diff --git a/.clang-format b/.clang-format
index 0b3aa238..56e516cd 100644
--- a/.clang-format
+++ b/.clang-format
@@ -3,10 +3,12 @@ AccessModifierOffset: -4
 AlignAfterOpenBracket: Align
 AlignConsecutiveAssignments: None
 AlignOperands: Align
+AlignTrailingComments: true
 AllowAllArgumentsOnNextLine: false
 AllowAllConstructorInitializersOnNextLine: false
 AllowAllParametersOfDeclarationOnNextLine: false
 AllowShortBlocksOnASingleLine: Always
+AllowShortEnumsOnASingleLine: true
 AllowShortCaseLabelsOnASingleLine: false
 AllowShortFunctionsOnASingleLine: None
 AllowShortLambdasOnASingleLine: All
@@ -16,7 +18,7 @@ AlwaysBreakAfterDefinitionReturnType: None
 AlwaysBreakAfterReturnType: TopLevelDefinitions
 AlwaysBreakBeforeMultilineStrings: false
 AlwaysBreakTemplateDeclarations: Yes
-AttributeMacros: ['__capability', '__output', '__ununsed', '_U_']
+AttributeMacros: ['__capability', '__output', '__unused', '_U_']
 BinPackArguments: false
 BinPackParameters: false
 BitFieldColonSpacing: None
@@ -28,9 +30,11 @@ BraceWrapping:
   AfterEnum: false
   AfterFunction: true
   AfterNamespace: false
+  AfterStruct: false
   AfterUnion: false
   BeforeCatch: false
   BeforeElse: false
+  BeforeWhile: false
   IndentBraces: false
   SplitEmptyFunction: false
   SplitEmptyRecord: true
@@ -69,7 +73,12 @@ IncludeCategories:
     Priority:        6
 IndentGotoLabels: false
 IndentPPDirectives: None
+IndentRequiresClause: false
+IndentExternBlock: AfterExternBlock
 IndentWidth: 4
+IndentWrappedFunctionNames: false
+InsertBraces: true
+InsertTrailingCommas: None
 KeepEmptyLinesAtTheStartOfBlocks: false
 MaxEmptyLinesToKeep: 1
 NamespaceIndentation: None
@@ -86,12 +95,16 @@ SpaceBeforeCtorInitializerColon: true
 SpaceBeforeInheritanceColon: true
 SpaceBeforeParens: ControlStatements
 SpaceBeforeRangeBasedForLoopColon: false
+SpaceInEmptyBlock: false
 SpaceInEmptyParentheses: false
 SpacesBeforeTrailingComments: 1
 SpacesInAngles:  Never
 SpacesInConditionalStatement: false
 SpacesInContainerLiterals: false
 SpacesInCStyleCastParentheses: false
+SpacesInLineCommentPrefix:
+  Minimum:         1
+  Maximum:         -1
 SpacesInParentheses: false
 SpacesInSquareBrackets: false
 TabWidth: 8
diff --git a/docs/CREDIT b/docs/CREDIT
index 26583a35..d21aee22 100644
--- a/docs/CREDIT
+++ b/docs/CREDIT
@@ -118,4 +118,5 @@ Bastian Triller <GitHub @btriller>
     - Linux SLL2
 
 GithHub @plangarbalint
+    - eBPF
     - nanosecond timers
\ No newline at end of file
diff --git a/src/common/timer.c b/src/common/timer.c
index 60e10bf2..38e123f9 100755
--- a/src/common/timer.c
+++ b/src/common/timer.c
@@ -47,7 +47,7 @@ init_timestamp(struct timespec *timestamp)
 int
 get_current_time(struct timespec *ts)
 {
-#if defined _POSIX_C_SOURCE && _POSIX_C_SOURCE >= 199309L
+#if defined CLOCK_MONOTONIC || defined _POSIX_C_SOURCE && _POSIX_C_SOURCE >= 199309L
     return clock_gettime(CLOCK_MONOTONIC, ts);
 #else
     struct timeval tv;
diff --git a/src/common/timer.h b/src/common/timer.h
index 0cf801fa..b4cbfa34 100755
--- a/src/common/timer.h
+++ b/src/common/timer.h
@@ -178,6 +178,4 @@ void timesdiv_float(struct timespec *tvs, float div);
 typedef struct timeval timestamp_t;
 
 void init_timestamp(struct timespec *timestamp);
-int get_current_time(struct timespec *ts);
-
-#endif /* _TIMER_H_ */
+int get_current_time(struct timespec *timestamp);
diff --git a/src/common/utils.c b/src/common/utils.c
index 8e69834e..520d40d2 100644
--- a/src/common/utils.c
+++ b/src/common/utils.c
@@ -307,8 +307,8 @@ format_date_time(struct timespec *when, char *buf, size_t len)
     if (!tm)
         return -1;
 
