eal: add timespec_get shim
[dpdk.git] / drivers / net / pcap / rte_eth_pcap.c
index 353538f..ef50d08 100644 (file)
 #include <pcap.h>
 
 #include <rte_cycles.h>
-#include <rte_ethdev_driver.h>
-#include <rte_ethdev_vdev.h>
+#include <ethdev_driver.h>
+#include <ethdev_vdev.h>
 #include <rte_kvargs.h>
 #include <rte_malloc.h>
 #include <rte_mbuf.h>
+#include <rte_mbuf_dyn.h>
 #include <rte_bus_vdev.h>
 #include <rte_string_fns.h>
 
 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
-#define RTE_ETH_PCAP_SNAPLEN ETHER_MAX_JUMBO_FRAME_LEN
+#define RTE_ETH_PCAP_SNAPLEN RTE_ETHER_MAX_JUMBO_FRAME_LEN
 #define RTE_ETH_PCAP_PROMISC 1
 #define RTE_ETH_PCAP_TIMEOUT -1
 
 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
 #define ETH_PCAP_IFACE_ARG    "iface"
 #define ETH_PCAP_PHY_MAC_ARG  "phy_mac"
+#define ETH_PCAP_INFINITE_RX_ARG  "infinite_rx"
 
 #define ETH_PCAP_ARG_MAXLEN    64
 
 #define RTE_PMD_PCAP_MAX_QUEUES 16
 
 static char errbuf[PCAP_ERRBUF_SIZE];
-static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
 static struct timeval start_time;
 static uint64_t start_cycles;
 static uint64_t hz;
 static uint8_t iface_idx;
 
+static uint64_t timestamp_rx_dynflag;
+static int timestamp_dynfield_offset = -1;
+
 struct queue_stat {
        volatile unsigned long pkts;
        volatile unsigned long bytes;
        volatile unsigned long err_pkts;
 };
 
+struct queue_missed_stat {
+       /* last value retrieved from pcap */
+       unsigned int pcap;
+       /* stores values lost by pcap stop or rollover */
+       unsigned long mnemonic;
+       /* value on last reset */
+       unsigned long reset;
+};
+
 struct pcap_rx_queue {
        uint16_t port_id;
        uint16_t queue_id;
        struct rte_mempool *mb_pool;
        struct queue_stat rx_stat;
+       struct queue_missed_stat missed_stat;
        char name[PATH_MAX];
        char type[ETH_PCAP_ARG_MAXLEN];
+
+       /* Contains pre-generated packets to be looped through */
+       struct rte_ring *pkts;
 };
 
 struct pcap_tx_queue {
@@ -78,10 +95,11 @@ struct pmd_internals {
        struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
        struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
        char devargs[ETH_PCAP_ARG_MAXLEN];
-       struct ether_addr eth_addr;
+       struct rte_ether_addr eth_addr;
        int if_index;
        int single_iface;
        int phy_mac;
+       unsigned int infinite_rx;
 };
 
 struct pmd_process_private {
@@ -101,6 +119,17 @@ struct pmd_devargs {
        int phy_mac;
 };
 
+struct pmd_devargs_all {
+       struct pmd_devargs rx_queues;
+       struct pmd_devargs tx_queues;
+       int single_iface;
+       unsigned int is_tx_pcap;
+       unsigned int is_tx_iface;
+       unsigned int is_rx_pcap;
+       unsigned int is_rx_iface;
+       unsigned int infinite_rx;
+};
+
 static const char *valid_arguments[] = {
        ETH_PCAP_RX_PCAP_ARG,
        ETH_PCAP_TX_PCAP_ARG,
@@ -109,6 +138,7 @@ static const char *valid_arguments[] = {
        ETH_PCAP_TX_IFACE_ARG,
        ETH_PCAP_IFACE_ARG,
        ETH_PCAP_PHY_MAC_ARG,
+       ETH_PCAP_INFINITE_RX_ARG,
        NULL
 };
 
@@ -119,12 +149,62 @@ static struct rte_eth_link pmd_link = {
                .link_autoneg = ETH_LINK_FIXED,
 };
 
-static int eth_pcap_logtype;
+RTE_LOG_REGISTER(eth_pcap_logtype, pmd.net.pcap, NOTICE);
 
 #define PMD_LOG(level, fmt, args...) \
        rte_log(RTE_LOG_ ## level, eth_pcap_logtype, \
                "%s(): " fmt "\n", __func__, ##args)
 
