +3051,7 @@ void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
                                FIO_IO_U_PLAT_NR * sizeof(uint64_t));
                        flist_add(&dst->list, &hw->list);
                        __add_log_sample(iolog, sample_plat(dst), ddir, bs,
-                                               elapsed, offset, priority_bit);
+                                        elapsed, offset, ioprio);
 
                        /*
                         * Update the last time we recorded as being now, minus
@@ -2880,8 +3068,8 @@ void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
 }
 
 void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
-                       unsigned long long nsec, unsigned long long bs, uint64_t offset,
-                       uint8_t priority_bit)
+                    unsigned long long nsec, unsigned long long bs,
+                    uint64_t offset, unsigned int ioprio)
 {
        const bool needs_lock = td_async_processing(td);
        struct thread_stat *ts = &td->ts;
@@ -2895,11 +3083,11 @@ void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
        add_stat_sample(&ts->slat_stat[ddir], nsec);
 
        if (td->slat_log)
-               add_log_sample(td, td->slat_log, sample_val(nsec), ddir, bs, offset,
-                       priority_bit);
+               add_log_sample(td, td->slat_log, sample_val(nsec), ddir, bs,
+                              offset, ioprio);
 
        if (ts->slat_percentiles)
-               add_lat_percentile_sample_noprio(ts, nsec, ddir, FIO_SLAT);
+               add_lat_percentile_sample(ts, nsec, ddir, FIO_SLAT);
 
        if (needs_lock)
                __td_io_u_unlock(td);
@@ -2907,7 +3095,7 @@ void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
 
 void add_lat_sample(struct thread_data *td, enum fio_ddir ddir,
                    unsigned long long nsec, unsigned long long bs,
-                   uint64_t offset, uint8_t priority_bit)
+                   uint64_t offset, unsigned int ioprio, bool high_prio)
 {
        const bool needs_lock = td_async_processing(td);
        struct thread_stat *ts = &td->ts;
@@ -2922,11 +3110,22 @@ void add_lat_sample(struct thread_data *td, enum fio_ddir ddir,
 
        if (td->lat_log)
                add_log_sample(td, td->lat_log, sample_val(nsec), ddir, bs,
-                              offset, priority_bit);
+                              offset, ioprio);
 
+       /*
+        * When lat_percentiles=1 (default 0), the reported high/low priority
+        * percentiles and stats are used for describing total latency values,
+        * even though the variable names themselves start with clat_.
+        *
+        * Because of the above definition, add a prio stat and prio lat
+        * percentile sample only when lat_percentiles=1. add_clat_sample() will
+        * add the prio stat and prio lat percentile sample when
+        * lat_percentiles=0.
+        */
        if (ts->lat_percentiles) {
-               add_lat_percentile_sample(ts, nsec, ddir, priority_bit, FIO_LAT);
-               if (priority_bit)
+               add_lat_percentile_sample(ts, nsec, ddir, FIO_LAT);
+               add_lat_percentile_prio_sample(ts, nsec, ddir, high_prio);
+               if (high_prio)
                        add_stat_sample(&ts->clat_high_prio_stat[ddir], nsec);
                else
                        add_stat_sample(&ts->clat_low_prio_stat[ddir], nsec);
@@ -2955,7 +3154,7 @@ void add_bw_sample(struct thread_data *td, struct io_u *io_u,
 
        if (td->bw_log)
                add_log_sample(td, td->bw_log, sample_val(rate), io_u->ddir,
-                              bytes, io_u->offset, io_u_is_prio(io_u));
+                              bytes, io_u->offset, io_u->ioprio);
 
        td->stat_io_bytes[io_u->ddir] = td->this_io_bytes[io_u->ddir];
 
@@ -2977,7 +3176,7 @@ static int __add_samples(struct thread_data *td, struct timespec *parent_tv,
        next_log = avg_time;
 
        spent = mtime_since(parent_tv, t);
-       if (spent < avg_time && avg_time - spent >= LOG_MSEC_SLACK)
+       if (spent < avg_time && avg_time - spent > LOG_MSEC_SLACK)
                return avg_time - spent;
 
        if (needs_lock)
@@ -3009,7 +3208,8 @@ static int __add_samples(struct thread_data *td, struct timespec *parent_tv,
                        if (td->o.min_bs[ddir] == td->o.max_bs[ddir])
                                bs = td->o.min_bs[ddir];
 
-                       next = add_log_sample(td, log, sample_val(rate), ddir, bs, 0, 0);
+                       next = add_log_sample(td, log, sample_val(rate), ddir,
+                                             bs, 0, 0);
                        next_log = min(next_log, next);
                }
 
@@ -3049,7 +3249,7 @@ void add_iops_sample(struct thread_data *td, struct io_u *io_u,
 
        if (td->iops_log)
                add_log_sample(td, td->iops_log, sample_val(1), io_u->ddir,
-                              bytes, io_u->offset, io_u_is_prio(io_u));
+                              bytes, io_u->offset, io_u->ioprio);
 
        td->stat_io_blocks[io_u->ddir] = td->this_io_blocks[io_u->ddir];
 
@@ -3070,13 +3270,16 @@ static int add_iops_samples(struct thread_data *td, struct timespec *t)
 int calc_log_samples(void)
 {
        struct thread_data *td;
-       unsigned int next = ~0U, tmp;
+       unsigned int next = ~0U, tmp = 0, next_mod = 0, log_avg_msec_min = -1U;
        struct timespec now;
        int i;
+       long elapsed_time = 0;
 
        fio_gettime(&now, NULL);
 
        for_each_td(td, i) {
+               elapsed_time = mtime_since_now(&td->epoch);
+
                if (!td->o.stats)
                        continue;
                if (in_ramp_time(td) ||
@@ -3087,17 +3290,34 @@ int calc_log_samples(void)
                if (!td->bw_log ||
                        (td->bw_log && !per_unit_log(td->bw_log))) {
                        tmp = add_bw_samples(td, &now);
-                       if (tmp < next)
-                               next = tmp;
+
+                       if (td->bw_log)
+                               log_avg_msec_min = min(log_avg_msec_min, (unsigned int)td->bw_log->avg_msec);
                }
                if (!td->iops_log ||
                        (td->iops_log && !per_unit_log(td->iops_log))) {
                        tmp = add_iops_samples(td, &now);
-                       if (tmp < next)
-                               next = tmp;
+
+                       if (td->iops_log)
+                               log_avg_msec_min = min(log_avg_msec_min, (unsigned int)td->iops_log->avg_msec);
                }
+
+               if (tmp < next)
+                       next = tmp;
        }
 
+       /* if log_avg_msec_min has not been changed, set it to 0 */
+       if (log_avg_msec_min == -1U)
+               log_avg_msec_min = 0;
+
+       if (log_avg_msec_min == 0)
+               next_mod = elapsed_time;
+       else
+               next_mod = elapsed_time % log_avg_msec_min;
+
+       /* correction to keep the time on the log avg msec boundary */
+       next = min(next, (log_avg_msec_min - next_mod));
+
        return next == ~0U ? 0 : next;
 }