-    strftime(tmp, sizeof tmp, "%Y-%m-%d %H:%M:%S.%%06u", tm);
-    return snprintf(buf, len, tmp, when->tv_nsec / 1000);
+    strftime(tmp, sizeof tmp, "%Y-%m-%d %H:%M:%S.%%09u", tm);
+    return snprintf(buf, len, tmp, when->tv_nsec);
 }
 
 /**
diff --git a/src/defines.h.in b/src/defines.h.in
index fe776552..1975f053 100644
--- a/src/defines.h.in
+++ b/src/defines.h.in
@@ -102,7 +102,7 @@ char dummy[0];
 #define COUNTER unsigned long
 #define COUNTER_SPEC "%lu"
 #endif
-#define COUNTER_OVERFLOW_RISK (((COUNTER)~0) >> 20)
+#define COUNTER_OVERFLOW_RISK (((COUNTER)~0) >> 23)
 
 #include <common/cidr.h>
 #include <common/list.h>
diff --git a/src/send_packets.c b/src/send_packets.c
index e8d41fbc..788341b4 100644
--- a/src/send_packets.c
+++ b/src/send_packets.c
@@ -56,10 +56,6 @@ extern tcpedit_t *tcpedit;
 #include "send_packets.h"
 #include "sleep.h"
 
-#ifdef DEBUG
-extern int debug;
-#endif
-
 static void calc_sleep_time(tcpreplay_t *ctx,
                             struct timespec *pkt_time,
                             struct timespec *last,
@@ -70,8 +66,11 @@ static void calc_sleep_time(tcpreplay_t *ctx,
                             COUNTER start_us,
                             COUNTER *skip_length);
 static void tcpr_sleep(tcpreplay_t *ctx, sendpacket_t *sp _U_, struct timespec *nap_this_time, struct timespec *now);
-static u_char *
-get_next_packet(tcpreplay_t *ctx, pcap_t *pcap, struct pcap_pkthdr *pkthdr, int file_idx, packet_cache_t **prev_packet);
+static u_char *get_next_packet(tcpreplay_opt_t *options,
+                               pcap_t *pcap,
+                               struct pcap_pkthdr *pkthdr,
+                               int file_idx,
+                               packet_cache_t **prev_packet);
 static uint32_t get_user_count(tcpreplay_t *ctx, sendpacket_t *sp, COUNTER counter);
 
 #ifdef HAVE_NETMAP
@@ -336,7 +335,7 @@ increment_iteration(tcpreplay_t *ctx)
 void
 send_packets(tcpreplay_t *ctx, pcap_t *pcap, int idx)
 {
-    struct timespec now, print_delta, last_pkt_ts, pkthdr_ts;
+    struct timespec now, print_delta, last_pkt_ts;
     tcpreplay_opt_t *options = ctx->options;
     tcpreplay_stats_t *stats = &ctx->stats;
     COUNTER packetnum = 0;
@@ -386,7 +385,9 @@ send_packets(tcpreplay_t *ctx, pcap_t *pcap, int idx)
      * Keep sending while we have packets or until
      * we've sent enough packets
      */
-    while (!ctx->abort && (pktdata = get_next_packet(ctx, pcap, &pkthdr, idx, prev_packet)) != NULL) {
+    while (!ctx->abort && read_next_packet &&
+           (pktdata = get_next_packet(options, pcap, &pkthdr, idx, prev_packet)) != NULL) {
+        struct timespec pkthdr_ts;
         TIMEVAL_AS_TIMESPEC_SET(&pkthdr_ts, &pkthdr.ts); // libpcap puts nanosec values in tv_usec
         now_is_now = false;
         packetnum++;
@@ -480,6 +481,7 @@ send_packets(tcpreplay_t *ctx, pcap_t *pcap, int idx)
                             &stats->end_time,
                             TIMESPEC_TO_NANOSEC(&stats->start_time),
                             &skip_length);
+
             /*
              * Track the time of the "last packet sent".
              *
@@ -1033,12 +1035,6 @@ calc_sleep_time(tcpreplay_t *ctx,
         if (timescmp(pkt_ts_delta, time_delta, >)) {
             /* pkt_time_delta has increased, so handle normally */
             timessub(pkt_ts_delta, time_delta, &nap_for);
-            // printf("pkt_ts_delta sec: %lu\n", pkt_ts_delta->tv_sec);
-            // printf("pkt_ts_delta nsec: %lu\n", pkt_ts_delta->tv_nsec);
-            // printf("time_delta sec: %lu\n", time_delta->tv_sec);
-            // printf("time_delta nsec: %lu\n", time_delta->tv_nsec);
-            // printf("nap_for sec: %lu\n", nap_for.tv_sec);
-            // printf("nap_for nsec: %lu\n", nap_for.tv_nsec);
             TIMESPEC_SET(&ctx->nap, &nap_for);
             dbgx(3, "original packet delta time: " TIMESPEC_FORMAT, ctx->nap.tv_sec, ctx->nap.tv_nsec);
             timesdiv_float(&ctx->nap, options->speed.multiplier);
@@ -1055,7 +1051,7 @@ calc_sleep_time(tcpreplay_t *ctx,
         if (now_ns) {
             COUNTER next_tx_ns;
             COUNTER bps = options->speed.speed;
-            COUNTER bits_sent = ((ctx->stats.bytes_sent + len) * 8); // PB: Ferencol miert bits sent?
+            COUNTER bits_sent = ((ctx->stats.bytes_sent + len) * 8);
             COUNTER tx_ns = now_ns - start_ns;
 