+static struct queue_missed_stat*
+queue_missed_stat_update(struct rte_eth_dev *dev, unsigned int qid)
+{
+       struct pmd_internals *internals = dev->data->dev_private;
+       struct queue_missed_stat *missed_stat =
+                       &internals->rx_queue[qid].missed_stat;
+       const struct pmd_process_private *pp = dev->process_private;
+       pcap_t *pcap = pp->rx_pcap[qid];
+       struct pcap_stat stat;
+
+       if (!pcap || (pcap_stats(pcap, &stat) != 0))
+               return missed_stat;
+
+       /* rollover check - best effort fixup assuming single rollover */
+       if (stat.ps_drop < missed_stat->pcap)
+               missed_stat->mnemonic += UINT_MAX;
+       missed_stat->pcap = stat.ps_drop;
+
+       return missed_stat;
+}
+
+static void
+queue_missed_stat_on_stop_update(struct rte_eth_dev *dev, unsigned int qid)
+{
+       struct queue_missed_stat *missed_stat =
+                       queue_missed_stat_update(dev, qid);
+
+       missed_stat->mnemonic += missed_stat->pcap;
+       missed_stat->pcap = 0;
+}
+
+static void
+queue_missed_stat_reset(struct rte_eth_dev *dev, unsigned int qid)
+{
+       struct queue_missed_stat *missed_stat =
+                       queue_missed_stat_update(dev, qid);
+
+       missed_stat->reset = missed_stat->pcap;
+       missed_stat->mnemonic = 0;
+}
+
+static unsigned long
+queue_missed_stat_get(struct rte_eth_dev *dev, unsigned int qid)
+{
+       const struct queue_missed_stat *missed_stat =
+                       queue_missed_stat_update(dev, qid);
+
+       return missed_stat->pcap + missed_stat->mnemonic - missed_stat->reset;
+}
+
 static int
 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
                const u_char *data, uint16_t data_len)
@@ -163,19 +243,41 @@ eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
        return mbuf->nb_segs;
 }
 
-/* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
-static void
-eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
+static uint16_t
+eth_pcap_rx_infinite(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 {
-       uint16_t data_len = 0;
+       int i;
+       struct pcap_rx_queue *pcap_q = queue;
+       uint32_t rx_bytes = 0;
+
+       if (unlikely(nb_pkts == 0))
+               return 0;
 
-       while (mbuf) {
-               rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
-                       mbuf->data_len);
+       if (rte_pktmbuf_alloc_bulk(pcap_q->mb_pool, bufs, nb_pkts) != 0)
+               return 0;
 
-               data_len += mbuf->data_len;
-               mbuf = mbuf->next;
+       for (i = 0; i < nb_pkts; i++) {
+               struct rte_mbuf *pcap_buf;
+               int err = rte_ring_dequeue(pcap_q->pkts, (void **)&pcap_buf);
+               if (err)
+                       return i;
+
+               rte_memcpy(rte_pktmbuf_mtod(bufs[i], void *),
+                               rte_pktmbuf_mtod(pcap_buf, void *),
+                               pcap_buf->data_len);
+               bufs[i]->data_len = pcap_buf->data_len;
+               bufs[i]->pkt_len = pcap_buf->pkt_len;
+               bufs[i]->port = pcap_q->port_id;
+               rx_bytes += pcap_buf->data_len;
+
+               /* Enqueue packet back on ring to allow infinite rx. */
+               rte_ring_enqueue(pcap_q->pkts, pcap_buf);
        }
+
+       pcap_q->rx_stat.pkts += i;
+       pcap_q->rx_stat.bytes += rx_bytes;
+
+       return i;
 }
 
 static uint16_t
@@ -188,7 +290,6 @@ eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
        struct rte_mbuf *mbuf;
        struct pcap_rx_queue *pcap_q = queue;
        uint16_t num_rx = 0;
-       uint16_t buf_size;
        uint32_t rx_bytes = 0;
        pcap_t *pcap;
 
@@ -211,11 +312,7 @@ eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
                if (unlikely(mbuf == NULL))
                        break;
 
-               /* Now get the space available for data in the mbuf */
-               buf_size = rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
-                               RTE_PKTMBUF_HEADROOM;
-
-               if (header.caplen <= buf_size) {
+               if (header.caplen <= rte_pktmbuf_tailroom(mbuf)) {
                        /* pcap packet will fit in the mbuf, can copy it */
                        rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
                                        header.caplen);
@@ -232,6 +329,11 @@ eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
                }
 
                mbuf->pkt_len = (uint16_t)header.caplen;
+               *RTE_MBUF_DYNFIELD(mbuf, timestamp_dynfield_offset,
+                       rte_mbuf_timestamp_t *) =
+                               (uint64_t)header.ts.tv_sec * 1000000 +
+                               header.ts.tv_usec;
+               mbuf->ol_flags |= timestamp_rx_dynflag;
                mbuf->port = pcap_q->port_id;
                bufs[num_rx] = mbuf;
                num_rx++;
@@ -243,6 +345,16 @@ eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
        return num_rx;
 }
 
+static uint16_t
+eth_null_rx(void *queue __rte_unused,
+               struct rte_mbuf **bufs __rte_unused,
+               uint16_t nb_pkts __rte_unused)
+{
+       return 0;
+}
+
+#define NSEC_PER_SEC   1000000000L
+
 static inline void
 calculate_timestamp(struct timeval *ts) {
        uint64_t cycles;
@@ -250,8 +362,14 @@ calculate_timestamp(struct timeval *ts) {
 
        cycles = rte_get_timer_cycles() - start_cycles;
        cur_time.tv_sec = cycles / hz;
-       cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
-       timeradd(&start_time, &cur_time, ts);
+       cur_time.tv_usec = (cycles % hz) * NSEC_PER_SEC / hz;
+
+       ts->tv_sec = start_time.tv_sec + cur_time.tv_sec;
+       ts->tv_usec = start_time.tv_usec + cur_time.tv_usec;
+       if (ts->tv_usec >= NSEC_PER_SEC) {
+               ts->tv_usec -= NSEC_PER_SEC;
+               ts->tv_sec += 1;
+       }
 }
 
 /*
@@ -268,6 +386,8 @@ eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
        uint32_t tx_bytes = 0;
        struct pcap_pkthdr header;
        pcap_dumper_t *dumper;
+       unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
+       size_t len, caplen;
 
        pp = rte_eth_devices[dumper_q->port_id].process_private;
        dumper = pp->tx_dumper[dumper_q->queue_id];
@@ -279,31 +399,24 @@ eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
         * dumper */
        for (i = 0; i < nb_pkts; i++) {
                mbuf = bufs[i];
-               calculate_timestamp(&header.ts);
-               header.len = mbuf->pkt_len;
-               header.caplen = header.len;
-
-               if (likely(mbuf->nb_segs == 1)) {
-                       pcap_dump((u_char *)dumper, &header,
-                                 rte_pktmbuf_mtod(mbuf, void*));
-               } else {
-                       if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
-                               eth_pcap_gather_data(tx_pcap_data, mbuf);
-                               pcap_dump((u_char *)dumper, &header,
-                                         tx_pcap_data);
-                       } else {
-                               PMD_LOG(ERR,
-                                       "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
-                                       mbuf->pkt_len,
-                                       ETHER_MAX_JUMBO_FRAME_LEN);
-
-                               rte_pktmbuf_free(mbuf);
-                               break;
-                       }
+               len = caplen = rte_pktmbuf_pkt_len(mbuf);
+               if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
+                               len > sizeof(temp_data))) {
+                       caplen = sizeof(temp_data);
                }
 
+               calculate_timestamp(&header.ts);
+               header.len = len;
+               header.caplen = caplen;
+               /* rte_pktmbuf_read() returns a pointer to the data directly
+                * in the mbuf (when the mbuf is contiguous) or, otherwise,
+                * a pointer to temp_data after copying into it.
+                */
+               pcap_dump((u_char *)dumper, &header,
+                       rte_pktmbuf_read(mbuf, 0, caplen, temp_data));
+
                num_tx++;
-               tx_bytes += mbuf->pkt_len;
+               tx_bytes += caplen;
                rte_pktmbuf_free(mbuf);
        }
 
@@ -317,7 +430,31 @@ eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
        dumper_q->tx_stat.bytes += tx_bytes;
        dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 
-       return num_tx;
+       return nb_pkts;
+}
+
+/*
+ * Callback to handle dropping packets in the infinite rx case.
+ */
+static uint16_t
+eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+       unsigned int i;
+       uint32_t tx_bytes = 0;
+       struct pcap_tx_queue *tx_queue = queue;
+
+       if (unlikely(nb_pkts == 0))
+               return 0;
+
+       for (i = 0; i < nb_pkts; i++) {
+               tx_bytes += bufs[i]->pkt_len;
+               rte_pktmbuf_free(bufs[i]);
+       }
+
+       tx_queue->tx_stat.pkts += nb_pkts;
+       tx_queue->tx_stat.bytes += tx_bytes;
+
+       return i;
 }
 
 /*
@@ -334,6 +471,8 @@ eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
        uint16_t num_tx = 0;
        uint32_t tx_bytes = 0;
        pcap_t *pcap;
+       unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
+       size_t len;
 
        pp = rte_eth_devices[tx_queue->port_id].process_private;
        pcap = pp->tx_pcap[tx_queue->queue_id];
@@ -343,39 +482,34 @@ eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 
        for (i = 0; i < nb_pkts; i++) {
                mbuf = bufs[i];
-
-               if (likely(mbuf->nb_segs == 1)) {
-                       ret = pcap_sendpacket(pcap,
-                                       rte_pktmbuf_mtod(mbuf, u_char *),
-                                       mbuf->pkt_len);
-               } else {
-                       if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
-                               eth_pcap_gather_data(tx_pcap_data, mbuf);
-                               ret = pcap_sendpacket(pcap,
-                                               tx_pcap_data, mbuf->pkt_len);
-                       } else {
-                               PMD_LOG(ERR,
-                                       "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
-                                       mbuf->pkt_len,
-                                       ETHER_MAX_JUMBO_FRAME_LEN);
-
-                               rte_pktmbuf_free(mbuf);
-                               break;
-                       }
+               len = rte_pktmbuf_pkt_len(mbuf);
+               if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
+                               len > sizeof(temp_data))) {
+                       PMD_LOG(ERR,
+                               "Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
+                               len, sizeof(temp_data));
+                       rte_pktmbuf_free(mbuf);
+                       continue;
                }
 