             /*
@@ -1063,7 +1059,7 @@ calc_sleep_time(tcpreplay_t *ctx,
              *
              * ensure there is no overflow in cases where bits_sent is very high
              */
-            if (bits_sent > COUNTER_OVERFLOW_RISK && bps > 500000)
+            if (bits_sent > COUNTER_OVERFLOW_RISK)
                 next_tx_ns = (bits_sent * 1000) / bps * 1000000;
             else
                 next_tx_ns = (bits_sent * 1000000000) / bps;
@@ -1154,9 +1150,8 @@ tcpr_sleep(tcpreplay_t *ctx, sendpacket_t *sp, struct timespec *nap_this_time, s
 #endif
 
     /* don't sleep if nap = {0, 0} */
-    if (!timesisset(nap_this_time)) {
+    if (!timesisset(nap_this_time))
         return;
-    }
 
     /* do we need to limit the total time we sleep? */
     if (timesisset(&(options->maxsleep)) && (timescmp(nap_this_time, &(options->maxsleep), >))) {
diff --git a/src/signal_handler.c b/src/signal_handler.c
index f8b85d9e..9bfd56ee 100644
--- a/src/signal_handler.c
+++ b/src/signal_handler.c
@@ -31,7 +31,7 @@
 #include <sys/time.h>
 #include <unistd.h>
 
-struct timeval suspend_time; // PB: do we need to modify this part of the code?
+struct timeval suspend_time;
 static struct timeval suspend_start;
 static struct timeval suspend_end;
 
diff --git a/src/sleep.c b/src/sleep.c
index 16234a43..27e3e85f 100644
--- a/src/sleep.c
+++ b/src/sleep.c
@@ -80,10 +80,4 @@ ioport_sleep(sendpacket_t *sp _U_, const struct timespec *nap _U_, struct timesp
 #else
     err(-1, "Platform does not support IO Port for timing");
 #endif
-
-#ifdef HAVE_NETMAP
-    if (flush)
-        ioctl(sp->handle.fd, NIOCTXSYNC, NULL);   /* flush TX buffer */
-#endif /* HAVE_NETMAP */
-    get_current_time(now);
 }
diff --git a/src/sleep.h b/src/sleep.h
index 71365ab5..ee12d8d2 100644
--- a/src/sleep.h
+++ b/src/sleep.h
@@ -54,9 +54,9 @@
 static inline void
 nanosleep_sleep(sendpacket_t *sp _U_, const struct timespec *nap, struct timespec *now, bool flush _U_)
 {
+#if defined _POSIX_C_SOURCE && _POSIX_C_SOURCE >= 200112L
     struct timespec sleep_until;
     timeradd_timespec(now, nap, &sleep_until);
-#if defined _POSIX_C_SOURCE && _POSIX_C_SOURCE >= 200112L
     clock_nanosleep(CLOCK_MONOTONIC, TIMER_ABSTIME, &sleep_until, NULL);
 #else
     nanosleep(nap, NULL);
@@ -77,36 +77,36 @@ nanosleep_sleep(sendpacket_t *sp _U_, const struct timespec *nap, struct timespe
  * Note: make sure "now" has recently been updated.
  */
 static inline void
-gettimeofday_sleep(sendpacket_t *sp _U_, struct timespec *nap, struct timespec *now, bool flush _U_)
+gettimeofday_sleep(sendpacket_t *sp, struct timespec *nap, struct timespec *now, bool flush _U_)
 {
-    struct timeval now_ms, sleep_until, nap_for, last;
-    TIMESPEC_TO_TIMEVAL(&nap_for, nap);
-    gettimeofday(&now_ms, NULL);
+    struct timespec sleep_until;
 #ifdef HAVE_NETMAP
+    struct timespec last;
     uint32_t i = 0;
-    TIMEVAL_SET(&last, &now_ms);
+
+    TIMESPEC_SET(&last, now);
 #endif /* HAVE_NETMAP */
 
-    timeradd(&now_ms, &nap_for, &sleep_until);
+    timeradd_timespec(now, nap, &sleep_until);
+
     while (!sp->abort) {
 #ifdef HAVE_NETMAP
-        if (flush && timercmp(&now_ms, &last, !=)) {
-            TIMESPEC_SET(&last, &now_ms);
+        if (flush && timescmp(now, &last, !=)) {
+            TIMESPEC_SET(&last, now);
             if ((++i & 0xf) == 0)
                 /* flush TX buffer every 16 usec */
                 ioctl(sp->handle.fd, NIOCTXSYNC, NULL);
         }
 #endif /* HAVE_NETMAP */
-        if (timercmp(&now_ms, &sleep_until, >=))
+        if (timescmp(now, &sleep_until, >=))
             break;
 
 #ifdef HAVE_SCHED_H
         /* yield the CPU so other apps remain responsive */
         sched_yield();
 #endif
-        gettimeofday(&now_ms, NULL);
+        get_current_time(now);
     }
-    get_current_time(now);
 }
 
 #ifdef HAVE_SELECT
@@ -120,8 +120,6 @@ static inline void
 select_sleep(sendpacket_t *sp _U_, struct timespec *nap, struct timespec *now_ns, bool flush _U_)
 {
     struct timeval timeout;
-    timeout.tv_sec = 0;
-    timeout.tv_usec = 0;
 #ifdef HAVE_NETMAP
     if (flush)
         ioctl(sp->handle.fd, NIOCTXSYNC, NULL); /* flush TX buffer */
@@ -136,6 +134,7 @@ select_sleep(sendpacket_t *sp _U_, struct timespec *nap, struct timespec *now_ns
     if (flush)
         ioctl(sp->handle.fd, NIOCTXSYNC, NULL); /* flush TX buffer */
 #endif
+
     get_current_time(now_ns);
 }
 #endif /* HAVE_SELECT */
@@ -152,5 +151,3 @@ select_sleep(sendpacket_t *sp _U_, struct timespec *nap, struct timespec *now_ns
 void ioport_sleep_init(void);
 
 void ioport_sleep(sendpacket_t *sp _U_, const struct timespec *nap, struct timespec *now, bool flush);
-
-#endif /* __SLEEP_H__ */
diff --git a/src/tcpreplay_api.c b/src/tcpreplay_api.c
index 319ae109..453b92d7 100644
--- a/src/tcpreplay_api.c
+++ b/src/tcpreplay_api.c
@@ -166,6 +166,7 @@ tcpreplay_post_args(tcpreplay_t *ctx, int argc)
 
     options->loop = OPT_VALUE_LOOP;
     options->loopdelay_ms = OPT_VALUE_LOOPDELAY_MS;
+    options->loopdelay_ns = OPT_VALUE_LOOPDELAY_NS;
 
     if (HAVE_OPT(LIMIT))
         options->limit_send = OPT_VALUE_LIMIT;
@@ -314,7 +315,6 @@ tcpreplay_post_args(tcpreplay_t *ctx, int argc)
             options->accurate = accurate_gtod;
         } else if (strcmp(OPT_ARG(TIMER), "nano") == 0) {
             options->accurate = accurate_nanosleep;
-            options->loopdelay_ns = OPT_VALUE_LOOPDELAY_NS;
         } else if (strcmp(OPT_ARG(TIMER), "abstime") == 0) {
             tcpreplay_seterr(ctx, "%s", "abstime is deprecated");
             ret = -1;
@@ -1120,6 +1120,7 @@ tcpreplay_prepare(tcpreplay_t *ctx)
     return ret;
 }
 