+               /* rte_pktmbuf_read() returns a pointer to the data directly
+                * in the mbuf (when the mbuf is contiguous) or, otherwise,
+                * a pointer to temp_data after copying into it.
+                */
+               ret = pcap_sendpacket(pcap,
+                       rte_pktmbuf_read(mbuf, 0, len, temp_data), len);
                if (unlikely(ret != 0))
                        break;
                num_tx++;
-               tx_bytes += mbuf->pkt_len;
+               tx_bytes += len;
                rte_pktmbuf_free(mbuf);
        }
 
        tx_queue->tx_stat.pkts += num_tx;
        tx_queue->tx_stat.bytes += tx_bytes;
-       tx_queue->tx_stat.err_pkts += nb_pkts - num_tx;
+       tx_queue->tx_stat.err_pkts += i - num_tx;
 
-       return num_tx;
+       return i;
 }
 
 /*
@@ -415,7 +549,8 @@ open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
         * with pcap_dump_open(). We create big enough an Ethernet
         * pcap holder.
         */
-       tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
+       tx_pcap = pcap_open_dead_with_tstamp_precision(DLT_EN10MB,
+                       RTE_ETH_PCAP_SNAPSHOT_LEN, PCAP_TSTAMP_PRECISION_NANO);
        if (tx_pcap == NULL) {
                PMD_LOG(ERR, "Couldn't create dead pcap");
                return -1;
@@ -447,6 +582,24 @@ open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
        return 0;
 }
 
+static uint64_t
+count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
+{
+       const u_char *packet;
+       struct pcap_pkthdr header;
+       uint64_t pcap_pkt_count = 0;
+
+       while ((packet = pcap_next(*pcap, &header)))
+               pcap_pkt_count++;
+
+       /* The pcap is reopened so it can be used as normal later. */
+       pcap_close(*pcap);
+       *pcap = NULL;
+       open_single_rx_pcap(pcap_q->name, pcap);
+
+       return pcap_pkt_count;
+}
+
 static int
 eth_dev_start(struct rte_eth_dev *dev)
 {
@@ -520,7 +673,7 @@ status_up:
  * Is the only place for us to close all the tx streams dumpers.
  * If not called the dumpers will be flushed within each tx burst.
  */
-static void
+static int
 eth_dev_stop(struct rte_eth_dev *dev)
 {
        unsigned int i;
@@ -529,9 +682,12 @@ eth_dev_stop(struct rte_eth_dev *dev)
 
        /* Special iface case. Single pcap is open and shared between tx/rx. */
        if (internals->single_iface) {
-               pcap_close(pp->tx_pcap[0]);
-               pp->tx_pcap[0] = NULL;
-               pp->rx_pcap[0] = NULL;
+               queue_missed_stat_on_stop_update(dev, 0);
+               if (pp->tx_pcap[0] != NULL) {
+                       pcap_close(pp->tx_pcap[0]);
+                       pp->tx_pcap[0] = NULL;
+                       pp->rx_pcap[0] = NULL;
+               }
                goto status_down;
        }
 
@@ -549,6 +705,7 @@ eth_dev_stop(struct rte_eth_dev *dev)
 
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                if (pp->rx_pcap[i] != NULL) {
+                       queue_missed_stat_on_stop_update(dev, i);
                        pcap_close(pp->rx_pcap[i]);
                        pp->rx_pcap[i] = NULL;
                }
@@ -562,6 +719,8 @@ status_down:
                dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 
        dev->data->dev_link.link_status = ETH_LINK_DOWN;
+
+       return 0;
 }
 
 static int
@@ -570,7 +729,7 @@ eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
        return 0;
 }
 