+static bool apply_loop_delay(tcpreplay_t *ctx);
 /**
  * \brief sends the traffic out the interfaces
  *
@@ -1167,12 +1168,12 @@ tcpreplay_replay(tcpreplay_t *ctx)
                             loop, total_loops,
                             ctx->unique_iteration);
             }
-            if ((rcode = tcpr_replay_index(ctx)) < 0) {
+            if ((rcode = tcpr_replay_index(ctx)) < 0)
                 return rcode;
-            }
             if (ctx->options->loop > 0) {
-                apply_loop_delay(ctx);
-                get_current_time(&ctx->stats.end_time);
+                if (apply_loop_delay(ctx))
+                    get_current_time(&ctx->stats.end_time);
+
                 if (ctx->options->stats == 0) {
                     packet_stats(&ctx->stats);
                 }
@@ -1195,11 +1196,12 @@ tcpreplay_replay(tcpreplay_t *ctx)
                     printf("Loop " COUNTER_SPEC " (" COUNTER_SPEC " unique)...\n", loop,
                             ctx->unique_iteration);
             }
-            if ((rcode = tcpr_replay_index(ctx)) < 0) {
+            if ((rcode = tcpr_replay_index(ctx)) < 0)
                 return rcode;
-            }
-            apply_loop_delay(ctx);
-            get_current_time(&ctx->stats.end_time);
+
+            if (apply_loop_delay(ctx))
+                get_current_time(&ctx->stats.end_time);
+
             if (ctx->options->stats == 0 && !ctx->abort)
                 packet_stats(&ctx->stats);
         }
@@ -1381,19 +1383,20 @@ int tcpreplay_get_flow_expiry(tcpreplay_t *ctx)
     return ctx->options->flow_expiry;
 }
 
-void
-apply_loop_delay(tcpreplay_t *ctx)
-{
-    if (ctx->options->accurate == accurate_nanosleep) {
-        if (!ctx->abort && ctx->options->loopdelay_ns > 0) {
-            struct timespec nap;
-            nap.tv_sec = 0;
-            nap.tv_nsec = ctx->options->loopdelay_ns;
-            nanosleep_sleep(NULL, &nap, &ctx->stats.end_time, NULL);
-        }
-    } else {
-        if (!ctx->abort && ctx->options->loopdelay_ms > 0) {
-            usleep(ctx->options->loopdelay_ms * 1000);
-        }
+bool
+apply_loop_delay(tcpreplay_t *ctx) {
+    if (!ctx->abort && ctx->options->loopdelay_ms > 0) {
+        usleep(ctx->options->loopdelay_ms * 1000);
+        return true;
+    }
+
+    if (!ctx->abort && ctx->options->loopdelay_ns > 0) {
+        struct timespec nap;
+        nap.tv_sec = 0;
+        nap.tv_nsec = ctx->options->loopdelay_ns;
+        nanosleep_sleep(NULL, &nap, &ctx->stats.end_time, NULL);
+        return true;
     }
+
+    return false;
 }
diff --git a/src/tcpreplay_api.h b/src/tcpreplay_api.h
index 321c8f18..6cff96b7 100644
--- a/src/tcpreplay_api.h
+++ b/src/tcpreplay_api.h
@@ -272,7 +272,6 @@ int tcpreplay_set_verbose(tcpreplay_t *, bool);
 int tcpreplay_set_tcpdump_args(tcpreplay_t *, char *);
 int tcpreplay_set_tcpdump(tcpreplay_t *, tcpdump_t *);
 
-void apply_loop_delay(tcpreplay_t *ctx);
 /*
  * These functions are seen by the outside world, but nobody should ever use them
  * outside of internal tcpreplay API functions
diff --git a/src/tcpreplay_opts.def b/src/tcpreplay_opts.def
index a17dff73..e3536fde 100644
--- a/src/tcpreplay_opts.def
+++ b/src/tcpreplay_opts.def
@@ -320,6 +320,7 @@ flag = {
 
 flag = {
     name        = loopdelay-ms;
+    flags-cant  = loopdelay-ns;
     flags-must  = loop;
     arg-type    = number;
     arg-range   = "0->";
@@ -330,7 +331,8 @@ flag = {
 
 flag = {
     name        = loopdelay-ns;
-    flags-must  = loop, timer;
+    flags-cant  = loopdelay-ms;
+    flags-must  = loop;
     arg-type    = number;
     arg-range   = "0->";
     descrip     = "Delay between loops in nanoseconds";