-static void
+static int
 eth_dev_info(struct rte_eth_dev *dev,
                struct rte_eth_dev_info *dev_info)
 {
@@ -582,6 +741,8 @@ eth_dev_info(struct rte_eth_dev *dev,
        dev_info->max_rx_queues = dev->data->nb_rx_queues;
        dev_info->max_tx_queues = dev->data->nb_tx_queues;
        dev_info->min_rx_bufsize = 0;
+
+       return 0;
 }
 
 static int
@@ -589,6 +750,7 @@ eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 {
        unsigned int i;
        unsigned long rx_packets_total = 0, rx_bytes_total = 0;
+       unsigned long rx_missed_total = 0;
        unsigned long tx_packets_total = 0, tx_bytes_total = 0;
        unsigned long tx_packets_err_total = 0;
        const struct pmd_internals *internal = dev->data->dev_private;
@@ -599,20 +761,21 @@ eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
                stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
                rx_packets_total += stats->q_ipackets[i];
                rx_bytes_total += stats->q_ibytes[i];
+               rx_missed_total += queue_missed_stat_get(dev, i);
        }
 
        for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
                        i < dev->data->nb_tx_queues; i++) {
                stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
                stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
-               stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
                tx_packets_total += stats->q_opackets[i];
                tx_bytes_total += stats->q_obytes[i];
-               tx_packets_err_total += stats->q_errors[i];
+               tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
        }
 
        stats->ipackets = rx_packets_total;
        stats->ibytes = rx_bytes_total;
+       stats->imissed = rx_missed_total;
        stats->opackets = tx_packets_total;
        stats->obytes = tx_bytes_total;
        stats->oerrors = tx_packets_err_total;
@@ -620,7 +783,7 @@ eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
        return 0;
 }
 
-static void
+static int
 eth_stats_reset(struct rte_eth_dev *dev)
 {
        unsigned int i;
@@ -629,6 +792,7 @@ eth_stats_reset(struct rte_eth_dev *dev)
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                internal->rx_queue[i].rx_stat.pkts = 0;
                internal->rx_queue[i].rx_stat.bytes = 0;
+               queue_missed_stat_reset(dev, i);
        }
 
        for (i = 0; i < dev->data->nb_tx_queues; i++) {
@@ -636,11 +800,58 @@ eth_stats_reset(struct rte_eth_dev *dev)
                internal->tx_queue[i].tx_stat.bytes = 0;
                internal->tx_queue[i].tx_stat.err_pkts = 0;
        }
+
+       return 0;
 }
 
-static void
-eth_dev_close(struct rte_eth_dev *dev __rte_unused)
+static inline void
+infinite_rx_ring_free(struct rte_ring *pkts)
+{
+       struct rte_mbuf *bufs;
+
+       while (!rte_ring_dequeue(pkts, (void **)&bufs))
+               rte_pktmbuf_free(bufs);
+
+       rte_ring_free(pkts);
+}
+
+static int
+eth_dev_close(struct rte_eth_dev *dev)
 {
+       unsigned int i;
+       struct pmd_internals *internals = dev->data->dev_private;
+
+       PMD_LOG(INFO, "Closing pcap ethdev on NUMA socket %d",
+                       rte_socket_id());
+
+       eth_dev_stop(dev);
+
+       rte_free(dev->process_private);
+
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return 0;
+
+       /* Device wide flag, but cleanup must be performed per queue. */
+       if (internals->infinite_rx) {
+               for (i = 0; i < dev->data->nb_rx_queues; i++) {
+                       struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
+
+                       /*
+                        * 'pcap_q->pkts' can be NULL if 'eth_dev_close()'
+                        * called before 'eth_rx_queue_setup()' has been called
+                        */
+                       if (pcap_q->pkts == NULL)
+                               continue;
+
+                       infinite_rx_ring_free(pcap_q->pkts);
+               }
+       }
+
+       if (internals->phy_mac == 0)
+               /* not dynamically allocated, must not be freed */
+               dev->data->mac_addrs = NULL;
+
+       return 0;
 }
 
 static void
@@ -671,6 +882,63 @@ eth_rx_queue_setup(struct rte_eth_dev *dev,
        pcap_q->queue_id = rx_queue_id;
        dev->data->rx_queues[rx_queue_id] = pcap_q;
 
+       if (internals->infinite_rx) {
+               struct pmd_process_private *pp;
+               char ring_name[NAME_MAX];
+               static uint32_t ring_number;
+               uint64_t pcap_pkt_count = 0;
+               struct rte_mbuf *bufs[1];
+               pcap_t **pcap;
+
+               pp = rte_eth_devices[pcap_q->port_id].process_private;
+               pcap = &pp->rx_pcap[pcap_q->queue_id];
+
+               if (unlikely(*pcap == NULL))
+                       return -ENOENT;
+
+               pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
+
+               snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu32,
+                               ring_number);
+
+               pcap_q->pkts = rte_ring_create(ring_name,
+                               rte_align64pow2(pcap_pkt_count + 1), 0,
+                               RING_F_SP_ENQ | RING_F_SC_DEQ);
+               ring_number++;
+               if (!pcap_q->pkts)
+                       return -ENOENT;
+
+               /* Fill ring with packets from PCAP file one by one. */
+               while (eth_pcap_rx(pcap_q, bufs, 1)) {
+                       /* Check for multiseg mbufs. */
+                       if (bufs[0]->nb_segs != 1) {
+                               infinite_rx_ring_free(pcap_q->pkts);
+                               PMD_LOG(ERR,
+                                       "Multiseg mbufs are not supported in infinite_rx mode.");
+                               return -EINVAL;
+                       }
+
+                       rte_ring_enqueue_bulk(pcap_q->pkts,
+                                       (void * const *)bufs, 1, NULL);
+               }
+
+               if (rte_ring_count(pcap_q->pkts) < pcap_pkt_count) {
+                       infinite_rx_ring_free(pcap_q->pkts);
+                       PMD_LOG(ERR,
+                               "Not enough mbufs to accommodate packets in pcap file. "
+                               "At least %" PRIu64 " mbufs per queue is required.",
+                               pcap_pkt_count);
+                       return -EINVAL;
+               }
+
+               /*
+                * Reset the stats for this queue since eth_pcap_rx calls above
+                * didn't result in the application receiving packets.
+                */
+               pcap_q->rx_stat.pkts = 0;
+               pcap_q->rx_stat.bytes = 0;
+       }
+
        return 0;
 }
 
@@ -908,7 +1176,19 @@ select_phy_mac(const char *key __rte_unused, const char *value,
        return 0;
 }
 
-static struct rte_vdev_driver pmd_pcap_drv;
+static int
+get_infinite_rx_arg(const char *key __rte_unused,
+               const char *value, void *extra_args)
+{
+       if (extra_args) {
+               const int infinite_rx = atoi(value);
+               int *enable_infinite_rx = extra_args;
+
+               if (infinite_rx > 0)
+                       *enable_infinite_rx = 1;
+       }
+       return 0;
+}
 
 static int
 pmd_init_internals(struct rte_vdev_device *vdev,
@@ -953,7 +1233,7 @@ pmd_init_internals(struct rte_vdev_device *vdev,
         * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
         * where the middle 4 characters are converted to hex.
         */
-       (*internals)->eth_addr = (struct ether_addr) {
+       (*internals)->eth_addr = (struct rte_ether_addr) {
                .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
        };
        (*internals)->phy_mac = 0;
@@ -962,6 +1242,9 @@ pmd_init_internals(struct rte_vdev_device *vdev,
        data->nb_tx_queues = (uint16_t)nb_tx_queues;
        data->dev_link = pmd_link;
        data->mac_addrs = &(*internals)->eth_addr;
+       data->promiscuous = 1;
+       data->all_multicast = 1;
+       data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
 
        /*
         * NOTE: we'll replace the data element, of originally allocated
@@ -993,7 +1276,7 @@ eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
                return -1;
        }
 
-       mac_addrs = rte_zmalloc_socket(NULL, ETHER_ADDR_LEN, 0, numa_node);
+       mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
        if (!mac_addrs) {
                close(if_fd);
                return -1;
@@ -1002,7 +1285,7 @@ eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
        PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
        eth_dev->data->mac_addrs = mac_addrs;
        rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
-                       ifr.ifr_hwaddr.sa_data, ETHER_ADDR_LEN);
+                       ifr.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
 
        close(if_fd);
 
@@ -1040,7 +1323,7 @@ eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
        ifm = (struct if_msghdr *)buf;
        sdl = (struct sockaddr_dl *)(ifm + 1);
 
-       mac_addrs = rte_zmalloc_socket(NULL, ETHER_ADDR_LEN, 0, numa_node);
+       mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
        if (!mac_addrs) {
                rte_free(buf);
                return -1;
@@ -1049,7 +1332,7 @@ eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
        PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
        eth_dev->data->mac_addrs = mac_addrs;
        rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
-                       LLADDR(sdl), ETHER_ADDR_LEN);
+                       LLADDR(sdl), RTE_ETHER_ADDR_LEN);
 
        rte_free(buf);
 
@@ -1061,19 +1344,16 @@ eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
 
 static int
 eth_from_pcaps_common(struct rte_vdev_device *vdev,
-               struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
-               struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
+               struct pmd_devargs_all *devargs_all,
                struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
 {
        struct pmd_process_private *pp;
+       struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
+       struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
+       const unsigned int nb_rx_queues = rx_queues->num_of_queue;
+       const unsigned int nb_tx_queues = tx_queues->num_of_queue;
        unsigned int i;
 
-       /* do some parameter checking */
-       if (rx_queues == NULL && nb_rx_queues > 0)
-               return -1;
-       if (tx_queues == NULL && nb_tx_queues > 0)
-               return -1;
-
        if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
                        eth_dev) < 0)
                return -1;
@@ -1103,16 +1383,16 @@ eth_from_pcaps_common(struct rte_vdev_device *vdev,
 
 static int
 eth_from_pcaps(struct rte_vdev_device *vdev,
-               struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
-               struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
-               int single_iface, unsigned int using_dumpers)
+               struct pmd_devargs_all *devargs_all)
 {
        struct pmd_internals *internals = NULL;
        struct rte_eth_dev *eth_dev = NULL;
+       struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
+       int single_iface = devargs_all->single_iface;
+       unsigned int infinite_rx = devargs_all->infinite_rx;
        int ret;
 
-       ret = eth_from_pcaps_common(vdev, rx_queues, nb_rx_queues,
-               tx_queues, nb_tx_queues, &internals, &eth_dev);
+       ret = eth_from_pcaps_common(vdev, devargs_all, &internals, &eth_dev);
 
        if (ret < 0)
                return ret;
@@ -1125,19 +1405,29 @@ eth_from_pcaps(struct rte_vdev_device *vdev,
 
                /* phy_mac arg is applied only only if "iface" devarg is provided */
                if (rx_queues->phy_mac) {
-                       int ret = eth_pcap_update_mac(rx_queues->queue[0].name,
-                                       eth_dev, vdev->device.numa_node);
-                       if (ret == 0)
+                       if (eth_pcap_update_mac(rx_queues->queue[0].name,
+                                       eth_dev, vdev->device.numa_node) == 0)
                                internals->phy_mac = 1;
                }
        }
 
-       eth_dev->rx_pkt_burst = eth_pcap_rx;
+       internals->infinite_rx = infinite_rx;
+       /* Assign rx ops. */
+       if (infinite_rx)
+               eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
+       else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
+                       single_iface)
+               eth_dev->rx_pkt_burst = eth_pcap_rx;
+       else
+               eth_dev->rx_pkt_burst = eth_null_rx;
 
-       if (using_dumpers)
+       /* Assign tx ops. */
+       if (devargs_all->is_tx_pcap)
                eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
-       else
+       else if (devargs_all->is_tx_iface || single_iface)
                eth_dev->tx_pkt_burst = eth_pcap_tx;
+       else
+               eth_dev->tx_pkt_burst = eth_tx_drop;
 
        rte_eth_dev_probing_finish(eth_dev);
        return 0;
@@ -1147,14 +1437,19 @@ static int
 pmd_pcap_probe(struct rte_vdev_device *dev)
 {
        const char *name;
-       unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
        struct rte_kvargs *kvlist;
        struct pmd_devargs pcaps = {0};
        struct pmd_devargs dumpers = {0};
        struct rte_eth_dev *eth_dev =  NULL;
        struct pmd_internals *internal;
-       int single_iface = 0;
-       int ret;
+       int ret = 0;
+
+       struct pmd_devargs_all devargs_all = {
+               .single_iface = 0,
+               .is_tx_pcap = 0,
+               .is_tx_iface = 0,
+               .infinite_rx = 0,
+       };
 
        name = rte_vdev_device_name(dev);
        PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
@@ -1163,6 +1458,13 @@ pmd_pcap_probe(struct rte_vdev_device *dev)
        start_cycles = rte_get_timer_cycles();
        hz = rte_get_timer_hz();
 
+       ret = rte_mbuf_dyn_rx_timestamp_register(&timestamp_dynfield_offset,
+                       &timestamp_rx_dynflag);
+       if (ret != 0) {
+               PMD_LOG(ERR, "Failed to register Rx timestamp field/flag");
+               return -1;
+       }
+
        if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
                eth_dev = rte_eth_dev_attach_secondary(name);
                if (!eth_dev) {
@@ -1202,7 +1504,7 @@ pmd_pcap_probe(struct rte_vdev_device *dev)
 
                dumpers.phy_mac = pcaps.phy_mac;
 
-               single_iface = 1;
+               devargs_all.single_iface = 1;
                pcaps.num_of_queue = 1;
                dumpers.num_of_queue = 1;
 
@@ -1210,36 +1512,94 @@ pmd_pcap_probe(struct rte_vdev_device *dev)
        }
 
        /*
-        * We check whether we want to open a RX stream from a real NIC or a
-        * pcap file
+        * We check whether we want to open a RX stream from a real NIC, a
+        * pcap file or open a dummy RX stream
         */
-       is_rx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
+       devargs_all.is_rx_pcap =
+               rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
+       devargs_all.is_rx_iface =
+               (rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) +
+                rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_IN_ARG)) ? 1 : 0;
        pcaps.num_of_queue = 0;
 
-       if (is_rx_pcap) {
+       devargs_all.is_tx_pcap =
+               rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
+       devargs_all.is_tx_iface =
+               rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
+       dumpers.num_of_queue = 0;
+
+       if (devargs_all.is_rx_pcap) {
+               /*
+                * We check whether we want to infinitely rx the pcap file.
+                */
+               unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
+                               ETH_PCAP_INFINITE_RX_ARG);
+
+               if (infinite_rx_arg_cnt == 1) {
+                       ret = rte_kvargs_process(kvlist,
+                                       ETH_PCAP_INFINITE_RX_ARG,
+                                       &get_infinite_rx_arg,
+                                       &devargs_all.infinite_rx);
+                       if (ret < 0)
+                               goto free_kvlist;
+                       PMD_LOG(INFO, "infinite_rx has been %s for %s",
+                                       devargs_all.infinite_rx ? "enabled" : "disabled",
+                                       name);
+
+               } else if (infinite_rx_arg_cnt > 1) {
+                       PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
+                                       "argument has been provided more than once "
+                                       "for %s", name);
+               }
+
                ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
                                &open_rx_pcap, &pcaps);
-       } else {
+       } else if (devargs_all.is_rx_iface) {
                ret = rte_kvargs_process(kvlist, NULL,
                                &rx_iface_args_process, &pcaps);
-       }
+       } else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
+               unsigned int i;
+
+               /* Count number of tx queue args passed before dummy rx queue
+                * creation so a dummy rx queue can be created for each tx queue
+                */
+               unsigned int num_tx_queues =
+                       (rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
+                       rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
 
+               PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
+
+               /* Creating a dummy rx queue for each tx queue passed */
+               for (i = 0; i < num_tx_queues; i++)
+                       ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
+                                       NULL);
+       } else {
+               PMD_LOG(ERR, "Error - No rx or tx queues provided");
+               ret = -ENOENT;
+       }
        if (ret < 0)
                goto free_kvlist;
 
        /*
-        * We check whether we want to open a TX stream to a real NIC or a
-        * pcap file
+        * We check whether we want to open a TX stream to a real NIC,
+        * a pcap file, or drop packets on tx
         */
-       is_tx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
-       dumpers.num_of_queue = 0;
-
-       if (is_tx_pcap)
+       if (devargs_all.is_tx_pcap) {
                ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
                                &open_tx_pcap, &dumpers);
-       else
+       } else if (devargs_all.is_tx_iface) {
                ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
                                &open_tx_iface, &dumpers);
+       } else {
+               unsigned int i;
+
+               PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
+
+               /* Add 1 dummy queue per rxq which counts and drops packets. */
+               for (i = 0; i < pcaps.num_of_queue; i++)
+                       ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
+                                       NULL);
+       }
 
        if (ret < 0)
                goto free_kvlist;
@@ -1276,7 +1636,7 @@ create_eth:
 
                eth_dev->process_private = pp;
                eth_dev->rx_pkt_burst = eth_pcap_rx;
-               if (is_tx_pcap)
+               if (devargs_all.is_tx_pcap)
                        eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
                else
                        eth_dev->tx_pkt_burst = eth_pcap_tx;
@@ -1285,8 +1645,10 @@ create_eth:
                goto free_kvlist;
        }
 
-       ret = eth_from_pcaps(dev, &pcaps, pcaps.num_of_queue, &dumpers,
-               dumpers.num_of_queue, single_iface, is_tx_pcap);
+       devargs_all.rx_queues = pcaps;
+       devargs_all.tx_queues = dumpers;
+
+       ret = eth_from_pcaps(dev, &devargs_all);
 
 free_kvlist:
        rte_kvargs_free(kvlist);
@@ -1297,28 +1659,16 @@ free_kvlist:
 static int
 pmd_pcap_remove(struct rte_vdev_device *dev)
 {
-       struct pmd_internals *internals = NULL;
        struct rte_eth_dev *eth_dev = NULL;
 
-       PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
-                       rte_socket_id());
-
        if (!dev)
                return -1;
 
-       /* reserve an ethdev entry */
        eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
        if (eth_dev == NULL)
-               return -1;
+               return 0; /* port already released */
 
-       if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
-               internals = eth_dev->data->dev_private;
-               if (internals != NULL && internals->phy_mac == 0)
-                       /* not dynamically allocated, must not be freed */
-                       eth_dev->data->mac_addrs = NULL;
-       }
-
-       rte_free(eth_dev->process_private);
+       eth_dev_close(eth_dev);
        rte_eth_dev_release_port(eth_dev);
 
        return 0;
@@ -1338,11 +1688,5 @@ RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
        ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
        ETH_PCAP_TX_IFACE_ARG "=<ifc> "
        ETH_PCAP_IFACE_ARG "=<ifc> "
-       ETH_PCAP_PHY_MAC_ARG "=<int>");
-
-RTE_INIT(eth_pcap_init_log)
-{
-       eth_pcap_logtype = rte_log_register("pmd.net.pcap");
-       if (eth_pcap_logtype >= 0)
-               rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);
-}
+       ETH_PCAP_PHY_MAC_ARG "=<int>"
+       ETH_PCAP_INFINITE_RX_ARG "=<0|1>");